Springer Series in Supply Chain Management

Albert Y. Ha
Christopher S. Tang Editors

Handbook of
Information
Exchange in
Supply Chain
Management
Springer Series in Supply Chain Management
Volume 5

Series Editor
Christopher S. Tang
University of California
Los Angeles, CA, USA

More information about this series at http://www.springer.com/series/13081

Albert Y. Ha • Christopher S. Tang
Editors

Handbook of Information
Exchange in Supply Chain
Management

123
Editors
Albert Y. Ha Christopher S. Tang
Hong Kong University of Science University of California
and Technology Los Angeles, CA, USA
Kowloon, Hong Kong

ISSN 2365-6395 ISSN 2365-6409 (electronic)

Springer Series in Supply Chain Management
ISBN 978-3-319-32439-5 ISBN 978-3-319-32441-8 (eBook)
DOI 10.1007/978-3-319-32441-8

Library of Congress Control Number: 2016943081

© Springer International Publishing Switzerland 2017

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG Switzerland
Preface

To mitigate the “bullwhip effect” along a supply chain, Lee et al. (1997) develop
stylized models so that they can generate analytical results to argue for the impor-
tance of information sharing among supply chain partners. Since then, many supply
chain academic research has focused on the interaction between information sharing
and operations planning. For example, Lee et al. (2000) investigate further about
the conditions under which sharing information about market demand is beneficial;
and Aviv (2001) examines the benefits of collaborative forecasting (via sharing in-
formation about a firm’s demand forecast with other firms).
On September 8, 2015, our simple search of the keywords “information sharing”
and “supply chain” through Google Scholar yields 42,700 articles. This search result
reveals that many researchers are interested in several fundamental questions about
information sharing in supply chains. For instance, what is the magnitude of the
bullwhip effect observed in practice? Do different industries experience different
magnitudes of the bullwhip effect? Besides information about demand and fore-
casts, should firms share information about its capacity, inventory level, and product
development with its supply chain partner? How do firms facilitate collaborative
forecasting in practice?
To gain a better understanding about these questions, we have invited leading
scholars who have shaped their respective fields of research. This book contains
the state-of-the art research in operations management that deals with information
sharing in supply chains. It consists of comprehensive surveys of empirical and
analytical studies and novel models of various problems arising from information
asymmetry so that there is a need to exchange information.
We enjoyed the experience working on this book, and we hope that this book will
stimulate further work in this exciting area of research.

Kowloon, Hong Kong Albert Y. Ha

Los Angeles, CA, USA Christopher S. Tang
October 2015

v
vi Preface

References

Aviv Y (2001) The effect of collaborative forecasting on supply chain performance. Manag Sci
47(10):1326–1343
Lee HL, Padmanabhan V, Whang S (1997) Information distortion in a supply chain: the bullwhip
effect. Manag Sci 43(4):546–558
Lee HL, So KC, Tang CS (2000) The value of information sharing in a two-level supply chain.
Manag Sci 46(5):626–643
Introduction

In the instant classic “Bullwhip” paper, Lee et al. (1997) suggested that firms can
mitigate the bullwhip effect if demand information is shared among supply chain
partners. Since then, the rapid advancement in Information and Communication
Technology (ICT) has enabled supply chain partners to explore different mecha-
nisms to share different types of information (product design, production capacity,
inventory status, shipment status, realized demand, and forecasted demand). At the
same time, the research community expressed interest in evaluating the benefits of
sharing different types of information by using different mechanisms (including in-
centive contracts). Based on our Scopus search, there were over 1400 journal articles
published between 1997 and 2014 that contain keywords “information sharing” and
“supply chain.” (See the Fig. 1 below for details.) This finding revealed that infor-
mation sharing in supply chain management continued to be an important research
area of interest. This observation motivated us to invite leading scholars in this area
to share their perspectives through comprehensive surveys or novel models so that
our community can gain a clear understanding about the state-of-the-art research in
this area.

Overall Structure

This book is comprised of 17 chapters that are divided into 4 sections. The first sec-
tion (Chaps. 1–3) provides comprehensive reviews of research literature that deal
with different issues arising from information sharing in supply chains. The sec-
ond section (Chaps. 4–9) investigates different types of incentive contracts that are
intended to enable the principal to entice the agents to behave in a certain manner
when certain information about the agents are not perfectly known to the princi-
pal. The third section (Chaps. 10–13) examines situations when a firm can credibly
convey unverifiable information (e.g., demand forecast) via signaling or cheap talk

vii
viii Introduction

Fig. 1 Number of published journal articles with keywords “information sharing” and “supply
chain” (from 1997 to 2014)

to the recipients (e.g., customers). Finally, the fourth section (Chaps. 14–17) ana-
lyzes how certain issues (e.g., competition, trust, etc.) can affect the incentive for
information sharing.

Chapter Highlights

Part I: Value of Information Sharing

In Chap. 1, Li Chen and Hau Lee review both theoretical and empirical research
studies that are intended to examine the following questions: What causes the bull-
whip effect in a supply chain? What is the magnitude of the bullwhip effect in
practice? In addition, they identify approaches that are necessary to measure and
evaluate the bullwhip effect properly. In Chap. 2, Justin Ren assesses different em-
pirical studies of information sharing in supply chains. Specifically, he classifies the
relevant literature according to the type of information (realized customer demand,
demand forecasts, inventory status, and product development) that firms can share
in order to improve various supply chain performance measures. By focusing on
the issue of information sharing in retail supply chains, Mümin Kurtuluş in Chap. 3
reviews the existing literature on collaborative forecasting (CF): a process that en-
ables supply chain partners to share their individual demand forecasts to develop a
“joint” demand forecast. By highlighting the models that capture the essence of CF,
he summarizes the key findings and insights about the value of CF.
Part I sets the stage by reviewing the literature that examines the following fun-
damental questions: What causes the bullwhip effect? What is the proper way to
measure the bullwhip effect? How can firms use the shared information to improve
supply chain performance? What type of information should a firm share with its
Introduction ix

supply chain partners? However, due to self-interest, a firm is reluctant to share or

to reveal private information. Therefore, how can one firm provide incentives for
other firms to either share information truthfully or to behave in a certain manner?
This is the focus of the remainder of this book.

Part II: Contracting Under Asymmetric Information

In this section, different contributing authors analyze different types of incentive

contracts arising from different contexts. First, motivated by dissatisfaction with the
level of after-sales support from key suppliers, many firms are considering chang-
ing the service contracts from the traditional resource-based contracts (RBC) to
performance-based contracts (PBC). In Chap. 4, Sang-Hyun Kim, Morris Cohen,
and Serguei Netessine investigate the value of performance-based contracting (PBC)
in the context of after-sales product support services. Specifically, under RBC, the
price of the after-sales service is based on pre specified unit prices of the service
parts, labor, and other consumable resources. However, under PBC, the compensa-
tion is based on the actual availability of the product realized by the customer. By
considering the case when the supplier can exert costly effort to improve product re-
liability, the authors find that PBC is more cost-efficient to the customer especially
because PBC incentivizes the supplier to invest more in reliability. Also, PBC can
generate maximum benefit when the spare parts are fully owned by the supplier.
In the operations management literature, most analytical models tend to focus
on the dynamics between one customer and one supplier. However, in the context
of project management, many projects involve multiple suppliers (agents) with un-
observable efforts. Because of potential free riding, it is challenging for the cus-
tomer (principal) to develop effective incentive contracts that yield maximum pay-
off. George Georgiadis and Christopher Tang tackle this problem in Chap. 5 by con-
sidering a team dynamics problem in which multiple agents collaborate over time to
complete a project. They show that, when the size of project is exogenously given,
it is optimal for the customer to compensate the agents only upon completion of the
project. In other words, the “pay as you go” contract is not optimal. Also, when the
size of the project is endogenous and when the customer cannot commit to the size
of the project, they find that the customer is better off delegating the decision rights
over the project size to the agents.
In project management (PM), customers do not have perfect information about
the capability and the cost efficiency of different potential contractors. As such, how
can a customer develop incentive contracts to entice the contractor to exert costly
effort so that the project can be completed in a time-efficient and cost-effective man-
ner? Recently, a novel reverse auction idea has emerged under which each contractor
is invited to submit a bid that consists of two elements: a quoted price and a quote
completion time. In this two-bid auction, the winning bid is the one that yields the
lowest composite score (i.e., a weighted average of the quoted price and the quoted
completion time). However, the winning contractor is subject to a penalty if the
x Introduction

project is completed late. Christopher Tang, Kairen Zhang, and Sean Zhou evalu-
ate this time-based incentive contract in Chap. 6 for the case when the amount of the
work involved is inherent uncertain. They show that this two-bid auction is efficient;
however, they find that a simple auction mechanism can yield the same benefit.
Rapid new product development (NPD) is an important strategy for a firm to sus-
tain its revenue growth. Because NPD projects often involve internal stakeholders
and external partners, a firm needs to develop contracts to align the incentives of
different agents. For instance, if the incentive is based on the number of new prod-
uct launches in a year, then the NPD program manager would focus on developing
incremental products that can be launched quickly. Similarly, if the development in-
volves the collaboration of two different firms, then the free-riding issue will emerge
so that both firms will under invest in the NPD project. In Chap. 7, Sameer Hasija
and Shantanu Bhattacharya review the existing NPD literature that deals with in-
centive contracts for coordinating the agents’ efforts to attain higher profits. They
also discuss different types of incentive contracts examined in the literature that are
intended to deal with the issue of information asymmetry, the timing of different
decisions, etc.
Most supply contract models tend to focus on ways to govern interactions be-
tween a manufacturer and a supplier. In the OM literature, it is well known that the
buy back contracts and revenue-sharing contracts can coordinate the supply chain
by sharing the demand risks. Volodymyr Babich and Zhibin Yang examine the is-
sue of supply disruptions (e.g., bankruptcy) and the role of procurement contracts
to mitigate this form of supply risks in Chap. 8. Specifically, they discuss how firms
deal with the issues of supply chain risks and contracting in practice. Also, they
present various procurement contracts that can enable firms to cope with various
types of supply chain risks.
In the final chapter of Part II (Chap. 9), Guoming Lai and Wenqiang Xiao con-
sider a situation in which a manufacturer sells its product through a retailer who
engages in two types of activities: collect market information for developing more
accurate demand forecast, and generate sales for creating more revenue. When both
activities require costly efforts, what kind of incentive contract should the manufac-
turer offer to the retailer? The authors compare the performance of forecast-based
contracts and the traditional linear contracts. They show that there are situations
under which forecast-based contracts dominate linear contracts.

Part III: Information Signaling and Cheap Talk

In this section, several contributing authors investigate how a firm can credibly con-
vey private and unverifiable information to outside parties such as its supply chain
partners, competitors, customers, and investors. This is done by either signaling (i.e.,
undertaking costly actions such as inventory or committing on a contract) or cheap
talk (i.e., costless communication).
Introduction xi

In Chap. 10, Mehmet Gümüş considers the issue of signaling information about
demand and supply in a supply chain with two suppliers competing to sell to a single
buyer. The suppliers differ in reliability and cost. The author obtains conditions
under which it is beneficial for a firm to share private demand or supply information
with other firms in the supply chain. He shows that quantity flexible contract and
price and quantity guarantee contract can be used as an instrument for signaling
information. He also evaluates how the provision of these contracts can affect a
firm’s performance.
In Chap. 11, Guoming Lai and Wenqiang Xiao study a firm’s problem of convey-
ing unverifiable demand information to investors so that they can properly determine
the firm’s market valuation. They show that when the firm uses inventory to signal
demand information, the inventory decision is distorted which leads to operational
inefficiency. They also show that it is possible for a supplier to design a contract
menu that allows the firm to signal demand information via its contract choice deci-
sion, thus restoring inventory to the efficient level.
Gad Allon and Achal Bassamboo in Chap. 12 examine a retailer’s strategy of
disclosing inventory availability information to customers who are strategic in pro-
cessing the information and making their purchasing decisions. They show that the
retailer cannot credibly convey any information when customers are homogenous
and establish the conditions under which the retailer can do so to influence customer
behavior when they are heterogeneous.
In Chap. 13, Noam Shamir and Hyoduk Shin investigate how firms in a supply
chain can exchange information using cheap talk under linear wholesale price. In
a single supply chain, they show that a retailer cannot credibly share unverifiable
demand forecast with a manufacturer who makes either capacity or wholesale price
decision. However, when the manufacturer makes both decisions, it is possible for
credible information exchange to occur. In two competing supply chains, even when
each manufacturer makes only capacity decision, the authors show that credible in-
formation exchange can occur if a retailer publicly announces the demand forecast.
These results hold because the incentives for inflating and deflating demand forecast
balance each other.

Part IV: Incentives for Information Sharing

As shown in the previous sections, the benefits of information sharing to a supply

chain are well documented. However, firms may not want to share information be-
cause of self-interests. This section considers factors that influence the incentive for
a firm to share information in a supply chain.
One major hurdle to information sharing is the lack of trust between supply chain
partners. Özalp Özer and Karen Zhang in Chap. 14 examine the role of trust and
trustworthiness in facilitating credible information sharing in a supply chain. Based
on some recent research findings, they develop a set of principles that can help firms
establish trusting and cooperative supply chain relationships.
xii Introduction

Another major hurdle to information sharing is the fear of leaking information to

unintended parties such as the competitors. In Chap. 15, Guangwen Kong, Sampath
Rajagopalan, and Hao Zhang review the literature on information leakage in supply
chains. They propose a framework that classifies some existing papers according to
several factors such as the timing of information sharing decision, how information
is leaked, types of competition, etc. They highlight the main findings of these pa-
pers, provide a summary of the key insights from the literature, and offer several
directions for future research.
In Chap. 16, Anthony Dukes, Esther Gal-Or, and Tansev Geylani consider bilat-
eral information sharing in a supply chain with a manufacturer selling to one or
more retailers. They develop a modeling framework that allows both the manufac-
turer and the retailer to have private and imperfect information about the market
demand. They analyze the effect of sharing either the manufacturer’s or the retail-
ers’ private information and conclude that the manufacturer generally benefits from
information sharing at the expenses of the retailers and consumers.
Albert Ha and Hongtao Zhang in Chap. 17 explore how non linear production
cost and competition impact the incentive for demand information sharing in a sup-
ply chain. The authors first investigate the role of non linear production cost in
determining the value of information sharing in a single supply chain. Then they
extend their analysis to the case of two competing supply chains, demonstrate that
information sharing now has a competitive effect due to the rival chain’s reaction,
and show how this competitive effect depends on non linear production cost and
type of competition (price or quantity). They also discuss how the main insights can
be applied to the many-to-one and two-to-one supply chain structures.

References

Lee HL, Padmanabhan V, Whang S (1997) Information distortion in a supply chain: the bullwhip
effect. Manag Sci 43(4):546–558
Acknowledgements

This book cannot exist without the strong commitment from our colleagues. On this
note, we are grateful to each of the contributing authors for sharing their cutting-
edge research with us (see table below). Also, we are indebted to Mirko Janc for
typesetting each chapter beautifully and expeditiously. Of course, we are responsible
for any errors that may occur in this book.

Affiliation (in alphabetical order) Contributing authors (in alphabetical order)

Boston University Z. Justin Ren

Chinese University of Hong Kong Kairen Zhang, Sean X. Zhou
Cornell University Li Chen
Georgetown University Volodymyr Babich
Hong Kong University Albert Y. Ha, Hongtao Zhang
of Science and Technology
INSEAD Sameer Hasija, Serguei Netessine
McGill University Mehmet Gümüş
MIT Karen Zhang
New York University Wenqiang Xiao
Northwestern University Gad Allon, Achal Bassamboo, George Georgiadis
Singapore Management University Shantanu Bhattacharya
Stanford University Hau L. Lee
Tel Aviv University Noam Shamir
University of British Columbia Hao Zhang
University of California, Los Angeles Christopher S. Tang
University of California, San Diego Hyoduk Shin
University of Minnesota Guangwen Kong
University of Oregon Zhibin (Ben) Yang
University of Pennsylvania Morris A. Cohen
University of Pittsburgh Esther Gal-Or, Tansev Geylani
University of Southern California Anthony Dukes, Sampath Rajagopalan
University of Texas, Austin Guoming Lai
University of Texas, Dallas Özalp Özer
Vanderbilt University Mümin Kurtuluş
Yale University Sang-Hyun Kim

xiii
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

Part I Value of Information Sharing

1 Modeling and Measuring the Bullwhip Effect . . . . . . . . . . . . . . . . . . . . 3

Li Chen and Hau L. Lee
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Survey of Empirical Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Modeling the Bullwhip Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 Overview and Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.2 AR(1) Demand Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.3 IMA(0,1,1) Demand Process . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.4 General MMFE Demand Process . . . . . . . . . . . . . . . . . . . . 15
1.4 Measuring the Bullwhip Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.1 Seasonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4.2 Time Aggregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4.3 Product and Location Aggregation . . . . . . . . . . . . . . . . . . . 20
1.4.4 Material Flow Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5 Future Research Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2 Empirical Studies in Information Sharing . . . . . . . . . . . . . . . . . . . . . . . 27

Z. Justin Ren
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.1 Types of Information in a Supply Chain . . . . . . . . . . . . . . . 27
2.1.2 Key Questions for Empirical Research . . . . . . . . . . . . . . . . 29
2.2 Value of Demand Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

xv
xvi Contents

2.3 Value of Advance Orders Information . . . . . . . . . . . . . . . . . . . . . . . . 30

2.4 Sharing Forecast Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 Incentives and Contracts in Sharing Forecast Information . . . . . . . . 31
2.6 Value of Inventory Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.7 Value of Sharing Order Status Information . . . . . . . . . . . . . . . . . . . . 34
2.8 Sharing Information About Product Development . . . . . . . . . . . . . . 34
2.9 Sharing Inaccurate Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3 Collaborative Forecasting in Retail Supply Chains . . . . . . . . . . . . . . . . 39

Mümin Kurtuluş
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Value of CF When Supply Chain Members are Cooperative . . . . . . 44
3.2.1 Value of CF Relative to Decentralized Forecasting . . . . . . 45
3.2.2 Value of CF Relative to Vendor Managed Inventory
(VMI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.3 Value of CF Due to Inventory, Production Smoothing,
and Adherence-to-Plans . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3 Value of CF When Supply Chain Members are Strategic . . . . . . . . . 52
3.3.1 Incentives for Sharing Forecasts and Value of CF
Due to Better Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.2 Value of CF When the Forecasting Investments
are Endogenous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4 Conclusions and Future Research Directions . . . . . . . . . . . . . . . . . . 57
3.4.1 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . . . . 59
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

Part II Contracting and Information

4 Reliability or Inventory? An Analysis of Performance-Based

Contracts for Product Support Services . . . . . . . . . . . . . . . . . . . . . . . . . 65
Sang-Hyun Kim, Morris A. Cohen, and Serguei Netessine
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3.1 Repair Process and Performance Measurement . . . . . . . . . 71
4.3.2 Cost Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3.3 Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4.1 The First-Best Benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4.2 Resource-Based Contract . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4.3 Performance-Based Contract . . . . . . . . . . . . . . . . . . . . . . . . 76
4.4.4 Comparisons of the Contracts . . . . . . . . . . . . . . . . . . . . . . . 78
Contents xvii

4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Appendix 1: Mathematical Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Appendix 2: Proofs of the Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

5 Project Contracting Strategies for Managing Team Dynamics . . . . . . 89

George Georgiadis and Christopher S. Tang
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.1.1 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.3 Agents’ Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.3.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.3.2 Markov Perfect Equilibrium (MPE) . . . . . . . . . . . . . . . . . . 95
5.3.3 Comparative Statics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.4 Manager’s Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.5 Project Design with and Without Commitment . . . . . . . . . . . . . . . . . 100
5.5.1 Full Commitment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.5.2 No Commitment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.5.3 Optimal Delegation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

6 Time-Related Incentive Contracts for Managing Projects with

Uncertain Completion Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Christopher S. Tang, Kairen Zhang, and Sean X. Zhou
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.2 Model Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.3 The Combined Contract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.3.1 Bidding Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.3.2 Optimal Combined Contracts . . . . . . . . . . . . . . . . . . . . . . . . 115
6.4 An Extension: Delaying Work Rate Decision . . . . . . . . . . . . . . . . . . 116
6.4.1 Bidding Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.4.2 Optimal Combined Contracts . . . . . . . . . . . . . . . . . . . . . . . . 119
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

7 Contracting for New Product Development . . . . . . . . . . . . . . . . . . . . . . 123

Sameer Hasija and Shantanu Bhattacharya
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.1.1 Notation and General Model . . . . . . . . . . . . . . . . . . . . . . . . 125
7.2 Internal Incentives for NPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.3 Contracts for External NPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.3.1 Management of Contractual Collaborations
for External NPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.3.2 Contests for Contracting on External NPD by Freelance
Innovators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
xviii Contents

7.4 Conclusions and Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

8 Supply Disruptions and Procurement Contracting . . . . . . . . . . . . . . . . 145

Volodymyr Babich and Zhibin (Ben) Yang
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.2 Practice: Supply Chain Risk Management and Contracting . . . . . . . 147
8.3 Research: Standard Assumptions to be Reexamined
in the Presence of Supply Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.4 Insights on Contracting Under Asymmetric Information in the
Presence of Supply Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
8.4.1 Benchmark: Screening for Supplier Cost . . . . . . . . . . . . . . 155
8.4.2 Equivalent Model with Risk: What Assumptions
are Necessary? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.4.3 Screening for Supplier Risk: Why the Model
and the Insights are Different? . . . . . . . . . . . . . . . . . . . . . . . 158
8.4.4 Risk Management Tools and the Value of Information . . . 161
8.5 Multiple Dimensions of Asymmetric Information . . . . . . . . . . . . . . 165
8.6 Future Research Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
9 Contracting for Information Acquisition . . . . . . . . . . . . . . . . . . . . . . . . 171
Guoming Lai and Wenqiang Xiao
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
9.2 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
9.3 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.4 The First-Best Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.5 The Menu of Linear Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.6 The Forecast-Based Contract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
9.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

Part III Information Signaling and Cheap Talk

10 A Tale of Two Information Asymmetries in Competitive Supply

Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Mehmet Gümüş
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
10.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
10.3 Model Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
10.4 Symmetric Information Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
10.5 Asymmetric Demand Information . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
10.6 Asymmetric Capacity Information . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.7 Conclusions, and Future Research Opportunities . . . . . . . . . . . . . . . 208
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Contents xix

11 Supply Chain Information Signaling and Capital Market . . . . . . . . . . 215

Guoming Lai and Wenqiang Xiao
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
11.2 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
11.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
11.4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.4.1 The Single Contract Case . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.4.2 Restoring Efficiency with Operational Mechanism . . . . . . 226
11.5 Other Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
11.5.1 Capacity Constraint and Equilibrium Refinement . . . . . . . 229
11.5.2 Characteristics of Information Asymmetry . . . . . . . . . . . . 230
11.5.3 Sales Channel Stuffing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
11.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

12 Buying from the Babbling Retailer? The Impact of Availability

Information on Customer Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Gad Allon and Achal Bassamboo
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
12.1.1 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
12.1.2 Classical Cheap Talk Game . . . . . . . . . . . . . . . . . . . . . . . . . 240
12.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
12.3 Providing No Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
12.4 Cheap Talk Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
12.4.1 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
12.5 Multiple Customer Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
12.6 Multiple Channels of Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
12.7 Concluding Remarks and Future Directions . . . . . . . . . . . . . . . . . . . 256
Appendix: Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

13 Incentives for Forecast Information Sharing Under Simple Pricing

Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Noam Shamir and Hyoduk Shin
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
13.2 Information Sharing with Multiple Decisions . . . . . . . . . . . . . . . . . . 266
13.2.1 The Basic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
13.2.2 Two-Point Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
13.2.3 General Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
13.3 Competition Between Supply-Chains . . . . . . . . . . . . . . . . . . . . . . . . . 274
13.3.1 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
13.3.2 Benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
13.3.3 Public Forecast Sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
13.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
xx Contents

Part IV Incentives for Information Sharing

14 Establishing Trust and Trustworthiness for Supply Chain

Information Sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Özalp Özer and Yanchong Zheng
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
14.2 Personal Values and Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
14.3 Market Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
14.4 Business Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
14.4.1 Culture and Institutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
14.4.2 Social Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
14.5 Business Process Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
14.5.1 Information Exchange, Advice, or Delegation . . . . . . . . . . 303
14.5.2 Process-Driven Information Exchange . . . . . . . . . . . . . . . . 306
14.5.3 Engineering Reputation Information . . . . . . . . . . . . . . . . . . 307
14.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309

15 Information Leakage in Supply Chains . . . . . . . . . . . . . . . . . . . . . . . . . 313

Guangwen Kong, Sampath Rajagopalan, and Hao Zhang
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
15.1.1 Basic Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
15.2 Ex-Ante Information Sharing Arrangement . . . . . . . . . . . . . . . . . . . . 319
15.2.1 Li (2002) and Zhang (2002) . . . . . . . . . . . . . . . . . . . . . . . . . 320
15.2.2 Li and Zhang (2008) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
15.2.3 Jain et al. (2011) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
15.2.4 Gal-Or et al. (2008) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
15.2.5 Jain and Sohoni (2015) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
15.2.6 Shin and Tunca (2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
15.2.7 Shamir (2012) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
15.3 Ex-Post Information Sharing Arrangement . . . . . . . . . . . . . . . . . . . . 328
15.3.1 Anand and Goyal (2009) . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
15.3.2 Kong et al. (2013) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
15.3.3 Shamir (2015) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
15.4 Other Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
15.4.1 Uncertain Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
15.4.2 Auction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
15.4.3 Competing Supply Chains . . . . . . . . . . . . . . . . . . . . . . . . . . 334
15.5 Discussion and Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
15.5.1 Discussion of the Existing Literature . . . . . . . . . . . . . . . . . 335
15.5.2 Other Supply Chain Structures . . . . . . . . . . . . . . . . . . . . . . 337
15.5.3 Other Types of Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
15.5.4 Long-Term Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
Contents xxi

16 Bilateral Information Sharing and Pricing Incentives in a Retail

Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Anthony Dukes, Esther Gal-Or, and Tansev Geylani
16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.2 Modeling Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
16.3 Single Retailer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
16.3.1 No Sharing and Downward Sharing . . . . . . . . . . . . . . . . . . 348
16.3.2 Bilateral (Full) Sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
16.3.3 The Sharing Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
16.4 Competing Retailers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
16.5 Conclusions and Directions for Future Research . . . . . . . . . . . . . . . 358
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366

17 Sharing Demand Information Under Simple Wholesale Pricing . . . . . 369

Albert Y. Ha and Hongtao Zhang
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
17.2 Single Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
17.2.1 Linear Production Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
17.2.2 Production Diseconomies of Scale . . . . . . . . . . . . . . . . . . . 374
17.2.3 Production Economies of Scale . . . . . . . . . . . . . . . . . . . . . . 376
17.2.4 Make-to-Stock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
17.3 Two Supply Chains Competing in Price . . . . . . . . . . . . . . . . . . . . . . . 380
17.4 Two Supply Chains Competing in Quantity . . . . . . . . . . . . . . . . . . . . 384
17.5 One-to-Many Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
17.6 Two-to-One Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
17.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
 Part I
Value of Information Sharing
Chapter 1
Modeling and Measuring the Bullwhip Effect

Li Chen and Hau L. Lee

Abstract The bullwhip effect is a phenomenon commonly observed in industry. It

describes how the distortion of demand information in a supply chain amplifies de-
mand variance as it moves from consumption point up the supply chain to layers of
suppliers. The bullwhip effect has been a subject of intensive research activities. Re-
searchers have tried to address questions such as: What causes the bullwhip effect?
How would different types of demand signal processing in forecasting and replen-
ishment decisions affect the bullwhip effect? Can we explain the magnitude of the
bullwhip effect in terms of the characteristics of the product and the supply chain?
What is the magnitude of the bullwhip effect in practice, how does it differ across
industries and products, and how prevalent is the phenomenon? In this chapter, we
review both theoretical and empirical research done to address these questions, as
well as research done to identify important approaches and specifications that are
necessary to correctly measure and evaluate the true extent of the bullwhip effect.

1.1 Introduction

Demand variability and uncertainty is a driver of supply chain inventory. Manag-

ing supply chains can be a challenge when demand variability and uncertainty is
high. For a company in a supply chain consisting of multiple stages, each of which

L. Chen ()
Samuel Curtis Johnson Graduate School of Management, Cornell University, Ithaca,
NY 14853, USA
e-mail: [email protected]
H.L. Lee
Graduate School of Business, Stanford University, Stanford, CA 94305, USA
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 3

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 1
4 L. Chen and H.L. Lee

is run by a separate organization (or company), the variability of demand faced by

this company can be much higher than the variability faced by downstream stages
(where “downstream stages” refers to the stages closer to the final consumption of
the product). The bullwhip effect refers to the phenomenon where demand vari-
ability amplifies as one moves upstream in a supply chain (Lee et al. 1997a, or
LPW). LPW described this as a form of demand information distortion. Lee et al.
(1997b) further discussed the managerial and practical aspects of the bullwhip ef-
fect, giving more industry examples. The bullwhip effect phenomenon is closely re-
lated to studies in systems dynamics (Forrester 1961; Sterman 1989; Senge 1990).
Sterman (1989) observed a systematic pattern of demand variation amplification
in the Beer Game, and attributed it to behavioral causes (i.e., misperceptions of
feedback). Macroeconomists have also studied the phenomenon (Holt et al. 1968;
Blinder 1981; Blanchard 1983).
According to LPW, the bullwhip effect has been observed extensively in many
industries. However, they provided only anecdotal references. LPW developed sim-
ple mathematical models to explain how the bullwhip effect could arise, and iden-
tified four causes: demand signal processing, order batching, price variations, and
the rationing game. Demand signal processing refers to a company using forecast
updates, and such updates would automatically lead to larger order fluctuations than
demand. Order batching refers to companies not ordering in every single time unit,
and, as a result, order variance would naturally be larger than demand variance.
Price variations result in companies making more than “normal” order quantities
when prices are lower than normal, and vice versa, leading to higher order fluc-
tuations. Rational game refers to companies anticipating supply shortages, and to
ensure adequate supply, exaggerate their needs through placing larger order quanti-
ties than otherwise. Hence, these models showed that the bullwhip effect could be
a result of “rational” decision making under limited information and myopic objec-
tives. Croson and Donohue (2006), using clever experimental setups, demonstrated
that there could also be additional behavioral causes, namely, the under-reaction to
lags and coordination stock. Under-reaction to lags refers to players ignoring inven-
tory in the pipeline when they made ordering decisions. Coordination stock refers
to players increasing their orders because they wanted to have higher safety stocks,
resulting from experiences of past delays in shipments from suppliers.
The main contention of LPW is that one needs to understand the causes of the
bullwhip effect in order to devise counter-strategies. Hence, each of the causes re-
quires a set of strategies for companies to use. These counter-strategies tend to fall
into two broad types. First, one needs to recognize the existence of the bullwhip
effect so that one is not “fooled” by the distorted demand information. Companies
would therefore be smarter in making capacity and inventory decisions accordingly.
Second, companies need to work, often collaboratively, to reduce the magnitude of
the bullwhip effect. One of the most commonly cited counter-strategies is informa-
tion sharing, a topic covered in other chapters of this book. In the ideal situation,
by information sharing, a company might not be misled by the distorted demand
information, and, as a result, might not “bullwhip” its upstream supplier.
1 Modeling and Measuring the Bullwhip Effect 5

Since the work of LPW, two streams of research have emerged: modeling
and empirical. In the modeling stream, researchers have expanded the LPW work
through more complex modeling of the demand process to show how the bullwhip
effect could arise. In the empirical stream, instead of anecdotal evidence, researchers
have tried to measure the extent of the bullwhip effect in real industry cases.
These two streams of research have reinforced each other in our deepened un-
derstanding of the bullwhip effect. Modeling research generates insights and forms
the bases of hypotheses in empirical research. Empirical research serves to confirm
or refute some of the results derived in modeling research, but it can also suggest
potential additional causes of the bullwhip effect or additional phenomena that can
lead to new modeling research. Hence, the two streams together have provided a
healthy path for both streams of research. In this chapter, we review both streams
of work.

Fig. 1.1 Illustration of information and material flows at a supply chain stage

In studying the bullwhip effect, we note that there have been two primary def-
initions of bullwhip effect measurement used in the literature. It is worthwhile for
us to clarify these two definitions as they affect how one interprets the results in the
literature. LPW originally described the bullwhip effect as a form of “demand infor-
mation distortion.” The amplification of demand variance is based on the measure of
demand variance faced by each stage in the supply chain. Hence, consider one stage
of the chain facing its own demand variance. This stage in turn makes its ordering
decision (where order can also be interpreted as production release in a manufac-
turing setting). The orders then become the demand faced by the upstream stage.
The existence of the bullwhip effect implies that the order variance is larger than
the original demand variance. This definition captures the distortion of information
flow that goes upstream (see Fig. 1.1). A second definition, used in many empirical
studies, compares the variance of order receipts (or shipments) with the variance of
sales. Sales represent the material outflow from the current stage under consider-
ation, while order receipts (or shipments) represent the material outflow from the
upstream stage, which become the material inflow to the current stage. In some
cases, the order receipt information, if not available, is inferred from the sales and
6 L. Chen and H.L. Lee

inventory data (see Blinder 1981; Cachon et al. 2007). This definition essentially
captures the distortion of material flow that goes downstream (see Fig. 1.1).
When the upstream stage can always supply perfectly the orders placed by the
current stage, and the current stage can always satisfy the demands that it faces, then
orders and order receipts are the same, and sales and demands also coincide. In that
case, the measures of information-flow and material-flow bullwhip effects would be
identical. But once there are shortages in either the upstream or the current stages,
the two measures would diverge.
The bullwhip effect measurements based on these two definitions also differ in
concept. The information-flow based definition has a direct linkage to supply chain
cost because the orders issued by a stage become a driver to the upstream inven-
tory/capacity decision. Hence, the information-flow bullwhip effect is a cost driver.
In contrast, order receipts (or shipment) information is the outcome of the upstream
order-fulfillment decision process, rather than an input to the decision process.
Hence, the material-flow bullwhip effect is the consequence of the information-flow
bullwhip effect. Moreover, in the information-flow based definition, the bullwhip
effect is a result of one decision maker, i.e., the stage in question. This decision
maker observes demand, and then makes order decisions based on various struc-
tural and economic factors. In the material-flow based definition, however, there
are three decision effects involved. First, the sales data is determined by the ac-
tual demand and the on-hand inventory, where the latter is a result of the inven-
tory decisions made in previous periods. Second, as in the information-flow based
case, the unit makes order decisions, based on structural and economic conditions.
Third, the actual order receipts from the supplier are the result of the supplier’s
previous production/stocking decisions, where the order receipts may not exactly
equal the orders (e.g., production shortfall, transportation constraints, etc.). In view
of these differences, we believe the information-flow based definition is more suit-
able for theoretical analysis purposes. However, we recognize the need for using
the material-flow based definition as an empirical surrogate in some cases, and thus
include a discussion of the implications of such an approximation in Sect. 1.4.
The rest of the chapter is organized as follows. In Sect. 1.2, we review the empir-
ical findings of the bullwhip effect. In Sect. 1.3, we review the literature of bullwhip
effect modeling, with an emphasis on the demand process modeling. Section 1.4
discusses various issues related to the empirical measurement of the bullwhip ef-
fect. We conclude the chapter by discussing some future research opportunities in
Sect. 1.5.

1.2 Survey of Empirical Findings

Empirical research concerning the bullwhip effect is a large literature and, rather
than giving a comprehensive review, we highlight some significant findings. There
are two classes of empirical research. The first one is closer to the field-based
approach, in which a single supply chain is the unit of analysis. The demand
1 Modeling and Measuring the Bullwhip Effect 7

information in this single supply chain, often with a single class of products in focus,
is analyzed to explore the existence of the bullwhip effect and measure its magni-
tude. The second one uses secondary data of many companies, often aggregated, to
pursue statistical analysis of the bullwhip effect.
LPW started with their anecdotal observation of excessive volatility in weekly or-
ders in both Procter & Gamble’s diaper supply chain and Hewlett-Packard’s printer
supply chain.
We highlight a few sample studies that are based on single supply chains. In a
landmark teaching case, Hammond (1994) documented how Barilla SpA, the largest
pasta producer in the world, observed strong bullwhip effects. The supply chain
members—Barilla and its customers—all processed demand signals, orders were
batched, and promotions were common. At one of the distribution centers (DC) of
its largest retail customer, the weekly orders placed by this DC to Barilla had a
mean of 300 quintals and a standard deviation of 227. But the weekly sell-through
at the DC (which can be viewed as shipments to the stores) had a mean of 300
and standard deviation of 60. Suppose that we define the “bullwhip ratio” to be
CVout /CVin , where CVout is the coefficient of variation (CV) of the outgoing orders,
and CVin is the CV of the incoming orders. The bullwhip effect exists if the bullwhip
ratio is greater than one. The bullwhip ratio at the Barilla case was 3.75, i.e., 73 %
of the variation at the DC could be explained as the distortion within the supply
chain, while the remaining 27 % was the variation faced by the DC. Through VMI
(vendor-managed inventory), Barilla was able to reduce the inventory at this DC by
47 %, while shortage rates dropped by 7 % to almost zero.
Lai (2005) also studied a single supply chain, that of a Spanish grocery chain
Sebastian de la Fuente. The study was based on monthly product-level data at the
DC, showing prices, markups, sales delivered to stores, supplies from suppliers,
inventories and promotion. The data set contained records of 3745 products over 29
months that pass through the DC. Regression analysis by Lai (2005) demonstrates
that the bullwhip effect existed and was mainly driven by batching by the store, as
well as two behavioral causes identified by Croson and Donohue (2006).
Fransoo and Wouters (2000) studied two supply chains of convenience foods
(salads and ready-made pasteurized meals) involving four companies in The Nether-
lands. The supply chain consists of three stages: the producer, the regional DC and
retail franchisees. Using the filtered daily sales data (from March 23 until June 5,
1998), they found that the bullwhip effect was prominent across the supply chain.
The bullwhip ratios found in their study are shown in Table 1.1.

Table 1.1 Summary of bullwhip ratios of different supply chain stages

Supply chain stage Bullwhip ratio (meals) Bullwhip ratio (salads)

Production 1.75 1.23

Distribution Center 1.26 2.73
Retail Franchisee 1.67 2.09
8 L. Chen and H.L. Lee

Note that both Lai (2005) and Fransoo and Wouters (2000) used sales data, and so
the bullwhip ratios that they measured were based on the material flow. Hammond’s
case was based on orders placed by the DC (information flow) and the sales to the
DC (material flow). It showed the challenges faced by empirical researchers as it is
not easy to truly measure the information distortion aspect of the bullwhip effect,
i.e., the information-flow bullwhip effect. The study of Fransoo and Wouters (2000)
also highlights how the bullwhip effect can vary across products. Hence, one has to
be careful when conducting empirical research with data aggregated over products.
Moving from the single supply chain approach, there also have been empiri-
cal accounts of the bullwhip effect that are based on two levels of aggregation of
secondary data. These studies usually used monthly or quarterly data aggregated
across various products or firms. The first level of aggregation was on time units.
In the sample studies of single supply chains described earlier, the time unit of the
data was based on the timing of order decisions. For example, it was a week in the
case of Barilla, since orders were generated on a weekly basis; and it was monthly
in the case of Sebastian de la Fuente, since the supply chain members order on a
monthly basis. The time unit used in the convenience food supply chain study was
a day, since the supply chain ordered on a daily basis, which was necessary given
the perishable nature of meals and salads. If companies orders on a monthly basis
or a quarterly basis, then the use of monthly or quarterly data may pose no poten-
tial problem. But this is rarely the case in real life. The second level of aggregation
was on products and often across firms as well, which can also be problematic on
the validity of bullwhip effect measurement. These two aggregation problems are
discussed in detail in Sect. 1.4.
Industry-based studies are anchored on data that are aggregated over products
and firms. High production volatility was found in the TV set industry (Holt et al.
1968), retail industry (Blinder 1981; Mosser 1991), automobiles (Blanchard 1983;
Kahn 1992), cement industry (Ghali 1987), high tech consumables (Hanssens 1998),
paper products (Carlsson and Fuller 1999), semiconductors (De Kok et al. 2005),
semiconductor equipment (Terwiesch et al. 2005), and many other industries (Miron
and Zeldes 1988; Fair 1989). In these studies, researchers searched for explana-
tions to reconcile the bullwhip effect with the classic production-smoothing theory
that posits that the motive for keeping inventory is to smooth production variability
rather than to amplify it. One of the leading explanations is that production smooth-
ing was missing because seasonality had been excluded from the data (e.g., Ghali
1987).
There are also studies that are at the multi-industry or economy level (Blinder
1986; Bivin 1996; Cachon et al. 2007). Here, there was even more extensive aggre-
gation of products and firms, as data from products and firms in different industries
have also been combined. Fine (1999) demonstrated the bullwhip effect using this
multi-industry approach, as shown in Fig. 1.2. The fluctuation of automobile pro-
duction was clearly greater than that of the GDP (mimicking the downstream stage
of the automobile industry), while the fluctuation of machine tools (mimicking the
upstream stage of the automobile industry) was even greater.
1 Modeling and Measuring the Bullwhip Effect 9

Fig. 1.2 Illustration of industry volatility

Several recent empirical studies are worth highlighting. Cachon et al. (2007) used
monthly sales and inventory data from the U.S. Census Bureau of 1992–2004 to ex-
amine the bullwhip effect in the manufacturing, wholesale and retail sectors. Hence,
there was aggregation of time units, and aggregation across products. The bullwhip
effect analysis was based on the material flow concept. They found that if season-
ality is included in the measurement, production smoothing indeed exists in the
retail industry and in some manufacturing industries, but not in the wholesale in-
dustry. With seasonally unadjusted data: 62 % of manufacturers have a bullwhip ra-
tio less than one, 86 % of retailers had a bullwhip ratio less than one, while 84 % of
wholesalers have a bullwhip ratio larger than one. Hence, there is empirical evidence
that while there was a tendency for companies to bullwhip the upstream suppliers,
sometimes the desire to smooth production may be even stronger, dampening the
bullwhip effect.
Bray and Mendelson (2012) reported that about two-thirds of firms bullwhip in
a sample of 4689 public U.S. companies over 1974–2008. Building on the model
of a generalized order-up-to policy proposed by Chen and Lee (2009), the authors
decomposed the bullwhip effect by information transmission lead time. They found
that demand signals with shorter time notice have greater impact on the bullwhip
effect.
Bray and Mendelson (2015) further investigated the bullwhip effect and produc-
tion smoothing in an automotive manufacturing sample comprising 162 car models
and found that 75 % of the sample smooth production by at least 5 %, despite the
fact that 99 % of the sample exhibit the bullwhip effect. They measured production
smoothing with a structural econometric production scheduling model based on the
generalized order-up-to policy. According to their structural estimation, there exist
both a strong bullwhip effect (on average, production is 220 % as variable as sales)
and robust smoothing (on average, production would be 22 % more variable absent
volatility penalty costs).
10 L. Chen and H.L. Lee

Shan et al. (2014) investigated the bullwhip effect using data from over 1200 pub-
lic companies in China during 2002–2009. They found that more than two-thirds of
the companies exhibit the bullwhip effect. Their regression analysis suggests that
the bullwhip effect magnitude is positively associated with average on-hand inven-
tory and persistence of demand shocks, and is negatively associated with degree of
demand seasonality.
Duan et al. (2015) collected a daily product-level dataset consisting of 487 indi-
vidual products from a supermarket chain in China. They found that the magnitude
of the bullwhip effect at the product level is much more significant than those mea-
sured at the firm or industry level, suggesting that product and time aggregation may
mask the bullwhip effect measurement.
Osadchiy et al. (2016) investigated the systematic risk in demand for different
industries and firms in the U.S. economy, including retail, wholesale, and manufac-
turing sectors. They used sales as a proxy for demand, and defined the systematic
risk in sales as the correlation coefficient of sales change with the contemporane-
ous market return. They found that demand signal processing does not amplify the
systematic risk, however, aggregation of orders from multiple customers and aggre-
gation of orders over time can result in the amplification of systematic risk upstream
along the supply chain.

1.3 Modeling the Bullwhip Effect

There are multiple ways of modeling the bullwhip effect along a supply chain.
For example, the bullwhip effect can be modeled as a result of judgmental errors
by human decision makers (e.g., Sterman 1989; Chen 1999; Steckel et al. 2004;
Croson and Donohue 2006). It can also be modeled as a result of suboptimal in-
ventory policies. Chen et al. (2000a,b) showed that, when certain demand forecast
methods, such as moving average and exponential smoothing, are used to determine
a (suboptimal) inventory policy for an AR(1) demand process, the resulting order
variability exceeds demand variability. De Kok (2012) considered a two-echelon
supply chain in which the downstream demand is stationary but the retailers fore-
cast demand using the exponential smoothing method. He quantified the bullwhip
effect as a function of the exponential smoothing parameter. Dejonckheere et al.
(2003) modeled a supply chain as an engineering system and studied the bullwhip
effect under certain (suboptimal) system replenishment rules. We refer the reader to
Geary et al. (2006) for a survey of the control engineering approach for modeling
the bullwhip effect.
While the above approaches can all account for the bullwhip effect, there is a
certain degree of arbitrariness in the assumed irrational human behaviors and sub-
optimal inventory policies. To eliminate such arbitrariness, a normative approach is
needed, whereby the decision maker is assumed to make rational decisions in opti-
mizing the system cost performance. In the remainder of this section, we shall take
this normative approach for modeling the bullwhip effect.
1 Modeling and Measuring the Bullwhip Effect 11

1.3.1 Overview and Model Setup

An optimal order quantity from a rational decision maker is a response to the supply
and demand uncertainties of a given situation. On the demand side, LPW showed
that positively-correlated demand coupled with long lead time would amplify the
order variability, while negatively-correlated demand would dampen it. On the sup-
ply side, LPW also showed that potential supply shortages would cause downstream
stages to inflate orders and thus trigger the bullwhip effect. However, random supply
disruptions may also dampen the bullwhip effect (Rong et al. 2009) and a capacity
constraint can smooth the order quantity (Chen and Lee 2012).
The underlying cost structure also drives the order variability. For example, fixed
ordering costs, such as full truckload and machine setup costs, will lead to large
batch orders and cause the bullwhip effect (LPW; Cachon 1999; Chen and Lee
2012). External cost shocks, such as promotional discounts, will induce forward-
buying behavior, which again causes the bullwhip effect (Blinder 1986; LPW). Con-
versely, explicit penalty costs for order variability will force the decision maker to
smooth order quantities (Sobel 1969; Aviv 2007; Cantor and Katok 2012; Bray and
Mendelson 2015).
To mitigate the bullwhip effect, one can thus either encourage information shar-
ing among supply chain partners to reduce the supply and demand uncertainties,
or modify the supply chain cost structure to provide economic incentives for or-
der smoothing. In what follows, we highlight some bullwhip effect models that are
closely related to the study of demand information sharing.
Consider a supply chain stage for a single product. Inventory is reviewed period-
ically at this stage. Time is divided into periods of length one and indexed forwards
(i.e., t = 0, 1, 2, . . .). Let Dt denote the customer demand in period t and let μ (> 0)
denote the mean of demand in a period. Customer demand is fulfilled immediately
if the stage has enough on-hand inventory; unmet demand is fully backlogged. Unit
holding cost h and stockout penalty cost p are assessed and charged to the stage
at the end of each period. Inventory is replenished from an upstream stage with a
constant lead time L. The upstream stage is assumed to have ample supply. For ease
of exposition, we assume that the manager minimizes the long-run average cost;
we note that assuming a discounted cost objective function would yield the same
insights (e.g., LPW).
It is known in the literature that the optimal policy for such an inventory system
is a state-dependent base-stock policy; a static base-stock policy can be viewed as
a special case of the state-dependent policy. Let St denote the state-dependent base-
stock level in period t. Under the base-stock policy, the order quantity in period t,
denoted by Ot , can be written as

Ot = St − (St−1 − Dt−1 ) = St − St−1 + Dt−1 . (1.1)

In the above expression, we have implicitly assumed that the order quantity in
each period can be negative, such that the base-stock level is achievable in each
period. This is equivalent to assuming that excess inventory can be freely returned to
12 L. Chen and H.L. Lee

the supplier. We shall make this assumption throughout this chapter for tractability
reasons. We note that the chance of a negative order quantity becomes negligible
when the order mean is sufficiently greater than the order variance. Justifications for
this assumption can be found in LPW, Aviv (2003, 2007), and Chen and Lee (2009).
A few quick observations can be made based on expression (1.1). First, when
the demand process Dt is independent and identically distributed (i.i.d.), the op-
timal policy is a static base-stock policy, i.e., St = St−1 for any t. It follows that
Ot = Dt−1 , and hence var(Ot ) = var(Dt−1 ). Therefore, there is no bullwhip effect
in such a system. Second, when the demand process Dt is not i.i.d., the optimal
base-stock policy is state dependent. In this case, var(Ot ) may be greater or less
than var(Dt−1 ), depending on the covariance between St − St−1 and Dt−1 . Below
we consider several different demand processes, to quantify the variance ratio be-
tween order and demand.

1.3.2 AR(1) Demand Process

LPW considered an autoregressive AR(1) demand process for modeling the bull-
whip effect. Specifically, the demand in a period is defined as

Dt − μ = ρ (Dt−1 − μ ) + εt ,

where |ρ | < 1 and εt is an i.i.d. normal random variable with N(0, σ02 ). Let d =
(1 − ρ )μ . We can rewrite the above equation as follows:

Dt = d + ρ Dt−1 + εt . (1.2)

To ensure the chance of a negative demand is negligible, we assume σ0  d.

Under the “free-return” assumption discussed above, the optimal base-stock level
can always be achieved in each period. As a result, we can determine the optimal
base-stock level St by solving the following long-run average cost minimization
problem:
  L + L + 
min E h St − ∑ Dt+i + p ∑ Dt+i − St ,
St i=0 i=0

where (x)+= max(x, 0). The above problem is a standard newsvendor problem, and
we know the optimal base-stock level is given by
 
∗ −1 p
St = G , (1.3)
h+ p

where G(·) is the cumulative distribution function of the lead time demand ∑Li=0 Dt+i .
Thus, it remains to determine to the distribution of the lead time demand.
1 Modeling and Measuring the Bullwhip Effect 13

By leveraging the recursive expression (1.2), with some algebra, we can show
that, for k ≥ 0,

Dt+k = d + ρ Dt+k−1 + εt+k

= d + ρ (d + ρ Dt+k−2 + εt+k−1 ) + εt+k
= ···
1 − ρ k+1 k
=d + ρ k+1 Dt−1 + ∑ ρ k−i εt+i .
1−ρ i=0

Because εt+i is an i.i.d. normal random variable, the lead time demand ∑Li=0 Dt+i ,
conditional on Dt−1 , also follows a normal distribution, with the conditional mean
and variance given as follows:

   L+1
1 − ρ k ρ (1 − ρ L+1 )
E ∑Li=0 Dt+i  Dt−1 = d ∑ + Dt−1 ,
k=1 1 − ρ 1−ρ
 2
   L+1 k
var ∑Li=0 Dt+i  Dt−1 = ∑ ∑ ρ k−i σ02 .
k=1 i=1

Substituting the above result into expression (1.3), we arrive at

L+1 k 2
L+1
1 − ρ k ρ (1 − ρ L+1 )
St∗ = d ∑ 1 − ρ + 1 − ρ Dt−1 + zσ0 ∑ ∑ ρ k−i ,
k=1 k=1 i=1

where z = Φ −1 (p/(h + p)), with Φ −1 (·) being the inverse standard normal cumu-
lative distribution function. Under the above optimal base-stock policy, the order
quantity in period t is given by

ρ (1 − ρ L+1 )
Ot = St∗ − St−1
∗
+ Dt−1 = (Dt−1 − Dt−2 ) + Dt−1 .
1−ρ

Using the recursive expression (1.2) again, with some algebra, we can show that

var(Ot ) 2ρ (1 − ρ L+1 )(1 − ρ L+2 )

= 1+ . (1.4)
var(Dt−1 ) (1 − ρ )

The above expression provides a characterization of the bullwhip effect under the
AR(1) demand process. We note that the AR(1) demand process reduces to an
i.i.d. process when ρ = 0. In this case, the above expression reduces to var(Ot ) =
var(Dt−1 ), which is the same as what we obtained in the i.i.d. demand case. Thus,
expression (1.4) can be viewed as a generalization of the bullwhip effect result from
the i.i.d. demand case. From the expression, we observe that, if the demand pro-
cess has a positive temporal correlation, i.e., ρ > 0, we have var(Ot ) > var(Dt−1 ).
Order variability is amplified and the bullwhip effect exists in this case. On the
other hand, if the demand process is negatively correlated, i.e., ρ < 0, we have
14 L. Chen and H.L. Lee

var(Ot ) < var(Dt−1 ). Order variability is dampened instead. Moreover, when ρ > 0,
it can be shown that the ratio in (1.4) is increasing in the replenishment lead time L,
suggesting that longer lead time induces greater bullwhip effect.

1.3.3 IMA(0,1,1) Demand Process

Graves (1999) considered another simple demand process for modeling the bullwhip
effect. Specifically, demand Dt is assumed to follow an integrated moving-average
IMA(0, 1, 1) process, which is defined as follows:
Dt = Dt−1 − (1 − α )εt−1 + εt , (1.5)
where |α | < 1 and εt is an i.i.d. normal random variable with N(0, σ02 ).
We note that
the demand reduces to an i.i.d. process when α = 0, and that the demand process
becomes a random walk when α = 1.
Let Ft−1 = Dt−1 − (1 − α )εt−1 . From (1.5), we have Dt = Ft−1 + εt . Thus, Ft−1
is the best linear predictor for Dt at the end of period t − 1. With some algebra, we
can show that Ft−1 satisfies the following equation:
Ft−1 = α Dt−1 + (1 − α )Ft−2 ,
which has the same form as an exponential smoothing forecast.
By leveraging the recursive expression (1.5), with some algebra, we can show
that, for k ≥ 0,
Dt+k = Dt+k−1 − (1 − α )εt+k−1 + εt+k
= Dt+k−2 − (1 − α )εt+k−2 + αεt+k−1 + εt+k
= ···
k−1
= Ft−1 + ∑ αεt+i + εt+k .
i=0

Because εt+i is an i.i.d. normal random variable, the lead time demand ∑Li=0 Dt+i ,
conditional on Ft−1 , also follows a normal distribution, with the conditional mean
and variance given as follows:
  
E ∑Li=0 Dt+i  Ft−1 = (L + 1)Ft−1 ,
   L+1
var ∑Li=0 Dt+i  Ft−1 = ∑ (1 − α + α k)2 σ02 .
k=1

Following an argument analogous to that of the AR(1) demand process, we obtain

the optimal base-stock level under the IMA(0,1,1) demand process as

L+1
St∗ = (L + 1)Ft−1 + zσ0 ∑ (1 − α + α k)2 ,
k=1
1 Modeling and Measuring the Bullwhip Effect 15

where z = Φ −1 (p/(h + p)). Under the above optimal base-stock policy, the order
quantity in period t is given by
Ot = St∗ − St−1
∗
+ Dt−1 = (L + 1)(Ft−1 − Ft−2 ) + Dt−1 .

We note that the above optimal order quantity is the same as the adaptive ordering
policy proposed by Graves (1999). Here we have shown that this ordering policy is
the outcome of an optimal state-dependent base-stock policy.
By using the recursive expression (1.5) again, with some algebra, we can
show that
var(Ot |Ft−2 )
= (1 + α + α L)2 . (1.6)
var(Dt−1 |Ft−2 )
From the above expression, we observe that when α > 0, var(Ot |Ft−2 ) >
var(Dt−1 |Ft−2 ) and the bullwhip effect exists. Moreover, when α > 0, the ratio in
(1.6) is increasing in L, suggesting again that longer lead time induces greater bull-
whip effect—the same insight as shown earlier under the AR(1) demand process. It
is worth commenting that in the above expression (1.6), the conditional variances of
order and demand are used. This is because the unconditional variances of order and
demand under the IMA(0, 1, 1) demand process are both unbounded. Therefore, we
have to resort to the conditional variance measure, which captures the order and de-
mand uncertainties conditional on the most up-to-date demand forecast information.

1.3.4 General MMFE Demand Process

Chen and Lee (2009) proposed a demand model that generalizes the demand pro-
cesses discussed above. Specifically, they assumed that the demand process evolves
according to the martingale model of forecast evolution (MMFE) process (Hausman
1969; Graves et al. 1986; Heath and Jackson 1994). Under the MMFE model, the
demand Dt is defined as
∞
Dt = μ + ∑ εt−i,t , (1.7)
i=0

where εt−i,t is the demand information obtained in period t − i with respect to de-
mand Dt . For all i ≥ 0, εt−i,t is an independent normal random variable with with
N(0, σi2 ). Let σ 2 = ∑∞
i=0 σi . For ease of exposition, we shall assume σ < ∞ be-
2 2

low; in the case of σ = ∞, the bullwhip effect results can be modified with the
2

conditional variance as in Sect. 1.3.3.

From (1.7), the best linear predictor for demand Dt at the end of period t − i
(i ≥ 0) can be defined as
∞
Ft−i,t = μ + ∑ εt− j,t , (1.8)
j=i
16 L. Chen and H.L. Lee

with Ftt being the actual demand Dt itself. With some algebra, we can show

Ft−i,t = Ft−i−1,t + εt−i,t .

Hence, εt−i,t can be viewed as the forecast revision with regard to demand Dt made
at the end of period t − i, and εtt (also written as εt below) is the final uncertainty
resolved during period t after demand realization. An illustration of the MMFE
demand process is given in Table 1.2.

Table 1.2 Illustration of the MMFE process

Forecast revision for future demands in each period
Demand ··· 0 1 2 3 4 5 6 ···

D0 = μ + ··· + ε0
D1 = μ + ··· + ε0,1 + ε1
D2 = μ + ··· + ε0,2 + ε1,2 + ε2
D3 = μ + ··· + ε0,3 + ε1,3 + ε2,3 + ε3
D4 = μ + ··· + ε0,4 + ε1,4 + ε2,4 + ε3,4 + ε4
D5 = μ + ··· + ε0,5 + ε1,5 + ε2,5 + ε3,5 + ε4,5 + ε5
D6 = μ + ··· + ε0,6 + ε1,6 + ε2,6 + ε3,6 + ε4,6 + ε5,6 + ε6
.. .. .. .. .. .. .. .. .. ..
. . . . . . . . . .

The demand information obtained at the end of period t with regard to all future
demands can be summarized in a forecast revision vector ε t = [εt , εt,t+1 , εt,t+2 , . . .]T ,
where ε t is assumed to be i.i.d. with a multivariate normal distribution N(0, Σ ). The
variance-covariance matrix is given by Σ = E{ε t ε tT }.
The above MMFE demand model generalizes many commonly used demand
models. For example, if ε t = [εt , 0, 0, . . .]T for all t, then we have an i.i.d. demand
process. If ε t = [εt , ρεt , ρ 2 εt , . . .]T (with 0 ≤ |ρ | < 1) for all t, then we obtain
the AR(1) demand process described in Sect. 1.3.2. If ε t = [εt , αεt , αεt , . . .]T (with
0 < α ≤ 1) for all t, then we obtain the IMA(0,1,1) demand process described in
Sect. 1.3.3. It can also be shown that the MMFE model is general enough to cover
the ARIMA(p, d, q) model (Box et al. 1994; Gilbert 2005; Gaur et al. 2005), the
linear state-space model (Aviv 2003), and the advance demand information model
(Gallego and Özer 2001). We refer the reader to Chen and Lee (2009) for details.
Under the MMFE demand model, we have, for k ≥ 0,
k
Dt+k = Ft−1,t+k + ∑ εt+k−i,t+k .
i=0

Because εt+k−i,t+k is an i.i.d. normal random variable, conditional on {ε τ : τ ≤

t − 1}, the lead time demand ∑Li=0 Dt+i , also follows a normal distribution, with the
conditional mean and variance given as follows:
1 Modeling and Measuring the Bullwhip Effect 17
  
L  L

E ∑ Dt+i  τ , τ ≤ t − 1 = ∑ Ft−1,t+i ,
ε
i=0 i=0
L   L+1

var ∑ Dt+i  ε τ , τ ≤ t − 1 = ∑ ei1 Σ ei1 ,
T

i=0 i=1

where ek is the unitary vector with the k-th element being one and ei1 = ∑ik=1 ek .
Therefore, we obtain the optimal base-stock level under the MMFE demand pro-
cess as
L L+1
St∗ = ∑ Ft−1,t+i + z ∑ ei1
T
Σ ei1 ,
i=0 i=1

where z = Φ −1 (p/(h + p)). Under the above optimal base-stock policy, the order
quantity in period t is given by
L
Ot = St∗ − St−1
∗
+ Dt−1 = ∑ (Ft−1,t+i − Ft−2,t+i ) + Dt−1 .
i=0

From the expressions (1.7) and (1.8), with some algebra, we can show that

var(Ot ) (eL+2 )T Σ eL+2 i=1 (ei ) Σ ei

− ∑L+2 T
= 1+ 1 1
. (1.9)
var(Dt−1 ) σ2

The above expression provides a general, unifying formula for the bullwhip effect
for demand processes that can be represented by the MMFE model. For example,
the bullwhip effect result (1.4) under the AR(1) demand process can be recovered
from (1.9) by setting ε t = [εt , ρεt , ρ 2 εt , · · · ]T .
When the demand process is either AR(1) or IMA(0,1,1), we have shown that
longer lead time induces greater bullwhip effect. However, from (1.9), we cannot
claim such a result without additional assumptions on the variance-covariance ma-
trix Σ of the forecast evolution process. The general expression (1.9) indicates that
it is actually the overall forecast correlation during the lead time period that drives
the magnitude of the bullwhip effect.

1.4 Measuring the Bullwhip Effect

As discussed in the introduction, measurement of the bullwhip effect is another

important topic. In this section, we use analytical models to demonstrate various
issues that one can encounter in measuring the bullwhip effect with empirical data.
When one compares the variances of the order and demand, the first question
is whether a ratio or an absolute difference metric should be used. If the objective
is to determine whether the bullwhip effect exists or not, then both the ratio and
difference metrics can be used. However, if one wants to compare the bullwhip effect
for different products, the ratio metric, being unit-independent, appears to be a better
18 L. Chen and H.L. Lee

choice. For example, consider two products: one with demand variance of 10 and
order variance of 20, and the other with demand variance of 40 and order variance
of 80. With the ratio metric, the amplification ratio is 2 for both products. However,
with the difference metric, the second product has greater amplification than the first
one (40 versus 10). Furthermore, if one tries to calculate the aggregated bullwhip
effect measure over these two products (assuming the demands are independent),
the ratio would remain 2, but the difference would increase to 50 (10 + 40). The
ratio metric is thus more suitable for comparison purposes.
In some empirical studies, the standard deviation ratio metric and/or the coef-
ficient of variation ratio metric have been used (see the Barilla example given in
Sect. 1.2). We note that the standard deviation ratio and the variance ratio contain
essentially the same information for comparison purposes, as the former metric is
just a squared root value of the latter. The coefficient of variation ratio is equivalent
to the standard deviation ratio when the order mean is the same as the demand mean.
When the order mean is different from the demand mean, one needs to normalize the
variability measure based on the different mean values. For example, suppose that
only demand data from a partial set of customers are available in a distribution net-
work. Then the order mean may be greater than the demand mean. In this case, the
coefficient of variation ratio is more appropriate for measuring the bullwhip effect.
In what follows, we shall consider a single-stage model with the order mean equal
to the demand mean. Thus, the variance ratio metric is sufficient for our analysis.

1.4.1 Seasonality

Cachon et al. (2007) found that including seasonality in the measurement of the
bullwhip effect leads to a much lower bullwhip effect measurement result. The fol-
lowing model is used by Chen and Lee (2012) to demonstrate such an effect.
Let s0 , . . . , sT −1 denote the additive seasonality with a regular cycle of T periods,
where s0 corresponds to the seasonal factor of demand D0 . Without loss of general-
T −1
ity, we assume that the seasonal factors are normalized, such that ∑i=0 si = 0. We
can define the variability of seasonality as
T −1
1
Vs =
T ∑ s2i .
i=0

Let Dt (Dt ) and Ot (Ot ) denote the seasonal (deseasonalized) demand and order
quantity in period t, respectively. Thus, the demand in a period can be expressed as

Dt = Dt + s(t mod T ) .

1 Modeling and Measuring the Bullwhip Effect 19

Under the “free-return” assumption, the optimal base-stock level in a period is

given by
L
St∗ = St + ∑ s(t+i mod T ) ,
i=0

St
where is the state-dependent base-stock level for the deseasonalized demand pro-
cess. Therefore, the seasonal order quantity in a period is given by

Ot = St∗ − (St−1
∗
− Dt−1 ) = St − St−1
 
+ Dt−1 + s(t+L mod T ) = Ot + s(t+L mod T ) ,

where the last equality follows from relationship Ot = St − St−1
  . Hence, we
+ Dt−1
can show that
var(Ot ) Vs + var(Ot ) var(Ot ) − var(Dt−1
 )
=  ) = 1 +  ) ,
var(Dt−1 ) Vs + var(Dt−1 Vs + var(Dt−1

where Dt and Ot are the deseasonalized demand and order quantities. Thus, if
the bullwhip effect exists in the deseasonalized demand process, i.e., var(Ot ) −
 ) > 0, and if the variability of seasonality dominates the deseasonalized
var(Dt−1
 ), then including seasonality in the bullwhip
demand variability, i.e., Vs  var(Dt−1
effect measurement will drive the ratio close to one. Chen and Lee (2012) further
showed that the ratio may go below one when there is a capacity limit in the system.

1.4.2 Time Aggregation

Measuring the bullwhip effect based on aggregate data also prompts the question of
potential aggregation biases in the measurement. The following model is used by
Chen and Lee (2012) to investigate the effect of data aggregation on the bullwhip
effect measurement.
Consider first the time aggregation effect. Define the M-period aggregation of
M = M−1 D
demand and order as Dt−1 ∑τ =0 t−1+τ and OtM = ∑M−1
τ =0 Ot+τ , respectively. By
the relationship (1.1), it follows that

OtM = St+M−1 − St−1 + Dt−1

M
.

For most common demand models, it can be shown that limM→∞ var(Dt−1 M ) = ∞.

For example, it can be verified that this condition holds for an AR(1) demand pro-
M will
cess. Thus, under this condition, the variance of the aggregated demand Dt−1
eventually dominate the finite variance of St+M−1 − St−1 as M increases. Therefore,
it is straightforward to show that limM→∞ var(OtM )/ var(Dt−1M ) = 1. That is, time

aggregation has a masking effect.

Moreover, suppose that the demand follows an ARMA(1,1) process given by

Dt − μ = ρ (Dt−1 − μ ) + εt + θ εt−1 ,
20 L. Chen and H.L. Lee

where ρ > 0, ρ + θ > 0. Chen and Lee (2012) showed that the bullwhip effect under
time aggregation is given by

var(OtM ) 2(ρ + θ )(1 − ρ L+1 )(1 − ρ L+2 + θ ρ − θ ρ L+1 )

= 1 + .
M )
var(Dt−1 (M/(1 − ρ M ))(1 − ρ 2 )(1 + θ )2 − 2(ρ + θ )(1 + θ ρ )

It is easy to verify that the above ratio decreases to one monotonically as M

increases.

1.4.3 Product and Location Aggregation

Besides time aggregation, empirical data are also subject to product and location
aggregation. Since location aggregation is mathematically equivalent to product ag-
gregation, below we will present the analysis on product aggregation from Chen
and Lee (2012). Define the N-product aggregation of demand and order quantities
N = N D
as Dt−1 ∑n=1 t−1,n and OtN = ∑Nn=1 Ot,n , respectively, where Dt−1,n is the de-
mand for product n and Ot,n is the order quantity for product n. Also, define the
aggregated base-stock level as StN = ∑Nn=1 St,n .
Consider first the case in which the demands of the N products are spatially-

independent but share a common additive seasonality pattern s0 , . . . , sT −1 . Let Dt,n

and Ot,n denote the deseasonalized demand and order quantity for product n, respec-
tively. Thus, the demand for product n in a period can be expressed as

Dt,n = Dt,n + αn · s(t mod T ) ,

where the multiplicative factor αn captures the heterogenous magnitude of season-

ality across the N products. Without loss of generality, assume αn ≥ 1 for all n. Also
assume the replenishment lead time is a constant of L periods for all products. The
optimal order quantity for product n is given by

Ot,n = Ot,n + αn · s(t+L mod T ) .

We can show that the aggregate bullwhip ratio is given by

N

var(OtN )
∑ {var(Ot,n
 
) − var(Dt−1,n )}
N )
= 1 + n=1
N 2 N
.
var(Dt−1
∑ αn Vs + ∑ 
var(Dt,n )
n=1 n=1

The above ratio approaches to one as N goes to infinity. Therefore, if the products
under aggregation share a common seasonal profile (for example, the Christmas
seasonality in the retail industry), including seasonality in the measurement will
drive the aggregated bullwhip ratio to one and thus mask the bullwhip effect of
individual products.
1 Modeling and Measuring the Bullwhip Effect 21

Consider another case in which the demands of the N products are spatially-
independent but all belong to the AR(1) process family. Specifically, let us assume
that the demand for product n is given by:
Dt,n − μn = ρn (Dt−1,n − μn ) + εt,n ,
where 0 ≤ ρn < 1 and εt,n is an i.i.d. normal random variable with N(0, σn ). Thus,
we have var(Dt,n ) = σn2 /(1 − ρn2 ). By setting ρ = ρn in the result of (1.4), with some
algebra, we have the following:
N
1 − ρnL+1 1 − ρnL+2 2
var(OtN )
∑ 2ρn · 1 − ρn
·
1 − ρn2
· σn
N )
= 1 + n=1 .
var(Dt−1 N
σ2
∑ 1 −nρn2
n=1

Thus, if limN→∞ ∑Nn=1 ρn σn2 / ∑Nn=1 [σn2 /(1 − ρn )] = 0, the above aggregate bullwhip
ratio approaches to one as N goes to infinity. This condition can be satisfied, for
example, when ρn = 1/(n + 1) and σn = 1 for all n, which means there is an in-
creasing portion of the products with an autocorrelation coefficient close to zero.
Another example is when ρn = n/(n + 1) and σn = 1 for all n. In this case, there is
an increasing portion of products with autocorrelation coefficient approaching one.
In both cases, the bullwhip ratios for individual products are all strictly greater than
one (because ρn > 0), but product aggregation can mask the severity of the bullwhip
effect of individual products.

1.4.4 Material Flow Data

So far, we have considered bullwhip effect measurement based on the variances of

order and demand (i.e., the information flow). As discussed in Sect. 1.2, in many
empirical studies, material flow data (such as shipments and sales) are used as prox-
ies for the order and demand data to measure the bullwhip effect. Chen et al. (2016)
provided an analytical comparison of these two bullwhip effect measurements.
Consider a simple i.i.d. demand process. The optimal policy in this system is a
static base-stock policy S. From (1.1), it follows that Ot = Dt−1 , and there is no
bullwhip effect based on the information flow data. Now let us examine the material
flow in such a system. Let M1 (t) denote the shipment from an upstream stage in
period t. Since the upstream stage is assumed to have ample supply, we have
M1 (t) = Ot = Dt−1 . (1.10)
Let M0 (t) denote the sales to the downstream customer in period t. It can be
shown that
L +  L +
M0 (t) = Dt + ∑ Dt−L−1+i − S − ∑ Dt−L+i − S , (1.11)
i=0 i=0
22 L. Chen and H.L. Lee

where the last two terms represent the demand backlogs in periods t − 1 and t,
respectively. From the above expressions, the sales M0 (t) can be written as a sum-
mation of Dt and two additional random variable terms. Intuitively, one would ex-
pect the variance of M0 (t) to be greater than the variance of Dt . However, it can be
shown that the opposite is true. Specifically, from (1.11), the following expression
can be obtained:
 L +  L + 
var(Dt ) − var(M0 (t)) = 2E S − ∑ Dt−L−1+i ∑ Dt−L+i − S ≥ 0.
i=0 i=0
(1.12)
From (1.10), we know that var(M1 (t)) = var(Dt ). Thus, it follows that var(M1 (t)) ≥
var(M0 (t)). In other words, measurement based on the material flow data would
overstate the underlying information-flow bullwhip effect in such a system. Kahn
(1987) derived a similar insight in a model with zero lead time. The above result
generalizes the insight to the case with general, positive lead time. It is clear from
the expression that the shipment variance equals the sales variance only when S = 0
or S = ∞. Thus, measurement based on the material flow data may provide a good
approximation to the underlying information-flow bullwhip effect when the base-
stock level is either (sufficiently) high or low.
When demand Dt follows an i.i.d. normal distribution N(μ , σ 2 ), it can be shown
that the expression (1.12) is unimodal in S, reaching a peak value at S = (L + 1)μ
(where the system service level is 50 %). It can be further shown that the expression
(1.12) is decreasing in lead time L (under the same service level), suggesting that
measurement based on the material flow data may provide a good approximation to
the information-flow bullwhip effect when the replenishment lead time is long.
Chen et al. (2016) further considered the AR(1) demand process for comparing
the bullwhip effect measurements. They showed that the autocorrelation parame-
ter ρ in the AR(1) demand model has a direct impact on the measurement biases.
When the demand is moderately correlated, measurement based on the material flow
data may overstate the information-flow bullwhip effect, which generalizes the in-
sight from the i.i.d. demand case. However, when the demand has a strong temporal
correlation, the result reverses, i.e., measurement based on the material flow data
would understate the information-flow bullwhip effect. To reduce the measurement
discrepancy from the material flow data, Chen et al. (2016) proposed an estimation
method based on the sample variances of aggregated sales data. Their method works
for common demand processes with short-range dependence, and does not require
the knowledge of the underlying base-stock policy

1.5 Future Research Opportunities

The vast literature of theoretical and empirical research on the bullwhip effect
provides several important observations. First, we need to be careful to distin-
guish between information-flow and material-flow bullwhip effects. Second, the
1 Modeling and Measuring the Bullwhip Effect 23

magnitudes of the bullwhip effects of different products do differ. Third, data ag-
gregation over time, product, firms and industry sectors may play a role in masking
the bullwhip effect. For practical managerial insight, a single firm has to deal with
bullwhip effects for each of its products (since production and inventory are based
on a product), and the demand orders arrive or are issued in accordance with the
time units that the partners in the supply chain use as their decision points. Hence,
when products and firms data are aggregated, the resulting aggregate measurement
may not be that meaningful, unless one is dealing with capacities that are shared
by multiple products, or in the extreme cases, resources that are shared by multiple
firms.
The second and third observations above imply that much research is still needed
for a full understanding of the bullwhip effect. Richer demand models, as well as
analyses of the impacts of different levels of aggregation on the measurement of the
bullwhip effect, are needed. We have discussed some of the recent developments in
this direction in Sects. 1.3 and 1.4. Most of the models assume a linear supply chain.
It would be interesting to expand the models to include more complex distribution
networks. For example, in a system where multiple customers are supplied by a
common supplier, how would the inventory allocation rule at the supplier affect the
bullwhip effect and its measurement? How does the inventory pooling effect at the
supplier interact with the demand variability amplification in such a system? How
does the demand correlation among the downstream customers affect the bullwhip
effect? When each customer has a different ordering process resulting from a dif-
ferent demand process, decision time unit, and ordering economics, how does this
affect the bullwhip effect?
The empirical research indicated that the tendency toward demand amplification
through demand signal processing is sometimes counteracted by the tendency to-
ward production smoothing, so that the bullwhip effect may be dampened. Chen and
Lee (2009) was one of the articles that modeled both such drivers, using a simple
way of smoothing by postponing some orders to a later time. To gain a deeper under-
standing of the interactive effects of demand amplification and production smooth-
ing, more complex production smoothing models can be constructed (e.g., Bray and
Mendelson 2015).
Finally, we need empirical research that is based on the right units (time, prod-
ucts, or firm) to truly measure the bullwhip effect. We also need empirical research
to explore the drivers or characteristics of the products and supply chains that
give rise to different bullwhip effect magnitudes. In addition, Özer et al. (2014)
have found that cultural differences play a role in trusts and information sharing.
Since the bullwhip effect is closely tied to trust factors between the buyer and the
supplier, it would also be of interest to conduct behavioral research to study how
cultural differences may affect the magnitudes of the bullwhip effect.
24 L. Chen and H.L. Lee

References

Aviv Y (2003) A time-series framework for supply chain inventory management. Oper Res
51(2):210–227
Aviv Y (2007) On the benefits of collaborative forecasting partnerships between retailers and man-
ufacturers. Manag Sci 53(5):777–794
Bivin D (1996) Bunching in the production process. Econ Lett 50:259–263
Blanchard OJ (1983) The production and inventory behavior of the American automobile industry.
J Polit Econ 91(3):365–400
Blinder AS (1981) Retail inventory behavior and business fluctuations. Brook Pap Econ Act
2:443–520
Blinder AS (1986) Can the production smoothing model of inventory behavior be saved? Q J Econ
101(3):431–454
Box GEP, Jenkins GM, Reinsel GC (1994) Time series analysis forecasting and control, 3rd edn.
Prentice-Hall, Englewood Cliffs
Bray RL, Mendelson H (2012) Information transmission and the bullwhip effect: an empirical
investigation. Manag Sci 58(5):860–875
Bray RL, Mendelson H (2015) Production smoothing and the bullwhip effect. Manuf Serv Oper
Manag 17(2):208–220
Cachon G (1999) Managing supply chain demand variability with scheduled ordering policies.
Manag Sci 45(6):843–856
Cachon G, Randall T, Schmidt G (2007) In search of the bullwhip effect. Manuf Serv Oper Manag
9(4):457–479
Cantor D, Katok E (2012) Production smoothing in a serial supply chain: a laboratory investigation.
Transp Res Part E 48(4):781–794
Carlsson C, Fuller R (1999) Soft computing and the bullwhip effect. Econ Complex 2(3):1–26
Chen F (1999) Decentralized supply chains subject to information delays. Manag Sci 45(8):
1076–1090
Chen L, Lee HL (2009) Information sharing and order variability control under a generalized
demand model. Manag Sci 55(5):781–797
Chen L, Lee HL (2012) Bullwhip effect measurement and its implications. Oper Res 60(4):
771–784
Chen YF, Drezner Z, Ryan JK, Simchi-Levi D (2000a) Quantifying the bullwhip effect in a simple
supply chain: The impact of forecasting, lead times and information. Manag Sci 46(3):436–443
Chen YF, Ryan JK, Simchi-Levi D (2000b) The impact of exponential smoothing forecasts on the
bullwhip effect. Nav Res Logist 47(4):269–286
Chen L, Luo W, Shang KH (2016) Measuring the bullwhip effect: Discrepancy and alignment
between information and material flows. Manuf Serv Oper Manag (Forthcoming)
Croson R, Donohue K (2006) Behaviorial causes of the bullwhip effect and the observed value of
inventory information. Manag Sci 52(3):323–336
Dejonckheere J, Disney SM, Lambrecht MR, Towill DR (2003) Measuring and avoiding the bull-
whip effect: a control theoretic approach. Eur J Oper Res 147(3):567–590
De Kok AG (2012) The return of the bullwhip. Rev Bus Econ Lit 57(3):381–393
De Kok AG, Janssen F, van Doremalen J, van Wachem E, Clerkx M, Peeters W (2005) Philips
Electronics synchronizes its supply chain to end the bullwhip effect. Interfaces 35(1):37–48
Duan Y, Yao Y, Huo J (2015) Bullwhip effect under substitute products. J Oper Manag 36:75–89
Fair RC (1989) The production-smoothing model is alive and well. J Monet Econ 24:353–370
Fine CH (1999) Clockspeed: winning industry control in the age of temporary advantage. Perseus
Books, Cambridge
Forrester JW (1961) Industrial dynamics. MIT Press, Cambridge
Fransoo JC, Wouters MJF (2000) Measuring the bullwhip effect in the supply chain. Supply Chain
Manag 5(2):78–89
1 Modeling and Measuring the Bullwhip Effect 25

Gallego G, Özer Ö (2001) Integrating replenishment decisions with advance demand information.
Manag Sci 47(10):1344–1360
Gaur V, Giloni A, Seshadri S (2005) Information sharing in a supply chain under ARMA demand.
Manag Sci 51(6):961–969
Geary S, Disney SM, Towill DR (2006) On bullwhip in supply chains—historical review, present
practice and expected future impact. Int J Prod Econ 101(1):2–18
Ghali MA (1987) Seasonality, aggregation and the testing of the production smoothing hypothesis.
Am Econ Rev 77(3):464–469
Gilbert K (2005) An ARIMA supply chain model. Manag Sci 51(2):305–310
Graves SC (1999) A single-item inventory model for a nonstationary demand process. Manuf Serv
Oper Manag 1(1):50–61
Graves SC, Meal HC, Dasu S, Qiu Y (1986) Two-stage production planning in a dynamic environ-
ment. In: Axsäter S, Schneeweiss C, Silver E (eds) Multi-stage production planning and control.
Lecture notes in economics and mathematical systems, vol 266. Springer, Berlin, pp 9–43
Hammond J (1994) Barilla SpA (A). Harvard Business School Case No. 9-694-046
Hanssens DM (1998) Order forecasts, retail sales, and the marketing mix for consumer durables.
J Forecast 17(3):327–346
Hausman WH (1969) Sequential decision problems: a model to exploit existing forecasters. Manag
Sci 16:B93–B111
Heath DC, Jackson PL (1994) Modelling the evolution of demand forecasts with application to
safety stock analysis in production/distribution systems. IIE Trans 26(3):17–30
Holt CC, Modigliani F, Shelton JP (1968) The transmission of demand fluctuations through a
distribution and production system, the TV-set industry. Can J Econ 1(4):718–739
Kahn JA (1987) Inventories and the volatility of production. Am Econ Rev 77(4):667–679
Kahn JA (1992) Why is production more volatile than sales? Theory and evidence on the stockout-
avoidance motive for inventory-holding. Q J Econ 107(2):481–510
Lai RK (2005) Bullwhip in a Spanish shop. Harvard NOM Working Paper No. 06-06
Lee HL, Padmanabhan V, Whang S (1997a) Information distortion in a supply chain: the bullwhip
effect. Manag Sci 43(4):546–558
Lee HL, Padmanabhan V, Whang S (1997b) The bullwhip effect in supply chains. Sloan Manag
Rev 38(4):93–102
Miron JA, Zeldes SP (1988) Seasonality, cost shocks, and the production smoothing model of
inventories. Econometrica 56(4):877–908
Mosser PC (1991) Trade inventories and (S, s). Q J Econ 106(4):1267–1286
Osadchiy N, Gaur V, Seshadri S (2016) Systematic risk in supply chain networks. Manag Sci
62(6):1755–1777
Özer Ö, Zheng K, Ren Y (2014) Trust, trustworthiness, and information sharing in supply chains
bridging China and the United States. Manag Sci 60(10):2435–2460
Rong Y, Shen Z-JM, Snyder LV (2009) The impact of ordering behavior on order-quantity vari-
ability: a study of forward and reverse bullwhip effects. Flex Serv Manuf J 20(1):95–124
Senge P (1990) The fifth discipline: the art and practice of the learning organization. Doubleday
Currency, New York
Shan J, Yang S, Yang S, Zhang J (2014) An empirical study of the bullwhip effect in China. Prod
Oper Manag 23(4):537–551
Sobel M (1969) Production smoothing with stochastic demand I: finite horizon case. Manag Sci
16(3):195–207
Steckel J, Gupta S, Banerji A (2004) Supply chain decision making: Will shorter cycle times and
shared point-of-sale information necessarily help? Manag Sci 52(4):458–464
Sterman JD (1989) Modeling managerial behavior: Misperceptions of feedback in a dynamic de-
cision making experiment. Manag Sci 35:321–339
Terwiesch C, Ren Z, Ho T, Cohen MA (2005) An empirical analysis of forecast sharing in the
semiconductor equipment supply chain. Manag Sci 51:208–220
Chapter 2
Empirical Studies in Information Sharing

Z. Justin Ren

Abstract This chapter provides a non-technical literature survey of empirical stud-

ies in information sharing during the past two decades. We categorize by types of
information that can be shared in a supply chain environment, such as demand,
forecasts, inventory status, and product development. We encourage more empirical
research to be done to further advance knowledge in this field.

2.1 Introduction

No supply chain can run smoothly without information. In this chapter, we

provide a literature review related to sharing different types of information in a sup-
ply chain. We focus on empirical studies, however our definition of “empirical” is
rather broad—any relevant work that involves collecting data is permissible. Hence
it includes experimental work, surveys, and even case studies. Moreover, it is often
necessary to discuss theoretical models as they serve as the motivation or guide for
some empirical work. When that happens, we will do so briefly.

2.1.1 Types of Information in a Supply Chain

Think of a typical three-stage supply chain with a manufacturer, a distributor, and a

retailer. Products flow from upstream (manufacturer) to downstream (end demand).
But what really enables the supply chain is the flow of information. For example,

Z.J. Ren ()

Questrom School of Management, Boston University, Boston, MA, USA
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 27

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 2
28 Z.J. Ren

demand information arrives at the retailer, who orders from its distribution channel,
who in turn orders from the manufacturer. Without such information, the supply
chain cannot do what it is supposed to do, which is to match supply with demand.
There could be many types of information being transmitted in a supply chain—
which we categorize below—but a central question for each supply chain party is:
What information is needed in order for it to optimize its operations?
We could categorize the main types of information that are explicit or implicit in
the supply chain, somewhat in order of prevalence, as follows:
• Demand. Demand (or order information) is perhaps the most important type of
information in a supply chain. It should come as no surprise because the reason
supply chains exist is to match demand with supply. However, exactly how im-
portant is it to have accurate demand information? What can supply chain parties
do if such information is not available? Those are some of the key supply chain
research questions.
• Forecasts. Sometimes what was received from the downstream party or what
was sent to its upstream party are not actual orders, but just order forecasts (or
mere intention of orders) either in the form of a point estimate, or a statistical
distribution (“probably x, but maybe y . . . ”). To what extent those forecasts are
accurate (inaccurate, and why so) are then relevant questions. More importantly,
do supply chain parties always share forecast information truthfully? Why or
why not?
• Inventory status information. This includes order status information such as the
quantity of inventory at each stage in the system. It may also include information
about inventory availability at each stage of the system.
• Product development information. Companies follow very different practices in
announcing products in development: Microsoft regularly announces what will
be coming in the next few years, while Apple holds tight lips on what their next
product (or even next versions of an existing product) will be. So an interesting
question will be: Does it offer any benefit if a manufacturer shares its product
development information with its downstream parties?
• Information accuracy. Most of the research in information sharing assumes that
information existing in the supply chain is accurate (we distinguish inaccurate
information from information generated from strategic information-sharing be-
havior). However, that is frequently not true in practice. Supply chain parties
have to deal with inaccurate information. What if we know some information is
subject to errors, and may be inaccurate?
This list is not exhaustive by any means, but here we focus on topics where sub-
stantive research has been done. Next, we briefly summarize major research themes
in the literature that are related to information sharing in supply chains.
2 Empirical Studies in Information Sharing 29

2.1.2 Key Questions for Empirical Research

Related to information and information sharing, there are a number of research ques-
tions that can be asked. The following is a partial list:
1. What is the value of having a certain type of information in a supply chain, such
as demand forecast or order information?
2. What are the incentive issues in information sharing?
3. How can information be used in improving system efficiency?
4. How is the value of information changed by contracts, organizational factors of
supply chain parties, and market structure?
We now review relevant empirical literature by each type of information catego-
rized above.

2.2 Value of Demand Information

Over the past two decades, along with the increasing presence of information tech-
nology (IT), there has been increasing interest on studying the value of demand-
related information in various supply chains. A significant body of theoretical
knowledge has been built in this area.
Through various supply chain models, researchers have demonstrated that shar-
ing information on demand will overall improve the efficiency of the whole system,
though the exact magnitude of improvement vary depending on model specifics.
Representative work include Chen (1998) (N-stage serial system with recorder pol-
icy; benefit ranges from 0 to 9 %), Gavirneni et al. (1999) (two-stage capacitated
inventory system with (s, S) policy; costs decrease vary from 1 to 90 %), Lee et al.
(2000) (two-stage serial supply chain with autoregressive demand), Gaur et al.
(2005) (serial system autoregressive moving average demand), Aviv and Federgruen
(1998) (Vendor Managed Inventory (VMI) program, cost saving from 2 to 4.7 %).
Do those theoretical improvement pan out in reality? So far as we know, there
has been few empirical studies that aim to test directly the theoretical predictions
put forth by the aforementioned papers (it will be very interesting if some future
research aims to do just that). As far as we know, Zhou and Benton (2007) offer
the closest alternative. The authors collected surveys from over a hundred compa-
nies asking about their practice on information sharing with their suppliers and their
customers and the effectiveness of their information sharing practice. They consider
three aspects of information sharing: information sharing support technology, infor-
mation content, and information quality. The results are overall positive. For exam-
ple, they find strong evidence that effective information sharing enhances effective
supply chain practice.
Another stream of empirical research investigate the effectiveness and efficiency
of such supply chain collaboration initiatives as EDI or VMI. Daugherty et al. (1999)
survey executives of about 100 firms with an average annual revenue of $2.3 billion,
30 Z.J. Ren

and find that VMI practices are positively correlated with firm performance. Kuk
(2004) looks at the effect of organizational and supply chain factors such as em-
ployee involvement and supply chain integration on the effectiveness of VMI. Ana-
lyzing results from a survey in the electronics industry, the author finds that supply
chain members with higher levels of employee involvement and logistics integration
were more likely to realize the potential benefit of VMI. Moreover, it is also found
that VMI seems to have benefited small organizations more than large ones.

2.3 Value of Advance Orders Information

Another class of demand information is advanced orders, or “pre-orders.” Some-

times, customers are willing to wait for a certain length of time before receiving and
consuming a product. This advance demand information should permit suppliers to
better forecast demand and replenish inventory. Tang et al. (2004) documents the
case of a large bakery in Hong Kong on the production of moon cake, a type of pas-
try that is consumed for the annual “mid-autumn festival” in many Asian cultures.
The retailer can induce customers to pre-order for a discounted price before the
selling season, which the authors call “Advance Book Discount” (ABD) program.
The time between placement and fulfillment of these pre-committed orders provides
an opportunity for the retailer to update demand forecasts. This paper, by demon-
strating the benefit advance demand information, is rather prescient, because such
a benefit is increasingly being realized in global commerce with the help of infor-
mation technology. Fast forward to 2013, Alibaba (the world’s largest e-commerce
platform) has started such ABD program where Chinese consumers first place or-
ders for produce products from US such as cherries and then wait for delivery.
Such advance order tactics are also routinely employed by consumer-goods man-
ufacturers and retailers. Moe and Fader (2002) use actual advance sales and total
sales data of music albums obtained from CDNOW to study the effectiveness of
advance purchase orders on modeling total sales. They show that advance sales pro-
vide meaningful information in terms of better forecasting sales. With better demand
information, cost savings and increased margins should follow. The distribution of
the benefits will depends on the contracts between supply chain parties.
Fisher and Raman (1996) provides another anecdotal example in the apparel in-
dustry, and show that early orders indeed provide an important tool in forecasting
sales and managing production and global supply chains.
For the apparel manufacturer in this study, demand for a particular item is highly
unpredictable, and supply lead time is long. Initial forecasts of demand does not of-
fer great help in terms of accurately managing supply, but by observing early orders
the manufacturers are able to remarkably better allocate capacity to meet supply
and therefore increase profitability. See Fig. 1 in Fisher and Raman (1996) at http://
dx.doi.org/10.1287/opre.44.1.87, showing the value of early demand information
(a) with no early demand information and (b) with early demand information. Even
though in this example, early orders are technically different from advance orders in
2 Empirical Studies in Information Sharing 31

that the former are sold while in stock, while the latter are sold before products are
available, the idea is the same: Early demand signal serves as a powerful instrument
in managing supply with demand.

2.4 Sharing Forecast Information

Forecast information is different from “real” information. Forecasts are, by defini-

tion, inaccurate. More often than not, forecasts represent forecasters’ mere inten-
tions to order, and are not contractible. Hence, incentives play a huge role in sharing
forecast information.
With the increasing adoption of Material Requirement Planning (MRP) system
and Enterprise Resource Planning (ERP) system in industries all over the world,
incorporating demand forecasts into production and inventory management system
is become more and more important. In this area, research has been quite fruitful.
As one of the earliest works in this area, Graves et al. (1998) developed a mar-
tingale model of forecast evolution model (MMFE) to study requirement planning
in multistage production-inventory system. Motivated by an industry study, Heath
and Jackson (1994) propose another MMFE model to characterize the process of
forecast evolution, and apply this model in a simulation study to analyze safety
stock levels for a production and distribution system. Later works that use MMFE
are various but they typically model forecasts to become more accurate over time.
However, Cattani and Hausman (2000) question such an assumption. With real data
from companies they show that demand forecasts do not necessarily become more
accurate as they are updated. This is illustrated in Fig. 1 of Cattani and Hausman
(2000) at http://dx.doi.org/10.1287/msom.2.2.119.12354. As is shown in the figure,
the forecasts are changing, but are not converging. That is, they do not get closer to
the actual points as time goes on, a phenomenon the author call “forecast churning.”
They argue that such forecast churning can cause inefficiencies if the firm reacts to
the wrong forecast update. Another survey study by Wacker and Hanson (1997) of
large global manufacturing firms further show that forecast errors persist, however
they do not create significant impact on firm competitiveness.

2.5 Incentives and Contracts in Sharing Forecast Information

But there is more story behind forecast error and forecast churning. Cohen et al.
(2003) and Terwiesch et al. (2005) empirically study the evolution of forecasts based
on their study of the semiconductor equipment industry. First of all, consistent with
Cattani and Hausman (2000), they find that forecast accuracy does not improve over
time. See below a typical picture of different forecasts shared by a major semicon-
ductor manufacturer to one of its major suppliers, as well as the actual demand
(Fig. 2.1).
32 Z.J. Ren

Fig. 2.1 Forecasts show a over-forecasting bias, Cohen et al. (2003) (reproduced with permission
from INFORMS)

But the picture reveals something more than just random fluctuations of fore-
casts: the actual orders seem to be lying below most of the forecasts. In other words,
the manufacturer seems to be consistently over-forecasting. Why is that? The an-
swer to this question is the key contribution of Cohen et al. (2003). They find that the
manufacturer has a strong incentive to over-forecast. The reason can be explained
with a “newsvendor-type” logic: the cost of inflating forecasts and canceling is much
less than the cost of truthful forecasts but having to deal with the possible supply
shortage.
Speaking of incentives in information sharing in supply chains, Lee et al. (1997)
provide one of the earliest academic discussions of problems related to order fore-
casts and their implication on supply chain coordination. They call order forecasts
that are eventually cancelled “phantom orders,” and see them as one of the key fac-
tors contributing to the now well-known bullwhip effect in supply chains. Given that
supplier capacity may be constrained, a buyer has a strong incentive to forecast ex-
tra orders (phantom orders), especially if scarce capacity is rationed based on placed
orders. Armony and Plambeck (2005) investigate how such false orders can lead a
manufacturer to overestimate the demand and make faulty decisions about capacity
investment. They motivate their study by citing a high-profile news story of Cisco,
writing off assets worth of billions of dollars due to phantom orders placed by its
suppliers.
When a supply chain party possesses private information that the other parties
do not know, and given incentives that may be in place for distorting information,
how to induce truth-telling in sharing forecast is an important topic. Cachon and
Lariviere (2001) is one of the earliest study of incentives in information sharing
using a game-theoretical approach. They formulate a capacity procurement game,
2 Empirical Studies in Information Sharing 33

where a high-demand type customer has to dispense extra incentive as a signaling

device, such as fixed payment or a large firm commitment, in order to induce the
supplier to provide enough capacity for high demand. Later papers by Özer and
Wei (2006) show that channel coordination is possible when combining capacity
commitment contract with payback contract in the setting of Cachon and Lariviere.
Ren et al. (2010) propose a long-term perspective of this problem, where supply
chain parties expect to transact repeatedly over a long horizon. They show that un-
der some conditions supply chain parties can develop a ‘harmonious’ relationship
where a simple wholesale price can achieve truthful information sharing and sys-
tem coordination asymptotically. But can such a results be proven empirically or
experimentally?
Özer et al. (2011) devise a series of clever experiments to test if such a “trust-
worthy” relationship can be obtained from repeated interactions. In their benchmark
model they are able to replicate the over-forecasting phenomenon predicted by one-
shot game-theoretical models confirming the over-forecasting bias. See the top part
of Fig. 1 in Özer et al. (2011) at http://dx.doi.org/10.1287/mnsc.1110.1334.
Interestingly, as supply chain parties interact repeatedly over a longer horizon,
the authors observe a continuum of trust among supply chain dyads. This result
shows that in reality we may observe various degrees of supply chain coordination
that are much richer than what previous theoretical model. They then fashion a par-
simonious ‘trust-embedded’ model which fits experimental data remarkably well.

2.6 Value of Inventory Information

The advance of information technology such as RFID, Internet and cloud computing
has made inventory information available and therefore relevant to supply chain
management practice. But the basic information remains: “Is it helpful to know
inventory information of downstream supply chain parties?”
Using a series of lab experiments mimicking supply chain environments, Croson
and Donohue (2006) study the benefit of sharing inventory information in “the beer
game.” First, they find that decision makers consistently under-weigh the pipeline
stock even when the normal operational causes (e.g., batching, price fluctuations,
and demand estimation) are removed, and as a result the bullwhip effect persists.
Will sharing inventory information help remove some of the variability? It turns out
sharing inventory information has little effect on the orders of downstream chain
members. On the positive side, inventory information seems to substantially reduce
the variance of orders for upstream members.
Empirical findings related to the effect of sharing inventory information can be
found in studies on supply chain coordination initiatives such as Vendor Managed
Inventory (VMI), where sharing inventory information is a prerequisite. Clark and
Hammond (1997) find a correlation of VMI practices with performance improve-
ments. Kulp et al. (2004) conduct an extensive study with the food and consumer
34 Z.J. Ren

packaged goods industry, and find that sharing retail store inventory levels and col-
laborative planning on replenishment practices such as VMI provide benefit to the
manufacturer’s profitability.

2.7 Value of Sharing Order Status Information

Does it help to know where exact an order is in a supply chain, whether it is in

transit or in process? Before the advent of the Internet and sophisticated information
technology such information was unobtainable. However, as technologies such as
RFID and GPS systems, as well as Internet access are becoming widely available, it
becomes interesting to ask what the value of having order status information is.
Theoretical research on this topic overall demonstrate that there is positive value
in such type of information. For example, Gaukler et al. (2008) find that knowing
how far along an outstanding order is being processed in a supply chain not only
helps the retailer in deciding when to order and how much to order, but also allows
the retailer to place an emergency order when it expects a delay is likely to occur.
They calculate that the cost savings can be large, especially in supply chains with
longer and more volatile lead times. Jain and Moinzadeh (2005) consider the value
of manufacturers sharing inventory availability information with retailers. In their
model a manufacturer can inform its retailer if a product will be on backorder before
the retailer places its order. They find that sharing such information will actually
induce the retailer to order more frequently.
Empirical studies on the benefit of sharing such information can be found in the
area of collaborative planning, forecasting, and replenishment (CPFR). Yao et al.
(2013) collected data on CPFR activities from Motorola and its suppliers, and find
that forecast accuracy tends to improve immediately after CPFR implementation,
but then the rate of improvement slows over time. Inventory levels initially increase
after implementation and then begin to decrease. It seems that more empirical stud-
ies are needed to ascertain the benefit of sharing inventory status information.

2.8 Sharing Information About Product Development

Product development sits at the top of a supply chain. Therefore, it is relevant to

study information sharing that is related to product development. There are two
dimensions to this. The first is information sharing within the process of product
development. There, the research question is how to best share information in order
to come up with successful products with faster lead times. The second dimension
is about the value of sharing information about product development, with other
echelons of the supply chain. Below we review both areas of research.
Sharing product information is similar to sharing forecast information. In both
cases, what is being shared is not-yet-finalized preliminary information.
2 Empirical Studies in Information Sharing 35

Development teams frequently begin their work on a new product prior to receiving
detailed design specifications from the customer or from adequate feedback from
market research. How to utilize preliminary information is a central question. In
this line of research, Loch and Terwiesch (1998) study the situation faced by a
concurrent engineering team of a European auto company where an information-
receiving team must decide on how to rely on the preliminary information provided
by the information sender. The information receiver has an incentive to start early,
but starting early means using a lower quality of information and thus has a higher
chance of costly rework. Thus it faces the “Rush and be Wrong or Wait and be
Late” dilemma, which is similar to the supplier’s problem as described in Cohen
et al. (2003).
Empirically identifying the value of sharing product development-related infor-
mation is an interesting topic as firms take very different approaches in practice.
Take product pre-announcement, for example. In the high-tech industry, it is well
known that there are firms who never give product announcements (e.g., the fa-
mous Apple secrecy), and there are others who tend to pre-announce their pipeline
products well ahead of time (e.g., Microsoft). What is the impact on their com-
petitors and on their supply chains? Bayus et al. (2001) analyze data on product pre-
announcement in software development industry (also called “vaporware”), and find
that software development process is inherently uncertain and delays are inevitable.
Moreover, smaller firms tend to pre-announce products strategically in response
to product development status of larger competitors. In a supply chain context,
research seems to favor sharing product development information with their supply
chain partners. Kulp et al. (2004) collect data from the food and consumer packaged
goods industry, and find that when manufacturers work closely with their retailers
by sharing product development-related information, they are more likely to have
higher profit margins.

2.9 Sharing Inaccurate Information

Most of the supply chain modeling work assume that information shared in supply
chains is accurate. But to what extent is that true? Empirical evidence suggests that
such an assumption needs to be reevaluated. Raman et al. (2001) find that data in-
accuracy persists in the retailing inventory, which results in substantial operational
inefficiency and profit loss. DeHoratius and Raman (2008) study inventory record
accuracy with one nationwide retailer, and find that 65 % of the records have some
forms of inaccuracy.
Recent analytical research has taken notice, and has begun to address this issue
(e.g., Kok and Shang 2007). Lee and Özer (2007) study the value of RFID tech-
nology in improving information accuracy. They distinguish inventory records from
actual inventory, and attribute inventory inaccuracy to two causes: inventory mis-
placement, and transaction errors. They also show that RFID technologies allow for
better control and can reduce inventory-related costs.
36 Z.J. Ren

Fig. 2.2 Presentation slide from Fisher M (2005) What can we learn about research style from
physics, medicine and finance? Plenary speech, POMS Conference 2005, Chicago

2.10 Summary

Supply chain innovation constantly brings about new questions and challenges.
While in the area of operations management and supply chain management there
is an overall healthy balance of theoretical and empirical research, we argue that the
area could benefit from more empirical research.
In his keynote address at the 2005 Production and Operations Management
Society (POMS) annual conference, Professor Marshall Fisher delineates an em-
pirical research cycle for the OM research community (Fig. 2.2).
With more empirical research helping us identifying and answering questions
from the real world, we will be able to deepen our knowledge body, better guide
business practice, and also spur further analytical research.
We are also in need of empirical research that directs test and validate the theo-
retical modeling work that has been in blossom in recent decades. For example, as
we mentioned earlier, there has been few empirical or experimental studies that aim
to direct validate the various models measuring the value of demand information.
It will be immensely gratifying if more researchers could step up and take on these
important challenges and contribute to the larger OM community.

References

Armony M, Plambeck EL (2005) The impact of duplicate orders on demand estimation and capac-
ity investment. Manag Sci 51(10):1505–1518
Aviv Y, Federgruen A (1998) The operational benefits of information sharing and vendor managed
inventory (VMI) programs. Working paper, The John M. Olin School of Business, Washington
University
Bayus BL, Jain S, Rao AG (2001) Truth or consequences: an analysis of vaporware and new
product announcements. J Market Res 38(1):3–13
2 Empirical Studies in Information Sharing 37

Cachon GP, Lariviere MA (2001) Contracting to assure supply: how to share demand forecasts in
a supply chain. Manag Sci 47(5):629–646
Cattani K, Hausman W (2000) Why are forecast updates often disappointing? Manuf Serv Oper
Manag 1(2):119–127
Chen F (1998) Echelon reorder points, installation reorder points, and the value of centralized
demand information. Manag Sci 44(12):S221–S234
Clark TH, Hammond J (1997) Reengineering channel reordering process to improve total supply
chain performance. Prod Oper Manag 6(3):248–265
Cohen MA, Ho TH, Ren ZJ, Terwiesch C (2003) Measuring imputed cost in the semiconductor
equipment supply chain. Manag Sci 49(12):1653–1670
Croson R, Donohue K (2006) Behavioral causes of the bullwhip effect and the observed value of
inventory information. Manag Sci 52(3):323–336
Daugherty PJ, Myers MB, Autry CW (1999) Automatic replenishment programs: an empirical
examination. J Bus Logist 20(2):63–82
DeHoratius N, Raman A (2008) Inventory record inaccuracy: an empirical analysis. Manag Sci
54(4):627–641
Fisher M, Raman A (1996) Reducing the cost of demand uncertainty through accurate response to
early sales. Oper Res 44(1):87–99
Gaukler GM, Ozer O, Hausman WH (2008) Order progress information: improved dynamic emer-
gency ordering policies. Prod Oper Manag 17(6):599–613.
Gaur V, Giloni A, Seshadri S (2005) Information sharing in a supply chain under ARMA demand.
Manag Sci 51(6):961–969
Gavirneni S, Kapuscinski R, Tayur S (1999) Value of information in capacitated supply chains.
Manag Sci 45(1):16–24
Graves S, Kletter DB, Hetzel WB (1998) A dynamic model for requirements planning with appli-
cation to supply chain optimization. Oper Res 46(3):S35–S49
Heath DC, Jackson PL (1994) Modeling the evolution of demand forecasts with application to
safety stock analysis in production distribution-systems. IIE Trans 26(3):17–30
Jain A, Moinzadeh K (2005) A supply chain model with reverse information exchange. Manuf
Serv Oper Manag 7(4):360–378
Kok AG, Shang KH (2007) Inspection and replenishment policies for systems with inventory
record inaccuracy. Manuf Serv Oper Manag 9(2):185–205
Kuk G (2004) Effectiveness of vendor-managed inventory in the electronics industry: determinants
and outcomes. Inform Manag 41(5):645–654
Kulp SC, Lee HL, Ofek E (2004) Manufacturer benefits from information integration with retail
customers. Manag Sci 50(4):431–444
Lee H, Özer Ö (2007) Unlocking the value of RFID. Prod Oper Manag 16(1):40–64
Lee HL, Padmanabhan V, Whang SJ (1997) Information distortion in a supply chain: the bullwhip
effect. Manag Sci 43(4):546–558
Lee HL, So KC, Tang CS (2000) The value of information sharing in a two-level supply chain.
Manag Sci 46(5):626–643
Loch CH, Terwiesch C (1998) Communication and uncertainty in concurrent engineering. Manag
Sci 44(8):1032–1048
Moe WW, Fader PS (2002) Fast-track: article using advance purchase orders to forecast new prod-
uct sales. Market Sci 21(3):347–364
Özer Ö, Wei W (2006) Strategic commitment for optimal capacity decision under asymmetric
forecast information. Manag Sci 52(8):1238–1257
Özer Ö, Zheng Y, Chen K-Y (2011) Trust in forecast information sharing. Manag Sci 57(6):
1111–1137
Raman A, DeHoratius N, Ton Z (2001) Execution: the missing link in retail operations. Calif
Manag Rev 43(3):136–152
Ren ZJ, Cohen MA, Ho TH, Terwiesch C (2010) Information sharing in a long-term supply chain
relationship the role of customer review strategy. Oper Res. 58(1):81–93
Tang CS, Rajaram K, Alptekinoğlu A, Ou J (2004) The benefits of advance booking discount
programs: model and analysis. Manag Sci 50(4):465–478
Terwiesch C, Ren ZJ, Ho TH (2005) An empirical analysis of forecast sharing in the semiconductor
equipment supply chain. Manag Sci 51(2):208–220
38 Z.J. Ren

Wacker JG, Hanson M (1997) Some practical advice for manufacturing managers: empirical results
from the global manufacturing research group. Prod Inv Manag J 38(3):64–71
Yao Y, Kohli R, Sherer SA, Cederlund J (2013) Learning curves in collaborative planning, fore-
casting, and replenishment (CPFR) information systems: an empirical analysis from a mobile
phone manufacturer. J Oper Manag 31(6):285–297
Zhou H, Benton WC (2007) Supply chain practice and information sharing. J Oper Manag
25(6):1348–1365
Chapter 3
Collaborative Forecasting in Retail
Supply Chains

Mümin Kurtuluş

Abstract Collaborative forecasting (CF) is a supply chain information sharing ini-

tiative in which two or more members in a supply chain (e.g., a retailer and a man-
ufacturer) share their demand forecasts to form a single shared demand forecast,
which is used as the basis for production and replenishment decisions. This chapter
provides an overview of the existing academic research on CF practices in retail
supply chains. In particular, this chapter focuses on highlighting the models used to
capture the essence of CF practices and summarizes the key findings and insights
about the value of CF.

3.1 Introduction

Over the last three decades, a number of factors such as the proliferation of products,
limited shelf space, changing consumer tastes, and intense competition have resulted
in increased management complexity for the retailers. This management complex-
ity has contributed toward an array of operational inefficiencies such as stock-outs,1
and reveals itself in the form of razor thin profit margins for the retailers. In an effort
to reduce the operational inefficiencies, retailers have experimented with a number
of information sharing arrangements that emphasize the role of collaboration with
manufacturers. One such initiative that has been publicized by Wal-Mart and Procter

M. Kurtuluş ()
Owen Graduate School of Management, Vanderbilt University, Nashville, TN 37203, USA
e-mail: [email protected]
1 In a retail industry report, Gruen et al. (2002) find that stock-out rates vary wildly among retailers
but mostly fall in the range of 5–10 %. In studies that examine faster selling and/or promoted
products, the stock-out rate regularly exceeds 10 %. The overall average stock-out rate worldwide
is estimated at 8.3 %.

© Springer International Publishing Switzerland 2017 39

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 3
40 M. Kurtuluş

and Gamble (P&G) is Vendor Managed Inventory (VMI).2 In a VMI program, the
manufacturer takes the responsibility for managing the retailer’s inventory by de-
ciding on the timing and quantity of replenishments (Fry et al. 2001; Çetinkaya and
Lee 2000; Bernstein et al. 2006). The manufacturer can use information technology
such as Electronic Data Interchange (EDI) to monitor the retailer’s inventory and
sales information.
Wal-Mart deployed the Retail Link in 1992 to facilitate the implementation of
VMI with all of its vendors by allowing vendors to monitor their products’ sales in
real time. Essentially, Retail Link is a web-based platform which allows Wal-Mart
to share all sell through data by stock keeping unit (SKU), by hour, by store as well
as on-hand inventory by SKU, gross margin, inventory turns, and in-stock % with
vendors (Petersen 2013). This detailed information allows the vendors to decide on
when and how much to replenish Wal-Mart distribution centers and stores.
VMI can be viewed as transforming a decentralized supply chain into a central-
ized one where inventory is managed by a central entity (i.e., the vendor). VMI
was built on the premise that it can increase the service level provided to the end
consumers due to the vendor’s ability to better plan and execute replenishments.
Although VMI programs have been successfully adopted in many supply chains
(e.g., Clark and McKenney 1994; Hammond 1994), these programs have also been
scrutinized for not being able to incorporate all information available to supply chain
partners. Specifically, the point-of-sale (POS) data does not convey all relevant in-
formation about demand that is at the retailer’s disposal (Aviv 2002). For instance,
the retailer may have important information about an upcoming event or promotion
which may influence the vendor’s replenishment decision. Aviv (2002) claims that
this lack of complete information by the vendors making the replenishment deci-
sions may explain why a number of VMI programs did not succeed. For instance,
the grocery chain Spartan Stores abandoned a VMI program blaming the vendor for
not incorporating the retail promotions data into its forecasts (Mathews 1995), and
K-Mart abandoned VMI with some of its vendors blaming their poor forecasting
skills (Fiddis 1997).
Efficient Consumer Response (ECR) movement that started in the mid-90s is yet
another initiative that recognizes that trading partners (manufacturers, retailers, etc.)
in a supply chain can serve consumers better, faster and at a lower cost by working
together and harnessing their complementary skills. Essentially, the objective of the
ECR initiative has been to create a conscience about eliminating all activities that
do not add value and focusing on activities that maximize value and efficiency by
placing the needs of the consumers first. The ECR movement focused on several
areas such as (1) increasing the efficiency of the category management3 process via
better promotions, new product introductions and product assortment; (2) efficient

2 VMI is sometimes also referred to as Continuous Replenishment Program or Supplier Managed

Replenishment.
3 Category is defined as a group of interrelated products (e.g., oral care category) and the idea

behind category management is to manage each category as a strategic business unit.

3 Collaborative Forecasting in Retail Supply Chains 41

product replenishment that leads to cost savings through minimizing the amount of
inventory while meeting required service levels; and (3) the development of enabling
technologies (ECR Europe 1997; Seifert 2003; Sheffi 2002).
The Collaborative Planning, Forecasting and Replenishment (CPFR) initiative
builds on both VMI and ECR initiatives that were introduced and implemented with
the goal of mitigating the operational inefficiencies via better information exchange
between trading partners. One can think of CPFR as being the next natural step (fol-
lowing ECR and VMI) in the streamlining of business processes in the supply chain
(Sheffi 2002; Seifert 2003). CPFR is a set of business processes that enable trad-
ing partners in a supply chain to have visibility into one another’s critical demand
forecast information through systematic sharing of plans and information, and iden-
tification and resolution of differences in opinions (VICS 1999).
Essentially, CPFR aims to improve production and replenishment decisions
through the use of a shared demand forecast that is agreed on by the retailer and
the manufacturer. The CPFR process suggests that the retailer and the manufacturer
each create a demand forecast and share this forecast via the use of a collabora-
tion platform. The discrepancies (if any) between the forecasts are resolved through
discussion and information sharing among managers, and a single shared demand
forecast is generated. This single shared demand forecast then serves as the basis
for the planning and replenishment decisions in the supply chain. Typical benefits
of CPFR include improved forecast accuracy, improved reaction times to consumer
demand, improved sales, reduction in inventory, and improved relationship between
trading partners (KJR Consulting 2002; Seifert 2003).
The CPFR initiative was born in 1995, when Wal-Mart partnered with Warner-
Lambert Company to pilot a new model for collaboration on the forecasting and
replenishment of Listerine mouthwash products. This collaborative process was ini-
tially referred to as Collaborative Forecasting and Replenishment (CFAR). The trad-
ing partners shared and compared their sales and order forecasts and resolved the
differences in their forecasts when there were inconsistencies. Such inconsistencies
were usually due to lack of knowledge about events such as Wal-Mart promotions,
which can create significant variability in consumer demand (Seifert 2003). Before
this collaboration, Warner-Lambert would often be unaware of promotions at Wal-
Mart and was often required to hold a lot of inventory to hedge the uncertainty in
demand and prevent stock-outs. During the implementation, Wal-Mart and Warner-
Lambert separately forecasted weekly demand by store and by SKU for the follow-
ing 6 months. After sharing the weekly forecasts for the next 6 month, the partners
would resolve discrepancies on a weekly basis. As a result of the pilot, Wal-Mart
started to place its orders 6 weeks before the promotion to match the 6 week pro-
duction lead-time. By receiving the orders 6 weeks in advance, Warner-Lambert was
able to devise and execute a smoother production plan (Seifert 2003).
Wal-Mart reported that this collaboration initiative increased the service levels
from 85 to 98 % while reducing inventory by 25 % for Listerine mouthwash products
in its stores. The partners also reported benefits such as increased sales, streamlined
operations, and overall increase in consumer satisfaction (Seifert 2003). This pilot
42 M. Kurtuluş

represented a substantial shift in the relationship between trading partners and was
the first in a series of CF practices that demonstrated the potential benefits of sharing
demand forecasts and paving the way for what is known as CPFR today.
Following the successful CF pilot by Wal-Mart and Warner Lambert, Voluntary
Interindustry Commerce Standards (VICS) took a leadership role in developing the
CPFR standards and defining a framework and guidelines for implementing CPFR.
VICS is an industry association established by retailers and manufacturers with the
goal of developing business process standards to improve the efficiency of the retail
supply chains. Since its inception in 1986, VICS has been involved in establish-
ing cross-industry standards that enable information sharing partnerships in retail
supply chains. VICS CPFR committee, which includes retailers such as Wal-Mart
and Walgreens and manufacturers such as Kellogg’s, P&G, and Unilever published
the CPFR implementation guidelines, which describe the suggested implementation
process for CPFR, in 1998.
The VICS CPFR process, which is often referred to as the “nine-step” process,
suggests implementing CPFR in three phases (planning, forecasting and replen-
ishment) which constitute nine steps (Seifert 2003). The nine-step process con-
sists of the following: (1) Develop the front-end collaboration agreement; (2) Cre-
ate the joint business plan; (3) Create the sales forecast based on retailer’s POS
data; (4) Identify exceptions for sales forecast; (5) Resolve/collaborate on exception
items; (6) Create order forecast; (7) Identify exceptions for order forecast; (8) Re-
solve/collaborate on exception items; (9) Generate order. Steps 1 and 2 constitute
the planning phase, steps 3–8 constitute the forecasting phase and step 9 constitutes
the replenishment phase (VICS 1999). The experience gained from various CPFR
pilot implementations during the late nineties have yielded many valuable insights
and led the VICS CPFR committee to revise the original “nine-step” CPFR process
and publish a set of simplified guidelines in 2004 (VICS 2004).
Other successful CPFR pilot implementations include P&G’s partnerships with
retailers such as Metro and Tesco, and Wal-Mart’s partnerships with some of its
suppliers such as Sara Lee (VICS 1999; Accenture 2002). For example, the pilot
between Metro and P&G focused on the planning of promotions in the paper and
home care products category. Metro and P&G agreed on a collaborative process for
managing the promotions, which included areas such as planning and jointly fore-
casting the promotion volumes, monitoring the sales during a promotion and jointly
evaluating the effectiveness of the promotions. The promotion planning process at
Metro and P&G was as follows: P&G submits a sales forecast 12 weeks prior to
the actual promotion. A month later, Metro would add a sales forecast. If the dif-
ference between these two forecasts is substantial, the software (GNX Collaborate
developed by Manugistics) would generate an exception notification which would
require the trading partners to discuss and resolve the differences in forecasts. This
process would result in a sales forecast that was agreed on by both parties. The
inventories, orders and replenishments for each store are then managed based on
this single shared demand forecast generated by trading partners. Throughout the
promotion, the POS data would be jointly monitored by Metro and P&G and ad-
ditional actions would be taken if necessary. The partners have reported significant
improvements in forecast accuracy and service levels as well as significant inventory
reductions (VICS 1999; Seifert 2003).
3 Collaborative Forecasting in Retail Supply Chains 43

In another implementation, Wal-Mart and Sara Lee Branded apparel, piloted

CPFR with twenty-three branded women’s underwear items (including some new
product introductions). This pilot mainly focused on creating a joint sales forecast,
and identifying and resolving the exceptions to the sales forecast. After 24 weeks
of implementation, the involved parties reported a 2 % improvement in retail store
in-stock, a reduction of 14 % in store inventory, a 32 % increase in sales, and a 17 %
increase in inventory turns on the pilot items (VICS 1999). In addition to these
examples, numerous other CPFR success stories from Europe, North and South
America have been documented by Bowman (2002), Smaros (2007), Accenture
(2002), and Seifert (2003).
Since the publication of VICS CPFR guidelines in 1998, numerous companies
have implemented the process. It has been reported that CPFR implementations
have resulted in in-stock percentage improvements of 2–8 %, inventory reductions
of 10–40 % across the supply chain, and 3–4 % reduction in logistics costs (AMR
Research 2001; VICS 2004). Many of the early adopters also reported benefits such
as improved relationships with trading partners as one of the key benefits of CPFR.
However, it was also reported that many of the collaborative forecasting initiatives
did not move beyond pilot implementations (KJR Consulting 2002; SCDigest 2008).
According to a study conducted KJR Consulting, “67 % of Grocery Manufacturers
of America (GMA) member companies were engaged in some form of collaborative
planning, forecasting, and replenishment activity, with only 19 % moving beyond pi-
lot studies.” (KJR Consulting 2002). The cost of setting up a collaboration platform
(which can be significant), the complexity of the (nine-step) process, lack of trust
between trading partners have often been cited as possible reasons for the lack of
adoption of collaborative forecasting (Seifert 2003; White and Roster 2004; SCDi-
gest 2008). There is an ongoing debate on whether CPFR can be scaled beyond pilot
implementations to deliver significant benefits to trading partners (VICS 2007).
Motivated by this debate regarding the value of CPFR practices, the academic
literature on collaborative forecasting has mainly focused on quantifying the value
of CF practices and identifying the conditions under which CF delivers substantial
benefit to trading partners. In the rest of this chapter, we provide an overview of the
academic literature on CF. While the information sharing in supply chains literature
is extensive [see Cachon (2003) and Chen (2003)], in this chapter, we focus on
a subset of this literature where the information sharing is bi-directional and the
information shared is restricted to demand related information. In particular, this
chapter focuses on forecast collaboration practices where two or more members in
a supply chain (e.g., a retailer and a manufacturer) share their demand forecasts
to form a single shared demand forecast which is the basis for the ordering and/or
production decisions in the supply chain. Strategic issues around uni-directional
forecast sharing in a supply chain where forecast information is transmitted from
the retailer to the manufacturer or vice versa have been extensively studied in the
recent years (e.g., Cachon and Lariviere 2001; Terwiesch et al. 2005; Özer and Wei
2006; Taylor and Xiao 2009, 2010; Kurtuluş and Nakkas 2011) but are outside the
scope of this chapter.
44 M. Kurtuluş

The existing research on CF mainly focuses on identifying the environments

where CF programs deliver the highest value to the trading partners and can be
grouped in two broad areas. First, in Sect. 3.2 we provide an overview of research
that is rich in modeling the planning and replenishment aspects and investigates the
benefits of CF in multi-period settings where supply chain partners are coopera-
tive (e.g., Aviv 2001, 2002, 2007). Then, in Sect. 3.3 we provide an overview of
the research that studies the value of CF in a context where supply chain partners
are strategic (e.g., Mishra et al. 2007, 2009; Kurtuluş et al. 2012). We will focus
on highlighting the models used to capture the essence of collaborative forecasting
and summarize the key findings and insights about the value of CF. The reader is
referred to the specific papers for the technicalities and derivations of the results.
Section 3.4 concludes by providing a summary and some future research directions.

3.2 Value of CF When Supply Chain Members are Cooperative

Aviv (2001, 2002, 2007) consider the value of CF in the context of a two stage
supply chain where a manufacturer sells a product to a retailer who then sells it
to the end consumers. The retailer’s and manufacturer’s inventories are replenished
periodically. Orders placed by the manufacturer are received after a lead-time of Lm
periods, and orders placed by the retailer are received after lead-time of Lr periods.
At the end of each period, the retailer and manufacturer incur holding costs hr and
hm per unit of inventory, respectively. In addition, it is assumed that all shortages
are backlogged and the retailer incurs a shortage cost of pr per unit of backlogged
demand. All system parameters are common knowledge and both the retailer and
the manufacturer know the characteristics of the forecasting and demand processes.
Aviv’s research utilizes a first order autoregressive process [i.e., AR(1)] to model
demand evolution (Kahn 1987; Lee et al. 2000). The demand at the retailer in period
n, denoted by dn , is given by

dn − μ = α (dn−1 − μ ) + εn (3.1)

where μ is the long run average demand per period, {εn } are independent and identi-
cally distributed (i.i.d.) random variables, which are normally distributed with mean
zero. The parameter α captures the inter-temporal correlation between demands in
consecutive periods. Aviv (2001) considers a special case of this demand process
where α = 0 so that demand in each period is i.i.d. As the parameter α increases,
the inter-temporal correlation between demand in different periods also increases.
Aviv (2002, 2007) consider the general case where α ∈ (0, 1].
Aviv uses the following linear decomposition model to capture the manufac-
turer’s and the retailer’s ability to collect advanced demand information:
τ τ
εn = ∑ δn,i
r
+ ∑ δn,i
m
+ εn0 (3.2)
i=1 i=1
3 Collaborative Forecasting in Retail Supply Chains 45

where δn,i a is the information obtained by member a ∈ {r, m} about ε during

n
period n − i. That is, the demand variations (εn ) can be statistically explained in
advance by using information collected by the retailer (δn,ir ) and manufacturer (δ m ),
n,i
which are both normally distributed, i.e., δn,i ∼ N(0, σa,i ). The forecast adjustments
a 2

δn,i
a for a ∈ {r, m} can be based on various factors such as the weather conditions,

planned promotions, and events that may affect demand for the product. The values
{σa,i : i ≥ 0} represent the forecasting capability of member a ∈ {r, m} in the supply
chain. The parameter τ is the maximum time in advance from which the members
of the supply chain can start collecting information about a specific ε . The last term
in (3.2), εn0 ∼ N(0, σ02 ), represents the demand uncertainty that can not be explained
in advance at any time.
This demand model has several properties which make it an ideal candidate to
model forecast information sharing (Aviv 2002, 2007). First, this model allows in-
formation to be decentralized. That is, the retailer and manufacturers might have
different forecasts for making the operational decisions. Second, the information
collected by each member in the supply chain is collected over time. Third, this
model allows the inclusion of cross-correlation between each member’s informa-
tion. This is important to capture because the retailer’s and the manufacturer’s fore-
casts may be based on similar information sources. This cross-correlation is defined
as Cor(δn,ir , δ m ) = ρ and applies only if n = n and i = i and is zero otherwise.
n ,i

3.2.1 Value of CF Relative to Decentralized Forecasting

Aviv (2001) considers a special case of the autoregressive demand model introduced
in (3.1) where α = 0 and demands dn = μ + εn , n ≥ 1 are i.i.d. normally distributed
random variables with a known mean μ and standard deviation σ . The sequence of
events in each period is as follows: (1) the manufacturer and the retailer adjust their
demand forecasts; (2) the manufacturer places his order followed by the retailer
placing his order; (3) the manufacturer delivers the retailer’s order for that period
(including any prior backorders); (4) demand for the period is realized and each
party incurs costs.
The main goal of Aviv (2001) is to study how key system parameters affect the
performance and the benefits of CF. To do so, he first considers a local forecasting
(LF) model where the information structure is decentralized. That is, each party
in the supply chain maintains their own forecast, and each party can incorporate
the forecast updates into their own replenishment process. Then, Aviv studies a CF
model where the retailer and the manufacturer jointly maintain a single forecast of
demand, which is used in their individual replenishment policies.
The paper assumes that the retailer and the manufacturer are cooperative, so
that they set their replenishment policies to optimize the long-run cost performance
of the supply chain. Furthermore, Aviv (2001) assumes that the retailer and the
46 M. Kurtuluş

manufacturer use installation based policies where replenishment orders are gen-
erated according to local inventory and demand forecasts. In particular, the paper
focuses on myopic, state-dependent order-up-to policies.
Let Δna denote member a’s history of forecast adjustments up until period n. Let
also ΔnCF be the forecasts adjustments in the CF setting. Then the retailer’s and
manufacturer’s inventory decisions in the LF model are based on Δnr and Δnm whereas
both parties’ inventory policies are based on ΔnCF in the CF model. The forecasting
process in CF can be interpreted as follows: At the beginning of every period n, each
member of the supply chain offers adjustments to the forecasts of future demand.
The retailer and the manufacturer reach the best possible adjustments they can, in
the sense of minimizing their forecast errors. Because the CF model is a special case
of the LF model, the paper only develops and solves the LF model, and then uses it
to study the CF model.
Aviv (2001) shows that the retailer’s demand over the lead-time Lr conditional
upon the history of forecast adjustments, ∑Ll=0
r
dn+l |Δnr , is normally distributed with
r r
mean Mn and standard deviation S . The retailer’s order-up-to level at the beginning
of period n is given by βnr = Mnr + γ r where Mnr = E[∑Ll=0 r
dn+l |Δnr ] is the expected
lead-time demand for the retailer and γ is a fixed safety stock quantity. Manu-
r

facturer’s inventory decisions, on the other hand, are based on the retailer’s order
process which is denoted by {An : n ≥ 1}. Since the retailer places orders accord-
ing to the order-up-to policy described above, the retailer’s order quantities can be
expressed as An = dn−1 + Mnr − Mn−1 r . Similar to the retailer, the manufacturer also

uses a myopic, state-dependent order-up-to policy which is based on manufacturer’s

information Δnm . Aviv (2001) shows that the aggregate anticipated demand (i.e., re-
tailer orders) during the lead-time Lm conditional on adjustments, ∑Ll=0 m
An+l |Δnm , is
m m
normally distributed with mean Mn and standard deviation S . The manufacturer’s
order-up-to level is given by βnm = Mnm + γ m where Mnm = E[∑Ll=0 m
An+l |Δnm ] is the ex-
pected lead-time demand for the manufacturer and γ is a fixed safety stock quan-
m

tity. Aviv assumes that the both the retailer and the manufacturer set safety stock
levels γ r and γ m to minimize the systemwide long-run average cost. The analysis of
the CF model is identical with the only exception being that both parties use ΔnCF
to calculate demand over the lead-time (i.e., inventory decisions are based on better
information).
Aviv (2001) conducts a numerical study to examine the benefits of CF, and
whether CF is expected to be more successful under short or long lead times. The
results, which are based on a comparison of the inventory related operational costs
in LF and CF models, are summarized below.
First, Aviv (2001) finds that, on average, CF improves the supply chain perfor-
mance by reducing the costs by 9.56 %. The paper finds that it is the diversification
of forecasting capabilities that matter and that CF is more beneficial when forecast-
ing capabilities of the supply chain members are more diversified. In Aviv’s own
words, the main driver of the value of CF is “whether or not the partners can bring
something unique to the table.” Suppose that the manufacturer bases its demand
forecast on the past consumer trends and sales, and the retailer bases its demand
forecast on the planned promotions schedule. Then, if the impact of consumer trends
3 Collaborative Forecasting in Retail Supply Chains 47

and of product promotions on demand have a low correlation, CF will likely improve
forecast accuracy and will be more valuable. On the contrary, if both members base
their forecasts on the expected consumer trends only, they are less likely to benefit
from CF (Aviv 2001).
Second, comparing the value of CF in short and long lead-time supply chains,
Aviv (2001) finds that the absolute and the marginal benefits of CF are larger when
the lead-times are smaller. The implication of this result is that supply chain ini-
tiatives that aim at reducing lead-times (e.g., Quick Response) and CF are comple-
mentary. This is because with shorter lead-times, the retailer and the manufacturer
can better utilize the demand related information that they share. For example, if the
manufacturer has long replenishment lead-times and the retailer notifies the man-
ufacturer about an upcoming promotion for his product, neither party will benefit
from this information sharing arrangement because the long lead time prevents the
manufacturer from responding to this information.

3.2.2 Value of CF Relative to Vendor Managed Inventory (VMI)

Aviv (2002) builds on Aviv (2001) by considering the value of CF in a setting where
demand is given by an auto-regressive process in (3.1) with α ∈ (0, 1]. Aviv (2002)
not only compares CF to a decentralized information model but also offers insights
on the value of CF as compared to a VMI type of program where the manufacturer
decides on the retailer’s replenishments.
To this end, Aviv (2002) examines three supply chain information sharing mod-
els. In the first model, which is referred to as locally managed inventory (LMI),
inventory is controlled by each party using local information only (i.e., forecasts
are not shared). In the second model, which is referred to as manufacturer-managed
inventory (MMI), the manufacturer manages the supply chain’s inventory, but the
retailer’s demand forecasts are not available to the manufacturer. This model is sim-
ilar to VMI programs where the demand information available to the retailer is not
entirely shared with the manufacturer. In the third model, which is referred to as col-
laborative forecasting and replenishment (CFAR), inventory is managed centrally,
and all demand forecasts are shared and replenishment decisions are based on a
joint demand forecast. By comparing the cost performance in these three settings,
Aviv (2002) provides guidelines as to when each of these settings is expected to be
beneficial to the trading partners in a supply chain.
The sequence of events is as follows: (1) Forecasts are adjusted: Each party fore-
casts demand separately in the LMI, only the manufacturer’s forecast adjustments
are used in MMI, and parties maintain a single, joint forecast that is based on the
retailer’s and manufacturer’s forecast adjustments in CFAR; (2) Replenishment or-
ders and deliveries take place: The retailer orders are placed before the manufacturer
places his orders in LMI, the manufacturer is responsible for managing inventory at
48 M. Kurtuluş

both levels in the MMI model, and a centralized decision maker is responsible for
managing the inventories at both levels in CFAR model; (3) Demand is realized, and
costs are incurred.
In all settings, orders for each echelon are placed according to a base-stock policy
where each party orders enough units to bring the inventory position as close as
possible to their best estimate of demand over the lead-time, plus a predetermined,
fixed safety stock. The replenishment policy parameters are set to minimize the
long-run average total supply chain cost.
In the LMI setting, each member considers his local inventory position, defined
as the number of units on hand, minus backlogs, plus all outstanding orders. Similar
to Aviv (2001), the order-up-to level is given by βna = Mna + γ a where Mna is the ex-
pected lead-time demand conditional on the information available to each member
for a ∈ {r, m} and γ a is a safety stock term independent of the demand forecast. In
the MMI and CFAR models, the manufacturer acts as a centralized decision maker
and is responsible for replenishing the inventories both at the retailer and at the man-
ufacturer in an echelon-based manner. The characterization of the MMI and CFAR
models is easier than the LMI model because decisions are based on centralized in-
formation. The model used for explaining the errors in (3.2) can be rewritten for the
MMI and CFAR models as follows:
τ
εn = ∑ δ̂n,i
E
+ ε̂n0 ,
i=1

where δ̂n,i
E is based on manufacturer’s signal in the MMI and both manufacturer’s

and retailer’s forecast adjustments in CFAR model. The superscript E is used to

denote echelon-based settings.
In an echelon-based policy, the manufacturer monitors the inventory position of
the entire supply chain, instead of his local inventory position (Clark and Scarf
1960). The echelon base stock levels are given by Ban = Mna + Γ a for a ∈ {r, m}.
Mnr is the expected lead-time demand conditional on the information available to
the retailer and Mnm is the expected lead-time demand over Lr + Lm periods con-
ditional on the information available to the central decision maker. Γ a , a ∈ {r, m}
is a safety stock term independent of the demand forecasts. The reader is referred
to Aviv (2002) for the characterization of the base-stock levels for each model. In
what follows, we focus on some of the main insights on the value of VMI and CF
programs.
First, comparing the LMI and MMI models, Aviv finds that the cost under MMI
increases as the manufacturer’s relative (to the retailer) explanatory power declines,
and there exists a threshold for the relative explanatory power of the retailer below
which LMI performs better than MMI. It is intuitive that when the manufacturer’s
demand forecast is not good, the MMI setting would be less beneficial, and it ac-
tually may result in a loss when it replaces LMI. This observation is in line with
several VMI implementations in practice where the manufacturer’s poor forecasting
performance was one of the main reasons behind dissatisfaction with VMI programs
[see Mathews (1995) and Fiddis (1997) for examples].
3 Collaborative Forecasting in Retail Supply Chains 49

Second, comparing the MMI and CFAR, Aviv finds that the value of CF relative
to MMI depends on (1) inter-temporal correlation between demand in consecutive
periods (i.e., the parameter α ); (2) the correlation between the retailer’s and man-
ufacturer’s demand forecasts (i.e., the parameter ρ ); and (3) the retailer’s relative
explanatory power. Building on Aviv (2001) which suggests that significant bene-
fits can be achieved by taking into account early demand information in a setting
where demand in consecutive periods is i.i.d. (i.e., α = 0), Aviv (2002) shows that
when α > 0, the benefits are even larger. Not only in absolute terms but also the
value of CFAR relative to MMI increases as the inter-temporal correlation between
demand in consecutive periods increases. In addition, the performance gap between
CFAR and MMI is maximized as the correlation between the retailer’s and manufac-
turer’s demand forecasts gets smaller. As the correlation gets closer to one, all three
models LMI, MMI and CFAR become equivalent because the manufacturer and the
retailer observe the same signals. Furthermore, the value of CFAR relative to MMI
is greater when the retailer’s relative explanatory power is high. This is because the
manufacturer does not take into account the retailer’s demand forecast in MMI but
the central decision maker does use the retailer’s information in the CFAR.
To summarize, Aviv (2002) concludes that “the consideration of MMI and CFAR
programs becomes more important as the demand process is more correlated across
periods, and as companies are able to explain a larger portion of the demand
uncertainty through early demand information.” When the manufacturer’s relative
explanatory power is large (compared to the retailer’s), the MMI model may be the
best model. When the relative explanatory power of the retailer and the manufac-
turer are equal, the best model depends on the correlation between demand signals:
If the correlation is small, CFAR may be the option that leads to the best perfor-
mance, whereas if the correlation is high, the three settings perform similar. Finally,
when the retailer’s relative explanatory power is large, the CFAR may be the best
model.

3.2.3 Value of CF Due to Inventory, Production Smoothing,

and Adherence-to-Plans

Aviv (2001, 2002) consider the value of CF in a cooperative supply chain and fo-
cus on inventory related cost benefits of CF. Building on these papers, Aviv (2007)
studies the value of CF in a decentralized supply chain where the manufacturer is
not only concerned with inventory costs but also cares about the cost of production
smoothing and keeping a stable production schedule (Graves et al. 1998). The re-
duction of costs associated with production smoothing and adherence to plans have
been frequently cited as some of the key benefits of using a single shared demand
forecast in collaborative partnerships such as CPFR (AMR Research 2001; KJR
Consulting 2002).
Similar to his prior papers on CF, Aviv (2007) also uses the autoregressive de-
mand model in (3.1). The paper uses the following additional notation: Inr denotes
50 M. Kurtuluş

the retailer’s inventory position (= on hand, minus backlogs, plus all outstanding or-
ders) and Inm denotes the manufacturer’s net inventory (= on hand, minus backlogs)
at the beginning of period n. At the beginning of period n, the manufacturer decides
on a production plan for the immediate planning horizon of T periods which is de-
.
noted by qn = (qn,n , qn,n+1 , . . . , qn,n+T ) where qn,i is the production plan decided at
the beginning of period n for period i.
Operational Efficiency Metrics The first metric of operational efficiency relates
r (I r ) = .
to the inventory holding cost, which is defined as CI,h n hr max{Inr , 0}, where
In is the net inventory at the retailer at the end of period n. For the manufacturer,
r
.
m (I m ) =
the inventory cost metric is defined as CI,h n hm max{Inm , 0}. The second metric
relates to the manufacturer’s actual use of production capacity, which is captured
through the cost function CPm (qn,n ) = cP (qn,n − μ )2 . The parameter cP captures the
cost due to variability in production at the manufacturer. This convex cost function
represents an increasing marginal cost of production that results from variations in
production quantities and provides an incentive for production smoothing (Kahn
1987). When cP is larger compared to manufacturer’s holding cost hm , it is more
reasonable to produce at a steady rate, and manage supply demand mismatches via
holding inventory or backordering. The third metric of operational efficiency relates
to the adherence to production plans and measures the degree managers stick to
production plans specified in previous periods and reflects that changes made at
the last moment are more costly than those made in advance. The adherence to the
production plans cost function is defined as CAm (un,0 , . . . , un,T ) = ∑Tj=0 cA, j u2n, j where
un, j = qn,n+ j − qn−1,n+ j if j ∈ [0, T ) and un, j = qn,n+T − μ if j = T and parameters
cA, j > 0, ∀ j, capture the rigidity of the manufacturing process.
Service Performance Metrics Aviv uses the following two service performance
r (I r ) =. .
metrics: CI,p n pr max{−Inr , 0} and CI,p
m (I m ) =
n pm max{−Inm , 0} where pm is
a per-unit penalty cost that the manufacturer assigns for backorders. These metrics
ensure that both the retailer and the manufacturer have incentives to avoid shortages
(i.e., backorders). Furthermore, Aviv refers to the ratio pm /(hm + pm ) as the internal
service parameter, which is exogenously given.
To evaluate the value of CF in decentralized supply chains, Aviv focuses on
cases where the retailer minimizes the sum of long-run average cost, C̄I,h r + C̄r =
I,p
limN→∞ ∑n=1 [CI,h (In ) +CI,p (In )], and makes a trade-off between carrying inven-
N r r r r

tory and the risk of shortage. The manufacturer, on the other hand, focuses on mini-
m + C̄m + C̄m + C̄m where the first three
mizing the long-run average cost given by C̄I,h P A I,p
components capture the efficiency related metrics and the last component captures
the service performance metric. The paper considers a base case model where both
parties make decisions on local information and a CF model where they maintain
a single joint demand forecast which serves as the basis for retailer’s and manu-
facturer’s decisions. The results on value of CF are based on the cost performance
comparison in these two models.
The retailer uses a base-stock policy of the form βnr = Mnr + γ r where the first
component is the expected demand over the lead-time Lr and the second component
is a safety stock independent of the demand forecasts but depends on the internal
3 Collaborative Forecasting in Retail Supply Chains 51

service rate. The retailer sets γ r that minimizes the long-run average cost. The man-
ufacturer, on the other hand, uses a production strategy that is based on a class of
policies that are derived by defining and solving a surrogate linear-quadratic Gaus-
sian (LQG) problem. The readers is referred to Aviv (2007) for the details of the
production planning policy used by the manufacturer and the algorithm used to cal-
culate the retailer’s safety stock γ r . In what follows, we focus on the insights about
the value of CF.
Aviv (2007) concludes that CF leads to 4 % increase in supply chain perfor-
mance. However, the benefits of CF vary substantially depending on the member
in the supply chain and characteristics such as (1) the relative explanatory power
of the retailer and the manufacturer; (2) the supply side agility; and (3) the internal
service rate.
First, Aviv (2007) finds that the value of CF is higher for the supply chain as
a whole when the manufacturer has the largest relative explanatory power. This is
because an accurate demand forecast is more important for the retailer compared
to the manufacturer because lack of inventory at the retailer is more costly to the
supply chain compared to lack of inventory at the manufacturer. Hence, when the
manufacturer has higher explanatory power relative to the retailer, the supply chain
benefits more from CF. In addition, the manufacturer can extrapolate some informa-
tion about consumer demand from the retailer’s orders even without CF.
Second, the value of CF increases as the supply side becomes more agile. Aviv
argues that this is because improved demand information in a CF program is valu-
able only if the supply chain members can act upon it. If the manufacturer responds
very slowly to information, the value of CF will be negligible. This finding is sim-
ilar in spirit to the finding in Aviv (2001), which suggests that CF is more valuable
when the supply chain partners can react to improved demand information (i.e.,
when lead-times are short).
Third, Aviv examines the natural split of benefits due to CF and observes that the
benefits of CF are not split proportionally. In fact, in most cases where the benefits
of CF are the highest for the supply chain, the CF partnership is not valuable for the
manufacturer. This is because when the manufacturer collects highly explanatory
demand information in advance and the retailer does not have access to such infor-
mation, the manufacturer can predict the retailer’s orders quite well. As a result, the
manufacturer can manage the production in an efficient manner. When the manufac-
turer starts to share demand information in a CF initiative, the manufacturer might
be giving up a substantial degree of his ability to anticipate retailer’s future orders.
Furthermore, Aviv (2007) suggests that the way in which the total benefits are split
between the retailer and the manufacturer is highly dependent on the internal ser-
vice rate. Aviv’s analysis assumes that the internal service rate is the same before
and after CF. However, Aviv’s analysis reveals that revising the internal service rate
can not only result in an increase in the benefits of CF, but also help reposition an
unattractive CF partnership closer to a win–win outcome. The key implication of
this finding is that CF partnership alone may not lead to a beneficial outcome for
both partners and hence it may be important to complement the CF partnership with
a mechanism that could facilitate the sharing of benefits.
52 M. Kurtuluş

3.3 Value of CF When Supply Chain Members are Strategic

Aviv’s papers are rich in modeling the replenishment and forecasting processes in
multi-period supply chain models. Both Aviv (2001, 2002) consider centralized set-
tings in the sense that the retailer and the manufacturer coordinate their policy pa-
rameters in an attempt to minimize systemwide costs. Aviv (2007) deviates from this
assumption by considering a decentralized supply chain where both the retailer and
the manufacturer take actions to minimize their individual costs. More recent papers
(e.g., Mishra et al. 2007, 2009; Kurtuluş et al. 2012), however, focus on strategic in-
teractions in the context of CF. The papers in this stream capture the essence of the
quantity/replenishment decisions by using a one-period newsvendor model but are
rich in modeling the strategic interaction between the retailer and the manufacturer.
In what follows, we summarize the key findings related to the value of CF in the
context where the retailer and the manufacturer are strategic.
All papers in this stream focus on a supply chain where a manufacturer sells a
product to consumers via a retailer. The production cost to the manufacturer is c
per unit and the manufacturer sells the product to the retailer at a wholesale price w
per unit. The retailer sells the product to consumers at a retail price p per unit. The
papers in this stream use a forecasting model based on Winkler (1981) and Clemen
and Winkler (1985).
It is assumed that the manufacturer and the retailer have common prior informa-
tion about some aspect of consumer demand, X, which is normally distributed with
mean μ and variance σ02 . The manufacturer and the retailer each observe a signal
denoted by xm and xr , respectively, that carry information about the random vari-
able X. The manufacturer’s and the retailer’s forecast errors, xm − X and xr − X, are
normally distributed with mean zero and variances σm2 and σr2 , respectively. A fore-
cast error with large variance is associated with a less precise forecast. The errors in
the manufacturer’s and the retailer’s forecasts can be correlated but are independent
of X. The correlation, ρ , is due to the fact that both the manufacturer and the re-
tailer might utilize some common data, share common assumptions, or have access
to some of each other’s opinions. If the retailer’s and the manufacturer’s forecasts
are based on similar information, the correlation between the forecasts will be high.
The papers consider non-negative correlations (i.e., ρ ∈ [0, 1)) with ρ = 0 implying
that the retailer and the manufacturer’s forecasts are independent and ρ → 1 imply-
ing that the retailer and the manufacturer observe the same forecast. The variances
and correlation between the forecast errors capture the quality of the additional in-
formation obtained from a signal; a higher variance implies a less accurate signal,
whereas a higher correlation implies substantial information overlap between the
signals.
In the CF, the retailer and the manufacturer share their demand forecasts to form
a single shared demand forecast which is denoted by X|xm , xr . The shared forecast
X|xr , xm is normally distributed with mean E[X|xm , xr ] = z1 μ + z2 xm + z3 xr where
z1 + z2 + z3 = 1 and variance
3 Collaborative Forecasting in Retail Supply Chains 53

(1 − ρ 2 )σ02 σm2 σr2

Var[X|xm , xr ] = .
(1 − ρ )σm σr2 + σ02 (σr2 + σm2 − 2ρσm σr )
2 2

Note that E[X|xm , xr ] is a weighted average of the prior mean μ and the observed sig-
nals xm and xr and the variance Var[X|xm , xr ] is independent of the signal realizations
(Clemen and Winkler 1985). Following Winkler (1981), it is assumed that the co-
variance is not greater than individual variances, i.e., ρ ≤ min{σr /σm , σm /σr } (i.e.,
the covariance between the signals is smaller than the variance). This assumption
ensures that each party’s forecast is given non-negative weight in the final forecast.
All parameters (i.e., μ , σ02 , σm2 , σr2 , and ρ ) are common knowledge.

3.3.1 Incentives for Sharing Forecasts and Value of CF

Due to Better Pricing

Mishra et al. (2007, 2009) focus on (1) incentives to share forecasts in CF and
(2) the value of CF due to both making better pricing and inventory decisions in a
supply chain operating under different production strategies such as make-to-order
and make-to-stock. In the make-to-order model, neither the manufacturer nor the
retailer keeps any inventory because the manufacturer makes the production deci-
sion (and the retailer makes the ordering decision) after demand is realized. Most
computer manufacturers such as Dell and HP that sell their products online fit this
model. In the make-to-stock model, the manufacturer makes both production and
pricing decisions before, and the retailer makes the pricing decision before but the
ordering decision after demand is realized. There are many apparel and auto manu-
facturers that produce in a make-to-stock fashion. Effectively, in the make-to-order
scenario, demand forecast sharing affects the pricing decisions only allowing the
authors to isolate the value of CF which is due to better pricing decisions whereas in
the make-to-stock scenario, information sharing affects both pricing and production
(i.e., manufacturer’s inventory) decisions.
The consumer demand, denoted by d, is given by d = X − bp where X is the
market potential, p is the retail price, and b is the price sensitivity of demand. The
model captures uncertainty in demand by assuming that the demand intercept X is a
random variable. Specifically, the demand intercept X is given by X = μ + e where e
is normally distributed with mean zero and variance σ02 (Vives 1984; Gal-Or 1985).
The retailer and the manufacturer each observe a signal about the demand intercept,
X, which are denoted as xr and xm , respectively.
In each scenario, the authors derive the equilibrium profits when there is (1) no
forecast sharing (i.e., non-collaborative model) and (2) collaborative forecasting
where forecasts are shared. In the non-collaborative model, the manufacturer only
uses its own forecast xm , but the retailer has its own forecast xr and the wholesale
price w (from which the retailer may be able to infer xm ). In the CF model, both
parties share their forecasts xr and xm . The authors compare the profits in these two
models for both make-to-order and make-to-stock scenarios to identify conditions
under which forecast sharing will occur and study the value of CF.
54 M. Kurtuluş

In the make-to-order scenario, the sequence of events is as follows: (1) each

party forecasts the demand; (2) the manufacturer sets the wholesale price w and
then the retailer sets its retail price p; and (3) demand is realized, and production
and sales take place. Each model is solved via backward induction. In the non-
collaborative model, the retailer sets the retail price to maximize expected profit and
solves max p (p − w)(E[X|xr , w] − bp), and the manufacturer sets wholesale price by
solving maxw (w − c)(E[X|xm ] − bE[p(xr , w)|xm )). In the CF model, the retailer sets
the retail price by solving max p (p − w)(E[X|xr , xm , w] − bp), and the manufacturer
sets wholesale price by solving maxw (w − c)(E[X|xr , xm ] − bE[p(xr , xm , w)]).
In the make-to-stock scenario, the sequence of events is as follows: (1) each party
forecasts demand; (2) the manufacturer sets the wholesale price w and the produc-
tion quantity Q, and production occurs; (3) the retailer sets retail price p; (4) demand
is realized, the retailer places the order, sales take place and inventory or shortage
costs are incurred. It is important to note that, the manufacturer produces before the
demand is realized but the retailer places its order after demand is realized. Hence,
the inventory holding and shortage costs are incurred by the manufacturer only. If
the demand is less than Q, then the manufacturer incurs a holding cost and if demand
exceeds Q, the manufacturer can obtain additional units from an external source at
an additional cost. The reader is referred to Mishra et al. (2009) for the derivation
of the equilibrium prices, quantities and profits in both make-to-order and make-to-
stock scenarios. In what follows we highlight the key insights regarding the value
of CF in make-to-order and make-to-stock systems.
The authors show that while information sharing benefits the manufacturer, it
hurts the retailer. Hence, the retailer does not have incentive to share its forecast
voluntarily but the manufacturer always prefers to have access to the retailer’s fore-
cast. The question in this context becomes whether the manufacturer can induce the
retailer to share its forecast. Mishra et al. (2009) find that when the supply chain
operates in a make-to-order fashion, the manufacturer cannot buy the retailer’s fore-
cast by paying a fixed fee but the manufacturer can design a contract that offers a
discount to the retailer under certain conditions in return for the retailer’s forecast,
which will induce the retailer to share its forecast. The proposed discount contract
improves the consumer surplus, supply chain profit, and social welfare. Further-
more, Mishra et al. (2009) find that collaborative forecasting can be especially valu-
able for both the retailer and the manufacturer when the consumer demand is highly
uncertain, the accuracy of the retailer’s forecast is high, and the correlation between
forecasts is low.
When the supply chain operates in a make-to-stock fashion, on the other hand,
Mishra et al. (2009) find that the manufacturer may be able to obtain the retailer’s
forecast for a fee if the manufacturer’s inventory holding and shortage cost savings
are sufficiently high. Hence, the authors conclude that while a side payment con-
tract does not induce forecast sharing in the make-to-order scenario, a side payment
contract may be sufficient to induce forecast sharing in the make-to-stock scenario.
Finally, comparing the value of CF in the make-to-order and make-to-stock supply
chains, Mishra et al. (2009) find that the value of CF for the manufacturer is higher
3 Collaborative Forecasting in Retail Supply Chains 55

in a make-to-stock than in a make-to-order setting. This is because in addition to bet-

ter pricing, the improved demand forecast in CF allows the manufacturer to make
better quantity decision in the make-to-stock model.

3.3.2 Value of CF When the Forecasting Investments

are Endogenous

All of the papers described so far investigate the value of CF in a setting where
the retailer and the manufacturer combine their forecasts with exogenous accuracy.
Moreover, the accuracies of the forecasts are independent of whether CF is im-
plemented or not. Kurtuluş et al. (2012) bring a new dimension to the forecasting
process by assuming that supply chain partners can invest in improving their fore-
casts by making costly investments into forecasting (i.e., forecast accuracies are en-
dogenously determined). This is important because in the context of CF, the supply
chain parties must exert costly effort to obtain relevant data, improve data quality,
and generate the forecasts. In addition, forecast improvements by one party usu-
ally benefit both parties in the supply chain, and implementing CF can lead to a
change in each party’s incentive to exert effort into forecasting. Furthermore, CF is
typically implemented on top of an existing supply chain relationship. The nature of
this relationship (terms of trade and power balance) may have an important effect on
its success and value of CF initiatives. In this context, Kurtuluş et al. (2012) study
(1) the conditions that are favorable to the adoption of CF and the characteristics
that yield substantial benefits from CF for the retailer and the manufacturer; (2) the
impact of terms of trade and power balance on the value of CF for the involved par-
ties; and (3) whether CF is more valuable in settings where a simple wholesale price
contract is used or where a coordinating contract is used.
Similar to Mishra et al. (2007, 2009), Kurtuluş et al. (2012) also consider a supply
chain where a manufacturer sells its product to a retailer who serves end-consumers.
Consumer demand is uncertain and both the manufacturer and the retailer have com-
mon prior information and believe that demand X is normally distributed. The man-
ufacturer and the retailer have forecasting capabilities and can observe signals xm
and xr to resolve some of the uncertainty about consumer demand X. Both the man-
ufacturer and the retailer can make costly forecasting investments to improve the
quality of their demand forecasts. A higher investment by either party leads to a
more accurate forecast in the sense that the forecast error has a lower variance. For
example, as the retailer increases its forecasting investment, the retailer observes
signal xr with a smaller forecast error variance σr2 . A forecasting investment of yi ,
i ∈ {r, m} will result in forecast with error variance of σi2 (yi ) = σ 2 /yi which is
decreasing in yi . The cost of making investment yi is given by ki yqi for i ∈ {r, m}
where ki refers to firm i’s forecasting capability with lower k referring to a more
capable party. The parameter q refers to the forecasting technology and captures the
diseconomies in the forecasting investments.
56 M. Kurtuluş

The authors first consider a non-collaborative model where the manufacturer and
the retailer can invest in forecasting but do not share their forecasts. Then, they con-
sider a CF model where the manufacturer and the retailer combine their information
to form a single shared demand forecast. The authors compare these two models
to study the value of CF. Furthermore, this comparison is done under three con-
tractual forms that are widely used in practice: (1) retailer managed inventory with
wholesale pricing (RMI), where the retailer makes the quantity decision; (2) man-
ufacturer managed inventory with wholesale pricing (MMI), where the manufac-
turer makes the quantity decision; and (3) retailer managed inventory with buyback
(BB) where the retailer makes the quantity decision but is allowed to returns unsold
inventory to the manufacturer for a refund.
In the absence of CF, only one party invests into forecast improvement: The
party making the quantity decision in the face of uncertainty, which is referred
to as “the newsvendor,” benefits from improving the forecast and the other party
gains nothing from its own forecast investment because its profits are solely de-
termined by the newsvendor’s√decision. In RMI, the retailer’s investment is given
q
by maxyr (p − w)μ − HrRMI σ / yr − k√ r yr . In MMI, the manufacturer’s investment is
given by maxym (w − c)μ − Hm σ / ym − km yqm . In BB, the retailer’s investment
MMI
√
is given by maxyr (p − w)μ − HrBB σ / yr − kr yqr . The expressions Hij for i ∈ {r, m}
and j ∈ {RMI, MMI, BB} refer to the unit cost of uncertainity for member i in sce-
nario j [see Kurtuluş et al. (2012) for derivations of Hij terms for each member i in
scenario j]. The unit cost of uncertainty is a function of the cost parameters and the
terms of trade under which the supply chain partners operate.
While only the newsvendor (the party facing demand uncertainty) makes the
forecast investment in the non-collaborative scenario, both the manufacturer and the
retailer can invest into forecasting in the CF scenario. This is because both parties
benefit from a better demand forecast. However, the investments into forecasting be-
come a strategic decision in the CF model because party i’s profit is not only a func-
tion of its own investment but also a function of the other supply chain √ member. The
retailer decides on its investment by solving maxyr (p−w)μ −Hrj σ / yr + ym −kr yqr
√
and the manufacturer decides on its investment maxym (w − c)μ − Hmj σ / yr + ym −
km yqm for each contractual structure j ∈ {RMI, MMI, BB}. The authors show that
there exists a unique equilibrium in forecasting investments.
First, Kurtuluş et al. (2012) show that the adoption potential of CF is product-
and relationship-specific. The paper shows that CF may not be adopted because
one of the trading partners can be worse off after implementing CF. In particular,
the adoption of CF requires either that the terms of trade are such that the non-
newsvendor’s relative normalized cost of uncertainty is sufficiently large relative
to the newsvendor’s, or that the diseconomies in the forecast investment cost is
sufficiently large. Furthermore, similar to Aviv (2001, 2002, 2007) and Mishra et al.
(2009), Kurtuluş et al. (2012) confirm that the value of CF always increases for both
the retailer and the manufacturer when the information available to the manufacturer
and the retailer is more distinct (i.e., low correlation between signals), and as the
uncertainty of the prior increases.
3 Collaborative Forecasting in Retail Supply Chains 57

Second, the authors study how power balance in the supply chain impacts the
value of CF by using the relative share of operating profits appropriated by each
party as a proxy for supply chain power. The authors find that the value of CF for
the newsvendor is non-monotonic in its power. For example, consider the benefit of
CF for the retailer under a coordinating buyback contract. If the retailer is powerful
and appropriates a large share of the operating profit, the manufacturer’s forecast
investment is low, limiting the value of CF for both parties. As the manufacturer’s
share of the operating profit increases, its investment level increases; the retailer’s
benefit from CF initially does increase but starts decreasing as the manufacturer
becomes too powerful.
Finally, comparing the value of CF under a coordinating (e.g., buyback) and
non-coordinating (e.g., wholesale price) contracts, the adoption of CF is boosted
by quantity coordination. However, the magnitude of the benefits from CF are in
many cases higher in the absence of quantity coordination: When the manufacturer
appropriates a larger share of the operating profit, he gains more from CF under the
wholesale price contract. Since a larger forecasting investment by the manufacturer
translates into a higher positive spillover effect on the retailer (in the form of higher
accuracy achieved at lower cost), the retailer’s gain from CF is higher under the
non-coordinating contract as well. This finding suggests that in addition to reducing
the cost of the supply-demand mismatch, the improved demand information due to
CF has the added benefit of countering the adverse effects of double marginalization
inherent in the wholesale price contract.

3.4 Conclusions and Future Research Directions

The main focus of the research on CF has been to identify the environments where
CF delivers substantial benefits to the trading partners in a supply chain. The key
findings of this research stream are as follows. First, CF is more valuable when im-
plemented with products with highly uncertain demand (Aviv 2001, 2002, 2007;
Kurtuluş et al. 2012). The implication of this finding is that CF can provide substan-
tial benefits when implemented in the context of “irregular” one time events with
relatively large uncertainty in demand such as promotion planning and new product
introductions. On the contrary, CF is likely to be less valuable with functional prod-
ucts for which future demand patterns can be easily extrapolated from past demand.
Second, CF is more valuable when the correlation between trading partners’ in-
formation is low (Aviv 2001, 2002, 2007; Mishra et al. 2009; Kurtuluş et al. 2012).
In other words, the value of CF for trading partners increases when the information
available to the manufacturer and the retailer are more distinct. In Aviv’s words, “. . .
CF is more valuable when partners can put something unique on the table.” The im-
plication of this finding is that while negotiating the up-front agreement, the trading
partners should discuss the specific areas in which each partner should be focusing
on and making sure that the overlap in information conveyed by the retailer’s and
manufacturer’s forecasts is minimized.
58 M. Kurtuluş

Third, CF is more valuable when partners can respond to better information (Aviv
2001, 2007). For example Aviv (2001) finds that CF is more valuable when the
lead-times are shorter and Aviv (2007) finds that CF is more valuable when the
manufacturer is more agile (i.e., has the ability to adjust production). This is because
CF leads to a better demand forecast and trading partners can not benefit from CF
unless they can respond to this better forecast. The implication of this finding is
that, investing in lead-time reduction or agility improvement at the manufacturer
are complementary to a CF program and can be very valuable in a setting where
partners are in a collaborative relationship.
Fourth, the trade literature has focused on the operational benefits of CF such
as lower inventory levels and improved service levels. Most of the academic liter-
ature on CF also focuses on quantifying the benefits of CF due to better inventory
management in terms of matching supply and demand in a more cost effective way.
However, Mishra et al. (2009) demonstrates that CF can be valuable because it can
facilitate better pricing decisions and improve revenue for the trading partners. The
implication of this result is that trading partners should adopt a broader perspective
in evaluating the benefits of CF and look beyond operational improvements to get a
more accurate assessment of the value of CF.
Finally, it has been shown that strategic interactions in the context of CF can
play an important role in the value each member derives from CF (e.g., Mishra et al.
2009; Kurtuluş et al. 2012). For example, Mishra et al. (2009) shows that when
implemented in the context of a simple wholesale price contract, the manufacturer
may have an incentive to distort its forecast and suggest a discount scheme that
can overcome this challenge. Kurtuluş et al. (2012) shows that the strategic nature
of the investments into forecasting and the terms of trade might lead to situations
where one of the trading partners may be worse off after implementing CF. The
implication of these findings is that trading partners should consider the impact of
these strategic issues and address them before implementing CF. As emphasized in
the trade literature on CF, “trust” between trading partners can play an important
role in overcoming the barriers due to strategic interactions in the supply chain.
As discussed in the introduction, CPFR can be viewed as a natural evolution
of the VMI and ECR movements that emerged in the 1990s. Although CPFR was
very popular in the late 1990s and 2000s, CPFR itself has paved the way for other
collaborative relationships in the retailing industry. For example, many retailers col-
laboratively manage the category management process with some of their leading
manufacturers. In fact, many retailers have partnered with one manufacturer in the
category for not only collaborative forecasting but also other aspects of category
management such as assortment planning and pricing (e.g., Kurtuluş and Toktay
2011; Kurtuluş and Nakkas 2011). Therefore, it should be noted that CPFR has
evolved into other practices which are extensively used by retailers and manufactur-
ers in the consumer goods industry.
3 Collaborative Forecasting in Retail Supply Chains 59

3.4.1 Future Research Directions

Various aspects of CF deserve further attention. First, all existing research focuses
on investigating the value of CF in a supply chain with one manufacturer and one
retailer. One avenue for future research would be to evaluate the value of CF in a
supply chain with multiple retailers and/or multiple manufacturers. In this context,
partners involved in CF might fear the leakage of valuable information to competi-
tors, which can deter the trading partners from implementing CF. Future research
can explore the incentives for information leakages in CF and mechanisms to pre-
vent information leakages.
Second, one of the concerns raised in practice about implementing CF is the
amount of effort parties put in discussing and resolving the discrepancies in the
forecasts to form a consensus forecast. For this reason, automation of the reso-
lution of the discrepancies could be very useful. An interesting avenue for future
research would be to explore and compare the performance of various automation
rules. Third, although there are several surveys evaluating the state of CPFR prac-
tices from the early 2000s, there is a need for a more recent extensive industry survey
to assess the state of CF practices. Finally, as discussed before, CF has higher po-
tential to deliver substantial benefits when implemented in promotion planning and
new product introductions. Hence, practitioners could benefit from both theoretical
and empirical research that evaluates and quantifies the value of CF in a context
which specifically captures the key elements of the promotion planning and new
product introduction contexts.

References

Accenture (2002) European CPFR insights. Report prepared by Accenture and ECR Europe. http://
ecr-all.org/wp-content/uploads/pub 2002 cpfr european insights.pdf. Accessed 9 Apr 2015
AMR Research (2001) Beyond CPFR: collaboration comes of age – the report on retail e-business.
http://www.amrresearch.com
Aviv Y (2001) The effect of collaborative forecasting on supply chain performance. Manag Sci
47(10):1326–1343
Aviv Y (2002) Gaining benefits from joint forecasting and replenishment processes: the case of
auto-correlated demand. Manuf Serv Oper Manag 4(1):55–74
Aviv Y (2007) On the benefits of collaborative forecasting partnerships between retailers and man-
ufacturers. Manag Sci 53(5):777–794
Bernstein F, Chen F, Federgruen A (2006) Coordinating supply chains with simple pricing
schemes: the role of vendor-managed inventories. Manag Sci 52(10):1483–1492
Bowman RJ (2002) European grocery supplier shows how CPFR really works. Supply Chain Brain.
http://www.supplychainbrain.com. Accessed 9 Apr 2015
Cachon GP (2003) Supply chain coordination with contracts. In: Handbooks in operations research
and management science, vol. 11. Elsevier, Amsterdam, p 227–339
Cachon GP, Lariviere MA (2001) Contracting to assure supply: how to share demand forecasts in
a supply chain. Manag Sci 47(5):629–646
60 M. Kurtuluş

Çetinkaya S, Lee CY (2000) Stock replenishment and shipment scheduling for vendor-managed
inventory systems. Manag Sci 46(2):217–232
Chen F (2003) Information sharing and supply chain coordination. Handbooks in operations re-
search and management science, vol 11. Elsevier, Amsterdam, p 341–421
Clark TH, McKenney JL (1994) Campbell Soup Company: a leader in continuous replenishment
innovations. Harvard Business School Case #195124. Harvard Business School, Harvard Uni-
versity, Cambridge, MA
Clark AJ, Scarf H (1960) Optimal policies for a multi-echelon inventory problem. Manag Sci
6(4):475–490
Clemen RT, Winkler RL (1985) Limits for the precision and value of information from dependent
sources. Oper Res 33(2):427–442
ECR Europe (1997) Category management best practices report. ECR Europe
Fiddis C (1997) Manufacturer–retailer relationships in the food and drink industry: strategies and
tactics in the battle for power. FT Management Report. Financial Times Retail & Consumer
Publishing, London
Fry MJ, Kapuscinski R, Olsen TL (2001) Coordinating production and delivery under a (z, Z)-type
vendor-managed inventory contract. Manuf Serv Oper Manag 3(2):151–173
Gal-Or E (1985) Information sharing in oligopoly. Econometrica 53:329–343
Graves SC, Kletter DB, Hetzel WB (1998) A dynamic model for requirements planning with ap-
plication to supply chain optimization. Oper Res 46(S3):35–49
Gruen TW, Corsten DS, Bharadwaj S (2002) Retail out-of-stocks: a worldwide examination of
extent, causes and consumer responses. Grocery Manufacturers of America, Washington, DC
Hammond J (1994) Barilla SpA (A) and (B). Harvard Business School Case #694046. Harvard
Business School, Harvard University, Cambridge, MA
Kahn JA (1987) Inventories and the volatility of production. Am Econ Rev 77(4):667–679
KJR Consulting (2002) CPFR baseline study: manufacturer profile. Grocery Manufacturers of
America, Washington, DC
Kurtuluş M, Nakkas A (2011) Retail assortment planning under category captainship. Manuf Serv
Oper Manag 13(1):124–142
Kurtuluş M, Toktay LB (2011) Category captainship vs. retailer category management under lim-
ited retail shelf space. Prod Oper Manag 20(1):47–56
Kurtuluş M, Ülkü S, Toktay BL (2012) The value of collaborative forecasting in supply chains.
Manuf Serv Oper Manag 14(1):82–98
Lee HL, So KC, Tang CS (2000) The value of information sharing in a two-level supply chain.
Manag Sci 46(5):626–643
Mathews R (1995) Spartan pulls the plug on VMI. Progress Groc 74(11):64–65
Mishra BK, Raghunathan S, Yue X (2007) Information sharing in supply chains: incentives for
information distortion. IIE Trans 39(9):863–877
Mishra BK, Raghunathan S, Yue X (2009) Demand forecast sharing in supply chains. Prod Oper
Manag 18(2):152–166
Özer Ö, Wei W (2006) Strategic commitments for an optimal capacity decision under asymmetric
forecast information. Manag Sci 52(8):1238–1257
Petersen C (2013) Walmart’s secret sauce: how the largest survives and thrives. http://www.
\penalty0retailcustomerexperience.com. Accessed 1 Jun 2015
SCDigest (2008) The on-going battle over CPFR: does it work? Can it scale? The debate continues;
Is the real difference formal versus informal CPFR? http://www.scdigest.com/\penalty0assets/
\penalty0on target/08-03-05-2.php?cid=1529. Accessed 0 Apr 2015
Seifert D (2003) Collaborative planning, forecasting, and replenishment: how to create a supply
chain advantage, 4th edn. AMACOM, New York
Sheffi Y (2002) The value of CPFR. http://web.mit.edu/sheffi/www/documents/genMedia.
\penalty0theValueOfCPFR.pdf. Accessed 2 Jul 2015
Smaros J (2007) Forecasting collaboration in the European grocery sector: observations from a
case study. J Oper Manag 25(3):702–716
3 Collaborative Forecasting in Retail Supply Chains 61

Taylor TA, Xiao W (2009) Incentives for retailer forecasting: rebates vs. returns. Manag Sci
55(10):1654–1669
Taylor TA, Xiao W (2010) Does a manufacturer benefit from selling to a better-forecasting
retailer?. Manag Sci 56(9):1584–1598
Terwiesch C, Ren ZJ, Ho TH, Cohen MA (2005) An empirical analysis of forecast sharing in the
semiconductor equipment supply chain. Manag Sci 51(2):208–220
VICS (1999) The roadmap to CPFR: the case studies. http://www.gs1us.org/industries/
apparel-general-merchandise/workgroups/cpfr. Accessed 9 Apr 2015
VICS (2004) CPFR update. Overview: collaborative planning, forecasting, and replenish-
ment (CPFR). http://www.gs1us.org/industries/apparel-general-merchandise/workgroups/cpfr.
Accessed 9 Apr 2015
VICS (2007) Implementing successful large scale CPFR programs and onboarding trading part-
ners business process guide. http://www.gs1us.org/industries/apparel-general-merchandise/
\penalty0workgroups/cpfr. Accessed 9 Apr 2015
Vives X (1984) Duopoly information equilibrium: Cournot and Bertrand. J Econ Theory 34(1):
71–94
Winkler RL (1981) Combining probability distributions from dependent information sources.
Manag Sci 27(4):479–488
White A, Roster J (2004) VICS’s CPFR model receives a much needed makeover. http://www.
\penalty0gartner.com/. Accessed 2 Jul 2015
 Part II
Contracting and Information
Chapter 4
Reliability or Inventory? An Analysis
of Performance-Based Contracts
for Product Support Services

Sang-Hyun Kim, Morris A. Cohen, and Serguei Netessine

Abstract Traditional sourcing arrangements for after-sales product support have

centered around physical assets. Typically, a customer would pay the supplier of
maintenance services in proportion to the resources used, such as spare parts, that
are needed to maintain the product. In recent years, we have witnessed the emer-
gence of a new service contracting strategy called performance-based contracting
(PBC). Under such a contractual relationship, the basis of supplier compensation
is actual realized uptime of the product. In this study we build a game-theoretic
model and compare the inefficiencies arising under the traditional resource-based
contract (RBC) and PBC. In both cases, the customer sets the contract terms, and
as a response, the supplier sets the base-stock inventory level of spares as well as
invests in increasing product reliability. We find that PBC provides stronger incen-
tives for the supplier to invest in reliability improvement, which in turn leads to
savings in acquiring and holding spare product assets. Moreover, the efficiency of
PBC improves if the supplier owns a larger portion of the spare assets. Our analysis
advocates the view that the full benefit of a PBC strategy is achieved when suppli-
ers are transformed into total service providers who take the ownership of physical
assets.

S.-H. Kim ()

Yale School of Management, Yale University, New Haven, CT 06511, USA
e-mail: [email protected]
M.A. Cohen
The Wharton School, University of Pennsylvania, Philadelphia, PA 19104, USA
e-mail: [email protected]
S. Netessine
INSEAD, 1 Ayer Rajah Avenue, Singapore 138676, Singapore
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 65

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 4
66 S.-H. Kim et al.

4.1 Introduction

The importance of after-sales product support business cannot be understated. In the

defense industry, for example, only about 28 % of a weapon system’s total owner-
ship cost is attributed to development and procurement, whereas the costs to operate,
maintain, and dispose of the system account for the remaining 72 % (Government
Accountability Office 2003). Given that the U.S. Department of Defense’s (DoD)
annual spending for product support amounted to nearly $132B in 2008 (Depart-
ment of Defense 2009), it is not surprising that the manufacturers of military air-
craft, engines, and avionic equipment (e.g., Boeing, GE, Rockwell Collins, Lock-
heed Martin, Pratt & Whitney, and Rolls-Royce) consider the provision of service
parts and repair/maintenance services to be an important component of their com-
petitive strategies.
Traditionally, many after-sales contractual relationships for mature products were
governed by resource-based contracts (RBC), such as the time and material (T&M)
contracts, that specified the unit prices of the service parts, labor, and other con-
sumable resources that need to be utilized in order to satisfy a required service
level, such as product availability. However, increasing pressure to reduce spending
among customer organizations as well as dissatisfaction with the level of after-sales
support from key suppliers have led to reevaluation of these arrangements. In recent
years, a novel strategy for aligning interests in the after-sales service supply chains
has emerged: performance-based contracting. Its premise is simple. Instead of pay-
ing suppliers for parts, labor, and other resources used to provide after sales support,
the compensation is based on the actual availability of the product realized by the
customer. The key idea behind such performance-based contracts (PBC) is to align
the incentives of all parties by tying suppliers’ compensation to the same service
value that the customer cares about. Evidence supports the view that PBC improves
product availability: the U.S. Navy’s implementation of Performance-Based Logis-
tics for its fleet of F/A-18 E/F fighter jets, for example, has resulted in an availability
increase from 67 to 85 %, while a similar effort has seen the material availability of
Aegis guided missile cruisers rise from 62 to 94 % (Geary 2006). (In the commer-
cial aviation sector PBC is known as Power by the Hour™, a term copyrighted by
Rolls-Royce. In the technology sector, the elements of PBC are often included in
Service Level Agreements.)
The ultimate goal of PBC—providing incentives to suppliers to attain high prod-
uct availability at a lower cost—can be achieved through a variety of actions. Exam-
ples include service parts deployment across multiple stocking locations, R&D ef-
fort to improve product reliability, investment in capacity for scheduled/unscheduled
maintenance activities, and parts cannibalization. In this study, our focus is on the
trade-off and interaction between two such actions: investment in spare assets and
in product reliability. Industry practitioners in both government agencies and com-
mercial enterprises identify these two actions as key strategic decisions. Given these
considerations, we aim to address the following research questions. How does PBC
4 Reliability or Inventory? 67

differ from RBC in motivating suppliers to improve reliability and to manage the
inventory of spares? What kind of inefficiencies arise under these two contracts?
Does the ownership structure of the spare assets affect the answers to these ques-
tions?
To answer these questions, we develop a stylized economic model that draws
upon two distinct bodies of literature. We employ the classical service parts inven-
tory management model to represent repair and maintenance processes. This model
is further enriched by a novel feature which has not been previously studied in-
depth: endogenous product reliability improvement effort. By introducing this new
decision variable, we demonstrate a new perspective on after-sales support planning.
The relationship between the customer and the supplier is modeled using a sequen-
tial game formulation, in which the customer sets the terms of the contract in order
to minimize her total cost subject to a minimum product availability requirement.
The supplier’s goal is to set the profit-maximizing levels of reliability and spares
inventory given these contract terms. We allow for an arbitrary allocation of spare
inventory ownership between the customer and the supplier, and compare the im-
pacts of employing two types of contracts. Under RBC the supplier is compensated
for the resources used (spare units, labor, and other materials), and under PBC the
compensation is based on product availability. These two contracting approaches
are widely adopted in practice, yet there has not been a rigorous evaluation of their
relative merits.
In our model we assume that the availability target can be achieved by two means:
investment in spares inventory or investment in product reliability. We find that RBC
results in inefficiencies that lead the supplier to invest less in reliability and more in
the inventory of spares than an integrated firm would. Compared to RBC, we demon-
strate that PBC incentivizes the supplier to achieve the product availability target by
investing more in reliability and simultaneously achieving savings in inventory in-
vestment. As a direct consequence, contracting efficiency is higher under PBC than
under RBC. We also find that the allocation of spare asset ownership between the
customer and the supplier affects efficiency of the two contracts in an opposite way.
Namely, under PBC, the supplier invests more in reliability and less in inventory as
his share of asset ownership increases, whereas under RBC, the opposite occurs.
One of the main conclusions of our analysis is that the maximum benefit of PBC
is realized when spare assets are fully owned by the supplier. While this conclu-
sion provides clear policy guidance, implementing this idea is not trivial. Indeed,
contrary to what our model results advocate (i.e., transfer asset ownership to sup-
plier), it is a common practice for customer organizations to own spare assets even
though key decisions such as spares inventory management are often delegated to
suppliers. Such an ownership structure is understandable given that customer orga-
nizations are reluctant to cede control of their assets, but our findings indicate that
this practice may actually be an impediment to achieving the full benefits of PBC.
Thus, we argue that a significant efficiency improvement can be attained by trans-
forming traditional after-sales product support suppliers into total service providers
who assume control of the majority of service functions, including asset ownership.
68 S.-H. Kim et al.

4.2 Related Literature

Our study presents a game-theoretic model applied to a service parts inventory man-
agement problem. Sherbrooke (1968) introduced the METRIC model for service
parts (repairables) which led to numerous multi-echelon, multi-indenture inventory
model extensions. In METRIC, the repair process for each part is represented by an
M/G/∞ queueing system, and the decision is to optimize the number of spares in
stock given an exogenous part failure rate. Over the years the METRIC and related
models have become the basis for a number of decision support systems (see, for
example, Cohen et al. 1989, 1990, 2006). In comparison to the large volume of liter-
ature in this field that focus on optimization and other implementation aspects, only
recently have the issues of contracting and outsourcing started to receive attention.
We adopt a simplified version of the classic service parts inventory management
models, combining it with a game-theoretic analysis.
One of the distinguishing features of our model is endogenizing the product fail-
ure rate, which has been traditionally assumed to be exogenous in the service parts
inventory management literature. Endogenizing the failure rate as a firm decision
allows us to model the interaction between reliability improvement and inventory
level decisions made by the supplier, the main focus of our study. We note that sev-
eral studies have examined joint optimization of product reliability and service parts
inventory levels for repairables. Öner et al. (2010) introduced a model that includes
the provision for an emergency shipment from a supplier for the case when there is a
stock-out when the supplier operates under a performance-based contract. They de-
velop analytical results and develop an optimization algorithm for solving the joint
optimization problem. In a more recent paper, Öner et al. (2013) develop a model
that considers the option to introduce redundancy in product design and evaluates
the tradeoff between doing so and inventory-related costs. This paper is related to
ours since redundancy can be thought of as a mechanism for improving product re-
liability. Finally we note that Kumar et al. (2007) developed a goal programming
model that jointly optimizes reliability and inventory stocking levels where goals
include minimization of cost of ownership, maximization of spares availability and
minimizing the mean time to repair.
While all of these papers develop models and algorithms that consider the joint
optimization of inventory stocking levels with decisions that relate to product relia-
bility, none develop managerial insights concerning the impact of contract incentives
and asset ownership on the jointly optimal stocking and reliability level solution as
we do in this paper. The papers that come closest to ours in this spirit are Kim (2010),
who also considers reliability decision but focuses on interactions among suppliers,
and Bakshi et al. (2014), who study how RBC or PBC can be used as a mechanism
for signaling product reliability. Guajardo et al. (2012) present an empirical study of
how different contracting mechanisms influence product reliability using a dataset
from a major commercial aircraft engine manufacturer. Their conclusions support
one of our findings, namely, PBC promotes reliability improvement more than RBC
4 Reliability or Inventory? 69

does. Chan et al. (2014) is another empirical study that examines the issue similar
to that Guajardo et al. (2012), but they focus on maintenance service contracts for
medical equipment.
To represent the contractual relationship between the customer and the supplier,
we employ modeling approaches commonly found in the existing supply chain con-
tracting papers; see Cachon (2003) for an overview. The contracts we analyze fall
under the class of contracts found in this stream of literature, but our model is dis-
tinctive in that we focus on the current practices found in the context of the after-
sales support businesses, especially those commonly observed from the aerospace
and defense industry. Our model is closely related to the multitasking literature (e.g.,
Holmström and Milgrom 1991; Gibbons 2005), in which the agent controls more
than one action (reliability and inventory in our case).
A number of papers discuss incentives and contracting in the defense industry,
one of the industrial settings that motivated this research. Early papers include Cum-
mins (1977) and Rogerson (1994). More recently, Kang et al. (2010) propose a
decision-support model that can help support PBC relationships by trading off re-
liability and maintenance tasks. While the last paper investigates a similar problem
context as ours, it does not present an economic analysis where incentives play a
central role. Mirzahosseinian and Piplani (2011) examine other issues related to
PBC via numerical analysis.
The work that is most related to our study is Kim et al. (2007), who consider
how cost reduction and performance incentives interact under a general contracting
arrangement that includes PBC when significant cost uncertainty is present, while
ignoring asset ownership issues. The theme of our research is quite different since
we focus mainly on reliability improvement and its interaction with inventory man-
agement decisions under varying asset ownership structures. Another related paper
is Kim et al. (2010), who specifically study the contracting challenge arising from
the infrequent nature of product failures, as they provide severely limited informa-
tion about the supplier’s effort to maintain equipment. Jain et al. (2013) study a
similar setting but that involves double moral hazard. While these papers do not
provide a comprehensive picture of the complex dynamics created by PBC in after-
sales support environments, they do include complementary analyses of different
aspects of the problem and thus provide insights that are relevant to practitioners.
In summary, the analytical contributions of this research are twofold. First, we
endogenize reliability improvement decisions in a classical repairable inventory
management model and highlight how reliability interacts with inventory. Second,
we study and compare two frequently used contractual arrangements (RBC and
PBC), evaluate their inefficiencies, and identify the factors that cause them. From
a managerial perspective, our study sheds light on how performance-based incen-
tives can lead to reliability improvement and how it affects the ideal asset ownership
structure to achieve an efficient solution.
70 S.-H. Kim et al.

4.3 Model

A risk-neutral customer owns and operates a fleet of N identical products, whose

continued usage is disrupted by random product failures. A failed unit is immedi-
ately sent to the supplier for a repair, while a working spare unit is pulled from
the inventory, if one exists. The supplier performs three kinds of activities to sup-
port the customer’s fleet of products: (1) repair defective units, (2) manage spare
product inventory, and (3) manage product reliability. The duration of the contract-
ing relationship between the customer and the supplier is normalized to one. The
failures occur at a rate λ ≡ E[Λ ], where Λ is the total number of product failures
within the contracting horizon. It takes  j amount of time to repair the jth failure.
The expected repair lead time l ≡ E [ j ], or equivalently, the repair rate 1/l, is as-
sumed to be fixed and not impacted by the supplier’s effort (e.g., repairs are always
performed at the maximum speed). In contrast, we assume that the Mean Time Be-
tween Failures (MTBF) 1/λ , a measure of product reliability, can be increased by
the supplier’s effort. In our analysis we represent the reliability improvement effort
by the normalized MTBF τ ≡ (λ l)−1 , and henceforth refer to it simply as reliability.
The range in which τ can vary is assumed to be between τ and τ . The lower limit τ
represents the existing level of reliability, whereas τ is the theoretical upper limit of
reliability that can be achieved. For simplicity, we only consider a single indenture
level for the product, i.e., spares inventory is managed at the product level, not at
the component level.
The customer moves first as a Stackelberg leader by offering a contract, either
RBC or PBC, that influences the supplier’s simultaneous decision on product re-
liability τ and the stocking level s of spare products. Because spare products are
repairable items, i.e., they are repaired upon failure and returned to the system in-
stead of being scrapped, the quantity s (the number of spares that are produced
initially) remains constant after its value is chosen. Therefore, at any given moment,
there are N + s products in the system. Without loss of generality, we assume that
N products have been already produced and paid for and that s = 0 at the outset
of the game. By assuming that s is the supplier’s choice, we only consider the case
of Vendor Managed Inventory (VMI), which is the prevailing practice in after-sales
product support environments. We assume that the ownership of spare assets is split
between the customer and the supplier by introducing the parameter δ ∈ (0, 1] that
represents the fraction of spares owned by the supplier. Therefore, (1 − δ ) s and δ s
are the spares quantities belonging to the customer and the supplier, respectively.
Holding costs are incurred by the two parties in proportion to the number of spare
units each owns. Throughout the analysis we treat δ as an exogenous parameter, in
order to reflect on the spectrum of ownership structures observed in practice and to
highlight the consequences of varying the ownership allocation.
4 Reliability or Inventory? 71

4.3.1 Repair Process and Performance Measurement

To model the repair process, we adopt the standard assumptions in the classical
service parts inventory management literature (see, for example, Sherbrooke 2004;
Muckstadt 2005). The repair facility is modeled as an M/G/∞ queue. Product fail-
ures occur according to a Poisson process, and a failed product is replaced by a
working unit from the spares inventory, if one is available. Otherwise, a backorder
is logged. A one-for-one base stock inventory policy is used for replacement of de-
fective units, with each failed product immediately undergoing a repair that takes
a random amount of time. Note that the Poisson failure process is not an exact
representation in a closed-loop repair cycle, but it is a good approximation in an
environment where products fail relatively infrequently.
The Poisson failure assumption implies that the steady-state inventory on-order
O(τ ), the number of units that are being repaired at a random point in time, is
Poisson-distributed with mean λ l = 1/τ . Two important random variables are on-
hand inventory I and backorder B, which are related to O(τ ) and s by I | τ , s =
(s − O(τ ))+ and B | τ , s = (O(τ ) − s)+ , where (·)+ ≡ max {0, ·}. There is a one-
to-one correspondence between the performance measure of our interest, the ex-
pected product availability E [A | τ , s], and the expected backorder: E [A | τ , s] = 1 −
E[B | τ , s]/N. Consistent with common practices, we assume that the customer faces
an explicit service requirement E [A | τ , s] ≥ α (e.g., expected availability should be
95 % or more), which can be translated into the backorder constraint E[B | τ , s] ≤ β .
As it turns out, the discrete nature of the Poisson distribution in O(τ ) limits
analytical tractability of our game-theoretic model setup. To circumvent this dif-
ficulty, we restrict the range of τ to 1/N  τ < τ  0.1 in order to apply a con-
tinuous approximation of the Poisson distribution. This assumption is valid if the
fleet size N is sufficiently large and τ is sufficiently small. In this range of τ , we can
treat the distribution of O(τ ) as normal (with E[O(τ )] =Var[O(τ )] = 1/τ ), which
yields a good approximation for the exact values of E[B | τ , s] and E[I | τ , s]. To
that end, let φ and Φ be the pdf and the cdf of the standard normal distribution.
Define Φ (·) ≡ 1 − Φ (·). In addition, let f (·) ≡ φ (x)/Φ (x) be the hazard function
and L(x) ≡ φ (x) − xΦ (x) be the loss function. √ The normal
√ √τ
z-statistic for a given
and s is z ≡ (s − E[O(τ )]) / Var[O(τ )] = τ s − 1/ τ . Hence, s = 1/τ + z/ τ .
The expected backorder
√ and the expected inventory
√ on-hand are, respectively,
E[B | τ , s] = L(z)/ τ and E[I | τ , s] = (z + L(z)) / τ . Note that the expression for
E[I | τ , s] contains the negative domain of s, but its effect is negligible in the range
of τ we consider.

4.3.2 Cost Structure

We consider the following costs that are significant in the product support setting:
(1) K(τ ), cost of improving reliability τ ; (2) κ , cost of repairing a defective product
72 S.-H. Kim et al.

per unit time; (3) c, cost of producing a spare unit; (4) hg , cost of carrying a func-
tional product per unit time; (5)‘hb , cost of carrying a defective product per unit
time. All cost parameters are assumed to be public knowledge.
K(τ ) represents the dollar amount of investment the supplier incurs in or-
der to improve reliability to τ . We assume that K(τ ) has the following intu-
itive properties: K  (τ ) > 0, K  (τ ) > 0, K  (τ ) > 0, K(τ ) = 0, and limτ →τ K(τ ) =
limτ →τ K  (τ ) = ∞. The cost κ is incurred by the supplier per unit time while a unit is
undergoing a repair. Since the total expected duration of repairs over the contracting
horizon is λ l = 1/τ , the expected repair cost is κ /τ .
We assign two different values for the holding cost, hg and hb , each correspond-
ing to the state that a product is in: at any given time, a product is either func-
tional (“good” unit) or defective (“bad” unit). Good units include those deployed
in the fleet and the spares stored in the inventory, while the bad units are those
undergoing repairs in the repair facility. Since the value of a good unit is higher
than that of a bad unit, we assume hb < hg < c. The holding costs incurred by the
customer and the supplier are proportional to the number of products each owns,
and thus they depend on the parameter δ that represents the supplier’s portion of
the total spare assets. We adopt the convention that the products are indistinguish-
able as long as they are in the same state (good or bad). Under this assumption,
the number of good units that the customer owns at any given moment is equal to
N − (O(τ ) − s)+ + (1 − δ ) (s − O(τ ))+ . Similarly, the number of bad units that the
customer owns is equal to (O(τ ) − s)+ + (1 − δ ) min{O(τ ), s}. The expected to-
tal holding costs for the customer and the supplier are then, respectively, H(τ , s) ≡
hg (N − E[B | τ , s] + (1 − δ )E[I | τ , s])+hb (E[B | τ , s] + (1 − δ )E[min{O(τ ), s}]) and
η (τ , s) ≡ δ hg E[I | τ , s] + δ hb E[min{O(τ ), s}].
Adding all cost components described thus far, the total expected internal costs
of the customer and the supplier are, respectively, Ψ (τ , s) ≡ H(τ , s) and ψ (τ , s) ≡
K(τ ) + κ /τ + cs + η (τ , s). Note that the supplier’s production cost is cs, not δ cs, be-
cause the stocking level s is the supplier’s decision and hence he has to bear the full
cost of production. In the remainder of our analysis, we make the following tech-
nical assumptions regarding the cost parameters which ensure that the problem is
well-defined and allow us to focus on the most interesting and managerially relevant
cases: (1) τ < 1/β ; (2) τ 2 K  (τ ) < κ − (hg − hb ); (3) κ + c + hb < (1/β )2 K  (1/β );
(4) 2 (hg − hb ) < τ 3 K  (τ ).

4.3.3 Contracts

At the beginning of the contractual relationship the customer offers to the supplier a
contract that defines the payment to the supplier, denoted by T . Anticipating the sup-
plier’s optimal response τ ∗ and s∗ , the customer determines compensation terms that
would minimize her total cost E[T | τ ∗ , s∗ ]+Ψ (τ ∗ , s∗ ) subject to the availability con-
straint while ensuring the supplier participation in the trade. In response, the supplier
sets τ ∗ and s∗ that maximize his expected profit E[T | τ , s] − ψ (τ , s). As the names
4 Reliability or Inventory? 73

suggest, the customer pays the supplier based on the amount of resources used for
repair and maintenance activities under RBC, whereas it is the performance outcome
(availability) that is the basis of compensation under PBC.
Following the common assumption found in the supply chain contracting litera-
ture, we assume that the customer cannot contract directly on the desired levels of
τ and s, which are the supplier’s decision variables. Consequently, indirect contrac-
tual levers are used to induce the supplier’s decisions that agree with the customer’s
expectations. Under RBC, this lever consists of the payment for resource usage; un-
der PBC, it is the payment for performance realization. The basis of payment under
RBC is the number of spares s and the cumulative repair time ∑Λj=1  j , whereas un-

der PBC it is the cumulative backorder 01 B(t)dt, which maps to product downtimes
(B(t) denotes a backorder logged at time t). Note that, under RBC, the supplier re-
veals his choice of s later when he bills the customer for compensation. In this study
we focus on linear functions for the payments: T = w + ps + r ∑Λj=1  j under RBC

and T = w − v 01 B(t)dt under PBC. The contract terms are interpreted as follows:
w ≥ 0 is a lump-sum payment, p ≥ 0 is the unit price for the spares produced, r ≥ 0
is the compensation rate per unit time for repairing each defective unit, and v ≥ 0
is the penalty rate for realized backorders. In expectation, E[T | τ , s] = w + ps + r/τ
for RBC and E[T | τ , s] = w − vE[B | τ , s] for PBC.
Note that, under RBC, the supplier is paid p dollars for each of s units of spares
that he produces, not just for the (1 − δ ) s units that are transferred to the customer;
the supplier is also compensated for the rest (δ s of them), which are destined to
the supplier’s reserve inventory. Hence, p represents the reservation price that the
customer uses as a lever to incentivize the supplier to secure a desired number of
spares for the entire supply chain. (An alternative modeling approach is to assume
that the customer pays the amount (1 − δ ) ps only for the spare units that she takes
the ownership of. It can be demonstrated that this alternative assumption has little
impact on the results.) Note that the time and materials (T&M) contract can be
viewed as a special case of RBC with w = 0, p > c, and r > κ . In addition, both RBC
and PBC reduce to the traditional fixed price contract when w > 0 and p = r = v = 0.
In our analysis we mention these special cases whenever it is appropriate.
To summarize, the customer’s problem can be written as

(C ) min {E[T | τ ∗ , s∗ ] + Ψ (τ ∗ , s∗ )}
P
s.t. E[B | τ ∗ , s∗ ] ≤ β ,
E[T | τ ∗ , s∗ ] − ψ (τ ∗ , s∗ ) ≥ u,
(τ ∗ , s∗ ) ∈ arg max {E[T | τ , s] − ψ (τ , s)} ,

where T , Ψ , and ψ are defined above and P = {w, p, r} for RBC and P = {w, v} for
PBC. The first constraint in (C ) represents the customer’s availability requirement,
expressed in terms of the upper limit on the expected backorder (thanks to one-to-
one mapping between the two performance measures). The last two constraints, the
individual rationality (IR) and the incentive compatibility (IC) constraints, describe
that the supplier’s participation is ensured and that he decides (τ ∗ , s∗ ) to maximize
74 S.-H. Kim et al.

his expected profit given the contract parameters p, r, and v. It is important to rec-
ognize that the supplier is not subject to the same backorder constraint that the
customer faces, and as a result, the customer has to use contract terms as a lever to
influence the supplier’s decisions in order to satisfy the constraint. The supplier’s
constant reservation utility level u represents the profit that he can generate in an
outside opportunity, and without loss of generality, we assume that its value is suf-
ficiently high so that the lump-sum payment w is always nonnegative. (C ) can be
simplified after recognizing that the (IR) constraint is always binding at the optimum
by adjusting w accordingly. Then the problem is reduced to

(C) min C(τ ∗ , s∗ ), C(τ ∗ , s∗ ) ≡ u + Ψ (τ ∗ , s∗ ) + ψ (τ ∗ , s∗ )

P\{w}
s.t. E[B | τ ∗ , s∗ ] ≤ β ,
(τ ∗ , s∗ ) ∈ arg max u(τ , s),

where we used the notation u(τ , s) ≡ E[T | τ , s] − ψ (τ , s) for the supplier’s expected
profit. Therefore, the problem becomes that of minimizing the total expected supply
chain cost under the constraints that the backorder target should be met and the
reliability and inventory levels are optimally set by the supplier.

4.4 Analysis

4.4.1 The First-Best Benchmark

To establish the benchmark, we first analyze the case in which the customer and the
supplier are assumed to be integrated as a single firm minimizing its total cost sub-
ject to the availability requirement (first-best or FB), i.e., the integrated firm solves
the problem minτ ,s u+Ψ (τ , s)+ ψ (τ , s) subject to E[B | τ , s] ≤ β . (u is included here
to permit a fair comparison with the results of decentralized cases from Sects. 4.4.2
and 4.4.3.) The following two quantities appear frequently in the analysis below and
are defined here for convenience:
 √  κ + c + hb c + hg
ζ (τ ) ≡ L−1 β τ and Γ (τ ) ≡ + 3/2 f (ζ (τ )) .
τ2 2τ
Note that both are decreasing functions. The first-best optimal values of τ and s are
characterized as follows.

Proposition 1. The backorder constraintbinds at

√the optimum. The integrated firm
chooses τ FB > τ and sFB = 1/τ FB + ζ τ FB / τ FB > 0, where τ FB is uniquely
determined from the equation Γ (τ ) = K  (τ ).

The integrated firm finds it optimal to invest in both reliability and spares
inventory in order to satisfy the specified backorder target β . The quantity Γ (τ )
4 Reliability or Inventory? 75

represents the firm’s marginal benefit of improved reliability; hence, the first-order
condition Γ (τ ) = K  (τ ) appearing in Proposition 1 points to the optimal level at
which the marginal benefit is equal to the marginal cost of improving reliability. At
this level τ FB , the optimal stocking quantity sFB is determined from the backorder
constraint, which is shown to bind at the optimum.

4.4.2 Resource-Based Contract

Under RBC, the supplier is compensated in proportion to the resources used for
repair and maintenance services provided to the customer. The supplier chooses the
optimal reliability and spares stocking levels τ and s in response to the contract
terms p and r, which are the unit prices for spares quantity and repair times. As
noted in Sect. 4.3.3, the expected payment under RBC is E[T | τ , s] = w + ps + r/τ .
We now study how the supplier optimally responds to the proposed contract terms
p and r.

Lemma 1. Suppose c + δ hb < p < c + δ hg , the range of p in which a finite feasible

solution of (C) exists under RBC. Define a ≡ κ − τ 2 K  (τ ) − δ (hg − hb ) > 0 and
b ≡ κ − τ 2 K  (τ ) > a and let τ be the unique solution of

κ + c + δ hb − r − p δ (hg − hb )
+ φ (z∗ ) = K  (τ ), (4.1)
τ2 2τ 3/2
where  
c + δ hg − p
z∗ = Φ −1 1 − .
δ (hg − hb )
√
The supplier chooses τ ∗ and s∗ = 1/τ ∗ + z∗ / τ ∗ as follows: (i) if 0 ≤ r ≤ a
then τ ∗ = τ > τ for all p ∈ (c + δ hb , c + δ hg ); (ii) if a < r < b then there exists
p(r) ∈ (c + δ hb , c + δ hg ) such that τ ∗ = τ > τ if p ∈ (c + δ hb , p(r)) and τ ∗ = τ if
p ∈ [p(r), c + δ hg ); (iii) if r ≥ b then τ ∗ = τ for all p ∈ (c + δ hb , c + δ hg ). Further-
more, ∂ τ ∗ /∂ p ≤ 0, ∂ s∗ /∂ p > 0, ∂ τ ∗ /∂ r ≤ 0, where the equalities hold if and only
if τ ∗ = τ , and ∂ s∗ /∂ r = 0.

Several interesting observations are made from the lemma. First, it is clear that,
in order to motivate the supplier to invest in reliability improvement, the customer
should not offer large prices for either p or r or both. Second, the T&M contract,
which we defined in Sect. 4.3.3 as a RBC with positive margins p − c > 0 and
r − κ > 0 for both spares and time-based resources, never incentivizes the sup-
plier to improve reliability, as is evident from the condition stated in part (iii). These
results are direct consequences of the unique incentive structure inherent in the after-
sales support contracting environment. Namely, for the supplier who provides repair
and maintenance services and gets compensated for the invested resources to sup-
port them, his business grows if the products are less reliable: the more frequently
the products fail, the higher the supplier’s revenue under RBC. While this may
76 S.-H. Kim et al.

benefit the supplier, it has a negative impact on product availability and consequently
on the customer’s ability to generate value through product use. Therefore, increas-
ing the compensation rate for resource usage only exacerbates this skewed incentive
as the supplier earns higher margins for a higher rate of product failures. This insight
is summarized by the comparative statics results shown in the lemma. Among them,
the conflicting dual roles of p is particularly noteworthy: increasing p induces higher
stocking quantity but lower reliability. These results indicate that, in order to pro-
duce high product availability (achieved by high levels of reliability and inventory),
the customer has to offer a small value for r and a sufficiently large value for p.
Armed with the insights into the supplier’s optimal response, now we turn to
the customer’s contract design problem and the optimal solution that emerges in
equilibrium. From (C), we see that the customer’s problem under RBC is equivalent
to choosing the optimal values for the contract terms p and r so as to minimize the
total expected supply chain cost C(τ ∗ , s∗ ) = u + Ψ (τ ∗ , s∗ ) + ψ (τ ∗ , s∗ ) subject to
the backorder constraint E[B | τ ∗ , s∗ ] ≤ β , where τ ∗ and s∗ are determined by the
supplier as in Lemma 1. The equilibrium solution, denoted by the superscript R, is
specified as follows.
Proposition 2. In equilibrium the backorder constraint binds. The customer offers
r = 0 and p = c + δ hb + δ (hg − hb ) Φ (ζ (τ R )), where τ R > τ is the equilibrium
reliability chosen by the supplier that is uniquely determined from the equation

κ − δ (hg − hb ) Φ (ζ (τ )) δ (hg − hb )
+ φ (ζ (τ )) = K  (τ ). (4.2)
τ2 2τ 3/2
We find that the customer who employs RBC should not pay the supplier based
on time-based resource usage: r should be set to zero. Therefore, although we be-
gan with a general contract form that includes two contract parameters p ≥ 0 and
r ≥ 0, only p turns out to be a useful lever that enables satisfaction of the backorder
constraint; r is ineffective in incentivizing reliability improvement. Given that many
existing support contracts found in practice include time-based compensations, this
result is quite striking. Proposition 2 suggests that such a practice impedes the sup-
plier’s motivation to improve product reliability and therefore should be suppressed
when reliability is a concern. The reservation price p, on the other hand, is an im-
portant (and the only) instrument under RBC that makes it possible to achieve high
availability through investment in spares inventory, although it does not promote
reliability improvement, either.

4.4.3 Performance-Based Contract

Next, we analyze PBC. The supplier’s optimal response to the contract terms is as
follows.
Lemma 2. Suppose v > c + δ hb , the range of p in which a finite feasible solution
of (C) exists under PBC. The supplier chooses τ ∗ > τ which is a unique solution of
the equation
4 Reliability or Inventory? 77

κ + c + δ hb v + δ (hg − hb )
+ φ (z∗ ) = K  (τ ), (4.3)
τ2 2τ 3/2
where  
c + δ hg
z∗ = Φ −1 1 − ,
v + δ (hg − hb )
√
setting s∗ = 1/τ ∗ + z∗ / τ ∗ > 0. Furthermore, ∂ τ ∗ /∂ v > 0 and ∂ s∗ /∂ v > 0.

Notice that, unlike what we found in Lemma 1 for RBC, there is no upper bound
on the contract term v that ensures feasibility of the solution. This provides the first
hint at the qualitative difference between RBC and PBC. We see a stark contrast
between RBC and PBC from the comparative statics results. Recall from Lemma 1
that, under RBC, increasing the reservation price p induces the supplier to choose
a higher spares stocking level s but lower reliability τ . In contrast, the backorder
penalty v under PBC induces the supplier to increase both τ and s. Thus, Lemmas 1
and 2 highlight the key difference between RBC and PBC. Namely, the two compa-
rable terms in these contracts, p and v, induce opposite reactions from the supplier
with respect to the reliability improvement decision.
This difference arises from the relationship between availability—the measure
of performance that the customer wants to enhance—and the two intermediate out-
comes that each contract term is designed to influence, namely the spares inventory
(influenced by p) and the backorders (by v). Availability can be increased in differ-
ent ways: higher reliability, more spares, or a combination of both. Under RBC only
one component of this mix (i.e., spares inventory) receives the supplier’s attention,
whereas under PBC both do, as the backorder penalty is reduced by higher levels of
both reliability and spares inventory. RBC does contain an additional contract term r
that influences the supplier’s reliability decision, but it does not compensate for the
shortcoming of p since increasing it goes counter to the direction that the customer
desires: reliability is reduced with higher r.
We infer from these observations that PBC is superior to RBC in incentivizing
the supplier to improve product reliability. However, since availability is a function
of both reliability and inventory, it is still unclear if PBC leads to lower cost than
RBC does. We answer this question in the next subsection. As a prerequisite, we
derive the equilibrium outcome under PBC. The solution approach is similar to that
of Proposition 2. That is, we solve the optimization problem (C) by incorporating
the supplier’s optimal responses τ ∗ and s∗ as specified in Lemma 2.

Proposition 3. In equilibrium the backorder constraint binds. The customer offers

v = (c + δ hg )/Φ (ζ (τ P )) − δ (hg − hb ), where τ P is the equilibrium reliability cho-
sen by the supplier that is uniquely determined from the equation
 
hb hg
Γ (τ ) − (1 − δ ) 2 + 3/2 f (ζ (τ )) = K  (τ ). (4.4)
τ 2τ
78 S.-H. Kim et al.

4.4.4 Comparisons of the Contracts

Having analyzed the structures of optimal contracts and the equilibrium outcomes
under RBC and PBC, we are now in a position to compare relative performances of
each contracting approach and evaluate how they fare against the first-best bench-
mark. This is summarized in the following proposition. Here, we use the notations
CR , CP , and CFB to represent the customer’s (equivalently, the supply chain’s) ex-
pected cost in equilibrium for RBC, PBC, and the first-best cases.

Proposition 4. The following results hold in equilibrium: (i) τ R < τ P ≤ τ FB , sR >

sP ≥ sFB , and CR > CP ≥ CFB ; (ii) ∂ τ R /∂ δ < 0, ∂ sR /∂ δ > 0, ∂ CR /∂ δ > 0,
∂ τ P /∂ δ > 0, ∂ sP /∂ δ < 0, and ∂ CP /∂ δ < 0; (iii) τ P → τ FB , sP → sFB , and
CP → CFB as δ → 1, whereas τ R , sR , and CR never approach the first-best levels.

The insights we gained from our discussion of RBC and PBC above point to
PBC’s superiority in promoting reliability improvement, as stated in part (i) of the
proposition. Additionally, the proposition demonstrates that it is not only reliability
for which PBC brings an advantage; compared to RBC, it also lowers inventory.
Therefore, a win-win situation marked by higher reliability and lower inventory can
be attained through PBC. This reflects the fundamental relationship between relia-
bility and inventory: they are substitutes in achieving a given level of availability. In
other words, less frequent product failures lessens the need to maintain a large stock
of spares and other physical resources. See Fig. 4.1 for a schematic illustration of
this relationship. The proposition also reveals that contracting efficiency, measured
 −1  −1
by the inverse cost ratios CR /CFB and CP /CFB , is higher under PBC. This
is quite intuitive given what we have learned. As products fail less frequently, there

Feasible region:
RBC E[B|τ, s] < β
sR

PBC
sP FB
sFB E[B|τ,s] = β

τR τ P τ FB τ
Fig. 4.1 Substitutable relationship between product reliability (τ ) and inventory (s) with respect
to a fixed availability target, expressed in terms of the backorder constraint. The equilibrium levels
of τ and s under each contracting scenario are marked in the diagram. The arrows represent the
direction to which the optimal combination moves as δ increases
4 Reliability or Inventory? 79

is a smaller need for the resources that are used to counter the adverse effects of
failures, resulting in cost savings. Although it is costly to improve reliability, the
contract terms under PBC brings “more bang for buck,” and therefore, contributes
more to savings.
Another important insight from Proposition 4 is that the spares asset owner-
ship structure, represented by the parameter δ , impacts system efficiency differently
across the two contracting cases. In particular, first-best can be achieved under PBC
in the limit δ → 1, i.e., when the supplier owns the entire spare assets (see part (iii)).
Under RBC, by contrast, first-best can never be achieved. This observation makes
it clear that incentives between the customer and the supplier are better aligned un-
der PBC, and moreover, a transfer of asset ownership to the supplier facilitates it.
Perfect incentive alignment is attained with δ = 1 because of the combination of
two factors: (a) a complete ownership transfer forces the supplier to absorb the en-
tire cost existing in the supply chain and (b) PBC effectively converts the stochastic
performance outcome into financial consequences for the supplier. As a result, the
supplier bears the full risks of product downtimes and the associated loss of value,
as the integrated supply chain would.
This argument suggests that PBC is not as effective when the supplier only has a
partial ownership of assets (δ < 1). Indeed this is what we find. According to part (ii)
of Proposition 4, under PBC, lowering δ from one results in lower τ , higher s,
and higher supply chain cost—in other words, all equilibrium numbers move away
from the first-best levels. Interestingly, we find the opposite behavior under RBC:
reliability becomes worse, inventory goes up, and the supply chain cost increases
with a larger value of δ , i.e., as the supplier’s share of asset ownership becomes
larger, not smaller. See Fig. 4.2 for an illustration. Intuitively this happens under
RBC because, as the supplier’s profitability is eroded by a higher ownership cost,
he compensates for the loss by letting the products fail more often and increasing
the revenue originating from resource usage. This observation again highlights the
contrasting incentive structures that are present under RBC and PBC.

0.055 30

28
0.05 FB
RBC
26

0.045 PBC 24
τ s
0.04 22

20
FB
0.035
RBC PBC
18

0.03 16
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

d d
Fig. 4.2 An example showing the changes in the equilibrium levels of τ and s under RBC and PBC
as a function of δ
80 S.-H. Kim et al.

Summarizing, we find that spare asset ownership allocation plays an important

role in effective management of repair and maintenance services outsourcing. If a
situation dictates that RBC has to be implemented, then it is best that the customer
retains the majority share of assets. On the other hand, organizations considering
switching to PBC can maximize the benefit by transferring the ownership of assets
as much as possible, thereby converting the supplier into a total service provider who
not only delivers requested services but also actively manages physical resources
that are needed to support them.

4.5 Conclusion

Using a stylized economic model, we study the trade-off between investing in

reliability improvement and in spare assets under two contracts that are commonly
observed in after-sales support for complex equipment that requires repair and main-
tenance over the course of its product life. Our main focus is on performance-based
contracting, a contracting strategy that has gained wide acceptance especially in the
aerospace and defense industry. Performance-based contracts (PBC) are designed
to replace more conventional resource-based contracts (RBC) in an attempt to bet-
ter align the incentives of customers and suppliers. However, even after years of
PBC implementations, debates persist on effectiveness and practicality of this strat-
egy. Our study sheds light on one of the most important challenges in product sup-
port, namely the trade-off between reliability and spare inventory needed to enhance
product availability.
Our analysis suggests that RBC is not as effective as PBC in incentivizing sup-
pliers to invest in reliability improvement. Under RBC, suppliers tend to meet the
availability target by increasing the inventory of spares at the expense of reliability
investment. Under PBC, by contrast, the supplier achieves the availability target by
improving reliability as well as by increasing the size of the spares inventory. In gen-
eral, both contracts result in inefficiencies manifested in less reliable products and
more inventory than the first-best solution prescribes. Compared to RBC, however,
PBC enables a potential win-win scenario under which products are more reliable
and a lower inventory investment is needed.
Moreover, we find that successful implementation RBC and PBC depends cru-
cially on the asset ownership structure. Our analysis shows that the optimal own-
ership structures under these two contracting approaches are opposite: under RBC,
it is best if the customer retains the majority of spare assets, whereas under PBC,
a full transfer of ownership to the supplier is recommended, if it is viable. Under
PBC, incentives between the two parties are better aligned with ownership trans-
fer if the supplier fully internalizes the cost of maintaining the physical resources
that are used to support after-sales services, leading to the maximum levels of prod-
uct reliability and savings in material use. Therefore, our analysis advocates giving
suppliers full ownership responsibility and thereby transforming them into total ser-
vice providers. When this is done, PBC will coordinate the supply chain. Practical
4 Reliability or Inventory? 81

implementation of our policy recommendation will not be straightforward, however,

since many customer organizations believe that asset ownership protects them from
mismanagement by the supplier and endows them with more control over product
availability. Despite such difficulties, we see evidence that customer organizations
are moving towards increased levels of asset ownership transfer to their suppliers.
For example, in a case we are familiar with, a foreign military service has negotiated
with one of its U.S. aircraft suppliers to transfer the title of its spares assets.
Finally, our analysis generates a number of hypotheses that naturally lend them-
selves to empirical examination. We predict that PBC will result in greater product
reliability, lower inventory, and lower contracting cost than RBC does. Further, we
predict that PBC with a larger share of inventory owned by the supplier will result in
greater efficiency. In a separate study that analyzes data provided by a commercial
aircraft engine manufacturer (Guajardo et al. 2012), it has been empirically veri-
fied that PBC significantly increases the mean time between unscheduled repairs, a
proxy for reliability. More empirical findings will shed further light on the insights
obtained from our theoretical work.

Appendix 1: Mathematical Preliminaries

In the proofs, we use the following conventions related to Normal approximation.

To circumvent the conceptual difficulty of having negative s, we will regard 0 as
the lower bound on s and, as √ a consequence, we define the lower bound on z as
z ≡ (0 − 1/τ )/ 1/τ = −1/ τ . This definition does not √ cause a problem because
all quantities of our interest on the domain (−∞, −1/ τ ) are insignificant
√ in the
range 1/N  τ < τ  0.1 we consider (see Sect. 4.3.1). Thus, τφ (z)  0 and
Φ (z)  0. These approximations at the lower√ bound τ require us√to use the
√ follow-
ing conventions: (1) L(z) = φ (z) − zΦ (z) = τφ (z) + Φ (z) / τ  1/ τ = −z;
√
(2) Φ −1 (0)  z = −1/ τ . Throughout the analysis we replace the notation  with
an equality, with an understanding that some of them represent approximations. Un-
der these conventions, the following relationships hold.

Lemma 3. (i) ∂ E[B | τ , s]/∂ s = −Φ (z) < 0, (ii) ∂ E[B | τ , s]/∂ τ = −φ (z)/2τ 3/2 −
Φ (z)/τ 2 < 0, (iii) ∂ E[I | τ , s]/∂ s = Φ (z) > 0, (iv) ∂ E[I | τ , s]/∂ τ = −φ (z)/(2τ 3/2 )+
Φ (z)/τ 2 > 0.
In the next lemma we state the property of a probability distribution exhibiting
increasing failure rate (IFR) that becomes useful in the proofs.

Lemma 4. Let X be a random variable with an IFR property whose pdf g is differ-
entiable and vanishes at both extremes of its support [y, y], where y = −∞ and y = ∞
are permitted. Let G be the cdf of X and G(·) ≡ 1 − G(·). Then

g(y)E[(X − y)+ ]
ω (y) ≡ ≤ 1.
G(y)2
82 S.-H. Kim et al.

Proof. IFR means

 
d g(y) g (y)G(y) + [g(y)]2
= ≥ 0,
dy G(y) G(y)2

which in turn implies −g (y)/g(y) ≤ g(y)/G(y). It can easily be shown that
ω  (y) ≥ 0. To derive the upper bound, we only need to show m ≡ limy→y ω (y) ≤ 1.
Applying l’Hopital’s rule, we obtain:

g(y)E[(X − y)+ ] g (y)E[(X − y)+ ] − g(y)G(y)

m = lim = lim
y→y G(y)2 y→y −2g(y)G(y)
1 1  +
g (y)E[(X − y) ] 1 1 g(y)E[(X − y)+ ] 1 m
= − lim ≤ + lim = + ,
2 2 y→y g(y)G(y) 2 2 y→y G(y)2 2 2

where the inequality follows from the earlier result −g (y)/g(y) ≤ g(y)/G(y). Ar-
ranging both sides yields m ≤ 1.

Appendix 2: Proofs of the Main Results

Proof (Proof of Lemma 1). The supplier’s expected profit under RBC is u(τ , s) =
E[T | τ , s]− ψ (τ , s) = w− K(τ )+ (r − κ )/τ + (p − c − δ hb ) s− δ (hg − hb ) E[I | τ , s].
√ 3) we have ∂ u/∂ s = p−c− δ hb −
Differentiating this with respect to s, (see Lemma
δ (hg − hb ) Φ (z) and ∂ 2 u/∂ s2 = −δ (hg − hb ) τφ (z) < 0, i.e., u(τ , s) is concave in
s for a fixed τ . It can be shown that a finite feasible solution of (C) exists only if
c + δ hb < p < c + δ hg (proof is omitted). With δ > 0 and c + δ hb < p < c + δ hg , the
profit-maximizing s is found in the interior. The first-order condition ∂∂ us = 0 yields
z∗ = Φ −1 [1 − (c√+ δ hg − p)/(δ (hg − hb ))], which is independent
√ of τ . Substituting
s∗ = 1/τ + z∗ / τ and noting E[I | τ , s] = (φ (z) + zΦ (z))/ τ , u(τ , s) becomes

r + p − κ − c − δ hb δ (hg − hb ) φ (z∗ )
u(τ ) ≡ u(τ , s∗ ) = w − K(τ ) + − √ .
τ τ

Note that limτ →τ u(τ ) = −∞. Differentiating u(τ ) yields

r + p − κ − c − δ hb δ (hg − hb ) φ (z∗ )
u (τ ) = −K  (τ ) − + and
τ2 2τ 3/2
2(r + p − κ − c − δ hb ) 3δ (hg − hb ) φ (z∗ )
u (τ ) = −K  (τ ) + − .
τ3 4τ 5/2
Let τ be the solution of the first-order condition u (τ ) = 0. Multiplying this condition
by 2/τ and adding it to u (τ ), we get
4 Reliability or Inventory? 83

2K  (τ) δ (hg − hb ) φ (z∗ )

u (τ) = −K  (τ) − +
τ 4τ5/2

 
2K (τ) τ hg − hb 2K  (τ) hg − hb
<− + −2 + √ < − − (1.99) < 0.
τ 4 2π 
τ 3 
τ τ3

In the first inequality of this result, we used the assumption 2 (hg − hb ) < τ 3 K  (τ )
from Sect. 4.3.2, δ ≤ 1, and the upper bound on the standard normal pdf, i.e.,
φ (z) ≤ 1/(2π Var[O(τ )]) = τ /(2π ). In the second inequality, we used the up-
per bound τ < τ  0.1 as specified in Sect. 4.3.1. Suppose that the solution is in the
interior, i.e., τ < τ < τ . Then u (τ) < 0 implies that any interior critical point, if it
exists, should be a maximizer. Let us consider two cases: u (τ ) > 0 and u (τ ) ≤ 0.
If u (τ ) > 0 then τ > τ since u(τ ) initially increases and approaches −∞ as τ → τ .
Therefore τ is a unique maximizer since more than one maximizer requires at least
one interior minimizer (as u(τ ) is continuous), which contradicts our earlier obser-
vation that any interior critical point should be a maximizer. Now suppose u (τ ) ≤ 0.
Again, if an interior critical point exists, then it should be a maximizer. But this
means a minimizer should exist to the left of the maximizer, since u(τ ) initially
decreases. This leads to a contradiction, and therefore, no interior critical point
exist in this case; u(τ ) decreases monotonically if u (τ ) ≤ 0. Then u(τ ) is maxi-
mized at τ = τ . Summarizing, the supplier uniquely chooses τ ∗ = τ if u (τ ) ≤ 0 and
τ ∗ = τ > τ if u (τ ) > 0, where τ is obtained from the first-order condition (4.1). To
establish a connection between this result and the three cases stated in the lemma,
observe that, for pmin ≡ c + δ hb < p < c + δ hg ≡ pmax and a fixed τ , we have:
(i) m(τ ) ≡ lim p→pmin u (τ ) = −K  (τ ) − (r − κ )/τ 2 ; (ii) m(τ ) ≡ lim p→pmax u (τ ) =
−K  (τ )−(r − κ + δ (hg − hb ))/τ 2 < m(τ ); (iii) ∂ u (τ )/∂ p = −1/τ 2 −z∗ /(2τ 3/2 ) =
−1/(2τ 2 )−s∗ /(2τ ) < 0. At τ = τ , m(τ ) = (b − r)/τ 2 and m(τ ) = (a − r)/τ 2 , where
a and b are defined in the lemma. If 0 ≤ r ≤ a, we have m(τ ) > m(τ ) ≥ 0. Since u (τ )
decreases in p from m(τ ) > 0 to m(τ ) ≥ 0, u (τ ) > 0 for all p ∈ (pmin , pmax ), imply-
ing that u(τ ) is increasing initially at τ = τ . Therefore, the optimal τ is found in the
interior, i.e., τ ∗ = τ > τ . Next, assume a < r < b. Then m(τ ) > 0 and m(τ ) < 0, im-
plying that there exists p (r) ∈ (pmin , pmax ) such that u (τ ) > 0 for p ∈ (pmin , p (r))
and u (τ ) ≤ 0 for p ∈ [p (r) , pmax ). As we found above, τ ∗ = τ > τ in the former
case and τ ∗ = τ in the latter case. p (r) is determined from the equation u (τ ) = 0.
Finally, assume r ≥ b. Then we have 0 ≥ m(τ ) > m(τ ), which implies that u (τ )
remains in the negative region as p increases from pmin to pmax . So u (τ ) ≤ 0 for
all p ∈ (pmin , pmax ). Then by the finding above, τ ∗ = τ . The comparative statics re-
sults in the lemma are shown via implicit differentiation of the first-order condition
u (τ) = 0.

Proof (Proof of Proposition 2). The following statement can be proved (proof is
omitted). Under RBC, one of the√following three outcomes emerges in equilibrium,
along with the condition L(z∗ )/ τ ∗ = β : (i) τ ∗ > τ that solves (4.1) with r = 0,
(ii) τ ∗ > τ that solves (τ ∗ )2 K  (τ ∗ ) = κ − hg + hb and (4.1), or (iii) τ ∗ = τ . That the
backorder constraint binds in equilibrium follows directly from this statement. Bind-
ing constraint reduces the original two-dimensional problem to a single-dimensional
84 S.-H. Kim et al.
√
one, as the equation L(z)/ τ = β establishesa one-to-one
√  correspondence between
τ and s. Writing this condition as z = L−1 β τ = ζ (τ ) and substituting it in

the customer’s expected cost expression yields the √ reduced cost function C(τ ) =
u + hg N + K(τ ) + (κ + c + hb )/τ + (c + hg )ζ (τ )/ τ . As expected, this function is
convex and minimized at the first-best solution τ FB that we derived in Proposition 1.
(This can be verified by noting that ζ  (τ ) = (− f (ζ (τ )) + ζ (τ ))/(2τ ).) It remains
to choose the optimal solution among the three candidates for the equilibrium iden-
tified in the above statement. Consider each case (i)–(iii):
(i) In Lemma 1 we derived the optimality condition z∗ = Φ −1 [1 − (c + δ hg − p)/
(δ (hg − hb ))]. Inverting this and letting z∗ = ζ (τ ), i.e., the necessary condition
for the binding backorder constraint, yields p = c+ δ hb + δ (hg − hb ) Φ (ζ (τ )).
Substituting this in (4.1) and setting r = 0, we get
 √ 
τ 2 K  (τ ) = κ − δ (hg − hb ) Φ (ζ (τ )) − ( τ /2)φ (ζ (τ )) . (4.5)
√ √
Note that Φ (ζ (τ )) − ( τ /2)φ (ζ (τ )) = 1 − Φ (ζ (τ )) − ( τ /2)φ (ζ (τ )) <
1. Hence, the solution of (4.5), which we call τ1 , satisfies τ12 K  (τ1 ) > κ −
δ (hg − hb ).
(ii) Optimal τ for this case, called τ2 , is given by the stated condition τ 2 K  (τ ) =
κ − (hg − hb ). It is clear that τ22 K  (τ2 ) = κ − (hg − hb ) < κ − δ (hg − hb ) <
τ12 K  (τ1 ), where the last inequality is from (i) above.
(iii) In this case we have τ ∗ = τ . From the assumption τ 2 K  (τ ) < κ − (hg − hb )
found in Sect. 4.3.2, we have τ 2 K  (τ ) < κ − (hg − hb ) = τ22 K  (τ2 ).
Combining the inequalities we derived above, we have τ 2 K  (τ ) < τ22 K  (τ2 ) <
τ12 K  (τ1 ). Since τ 2 K  (τ ) is increasing, this implies τ < τ2 < τ1 . Next, we show τ1 <
τ FB . Using the definition of Γ (τ ), the optimality condition (4.5) can be rewritten as

(1 − δ ) hb + p (1 − δ ) hg + p
Γ ( τ1 ) − − f (ζ (τ1 )) = K  (τ1 ). (4.6)
τ12 2τ
3/2
1

The derivative of the customer’s cost is C (τ ) = K  (τ ) − Γ (τ ). Evaluating this at τ1

using (4.6), it is immediate that C (τ1 ) < 0. Since C(  τ ) is convex and minimized
at τ , C (τ1 ) < 0 implies τ1 < τ . Therefore, we have τ < τ2 < τ1 < τ FB and
FB  FB

among the three candidate equilibrium outcomes (i)–(iii) above, (i) produces the
lowest customer cost at τ1 since C(  τ ) is decreasing for τ < τ FB . The optimality
condition (4.2) is obtained by rearranging (4.5).

Proof (Proof of Lemma 2). The supplier’s expected profit under PBC is u(τ , s) =
E[T | τ , s] − ψ (τ , s) = w − K(τ ) − κ /τ − (c + δ hb ) s − δ (hg − hb ) E[I | τ , s] −
vE[B | τ , s]. Differentiating this with respect to s (see Lemma 3), we get
√ ∂ u/∂ s =
v−c− δ hb −(v + δ (hg − hb )) Φ (z) and ∂ 2 u/∂ s2 = − (v + δ (hg − hb )) τφ (z) < 0,
i.e., for a fixed τ , the supplier’s expected profit is concave in s. It can be shown
that a finite feasible solution of (C) exists only if v > c + δ hb (proof is omitted).
With v > c + δ hb , the profit-maximizing s is found in the interior. The first-order
4 Reliability or Inventory? 85

condition ∂ u/∂ s = 0 yields z∗ = Φ −1√ [1 − (c + δ hg )/(v + δ (hg − hb ))], which is

independent of τ . With s∗ = 1/τ + z∗ / τ , the supplier’s expected profit√becomes
u(τ ) ≡ u(τ , s∗ ) = w − K(τ ) − (κ + c + δ hb )/τ − (v + δ (hg − hb )) φ (z∗ )/ τ . Note
that limτ →τ u(τ ) = −∞. Differentiating u(τ ), we get

κ + c + δ hb (v + δ (hg − hb )) φ (z∗ )
u (τ ) = −K  (τ ) + + and
τ2 2τ 3/2
2(κ + c + δ hb ) 3 (v + δ (hg − hb )) φ (z∗ )
u (τ ) = −K  (τ ) − − < 0,
τ3 4τ 5/2
which shows that u(τ ) is concave. Evaluating u (τ ) at τ = τ and using the
assumption τ 2 K  (τ ) < κ − (hg − hb ) from Sect. 4.3.2 and the condition v > c + δ hb
in Lemma 2, we have

κ + c + δ hb (v + δ (hg − hb )) φ (z∗ )
u (τ ) = −K  (τ ) + +
τ2 2τ 3/2
c + hg − (1 − δ ) hb (c + δ hg ) φ (z∗ )
> + > 0.
τ2 2τ 3/2

Since u(τ ) is a concave function that initially increases at τ = τ and approaches −∞

as τ → τ , we conclude that it is maximized at τ > τ which is uniquely determined
from the first-order condition u (τ ) = 0, as written in (4.3). To obtain sensitivity
results, first note that
∂ z∗ c + δ hg 1
= > 0.
∂v (v + δ (hg − hb )) φ (z∗ )
2

Using this result and via implicit differentiation of the first-order condition
u (τ) = 0, we can show that

∂ τ L(z∗ )   −1
= u (τ)
− > 0.
∂ v 2τ3/2
To show ∂ s∗ /∂ v > 0, we first derive two intermediate results. Observe

2(κ + c + δ hb ) 3 (v + δ (hg − hb )) φ (z∗ )

u (τ) = K  (τ) +
− +
τ3 4τ5/2
2(κ + c + δ hb ) 3 (c + δ hg) ∗
> K  (τ) + + z
τ3 4τ5/2
 
8κ + 5 (c + δ hg ) 3 (c + δ hg ) ∗ c + δ hg 1 ∗
> + s > +s ,
4τ3 4τ2 4τ2 τ

where we used
c + δ hg
L(z∗ ) = φ (z∗ ) − z∗ Φ (z∗ ) = φ (z∗ ) − z∗ > 0
v + δ (hg − hb )
86 S.-H. Kim et al.

in the first inequality and the assumption 2 (hg − hb ) < τ 3 K  (τ ) from Sect. 4.3.2 in
the second. Also from Lemma 4, we obtain
 2
c + δ hg 1
−L (z∗ ) + ≥ 0.
v + δ (hg − hb ) φ (z∗ )

Using these two results, we can verify ∂ s∗ /∂ v > 0.

Proof (Proof of Proposition 3). We first show that the backorder constraint binds at
the optimum. First observe that, for τ ∗ = τ > τ and s∗ > 0 found in Lemma 2,
 
∂ E[B | τ, s∗ ] ∂ L(z∗ ) L(z∗ ) ∂ τ Φ (z∗ ) ∂ z∗
= √ = − 3/2 − √ < 0,
∂v ∂v τ 2τ ∂ v τ ∂ v
since ∂ τ/∂ v > 0 and ∂ z∗ /∂ v > 0, as we showed in the proof of the lemma. Com-
bined with limv→c+δ hb E[B | τ, s∗ ] > β and limv→∞ E[B | τ, s∗ ] = 0, which are straight-
forward to show, ∂ E[B | τ, s∗ ]/∂ v < 0 implies that the feasible region for the back-
order constraint E[B | τ, s∗ ]√≤ β can be expressed as v ≥ vmin , where vmin > c + δ hb
∗ ∗
solves E[B | τ , s ] = L(z )/ τ = β . Second, differentiating the customer’s expected

cost C(τ, s∗ ) = u + hg N + K(τ) + (κ − hg + hb )/τ + (c + hg )s∗ and substituting the
supplier’s optimal response τ given by the first-order condition (4.3) yields
 
∂C   κ − hg + hb ∂ τ ∂ s∗
= K (τ ) − + (c + hg )
∂v τ2 ∂v ∂v
 
c + hg − (1 − δ ) hb v + δ (hg − hb ) ∗ ∂ τ ∂ s∗
= + φ (z ) + (c + hg ) > 0,
τ2 2τ3/2 ∂v ∂v

since ∂ τ/∂ v > 0 and ∂ s∗ /∂ v > 0. This monotonicity implies that the C(τ, s∗ ) is min-
imized at the smallest feasible value of√ v, i.e., it is optimal to set vP = vmin , at which
∗
the backorder constraint binds (L(z )/ τ = β ). The equilibrium values vP and τ P ,
determined
√ from (4.4), are obtained by combining the optimality conditions (4.3),
L(z∗ )/ τ = β , and z∗ = Φ −1 [1 − (c + δ hg )/(v + δ (hg − hb ))].

Proof (Proof of Proposition 4). In all three cases√(FB, RBC, and PBC) the back-
order constraint binds in equilibrium, i.e., L(z)/ τ = β . Applying this, the cus-
tomer’s expected cost C(τ , s) = u + hg N + K(τ ) + (κ − hg + hb )/τ + (c + hg )s be-
√
comes C(  τ ) = u+hg N +K(τ )+(κ + c + hb )/τ +(c+hg )ζ (τ )/ τ , which is convex
and minimized at τ FB (see the proof of Proposition 2 for more details). Its derivative
is C (τ ) = K  (τ ) − Γ (τ ).
(i) Substituting the optimality conditions for RBC and PBC found in (4.6) and
(4.4) in C (τ ), along with the feasibility condition c + δ hb < p from Lemma 1,
it is easy to verify C (τ R ) < C (τ P ) ≤ C (τ FB ) = 0. This, combined with con-
vexity of C(  τ ), implies τ R < τ P ≤ τ FB . The next result follows immediately
4 Reliability or Inventory? 87
√
from the√ binding backorder√ constraint, as τ R < τ P ≤ τ FB and β = L(zR )/ τ√R =
L(zP )/ τ P = L(zFB )/ τ FB imply zR > zP ≥ zFB , and from s∗ = 1/τ ∗ + z∗ / τ ∗ ,
we have sR > sP ≥ sFB . The next result CR > CP ≥ CFB is implied by convexity of
 τ ), C (τ FB ) = 0, and τ R < τ P ≤ τ FB .
C(
(ii) Implicit differentiation of the first-order condition (4.2) yields
 √ 
∂ τ R (hg − hb )Φ (ζ (τ R )) τ R φ (ζ (τ R ))
= −2 + .
∂δ 2(τ R )2C (τ R ) Φ (ζ (τ R ))

In the lower limit z → z = − √1τ (see Sect. 4.5),

φ (z) −zφ (z) 1

lim = lim =√ ,
z→z Φ (z) z→z φ (z) τ
√
where we used √ rule. Since φ (z)/Φ (z) is decreasing, φ (z)/Φ (z) < 1/ τ
√ l’Hopital’s
for any z ≡ τ s − 1/ τ that corresponds to given τ and s. Hence,
√ √
τ R φ (ζ (τ R )) τR
−2 + < −2 + √ = −1.
Φ (ζ (τ ))
R
τR

Together with C (τ R ) > 0, this implies ∂ τ R /∂ δ < 0. Then ∂ sR /∂ δ > 0 immediately
follows from the binding backorder constraint. ∂ CR /∂ δ > 0 can be shown using the
envelope theorem. Similarly, implicit differentiation of (4.4) yields
 
∂ τP 1 hb hg   P  ∂ sP ∂ CP
= + f ζ τ > 0, < 0, and < 0.
∂δ C (τ P ) τ 2 2τ 3/2 ∂δ ∂δ

(iii) In the limit δ → 1, (4.4) becomes Γ (τ P ) = K  (τ P ), the same optimality

condition as the first-best. Hence, τ P → τ FB in this limit. Then sP → sFB and CP →
CFB follow. From (i) and (ii), we have τ R < τ FB and ∂ τ R /∂ δ < 0, implying that
the gap between τ R and τ FB is narrowed with smaller δ . In the limit δ → 0, (4.2)
becomes
c + hb c + hg
Γ (τ R ) − R 2 − f (ζ (τ R )) = K  (τ R ),
(τ ) 2(τ R )3/2
which is not equal to the first-best condition. Therefore, τ R never approaches its
first-best counterpart and neither do sR and CR .

References

Bakshi N, Kim S-H, Savva N (2014) Signaling new product reliability with after-sales service
contracts (2015). Manag Sci 61(8):1812–1829
Cachon GP (2003) Supply chain coordination with contracts. In: Graves S, de Kok T (eds) Hand-
books in operations research and management science: supply chain management. North Hol-
land, Amsterdam
88 S.-H. Kim et al.

Chan T, de Véricourt TF, Besbes O (2014) Contracting in medical equipment maintenance services:
an empirical investigation. Working paper
Cohen MA, Kleindorfer PR, Lee HL (1989) Near-optimal service constrained stocking policies for
service parts. Oper Res 37(1):104–117
Cohen MA, Kamesam P, Kleindorfer P, Lee H, Tekerian A (1990) Optimizer: a multi-echelon
inventory system for service logistics management. Interfaces 20(1):65–82
Cohen MA, Agrawal N, Agrawal V (2006) Achieving breakthrough service delivery through
dynamic asset deployment strategies. Interfaces 36(3):259–271
Cummins JM (1977) Incentive contracting for national defense: a problem of optimal risk sharing.
Bell J Econ 8:168–185
Department of Defense (2009) DoD weapon system acquisition reform product support assess-
ment. https://acc.dau.mil/CommunityBrowser.aspx?id=328610
Geary S (2006) Ready for combat. DC Velocity 4(7):75–80
Gibbons R (2005) Incentives between firms (and within). Manag Sci 51(1):2–17
Government Accountability Office (2003) Best practices: setting requirements differently could
reduce weapon systems’ total ownership costs. GAO-03-57. http://www.gao.gov/new.items/
d0357.pdf
Guajardo J, Cohen M, Kim S-H, Netessine S (2012) Impact of performance-based contracting on
product reliability: An empirical analysis. Manag Sci 58(5):961–979
Holmström B, Milgrom P (1991) Multitask principal-agent analyses: incentive contracts, asset
ownership, and job design. J Law Econ Org 7:24–52
Jain N, Hasija S, Popescu DG (2013) Optimal contracts for outsourcing of repair and restoration
services. Oper Res 61(6):1295–1311
Kang K, Doerr KH, Apte U, Boudreau M (2010) Decisions support models for valuing improve-
ments in component reliability and maintenance. Mil Oper Res 15(4):55–68
Kim S-H (2010) Incentives in multi-indenture service supply chains. Working paper, Yale Univer-
sity
Kim S-H, Cohen MA, Netessine S (2007) Performance contracting in after-sales service supply
chains. Manag Sci 53(12):1843–1858
Kim S-H, Cohen MA, Netessine S, Veeraraghavan S (2010) Contracting for infrequent restoration
and recovery of mission-critical systems. Manag Sci 56(9):1551–1567
Kumar UD, Nowicki D, Ramirez-Marquez JE, Verma D (2007) A goal programming model for
optimizing reliability, maintainability and supportability under performance based contracts.
Int J Reliab Qual Saf Eng 14(3):251–261
Mirzahosseinian H, Piplani R (2011) A study of repairable parts inventory system operating under
performance-based contract. Eur J Oper Res 214(2):256–261
Muckstadt JA (2005) Analysis and algorithms for service parts supply chains. Springer, New York
Öner KB, Kiesmüller GP, van Houtum G-J (2010) Optimization of component reliability in the
design phase of capital goods. Eur J Oper Res 205:615–624
Öner KB, Scheller-Wolf A, van Houtum G-J (2013) Redundancy optimization for critical compo-
nents in high-availability technical systems. Oper Res 61(1):244–264
Rogerson WP (1994) Economic incentives and the defense procurement process. J Econ Perspect
8(4):65–90
Sherbrooke CC (1968) Metric: a multi-echelon technique for recoverable item control. Oper Res
16:122–141
Sherbrooke CC (2004) Optimal inventory modeling of systems: multi-echelon techniques.
Springer, New York
Chapter 5
Project Contracting Strategies for Managing
Team Dynamics

George Georgiadis and Christopher S. Tang

Abstract In this chapter we study a team dynamic problem in which a group of

agents collaborate over time to complete a project. The project progresses at a rate
that depends on the agents’ efforts, and it generates a payoff upon completion. First,
we show that agents work harder the closer the project is to completion, and mem-
bers of a larger team work harder than members of a smaller team—both individu-
ally and on aggregate—if and only if the project is sufficiently far from completion.
Second, we analyze the problem faced by a manager who is the residual claimant of
the project and she chooses the size of the team and the agents’ incentive contracts
to maximize her discounted payoff. We show that the optimal symmetric contract
compensates the agents only upon completion of the project. Finally, we endogenize
the size of the project, where a bigger project is one that requires greater cumula-
tive effort and generates a larger upon completion. We show that if the manager can
commit to her optimal project size at the outset of the game, then she will choose
a smaller project relative to the case without commitment. An implication of this
result is that without commitment, the manager is better off delegating the decision
rights over the project size to the agents.

G. Georgiadis
Kellogg School of Management, Northwestern University, Evanston, IL, USA
e-mail: [email protected]
C.S. Tang
UCLA Anderson School of Management, Los Angeles, CA, USA
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 89

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 5
90 G. Georgiadis and C.S. Tang

5.1 Introduction

Teamwork and projects are omnipresent. Lawler et al. (2001) reported that most
large corporations engage a substantial proportion of their workforce in teams. This
is because teamwork has been shown to increase productivity in both manufacturing
and service firms (Ichniowski and Shaw 2003). Moreover, the use of teams is espe-
cially common in situations when the task at hand will result in a defined deliverable
(Harvard Business School 2004). A key component of most projects is choosing the
features that must be included before the decision maker deems the product ready to
market. When choosing these features, the decision maker must trade off the added
value derived from a bigger or a more complex project (i.e., one that contains more
features) against the additional cost associated with designing and implementing the
additional features.
Motivated by these observations, we analyze a team dynamic problem in which
a group of agents collaborate over time to complete a project, and we address a
number of questions that naturally arise in this environment. In particular, what is
the impact of group size on the agents’ incentives? How should a manager determine
the agents’ incentive contracts to maximize her profit; for example, should they
be rewarded for reaching intermediate milestones? How and by whom should the
optimal project size be chosen; for example, can the manager benefit by delegating
the decision rights over the project size to the agents?
Our model can be applied to both within firms (e.g., research teams in new prod-
uct development or consulting projects) and across firms (e.g., R&D joint ventures).
More broadly, the model is applicable to settings in which a group of agents collabo-
rate to complete a project, which generates a payoff upon completion. The expected
project completion time is sufficiently long such that the agents discounting time
matters. A natural example is the Myelin Repair Foundation (MRF): a collabora-
tive effort among a group of leading scientists in quest of a treatment for multiple
sclerosis (Lakhani and Carlile 2012). This is a long-term venture, progress is grad-
ual, each principal investigator incurs an opportunity cost by allocating resources
to MRF activities (which gives rise to incentives to free-ride), and it will pay off
predominantly when an acceptable treatment is discovered.
We use a parsimonious model to analyze a dynamic contribution game in which
a group of agents collaborate to complete a project. The project progresses at a rate
that depends on the agents’ costly effort, and it generates a payoff upon comple-
tion. Formally, the state of the project qt starts at 0, and it progresses according to
dqt = ∑ni=1 ai,t dt, where ai,t denotes the effort level of agent i at time t. The project
generates a payoff at the first stopping time τ such that qτ = Q, where Q is a one-
dimensional parameter that captures the project requirements, or equivalently, the
project size. The manager is the residual claimant of the project, and she possesses
the decision rights over its size, as well as the agents’ contracts. The model is very
tractable, and payoffs and strategies are derived in closed-form.
In Sect. 5.3, we analyze the agents’ problem given a fixed project size. We
characterize the (essentially) unique Markov perfect equilibrium, wherein at every
moment, each agent’s strategy depends solely on the current state of the project.
5 Project Contracting Strategies for Managing Team Dynamics 91

A key result is that the agents exert greater effort the closer the project is to comple-
tion. Intuitively, this is because they discount time and they are compensated upon
completion; hence, their incentives become stronger as the project progresses. An
implication of this result is that efforts are strategic complements (across time). This
is because by raising his effort, an agent brings the project closer to completion, thus
incentivizing the other agents to raise their future efforts.
We also examine the impact of team size on the agents’ incentives. In contrast
to static moral hazard in teams models where an increase in the group size leads
to lower effort levels (Olson 1965), we obtain a partially opposite result in our dy-
namic model. In particular, members of a larger team work harder than members
of a smaller team—both individually and on aggregate—if and only if the project is
sufficiently far from completion. Intuitively, when the project is close to completion,
then the game resembles the static one, and the standard free-riding effect which as-
serts that smaller teams work harder than larger ones holds. However, in contrast
to the static game and because at every moment each agent observes the state of
the project before choosing his effort, the strategic complementarity is stronger in a
larger team. That is because by raising his effort, each agent induces a greater num-
ber of other agents to raise their future efforts, which in turn renders him better off.
Noting that this effect is stronger at the early stages of the project, because the bene-
fits from greater effort have a longer lasting effect, it follows that this encouragement
effect dominates the free-riding effect when the project is far from completion.
In Sect. 5.4, we analyze the manager’s problem, and we show that the optimal
symmetric contract compensates the agents only upon completion. The intuition
is that by backloading payments (compared to rewarding the agents for reaching
intermediate milestones), the manager can provide the same incentives at the early
stages of the project (via continuation utility), while providing stronger incentives
when the project is close to completion. This result simplifies the manager’s problem
to determining the team size and her budget for compensating the agents.
In Sect. 5.5, we endogenize the size of the project, where a larger project re-
quires greater cumulative effort, and delivers a bigger payoff upon completion. We
consider the case in which the manager can commit to her optimal project size at
the outset of the game, and the case in which she cannot and at every moment, she
observes the state of the project and decides whether to complete it immediately,
or to let the agents continue working and re-evaluate her decision to complete the
project a moment later. To motivate this case, note that an intrinsic challenge in-
volved in choosing the requirements of a project is that the manager may not be
able to commit to them in advance. This can be due to the fact that the requirements
are difficult to describe; for example, if the project involves significant novelty in
quality or design. What we have in mind about the incontractibility of the project
requirements was eloquently posed by Tirole (1999):
In practice, the parties are unlikely to be able to describe precisely the specifics of an in-
novation in an ex ante contract, given that the research process is precisely concerned with
finding out these specifics, although they are able to describe it ex post.

For example, anecdotal evidence from the development of Apple’s first genera-
tion iPod indicates that Steve Jobs kept changing the requirements of the iPod as
92 G. Georgiadis and C.S. Tang

the project progressed. In particular, the development team would get orders such
as “Steve doesn’t think it is loud enough,” or “the sharps are not sharp enough,” or
“the menu is not coming up fast enough” (Wired Magazine 2004). This suggests
that committing to a set of features and requirements early on was not desirable in
the development of an innovative new product such as the iPod back in 2001.
We show that without commitment, the manager chooses a larger project rela-
tive to the case with commitment. Practically, this result asserts that the manager
extends the project; for example, by introducing additional requirements. The man-
ager chooses the project size by trading off the marginal benefit of a larger project
against the marginal cost associated with a longer wait until the larger project is
completed. However, as the project progresses, the agents increase their effort, so
that this marginal cost decreases, while the respective marginal benefit does not
change. As the project size will be chosen such that the two marginal values are
equal, it follows that if the manager cannot commit to her optimal project size at the
outset, she will end up choosing a bigger project.
Anticipating that if the manager cannot commit, then she will choose a larger
project, the agents decrease their effort, which renders the manager worse off. We
show that without commitment and assuming that the agents receive a share of the
project’s worth upon completion (i.e., an equity contract), the manager finds it opti-
mal to delegate the decision rights over the project size to the agents. In this case, the
agents will choose a smaller project than is optimal for the manager, but their prefer-
ences are time-consistent. Intuitively, because (unlike the manager) they also trade
off the cost of effort when choosing the project size, their marginal cost associated
with a larger project does not decrease as the project progresses.

5.1.1 Related Literature

First and foremost, this chapter is related to the literature on dynamic contribution
games. The general theme of these games is that a group of agents interact repeat-
edly, and in every period (or moment), each agent chooses his contribution (or effort)
to a joint project at a private cost. Contributions accumulate until they reach a cer-
tain threshold, at which point the game ends. Agents receive flow payoffs while the
game is in progress, a lump-sum payoff at the end, or a combination thereof. Admati
and Perry (1991) and Marx and Matthews (2000) show that contributing little by lit-
tle over multiple periods, each conditional on the previous contributions of the other
agents helps mitigate the free-rider problem. More recently, Yildirim (2006) and
Kessing (2007) show that in contrast to the case in which the project generates flow
payments while it is in progress as studied by Fershtman and Nitzan (1991), efforts
are strategic complements when the agents receive a payoff only upon completion.
A second strand of related literature is that on incomplete contracting. In partic-
ular, our article is closely related to the articles that study how ex-ante contracting
limitations generate incentives to renegotiate the initial contract ex-post (Grossman
and Hart 1986; Aghion and Tirole 1994; Tirole 1999, and others). A subset of this
5 Project Contracting Strategies for Managing Team Dynamics 93

literature focuses on situations wherein the involved parties have asymmetric infor-
mation. Here, ratchet effects have been shown to arise in principal-agent models in
which the principal learns about the agent’s ability over time, and the agent reduces
his effort to manipulate the principal’s beliefs about his ability (Freixas et al. 1985;
Laffont and Tirole 1988). Another thread of this strand includes articles that con-
sider the case in which the agent is better informed than the principal, or he has
better access to valuable information. A common result is that delegating the deci-
sion rights to the agent is beneficial as long as the he is sufficiently better informed
and the incentive conflict is not too large (Aghion and Tirole 1997; Dessein 2002). In
our model however, all parties have full and symmetric information, so that ratchet
effects and the incentives to delegate the decision rights to the agents arise purely
out of moral hazard.
The chapter is organized as follows. We introduce the model in Sect. 5.2, and in
Sect. 5.3 we analyze the agents’ problem given a fixed project size. In Sect. 5.4, we
study the manager’s problem, and we characterize the optimal contract. In Sect. 5.5
we endogenize the project size, and we conclude in Sect. 5.6. This paper unifies
results from Georgiadis et al. (2014) and Georgiadis (2015a) and the proofs are
provided therein.

5.2 The Model

A group of n identical agents contracts with a manager to undertake a project. The

agents exert (costly) effort over time to complete the project, they receive a lump-
sum compensation upon completing the project, and they are protected by limited
liability.1 The manager is the residual claimant of the project, and she possesses the
decision rights over its size, as well as the agents’ compensation. A project of size
Q ≥ 0 generates a payoff equal to Q upon completion. Time t ∈ [0, ∞) is continuous;
all parties are risk neutral, they discount time at rate r > 0, and have outside option 0.
The project starts at state q0 = 0. At every moment t, each agent observes the state
of the project denoted by qt , and exerts costly effort to influence the progress of the
project according to:
n 
dqt = ∑ ai,t dt ,
i=1

1 In practice, the relevant employees are rewarded by a combination of flow payments (i.e., peri-

odic salary) and compensation after completion of the project. The latter can take the form of bonus
lump-sum payments, stock options (that are correlated to the profit generated by the project), and
reputational benefits. In the base model, we assume (for tractability) that the agents are compen-
sated only by a lump-sum upon completion of the project. Georgiadis (2015a) also considers the
case in which, in addition to a lump-sum payment upon completion, they also receive a per unit of
time compensation while the project is ongoing.
94 G. Georgiadis and C.S. Tang

where ai,t denotes the (unverifiable) effort level of agent i at time t.2 The project
is completed at the first stopping time τ such that qτ = Q.3 Each agent is credit
constrained, his effort choices are not observable to the other agents, and his flow
cost of exerting effort level a is a2 /2. Finally, we assume that (1) strategies are
Markovian, so that at every moment, each agent chooses his effort level as a function
of the current state of the project qt , and (2) incentive contracts are symmetric.4

5.3 Agents’ Problem

In this section, we study the agents’ problem, and we characterize the unique
project-completing Markov Perfect equilibrium (hereafter MPE) wherein each agent
conditions his strategy at t only on the current state of the project qt . Through-
out Sects. 5.3 and 5.4 we take the project size Q as given, and we endogenize Q
in Sect. 5.5.

5.3.1 Preliminaries

We assume (for now) that each agent receives a lump sum reward V /n upon com-
pletion of the project and no intermediate compensation. We then show in Sect. 5.4
that the optimal symmetric contract rewards the agents only upon completion of
the project. Moreover, we will carry out the analysis in Sect. 5.3 assuming that the
project size Q is given, and we will endogenize it in Sect. 5.5.
Given the current state of the project qt , and others’ strategies, agent i’s dis-
counted payoff function is given by
  τ  
a2 
Πi,t (q) = max e −r(τ −t)
V− e−r(s−t) i,s
ds  {a−i,s }s≥t , (5.1)
{ai,s }s≥t t 2

2 The assumptions that efforts are perfect substitutes and the project progresses deterministically
are made for tractability. Georgiadis et al. (2014) and Georgiadis (2015a) also examine the case in
which efforts are complementary and the project progresses stochastically, and they show that all
results continue to hold.
3 We implicitly assume that the agents do not face a deadline to complete the project. This assump-

tion is made (1) for simplicity, and (2) because deadlines are generally not renegotiation proof. As
a result, if the project has not been completed by the deadline, the agents find it mutually beneficial
to extend the deadline. For a treatment of deadlines, see Georgiadis (2015b).
4 When progress is deterministic, as Georgiadis et al. (2014) show, the game also admits non-

Markovian equilibria where at every moment t, each agent’s strategy is a function of the entire
path of the project {qs }s≤t . We use the deterministic specification as a reduced form for a stochas-
tic process, in which case as Georgiadis (2015a) conjectures, the agents’ payoffs from the best
symmetric Public Perfect equilibrium are equal to the payoffs corresponding to the MPE.
5 Project Contracting Strategies for Managing Team Dynamics 95

where τ denotes the completion time of the project and it depends on the agents’
strategies. The first term captures the agent’s net discounted payoff upon completion
of the project, while the second term captures his discounted cost of effort for the
remaining duration of the project. Because payoffs depend solely on the state of
the project (i.e., q) and not on the time t, this problem is stationary, and hence the
subscript t can be dropped. Using standard arguments (Dixit 1999), one can derive
the Hamilton-Jacobi-Bellman equation for the expected discounted payoff function
for agent i  2  n  
a
rΠi (q) = max − i + ∑ a j Πi (q) , (5.2)
ai 2 j=1

subject to the boundary conditions

Πi (q) ≥ 0 for all q and Πi (Q) = V. (5.3)

The first boundary condition captures the fact that each agent’s discounted payoff
must be non-negative as he can guarantee himself a payoff of 0 by exerting no effort
and hence incurring no effort cost. The second boundary condition states that upon
completing the project, each agent receives his reward and exerts no further effort.5

5.3.2 Markov Perfect Equilibrium (MPE)

In a MPE, at every moment, each agent i observes the state of the project q, and
chooses his effort ai to maximize his discounted payoff while accounting for the
effort strategies of the other team members. It follows from (5.2) that the first or-
der condition for agent i’s problem yields that ai (q) = Πi (q): at every moment, he
chooses his effort such that the marginal cost of effort is equal to the marginal benefit
associated with bringing the project closer to completion. By noting that the second
order condition is satisfied and that the first order condition is necessary and suffi-
cient, it follows that in any differentiable, project-completing MPE, the discounted
payoff for agent i satisfies
 n 
1 
rΠi (q) = − [Πi (q)] + ∑ Π j (q) Πi (q)
2 
(5.4)
2 j=1

subject to the boundary conditions (5.3). Proposition 1 characterizes the MPE, and
establishes conditions under which it is unique.

Proposition 1. For any given project size Q, there exists a Markov Perfect equi-
librium for the game defined by (5.1). This equilibrium is symmetric, each agent’s
effort strategy satisfies

5 Because the agents’ rewards are independent of the completion time, the game is stationary, and
so we can drop the subscript t.
96 G. Georgiadis and C.S. Tang

r 2V 2n − 1
a(q) = [q −C]+ , where C = Q − , (5.5)
2n − 1 r n

and the project is completed at τ (Q) = ((2n − 1)/rn) ln[1 − Q/C].6 In equilibrium,
each agent’s discounted payoff is given by

r ([q −C]+ )2
Π (q) = . (5.6)
2 2n − 1

If Q2 < (2V /r) · (2n − 1)/n, then this equilibrium is unique, and the project is com-
pleted in finite time. Otherwise, there also exists an equilibrium in which no agent
ever exerts any effort and the project is never completed.

Because the project starts at q0 = 0, notice that if C ≥ 0, or equivalently if Q2 ≥

(2V /r) · (2n − 1)/n, then there exists no MPE in which the project is completed.
This is because the discounted cost to complete the project exceeds its discounted
net payoff, and hence the agents are better off not exerting any effort, in which
case the project is never completed. On the other hand, if C < 0, then the project is
completed, and each agent’s effort level increases in the state of the project q. This is
due to the facts that agents are impatient and they incur the cost of effort at the time
the effort is exerted, while they are compensated only when the project is completed.
As a result, their incentives are stronger, the closer the project is to completion. An
implication of this observation is that efforts are strategic complements (across time)
in this game. That is because by raising his effort, an agent brings the project closer
to completion, which induces the other agents to raise their future efforts.
It is worth emphasizing that the MPE is always symmetric, but it need not be
unique. In particular, if C|n=1 ≥ 0, or equivalently if Q2 ≥ 2V /r, then there exists
an equilibrium in which no agent ever exerts any effort and the project is never
completed.7 However, it turns out that when the project size Q is endogenous, the
manager will always choose it such that the equilibrium is unique. As such, we shall
restrict attention to the project-completing MPE characterized in Proposition 1.
Lastly, one can show that in the MPE, each agent exerts less effort and is worse
off relative to the first best outcome; i.e., the case in which at every moment, a social
planner chooses each agent’s effort to maximize their total discounted payoff. This
in a direct implication of the free-rider problem: in equilibrium, each agent chooses
his effort by trading off the marginal cost of effort and his marginal benefit from
bringing the project closer to completion, but he ignores the positive externality of
his effort on the other agents.

6 To simplify notation, because the equilibrium is symmetric and unique, the subscript i is dropped

throughout the remainder of this article. Moreover, [·]+ = max{·, 0}.

7 So see why such an equilibrium exists, suppose that all agents except for i exert no effort at

q0 = 0. Then agent i finds it optimal to also exert no effort, because he is not willing to undertake
the entire project single-handedly (since C|n=1 ≥ 0).
5 Project Contracting Strategies for Managing Team Dynamics 97

5.3.3 Comparative Statics

Here we examine how each agent’s payoff and effort level depends on the parame-
ters of the problem. The following result establishes some comparative statics about
how each agent’s effort level depends on the parameters of the problem.

Proposition 2. All other parameters held constant, each agent’s effort level a(q)
possesses the following properties:
(i) a(q) is increasing in V ; and
(ii) there exists a threshold Θr such that a(q) is increasing in r if and only
if q ≥ Θr .

Statement (i) is intuitive: if the agents receive a larger reward upon completion,
then they have stronger incentives. Statement (ii) asserts that as agents become less
patient, they tend to work harder when the project is close to completion, and less
hard when it is far from completion. To see the intuition, notice that the marginal
benefit of bringing the completion time forward (which occurs if the agents raise
their effort) is equal to −(d/d τ )(e−rτ V /n) = (rV /n)e−rτ , and it increases in r if τ
is sufficiently small; i.e., if the project is sufficiently close to completion.
We next consider the effect of the team size on the agents’ incentives. For this
analysis, it is necessary to consider how each agent’s reward depends on the team
size. We assume that upon completion of the project, each agent receives reward
Vn = V /n, so that the total rewards disbursed to the agents is independent of n.

Proposition 3. Consider two teams comprising of n and m > n identical agents.

Other things equal, there exist thresholds Θn,m and Φn,m such that
(i) am (q) ≥ an (q) if and only if q ≤ Θn,m ; and
(ii) mam (q) ≥ nan (q) if and only if q ≤ Φn,m .8

By increasing the size of the team, two opposing forces influence the agents’ in-
centives: First, as is well known in the static moral hazard in teams literature starting
with Olson (1965), the agents obtain stronger incentives to free-ride. Moreover, be-
cause the agents’ incentives to free-ride are proportional to the cost of effort they are
incurring, effort increases with progress, and effort costs are convex, the free-riding
effect becomes more intense as the project progresses. Second, because efforts are
strategic complements in this game, when agents are part of a larger team, they have
incentives to raise their effort, because doing so will induce a greater number of
other agents to raise their future efforts, which in turn makes them better off. This
encouragement effect is strong at the outset of the game where a lot of progress
remains to complete the project, and it becomes weaker as the project nears com-
pletion.

8This result follows by noting that C = Q − (2V /r) · (2n − 1)/n decreases in n, while both a (Q)
and na (Q) decrease in n.
98 G. Georgiadis and C.S. Tang

In summary, the encouragement effect dominates the free-riding effect if and

only if the project is sufficiently far from completion. This implies that by increasing
the team size, the agents obtain stronger incentives when the project is far from
completion, while their incentives become weaker near completion.

5.4 Manager’s Problem

In this section, we analyze the manager’s problem who is the residual claimant of
the project, and his objective is to choose the team size and the agents’ (symmetric)
incentive contracts to maximize her ex-ante discounted profit. As in the previous
section, we will keep the project size fixed. Moreover, we will restrict attention to
incentive contracts that specify a set of milestones q0 < Q1 < . . . < QK = 0 (where
K ∈ N), and for every k ∈ {1, . . . , K}, allocates non-negative payments {Vk }ni=1 that
are payable to the agents upon reaching milestone Qk for the first time.9
We set out by considering the case in which the manager compensates the agents
only upon completing the project, and we characterize the manager’s discounted
profit function. Then we explain how this result extends to the case in which the
manager also rewards the agents for reaching intermediate milestones.
Given the team size and the agents’ rewards that are due upon completion of the
project (where we can assume without loss of generality that V ≤ Q), the manager’s
expected discounted profit function can be written as

W (qt ) = e−r(τ −t) (Q −V ) ,

where τ denotes the completion time of the project and it depends on the agents’
strategies as defined in Proposition 1. By using the first order condition for each
agent’s equilibrium effort as determined in Sect. 5.3, the manager’s expected dis-
counted profit at any given state of the project satisfies

rW (q) = na(q)W  (q) (5.7)

subject to the boundary conditions

W (q) ≥ 0 for all q and W (Q) = Q −V. (5.8)

To interpret these conditions, note that manager’s discounted profit is non-negative

at every state of the project, because she does not incur any cost or disburse any

9 The manager’s contracting space is restricted. In principle, the optimal contract should condition

each agent’s payoff on the path of qt (and hence on the completion time of the project). However,
when the project progresses deterministically, the problem becomes trivial as efforts effectively
become contractible, and in the stochastic case the problem is not tractable. For example, the
contracting approach developed in Sannikov (2008) boils down a partial differential equation with
at least variables (i.e., the state of the project q and the continuation value of each agent), which is
intractable.
5 Project Contracting Strategies for Managing Team Dynamics 99

payments to the agents while the project is in progress. On the other hand, she
receives her net profit Q−V , and the game ends as soon as the state of the project hits
Q for the first time. After substituting (5.5) and solving the above ODE, it follows
that
 (2n−1)/n 
[q −C]+ 2V 2n − 1
W (q) = (Q −V ) , where C = Q − . (5.9)
Q −C r n

We now discuss how each agent’s payoff function and the principal’s profit func-
tion extend to the case in which the manager rewards the agents upon reaching in-
termediate milestones. Recall that she can designate a set of milestones, and attach
rewards to each milestone that are due as soon as the project reaches the respective
milestone for the first time. Let Πk (·) denote each agent’s discounted payoff given
that the project has reached k − 1 milestones, which is defined for q ≤ Qk , and note
that it satisfies (5.4) subject to Πk (q) ≥ 0 for all q and Πk (Qk ) = Vk /n + Πk+1 (Qk ),
where ΠK+1 (0) = 0. The second boundary condition states that upon reaching mile-
stone k, each agent receives the reward attached to that milestone, plus the continu-
ation value from future rewards. Starting with ΠK (·) and proceeding by backward
induction, it is straightforward to derive each agent’s discounted payoff by solving
(5.4) subject to the corresponding boundary conditions.
To examine the manager’s problem, let Wk (·) denote her expected discounted
profit given that the project has reached k − 1 milestones, which is defined for
q ≤ Qk , and note that it satisfies (5.7) subject to Wk (q) ≥ 0 for all q and Wk (Qk ) =
Wk+1 (Qk ) − Vk , where FK+1 (Qk ) = Q. The second boundary condition states that
upon reaching milestone k, the manager receives the continuation value of the
project, less the payments that she disburses to the agents for reaching this mile-
stone. Again starting with k = K and proceeding backwards, it is straightforward to
derive the principal’s discounted profit.
Proposition 4 shows that one can without loss of generality restrict attention to
those that compensate the agents only upon completion of the project.
Proposition 4. The optimal symmetric contract compensates the agents only upon
completion of the project.
To prove this result, Georgiadis (2015a) considers an arbitrary set of milestones
and arbitrary rewards attached to each milestone, and constructs an alternative con-
tract that rewards the agents only upon completing the project and renders the man-
ager better off. Intuitively, because rewards are sunk in terms of incentivizing the
agents after they are disbursed, and all parties are risk neutral and they discount time
at the same rate, by backloading payments, the manager can provide the same in-
centives at the early stages of the project, while providing stronger incentives when
it is close to completion.
The value of this result lies in that it reduces the infinite-dimensional problem
of determining the team size, the number of milestones, the set of milestones, and
the rewards attached to each milestone into a two-dimensional problem, in which the
manager only needs to determine her budget V for compensating the agents and the
team size n to maximize her ex-ante discounted profit. In other words, the manager
solves
100 G. Georgiadis and C.S. Tang
   
−C (2n−1)/n
max (Q −V ) s.t. C < 0 .
V, n Q −C
Solving this optimization problem analytically is not tractable. Figure 5.1 illustrates
how the optimal contract depends on the parameters of the problem: the discount
rate r and the project size Q.
By examining the left panel in Fig. 5.1, one observes that if the project is larger,
then the manager optimally hires a larger team, and she allocates a larger proportion
of her budget to compensate the agents. From the right panel, notice that if the
discount rate r increases, then the manager show employ a bigger team and allocate
more of her budget to compensate the agents.

5.5 Project Design with and Without Commitment

In this Section, we endogenize the project size Q. To obtain tractable results, we

assume that the manager’s budget V = β Q; i.e., each agent receives a fixed propor-
tion of the payoff that the project generates, and this proportion does not depend
on Q. We consider the case in which the manager is able to commit to a particular
project size at the outset of the game, as well as the case in which she cannot, in
which case at every moment she observes the current state of the project q, and she
decides whether to stop work and collect the net profit (1 − β )q or to let the agents
continue working and re-evaluate her decision to complete the project a moment
later.
To analyze the case in which the project size Q is endogenous and the manager
cannot commit to a particular Q at the outset of the game, the agents must form
Optimal (Relative) Budget ( V / Q )

Optimal (Relative) Budget ( V / Q )

0.7 0.75
0.7
0.6
0.65
0.5 0.6
0.4 0.55
0.5
0.3 0.45
0.2 0.4
0.35
0.1
0.3
0 0.25
1 2 3 4 5 6 7 8 9 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Project Size ( Q ) Discount Rate ( r )

50 50
45 45
Optimal Team Size ( n )

Optimal Team Size ( n )

40 40
35 35
30 30
25 25
20 20
15 15
10 10
5 5
0 0
1 2 3 4 5 6 7 8 9 0.02 0.04 0.06 0.08 0.1 0.12 0.14

Project Size ( Q ) Discount Rate ( r )

Fig. 5.1 Optimal contract. The left panel illustrates how the optimal (relative) budget (i.e., V ∗ /Q)
and the optimal team size n∗ depend on the project size, given r = 0.15. The right panel illustrates
how optimal (relative) budget and the optimal team size depend on the discount rate, given Q = 10
5 Project Contracting Strategies for Managing Team Dynamics 101

beliefs about the project size that the manager will choose at a later date and choose
their efforts accordingly. Following rational expectations, we assume that in equilib-
rium, the agents can correctly anticipate the manager’s choice. Therefore, we write
the manager’s discounted profit function as
 
[q −C(Q̃)]+ (2n−1)/n   2β Q̃ 2n − 1
W (q; Q, Q̃) = (1 − β )Q , where C Q̃ = Q̃ −
Q −C(Q̃) r n
(5.10)
and Q̃ denotes the agents’ (common) belief about the project size that the manager
will choose.

5.5.1 Full Commitment

If the manager can commit to a project size at the outset of the game, then at q0 = 0,
she leads a Stackelberg game in which she chooses the project size that maximizes
her discounted profit and the agents follow by adopting the equilibrium strategy
characterized in Proposition 1. As a result, her optimal project size with full commit-
ment (FC) satisfies QMFC ∈ arg maxQ W (0; Q, Q). Noting from (5.10) that W (0; Q, Q)
is concave in Q, taking the first order condition with respect to Q yields
 2
β 2n − 1 4n
FC =
QM .
r 2n 4n − 1

5.5.2 No Commitment

If the manager has no commitment power, then at every moment she observes
the current state of the project q, and she decides whether to stop work and col-
lect the net profit (1 − β ) q or to let the agents continue working and re-evaluate
her decision to complete the project a moment later. In this case, the manager
and the agents engage in a simultaneous-action game, where the manager chooses
Q to maximize her discounted profit given the agents’ belief Q̃ and the corre-
sponding strategies, and the agents form their beliefs by anticipating the manager’s
choice Q. Therefore, her optimal project size with no commitment (NC) satisfies
QMNC ∈ arg maxQ {W (q ; Q, Q̃)}, where  in equilibrium beliefs must be correct; i.e.,
Q = Q̃. By solving ∂ W (q; Q, Q̃)/∂ Qq=Q̃=Q = 0, we have

β 2n
NC =
QM .
r 2n − 1
102 G. Georgiadis and C.S. Tang

Observe that with no commitment, the manager will choose a strictly larger project
NC > QFC .
relative to the case in which she commitment power: QM M 10,11

Proposition 5. If the manager can commit to her optimal project size at the outset,
then she finds it optimal to choose
 2
β 2n − 1 4n
QM
FC = .
r 2n 4n − 1

In contrast, without commitment, she completes the project at

β 2n
NC =
QM > QM
FC .
r 2n − 1
Moreover, the manager’s ex-ante discounted profit is higher with commitment than
without; i.e., W (0; QM
FC , QFC ) > W (0; QNC , QNC ).
M M M

Intuitively, the manager chooses the project size by trading off the marginal bene-
fit of a larger project against the marginal cost associated with a longer wait until the
larger project is completed. However, as the project progresses, the agents increase
their effort, so that this marginal cost decreases, while the respective marginal ben-
efit does not change. As the project size will be chosen such that the two marginal
values are equal, it follows that the manager’s optimal project size increases as the
project progresses.12 As a result, without commitment, the manager chooses a big-
ger project relative to the case with commitment.

5.5.3 Optimal Delegation

The manager’s limited ability to commit, in addition to disincentivizing the agents

from exerting effort, is detrimental to her ex-ante discounted profit; i.e.,
NC , QNC ) < W (0; QFC , QFC ). Thus, unable to commit, the manager might
W (0; QM M M M

consider delegating the decision rights over the project size to the agents.

10 This case raises the question of what happens to the agents’ beliefs off the equilibrium path if
the manager does not complete the project at QM NC . Suppose that the manager did not complete
the project at QM NC so that q > QNC . Clearly, Q, Q̃ > QNC , and it is straightforward to verify that
M M

∂ W (q; Q, Q̃)/∂ Q < 0 for all q, Q, Q̃ > QNC , which implies that the manager would be better off
M

had she completed the project at QM NC irrespective of the agents’ beliefs.

11 Conceptually, this commitment problem could be resolved by allowing β to be contingent on the

project size. In particular, suppose that the manager can fix β , and let β̂ (Q) equal β if Q = QM
FC , and
1 otherwise. Then, her optimal project size is equal to QM FC regardless of her commitment power
because any other project size will yield her a net profit of 0. However, this implicitly assumes that
QMFC is contractible at q = 0, which is clearly not true without commitment. Therefore, we rule out
this possibility by assuming that β is independent of Q.
12 Letting QM (q) = arg max
Q≥q {W (q ; Q, Q)}, one can show that Q (q) increases in q.
M
5 Project Contracting Strategies for Managing Team Dynamics 103

We begin by examining how the agents would select the project size. Let QA ∈
arg maxQ {Π (q; Q)} denote the agents’ optimal project size given the current state x.
Solving this maximization problem yields

β 2n − 1
QA = .
r 2n
Observe that the agents’ optimal project size is independent of the current state q.
Intuitively, this is because the agents incur the cost of their effort, so that their effort
cost increases together with their effort level as the project progresses. As a result,
unlike the manager, their marginal cost associated with choosing a larger project
does not decrease as the project evolves, and consequently they do not have incen-
tives to extend the project as it progresses.
Second, observe that the agents always prefer a smaller project than the manager;
i.e., QA < QM FC < QNC . This is because they incur the cost of their effort, so that their
M

marginal cost associated with a larger project is greater than that of the manager’s.
Proposition 6 shows that without commitment, the manager finds it optimal to
delegate the decision rights over the project size to the agents.

Proposition 6. If the manager can commit to her optimal project size at the outset,
then she find it optimal to retain the decision rights over the project size. In contrast,
without commitment, she should delegate the decision rights over Q to the agents;
NC , QNC ) < W (0; Q , Q ) < W (0; QFC , QFC ).
i.e., W (0; QM M A A M M

Since QMNC maximizes the manager’s ex-ante discounted profit, it is straightfor-

ward that she cannot benefit by delegating the decision rights over the project size to
the agents. Without commitment however, because the agents’ preferences over Q
are time-consistent whereas the manager’s are not, delegation turns out to be always
optimal.

5.6 Concluding Remarks

We use a tractable model to study the interaction between a group of agents who col-
laborate over time to complete a project and a manager who chooses its size. First,
we analytically characterize the Markov Perfect equilibrium, and we show that in
contrast to the static moral hazard in teams, larger teams may be more effective
in completing the project than smaller ones. We then study the manager’s problem
who chooses the size of the team and the contracts of the team members to max-
imize her ex-ante discounted profit. We show that the optimal symmetric contract
compensates the agents only upon completion of the project. This result reduces the
infinite-dimensional contracting problem to a two-dimensional one where the man-
ager needs to only choose the team size and her budget for compensating the agents.
Lastly, we endogenize the size of the project, and we show that without the ability
to commit to a particular project size at the outset of the game, the manager will end
104 G. Georgiadis and C.S. Tang

up choosing a larger project relative to the case in which she can commit. An impli-
cation of this result is that without commitment, she finds it optimal to delegate the
decision rights over the project size to the agents, who will choose a smaller project
but their preferences are time-consistent.
In subsequent work, Cvitanić and Georgiadis (2015) characterize a mechanism
that induces each agent to at every moment exert the efficient effort level in a
MPE. Bowen et al. (2015) study a dynamic contribution game with two heteroge-
neous agents and endogenous project size, and they analyze how different collective
choice institutions influence the size of the project that is implemented in equilib-
rium. Georgiadis (2015b) considers the effect of deadlines and the agents’ ability to
monitor the state of the project on the agents’ incentives and their payoffs. Finally,
Ederer et al. (2015) examine how the team size and the monitoring structure affects
incentives in a discrete public good contribution game using laboratory experiments.
Preliminary results support the theoretical predictions.

References

Admati AR, Perry M (1991) Joint projects without commitment. Rev Econ Stud 58:259–276
Aghion P, Tirole J (1994) The management of innovation. Q J Econ 109:1185–1209
Aghion P, Tirole J (1997) Formal and real authority in organizations. J Polit Econ 105:1–29
Bowen TR, Georgiadis G, Lambert N (2015) Collective choice in dynamic public good provision:
real versus formal authority. Working paper
Cvitanić J, Georgiadis G (2015) Achieving efficiency in dynamic contribution games. Working
paper
Dessein W (2002) Authority and communication in organizations’. Rev Econ Stud 69:811–838
Dixit A (1999) The art of smooth pasting. Taylor & Francis, London
Ederer FP, Georgiadis G, Nunnari S (2015) Team size effects in dynamic contribution games:
experimental evidence. Working paper
Fershtman C, Nitzan S (1991) Dynamic voluntary provision of public goods. Eur Econ Rev
35:1057–1067
Freixas X, Guesnerie R, Tirole J (1985) Planning under incomplete information and the ratchet
effect. Rev Econ Stud 52:173–191
Georgiadis G (2015a) Projects and team dynamics. Rev Econ Stud 82(1):187–218
Georgiadis G (2015b) Deadlines and infrequent monitoring in the dynamic provision of public
goods. Working paper
Georgiadis G, Lippman SA, Tang CS (2014) Project design with limited commitment and teams.
RAND J Econ 45(3):598–623
Grossman SJ, Hart OD (1986) The cost and benefits of ownership: a theory of vertical and lateral
integration. J Polit Econ 94:691–719
Harvard Business School (2004) Managing teams: forming a team that makes a difference. Harvard
Business School Press, Boston
Ichniowski C, Shaw K (2003) Beyond incentive pay: insiders’ estimates of the value of comple-
mentary human resource management practices. J Econ Perspect 17(1):155–180
Kessing SG (2007) Strategic complementarity in the dynamic private provision of a discrete public
good. J Public Econ Theory 9:699–710
Laffont J-J, Tirole J (1988) The dynamics of incentive contracts. Econometrica 56:1153–1175
Lakhani KR, Carlile PR (2012) Myelin repair foundation: accelerating drug discovery through
collaboration. HBS Case No. 9-610-074, Harvard Business School, Boston
5 Project Contracting Strategies for Managing Team Dynamics 105

Lawler EE, Mohrman SA, Benson G (2001) Organizing for high performance: employee involve-
ment, TQM, reengineering, and knowledge management in the fortune 1000. Jossey-Bass,
San Francisco
Marx LM, Matthews SA (2000) Dynamic voluntary contribution to a public project. Rev Econ
Stud 67:327–358
Olson M (1965) The logic of collective action: public goods and the theory of groups. Harvard
University Press, Cambridge
Sannikov Y (2008) A continuous-time version of the principal-agent problem. Rev Econ Stud
75:957–984
Tirole J (1999) Incomplete contracts: where do we stand? Econometrica 67:741–781
Wired Magazine (2004) http://www.wired.com/gadgets/mac/news/2004/07/64286? Accessed 28
June 2015
Yildirim H (2006) Getting the ball rolling: voluntary contributions to a large scale public project.
J Public Econ Theory 8:503–528
Chapter 6
Time-Related Incentive Contracts for Managing
Projects with Uncertain Completion Time

Christopher S. Tang, Kairen Zhang, and Sean X. Zhou

Abstract Due to contractor’s unintentional and intentional poor planning, project

delays are common. When project delays affect the manager’s payoff negatively,
what kind of incentive contracts should the manager offer especially when she does
not have perfect information about the amount of work involved or the contractor’s
cost structure? In this chapter, we evaluate two commonly used time-related incen-
tive contracts (C1 and C2 contracts) when the manager conducts a reverse auction
to select the contractor for a project. Under the C1 contract, the contractor with the
lowest bid price wins; however, the manager imposes a linear and symmetric in-
centive/disincentive for early/late completion. Under the C2 contract, the winning
contractor has the lowest composite score that is based on the quoted price and the
quoted due date; meanwhile, the contractor is subject to a late completion penalty.
Our analysis reveals that, unless the project is truly urgent and the winning contrac-
tor can set his work rate after observing actual workload, the more complicated C2
contract adds no value to the manager: the simple C1 contract will suffice.

C.S. Tang ()

UCLA Anderson School of Management, 110 Westwood Plaza, Los Angeles,
CA 90095-1481, USA
e-mail: [email protected]
K. Zhang
Department of Systems Engineering and Engineering Management, The Chinese
University of Hong Kong, Shatin, NT, Hong Kong, China
e-mail: [email protected]
S.X. Zhou
Department of Decision Sciences and Managerial Economics, CUHK Business School,
The Chinese University of Hong Kong, Shatin, NT, Hong Kong, China
e-mail: [email protected]

The revision(s) of this paper were prepared during a sabbatical leave at the Haas Business School
of University of California, Berkeley.

© Springer International Publishing Switzerland 2017 107

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 6
108 C.S. Tang et al.

6.1 Introduction

Project delays happen all the time. For example, Boeing experienced a series of
delays (4 years) and a $6 billion cost overrun when managing the 787 Dreamliner
development project. The delays associated with this complex development project
were expected due to many uncertain elements such as unproven technologies (due
to the use of composite materials), unprecedented outsourcing of design (in addition
to the traditional outsourcing of manufacturing), untested multi-tier supply chain (as
opposed to the traditional one-tier supply chain), and unprecedented risk-sharing
contract.1 Not only large and complex projects face major delays, well-defined
projects such as construction projects can experience long delays. For example,
Al-Momani (2000) examined 130 public construction projects (such as residential
or office buildings, school buildings, medical centers, etc.) in Jordan. Out of these
130 projects, 106 projects were completed late, for various reasons, ranging from
poor design, poor planning, orders change (by customers), etc.
What causes project delays? Unforeseen events such as natural/man-made dis-
asters are certainly a key factor, but contractor’s poor planning is another common
cause. Contractor’s poor planning can be unintentional or intentional. Unintentional
poor planning is usually due to contractor’s inexperience or poor execution: fail
to schedule project tasks properly, fail to communicate/coordinate properly with
his vendors/crew members, etc. At the same time, intentional poor planning can
take place when the contractor does not have the right incentive to exert effort in
screening the capability of his vendors/crew members or in coordinating different
requisite tasks properly. Through screening (i.e., references, reputation, etc.), the
manager may be able to assess the contractor’s (unintentional) planning skill. How-
ever, the contractor’s (intentional) poor planning is difficult to detect because it is
project-specific and contract-specific. For this reason, the manager needs to develop
incentive contracts to entice the contractor to exert the right amount of effort. What
kind of incentive contracts should the manager offer especially when she does not
have perfect information about the amount of work involved or the contractor’s cost
structure?
In this chapter, we examine two types of incentive contracts which are commonly
observed in practice when the manager conducts a reverse auction to select the con-
tractor for a project. For the first type, the manager offers payment to the winning
contractor that takes the form of p − α (T − d), where p is the price (i.e., the winning
bid) submitted by the winning contractor whereas α ≥ 0 is the daily penalty/bonus
rate and d is the due date (or deadline), both of which are selected by the project
manager. Observe that the due date d is redundant because the term α d can be ab-
sorbed into the price p. Hence, the first type of incentive contracts can be reduced
to a simpler form: the manager only chooses the daily penalty rate α ≥ 0 and the

1 Under the risk-sharing contract, each supplier will receive his payment only after all suppliers
have completed their development tasks. The reader is referred to Tang and Zimmerman (2009)
and Kwon et al. (2010) for a detailed description of the Boeing 787 development project and the
analysis that examines the impact of risk-sharing contracts on the project completion time.
6 Time-Related Incentive Contracts 109

payment to the winning contractor is p − α T . We shall refer to this type of con-

tracts as C1 contract because each bid has only 1 component p. For example, after
the Northridge earthquake in 1994, the City of Los Angeles (the manager) wanted
to repair the Santa Monica Freeway within d = 180 days and offered an incentive
(penalty) of α = $200,000 per day for early (late) completion. In this case, the
payment takes the form of p − 200,000(T − 180) = (p + 36,000,000) − 200,000T ,
where p is the lowest bid submitted by the winning contractor and T is the (actual)
project completion time that is uncertain ex ante (Boarnet 1998). For the second
type of contracts, which we shall refer to as C2 contract, each contractor submits a
bid with two components (p, d), where p is the quoted price and d is the quoted due
date. Under a C2 contract, the project is awarded to the contractor whose bid has the
lowest composite score s = p + ρ d, where ρ is the weight on quoted due date. Upon
completing the project at time T , the winning contractor receives a payment that
takes the form of p − β (T − d)+ , where the second term represents the penalty for
late completion and β is the late penalty rate. Under C2 contract, the manager se-
lects (β , ρ ) ≥ 0. For example, C2 contract is used for state highway repair projects
under which the winning contractor is subject to a late penalty (Lewis and Bajari
2011).
While both C1 and C2 contracts have been adopted in practice, these two con-
tracts are different in the following sense. First, from the contractors’ perspectives,
each bid involves only the quoted price p under a C1 contract; however, under a
C2 contract, each bid must contain the quoted price p and the quoted due date d.
Second, from the project manager’s perspective, she needs to determine a single de-
cision variable α under a C1 contract, but she needs to make two decisions under a
C2 contract: (a) ρ for specifying the scoring process; and (b) β for specifying the
penalty. Hence, each entity makes one decision under a C1 contract, but each party
makes two decisions under a C2 contract. However, one can view C1 contract as
special cases of C2 contract. To elaborate, consider the case when the manager sets
a C2 contract (β , ρ ) that has ρ = ∞ and β = α . In this case, it can be seen that each
contractor will bid d = 0 under the C2 contract. Consequently, the winning bid is the
one with the lowest p, and the winning contractor will receive a payment of p − α T .
Hence, C2 contract should (weakly) dominate C1 contract. However, it is unclear if
and/or when C2 contract will strictly dominate C1 contract from the manager’s per-
spective. Also, if a contract that involves more decision variables is more beneficial,
we wonder if a contract that combines the features of both types of contracts will
lead to an even better outcome for the manager. Under this “combined” contract,
all interested contractors still submit their bids (p, d) and the bid with the lowest
score s = p + ρ d is the winning bid. Then the winning contractor receives a total
payment that “combines” the penalty terms of both C1 and C2 contracts; namely,
p − α T − β (T − d)+ . Clearly, the combined contract involves more decisions: the
contractors need to decide on (p, d) and the project manager selects (α , β , ρ ).
In this chapter, we determine the contractors’ equilibrium bidding strategies, the
manager’s optimal contract terms, and their corresponding expected payoffs asso-
ciated with three types of project contracts: C1, C2, and combined. In addition, we
compare the equilibrium outcomes across different contracts. By noting that C1 con-
110 C.S. Tang et al.

tract is a special case of combined contract when β = ρ = 0 and that C2 contract is a

special case of combined contract when α = 0, it suffices for us to focus our analysis
on combined contract. As an initial attempt to analyze combined contract, we model
a situation in which a manager needs to manage a project with two features: (1) the
cost information of each contractor and the work rate selected by each contractor
are private (i.e., unobservable by the manager); and (2) the completion time T is
uncertain, where the uncertainty arises from the uncertain amount of workload that
is involved in the project. The value (social welfare or utility) of the project to the
manager V (T ) is decreasing in T . The payment to the winning contractor takes the
form of p − α T − β (T − d)+ , where the second term α T corresponds to the penalty
under C1 contract, and the third term β (T − d)+ corresponds to the penalty under
C2 contract. Upon announcing the payment terms, the manager invites two inde-
pendent contractors with private cost information to submit their bids (p, d).2 The
contract is awarded to the contractor with the lower composite score s = p+ ρ d. The
winning contractor then starts the project after which the actual project completion
time T realizes and the payoff for each party is settled. In our model, the comple-
tion time T depends on the work rate r selected by the winning contractor, where the
work rate r cannot be adjusted once selected. This assumption is reasonable in many
practical settings. Also, we assume that the work rate r is not contractible (e.g., r is
not observable and/or verifiable by the project manager).
Our equilibrium analysis reveals that the combined contract is redundant. Hence,
using all three parameters (α , β , ρ ) > 0 does not provide additional benefit for the
project manager. More importantly, the optimal C1 contract and the optimal C2 con-
tract yield the same expected payoff and expected completion time: the simple C1
contract will suffice. However, based on a specific extension of our base model, we
find that the combined contract dominates C2 contract, and C2 contract (strictly)
dominates C1 contract when the following conditions hold: (a) the winning contrac-
tor can fully (or partially) observe the actual workload before determining his work
rate; and (b) the project is urgent; i.e., the project’s value V (T ) takes the form of
a − bT m with m > 1.
Our results complement a recent paper by Gupta et al. (2015) who study a more
general form of C2 contract. Under C2 contract, they consider a project procurement
problem arising from a multi-dimensional bidding process. While our C2 contract is
a special case of theirs, we have a different intent and a different focus. First, Gupta
et al. (2015) focus on the analysis of C2 contract from the contractors’ perspectives,
while we focus on the comparison between C1 and C2 contracts from the manager’s
perspective. Second, we consider a simpler version of C2 contract so that we can
derive closed form expressions for the equilibrium solutions and payoffs for both the
manager and the contractors. These closed form expressions enable us to compare
the expected payoffs associated with C1 and C2 contracts analytically. One of our
important findings is that, in many instances, having more decision variables for
the manager does not bring additional benefit: the simple C1 contract is sufficient.

2 Our model can be extended to the case of n > 2 contractors but the analysis becomes more
complex.
6 Time-Related Incentive Contracts 111

However, when the project is urgent (formally, V (T ) = a − bT m with m > 1) and

the winning contractor determines his work rate after observing actual workload,
we find that C2 contract dominates C1 contract.

6.2 Model Preliminaries

A project manager invites two contractors to bid for a project with uncertain com-
pletion time T ,3 where T depends on the random workload W associated with the
project as well as the work rate r selected by the winning contractor so that T = W /r.
Moreover, all players in our model know that W follows a probability distribution
Φ (·) with support [l1 , l2 ] (0 < l1 < l2 ≤ ∞) and mean μ . Let Φ̄ (·) and φ (·) denote
the complementary cumulative distribution function of W and the probability den-
sity function of W , respectively. In addition, we assume that the work rate r is neither
observable nor contractable by the project manager. If the winning contractor sets
work rate equal to r, his operation cost equals kr2 /2 per unit of time, where the cost
factor k is his private information. The cost factors of contractors, K1 and K2 , are
independently drawn from the same probability distribution F(·) with support [k, k̄]
(we shall drop the subscript i of Ki whenever it is irrelevant). This assumption is
standard in the auction literature (Krishnan 2002).
If the project is completed at time T , its value to the manager is V (T ) and the
manager’s payment to the winning contractor under the combined contract (α , β , ρ )
is p − α T − β (T − d)+ , where (p, d) is the winning bid. Throughout most of the
paper, the value function of the project V (·) is assumed to be decreasing and differ-
entiable. Thus, when the contractor with cost factor k wins the contract with a bid
(p, d) and completes the project at time T , the manager’s payoff is

Π (α , β , ρ ) = V (T ) − [p − α T − β (T − d)+ ], (6.1)

and the winning contractor’s payoff is

1
π (d, p, r | k) = p − α T − β (T − d)+ − kr2 T
2
 +
W W 1
= p−α −β − d − krW. (6.2)
r r 2

The sequence of events is as follows. The manager announces the value of

(α , β , ρ ) associated with the combined contract, and each contractor submits his bid
(p, d). Then the manager calculates the score s = p + ρ d associated with each bid,
and the contract is awarded to the contractor with the lower score. Upon winning

3 We also analyzed the case where workload is deterministic in Tang et al. (2013) and the qualitative
results are the same.
112 C.S. Tang et al.

the contract, the contractor starts the project according to work rate r. Finally, the
amount of work W is realized (so is the completion time) and the payoff to each
party is settled upon the completion of the project.

6.3 The Combined Contract

Our analysis is based on backward induction as follows. First, for any given com-
bined contract (α , β , ρ ), we determine each contractor’s bidding strategy as well as
work rate decision in the equilibrium of the associated subgame. Then, we deter-
mine the manager’s choice of contract terms in equilibrium.

6.3.1 Bidding Equilibrium

Under a given combined contract (α , β , ρ ), the bidding strategy of a contractor is a

quoted price p(·) and a quoted due date d(·) given his cost factor k ∈ [k, k̄]. Since a
bid (p, d) is evaluated according to s = p + ρ d, bidding (p, d) is equivalent to bid-
ding the composite score s with optimally chosen (p, d). Hence, we can focus our
analysis on s(·), and the corresponding (p(·), d(·)) could be retrieved via maximiz-
ing the contractor’s payoff subject to p(k) + ρ d(k) = s(k) for k ∈ [k, k̄].
Given his competitor’s strategy s̃(·), a contractor who bids s with cost factor
k will win the contract with probability P(s ≤ s̃(K)) and his expected payoff is
maxd,p,r E {π (d, p, r | k)} s.t. p + ρ d = s, d ≥ 0, r > 0. Thus, his optimal bidding
score is the solution to the following problem:
 
max max E {π (d, p, r | k)} P(s ≤ s̃(K)) s.t. p + ρ d = s, d ≥ 0, r > 0 . (6.3)
s d,p,r

Given s, it suffices for the contractor to decide on either p or d in the second maxi-
mization. Hence, by substituting p = s − ρ d, we can simplify Problem (6.3) as:

max[s − χ (k | α , β , ρ )]P(s ≤ s̃(K)), (6.4)

s
  + 
W W 1
where χ (k | α , β , ρ ) = min E ρ d + α + β − d + krW . (6.5)
d≥0,r>0 r r 2
By viewing ρ d as the “implicit” cost associated with losing the auction (when quot-
ing a longer due date d) and by noting that the other terms are the disincentives and
the operation cost, we can treat χ (k | α , β , ρ ) as the contractor’s “effective” cost for
completing the project. Since χ (k | α , β , ρ ) is strictly increasing in k, the contractor
bids as if he were privately informed about χ (k | α , β , ρ ) instead of k. Therefore,
the bidding strategies, the contractors’ expected payoffs, and the manger’s expected
6 Time-Related Incentive Contracts 113

payoff in equilibrium are the same as those in a standard single-unit reverse auction
with the costs of bidders replaced by the contractors’ effective costs χ (k | α , β , ρ )
(see, Corollary 1 of Asker and Cantillon 2008 for a more general result).
According to Problem (6.4), the contractor determines his quoted due date and
work rate via solving Problem (6.5) independent of his bidding score as well as
his competitor’s bidding strategy s̃(·). This observation generates Proposition 1.
(All proofs are available in Tang et al. 2013.)
Proposition 1. Under the combined contract (α , β , ρ ), a contractor with cost factor
k sets his quoted due date d(k | α , β , ρ ) in equilibrium that satisfies:
⎧    −1/2

⎨√ −1 ρ
⎪ β ∞
kΦ̄ 2 α+ wφ (w) dw , if ρ < β ,
d(k | α , β , ρ ) = β μ Φ̄ −1 (ρ /β )
⎪
⎩0, otherwise.
(6.6)
If he wins the contract, he will set his work rate r(k, α , β , ρ ) = θ (α , β , ρ )k−1/2
(r(k), for ease of exposition), where
⎧   
⎪
⎨ β ∞
2 α+ wφ (w) dw , if ρ < β ,
θ (α , β , ρ ) = μ Φ̄ −1 (ρ /β ) (6.7)
⎪
⎩
2(α + β ), otherwise.

Proposition 1 confirms our intuition: the contractor with a lower cost factor k
will submit a bid with a shorter due date and will work faster if he wins the con-
tract. Moreover, the proposition implies that the quoted due date d(k | α , β , ρ ) is
decreasing in α , ρ (it is increasing in β if φ (·) has an increasing failure rate) and
the contractor’s work rate r(k, α , β , ρ ) (equivalently, θ (α , β , ρ )) is increasing in α ,
β , ρ . In other words, if the manager imposes a heavier penalty rate or puts more
weight on due date in determining the composite score, the contractor will bid a
smaller quoted due date and work faster; however, if the manager imposes a heavier
late penalty rate β , the contractor will bid a greater quoted due date and work faster.
Moreover, Eq. (6.7) shows that the late penalty rate β essentially has the same
effect as the daily penalty rate α on the contractor’s work rate, i.e., the change of
θ via adjustment of α can always be attained by adjustment of β . This is not very
intuitive at the first glance since the penalty corresponding to β is effective only after
the project lasts beyond the contractor’s quoted due date while that corresponding to
α is effective once the project starts. The reason is that the quoted due date is chosen
by the contractor himself under the pressure of lowering his score in the auction. If
the contractor works slower and hence lengthens his project completion time, he
can avoid the increase in the penalty corresponding to β via increasing his quoted
due date by some Δ d. However, he has to cut his price by ρΔ d so as to maintain
the same bidding score as before. This essentially amounts to a penalty for each day
that the project lasts.
Why is it the case d(k | α , β , ρ ) = 0 when ρ ≥ β ? If ρ ≥ β , the reduction in
price (p = s − ρ d) due to bidding a longer quoted due date will be higher than
114 C.S. Tang et al.

the increase in the penalty corresponding to β for any given work rate. Thus, the
contractor would rather bid a quoted due date equal to zero and incur the penalty
corresponding to β from the start of the project than bid a positive quoted due date
and avoid the penalty within the period specified by his quoted due date.
Next, we examine the impact of uncertain workload W on the contractor’s work
rate r(k, α , β , ρ ). Consider the case when W ∼ U[μ − x, μ + x] (0 < x < μ ) so that
E(W ) = μ and the level of uncertainty is captured by x (the variance of W is x2 /3).
By using the expression of r(k, α , β , ρ ) given in Proposition 1, we obtain the fol-
lowing result.
Corollary 1. When W ∼ U[μ − x, μ + x], the contractor’s work rate r(k, α , β , ρ ) is
increasing in the workload uncertainty level x.
When W becomes more uncertain (i.e., as x increases), the above corollary re-
veals that the contractor will work faster. Consequently, the corresponding expected
completion time E[W /r(k, α , β , ρ )] will become shorter when W becomes more un-
certain. This is because higher workload uncertainty level increases the possibility
of incurring extra penalty for late completion and hence makes working slower less
desirable.
By substituting the quoted due date d(k | α , β , ρ ) and the work rate r(k, α , β , ρ )
(given in Proposition 1) into Problem (6.4), we can determine the bidding strategy
of each contractor s(· | α , β , ρ ) under the combined contract as follows.
Proposition 2. Under the combined contract (α , β , ρ ), each contractor adopts the
same strictly increasing and differentiable bidding strategy s(· | α , β , ρ ) in equilib-
rium, where s(· | α , β , ρ ) satisfies:
 k̄
1
s(k | α , β , ρ ) = χ (k | α , β , ρ ) + F̄(v)E[r(v, α , β , ρ )W ] dv, (6.8)
2F̄(k) k

for any k ∈ [k, k̄]. Moreover, a contractor with cost factor k has an expected payoff:
 k̄
1
π (k | α , β , ρ ) = F̄(v)E[r(v, α , β , ρ )W ] dv. (6.9)
2 k

Finally, the manager’s expected payoff satisfies:

  
W
Π (α , β , ρ ) = E V
r(K̂)
  k̄ 
1 1
− E K̂r(K̂)W + F̄(v)E {r(v)W } dv , (6.10)
2 F̄(K̂) K̂

where K̂ = min{K1 , K2 }.
Because the contractors adopt the same strictly increasing bidding strategy in
equilibrium, Proposition 2 implies that the contractor with the lowest cost factor
wins the contract. Moreover, Eq. (6.9) reveals that the lower the contractor’s cost
factor k is, the higher his expected payoff is. Finally, since both χ (k | α , β , ρ ) and
6 Time-Related Incentive Contracts 115

r(v, α , β , ρ ) are increasing in α , β , ρ , Eqs. (6.8) and (6.9) imply that each contrac-
tor will bid a higher score and earn a higher payoff if the manager imposes higher
penalty rates or puts more weight on due date in determining the score. The intuition
is that higher penalty rates or weight on due date dampen the competition between
the contractors. When α , β , or ρ increases, each contractor knows that, if he wins,
the difference between his “effective” cost and his competitor’s will be greater on
average, since the contractor with the higher cost factor suffers a greater increase in
“effective” cost. Since each contractor bids as if he were privately informed about
his “effective” cost, this knowledge shall induce each contractor to bid less aggres-
sively (i.e., to bid a higher composite score). Consequently, the winning contractor
obtains a higher expected payoff when α , β or ρ increases.
By noting that the C1 (C2) contract is a special case of the combined contract,
we obtain similar results under C1 (C2) contract by setting β = ρ = 0 (α = 0).
Specifically, under C1 contract, the winning contractor works faster and obtains a
higher expected payoff if α increases; under C2 contract, the winning contractor
works faster and obtains a higher expected payoff if β or ρ increases.

6.3.2 Optimal Combined Contracts

By using the analysis of the bidding equilibrium in the previous section, we now
determine the combined contract of the manager’s choice in equilibrium, i.e., the
optimal combined contract(s).
In anticipation of the bidding equilibrium, the manager’s problem could be for-
mulated as:
max Π (α , β , ρ ), (6.11)
(α ,β ,ρ )≥0

where Π (α , β , ρ ) is given in Eq. (6.10). Since Π (α , β , ρ ) solely depends on (α , β , ρ )

through the contractors’ work rate decisions r(·, α , β , ρ ), two different combined
contracts yield the same payoff for the manager as long as they induce the same
work rate decision. Combining this observation with the equivalence of α and β in
terms of their impacts on r(·, α , β , ρ ) [see, Eq. (6.7)], we can readily establish the
following result.

Corollary 2. Combined contract is redundant: it is sufficient for the manager to

consider offering C1 contract (with one parameter α ) or C2 contract (with two pa-
rameters β and ρ ). Moreover, the optimal C2 contract and the optimal C1 contract
yield the same expected payoff for the manager and for the winning contractor.

Intuitively, the result stated in Corollary 2 is due to the following reason: the
only information advantage that a contractor possesses over the manager is his cost
factor. Therefore, to overcome her information disadvantage, it is sufficient for the
manager to use a single parameter α under C1 contract.
116 C.S. Tang et al.

Next, we examine the impact of the project’s time sensitivity on optimal com-
bined contracts via considering a specific form of V (·), i.e. V (T ) = a−bT (a, b > 0).
Plugging V (T ) = a − bT into Problem (6.11) and solving the derived problem, we
obtain the following result.

Corollary 3. If V (T ) = a − bT , a combined contract (α , β , ρ ) is optimal if and only

if it solves θ (α , β , ρ ) = {(1/κ )bE[K̂ 1/2 ]}1/2 , where θ (α , β , ρ ) is given in Eq. (6.7)
and   k̄ 
1 1
κ= E K̂ + v−1/2 F̄(v) dv . (6.12)
2 F̄(K̂) K̂

Since θ (α , β , ρ ) is increasing in α , β , and ρ , Corollary 3 implies that, when the

project becomes more time-sensitive (i.e., b increases), the manager will set a greater
(α , β , ρ ) so as to induce the winning contractor to complete the project earlier. In
particular, the manager will set a higher α ((β , ρ )) under C1 (C2) contract if the
project becomes more time-sensitive.
In summary, we have determined the equilibrium bidding strategy of each con-
tractor under any given combined contract (or under any given C1 (C2) contract as
a special case). Also, we have shown that, regardless of the functional form of V (·),
the combined contract is redundant, and the optimal C1 contract and the optimal C2
contract generate the same expected payoff for the manager and for the winning con-
tractor. Essentially, the manager does not gain any additional benefit by considering
more incentive parameters: it suffices to offer the simple C1 contract.

6.4 An Extension: Delaying Work Rate Decision

To examine the robustness of our results, we now extend our base model to the case
when the winning contractor can delay his work rate decision in the following sense.
Even though each contractor can develop a tentative plan for his work rate, he does
not need to commit his actual work rate until he fully observes the actual workload
W after the auction.4 This setting can happen in practice, because the contractor may
conduct more due diligence, e.g., independent tests, market information acquisition,
“after” winning the contract, through which he can know extra information about
the workload. For example, in deep sea oil exploration, the winning contractor may
conduct additional seismic tests before he commits to a specific work rate.
In this extension, the modified sequence of events goes as follows. The manager
first announces (α , β , ρ ) associated with the combined contract. Then, each con-
tractor submits his bid (p, d) based on his estimation of W and the bid with the
lowest score s = p + ρ d is the winning bid. After that, the winning contractor ob-
serves the actual workload W , which is not observable to the manager, and then

4 While we present the case when the contractor can fully observe the actual workload W for
ease of exposition, our main results in this section continue to hold even when the contractor only
partially observes the actual workload W before committing his work rate.
6 Time-Related Incentive Contracts 117

selects the work rate according to the realized workload. Hence, unlike the previous
analysis, the work rate to be committed depends on the realized workload W and
so the winning contractor has an additional information advantage (i.e., the actual
workload W ) over the manager. Finally, both the manager and the contractor receive
their payoffs when the project is completed.

6.4.1 Bidding Equilibrium

In this section, we determine the bidding equilibrium of the subgame associated

with a given combined contract (α , β , ρ ).
Analogous to Sect. 6.3.1, it suffices to focus our analysis on how each contrac-
tor sets his composite score s(·) based on his cost factor k, and then retrieve the
corresponding bidding strategy (p(k), d(k)) via maximizing the contractor’s payoff
subject to p(k) + ρ d(k) = s(k) ≡ s for k ∈ [k, k̄].
Given his competitor’s strategy s̃(·), a contractor who has a cost factor k and
bids s will win the contract with probability P(s ≤ s̃(K)) and his expected payoff is
maxd,p E{maxr>0 π (d, p, r | k)} s.t. p + ρ d = s, d ≥ 0, where π (d, p, r | k) is given
in Eq. (6.2). Different from Sect. 6.3.1, in the above formula of the contractor’s ex-
pected payoff the optimization of work rate is within the expectation operator since
the contractor, if he wins, sets work rate after observing the realized workload. Then,
his optimal bidding score is the solution to the following problem:
   
max max E max π (d, p, r | k) P(s ≤ s̃(K)) s.t. p + ρ d = s, d ≥ 0 . (6.13)
s d,p r>0

Replacing p with s − ρ d, we can simplify Problem (6.13) as:

max {[s − χ̌ (k | α , β , ρ )]P(s ≤ s̃(K))} , (6.14)

s

where
   + 
W W 1
χ̌ (k | α , β , ρ ) = min E min ρ d + α + β − d + krW . (6.15)
d≥0 r>0 r r 2

The only difference between Problems (6.14) and (6.4) lies in where the expectation
operator E(·) is applied in χ̌ (k | α , β , ρ ) and χ (k | α , β , ρ ). Therefore, we can apply
the same approach as that in Sect. 6.3.1 so as to establish the bidding equilibrium
below. (For ease of reference, we use “x̌” to denote any quantity x associated with the
case when the winning contractor can set his work rate after observing the realized
workload.)
Proposition 3. If the winning contractor can delay his work rate decision after
observing actual workload, a contractor with cost factor k will set his work rate
ř(w | k, α , β , ρ ) to θ̌ (w | α , β , ρ ) · k−1/2 after winning the contract and observing
W = w, where
118 C.S. Tang et al.

θ̌ (w | α , β , ρ )
⎧    
⎨ 2(α + β ) max min 1, W α
, , if ρ < β ,
= h−1 ((α + ρ )/(α + β )) α +β
⎩
2(α + β ), otherwise,
(6.16)

and 
α 1
h(x) = +2 √ uΦ̄ (xu) du
α +β α /(α +β )

ˇ | α , β , ρ ) as
for any x > 0. Moreover, he sets his quoted due date d(k
⎧  
⎪
⎨ k −1 α + ρ
ˇ | α, β , ρ) = h , if ρ < β ,
d(k 2(α + β ) α +β (6.17)
⎪
⎩
0, otherwise,

and bids a composite score that satisfies:

 k̄
1
š(k | α , β , ρ ) = χ̌ (k | α , β , ρ ) + F̄(v)E[ř(W | v)W ] dv, (6.18)
2F̄(k) k

which is differentiable and strictly increasing in k. Finally, the winning contractor’s

expected payoff satisfies:
 k̄
1
π̌ (k | α , β , ρ ) = F̄(v)E[ř(W | v)W ] dv, (6.19)
2 k

and the manager’s expected payoff satisfies:

  
W
Π̌ (α , β , ρ ) = E V
ř(W | K̂)
  k̄ 
1 1
− E K̂ ř(W | K̂)W + F̄(v)E{ř(W | v)W } dv , (6.20)
2 F̄(K̂) K̂

where K̂ = min(K1 , K2 ).
Proposition 3 resembles Propositions 1 and 2. Specifically, all results in Propo-
sition 3 except the quoted due date follow from replacing r and χ with ř and χ̌
respectively in the corresponding results in Propositions 1 and 2. Thus, most prop-
erties of the bidding equilibrium presented in Sect. 6.3.1 (for example, the contractor
with a lower cost factor bids a smaller composite score, sets a smaller quoted due
date and works faster after winning the contract) still hold even if the winning con-
tractor can delay his work rate decision after observing the actual workload, and we
shall single out those that do not hold in the following.
First, Eq. (6.16) demonstrates that the late penalty rate β no longer has the same
impact as the incentive/penalty rate α on the work rate decision ř(· | k, α , β , ρ ).
6 Time-Related Incentive Contracts 119

Specifically, if ρ < β , the extra late penalty corresponding to β deters the winning
contractor from completing the project late and therefore induces him to complete
on time (i.e. W /ř(W | k) = d(k))
ˇ when he observes an intermediary workload, i.e.
    
α −1 α + ρ −1 α + ρ
h <W <h .
α +β α +β α +β

In contrast, without the late penalty rate β (i.e., β = 0), the winning contractor works
at the same rate irrespective of the realized workload and hence never completes the
project in the same amount of time under two different workload realizations.
Second, Eq. (6.16) shows that the contractor may work slower if the manager
imposes a heavier late penalty rate β . The intuition is as follows. Note that the
contractors tend to bid larger quoted due dates to avoid the penalty when β is large.
Consequently, whenever the winning contractor observes a moderate workload, he
will work slower as long as he can complete the project on time and avoid the
penalty.
Finally, the possibility of a lower work rate for a higher β and Eq. (6.19) imply
that a higher late penalty rate β may decrease the contractor’s expected payoff.

6.4.2 Optimal Combined Contracts

In anticipation of the bidding equilibrium, the manager determines the combined

contract to offer via solving the following problem:

max Π̌ (α , β , ρ ), (6.21)
(α ,β ,ρ )≥0

where Π̌ (α , β , ρ ) is given in Eq. (6.20).

First, let us consider the case when the project’s value to the manager drops at a
constant rate as the project progresses, i.e. V (T ) = a − bT (a, b > 0). In practice, it
corresponds to managing non-urgent projects such as regular highway maintenance.
We obtain the following result via solving Problem (6.21).
Proposition 4. When V (T ) = a − bT , the combined contract is redundant. Specif-
ically, the optimal combined contract, the optimal C2 contract and the optimal C1
contract all yield the same expected payoff for the manager.
Proposition 4 shows that, if the project is not urgent, our result about the equiv-
alence among combined contract, C1 contract and C2 contract when the winning
contractor sets work rate before observing the actual workload continues to hold
even when he can defer his work rate decision till observing the actual workload.
Will such result continue to hold if the project is urgent, i.e., the project’s value
to the manager drops at an increasing rate as the project progresses? To do so, let us
120 C.S. Tang et al.

consider the case when V (T ) = a − bT m with m > 1. By plugging V (T ) = a − bT m

into Problem (6.21) and solving the derived problem, we can characterize the opti-
mal combined contract as follows.

Proposition 5. When V (T ) = a − bT m with m > 1, C1 contract is dominated by

C2 and combined contracts. Also, any optimal combined contract (α ∗ , β ∗ , ρ ∗ ) must
satisfy β ∗ > ρ ∗ > 0.

Proposition 5 offers a scenario (i.e., urgent project and delayed work rate deci-
sion) under which the manager could be better off by imposing a contract with more
incentive parameters. The intuition behind could be explained as follows. When the
project is urgent, the manager would like to induce the winning contractor to work
faster upon observing a heavier workload so as to prevent the loss of the project’s
value from becoming exceedingly high. She can achieve this via employing a C2
contract or a combined contract with β > ρ > 0 since ř(w | k, α , β , ρ ) is increas-
ing in w; however, she cannot achieve this via employing any C1 contract since
ř(w | k, α , 0, 0) is independent of w (see, Proposition 3). Thus, combined contract
and C2 contract dominate C1 contract when the project is urgent.
We now compare the expected payoff of the manager and the expected pay-
off of the contractor under all three types of optimal contracts for the case when
V (T ) = a − bT m with m > 1. To facilitate the comparison of these payoffs, let Π̌C∗ ,
∗ , and Π̌ ∗ denote the maximal expected payoff of the manager under combined,
Π̌C2 C1
C2 and C1 contracts, respectively. Similarly, we let π̌C∗ (k), π̌C2
∗ (k) and π̌ ∗ (k) be
C1
the expected payoffs of a contractor with cost factor k under the optimal combined
contract, the optimal C2 contract and the optimal C1 contract, respectively. Maxi-
mizing the manager’s payoff Π̌ (α , β , ρ ) within each type of contracts and retrieving
the corresponding contractor’s payoffs via Eq. (6.19), we can establish the following
corollary.

Corollary 4. When V (T ) = a − bT m with m > 1, the manager earns the highest

payoff under the optimal combined contract and earns the lowest payoff under the
optimal C1 contract (i.e., Π̌C∗ ≥ Π̌C2
∗ > Π̌ ∗ ). On the contrary, the contractor earns
C1
the highest payoff under the optimal C1 contract and earns the lowest payoff under
the optimal combined contract (i.e., π̌C∗ (k) ≤ π̌C2
∗ (k) < π̌ ∗ (k)). Moreover, the con-
C1
tractor’s equilibrium payoff π̌i (k) and the manager’s optimal payoff Π̌i∗ associated
∗

with contract type i satisfy:

 k̄
m
π̌i∗ (k) = v−1/2 F̄(v) dv · (a − Π̌i∗ ), for i = C,C2,C1. (6.22)
2κ (m + 1) k

According to Corollary 4, when the project is urgent, the manager prefers com-
bined contract over C2 contract, and strictly prefers C2 contract over C1 contract.
However, the contractors’ preference over these three types of contracts is the oppo-
site. This tension between the manager and the contractor is captured by Eq. (6.22),
where the contractor’s payoff decreases as the manager’s payoff increases.
6 Time-Related Incentive Contracts 121

In summary, when the winning contractor can delay his work rate decision and
the project is urgent, the manager can be better off by imposing a contract with
more incentive parameters. In this case, combined contract dominates both C1 and
C2 contracts.

6.5 Conclusion

When project completion time is an important factor, it is common for the project
manager to provide incentives to contractors for early completion. In this chapter,
we have analyzed three different types of contracts. The first two types (C1 and C2
contracts) are observed in practice, while the third type (combined contract which
combines both C1 and C2 contracts) is new. By considering a reverse auction in
which the winning contractor has the lowest composite score, we have determined
the optimal bid in equilibrium and characterized the optimal combined contract in
closed form expressions. More importantly, we have shown that, when the winning
contractor determines work rate before observing actual workload, combined con-
tract is redundant. Specifically, both the optimal C2 contract and the optimal C1
contract yield the same expected payoff and the same expected completion time for
the project manager. Hence, simple C1 contract will suffice.
To examine the robustness of our result, we extend our model to the case when
the winning contractor can fully (or partially) observe the actual workload before he
commits his work rate. We find that our result continues to hold when the value of
the project is linearly decreasing in the completion time (i.e., non-urgent project).
However, when the value of the project is exponentially decreasing in the comple-
tion time (i.e., urgent project), combined contract may dominate C2 contract, and
C2 contract may strictly dominate C1 contract.
As stated at the beginning of this chapter, the literature on project contracts re-
mains scant and there are plenty opportunities to explore further. For instance, we
have assumed that the contractors’ cost factors are not known to the project man-
ager. However, it is of interest to examine the implications of having the manager
to request contractors to reveal their cost factors and the winning contractor may be
subject to an audit. Also, we have assumed that the work rate remains fixed once
committed. It is of interest to examine the case when the contractor can adjust his
work rate as the project progresses. This is especially interesting when the actual
workload W involves unexpected events, setbacks and surprises that may happen in
highly uncertain projects such as new drug development projects. Finally, we have
only examined the case when the project is a single task that can be done by a single
contractor. It would be of interest to examine the case when the project involves sub-
stitutable and/or complementary efforts of multiple contractors who perform their
tasks simultaneously.
122 C.S. Tang et al.

References

Al-Momani A (2000) Construction delay: a quantitative analysis. Int J Proj Manag 18(1):51–59
Asker J, Cantillon E (2008) Properties of scoring auctions. Rand J Econ 39(1):69–85
Boarnet MG (1998) Business losses, transportation damage, and the Northridge earthquake.
J Transp Stat 1(2):49–63
Gupta D, Snir EM, Chen Y (2015) Contractors’ and agency decisions and policy implications in
A+B bidding. Prod Oper Manag 24(1):159–177
Krishnan A (2002) Auction theory. Academic, New York
Kwon D, Lippman SA, McCardle KF, Tang CS (2010) Project management contracts with delayed
payments. Manuf Serv Oper Manag 12(4):692–707
Lewis G, Bajari P (2011) Procurement contracting with time incentives: theory and practice.
Q J Econ 126:1173–1211
Tang CS, Zimmerman JD (2009) Managing new product development and supply chain risks: the
Boeing 787 case. Supply Chain Forum Int J 10(2):74–85
Tang CS, Zhang KR, Zhou S (2013) Evaluating project contracts with time-related incentives.
Unpublished Manuscript, UCLA Anderson School, 2013
Chapter 7
Contracting for New Product Development

Sameer Hasija and Shantanu Bhattacharya

Abstract In this chapter, we outline the findings of the research on contracting in

new product development from the perspective of mitigating agency issues in both
the internal and external contexts to the firm. The literature has studied the role
of contracting in aligning the incentives of both internal stakeholders and external
partners, and the nature of decisions being made by the principal. A common set of
agency issues studied in the literature is the use of contracts for coordinating single
or multiple efforts to attain higher profits for the supply chain, and the division of
profits between the different partners in the supply chain. Additionally, the mitiga-
tion of information asymmetry effects, the decision rights for exercising contracts,
group and individual incentives, different contracting levers and monitoring of agent
efforts have also been considered. Finally, simultaneous versus sequential decision-
making, the use of informal contracts in addition to formal contracts, and the impact
of risk profiles have been studied. We also identify critical gaps in the literature and
propose directions for future research.

7.1 Introduction

The area of new product development has traditionally been studied from the per-
spective of the focal firm investing in the development of new products and services
(Krishnan and Ulrich 2001). This stream of literature largely focused on product

S. Hasija ()
INSEAD, 1 Ayer Rajah Avenue, Singapore 138676, Singapore
e-mail: [email protected]
S. Bhattacharya
Lee Kong Chian School of Business, Singapore Management University, 50 Stamford Road,
Singapore 178899, Singapore
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 123

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 7
124 S. Hasija and S. Bhattacharya

development decision-making at the product level and at the organizational level to

aid decision-making in the areas of concept development, product design, process
design, and production ramp-up and launch. However, there has been an increasing
recognition recently that (1) new product development (NPD) decisions in a fo-
cal firm are taken by multiple players, and (2) product development activities span
firm boundaries in many cases, and different firms may be responsible for the prod-
uct prototyping (concept development and product design activities) and large-scale
manufacturing (process design and production activities).
Given the involvement of different parties in NPD activities, a number of agency
issues can be identified that can make the management of the new product develop-
ment process and the associated decision-making difficult. For instance, if a focal
firm is developing a new product or service, the firm may have a certain goal (e.g.,
profits from the new product). However, if the NPD program manager is provided
incentives for the number of new product launches in that year, s/he may prefer
allocating resources to the development of products that are incremental but can
be developed quickly and with a higher percentage of success, which would give
him/her a better compensation. Hence, incentivizing managers to align their objec-
tive with the firm is an important agency issue. Similarly, if two different firms are
collaborating in developing a new product, if the outcome of the NPD process can-
not be verifiably separated in the efforts/investments made by the firms, then the
firms have an incentive to free-ride on the other firm’s effort, and both partners may
under-invest in the NPD project. Hence, providing the right set of incentives to both
firms so that both of them exert system-optimal efforts is an important issue for
research. The existence of these agency issues has led to a nascent but important
stream of literature that investigates the design of contracts to align the incentives
of managers internally and partner firms externally. Also, the objectives of the firm
may not always be profit maximization; in the early stage of the product lifecycle,
firms may want to maximize their chances of survival, or the revenues generated by
a new product rather than the profits. The design of contracts then becomes critical
to the reduction or the elimination of the agency issues to manage the alignment of
the actions of internal decision-makers and external partners to the objectives of the
focal firms.
In this chapter, we take the approach of studying the literature on contracting in
new product development from the perspective of mitigating agency issues in both
the internal and external contexts. We divide the literature based on the primary par-
ties involved in the new product development process, and the nature of decisions
being made by the focal firm. Figure 7.1 outlines the classification of contracting
models in new product development that we will summarize in this paper. While
the literature on contracting for NPD has looked at multiple objectives and contrac-
tual structures, there are some common themes in this literature that we will use
for classifying the work done in this area. A common set of agency issues studied
in the literature is the use of contracts for coordinating single or multiple efforts
to attain higher profits for the supply chain, and the division of profits between the
different partners in the supply chain. Additionally, the literature has also studied
the mitigation of information asymmetry effects, the decision rights for exercising
7 Contracting for New Product Development 125

Internal External

Freelance Contractual
innovators collaborations

Single-party Multiple-party
effort effort

Fig. 7.1 Classification of literature on contracting for NPD

contracts, group and individual incentives, different contracting levers and moni-
toring of agent efforts. Finally, the literature has considered simultaneous versus
sequential decision-making, the use of informal contracts in addition to formal con-
tracts, and risk profiles. We will delineate the use of different kinds of contracts in
the different contexts that we study.
The chapter is structured as follows. In Sect. 7.1.1, we provide a general model
that may be used to capture a wide variety of issues in contracting for NPD. In
Sect. 7.2, we study the agency issues studied and the corresponding contracts in the
case where firms offer contracts internally to align the objectives of their managers
to the goals of the firm. Following that, in Sect. 7.3, we study the related agency
issues and proposed contracts in an external NPD setting, where firms work to-
gether in a supply chain with other firms. In Sect. 7.3.1, we discuss the issue of
collaborative efforts, where firms collaborate with other firms in an NPD setting. In
Sect. 7.3.1.1, we consider cases where only agents invest in the NPD effort, while in
Sect. 7.3.1.2, we study cases where the principal and agents invest in the NPD effort.
In Sect. 7.3.2, we discuss the outsourcing of NPD to freelancers. We conclude and
provide directions for future research in Sect. 7.4.

7.1.1 Notation and General Model

Broadly speaking, an NPD process can be considered to be a knowledge-intensive

service process. Therefore, like services, it has certain characteristics that may
make it different from other standard processes (e.g., manufacturing). Using the
126 S. Hasija and S. Bhattacharya

terminology commonly used to describe service operations, the following charac-

teristics are important in describing the NPD process: (1) co-production (multiple
parties involved in the process), (2) intangibility of efforts and some intermediate
outcomes (e.g., quality of an idea), (3) information asymmetry (due to high knowl-
edge specificity), and (4) a high degree of uncertainty. Therefore, analytical models
that aim at studying incentive design for NPD initiatives, should be able to capture
these characteristics in order to provide managerial insights on how optimal incen-
tive design varies with them. In this section we build a very general model that, we
believe, can capture a wide range of economic issues in the NPD incentive design
domain. We use the following notation in the chapter:
ΩPt information set of the principal at time t.
Ωat i information set of agent i at time t.
d tp decisions made by the principal at time t.
dat i decisions made by agent i at time t.
Π t (dtp , dta ) revenue (potentially stochastic) generated from the project at time
t, where dtp = {d 1p , d 2p , . . . , d tp }, and dta = {da11 , da21 , . . . , dat 1 , . . . , da1M ,
da2M , . . . , dat M }.
Θ t (dtp , dta ) vector of verifiable, hence contractible, signals (potentially stochastic)
at time t, where dtp = {d 1p , d 2p , . . . , d tp }, and dta = {da11 , da21 , . . . , dat 1 , . . . ,
da1M , da2M , . . . , dat M }.
Wati (Θ t ) contractual payoff of agent i at time t.
Ui (·) utility function of the agent i.
cti (dat i ) cost incurred by the agent i at time t by making decisions dat i .
ζ discount rate.

7.1.1.1 Contract Design Problem

The principal’s contract design problem aims at maximizing his/her surplus, i.e.,
choosing Wati (Θ t ) that maximizes ∑t=0
N
e−ζ t E[Π t (dtp , dta ) − ∑M
i=1 wai (Θ ) | Ω p ],
t t t

where the following constraints need to be considered:

1. The decisions are the equilibrium outcome of self-interested parties making de-
cisions, i.e., d τp and daτi are the equilibrium actions from maximizing
  
N M 
t  t
∑e −ζ (t−τ )
E Π (d p , da ) − ∑ wai (Θ )  Ω p
t t t t
and
t=τ i=1
N  
∑ e−ζ (t−τ ) E Ui (wtai (Θ t ) − ctai (dat i )) | Ωat i ,
t=τ

determined using backward induction starting with t = N. The principal’s inf-

ormation set, Ω pt , includes all priors on the information sets of the agents and
all future (exogenous and endogenous) outcomes, and the updated information
based on information revealed by the agents (via observable elements of dta ).
7 Contracting for New Product Development 127

The agents information set, Ωat i , includes all priors on the information sets of all
other agents and of the principal, priors on all future (exogenous and endogenous)
outcomes, and the updated information based on information revealed by other
agents.
2. The transfer payment to agent i in period t, wtai (Θ t ) may be equal to the one set
by the initial contract, Wati (Θ t ), or determined by an outcome of a renegotiation
process dependent on {Ω pt , Ωat i }, if doing so is Pareto efficient.
3. The individual rationality constraint of the agent needs to be satisfied. This may
be over the entire project horizon, i.e.,
N
∑ e−ζ t E[Ui (wtai (Θ t ) − ctai (dat i )) | Ωat i ] ≥ ū,
t=0

or may be at each decision epoch, i.e.,

N
∑ e−ζ (t−τ ) E[Ui (wtai (Θ t ) − ctai (dat i )) | Ωat i ] ≥ ūΩτ aτi ,
t=τ

τ
where ūΩ τ is the reservation utility of the agent at time τ given its current infor-
ai
mation set. Other constraints in individual rationality may also be imposed, such
as one period cash-flow constraints

E[Ui (wτai (Θ τ ) − cτai (daτi )) | Ωaτi ] ≥ ûΩ

τ
τ ,
ai

inter-temporal fairness
N
∑ e−ζ (t−τ ) E[Ui (wtai (Θ t ) − ctai (dat i ))|Ωat i ] ≥ ūτΩaτi−1 ,
t=τ

etc.
This general model can easily be reduced to capture a wide range of issues that
are relevant to the NPD process. The important characteristics of the NPD process
are easily captured by this model: (1) Π (·) is a joint function of decisions made by
different parties and can be modeled to represent efforts that are substitutes or com-
plements. It can also be made separable in efforts to represent cases that are more
modular in nature. (2) The verifiable signal, Θ (·) can be written as a stochastic map-
ping between decisions and some performance measures (again, may be separable
in efforts to represent modularity). (3) The information sets of different parties can
be different, thus representing information asymmetry. (4) Finally, uncertainty can
be easily represented in terms of the revenues and the verifiable signals.
128 S. Hasija and S. Bhattacharya

7.2 Internal Incentives for NPD

Managing an NPD project within a firm is inherently complex due to the involve-
ment of highly skilled employees in a knowledge-intensive, complex, and uncertain
process. These characteristics of an NPD process lead to incentive alignment issues
within an organization, where optimal incentives need to be designed to incentivize
employees to exert appropriate efforts during the process and reveal relevant infor-
mation truthfully, while accounting for the exogenous uncertainties in the process.
In this section, we look at such issues faced by a firm.
Within an organization, the successful completion of NPD initiatives requires
an appropriate delegation of corresponding resources. Moreover, managers have to
make allocation decisions based on the risk-reward trade-off of different projects
(e.g., between incremental and radical innovation projects). The senior management
of the firm may not have visibility of how managers of different business units allo-
cate resources between different types of NPD initiatives and hence, need to design
incentive schemes that appropriately incentivize managers of individual business
units to make decisions so that the overall portfolio of the firms NPD initiative is
balanced between these different types of NPD initiatives. Our general model can
be reduced to a setting where managers of individual business units make resource
allocation decisions between different NPD projects (incremental and new), and the
firm’s objective is to design the optimal incentives from its perspective. In addi-
tion, it is also possible to include decision on funding authority made by the firm,
wherein it may or may not give autonomy to the managers of the business unit
over NPD funding. Such a setting is modeled in Chao et al. (2009), where the
authors show that giving managers funding authority based on the turnover has dif-
ferent results depending on whether the funding authority is fixed or variable. While
variable funding dominates fixed funding in terms of incentive structure, variable
funding has a tendency to incentivize managers to focus on incremental innovation
at the expense of radical innovation. They also find that significant career concerns
for managers based on potential failure of NPD projects induces managers to under-
invest in innovation, hence, managers should not be penalized for budget deficits, as
this provides them with a negative incentive for investing in NPD. In a related vein,
Schlapp et al. (2015) find that when interdepartmental coordination (via informa-
tion sharing across multiple departments) is vital to the success of NPD projects
but departments compete for resources, contracts play a key role in aligning the in-
centives of the different departments. First, group or shared incentives are better for
engendering radical innovation, while individual incentives are better for motivating
incremental NPD. They also find that information asymmetries increase the effective
cost of product evaluation, resulting in under-investment by the firm in information
acquisition. In contrast, for intermediate information acquisition costs and precise
information, information asymmetries result in the firm over-investing in informa-
tion acquisition. Finally, firms that have a high degree of information asymmetry
internally between departments have a tendency to develop more products than the
first-best solution, and a higher degree of truthful information exchange can signifi-
cantly rationalize the firm’s NPD portfolio.
7 Contracting for New Product Development 129

In addition to hidden actions of managers, NPD initiatives may involve managers

with domain specialization leading to asymmetric information between the firm and
the managers. In particular, often an NPD project involves a stage-gate process,
where employees (agents) of a firm learn about potential ideas from the research
stage of the process and need to exert efforts during the developmental stage. The
firm (principal) typically makes decisions to proceed or not with the project idea at
the evaluation stage. Due to knowledge-specificity of such projects, it is likely that
the agents have superior information about the quality of the initial idea (leading
to information asymmetry). Moreover, during the development stage, often efforts
exerted by the agents which influence the outcome of the project are unverifiable
(leading to moral hazard). Hence, designing incentive contracts that overcome in-
efficiencies due to these agency issues is imperative for the principal. Chao et al.
(2014) study such a problem, where the agent has perfect information but the prin-
cipal has imperfect information about the quality of the output from the research
stage (idea). The principal makes the decision to proceed or not with the project at
the evaluation stage. If the idea is pursued further, the agent exerts efforts (unob-
servable), and the firm exerts fixed resources for development. Their setting can be
represented by our model by making the agent’s information set perfectly informed,
and the principal’s information set partially informed about the quality of the idea
at the point when the incentive contract is designed. The decisions of the agent are
the information he/she reveals about the quality of the idea after the research stage,
and the effort in the development stage if the idea is pursued further. The decision
of the principal is to pursue the idea further, contingent on the information revealed
by the agent regarding the quality of the idea. The revelation principle allows the
authors to restrict their attention to contracts such that the agent’s decision post-
contracting is to truthfully reveal the quality of the idea. At this point in time, both
the agent and the principal have perfect information about the quality of the idea.
The incentive contract, which pays the agent after the entire project is executed, not
only incentivizes the agent to truthfully reveal information, but also determines the
effort exerted by the agent during the development stage. Restricting their attention
to linear contracts, the authors find that the information rent extracted by the agent
increases in the complexity of the project, and the principal has to incentivize the
agent to devote more time to the search space in the presence of greater complexity.
Information asymmetry may also lead to inefficiency in the system when agents
with superior information may derive an implicit utility from an NPD initiative than
is not aligned with the goals of the firm. For example, “engineers may be inclined to
value their own beauty-ideals” Mihm (2010) that may lead to decisions that are not
profit maximizing for the firm. To study such a setting, our general model may be re-
duced to one where there is a mismatch in the utility of the firm and multiple agents.
Mihm (2010) studies such a setting under different control regimes and incentive
structures in terms of cost compliance, and shows that different contract structures
are more effective under different control regimes in aligning agent incentives with
the firm. The information set of each agent includes his/her private information of
cost-performance trade-off for the component that they are responsible for and prior
beliefs on the same trade-off for all other components. The principal has imperfect
130 S. Hasija and S. Bhattacharya

information about the trade-off for all the components. The decisions made by the
agents before each milestone stage is to announce the spending level on their com-
ponent. The decision of the principal makes the decision to proceed further or not
with the NPD project based on its updated information. The analysis of such a model
shows that that heavyweight project management (as a control mechanism) can in-
crease firm profits significantly, but has the disadvantage of delaying the project by
having multiple iterations due to cost-gaming. Providing incentives at the individual
level such as career advancement or providing targets to the agent can align the ac-
tions of the agent to the firm, but transfer mechanisms can be expensive in terms of
information gathering and exchange; while target outcome mechanisms like target
costing are feasible if the targets can be determined accurately. Profit sharing mech-
anisms can be used if the surplus to be shared is reasonably high and if the number
of agents is small.
In addition to the issues discussed above, NPD projects pose additional man-
agerial challenges due to uncertainty of future events that may be hard to calibrate
ex-ante. In such situations, the principal also needs to account for the risk exposure
of the agents to such unforeseeable events to incentivize them to invest appropriate
effort levels. Further, such incentive contracts may need to satisfy additional con-
straints on inter-temporal fairness, wherein the agents should not be made worse
off in expectation due to such events. Such a setting can be captured by the general
model presented in Sect. 7.1.1 of this chapter by updating the information set of the
principal and the agent with time, as more information is revealed about the future.
Different types of individual rationality constraints may also be captured to reflect
the economic tension in the system (as discussed in Sect. 7.1.1). Sommer and Loch
(2009) study this exact setting and show that under unforeseen uncertainty, target
oriented contracts may be inadequate as firm plans may change due to unforeseen
events, fairness dictates that employee compensation cannot be shifted downwards,
and efforts and unforeseen events may be unobservable. Under such conditions,
contracts have to be adjusted ex-post to reflect process monitoring, downside pro-
tection and upside rewards for the employee. If the firm can observe actions, then
monitoring efforts and effort-based incentive contracts are optimal, and if the firm
cannot observe actions but can observe unforeseen events, it readjusts the contract
for the impact of the unforeseen event with downside protection (no wage loss due
to circumstances beyond the agent’s control). If the firm cannot observe either agent
actions or unforeseen events, then the firm provides a contract with upward incen-
tives and downside protection.
The discussion above provides insights on how firms can incentivize its employ-
ees to overcome agency issues that affect the efficiency of NPD initiatives. These in-
sights are derived using models that are based upon axiomatic assumptions of game-
theoretic behavior on part of all parties. However, for the insights to have practical
relevance, it is important to understand the behavioral implications of incentives that
are designed to align employee actions with firm objectives. Using experiments with
human subjects, Ederer and Manso (2013) investigate if pay-for-performance is in-
deed the optimal way for incentivizing internal managers, and find that the horizon
of offering the incentives are a key element of the efficacy of a contract. They find
7 Contracting for New Product Development 131

that contracts that allow for early exploration by not penalizing their experimen-
tal subjects for failure and provide incentives for long-term performance encourage
the most exploration of innovative business strategies and tend to dominate fixed-
wage and pay-for-performance contracts. As expected, subjects who are offered the
fixed wage tend to exert the least effort, and pay-for-performance contracts do not
perform well under subject risk-aversion. Finally, early termination threat affects
exploration behavior and innovation success negatively, but golden parachutes al-
leviate these effects. In an empirical study, Samila and Sorenson (2011) find that
non-compete covenants have a limiting effect on innovation and entrepreneurship in
the innovation community, and they impede entrepreneurial and innovative behavior
as measured by patent count and number of new ventures started. The spillover ef-
fect is an important enabler of innovation in a community, and providing employees
with the potential for an exit increases the innovative practices within a community.

Table 7.1 Literature on internal contracting for NPD

Paper Agency issues Contracting levers Decisions made

Chao et al. Agent effort, Explicit (linear wage) Resource allocation for
(2009) adverse selection, and implicit (career NPD projects
risk aversion growth) (radical/incremental)
Chao et al. Effort coordination, Linear wage contract Contract design, agent
(2014) adverse selection, effort
risk aversion
Ederer and Effort coordination Pay-for-performance, Contract design, business
Manso (2013) fixed-wage, termination, strategy choice
golden parachute
Mihm (2010) Effort coordination Linear utility, group Project management,
incentives agent effort
Samila and Effort coordination Non-compete covenants Non-compete
Sorenson (2011) enforcement, external
investment
Schlapp et al. Effort coordination, Fixed wage, own project Agent effort, contract
(2015) adverse selection performance, peer design
project performance,
group incentives
Sommer and Effort coordination, Order-based, Contract design,
Loch (2009) risk mitigation, target-based with monitoring, agent efforts
adverse selection downside and upside
rewards

Table 7.1 summarizes the key literature on contracting in facilitating internal

NPD. Although the extant literature has studied different economic tensions in the
NPD process, some commonalities can be found in these studies. The modeling lit-
erature on the subject generally considers linear wage contracts for the mitigation of
132 S. Hasija and S. Bhattacharya

agency issues, comprising of a fixed component of wage (salary) and a performance-

based component (where the performance-based component can be linked either
explicitly or implicitly to the total reward or profit from the project). Linear con-
tracts have been studied extensively in the literature on supply chain contracting
(e.g., revenue sharing contracts), and been shown to be implemented extensively in
practice as well. In addition to the use of linear wage contracts, the literature has
also studied the use of implicit incentives like career growth (Chao et al. 2009), the
threat of termination with or without additional compensation (Ederer and Manso
2013), non-compete clauses (Samila and Sorenson 2011), peer project performance
based measures (Schlapp et al. 2015), and target-based compensation with down-
side protection (Sommer and Loch 2009). Additionally, a number of papers in the
literature consider the issue of adverse selection, usually with the project manager
having better information on the quality of the NPD project than the senior manager
representing the firm, or the firm itself (Chao et al. 2014; Schlapp et al. 2015).
The generic contract formulation in all these papers that consider the agency issue
of adverse selection has a menu of contracts that can be offered to the agent (the
employee), and the agent has to self-select (choose the contract based on his/her
incentive compatibility and individual rationality constraints) the best contract for
him/her. The menu of contracts is designed to elicit truthful information from the
agent based on the revelation principle (Myerson 1981), and the contract is optimal
for the principal, though the agent can extract some information rent based on the
degree of asymmetric information between the firm and him/her.
While most of the modeling literature also provides individual incentives for the
agent only, Schlapp et al. (2015) and Mihm (2010) also include a group incentive,
in that agents are rewarded partially based on the performance of projects evaluated
by peers or firm surplus (profit-sharing). Schlapp et al. (2015) finds that in environ-
ments where the degree of information asymmetry is low, or where there is a signif-
icant disparity in the potential outcomes between projects competing for NPD re-
sources, individual contracts tend to outperform group-based contracts, while Mihm
(2010) finds that group incentive contracts are effective if the projected profit sur-
plus of the firm is above a certain threshold value, however, profit sharing contracts
are only effective when the number of agents is small. Finally, when the horizon
of the problem being studied is multiple periods versus a single period, the design
of the contract crucially depends on the information revealed during the horizon.
For instance, Sommer and Loch (2009) find that incentive contracts for individuals
have to take into account the events of unforeseeable uncertainty that are revealed
during the horizon, while Mihm (2010) models that agents sink at most the same
effort that they sank in the first period.
This chapter has looked at some issues and corresponding insights regarding
a firm’s incentive design problem internally. Naturally, economic tensions due to
agency issues arise in settings of external NPD as well, making the contract design
problem an important step to ensure a favorable outcome of the NPD initiative. In
the following chapter, we discuss issues and present corresponding insights from
the literature pertaining to the setting where firms collaborate with external firms
for NPD.
7 Contracting for New Product Development 133

7.3 Contracts for External NPD

Partnering with external firms (vendors) for NPD is a rising trend, even in industries
that have traditionally conducted NPD in-house (Cassiman and Veugelers 2006).
Just as in internal NPD, in such settings, designing contracts to align incentives of
different parties may be challenging due to factors such as unobservable efforts,
different risk appetites of the involved parties, and information asymmetry between
these parties. We now summarize the results for contracts governing existing exter-
nal NPD relationships.

7.3.1 Management of Contractual Collaborations

for External NPD

The contracting literature on governing NPD relationships can be divided into two
sets: one set studies scenarios where only the agents exert effort, and a second set
that studies the agents and the principal exerting efforts. Many of the economic ten-
sions discussed in Sect. 7.2 of this chapter are also relevant to the setting where client
firms contract external vendors to participate in their NPD initiatives. By consider-
ing the client firm as the principal and the external vendors as the agents, the insights
provided in Sect. 7.2 readily port to this section as well. That said, the external NPD
setting may allow for certain context-specific characteristics (e.g., competition be-
tween multiple vendors, use of negative financial transfer payments to agents, buy-
out clauses in contracts, payments contingent on project tardiness), which may not
easily be applicable in the case of internal contracting between a firm and its em-
ployees. In this section, we discuss some issues around contractual collaborations
for external NPD and corresponding insights from the literature, with the caveat that
some of the discussion may be relevant for the case of internal NPD and vice-versa.
We divide this section into two parts, first, where only the agents invest in NPD ef-
forts, and the second where both the principal and the agents do so. We provide the
key lessons from the extant research below.

7.3.1.1 Only Agents Invest in NPD Effort

In the absence of other agency issues due to information asymmetry and risk-
aversion, when only the agents exercise efforts, the principal can always design
contracts that yield the first-best outcome, even if the verifiable signal (Θ (·), in our
general model) is different from the final revenues (Π (·), in our general model).
This is also true when multiple agents exert efforts and Θ (·) may or may not be
separable in the efforts of these agents. This is because in such a setting the princi-
pal, in effect, plays the role of a “budget breaker” (Holmstrom 1982), which is not
possible when the principal is also required to exert efforts for the project’s success.
Therefore, we discuss the latter scenario separately.
134 S. Hasija and S. Bhattacharya

However, obtaining the first-best profit when only agents exert efforts may
require the principal to include a negative fixed-fee in the contract to extract the sur-
plus of the project from the agents (Laffont and Martimort 2002, p. 156). Whereas,
this may be possible to implement in settings that involve contracting between firms,
this seems unrealistic to implement when firms contract with its own employees.
Moreover, if agents are protected by a limited liability clause, then there exists a
bound on such a negative fixed-fee, imposing some limitations on the principals abil-
ity to attain the first-best outcome. Furthermore, such cases may also limit the prin-
cipals ability to impose performance-based penalty clauses in the contracts, leaving
it with limited contractual levers to incentivize appropriate effort levels from the
agents, making the contract design problem an interesting challenge.
Kwon et al. (2010) consider one such interesting case, where multiple vendors
work on components of the same project. They study two types of contracts, one
where each vendor receives a payment upon completion of their task, and second
where each vendor is paid only when all vendors have completed their tasks. Such a
setting, in terms of our general model implies, that the principal can contract with the
agents on verifiable signals which are only a function of their efforts or on a signal
which is a function of efforts of all agents. The authors find that when the processing
rates of vendors are not adjustable, offering all vendors a contract where the payment
will be made by the overall project completion time (delayed payment contract)
has the detrimental effect of delaying the NPD project. However, when vendors can
adjust their processing rates, the delayed payment contract has an equilibrium where
all vendors process at the same rate, and hence, this leads to a higher payoff for the
vendors and the client. This effect is strong when the revenue of the NPD project is
small, or the number of vendors is large.
When client firms use multiple vendors for their NPD initiatives, competition be-
tween these vendors creates an interesting setting that may not be readily applicable
to the case with internal NPD. Although, the general model presented in Sect. 7.1
does not directly cover such a situation, it has received some attention in the liter-
ature. Wang and Shin (2015) model the case of competing vendors who invest in
NPD and provide the innovative component to the manufacturer. They find that if
there is a sole vendor that is investing in NPD, the revenue-sharing contract out-
performs the wholesale-pricing contracts for coordinating the effort of the vendor.
However, if the market values quality, then the manufacturer may be better off by
using a quality-dependent wholesale-pricing contract, even though the vendor may
under-invest in NPD. When competing vendors invest in NPD, the wholesale pric-
ing contract does better than the case with only one vendor, and the revenue-sharing
contract again alleviates the under-investment problem on the part of the vendor,
and is optimal when the vendors are symmetric.
In contrast to Wang and Shin (2015) who consider competing vendors, when
the competition is at the retail level and there is a single vendor who invests in in-
novation, the resulting insights for the NPD efforts can be very different. Williams
et al. (2011) model this case, and find that the market characteristics (specifically,
consumers who are divided either as price-sensitive or insensitive) tend to determine
the design chosen by the vendor, and the presence of competing retailers tends to
7 Contracting for New Product Development 135

make markets efficient from the perspective of consumer welfare and market cov-
erage. In contrast, in a retail monopoly, the retailer has more power and can force
the vendor to provide product designs that increase retailer profits at the expense of
the manufacturer’s. They also find that when NPD efforts are made based on tech-
nologies being developed by other providers, the choice of generation of technology
by the client is an important component of the vendor’s decision to invest in NPD.
Erat and Kavadias (2006) also study competition at the client level and find that the
prices offered by the vendor, who develops a technology, can be used to either sep-
arate the competing clients into two sets of immediate adopters and future adopters,
or induce all clients to adopt current and future generations of the technology. The
strategy of inducing some clients to commit to the future version of the technology
is optimal if the rate of improvement of the technology is low.
So far have discussed cases where agent efforts are exerted after the offering of
the contract by the principal. Many instances exist where an innovator has already
developed an innovation, but is not capable of commercializing it without the help
of an external partner. In such cases, the innovator may be considered as the prin-
cipal and the commercial partner may be considered as an agent (e.g., a biotech
firm that has developed a new drug and needs a pharmaceutical firm to commer-
cialize it). Owing to its potential lack of commercial know-how, the innovator may
not know the true commercial value of the innovation. In such a case, the pres-
ence of downstream competition enables the innovator to sell the innovation (either
outrightly or partially via royalty terms) using an optimal auction design. Outright
sale of the innovation is not an uncommon practice, especially in the presence of
intermediary firms (patent intermediaries), that have a significant cost advantage in
terms of the administrative costs associated with patenting the innovation. Agrawal
et al. (2016) study a setting in which two incumbent firms and a patent intermedi-
ary form compete to obtain an innovation from an innovator who does not know the
true commercial value of the innovation. After the innovation is sold via an auction,
the winner may choose to cross-license the innovation to the other incumbent (if the
winner is one of the incumbent firms). If the winner is the patent intermediary firm,
then it may choose to license the innovation to one or both incumbents. This paper
shows that the equilibrium outcome of the setting critically depends on the degree
of innovation, and that the presence of the patent intermediary in the equilibrium is
more likely when the degree of innovation is low, and the patent intermediary firm
serves to make the overall market more efficient.
If there is no downstream competition, then the innovator may not be able to
extract a positive surplus by outrightly selling the innovation. Crama et al. (2008)
model the innovator as the principal who offers the innovation to an agent via a
licensing contract. When the innovator does not have perfect knowledge of the value
of the innovation, the innovator should increase the upside to the agent from the
contract (via royalties) as the agent’s valuation of the innovation increases. A risk-
averse innovator should also include fixed fees in the contract to avoid the potential
stochasticity of payoffs from milestone and royalty payments. In many cases, after
obtaining the license of the innovation, the agent also needs to exert some effort in
commercializing it. Crama et al. (2008) show that when the agent has to exert an
136 S. Hasija and S. Bhattacharya

effort and the innovator and agent have asymmetric information, it leads to a further
loss for the innovator. The effort of the agent in commercializing the innovation
may also be looked as a real-option, where the agent may choose to terminate the
license at some stage if commercialization does not seem profitable. Dechenaux
et al. (2008) find that the risk of terminating a license to an invention is decreasing
in the effectiveness of patent strength and secrecy, while for inventions that have
a high lead time to commercialization, agents should defer the commercialization
to minimize development risk. However, when there is a high reliance on learning
for inducing profitability from an invention, the agent should commercialize the
invention early. Hence, the agent balances between market and development risks
in deciding whether to continue with a license, terminate the license, or defer the
decision. Together, these choices comprise a real option. In addition to deferring
decisions on licensing innovations, agents can also use the option of deferring the
date of payment for suppliers based on the project completion time, to ensure that
projects are not tardy.
Table 7.2 summarizes the key literature used for the basis for the discussion in
Sect. 7.3.

Table 7.2 Literature on agent only investments in collaborative NPD

Paper Agency issues Contracting levers Decisions made

Agrawal Asymmetric information, Licensing agreements, Cross-licensing, single

et al. downstream competition auction or dual licensing, bidding
(2016)
Crama et al. Effort coordination, Fixed fee, Contract design,
(2008) adverse selection, milestone payment, agent effort
risk aversion royalties
Dechenaux Risk mitigation Licensing fee Commercialize,
et al. terminate or defer
(2008)
Erat and Risk mitigation, Licensing prices Technology adoption,
Kavadias uncertainty introduction sequence
(2006)
Kwon et al. Effort coordination, Fixed fees, Contract design,
(2010) risk management timing of payment vendor processing rates
Wang and Effort coordination Wholesale price, Contract design,
Shin (2015) quality-dependent agent effort
wholesale price,
revenue sharing contract
Williams Effort coordination Pricing Contract design,
et al. channel structures
(2011)
7 Contracting for New Product Development 137

Although the above literature considers only agents exerting efforts in the NPD
process, there are a number of examples in practice where the principal and the
agents exert efforts (simultaneously or sequentially) during the NPD process. We
now focus on the main findings of this stream of the literature.

7.3.1.2 Principal and Agents Invest in NPD Effort

Contractual collaborations between the principal and agents for NPD in many cases
may involve the notion of co-production, where the principal is also required to exert
efforts along with the agent to facilitate successful outcomes. Such common settings
may give rise to new economic tensions (e.g., holdup, free-riding) that may result
in loss of economic value of the project. In this section of the chapter, we focus
on these setting and discuss some insights from the literature that have provided
guidelines on how firms can eliminate/attenuates such losses.
The literature on collaborative investments in NPD with joint efforts has been
studied from the perspective of (1) the timing of the investments of the partners
(simultaneous versus sequential), (2) design of alliances, (3) informal contracting
relationships. Table 7.3 summarizes the key literature on joint collaborative invest-
ments in external NPD.
An important aspect of collaborative investments in NPD is the timing of the
investments made by the partners: investments can either be made sequentially, or
simultaneously. When partners make sequential investments, there is a propensity
for holdup, i.e., the second-mover in the sequential game has incentives to hold up
its partner and renegotiate the contract by threatening to under-invest in its effort
if the contract is not renegotiated. The principal has to account for this holdup in
its contract design if its the first-mover, as it has to align incentives in such a way
that it can still attain the first-best outcome for itself under any renegotiated contract
after it has sunk its effort. If the principal is the second-mover, it has to design the
contract in a way that the agent (first-mover) will exert its first-best effort even if the
principal threatens to renegotiate the contract after the agent has sunk its effort, as
the agent will account for the holdup problem when it exerts its effort. In our general
model, w(·) = W (·) would represent the notion of renegotiation. Bhattacharya et al.
(2015) consider the case where the principal is the second-mover and the agent is
the first-mover. They find that it is optimal for the principal to offer a contract to
the agent where the offered contract is endogenously renegotiation-proof, i.e., if the
agent exerts its first-best effort, the principal will not renegotiate the contract when
it exerts its own effort. The threat of renegotiation is then used by the principal to en-
sure that the agent exerts its first-best effort. The authors show that options contracts
based on intermediate verifiable outcomes (e.g., FDA approval for new drug devel-
opment) attain such an outcome for the principal. They also show that such contracts
are able to overcome any inefficiency in the system due to risk-aversion on part of
the agent. When the effort of the first-mover is not observable to the second-mover,
there is no holdup issue. However, since both the principal and the agent are needed
to exert efforts, this leads to the free-rider problem and hence the principal cannot
138 S. Hasija and S. Bhattacharya

Table 7.3 Literature on joint collaborative investments in external NPD

Paper Agency issues Contracting levers Decisions made

Amaldoss and Effort coordination Profit sharing Partner efforts

Staelin (2010)
Bhaskaran and Effort coordination Cost, development work Contract design
Krishnan (2009) and revenue sharing
contracts
Bhattacharya et al. Effort, holdup, Fixed fee, Contract design,
(2015) risk aversion milestone payment agent effort
Bhattacharya et al. Effort, holdup Cost-plus, gain-share Contract design,
(2014) agent effort
Cassiman and Effort coordination licensing, fixed fee, Self and partner R&D
Veugelers (2006) acquisition mix
Ge et al. (2014) Effort coordination Wholesale price, alliance Alliance structure
Kalaignanam et al. Value split Alliance, profit sharing Alliance formation
(2007)
MacCormack and Effort coordination Fixed fee, Contract design
Mishra (2015) performance-based, time
and materials, relational
contracts
Savva and Effort coordination, Fixed fee, milestone Contract design,
Scholtes (2014) risk mitigation payment, royalty, partner efforts
opt-out
Xiao and Xu Effort coordination, Fixed fee, royalty Contract design
(2012) adverse selection
Yang (2010) Effort coordination Profit sharing, Contract design
payment timing

attain the first-best outcome. Interestingly, although holdup introduces the notion
of potential renegotiation in the system after the initial contract is signed, design-
ing contracts that account for such behavior can actually make such game-theoretic
dynamics in the setting Pareto improving, i.e., favorable for the principal and not
detrimental for the agent (Bhattacharya et al. 2014). Xiao and Xu (2012) study the
design of the contract under the threat of renegotiation where the client (principal)
and the vendor make investments in both the R&D stage (first) and the marketing
stage (second), and find that if the vendor has a significant effort to make in the mar-
keting stage as well, the client should offer the vendor a lower royalty rate upfront
and then revise the royalty rate upwards to incentivize the agent to make its first-best
effort in the second period. Yang (2010) considers the sequential game between an
entrepreneur who makes an upfront investment, and a manager and another agent
who make continuing investments in a continuous game, and find that the optimal
design of the contract entails all the reward of the collaborative effort up to a certain
7 Contracting for New Product Development 139

threshold date being extracted by the entrepreneur, and all subsequent rewards from
the collaborative effort being shared by the manager and the other agent. Plambeck
and Taylor (2007) consider bilateral investments where the firm invests in inno-
vation, and the supplier invests in capacity. Since supply quantity is verifiable in
their model, a quantity enforcing mechanism can ensure the first-best outcome in
the one-buyer, one-supplier case.
The design of alliances for NPD collaborations can also be a challenging task,
as partners will behave differently if (1) the other partner has complementary ver-
sus substitutable skills, (2) the partners have significantly different sizes (implying
differential levels of bargaining power), and (3) the partners are highly integrated in
the alliance. Hence, alliances and associated contracts between the partners have to
be designed in a way that can align the actions of the partners to the overall profits
for the channel, and attain a division of profits based on the respective bargaining
power of the partners. When firms with diverse sets of resources make alliances for
NPD, they have the benefit of complementary skills, enabling them to develop NPD
projects that each partner could not perform by themselves, while when firms with
similar sets of resources form alliances, they have the advantage of scale. When
firms of differing sizes form an alliance for NPD, the characteristics of the firms is
very important, as the difference in bargaining power may make the smaller firm
wary of entering such an alliance, and the larger firm may find it difficult to con-
vince the smaller firm of the benefits of the alliance. Amaldoss and Staelin (2010)
consider the first issue, and find that when firms with diverse sets of resources form
alliances (cross-function alliances), they invest more than corresponding firms in
alliances with similar resources (same-function alliances). This effect is robust in
the number of firms in the alliances. Kalaignanam et al. (2007) study alliances of
firms of disparate sizes, and find that broad scope alliances (cross-function or link
alliances) are better for financial gains for large firms, but same-function alliances
(scale alliances) are better for financial gains for smaller firms in the long run. In a
similar vein, Savva and Scholtes (2014) study the incentives of smaller firms to col-
laborate with larger firms on NPD, and find that when smaller firms face bankruptcy
risk, giving the option of opting out of the collaborative effort to the smaller firm
with a pre-agreed licensing agreement greatly reduces the risk of bankruptcy for the
smaller firm. The opt-out clause of the contract has the further advantage of elim-
inating the possibility that profitable projects are abandoned owing to bankruptcy
risk. When firms search for external R&D sources as a complement to augment
their NPD output in addition to internal R&D, Cassiman and Veugelers (2006) find
that having basic in-house R&D capabilities is critical to attaining the full bene-
fits of external R&D knowledge. The contractual forms used by NPD alliances also
depend significantly on the nature of innovation. Bhaskaran and Krishnan (2009)
find that investment cost sharing contracts and development work sharing contracts
are better for developing brand new products. The investment cost sharing contract
performs best for new-to-the-world projects with timing uncertainty, while develop-
ment work sharing contracts work well for NPD projects with quality uncertainty. In
contrast, revenue sharing contracts may perform better for incremental new product
development. Ge et al. (2014) investigate the formation of NPD cartels, and show
140 S. Hasija and S. Bhattacharya

that sustainable cartelization with a win-win effect for both partners can only be
achieved if their efforts are comparable to each other, and the existence of spillovers
will only benefit the firm with the lower contribution, hence, spillovers strengthen
the moral hazard effect.
Relational contracting is a relatively new way of constructing social rather than
formal economic contracts for the purpose of managing relationships. These infor-
mal mechanisms take an information exchange approach for coordinating relation-
ships and reducing information asymmetry and aligning the goals of the partners
in the collaborative NPD process. While NPD alliances may differ in the degree of
integration of the partners in the alliance, the nature of the relationship has impor-
tant ramifications for the performance of NPD alliances. MacCormack and Mishra
(2015) find that a higher degree of integration between NPD alliance partners results
in higher coordination costs, but also leads to a higher new product or service qual-
ity if time-and-materials or pay-for-performance contracts are used. They also find
that the benefits of relational contracts are not reaped by the partners if the choice of
contract is misaligned with the project outcomes, i.e., when the choice of contract
is misaligned, they find that the NPD project coordination cost increases, and there
is no effect on the product quality from the collaborative effort.
However, while some firms seek contractual collaborations for NPD with exter-
nal vendors, recently, many firms also utilize pools of freelance problem solvers to
aid their NPD initiatives. Such settings typically involve the design of innovation
contests (including incentive structures) to help firms to effectively tap the benefits
of crowd-sourcing. In the next section, we discuss issues around contest and incen-
tive design when firms use freelance problem solvers and then discuss the manage-
ment of contractual collaborations which involves issues around contract design.

7.3.2 Contests for Contracting on External NPD by Freelance

Innovators

The literature in this field is in its nascent stage, but a base exists for researchers to
build on our understanding of this important topic. Table 7.4 summarizes the key
literature in this field.

Table 7.4 Literature on contracting for external NPD by freelance innovators

Paper Agency issues Contracting levers Decisions made

Che and Gale Effort coordination Fixed prize, auctions, Contract design, agent
(2003) contract menus effort, vendor selection
Erat and Krishnan Effort coordination, Fixed prize, number of Contract design,
(2012) specification risk prizes agent effort
Terwiesch and Xu Effort coordination Fixed prize, Contract design,
(2008) performance-contingent agent effort
7 Contracting for New Product Development 141

Obviously, the incentives in most contests are individual-driven rather than

group-driven, apart from team-oriented contests. The contests are also conducted
as single-stage contests, in that there is no mid-stage screening of ideas with a
final contest to determine the winner. All the three papers mentioned in Table. 7.4
consider the question of the client eliciting the best effort from the competing free-
lance problem solvers (agents), and then choosing the one with the best new product
solution.
The key questions addressed by this stream of literature are: (1) what is the award
structure that the client1 should use to incentivize competing innovators? (2) What is
the number of awards that should be given and the number of agents that should be
considered? In a two-agent first-price auction setting, Che and Gale (2003) find that
under certain weak conditions, the agent that is more efficient (has a lower marginal
cost of innovation for a given level of quality) will bid for a higher award. They
find that if only two agents are used by the client, then the first-price auction is the
best contract design that can be used by the client. Terwiesch and Xu (2008) build
on this research by considering the trade-off between having multiple agents com-
peting in the innovation contest leading to a lower effort in equilibrium from each
agent (which hurts the client), versus the benefit of obtaining a more diverse set of
solutions, which can sometimes compensate for the under-investment of effort. Fur-
ther, this effect of under-investment of effort on the part of multiple agents can be
reduced by considering pay-for-performance awards in place of fixed-prize awards.
The characteristics of the search space are also an important element for the design
of contests. Terwiesch and Xu (2008) divide the solution space into expertise-based
projects (projects with well-defined performance and solution spaces), trial and er-
ror projects (projects with well-defined goals but rugged solution landscapes), and
ideation projects (ill-defined performance and solution spaces). They find that the
under-investment in effort effect is more severe for expertise-based projects, less
severe for ideation projects, and least severe for trial-and-error projects. Addition-
ally, they also investigate if the agents should pay a fixed fee for participation and
for the performance of performance-contingent versus fixed fee awards, and find
that for ideation and trial-and-error projects, the fixed fee should not be charged and
performance-contingent prizes perform better for the client than fixed-fee prizes.
Erat and Krishnan (2012) study the contract design problem in an innovation con-
test setting, and find that if a single fixed-prize award is used, the size of the award
has to be higher if the search for better new product solutions is costly and if the
search space is broad; if there are multiple agents competing in the contest, decreas-
ing and then increasing in the probability that the client and the agent have the same
understanding of the problem’s specifications; and increasing in the ex-ante simi-
larity of the design search space. They find that the number of agents searching a
design space depends on the potential of the design approach, search cost, award
size, and the specification clarity provided by the client. The client may also offer
two awards (first and second prizes) to incentivize agents to broaden the search.

1 We avoid using the term “principal” here to differentiate this setting from a typical principal agent
setting, which is more amenable to the setting with external NPD involves a contractual vendor.
142 S. Hasija and S. Bhattacharya

We now conclude the chapter with an assessment of the state-of-the-art knowl-

edge of contracting in NPD and provide guidelines for future research.

7.4 Conclusions and Future Research

The area of contracting in NPD is fairly new in the Operations literature, however,
in recognition of the importance of the problem, there has been an increase in the
research attention given to this topic in the last few years. The research on the area
has looked at mitigating agency issues like moral hazard, adverse selection, and risk
aversion in determining methods by which contracts can increase the total profits
for all the parties involved in the NPD process, and then divide the total profits be-
tween the parties in accordance with bargaining or other splitting rules. The parties
involved in the NPD process can be internal (firms contracting with employees), ex-
ternal (firms contracting with other firms where either one firm or both firms exert
effort), or freelancers participating in outsourced NPD in the form of contests. Con-
tracts can be designed for either individual agents or for groups of agents with levers
that are either guaranteed (e.g., fixed fees), or are based on pay-for-performance
(e.g., milestone payments and royalties), or can be relational contracts rather than
formal contracts. The timeline of the contractual collaboration (repeated versus sin-
gle interactions, sequential versus simultaneous efforts) also has an important role
to play in the design of contracts.
As the research on this area is in its early phase, we can identify a number of
avenues for future research. First, most of the research on this field has considered
NPD with the objective of maximizing firm profitability or the profitability of the
partnership. There is a developing field in social entrepreneurship in practice that
has an objective of maximizing consumer and social welfare, however, contracting
on new product or service development has been largely unexplored so far. Contracts
that govern relationships in social entrepreneurship would deal with a different set
of motivating factors and contracting levers, and the insights would be different
from contracts that govern for-profit relationships. Similarly, NPD projects can be
collaborated on by a for-profit firm, and a non-profit partner like the government or
an NGO.
There is also a lack of understanding of the use of phenomena like crowdsourc-
ing in NPD that is better established in practice. While the literature has begun
studying the impact of crowdsourced innovations (the innovation contest literature
summarized in Sect. 7.3.2), consumer feedback is being elicited in practice on vari-
ous parts of the NPD process from understanding market requirements to providing
new product concepts to testing prototypes and delivery of new products. Future
work should consider the role of contracting in incentivizing consumers to collab-
orate with firms on the NPD effort. Consumers may also provide useful feedback
to firms that can be gathered from secondary sources of data like social media, the
research on contracting in this field is sparse. Network theory provides a lot of in-
sights on gathering information and disseminating product information better using
7 Contracting for New Product Development 143

social media by harnessing the connectivity of different components in a network,

future research can investigate how contracts can be used to facilitate these pro-
cesses of information gathering and marketing. Moreover, it would be interesting
to see how client firms can create multiparty NPD collaborations with (1) experts
who can mine, analyze, and systematically report insights obtained from unconven-
tional sources of data (e.g., social media), (2) firms that use traditional methods of
information collection (e.g., focus group based market research), and (3) other part-
ners in the supply chain (suppliers, buyers) who may have better information about
certain aspects of the NPD process (e.g., due to knowledge specialization or due to
closer contact with the end-user due to their position in the supply chain).
Finally, while the literature on contracting in NPD has studied issues like group
incentives, the dominant mode of incentivizing agents remains the individual mode
in the literature. Future work can take a look at group incentives in more detail. Also,
the literature has almost completely focused on formal contracts for governing NPD
collaborations, future research should also investigate the role of relational contracts
in aligning incentives, as relational contracts play an important role, especially in
internal NPD scenarios.

References

Agrawal A, Bhattacharya S, Hasija S (2016) Cost-reducing innovation and the role of patent inter-
mediaries. Prod Oper Manag 25(2):173–191
Amaldoss W, Staelin R (2010) Cross-function and same-function alliances: how does alliance
structure affect the behavior of partnering firms? Manag Sci 56(2):302–317
Bhaskaran SR, Krishnan V (2009) Effort, revenue, and cost sharing mechanisms for collaborative
new product development. Manag Sci 55(7):1152–1169
Bhattacharya S, Gupta A, Hasija S (2014) Joint product improvement by client and customer sup-
port center: the role of gain-share contracts in coordination. Inf Syst Res 25(1):137–151
Bhattacharya S, Gaba V, Hasija S (2015) A comparison of milestone-based and buyout options
contracts for coordinating R&D partnerships. Manag Sci 61(5):963–978
Cassiman B, Veugelers R (2006) In search of complementarity in innovation strategy: internal
R&D and external knowledge acquisition. Manage Sci 52(1):68–82
Chao RO, Kavadias S, Gaimon C (2009) Revenue driven resource allocation: funding authority,
incentives, and new product portfolio management. Manag Sci 55(9):1556–1569
Chao RO, Lichtendahl KC Jr, Grushka-Cockayne Y (2014) Incentives in a stage-gate process. Prod
Oper Manag 23(8):1286–1298
Che Y-K, Gale I (2003) Optimal design of research contests. Am Econ Rev 93(3):646–670
Crama P, De Reyck B, Degraeve Z (2008) Milestone payments or royalties? Contract design for
R&D licensing. Oper Res 56(6):1539–1552
Dechenaux E, Goldfarb B, Shane S, Thursby M (2008) Appropriability and commercialization:
evidence from MIT inventions. Manag Sci 54(5):893–906
Ederer F, Manso G (2013) Is pay for performance detrimental to innovation? Manag Sci
59(7):1496–1513
Erat S, Kavadias S (2006) Introduction of new technologies to competing industrial customers.
Manag Sci 52(11):1675–1688
Erat S, Krishnan V (2012) Managing delegated search over design spaces. Manag Sci 58(3):
606–623
144 S. Hasija and S. Bhattacharya

Ge Z, Hu Q, Xia Y (2014) Firms’ R&D cooperation behavior in a supply chain. Prod Oper Manag
23(4):599–609
Holmstrom B (1982) Moral hazard in teams. Bell J Econ 13(2):324–340
Kalaignanam K, Shankar V, Varadarajan R (2007) Asymmetric new product development al-
liances: win–win or win–lose partnerships? Manag Sci 53(3):357–374
Krishnan V, Ulrich KT (2001) Product development decisions: a review of the literature. Manag
Sci 47(1):1–21
Kwon HD, Lippman SA, McCardle KF, Tang CS (2010) Project management contracts with de-
layed payments. Manuf Serv Oper Manag 12(4):692–707
Laffont J-J, Martimort D (2002) The theory of incentives: the principal-agent model. Princeton
University Press, Princeton, NJ
MacCormack A, Mishra A (2015) Managing the performance tradeoffs from partner integration:
implications of contract choice in R&D projects. Prod Oper Manag 24(10):1552–1569
Mihm J (2010) Incentives in new product development projects and the role of target costing.
Manag Sci 56(8):1324–1344
Myerson RB (1981) Optimal auction design. Math Oper Res 6(1):58–73
Plambeck EL, Taylor TA (2007) Implications of breach remedy and renegotiation for innovation
and capacity. Manag Sci 53(12):1859–1871
Samila S, Sorenson O (2011) Noncompete covenants: incentives to innovate or impediments to
growth. Manag Sci 57(3):425–438
Savva N, Scholtes S (2014) Opt-out options in new product co-development partnerships. Prod
Oper Manag 23(8):1370–1386
Schlapp J, Oraiopoulos N, Mak V (2015) Resource allocation decisions under imperfect evaluation
and organizational dynamics. Manag Sci. doi:10.1287/mnsc.2014.2083
Sommer SC, Loch CH (2009) Incentive contracts in projects with unforeseeable uncertainty. Prod
Oper Manag 18(2):185–196
Terwiesch C, Xu Y (2008) Innovating contests, open innovation, and multi-agent problem solving.
Manag Sci 54(9):1529–1543
Wang J, Shin H (2015) The impact of contracts and competition on upstream innovation in a supply
chain. Prod Oper Manag 24(1):134–146
Williams N, Kannan PK, Azarm S (2011) Retail channel structure impact on strategic engineering
product design. Manag Sci 57(5):897–914
Xiao W, Xu Y (2012) The impact of royalty contract revision in a multistage strategic R&D al-
liance. Manag Sci 58(12):2251–2271
Yang J (2010) Timing of effort and reward: three-sided moral hazard in a continuous-time model.
Manag Sci 56(9):1568–1583
Chapter 8
Supply Disruptions and Procurement
Contracting

Volodymyr Babich and Zhibin (Ben) Yang

Abstract We discuss challenges that arise in practice of procurement contracting

from supply disruptions and suggest how the theory should be adjusted to address
these challenges. The challenges range from the loss of contract power when unex-
pected shocks occur to the need for contracts to govern joint risk management efforts
of buyers and sellers. We highlight which traditional contracting modeling assump-
tions need to be reexamined and how our intuition based on the classical results
might change. We point out a new direction of impactful research—coordination of
supply risk management efforts among firms.

8.1 Introduction

Contract theory in economics has been recognized for its successes in describing
interaction between economic actors, such as firms and workers, used-car sellers and
buyers, insurance companies and insurance buyers, and auction bidders and sellers.
There is a remarkably long list of winners of the Nobel Prize in economics1 who
made significant contributions to contract theory and the related field of mechanism
design.2 The Operations Management (OM) field has successfully adopted and

V. Babich
McDonough School of Business, Georgetown University, Washington, DC 20057, USA
e-mail: [email protected]
Z. (Ben) Yang
Lundquist College of Business, University of Oregon, Eugene, OR 97403, USA
e-mail: [email protected]
1 The official prize title is “The Sveriges Riksbank Prize in Economic Sciences in Memory of
Alfred Nobel”.
2 This list includes: Jean Tirole, Peter A. Diamond, Oliver E. Williamson, Leonid Hurwicz, Eric

S. Maskin, Roger B. Myerson, Robert J. Aumann, Thomas C. Schelling, George A. Akerlof, A.

© Springer International Publishing Switzerland 2017 145

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 8
146 V. Babich and Z. Yang

applied contract theory and mechanism design to problems that are relevant to
OM practices, including that of procurement. Numerous articles, book chapters,
and books (including this one) have been published to explore questions regarding
how to govern interaction between firms that exchange goods and services.3
This prior theoretical work is useful in practice. For example, suppose that
Chrysler would like to acquire plastic moldings for a new car model. OM schol-
ars can assist Chrysler in answering the following questions. How should Chrysler
manage contracting process with its suppliers? Which supplier should it use? What
terms should contracts have? What values should these term take? OM researchers
can advise Chrysler on how to account for demand uncertainty for Chrysler cars,
cost uncertainty of the suppliers, asymmetric information about supplier costs, com-
petition among suppliers, and competition of Chrysler with other automakers.
Much of the existing OM research on contracting helps firms to share demand
risk. However, economic and business developments of the last decade brought sup-
ply risk of the firms to the attention of the OM scholars and presented new research
opportunities.
Let’s continue with the Chrysler example. In 2008, Plastech, the supplier of
plastic moldings for Chrysler, filed for bankruptcy protection in 2008 and stopped
shipping moldings to Chrysler (Gillenwater 2008). Thus, the contracts that existed
between Chrysler and Plastech stopped working. Disruption to parts supply seems
to have caught Chrysler by surprise, and the automaker had to shut down production
on four plants. Several questions arise. What should Chrysler have done before the
disruption? What should Chrysler do in response? Should Chrysler have contracted
with other suppliers? Would having an internal or another external fabrication cap-
acity been a viable option? How to assess Chrysler’s risk exposure and the costs and
benefits of various mitigation options? How does the presence of risk-mitigation
tools affect the contracting process?
Plastech’s bankruptcy was precipitated by Chrysler’s terminating its future busi-
ness with Plastech and was followed by a dispute between Chrysler and Plas-
tech regarding the ownership of molds used in Plastech’s production. Interestingly,
throughout the dispute with Chrysler, Plastech continued to supply parts to other au-
tomakers. This also raises several questions. Should the contracts between Chrysler
and Plastech accounted for the possibility of disruption to parts supply? Could
Chrysler have used contracts to stave off the supplier’s bankruptcy? Could Chrysler
have inferred the likelihood of the supplier’s bankruptcy from financial contracts
that Plastech issued (e.g., debt contracts)? Should the contracts between Chrysler
and Plastech have been more explicit regarding the ownership of production assets,
in order to avoid costly court hearings? Should Chrysler and Plastech have coordi-
nated their efforts to manage Chrysler’s supply risk? These are all important ques-
tions, and OM researchers are only beginning to understand how to answer them.

Michael Spence, Joseph E. Stiglitz, James A. Mirrlees, William Vickrey, and for the contributions
to the game theory John C. Harsanyi, John F. Nash Jr., and Reinhard Selten.
3 For example, Tsay et al. (1999), Graves and de Kok (2003), Simchi-Levi et al. (2004), and the

references therein.
8 Supply Disruptions and Procurement Contracting 147

This example highlights that supply risk management is important. It also under-
scores the importance of capturing interaction between physical, information, and
financial flows in the supply chain. In this chapter, we shall discuss how supply risk
affects our understanding of procurement contracts and present a few ideas on how
Chrysler and other companies can deal with such new challenges.

8.2 Practice: Supply Chain Risk Management and Contracting

Supply chain risks are paramount concern of today’s managers. In recent years,
numerous supply disruptions and other supply risks have prominently featured in
media coverages, such as 2011 Japanese earthquake and tsunami (Lohr 2011), 2011
flood in Thailand (Kelly 2011), and 2007 recalls of various products manufactured
in China due to adulteration (e.g., Mattel toy recalls; see Story and Barboza 2007).
Port labor strikes, supplier bankruptcies, fluctuations in exchange rates and com-
modity prices, unexpected changes in government regulation, market manipulations,
political unrest, data security breaches, intellectual property breaches, epidemics—
the list of supply risks is inexhaustible. The consequences of these supply events
are significant, including delays in delivery, production disruption, loss of the firm’s
value, financial bankruptcy, and losses of health and lives (Sheffi 2005).
There exists a significant body of academic work that studies challenges to sup-
ply risk management from a single firm perspective.4 There are papers that look
at the multi-echelon setting, but still with a single firm in charge of supply risk
management (e.g., see Sobel and Babich 2012). There is a much smaller but sub-
stantial literature that studies how interactions among firms is affected by supply
risk.5 However, there are not many studies that focus specifically on the effects of
supply risk on contracting.
To see how contracting and supply risk interacts, it is useful to review the supply
chain risk management (SCRM) process. Based on ISO31000, a SCRM comprises
of five steps shown in Fig. 8.1: (1) identify risks, (2) assess and prioritize risks,
(3) prepare a mitigation plan, (4) implement mitigation plan, and (5) review results,
learn and adapt.
Let’s start with the step “Identify risks.” Risks in supply chains often stem from
the misalignment of incentives among companies, usually exacerbated by lack of
information. For example, suppliers that used lead paint in Mattel’s toys (Story and
Barboza 2007), or melamine in Menu Food’s pet food (Myers 2007), or melamine
in Baxter’s heparin (Reinberg 2008), were concerned about immediate financial
gains, the survival of their firms until the next quarter, but not about the long-term
4 Earlier work on random yield management falls into this category, see review article by Yano
and Lee (1995). More recently, a number of papers on supply disruptions and the value of such
strategies as diversification, backup production, inventory, insurance have appeared. See review
articles and chapters by Tang (2006), Tomlin and Wang (2011), Snyder et al. (2012).
5 For example, see the review chapter by Aydin et al. (2011) and papers by Yang et al. (2009,

2012), Babich and Tang (2012), Wadecki et al. (2012), Tang and Kouvelis (2011), Gümüş et al.
(2012), Chaturvedi and Martı́nez-de Albéniz (2011), Yang and Babich (2015).
148 V. Babich and Z. Yang

Fig. 8.1 ISO31000 SCRM

process

brand damage or even legal and regulatory responses. The opposite is true for their
buyers (Mattel, Menu Foods, and Baxter). This misalignment of incentives leads to
supply risk. Contracts can be used to align the supply chain firms’ incentives and to
manage the exchange of information. In the example of Chrysler, financial contracts
can be used to assess the supplier’s bankruptcy probability.
Step “Assess and prioritize” requires the collection of information to quantify
risk exposure. This information is distributed across the supply chain. For example,
to secure orders, suppliers could be reluctant to reveal risks they are exposed to.
Conversely, if the buyer provides an assistance to suppliers in need, suppliers could
exaggerate the extent of their risk mitigation needs. Contracts can govern the sharing
of information among supply chain partners.
Step “Prepare a mitigation plan” involves evaluating how mitigation actions
will affect incentives. For example, if multi-sourcing is used, incentives to com-
pete could be lower. Contracts provide both mitigation mechanisms (e.g., non-
performance penalties, making payments contingent on certain quality levels, making
payments contingent on adherence to certain manufacturing and risk management
processes) and tools for managing incentives, information, and relationships. It
appears that Chrysler did not have supplier bankruptcy in its risk mitigation plan, if
there was one. Contracts also specify how the costs and benefits of risk management
actions will be shared and who is responsible for what actions and how compliance
will be ensured.
Step “Implement and execute” includes installing actual contracts and execut-
ing them. Execution of a risk management plan may involve decisions on how scarce
resources should be allocated. For Chrysler, the ownership of the molds should have
been specified in the contract. In the famous Nokia vs. Ericsson case (Sheffi 2005),
one of the components of Nokia’s successful response to supply disruption was the
ability to secure entire existing inventories of micro-chips. Contracts can be used
8 Supply Disruptions and Procurement Contracting 149

to regulate such contingencies. The timely detection of the crisis also requires an
alignment of incentives and contracts can be used to assist that.
Step “Review, learn, and adapt” completes the risk management cycle with
lessons that can be used to manage future adverse events. Learning should hap-
pen along supply chains and contracts can govern sharing of relevant information.
Learning must happen within the firms and contracts can manage agency problems
that can surface in the process.
The effects of supply risks on contracting is manifold. Adverse events can affect
one of the contracting parties or several. Adverse events can be exogenous or be
caused by actions of contract parties. They can change incentives of the firms or
information structure. One can view the interactions between supply risk and con-
tracting from multiple angles. We will define four levels of such interactions, in the
increasing order of awareness of the contract designer about the supply risk, as illus-
trated in Fig. 8.2: (1) Contracts do not perform as intended because of a supply risk
event, (2) Contracts are adjusted to accommodate the risk mitigation steps taken by
the firms, (3) Contracts are used to manage supply risk directly, and (4) Contracts
are used to coordinate risk management activities among firms in supply chains. The
following are a few examples from practice.

Fig. 8.2 Four levels of inter-

action between supply risk
and contracting

(1) Contracts do not perform as intended because of a supply risk event. In the
summer of 2010 Russian government instituted a wheat export ban (Kramer 2010).
That meant that Russian wheat traders could not fulfill their contractual obligations
to the buyers outside of the country. According to Clyde & Co (2012), most grain
from that region is sold on a standard GAFTA forward grain contract.6 Such con-
tract contains a “prohibition clause” that states that a seller is excused from perfor-
mance of its obligations if it is prohibited from exporting by an act of government.
Therefore, buyers of wheat not only suffered from the counterparty default, but also

6 Source: http://www.gafta.com/contracts.
150 V. Babich and Z. Yang

could not claim damages from their Russian suppliers. Actions of government are
often unpredictable. For example, Ukrainian government decided to take a different
approach during 2010 wheat crisis. Instead of officially banning wheat exports, it
slowed down customs approval process and artificially created shortages of wag-
ons and containers. Thus, Ukrainian wheat suppliers failed to meet their contractual
obligations, but could not claim an exemption from the contract breach penalties.
(2) Contracts are adjusted because of the risk mitigation steps taken by the firms.
In March of 2011, the Toyota car production in Japan dropped 63 % following the
magnitude nine earthquake and tsunami events (Sasahara 2011). Toyota was af-
fected by the disaster particularly badly because it single-sources many parts from
suppliers that had production facilities in the affected areas. According to reports,
the delivery of more than 1200 different parts to Toyota’s were affected. Reflecting
on the event 6 month later, Shinichi Saki, Toyota’s manager in charge of bring-
ing the automaker’s supply chain back on-line, commented to “Automotive News”
that the new strategy for Toyota is to make sure that in the future more than one
factory will be able to supply every Toyota part (Greimel 2011). However, chang-
ing from single-sourcing to diversification changes competition between suppliers
(Yang et al. 2012) and relationships between Toyota and suppliers, which will be
reflected in the contracts.
(3) Contracts are used to manage supply risk. In contracting with strategic suppli-
ers when developing the 787 Dreamliner airplane, Boeing adopted contracts featur-
ing delay payment clauses (Tang and Zimmerman 2009). Such clauses specify that
the suppliers will not receive payment until the first 787 is successfully delivered to
the customer and were intended to enable Boeing to share the project delay risk with
its suppliers. This is an example of the use of contracts for supply risk management.
Ironically, these clauses are now believed to have contributed to delays in Dream-
liner’s completion. Studies show that the delayed payment discouraged suppliers
from making efforts to deliver sooner than other suppliers.
(4) Contracts are used to coordinate risk-management activities among firms
in supply chains. Such coordination is clearly beneficial, and it plays an essential
role in achieving the highest marks on the Supply Chain Risk Maturity tool from
the Supply Chain Risk Leadership Council (see example output in Fig. 8.37 ). This
model ranks the supply chain according to five categories: Leadership, Planning,
Implementation, Evaluation, and Improvement. Each category is broken into mul-
tiple subcategories and the assessment is done on a scale from 1 to 5. In each of
the subcategories, the difference between grades 4 and 5 (the highest) is based on
whether there is integration across the supply chain with respect to supply chain
risk management. Contracts help to articulate how such integration happens, specify
roles and responsibilities, and divide benefits and costs.

7 Source: http://www.scrlc.com/articles/SCRLCMaturityModel-2April2013.xlsx.
8 Supply Disruptions and Procurement Contracting 151

Fig. 8.3 Output from the

Supply Chain Risk Maturity
Model

8.3 Research: Standard Assumptions to be Reexamined

in the Presence of Supply Risk

There is a set of standard assumptions that OM and economics literatures on pro-

curement contracts take for granted.8 In the presence of supply risk, some of these
assumptions might no longer apply. In this subsection, we shall discuss a few of
these assumptions and point out alternative assumptions to be considered.

Assumption 1. Once a contract is signed, it will be executed as agreed on.

Why this assumption can be violated should be pretty obvious. A disruption to

supply chains can stop or delay deliveries of products and services. (Just recall the
wheat shortage crisis in 2011, or Chrysler’s case with Plastech.) The suppliers’ non-
delivery is particularly relevant if they have already received the full payments from
the buyer and thus have less incentive to fulfill their part of the contract. For exam-
ple, a simple contract (X, q) specifying the full payment X, to be made up-front, and
quantity q, to be delivered later, is vulnerable to such non-compliance risk. How
large this risk is depends on the relationship between the buyer and the seller. If
the seller is expecting to continue doing business with the buyer in the future, the
threat of losing future business may provide the required compliance incentive. If
not, then in some circumstances contracts can be enforced by courts. However, re-
lying on legal solution is not always feasible or cost efficient. For example, when
buying from overseas suppliers, the buyers might not have much recourse in courts
of the suppliers’ home countries. Court proceedings are expensive and court cases
can take many years and be subject to political uncertainties. Even in high-visibility
cases of product adulteration, lawsuits against overseas suppliers are rarely success-
ful (Yang 2007).
Another solution to the supplier’s non-performance is to add a clause to the
contract that ties payments with the quantity actually delivered to the buyer.
8A synthesis of economics literature is given in Bolton and Dewatripont (2005) and the same for
OM literature is in Cachon (2003).
152 V. Babich and Z. Yang

For example, the contract can take form (X, q, p), where p is the penalty per unit
by which the supplier falls short from the promised amount. Such non-performance
clauses are relatively common in procurement contracts (e.g., see Baiman et al.
2000). But, this solution is not without its faults. If the supplier has already been
paid, “clawing back” a part of that payment as a non-performance penalty can be
difficult because the supplier might no longer be in a financial position to pay it, or
because the supplier might be unwilling to pay it (Babich and Tang 2012). More-
over, for some disruption types, the suppliers might claim that a “force majeure”
event has happened, liberating them from contractual obligations. This brings us to
another common assumption.
Assumption 2. The timing of payments is irrelevant.
Consider one of the simpler contracts studied in the OM literature—the whole-
sale price contract (w, q), where w stands for the price per unit and q for the buyer’s
requirement. If the supplier and buyer are perfectly reliable, there is no difference
whether the buyer pays the supplier before the delivery or after. It is only a matter
of accounting for the time value of money and discounting for future cash flows.
With supply risk, however, it becomes important to decide who bears that risk.
Paying the supplier upfront shifts the risk of supply disruption to the buyer. If the
supplier defaults (e.g., goes out of business), the buyer loses both the components
it was expecting and the payment. Conversely, making payments after the delivery
shifts the risk to the supplier. The supplier will have invested resources in produc-
tion. If the supplier fails to deliver components to the buyer due to disruption, the
supplier might not recover the initial production investment. Babich (2006) dis-
cusses how supply disruption risks are reflected in the equilibrium prices of the
wholesale price contracts, depending on the timing of payments.
Furthermore, the buyer could be subject to disruption as well, such as bankruptcy.
By the time the products are ready to be delivered (after the supplier’s production),
the buyer might be out of business. The supplier could lose the investment in produc-
tion and receive only the salvage value for the products produced. To make matters
worse, this salvage value can be negatively correlated with the buyer’s default. For
example, the buyer might be out of business because the demand for the products
has declined. This means that other buyers in the same market are not interested in
purchasing the products from the supplier either.
Similar concerns apply to other procurement contracts, such as the buyback
contract, quantity discount contract, and revenue sharing contract. Under these
contracts, the exchange of goods and money happens over time, making
those contracts susceptible to the counterparty risk. For example, consider a buy-
back contract (w, q, b), which allows the buyer to return items to the supplier at
the buyback price b. By design, there is a time-gap from the moment when the
buyer receives and pays w per unit for the supplier’s goods till the moment when the
buyer invokes the buyback clause and returns unsold items to the supplier. Because
the supplier’s financial status may deteriorate during the selling season, the supplier
may become insolvent and thus unable to fulfill its buyback obligation. To make this
8 Supply Disruptions and Procurement Contracting 153

scenario more concrete, consider a situation where the buyer is faced with random
demand D and collects the revenue r per unit. The buyer’s expected profit is:
 
E r min(D, q) − wq + b(q − D)+ 1(supplier exists) . (8.1)

The indicator 1(supplier exists) represents the random event that supplier bankruptcy
may occur. Assume that E 1(supplier exists) = δ and that the supplier’s bankruptcy
event is independent of the buyer’s random demand D. On the bright side, the co-
ordinating condition for this problem is a slight variation of the traditional buyback
coordinating condition without accounting for the supplier’s bankruptcy risk.9 The
coordinating condition in the presence of supplier risk is:
r−w r−c
= . (8.2)
r − bδ r
However, observe from this condition that if the buyback contract is constructed
without accounting for the counterparty risk (i.e., presuming δ = 1), it will under-
estimate the value of b needed for coordinating the supply chain.
Unfortunately, accounting for the counterparty risk is not so simple, if the sup-
plier’s survival and demand are correlated. For example, a low demand means that
the supplier’s product is not popular and the supplier is likely to declare bankruptcy.
But, this is precisely the occasion when the buyer needs the buyback clause the most.
Therefore, the more correlation between low demand and the supplier’s bankruptcy,
the less valuable the buyback option is to the buyer, and the less the buyback contract
is effective in coordinating the supply chain.

Assumption 3. The choice of the contract does not affect the total risk in the system.

Typically, the choice of the contracts does not affect the demand uncertainty
(which is assumed to be exogenously specified). In contrast, certain forms of sup-
ply uncertainty (e.g., product quality; see Babich and Tang 2012; Tang and Babich
2014) depend on the decision makers’ incentives and on the contracts used to man-
age those. For example, on August 2, 2014, an explosion occurred at the facility of
Zhongrong in Kunshan, China. The explosion caused 64 deaths and disrupted the
wheel supply to Dicastal, a global supply chain partner of GM (Yan 2014). Anec-
dotal evidences indicate that the managers of Zhongrong refused to install sufficient
safety measures even after they became aware of the looming explosion risk. One
of the Zhongrong’s managers later argued that they were under pressure to make
on-time delivery to Dicastal. This suggests that the heightened supply risk was due
to harsh contracting practices of the U.S. automobile manufactures. The big U.S.
automobile manufacturers relentlessly pressure their suppliers to reduce costs while
meeting stringent delivery targets. This forces suppliers to cut corners in choosing
where to source the materials, quality control, and labor practices.
Even a contracting scheme that is designed to reduce supply risk may increase
risk. Recall that in the Boeing example discussed earlier, the delayed-payment

9 Such condition ensures that the buyer orders the system’s optimal quantity.
154 V. Babich and Z. Yang

clauses created disincentives for the suppliers to deliver ahead of other suppliers.
As yet one more example, consider again the buy-back contract (w, q, b). Investors
of the supplier firm might deduce from market signals that if the demand for the
product is not high, then there will be a large future expense on buyback. Before
this happens, investors of the supplier firm may pull out, possibly leading to the
supplier’s bankruptcy. This observation leads to the next questionable assumption.

Assumption 4. The decisions are made by firms, which are uniform, monolith
entities.

Firms are not uniform entities, but collectives of decision makers with different
objectives, which are often conflicting each other (classical references are Alchian
and Demsetz 1972; Jensen and Meckling 1976; Myers 1977; Myers and Majluf
1984). Procurement contracts affect these decision makers differently, so that the
choice of the contract can bring to the surface agency problems within firms. For
example, there exists a fundamental agency problem between the equity holders
and debt holders of a firm and this affects procurement decisions (e.g., see Chod
and Zhou 2013). The debt holders have incentives to reduce the risk of the firm’s
cash flows, while the equity holders have incentive to increase risk. This conflict can
affect the firm’s choice of supplier, risk-mitigation schemes, and contracts.
A major part of contract theory is dedicated to understanding interaction among
firms under asymmetric information. Because this is such an important topic, we
shall dedicate the entire next section to discussing it.

8.4 Insights on Contracting Under Asymmetric Information

in the Presence of Supply Risk

Asymmetric information about risk is ubiquitous. The buyer may not have perfect
visibility into the supplier’s safety practices, financial status, operational conditions,
and supply chain structure. For example, in the case of Menu Foods’ supply disrup-
tion (Yang et al. 2009), a first-tier supplier outsourced a part of its production to a
second-tier Chinese supplier with low production cost and high risk of unsafe pro-
duction practices. The first-tier supplier did so without notifying the buyer, Menu
Foods, creating an information gap between the two firms. This left Menu Foods
unaware of the risk of supply disruption caused by the unsafe ingredient used by the
second tier supplier.
The risk management process (see Fig. 8.1) involves acquisition, processing, and
use of information about the suppliers’ risks. Without knowing the true risk profile
of the suppliers, the buyer may create risk management plans that are either inef-
fective or inefficient. The information gap creates opportunities for the supplier to
exploit it. For example, a supplier with high risk may intentionally under-report its
risk level to win the buyer’s contract. Conversely, a supplier already selected by the
buyer may exaggerate the production difficulty to bargain for an increased payment.
8 Supply Disruptions and Procurement Contracting 155

Traditional procurement contracting focused on managing asymmetric informa-

tion about costs. It is reasonable to ask: Can results from this literature be applied
to problems with asymmetric information about risk? The answer is: sometimes.
In general, asymmetric information about costs and asymmetric information about
supply risk are different. In this section we shall analyze these differences.

8.4.1 Benchmark: Screening for Supplier Cost

Consider a classical procurement problem10 with a buyer (we will refer to the buyer
as she) purchasing from a supplier (we will refer to the supplier as he). The buyer
faces demand D (assume it is constant for the sake of this example) with revenue r
per unit (assume this is constant as well). The buyer’s revenue has the newsvendor
form r min{q, D}.
The supplier knows his production cost c, but the buyer only has a belief about it.
Her beliefs are consistent with the distribution of suppliers in the economy, where
fraction α of the suppliers has cost cH (we shall refer to these as high-type suppliers)
and the rest have cost cL > cH (we shall refer to these as low-type suppliers).
A fundamental and general result, called the Revelation Principle, allows us to
restrict the search for the optimal contracts to direct and truth-telling contracts only
(if we are interested in the game outcome but not implementation, and if certain
conditions are satisfied; see Myerson 1979). Specifically, we can consider these
contracts customized to the supplier’s type: (X, q)(t) for t ∈ {H, L}.
The following simple calculations illustrate how suppliers might derive value
from the informational gap and what the buyer can do about it. If the buyer could
observe the type of the supplier, she would offer contracts (cH D, D) to the high-
type supplier and (cL D, D) to the low type supplier and extract the entire system
profit. But, as is well-known, when the supplier’s type is not observable, the buyer
will need to manage the supplier’s strategic behavior of misrepresenting his type.
Specifically, if contracts (cH D, D) and (cL D, D) are offered, because cL D > cH D,
the high-type supplier has an incentive to claim to be low type to receive higher
payment.
The buyer has two choices. She can either offer the same contract (cL D, D) reg-
ardless of the type and thus pay extra cL D − cH D to the high-type supplier. With
this payment the high-type supplier does not have an incentive to misrepresent his
type. Thus, the buyer effectively pays the supplier to reveal his type truthfully. This
payment is known as the information cost to the buyer. Accounting for the frac-
tion of high-type suppliers in the economy, this comes to an expected payment of
α (cL D − cH D), referred to as the information rent to the supplier.
Another alternative is for the buyer to stop offering a contract to the low-type sup-
plier altogether, forcing the high type to accept his contract (cH D, D). But the buyer
loses out on the expected profit (1 − α )(r − cL )D she could have earned working

10 For example, refer to the model considered in Sect. 2.1 of Laffont and Martimort (2002).
156 V. Babich and Z. Yang

with a low-type supplier. Depending on which loss is dearer, the informational rents
to the high-type or the loss of profit from not working with the low-type, the buyer
chooses the appropriate contract menu. Specifically, the buyer would choose to stop
working with the low-type supplier if

(1 − α )(r − cL ) < α (cL − cH ). (8.3)

The discussion above can be formalized in the mechanism design program (8.4),
which describes the process of choosing the optimal contracts:
    
max α r min(q(H), D) − X(H) + (1 − α ) r min(q(L), D) − X(L)
X(·)≥0,q(·)≥0

subject to: π (H |H) ≥ π (L|H) and π (L|L) ≥ π (H |L) (8.4)

(i.e., incentive compatibility)
π (H |H) ≥ 0 and π (L|L) ≥ 0 (i.e., individual rationality),

where π (s|t) = X(s) − ct q(s) is the expected profit of the supplier who reports his
type to be s ∈ {H, L} when his true type is t ∈ {H, L}.
This program can be solved following the standard solution procedure of mecha-
nism design (for example, see Laffont and Martimort 2002) by separating the prob-
lem into two subproblems for the high- and low-type contracts. To find the optimal
q(H) the buyer solves program (8.5):
 
α max r min(q(H), D) − cH q(H) . (8.5)
q(H)≥0

Observe that this is the problem of the buyer when the supplier is known to have high
type. Thus, there is no distortion in the order quantity with the high-type supplier
to account for asymmetric information. To find the optimal q(L), the buyer solves
program (8.6):
   
max (1 − α ) r min(q(L), D) − cL q(L) − α (cL − cH )q(L) . (8.6)
q(L)≥0

From the optimal q(H) and q(L), the optimal payments can be computed using these
equations:

X(H) = cH q(H) + (cL − cH )q(L) and X(L) = cL q(L). (8.7)

The term of α (cL − cH )q(L) in program (8.6) represents the information rent
(which we mentioned above) the buyer pays in expectation for revealing the sup-
plier’s true cost information. In particular, if q(L) = 0 the buyer does not pay in-
formation rent, but she misses an opportunity of making profit by working with a
low-type supplier.
8 Supply Disruptions and Procurement Contracting 157

8.4.2 Equivalent Model with Risk: What Assumptions

are Necessary?

Consider a variation of the classical contracting problem by assuming that there

is asymmetric information about supplier risk, instead of asymmetric information
about supplier cost. Specifically, assume that the outcome of the supplier’s produc-
tion is uncertain with Bernoulli random production yield ρ (θ ), where ρ (θ ) = 1
with probability θ and ρ (θ ) = 0 with probability 1 − θ . The supplier knows the
value of θ , but the buyer only knows its distribution, that is θ = h (i.e., the high
(H) type) with probability α and θ = l < h (i.e., the low (L) type) with probability
1 − α.
We will make the following important assumption.
Assumption 5. A supplier with unreliable production can make an infinite number
of production attempts until the production target is successfully met. The outcomes
of attempts are independent.
In the benchmark model in Sect. 8.4.1, the supplier is passive. He either accepts
terms from the buyer and follows the proposed production plan or not. Here, we
need to model the supplier’s decision on how much to produce z (possibly, he may
not produce at all) and how much to deliver y.
As we discussed earlier, to provide economic incentives for the supplier to de-
liver the contract must include a term contingent on the delivery. For instance, we
will consider contract (X, q, p) where terms (X, q) have the same meaning as in
the benchmark model and p represents penalty the supplier pays for every unit the
delivery that is short from the buyer’s requirement q.
With any positive penalty p > 0, if the supplier has units to deliver, he will do
so. The question we need to answer is whether the supplier will produce or not
in the first place. That is, knowing penalty p, the supplier has to solve a simple
dynamic control (optimal stopping) problem regarding whether to produce or to give
up and incur penalty. The problem is dynamic, because if the supplier’s production
fails it will face the same decision in the next period. Because production yield is
Bernoulli and production attempts are independent, we can assume that the size of
the production run is z = q (i.e., the supplier tries to produce exactly what the buyer
requested until a success or the supplier gives up on the entire order). Then, the
total production plus penalty cost per unit to the supplier, denoted as C, satisfies the
following Bellman DP recursion:
 
Cq = min pq, cq + (1 − θ )Cq . (8.8)

From this equation it follows that for p ≥ c/θ , the supplier will keep trying to
produce the required units until he succeeds. In this case, the supplier’s expected
cost for producing one good unit is C = c/θ . For p < c/θ the supplier will not even
try once and will simply pay the penalty.
Observe that even though there is supply risk in the model, under Assumption 5,
this risk does not affect the buyer, as long as the penalty p is high enough. The
158 V. Babich and Z. Yang

supplier’s optimal profit is π (X, q, p) = X − (c/θ )q. What is interesting is that by

defining cL = c/h and cL = c/l we recover the benchmark model in Sect. 8.4.1,
with the asymmetric information about costs rather than risks. That is, under the
assumption of infinite production attempts, the unreliable supplier behaves as if its
production is perfectly reliable with the unit production costs c/h or c/l. The buyer’s
contract design problem is the same as that of the benchmark model with cH = c/h
and cL = c/l. The buyer faces a fundamental tradeoff: she will stop ordering from
the low-type supplier if and only if

(1 − α )(r − c/l) < α (c/l − c/h). (8.9)

8.4.3 Screening for Supplier Risk: Why the Model

and the Insights are Different?

Clearly, the assumption of the supplier trying infinitely many times to produce a
good unit (i.e., Assumption 5) is not practical. There are production lead-times and
delivery deadlines. Therefore, we will make a more realistic assumption:
Assumption 6. The supplier has only one production attempt.
Again, we will consider contract (X, q, p) where terms (X, q) have the same
meaning as in the benchmark model and p represents the penalty the supplier pays
for every unit the delivery that is short from the required amount q. Let c be the sup-
plier’s unit production cost (the same cost for both types). The supplier of type θ ,
given contract (X, q, p), solves optimization problem (8.10):
 
π (X, q, p| θ ) = max X − cz − Eρ (θ ) max [p(q − y)+ ] . (8.10)
z≥0 0≤y≤ρ (θ )z

The supplier’s optimal production quantity z∗ (θ ), delivery quantity y∗ (θ ) and profit

π (X, q, p| θ ) are easily derived and are given in Table 8.1.11 As the penalty p
increases to exceed the effective expected production cost per unit c/θ , the sup-
plier commits to producing for the buyer’s requirement q. The supplier’s profit is
maximized at p = c/θ and equal to X − (c/θ )q.

Table 8.1 The optimal solution profit of the supplier from the production problem (8.10)
Parameter values z∗ ( θ ) y∗ ( θ ) π (X, q, p| θ )
p < c/θ 0 0 X − pq
p ≥ c/θ q ρ (θ )q X − cq − (1 − θ )pq (which is ≤ X − c/θ q)

11 Adapted from Proposition 1 of Yang et al. (2009).

8 Supply Disruptions and Procurement Contracting 159

Again, one may suspect that transformation cθ = c/θ will yield the benchmark
model without risk in Sect. 8.4.1. Recall that in that model, the supplier’s profit
is X − cq.
One can argue that there is a difference between these models, because under sup-
ply risk the delivery depends on the value of the penalty. But, this difference is not
essential at this point, because we can modify the benchmark model in Sect. 8.4.1
to incorporate penalty for non-delivery in a trivial way:
 
π (X, q, p) = max X − cz − max [p(q − y)+ ] . (8.11)
z≥0 0≤y≤z

As long as p ≥ c, production of q units and full delivery will happen. We obtain the
supplier’s optimal profit to be X − cq. Therefore, we can assume that p = c.
To summarize, the supplier’s problem is similar to that in the benchmark model.
Now let’s consider the buyer’s problem. As before, it is instructive to first discuss
what the optimal contract would look like, if the buyer knew the supplier’s type. The
optimal contract terms (X, q, p) and the supplier’s and buyer’s profits are presented
in Table 8.2.12

Table 8.2 The optimal contract and the supplier’s and buyer’s profits, when the buyer knows the
supplier’s risk type
Revenue r p q X Supplier’s profit Buyer’s profit
r > c/θ Any p ≥ c/θ D [c + (1 − θ )p]D (≥ c/θ D) 0 θr −c
r ≤ c/θ 0 0 0 0 0

From Table 8.2, it is possible that some suppliers have production that is too
unreliable and the buyer would not work with them if she knows who they are. We
shall assume that suppliers are sufficiently reliable, so that the buyer would work
with both types under symmetric information.
Under asymmetric information, we observe that, as before, if the supplier selects
the contract intended for his type θ , he will make zero profit. However, if the high-
type supplier type selects the contract intended for the low type, he will make profit
(h − l)p(l)D > 0. From this expression it follows that reducing penalty p(l) reduces
high-type supplier’s incentive to lie and his profit. Therefore, the buyer will set the
penalty at the lowest level that provides just enough incentives for the low-type
supplier to deliver goods. That is, p(l) = c/l.
As in the benchmark model in Sect. 8.4.1, the buyer can accept the fact that the
high-type suppliers will lie and pay these suppliers information cost (h − l)(c/l)D.
With fraction α of high-type suppliers in the economy this generates the ex-
pected value (informational rents) of α (h − l)(c/l)D. Alternatively, the buyer can
stop working with the low-type suppliers. This will eliminate information rents
to the high-type supplier, but the buyer will be giving up the expected profit of
(1 − α )(lr − c)D she would have earned working with the low-type suppliers. Thus,
the condition for the buyer to stop working with the low-type suppliers in order to
12 The results in this table is obtained by applying b = ∞ to Proposition 2 of Yang et al. (2009).
160 V. Babich and Z. Yang

avoid paying information rents to the high type ones under asymmetric information
about supply risk is:  c
(1 − α )(lr − c) < α (h − l) . (8.12)
l
Recall that the condition for such tradeoff in the benchmark model with asym-
metric information about the supplier’s cost is [replicated from (8.3)]: (1 − α ) ·
(r − cL ) < α (cL − cH ). Using cθ = c/θ we write condition (8.3) as
 c l
(1 − α )(lr − c) < α (h − l) . (8.13)
l h
This is almost like condition (8.12) in the model with risk, but not quite, because
the value l/h < 1 on the right-hand side.
Why is there a difference? In the model with asymmetric information about costs
or in the model with asymmetric information about risk under Assumption 5, the
effect of asymmetric information is only on the procurement costs side and the rev-
enues are not affected. In the model with asymmetric information about supply risk
under Assumption 6, the asymmetric information about risk also affects the buyer’s
revenue. This is an important difference between the two models.
Why does it matter? The optimal contract and the actions of the players under
the optimal contracts are different under Assumptions 5 and 6. Compare conditions
(8.12) and (8.13) and recall that l/h < 1. If the buyer decides to stop working with
the low-type supplier in the model under Assumption 5, she will also stop work-
ing with the low-type supplier in the model under Assumption 6. But, under the
condition  
c l c
α (h − l) ≤ (1 − α )(lr − c) < α (h − l) , (8.14)
l h l
in the model under Assumption 5 the buyer will pay informational rent but continue
working with the low-type supplier, whereas in the model under Assumption 6 the
buyer will stop working with the low-type supplier.
The following is the formalized presentation of the buyer’s contract design prob-
lem and the resulting optimal contract. Let’s denote the supplier’s optimal profit of
reporting risk type θ  as: π (θ  | θ ) = π ((X, q, p)(θ  )| θ ). The buyer’s mechanism
def

design problem is program (8.15):

!   "
α rE min(y∗ (h), D) − X(h) + p(h)Eρ (h) [q(h) − y∗ (h)]+
max 
X(·)≥0, q(·)≥0, p(·)≥0 + (1 − α )[rE min(y∗ (l), D) − X(l) + p(l)Eρ (l) [q(l) − y∗ (l)]+

subject to: π (h|h) ≥ π (l |h), π (l |l) ≥ π (h|l); π (h|h) ≥ 0, π (l |l) ≥ 0.

(8.15)
The solution to this problem is presented in Table 8.3.13
Let’s take stock of what we have learned in this section so far. We have shown
which assumption (i.e., Assumption 5) makes models with asymmetric information
about risk and asymmetric information about costs equivalent. This assumption of

13 The results can be derived by applying b = ∞ and cL = cH = c to Proposition 3 of Yang et al.

(2009).
8 Supply Disruptions and Procurement Contracting 161

Table 8.3 The solution to the model with supply risk (8.15)
If (1 − α )(lr − c) > α [(h − l)(c/l)], then
Penalties p(h) ≥ c/l p(l) = c/l
Quantities q(h) = q(l) = D
Payments X(h) = [h(c/l) + (1 − h)p(h)]D X(l) = (c/l)D
Supplier profits π (h|h) = [(h − l)(c/l)]D π (l |l) = 0
Buyer profit α [h(r − c/l)]D + (1 − α )[l(r − c/l)]D
If (1 − α )(lr − c) ≤ α [(h − l)(c/l)], then
Penalties p(h) ≥ c/h p(l) = 0
Quantities q(h) = D q(l) = 0
Payments X(h) = [c + (1 − h)p(h)]D X(l) = 0
Supplier profits π (h|h) = π (l |l) = 0
Buyer profit α (hr − c)D

infinitely many production attempts is not practical and we have shown that under
a more practical assumption (i.e., Assumption 6) equilibrium actions of the firms in
the two models diverge.
But this is only the tip of the iceberg of the differences between the two frame-
works. As we discussed earlier, in the presence of supply risk, the buyer can use var-
ious risk mitigation tools to control the effects of risk on her revenues. For example,
the buyer can use backup production, either internally or at the supplier. The buyer
can diversify risk by working with multiple suppliers. The buyer can outsource pro-
duction decision to a knowledgeable Procurement Service Provider (PSP). These
risk-mitigation tools affect strategic interactions between the buyer and the supplier
and among suppliers. These tools usually have no place in the procurement models
with asymmetric information about costs (or are used differently). Therefore, the in-
sights for the model with supply risk are unique. The complexity of the problem also
changes drastically, when such tools are added to the model. For example, with the
use of a PSP the buyer’s procurement model becomes intrinsically multidimensional
(see Yang and Babich 2015).
In the following subsections we will discuss the role of various risk-mitigation
tools, their effect on the strategic interactions among firms, and contracting
outcomes.

8.4.4 Risk Management Tools and the Value of Information

We will discuss two commonly used and studied risk-mitigation tools: backup
production and multi-sourcing. Let’s start with backup production.
162 V. Babich and Z. Yang

8.4.4.1 Backup Production

While Assumption 5 of infinite production attempts is too unrealistic, perhaps

Assumption 6 of a single attempt is too strict. Assumption 7 could be viewed as
a compromise between the two:

Assumption 7. The supplier has access to a perfectly reliable backup production

with cost b.

Now the buyer has a more refined control over supply risk. By choosing penalty
term of the contract with the supplier, the buyer can control whether she wants the
supplier to produce at all, produce once using regular production (which may fail),
or produce using backup production (if regular production fails). But, the contract
now has multiple roles. It is used to screen supplier’s private information. It is used
to maximize the profit of the buyer. It is also used to control the use of the risk
management tool—namely backup production.
These multiple functions of the contract make the design problem more complex
(see Yang et al. 2009) because of the conflicts between these functions. For exam-
ple, Yang et al. (2009) show that informational cost that contracts manage depends
on whether backup production is used or not and on the backup cost b. Conse-
quently, sometimes the buyer would rather not have the supplier using backup pro-
duction, because this increases the informational rent to the high-type supplier. Put
another way, asymmetric information creates additional costs on the use of backup
production. Therefore, in the model with asymmetric information backup produc-
tion is less likely to be used. Figure 8.4 illustrates the effect of asymmetric informa-
tion on the use of backup production. In regions (II) and (III), the buyer will induce
the low type supplier to run backup production under symmetric information, but
will not under asymmetric information.
Having the backup production option may make information more or less valu-
able for the buyer. The solid line in Fig. 8.5 illustrates the value of information for
the buyer in relation to b at the revenue r = r > r̄, where both r and r̄ are marked
in Fig. 8.4. The dashed line represents the value of information in the absence of the
backup option. As Yang et al. (2009) show, when the backup cost is very large, the
buyer will never consider using backup production. The backup option is irrelevant
and thus has no effect. To the other extreme, when the backup cost is very small,
the buyer will use the backup option regardless of the supplier’s type. The backup
option makes the supplier’s information less relevant and thus reduces the value of
information. Between the two extreme cases, the backup option with a moderate
cost (i.e., around b = b , which is also marked in Fig. 8.4) increases the value of
information to be the greatest. At such backup cost, the uninformed buyer struggles
the most in balancing between information cost vs. risk management benefit of us-
ing the backup option with the low-type supplier. Information about the supplier’s
type is most beneficial for the buyer.
Backup production mitigates supply risk and is valuable for the buyer (even if it is
used less under asymmetric information). Interestingly, the value of backup produc-
tion can be higher under asymmetric information than under symmetric information.
8 Supply Disruptions and Procurement Contracting 163

Fig. 8.4 Effect of information asymmetry on the use of backup production (adapted from Fig. 3 of
Yang et al. 2009 with cH = cL = c)

Fig. 8.5 The value of information for the buyer

Specifically, when the backup cost is small, the use of backup production under
asymmetric information has an additional benefit of reducing the informational rent
to the high type. This is because the low-type supplier running backup production
becomes so cost-efficient that the high-type supplier has a weak incentive to mimic
the low type.

8.4.4.2 Multi-Sourcing

Multi-sourcing is another well-studied risk-management tool (Babich et al. 2012).

The buyer may decide to place orders with several suppliers, if there is significant
164 V. Babich and Z. Yang

risk that one supplier may experience a disruption. Because of the need to manage
risk, the buyer deviates from the simple one-supplier procurement framework. Thus,
contracts need to account for the presence of other suppliers and more importantly
for the strategic interactions among those suppliers.
A simple example from Yang et al. (2012) illustrates this point. Suppose the
buyer can commit to ordering from only one supplier from several available for a
particular product. Then, to win the buyer’s business these suppliers will fiercely
compete with each other. This competition will erode away supplier’s informational
rents, and the buyer would not have to alter contracts too much from the first best
to control informational costs. However, in the presence of supply risk, the buyer
might want to place orders with several suppliers. Thus, each supplier will win a
smaller share of the business and probability of winning increases. This weakens
the competition among suppliers, and the contracts must work harder to control
informational costs.
Therefore, there is a tension for the buyer to design contracts that would either
promote supplier competition or use diversification to manage supply risk. The
buyer’s contracting tradeoff with low-type supplier 1 and high-type supplier 2 is
illustrated in Fig. 8.6. The solid line represents the buyer’s benefit of competition,
which is made up of the benefit of selecting the better supplier and reduction of inf-
ormation cost. The dashed line represents the buyer’s diversification benefit, which
arises from the reduced risk of no supply. As the cost of disruption (i.e., the rev-
enue r) increases to surpass the threshold, the buyer prefers diversification over
supplier competition.

Fig. 8.6 The buyer’s con-

tracting tradeoff when the
two suppliers are high- and
low-type (adapted from Fig. 4
of Yang et al. 2012)

The loss of competition benefits due to diversification under asymmetric infor-

mation is greater than under symmetric information, because in the former case
competition also helps reduce informational cost. Therefore, as Yang et al. (2012)
show, under asymmetric information diversification will be used less than under
symmetric information.
8 Supply Disruptions and Procurement Contracting 165

8.5 Multiple Dimensions of Asymmetric Information

It is very easy to come up with a model that is too complex to solve. Informa-
tion asymmetry about supply risk complicates the design of the contract menu by
complicating the supplier’s incentive of misrepresentation. Specifically, the supplier
has two dimensions of private information: cost and risk. To fully reveal the sup-
plier’s private information, the buyer needs to specify a contract for each possi-
ble combination of (c, θ ). In other words, the buyer designs the contract menu
of (X, q, p)(c, θ ). Let the supplier’s profit be π (c , θ  |c, θ ), when it has true cost
and reliability (c, θ ) but reports (c , θ  ) and receives the contract of (X, q, p)(c , θ  ),
given that the supplier will run production and recourse optimally. The buyer’s con-
tract design problem is represented by the following program.
! "
rEρ (θ ) min(y∗ (c, θ ), D) − X(c, θ )
max E(c,θ )
X(·,·)≥0, q(·,·)≥0, p(·,·)≥0 + p(c, θ )Eρ (θ ) [q(c, θ ) − y∗ (c, θ )]+
(8.16)
subject to: for all (c, θ ) and (c , θ  ), π (c, θ |c, θ ) ≥ π (c , θ  |c, θ )
and π (c, θ |c, θ ) ≥ 0.

In general, when the agent has more than one dimension of private information,
the mechanism design problem suffers from increased complexity in maintaining
incentive compatibility. Consider the simplest scenario in which the supplier’s cost
c can be one of two values cH or cL , and the supplier’s reliability θ can be h or l.
In combination, there are a total of four types of suppliers: (cH , h), (cH , l), (cL , h),
and (cL , l). Each supplier type has three possible ways to misrepresent itself. To
make the contract menu fully incentive compatible, one must specify as many as 12
incentive compatibility constraints.
The complexity may be reduced, and thus the tractability may be retained, if there
is a strong correlation between the cost and reliability of the supplier. For example,
Yang et al. (2009) consider the situation where the high reliability type has cost cH
and the low reliability type has cost cL ; moreover, the supplier with high reliability
has a lower expected production cost per unit, that is, cH /h ≤ cL /l. Under these
assumptions, the contract design problem is fully tractable.
This trick does not work, however, if one is to study the use of a Procurement
Service Provider (PSP) for risk mitigation. Buyers use a PSP for sourcing, because
it is often more knowledgeable about the suppliers than the buyers. Yang and Babich
(2015) study a supply chain that comprises of the buyer, a PSP and two ex ante iden-
tical suppliers, as shown in Fig. 8.7. Instead of dealing with the supplier directly, the
buyer contracts with the PSP, who then contracts with the suppliers. This paper as-
sumes that the suppliers and the PSP (but not the buyer) have perfect information
about the suppliers’ reliability types. Even if each supplier has one dimension of pri-
vate information with only two possible reliability types—high (H) vs. low (L), the
buyer faces a grand agent (i.e., the PSP) with as many as four possible types. Specifi-
cally, in the order of decreasing reliability, the PSP’s type can be (H,H), (H,L), (L,H)
[which is identical to (L,H)], or (L,L), where each duplet indicates the types of the
166 V. Babich and Z. Yang

Fig. 8.7 A sourcing model

with a Procurement Service
Provider (PSP) (replicated
from Fig. 1 of Yang and
Babich 2015)

two suppliers. With more than two PSP-types, the mechanism design program must
specify both local and global incentive compatibility constraints. The local con-
straints ensure that any two adjacent PSP-types [i.e., (H,H) vs. (H,L) or (L,H); and
(H,L) or (L,H) vs. (L,L)] will not mimic each other. The global constraints ensure
that any two non-adjacent PSP-types [i.e., (H,H) vs. (L,L)] will not mimic each
other. Depending on the model parameters, the global incentive constraints may be
binding at the optimal solution, leading to complicated incentive structure among
the PSP’s types. To make this program tractable using the canonical mechanism de-
sign approach, Yang and Babich (2015) had to make addition assumptions about
the model parameters.

8.6 Future Research Opportunities

In this final section we will discuss a promising new research area: coordination of
risk-management activities. As we mentioned above, the state of the art in supply
risk management is considered to be joint programs among buyers and suppliers.
Such joint programs involve coordination of disaster response plans, sharing costs of
resource deployment, and sharing information about the status of the infrastructure,
the product, the components, or the firms themselves, etc.
It might seem obvious that firms should strive to coordinate supply chain risk
activities. However, the reality is that very few firms do it. It is natural to ask why.
What obstacles are there? What can one do to remove these obstacles? How should
we think about this problem and can our knowledge of contracting help? Is there a
need for new theoretical developments?
Some obstacles to risk-management coordination are obvious. Firms’ costs of
deploying risk-management resources might not match the benefits they expect
to derive from that. For example, a supplier could bear the brunt of performing
8 Supply Disruptions and Procurement Contracting 167

thorough quality inspections, but benefit very little from the reduction of reputation
risk in an unlikely event that product defect is discovered by the consumers over-
seas. This is probably one of the factors behind many cases of product adulteration
by Chinese suppliers in the late 2000s. Contract theory should be able to help firms
resolve this misalignment of incentives. Each firm must take action ai that costs
ci ai , for i = S or B, and that affects the value of the risk management benefits for
the system ρ (aS , aB ). These benefits then must be divided between the firms, that is,
ρ (aS , aB ) = ρS + ρB .
There are geographic, legal, language, political, and cultural distances that exa-
cerbate transaction costs of coordination (Tang and Babich 2014). For instance,
US FDA investigators did not have Chinese speakers on the team and had to rely
on the translations from the factory director, when investigating a case of product
adulteration. When designing contracts, the supply chain firms need to be aware of
these transaction costs.
Other obstacles to coordination are more subtle. Firms may be reluctant to pub-
licly admit the true magnitude of the risk exposure (or the magnitude of the disaster)
out of fear that the buyer might switch to another supplier. Firms might be reluctant
to share information with their suppliers or buyers out of the fear that this will benefit
the competitors. For example, Tang and Kouvelis (2011) and Wadecki et al. (2012)
find that competition between buyers may cause them to rely less on diversification,
when diversification means sharing the same supply base.
Although there is some research that can be classified as relating to coordination
of risk-management activities, the problem presents many opportunities still. We
would like to mention three of them specifically: coordination among efforts of
multiple suppliers, coordination among multiple tiers of supply chains, and dynamic
contracting for coordination.
There have been attempts to look at the problem of coordinating actions of mul-
tiple suppliers that have supply risk consequences. For example, Mu et al. (2015)
study a problem of milk adulteration by milk wholesalers buying from multiple milk
farmers. Because information asymmetry about the quality of inputs, the suppliers
have incentive to take “free ride” of other suppliers’ higher quality, leading to poor
quality in the blended milk. To solve the problem, Mu et al. (2015) require that
the payments to individual suppliers can be deferred and tied to the quality tests of
the blended milk from all suppliers. With appropriate contract design, in equilib-
rium, farmers will not adulterate. In practice, such coordination could be difficult to
achieve, as Boeing 787 Dreamliner case demonstrates. Boeing tied the payments to
all suppliers to the latest delivery among the supplier group, hoping that the suppli-
ers will be motivated to coordinate their development efforts for fast deliveries. But
this contract failed the purpose, because it created a disincentive for an individual
supplier to deliver sooner than other suppliers.
Coordination of multiple tiers in supply chains requires looking at the nested
information structures, for example, the buyer knows who the supplier is, but not
the structure of the supply chain further up. The supplier knows his supplier, but not
suppliers further up, and so on, recursively. Such nested information structures can
be difficult to model, and they may require new theoretical breakthroughs.
168 V. Babich and Z. Yang

With multiple firms involved in the coordination of risk management, it might

be helpful to consider alternative modeling frameworks. In modeling coordination,
the reality may be better described as cooperative games (for example, Nash Bar-
gaining; see Nash 1950; Binmore et al. 1986), instead of non-cooperative games.
Therefore, a coordination problem may entail a hybrid model with cooperative and
non-cooperative actions of the firms, which is sometimes referred to as “coopeti-
tion”. Examples of hybrid models in OM include Anupindi et al. (2001) and Tian
et al. (2014), but with focus in contexts other than coordination of risk management.
Finally, in many situations of supply disruption management, contracts need to
manage timely information sharing, as new information arrives over time. This
may require tools like dynamic contracting and mechanisms design to be applied
(e.g. Zhang and Zenios 2008; Oh and Özer 2013).

References

Alchian AA, Demsetz H (1972) Production, information costs, and economic organization. Am
Econ Rev 62(5):777–795
Anupindi R, Bassok Y, Zemel E (2001) A general framework for the study of decentralized distri-
bution systems. Manuf Serv Oper Manag 3(4):349–368
Aydin G, Babich V, Beil DR, Yang Z (2011) Decentralized supply risk management. In: Kouvelis
P, Boyabatli O, Dong L, Li R (eds) Handbook of integrated risk management in global supply
chains, Chap. 14. Wiley, New York, pp 389–424
Babich V (2006) Vulnerable options in supply chains: effects of supplier competition. Nav. Res.
Logist. 53(7):656–673
Babich V, Tang CS (2012) Managing opportunistic supplier product adulteration: deferred pay-
ments, inspection, and combined mechanisms. Manuf. Serv. Oper Manag. 14(2):301–314
Babich V, Aydin G, Brunet P-Y, Keppo J, Saigal R (2012) Risk, financing, and the optimal number
of suppliers. In: Gurnani H, Mehrotra A, Ray S (eds) Managing supply disruptions, Chap. 8.
Springer, London, pp 195–240
Baiman S, Fischer PE, Rajan MV (2000) Information, contracting, and quality costs. Manag Sci
46(6):776–789
Binmore K, Rubinstein A, Wolinsky A (1986) The Nash bargaining solution in economic
modelling. RAND J Econ 17(2, Summer):176–188
Bolton P, Dewatripont M (2005) Contract theory. MIT Press, Cambridge, MA
Cachon GP (2003) Supply chain coordination with contracts. In: de Kok AG, Graves SC (eds)
Handbooks in operations research and management science, vol. 11, Chap. 6. Elsevier,
Amsterdam, pp 227–339
Chaturvedi A, Martı́nez-de Albéniz V (2011) Optimal procurement design in the presence of sup-
ply risk. Manuf Serv Oper Manag 13(2):227–243
Chod J, Zhou J (2013) Resource flexibility and capital structure. Manag Sci 60(3):708–729
Clyde & Co (2012) 50 shades of grain. Technical report. Clyde & Co, London
Gillenwater G (2008) Supplier for Chrysler files bankruptcy. http://www.lawyersandsettlements.
\penalty0com/articles/bankruptcy debt help/bankruptcy-chrysler-01889.html#.Vc54e3j92M4
Graves SC, de Kok A (eds) (2003) Handbooks in operations research and management science.
Supply chain management: design, coordination and operation, vol. 11. Elsevier, London
Greimel H (2011) Exec moves Toyota toward quakeproof supply chain. Automotive News
(September 19)
Gümüş M, Ray S, Gurnani H (2012) Supply side story: risks, guarantees, competition and infor-
mation asymmetry. Manag Sci 58(9):1694–1714
8 Supply Disruptions and Procurement Contracting 169

Jensen MC, Meckling WH (1976) Theory of the firm: managerial behavior, agency costs and
ownership structure. J Financ Econ 3(4):305–360
Kelly T (2011) UPDATE 2-Hondas Thai factory shut as floods hit Japan firms. http://www.reuters.
com/article/2011/10/12/honda-flood-idUSL3E7LC0VO20111012
Kramer AE (2010) Russia, crippled by drought, bans grain exports. The New York Times
(August 5)
Laffont J-J, Martimort D (2002) The theory of incentives. Princeton University Press. Princeton, NJ
Lohr S (2011) Stress test for the global supply chain. The New York Times (March 19)
Mu L, Dawande M, Geng X, Mookerjee V (2015) Milking the quality test: improving the milk
supply chain under competing collection intermediaries. Manag Sci 62(5):1259–1277
Myers SC (1977) Determinants of corporate borrowing. J Financ Econ 5(2):147–175
Myers R (2007) Food fights. CFO Magazine (June 1)
Myers SC, Majluf NS (1984) Corporate financing and investment decisions when firms have
information that investors do not have. J Financ Econ 13(2):187–221
Myerson RB (1979) Incentive compatibility and the bargaining problem. Econometrica 47(1):
61–74
Nash JF (1950) The bargaining problem. Econometrica 18(2):155–162
Oh S, Özer Ö (2013) Mechanism design for capacity planning under dynamic evolutions of asym-
metric demand forecasts. Manag Sci 59(4):987–1007
Reinberg S (2008) FDA finds contaminant in Baxter’s recalled heparin products. http://
consumer.healthday.com/respitory-and-allergy-information-2/misc-allergy-news-17/
fda-finds-contaminant-in-baxter-s-recalled-heparin-products-613295.html
Sasahara K (2011) Toyota car production plummets after tsunami. USA Today (April 2)
Sheffi Y (2005) The resilient enterprise: overcoming vulnerability for competitive advantage. MIT
Press, Cambridge, MA
Simchi-Levi D, Wu SD, Shen Z-JM (eds) (2004) Handbook of qantitative supply chain analysis:
modeling in the e-business era. Springer, New York
Snyder LV, Atan Z, Peng P, Rong Y, Schmitt AJ, Sinsoysal B (2012) OR/MS models for supply
chain disruptions: a review. Available at SSRN 1689882
Sobel MJ, Babich V (2012) Optimality of myopic policies for dynamic lot-sizing problems in serial
production lines with random yields and autoregressive demand. Oper Res 60(6):1520–1536
Story L, Barboza D (2007) Mattel recalls 19 million toys sent from China. The New York Times
(August 15)
Tang CS (2006) Robust strategies for mitigating supply chain disruptions. Int J Logist: Res Appl
9(1):33–45
Tang CS, Babich V (2014) Using social and economic incentives to discourage Chinese suppliers
from product adulteration. Bus Horizons 57(4):497–508
Tang SY, Kouvelis P (2011) Supplier diversification strategies in the presence of yield uncertainty
and buyer competition. Manuf Serv Oper Manag 13(4):439–451
Tang CS, Zimmerman JD (2009) Managing new product development and supply chain risks: the
Boeing 787 case. KEDGE Business School. Supply Chain Forum: Int J 10(2):74–86
Tian F, Sošić G, Debo LG (2014) Manufacturers’ competition and cooperation in sustainability:
stable recycling alliances. Available at SSRN 2459656
Tomlin B, Wang Y (2011) Operational strategies for managing supply chain disruption risk. In:
Kouvelis P, Boyabatli O, Dong L, Li R (eds) Handbook of integrated risk management in global
supply chains, Chap. 4, vol. 1. Wiley, London, pp 79–101
Tsay AA, Nahmias S, Agrawal N (1999) Modeling supply chain contracts: a review. In: Tayur S,
Ganeshan R, Magazine M (eds) Quantitative models for supply chain management, Chap. 10,
vol. 17. Springer, New York, pp 301–336
Wadecki AA, Babich V, Wu OQ (2012) Manufacturer competition and subsidies to suppliers. In:
Gurnani H, Mehrotra A, Ray S (eds) Supply chain disruptions. Springer, New York, pp 141–163
Yan A (2014) Kunshan explosion factory ignored several danger warnings, says regu-
lator. http://www.scmp.com/news/china/article/1565751/kunshan-explosion- factory-ignored-
several- warnings-says-regulator
170 V. Babich and Z. Yang

Yang X (2007) Liability lawyers struggle to pierce the Chinese curtain. The Washington Post
(July 28)
Yang Z, Babich V (2015) Does a procurement service provider generate value for the buyer through
information about supply risks? Manag Sci 61(5):979–998
Yang Z, Aydın G, Babich V, Beil DR (2009) Supply disruptions, asymmetric information, and a
backup production option. Manag Sci 55(2):192–209
Yang Z, Aydın G, Babich V, Beil DR (2012) Using a dual-sourcing option in the presence of
asymmetric information about supplier reliability: competition vs. diversification. Manuf Serv
Oper Manag 14(2):202–217
Yano CA, Lee HL (1995) Lot sizing with random yields: a review. Oper Res 43(2):311–334
Zhang H, Zenios S (2008) A dynamic principal-agent model with hidden information: sequential
optimality through truthful state revelation. Oper Res 56(3):681–696
Chapter 9
Contracting for Information Acquisition

Guoming Lai and Wenqiang Xiao

Abstract This chapter considers a supply chain scenario where a producer employs
a seller to sell a product over a single sales season. The seller is engaged in two
types of activities, gathering market information and generating sales. The producer
benefits from both of these activities. In particular, the market information improves
the producer’s demand forecast and thus her production planning. However, both
activities require the seller to exert costly effort. Employing a stylized principal-
agent model, we characterize the optimal performance of forecast-based contracts
and menus of linear contracts. While forecast-based contracts have been widely
used in practice, the academic literature suggests that menus of linear contracts are
superior when agents are costlessly endowed with private information about the
market condition. This chapter discusses that the relative performance of these two
classes of contracts is more nuanced than what the literature suggests. In particular,
when information acquisition is costly, forecast-based contracts may outperform
menus of linear contracts.

9.1 Introduction

A prevalent way to organize supply chain activities is to have production and sell-
ing done by different entities. For example, a manufacturer of fashion apparel may
engage a department store to sell its seasonal product lines; a high-end bicycle
manufacturer may contract with a dealer to sell its new designs; a home appliance

G. Lai ()
McCombs School of Business, The University of Texas, Austin, TX 78712, USA
e-mail: [email protected]
W. Xiao
Stern School of Business, New York University, New York 10012, NY, USA
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 171

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 9
172 G. Lai and W. Xiao

producer may sell its products through telemarketers. In all these examples, an im-
portant element in the producer-seller relationship is incentive alignment, e.g., the
producer pays the seller on a commissions basis, which serves to correct the conflict
between the producer’s desire for more sales and the seller’s aversion to exerting
selling effort. However, the problem of incentive alignment between the producer
and the seller goes beyond simply motivating sales: the seller has close contacts
with the end customers and thus is in an ideal position to gather market information,
and fresh information from the field helps the producer make the right products and
produce the right quantity. Here lies another incentive problem that is often over-
looked: the fresh information from the field, while highly valuable to the producer,
is costly for the seller to gather. Therefore, a better characterization of the incentive
problem in the producer-seller relationship is: how can the seller be motivated to
gather market information about the producer’s product, share that information with
the producer, and work hard to promote sales of the product?
We employ a stylized principal-agent model to study the contract-design prob-
lem that arises in the producer-seller relationship. In particular, a principal (the pro-
ducer, she) engages an agent (the seller, he) to sell a product over a single sales
season. Both the principal and the agent are risk neutral. The total demand for the
product in the season is influenced by the market condition, the agent’s sales ef-
fort, and a random noise in an additive manner. A key feature of this model is that
the agent has an opportunity to collect a signal about the market condition, before
the sales season. The agent can improve the quality of the signal by expending an
information-acquisition effort. The principal is interested in improving her knowl-
edge about the market condition for better production planning. The question is how
the principal should design a compensation scheme for the agent that maximizes her
expected profit while motivating the agent to gather market information, to share the
information with her, and to increase sales.
Ideally, the principal can compensate the agent for the amounts of efforts he has
devoted to the job, both for information acquisition and for selling. However, as
in any principal-agent setting, the agent’s efforts are not observable and thus not
contractible. In our setting the only observable measure related to the agent’s efforts
is the total sales, on which the compensation scheme must be based. We focus on
two types of contracts that are widely studied in the agency literature: the forecast-
based contract (FC) and the menu of linear contracts (MLC).
Under the FC, the agent is asked to submit a demand forecast before the sales
season, and the agent is penalized if the final sales deviates from his forecast. In
addition to the penalty term, the contract also contains a linear term that is increas-
ing in total sales, rewarding the agent for increasing sales and thus inducing sell-
ing effort. Because of the penalty term, the agent also has an incentive to invest
in collecting a market signal, which improves the accuracy of his demand fore-
cast. Furthermore, by submitting the demand forecast, the agent shares his updated
market knowledge with the principal, who can then use it to fine tune her produc-
tion decision. The forecast-based contract is a formalization of the Gonik (1978)
scheme, which was originally designed to extract information and motivate selling
effort from a salesforce and was implemented by IBM’s Brazilian sales office many
9 Contracting for Information Acquisition 173

years ago. Interestingly, the Gonik scheme is alive and well, see Turner et al. (2007)
who describe how this forecast-based contract is currently used to compensate over
100,000 salespeople and managers in the pharmaceutical industry in Europe.
Under the MLC, the principal provides the agent with a list of contracts (possibly
an infinite number of contracts), and each contract is a distinct linear, nondecreasing
function that maps the total sales to the agent’s compensation. The agent has an
opportunity to collect a signal about the market condition before choosing a contract
from the menu. Because a better signal enables the agent to make a better contract
choice, the agent is motivated to exert information-acquisition effort. After the agent
has chosen a contract (and the sales season has begun), he has an incentive to exert
selling effort because the more he sells the higher his compensation. Moreover, the
agent’s contract choice conveys to the principal his updated market information,
which she uses in making her production decision.
Despite the popularity of the forecast-based contract in practice, the academic
literature seems to suggest that the menu of linear contracts is superior. In particular,
under various settings where the risk-neutral agent holds private market information,
the menu of linear contracts has been shown to be optimal among all possible menu
contracts (which of course includes the forecast-based contract), see, e.g., Laffont
and Tirole (1986), McAfee and McMillan (1987), Picard (1987), and Rao (1990).
Under a closely related model setup, Chen (2005) shows that when the agent is
risk averse, the principal also prefers the menu of linear contracts to the forecast-
based contract. In all of the above-mentioned papers, the agent is costlessly endowed
with private information unobservable to the principal. Interestingly, by considering
costly information acquisition, this chapter discusses that the principal may actually
prefer the forecast-based contract.
The rest of the chapter is organized as follows. In Sect. 9.2, we review the rel-
evant literature. In Sect. 9.3, we describe the model and present some preliminary
results. Sections 9.4–9.6 contain, respectively, the analysis of the first-best solution,
the menu of linear contracts, and the forecast-based contract. Section 9.7 concludes.

9.2 Literature

The moral hazard models assume that the principal and the agent possess the same
information before contracting and focus on motivating hidden actions, while the
adverse selection models stipulate that the agent possesses superior information and
focus on information screening through mechanism designs. There is substantial lit-
erature that adopts the adverse selection models to study the incentive contracts in
supply chain settings where members have distinct information about market de-
mand or production costs. A typical theme of this stream of work is to explore how
contracts should be designed and then to evaluate the performance of optimal con-
tracts and/or simple and commonly-used contracts (see, e.g. Cachon and Lariviere
2001; Ha 2001; Cachon and Zhang 2006; Özer and Wei 2006; Burnetas et al. 2007).
See Cachon (2003) and Chen (2003) for comprehensive reviews of this literature.
174 G. Lai and W. Xiao

While some agents simply come endowed with superior information, what we see
in practice is often that the agents need to expend effort to obtain the information.
The more the effort, the more or better the information. In other words, the agent
can determine the extent of information asymmetry in the principal-agent relation-
ship. This model feature places our work in the category of endogenous adverse
selection.
One of the first papers paying attention to the agent’s acquisition of information
is Cremer and Khalil (1992), where a principal offers a contract to an agent to pro-
duce a good. Before deciding whether or not to accept the contract, the agent can
exert a costly effort to acquire information on the cost of the job. An interesting
feature of this paper is that the cost information can be had for free after the agent
accepts the contract. Therefore obtaining the information is socially wasteful, and
is done purely for strategic reasons. The main finding of the paper is that although
in equilibrium, the principal does not offer a contract to induce the agent to acquire
information, the agent’s ability to acquire information significantly alters the terms
of the contract. Lewis and Sappington (1997) couple the information acquisition
with a subsequent moral hazard problem, i.e., after acquiring information about the
operating environment, the agent takes a hidden action to reduce the cost of produc-
tion. Moreover, the principal also benefits from the information the agent gathers
because she can make a state-dependent effort to increase the value of the project.
Therefore, the principal’s task is to find the best way to motivate the agent to acquire
information, to share it with the principal, and to exert cost-reduction effort. In this
sense, our model is very similar to the Lewis-Sappington model (LS), but there are
some subtle differences between them. The Lewis-Sappington model assumes that
the principal always induces information acquisition by the agent, whereas we leave
that decision to the principal’s optimization problem. Moreover, LS consider only
menus of linear contracts without addressing the principal’s first-best solution. In
contrast, we go beyond that by considering forecast-based contracts. We show that
the first-best solution can be achieved if contract negotiation can be based on the ini-
tial menu offered and concluded before the agent’s information acquisition. This is a
plausible scenario in a supply chain setting with a powerful producer (e.g., with a hot
product or brand). For further extensions of the Cremer-Khalil model, see Cremer
et al. (1998a,b), where the agent’s information acquisition can take place even be-
fore the principal’s contract offer or the information gathered is “productive” (rather
than “strategic”) in the sense that if the agent decides not to gather information and
accepts the contract, he has to produce without the knowledge of the state of nature.
Recently, we have seen some interesting papers that study how manufacturers
should design effective incentive contracts when retailers have an opportunity to ac-
quire information about the market demand. For instance, Taylor and Xiao (2009)
compare rebate versus return contracts when the retailer can improve the quality of
her demand information by exerting costly forecasting effort. Fu and Zhu (2010)
reveal that the commonly used supply chain contracts such as buy-back and revenue
sharing can achieve channel coordination when the retailer’s acquisition of informa-
tion is costly, while Shin and Tunca (2010) show that a market-based pricing scheme
9 Contracting for Information Acquisition 175

is efficient in inducing multiple competing retailers to acquire demand information.

With a slightly different focus, Taylor and Xiao (2010) investigate whether it is ben-
eficial to sell to a retailer with more accurate demand forecast.
Finally, we make note of the connections to the vast marketing literature on sales-
force compensation, which deals with the design of compensation schemes for sales
people. Coughlan and Sen (1989) and Albers (1996) provide comprehensive re-
views of this literature. The two contract forms studied in this paper, namely the
forecast-based contract and the menu of linear contracts, have been widely studied
in the salesforce literature. As mentioned before, the forecast-based contract extends
the Gonik scheme originally designed to motivate a salesforce. It is also related to
the so-call bottom-up approach that many firms use to gain valuable information
from their salesforce. Under this approach, the firm requires its salesforce to submit
forecast information and ties this information to their sales quotas and performance
goals, which in turn have impact on their final compensation. Therefore our forecast-
based contract can be viewed as a specific incarnation of the bottom-up approach
(Mishra and Prasad 2004). On the other hand, Rao (1990) studies the menu of linear
contracts in a model with a risk-neutral sales agent whose selling skills are unknown
to the firm. The sales volume is jointly determined by the agent’s selling skills and
sales effort. Rao provides an intuitive implementation of the menu of linear con-
tracts: The agent is asked to submit a quota, which determines a fixed payment for
meeting the quota and a commission rate for over/under-fulfilling the quota. Rao
shows that under certain assumptions on the sales response function and the distri-
bution function of the agent’s selling skills, the menu of linear contracts achieves
the maximum profit for the firm among all possible forms of menu contracts.

9.3 The Model

A risk-neutral principal produces a single product and employs a risk-neutral agent

to sell it. The sales volume x is jointly determined by a market condition θ , the
agent’s sales effort a, and a random noise ε via the following additive form: x =
θ + a + ε . Assume that θ and ε are two independent, normally distributed, random
variables with θ ∼ N(μθ , σθ2 ) and ε ∼ N(0, σε2 ). The agent incurs a cost of V (a) for
exerting sales effort a. For simplicity, let V (a) = a2 /2.1
Before determining the level of sales effort, the agent has an opportunity to ac-
quire a private signal s about the market condition. Assume that s = θ + η , where
η is independent of θ and ε with η ∼ N(0, ση2 ). Let σ 2 ≡ ση2 /(ση2 + σθ2 ). From the
conjugate property of the normal distribution, the posterior distribution of θ for a
given signal s is
θ |s ∼ N(μ ps , σ ps
2
), (9.1)

1 These assumptions are often made in agency models to evaluate the performance of specific
contracts, see, e.g., Lal and Srinivasan (1993).
176 G. Lai and W. Xiao

where
(μ ps , σ ps
2
) ≡ (σ 2 μθ + (1 − σ 2 )s, σθ2 σ 2 ). (9.2)
Note that ση is a measure of the signal’s precision in predicting the market con-
dition. The larger the value of ση , the less information the signal has about the
market condition. The agent can select a value of ση , or equivalently σ , by exert-
ing information-acquisition effort. If σ = 0, the signal reveals the exact value of the
market condition. On the other hand, as σ → 1, the posterior distribution of the mar-
ket condition is identical to its prior distribution, i.e., the signal contains no useful
information. Let Γ (σ ) be the cost the agent incurs for collecting a signal of preci-
sion σ , σ ∈ (0, 1], with Γ (1) = 0. We assume that Γ (σ ) is strictly decreasing and
convex, i.e., Γ  (σ ) < 0 and Γ  (σ ) > 0 for σ ∈ (0, 1].2
The sequence of events is as follows. (1) The principal offers a menu of con-
tracts to the agent. (The menu can of course contain only a single contract.) (2) The
agent decides whether or not to accept this menu. If he rejects it, the game is over.
Otherwise, the game continues to the next step. (3) The agent decides how much
effort to expend to acquire market information. He then receives a signal. (4) Based
on the signal, the agent chooses a contract from the menu. The principal produces.
The sales season starts. (5) The agent decides how much effort to exert on selling.
(6) The sales are realized, and the agent is compensated according to the contract
he has chosen in Step 4. Notice that the agent’s participation is determined before
information acquisition.
The agent’s objective is to maximize his expected net income, which is equal
to the wage received (w) minus the cost of sales effort (V (a)) and the cost of
information-acquisition effort (Γ (σ )). The agent’s minimum requirement for his
expected net income is normalized to 0. Thus the menu of contracts offered by the
principal must generate for the agent a nonnegative expected net income in order
for him to accept it.
As in standard agency theory, the agent’s information-acquisition effort and sales
effort are both unobservable to the principal. Moreover, we assume that the signal
the agent collects is also unobservable to the principal. Consequently, the contract
between the principal and the agent can only be based on the total sales, which is
observable and thus contractible. To facilitate the extraction of market information,
we allow the principal to offer a menu of contracts. The menu of contracts is of the
form: {w(x, y) | y ∈ Y }, i.e., the wage w(x, y) the agent receives is a function of the
total sales x and a parameter y chosen by the agent. By observing the agent’s contract
choice y, the principal may gain some information about the market condition.
The principal must decide how much to produce before the sales season. This is
often true where the production leadtime is long relative to the length of the sales
season, rendering a make-to-order system impractical. Denote the unit production
cost by c. The sales price is 1 + c (hence the profit margin is normalized to 1). If
the total demand exceeds the production quantity, the excess demand is satisfied
by emergency production with unit cost c . Otherwise, the leftover inventory is sal-
vaged for v per unit. To avoid trivial cases, we assume that v < c < c < 1 + c.

2 This model of information acquisition was developed by Winkler (1981).

9 Contracting for Information Acquisition 177

Thus, if the demand is x and the production quantity is q, then the principal’s profit
(excluding the agent’s wage) is x − L(q − x), where L(·) represents the total cost
of supply/demand mismatch, with L(z) = (c − v) max{0, z} + (c − c) max{0, −z}.
Let I be the principal’s information at the time of making the production deci-
sion. The principal’s objective is to maximize her expected profit, E[x − w(x, y)] −
E[E[L(q − x)]|I], where y represents the agent’s contract choice.

9.4 The First-Best Solution

In this section, we assume that the agent’s information-acquisition effort, the real-
ized value of the signal, and the agent’s sales effort are all observable to the prin-
cipal. These additional assumptions simplify the principal’s problem significantly,
with the resulting maximum expected profit for the principal being the first-best
solution.
When the principal is omniscient, the optimal strategy is to centralize all the de-
cisions. And the implementation of these decisions can be guaranteed by imposing
severe penalties on the agent for any noncompliance. In the context of our model,
this means that it is optimal for the principal to use a forcing contract, i.e., the prin-
cipal tells the agent what to do and how he will be compensated given that the agent
follows the principal’s instructions.
Take any forcing contract , which consists of σ , the signal precision; a(s), the
sales effort as a function of the realized signal s; and w(s, a), the wage as a function
of the realized signal s and the agent’s sales effort a.
First consider the principal’s production decision. Recall that the production de-
cision is made after the signal has been collected. The distribution of the mar-
ket condition θ given s is normal with mean μ ps and variance σθ2 σ 2 [see (9.1)
and (9.2)]. Since x = θ + a(s) + ε and ε is a normal random variable indepen-
dent of s, the distribution of x given s is also normal with mean μ ps + a(s) and
variance σθ2 σ 2 + σε2 . Note that, given s, the problem facing the principal is simply
the standard newsvendor problem. Consequently, the principal’s optimal produc-
tion quantity is q(s) = μ ps + a(s) + σθ2 σ 2 + σε2 Φ −1 ((c − c)/(c − v)) and the
total cost of supply/demand mismatch is E[L(q(s) − x)|s] = ρ σθ2 σ 2 + σε2 , where
ρ = (c − v)φ (Φ −1 ((c − c)/(c − v))) with φ (·) being the standard normal den-
sity function and Φ −1 (·) the inverse of the standard normal distribution function.
Under the above optimal production decision, the principal’s expected profit is
μθ + E[a(s)] − E[w(s, a(s))] − ρ σθ2 σ 2 + σε2 . The principal’s optimization prob-
lem can be written as
 
(P1) max μθ + E[a(s)] − E[w(s, a(s))] − ρ σθ2 σ 2 + σε2
σ ∈(0,1], a(·)≥0, w(·,·)

s.t. E[w(s, a(s)) −V (a(s)) − Γ (σ )]  0

178 G. Lai and W. Xiao

where the constraint ensures the agent’s participation. The following proposition
characterizes an optimal solution to (P1). All the proofs can be found in Chen et al.
(2016).

Proposition 1. An optimal solution {σo , ao (·), wo (·, ·)} to (P1) is as follows:

 
σo = arg min Γ (σ ) + ρ σθ2 σ 2 + σε2 (9.3)
σ ∈(0,1]
ao (·) = ao = arg max{a −V (a)} = 1 (9.4)
a≥0
!
wo = V (ao ) + Γ (σo ) if σ = σo and a = ao
wo (·, ·) = (9.5)
−M otherwise

where M > 0 is sufficiently large. The principal’s maximum expected profit is

#
Πo = μθ + 1/2 − Γ (σo ) − ρ σθ2 σo2 + σε2 . (9.6)

Next we provide intuitive explanations for Proposition 1. First, a more precise

signal is more costly to the agent, and the information-acquisition cost is ultimately
paid by the principal. The benefit of having a more precise signal is a smaller
posterior variance of the market condition and thus a lower expected cost of sup-
ply/demand mismatch. Hence the optimal level of the signal precision is obtained
by minimizing the total cost [see Eq. (9.3)]. Due to the additive sales response, the
optimal sales effort remains constant for any value of the realized signal. The opti-
mal level of sales effort is derived by balancing the gross profit due to sales effort
(profit margin is 1) and the cost of the effort. This explains Eq. (9.4). The agent re-
ceives a fixed wage, which consists of the cost of sales effort (V (ao )) and the cost
for information acquisition (Γ (σo )). See Eq. (9.5). Finally, the principal’s profit Πo
is simply its gross profit (μθ + ao ) minus the agent’s wage minus the expected cost
of supply-demand mismatch. See Eq. (9.6).

9.5 The Menu of Linear Contracts

By giving the agent a list of linear contracts and asking him to choose one from
the list after he observes the market signal, it creates an incentive for the agent
to acquire market information because the information may help him choose the
optimal contract. The principal can now learn something about the market condition
by observing the agent’s contract choice. The number of linear contracts on the
menu determines the amount of information communicated to the principal through
the agent’s contract choice: the higher the number of contracts on the menu, the
finer the information revealed to the principal. Because there are an infinite number
of possible market states (θ is a continuous random variable), it makes sense to have
an infinite number of linear contracts on the menu.
9 Contracting for Information Acquisition 179

We consider the following menu:

w(x, μ ps ) = α (μ ps )x + β (μ ps )

where the agent’s wage is a linear function of the final sales x, with the commission
rate α (μ ps ) ≥ 0 and the fixed transfer β (μ ps ) each being a function of μ ps , the
posterior mean of the market condition reported by the agent. We denote the menu
by {α (·), β (·)}.
Recall from Sect. 9.3 that after accepting the menu w(·, ·), the agent first decides
the signal precision σ , then observes the signal s and thus μ ps , then chooses a con-
 ps ), and finally makes the sales effort decision. Next we derive
tract (by reporting μ
the agent’s optimal decisions using backward induction.
We begin with the agent’s sales effort decision. The agent who observed μ ps is
called the type-μ ps agent. Recall that x| μ ps ∼ N(μ ps + a, σθ2 σ 2 + σε2 ). Thus, the
type-μ ps agent’s expected wage is equal to α (μ  ps )(μ ps + a) + β (μ
 ps ), if he reports
 ps and exerts sales effort a. Subtracting the cost of sales effort V (a) = a2 /2 from
μ
the expected wage, we have the type-μ ps agent’s expected profit (excluding the sunk
information-acquisition cost). Maximizing this profit over a, we obtain

 ps ) = max[α (μ
π (μ ps , μ  ps )(μ ps + a) + β (μ
 ps ) − a2 /2]
a≥0

which is the type-μ ps agent’s maximum expected profit if he reports μ  ps . From the
revelation principle, we can without loss of generality restrict to the menus that
induce the agent to truthfully report his type, i.e., π (μ ps , μ ps ) ≥ π (μ ps , μ  ps ) for
 ps . Let π (μ ps ) ≡ π (μ ps , μ ps ). Clearly, with truth telling, the type-μ ps
any μ ps and μ
agent’s optimal sales effort, denoted by a(μ ps ), is equal to the commission rate of
the chosen contract, i.e., a(μ ps ) = α (μ ps ).
Now consider the agent’s decision on σ , the signal precision. Let π(σ ) be the
agent’s expected profit as a function of σ . Thus

π(σ ) = Eμ ps π (μ ps ) − Γ (σ ),

where μ ps ∼ N(μθ , σθ2 (1 − σ 2 )) [see Eq. (9.2)]. Let σ = arg maxσ  ∈(0,1] π(σ  ), the
optimal signal precision.
We are now ready to consider the principal’s optimization problem. Let I rep-
resent the principal’s information related to the market demand at the time of the
production decision. Because of truth telling, I = {μ ps , σ }, where σ is inferred
from the agent’s optimization problem. Therefore, the minimum expected cost of
supply/demand mismatch is ρ σθ2 σ 2 + σε2 (similar to the first-best solution). As a
result, the principal’s expected profit under the menu {α (·), β (·)} can be written as

Π (α (·), β (·)) = Eμ ps [μ ps + α (μ ps )] − Eμ ps [α 2 (μ ps )/2]

#
− Γ (σ ) − ρ σθ2 σ 2 + σε2 − π(σ ),
180 G. Lai and W. Xiao

which can be interpreted as the supply chain’s expected profit (the first four terms:
gross profits, cost of sales effort, cost of information-acquisition, and cost of sup-
ply/demand mismatch) minus the agent’s expected profit (the last term). Conse-
quently, the optimal menu of linear contracts is the solution to:

(P2) max Π (α (·), β (·))

α (·)≥0,β (·)
 ps ), ∀μ ps , μ
s.t. π (μ ps ) ≥ π (μ ps , μ  ps (IC1)

σ = arg max π(σ ) (IC2)
σ  ∈(0,1]

π(σ ) ≥ 0. (IR)

Note that (IC1) and (IC2) are incentive-compatibility constraints, whereas (IR) is
the (ex ante) participation constraint. Denote the optimal menu by {α ∗ (·), β ∗ (·)}.
Before we present the optimal menu, define
   
1 z − μθ z − μθ
φ (z, σ ) ≡ √ φ √ and Φ (z, σ ) ≡ Φ √ ,
σθ 1 − σ 2 σθ 1 − σ 2 σθ 1 − σ 2
which are the density and the residual distribution
√ of the normal random variable μ ps
with mean√ θμ and standard deviation σθ 1 − σ 2 . Define λ (σ ) ≡ {λ | λ Φ (−1/λ ) =

−Γ  (σ ) 1 − σ 2 /(σθ σ )} for σ ∈ (0, 1]. Note that λ Φ (−1/λ ) strictly increases

from 0 to infinity
√ as λ increases from 0 to infinity. Because Γ  (σ ) < 0 and Γ  (σ ) >
0, −Γ  (σ ) 1 − σ 2 /(σθ σ ) is positive and strictly decreases in σ for σ ∈ (0, 1].
Therefore, λ (σ ) is well defined and strictly decreasing in σ . Let x+ ≡ max{x, 0}.
Proposition 2. The optimal menu of linear contracts is
 
μ ps − μθ +
α ∗ (μ ps ) = 1 + λ (σ ∗ ) √
σθ 1 − σ ∗2
and
1
β ∗ (μ ps ) = − [α ∗ (μ ps )]2 − μ ps α ∗ (μ ps ) + Γ (σ ∗ )
2
 +∞  μ ps
− Φ (z, σ ∗ )α ∗ (z)dz + α ∗ (z)dz,
−∞ −∞

where
 +∞
  # 
∗ α 2 (z, σ )
σ = arg max α (z, σ ) − φ (z, σ ) dz − Γ (σ ) − ρ σθ σ + σε ,
2 2 2
σ ∈(0,1] −∞ 2
(9.7)
√ +
where α (μ ps , σ ) ≡ [1 + λ (σ )(μ ps − μθ )/(σθ 1 − σ ] ). When the principal offers
2

this menu of linear contracts to the agent, he will accept it, choose signal preci-
sion σ ∗ , collect a market signal (and thus observe μ ps ), reveal μ ps to the princi-
pal (and thus choose the linear contract with commission rate α ∗ (μ ps ) and fixed
9 Contracting for Information Acquisition 181

transfer β ∗ (μ ps )), and finally, exert sales effort a∗ (μ ps ) = α ∗ (μ ps ). The principal’s

maximum expected profit is
 +∞   #
∗ ∗ [α ∗ (z)]2
ΠMLC = μθ + α (z) − φ (z, σ ∗ ) dz − Γ (σ ∗ ) − ρ σθ2 σ ∗2 + σε2 .
−∞ 2
(9.8)

The following corollaries point out the key differences between the optimal MLC
and the first-best solution. Recall that ao (= 1) and σo are the first-best sales effort
and information-acquisition effort, respectively.

Corollary 1. If ρ > 0, the optimal MLC leads to underinvestment in information

acquisition, i.e., σ ∗ > σo . Moreover, limρ →0 σ ∗ = limρ →0 σo = 1.

Corollary 2. If ρ > 0, the optimal MLC leads to overinvestment in sales effort

when the market signal is favorable, and underinvestment in sales effort other-
wise. That is, a∗ (μ ps ) > ao for μ ps > μθ and a∗ (μ ps ) < ao for μ ps < μθ . Moreover,
limρ →0 a∗ (μ ps ) = limρ →0 ao = 1.

Corollary 3. Let Γk (σ ) ≡ kΓ (σ ) for k > 0 and σ ∈ (0, 1]. Let Πo (k) and ΠMLC∗ (k)

be the principal’s expected profit under the first-best solution and the optimal MLC,
respectively, assuming that the agent’s cost of information acquisition is Γk (·). Then
∗ (k) = lim
limk→0 ΠMLC k→0 Πo (k). In other words, as the cost of information acqui-
sition decreases, the principal’s optimal expected profit under the menu of linear
contracts approaches to her first-best profit.

The above results provide intuition as to why the menu of linear contracts fails
to achieve the first best. Recall that the principal needs to motivate the agent to ex-
ert two hidden efforts: gathering information and generating sales. Fundamentally,
the MLC fails because these two goals are in conflict with each other. To motivate
the agent to gather information, the principal must offer a broader menu, i.e., the
linear contracts on the menu should offer a wide range of slopes so that the optimal
contract choice for the agent varies greatly depending on the market signal. In other
words, the contract the agent picks after receiving good news from the market should
be very different from the contract chosen when the news is bad. This suggests that
if the agent does not exert effort to gather precise market information, the risk of
choosing a wrong contract is very high, thus motivating information acquisition. On
the other hand, the slope of the linear contract chosen by the agent is the only thing
that motivates the agent to exert selling effort. Moreover, due to the additive sales
response function, the first-best sales effort is a constant and is independent of the
market signal. Therefore, in order to achieve the first-best sales effort, the principal
should simply offer a single linear contract, which of course completely removes
the incentives for information acquisition (see Sect. 9.4). In sum, motivating sales
effort calls for a narrower menu, whereas motivating information-gathering effort
requires a broader menu. It is this conflict that leads to the principal’s profit being
strictly less than the first-best under the menu of linear contracts.
182 G. Lai and W. Xiao

The above discussions also suggest that if motivating information gathering be-
comes either less useful or easier, then the principal’s profit will approach the first-
best. The intuition is clear because both of these scenarios make it unnecessary to
have a broad menu. The theoretical arguments are provided in Corollaries 1 and 2
when the market information becomes less useful as the cost of supply-demand
mismatch decreases, and in Corollary 3 when the cost of information acquisition
decreases.

9.6 The Forecast-Based Contract

In this section, we consider an alternative contract form that is obtained by adding

a penalty term to the linear contract. Specifically, the agent is required to submit
a sales forecast and his compensation is reduced if the actual sales turns out to be
different from the forecast. Mathematically, the agent’s compensation is determined
by the following formula:

w(x, y) = α x + β − γ h(x − y) (9.9)

where x and y are the actual sales and the submitted forecast respectively, {α , β , γ }
are contract parameters chosen by the principal, and h(·) is a penalty function with
h(0) = 0, h (z) > 0 for z > 0, and h (z) < 0 for z < 0. Examples of the penalty
function include: h(z) = |z|, h(z) = z2 , etc. We will refer to this contract as the
forecast-based contract. It is also a menu contract: the agent chooses a contract from
the menu by specifying y. Notice that the Gonik (1978) scheme is a special case of
the above contract form with h(z) = −uz for z < 0 and h(z) = vz for z > 0 for some
positive constants u and v. We seek the optimal contract parameters {α ∗ , β ∗ , γ ∗ } that
maximize the principal’s expected profit. As we will see, the forecast-based contract
achieves the first-best outcome.
Suppose the principal offers the contract (9.9) to the agent. Assume that the agent
decides to participate, i.e., accept the contract. He then faces a two-stage decision
problem. The first stage is to decide the signal precision σ or equivalently the effort
for gathering information. The second stage is after collecting the market signal.
Here the agent decides a sales forecast y and determines the sales effort a. Obviously,
the forecast and the sales effort can both be functions of the market signal. As before,
we substitute μ ps for the market signal. At the second stage, given σ and μ ps , the
agent reports the forecast y and exerts sales effort a to maximize his expected profit
(excluding the sunk information-acquisition cost). That is,

π (μ ps ) = max [α (μ ps + a) + β − γ Ex [h(x − y)| μ ps ] − a2 /2] (9.10)

y,a≥0

where π (μ ps ) is thus the agent’s maximum expected profit going forward after
observing μ ps . Recall that x|μ ps ∼ N(μ ps + a, σθ2 σ 2 + σε2 ). Thus we can rewrite
9 Contracting for Information Acquisition 183

x = μ ps + a + ξ σθ2 σ 2 + σε2 , where ξ ∼ N(0, 1). Using this expression in (9.10)

and defining Δ = y − μ ps − a, we have
  
π (μ ps ) = max[α (μ ps + a) + β − a2 /2] − min γ Eξ h ξ σθ2 σ 2 + σε2 − Δ . (9.11)
a≥0 Δ

Notice that the optimization is greatly simplified because the objective function is
separable in a and Δ . The optimal solution is ã = α and Δ̃ = arg minΔ Eξ [h(ξ ·
σθ2 σ 2 + σε2 − Δ ]. It is important to note that ã is independent of the first-stage de-
cision (σ ) and the market signal, and that Δ̃ is independent of the market signal (but
may depend on σ ). Consequently, the agent’s optimal forecast is ỹ = μ ps + α + Δ̃ .
Having characterized the agent’s decisions at stage 2, we step back to Stage 1, where
the decision is σ . Denote by π(σ ) the agent’s expected profit as a function of σ .
Clearly,
π(σ ) = Eμ ps π (μ ps ) − Γ (σ ), (9.12)
where μ ps ∼ N(μθ , σθ2 (1 − σ 2 )). Maximizing the above expression over σ leads to
the optimal precision level.
We proceed to consider the principal’s optimization problem. Let I be the infor-
mation the principal has that is related to the demand during the sales season at the
time of the production decision. Because ỹ = μ ps + α + Δ̃ , the principal can first
solve the agent’s first- and second-stage problems to determine the value of Δ̃ , from
which she infers the value of μ ps from the agent’s submitted forecast ỹ. Therefore,
I = {μ ps , σ }. Given this information, the minimum expected cost of supply/demand
mismatch is ρ σθ2 σ 2 + σε2 (see Sect. 9.4). This, together with the fact that the agent
exerts sales effort ã = α , implies that the principal’s expected profit, denoted by
Π (α , β , γ ), can be written as:
#
α2
Π ( α , β , γ ) = μθ + α − − Γ (σ ) − π(σ ) − ρ σθ2 σ 2 + σε2 .
2
The principal’s optimization problem can thus be formulated as:

(P3) max Π (α , β , γ )
α ≥0,β ,γ ≥0

s.t. σ = arg max π(σ  )

σ  ∈(0,1]

π(σ ) ≥ 0.

Let α ∗ = 1, γ ∗ be the solution to

 
  
arg min γ ∗ min Eξ h ξ σθ2 σ 2 + σε2 − Δ + Γ (σ ) = σo
σ ∈(0,1] Δ
184 G. Lai and W. Xiao

where ξ ∼ N(0, 1) and σo is the first-best signal precision, and

  
β ∗ = −μθ − 1/2 + γ ∗ min Eξ h ξ σθ2 σo2 + σε2 − Δ + Γ (σo ).
Δ

Proposition 3. The forecast-based contract with parameters {α ∗ , β ∗ , γ ∗ } is accept-

able to the agent. Under this contract, the agent’s optimal decisions on the sig-
nal precision and the sales effort are the same as in the first-best solution, i.e.,
σo and ao (= 1). Moreover, for any given μ ps , the# agent’s optimal forecast is
y∗ = μ ps + ao + Δ ∗ , where Δ ∗ = arg minΔ Eξ [h(ξ σθ2 σo2 + σε2 − Δ )] with ξ ∼
N(0, 1). Therefore, the principal can fully infer the value of μ ps from the agent’s
forecast. Finally, under the contract, the agent’s expected profit is equal to his reser-
vation profit and the principal’s expected profit, denoted by ΠFC ∗ , is equal to the
∗
first-best profit, i.e., ΠFC = Πo .

We saw in the previous section that the menu of linear contracts does not provide
the first-best outcome. Therefore, it is all the more interesting that the forecast-based
contract does. To see the intuition why this is the case, first recall that the principal
aims to provide incentives for the agent to gather market information and to increase
sales. Whereas under the menu of linear contracts these two objectives are in con-
flict, the forecast-based contract actually decouples these two tasks and makes it
possible to achieve the first-best outcome along both dimensions. Note that the FC
contract consists of two parts: a linear contract, which is parameterized by α and β ,
and a penalty term parameterized by γ . An important feature is that the penalty is as-
sessed based on x − y, the difference between the actual sales and the forecast. This
feature, together with the additive form of the sales response function, suggests that
the agent will always include his subsequent sales effort a in his forecast. After all,
this is the easiest part (actually a “sure” part) to forecast in the total sales. The con-
sequence of including a in the forecast is that the penalty term now is independent of
the sales effort, making the market condition (θ ) and the noise (ε ) the only drivers
of the expected penalty. Although there is nothing to be done about ε , the agent
can, at some cost to himself, gather market information to improve his prediction
of θ , which in turn leads to a better forecast and thus a lower penalty cost. The prin-
cipal can control the agent’s information-acquisition decision (σ ) by adjusting the
value of γ . Now consider the agent’s sales effort decision. As mentioned, the penalty
term is independent of this decision and can now be safely ignored. What is left of
the FC contract is a linear function of x, and the agent’s sales effort is completely
controlled by α . In sum, the principal has two “dials,” one controls the information-
acquisition effort and the other controls the sales effort, and these controls do not
interfere with each other. By setting the right values of α and γ , the principal can
induce the first-best effort decisions. Finally, the principal uses β to extract all the
profits while ensuring the agent’s participation. In essence, the FC contract succeeds
in achieving the first-best outcome because it has the capability to motivate the two
hidden efforts separately.
9 Contracting for Information Acquisition 185

9.7 Conclusions

Most studies of decentralized supply chains assume that the actors are simply
endowed with information (some may be private) about the operating environment.
But reality is often that valuable information needs to be gathered and doing so is
costly. This chapter fills this important gap by considering a producer-seller rela-
tionship where the seller is uniquely positioned to gather market information, which
can be used to improve production planning by the producer.
We have selected two widely studied contract forms and examined their per-
formance in the above principal-agent model. One is the forecast-based contract,
where for each possible sales forecast the agent may provide, the agent receives a
payment linear in the total sales and pays a penalty if the realized total sales differs
from the forecast. The other is the menu of linear contracts, with each contract on
the menu specifying a specific linear relationship between the agent’s compensation
and the realized total sales. A key message is that the forecast-based contract out-
performs the menu of linear contracts when information acquisition is costly, even
though the extant literature points to the contrary when the agent is costlessly en-
dowed with private market information. Our analysis reveals an important difference
between the two contract forms in their abilities to manage the tension between mo-
tivating for information acquisition and motivating for sales effort. Specifically, the
forecast-based contract decouples the two motivation tasks and thus allows both the
information-acquisition effort and the selling effort to reach their first-best levels,
whereas the menu of linear contracts creates friction between the two motivation
tasks making it impossible to achieve the first-best outcome.
Another important aspect of the contracting relationship between the producer
and the seller is the timing of contract signing. When the contract signing takes
place before the agent’s information acquisition, the principal is indifferent between
the two contracts when information acquisition is costless, and strictly prefers the
forecast-based contract when information acquisition is costly. In contrast, when the
contract signing takes place after the agent’s information acquisition, the menu of
linear contracts is superior to the forecast-based contract if information acquisition
is costless, which is consistent with a main finding in the literature. As information
acquisition becomes more costly, the preference shifts back to the forecast-based
contract if the cost of supply-demand mismatch is large. This result stems from the
fact that ensuring the participation of every agent type (after receiving the market
signal) severely limits the ability of the forecast-based contract to provide incentives
for the agent to exert selling effort, whereas the menu of linear contracts can take
advantage of its flexibility to tailor a linear contract for each agent type.

References

Albers S (1996) Optimization models for salesforce compensation. Eur J Oper Res 89:1–17
Burnetas A, Gilbert SM, Smith CE (2007) Quantity discounts in single-period supply contracts
with asymmetric demand information. IIE Trans 39(5):465–479
186 G. Lai and W. Xiao

Cachon GP (2003) Supply chain coordination with contracts. In: Graves S, de Kok T (eds) Hand-
books in operations research and management science: supply chain management. Elsevier
Science Publisher, North-Holland, Amsterdam
Cachon GP, Lariviere MA (2001) Contracting to assure supply: how to share demand forecasts in
a supply chain. Manag Sci 47(5):629–646
Cachon GP, Zhang F (2006) Procuring fast delivery: sole-sourcing with information asymmetry.
Manag Sci 52(6):881–896
Chen F (2003) Information sharing and supply chain coordination. In: Graves S, de Kok T (eds)
Handbooks in operations research and management science: supply chain management. Else-
vier Science Publisher, North-Holland, Amsterdam
Chen F (2005) Salesforce incentives, market information, and production/inventory planning.
Manag Sci 51(1):60–75
Chen F, Lai G, Xiao W (2016) Provision of incentives for information acquisition: forecast-based
contracts versus menus of linear contracts. Manag Sci 62(7):1899–1914
Coughlan AT, Sen SK (1989) Salesforce compensation: theory and managerial implications. Mark
Sci 8(4):324–342
Cremer J, Khalil F (1992) Gathering information before signing a contract. Am Econ Rev
82(3):566–578
Cremer J, Khalil F, Rochet JC (1998a) Strategic information gathering before a contract is offered.
J Econ Theory 81:163–200
Cremer J, Khalil F, Rochet JC (1998b) Contracts and productive information gathering. Games
Econ Behav 25:174–193
Fu Q, Zhu K (2010) Endogenous information acquisition in supply chain management. Eur J Oper
Res 201(2):454–462
Gonik J (1978) Tie salesmen’s bonuses to their forecasts. Harvard Business Review (May-June)
78305-PDF-ENG
Ha AY (2001) Supplier-buyer contracting: asymmetric cost information and the cut-off level policy
for buyer participation. Nav Res Logist 48(1):41–64
Laffont JJ, Tirole J (1986) Using cost observation to regulate firms. J Polit Econ 94(3):614–641
Lal R, Srinivasan V (1993) Compensation plans for single- and multi-product salesforces: an ap-
plication of the Holmstrom-Milgrom model. Manag Sci 39(7):777–793
Lewis TR, Sappington DEM (1997) Information management in incentive problems. J Polit Econ
105(4):796–821
McAfee RP, McMillan J (1987) Auctions and bidding. J Econ Lit 25:699–738
Mishra BK, Prasad A (2004) Centralized pricing versus delegating pricing to salesforce under
information asymmetry. Mark Sci 23(1):21–28
Özer Ö, Wei W (2006) Strategic commitment for optimal capacity decision under asymmetric
forecast information. Manag Sci 52(8):1238–1257
Picard P (1987) On the design of incentive scheme under moral hazard and adverse selection.
J Public Econ 33:305–331
Rao R (1990) Compensating heterogeneous salesforces: some explicit solutions. Mark Sci
9(4):319–341
Shin H, Tunca T (2010) The effect of competition on demand forecast investments and supply
chain coordination. Oper Res 58(6):1592–1610
Taylor TA, Xiao W (2009) Incentives for retailer forecasting: rebates vs. returns. Manag Sci
55(10):1654–1669
Taylor TA, Xiao W (2010) Does a manufacturer benefit from selling to a better-forecasting retailer?
Manag Sci 56(9):1584–1598
Turner R, Lasserre C, Beauchet P (2007) Innovation in field force bonuses: enhancing motivation
through a structured process-based approach. J Med Mark 7(2):126–135
Winkler RL (1981) Combining probability distributions from dependent information sources.
Manag Sci 27(4):479–488
 Part III
Information Signaling and Cheap Talk
Chapter 10
A Tale of Two Information Asymmetries
in Competitive Supply Chains

Mehmet Gümüş

Abstract While globalization leads to more efficient supply chain systems, at the
same time, it makes them more susceptible to demand and supply-side shocks.
Moreover, these risks are exacerbated by the reduced end-to-end visibility due to
the fragmentation of the supply chains. Hence, oftentimes, suppliers and buyers
find themselves having to make pricing and sourcing decisions without knowing
the extent of demand and supply risks affecting the supply chain. In this chapter,
we analyze the effectiveness of demand- and supply-side signals in enhancing the
downstream and upstream information flows and their consequences on the perfor-
mances of the firms in competitive supply chains.

10.1 Introduction

“The world is flat” claims Thomas Friedman in his bestseller. Obviously, he did
not intend to criticize Galilean view of global world. Rather, with this title, he was
referring to the globalization forces1 that leveled the playing field for the supply
chains and expanded them to the new and sometimes unproven geographies. The
very same forces also increase the fragmentation of supply chains, which has two
important consequences for the managers of the global supply chains.
First, it decreased the demand visibility across the chain. Even though there is
little doubt in the supply chain management (SCM) circles on the potential benefits
of sharing information (Seifert 2003; Helms et al. 2000), recent surveys suggest that

M. Gümüş ()
Desautels Faculty of Management, McGill University, Montreal, QC, Canada H3A 1G5
e-mail: [email protected]
1 In his bestseller, Thomas Friedman discusses ten forces that flattened the world. Four of them—

outsourcing, offshoring, supply-chaining, and insourcing—are directly related to supply chain

management (Friedman 2007).

© Springer International Publishing Switzerland 2017 189

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 10
190 M. Gümüş

firms are still quite reluctant in sharing demand-side information with their sup-
pliers (Barratt 2004). This inevitably led to lack of demand-side visibility, which
in turn degraded the accuracy of demand information across the supply chain as
one moves in upstream direction. Research report published by Kaipia and Hartiala
(2006) shows that demand forecast error increases from below 10 % at retail level
to 15 % at tier-1 and 80 % at tier-2 level. There are several factors leading to slow
adoption of information-sharing systems. According to a survey conducted with 120
companies by Fraser (2003), lack of trust over sharing sensitive demand information
is listed by 42 % of the respondents as one of the top three challenges that hinder in-
formation sharing among supply chain firms. The other two include internal process
change (by 60 % of respondents) and cost of implementation (by 50 % of respon-
dents). The concerns with regards the lack of trust can be due to many reasons, such
as the potential loss of control, and security breaches. Our first goal, in this chapter,
is to investigate the impact of a particular one, which is called “fear of collusive
behaviour” by Fliedner (2003). Stein (1998) considers this fear behind lack of trust
between supply chain partners because many organizations have a real and justified
fear that information sharing can turn into a competitive disadvantage. An analyst
from Yankee Group Inc., a market-research information company, reflects a typical
perspective of a supply chain manager on this in the following quote:
If one of your customers knows you are behind on your production schedule, they may turn
around and try to negotiate a more favourable price.

These concerns are also shared by Verity (1996) and Fliedner (2003). They add
that by coordinating forecasts, two or more suppliers might influence the supply
and increase its price. Fliedner (2003) notes that such collusive behaviour can par-
ticularly emerge when the item being purchased is custom made or possessing a
proprietary nature. Indeed, these fears have led researchers to design secure supply
chain protocols to prevent such collusive behaviour (refer to Atallah et al. 2003,
2004, and references therein).
Second issue which is exacerbated by the fragmentation of supply chains is re-
lated to reduced supply-side visibility. Based on IndustryWeek (2009), more than
40 % of buyers lack visibility even into their tier-1 suppliers, and the percentage in-
creases to 75 % for tier-2 suppliers. Similar to demand-side visibility, supply-side in-
formation asymmetry among supply chain parties naturally creates an incentive for
information distortion in order to influence the decision of uninformed parties. This
in turn reduces the supply-side visibility as one moves into downstream direction.
A number of strategies have been proposed to deal with such increases in demand
and supply risks. Among the strategies used to increase demand- and supply-side
visibilities, commitment- and performance-based contracts provided by the buy-
ers and suppliers, respectively, are a popular one (O’Marah 2009; Bernstein and
de Vericourt 2008). We observe various forms of such contracts in industries such
as the automobile industry, semiconductors, electronic equipments, energy and nat-
ural resources (Tsay 1999). In this paper, we focus on two contractual formats called
quantity flexibility (hereinafter referred to as QF contract) for the downstream par-
ties and price and quantity guarantees (hereinafter referred to as P&Q guarantee)
for the upstream parties.
10 A Tale of Two Information Asymmetries 191

QF contracts come in many forms (e.g., see Farlow et al. 1995, and Bassok et al.
1997 for the practical uses of these contracts in industries such as the automobile
industry, semiconductors, electronic equipments, energy and natural resources). In
its most general form, a QF contract specifies an initial forecast, and lower and
upper bound constraints on the amount of products that can be ordered from the
supplier. For example, a sample QF contract used by Sun MicroSystems for its key-
board, monitor, and other workstation components (see Farlow et al. 1995) stipu-
lates in January 10,000 units for the initial forecast for the demand in April, and
9975 and 11,025 units for the lower and upper bounds, respectively, on March’s
purchase order (PO) for the April delivery. Similarly, P&Q guarantee contracts are
used by the upstream parties in many different formats. For the variants of such con-
tracts in commodity sectors such as electricity/metal/coffee/natural gas, see Steven-
son (2006) and Creti and Fabra (2007). Basically, these guarantees are contractual
assurances from suppliers to provide a certain minimum quantity/capacity at a fixed
price for the buyer.
Interestingly, so far, there has been little attempt in the literature to understand
what role contractual methods play in a risky and asymmetric information environ-
ment, especially when there are multiple suppliers of varying capabilities competing
for the buyer’s order. Moreover, whether they lead to collusive behaviour or not or,
generally speaking, their impact on the competitiveness of the supply chains did not
receive the due attention. The motivation for our research stems from our interest
in addressing these gaps. Specifically, our objective in this chapter is to study the
following research questions in the context of a decentralized supply chain.
• What motivates a buyer, facing a private demand risk, and a supplier, facing
a private supply risk to share information with his/her supply chain partners?
Under what conditions will they do so? Does demand- and supply-side informa-
tion sharing lead to collusive behaviour in the supply chain? If yes, when?
• How effective are commitment- and performance-based contracts in dealing with
upstream and downstream information asymmetries across the supply chain?
Especially, can QF and P&Q contracts be used as signalling devices providing
the demand-side and supply-side visibilities into the supply chain system? If so,
under which conditions?
• How does the provision of such contracts affect the cost or profit performance of
the supply chain partners?
In order to answer the above questions, we propose a model framework that
consists of a buyer, who uses two competing suppliers to satisfy a random demand
in a risk-neutral setting. See Fig. 10.1 for the high-level representation of the supply
chain model considered in this paper. The extent of demand risk facing buyer is
private information, which is unknown to the two suppliers. The two suppliers differ
in their marginal costs as well as the degree and the amount of information available
about their supply risks. Specifically, one of them (supplier R) is a known and trusted
supplier who can satisfy the buyer’s demand without any risk, although her risk-
free capacity is associated with her supply being more costly. The other supplier
(supplier U), who is relatively new and unproven as far as the buyer is concerned, is
192 M. Gümüş

Spot Market

Unreliable &
Supply Risk
Low-Cost
(Private Info)
Supplier (U)

Demand Risk
Buyer (Private Info)

Reliable &
High-Cost
Supplier (R)

Fig. 10.1 Supply chain model analyzed in this chapter

cheaper than R, but is less “reliable” with higher disruption risk. Furthermore, the
level of capacity risk facing supplier U is private information, which is unknown to
the buyer and supplier R.
We also include in our model framework a spot market, which the buyer and sup-
plier U can access for procurement purposes. The price in the spot market is random,
the origins of randomness being both exogenous and endogenous. The endogene-
ity arises from the fact that the spot market price decreases in the total amount of
mismatch between capacity and demand, and, hence, is correlated with buyer’s de-
mand risk and supplier U’s supply risk. This implies that lack of exact information
about demand risk of buyer and supply risk of supplier U also results in information
asymmetry about the spot market price. Lastly, note that the buyer can also use the
spot market to sell any inventory that he receives from the suppliers in excess of his
requirements, under the caveat that the selling price will be lower than the purchase
price.
The buyer initiates the procurement process by issuing a QF contract to the two
suppliers—U and R. The suppliers then compete horizontally (i.e., play a Nash
game) to decide on their contract terms, which they then submit to the buyer. Based
on the contract terms, the buyer then decides on the optimal order allocation strat-
egy. Both suppliers include their per-unit prices in the contracts. However, U may
decide to include an additional P&Q guarantee as a supply assurance to the buyer.
In our setting, supplier U’s P&Q guarantee involves promising to provide a certain
minimum quantity at a fixed price, irrespective of the level of risk she is facing. If
U offers a guarantee, she may need to use the same stochastic spot market defined
above to satisfy any shortfall (relative to the guaranteed amount).
We develop three models in Table 10.1 on the basis of whether demand and
capacity risks are private information or shared with uninformed parties. We analyt-
ically characterize the equilibrium strategies for all the three models using appro-
priate equilibrium concepts listed in Table 10.1 (refer to Fudenberg and Tirole 2000
for the definitions of these equilibrium concepts). Then, we compare Models 1 and
2 with benchmark model in order to answer our research questions.
10 A Tale of Two Information Asymmetries 193

Table 10.1 Models analyzed and compared in this chapter

Demand Supply Section
Model risk info risk info Equilibrium concept used No

Benchmark Known Known Subgame Perfect Nash Equilibrium (SPNE) Section 10.4
Model 1 Private Known Section 10.5
Perfect Bayesian Nash Equilibrium (PBNE)
Model 2 Known Private Section 10.6

The comparative analysis of Model 1 with the benchmark model from buyer’s
perspective in Sect. 10.5 provides us conditions under which the buyer benefits from
demand information sharing. Specifically, we identify two cases. In the first case,
when one of the suppliers can never qualify for the order allocation, the buyer does
not benefit from demand sharing because it leads to higher prices as expected from
a monopolist supplier. In the second case, when both suppliers can compete, we
identify two essential roles of demand sharing, both of which are of strategic nature.
First one is that, contrary to the collusion fears, buyer can indeed use demand shar-
ing to fire up the degree of competition at the upstream level, and prevent them from
charging higher prices. Second one is that QF can be used as a signaling device for
the buyer to share demand information in a credible fashion. We also analyze the
impact of credible demand sharing on the suppliers’ profits, and identify conditions
under which they benefit or lose from it and conclude with a discussion about how
credible demand sharing affects the total supply chain efficiency.
Next, we compare Model 2 with the benchmark model in Sect. 10.6. Our analysis
reveals that a P&Q guarantee has two roles, both of which are of strategic nature.
First, we establish that one of the primary motivations for supplier U to provide
such a guarantee is that she can use it to credibly signal her true level of risk to
the buyer. Indeed, such guarantees may afford perfect visibility into the risk of the
supply system and expected spot market price for the buyer, especially when i)
demand is relatively low, or when ii) demand is high while there is considerable
uncertainty about supplier U’s potential risk. While the visibility helps the buyer in
the former case, unfortunately, it may be harmful for the buyer in the latter scenario.
Specifically, a P&Q guarantee can then weaken competition between the suppliers,
allowing both to raise their contract prices. This results in higher expected costs for
the buyer. Interestingly, this also implies that, in an asymmetric setup, supplier R
may prefer to compete with a reliable (guarantee-offering) supplier U.
The rest of the chapter is organized as follows. In the next section, we review the
related research streams, and in Sect. 10.3, we develop the basic model framework.
Subsequently, we analyze the benchmark model in Sect. 10.4, and analyze and com-
pare Models 1 and 2 with the benchmark model in Sects. 10.5 and 10.6, respectively.
Finally, Sect. 10.7 concludes and discusses future research directions.
194 M. Gümüş

10.2 Related Literature

Our research falls within the general theme of managing demand and supply risk in
an asymmetric information setting, an area with a growing body of literature. In con-
sistent with our focus, we divide our literature review based on whether information
asymmetry is on the demand or supply side.
Demand-Side Information Asymmetry There are two streams of research in the
literature that consider the impact of demand-side information asymmetry on the
supply chains.
Papers in the first stream assume that supply chain partners make their decisions
in a cooperative manner (i.e., they all minimize same system-wide costs), however,
they differ in terms of amount of demand information available to them before
making operational decisions. In general, the demand information asymmetry be-
tween supply chain partners leads to a variance increase in upstream direction; a
phenomenon known as bull-whip effect (Lee et al. 1997). Several models are devel-
oped to quantify the magnitude of bull-whip effect under various demand models;
Chen (2003) provides a comprehensive review on the relation between information
sharing and bullwhip effect.
The second stream of research explores the role of contracts in dealing with in-
formation asymmetry in the demand forecasts. There are two categories of papers
in this stream: (1) analytical models, and (2) mixture of analytical and empirical
models. Among the papers in the first category, Cachon and Lariviere (2001) ana-
lyzes buyer’s incentives to share demand forecast information and the role of com-
pliance regimes in this context. Özer and Wei (2006) develop both screening and
signaling contracts to elicit demand forecast information from the buyer and evalu-
ate them in terms of their effects on supply chain efficiency. Ebrahim-Khanjari et al.
(2012) build a multi-period model between an informed party (a salesperson acting
on the behalf of manufacturer who has private demand forecast information) and
uninformed one (retailer). They analyze the trust relationship between parties in the
context of forecasting sharing. In the second category, Özer et al. (2011) develop
an analytical model to investigate the credibility of the informed party (buyer), and
conduct human-subject experiments to test the hypothesis obtained from the equi-
librium analysis of their analytical model. Also, using a dyadic supply chain model,
Oh and Özer (2013) analyze the role of time in forecast information sharing in a dy-
namic setting where firms’ forecast information and hence, the asymmetry among
such information evolves over time.
There are two distinguishing features between our model (Model 1) and the
above ones. First, our main objective is to analyze the impact of demand infor-
mation sharing on the degree of competition in the supply chain. Therefore, we
explicitly consider a supply chain, where the buyer sources from two competing
suppliers at the upstream level, whereas the above papers consider a dyadic sup-
ply chain configuration that consists of a single buyer and a single supplier. Sec-
ond, we model a secondary market which, if needed, serves to clear the supply-
demand mismatch at an endogenous price that depends on the degree of mismatch.
10 A Tale of Two Information Asymmetries 195

The above models consider either no clearance mechanism or an exogenously det-

ermined salvage value for the excess units. Considering the endogenous clearance
mechanism enables us to explore the link between demand information and supply-
demand mismatch risk, which in turn affects the buyer’s incentive for sharing infor-
mation with his suppliers.
Supply-Side Information Asymmetry There are two streams of research directly
related to our model. We refer the readers to excellent reviews by Tang (2006) and
Vakharia and Yenipazarlı (2009) for a detailed discussion of other relevant supply
risk management strategies.
The first relevant stream of literature studies issues related to exogenous supply
risk in centralized decision-making setting (e.g., Ciarallo et al. 1994; Farmer 1994,
and references therein). Recently, several papers extend this stream by modeling
exogenous supply risk in the form of supplier default by using non-cooperative (e.g.,
Babich et al. 2007) and cooperative game theories (e.g., Huang et al. 2015). How-
ever, in these papers, the extent of risk faced by the suppliers is common knowledge
in the supply chain. Evidently, we add to this literature stream by incorporating
information asymmetry about supplier U’s disruption risk in the model framework.2
Indeed, the most relevant stream of research for us is the one that addresses the
issue of information asymmetry among decentralized channel partners. Our research
is more closely associated with papers that model information asymmetry about sup-
plier risk or reliability (see Gurnani and Shi 2006; Tomlin 2009; Yang et al. 2009;
Nikoofal and Gümüş 2014, 2015). Gurnani and Shi (2006) consider a bargaining
approach where a buyer and a supplier have different estimates about supply relia-
bility. In their model, the players do not update their beliefs and the contract terms
reflect their relative beliefs about supply reliability. In contrast, Tomlin (2009) con-
siders the case of a buyer who has forecast of a supplier’s yield distribution, which
is updated based on the buyer’s experience with the supplier. A Bayesian model of
supply learning is analyzed to evaluate the effects of supply learning on sourcing
and inventory strategies. In Yang et al. (2009), the buyer designs a menu contract,
and subsequently private information about supplier reliability is revealed through
contract choices made by the supplier. Nikoofal and Gümüş (2014) analyze and
compare the incentive- and audit-based mechanisms for managing supplier’s relia-
bility improvement when the supplier has private information about his true relia-
bility. Nikoofal and Gümüş (2015) build a dyadic supply chain model with a buyer
who contracts manufacturing of a new product to a supplier. Due to the lack of
experience in manufacturing, extent of supply risk is not known to both buyer and
supplier at the time of contract. Using this setting, they evaluate impact of diagnostic
test on new product development projects. Note that most of the above papers use
a screening approach to model the buyer’s contracting problem. A novel feature of
our model is that, since the P&Q contract is offered by the supplier (U), the resulting
problem is a signaling game as opposed to a screening one. In screening games, the

2 There are a number of papers within this stream that model endogenous supply uncertainty. This
stream is known as random yield literature, and we refer the readers to Huh and Nagarajan (2010)
and references therein for details.
196 M. Gümüş

buyer devises contract mechanisms to identify the supplier type by eliciting private
information about her reliability level. In contrast, in our model, supplier U, who
has private information, offers a guarantee contract to signal her degree of reliabil-
ity to the buyer. Moreover, suppliers U and R actively compete for the buyer’s order
through their contract terms in our framework; such competition is less relevant in
a screening contract (the suppliers either accept or reject a contract). Note that a
significant body of economics literature examining signaling contracts has emerged
since the seminal paper on job-market signaling by Spence (1973). We also refer
the readers to Riley (2001) for a detailed review of economics literature related to
both signaling and screening contracts. However, this stream of literature is not con-
cerned with the issue of supply risk.

10.3 Model Framework

In order to model the impact of demand- and supply-side information asymmetries

on competitive supply chains, we consider a typical bi-level decentralized supply
chain model with two competing suppliers at the upstream level and a buyer at the
downstream level. The buyer is exposed to demand uncertainty, whose distribution
is known only to him.3 In response to that, two suppliers at the upstream layer
engage in price competition to be able to get order allocation from the buyer. How-
ever, they differ in terms of their exposures to the capacity risks and cost structure.
Specifically, the unreliable one (supplier U) has access to a random capacity at a
lower marginal cost of cU per unit. Furthermore, the level of capacity risk facing
supplier U is private information, which is unknown to the buyer and supplier R.
In contrast, the reliable one (supplier R) can always secure ample (i.e., risk-free)
capacity to satisfy the demand of the buyer albeit at a higher marginal cost of cR per
unit, where 0 ≤ cU ≤ cR . Finally, an important entity of our model is the existence of
a platform (spot market) which enables the buyer to clear all sorts of supply-demand
mismatches. Before modelling the interaction of all these four entities in more de-
tails, we define the distribution of information among supply chain partners, the
specific forms of demand and supply uncertainties, and the price formation in the
spot market as follows:
Information Asymmetry For the sake of analytical tractability, we assume that
the capacity and demand uncertainties can be of two types; high (h) and low (l), and
use superscripts θ ∈ {l, h}, and φ ∈ {l, h} to denote the true capacity and demand
uncertainty types, respectively. Furthermore, we assume that the true capacity and
demand uncertainty types are known only to supplier U and the buyer, respectively.

3 Throughout the chapter, we use masculine and feminine pronouns for the buyer and suppliers,

respectively.
10 A Tale of Two Information Asymmetries 197

Uninformed parties (buyer and R for θ and U and R for φ ) have a-priori beliefs4 on
φ φ
the type, denoted by rSθ , and rD , where rSθ and rD are the probabilities of the capacity
and demand uncertainty types being equal to θ ∈ {l, h}, and φ ∈ {l, h}, respectively.
φ
We also assume that rSθ and rD are common knowledge.
φ
Demand and Capacity Uncertainties Let εSθ and εD represent the capacity and
demand uncertainties faced by θ -type supplier U and φ -type buyer, respectively.
In order to develop analytical managerial insights, we assume that both demand
and capacity uncertainties are independently distributed and have two-point distri-
butions, where low and high states in the support of the distributions are represented
by 0 and Q, respectively5 as follows:
! !
θ 0 with prob. α θ φ 0 with prob. 1 − β φ
εS = θ
and εD = (10.1)
Q with prob. 1 − α Q with prob. β φ .

Note that α θ ∈ [0, 1] represents the probability of θ -type capacity distribution being
equal to low state (i.e., 0), and β φ ∈ [0, 1] represents the probability of φ -type
demand distribution being equal to high state (i.e., Q). In order to make sure that
h-type capacity uncertainty is stochastically riskier6 than l-type capacity, and h-type
demand uncertainty is stochastically larger than l-type demand, we assume α h ≥ α l
and β h ≥ β l .
Price Formation in Spot Market Depending on capacity and demand realizations,
the spot market’s price will be equal to
φ φ
pS (εSθ , εD ) = p̄S + (1 − ρ )ΔS + ρΔE (εSθ , εD ). (10.2)

Note that spot market price is determined by the combination of three components:
(1) a deterministic term, denoted by p̄S ≥ 0, (2) an exogenous stochastic term, ΔS ,
where ΔS is distributed by FS with mean zero (E [ΔS ] = 0), and finally, (3) an en-
φ
dogenous stochastic term ΔE (εSθ , εD ). The first two terms in Eq. (10.2) determine
the spot market price in the absence of the supply-demand mismatch, and account
for exogenous factors such as macroeconomic conditions, industry or nature-related
shocks (refer to Frankel and Rose 2009, for examples). The last term accounts for
the price volatility due to the supply-demand mismatch:

φ
4 Later, we will also define posteriors, denoted by r̂Sθ , and r̂D , to represent the updated beliefs for
type-θ capacity, and type-φ demand, respectively. Posteriors are derived from priors via Bayesian
updating after the suppliers observe buyer’s actions.
5 “Two-point” distributions are commonly used in asymmetric information literature for analytical

tractability. Refer to Yang et al. (2009) and references therein for examples.
6 “h-type capacity distribution is stochastically more risky than l-type capacity distribution” im-

plies that h-type capacity is more likely to be in low state than l-type capacity. Similarly, “h-type
demand distribution is stochastically larger than l-type demand distribution” implies that h-type
demand is more likely to be in high state than l-type capacity. For the rigorous definition and
treatment of stochastic orders, see Shaked and Shanthikumar (1994).
198 M. Gümüş
⎧ φ
⎪
⎨+Δ if εSθ − εD = −Q (with prob. α θ β φ )
φ φ
ΔE (εSθ , εD ) = 0 if εSθ − εD = 0 (with prob. α θ − 2α θ β φ + β φ ) (10.3)
⎪
⎩ φ
−Δ if εSθ − εD = +Q (with prob. (1 − α θ )(1 − β φ )),

φ
where Δ ≥ 0. Note that ΔE (εSθ , εD ) increases or decreases the spot market price
by Δ , respectively, depending on the degree of the supply-demand mismatch. The
former happens when “low supply and high demand” state (supply scarcity) occurs
with probability α θ β φ and the latter happens when “high supply and low demand”
state (supply abundance) occurs with probability (1 − α θ )(1 − β φ ) (refer to Kazaz
and Webster 2011, and references therein for examples on the relationship between
spot market price and supply-demand mismatches). As indicated in the price ex-
pression in Eq. (10.2), the relative impact of supply-demand mismatch on the spot
market price is also controlled by another parameter: ρ ∈ [0, 1]. Combining the joint
effect of ρ and Δ , we can define ρΔ which measures the volatility in the spot market
induced by the supply-demand mismatches. As a final note, we assume that there
is a positive difference (a.k.a. spread) between buying and selling prices in order to
prevent the arbitrage opportunity through continual buying and selling from/to the
spot market. This spread is denoted by δ > 0. One can also interpret δ as a parame-
ter measuring the transaction costs involved in and/or the ease of access to the spot
market.
The four entities introduced above (i.e., the buyer, two suppliers and the sec-
ondary market) interact with each other in three decision stages. The timing of
stages, decisions and events is illustrated in Fig. 10.2.
We discuss each stage in detail as follows:
• Buyer and supplier U observe true demand and capacity distributions,
respectively.
• Commitment and Competition Stages:
– The buyer requests from the suppliers to send their price quotations by issuing
a QF contract, in which he decides on m, and M, where m and M correspond
to, respectively, the upper and lower bounds on the total quantity requested
from the suppliers.

Fig. 10.2 Timeline of decisions and events

10 A Tale of Two Information Asymmetries 199

– In response to the request from the buyer, supplier R and U simultaneously

decide on their contract terms; pR and (pU , pG , qG ), respectively.
• Order Allocation Stage:
– Based on the contract terms from the suppliers, the buyer decides on how
much to order from each supplier. Let qU and qR be the order quantities for
unreliable and reliable suppliers, respectively. Then, buyers total order quan-
tity should be between m and M, i.e., m ≤ qU + qR ≤ M.
• Clearance Stage:
– Demand and supply uncertainties are realized, and spot market price pS (εSθ , εDθ )
is formed as provided in Eq. (10.2).
– Supplier R delivers qR . If supplier U’s realized capacity turns out to be
less than what she committed to deliver, i.e., εSθ ≤ min(qU , qG ), she pur-
φ
chases the shortfall from the spot market at the price pS (εSθ , εD ) and delivers
θ
min(qU , max(qG , εS )). Finally, if the total amount delivered by the suppliers
turns out to be less than the realized demand, i.e.,
φ
qR + min(qU , max(qG , εSθ )) ≤ εD
φ
then, the buyer purchases the shortfall, i.e., εD −(qR +min(qU , max(qG , εSθ ))),
φ
from the spot market at the price pS (εSθ , εD ). Otherwise, if the total amount
delivered by the suppliers exceeds the realized demand, then the buyer dis-
φ
poses the excess inventory, i.e., (qR + min(qU , max(qG , εSθ ))) − εD , in the spot
φ
market at the price pS (εSθ , εD ) − δ .
Before analyzing the equilibrium strategies for each model, we summarize the
list of notations for the problem parameters and decision variables in Table 10.2.

10.4 Symmetric Information Case

In this section, we restrict our attention to the full information case. Note that this
assumption transforms the problem to a symmetric information Stackelberg game
between two competing suppliers at the upstream level and the buyer at the down-
stream level. Hence, we start the characterization of the equilibrium strategy from
the last stage, i.e., the buyer’s allocation problem given that suppliers U and R offer
their contracts. First, we can show that under full information case, QF and P&Q
guarantee contracts do not play any role. Therefore, without loss of generality, we
assume that M = Q, and m = pG = qG = 0. Letting K θ ,φ (qU , qR ; pU , pR ) denote the
expected total cost of φ -type buyer if he orders qU and qR from U and R, respec-
tively, buyer’s order allocation problem, then, can be written as follows:

min K θ ,φ (qU , qR ; pU , pR ) (10.4)

qU ,qR
200 M. Gümüş

Table 10.2 Notation of problem parameters and decision variables

Parameters

θ ; rSθ Type of capacity uncertainty observed by U; a-priori probability of supply type being
equal to θ
φ
φ ; rD Type of demand uncertainty observed by the buyer; a-priori probability of demand un-
certainty type being equal to φ
φ
εD ; β φ Random variable representing the demand uncertainty of the φ -type buyer and its high-
state probability
εSθ ; α θ Random variable representing capacity uncertainty of θ -type supplier U and its low-state
probability
cU ; cR Unit capacity costs of supplier U and R, respectively
p̄S Expected spot market price in the absence of exogenous and endogenous uncertainties
Δ; ρ Parameters representing the impact of supply-demand mismatch on the spot market price
δ The positive spread between buying and selling price in the spot market
Decision variables

m; M Minimum and maximum total quantities requested by the buyer.

pG ; qG Price and quantity guarantee offered by the supplier U.
pU ; pR Unit prices quoted by the supplier U and R, respectively.
qU ; qR Buyer’s order allocation decisions for the supplier U and R, respectively.

subject to 0 ≤ qU ≤ Q and 0 ≤ qR ≤ Q. Since buyer’s total cost is a linear function

of qU and qR , depending on the comparison between pU and pR , on the equilibrium,
either supplier U gets the order, i.e., qU = Q and qR = 0 or supplier R gets the order,
i.e., qU = 0 and qR = Q or neither of them get, i.e., qU = qR = 0 (in this case, the
buyer purchases from the spot market).
Given buyer’s allocation decision, the competition at the upstream level can be
analyzed as a pure Bertrand price competition game between two suppliers with
different marginal costs. This implies that the suppliers undercut each other’s offers
continuously, which drives down the prices to the marginal cost of the inefficient
supplier (i.e., the supplier with higher marginal cost). Hence, on the equilibrium, the
supplier that can deliver the units to the buyer at the lowest (total landed) cost wins
the order, and offers a unit price that is epsilon below the buyer’s indifference point.
The resulting equilibrium strategies for the supply chain parties are characterized in
the following proposition7 :

Proposition 1. Under full information, in equilibrium, the buyer sources from

⎧
⎪ φ θ
⎨spot market, when 0 ≤ β ≤ min(γU , γR − α γδ ),
supplier U, when γU < β φ ≤ γU − γδ + (γR − γU )/α θ ,
⎪
⎩
supplier R, otherwise,

7Since the analyses throughout the chapter are based on Gümüş et al. (2012) and Gümüş (2014),
we omit the proofs and refer the readers to them.
10 A Tale of Two Information Asymmetries 201

−
−

−
−

Fig. 10.3 Equilibrium regions and order allocation under symmetric information case

where
p̄S − cU p̄S − cR δ
γU = 1 − , γR = 1 − , γδ = 1 − .
ρΔ + δ ρΔ + δ ρΔ + δ
The equilibrium sourcing regions are plotted in Fig. 10.3. Note under full infor-
mation, all the firms in the supply chain know true demand and capacity uncertainty
types, hence, the regions can be illustrated on an α θ -β φ plane.

10.5 Asymmetric Demand Information

In this section, we analyze the case where supply risk information is known to all
the supply chain parties but the true state of demand risk is only known to the buyer.
φ
Hence, the suppliers U and R need to rely on their a-priori beliefs, rD , for demand
uncertainty types φ ∈ {l, h} to come up with their contractual terms.
Again, we start the analysis from buyer’s allocation problem. Similar to the full-
information model, without loss of generality, we can assume that pG = qG = 0.
Following the same approach in Sect. 10.4, we can formulate φ -type buyer’s opti-
mal allocation problem given pU and pR . Indeed, φ -type buyer’s allocation decision
is exactly same as in the symmetric information setting. However, in contrast to
symmetric information case, there are two allocation policies associated with two
demand types, and suppliers don’t know which allocation policy eventually to be
used by the buyer. Exactly the same price pair, (pU , pR ), may lead to different sourc-
ing decisions depending on the true demand uncertainty type. In such cases, we will
show that a simple use of price undercutting argument may not always provide us
with an equilibrium characterization for pU and pR .
202 M. Gümüş

Before we go into details of characterizations of such complicated cases, we will

first restrict our attention to a simpler case, in which at least one of the suppliers can
never qualify as a sourcing partner for the buyer, irrespective of the demand type.
We call these cases as non-competitive cases, which is formally defined as follows:

Definition 1 (Non-competitive). Suppose that there exists at least one supplier who
is undercut by the spot market when the demand uncertainty is of l-type. Such case
is called non-competitive if the same supplier is undercut by either the spot market
or the other supplier when the demand uncertainty is of h-type.

Using order allocation functions under l and h-type demand uncertainties, we can
derive necessary and sufficient condition for non-competitive cases in terms of cost
and risk parameters. For instance, when supply risk is low (i.e., 0 ≤ α θ ≤ (γR − γU )/
(1 − γU + γδ )), supplier R is undercut by the spot market irrespective of the demand
uncertainty type if and only if β l ≤ β h ≤ γR − α θ γδ .
Based on Definition 1, we can show that the suppliers satisfy the non-competitive
condition if and only if β l ≤ NC(β h ), where
⎧ h
⎪
⎪ β if 0 ≤ β h ≤ max(γR − α θ γδ , γU )
⎪
⎨β  ( β h ) if max(γR − α θ γδ , γU ) < β h
NC(β h ) = (10.5)
⎪
⎪
⎪ ≤ max(γU − γδ + (γR − γU )/α θ , γU )
⎩
γU if max(γU − γδ + (γR − γU )/α θ , γU ) < β h ≤ 1

where
γR − α θ γδ αθ
β  (β h ) = − β h.
1 − αθ 1 − αθ
Using the above condition, we can decide which supplier(s) can never qualify
for the order. We call this supplier passive and the other one active supplier. So, the
order goes to either the active supplier or the spot market. Based on this definition,
we can divide non-competitive cases into three subcases: (1) subcase NC1 , where
neither of the suppliers are active, (2) subcase NC2 , where only supplier U is active,
and (3) subcase NC3 , where only supplier R is active. These subcases are plotted in
Fig. 10.4. In NC1 , irrespective of the demand uncertainty type, buyer always sources
from the spot market. However, in NC2 and NC3 , the order goes to either the active
supplier or spot market.
Next, we analyze the buyer’s incentive for signalling the true state of information
to the suppliers. Recall that when the non-competitive condition holds, one of the
suppliers can never qualify for the order. Under such cases, if the demand infor-
mation is truthfully signalled, the active supplier can customize her price based on
the signal. However, if the demand information is not signalled, the active supplier
needs to rely on her a-priori belief to decide her optimal pricing strategy. Therefore,
the absence of the credible demand signal leads to the active supplier charging a
price which is less than what she would have charged if she received the signal.
For example, consider subcase NC2 in Fig. 10.4. The active supplier in this region
is supplier U, and if she is sufficiently convinced that the demand uncertainty type
10 A Tale of Two Information Asymmetries 203

a b

−
−
−

−
−
− − −

Fig. 10.4 Equilibrium regions when the demand uncertainty is private information. (a) Low
α θ : 0 ≤ α θ < (γR − γU )/(1 − γU + γδ ); (b) Medium α θ : (γR − γU )/(1 − γU + γδ ) ≤ α θ <
(γR − γU )/γδ ; (c) High α θ : (γR − γU )/γδ ≤ α θ ≤ 1. Panels (a), (b) and (c) above correspond to,
respectively, low, medium, and high α θ values shown in this figure

l is high), she will offer a price low enough to undercut spot market. In this
is l (rD
case, h-type buyer would obtain a surplus by hiding behind l-type, and hence, would
never signal demand information to the supplier. Similar approach can be used to
characterize the equilibrium for subcase NC3 , where the active supplier is R.
Now, we consider the cases in which we can not eliminate any of the suppliers
from the competition. We call these cases as competitive cases. More specifically,
both suppliers are “active” and can undercut spot market. By inverting the condition
in Eq. (10.5), we can show that the competitive condition is satisfied if and only
if NC(β h ) ≤ β l ≤ β h . However, the equilibrium characterization for competitive
cases is considerably more complicated than for the non-competitive cases. This
complication arises due to the following two reasons. First, since both suppliers are
active and competing with each other, suppliers’ prices are interrelated. Secondly,
and more importantly, depending on demand information is credibly signalled or
not, suppliers may or may not know whether they are competing for a buyer facing
l- or h-type demand uncertainties. In the former case, they can uniquely decide on
what prices to charge, whereas in the latter case, they can not.
204 M. Gümüş

This second point plays a pivotal role in shaping the degree of the upstream sup-
plier competition. First, consider that the buyer credibly signals the true type of de-
mand uncertainty. In this case, the suppliers engage in an aggressive (a la Bertrand)
price competition and settle on an equilibrium where the more competitive one gets
the order and charges a price which is epsilon lower than the other supplier’s offer.
However, the above price undercutting argument does not lead to a stable equilib-
rium among suppliers when the true type of demand uncertainty is not credibly
signalled by the buyer. Similar to non-competitive cases, we divide the equilibrium
analysis for the competitive cases into three subcases: C1 , C2 , and C3 as shown in
Fig. 10.4. In what follows, we consider C1 and delegate the details of other cases to
Gümüş (2014).
Recall that in C1 , supplier U can always (i.e., for both l and h-type demand un-
certainty types) undercut supplier R under full information. However, equilibrium
price for U is different and depends on demand type. Note that undercutting price for
h-type demand uncertainty is always less than that for l-type demand uncertainty.
This implies that in the presence of asymmetric information, an l-type buyer always
has an incentive to imitate the contract offered by an h-type buyer. So, in order for
the h-type to credibly separate himself from the l-type, he needs to design a QF
contract that is too costly for an l-type to mimic. This can be accomplished by the
h-type buyer via a costly self-imposed lower bound on the total order quantity.
We can formalize this as follows: let (ml∗ , M l∗ ) and (mh∗ , M h∗ ) be two different
QF contracts offered by l- and h-type buyers, respectively. In response to this, both
suppliers update their a-priori beliefs on the type of demand uncertainty accordingly
and charge (pUl∗ , pl∗
R ) and (pU , pR ) for l- and h-type buyers, respectively. In order
h∗ h∗

for these contracts to be credible in equilibrium, and updated beliefs to be consis-

tent with the buyer’s contracting decisions, equilibrium contracts and prices should
satisfy the following incentive compatibility (IC) constraints:

IC constraint for h-type buyer: TCBh (pUh∗ , ph∗

R |m , M ) ≤ TCB (pU , pR |m , M )
h∗ h∗ h l∗ l∗ l∗ l∗

IC constraint for l-type buyer: TCBl (pUl∗ , pl∗

R |m , M ) ≤ TCB (pU , pR |m , M )
l∗ l∗ l∗ h∗ h∗ h∗ h∗

φ φ ∗ φ ∗
where TCB (pU , pR |mφ ∗ , M φ ∗ ) denotes the expected total cost incurred by a φ -
type buyer provided that he offers a QF contract designed by a φ  -type buyer, where
φ ∈ {l, h}, φ  ∈ {l, h} and φ = φ  . In other words, the above two conditions ensure
that neither l- nor h-type buyer has any incentive to deviate from his respective QF
contract even if doing so would lead to lower prices. Note that there can be multiple
solutions to the above conditions. In order to eliminate the multiplicity of the equi-
libria, we employ the intuitive criterion, a commonly used equilibrium refinement
concept developed for the signalling games (Cho and Kreps 1987). Indeed, this re-
finement enables us to determine a unique equilibrium contract that is both incentive
compatible and least costly from both the l- and h-type buyers’ perspectives.

Proposition 2. There exist separating equilibria if and only if (β l , β h ) ∈ C1 or

(β l , β h ) ∈ C2 . The equilibrium allocation and prices for these two regions are same
as the symmetric information case provided in Proposition 1.
10 A Tale of Two Information Asymmetries 205

There are two main takeaways from Proposition 2. First, recall that one of the
main concerns shared by many companies for demand information sharing is that
they may lead to collusive behaviour among the suppliers, resulting in higher prices.
Our finding is in contrast to this view and indicates that demand information shar-
ing indeed helps intensify the upstream competition and prevent the suppliers from
increasing their prices.
Second, a buyer who has private information about his true demand risk can con-
vey this information in a credible way via a QF contract. The answer to when credi-
ble information sharing is sustainable in equilibrium depends on two things: (1) the
impact of offering a separating contract on the strength of price competition and
(2) the cost of sustaining a separating equilibrium. Recall that the lack of credibility
may weaken the price competition, which can be restored via a QF contract offering.
However, in order to make the contract offering credible in the eyes of suppliers, the
party who is offering this contract has to burden a cost, so-called signalling cost,
which increases in the degree of demand information asymmetry between upstream
and downstream parties (measured by β h − β l ). Hence, if the degree of information
asymmetry is relatively low (i.e., C1 and C2 ), the signalling cost would also be low,
and the separating contracts can be sustained in equilibrium. On the other hand, if
the degree of information asymmetry is very high (i.e., C3 ), then, the buyer would
find it too costly to sustain separating contracts even if these contracts strengthen
the degree of competition among the suppliers U and R.
Next, we consider effects of demand information sharing in two cases: (1) non-
competitive and (2) competitive cases. Under non-competitive cases, both l and h-
type buyers always (weakly) lose from information sharing. The rationale behind
this relates to the fact that one of the suppliers is always eliminated from the compe-
tition and the other (active) supplier is left with satisfying either both l- and h-type
demand uncertainty scenarios (mass-market strategy) or only one of the demand
uncertainty scenarios (niche-market strategy). Since in both cases the active sup-
plier would end up lowering her price in the absence of credible demand signal, the
buyer who faces a private demand information would never choose to share it with
the suppliers. This implies that the buyer is always better off and the suppliers are
always worse off with asymmetric demand information compared to the symmetric
case under non-competitive cases.
On the other hand, when the demand uncertainties are such that both suppli-
ers actively compete with each other (i.e., when the competitive condition holds
true), the effect of demand information sharing is not always harmful for the buyer
and beneficial for the suppliers. Specifically, as the expected demands facing l- and
h-type buyers get dissimilar, it becomes too costly for suppliers to compete with
each other for the mass-market. Consequently, the suppliers have a greater incen-
tive to follow niche-market scenario by splitting the demand uncertainties among
each other, which in turn weakens the competition. This creates a tradeoff for the
buyer. Namely, the buyer can strengthen the competition by signalling true demand
type via a QF contract, but he has to incur a cost to make it credible. Cost-benefit
analysis indicates that the buyer gains more from a QF contract than he loses as the
degree of demand information asymmetry increases. On the other hand, from the
206 M. Gümüş

suppliers’ perspective, they are generally better off with no information sharing and
prefer a credible demand signal only when the degree of information asymmetry is
sufficiently low.

10.6 Asymmetric Capacity Information

We now focus on analyzing the case where demand information is shared among
all the supply chain parties but the supply uncertainty θ information is known only
to supplier U. That means, the buyer and supplier R need to rely on their a-priori
probabilistic beliefs about θ , given by rSθ in order to make their ordering and pric-
ing decisions, respectively. Also, since the spot market price partially depends on
supplier U ’s capacity uncertainty type, uninformed parties faces information asym-
metry regarding the exact spot market price distribution.
Recall that without loss of generality, in Sects. 10.5 and 10.6, we assumed
pG = qG = 0. However, here, we can no longer make the same assumption. Indeed,
we can show that in the no-guarantee case (i.e., when pG = qG = 0) and asym-
metric capacity information scenario, the only equilibrium is of the pooling type
where both h- and l-type supplier U charge the same price. The underlying reason
for the above result is as follows. In this model setting, U’s marginal supply cost
does not change with respect to her type. That means, both l- and h-type supplier
U’s profit expressions increase in pU . Therefore, if l-type supplier U increases her
price, h-type also follows or vice versa. Consequently, in equilibrium, it would be
impossible to separate l- and h-type supplier U from each other by only considering
their pU ’s.8
Next, we focus on equilibrium characterization under a P&Q guarantee (i.e.,
pG ≥ 0 and qG ≥ 0). The analysis for this scenario becomes quite involved because
the price competition between suppliers U and R as well as the order allocation
decision of the buyer depend on whether or not supplier U would signal the true
type of supply uncertainty to the buyer. However, we are able to exactly character-
ize the equilibria associated with both pooling and separating cases as shown in the
following proposition.

Proposition 3. The type of equilibrium for different parameter ranges are charac-
terized below.
• It is possible to have either a pooling or a separating equilibrium depending
on both the absolute as well as the relative values of the two suppliers’ default
risks. Under a pooling equilibrium, both h- and l-type supplier U charge the
same price, while under a separating one, they charge different prices. More
specifically,

8 When supplier U faces a yield risk (rather than a capacity risk), then Gurnani et al. (2012) shows
that supplier U can signal the true yield risk type by her unit price pU .
10 A Tale of Two Information Asymmetries 207
−
a b

−
−
− −
1 −
1
Fig. 10.5 Equilibrium regions when the capacity uncertainty is private information. (a) β φ ≤ γU ,
b β φ > γU . Panels (a) and (b) in the above figure correspond to β φ ≤ γU and β φ > γU in Fig. 10.3,
respectively

– When expected demand β φ is low (see Fig. 10.5a), the unique equilibrium is
of the separating type.
– When expected demand β φ is high (see Fig. 10.5b), the unique equilibrium
is of the pooling type if h- and l-type suppliers’ default risks—αUh and αUl ,
respectively—are close to each other; otherwise, it is of the separating type.

There are two main take-aways from the above proposition:

• First, unlike no-guarantee case, supplier U can signal her supply risk via P&Q
guarantees. The rationale behind this relates to the fact that the P&Q guarantees
make the marginal supply cost of U contingent on her true type. Namely, the
marginal cost of l-type supplier when she offers guarantee is lower than that of
h-type. This in turn makes mimicking l-type supplier U too costly for h-type.
Therefore, l-type (more reliable) supplier U would be able to employ guarantee
contracts to separate herself from h-type (less reliable) supplier U.
• Second, the signalling ability of U also indirectly affects the degree of upstream
competition as follows. Recall that in no-guarantee case, supplier R knows that
the buyer cannot differentiate l-type supplier U from h-type. Therefore, supplier
R charges a price pR that would make her compete with an average-type sup-
plier U. However, under guarantee contract, even though supplier R does not
receive the signal from supplier U, she knows that buyer would receive it. This
forces supplier R to make a choice between two pricing alternatives: she charges
either (1) a high price, which would enable her win the order against h-type (more
risky) supplier U or (2) a low price that would help her win against both l- and
h-types. Similar to the case of demand-side information asymmetry, we call the
former as niche-market strategy and the latter as mass-market strategy.
Another interesting feature of the guarantee offer is on how the resulting sig-
nal is used by the buyer. For instance, when expected demand is low, the signal
is used to communicate the expected spot market price to the buyer. In that case,
208 M. Gümüş

Table 10.3 Summary of results

Their impacts on
Signals Competition and signaling ability The performance of Supply Chain Firms
Demand-side
signals • Competitive effect: It intensifies • Non-competitive case: The buyer is
the upstream competition. worse off while U and R are better off
• Signalling effect: QF contracts with demand information sharing.
enable the buyer to signal his • Competitive case: In general, the
demand. buyer is better off, while U and R are
worse off with demand signal due to
increased upstream competition.

Supply-side
signals • Competitive effect: It weakens • Low-demand case: In general, buyer
the upstream competition. and l-type are better off while R and
• Signalling effect: P&Q h-type are worse off.
guarantees enable the riskier • High-demand case: Buyer can be
supplier to signal her capacity worse off, while R can be better off
risk level. due to reduced upstream competition.

guarantees generate lower costs for the buyer and lower profits for supplier R. But,
when expected demand is high and the two supplier U types have quite different de-
fault risks, the l-type supplier U is able to use a guarantee contract to get allocation
from the buyer. Such guarantees might then actually reduce the competitive inten-
sity between the two suppliers. Consequently, both of them can charge the buyer a
premium. So, interestingly, guarantees can be harmful for the buyer, and beneficial
for supplier R.

10.7 Conclusions, and Future Research Opportunities

In this chapter, our objective is to understand the impacts of demand- and supply-
side signals on the degree of competition, signalling ability of the informed parties
and the expected costs/profits of the chain partners. We develop three models to
address the above issues and fully characterize the equilibrium solution for each of
them. The analysis of these models provides the insights summarized in Table 10.3
into the effects of demand- and supply-side signals. Note that demand- and supply-
side signals have exactly opposite impacts on the degree of upstream competition.
The former does strengthen the degree of competition, whereas the latter reduces.
The above-mentioned results shed some managerial insights on the role of
demand- and supply-side signals in competitive supply chains. On one hand,
demand-side signal rectifies the competition by pulling the strings of suppliers and
10 A Tale of Two Information Asymmetries 209

increases supply chain efficiency (we call this competition effect). On the other hand,
the buyer may have to incur additional cost in order to establish the credibility of
the signal (we call this signalling cost). Both factors concurrently play a role on the
ultimate information sharing decision. For example, at the initial stage of the rela-
tion among supply chain firms when the degree of information asymmetry is quite
high, it may not be sustainable for the buyer to try to credibly signal about the true
demand information due to the fact that doing so entails incurring high signalling
cost. As the relationship gets more and more established, downstream parties can
lower the signalling cost, and at the same time, gain from the increased competi-
tion, benefiting from credible demand information sharing. The spot market volatil-
ity is also an important factor on demand information sharing decision. A relatively
stable spot market would cap the suppliers’ prices and limits the buyer’s potential
gains from demand information sharing. On the other hand, a highly volatile spot
market increases the cost of outside options for the buyer, which, in turn, enables
the suppliers to increase their prices. Finally, the buyer needs to take into account
the extent of supply risk facing the supplier. Specifically, if the buyer is interacting
with an unreliable supplier (i.e., supplier U), then he would need to impose a lower
bound on the amount of products to be purchased from the supplier. Indeed, mini-
mum capacity reservations, a common contractual form used in the semiconductor
industry, where producers’ yields are not too high due to the high degree of produc-
tion risks, are similar to the agreements with a lower bound on the order quantity.
On the other hand, if the buyer is interacting with a reliable supplier (i.e., supplier
R), then he would instead offer an upper bound on the amount of products to be
purchased. Indeed, as indicated in Farlow et al. (1995), Sun MicroSystems used a
specific form with only an upper bound constraint for the suppliers of relatively low
risk items such as keyboards and monitors.
Next, we briefly discuss the future research opportunities. First, we analyze
demand- and supply-side information asymmetries between the supply chain firms
one at a time, i.e., either the extent of demand uncertainty or the extent of supply
uncertainty is unknown. When both are unknown, some of the results discussed in
the chapter could change. For example, we show that supply-side signals reduce the
upstream competition, whereas the demand-side signals increase it. When both sig-
nals are used simultaneously, the effects may counter-balance each other, resulting
in higher or lower upstream competition depending on whether supply-or demand-
side effect is more dominant. Also, consideration of demand- and supply-side sig-
nals simultaneously would lead to externalities between upstream and downstream
information flows. Analysis of these externalities would help to identify positive
and negative interactions between demand- and supply-side signals in the presence
of dual information asymmetries. Another related question is whether the credible
supply-side visibility fosters or hinders the credible demand-side visibility. This has
practical implications for supply chain management because the central tenets of
collaborative, planning, forecasting and replenishment (CPFR) are built on credible
and end-to-end supply chain visibilities, which requires both upstream and down-
stream information flows. Therefore, understanding whether there exists a reinforc-
ing or attenuating relationship between the credibility of upstream and downstream
information flows is critical in successful implementation of CPFR.
210 M. Gümüş

Second, we consider supply chains where competition exists only at the upstream
level. This is in general the case for a typical B2B transaction, where the buyer
initiates the procurement by issuing a request for proposal (RFP) or request for
quote (RFQ). However, the emergence of e-marketplaces (such as Alibaba.com)
where multiple upstream and downstream parties interact simultaneously necessi-
tates for an explicit consideration of supply chains with both upstream and down-
stream competitive structures. Moreover, under such many-to-many supply chain
configurations, the impact of credible supply- and demand-side signals on the de-
gree of vertical and horizontal competitions would deepen our understanding about
the management of information flows across the supply chains.
Third, throughout the chapter, we restrict our attention to risk-neutral setting.
However, the use of signals do not only affect expected profits and/or costs of the
supply chain partners. For example, we show that P&Q guarantees may increase
the buyer’s expected costs under asymmetric information settings. But, they also
reduce the variability in terms of the buyer’s costs. This suggests that if we consider
the implications on both the mean and the variability of the buyer’s costs, it becomes
more likely that a risk-averse buyer would benefit from a guarantee contract. Sim-
ilarly, even though the absence of credible demand signal enables the suppliers to
raise their prices, it also causes them to win the order only under restricted demand
cases, which increases the variability of their profits. Therefore, a risk-averse sup-
plier would prefer to receive a credible demand signal in order to secure the order
under all demand scenarios.
Finally, we assume that both demand- and supply-side information flows across
the supply chain in a non-verifiable form. That means, the informed party (the buyer
or the supplier) has to convey information to the uninformed parties via a credible
(and costly) signal. In information economics literature, this form of information
is called soft information. The other possibility corresponds to the case of hard in-
formation, where the uninformed party can easily verify the validity of information
conveyed by the informed party if the latter decides to disclose it. So, the key ques-
tion is whether or not an informed party would disclose voluntarily the hard infor-
mation, and when the disclosure is sustainable in equilibrium. Seminal papers by
Grossman (1981) and Milgrom (1981) prove a famous result known as information
unravelling principle, which states that verifiability of hard information induces the
informed parties to voluntarily share all the information with the uninformed par-
ties. The follow-up studies then identify the specific conditions under which unr-
avelling principle fails to hold (see Dranove and Jin 2010, for a recent literature
review). One of these conditions that may hinder unravelling of information is the
existence of competitive forces between informed and/or uninformed parties (Board
2009). Although to the best of our knowledge, unravelling of hard information has
not been studied in the context of supply chain management, we think that com-
petitive supply chains could create a rich framework to analyze the interactions
between upstream/downstream competition and information flows in the case of
hard information.
10 A Tale of Two Information Asymmetries 211

We end by hoping that this research spurs work on the above extensions that
provides guidance to supply chain managers regarding how to address two of their
main concerns—managing information flows and dealing with demand and supply
risks in supply chains.

References

Atallah M, Elmongui H, Deshpande V, Schwarz L (2003) Secure supply-chain protocols. In: IEEE
international conference on E-Commerce – CEC 2003. IEEE, New York, pp 293–302
Atallah M, Bykova M, Li J, Frikken K, Topkara M (2004) Private collaborative forecasting and
benchmarking. In: Proceedings of the 2004 ACM workshop on privacy in the electronic society,
ACM, New York, pp 103–114
Babich V, Burnetas A, Ritchken P (2007) Competition and diversification effects in supply chains
with supplier default risk. Manuf Serv Oper Manag 9(2):123–146
Barratt M (2004) Unveiling enablers and inhibitors of collaborative planning. Int J Logist Manag
15(1):73–90
Bassok Y, Bixby A, Srinivasan R, Wiesel HZ (1997) Design of component-supply contract with
commitment-revision flexibility. IBM J Res Dev 41(6):693–703
Bernstein F, de Vericourt F (2008) Competition for procurement contracts with service guarantees.
Oper Res 56(3):562–575
Board O (2009) Competition and disclosure*. J Ind Econ 57(1):197–213
Cachon G, Lariviere M (2001) Contracting to assure supply: how to share demand forecasts in a
supply chain. Manag Sci 47(5):629–646
Chen F (2003) Information sharing and supply chain coordination. In: Handbooks in operations
research and management science, vol 11. Elsevier, Amsterdam, pp 341–421
Cho IK, Kreps DM (1987) Signaling games and stable equilibria. Q J Econ 102(2):179–221
Ciarallo F, Akella R, Morton T (1994) A periodic review, production planning model with un-
certain capacity and uncertain demand-optimality of extended myopic policies. Manag Sci
40(3):320–332
Creti A, Fabra N (2007) Supply security and short-run capacity markets for electricity. Energy
Econ 29(2):259–276
Dranove D, Jin GZ (2010) Quality disclosure and certification: theory and practice. J Econ Lit
48(4):935–963
Ebrahim-Khanjari N, Hopp W, Iravani SM (2012) Trust and information sharing in supply chains.
Prod Oper Manag 21(3):444–464
Farlow D, Schmidt G, Tsay A (1995) Supplier management at Sun Microsystems. Case study.
Graduate School of Business. Stanford University, Palo Alto, CA
Farmer A (1994) Information sharing with capacity uncertainty: the case of coffee. Can J Econ/Rev
canadienne d’Econ 27(2):415–432
Fliedner G (2003) CPFR: an emerging supply chain tool. Ind Manag Data Syst 103(1):14–21
Frankel JA, Rose AK (2009) Determinants of agricultural and mineral commodity prices. Working
Paper, Haas School of Business, UC Berkeley
Fraser J (2003) CPFR—Status and Perspectives: key results of a CPFR survey in the consumer
goods sector and updates. Collaborative planning, forecasting, and replenishment—how to
create a supply chain advantage, American Management Association, New York, pp 70–93
Friedman T (2007) The world is flat: a brief history of the twenty-first century. Picador (imprint).
Farrar, Straus and Giroux, New York
Fudenberg D, Tirole J (2000) Game theory. The MIT Press, Cambridge
Grossman SJ (1981) The informational role of warranties and private disclosure about product
quality. J Law Econ 24(3):461–483
212 M. Gümüş

Gurnani H, Shi M (2006) A bargaining model for a first-time interaction under asymmetric beliefs
of supply reliability. Manag Sci 52(6):865–880
Gurnani H, Gümüş M, Ray S, Ray T (2012) Optimal procurement strategy under supply risk. Asia
Pac J Oper Res 29(1):1240006/1-1240006/31
Gümüş M (2014) With or without forecast sharing: Competition and credibility under information
asymmetry. Prod Oper Manag 23(10):1732–1747
Gümüş M, Ray S, Gurnani H (2012) Supply-side story: risks, guarantees, competition, and infor-
mation asymmetry. Manag Sci 58(9):1694–1714
Helms M, Ettkin L, Chapman S (2000) Supply chain forecasting—collaborative forecasting sup-
ports supply chain management. Bus Process Manag J 6(5):392–407
Huang X, Boyacı T, Gümüş M, Ray S, Zhang D (2015) United we stand, divided we fall: strategic
supplier alliances under default risk. Manag Sci. Published online. doi: 10.1287/mnsc.2015.
2175
Huh W, Nagarajan M (2010) Technical note—Linear inflation rules for the random yield problem:
Analysis and computations. Oper Res 58(1):244–251
IndustryWeek (2009) Lack of visibility causing elevated supply chain risk. (December
23). http://www.industryweek.com/articles/lack of visibility causing elevated supply chain
risk \hfill\penalty-\@M20696.aspx
Kaipia R, Hartiala H (2006) How to benefit from visibility in supply chains. IJAM 9(1):9–18
Kazaz B, Webster S (2011) The impact of yield-dependent trading costs on pricing and production
planning under supply uncertainty. Manuf Serv Oper Manag 13(3):404–417
Lee H, Padmanabhan V, Whang S (1997) Information distortion in a supply chain: The bullwhip
effect. Management Sci 43(4):546–558
Milgrom PR (1981) Good news and bad news: representation theorems and applications. Bell J
Econ 12(2):380–391
Nikoofal ME, Gümüş M (2014) The value of audit in managing supplier’s process improvement.
Technical report, Working paper, McGill University, Montréal, Canada
Nikoofal ME, Gümüş M (2015) Supply diagnostic incentives in new product launch. Technical
report, Working paper, McGill University, Montréal, Canada
Oh S, Özer Ö (2013) Mechanism design for capacity planning under dynamic evolutions of asym-
metric demand forecasts. Manag Sci 59(4):987–1007
O’Marah K (2009) Supply chain risk: Kanban won’t cut it (July 16). http://www.businessweek.
com/technology/technology at work/archives/2009/07/supply chain ri.html
Özer Ö, Wei W (2006) Strategic commitments for an optimal capacity decision under asymmetric
forecast information. Manag Sci 52(8):1238–1257
Özer Ö, Zheng Y, Chen K (2011) Trust in forecast information sharing. Manag Sci 57(6):
1111–1137
Riley J (2001) Silver signals: twenty-five years of screening and signaling. J Econ Lit 39(2):
432–478
Seifert D (2003) Collaborative planning, forecasting, and replenishment: how to create a supply
chain advantage. Amacom Books, New York
Shaked M, Shanthikumar J (1994) Stochastic orders and their applications: probability and
mathematical statistics. Academic, Boston
Spence M (1973) Job market signaling. Q J Econ 87(3):355–374
Stein T (1998) Extending ERP. Information Week http://www.informationweek.com/686/86iuext.
htm. No. 686(June 15), pp 75–82
Stevenson J (2006) Petro-Canada and Gazprom sign pact for Baltic LNG plant. http://www.
resourceinvestor.com/2006/03/13/petro-canada-and-gazprom-sign-pact-baltic-lng-plant
(March 14)
Tang C (2006) Perspectives in supply chain risk management. Int J Prod Econ 103(2):451–488
Tomlin B (2009) Impact of supply learning when suppliers are unreliable. Manuf Serv Oper Manag
11(2):192–209
Tsay AA (1999) The quantity flexibility contract and supplier-customer incentives. Manag Sci
45(10):1339–1358
10 A Tale of Two Information Asymmetries 213

Vakharia A, Yenipazarlı A (2009) Managing supply chain disruptions. Found Trends Technol
Inform Oper Manag 2(4):243–325
Verity J (1996) Clearing the Cobwebs from the Stockroom. Business Week http://www.
businessweek.com/1996/43/b3498166.htm (October 21)
Yang Z, Aydın G, Babich V, Beil D (2009) Supply disruptions, asymmetric information, and a
backup production option. Manag Sci 55(2):192–209
Chapter 11
Supply Chain Information Signaling
and Capital Market

Guoming Lai and Wenqiang Xiao

Abstract Empirical and anecdotal evidence shows that public firms may take into
account capital market reaction when making their operational decisions. The con-
ventional operations literature has mainly focused on maximizing firms profits.
However, when investors do not have complete information, their valuation may
deviate from the firms true profitability. This chapter discusses how the presence of
asymmetric information and the managers market value interest may influence their
operational decisions. It is shown that operational distortions may arise in equilib-
rium that can hurt the firms true performance. Then, a supply chain mechanism is
presented that can effectively signal the internal information to the investors and
alleviate the market friction. The chapter also discusses several related studies that
enrich the literature in the operations and finance interface.

11.1 Introduction

When investors value a firm, they often go beyond the current earnings figure to
factor in the prospect of the firm’s future performance. For instance, the stock
price of Nike, Inc. came under pressure on December 19, 2014, despite an up-
beat earnings report of its past quarter. The weakening number of futures orders
(scheduled for delivery from December through April) from Japan and emerg-
ing markets triggered investors’ concern about its ongoing performance. They
“watch futures orders closely as an indicator of demand for the current quarter”

G. Lai ()
McCombs School of Business, The University of Texas, Austin, TX 78712, USA
e-mail: [email protected]
W. Xiao
Stern School of Business, New York University, New York 10012, NY, USA
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 215

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 11
216 G. Lai and W. Xiao

(The Associated Press 2014). While in this example the firm’s future revenues can
be estimated from the reported futures orders, there are other scenarios where firms
do not report futures orders due to various uncertainties or such direct demand in-
formation is simply unavailable even for the firm itself.
On the other hand, a firm’s operations information (such as, procurement con-
tracts, inventory level, writeoffs) can often reflect the firm’s knowledge of its de-
mand prospect as well as its management capability, which has thus been used by
investors to guide their valuation. For instance, when it was revealed on Decem-
ber 18, 2009, that Zales, the second-largest U.S. jewelry retailer at that time, refused
to accept tens of millions of dollars of inventory at the end of November 2009, its
stock price plunged 12.7 %. “The cancellation of orders at a busy time of year is an
ominous sign for Zales’ sales prospects,” Milton Pedraza, Chief Executive of Lux-
ury Institute, said of the cancellation. “Anyone who thinks Christmas will be dra-
matically up is fooling themselves. It [cancellation] means they are in trouble, that
they’re not expecting sales to be as good as expected” (Wahba 2009). In another
example, investors eyed on Abercrombie & Fitch’s capability of inventory control
when it reported on May 28, 2013 a sharp decline of sales in the previous quarter
which triggered about a 10 % dive of its stock price. Abercrombie had been slowing
inventory increase in the past years in a bid to sustain a healthier growth. “However,
Abercrombie & Fitch seems to have gone tad too far” and “about 10 % of the 17 %
decline in comparable store sales was due to inventory issues” (Trefis Team 2013).
Translating operations information into accurate valuation is, however, a non-
simple task. As discussed by David Berman, general partner of Berman Capital
Management in New York and a pioneering investor of using inventory information
to pick stocks, the relationship between inventory and earnings is “astoundingly
powerful, but few understand why” (Raman et al. 2005). As for inventory, it fulfills
demand to realize revenues, but it is also costly. Having insufficient inventory risks
losing sales, while excess inventory can result in painful markdowns. In fact, finding
the right position of inventory sometimes is even difficult for a firm itself especially
when the market condition shifts or the management team reshuffles which gives
rise to more uncertainties. To the external investors, a firm’s reported operations
information may often carry different signals. For instance, an increasing inventory
level can either indicate a firm’s inefficiency in matching supply with demand or
rather signal a strong demand prospect (Monga 2012). On the other hand, as the
investors are trying to utilize more operations information to value firms, the internal
management teams of firms are the same sophisticated. For their interest in the short-
term stock performance, the internal management teams may purposely alter their
firms’ operations to provide more cheerful results, which is typically known as real
earnings management. For instance, before reporting, they may “try to artificially
reduce inventories, to make the balance sheet look favorable” (Monga 2012). They
may also overproduce to lower cost of goods sold, reduce discretionary spending,
liquidate unmature investments to report more short-term earnings, etc. (Graham
et al. 2005; Roychowdhury 2006). Such short-term activities not only make it more
challenging to interpret firms’ operations information but also sacrifice firms’ long-
term profits. Hence, there is a pressing need to understand the interaction between
market valuation and operations.
11 Supply Chain Information Signaling and Capital Market 217

The operations literature has mainly focused on the optimization of a firm’s

decisions to maximize its profit. It has little investigated how the interaction with
the capital market may affect a firm’s decisions, and from the investors’ perspective,
how a firm’s operations information can be used for valuation. Moreover, if such
a market interaction can cause operational inefficiencies as observed in practice, a
natural question also arises whether there exists any mechanism that can mitigate
the firm’s incentives. Motivated by these questions, we investigate the inventory
ordering decision of a public firm that not only has the conventional operational
considerations but also has a new financial consideration about its market value.
That is, different from the conventional studies where a firm cares only about its
profit, the firm considered in this chapter is publicly traded and is interested in not
only its true profit but also its market value. Furthermore, there is asymmetric in-
formation between the firm and the investors about its future demand. Specifically,
before making the inventory decision, the firm is able to privately observe a signal
of the demand, which cannot be credibly communicated to the external investors.
Nevertheless, the investors will learn the firm’s inventory decision, based on which
they may infer the information about the demand and value the firm accordingly.
The firm’s objective is to maximize a combination of its short-term market value
and its expected true profit.
We find that the firm’s inventory decision in such a setting may deviate from the
classical inventory solution. Given that the firm’s market value is positively associ-
ated with its demand prospect and the inventory order size can reflect its information
about the potential demand, the firm may have an incentive to inflate its order size
to let the investors believe that its demand is strong even when the true prospect
is weak. On the other hand, the investors are rational who will take into account
the firm’s incentive when valuing the firm. Consequently, a typical signaling game
arises between the firm and the investors. Under certain regularity condition, we
identify a unique perfect Bayesian equilibrium in which ordering distortion may
arise when the firm observes a strong demand prospect and orders more than the
classical optimal level to signal its information to the investors. While the investors
can precisely infer the firm’s demand prospect through its inventory level and thus
correctly value the firm, the ordering distortion hurts the firm’s profitability. The re-
vealing of this operational inefficiency motivates us to further study a supply chain
setting where the firm’s supplier can offer a menu of contracts to the firm, instead
of a fixed wholesale price, for its procurement. Interestingly, we find that one can
properly design the contract menu so that the firm’s demand information can be sig-
naled to the investors through its choice of contract and the inventory decision can
be restored to the efficient level. These insights are useful for firm management and
market valuation.
The remainder of this chapter is organized as follows. Section 11.2 reviews the
related literature. We present the model and analysis in Sects. 11.3 and 11.4, respec-
tively. Section 11.5 discusses several closely related studies, and we conclude in
Sect. 11.6.
218 G. Lai and W. Xiao

11.2 Literature

The impact of market interaction with information signaling was first studied in the
economics and finance literature. In the presence of partially informed investors,
Ross (1977) and Miller and Rock (1985) show how short-term oriented managers
can use issuing debts and dividends to signal their firms’ expected cash flows to
the investors, to obtain consistent valuation. Similarly, Stein (1988) shows that the
manager of a more efficient firm can sell a part of the long-term asset to sepa-
rate from a less efficient firm. Stein (1989) portrays a signaling jamming scenario
where only the current-period earnings are reported to the investors while the man-
ager’s action is unobservable. He reveals that the manager’s short-term objective
can drive him to borrow some long-term profits to inflate the current earnings (such
as, liquidate unmature long-term investments) even though the investors correctly
anticipate it. As further elaborated by Bebchuk and Stole (1993), the type of in-
formation asymmetry plays an important role. They demonstrate with a unified
framework that: when the manager’s action is observable while the firm’s long-
term productivity is private information, a typical signaling game arises in which
the manager who cares about the firm’s short-term market value may overinvest in
the long-term project to signal the firm’s type; conversely, the manager will sim-
ply follow the investors’ expectation about his action to make under-investment in
the long-term project. Several accounting studies also investigate managerial “short-
termism” from the perspective of exploring more effective accounting policies for
information revelation (see, e.g., Dye and Sridhar 2004; Liang and Wen 2007). In all
these studies, the specification of the managers’ short-term objective is exogenous.
Differently, Dybvig and Zender (1991) study an endogenous setting and show that
the firm owner can design an efficient compensation contract to align the manager’s
short-term objective with the first-best solution for the firm. However, it requires that
the owner has the knowledge of all possible payoff realizations, when she designs
the compensation contract. As argued by Bebchuk and Stole (1993), it is often infea-
sible in practice for the owner to have all of the information or redesign the contact
in every period. Garvey et al. (1999) also show, based on a moral hazard setting, that
when the manager has the flexibility to decide the amount of his shares to sell, the
firm’s shareholders may choose an incentive contract which induces short-term bias.
The above studies mainly focus on firms’ strategic decisions by abstracting away
detailed operational factors, while the traditional operations literature has mainly
focused on optimizing operational decisions from a firm’s perspective to maximize
its profit. It is relatively underexplored, from the investors’ perspective, how one
can predict a firm’s performance through the operations information and how a firm
will be operated in the presence of capital market interaction. Some recent stud-
ies are filling this gap. Gaur et al. (2005) investigate the inventory turnover metric
which is often used by investors to evaluate a firm’s performance. They find that
a firm’s inventory turnover can be affected by its gross margin, capital intensity,
and sales surprise, based on which they introduce the “adjusted inventory turns”
concept to better detect a firm’s inventory productivity. Chen et al. (2005, 2007) in-
vestigate the inventory trend of the publicly traded U.S. manufacturing and retailing
11 Supply Chain Information Signaling and Capital Market 219

companies. They find that the average inventory holdings have been decreasing,
and those firms with abnormally high inventories often have abnormally poor long-
term stock returns. Similarly, Hendricks and Singhal (2009) show that inventory
write-offs which are interpreted as an indicator of demand and supply mismatch can
significantly reduce a firm’s stock price. The impact of excess inventory announce-
ments is more negative for firms with high growth prospects. The above studies
reveal the statistical linkages between firms’ financial performances and their in-
ventory metrics based on the industry wide data. It is also important to understand
the operations information of an individual firm. As discussed in Fisher and Raman
(2010), investors try to use the inventory and sales information to make buying and
selling decisions over a specific firm’s stocks. On the other hand, while the external
investors are trying to understand firms’ operations to make better investments, the
internal management teams of firms are sophisticated and they may intentionally
alter their operations for the reporting purpose (Graham et al. 2005; Roychowdhury
2006; Monga 2012). It is thus intriguing to understand how managers will make
their operational decisions and how operations information can be interpreted in the
presence of capital market interaction. This is the place that the studies discussed in
this chapter fit in.

11.3 Model

We consider a public firm whose operations are controlled by a representative man-

ager. The shares of the firm are publicly traded in a capital market with identical,
risk-neutral and rational investors. The external investors may not have complete in-
formation about the firm, and thus the firm’s market value assessed by the external
investors may deviate from its true value in case information inference is not pre-
cise. The manager is self-interested who cares not only about the firm’s long-term
profit but also about its short-term market value.
The firm faces a specific selling event which can have a significant effect on
its value. Before the selling event, the firm needs to procure a product from an
upstream supplier. While the demand is still uncertain, the manager is able to obtain
a forecast before he makes the ordering decision. Ex ante, the forecast, denoted
by i, can be either high (H) with probability λ ∈ (0, 1) or low (L) with probability
1 − λ , and the demand conditional on i is a nonnegative random variable Xi with a
strictly increasing distribution Fi (·) (density fi (·)) over R+ . A high forecast implies
a stochastically larger demand, with FH (x) < FL (x) for all x > 0. Let F̄i (·) ≡ 1−Fi (·).
We assume that there is no replenishment opportunity after the realization of the
demand, and thus any excess demand is lost. In the case of overage, the leftover
inventory has zero value (but can be returned to the supplier based on the contract).
While we assume that the setting of the selling event, the prior of the forecast as
well as the conditional distributions of the demand are all common knowledge, the
specific forecast obtained by the manager is his private information and there is not
220 G. Lai and W. Xiao

a credible way for him to directly communicate this information to the investors.
In practice, there certainly can be more asymmetric information. Here, without loss
of generality, we focus on one specific source of asymmetric information to show
its impact on public firms’ operations.
For simplicity, we assume that the selling price of the product is fixed at p per
unit, and the supplier’s production cost is c per unit. The trade between the firm
and its supplier is carried out through a general form of buy-back contract. We use
(w, b,t) to denote a single contract in which w is the per-unit wholesale price, b
is the per-unit buy-back price, and t is the transfer payment. If b and t are zero,
the contract reduces to a wholesale price only contract. Given a single contract, the
firm procures q units from the supplier. The trade can also be carried out through
a menu of two buy-back contracts, denoted as {(wH , bH ,tH ), (wL , bL ,tL )}, with the
subscripts corresponding to the possible values of the signal. Given a menu of two
contracts, the firm chooses one contract (wi , bi ,ti ) and procures q units from the
supplier. We impose no constraint on the choice of q. The contract, once taken, is
legally binding and is not renegotiable. We assume that the contract information is
accessible to the capital market.
To model the manager’s objective, we first simplify the setting by assuming that
the firm will be liquidated at its fair value at the end of the selling event and the firm
does not have other cash flows other than that from the selling event. We normal-
ize the time discount to zero. Therefore, the investors’ valuation of the firm is the
expectation of its ending profit conditioned on the information they can access. We
then adopt the following objective function: the manager places a weight β ∈ (0, 1)
on the firm’s short-term market value and a weight 1 − β on the firm’s long-term
true profit when he makes the ordering decision. This incentive scheme (captured
by β ) is known to the investors. Such a setting is commonly used in the finance
and accounting literature (see, e.g., Stein 1989; Liang and Wen 2007), and it can
be motivated, for instance, if the manager needs to obtain a certain amount of cash
from his stock-based compensation in the short term, he bears pressure from the
firm’s short-term shareholders, or the firm needs to sell a fraction of its shares to
raise capital.
Figure 11.1 describes the timeline. The supplier first offers the contract. The
manager obtains the demand forecast and makes the ordering decision. The investors
observe the firm’s ordering decision and infer the demand potential, which results
in the firm’s short-term market value. The manager realizes a short-term payoff

The manager observes the demand The capital market observes the firm’s The true value of the firm is
forecast, then he (chooses a contract and) contract choice and stocking level and realized, which determines the
decides the stocking level values the firm accordingly; then the manager’s long-term payoff
manager’s short-term payoff is determined

The supplier offers a single The supplier produces and delivers The demand is realized and the leftover
buy-back contract (or a menu the products to the firm inventory is returned to the supplier for
of buy-back contracts) refund according to the agreed contract

Fig. 11.1 The timeline of the model

11 Supply Chain Information Signaling and Capital Market 221

equal to the market value multiplied by the weight β . After that, the demand and
the corresponding cash flows are realized. The firm is liquidated and the manager
realizes another payoff equal to the firm’s true profit multiplied by the weight 1 − β .
Clearly, in absence of the capital market (or β = 0), what we have described
would be a classical selling to newsvendor problem. An optimized single contract
can maximize the total supply chain surplus. That is, when (w − b)/(p − b) = c/p,
the manager would order qoi ≡ F̄i−1 (c/p) that maximizes the total supply chain sur-
plus (Pasternack 1985). We call this quantity the first-best stocking level. However,
if the manager cares about the firm’s market value, his decision may deviate from
the classical newsvendor solution.

11.4 Analysis

In this section, we first analyze the model with a typical single contract. After reveal-
ing the inefficiencies, we present an operational approach to mitigate the manager’s
market value incentive.

11.4.1 The Single Contract Case

For each forecast i ∈ {H, L}, the expected profit of the firm with a stocking level q
under a single contract (w, b,t) follows:

π B (q; i) = pE [min(q, Xi )] + bE [max(q − Xi , 0)] − wq − t (11.1)

 q
= (p − b) F̄i (x)dx − (w − b) q − t.
0

Based on the classical newsvendor solution, we can find a unique maximizer of

π B (q; i): q∗i ≡ F̄i−1 ((w − b)/(p − b)). However, the manager may not follow this
solution.
To formulate the firm’s market value, we use j to denote the investors’ belief of
the demand forecast: !
H if q ∈ QH ,
j(q) =
L otherwise,
where QH is a subset of R+ . That is, if the observed stocking level q ∈ QH , then
the investors believe that the forecast is high; otherwise, it is low. Then, the firm’s
market value follows:
 q
P(q) = (p − b) F̄j(q) (x) dx − (w − b) q − t. (11.2)
0
222 G. Lai and W. Xiao

The manager’s objective is to maximize a combination of the firm’s profit and

market value:
max β P(q) + (1 − β )π B (q; i). (11.3)
q∈R+

Let q(i) denote the manager’s optimal decision. In equilibrium, the market belief
must be consistent with the manager’s decision. Hence, we introduce the following
equilibrium concept: Given a single contract (w, b,t), a separating equilibrium is
reached if the manager’s optimal stocking decision and the market belief satisfy
j(q(i)) = i so that P(q(i)) = π B (q(i); i) for each forecast i ∈ {H, L}. Notice that this
is a typical signaling game that can potentially have many separating and pooling
equilibria. However, some of the equilibria might not be reasonable or stable. In
such situations, researchers often use some refinement to rule out those equilibria.
Here, we adopt the intuitive criterion that is commonly used in the literature (Cho
and Kreps 1987). In fact, under the intuitive criterion, we can find a threshold β̂
such that given any single contract (w, b,t), a unique separating equilibrium exists,
in which the manager’s stocking decision follows
!
q̂ if i = H,
q(i) = (11.4)
q∗L if i = L,

where q̂ = q∗H when β ≤ β̂ and q̂ > q∗H when β > β̂ , and the market belief is speci-
fied as !
H if q = q̂,
j(q) = (11.5)
L otherwise.
The proof can be found in Lai et al. (2012). In this equilibrium, the manager
stocks q̂ with a high forecast and q∗L with a low forecast, and the market belief is
consistent with the manager’s decision. Notice that if the forecast is low, the stocking
decision coincides with the classical newsvendor solution; however, if the forecast
is high, overstocking can occur when β > β̂ . Given his interest in the firm’s market
value, the manager always wants the investors to believe that the firm has a good
potential. As a result, when observing a low forecast, he may mimic the ordering
decision associated with a high demand forecast. On the other hand, the investors
are rational who know the manager’s incentive from his objective function. They
may not give the firm a high valuation if they cannot infer the true demand forecast
from the firm’s stocking level. Consequently, when the manager observes a high
demand forecast, in order to achieve a fair valuation, the manager may have to stock
more than the classical newsvendor solution to a level that he would not mimic
when the demand forecast is low, which can thus provide a credible signal to the
investors that the demand forecast is truly high. Such an overstocking will arise if
β > β̂ , under which the manager’s mimicking incentive is strong when he observes
a low forecast. Given that the stocking level deviates from the classical newsvendor
solution, it hurts the firm’s profitability. Figure 11.2 illustrates the firm’s equilibrium
stocking level and its expected profit Π B = λ π B (q(H); H) + (1 − λ )π B (q(L); L).
11 Supply Chain Information Signaling and Capital Market 223

The firm’s Expected Profit

Equilibrium Stocking Level

q(H )= q

>
qH*
q(H)=qH*

q(L)=qL*

>
β
>

β β

Fig. 11.2 Demonstration of the equilibrium stocking level and the firm’s expected profit Π B as
functions of β . The parameters are: p = 20, c = 5, w = 8, b = 4, t = 0, λ = 0.5, and the demand
follows the gamma distribution with density fi (x) = (x/κi )θi −1 e−x/κi /(κi Γ (θi )) for i ∈ {H, L},
where (κH , θH ) = (1.5, 5) and (κL , θL ) = (1, 5)

The firm’s ordering distortion will likely affect the supplier’s performance too.
When the firm overstocks, on the one hand, the supplier seems to benefit because
more revenues can be collected; on the other hand, when a buy-back term is pro-
vided, more returns can occur, which is costly for the supplier. To assess these im-
pacts, we first formulate the supplier’s expected profit:
  q̂ 
Π = λ (w − c − b)q̂ + b
S
F̄H (x) dx
0
  q∗ 
L
∗
+ (1 − λ ) (w − c − b)qL + b F̄L (x) dx + t. (11.6)
0

Clearly, when β < β̂ , the supplier’s profit is not affected by the manager’s market
value interest since stocking distortion does not occur. When β > β̂ , the downstream
stocking distortion can either hurt or benefit the supplier, depending on the contract
terms. In particular, if b ≥ ((w − c)/(p − c))p (which can be rewritten as ((p − w)/
(p − b))b ≥ w − c), the downstream stocking distortion always hurts the supplier
(see the left plot in Fig. 11.3). Note that (p − w)/(p − b), which equals FH (q∗H ),
represents the probability that a unit product will be returned to the supplier when
the buyer stocks q∗H , and ((p − w)/(p − b))b captures the marginal refund cost.
Hence, if ((p − w)/(p − b))b ≥ w − c, then the marginal refund cost (FH (q)b) will
always outweigh the marginal revenue w − c when the firm stocks any q beyond
q∗H (given that FH (q) increases in q), which is costly for the supplier. In contrast,
if b < (w − c)/(p − c)p, some amount of overstocking at the downstream firm may
benefit the supplier in that it mitigates double marginalization. A threshold β  can be
determined such that the downstream overstocking will benefit (hurt) the supplier
224 G. Lai and W. Xiao
Supplier’s Expected Profit

Supplier’s Expected Profit

β'
>

>
β β

β β
Supplier’s Expected Profit

>

Fig. 11.3 Demonstration of the supplier’s expected profit Π S as a function of β . The common
parameters are: p = 20, c = 5, w = 8, t = 0, λ = 0.5, and the demand follows the gamma distribu-
tion with density fi (x) = (x/κi )θi −1 e−x/κi /(κi Γ (θi )) for i ∈ {H, L}, where (κH , θH ) = (1.5, 5) and
(κL , θL ) = (1, 5). In the left plot, b = 4; in the middle plot, b = 3.5; and in the right plot, b = 3

when β < (>)β  (see the middle plot in Fig. 11.3). Note that the value of β  can
possibly reach one if the double marginalization effect is strong (see the right plot
in Fig. 11.3). Therefore, for a given contract, the downstream stocking distortion is
beneficial for the supplier if it mitigates double marginalization in scenarios where
β is intermediate, and it is detrimental if the downstream stocking level is distorted
to a large extent when β is large. It is useful to notice that if p becomes larger rel-
ative to c, then the term ((w − c)/(p − c))p becomes smaller and the region where
the downstream stocking distortion is detrimental for the supplier becomes wider.
In other words, for high-margin products, the short-term market value interest at the
downstream firm is more likely to hurt the supplier.
11 Supply Chain Information Signaling and Capital Market 225
Expected Total Supply Chain Surplus

Expected Total Supply Chain Surplus

>

β β''

>
β

β β
Expected Total Supply Chain Surplus

>

Fig. 11.4 Demonstration of the expected total supply chain surplus Π SC as a function of β . The
common parameters are: p = 20, c = 5, t = 0, λ = 0.5, and the demand follows the gamma distri-
bution with density fi (x) = (x/κi )θi −1 e−x/κi /(κi Γ (θi )) for i ∈ {H, L}, where (κH , θH ) = (1.5, 5)
and (κL , θL ) = (1, 5). In the left plot, w = 8 and b = 4; in the middle plot, w = 8 and b = 0; and in
the right plot, w = 14 and b = 0

The impact of downstream stocking distortion on the overall supply chain per-
formance is similar. For a given contract, it can sometimes benefit and other times
hurt the supply chain, as illustrated in Fig. 11.4. More importantly, note that in the
absence of capital market interaction (or β = 0), the supply chain can be coordinated
if (w − b)/(p − b) = c/p (or identically, b = ((w − c)/(p − c))p). However, this is
not always true in the presence of capital market interaction. We find that when β
is greater than a threshold β̂ o , supply chain coordination cannot be achieved with
any single contract. That is, the downstream firm’s interest in its market value will
likely hurt the overall supply chain performance if one uses the conventional sup-
ply chain coordination mechanisms. Although similar operations distortions have
been revealed in the literature, little discussion appears about how to restore system
226 G. Lai and W. Xiao

efficiency. Notice that in our model, improving supply chain efficiency is in both
the downstream firm’s and the supplier’s interest because their profits can be in-
creased if supply chain coordination can be achieved. Naturally, one approach to
improve system efficiency would be to reduce β , for instance, structuring executive
compensation to include fewer market-based incentives; regulation policies might
also help. However, these approaches might not always be appealing or possible.
In Sect. 11.5, we investigate whether operational approaches exist that can improve
system efficiency.

11.4.2 Restoring Efficiency with Operational Mechanism

In the absence of capital market interaction, an properly designed single contract

would be sufficient to coordinate the supply chain even with information asymme-
try among the supply chain parties in our model, and offering a menu of contracts
would be redundant. However, in the presence of capital market interaction, the
single contract becomes ineffective for coordination. A natural question emerges:
whether a menu of contracts is now necessary and sufficient to achieve supply chain
coordination. We explore this question below.
First, notice that the contract design in our model differs from those in the tra-
ditional adverse selection context because our problem involves a third-party, the
capital market; as a result, we need to establish a downstream market equilibrium.
This equilibrium must be a separating equilibrium because full efficiency would not
be achieved otherwise. Let (wi , bi ,ti ) denote the menu of contracts corresponding to
the forecast i ∈ {H, L}. We use τ ∈ {H, L} to denote the manager’s contract choice.
The market infers the forecast from the manager’s decisions by a belief denoted by:
!
H if q ∈ QτH ,
j(τ , q) =
L otherwise,

τ ∈{H,L}
where QH is the set of stocking levels corresponding to the contract τ , for
which the market believes the signal to be high. Recall from Sect. 11.3 that qoi =
F̄i−1 (c/p) is the first-best stocking level. Under a menu of contracts, full efficiency
can be achieved in the supply chain if and only if the first-best stocking level can be
implemented and, at the same time, the market is able to correctly infer the forecast
from the manager’s decisions. Formally, we introduce the following equilibrium
concept: A market equilibrium with a menu of two buy-back contracts is system-
wise efficient if the manager’s decisions follow (τ , q)(i) = (i, qoi ) and the market
belief satisfies j((τ , q)(i)) = i for each forecast i ∈ {H, L}.
Notice that in a system-wise efficient market equilibrium, the set QH H must con-
tain qoH so that a high forecast can be correctly inferred if the manager with a high
forecast chooses the H contract and stocks qoH ; in contrast, QLH must not contain qoL
so that a low forecast can be correctly inferred if the manager with a low forecast
selects the L contract and stocks qoL . Given the structure of the game, any design of
11 Supply Chain Information Signaling and Capital Market 227

the contracts needs to be associated with the characterization of a market belief (i.e.,
τ ∈{H,L}
QH ). Importantly, we find that given any menu of two buy-back contracts, if a
system-wise efficient market equilibrium is achieved with a market belief (QH H , QH ),
L

then the equilibrium can also be achieved with the market belief ({qH }, 0).
o / Under
this restrictive market belief, if the manager chooses the H contract, then he must
stock qoH to be recognized as having a high forecast; if the manager chooses the L
contract then he is automatically considered to have a low forecast. This observation
is very useful, and we now can design the mechanism directly using this restrictive
market belief, which greatly shrinks the searching space.
Notice that for the manager to stock the first-best quantity for each forecast i ∈
{H, L}, the contract terms must satisfy:

wH − bH wL − bL c
= = .
p − bH p − bL p
With this condition, we can determine the wholesale price wi once the buy-back
price bi is given, and vice versa. Further, let F̄i j (x) ≡ β F̄j (x) + (1 − β )F̄i (x),
 q
c
gi j (q) ≡ F̄i j (x) dx − q, ∀i, j ∈ {H, L},
0 p

where the subscript i j is reduced to i when i = j, and qoij ≡ F̄i−1 j (c/p) for i = j.
We can now establish the conditions, under which a menu of buy-back contracts
can restore full efficiency. In particular, with the market belief ({qoH }, 0), / a system-
wise efficient market equilibrium can be achieved if the menu of buy-back contracts
(wi , bi ,ti ) satisfies: (wi − bi )/(p − bi ) = c/p for each i ∈ {H, L}, bL ≥ (1/K)bH +
p(K − 1)/K, and tH − tL ∈ [Δ t , Δ t ] where
 
gHL (qoHL ) − gL (qoL ) gHL (qoHL ) − gL (qoL )
K = max , ;
gH (qoH ) − gLH (qoH ) gH (qoH ) − gL (qoL )
Δ t = (p − bH ) max{gLH (qoH ), gL (qoL )} − (p − bL ) gL (qoL );
Δ t = (p − bH ) gH (qoH ) − (p − bL ) gHL (qoHL ).

In equilibrium, the manager observing the high (low) forecast shall prefer the
H (L) contract and stock the associated first-best quantity even if he takes the mar-
ket value into account. A key condition for achieving this result is the buy-back
prices chosen for the contracts. In general, the manager favors a generous return
term when the demand outlook is pessimistic. Thus, the buy-back price (bL ) of
the L contract shall be attractive enough for the manager with a low forecast to
choose this contract. Specifically, bL shall be no less than a particular threshold level
((1/k)bH + p(K − 1)/K) that is contingent on bH . When this condition is satisfied,
a pair of transfer payments can always be chosen (with their difference bounded
by the two thresholds Δ t and Δ t depending on the buy-back prices) that provides
sufficient incentives for the manager having each forecast to take the truth-telling
contract and stock the first-best quantity under the coordination condition.
228 G. Lai and W. Xiao

The above result indicates that in a supply chain context, the supplier might be
able to “correct” the downstream stocking distortion by offering alternative con-
tract choices. The manager credibly reveals his information through his choice of
contract, in contrast to the overstocking that would otherwise be necessary under a
single contract. A remaining issue to implement the above mechanism is to deter-
mine how the surplus can be divided between the parties in the supply chain. Notice
that the mechanism can be difficult to implement if the resulting payoff is not satis-
factory for one party (e.g., compared with the payoff that can be obtained under an
existing single contract). Interestingly, we find that with the market belief ({qoH }, 0),
/
there always exists a menu of buy-back contracts
!
bH = 0, wH = c, and tH = pgH (qoH ) − ε [gH (qoH ) − gL (qoL )] − T
bL = p − ε , wL = p − ε (1 − c/p), and tL = ε gL (qoL ) − T,

with any constant ε ≤ p/K and T , under which a system-wise efficient market equi-
librium can be reached. The supplier’s profit goes to the total supply chain surplus
as ε and T go to zero.
The above result provides a special menu of buy-back contracts that can achieve
a system-wise efficient market equilibrium. In particular, under this menu of con-
tracts, the supplier is able to obtain almost all of the supply chain surplus as ε and
T go to zero. Because T is a constant appearing in both tH and tL , any specific al-
location of the supply chain surplus can always be achieved by adjusting T . As a
result, it demonstrates that both parties can be feasibly made better off compared
to a single contract scenario (given that the total supply chain surplus is enlarged).
That is, Pareto improvement can be achieved. Note that the above result gives just
one example. One can design menus of contracts with other wholesale, buy-back
and transfer prices for the specific need of implementation, as long as the efficient
equilibrium conditions are satisfied.
Thus far, we have revealed that to offer an appropriately designed menu of buy-
back contracts can resolve the downstream stocking distortion. It is also interesting
to assess the benefit from using such a mechanism. For that purpose, we define a
specific benchmark where, instead of a menu, a single conventional coordinating
buy-back contract is used (that satisfies (w − b)/(p − b) = c/p; the supply chain
obviously would not be coordinated as the downstream buyer will overstock). We
derive the total supply chain surplus of this case, denoted by ΠSNGSC . We also derive

the total supply chain surplus of the case where the proposed mechanism is imple-
mented, that is, the supply chain surplus when it is truly coordinated, denoted by
ΠMNSC . Figure 11.5 provides an example of the benefit from using this mechanism

(the percentage of surplus increment). It is straightforward that the benefit of this

mechanism increases in β , the magnitude of the manager’s interest in market value,
and λ , the probability that the demand signal is high (the manager will distort the
stocking level only if he receives a high signal). The benefit also increases as the
(stochastic) difference of the demands under the two forecasts increases because
the manager will have a stronger incentive to distort the stocking level. Certainly,
the benefit also increases in the profitability of the business. This example shows
that the benefit by using the proposed mechanism can be quite substantial.
11 Supply Chain Information Signaling and Capital Market 229

λ = 0.4 λ = 0.7
Percent of Surplus Increment (%)

Percent of Surplus Increment (%)

β β

Fig. 11.5 Demonstration of the gain of supply chain surplus by using the mechanism. We calculate
the total supply chain surplus ΠSNG SC when a single traditional coordinating buy-back contract (w =

11, b = 8, t = 0) is used; we then calculate the total supply chain surplus ΠMN SC of the case where

the proposed mechanism is implemented (that is, the truly coordinated supply chain surplus) and
compute [(ΠMN SC − Π SC )/Π SC ]×100 %. The other parameters are: p = 20, c = 5, and the demand
SNG SNG
follows the gamma distribution with density fi (x) = (x/κi )θi −1 e−x/κi /(κi Γ (θi )) for i ∈ {H, L} with
(κH , θH ) = (1.5, 5) and (κL , θL ) = (1, 5) such that FH (x) < FL (x), ∀x > 0

11.5 Other Studies

In this section, we discuss several other studies that investigate closely related issues
under the interaction with capital market.

11.5.1 Capacity Constraint and Equilibrium Refinement

In Sect. 11.4, we have assumed that the firm does not have any capacity constraint
and the inventory level can be at any positive real number. In practice, however, firms
may have various constraints so that their inventory levels cannot be at any value.
For instance, due to specific lot size limit, a firm’s inventory may have to be at some
discrete values. Furthermore, the equilibrium analysis in Sect. 11.4 was based on
the intuitive criterion to refine the beliefs of the players in the game. However, there
are also other criteria that can be used to refine the beliefs, which may consequently
result in different equilibrium outcomes.
In a related study, Schmidt et al. (2015) investigate the equilibrium outcomes
of the problem presented in Sect. 11.4, when the firm’s inventory level needs to be
at specific discrete values and also when an alternative refinement—the undefeated
refinement—is used to refine the beliefs. Interestingly, they find that other than the
separating equilibrium we discussed in the above, there can exist one or multiple
pooling equilibria that also survive the intuitive criterion when the inventory level
230 G. Lai and W. Xiao

has to be discrete values. Some of these pooling equilibria can Pareto dominate the
separating equilibrium in the sense that the manager (the firm) is better off under
the pooling equilibria for both forecasts. Clearly, such pooling equilibria if they can
exist will be more appealing to the manager than the separating equilibrium. For the
investors, since in equilibrium their valuation of the firm is always consistent with
the firm’s expected true value, they are indifferent among these equilibria. How-
ever, in the pooling equilibria, the investors will not be able to perfectly infer the
true forecast and they will use the prior to value the firm. Schmidt et al. (2015)
further discuss the potential issues with the intuitive criterion. For instance, it is
well known that the intuitive criterion may result in some extreme cases. For the
problem presented in the above, under the intuitive criterion, even if the forecast
is almost surely to be high, the manager would still need to overstock inventory to
signal his information to the investors when he observes a high forecast. In other
words, the intuitive criterion might not be suitable if the prior of the private knowl-
edge is highly skewed. In view of such potential issues, Schmidt et al. (2015) apply
the undefeated refinement developed by Mailath et al. (1993) to refine the beliefs.
They find that under this alternative refinement, some pooling equilibria will sur-
vive while the separating equilibrium will not. Therefore, a Pareto improvement can
be achieved. These findings are useful to enrich our understanding of the impact of
capital market interaction on firms’ performance.

11.5.2 Characteristics of Information Asymmetry

For the problem presented in Sect. 11.5.1 and also the one studied in Schmidt et al.
(2015), information asymmetry between the manager and the external investors
arises about the demand prospect with respect to its size. More specifically, it is
generally assumed that the demand size in one possible state first-order stochasti-
cally dominates the size in the other state. This assumption implies that the firm’s
potential profits in different states are well ordered, and as discussed in Sect. 11.5.1,
applying the intuitive criterion results in a unique separating equilibrium which is
featured with operational distortion only in the high demand state for the signaling
purpose. However, in practice, a firm (or its demand) may have various characteris-
tics which in different states may not be well ordered. With this motivation, Lai and
Xiao (2015) study a scenario where there is asymmetric information about the firm’s
demand risk (i.e., its variance) between the manager and the external investors. For
instance, the manager may know better than the external investors about the firm’s
demand forecasting capability and risk management capability. They find that when
information asymmetry arises about the demand variance, the firm’s profits in dif-
ferent states are not perfectly ordered and thus the conventional regularity condi-
tion used in prior literature on signaling does not apply. Interestingly, their analysis
shows that in such a scenario, separating and pooling equilibria may coexist, even
after the beliefs are refined by the intuitive criterion. They characterize the con-
ditions under which only separating equilibrium, only pooling equilibrium, or both
11 Supply Chain Information Signaling and Capital Market 231

types of equilibria exist, and then they apply the Pareto dominance principle to select
the most efficient equilibrium. Interesting insights are revealed that in equilibrium a
firm’s inventory can be sometimes understocked and other times overstocked. They
discuss how to interpret a firm’s inventory in such scenarios to infer its operational
efficiency and also offer managerial implications for compensation design.

11.5.3 Sales Channel Stuffing

The studies discussed in the above all assume that after the demand is realized, the
firm’s reported sales are simply the true demand realization. In practice, however,
firms may inflate their reported sales which then exceed the true demand realization.
Beyond accounting manipulations, firms may also use real earnings management to
generate temporary and unsustainable sales by stuffing their downstream channel
(or pushing to individual customers) with unneeded inventory, for instance, by pro-
viding rebates, generous credit terms, future price discounts and other incentives. By
presenting more sales, the firm can not only report more profits in the current period
but also provide a better prospect of its future performance since the future demand
is often correlated with the current realization. As a result, asymmetric information
may also arise about the firm’s true demand that the internal managers know while
the external investors do not observe, which can significantly affect the firm’s mar-
ket value. Lai et al. (2011) study this phenomenon in a similar framework applied
in Sect. 11.5.2. They characterize a hybrid equilibrium with partial separating and
partial pooling, in which the magnitude of sales channel stuffing first monotonically
increases in the realized true demand when it is smaller than a threshold, and after
the demand exceeds the threshold the firm always pushes all the left inventory to the
downstream to inflate the sales. Correspondingly, the investors can perfectly infer
the true demand from the reported sales when the sales are below a threshold, and
they have to form an expectation about the true demand realization after the sales ex-
ceed a threshold. They also find that the incentive to inflate the sales may affect the
initial inventory stocking decision which may deviate either above or below from
the conventional optimal level. These insights are useful for understanding firms’
sales inflating incentives and also for interpreting the reported sales data.

11.6 Conclusion

This chapter discusses how a public firm manager’s short-term interest in the firm’s
market value may influence the performances of the firm and its supplier in the
supply chain. We first show that under a single contract, the manager may dis-
tort the stocking level, which hurts the firm’s profitability. We further find scenar-
ios where supply chain coordination cannot be achieved by any single contract.
Noticing this inefficiency, we discuss how an operational mechanism with a proper
232 G. Lai and W. Xiao

menu of contracts can mitigate the downstream manager’s incentive and reveal int-
ernal information to external investors. While prior research has explored different
purposes for using a menu of contracts, such as, to share and improve demand fore-
casting (Cachon and Lariviere 2001; Özer 2006; Taylor and Xiao 2009), or to elicit
cost and inventory information (Ha 2001; Zhang 2010; Zhang et al. 2010), the usage
and benefit discussed in this chapter appear to be novel and effective.
We conclude this chapter by discussing some directions for future research. First,
for the problem discussed in this chapter and those in other studies, only one firm is
modeled that is interested in the market value. In practice, a supply chain can consist
of public downstream and upstream firms that may all care about their market value
and, at the same time, possess private information. It is interesting to investigate how
the operations of the firms and their contracting decisions might be affected by the
interaction with the capital market. Second, the existing studies generally adopt a
one-period setting. In practice, the interaction between firms and the capital market
may arise repeatedly. It is thus useful to extend the current studies to multi-period
settings. Third, the current studies often assume a single source of asymmetric in-
formation (e.g., only the demand is unknown to the investors), which simplifies the
analysis but may also lose certain insights. It is helpful to explore settings where
multiple streams of asymmetric information may arise. Finally, prior studies have
focused on asymmetric information about the firm’s demand prospect and sales re-
alization. There are many other factors that are important for a firm’s performance
such as the innovation capability, product development processes and quality man-
agement. Asymmetric information may also arise about these factors to influence
a firm’s valuation and operational decisions. Hence, it is interesting to expand the
research scope in this direction.

References

Bebchuk LA, Stole L (1993) Do short-term managerial objectives lead to under- or over-investment
in long-term projects? J Financ 48(2):719–729
Cachon G, Lariviere M (2001) Contracting to assure supply: how to share demand forecasts in a
supply chain. Manag Sci 47(5):629–646
Chen H, Frank MZ, Wu OQ (2005) What actually happened to the inventories of American com-
panies between 1981 and 2000? Manag Sci 51(7):1015–1031
Chen H, Frank MZ, Wu OQ (2007) U.S. retail and wholesale inventory performance from 1981 to
2004. Manuf Syst Oper Manag 9(4):430–456
Cho I-K, Kreps DM (1987) Signaling games and stable equilibria. Q J Econ 102(2):179–221
Dybvig P, Zender J (1991) Capital structure and dividend irrelevance with asymmetric information.
Rev Financ Stud 4(1):201–219
Dye R, Sridhar S (2004) Reliability-relevance trade-offs and the efficiency of aggregation.
J Account Res 42:51–87
Fisher M, Raman A (2010) The new science of retailing: how analytics are transforming the supply
chain and improving performance. Harvard Business School Press, Boston
Garvey G, Grant S, King S (1999) Myopic corporate behaviour with optimal management incen-
tives. J Ind Econ 51(2):181–194
11 Supply Chain Information Signaling and Capital Market 233

Gaur V, Fisher M, Raman A (2005) An econometric analysis of inventory turnover performance in

retail services. Manag Sci 51(2):181–194
Graham JR, Harvey CR, Rajgopal S (2005) The economic implications of corporate financial re-
porting. J Account Econ 40(1–3):3–73
Ha A (2001) Supplier-buyer contracting: asymmetric cost information and the cut-off level policy
for buyer participation. Naval Res Logist 48:41–64
Hendricks KB, Singhal VR (2009) Demand-supply mismatch and stock market reac-
tion: evidence from excess inventory announcements. Manuf Syst Oper Manag 11(3):
509–524
Lai G, Debo L, Nan L (2011) Channel stuffing with short-term interest in market value. Manag Sci
57(2):332–346
Lai G, Xiao W, Yang J (2012) Supply chain performance under market valuation: an operational
approach to restore efficiency. Manag Sci 57(2):332–346
Lai G, Xiao W (2015) Inventory decisions and signals of demand management capability to
investors. Working paper
Liang PJ, Wen X (2007) Accounting measurement basis, market mispricing, and firm investment
efficiency. J Account Res 45(1):155–197
Mailath GJ, Okuno-Fujiwara M, Postlewaite A (1993) Belief-based refinements in signalling
games. J Econ Theory 60(2):241–276
Miller M, Rock K (1985) Dividend policy under asymmetric information. J Finance 40(4):
1031–1051
Monga V (2012) Unraveling inventory’s riddle. Wall Street J. http://online.wsj.com/news/articles/
SB20001424052702304587704577333731187746706. Accessed Apr 2012
Özer Ö, Wei W (2006) Strategic commitments for an optimal capacity decision under asymmetric
forecast information. Manag Sci 52(8):1238–1257
Pasternack BA (1985) Optimal pricing and return policies for perishable commodities. Mark Sci
4(2):166–176
Raman A, Gaur V, Kesavan S (2005) David Berman. HBS Case 605-081. Harvard Business School,
Boston
Ross S (1977) The determination of financial structure: the incentive-signalling approach. Bell J
Econ 8(1):23–40
Roychowdhury S (2006) Earnings management through real activities manipulation. J Account
Econ 42(3):335–370
Schmidt W, Gaur V, Lai R, Raman A (2015) Signaling to partially informed investors in the
newsvendor model. Prod Oper Manag 24(3):383–401
Stein J (1988) Takeover threats and managerial myopia. J Polit Econ 96(1):61–80
Stein J (1989) Efficient capital markets, inefficient firms: a model of myopic corporate behavior.
Q J Econ 104(4):655–670
Taylor TA, Xiao W (2009) Incentives for retailer forecasting: rebates vs. returns. Manag Sci
55(10):1654–1669
The Associated Press (2014) Nike shares fall despite 2Q beat. Yahoo Finance. http://finance.yahoo.
com/news/nike-shares-fall-despite-2q-191214115.html. Accessed Apr 2015
Trefis Team (2013) Abercrombie & Fitch misses expectations and lowers guidance. TREFIS.
http://www.trefis.com/stock/anf/articles/188715/abercrombie-fitch-misses-expectations-and-
lowers-guidance/2013-05-28. Accessed May 2013
Wahba P (2009) Zale cancelling orders, delaying payment. Reuters (Dec 18).
http://www.reuters.com/article/zale-idUSN1822010220091218. Accessed Apr 2015
Zhang, F (2010) Procurement mechanism design in a two-echelon inventory system with price-
sensitive demand. M&SOM, 12(4):608–626
Zhang H, Nagarajan M, Sošić G (2010) Dynamic supplier contracts under asymmetric inventory
information. Oper Res 58(5):1380–1397
Chapter 12
Buying from the Babbling Retailer? The Impact
of Availability Information on Customer
Behavior

Gad Allon and Achal Bassamboo

Abstract Provision of real-time information by a firm to its customers has become

prevalent in recent years in both the service and retail sectors. In this chapter, we
study a retail operations model where customers are strategic in both their actions
and in the way they interpret information, while the retailer is strategic in the way
it provides information. This chapter focuses on the ability (or the lack thereof)
to communicate unverifiable information and influence customers’ actions. We de-
velop a game-theoretic framework to study this type of communication and discuss
the equilibrium language emerging between the retailer and its customers. We show
that for a single-retailer and homogeneous customer population setting, the equilib-
rium language that emerges carries no information. In this sense, a single-retailer
providing information on its own cannot create any credibility with the customers.
We study how the results are impacted due to the heterogeneity of the customers.
We provide conditions under which the firm may be able to influence the cus-
tomer behavior. In particular, we show that the customers’ willingness-to-pay and
willingness-to-wait cannot be ranked in an opposite manner. However, even when
the firm can influence each customer class separately, the effective demand is not
impacted.

G. Allon () • A. Bassamboo

Kellogg School of Management, 2001 Sheridan Road, Evanston, IL 60208, USA
e-mail: [email protected]; [email protected]

This chapter is based on Allon and Bassamboo (2011).

© Springer International Publishing Switzerland 2017 235

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 12
236 G. Allon and A. Bassamboo

12.1 Introduction

Provision of real-time information by a firm to its customers has become prevalent

in recent years in both the service and retail sectors. Service providers use delay
announcements to inform customers about anticipated service delays, whereas re-
tailers provide the customers with information about the inventory level and the
likelihood of being out-of-stock. Often, this information cannot be credibly verified
by the customer. The question of how the information the firm shares with its cus-
tomers influences their buying behavior is a complex one, and its answer depends
both on the dynamics of the underlying operations and the customer behavior.
For example, the web-retailer sierratradingpost.com uses the tag “almost gone!”
for some of the products, and in its Frequently Asked Questions section explains
this tag as follows:
If an “almost gone!” label appears next to the item, the sell out risk is very high. We recom-
mend that you place your order immediately.

Several other web based retailers, such as BarnesandNoble.com and Circuit-

city.com, allowed customers to search for the availability of specific products for
in-store pick-up. Along the same lines, web-based travel agencies such as Expe-
dia.com allow customers to view the availability of airline tickets on specific flights,
prior to making the purchasing decision. Dickies.com (an apparel retailer for men)
and blackoutbmx.com (an online bike store) use the term “low inventory” for cer-
tain items, where they claim to carry low inventory.1 Similarly, brick-and-mortar
stores use different displays mode to inform customers about availability, which
range from showing ample stock per item to showing only a single available item
per product. In all of these examples, the information shared cannot be fully verified
by the customers. A customer in a brick and mortar does not know if there is more
than a single item available even if only one is displayed and cannot verify whether
the stock is indeed low, even if a tag “almost gone!” or a sign “limited availability”
is attached to an item.
In this chapter, we address these issues by proposing a model in which cus-
tomers treat information provided by the retailer as unverifiable and non-binding.
The model thus treats customers as strategic both in the way they make decisions
and process information, and the firm as strategic in the way it provides this infor-
mation. Prior work in Operations Management analyzes systems assuming that the
firm’s information sharing strategy is a-priori fixed and given. These typically lead
to contrasting systems where the firm provides either full information or no infor-
mation. Usually these papers assume that customers treat information as credible
and verifiable, and implicitly assume that the firm restricts itself to truth-telling. The
main issues with these assumptions are: (a) customers may not blindly trust the in-
formation provided to them by firms, and (b) in practice, the information provided
rarely translates directly into the inventory level. In this sense, the information may

1 The authors of this chapter refrain from claiming that these announcements are indeed made only

when the inventory is low.

12 Availability Information 237

not be quantifiable. For example, the tag “almost gone!” does not reveal the accurate
stock level, but it still carries some information. These issues are also interrelated
and stem from the assumption that customers do not further process the information
provided by the firm. In other words, the customers are not assumed to be strategic
with respect to the information.
We develop a model in which these issues are addressed by considering a re-
tailer which can provide various kinds of announcements (without restricting it to
providing full-information, no-information or making quantifiable announcements)
and customers who are able to process the announcements and make strategic de-
cisions. Specifically, we study a game played between the firm and its customers.
In our model, the firm starts with a fixed inventory which it tries to sell at a fixed
price during the regular season. Once the regular season is over, the price of the
product drops. This marks the commencement of the sales season. The firm needs
to decide what information to reveal to the customers during the regular season. The
customers receive this information and decide whether to purchase the product im-
mediately or wait for the sales season. The advantage of waiting is that the price of
the product will be lower, however there are two sources of disutility from buying
in the sales period: (a) the customer needs to wait for the product and thus incurs
a waiting cost; and (b) there is a chance that he might not be able to purchase the
product due to its limited availability during the sales period.
In characterizing the emerging equilibrium language for this game, we account
both for the strategic nature of the interested parties—the customers and the firm—
as well as the dynamics prevalent in retail operations. We shall begin by showing
that an influential language (in which the firm can influence customers using the
information that they provide) is not possible between a single retailer and its cus-
tomers when they are homogeneous in terms of their valuation of the product and the
cost of waiting. This result demonstrates that a single monopolistic retailer cannot
credibly communicate unverifiable availability information to its customers using
cheap talk. When the customers are homogeneous, one can show that while the
price may contain information regarding inventory availability, it cannot improve
the credibility of the cheap talk announcements.
We also study a setting in which the firm faces customers of multiple classes with
different valuations of the product and time. Even when the customers are heteroge-
neous, the firm cannot impact the effective demand using availability information.
That is, the overall demand realized for the product in both the regular and sales
season is independent of the availability information provided by the firm. In this
manner, the above result is robust. We also show that even though the firm cannot
influence the effective demand, it may be able to influence its composition. That is,
the firm may be able to influence each customer segment differently using avail-
ability information. In particular we show that the firm cannot influence customer
behavior unless customers with higher valuation of the product have a lower cost
of waiting. We then show that the crucial variable that customers infer from every
message is the extent by which the availability drops between the regular season
and the sales season. For the firm to be able to influence its customers, it must be
able to associate different availability drops with each message. Further, we show
238 G. Allon and A. Bassamboo

that the firm must signal this drop in availability to be above and below a specific
threshold to induce the influential behavior. In all other cases, the firm is not capable
of influencing customer behavior using announcements of the availability risk.
The model formulated in this chapter treats the information disclosure as “cheap
talk,” i.e., a pre-play communication that carries no cost. Cheap talk, as described
in the literature, consists of costless, non-binding, non-verifiable messages that may
affect the customers’ beliefs. It is important to note that while providing the in-
formation does not directly affect the payoffs, it does have an indirect implication
through the customers’ reaction and the equilibrium outcomes. However, the infor-
mation on its own has no impact on the payoffs of the different players per se, i.e.,
the payoffs of both sides depend only on the actions taken by the customer and in-
ventory dynamics. The firm can neither reward nor penalize a customer based on
whether or not he follows the firm’s recommendation. In Sect. 12.1.2 we discuss in
detail the different modeling assumptions, as well as the differences in the results
between our model and the classical cheap talk literature.
The key contributions of the chapter can be summarized as follows:
1. We develop a model that studies the strategic nature of the information trans-
mission in retail operations, where unverifiable and non-committal information
is provided by a self-interested retailer to selfish customers.
2. The analysis of this model provides what appears to be the first theoretical result
which shows that in any equilibrium that emerges in the single-retailer game with
homogeneous customers, the availability announcements are non-influential. In
other words, the firm can obtain no credibility regarding the information provided
about the inventory-on-hand and thus cannot influence a rational customer in
terms of his purchasing decision. (See Proposition 2.)
3. We show that when customers have diverse valuations of both their willingness-
to-wait and willingness-to-pay, the firm may be able to influence customer behav-
ior, albeit in a limited manner. We first show that, in all equilibria, the effective
demand (across all customer classes) is independent of the initial inventory and
the messages of the firm. We then show that the firm may be able to influence
each customer class individually. This can only happen if customers with higher
valuation for the product have a higher willingness-to-wait. We also characterize
the relationship between (a) the availability-drop associated with messages that
influence the customers and (b) the customer parameters.
4. We also prove that as the size of the market increases, the necessary conditions
for an equilibrium to have influential cheap talk become more stringent in regards
to the feasible parameters for the customer classes. In this sense, we expect that
it becomes increasingly more difficult for firms to influence customers as they
grow, assuming their demand and capacity grow at the same rate. We also discuss
settings where the demand and quantity grow disproportionately.
Outline of the Chapter We conclude this section with a review of the relevant lit-
erature. Section 12.2 describes the model for a single retailer. Section 12.3 analyzes
the game played among the customers when the firm provides no information. Sec-
tion 12.4 studies the cheap talk game played between the retailer and the customers.
12 Availability Information 239

Section 12.5 studies the impact of heterogeneity among the consumers. Section 12.6
provides discussion and conclusions. Proofs of results stated in Sects. 12.3 and 12.4
are in the main body of the chapter, while the proofs of the remaining results are
relegated to Appendix.

12.1.1 Literature Review

Recent literature in Operations Management analyzes and models the impact of

strategic customers. This literature can be broadly divided into two categories:
(a) models where no availability information is provided to the customer and
(b) models where customers are provided with complete information regarding
availability.
Our single retailer model is related to the model introduced in Aviv and Pazgal
(2008), which falls in the first category. The authors study pricing strategies for a
retailer facing a stochastic arrival stream of customers. When customers arrive, they
have no information about the current state of the inventory. A related model is
studied in Dana (2001), where the firm signals availability using prices. Cachon and
Swinney (2009) considers a model of a retailer that sells a product with uncertain
demand over a finite selling season. The authors characterize the Rational Expecta-
tion Equilibrium (REE) between the firm, who sets its initial quantity level, and the
strategic customers, who choose whether to buy during the selling season or during
the clearance season. The authors also study the impact of quick response and the
interplay between the existence of strategic customers and this option. Su and Zhang
(2008) shows that the presence of strategic customers can impact the performance
of a centralized supply chain when the customers form rational expectations regard-
ing quantities and prices. They show that, while firms cannot commit to specific
levels of inventory, decentralized supply chains can use contractual arrangements as
indirect commitment devices to attain the desired outcomes with commitment. Liu
and van Ryzin (2008) investigates whether it is optimal for a firm to create rationing
risk by deliberately understocking products. The authors develop a model that de-
termines the optimal amount of rationing risk to maximize their profits. In a related
paper, Debo and van Ryzin (2009) studies how customer infer the quality of the
product from stock-outs and availability, see Veeraraghavan and Debo (2009) for a
service setting.
Yin et al. (2009), and Su and Zhang (2009) belong to the second category. Yin
et al. (2009) considers a retailer that announces the regular price and the sales-season
clearance price (or a contingent price) at the beginning of the selling season, as in
our model. In the presence of either myopic customers or strategic customers, the
authors compare two display modes: one where the retailer displays all the avail-
able units (and corresponds to providing full information to the customers) and one
where it shows only one unit. Customers treat this one unit as verifiable proof that
the firm has at least one unit in stock. The authors show that the retailers will earn
higher expected profits under the “display one unit” format when the customers are
240 G. Allon and A. Bassamboo

strategic. Su and Zhang (2009) studies the role of availability and its impact on
consumer demand by analyzing a newsvendor model with strategic customers who
incur some search costs in order to visit the retailer. They contrast the REE in a game
where the availability information is not provided to the customer with the scenario
where such information is provided. It is shown that the retailer can improve its
profits in the latter. In order to deal with the lack of credibility of the above informa-
tion, the authors study availability guarantees, in which the seller compensates the
consumers in the event of stock-outs.
Our chapter contributes to the above papers by proving that, indeed, the firm
cannot influence its customers using availability information when customers are
homogeneous, thus showing that indeed “display one” (which is equivalent to a
babbling equilibrium in our setting) is the only viable option for a firm that would
like to announce information. Our chapter also contributes to the above by showing
that a firm may be able to influence customers using more refined information when
customers are heterogenous.
A closely related paper in terms of the underlying framework is Allon, Bassam-
boo, and Gurvich (2011), which appears to be the first paper in the operations man-
agement literature to consider a model in which a firm communicates unverifiable
information to its customers. Both papers focus on analyzing the problem of infor-
mation communication in an operational setting by considering a model in which
the firm and the customers act strategically: the firm in choosing its announcements,
and the customers in interpreting this information and in making the decision. The
settings considered, however, are very different, and the results are driven by the
characterizing features of service systems and inventory systems. The chapter con-
cludes with Sect. 12.7.

12.1.2 Classical Cheap Talk Game

In this section, we provide an overview of the cheap talk game introduced in Craw-
ford and Sobel (1982) and compare the model to the one studied in this chapter. The
classical cheap talk game is played between a sender who has some private infor-
mation and a receiver who takes the payoffs-relevant actions. The game proceeds
as follows: The Sender observes the state of the world, which we shall denote by Q.
The Sender then sends a signal (or a message) denoted by m ∈ M . (Here M de-
notes the set of all signals that can be used by the Sender.) The Receiver, who cannot
observe the state of the world Q, but does know its distribution, processes the signal
(using Bayes rule) and chooses an action y that determines the players payoff. Both
the Sender and the Receiver obtain utilities which depend on: (a) the action taken by
the Receiver y; and (b) the state of the world Q. A distinctive feature of their model
is that the distribution of the state of the world is exogenous and independent of the
actions of the players.2
2 A variety of chapters study mixed-motive economic interaction involving private information

and the impact of cheap talk on the outcomes. Farrell and Gibbons (1989) studies cheap talk in
12 Availability Information 241

Driven by the applications in operations management, our model has two novel
features: first, the game is played with multiple receivers (customers) whose actions
have externalities on other receivers; and second, the stochasticity of the state-of-
the-world (i.e., the state of the system) is not exogenously given but is determined
endogenously. In particular, the private information in this model (i.e., the availabil-
ity of inventory both in the regular and the sales season) is driven by the equilibrium
strategies of both the firm and the customers. In particular, the customers’ actions
are payoff-relevant as well as system-dynamic-relevant. As we shall see, the mul-
tiplicity of receivers with externalities as well as the endogenization of the uncer-
tainty impact both the nature of the communication, when one exists, as well as the
outcome for the various players. Hence, while the framework used in this chapter
echoes the cheap-talk model described in the literature, the above mentioned distin-
guishing features lead to different results.3

12.2 Model

We study a two period model in which a firm aims to maximize its revenue. We
will refer to the first period as the regular period and the second period as the sales
period. In our model, the firm starts with an initial inventory Q0 . The customers do
not know the initial inventory, however, they have beliefs regarding the actual inven-
tory. The customer believes that the initial inventory has a cumulative distribution
function FQ0 . Further we assume that F(0) = 0.
During the regular season, the potential customer demand is realized. The poten-
tial number of customers is Poisson with mean λ ; we denote the realized potential
demand by D1 . Each customer obtains a value v from the product. The potential
demand captures the number of customers who are interested in buying the product
but who will time their purchase to maximize their utility. The firm provides an an-
nouncement that signals the inventory level at the beginning of the regular period.
The price of the product during the regular is set to p.4 The customers (who form the
potential demand) decide whether to buy5 in the regular period or wait for the next
period. The customers who decide to buy during the regular season form the effec-
tive demand for the regular period. The firm allocates/satisfies (as much as possible)

bargaining; in political context cheap talk has been studied in multiple papers including Austen-
Smith (1990), and Matthews (1989).
3 One should note that the results in Crawford and Sobel and most of the cheap talk literature

are stated, based on the bias between the sender’s and the receiver’s preferred actions, which are
exogenously given. In our model, the extent of the misalignment depends endogenously on the
preferred action of the customers as it arises in equilibrium in the game they play. Thus, even the
most basic results cannot be directly borrowed from this literature.
4 We also provide a discussion of the setting where the pricing is a decision of the firm and contin-

gent on the initial inventory in Sect. 12.7.

5 Note that when we say that a customer decides to buy, we merely mean that the customer attempts

to buy, but they might not be able to purchase due to limited availability.
242 G. Allon and A. Bassamboo

the effective demand. We assume that, if the effective demand is higher than the
quantity, then the product would be rationed uniformly among the customers who
form the effective demand.
After the regular season, the firm is left with Qs units of inventory. Note that the
actual quantity-on-hand at the beginning of the sales period Qs is determined by
both the potential demand as well as the customers’ buying decisions. The latter de-
pends on the information they have, including among other things, the information
provided by the firm.
The price in the sales season may be contingent on the quantity left and is de-
noted by s(Qs ). During the sales season, new customers, which we refer to as
bargain hunters, are also interested in buying the product. We will denote by D2
the number of such customers, where the distribution of D2 is denoted by FD2 .6
Thus, the effective demand during the sales season is formed by both the cus-
tomers who decided to wait for the sales season and the bargain hunters. The cus-
tomers who arrived during the regular season but decided to wait for the sales
season will incur a cost of c for waiting. As before, the firm satisfies as much
of this overall demand in the sales period as possible. Thus, the firm’s revenue is
p(Q0 − Qs ) + S(Qs ) min(Qs , D1 + D2 − Q0 + Qs ).
Each customer arriving during the regular season faces a decision whether to
buy immediately or wait for the sales season. The main tradeoff customers face is
whether to buy now at a given (high) price with relative high availability or wait and
buy, facing much greater availability risk. This is driven by the customers’ param-
eters (v, c). If he decides to buy during the regular season, then he obtains a value
(v − p)A(Q0 , DE ) where DE is the effective demand and A(x, y) is the availability
function. We assume that, if the demand is higher than the quantity-on-hand, then
the likelihood of obtaining the product is identical among the customers who decide
to buy, i.e., if the demand for the product is x and the quantity-on-hand is y then
the likelihood is A(x, y) = min{x/y, 1}. Similarly, if he decides to wait for the sales
season, then his value is given by (v − s)A(Qs , D2 + D1 − DE ) − c, where c is the
cost of waiting between the regular and the sales season. Here c is associated with
the inconvenience of not obtaining the product immediately (we thus refer to 1/c as
the willingness-to-wait, since it is the amount of time the customer is willing to wait
for a dollar). Note that the quantity-on-hand in the sales period is given by Qs , and
the overall demand during the sales period is given by D1 + D2 − DE . We shall refer
to A(Qs , D2 + D1 − DE ) as the availability of the product during the sales season.
We can immediately make the following observation about the customer strate-
gies: The customer has the option to leave the market, and obtain zero utility, but it
can be easily seen since v > p that the option of leaving the market is strictly dom-
inated by the buying option during the regular season since Q > 0 with probability
one. Thus, the customer decision can be reduced to whether he buys immediately or
he waits for the sales season.

6 One can view a more detailed description of the bargain hunter. For instance, D2 can emerge as
an aggregate number of arrivals during the regular season of customers whose valuation is below
the regular price p.
12 Availability Information 243

12.3 Providing No Information

The main focus of this chapter is characterizing the ability or inability of a retailer
to communicate unverifiable information to strategic customers. In order to be able
to discuss the specific model of communication, we will initially discuss the cus-
tomers behavior when the firm provides no information about inventory availability.
In Sect. 12.4, we discuss whether this strategy of not providing information emerges
in equilibrium.
In this setting, we assume that the firm is not providing any information with
regards to the inventory position. Since the customers cannot observe the state of the
system, they have to rely on the their belief about the inventory level. Further, since
all agents are a-priori identical, we will be focusing on symmetric strategies for the
customers. We will represent their strategy by y ∈ [0, 1], which is the probability that
a customer tries to buy the product in the regular season. We next define the notion
of Bayesian Nash equilibrium (BNE) under-no-information (see Chaps. 6 and 13
of Fudenberg and Tirole (1991) for a definition of Bayesian Nash Equilibrium).

Definition 1. We say that the pair y∗ ∈ [0, 1] forms a Bayesian Equilibrium (BNE)
under no-information in the retail cheap talk game if and only if it satisfies the
following condition:

y ∈ arg max yE[(v − p)A(Q0 , Zy + 1) − (v − S(Q0 − Zy ))

y∈[0,1]
× A((Q0 − Zy )+ , D2 + D1 − Zy ) + c | D1 ≥ 1], (12.1)

where Zy is a Binomial random variable with (D1 −1)+ trials each with probability y
of success.

The above definition requires that the customers do not have any unilateral prof-
itable deviation from the strategy profile which defines the equilibrium. Specifically,
the condition requires that when fixing the strategy of the rest of the customers and
assuming the firm provides no information, a customer should not have any prof-
itable deviation. In the condition the random variable Zy is the effective demand
(excluding the customer who is making the decision). Since all customers random-
ize with probability y and the demand (excluding the deciding customer) is given
by (D1 − 1)+ , we obtain that the effective demand is the number of successes in
(D1 − 1)+ binomial trials, each with a success probability y. Thus, the objective in
(12.1) is the difference between the utility if the customer decides to buy during the
regular season and the utility if the customer decides to buy during the sales season.
In order to study whether such an equilibrium always exists let U R (y) and U S (y)
be the utility of a customer when he decides to buy during the regular season and
sales season, respectively, when other customers are buying with probability y dur-
ing the regular season. It is easy to verify the these functions are continuous decreas-
ing functions of y. Thus, we obtain that the following two cases.
244 G. Allon and A. Bassamboo

Case I If U R (0) > U S (0) and U R (1) < U S (1), then there exists U R (y) = U S (y). Let
us denote the solution to U R (y) = U S (y) by y∗ . Then y = y∗ forms a BNE. (In this
case, there is a mixed strategy BNE.)
Case II If Case I is not satisfied, it must be the case that either U R (0) < U S (0) or
U R (1) > U S (1) are true. Thus, either y = 0 or y = 1 (or both) form an equilibrium,
respectively. In this case, there is a pure strategy BNE. This completes the proof.
These are summarized by the following proposition, whose proof follows directly
from the above discussion.
Proposition 1. There exists a BNE for the game under no information.
The above theorem shows that there exists an equilibrium among the customers
when the firm does not provide any information.7 One can view this equilibrium as
self-organization of the customers among themselves in the absence of any infor-
mation.

12.4 Cheap Talk Equilibrium

In this section, we explore the game played between the firm and its customers,
where the firm is allowed to use any information provision strategy. To define the
single-retailer game formally, we shall start by defining the strategy of the customer
followed by the strategy of the firm.
Let M be the Borel set which is comprised of feasible signals that the firm can
use, and let Ω denote its σ -algebra. Let y : M → [0, 1] represent the strategy of a
customer. Here, y(m) is the probability that a customer arriving during the regular
season and receiving a signal m ∈ M , buys the product during the regular season.
Thus, this customer waits for the sales period with probability 1 − y(m). Let the
space of feasible strategies for the customers be denoted by Y .
Next, we describe the strategy of the firm. In doing so, we allow the firm to
randomize over the set of messages in the set M , i.e., given a specific quantity
on hand q, the firm may randomize among different messages. To capture this, let
ν : Z × Ω → [0, 1] represent the strategy of the firm. Here we require that ν (q, ·)
induces a probability measure on M from which the firm announces a realization,
if the quantity-on-hand is q. Thus, if the quantity-on-hand is q at the beginning of
the regular period, the probability that the firm
 signals a message from a measurable
Borel-subset S ⊆ M is given by ν (q, S) = m∈S d ν (q, m) (For example, if the firm
announces messages m1 and m2 with probability half when the quantity on hand is 5,
then the measure ν (5, ·) is defined as follows: ν (5, {m1 }) = ν (5, {m2 }) = 0.5, and
ν (5, S) = 0 for all S ⊂ Ω and S ∩ {m1 , m2 } = 0.) Let the space of feasible strategies
for the firm be denoted by G . Note that the quantity-on-hand at the beginning of
the sales period Qs is determined by the customer’s strategy as well as the firm’s
strategy ν . Let μν ,y represent the distribution of the signal transmitted if the firm
7 It is worth noting that the equilibrium may not be unique.
12 Availability Information 245

follows strategy ν and the customers follow strategy y. A random variable with
measure μ shall be represented by Xμ . Further, let the firm’s profit under the strategy
pair (ν , y) be written as Π (ν , y). Also, let Qν ,y,s be the inventory-on-hand at the
beginning of the sales period under the strategy pair (ν , y).

Definition 2. We say that the pair (ν ∗ , y∗ ) ∈ G × Y forms a Bayesian Equilibrium

(BNE) in the retail cheap talk game if and only if it satisfies the following two
conditions:
1. For all m ∈ M ,

y∗i (m) ∈ arg max yE (v − p)A(Q0 , Zy + 1) − (v − S(Q0 − Zy ))
y∈[0,1]

× A((Q0 − Zy )+ , D2 + D1 − Zy ) + c | D1 ≥ 1, Xμν ∗ , y = m ,

where Zy is a Binomial random variable with (D1 − 1)+ trials each with proba-
bility y of success.
2. Fixing y∗ , ν ∗ solves:
ν ∗ ∈ arg max Π (ν, y∗ ).
ν∈G

The above definition requires that both the firm and the customers do not have
any unilateral profitable deviation from the strategy profile which defines the equi-
librium. Specifically, the first condition in the definition requires that when fixing
the strategy of the rest of the customers and the firm, a customer should not have
any profitable deviation. Thus, given that all other customers interpret the messages
and purchase the product as prescribed by the equilibrium, and thus given the avail-
ability that is driven by such behavior, a customer has no incentive to deviate from
this prescription. Similarly, the second condition requires that given the customer’s
action rule y∗ as fixed, the firm maximizes its profit by using strategy ν ∗ . That is,
given that the firm knows how customers interpret its messages, the firm should not
have any incentives to use this messages on a different realization of capacity than
the one prescribed by the equilibrium.8

12.4.1 Main Result

Next, we characterize the emerging equilibria in the homogeneous customers cheap

talk game. We prove that it is impossible for the firm to influence the customer
behavior using availability information if all customers are homogeneous in their
valuation of the product and their cost of waiting and, in that sense, for the firm to
credibly communicate any meaningful information to its customers. In particular,
we show that any equilibrium that can emerge in the game played between the firm

8 We assume that, off the-equilibrium-path, a message that was not supposed to be used will result

in a “wait” decision by the customer.

246 G. Allon and A. Bassamboo

and its customers is such that the customers are not influenced by what the firm
announces. We shall refer to such equilibrium as non-influential (formal definition
is defined below). Such equilibria can manifest themselves in several ways: the firm
may either provide no information (in which case, naturally, the customers base their
decisions only on their expectations of the inventory levels), or provide any type of
information which is uncorrelated with the state of the system, yet the customers
disregard it and, again, base their decisions on their expectations of the inventory
levels. The key feature of a non-influential equilibrium is that the action taken by
the customers in equilibrium is independent of what the firm announces, due to the
lack of credibility of such an announcement. Note that the definition focuses on
the ability to induce different actions and these actions directly affect the profit of
the firm and the utility of the customers. We shall next formally define the class of
non-influential equilibria.

Definition 3. We say that the pair (y, ν ) ∈ Y × G forms a non-influential equilib-

rium if there exists a constant p such that for all q ∈ {0, 1, 2, . . . , Q(0)}, we have that
ν (q, {m : y(m) = p}) = 1, i.e., on the equilibrium path, irrespective of the firm’s
announcement, the probability that a customer buys during the regular season is p.

While in the previous section we showed that a non-influential equilibrium al-

ways exists in the sense that an equilibrium exists in the game with no informa-
tion, we have not yet excluded other more influential types of equilibria. We next
show that all equilibria that can arise in the single retailer cheap talk game are non-
influential.

Proposition 2 (The Non-influential Result). Under any BNE of the single retailer
cheap talk game, the customer’s realized buying behavior satisfies the following

y(Xμν ,y ) = y∗ a.s.,

where there exists an equilibrium with non-influential cheap talk under which the
customers purchase with probability y∗ during the regular season. Thus, any BNE
is non-influential.

Proof. The proof is based on the following observation: since the firm prefers cus-
tomers to buy in the regular season, it will provide a message that maximizes the
probability of buying. Thus, under any strategy profile in which customers react
differently to different messages, the firm will always have an incentive to deviate
and use the one that maximizes the buying probability. Hence, in equilibrium, the
customers behave as if the buying probability is fixed and is not impacted by the
message provided by the firm.
We next make this rigorous. Consider any BNE of the above cheap talk game,
(y, ν ) ∈ Y × G . When the firm signals m, the effective demand observed during the
regular season is Poisson with mean λ y(m). Thus the profit of the firm if it provides
message m is given by
12 Availability Information 247
  D1   D1 + 
E p min Q0 , ∑ I{ω L ≤y(m)} + S Q0 − ∑ I{ω L ≤y(m)}
i i
i=1 i=1
 D1 + D1 
× min Q0 − ∑ I{ω L ≤y(m)} , ∑ I{ω L >y(m)} + D2 .
i i
i=1 i=1

Let M ∗ = arg maxm y(m). Thus, it is easy to see that under any equilibrium the firm
must use signals from the set M ∗ only, i.e., any strategy where the firm signals a
message that is not in M ∗ cannot form an equilibrium. Based on this, we obtain
that the customer can only buy with probability y∗ = maxm y(m) during the regular
season in equilibrium and thus any equilibrium is non influential. This completes
the proof.

The above proposition shows that, in equilibrium, no matter what signaling rule
the firm uses, the customers would simply ignore all the signals and make their
buying decisions irrespective of any information provided. Thus, in this cheap talk
game no credibility whatsoever can be created. We next study how this result ex-
tends when the customers are heterogenous.

12.5 Multiple Customer Classes

While until now we have assumed that all the customers (those who actively shop
in the regular season) were homogeneous, we next consider a model where they
are heterogenous both in terms of the value they derive from the product as well as
the cost of waiting they incur when they decide to postpone purchase to the sales
season. Specifically, we assume that there are 2 customer classes; we shall denote
these classes by H and L and class specific attributes by subscript i ∈ {H, L}. Thus,
the value of the product is vi and the waiting cost per unit of time is ci for a class i
customer. We assume that the potential demand rate of each class is λi .
The key result of the previous section is that the firm cannot influence the cus-
tomer’s purchasing behavior by providing availability information. We next answer
the question of whether the heterogeneity of customers’ preferences allows the firm
to influence customer behavior or not. To answer this question, consider two set-
tings: (a) one in which the customer’s class is observable to the firm and the firm
can tailor the information provided to each class separately; and (b) one where the
classes are unobservable, and thus the firm cannot tailor its message. The former
may be applied by online retailers or catalog-based retailers, while the latter is more
applicable for brick-and-mortar retailers. For the first setting, it is easy to show that
our non-influential result from the previous setting extends. We will thus focus on
the latter case, in which the firm cannot observe the classes and thus need to use a
unified message.
Our description is reminiscent of Farrell and Gibbons (1989) that compares set-
tings in which the sender of information may choose between providing the infor-
mation that is class-dependent (they refer to such a setting as private messaging)
248 G. Allon and A. Bassamboo

to the setting where the information is provided without the knowledge of the cus-
tomer class (they refer to such settings as public messaging). They show that if the
sender can create credibility when providing information that is class-dependent, it
can also do so in the setting where the class is unobserved. However, if the sender
cannot create credibility in the observable case, Farrell and Gibbons (1989) do not
rule out the possibility of establishing credibility in the unobservable case.
We begin by demonstrating that even when customers are heterogenous, the firm
cannot impact the effective demand using messages on the availability of inventory.
To study the setting where the customer class is unobservable, we define the tu-
ple {ν , yL , yH } that represents the strategy of the various customer classes and the
signaling rule for the firm. Here yi (m) is the probability that a customer of class i
attempts to purchase the product in the regular season when the firm signals m. As
before, ν (q, ·) denotes the measure induced on the messages from which the firm
signals during the regular season when the quantity-on-hand is q. We can define the
BNE for the cheap talk game in an analogous manner to Definition 2.
As with the single class setting, our work is centered around the question of
whether an equilibrium is influential or not. To define the non influential equilib-
rium, we need the following two definitions: one focuses on the overall purchasing
probability while the other one is customer class specific.

Definition 4. We say that an equilibrium (ν , yL , yH ) is Aggregate Non-Influential

(AGNI) if there exists a constant θ

λL yL (Xν ,yL ,yH ) + λH yH (Xν ,yL ,yH ) = θ a.s. (12.2)

on the equilibrium path.

Definition 5. We say that an equilibrium (ν , yL , yH ) is Class-Wise Non-Influential

(CWNI) if there exist constants θi such

λi yi (Xν ,yL ,yH ,t) = θi a.s. (12.3)

for i = L, H on the equilibrium path.

Of course, based on the above definition, any equilibrium that is CWNI is also
AGNI. Also, if an equilibrium is not CWNI, we will refer to as Class-Wise Influen-
tial (CWI).
Note that the above definitions imply that, under an AGNI-equilibrium, the prob-
ability that a customer attempts to purchase the product in the regular season is
independent of the initial inventory level since it is independent of any message the
firm provides. Similarly, under CWNI-equilibrium, the probability that a customer
of class i attempts to purchase the product in the regular season is independent of the
inventory for every class i customer where i ∈ {L, H}. Note that if an equilibrium is
CWNI then it must be AGNI, but the converse is not true.
We will start by showing that any equilibrium must be AGNI, regardless of the
customer parameters. (One can show that the result holds even if not restricting
attention to symmetric strategies.)
12 Availability Information 249

Proposition 3. Every equilibrium for the multi-class cheap talk game, (ν , yL , yH ),

is AGNI.

The AGNI property says that the firm cannot impact the aggregate purchasing
probability using availability information. Loosely speaking, this property sets a
limit on the ability to influence customer behavior. Even if an equilibrium that is not
CWNI exists, it is still true that the overall purchasing probability is unaffected by
the messages provided the firm. We thus have the following immediately corollary.

Corollary 1. In the two-class cheap talk game, for a class-wise influential equilib-
rium to exist in pure strategies it must be the case that λL = λH .

The above results says that for a CWI to exists in pure strategies, it must be the
case that the demand rates of both classes are identical. The result stems immedi-
ately from the AGNI results since, for an influential equilibria, it must be the case
that there exist two messages such that one class buys during the regular season
with one and the other class buys during the regular season with the other. since the
overall buying rate with each message is the same, for an equilibrium to exist it must
be the case that both classes have the same demand rate. Otherwise, if restricted to
pure strategies, the firm would try to induce only the class with higher demand to
buy immediately, which would break the equilibrium.
Proposition 3 raises the question, however, whether a firm is capable of influ-
encing customer classes separately, within the limitation of AGNI. We next provide
necessary conditions for the existence of a class-wise influential equilibrium.

Proposition 4. In the two-class cheap talk game, for a class-wise influential equi-
librium to exist it must be the case that

cH − cL ≤ Δ vL , (12.4)

where Δ is a constant that depends only on λL , λH and the distributions of D1 , D2

and Q.

Before providing an intuition for the condition we state an immediate corollary

for similar condition when the market becomes large (i.e. the number of potential
buyers and items grow).

Corollary 2. For the two-class cheap talk game with large market9 , for a class-wise
influential equilibrium to exist it must be the case that

(cH − cL )(vH − vL ) ≤ 0, (12.5)

9 if the demand and the quantity do not grow proportionally, it is easy to see that the results are

quite trivial. For example, assume that p−s > c. If Q grows faster than the demand, then everybody
waits. If the demand grows faster, everybody purchases immediately. In both cases the messages
play no role
250 G. Allon and A. Bassamboo

Thus, the above corollary shows that it is possible for an equilibrium to be CWI
while being AGNI when the customers are heterogeneous. Recall first that the ef-
fective demand cannot be influenced by the retailer. Thus, the only way the firm
can influence each class differently is if the customers have different preferences
for time and value and thus may take different actions given the same information.
Moreover, we cannot influence customers if vH > vL and cH > cL , i.e. it must be the
case that customers with higher willingness-to-pay are those that are also willing to
postpone their purchase. The key factor contributing to the existence of an influen-
tial equilibrium is the fact that, in the regular season, the customers need to tradeoff
the availability of the product and the price drop in the sales period. Furthermore,
the conditions state that for an influential equilibrium to be possible, the different
classes must tradeoff the benefits and the costs of waiting differently.
One may observe that conditions in Proposition 4 are easier to satisfy than those
in Corollary 2. One of the main difficulties in sustaining an influential equilibrium
is the fact that when providing a message on availability, customers of different
classes may have the same beliefs regarding the inventory availability. Since the
effective demand is unaffected by the messages (due to AGNI), for an influential
equilibrium to exist, the same message must elicit different actions from different
classes. It turns out that when customers are non-atomistic, they may have different
beliefs even if provided with the same message. The key intuition is that the cus-
tomers in our model are non-atomistic. Thus the players from one customer class
will have a different belief about the availability even when they are given the same
message. This discrepancy in beliefs helps sustain an influential (class-wise) equi-
librium. However, as the market size grows, their views get closer and coincide.
Thus, it is difficult to influence the customers differently. This suggests that it is
harder to influence large markets than small. In the larger market, it is necessary to
have more markedly different valuations of the product and time in order to influ-
ence customer behavior.
In order to provide a better characterization of influential equilibria, one can
rewrite the problem a customer is facing when deciding whether to purchase the
product in the regular season or wait. In particular, a customer of class i, will pur-
chase the product in the regular season when provided with message m if

R,i − AS,i )v + c ≥ pAR,i − sAS,i ,

(Am m m m
(12.6)

where Am R is the availability during the regular season when the customer receives
the message m, Am S,i is the availability during the regular season when the customer
receives the message m. Note that Am m
R,i and AS,i depend not only on the customer
behavior and the message but also on the customer’s own class. We will denote the
βim := AmR,i − AS,i , i.e. the perceived availability drop in equilibrium associated with
m

message m from the perspective of a class i customer. As we described earlier, for an

equilibrium to be influential the customer must tradeoff the value of the product and
the cost of waiting. To convert this trade-off into an influential action, the firm must
use signals that have certain restrictions on the availability drop. We next provide
this characterization of an influential equilibrium.
12 Availability Information 251

Theorem 1. For an equilibrium to be influential, there must be two messages (that

the firm uses with positive probability) m1 and m2 such that for i = L, H
cL − cH Δ vH
Am1 m1
R,i − AS,i ≤ + (12.7)
vH − vL vH − vL
cL − cH Δ vH
− ≤ Am m2
R,i − AS,i .
2
(12.8)
vH − vL vH − vL
Before we provide an intuition we state the result for large markets.

Corollary 3. For the two-class cheap talk game with a large market, for an equilib-
rium to be influential, there must be two messages (that the firm uses with positive
probability) m1 and m2 such that for i = L, H

|cL − cH |
Am m1
R,i − AS,i ≤
1
≤ Am m2
R,i − AS,i .
2
(12.9)
|vL − vH |

Recall that for an equilibrium to be influential, there must be at least two mes-
sages that result in two different actions. In equilibrium, each of these messages
will be associated with availability in the regular and the sales season. The above
theorem shows that for an equilibrium to be influential, it must be the case that there
exists a threshold such that the availability drop associated with one message has
to be below this threshold and the availability drop of a second message has to be
above this threshold. The threshold depends on the value of the product and cost of
waiting for the two customer classes.
In order to better understand the intuition behind the characterization, we next
provide a graphical representation of the incentive-compatibility conditions of each
class with respect to each message for large markets. We shall represent the cus-
tomer classes based on their two characteristics: value of the product, v and the cost
of waiting c. (In Fig. 12.1 the horizontal axis is the value, v and the vertical axis is
the cost of waiting, c.) Note that the IC condition divides the value-cost space into
two sets via a line (labeled as mi for message mi ) whose slope is the availability
drop of the corresponding message. The set “above” the IC condition denotes the
parameters for the customer classes that will buy during the regular season when
they obtain a message m1 , the other set, i.e., the region below the IC condition de-
notes the parameters for the customer class that will postpone their purchase to the
sales period. Thus, for the CWI equilibrium, it must be the case that the IC con-
straints for two messages align as shown in the figure. Specifically, we need that
the two sets be formed by the intersection of the ‘Buy’ from one message and ‘Not
Buy’ from the other message, i.e., Regions I and II, are non empty. Combining this
with the fact that the slope of IC constraint is the drop in the availability, we obtain
the condition provided in the Theorem. As shown in Corollary 2, for an influential
equilibrium to exist, the two classes’ parameters (vH , cH ) and (vL , cL ) cannot be on
a line with a positive slope.
Example of Class-Wise Influential Equilibrium We next provide an example
where the customers do react to the announcements made by the firm and hence
252 G. Allon and A. Bassamboo

the equilibrium is Class-Wise influential. We assume that Q(0) = 20, 40 and 60 with
probability 1/3, 1/2 and 1/6, respectively. There are two customer classes, and both
have a demand that has Poisson distribution with mean 30. We will next outline the
equilibrium and the implied availability functions. When the firm has quantity 20 or
60, it announces m1 , otherwise the firm announces m2 . Based on the message mi for
i = 1, 2, class i customers purchase in the regular season whereas the other class’
customers wait for the sales period. The price in the regular season is $10 whereas
the sales season price is $5.
We next characterize the incentive compatibility conditions on the value (v1 , v2 )
and the cost of waiting (c1 , c2 ). To this end, note that when the message m1 is pro-
vided, the customers’ beliefs are that the quantity at the beginning of the regular
season is 20 and 60 with probability 2/3 and 1/3, respectively. Thus, following the
equilibrium behavior, we have that the availability during the regular season and
sales season is Am m1
R = 0.7912, and AS = 0.3045, respectively. Similarly, for mes-
1

sage m2 we have AR = 0.9979, and Am

m2
S = 0.3482. These numbers were computed
2

numerically by approximating the Poisson demand by a truncated Poisson random

variable. Thus, we obtain the following four incentive compatibility conditions:

v1 (0.4871) + c1 ≥ 6.3942 (12.10)

v2 (0.4871) + c2 ≤ 6.3942 (12.11)
v1 (0.6497) + c1 ≤ 8.2380 (12.12)
v2 (0.6497) + c2 ≥ 8.2380. (12.13)

One can observe that these reduce to the condition of (v1 , c1 ) and (v2 , c2 ) being on
opposite sides of the following lines two lines:

m1 : v(0.4871) + c = 6.3942 (12.14)

m2 : v(0.6497) + c = 8.2380. (12.15)

It is easy to see the existence of such (v1 , v2 , c1 , c2 ). For instance, v1 = 11, v2 = 13,
c1 = 1.05, and c2 = 0 is one such instance. This is in line with Corollary 2, where
v2 > v1 and c1 < c2 .
It is important to note that the chapter provides necessary conditions for the exis-
tence of CWI equilibrium. When these conditions are violated, any arising equilib-
rium is CWNI. However, even if these conditions are satisfied, it is not guaranteed
that a CWI equilibrium exists. Note that for an equilibrium with CWI cheap talk
to exist in pure strategies, it must be the case that there are two messages corre-
sponding to the two classes such that, when the firm announces these message, only
the customers of the corresponding class attempt to buy during the regular season.
Thus, one can compute the availability drops that correspond to the different real-
izations of initial quantity-on-hand. Using these, one can construct a linear program;
the solution of which provides the signaling rule that the firm can use to influence
customers. Further, if this set of linear equations do not have a solution then there is
no CWI.
12 Availability Information 253

Cost of waiting per unit of time

m2

Region II

m1

Region I

p Value of the good for the customer

Fig. 12.1 The figure depicts the (v, c) parameters. If the class 1 parameters lie in Region I and class
2 parameters lie in Region II then the firm can sustain an CWI equilibrium

12.6 Multiple Channels of Information

While the previous section showed that the only equilibrium that emerges in the sin-
gle retailer game is one that satisfies AGNI condition, we next study a decentralized
setting where the existence of a second information provider enables the retailers
to gain “some” credibility. In fact, under some technical conditions the retailer can
induce the full information behavior.
There are numerous cases in practice where multiple channels sell items from
the same pool of inventory and independently provide availability information (this
inventory may either be physically co-located or virtually pooled). For example,
Dicks.com and Modells.com—whose operations are both run by GSI commerce—
compete over the same potential customers yet provide information on the same
pool of inventory for the same items. Also, many businesses have multiple retail
channels catering to same pool of customers and sharing the same pool of inven-
tory. For example, J.Crew and L.L. Bean use both their websites and catalogs to sell
their products to customers. Usually these channels have some autonomy and are
run as profit maximizers. Demery (2004) explains, “Channels run under different re-
sponsibility centers and profit centers, so a dot-com, a catalog and brick-and-mortar
store were run as separate businesses.” There are two main characterizing features
of these practices: (1) partially shared inventory and (2) decentralized information
providers which are profit maximizers.
Our model incorporates these feature and we shall show that this multiplic-
ity of information sources can actually help the firms to achieve some credibility.
254 G. Allon and A. Bassamboo

In systems in which such practices are not yet implemented or used, allowing cus-
tomers to obtain information through multiple channels can be viewed as a remedy
to the inability to communicate un-verifiable information with only a single retailer.
To study this multiple-retailer setting and to explore how much credibility “de-
centralization” can create in this setting, we shall next define the formal model and
proceed to analyze it. We consider multiple autonomous sales channels of the same
retailer or multiple sellers sharing a common inventory. The customer cannot see or
verify the status of this inventory. In this setting the sellers’ signals are based on the
common inventory and the customers make their buying decisions based on both
signals. We assume that the utility function and profit of the firms are similar to the
previous section with the following modification: each firm receives the profits from
the products it sells. Note that similar analysis can be carried out for more general
systems, where the retailers carry some inventory “on-site” and share the rest.
To describe the game formally, we denote the strategies of the firms by functions
ν1 and ν2 to represent the signalling rule for the two sellers and y : M × M →
S ⊂ R2+ , where S = {(y1 , y2 ) : y1 + y2 ≤ 1, y1 ≥ 0, y2 ≥ 0} represents the purchasing
behavior, where y1 and y2 are the probability of purchasing from the retailer 1 and
retailer 2 during the regular season respectively.
Let ν1 and ν2 be the set of feasible strategies for retailers 1 and 2, respectively.
For i = 1, 2, let Π i (ν1 , ν2 , y) be the profit of the ith retailer if retailer 1 follows
strategy ν1 , retailer 2 follows strategy ν2 and the customers follow strategy y for
i = 1, 2.
We can define the BNE as a triplet (ν1 , ν2 , y) in a similar manner as in earlier
setting, where none of the players have a profitable deviation.

Definition 6. We say that the triplet (ν1 , ν2 , y) forms a BNE in the multi-retailer
game if and only if it satisfies the following three conditions:
1. For all m1 , m2 ∈ {M1 , M2 }, we have

(y∗1 (m1 , m2 ), y∗1 (m1 , m2 ))


∈ arg max (y1 + y2 )E (v − p)A(Q0 , Zy + 1) − (v − S(Q0 − Zy ))
{y1 +y2 ≤1,y1 ≥0,y2 ≥0}

× A((Q0 − Zy )+ , D2 + D1 − Zy ) + c | D1 ≥ 1, Xμν ∗ , y = m1 , Xμν ∗ , y = m2 ,

1 2

where Zy is a Binomial random variable with (D1 − 1)+ trials each with proba-
bility y of success.
2. Fixing ν2 and y, ν1 solves:

Π 1 (ν1 , ν2 , y) ∈ arg max Π 1 (ν , ν2 , y).

ν

3. Fixing ν1 and y, ν2 solves:

Π 1 (ν1 , ν2 , y) ∈ arg max Π 1 (ν1 , ν , y).

ν
12 Availability Information 255

In Sect. 12.4 we showed that a babbling equilibrium always exists. This equilib-
rium trivially exists also in the multi-retail game. The next proposition shows that
there also exists a BNE where the firms reveal complete information regarding their
inventory to their customers. To this end, let us consider the full information equi-
librium and denote it by yFI (Q) which denotes the probability of purchasing by the
customer when they have the full information about the initial stocking level in the
regular season. Note that yFI (Q) may not be unique. We shall show that there exists
a BNE in the multi-retailer game where the purchasing behavior of the customers is
identical to yF I.

Proposition 5. Suppose there exists a full information equilibrium where yFI (Q) is
monotone decreasing in Q. Then the strategy described below will form a BNE:
1. The two retailers reveal the truth, i.e, there are messages m1i , m2i , . . . that the
retailer i uses with probability 1 when the quantity on hand is q1 , q2 , . . . , respec-
tively, for i = 1, 2
2. Suppose the two retailer provides messages that correspond to the same quan-
tity then the customer buying probability is y1 = y2 = yFI (q)/2, where q is the
quantity on hand in the regular season and purchase with equal probability in
the sales period.
3. Suppose that the two retailer provided signals that correspond to different quanti-
ties q1 based on retailer 1’s signal and q2 based on retailer 2’s signal and if q2 <
q1 then the customer’s strategy is y1 = yFI (q2 )/2 and y2 = yFI (q1 ) − yFI (q2 )/2.
Thus, the overall buying probability is yFI (max(q1 , q2 )). Further during the sales
season the customer spilt between the two retailers to ensure that the overall sales
quantity across the regular and sales season is equal for the two retailers. (The
case where q1 < q2 the strategy is in a similar manner.)

Proof. To show that we indeed have a BNE, we need to rule out the following
possible deviations:
1. (Deviations for the customer) The customer has no profitable deviation given that
the retailers are revealing the truth and yFI (·) is a BNE.
2. (Deviation for the retailer) We will focus on retailer 2 deviating from the truth
telling strategy. A similar analysis holds for retailer 1. Suppose retailer 1 signals
a message that corresponds to q1 which is the actual inventory. The retailer 2’s
deviation can be either to signal a higher quantity or a lower quantity. So, we will
consider these two cases separately.
– If retailer 2 provides a message that corresponds to quantity q2 > q1 , then the
customer reacts by changing their purchasing probability for the two retailers.
However, the overall purchasing probability remains the same. Further retailer
2’s sale in the regular season and sales season are not changed and thus this is
not a profitable deviation for retailer 2.
– If retailer 2 provides a message that corresponds to quantity q2 < q1 , then the
customer reacts by changing their purchasing probability for the two retailers.
In this case, they also change the overall purchasing probability to yFI (q2 ).
256 G. Allon and A. Bassamboo

Given the monotonicity of yFI (·), we obtain that yFI (q2 ) > yFI (q1 ). Thus, we
have that the sales for retailer 2 will lower in the regular season. Also, overall
sales quantity from the two retailer is not changed but since p > s, thus this is
not a profitable deviation for retailer 2.
This completes the proof.
Thus, the strategy induced by ηi (·) = ηFI (·) for i = 1, 2 forms a MPBNE.
The importance of this result stems from the somewhat negative result that shows
that the purchasing behavior satisfies AGNI. Here, we show that the presence of an-
other retailer sharing a common inventory can induce full revelation of the quantity
in the common pool. Thus, we show that decentralization can move the information
sharing from being completely non-informative to being fully-informative.
The key driver for the existence of a fully revealing equilibrium, in this model,
even though the inventory status is one-dimensional, is the fact that the customer can
“punish” the two senders differently given the signals. Even though both senders are
identical, when faced with a signal which is off-the-equilibrium path, the customer
punishes the senders in a differential manner. For example, if one retailer attempts
to signal higher shortage during the regular season, the customer punishes that re-
tailer by not purchasing from him at all during the regular season. Such punishment
ensures that the retailer does not want to signal lower inventory, further given the
condition on profit margin, it is not in the retailer’s best interest to signal higher
inventory either.
The two key features used to generate the possibility of communication are the
decentralization of the information providers as well as the existence of a common
pool of products. Another setting where similar features are present, but does not
fall squarely in the above model, is web-based airlines tickets market. For example,
both Expedia.com and Travelocity.com provide seat availability for flights. From the
website’s perspective the setting is similar to that described above. Apart from the
effects outlined in the model above, the customer’s decision to buy or wait in these
settings are driven both by the price patterns as well as by the temporal variation
in customers’ valuation of the tickets. However, one can argue along the lines de-
scribed above that the features described in the model are the drivers for informative
communication in these settings as well. Note that both of the features, decentral-
ization and common pool of inventory, are crucial for the ability to communicate
credible information. For example, in settings where each firm independently man-
ages not only the information it provides, but also its own inventory, no information
can be credibly shared with customers, since the model reduces to the one discussed
in the earlier section.

12.7 Concluding Remarks and Future Directions

In this chapter, we study a retail operations model where customers are strategic
not only in their actions but also in the way they interpret information, while the
retailer is strategic in the way it provides information. This chapter focuses on the
12 Availability Information 257

ability to influence consumer behavior when providing unverifiable information on

inventory availability. We develop a game-theoretic framework to study this class of
communication and discuss the equilibrium language emerging between the retailer
and its customers. In this setting, we observe that the firm is limited in its capacity
to influence customers. In fact, in most circumstances, the firm cannot influence
the purchasing behavior using messages that signal the quantity on hand. This is
consistent with a recent New York Times article which quotes Mr. Tansky, president
and chief executive of Neiman Marcus saying “We have told our customers that the
availability is less than they’re used to seeing in the stores. . . we’ve suggested that
it would be prudent to shop early.” The article then adds that “some surveys have
found that, so far, the prospect of lean inventories is not prompting consumers to
hasten their holiday shopping,” demonstrating that the proclamation that the firm is
using lean inventories is non influential.
We do show, however, that the firm is capable of influencing customer behav-
ior if it targets customer populations with diverse preferences (that is, their waiting
time cost and product valuations do not “line-up”). We provide necessary condi-
tions for an equilibrium with influential cheap talk to exist, and demonstrate how
to construct one using a simple example. One may take one of two views of this
ability to influence customer behavior—seeing the glass as half full or empty. The
firm may indeed be able, under certain conditions, to influence customer behavior
using availability information. Yet, a careful look at the conditions on the emerging
equilibrium and the customer attributes shows that these require the firms’ signaling
rule to be very sophisticated, and the customer attributes to be ordered in a specific
manner, and moreover, the effective demand cannot be impacted using these mes-
sages. We show that a firm can also establish an influential equilibrium by selling
across decentralized and independent channels.
It is important to note that in the cases in which the firm is unable to influence
customer behavior, no matter what information the firm provides, or what technique
it employs to lure customers to buy the product, a rational customer will ignore
it. In these settings, if a customer reacts in any way to the availability information
provided or to the actions taken by the firm (such as displaying the item in a specific
format), it will continuously use this method to maximize the number of customers
buying in the regular season and, in turn, maximizing its profits. Thus, the only
equilibrium language is one where customers disregard any information provided
by the firm.

Appendix: Proofs

Proof of Proposition 3. The proof follows along the same line as the proof of Propo-
sition 2. As before the firm’s incentive compatibility condition requires that the firm
signals from the set of messages that maximizes

λL yL (m) + λH yH (m).
258 G. Allon and A. Bassamboo

Thus, given the equilibrium (ν , yL , yH ), we have that there exist a constant θ for all
realization of quantity on hand that satisfies the definition of AGNI. This completes
the proof.
Proof of Proposition 4. Let Am 1
R,i be the implied availability in equilibrium for class i
customer during the regular season when the firm signals m1 . Let Am 1
S,i be the implied
availability in equilibrium for class i customer during the sales season when the firm
signals m1 . We know from Lemma 1 that

|Am1 m1
R,1 − AR,2 | ≤ Δ ,
R
|Am m1
S,1 − AS,2 | ≤ Δ .
1 S
(12.16)

For the equilibrium to be CWI, there must be a message such that

(Am m1 m1 m1
R,L − AS,L )vL + cL ≥ pAR,L − sAS,L ,
1

(12.17)
(Am m1 m1 m1
R,H − AS,H )vH + cH ≤ pAR,H − sAS,H .
1

Using (12.16), we have that

|pAm1 m1 m1 m1
R,L − sAS,L − (pAR,H − sAS,H )| ≤ pΔ + sΔ ≤ vL (Δ + Δ ).
R S R S
(12.18)

Thus, (12.17) implies that

(Am m1 m1 m1
R,L − AS,L )vL + cL ≥ (AR,H − AS,H )vH + cH − vL (Δ + Δ ).
1 R S
(12.19)

Rearranging we obtain

cH − cL ≤ (Am m1 m1 m1
R,L − AS,L )vL − (AR,H − AS,H )vH + vL (Δ + Δ )
1 R S

≤ (Am m1 m1 m1 m1 m1
R,L − AS,L )vL − (AR,H − AS,H )vL + (AR,H − AS,H )(vL − vH )
1

+ vL (Δ R + Δ S )
(a)
≤ (Am m1 m1 m1
R,L − AS,L )vL − (AR,H − AS,H )vL + vL (Δ + Δ )
1 R S

(b)
≤ 2vL (Δ R + Δ S ),

where (a) follows by noting that vH > vL and Am m1

R,H ≥ AS,H , and (b) follows from
1

(12.16). This completes the proof.

Lemma 1. For any equilibrium, we have that for any message m
 
− λL − λH 1
|Am 1
R,L − Am1
R,H | ≤ Δ R
:= 2e + 2e + E , (12.20)
Q0
 
− λL 1
|Am 1 m1
S,L − AS,H | ≤ Δ := 2e
S
+ 2e−λH + E . (12.21)
D2

Proof of Lemma 1. Let y be the equilibrium behavior for the customers. Thus, for
the message m1 customer of class-L and class-H purchase during the regular season
with probability yL (m1 ) and yH (m1 ). Thus, we can represent the
12 Availability Information 259
$  %

Q0 
Am1
=E ∧ 1  NL ≥ 1 , (12.22)
R,L NL NH
∑i=2 I{ωiL ≤yL } + 1 + ∑ j=1 I{ω H ≤yH } 

j
$ %

Q0 
Am1
=E ∧ 1  NH ≥ 1 , (12.23)
R,H NL NH
∑i=1 I{ωiL ≤yL } + 1 + ∑ j=2 I{ω H ≤yH } 
j

where ω H
j and ω j are iid sequences of uniform random variables on [0, 1]. Let us
L

define
$ %
Q0
XL = NL
∧1 ,
∑i=2 I{ω L ≤yL } + 1 + ∑Nj=1
H
I{ω H ≤yH }
i j
$ %
Q0
XH = NL
∧1 .
∑i=1 I{ω L ≤yL } + 1 + ∑Nj=2
H
I{ω H ≤yH }
i j

Since,
 NL   NL 
 NH NH 
 ∑I L + 1 + ∑ {ω Hj ≤yH }
I − ∑ {ωiL ≤yL }
I + 1 + ∑ {ω Hj ≤yH } 
I
 {ωi ≤yL }
i=2 j=1 i=1 j=2
 
 
= I{ω H ≤yH } − I{ω L ≤yL } ≤ 1,
1 1

we obtain that
1
|XL − XH | ≤ . (12.24)
Q0
Using the definitions of XL and XH , we have

|Am m1
R,L − AR,H | = |E[XL |NL ≥ 1] − E[XH |NH ≥ 1]|
1

 
 E[XL I{NL ≥1} ] E[XH I{NH ≥1} ] 
=  − 
P(NL ≥ 1) P(NH ≥ 1) 
 
 E[XL I{NL ≥1, NH ≥1} ] E[XH I{NH ≥1 ,NL ≥1} ] 
≤  − 
P(NL ≥ 1) P(NH ≥ 1) 
   
 E[XL I{NL ≥1, NH =0} ]   E[XH I{NH ≥1, NL =0} ] 
+    
P(NL ≥ 1) + P(NH ≥ 1) 
 
(a)  E[XL I{N ≥1, N ≥1} ] E[XH I{NH ≥1 ,NL ≥1} ]  −λ
≤  L H
−  + e L + e−λH

P(NL ≥ 1) P(NH ≥ 1)
 
 E[XL P(NH ≥ 1)I{NL ≥1, NH ≥1} ] − E[XH P(NL ≥ 1)I{NH ≥1 ,NL ≥1} ] 
=  

P(NL ≥ 1)P(NH ≥ 1)
+ e−λL + e−λH
 
 E[(XL P(NH ≥ 1) − XH P(NL ≥ 1))I{NL ≥1, NH ≥1} ]  −λ
=   + e L + e−λH

P(NL ≥ 1)P(NH ≥ 1)
260 G. Allon and A. Bassamboo
 
= E[(XL P(NH ≥ 1) − XH P(NL ≥ 1))|NL ≥ 1, NH ≥ 1]
+ e−λL + e−λH
(b)  
≤ E[(XL − XH )|NL ≥ 1, NH ≥ 1] + 2(e−λL + e−λH )
 
(c) 1
≤E + 2(e−λL + e−λH ).
Q0

Here (a) and (b) follows by noting that XL and XH are bounded by 1, P(NL = 0) =
e−λL and P(NH = 0) = e−λH . The inequality in (c) follows by (12.24) and mutual
independence of random variables Q0 , NL and NH . This completes the proof for
(12.20). The proof of (12.21) follows along the same line and using the definitions:
$  %
Q0 − ∑Ni=2
L
I{ω L ≤yL } − ∑Nj=1
H
I{ω H ≤yH } 

Am =E ∧ 1  NL ≥ 1
1 i j
(12.25)
S,L
NB + 1 + ∑Ni=2
L
I{ω L >yL } + ∑Nj=1
H
I{ω H >yH } 
i j
$  %
Q0 − ∑Ni=1
L
I{ω L ≤yL } − ∑Nj=2
H
I{ω H ≤yH } 

Am
S,H = E ∧ 1  NH ≥ 1 .
1 i j
(12.26)
NB + 1 + ∑Ni=1
L
I{ω L >yL } + ∑Nj=2
H
I{ω H >yH } 
i j

Proof of Theorem 1. As in the proof of Proposition 4 , we let Am 1

R,i be the implied
availability in equilibrium for class i customer during the regular season when the
firm signals m1 , and Am 1
S,i be the implied availability in equilibrium for class i cus-
tomer during the regular season when the firm signals m1 . We know from Lemma 1
that

|Am1 m1
R,1 − AR,2 | ≤ Δ ,
R
|Am m1
S,1 − AS,2 | ≤ Δ .
1 S
(12.27)

For the equilibrium to be CWI, there must be a message m1 and m2 such that

(Am m1 m1 m1
R,L − AS,L )vL + cL ≥ pAR,L − sAS,L ,
1

(Am m1 m1 m1
R,H − AS,H )vH + cH ≤ pAR,H − sAS,H ;
1

(12.28)
(Am m2 m2 m2
R,L − AS,L )vL + cL ≤ pAR,L − sAS,L ,
2

(Am m2 m2 m2
R,H − AS,H )vH + cH ≥ pAR,H − sAS,H .
2

Further, using (12.18), we have

cH − cL ≤ (Am m1 m1 m1
R,L − AS,L )vL − (AR,H − AS,H )vH + vL (Δ + Δ )
1 R S

≤ (Am m1 m1 m1 m1 m1
R,L − AS,L )vL − (AR,H − AS,H )vL + (AR,H − AS,H )(vL − vH )
1

+ vL (Δ R + Δ S )
≤ (Am m1 m1 m1
R,L − AS,L )vL − (AR,H − AS,H )vL + vL (Δ + Δ )
1 R S

+ (Am m1
R,H − AS,H )(vL − vH )
1

≤ 2vL (Δ R + Δ S ) + (Am m1
R,H − AS,H )(vL − vH )
1
12 Availability Information 261

We thus obtain:
cL − cH vL
≥ −2(Δ R + Δ S ) + (Am m1
R,H − AS,H ),
1
vH − vL vH − vL
cL − cH vL
+ 2(Δ R + Δ S ) ≥ (Am m1
R,H − AS,H ).
1
vH − vL vH − vL
The other inequalities follows in the same manner. This completes the proof.

References

Allon G, Bassamboo A (2011) Buying from the babbling retailer? The impact of availability infor-
mation on customer behavior. Manag Sci 57(4):713–726
Allon G, Bassamboo A, Gurvich I (2007) “We will be right with you”: managing customers with
vague promises and cheap talk. Oper Res 59(6):1382–1394
Austen-Smith D (1990) Information transmission in debate. Am J Polit Sci 34(1):124–152
Aviv Y, Pazgal A (2008) Optimal pricing of seasonal products in the presence of forward-looking
consumers. Manuf Serv Oper Manag 10:339–359
Cachon G, Swinney R (2009) Purchasing, pricing and quick response in the presence of strategic
consumers. Manag Sci 55:497–511
Crawford VP, Sobel J (1982) Strategic information transmission. Econometrica 50:1431–1451
Dana JD Jr (2001) Competition in price and availability when availability is unobservable. RAND
J Econ 32(3):497–513
Debo L, van Ryzin G (2009) Creating sales with stock-outs. Available at SSRN 1923706 (2009)
Demery P (2004) The cross-channel ideal. Available at https://www.internetretailer.com/2004/04/
02/the-cross-channel-ideal
Farrell J, Gibbons R (1989) Cheap talk can matter in bargaining. J Econ Theory 48(1):221–237
Fudenberg D, Tirole J (1991) Game theory. MIT Press, Cambridge
Liu Q, van Ryzin G (2008) Strategic capacity rationing to induce early purchases. Manag Sci
54:1115–1131
Matthews S (1989) Veto threats: rhetoric in a bargaining game. Q J Econ 104(2):347–370
Su X, Zhang F (2008) Strategic customer behavior, commitment, and supply chain performance.
Manag Sci 55(10):1759–1773
Su X, Zhang F (2009) On the value of commitment and availability guarantees when selling to
strategic consumers. Manag Sci 55(5):713–726
Veeraraghavan S, Debo L (2009) Joining longer queues: information externalities in queue choice.
Manuf Serv Oper Manag 11(4):543–562
Yin R, Aviv Y, Pazgal A, Tang CS (2009) Optimal markdown pricing: Implications of inventory
display formats in the presence of strategic customers. Manag Sci 55(8):1391–1408
Chapter 13
Incentives for Forecast Information Sharing
Under Simple Pricing Mechanisms

Noam Shamir and Hyoduk Shin

Abstract In this chapter we discuss the ability of firms in supply chains to share
forecast information using simple pricing mechanisms. Empirical evidence suggests
that firms exchange non-verifiable forecast information via informal talk; this stands
in sharp contrast with research suggesting that sharing non-verifiable information in
this way invites firms to act in an opportunistic way, and, thus, such information
should be exchanged using sophisticated signaling or screening mechanisms. We
survey the challenges in sharing forecasts in supply chains, and some of the recent
answers to this apparent contradiction between the observed industry practice and
the suggested mechanisms in research. Specifically, we focus on the way competi-
tion between supply-chains serves as an enabler to share forecast information, and
the way multiple decisions that are being made based on the shared information
enables firms to share non-verifiable information in a “cheap-talk” manner.

13.1 Introduction

The estimated costs of mismatch between supply and demand are significant (Kurt
Salmon Associates, Inc 1993; Troyer 1996; Cohen et al. 2003). One of the main
tools firms in a supply-chain use in order to reduce these costs is to share fore-
casts between partners in the supply-chain. The initiative to share forecasts is rein-
forced by a growing stream of research that emphasizes the value of sharing forecast

N. Shamir ()
School of Management, Tel Aviv University, Tel Aviv 6997801, Israel
e-mail: [email protected]
H. Shin
Rady School of Management, University of California, San Diego, La Jolla,
CA 92093-0553, USA
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 263

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 13
264 N. Shamir and H. Shin

information on operational measures such as improved inventory control, the

alleviation of the bullwhip effect, and the ability to better match supply with demand
(Chen 1998; Aviv and Federgruen 1998; Lee and Whang 2000; Lee et al. 2000; Ca-
chon and Fisher 2000). An excellent survey of information sharing in supply chains
can be found in Chen (2003).
In spite of the clear benefits to a supply-chain of sharing forecasts, research also
highlighted a few obstacles that exist when firms attempt to share forecasts. First,
when each firm in a supply-chain acts strategically, a firm must be better-off shar-
ing forecasts than concealing them. In many cases, although better-information im-
proves the overall performance of the supply-chain, it is possible for one firm to be
better-off not sharing information or manipulating; in this case, a firm can decide
to withheld its information even when by doing so, the supply-chain is being hurt
(Li 2002; Li and Zhang 2002; Zhang 2002). In addition, when each firm shares in-
formation strategically and the accuracy of this information cannot be verified, it
also introduces forecast manipulation; for example, a firm may attempt to influence
the decision of another firm by sharing information that is inconsistent with its ob-
served private forecast. This opportunistic forecast manipulation has been observed
in many industries, including telecommunications, commercial aircraft, defense sys-
tems, and automotive industries (Oh and Özer 2013). For example, Solectron, a ma-
jor electronics supplier, had $4.7 billion in excess component inventory because of
inflated forecasts provided by its customers (Engardio 2001). As a result of this
problem of non-credible information exchange, firms sometimes do not fully take
the information provided by their partners into account in making their decisions.
However, such a strategy can result in even greater losses: Cole (1997) documented
how Boeing’s suppliers did not believe its optimistic forecasts; consequently, they
were unable to fulfill Boeing’s large orders in time.
Due to the challenge of sharing non-verifiable information, researchers have
proposed mechanisms that can restore accountability when sharing forecast in-
formation; researchers have proposed using screening contracts or signaling. In a
screening contract, the uninformed firm in the supply-chain designs a menu of con-
tracts such that when the informed party in the supply-chain selects a specific con-
tract out of this menu, it also reveals the private information. In a signaling game,
the informed firm moves first by taking a costly action that conveys its accountabil-
ity about the shared information; the informed firm chooses an action that cannot be
mimicked by other firms holding different information (Cachon and Lariviere 2001;
Ha 2001; Özer and Wei 2006).
Interestingly, despite the prolific research that studies screening or signaling
games, empirical and anecdotal evidence suggests that firms often employ a sim-
ple wholesale price contract (Hwang et al. 2015) and still interact by sharing
information in a cheap-talk fashion. Cheap-talk refers to a situation in which
when information is exchanged it does not have any direct costs associated with
it (Crawford and Sobel 1982). As Farrell and Rabin (1996) write: “We suspect,
most information sharing is not done through Spence-style signaling, through the
price system, nor through carefully crafted Hurwicz-style incentive-compatible
mechanisms: it is done through ordinary, informal talk (p. 104).” Although
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 265

information is exchanged in an informal, cheap-talk manner, firms make impor-

tant decisions based on this information. For example, in the market for mass disk
storage systems, public announcement of optimistic market information provided
by Storage Technology Corp. encouraged its rival EMC to develop a similar prod-
uct (Mohr et al. 2010); Aleton WebSystems Inc. revealed its production plans in
order to deter its competitors from entering the same market (Mohr et al. 2010).
This chapter covers some of the answers to this contradiction—how firms are able
to exchange forecasts in informal ways under the simple wholesale price contract.
Four possible explanations were raised regarding the ability of firms to exchange
non-verifiable information using cheap-talk under a simple wholesale price.
1. Repeated Interaction In many cases, the relationship between the firms in the
supply-chain lasts for more than one period. Under repeated relationship, a firm
must consider the trade-off between the short-term benefit by influencing the de-
cisions of the other firm during the current period and the long-term prospects of
future cooperation that can be jeopardized when an opportunistic behavior is ob-
served or even partially inferred based on market conditions. Ren et al. (2010) de-
velop an infinite-horizon model, and show that under this setting, when a retailer
places a high weight on the future and prefers the relationship to endure, he will
choose to truthfully reveal his private forecast information every period. The threat
of retaliation during future periods aligns the incentive of a retailer to share infor-
mation truthfully with the manufacturer.
2. Behavioral Aspects Özer et al. (2011) examine the ability to share non-verifiable
forecast information under the given wholesale price contract; they incorporate the
notion of trust and trustworthiness, and show both analytically and experimentally
how some level of information can be shared via informal talk. Özer et al. (2011)
argue that contrary to traditional view that either firms in the supply-chain com-
pletely trust each other, or alternatively lack any credibility, there is a continuum
of trust and trustworthiness level. This continuum reveals the option to share some
degree of information even when the forecast is shared in a ‘cheap-talk’ manner.
They further explore the conditions under which achieving trust and trustworthiness
is important and the way this notion influences operational decisions. Özer et al.
(2014) later extend their findings and explore the effect of culture on the ability to
share forecast information.
3. Multiple Decisions Prior research that demonstrate the challenge in sharing
forecast information in a static model assumes that the recipient of this informa-
tion (a manufacturer) determines only the capacity level based on the shared infor-
mation, i.e., the wholesale price is fixed in advance. Chu et al. (2016) extend the
decision space of the receiver and explore the ability to share forecasts when the
recipient of the forecast information makes multiple decisions based on the shared
forecasts. Specifically, Chu et al. (2016) consider the case in which the manufacturer
sets the wholesale price and determines the capacity level based on the shared fore-
casts. They show that these multiple decisions can actually weaken the incentive of
the sender (a retailer) to manipulate the forecasts; when the retailer shares optimistic
266 N. Shamir and H. Shin

forecasts about the future demand, he is being rewarded with high capacity but at the
same time being charged with a high wholesale price. On the other hand, when the
retailer shares pessimistic forecasts about low market potential, he is being charged
with a low wholesale price, but may also face capacity constraints since the man-
ufacturer invests in a low capacity level. These opposing incentives to manipulate
shared forecasts can balance each other and actually result in truthfully sharing fore-
cast information in a static model, in particular if the value of forecast information
is high.
4. Supply-Chains Competition Shamir and Shin (2015) extend the findings of
Chu et al. (2016) and consider the case in which, instead of multiple decisions made
by a single firm, there are multiple recipients to the shared information. They con-
sider a market with competing supply-chains, and one retailer that possesses more
accurate demand forecast. In their setting, each supply-chain determines only the
capacity level. Similar to previous research (e.g., Özer and Wei 2006; Özer et al.
2011), they show that when forecasts are shared within a supply-chain, the tempta-
tion of the retailer to inflate the forecast is so high that no-information can be shared
credibly. However, when forecasts are revealed publicly to both supply-chains, in-
flating the forecasts results in an increased competition that hurts the retailer; how-
ever, this negative aspect of inflating forecasts provides credibility to the retailer
with optimistic forecasts, which can then enable truthful forecast sharing.
The topic of repeated interaction and empirical studies in forecast sharing is dis-
cussed in Chap. 2 in this book, and the behavioral incentives in information sharing
in supply chains are covered in Chap. 16 in this book. In this chapter we focus on
the topic of multiple decisions (Sect. 13.2) and competition between supply-chains
(Sect. 13.3) on the ability to share non-verifiable forecast information.

13.2 Information Sharing with Multiple Decisions

In this section, we discuss the effect of making multiple decisions on the ability of
the firms in the supply-chain to exchange forecast information.

13.2.1 The Basic Model

Consider a supply chain with a single retailer (he) who sources a product from a
single manufacturer (she). The retailer sells the product at the consumer market,
where the linear demand function is:

Q = θ − p. (13.1)
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 267

In the demand function, θ is the unknown market potential, drawn from a distribu-
tion with CDF F(·). The retailer sets the retail price p to maximize his profit subject
to the capacity level K set by the manufacturer, and his profit is:

πR = max [(p − w) min(K, Q(θ , p))] , (13.2)

p≥w

where w is the wholesale price the retailer pays to purchase one unit from the man-
ufacturer. Let p∗ (K, w, θ ) denote the optimal price of the retailer.
We assume that since the retailer is closer to the consumer market he has the
ability to observe a signal s about θ prior to the beginning of the selling season. For
expositional purposes, we present in this chapter the case that the signal provides
perfect information about the future demand, i.e., s = θ . See Chu et al. (2016) for a
more general case in which the signal is noisy, and qualitative insights are robust.
Upon observing θ , the retailer can share this forecast with the manufacturer by
sending a message θ̂ about the value of θ . The shared information between the
retailer and the manufacturer is not verifiable or contractible, and consequently the
retailer is free to choose a message θ̂ different from the observed θ .
In preparation for the selling season, the manufacturer secures capacity K for the
unit cost of c0 and determines the wholesale price w. When the demand Q is realized,
the manufacturer produces to satisfy the demand up to her capacity level. The cost of
production is c per unit, and the manufacturer builds capacity and sets the wholesale
price to maximize her expected profit given her belief about the market condition
based on shared forecasts. Her belief can be influenced by the message the retailer
shares. To be more specific, the sequence of events is as follows:
1. The retailer observes forecast θ ;
2. The retailer sends the message θ̂ , and the manufacturer establishes her belief
about the true value of θ based on her prior belief and the retailer’s message θ̂ ;
3. Based on her updated belief, the manufacturer builds capacity K and determines
the wholesale price w;
4. The retailer sets the price p. Demand is realized and the manufacturer produces
to satisfy demand up to her capacity.
We ask the following question: Can the retailer’s forecast be shared truthfully
with the manufacturer via informal talk under wholesale pricing? An equilibrium in
which forecasts are always shared truthfully is called an informative equilibrium. In
an informative equilibrium, the retailer finds it in his best interest to share a message
which reveals his actual forecast, and the manufacturer uses this information when
determining her capacity and wholesale price.

13.2.2 Two-Point Distribution

In order to illustrate the key intuitions, we first consider the case in which the
demand forecast is drawn from a two-point distribution; specifically, θ = θH with
probability μ , and θ = θL with probability 1 − μ , where μ ∈ (0, 1). To study the
268 N. Shamir and H. Shin

retailer’s incentives to transmit his private forecast information strategically, we first

explore the case in which forecast information is shared after both the wholesale
price and the retail price are fixed, followed by the case in which forecast informa-
tion is shared after the capacity has been set. We then consider the case in which the
manufacturer’s capacity and wholesale price decisions as well as the retailer’s retail
price decision are made after forecast information sharing.
Information Sharing After the Prices Have Been Set In this case, the manufac-
turer’s wholesale price w and the retailer’s retail price p are fixed. After information
transmission, the manufacturer sets her capacity level K, and then retailer sells up
to this capacity level. Given that the wholesale price is fixed, the retailer prefers
the manufacturer to invest in high capacity level. Suppose that the manufacturer
believes the retailer’s message and builds up her capacity based on the transmit-
ted information; that is, the manufacturer increases her capacity level as the retailer
increases his message about forecasts. Knowing this manufacturer’s response, the
retailer has an incentive to inflate his forecast as much as possible, regardless of his
actual forecast. Understanding the retailer’s incentive to inflate, i.e., that the shared
message is independent of the retailer’s actual forecast, the manufacturer does not
believe the retailer’s message, and information cannot be shared. Formal proof of
this statement can be found in Özer et al. (2011) and Chu et al. (2016). As a re-
sult, the only equilibrium in this case is the uninformative one—if the wholesale
price is fixed in advance, for example, due to competition or long-term contracts,
the retailer’s forecast information cannot be shared by means of cheap-talk.
Information Sharing After the Capacity Has Been Set Suppose that the man-
ufacturer’s capacity level K is fixed. The manufacturer sets the wholesale price w
following the information sharing, and the retailer subsequently determines the retail
price p and sells up to the fixed manufacturer’s capacity level. The retailer’s opti-
mal profit becomes max p≥w πR , where πR = (p − w) min(θ − p, K), which strictly
decreases in the wholesale price w under the optimal price p. As a result, the re-
tailer prefers the wholesale price as low as possible. Suppose that the manufacturer
believes the retailer’s shared message and determines the wholesale price based on
the retailer’s transmitted information. Intuitively, the manufacturer sets the higher
wholesale price when the market condition is good, i.e., under the higher forecast θ .
Recognizing this manufacturer’s behavior, the retailer then deflates his forecast as
low as possible, independent of his actual forecast, to induce the manufacturer to
set the lower wholesale price. However, knowing this retailer’s incentive, the man-
ufacturer does not believe the retailer’s shared message to be truthful; hence, no
information can be shared between the retailer and the manufacturer.
In summary, if we consider only single dimensional decision for the manufac-
turer, either a capacity decision or a pricing decision, the retailer’s forecast informa-
tion cannot be shared credibly; if the manufacturer believes him, the retailer tends to
inflate his forecast facing the manufacturer’s subsequent capacity decision, whereas
he is inclined to deflate facing the manufacturer’s subsequent pricing decision.
We next explore what happens if forecasts are shared before both the pricing and
capacity decisions by the manufacturer.
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 269

Information Sharing Before the Capacity and the Prices Have Been Set In this
case, when the retailer transmits his forecast to the manufacturer, he faces a trade-
off. For example, consider the retailer with his forecast θH , and suppose that the
manufacturer believes the retailer’s transmitted information. To study the retailer’s
incentive to manipulate, we consider what happens if the retailer lies, i.e., reports θL .
After receiving the retailer’s transmitted information θL , the manufacturer sets a
lower wholesale price, which benefits the retailer. On the other hand, the manufac-
turer invests in a smaller capacity level, and this smaller capacity becomes a binding
constraint for the retailer’s pricing p, which hurts the retailer. The question is then
whether these benefits and costs can balance each other enough to incentivize the
retailer to report truthfully:

Proposition 1. If θH − c − c0 ≥ 3(θL − c − c0 ), truthful information sharing be-

tween the retailer and the manufacturer can occur in equilibrium. Otherwise, no
information can be shared in equilibrium, i.e., babbling equilibrium is the unique
equilibrium.

Instead of considering the manufacturer’s capacity and pricing decisions sepa-

rately, if we consider both decisions of the manufacturer after information shar-
ing, the retailer’s forecasts can be truthfully shared with the manufacturer. Define
v = [θH − (c + c0 )]/[θL − (c + c0 )], where v can be viewed as a proxy for the value
of forecast information. This proposition then implies that the retailer has an incen-
tive to share his information if the value of information becomes higher, i.e., v ≥ 3.
If the value of information to the supply chain v is high, it then aligns the incen-
tives of the retailer and the manufacturer better. For example, if v is high enough
(specifically, v ≥ 3), the reduction in the wholesale price to the retailer with high
forecast does not compensate him for the scarce capacity. Thus, he prefers to share
truthfully, in order to secure the high capacity level, although the wholesale price
will be higher.
This proposition answers our question: Is it possible for the retailer to truthfully
transmit his forecast information to the manufacturer under a simple wholesale pric-
ing contract? If we consider pricing decisions in addition to the capacity decision
after information sharing, the retailer’s forecast information can be shared truthfully
within a supply chain. In particular, if the value of information is high, the retailer’s
forecast can be shared in equilibrium under simple wholesale pricing. This implica-
tion can be one potential explanation of the reason why many firms in supply-chains
still choose to contract using the simple wholesale price and share their forecasts via
informal talk.
As the capacity build-up cost c0 increases, the value of information v increases;
hence, it is more likely for the retailer to be able to share his forecast truthfully with
the manufacturer. To understand the intuition, consider the retailer with forecast θH .
As c0 increases, θL − (c + c0 ) decreases, and as a result, the capacity level built
by the manufacturer decreases as well. Thus, the retailer with θH has less incen-
tive to deflate, because by deflating, the retailer faces stricter capacity constraints.
Consequently, the higher capacity build-up cost c0 leads to more effective forecast
information sharing.
270 N. Shamir and H. Shin

13.2.3 General Distribution

We generalize our model and study the case in which θ ∈ [θL , θH ] is drawn from a
general distribution.

13.2.3.1 Partial Information Sharing

Can the retailer and the manufacturer share forecast information truthfully even for
a general distribution of the retailer’s forecast information? The next proposition
demonstrates that for a discrete distribution case, the answer is positive; i.e., it is pos-
sible to generalize the results and under certain conditions, a perfectly informative
equilibrium can emerge. We consider a discrete distribution Θ = {θ1 , θ2 , . . . , θn }
where there are n possible demand states and θL = θ1 < θ2 < · · · < θn = θH with
the probability of θi being μi , such that ∑ni=1 μi = 1.
Proposition 2. For a general discrete distribution case, perfect information sharing
between the retailer and the manufacturer can occur in equilibrium if and only if
θi+1 − c0 − c ≥ 3(θi − c0 − c) for all i = 1, 2, . . . , n − 1.
This Proposition is a generalization of Proposition 1; Proposition 1 corresponds
to the case of n = 2. As the number of possible demand states increases, i.e., as
i increases, the minimum distance between two consecutive types, θi+1 − θi , must
increase as well. As a result, it becomes harder to reach a perfectly-informative
equilibrium as the number of possible demand states increases.
From this proposition, we can also infer that it is impossible to achieve full in-
formation sharing under a general continuous distribution because the necessary
condition requires the difference between neighboring types should get larger, but,
for a continuous distribution, such a separation between the possible types is not
possible.
Corollary 1. If θ is drawn from a general continuous distribution, a perfectly infor-
mative equilibrium does not exist.
Given the negative result of Corollary 1 on full information sharing, the follow-
ing natural question arises: Can the retailer and the manufacturer achieve some level
of forecast sharing even when they cannot share forecasts fully? We reveal that they
can still share forecasts partially. Specifically, we show that any possible pure strat-
egy equilibrium with finite possible outcomes must correspond to a partition on
the interval [θL , θH ]. In a partial information-sharing equilibrium (PIS), the retailer
does not transmit the exact value of θ , but rather reports a message m that describes
a sub-interval of the interval [θL , θH ]. A partition I of [θL , θH ] with z intervals is de-
fined by z + 1 points, such that θ0∗ = θL , θ1∗ , . . . , θz−1
∗ , θ ∗ = θ , where θ ∗ < · · · < θ ∗ .
z H 0 z
We use the notation I0 to denote the null partition that includes only one interval,
i.e., z = 1.
The PIS equilibrium that we focus on corresponds to the partition equilibrium
in Crawford and Sobel (1982); in a PIS equilibrium, the retailer chooses to share
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 271

the sub-interval containing θ truthfully, and the manufacturer takes this informa-
tion into account in her capacity and pricing decisions. For illustration, consider
the following example: Assume that θ is drawn from some continuous distribu-
tion over the support [0, 1], and in a PIS equilibrium, the partition I is defined as
I = {[0, 1/3], (1/3, 1]}. Then, in a PIS equilibrium, the reporting strategy of the re-
tailer is to report [0, 1/3], if θ ∈ [0, 1/3], and to report (1/3, 1], if θ ∈ (1/3, 1], so it
is truthful, but vague.

Proposition 3. Consider a continuous distribution with positive density on support

[θL , θH ]. The equilibria with finite possible outcomes have the following structure:
(a) There exists an integer z, and θL = θ0 < θ1 < · · · < θz = θH , such that we have
pooling among types between θ j and θ j+1 for all j = 0, 1, 2, . . . , z − 1;
(b) Denote by (K j+1 , w j+1 ) as the manufacturer’s best response given that the re-
tailer type is between θ j and θ j+1 for j = 0, 1, 2, . . . , z − 1. Then, K j < K j+1 and
w j ≤ w j+1 for all j = 1, 2, 3, . . . , z − 1;
(c) The retailer type θ j is indifferent between (K j , w j ) and (K j+1 , w j+1 ) for all
j = 1, 2, 3, . . . , z − 1.

To illustrate how to find a PIS, we present our analysis with a partition of the
interval [θL , θH ] into two sub-intervals. We further assume that production is always
strictly positive. Consider the set Θ to be partitioned into two sub-intervals using
the breaking-point θ ∗ . We call the sub-interval [θL , θ ∗ ] the lower sub-interval and
denote it by IL , and refer to the interval (θ ∗ , θH ] as the upper sub-interval and denote
it by IH . In a PIS equilibrium, upon receiving the message m, the manufacturer
updates her belief about the market condition in a Bayesian manner.
In a PIS equilibrium, the manufacturer sets the wholesale price and capacity level
to maximize her expected profit, given the information received from the retailer.
The partition is designed, such that a retailer with forecast θ ∗ is indifferent between
announcing mH and mL , where mH = IH and mL = IL . The set of wholesale prices
and the capacities is chosen, such that a retailer observing θ < θ ∗ is better-off an-
nouncing mL truthfully. Although such a retailer might face a capacity constraint,
the increased wholesale price from announcing mH offsets the gains from any in-
creased capacity. Similarly, a retailer observing θ > θ ∗ is better-off announcing mH
truthfully since a different announcement would result in a lower wholesale price,
but also lead to severe capacity constraints.

13.2.3.2 Numerical Study

We present a numerical example for equilibrium outcomes and the manufacturer’s

equilibrium expected profit under the uniform distribution of the retailer’s forecast
information, i.e., θ ∼ U[θL , θH ]. Considering U[100, 150] and c = 0, we illustrate
the structure of the equilibrium outcome as a function of the capacity build-up cost
c0 in Fig. 13.1: First, if c0 ∈ (0, 84], the only equilibrium is a babbling equilibrium,
i.e., information cannot be shared in equilibrium; second, if c0 ∈ (84, 97.5], partial
272 N. Shamir and H. Shin

Fig. 13.1 Equilibrium structure under uniform distribution U[100, 150] and c = 0

information sharing via two intervals also becomes an equilibrium. For example, if
c0 = 92, the retailer reports either high, [129, 150], or low, [100, 129), as illustrated
in vertical line A in Fig. 13.1; third, if c0 increases further, c0 ∈ (97.5, 100), partial
information sharing via three intervals is also an equilibrium. For example, if c0 =
99, the retailer truthfully reports among high, [120, 150], intermediate, [103, 120),
and low, [100, 103), as depicted along the vertical line B in Fig. 13.1.
One might conjecture that the manufacturer’s expected profit in equilibrium
would decrease as the capacity build-up cost c0 increases. This conjecture holds if
we restrict our attention to the babbling equilibrium, as illustrated in Fig. 13.2. How-
ever, surprisingly, this conjecture does not hold for partial information sharing, via
either two intervals or three intervals; that is, the manufacturer’s expected profit can
increase as her unit capacity-build up cost increases. For example, for c0 ∈ [84, 89] in
Fig. 13.2, the manufacturer’s expected profit in a partial information-sharing equilib-
rium via two intervals increases as c0 increases. The reason is closely related to the
expected value of the information contained in a shared message between a manu-
facturer and a retailer, which can be measured using entropy as depicted in Fig. 13.3.
Given the equilibrium partial information sharing structure, the Shannon entropy is
calculated as H = − ∑zi=1 pi log pi , where z is the number of intervals in equilib-
rium, and pi is the length of each interval. Note that for c0 ∈ [84, 85] in Fig. 13.1,
the threshold value of θ is close to the upper bound, which implies that the amount
of shard information is small, i.e., the corresponding entropy is close to zero. As
c0 increases further, the threshold value of θ decreases, and, as a result, the two
intervals become more evenly split. Thus, the retailer’s shared forecast information
becomes more informative; hence, the entropy increases as c0 increases, which ben-
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 273

Fig. 13.2 Manufacturer’s equilibrium expected profit under uniform distribution U[100, 150] and
c=0

Fig. 13.3 The corresponding Shannon entropy. Note that pi = |Ii |/50, where |Ii | is the length of
the ith interval

efits the manufacturer. In terms of informativeness, the entropy is maximized when

c0 is approximately 94; in this case, the cut-off point in Fig. 13.1 evenly splits the
whole interval. This benefit of more informative shared information dominates the
downside of the increased capacity build-up cost c0 ; and hence, the manufacturer’s
expected profit increases as c0 increases for c0 ∈ [84, 89] in a partial information-
sharing equilibrium. This also explains that the manufacturer’s expected profit in-
creases for the partial information sharing via three intervals.
274 N. Shamir and H. Shin

13.3 Competition Between Supply-Chains

We now provide another reason that enables forecast sharing under a simple contract
via informal talk. In the previous section, we have considered multiple decisions that
the supplier makes based on shared forecasts, including the capacity and wholesale
price, which makes credible forecast sharing possible under wholesale pricing. In
this section, we introduce multiple stake holders of supply chains that are exposed
to shared forecasts, in particular, including an upstream supplier within a channel
and a competitor. For example, Boeing has been releasing its 20 year long-term
forecasts publicly, i.e., making its private demand forecasts available to its suppliers
as well as its competitors, over the past 50 years.1 When Boeing announces opti-
mistic forecasts, it may potentially lead its competitor to act more aggressively. On
the other hand, when Boeing releases pessimistic forecasts, its suppliers may re-
duce their capacities, which may result in supply shortages and delayed deliveries.
Despite those potential negative consequences, why does Boeing make its forecast
publicly available?

13.3.1 The Model

In order to illustrate strategic interactions and impacts of public forecast sharing in

a competitive setting, we study a market consisting of two competing supply chains,
an incumbent supply chain (denoted by the index I) and a potential entrant (dented
by the index E). For a detailed analysis of this section, please see Shamir and Shin
(2015). Each supply chain is comprised of an upstream manufacturer and a down-
stream retailer. Before the selling season, the entrant retailer decides whether to
enter the market or not, denoted by aE ∈ {In, Out}. If the entrant retailer decides to
enter, he incurs a fixed entry cost F, and the retailers compete in a Cournot fashion;
specifically, the inverse demand curve is p(θ , qI , qE , aE ) = θ − qI − qE , if aE = In,
where qI and qE are the order quantities of the incumbent retailer and the entrant
retailer, respectively. Otherwise, i.e., if aE = Out, p(θ , qI , qE , aE ) = θ − qI . As in
the previous section, the intercept of the inverse demand curve θ captures the po-
tential market size and is uncertain, taking the value of θH with probability μ or θL
with probability 1 − μ . The incumbent retailer obtains a forecast about θ , prior to
the beginning of the selling season, denoted by the informative signal s, taking the
values sh or sl with the accuracy ρ = Pr(sh |θH ) = Pr(sl |θL ) > 0.5. Each manufac-
turer secures capacities Ki , i ∈ {E, I}, prior to observing the actual demand. Both
manufacturers have an identical constant marginal production cost c, and they both
sell to their respective retailer at the given unit wholesale price of w, up to their
capacity.

1 Long-Term Market: Current Market Outlook 2014–2033, can be found at http://www.boeing.

com/boeing/commercial/cmo/index.page (last accessed May 31st 2015).

13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 275

The sequence of events is as follows:

1. The incumbent retailer observes the private forecast s.
2. Forecast is shared/not shared.
3. The entrant retailer decides whether to enter the market (aE = In) or not
(aE = Out).
4. The incumbent manufacturer builds capacity KI . If the entrant retailer enters the
market, the entrant manufacturer also decides on her capacity level KE .
5. Both retailers observe θ , and determine order quantities qI and qE up to their
respective manufacturers’ capacities.
In the second stage when forecast is shared, the incumbent retailer decides to
send a message m, which may differ from his observed forecast s, since it is not
verifiable. Simultaneously, the other parties of the supply chains observe the incum-
bent retailer’s message m, and form a belief about the value of s conditional on this
message m, denoted by a probability measure μm .
The profit of the incumbent retailer can be written as πI (qI |KI , KE , aE , θ ) =
(p(θ , qI , qE , aE )−w)qI , and denote his optimal order quantity by q∗I (KI , KE , aE , θ ) ∈
arg maxqI ≤KI πI (qI |KI , KE , aE , θ ).
Similarly, in case aE = In, the entrant retailer’s profit is πE (qE |KE , KI , aE , θ ) =
(p(θ , qE , qI , aE ) − w) qE − F. Otherwise, i.e., aE = Out, it is zero. Denote his
optimal order quantity by q∗E (KE , KI , aE , θ ) ∈ arg maxqE ≤KE πE (qE |KE , KI , aE , θ ).
Furthermore, the incumbent manufacturer’s profit can be written as ΠI (KI |aE ) =
q∗I (KI , KE , aE , θ )w − cKI , and she decides her capacity level based on her belief
μm , i.e., the optimal capacity level of the incumbent manufacturer is KI∗ (aE ) ∈
arg maxKI E [ΠI (KI |aE )|μm ]. The entrant manufacturer’s profit is ΠE (KE |aE ) =
q∗E (KI , KE , aE , θ )w − cKE if aE = In, in which case the optimal capacity level of
the entrant manufacturer is KE∗ (aE ) ∈ arg maxKE E [ΠE (KE |aE )|μm ]. We adopt the
Perfect Bayesian Equilibrium as our equilibrium concept.

13.3.2 Benchmark

Before presenting the case of forecast sharing through public announcement, we

first provide two benchmark cases: (1) outcome under no-information sharing, and
(2) possibility of credible forecast sharing within the supply chain.

13.3.2.1 No-Information Sharing

Since both manufacturers have the same prior demand information before setting
their capacities, and the market uncertainties have been resolved prior to both re-
tailers’ deciding their order quantities, the incumbent retailer has no informational
advantage over the entrant retailer in decision-making in this case of no-information
sharing. The equilibrium outcome is given as follows:
276 N. Shamir and H. Shin

Fig. 13.4 Possible market outcome when no information is shared as a function of the prior belief
and the entry cost. Parameter values are: θH = 33, θL = 15, w = 10, c = 5, and ρ = 0.8

Proposition 4. There exist F1 (μ ) > F2 (μ ), and μ̄ such that:

(a) If F ≥ F1 (μ ) or if F2 (μ ) ≤ F < F1 (μ ) and μ ≤ μ̄ , then aE = out and
!
(θH − w)/2 if μ ≥ μ̄ ,
KI =
(θL − w)/2 if μ < μ̄ ,

(b) otherwise, aE = In and

!
(θH − w)/3 if μ ≥ μ̄ ,
KI = KE =
(θL − w)/3 if μ < μ̄ ,

where μ̄ = c/w,
   
θH − w 2 θL − w 2
F1 (μ ) = μ + (1 − μ ) , and
3 3
   
2θL w θL − w θL − w 2
F2 (μ ) = μ θH − − + (1 − μ ) .
3 3 3 3

If the entry cost is high, i.e., F2 (μ ) ≤ F, and it is also less likely that the demand
is high, i.e., (μ < μ̄ ), as depicted in Case A in Fig. 13.4, the entrant decides to stay
out of the market and the incumbent manufacturer invests in the low-capacity level
of monopoly (KI = (θL − w)/2). However, if the entry cost is high, but it is more
likely that the demand is high, i.e., (μ ≥ μ̄ ), as depicted in Case B in Fig. 13.4, the
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 277

entrant decides not to enter the market but the incumbent manufacturer invests in the
high-capacity level (KI = (θH − w)/2). In contrast, if the entry cost is low, in Cases
C and D, the entry occurs. Furthermore, depending on how likely that the market
is good, i.e., whether μ ≥ μ̄ holds or not, both manufacturers invest in the duopoly
capacity level accordingly; specifically, if μ ≥ μ̄ , KI = KE = (θH − w)/3 (case D),
and otherwise, KI = KE = (θL − w)/3 (case C).

13.3.2.2 Forecast Sharing Within the Supply Chain

Is it possible for the incumbent retailer to credibly share his forecasts with his own
manufacturer within a supply chain? In order to avoid trivial cases, define mean-
ingful information as information that changes the behavior of the recipient com-
pared to no-information sharing case. If meaningful information is shared in equi-
librium, we refer to this equilibrium as an influential equilibrium. Otherwise, we
call it uninformative.

Proposition 5 (Özer et al. 2011). The only equilibrium in the case of forecast
sharing within the supply chain is uninformative.

Proposition 5 emphasizes the difficulty of sharing meaningful information when

the forecast is shared only within the supply chain by informal talk under wholesale
pricing. This case has been discussed in Sect. 13.2.2 for the case of information
sharing after the prices have been set.

13.3.3 Public Forecast Sharing

This section analyzes how the exposure of the competing supply chain to the shared
forecast affects the ability of the incumbent retailer to truthfully share his forecast.
Specifically, we reveal that the incumbent retailer can achieve meaningful informa-
tion sharing by making his forecast publicly available. We show that the incumbent
retailer with high forecast sh is perceived as being accountable for the shared fore-
cast, by intensifying the competition between the supply chains, and thus can induce
the manufacturer to invest in the high-capacity level. In addition, in an influential
equilibrium, an incumbent retailer who observes the low forecast sl is able to cred-
ibly deter the competing supply chain from entering the market by simultaneously
reducing the capacity level of his manufacturer.
The following Lemma presents a unique outcome of an influential equilibrium
achieved by publicly sharing forecasts.
278 N. Shamir and H. Shin

Lemma 1. In an influential equilibrium, achieved by publicly sharing forecasts, the

unique outcome is:
(a) If the incumbent retailer announces high forecast sh , the entrant supply chain
decides to enter the market, and both manufacturers invest in the high-capacity
level under duopoly, i.e., (θH − w) /3.
(b) If the incumbent retailer announces low forecast sl , the entrant supply chain
decides to stay out of the market, and the incumbent manufacturer invests in the
low-capacity level under monopoly, i.e., (θL − w) /2.

An incumbent retailer has an incentive to report high forecasts sh only if this

announcement results in a high capacity-level investment carried by his manufac-
turer, and this retailer is perceived as credible only if his announcement is followed
by intense competition of the competing supply chain. In addition, an incumbent
retailer may have an incentive to truthfully report low forecasts sl , only if this an-
nouncement weakens the competition of the entrant supply chain (specifically, this
announcement induces the competing supply chain to stay out of the market), and
this retailer obtains credibility for the shared forecast only if this forecast also results
in reduced capacity level of his manufacturer.
Understanding what happens in an influential equilibrium, a natural following
question is when an influential equilibrium can arise, i.e., when credible forecast
sharing can happen in the case of public forecast sharing? In order to achieve mean-
ingful information sharing, a retailer with the high forecast should be better-off re-
vealing this forecast than mimicking the retailer with the low forecast. Similarly, a
retailer with the low forecast should be also better-off revealing this forecast rather
than mimicking the retailer with the high forecast. Based on Lemma 1, these condi-
tions can be written as:
   
θH − w 2 θL − w 2
μ (sh ) + (1 − μ (sh ))
3 3
   
θL w θL − w θL − w 2
≥ μ (sh ) θH − − + (1 − μ (sh )) , (IChl )
2 2 2 2
   
θL w θL − w θL − w 2
μ (sl ) θH − − + (1 − μ (sl ))
2 2 2 2
 2  2
θH − w θL − w
≥ μ (sl ) + (1 − μ (sl )) , (IClh )
3 3

where μ (si ) is the updated posterior belief in the following Bayesian manner:
⎧ μρ
⎪
⎨ if si = sh ,
μρ + (1 − μ )(1 − ρ )
μ (si ) = Pr(θH |si ) = μ (1 − ρ ) (13.3)
⎪
⎩ if si = sl .
μ (1 − ρ ) + (1 − μ )ρ
The condition (IChl ) represents the incentive compatibility constraint of an incum-
bent retailer with the high signal. The Left-Hand Side (LHS) denotes this retailer’s
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 279

expected profit when he reports truthfully that he observed the high signal, and the
Right-Hand Side (RHS) represents this retailer’s profit when he reports falsely that
he observed the low signal. If the retailer reports that he observed the high signal,
he does not face capacity constraints when demand is high, but he does suffer from
intense competition in the market. On the other hand, if the retailer reports that he
observed the low signal, he does not suffer from competition since the entrant sup-
ply chain decides to stay out of the market, but he faces capacity constraints when
demand is high. Similarly, condition (IClh ) represents the incentive compatibility
constraint of an incumbent retailer with the low signal.

Proposition 6. Meaningful information sharing, can be achieved in the case of pub-

lic forecast sharing if and only if the following two conditions are satisfied:
(a)
1& '
9+ 5 (9 + (1 − μ )(1 − ρ )/(μρ ))
4
1& '
≤Ψ ≤ 9 + 5 (9 + (1 − μ )ρ /(μ (1 − ρ ))) , (13.4)
4
where Ψ = (θH − w)/(θL − w).
(b) F2 (sl ) < F ≤ F1 (sh ) and μ (sh ) ≥ μ̄ > μ (sl ), where
 2  2
θH − w θL − w
F1 (sh ) = μ (sh ) + (1 − μ (sh )) ,
3 3
   
2θL w θL − w θL − w 2
F2 (sl ) = μ (sl ) θH − − + (1 − μ (sl )) .
3 3 3 3

An important factor that affects the ability to reach meaningful information shar-
ing is the measure of market variability, i.e., the distance between θH −w and θL −w.
Similar to v introduced in Part II, as a measure of this distance, we use the param-
eter Ψ = (θH − w)/(θL − w), which is a proxy to the value of forecast information.
When the incumbent retailer reports that he observed the high forecast, he knows
that in an influential equilibrium, it results in more intense competition from the
entrant supply chain, and that his manufacturer will invest in the higher capacity
level. If the realized demand is low, this announcement results in a profit loss to
the retailer. In order for this announcement to be profitable, the profit gain during
the high-demand state must be high enough. This level can be achieved if the pa-
rameter Ψ is greater than the lower bound prescribed in part (a) of Proposition 6.
Moreover, in order to achieve an influential equilibrium, the measure Ψ should be
also bounded from above. In an influential equilibrium, a retailer with the low fore-
cast should be better-off reporting truthfully than by mimicking the retailer with the
high forecast. However, if the profit that the low-type retailer (with forecast sl ) can
gain from mimicking the high-type retailer (with forecast sh ) is very high, which is
the case when Ψ is very high, the low-type retailer will choose to inflate the shared
information. Consequently, the value of forecast information Ψ should be bounded
from above to guarantee truthful forecast sharing.
280 N. Shamir and H. Shin

Part (b) imposes two additional conditions: (1) the entry cost should be between
F2 (sl ) and F1 (sh ), and (2) the posterior belief should satisfy μ (sl ) < μ̄ ≤ μ (sh ). The
first condition implies that given low signal and low-capacity investment, the entrant
retailer cannot cover his entry cost, and thus stays out of the market. However, in
the case of high signal and high-capacity level, the entrant enters the market leading
to intensified competition. The second condition means that upon learning that the
observed signal is sl , the manufacturers invests in the low-capacity level, and that
upon learning that the signal is sh , the manufacturers invest in the high-capacity
level.
Figure 13.5 depicts the possibility to credible forecast sharing with respect to the
prior belief μ and the entry cost F, in order to illustrate the results of Proposition 6.
In region I, no influential equilibrium can be achieved since either the entry costs are
so high that the entrant retailer stays out of the market regardless of the forecasts, or
alternatively the entry costs are so low that the entrant retailer always enters the mar-
ket. In both cases, the incentive compatibility constraints of the incumbent retailer
in Proposition 6 are not satisfied. In region II, similarly, no influential equilibrium
can be achieved; in this case, either the prior belief μ is so low that the incum-
bent manufacturer always invests in the low-capacity level, or it is so high that the
incumbent manufacturer always invests in the high-capacity level. Region III also
describes a scenario in which no influential equilibrium can be achieved. In this
case, the retailer with a high demand forecast is better off by mimicking the retailer
with a low forecast.

Fig. 13.5 Equilibrium outcome and comparison with the benchmark case. Parameter values are
the same as those given in Fig. 13.4
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 281

Region IV describes the area in which an influential equilibrium can be achieved,

and it is divided into four cases according to the market outcome without informa-
tion sharing (see Fig. 13.4); those four cases within Area IV, i.e., IV.A through IV.D
correspond to Area A though D in Fig. 13.4. For example, IV.A describes the case
in which without forecast sharing the entrant stays out of the market, and when the
optimistic forecast is shared, he decides to enter the market. In addition, without
forecast sharing, the incumbent manufacturer invests in the low capacity level, and
she increases the capacity when learning that the forecast is high. This region is
consistent with the example of the market for mass disk storage systems discussed
in the Introduction; specifically, public announcement of optimistic market infor-
mation provided by Storage Technology Corp. encouraged rival EMC to develop a
similar product (Mohr et al. 2010). On the other hand, firms sometimes provide pes-
simistic forecast that results in weakened competition, as in Area IV.D. For example,
Network equipment manufacturer, Aleton WebSystems Inc. revealed its production
plans to deter its competitors from entering the same market (Mohr et al. 2010).
By sharing the optimistic forecast publicly, the incumbent retailer benefits from
increased capacity of his own manufacturer. At the same time, it hurts him due to
increased competition; however, this negative impact then provides credibility of his
optimistic forecast which enables truthful forecast sharing. A natural question that
follows is: Who benefits, if any, from this forecast sharing compared to no-forecast
sharing?
We focus on the case in which the incumbent retailer obtains optimistic forecasts,
i.e., sh (please see Shamir and Shin 2015, for the other case). In this case, in an
influential equilibrium, the competing supply chain operates in the market, and both
manufacturers invest in the high-capacity level. In this equilibrium, the incumbent
retailer is able to influence his manufacturer to invest in the high-capacity level by
inducing the competing supply chain to intensify competition.

Proposition 7. Under the conditions of Proposition 6 when the incumbent retailer

observes the signal sh :
(a) Forecast sharing hurts the incumbent retailer, compared with no-information
sharing case, if and only if F > F1 (μ ), and μ ≥ μ̄ .
(b) Forecast sharing hurts the incumbent manufacturer, if and only if
(i) F > F2 (μ ), μ < μ̄ , and

w − c 2(1 − μ (sh ))(θH − θL )

≥ , or
w 2θH − 3θL + w

(ii) F > F1 (μ ) and μ ≥ μ̄ .

(c) The entrant retailer and the entrant manufacturer are better-off when the forecast
is shared.
(d) Meaningful information sharing makes the consumers better-off.
282 N. Shamir and H. Shin

First, the competing supply chain and consumers always become better-off as
a result of the public forecast sharing in this case, whereas the effect of forecast
sharing on the profits of the incumbent retailer and the incumbent manufacturer is
inconclusive. Interestingly, there is a case in which the incumbent retailer is worse-
off when forecast is shared. In this case, without forecast sharing, the entrant supply
chain stays out of the market and the incumbent manufacturer invests in the high-
capacity level, i.e., (θH − w)/2. However, when the optimistic forecast is shared,
the entrant retailer enters the market, and the incumbent manufacturer updates the
capacity level to (θH − w)/3 to reflect the new market structure. In this case, both
effects, of intensifying competition and reducing the capacity level of the incumbent
manufacturer, hurt the incumbent retailer compared with no-information sharing
case. In this case, it also hurts the incumbent manufacturer, case (ii) in part (b). In
all other possible cases, forecast sharing benefits the incumbent retailer.
Private forecast information can be shared within the supply chain through com-
plex contracts, e.g., an advance purchase contract. The last question we address is
this chapter is: How does public forecast sharing perform compared to complex
mechanisms like an advance purchase contract? We find that as the forecast gets
more accurate, the advance purchase contract becomes more attractive to the in-
cumbent retailer than public forecast sharing. Interestingly, this finding is aligned
with the example of Boeing. The Market Outlook that Boeing publishes every year
is an aggregated forecast for the next 20 years, i.e., its long-term forecast, which is
in nature less accurate than short-term forecasts. Consistent with our findings, Boe-
ing publicly shared its long-term forecast, and did not publicly announce its short-
term forecast, which is better to be shared within the supply chain via complicated
contracts.

13.4 Concluding Remarks

We demonstrate how forecast information can be shared truthfully in supply chains

via cheap talk under a wholesale price contract. First, the fact that the manufacturer
makes multiple decisions based on the shared forecasts brings interesting dynamics
into strategic interactions of the information exchange between the retailer and his
manufacturer. While with a single decision after information sharing, truthful in-
formation sharing cannot occur, the countervailing incentives the retailer faces with
the multiple manufacturer’s decisions after information sharing can result in truthful
information exchange. Second, by making forecasts publicly available including to
the competitors, the retailer can obtain the credibility of his forecasts, which then
enables him to share his forecast truthfully with his manufacturer.
Our findings have several important implications. First, we provide an explana-
tion for the prominent use of cheap-talk as a mechanism for information exchange
among firms in a supply chain under simple contracts. Second, firms that wish to
share information in a supply chain can plan when to share and whom to share
13 Incentives for Forecast Information Sharing Under Simple Pricing Mechanisms 283

accordingly. For example, a manufacturer can communicate to the retailer that he

is bundling the operational decision and the marketing decision together in order
to encourage the retailer to share forecasts truthfully. However, if the manufacturer
commits to his pricing decision in advance, forecast information is hard to be shared
via informal talk.

References

Aviv Y, Federgruen A (1998) The operational benefits of information sharing and vendor managed
inventory (VMI) programs. Working paper, Olin School of Business, Washington University
Cachon GP, Fisher M (2000) Supply chain inventory management and the value of shared
information. Manag Sci 46(8):1032–1048
Cachon G, Lariviere M (2001) Contracting to assure supply: how to share demand forecasts in a
supply chain. Manag Sci 47(5):629–646
Chen F (1998) Echelon reorder points, installation reorder points, and the value of centralized
demand information. Manag Sci 44(12):S221–S234
Chen F (2003) Information sharing and supply chain coordination, Chap 7. In: de Kok A, Graves S
(eds) Handbooks in operations research and management science – supply chain management:
design, coordination and operations. North-Holland, Amsterdam
Chu L, Shamir N, Shin H (2016) Strategic communication for capacity alignment with pricing in
a supply chain. Management Science (forthcoming)
Cohen M, Ho T, Ren Z, Terwiesch C (2003) Measuring imputed cost in the semiconductor
equipment supply chain. Manag Sci 49(12):1653–1670
Cole J (1997) Boeing, pushing for record production, finds part shortages, delivery delays. Wall
Street J http://www.wsj.com/articles/SB867281973861622000
Crawford V, Sobel J (1982) Strategic information transmission. Econometrica 50(6):1431–1451
Engardio P (2001) Why the supply chain broke down. Bus Week http://www.bloomberg.com/news/
articles/2001-03-18/why-the-supply-chain-broke-down
Farrell J, Rabin M (1996) Cheap talk. J Econ Perspect 10(3):103–118
Ha AY (2001) Supplier-buyer contracting: asymmetric cost information and cutoff level policy for
buyer participation. Nav Res Logist 48(1):41–64
Hwang W, Bakshi N, DeMiguel V (2015) Inducing reliable supply: simple versus simplistic
contracts. London Business School, Working paper
Kurt Salmon Associates, Inc (1993) Efficient consumer response: enhancing consumer value in the
grocery industry. Food Marketing Institute, Washington, DC
Lee H, Whang S (2000) Information sharing in a supply chain. Int J Manuf Technol Manag
1(1):79–93
Lee HL, So KC, Tang CS (2000) The value of information sharing in a two-level supply chain.
Manag Sci 46(5):626–643
Li L (2002) Information sharing in a supply chain with horizontal competition. Manag Sci
48(9):1196–1212
Li L, Zhang H (2002) Supply chain information sharing in a competitive environment. In: Song
J-S, Yao DD (eds) Supply chain structures. Springer, New York, pp 161–206
Mohr JJ, Sengupta S, Slater SF (2010) Marketing of high-technology products and innovations.
Pearson Prentice Hall, Upper Saddle River
Oh S, Özer Ö (2013) Mechanism design for capacity planning under dynamic evolutions of
asymmetric demand forecasts. Manag Sci 59(4):987–1007
Özer Ö, Wei W (2006) Strategic commitments for an optimal capacity decision under asymmetric
forecast information. Manag Sci 52(8):1238–1257
284 N. Shamir and H. Shin

Özer Ö, Zheng Y, Chen K (2011) Trust in forecast information sharing. Manag Sci 57(6):
1111–1137
Özer Ö, Zheng Y, Ren Y (2014) Trust, trustworthiness, and information sharing in supply chains
bridging China and the US. Manag Sci 60(10):2435–2460
Ren Z, Cohen MA, Ho TH, Terwiesch C (2010) Information sharing in a long-term supply chain
relationship: the role of customer review strategy. Oper Res 58(1):81–93
Shamir N, Shin H (2016) Public information sharing in a market with competing supply chains.
Manag Sci (forthcoming)
Troyer C (1996) EFR: efficient food service response. Paper presented at the conference on
logistics, GMA, May 21, Palm Springs, California
Zhang H (2002) Vertical information exchange in a supply chain with duopoly retailers. Prod Oper
Manag 11(4):531–546
 Part IV
Incentives for Information Sharing
Chapter 14
Establishing Trust and Trustworthiness
for Supply Chain Information Sharing

Özalp Özer and Yanchong Zheng

“I know who I can depend on, I know who to trust

I’m watching the roads, I’m studying the dust.”
– Bob Dylan

Abstract In this chapter, we discuss when, how, and why trust and trustworthiness
arise to support credible information sharing and cooperation in a supply chain.
Synthesizing our learning, we identify the four building blocks of trust and trust-
worthiness as personal values and norms, market environment, business infrastruc-
ture, and business process design. We elaborate on these building blocks and offer
tangible insights into how to establish more trusting and cooperative supply chain
relationships.

14.1 Introduction

We are living in a supply chain society. Virtually every product in our lives comes
from supply chains that often transcend national borders. From the food we eat,
to the clothes we wear, the electronics we use, to the car we drive, the aircraft we
fly in and the medicine we take, all of them are the results of collaborative efforts
between business partners in global supply chains. Therefore, the efficiency of our
lives depends on the efficiency of these supply chains. One of the most critical
enablers of an efficient supply chain is to ensure timely and credible information
sharing among supply chain partners. The key question is then, how do we achieve
this enabler?

Ö. Özer ()

Jindal School of Management, The University of Texas at Dallas, Richardson, TX 75080, USA
e-mail: [email protected]
Y. Zheng ()
Sloan School of Management, Massachusetts Institute of Technology,
Cambridge, MA 02139, USA
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 287

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 14
288 Ö. Özer and Y. Zheng

The development of information technologies (IT) in the last two decades has led
to the blossoming of information management systems, which have been adopted
by large companies such as General Motors, Procter & Gamble, Neiman Marcus,
and many others, to facilitate information sharing within their supply chains (The
Economist 2008; CDC White Paper 2011). IT is a necessary facilitator for the exc-
hange of information; however, it is not sufficient to guarantee truthful information
sharing. As a result, the value of such IT systems heavily depends on the credibility
of the information fed through them and how well the information is put to use.
Consider the example of a two-tier supply chain with a supplier and a downstream
buyer, e.g., Hewlett-Packard (HP) and BestBuy, or a tool equipment supplier and
Intel.1 Due to long production leadtimes, the supplier needs to either secure costly
capacity or to build up and hold inventory before receiving binding orders from
the buyer. On the other hand, the buyer often has better demand forecast infor-
mation because of its proximity to the end market. In an ideal world, the buyer
would share its forecast information credibly with the supplier and the supplier
would utilize the shared information to make better capacity and inventory deci-
sions. However, when the financial incentives of the two parties are not well aligned,
the buyer may report overoptimistic forecasts to ensure an abundant supply. Yet,
the supplier, anticipating this tendency, may not find the information credible and
may discount the forecast provided by the buyer (even if the buyer were to provide
truthful information). Forecast manipulation and the resulting over-caution by up-
stream supply chain members are prevalent throughout industries ranging from elec-
tronics to semiconductors, medical equipment and commercial aircraft. For exam-
ple, personal computer (PC) and electronics manufacturers often submit “phantom
orders” to induce their suppliers to secure more component capacity (Lee et al.
1997). Researchers have empirically shown that semiconductor manufacturers
order on average 30 % less than the demand forecasts they share with their suppliers
(Cohen et al. 2003). General Motors (GM) admits that it “purifies” the demand fore-
cast information received from its dealers before using it to decide on the component
capacity of its assembly lines (Özer and Wei 2006). These conflicting incentives and
interactions prevent critical information from being shared effectively and put to the
best use within the supply chain. As a result, the supply chain suffers from either
having too much inventory tied up or missing potential demand. Ultimately, the lack
of credible supply chain information sharing also hurts consumers, who suffer from
high prices and stock-outs.
How could a supply chain that comprises of companies with potentially conflict-
ing financial incentives mitigate inefficiency due to a lack of credible information
sharing? One possible remedy is to design risk-sharing contracts that align the pec-
uniary incentives of the supply chain partners (Cachon and Lariviere 2001; Özer and
Wei 2006).2 Yet, designing and implementing such contracts often incurs high trans-
action costs due to the complexity of the contracts and, sometimes, to the necessity
1 The example applies to a general supply chain dyad (e.g., supplier-manufacturer or manufacturer-

retailer) in which the downstream member has better demand information than the upstream one
and the upstream production leadtime is longer than the downstream delivery leadtime.
2 We refer the reader to Part 2 of this book for a more detailed discussion on this line of research.
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 289

of auditing managerial actions. Alternatively, some argue that as long as companies

interact repeatedly over time, each company’s concerns about its own reputation
will solve the incentive problem. Concerns about reputation do indeed have an effect
(e.g., Ren et al. 2010). Nevertheless, relying solely on reputation concerns to solve
incentive conflicts may not be sufficient for a few reasons. First, the effective use
of reputation to deter opportunistic behavior relies on the ability to clearly identify
such behavior and impose penalties. However, such deterrence may not be possible
because uncertainty in the market environment often makes it difficult for a com-
pany to determine whether its partner has intentionally manipulated information or
it has simply experienced a “bad” market. Second, a strategy that uses reputation to
deter opportunistic behavior requires the company to have a good historical record
of its transactions with its partners (e.g., Kreps et al. 1982; Radner 1985; Dal Bó
2005). This information requirement can be challenging for companies that interact
with a large number of partners or that do not have the capability to collect and
process the data. Third, there are many business situations in which the parties prac-
tically transact only once. Examples include the majority of online transactions, the
establishment of a new business relationship, the practice of “lifetime buy,” in which
a manufacturer procures components for the entire life cycle of a product in a single
order. In these situations, repeated interactions and the associated reputation mech-
anism are unavailable. We suggest that there is another important factor, somewhat
overlooked historically but increasingly being studied—the nonpecuniary, behav-
ioral motives that may facilitate or deter information sharing among human deci-
sion makers. That is, as supply chain managers are human-beings, not machines
or algorithms, they may be inclined to be trustworthy and trusting in their infor-
mation exchange because deceiving and exploiting others often induce moral costs
(Erat and Gneezy 2012). However, the extent of an individual’s trustworthiness and
trust may vary. Hence, it is important to understand when, how, and why trust and
trustworthiness arise, resulting in truthful information sharing and cooperation in a
supply chain (see also Zipkin 2012). This is the focus of the current chapter.
Anthropologist Edward Hall argues in The Silent Language (1959) that trust
plays a stronger role than legal contracts in supporting cooperative relations. Nobel
Laureate Kenneth Arrow echoes Hall’s assertion, stating that “virtually every com-
mercial transaction has within itself an element of trust” (Arrow 1972, p. 357).
No two executives or firms begin talking with each other by first writing down a
legal contract. Just like our own relationships in life, all successful and sustain-
able business relationships are built upon a considerable degree of mutual trust and
trustworthiness. Indeed, countries and regions with overall higher levels of trust
and trustworthiness observe less friction in business transactions and more pros-
perous economic growth (Knack and Keefer 1997; La Porta et al. 1997; Zak and
Knack 2001). Practitioners are not oblivious to the important role of trust in business
relationships. For example, the successful implementation of the vendor managed
inventory (VMI) initiative between Wal-Mart and Procter & Gamble was built upon
the tremendous trust between the two companies, allowing for timely and credible
information sharing (Kumar 1996). Indeed, a recent extensive analysis of supply
chain projects (e.g., VMI, EDI, CPFR, and RFID projects), which examines over 60
implementations of such projects between multinational firms in various industries,
290 Ö. Özer and Y. Zheng

confirms that the success of such projects critically depends on the degree of trust
between parties (Brinkhoff et al. 2015). The researchers find, for example, that the
degree of trust is a stronger predictor of success than the degree of asymmetric
power/dependence between the firms. Incidentally, the Boeing Company, which
works with 17,500 suppliers in more than 50 countries, commented that “it has
sometimes been a job to persuade all these suppliers to invest enough to meet future
demand.” An effective way to do so, the company learned, is to build more trust in
the supply chain and be more open to sharing information with its suppliers (The
Economist 2012).
What is trust and how can we build trust in a supply chain? A definition of trust
commonly agreed upon across multiple disciplines stipulates that “trust is a psy-
chological state comprising the intention to accept vulnerability based upon posi-
tive expectations of the intentions or behavior of another” (Rousseau et al. 1998,
p. 395). In the context of supply chain information exchange, being trusting means
that the party receiving the shared information is willing to rely on that information
to make operational decisions even though the information may have been manipu-
lated, exposing itself to potentially significant financial vulnerabilities. Conversely,
being trustworthy means that the party owning the private information does not ma-
nipulate the information to its own benefit in the exchange.
For a long time, trust has been considered as something intangible. However, it
is indeed possible to quantify what it means to trust. To do so, one needs a specific
context and a three-way interaction; i.e., A trusts B in doing C. The statement “HP
trusts BestBuy” is not meaningful because there is no way to tangibly measure what
trust means in such a statement. However, the statement becomes concrete and com-
plete when the environment (e.g., information, decisions, and uncertainties) related
to trust and trustworthiness is specified. Thus by saying HP trusts BestBuy to pro-
vide accurate forecast information, we now have a concrete context to quantify and
articulate what it means to trust or to be trustworthy.
In our past and ongoing research, we have observed that trust and trustworthi-
ness are not necessarily genetic predispositions. Instead, they are shaped and influ-
enced by various environmental and procedural factors. Synthesizing our learning,
we have identified four building blocks of trust and trustworthiness, as shown in
Fig. 14.1: personal values and norms, market environment, business infrastructure,
and business process design.
The rest of this chapter will focus on these four building blocks in four subsec-
tions, each of which can be read independent of the others. Our goal is to discuss
what we have learned in our endeavor to uncover the role of trust and trustworthi-
ness in supply chain information exchange. We hope to offer tangible insights into
how to establish more trusting and cooperative supply chain relationships. As such,
we do not intend to provide a comprehensive review of the vast literatures on trust
and information sharing; the references cited in this chapter are only a starting point
for a curious reader wishing to explore further.
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 291

Fig. 14.1 Building blocks of trust and trustworthiness

14.2 Personal Values and Norms

Supply chain decisions and information sharing ultimately rely on input from people
embedded in the supply chain. Hence, one cannot talk about improving supply chain
decisions without understanding the “people” element of the supply chain. In con-
texts where a person’s action may impact the well-being of another—information
exchange being one example—an important behavioral aspect to keep in mind is
that people’s decisions and behavior are constantly influenced by the values and
norms they adhere to. In this section, we consider the personal values and norms
that pertain to trusting and trustworthy behavior.
Trust involves accepting risks or vulnerabilities in at least three distinct dimen-
sions: the risk of being worse off than not trusting, the risk of being worse off than
the trusted party, and the risk of being betrayed by the trusted party. The first dimen-
sion of risk is related to a person’s attitude toward natural risks, that is, risks due to
natural randomness (e.g., whether it will be a snowy winter). It is natural to expect
that the more risk averse a person is, the less likely he/she is to trust. The latter two
risks can be termed inequality risk and betrayal risk respectively (e.g., Hong and
Bohnet 2007) and can be considered as social risks, that is, risks created by another
person’s potentially self-interested actions.
292 Ö. Özer and Y. Zheng

One may wonder how we can tease apart the roles of these different risks in
determining an individual’s tendency to trust. Here is one way to disentangle the
effect of each using three simple but concrete contexts. Consider the three invest-
ment scenarios shown in Fig. 14.2: (a) retailer-specific investment, (b) supply chain
investment, and (c) supplier-specific investment.3 In retailer-specific investment, the
supply chain begins with a profit of $20. The supplier first decides whether to simply
split the current profit with the retailer or to further invest all profit in the retailer in
hopes of additional returns. If the supplier chooses not to invest, each party receives
$10. Conversely, if the supplier chooses to invest, the supply chain will gain a 50 %
return for $30 in total profit. However, now it is the retailer who has the right to de-
cide how to split this $30, either splitting it equally resulting in $15 for each party,
or exploiting the supplier by keeping $22 for itself and leaving $8 for the supplier.
Notice that by choosing to invest, the supplier faces all three types of risk: it may
be worse off than not investing, it may be worse off than the retailer, and it may
be betrayed by the retailer (if the retailer chooses to keep most of the final profit).
Therefore, the supplier’s willingness to invest indicates its willingness to trust the
retailer to equally split the final profit. Thus, to measure its willingness to trust, we
can ask the supplier the following question: “If the retailer will choose to equally
split the final profit with some probability, what is the minimum value of this prob-
ability such that you will prefer to invest?” This minimum acceptable probability
(MAP) measures the supplier’s aversion to the combined risks: the higher the MAP,
the less willing the supplier is to accept these risks, and hence, the less willing the
supplier is to trust/invest.

Fig. 14.2 Three investment scenarios to disentangle betrayal risk, inequality risk, and natural risk.
(a) Retailer-specific investment, (b) Supply chain investment, (c) Supplier-specific investment

Next, consider supply chain investment (Fig. 14.2b). Here we still have a supplier
interacting with a retailer based on an identical payoff structure as in retailer-specific
investment. The key difference is that now if the supplier chooses to invest, it is the
uncertainty in the invested market that will determine the share of the final profit
3 These scenarios are adapted from the Trust Game, the Risky Dictator Game, and the Decision
Problem originally designed by Bohnet and Zeckhauser (2004). We modify the terminologies from
the original design to fit into a supply chain context.
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 293

received by each party. Two possible outcomes may occur: (1) each party receives
$15, or (2) the supplier receives $8 while the retailer receives $22. Compared to
retailer-specific investment, in this case trusting/investing requires the supplier to
tolerate natural risk (uncertainty in the invested market) and inequality risk (being
worse off than the retailer), but betrayal risk is no longer relevant. Again, we can
ask the supplier what is the MAP of the ($15, $15) outcome being realized such
that it will prefer to invest. If the MAP for supply chain investment turns out to be
lower than the MAP for retailer-specific investment, then the difference in MAP will
measure the supplier’s aversion to betrayal risk.
Finally, consider supplier-specific investment (Fig. 14.2c). In this scenario, the
supplier’s investment decision impacts only its own profit. If the supplier chooses
not to invest, it receives $10. If the supplier chooses to invest, then the investment
outcome is uncertain and can be either $15 or $8. In this setting, the supplier only
faces natural risk (uncertainty in the market). Hence, the supplier’s aversion to nat-
ural risk can be elicited by asking the supplier the MAP of the $15 outcome being
realized such that it will prefer to invest. Furthermore, if aversion to inequality risk is
capable of impacting willingness to trust/invest, then we would expect the MAP for
supplier-specific investment to be lower than the MAP for supply chain investment.
Through a series of experiments (Hong and Bohnet 2007; Bohnet et al. 2008),
researchers have arrived at a few important conclusions. First, aversion to betrayal
risk is a rather universal phenomenon. It exists for both men and women, as well as
in at least six countries: the United States, Oman, Turkey, Switzerland, Brazil, and
China. Second, comparing the impact of betrayal risk with that of inequality risk,
women and young adults are more concerned about inequality risk whereas men
and middle-aged people are more concerned about betrayal risk. In addition, people
considered to have more power or control in social relations pay more attention to
betrayal risk, while people with less power are more worried about inequality risk.
These findings should provoke some careful thinking about how a company should
go about building trust with its business partners. For example, if your business part-
ner is the weaker one in terms of power (e.g., you are a multinational conglomerate
and your partner is a local manufacturer in a developing country), you should not
squeeze your partner’s margin too much, as this strategy will very likely discour-
age your partner’s trust. Instead, you should pursue a more equitable distribution of
profits, as the above results suggest that a strategy of equitable play is more critical
for inducing trust in a supply chain with unbalanced power than one where power
is balanced (in which case the focus should be on reducing betrayal risk). Similarly,
in a meeting with a procurement manager from a retailer and a sales manager from
a supplier, paying attention to the manager’s gender may not be a bad idea!
Turning to the other side of the equation, what motivates trustworthy behav-
ior, particularly for inducing truthful information exchange when there is room
for deception? This is equally important to understand because trust goes hand
in hand with trustworthiness. We cannot speak of one without the other. Think
about the retailer-specific investment (Fig. 14.2a) discussed above. Without a suf-
ficiently strong expectation that the retailer will be trustworthy and equally split the
final profit, there are no grounds for the supplier to trust. Similarly, in supply chain
294 Ö. Özer and Y. Zheng

information exchange, there is no reason to trust the shared information if the party
that owns the information is believed to be not trustworthy. Various norms and val-
ues contribute to trustworthy behavior, such as altruism (pure selflessness), warm
glow (positive feelings when helping others), inequality aversion (dislike for unfair
distribution of wealth), guilt (negative feelings when disappointing others), or sim-
ply lie aversion (distaste for lying; e.g., Ashraf et al. 2006; Attanasi et al. 2016). It
is established that people experience a moral cost for lying. This cost is influenced
by how much a person gains from lying relative to how much the person being de-
ceived is hurt by the lie (Gneezy 2005). Thus, the more a person gains from lying,
the more tempted he/she will be to lie. However, the counterforce also prevails: The
more the other party will be hurt by the lie, the less likely it is that the person will
lie. Interestingly, even when lying helps the other party and does not hurt anyone,
i.e., a white lie such as “You look great today!”, a significant portion of people still
prefer not to lie (Erat and Gneezy 2012). This good human nature of lie aversion is
good news. It implies that businesses, managed and organized by humans, need not
be paranoid about their partners exploiting them at every opportunity. It is indeed
worthwhile for companies to invest in trust towards their partners, because there
is evidence that demonstrating a clear intention to trust encourages more trustwor-
thy behavior from the trusted party (Ho and Weigelt 2005). Importantly, companies
should also learn to be circumspect about when to trust and rely on trustworthiness.
In the next section, we discuss how trusting and trustworthy behavior are influenced
by the environment in which the supply chain members operate.

14.3 Market Environment

Product and market characteristics are important environmental factors that impact
trust and trustworthiness in supply chain information exchange. These exogenous
factors influence behavior primarily through their effects on the perceived vulnera-
bility that trusting entails and the extent of social uncertainty that exists.
Consider the supply chain setting described earlier, in which the supplier seeks
private forecast information from the retailer to help it make capacity investment
decisions. In this setting, the supplier usually bears most, if not all, of the investment
cost. Conversely, the retailer is motivated by a desire to meet demand and would like
to secure more supply from the supplier. By trusting the retailer’s possibly inflated
forecast, the supplier may overinvest, exposing itself to financial vulnerability from
having too much capital tied up in excess capacity. The higher the investment cost,
the more costly it is to have excess capacity, and hence the more vulnerability the
supplier faces. As a result, when investment costs are higher, the supplier will be less
willing to trust and more conservative in making capacity investments. In return,
the retailer will have more reason to inflate its private forecast to ensure sufficient
supply. Therefore, credible information sharing is harder to achieve under higher
investment costs.
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 295

Another important factor to consider is the demand volatility of the product.

Companies can use forecasting methods to capture systematic patterns in market
demand, such as seasonality and trends. However, there always exist unsystem-
atic fluctuations that are not predictable. These unsystematic fluctuations consti-
tute the demand volatility and are often measured by the distribution of forecast
errors. When operating in an environment where the demand volatility is high (i.e.,
there is a greater distribution of forecast errors), managers are likely to be more cir-
cumspect about trusting the shared forecasts from their partners as a result of social
uncertainty. According to the sociology literature, social uncertainty exists for an in-
dividual when (1) his/her partner in an interaction has an incentive to take an action
that will hurt the individual, and (2) the individual cannot predict whether his/her
partner will indeed take such an action (e.g., Yamagishi et al. 1998a). The existence
of social uncertainty discourages trust. In our supply chain setting, higher volatil-
ity in product demand gives the retailer a stronger incentive to inflate its private
forecast because the retailer is more likely to face a demand hike and supply short-
age. Hence, higher demand volatility leads to higher social uncertainty in the supply
chain, reducing trust, and rendering credible information sharing less possible.
A group of researchers have designed and conducted a series of experiments to
examine how investment cost and demand volatility jointly impact the extent of
trust, trustworthiness, and information sharing in the above supplier-retailer supply
chain (Özer et al. 2011). They replicate the supply chain setup in a computer envi-
ronment and recruit undergraduate and graduate students from Stanford University
and the University of Texas at Dallas to participate in the experiment.4 Participants
make decisions as suppliers or retailers in the simulated supply chains, and are pro-
vided performance-based cash earnings to encourage careful decision-making. Let
us take a short digression here to establish the validity of using student participants
to understand manager behavior. First, experienced professionals often exhibit sim-
ilar behavioral biases as students (e.g., Cooper et al. 1999; Croson and Donohue
2006; Bolton et al. 2012; Massey and Thaler 2013). Moreover, today’s students are
tomorrow’s managers. Second, on an aggregate level, trust and trustworthiness are
fundamental behavioral factors that are found to be stable across different adult age
groups (Sutter and Kocher 2007). Third, Choi et al. (2015) replicate the same exp-
eriment by Özer et al. (2011) with high-ranking executives who have on average 17
years of professional experience and observe very similar results. Therefore, per-
forming experiments using student participants is a highly effective way to gain
insights into the behaviors of supply chain professionals.
The data collected in the above experiments show that investment cost has a
dominant effect on trust, trustworthiness, and information sharing. In particular,
when the investment cost is low, retailers are more trustworthy in the information
exchange, and suppliers in turn trust the shared information more. This is true
regardless of whether demand volatility is high or low. Conversely, when the

4 Readers who are interested in using the forecast information sharing game in their own organi-
zations are encouraged to visit the game website: forecastsharing.com. The authors have provided
easy-to-follow instructions on the website to explain how one can set up and run the game to
illustrate when and how trust and the environment interact to affect business decisions.
296 Ö. Özer and Y. Zheng

Fig. 14.3 Forecast information sharing strategies. (Notes: A high/low investment cost corresponds
to a low/high profit margin. A “simple” strategy relies on natural trust and trustworthiness to mo-
tivate credible information sharing. A “complex” strategy involves a complex contract or a trust-
building process or both between the supply chain parties.)

investment cost is high, higher demand volatility does indeed lead to significantly
less trusting and trustworthy behavior, thereby preventing information being cred-
ibly shared in the supply chain. We have repeatedly observed similar results when
the forecast information sharing game was used as a teaching tool at various MBA
programs (e.g., MIT, Columbia University, Stanford University, and University of
California, San Diego).
The above behavioral principles can help to improve forecast information
management. Figure 14.3 provides some example products according to the two
characteristics discussed: investment cost and demand volatility. Ink cartridges, for
example, are considered to have low investment cost and low demand volatility.
The low investment cost (or high profit margin) is due to the maturity of the tech-
nology and a highly standardized production process. Their demand is relatively
stable because ink cartridges are specialized for certain types of printers and have
predictable usage rates. Laptop computers, by contrast, are associated with a high
investment cost and high demand volatility due to frequent technological innova-
tion and severe market competition. Home appliances, such as microwaves and
washers/dryers, typically involve a high investment cost and low demand volatil-
ity. Finally, movie DVDs are examples of products with a low investment cost but
high demand volatility, because consumer preferences for these products are hard to
predict.
Figure 14.3 positions these four product categories within two information
sharing strategies: (1) a “simple” strategy in which the parties rely on each other’s
natural propensity to trust and be trustworthy to promote credible information shar-
ing; (2) a “complex” strategy which involves a complex contract, a trust-building
process, or both to improve information sharing (e.g., Ebrahim-Khanjari et al. 2012).
Recall from Sect. 14.1 that researchers have extensively studied the roles of con-
tracts and repeated interactions in ensuring credible information sharing in a supply
chain. A complex strategy will utilize these approaches when natural propensities
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 297

for trust and trustworthiness in the supply chain are low. Thus, understanding market
environment as a building block of trust and trustworthiness allows us to determine
the boundary of the simple strategy and the conditions under which the complex
strategy may be necessary.
The shaded area in Fig. 14.3 represents products for which the simple strategy
may be sufficient, while the non-shaded area represents those for which the com-
plex strategy may be better. When the investment cost is low and hence vulnerability
from trusting is low, the propensity to trust is high for all levels of demand volatil-
ity. In this situation, trust naturally induces cooperation and the simple strategy is
effective (e.g., for products like ink cartridges). A similar dynamic may explain why
we observe catastrophic failures in some industries (such as the telecommunication,
networking, and computer industries) because of a lack of credible information shar-
ing, whereas in others we do not.
The above findings also shed important light on how managers may induce more
trust and trustworthiness in their supply chains. If a manager hears from the sup-
plier or customer a statement such as “We don’t trust you in . . . [doing something]”
or “It is difficult to trust,” then the manager should investigate what vulnerabili-
ties and uncertainties are causing those “feelings.” Once barriers to trust have been
identified, the manager can figure out how to reduce or eliminate those barriers.
For example, risk-sharing agreements could be used to address vulnerabilities. To
establish such agreements, supply chain partners can divide and share costly invest-
ments according to which investments are customer-specific and which ones can
be distributed across multiple customers. A case in point is the way Ericsson (cus-
tomer) and Phillips Electronics (supplier) share investments into expensive equip-
ment in their supply chain. In their risk-sharing arrangement, Ericsson invests in
a specialized testing equipment that is used only for Ericsson products at Phillips
Electronics’ site. In return, Phillips Electronics agrees to invest in general standard
machinery that can be used for products of various customers, including Ericsson.
These shared investments help to cultivate more trusting and trustworthy relation-
ships within the supply chain.
On the supplier side, engaging in relationship-specific investments even before
any formal contracts are signed is an effective way to signal trustworthiness, and
often leading to a more collaborative supply chain (Beer et al. 2015). For example,
Kayene, an Argentinian apparel manufacturer, hired a special quality assurance team
dedicated to a particular retailer prior to having any written agreement with the
retailer. Cosmax, a leading general beauty original design manufacturer, invested
in special equipment recommended only by L’Oreal before it had a contract with
L’Oreal. In each case, the relationship-specific investment consequently led to more
collaborative interactions between the parties. These supplier-initiated, relationship-
specific investments are informative signals that help buying firms determine which
supplier they can trust. Indeed, many companies, including Delphi, Verizon, AT&T,
and Costco, explicitly encourage such supplier initiatives that are above and beyond
the minimum necessary business requirements. In this way, these companies are able
to identify trustworthy suppliers with which they can develop long-term strategic
partnerships.
298 Ö. Özer and Y. Zheng

14.4 Business Infrastructure

In this section, we proceed to discuss how the “infrastructure” of the business

society influences the extent of trust and trustworthiness in supply chain informa-
tion exchange. We consider business infrastructure to be the culture, institutions, and
networks within which business organizations operate and interact. As our society
becomes more and more interconnected, such infrastructure plays an increasingly
important role in shaping business interactions.

14.4.1 Culture and Institutions

Today’s supply chains often span across different continents and countries. Hence,
they frequently encompass diverse cultures and institutions. Cultures in western
countries are typically characterized by individualism that values independence and
self-reliance. People in individualistic cultures are driven by their own goals and
desires, and maintain that the interests of the individual surpass those of the state.
In contrast, eastern cultures are generally oriented towards collectivism that emp-
hasizes the significance of social groups and the importance of cohesion within
them. When a global supply chain consists of companies from both individualistic
and collectivist countries, the clash in cultural orientations adds further challenges
to the effective management of the supply chain. Thus, understanding cultural dif-
ferences, including how trust manifests itself in different cultures, is increasingly
critical in supply chain management.
An important case in point is the way the distinctive culture of China interacts
with cultures of the western world, e.g., the United States. A few reasons make
China–U.S. supply chains an important context to consider. First, as the world’s
largest manufacturer, China sits at the crossroads of the majority of global supply
chains. Second, the trade volume between China and the U.S. is among the highest
for all country pairs in the world. Third, these two countries are situated at the two
far ends of the cultural spectrum, China being a country with a collectivist culture
rooted in thousands of years of history, and the U.S. being an individualistic coun-
try since its inception (Hofstede 1980, 2001). When U.S. companies ignore China’s
distinct culture but simply replicate their U.S. business models, they often encounter
bitter failures in the Chinese market. A well-known example is the loss of eBay to
a local Chinese competitor, taobao.com. eBay’s loss is often attributed to a failure
to understand local consumers by eBay’s management team. For instance, eBay did
not allow online chats between buyers and sellers due to concerns that they might
close their transactions offline to circumvent fees. In contrast, taobao.com allowed
such conversations to help cultivate trust among the trading parties (Barboza and
Stone 2010; Wang 2010). As another example, executives from Quantum Corpo-
ration noted that its common practice of rotating management positions within the
company created a big hurdle to building a trusting relationship with its Japanese
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 299

partner (Hausman 2011). This is because Japanese culture, like Chinese culture,
is relationship-oriented and relies heavily on long-term contact with the same ind-
ividual to develop trusting relationships. Similar issues can arise even within an
organization. For example, senior executives at Hitachi in Japan pointed out dif-
ficulties in establishing relationships with managers and business analysts at Hi-
tachi’s U.S. headquarters because the U.S. management team rotates positions more
frequently than its counterpart in Japan (Dyer 2011). These and many other exam-
ples highlight the importance of understanding distinctive cultural and institutional
characteristics in promoting efficient operations and supply chain management.
How do different cultural expectations, e.g., under collectivism versus individu-
alism, influence trust? Social psychologists posit that the average propensity to trust
and to be trustworthy is lower in collectivist societies due to a strong in-group bias;
that is, collectivists exhibit high levels of trust and trustworthiness only among their
in-groups, i.e., parties to whom an individual is related through kinship or long-term
social ties. In contrast, these cultures tend to treat out-group members with a mix
of suspicion and opportunism (Triandis et al. 1988; Fukuyama 1995; Child 1998;
Yamagishi et al. 1998b; Chen et al. 2002). Conversely, individualists do not emb-
race a strong affinity for social groups and hence are less subject to the in-group
bias. For example, the World Values Survey (2009) shows that over 80 % of both
Chinese and U.S. individuals report that they trust people they know well, but only
11 % of Chinese (versus 40 % of U.S.) individuals said they would trust people that
they just met for the first time.
The dynamic discussed above suggests two important predictions. For one, when
there is no prospect for long-term relationships, we should expect that sponta-
neous trust and trustworthiness will be less likely to arise in a collectivist country
(e.g., China) than in an individualistic one (e.g., the U.S.) due to the strength of the
in-group bias. Yet, this strong in-group bias also implies that trust and trustworthi-
ness will be higher among individuals with long-term relationships in collectivist
societies (see Bohnet et al. 2010, 2012, for evidence of similar phenomena among
countries in the Gulf of Mexico and the Middle East). Another important reason
why long-term relationships are necessary to induce trust in China is the relatively
weak legal institutions in the country. In the U.S., there is a stronger legal system
that protects individuals’ rights. U.S. individuals tend to believe that their legal sys-
tem will take care of untrustworthy or corrupt actors. Hence, they tend to be trust-
ing from the get go because being untrustworthy or corrupt could be very costly.
However, such a legal system does not exist in China. Most people do not expect
justice and lose hope in the system. This loss of hope, combined with the excessive
greed of wanting too much of everything (an attitude that was itself encouraged in
China by the western world through “buy more and live large” advertising over the
past two decades), causes widespread untrustworthy and corrupt behavior in China.
As a result, trusting others entails much higher risk and vulnerability in China than
in the U.S. Therefore, to instigate trust in China, it becomes necessary to establish
long-term relationships and solid commitments.
300 Ö. Özer and Y. Zheng

A research team who conducted a forecast information sharing experiment with

both Chinese and U.S. participants has found support for the above two predictions
(Özer et al. 2014). When participants engage in single interactions (i.e., they do
not form a long-term relationship), Chinese retailers inflate their forecasts more and
Chinese suppliers invest in much less capacity; i.e., the Chinese participants are less
trustworthy and less trusting than their U.S. counterparts. In sharp contrast, when
the participants engage in repeated interactions, Chinese and U.S. participants begin
with the same levels of trust and trustworthiness. Importantly, while trust gradually
declines among the U.S. participants as some suppliers realize their retailers have
inflated the forecasts, Chinese suppliers are more tolerant of minor information ma-
nipulation by Chinese retailers and maintain high levels of trust towards their supply
chain partners. The better-sustained trust among the repeatedly-interacting Chinese
participants may be due to the important norm of guanxi that governs business inter-
actions in Chinese society. Since establishing long-term interactions is a prerequi-
site for building guanxi, engaging in repeated interactions likely induces the guanxi
norm, providing behavioral guidance for the Chinese participants. This norm sub-
stantially motivates trust and credible information sharing in Chinese supply chains.
The research team has also observed interesting phenomenon in experiments in-
volving a cross-country supply chain with a Chinese supplier and a U.S. retailer.
They have found that Chinese suppliers show significantly more trust in informa-
tion shared by U.S. retailers than information from Chinese ones. This higher level
of trust can primarily be ascribed to the positive perception among Chinese partici-
pants that westerners are more fair and trustworthy. This observation coincides with
widespread western favoritism in Chinese society. Since the early 1980s, increasing
media exposure and penetration of foreign brands into China has helped to shape a
positive western stereotype among the Chinese population (Batra et al. 2000; Zhou
and Hui 2003). Chinese individuals tend to regard westerners as more open-minded,
reliable, and honest, and having higher moral grounds (Bond 1986; Willnat et al.
1997; Huang 2011). These perceptions will likely lead to Chinese companies to be
more willing to cooperate with U.S. partners than with Chinese ones.
What can western companies learn from these results? Two mechanisms stand
out as excellent vehicles for motivating trust from potential Chinese business part-
ners. When the prospect of long-term interactions is less certain (e.g., when initi-
ating new business relationships with Chinese companies), western companies can
rely on and reinforce the Chinese population’s positive perceptions of westerners as
generally more trusting and trustworthy to encourage higher trust among their Chi-
nese partners. Possible actions include ensuring that products sold in the Chinese
market are of the same quality as those in western markets, and establishing living
and working conditions for Chinese workers in offshore factories that comply with
western standards. These actions would demonstrate the company’s commitment to
protecting the rights of trusting and trustworthy individuals, thus making Chinese
partners believe that being trusting and trustworthy will pay off. Moreover, when
long-term engagement is more likely (e.g., when initial transactions with a Chinese
partner signal the profitability of a longer-term partnership), western companies
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 301

should secure a long-term relationship (e.g., by signing long-term contracts and

appointing a fixed management contact) to induce the guanxi norm and motivate
trust from the Chinese partner.
In turn, Chinese and, more broadly, all Asian companies should proactively
cultivate an open and cooperative mindset among the younger generation and estab-
lish an environment conducive to trust. For example, more and more Asian compa-
nies are sponsoring their mid-rank managers to attend schools in the western world
for professional training, offering these managers a great opportunity to become
more open-minded to new business partners. Some large companies in China have
also started to cultivate a sense of community within their organizations, hoping
to transform the competition-oriented mindset of young professionals into a more
collaboration-oriented one. In addition, it is also critical for organizations, indus-
tries, and governments to develop, and enforce, reward/penalty systems that encour-
age trustworthy behaviors and punish corruption indiscriminately.

14.4.2 Social Networks

Another important factor within the business infrastructure is the social networks
that managers have been and are currently embedded in. The experiential view of
trust stipulates that a person’s disposition for trust is gradually formed through prior
experiences (Brehm and Rahn 1997; Hardin 2002). In parallel, sociologists pos-
tulate that networks of personal relationships play an important role in generating
trust and influencing economic activities. As Granovetter (1985) puts it, “continu-
ing economic relations often become overlaid with social content that carries strong
expectations of trust and abstention from opportunism” (p. 490). Integrating both
perspectives, the extent of trusting behavior in supply chain information exchange
is likely influenced by two types of trust propensity: (1) preconditioned trust, i.e., the
trust that supply chain managers have, in general, for business partners in their past
professional relationships, and (2) network trust, i.e., the expectation and strength
of mutual trust and trustworthiness that exist between the managers in the current
supply chain.
When both preconditioned trust and network trust are high in a supply chain,
it is natural that credible information sharing and cooperation will occur. Less ob-
vious is what will happen when one type of trust is high and the other is low. The
experiential view predicts that preconditioned trust will dominate over network trust
because trust is perceived as a stable, gradually formed psychological state that is
robust to short-term perturbations (e.g., encountering a partner with a low network
trust in a particular transaction). This view suggests that as long as the managers’
preconditioned trust is high, network trust does not play a significant role in shap-
ing behavior. Network trust matters only when the managers’ preconditioned trust
is low. Conversely, one can also consider a manager’s level of preconditioned trust
reflects his/her faith in social capital. Managers with high preconditioned trust have
302 Ö. Özer and Y. Zheng

had successful experiences of trusting their business partners. Hence, they recognize
the value of trust in businesses and are mindful of social capital. These managers
are thus more likely to shape their behavior according to how much trust and trust-
worthiness they expect from their current partners (i.e., network trust). In contrast,
managers with low preconditioned trust do not believe in the value of trust and
social capital. Thus, they are not willing to rely on network trust, but rather, refer to
environmental factors (e.g., whether vulnerability and/or uncertainty is high or low)
to determine how much they should trust.
A recent study examines how preconditioned trust and network trust affect trust-
ing behaviors in supply chain information exchange among a group of experienced,
high-ranking executives (Choi et al. 2015). These executives have, on average, 17
years of business experiences, and over half of them hold positions at the C-level
or above. The study design combines well-established network surveys from the
psychology and organization literatures (e.g., McAllister 1995; Levin and Cross
2004) with the forecast information sharing game (Özer et al. 2011). The network
surveys elicit the executives’ preconditioned trust and network trust levels while the
forecast information sharing game is used to examine trusting behavior in a concrete
information exchange context.
The first question one may ask is, do executives trust at all? Many people suspect
that as managers gain more experience in the business world and constantly make
decisions with high stakes, monetary objectives start to suppress moral, behavioral
motives and become the managers’ primary focus. Hence, they will only trust if
trusting is aligned with their pecuniary incentives. However, the executives in the
above study are shown to exhibit significant levels of trust and trustworthiness. In
addition, they seem to be more mindful of what factors in the supply chain environ-
ment can influence trust and trustworthiness, and hence more intelligent in deter-
mining when to trust. This observation is supported by evidence that their trusting
behavior is jointly shaped by preconditioned trust, network trust, and the environ-
ment in a nontrivial way.
When preconditioned trust and network trust are considered individually, each
is positively correlated with trusting behavior in the game. That is, an executive
with a higher preconditioned trust or higher network trust is also more trusting of
his/her partner in the game, relying more on the information shared to make in-
vestment decisions. Importantly, the data support the viewpoint that preconditioned
trust reflects how much a manager is willing to rely on social capital. In particular,
network trust plays a significant role in shaping behavior only for those executives
with high preconditioned trust to begin with. For those executives with low precon-
ditioned trust, network trust has no effect on behavior. Instead, their willingness to
trust their partners increases only through repeated interactions. These results shed
light on how social networks and environmental factors jointly influence trusting
relationships in business interactions.
One implication of the above results is that companies need to be mindful of
how to position managers in the various intra- and inter-organizational networks
that exist. When developing an external business network (e.g., a supply chain)
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 303

in an information-critical setting, establishing a mutually trusting and trustworthy

relationship with one’s partner is crucial. Because managers with high precondi-
tioned trust believe in the value of social capital, they are effective initiators of such
relationships. In addition, they are the “smart” ones who will leverage trust when
possible (as opposed to managers with low preconditioned trust, who rely more on
the external environment when making decisions). Hence, managers with high pre-
conditioned trust may be better positioned to handle new business partners. This is
particularly true when the supply chain environment is characterized by high levels
of vulnerability and uncertainty, and hence, trust in the relationship will need to be
consciously cultivated. For managers that begin with low preconditioned trust, how-
ever, it may be wise to inject them into an intra-organizational network with coher-
ent incentives and little rotation. Such a stable and coherent network can help these
managers understand how the building blocks discussed in this chapter (Fig. 14.1)
can influence the roles of trust and trustworthiness in producing mutually beneficial
business outcomes, thereby increasing their preconditioned trust in the long run.

14.5 Business Process Design

The previous sections have focused on how environmental factors that companies
cannot easily change influence trust and trustworthiness in supply chain informa-
tion exchange. In this section, we turn to factors that companies are more able to
control and adjust: the design of relevant business processes. We will discuss three
examples. The first example contrasts how different processes for eliciting private
information from a channel partner impact trust. The second example focuses on
structuring team discussions to improve information sharing in team-based decision
making. The third example addresses how to design an effective feedback system
that promotes trust and the truthful sharing of reputation information.

14.5.1 Information Exchange, Advice, or Delegation

The business process for sharing private information with one’s supply chain part-
ners can take multiple forms. Consider a manufacturer-retailer supply chain setting
in which a retailer needs to make service-level decisions such as assortment selec-
tion, shelf space allocation, and promotion planning. The manufacturer, having more
insights into its own products, may thus have better information about their market
potentials as well. Hence, the retailer would like to solicit this private information to
make better service decisions. However, the manufacturer may want to manipulate
the information in order to get more attention and investment from the retailer, or
to hide the information to prevent the retailer from identifying products with a low
market potential.
304 Ö. Özer and Y. Zheng

In this setting, there are three widely-used processes of information exchange.

The first process is direct information exchange, the focus of our earlier discussions.
Under this process, the manufacturer directly communicates its private information
to the retailer, letting the retailer make its own service decisions. The second pro-
cess is for the manufacturer to offer advice on relevant service decisions without
disclosing its private information to the retailer. For example, Carrefour, a lead-
ing France-based global retailer, receives advice from Colgate-Palmolive regarding
shelf design and consumer education initiatives for oral care products. Based on
this advice, Carrefour determines the final shelf layouts and promotional activities
in its stores (ECR Europe Conference 2004). The third process is for the retailer
to delegate the service decisions to the manufacturer, hoping that the manufacturer
will leverage its private information to make better decisions for the supply chain.
For example, the retailer may decide on the maximum assortment size it can carry
and the maximum shelf space it can allocate. The manufacturer then determines the
actual assortment and the shelf space allocated to each product. Delegation is com-
monly used in categories such as magazines and kitchen gadgets (Federal Trade
Commission 2001).
How do trust and the efficacy of information sharing vary, if at all, across these
different processes? To answer this question, we first need to understand the fun-
damental ways in which these processes differ. First, note that direct information
exchange differs from offering advice in terms of the type of information processing
the manufacturer needs to perform. Direct information exchange involves descrip-
tive information processing; i.e., the manufacturer’s job is simply to convey infor-
mation. By contrast, offering advice involves prescriptive information processing, in
that the manufacturer must formulate a service decision by considering the resulting
profits and recommend a decision to the retailer. How do descriptive and prescriptive
information processing impact trustworthiness and trust differently? When informa-
tion processing is descriptive, the manufacturer’s private information is an objective
benchmark. Any value deviating from the true information would be an obvious sign
of deception, making it hard for the manufacturer to ignore its own untrustworthy
behavior. However, there is no such salient benchmark for prescriptive information
processing. In this situation, the manufacturer also needs to evaluate the profit impli-
cations of different advice it could offer, thereby increasing the salience of monetary
objectives. As a result, the manufacturer may find it easier to justify untrustworthy
behavior when information processing is prescriptive. Thus, in light of the above
discussion, direct information exchange will likely lead to higher trust, and trust-
worthiness, as well as more effective information sharing in the supply chain, than
offering advice.
Next, let us turn to the delegation process. Delegation differs from both direct
information exchange and offering advice in that the manufacturer is given decision
rights under delegation; i.e., the manufacturer is the one responsible for making the
final decision. How the transfer of decision rights impacts trust and trustworthiness
can go both ways. On the one hand, if the manufacturer is entrusted with decision
rights, then it may be willing to reciprocate the retailer’s trust by being more trust-
worthy. On the other hand, the service limit (e.g., the maximum shelf space the
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 305

manufacturer can use) offered by the retailer under delegation may be taken by the
manufacturer as permission to exhaust the retailer’s resources up to that limit regard-
less of the information. Worrying about this possibility, the retailer may then set a
tight limit to reduce its vulnerability to potential opportunistic behavior from the
manufacturer. This tight limit would in turn send a signal of distrust, discouraging
the manufacturer from being trustworthy.
A group of researchers recently conducted a series of experiments with the above
supply chain setup to study how trust, trustworthiness, and information sharing be-
havior vary across these three different processes (Özer et al. 2016). The results
show that information sharing is most effective under direct information exchange
due to the high levels of trust and trustworthiness observed, whereas delegation
yields the least trust and trustworthiness. Thus, the data show behavioral evidence
that supports our earlier characterization of the contrast between direct information
exchange and offering advice. Interestingly, the results also show support for both
arguments about the delegation process. In particular, manufacturers who are en-
trusted by the retailers with a high service limit, do indeed repay the trust by being
trustworthy in their service decisions. However, the majority of the retailers are not
trusting, as is reflected by the low service limits they set to constrain the delegated
decision. Consequently, this lack of retailers’ trust leads the manufacturers to exploit
the delegation and set the service to a level closer to the maximum level allowed by
the retailer, regardless of the private demand information.
These observations have important implications for supply chain initiatives that
require one party to delegate design rights to another party, such as vendor managed
inventory (VMI). Although VMI is expected to increase supply chain efficiency by
reducing inventory costs and stock-outs, the actual results of this practice have been
mixed. Some researchers have suggested that much of the benefits associated with
VMI can be obtained through sharing information alone (e.g., Cachon and Fisher
1997). Others cite the downstream firm’s lack of trust in its supplier as an impor-
tant reason for poor performance under VMI (Blackhurst et al. 2006; Claassen et al.
2008; Brinkhoff et al. 2015). For instance, Blackhurst et al. (2006) find that “While
the companies have sought a collaborative relationship in the form of a VMI initia-
tive, there appears to be some evidence of lack of trust . . . [T]he downstream partner
is not allowing the partners to process the [demand] information or use their own
judgment relative to this information.” Similarly, based on interviews with several
purchasing managers, Claassen et al. (2008) report, “A surprising finding in this
qualitative exploration was that almost all buyers provided their supplier with unex-
pectedly tight upper and lower limits for the inventory level. This is at odds with the
theory of VMI, and significantly reduces the level of replenishment flexibility for
the supplier.” Consistent with these findings, results from the above study on infor-
mation sharing suggest that the retailer may not be delegating sufficient control to
the manufacturer in VMI arrangements. The results further suggest that the retailer
will be better served if it retains control of inventory decisions and simply asks the
manufacturer to share its demand information.
An important point to take away here is that, even though none of the processes
discussed above changes the pecuniary incentives for either party in the supply
306 Ö. Özer and Y. Zheng

chain, they do substantially impact the behavioral motives driving managers’ de-
cisions. Therefore, it would be advisable for companies to revisit and (re)design
current processes with an eye towards their impacts on trust and trustworthiness.

14.5.2 Process-Driven Information Exchange

Managers at a company often engage in internal discussions to collectively make

decisions. For example, a manufacturer may form a risk management team to assess
the risk of disruption to its upstream suppliers before making sourcing decisions.
Each member of the risk management team may be privately informed about the
disruption risk to the suppliers through his/her own personal connections. Ideally,
every member of the team would share their private information so that the team
could collectively make the best decisions for the supply chain. However, inducing
team members to share private information is often not an easy task (e.g., Guzzo
and Dickson 1996; Kerr and Tindale 2004; Lightle et al. 2009).
There are four main reasons why information sharing within a group is diffi-
cult to achieve when different members have different pieces of private information.
The first reason is known as “commitment bias,” meaning that people tend to defend
their own opinions or preferences, and hence will only share information that sup-
ports their own position. The second reason has to do with issues of status. When
a team is composed of people of different ranks on the organization ladder, lower-
rank members often find it risky to share information that goes against the opinions
of higher-rank members. The third reason is more generally related to building re-
lationships. Because conflicting information may create tensions between people
of different views, individuals who desire to develop personal networks within the
organization may refrain from sharing divergent information. The fourth reason is
the existence of potential incentive conflicts among team members, as well as be-
tween an individual and the organization. In these situations, people may strategi-
cally choose to reveal or withhold information according to their own best interests.
Researchers have extensively studied these four factors and suggested remedies
to correct them and improve group decisions. For the most part, their focus has been
on forming diverse teams and engineering the distribution of private and common
information among team members (see Stasser and Titus 2003, for a review). In
a recent study, researchers have taken a different approach, focusing on how pro-
cess design can be used to overcome commitment bias (Levi et al. 2015). They
consider two alternative designs. One design is the “structured discussion process.”
In this design, team members are guided through a structured discussion sheet on
which they are asked to write down all the risk information they know about each
supplier before the team proceeds to discuss sourcing recommendations. In the sec-
ond design, team members are advised to follow the structured discussion sheet
but are not required to write down any information. In other words, following the
discussion sheet is only a suggested framework for the group to discuss their infor-
mation/recommendations, but it is not an enforced process as is in the structured
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 307

discussion design. The researchers compare these two designs with a control con-
dition in which no process or advice is in place (i.e., team members are allowed to
freely discuss among themselves to make the final sourcing recommendation).
The results show that implementing the structured discussion process signifi-
cantly improves information sharing among team members and leads to the highest
proportion of teams choosing the best supplier. Advising teams to follow the struc-
tured process does improve information sharing and decision-making to some extent
but not as much as implementing the structured process itself. What is the secret of
the superior performance of the structured process? The answer seems to be that
the structured process encourages team members to both share and assimilate diver-
gent information before making the final decision. In contrast, while advising also
encourages information sharing, it does not facilitate information assimilation. Psy-
chologically, the structured process objectifies and decomposes the decision process
into an information collection stage and a decision stage. This decomposition helps
team members to become mentally and emotionally detached from the information
that supports their own opinions, thus facilitating both the sharing and assimila-
tion of different information. Extrapolating this finding to a supply chain setting
where the supply chain members have conflicting interests, it is likely that the struc-
tured process could be even more effective in improving information sharing. This
is because monetary objectives are salient in the decision stage, creating additional
strategic reasons for not sharing all information. In this case, isolating and objecti-
fying the information collection stage could help to reduce the salience of monetary
incentives and increase the chances that all information will be shared before the
members enter the decision stage.

14.5.3 Engineering Reputation Information

The availability of reputation information concerning past transactions often affects

one’s willingness to trust. It is natural that people are much more willing to trust
a business partner who has a good track record than another who in the past of-
ten failed to deliver on its promises (Doney and Cannon 1997). In addition, making
such reputation information available can be a useful tool for screening out deceitful
behavior and protecting the rights of trading partners. This approach is indeed crit-
ical for the prosperity of Internet trading, where the majority of transactions occur
between strangers (e.g., Resnick and Zeckhauser 2002; Resnick et al. 2006).
How should such a feedback system be designed to allow reliable reputation
information to be conveyed? One natural approach is to have an open, two-sided
system in which both companies provide feedback about their trading partners’
credibility and that feedback is publicly observable once it has been submitted. For
example, electronics manufacturers such as HP, Lenovo, Samsung, and Sony each
provide feedback about their transactions with electronics retailers such as BestBuy
and Micro Center, and vice versa. Indeed, this is the design most commonly used
in online marketplaces, such as eBay and Amazon.com. Can such an open system
308 Ö. Özer and Y. Zheng

deliver useful reputation information? If company A should report problems con-

cerning its transaction with company B, then how would company B respond? It is
likely that company B would look for opportunities to “retaliate” through its feed-
back about A. Given such concerns about retaliation, A would thus likely refrain
from leaving negative comments in the first place. Hence, reputation information in
such an open system is likely “inflated,” i.e., all companies have great reputations.
Researchers indeed find evidence of retaliatory feedback and inflated reputations in
online platforms using open feedback systems, such as eBay (Bolton et al. 2013).
One potential solution is to use a double-blind system in which a company
does not see its partner’s feedback about itself until it has also submitted feed-
back about its partner. However, a double-blind system may reduce feedback giv-
ing altogether, especially for mutually satisfactory transactions, because now giving
positive feedback about a partner cannot be used to motivate the partner to do the
same. An alternative design would be to supplement an open system with blinded
ratings for each company. Reputation scores from the open system would remain
publicly observable by all companies but the blinded ratings would be observable
only by trading partners; e.g., BestBuy never observes its own blinded rating, but
all electronics manufacturers who currently sell or want to sell their products in
BestBuy can observe BestBuy’s blinded rating. With such a design, companies still
have a good motivation to submit feedback. In the meantime, they are given a “safe”
place to truthfully rate their experiences. Therefore, this design would likely lead to
more informative, reliable reputation information that serves its intended purpose.
A group of researchers have compared this design with the open system and the
double-blind system in an eBay-like market in the laboratory (Bolton et al. 2013).
They find that the open system supplemented by blinded ratings does indeed signif-
icantly outperform the other designs in motivating trading partners both to submit
more truthful feedback and to rely more on the feedback when determining whether
or not to trust another party. As a result, when the open feedback system with blinded
rating is implemented, the same reputation score will lead to higher prices offered
by the buyers and higher quality offered by the sellers, thus improving the overall
efficiency of trades.

14.6 Conclusion

Trust is an important psychological and social element that supports the efficiency
of economic activities, especially in information-critical settings. Meanwhile, trust
is malleable and influenced by both environmental and procedural factors such as
individual values and norms, product and market characteristics, the institutions
and networks that form the business infrastructure, as well as intra- and inter-
organizational process designs. If supply chain managers are willing to invest in
and consciously leverage trust in a supply chain, then organizations can largely
reduce non-value-added activities that only serve to limit opportunistic behavior.
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 309

Consequently, companies can devote more valuable resources to making better

products, offering better service, and improving our lives and societies.
Building and managing trust can be challenging because it requires the trustor to
accept vulnerability and uncertainty. By synthesizing the broad spectrum of research
in this domain, we hope to shed light on when and how trust can be developed and
sustained to improve information sharing in a supply chain. We identify a number
of conditions under which trust will naturally arise. For example, when the potential
loss due to trusting is low, when managers have successful prior experience of trust-
ing, when a company appoints a fixed, long-term point of contact in its business
with Asian partners, or when information exchange is decomposed from decision
making, truthful information sharing is more likely to happen and supply chain ef-
ficiency is improved. We also suggest remedies to overcome barriers to trust and
information sharing when these conditions are absent. For example, a company in a
stronger power position within the supply chain can induce trust from its partner by
proposing more equitable sharing of the supply chain risks and profit. Offering risk-
sharing agreements (e.g., shared investments and advance purchase commitments)
can also help mitigate vulnerability and uncertainty. Finally, direct information ex-
change is often more effective at mitigating opportunistic behavior than other indi-
rect forms of soliciting information, such as seeking advice and transferring decision
rights.
To conclude, it may take a lot of effort to develop trust in a supply chain.
However, the return on investing in trust is often far beyond the cost. The princi-
ples discussed in this chapter concerning how to establish trust and trustworthiness
in a supply chain are relevant and applicable to a wide range of supply chain con-
texts, including information exchange and coordinating actions and decisions. Thus,
these principles have far-reaching implications on the efficiency of supply chains.
Now is the time to begin cultivating more trusting and trustworthy relationships in
your organization and your supply chains!

References

Arrow K (1972) Gifts and exchanges. Philos Public Aff 1(4):343–362

Ashraf N, Bohnet I, Piankov N (2006) Decomposing trust and trustworthiness. Exp Econ
9(3):193–208
Attanasi G, Battigalli P, Manzoni E (2016) Incomplete-information models of guilt aversion in the
trust game. Manag Sci 62(3): 648-667
Barboza D, Stone B (2010) China, where U.S. internet companies often fail. New York Times,
p B1
Batra R, Ramaswamy V, Alden D, Steenkamp J, Ramachander S (2000) Effects of brand local
and nonlocal origin on consumer attitudes in developing countries. Boston Univ Law Rev
9(2):83–95
Beer R, Ahn H, Leider S (2015) Can trustworthiness in a supply chain be signaled? Working paper,
Indiana University, Bloomington, IN
Blackhurst J, Craighead C, Handfield R (2006) Towards supply chain collaboration: an operations
audit of VMI initiatives in the electronics industry. Int J Integr Supply Manag 2(1/2):91–105
310 Ö. Özer and Y. Zheng

Bohnet I, Zeckhauser R (2004) Trust, risk and betrayal. J Econ Behav Organ 55(4):467–484
Bohnet I, Greig F, Herrmann B, Zeckhauser R (2008) Betrayal aversion: evidence from Brazil,
China, Oman, Switzerland, Turkey, and the United States. Am Econ Rev 98(1):294–310
Bohnet I, Herrmann B, Zeckhauser R (2010) Trust and the reference points for trustworthiness in
Gulf and Western countries. Q J Econ 125(2):811–828
Bohnet I, Herrmann B, Al-Ississ M, Robbett A, Al-Yahya K, Zeckhauser R (2012) The elasticity
of trust, Chap. 7. In: Kramer R, Pittinsky T (eds) Restoring trust in organizations and leaders:
enduring challenges and emerging answers. Oxford University Press, New York, pp 151–169
Bolton G, Ockenfels A, Thonemann U (2012) Managers and students as newsvendors. Manag Sci
58(12):2225–2233
Bolton G, Greiner B, Ockenfels A (2013) Engineering trust: reciprocity in the production of repu-
tation information. Manag Sci 59(2):265–285
Bond M (1986) Mutual stereotypes and the facilitation of interaction across cultural lines.
Int J Intercult Rel 10(3):259–276
Brehm J, Rahn W (1997) Individual-level evidence for the causes and consequences of social
capital. Am J Polit Sci 41(3):999–1023
Brinkhoff A, Özer Ö, Sargut G (2015) All you need is trust? An examination of inter-organizational
supply chain projects. Prod Oper Manag 24(2):181–200
Cachon G, Fisher M (1997) Campbell Soup’s continuous replenishment program: evaluation and
enhanced inventory decision rules. Prod Oper Manag 6(3):266–276
Cachon G, Lariviere M (2001) Contracting to assure supply: how to share demand forecasts in a
supply chain. Manag Sci 47(5):629–646
CDC White paper (2011) CDC TradeBeam global trade management overview,
October 17, retrieved at http://www.cdcsoftware.com/en/Resource-Library/Documents/
White-Papers-and-Articles/GL-GTM-GB-WP-GTMOverview-PM-09292011
Chen C, Peng M, Saparito P (2002) Individualism, collectivism, and opportunism: a cultural per-
spective on transaction cost economics. J Manag 28(4):567–583
Child J (1998) Trust and international strategic alliances: the case of Sino-foreign joint ventures. In:
Lane C, Bachmann R (eds) Trust within and between organizations. Oxford University Press,
Oxford, pp 241–272
Choi E, Özer Ö, Zheng Y (2015) Trust, social networks, and information sharing among executives.
Working paper, The University of Texas at Dallas, Dallas, TX
Claassen M, Weele Av, Raaij Ev (2008) Performance outcomes and success factors of vendor
managed inventory (VMI). Supply Chain Manag Int J 13(6): 406–414
Cohen M, Ho T, Ren Z, Terwiesch C (2003) Measuring imputed cost in the semiconductor equip-
ment supply chain. Manag Sci 49(12):1653–1670
Cooper D, Kagel J, Lo W, Gu Q (1999) Gaming against managers in incentive systems: experi-
mental results with Chinese students and Chinese managers. Am Econ Rev 89(4):781–904
Croson R, Donohue K (2006) Behavioral causes of the bullwhip effect and the observed value of
inventory information. Manag Sci 52(3):323–336
Dal Bó P (2005) Cooperation under the shadow of the future: experimental evidence from infinitely
repeated games. Am Econ Rev 95(5):1591–1604
Doney P, Cannon J (1997) An examination of the nature of trust in buyer-seller relationships.
J Mark 61(2):35–51
Dyer C (2011) Personal communications with Clint Dyer, former Chief Operations Officer, Hitachi
Global Storage Technologies
Ebrahim-Khanjari N, Hopp W, Iravani S (2012) Trust and information sharing in supply chains.
Prod Oper Manag 21(3):444–464
ECR Europe Conference (2004) Category management is here to stay. Presentation by Colgate–
Palmolive. ECR Conference, Brussels, Belgium
Erat S, Gneezy U (2012) White lies. Manag Sci 58(4):723–733
Federal Trade Commission (2001) Report on the Federal Trade Commission workshop on slotting
allowances and other marketing practices in the grocery industry. Available online through
www.ftc.gov. Last accessed June 27, 2014
14 Establishing Trust and Trustworthiness for Supply Chain Information Sharing 311

Fukuyama F (1995) Trust: the social virtues and the creation of prosperity, 28th edn. Free Press,
New York
Gneezy U (2005) Deception: the role of consequences. Am Econ Rev 95(1):384–394
Granovetter M (1985) Economic action and social structure: the problem of embeddedness. Am
J Sociol 91(3):481–510
Guzzo R, Dickson M (1996) Teams in organizations: recent research on performance and
effectiveness. Ann Rev Psychol 47:307–338
Hardin R (2002) Trust and trustworthiness. Russell Sage Foundation, New York
Hausman W (2011) Private conversation with Warren Hausman and Quantum executive
Ho T, Weigelt K (2005) Trust building among strangers. Manag Sci 51(4):519–530
Hofstede G (1980) Culture’s consequences: International differences in work-related values, 1st
edn. Sage Publications Inc, Newbury Park
Hofstede G (2001) Culture’s consequences: comparing values, behaviors, institutions and organi-
zations across nations, 2nd edn. Sage Publications, Inc, Thousand Oaks
Hong K, Bohnet I (2007) Status and distrust: the relevance of inequality and betrayal aversion.
J Econ Psychol 28(2):197–213
Huang X (2011) Stereotyping in any form is harmful. China Daily. Retrieved at http://www.
\penalty0chinadaily.\penalty0com.cn/opinion/2011-04/13/content 12319727.htm
Kerr N, Tindale R (2004) Group performance and decision making. Ann Rev Psychol
55(1):623–655
Knack S, Keefer P (1997) Does social capital have an economic payoff? A cross-country investi-
gation. Q J Econ 112(4):1251–1288
Kreps D, Milgrom P, Roberts J, Wilson R (1982) Rational cooperation in the finitely repeated
prisoner’s dilemma. J Econ Theory 27(2):245–252
Kumar N (1996) The power of trust in manufacturer-retailer relationships. Harvard Bus Rev
74(6):92–106
La Porta R, de Silanes FL, Shleifer A, Vishny R (1997) Trust in large organizations. Am Econ Rev
87(2):333–338
Lee H, Padmanabhan V, Whang S (1997) Information distortion in a supply chain: the bullwhip
effect. Manag Sci 43(4):546–558
Levi R, Wang S, Zheng Y (2015) Process-driven discussions in team-based decision making for op-
erational risk management. Working paper, Massachusetts Institute of Technology, Cambridge
Levin D, Cross R (2004) The strength of weak ties you can trust: the mediating role of trust in
effective knowledge transfer. Manag Sci 50(11):1477–1490
Lightle J, Kagel J, Arkes H (2009) Information exchange in group decision making: the hidden
profile problem reconsidered. Manag Sci 55(4):568–581
Massey C, Thaler R (2013) The loser’s curse: Decision making and market efficiency in the
National Football League draft. Manag Sci 59(7): 1479-1495
McAllister D (1995) Affect- and cognition-based trust as foundations for interpersonal cooperation
in organizations. Acad Manag J 38(1):24–59. http://www.jstor.org/stable/256727
Özer Ö, Wei W (2006) Strategic commitments for an optimal capacity decision under asymmetric
forecast information. Manag Sci 52(8):1239–1258
Özer Ö, Zheng Y, Chen K (2011) Trust in forecast information sharing. Manag Sci
57(6):1111–1137
Özer Ö, Zheng Y, Ren Y (2014) Trust, trustworthiness, and information sharing in supply chains
bridging China and the United States. Manag Sci 60(10):2435–2460
Özer Ö, Subramanian U, Wang Y (2016) Information sharing, advice provision, or delegation:
what leads to higher trust and trustworthiness? Manag Sci. Available at SSRN: http://ssrn.com/
abstract=2565577 or http://dx.doi.org/10.2139/ssrn.2565577
Radner R (1985) Repeated principal-agent games with discounting. Econometrica 53(5):
1173–1198
Ren Z, Cohen M, Ho T, Terwiesch C (2010) Information sharing in a long-term supply chain
relationship: the role of customer review strategy. Oper Res 58(1):81–93
312 Ö. Özer and Y. Zheng

Resnick P, Zeckhauser R (2002) Trust among strangers in Internet transactions: empirical anal-
ysis of eBay’s reputation system. In: Baye M (ed) Advances in applied microeconomics: the
economics of the internet & e-commerce, vol 11. Elsevier Science, Amsterdam, pp 127–157
Resnick P, Zeckhauser R, Swanson J (2006) The value of reputation on eBay: a controlled experi-
ment. Exper Econ 9(2):79–101
Rousseau D, Sitkin S, Burt R, Camerer C (1998) Not so different after all: a cross-discipline view
of trust. Acad Manag Rev 23(3):393–404
Stasser G, Titus W (2003) Hidden profiles: a brief history. Psychol Inq 14(3–4):304–313
Sutter M, Kocher M (2007) Trust and trustworthiness across different age groups. Games Econ
Beh 59(2):364–382
The Economist (2008) The power of information sharing: impact on product development. The
Economist Intelligence Unit (Oct 22). Retrieved at http://viewswire.eiu.com/\penalty0report
\penalty0dl.\penalty0asp\penalty0?mode=fi&fi=1593883744.PDF
The Economist (2012) Boeing – faster, faster, faster: the planemaker struggles to fulfill a rush of
orders. (Jan 28). Retrieved at http://www.economist.com/node/21543555
Triandis H, Bontempo R, Villareal M (1988) Individualism and collectivism: Cross-cultural per-
spectives on self-ingroup relationships. J Pers Soc Psychol 54(2):323–338
Wang H (2010) How eBay failed in China, excerpt from “The Chinese Dream: the rise of the
world’s largest middle class and what it means to you.” Bestseller Press, Brande, Denmark
Willnat L, Zhou H, Hao X (1997) Foreign media exposure and perceptions of Americans in Hong
Kong, Shenzhen, and Singapore. J Mass Commun Q 74(4):738–756
World Values Survey (2009) World Values Survey 1981–2008 Official aggregate v.20090901. Data
file, ASEP/JDS, Madrid. Retrieved at www.worldvaluessurvey.org
Yamagishi T, Cook K, Watabe M (1998a) Uncertainty, trust and commitment formation in the
United States and Japan. Am J Sociol 104(1):165–194
Yamagishi T, Jin N, Miller A (1998b) In-group bias and culture of collectivism. Asian J Soc
Psychol 1(3):315–328
Zak P, Knack S (2001) Trust and growth. Econ J 111(470):295–321
Zhou L, Hui M (2003) Symbolic value of foreign products in the People’s Republic of China. J Int
Mark 11(2):36–58
Zipkin P (2012) A reply to Williamson’s “Outsourcing . . . ”. Prod Oper Manag 21(3):465–469
Chapter 15
Information Leakage in Supply Chains

Guangwen Kong, Sampath Rajagopalan, and Hao Zhang

Abstract Information sharing within a supply chain has numerous benefits.

However, in the past decade, several works using game-theoretic models have
pointed out that: (1) a supply chain entity may not have an incentive to share inf-
ormation for fear of exploitation by the party (e.g., manufacturer) with whom they
share information as well as leakage of information to their competitors, (2) the neg-
ative effects of information leakage can be mitigated by using appropriate contracts
between supply chain entities. This chapter reviews this literature and provides a
framework for classifying it. The most common supply chain structure analyzed
in these works comprises of a manufacturer supplying a set of retailers who share
demand information with the manufacturer which may be leaked, either directly or
indirectly, to other retailers. The literature has shown that while vertical information
sharing with the manufacturer always has negative effects, the horizontal leakage of
information to other retailers can have positive or negative effects and the strength
of these effects depends on a number of factors. These include whether the compe-
tition among retailers is on price or quantity, whether their products are substitutes
or complements, whether information sharing arrangements are made before or af-
ter private information is revealed and the level of accuracy of private information
among different entities. To incentivize truthful information sharing despite the pot-
ential for leakage and its negative effects, the literature has come up with a variety

G. Kong ()
Industrial & Systems Engineering Department, University of Minnesota,
Minneapolis, MN 55455, USA
e-mail: [email protected]
S. Rajagopalan
Marshall School of Business, University of Southern California, Los Angeles, CA 90089, USA
e-mail: [email protected]
H. Zhang
Sauder School of Business, University of British Columbia, Vancouver, BC, Canada V6T 1Z2
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 313

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 15
314 G. Kong et al.

of solutions: side payments by manufacturers to retailers, different wholesale prices

charged to different retailers, revenue sharing contracts and market-based contracts,
etc. In addition, retailers can enter into a binding confidentiality agreement to pre-
vent leakage. Some of these solutions can coordinate the supply chain and some-
times benefit all the entities, including ones that do not have private information.
While anecdotal evidence from industry suggests that firms primarily fear direct
leakage of information to their competitors, two important insights from the aca-
demic literature are: (1) indirect leakage of information can have as significant an
effect on the incentives to share information as direct leakage, (2) information leak-
age to competitors can sometimes have positive effects.

15.1 Introduction

The information revolution has spawned a dramatic growth in the amount of in-
formation available to firms. The benefits of information sharing within a supply
chain have been well-documented in both the trade and academic literature over the
past two decades. The emergence of initiatives such as Vendor Managed Inventory
(VMI), Collaborative Planning, Forecasting and Replenishment (CPFR) has accel-
erated the trend towards sharing information to take advantage of these benefits. Lee
and Whang (2000) provide a nice summary of the benefits of information sharing
based on industry practices. However, they also point out that “information sharing
faces several hurdles” including loss of information rent and leakage of informa-
tion to unintended parties in the supply chain. Subsequently, several works have
pointed out using formal models that a supply chain entity may not have an incen-
tive to share information for fear of exploitation by the party (e.g. manufacturer)
with whom they share information as well as leakage of information to their com-
petitors. An example from Anand and Goyal (2009) illustrates the challenges faced
in sharing information given the possibility of information leakage.
Newbury Comics is a small, trendy chain in the Northeast that sells music
records. This retail chain is seen as a trendsetting retailer that is able to identify
early which artists are likely to break out and which records and type of music are
likely to be hot-sellers. SoundScan is a company that tracks music records sold by a
vast majority of the retailers in the US, including small ones like Newbury as well
as larger ones like Best Buy, and passes along this information to upstream record
labels such as Sony as well as middlemen such as Handleman Inc., who manage
the shelf inventories of large retailers such as Best Buy and Wal-Mart. Record la-
bels such as Sony found the sharing of information by the retailers valuable, espe-
cially given the hit-and-miss nature of this market and provided promotional and
co-advertising support to retailers to share this information. However, over time,
Newbury realized that it was losing its competitive edge in the market because the
valuable information it was sharing with SoundScan was in turn being used by Han-
dleman to make inventory planning and replenishment decisions at retailers such as
Wal-Mart. So, it stopped sharing information.
15 Information Leakage in Supply Chains 315

This example raises a number of questions that are the focus of this chapter. How
do the record labels benefit from the retailers sharing information with SoundScan?
Why do they provide incentives such as promotion and advertising support to the
retailers to share information? What is the loss to Newbury from sharing informa-
tion? What can the record labels do to incentivize Newbury to share information,
given its genuine concern about loss of market advantage?
There are other examples provided in Anand and Goyal (2009), Gal-Or et al.
(2008) and Kong et al. (2013) on the leakage of information in supply chains as
well as the reluctance to share information due to the threat of information leakage.
There are several forces that result in the leakage of information in a supply chain,
whether intentionally or otherwise. First, vertical information is shared by a retailer
with a supplier or an intermediary (such as SoundScan in the Newbury Comics
case) who aggregates information and shares it with other retailers. Second, the
emergence of category management in the retail industry has resulted in informa-
tion being shared by retailers with a leading manufacturer who manages a category,
called the “category captain”. In practice, a leading manufacturer serves as a cate-
gory captain for many retailers that are competing for the same consumers (Kurtulus
and Toktay 2008). While retailers find the benefits of category management and cat-
egory captainship attractive, this can result in valuable information being shared by
the category captain with competing retailers. As Kurtulus and Toktay (2008) point
out, “the trade-off that retailers face is the benefit from category captainship versus
the potential problems and loss of competitiveness that could arise from information
leakage.” Third, initiatives such as VMI and CPFR have accelerated the sharing of
information between manufacturers and retailers as well as between suppliers and
a manufacturer who uses their inputs. In turn, this has increased the likelihood of
the shared information being leaked to unintended recipients in a supply chain. The
leakage of information may occur by accident or due to deliberate efforts to obtain
proprietary information either by competitors or third parties. For example, hackers
recently managed to get into the computer systems at Foxconn, a major supplier of
consumer electronics to large corporations such as Apple and HP, and post informa-
tion about their client purchases (Mello 2012). Finally, the information may not be
leaked directly but observable actions based on that information taken by an entity
in the supply chain may unwittingly reveal the confidential information.
The fear and negative effects of information leakage may result in firms not shar-
ing information and reaping the corresponding benefits. Ron Ireland, who helped
develop CPFR processes at Wal-Mart points out that while Wal-Mart was willing
to share its forecasts and POS data with vendors, the sales teams within those ven-
dor organizations were scared to share it with their own corporate offices for fear
that information may leak to third parties and they would get into trouble with Wal-
Mart (Douglas 2004). Adewole (2005) points out that retailers in the UK clothing
industry are “reluctant to share information with suppliers, recognizing that those
suppliers might also be supplying competitors and could wittingly or unwittingly
divulge sensitive information.”
316 G. Kong et al.

The articles reviewed in this chapter provide insights into the issues raised above
using game-theoretic models. In particular, the focus of this chapter is on informa-
tion leakage. We would like to clarify some terms upfront. By the definition given
in the Oxford Dictionaries, “leakage” refers to the “deliberate disclosure of confi-
dential information.” We use the term “leakage” in a broader sense in that it can
be both intentional (deliberate) and unintentional because it is difficult to verify a
decision maker’s intentions. We also allow leakage to be direct or indirect. An act
of “direct leakage” means that the receiver of the confidential information passes
that information directly to a third party without the consent of the sender of that
information. An act of “indirect leakage” means that a third party can infer (per-
haps partially) the sender’s information from the receiver’s public actions indirectly.
As a common practice, direct leakage can be prevented or deterred by a binding
confidentiality agreement between the sender and the receiver. However, such an
agreement is often ineffective in preventing indirect leakage. In this article, we use
“confidentiality” and “no direct leakage” interchangeably. We also treat “degree of
confidentiality” and “degree of nonleakage” roughly the same.

15.1.1 Basic Framework

A typical model of a supply chain with (vertical) information sharing and (hori-
zontal) information leakage is illustrated in Fig. 15.1, first introduced by Li (2002).
The supply chain consists of a common manufacturer or supplier (“she”) at the up-
stream, denoted by M, and multiple retailers (“he”) at the downstream, labeled by
N = {1, 2, · · · , n}. The retailers compete in a common consumer market based on
either quantity or price, i.e., engaging in a Cournot or Bertrand competition.

Fig. 15.1 A common supply chain structure with multiple retailers

In the case of Cournot competition, the retail prices are determined from sales
quantities as follows:

pi = a + θ − qi − β ∑ q j , ∀i ∈ N, (15.1)
j=i
15 Information Leakage in Supply Chains 317

where pi and qi are the price and quantity of retailer i’s product. The intercept a + θ
represents the market condition or potential, where θ is a random variable with
zero mean and variance σ 2 . The parameter β (with |β | ≤ 1) captures the degree of
substitution or competition. When β = 1, the retailers’ products are perfect sub-
stitutes and the competition is most intensive, in which case the retail prices are
identical and can be denoted by p. When 0 < β < 1, the products are imperfect
substitutes and the competition is imperfect as well. The products are independent
when β = 0 and complements when −1 ≤ β < 0.
In the case of Bertrand competition, the sales quantities are determined from
retailer prices:
γ
n−1 ∑
qi = a + θ − (1 + γ )pi + p j, ∀i ∈ N. (15.2)
j=i

The products are imperfect substitutes when γ > 0 (with the degree of substitution
increasing in γ ), independent when γ = 0, and imperfect complements when − 12 <
γ < 0 (which satisfies |1 + γ | > |γ |).
It is often assumed that each retailer i observes a private signal Yi about the
uncertain θ . The signals have the following properties: (1) E(Yi |θ ) = θ , ∀i ∈ N;
(2) E(θ |Y1 , · · · ,Yn ) = α0 + ∑i∈N αiYi , for some constants αi ; (3) Yi , i ∈ N, are ind-
ependent and identically distributed, conditional on θ .1 These assumptions imply
(Li 1985, Lemma 1):
1
E(θ |Y j , j ∈ K) = E(Yi |Y j , j ∈ K) =
k+s ∑ Yj, i ∈ N\K, (15.3)
j∈K

where K ⊂ N is the set of retailers participating in information sharing, k = |K|, and

s = E(Var(Yi |θ ))/σ 2 is a measure of signal errors. This result means that ∑ j∈K Y j is
a sufficient statistic of (Y j ) j∈K for the purpose of estimating θ and unknown signals.
There are two basic classes of models in this body of literature. In one class,
the members of the supply chain determine how the private information will be
transmitted in the supply chain prior to the arrival of that information and their op-
erational activities, which is pioneered by Li (2002) and will be called the “ex-ante
information sharing arrangement.” In the other class, the members do not resort to
any formal agreement in advance and the flow of private information in the supply
chain is resolved after the information is available and through the interaction of the
parties. This setting is exemplified by the work of Anand and Goyal (2009) and will
be called the “ex-post information sharing arrangement.”
Table 15.1 lists the collection of papers that have addressed the issue of informa-
tion leakage in a supply chain. The second column in the table identifies the nature of
information sharing arrangement: whether it is ex-ante or ex-post. The third column
identifies whether information is shared directly or indirectly or both. It is interesting

1 Some prior-posterior conjugate distributions, e.g., normal-normal, gamma-Poisson, and beta-

binomial, satisfy these assumptions. However, as pointed out by Li and Zhang (2008), assumption
(2) imposes some restrictions on how much about θ can be learned from these signals.
318 G. Kong et al.

Table 15.1 Publication classification

Publication Information Retail Number

or working Sharing compe- of Type of
paper decision Leakage tition retailers product Other

Li (2002) Ex-ante Direct Cournot N≥2 Perfect Cost

substitutes information,
Side payments
Zhang (2002) Ex-ante Direct Cournot, 2 Imperfect Side payments
Bertrand substitutes,
comple-
ments
Li and Zhang Ex-ante Direct, Bertrand N≥2 Imperfect
(2008) Indirect substitutes
Gal-Or et al. Ex-ante Direct, Bertrand 2 Imperfect M has private
(2008) Indirect substitutes information
Anand and Ex-post Direct Cournot 2 Perfect Information
Goyal (2009) substitutes acquisition
Chen and Ex-ante Direct Cournot 2 Perfect Auction pricing
Vulcano substitutes
(2009)
Shin and Ex-ante Indirect Cournot, N≥2 Perfect Two-part tariff,
Tunca (2010) Bertrand substitutes Market-based
contract
Jain et al. Ex-ante Direct Cournot N≥2 Perfect, Different
(2011) imperfect wholesale prices
substitutes
Ha et al. Ex-ante Minimum Cournot, Two SCs Imperfect Competing SCs
(2011) Bertrand with one M substitutes
one R
Qian et al. Ex-ante Direct Cournot N≥2 Perfect M has limited
(2012) substitutes capacity
Shamir Ex-ante Direct Bertrand N≥2 Imperfect Mechanism
(2012) substitutes design
Kong et al. Ex-post Direct Cournot 2 Perfect Revenue sharing
(2013) substitutes
Jain and Ex-ante Direct, Cournot 2 Imperfect Different
Sohoni Indirect substitutes wholesale prices
(2015)
Shamir Ex-post Direct, Bertrand N≥2 Perfect Infinite horizon
(2015) Indirect substitutes
Shamir and Ex-post Direct Cournot Two SCs, Perfect Competing SCs
Shin (2015) with one M substitutes
one R
15 Information Leakage in Supply Chains 319

to note that all the works in the literature on information leakage have considered a
supply chain structure where the competition is at the downstream end (i.e., a manu-
facturer supplies multiple retailers) or two supply chains, each with a manufacturer
and retailer, compete with each other. At the retail level, the competition could be
based on quantity (Cournot) or price (Bertrand) and the fourth column in Table 15.1
identifies this aspect of the models. The fifth column identifies whether the model
has two or more retailers or if there are two supply chains competing with each
other. The sixth column examines whether the products are substitutes or comple-
ments. Most papers except Zhang (2002) focus on perfect or imperfect substitutes.
Finally, the last column identifies some unique aspects of the models considered in
a paper. For example, most of the papers assume a wholesale price contract between
the manufacturer and the retailers but a few papers have considered other types of
contracts or mechanisms, e.g., revenue sharing, side payments, market-based con-
tracts, and auction.

15.2 Ex-Ante Information Sharing Arrangement

A stream of papers share the basic model setup of Li (2002). The manufacturer and
retailers’ marginal costs are normalized to zero, without loss of generality. Events
take place in the following order:
1. The manufacturer (M) and retailers make an information sharing arrangement,
i.e., deciding the set K of retailers who will share their information with the
manufacturer;
2. Each retailer i observes signal Yi and, if i ∈ K, shares it with M;
3. M sets the wholesale price w;
4. Each retailer i chooses an order quantity qi (under Cournot competition) or retail
price pi (under Bertrand competition);
5. M delivers the products and the market is cleared.
Li and Zhang (2008) propose three scenarios of information sharing and leakage,
or degrees of confidentiality, as summarized in Table 15.2, which offers a useful
perspective to organize the existing literature.

Table 15.2 Information known or disclosed to different parties

Scenario Manufacturer Participating retailer i ∈ K Non-participating retailer i ∈
/K
S1 ∑ j∈K Y j Yi and ∑ j∈K Y j Yi and ∑ j∈K Y j
S2 ∑ j∈K Y j Yi and ∑ j∈K Y j Yi and w(∑ j∈K Y j )
S3 ∑ j∈K Y j Yi and w(∑ j∈K Y j ) Yi and w(∑ j∈K Y j )
320 G. Kong et al.

In scenario (S1), M leaks the information collected from participating retailers

(those in K) to all retailers directly. In other words, there is no confidentiality. In
(S2), M only leaks the collected information to the participating retailers directly,
which corresponds to partial confidentiality. In (S3), M does not leak any informa-
tion to any retailer directly, i.e., participating retailers’ information is kept by M
with full confidentiality. However, all retailers can infer ∑ j∈K Y j from the wholesale
price w indirectly (it is commonly assumed that w is strictly increasing in ∑ j∈K Y j
in the equilibrium). Information inferred indirectly is less reliable (and hence less
valuable) than that acquired directly, as the former is subject to manipulation and
incentive concerns. Thus, when ∑ j∈K Y j can be obtained directly, the wholesale
price w(∑ j∈K Y j ) only plays the traditional role of price setting, without the role
of signaling.
Note that the scenarios in Table 15.2 require less information than the ones
defined in Li and Zhang (2008): the full information (Y j ) j∈K is replaced by the agg-
regate information ∑ j∈K Y j , because the latter is a sufficient statistic of the former
as discussed earlier. Although the two sets of scenarios are mathematically equiva-
lent, the current ones are considerably easier to implement in practice. It has been
shown that aggregate information can be shared through secured protocols without
revealing private information held individually (Deshpande et al. 2010).

15.2.1 Li (2002) and Zhang (2002)

Li (2002) investigates the full-leakage scenario (S1) under Cournot competition with
perfect substitutes. Using backward induction, the paper shows the following nega-
tive result (Proposition 4): Given any information sharing arrangement K ⊂ N, the
manufacturer is better off by acquiring information from more retailers, but each
retailer is worse off by sharing information with the manufacturer; thus, no infor-
mation sharing, or K = 0,/ is the unique equilibrium.
This result is driven by two effects of information sharing. The direct effect
(loss of information rent) is that more information allows the manufacturer to ex-
tract more surplus from a retailer through the choice of wholesale price, as evident
from the equilibrium wholesale price: w∗ ((Y j ) j∈K ) = a/2 + 1/(2(k + s)) ∑ j∈K Y j .2
The indirect effect of information sharing (loss of competitive advantage) refers to
the leakage of a participating retailer’s demand information to his competitors and
the resulting information disadvantage. The expressions for the sales quantities are
given by (Proposition 1):

2 The paper shows that the manufacturer cannot gain from charging different w’s to participating

and non-participating retailers (if the wholesale price is determined after information sharing).
15 Information Leakage in Supply Chains 321
 
1
q∗i (Yi , w∗ ) = a − w∗ + Ak1 ∑ Y j , i ∈ K, (15.4)
n+1 j∈K
 
1
∗ ∗
qi (Yi , w ) = a − w + B1 ∑ Y j + B2Yi , i ∈ N\K,
∗ k k
(15.5)
n+1 j∈K

for some positive constants Ak1 , Bk1 , and Bk2 . Thus, a non-participating retailer ex-
ploits both the leaked information ∑ j∈K Y j and his private information Yi , while a
participating one can only utilize ∑ j∈K Y j .
Because of the misaligned incentives between the manufacturer and the retailers,
it may be beneficial for the manufacturer to buy information from the retailers. Stage
(1) of the sequence of events is modified as follows: M promises each retailer a fee δ
if he will share his information later, and each retailer decides whether to accept the
payment and commit to information sharing. The paper shows that (Proposition 5)
in the augmented game, there are only two possible equilibria, complete informa-
tion sharing (K = N) and no information sharing (K = 0), / and the former Pareto
dominates the latter if and only if s ≤ (n − 2)(n + 1)/2. It also shows that complete
information sharing hurts both the social benefits and consumer surplus (Proposi-
tion 7). Thus, information sharing should be discouraged from the standpoint of a
social planner.
Zhang (2002) extends the main finding of Li (2002) to more general types of
competition. The paper focuses on a supply chain with one manufacturer and two
retailers who engage in either Cournot or Bertrand competition with imperfect sub-
stitutes or complements. It investigates the full-leakage scenario (S1) in Li (2002)
and shows the following (Proposition 2): The manufacturer is always better off by
acquiring demand information from more retailers; Each retailer is always worse off
by disclosing his private information to the manufacturer; Therefore, no information
sharing is the unique equilibrium.
As in Li (2002), information sharing has both direct (loss of information rent) and
indirect (leakage) effects. When the products are substitutes (complements) under
Cournot (Bertrand) competition, both effects are negative for a retailer. When the
products are complements (substitutes) under Cournot (Bertrand) competition, the
leakage effect is positive, but not strong enough to overcome the negative direct
effect.
The paper further shows that the manufacturer may be able to induce informa-
tion sharing by a side payment, which is easier to achieve when the products are
closer to perfect complements (substitutes) in a Cournot (Bertrand) competition or
the demand signals are statistically less accurate (s above a threshold).

15.2.2 Li and Zhang (2008)

This paper studies all three leakage scenarios, (S1), (S2), and (S3), under Bertrand
competition with imperfect substitutes. The main differences among the three
322 G. Kong et al.

scenarios manifest themselves in the equilibrium retail prices. Let ri = pi − w denote

the margin of retailer i. In each scenario S ∈ {S1, S2, S3}, in the equilibrium, ri∗ are
given by:
1
ri∗ (w) = [a − w + Ak θPS + (1 − Ak )θNS ], i ∈ K, (15.6)
2+γ
1
ri∗ (w) = [a − w + BkYi + (1 − Bk )θNS ], i ∈/ K, (15.7)
2+γ

for some positive constants Ak and Bk . In the expressions, θPS is the estimate of θ
available to a participating retailer and θNS is that to a nonparticipating retailer in sce-
nario S. By the definition of the scenarios, we have: (1) θPS1 = θNS1 = E(θ | ∑ j∈K Y j );
(2) θPS2 = E(θ | ∑ j∈K Y j ), θNS2 = g(w); and (3) θPS3 = θNS3 = g(w), with the special
case that θPS3 = E(θ |Yi ) when K = {i}. The function g(w) represents the estimation
E(θ | ∑ j∈K Yj ) where ∑ j∈K Yj is inferred from w(∑ j∈K Y j ) (assuming a separating
equilibrium). In scenario (S1), the wholesale price w is a pure price instrument for
the manufacturer, while in scenarios (S2) and (S3), it also serves as a signaling
device.
In the full-leakage scenario (S1), the paper confirms the results of Li (2002)
and Zhang (2002), under oligopoly Bertrand competition. That is, information shar-
ing benefits M but harms the retailers and, therefore, no information sharing is the
unique equilibrium.
In the nonleakage scenario (S3), the following results are found (Propositions 3,
4, 7, and 8): When retail competition is intense (γ large enough), the only possible
equilibria are complete information sharing (K = N) and no information sharing
(K = 0);/ The former can be induced voluntarily if (in addition) the demand infor-
mation is less accurate (s large enough) or through a side payment from M to the
retailers; In any case, complete information sharing Pareto-dominates no informa-
tion sharing; Under complete information sharing, the supply chain is coordinated,
and no retailer will misreport his signal if all other retailers do it truthfully.
These positive results stem from the weakened direct effect of information shar-
ing (double marginalization) in scenario (S3). An increase (or decrease) of w would
signal higher (or lower) demand and induce the retailers to raise (or reduce) their
margins, which would dampen (or boost) the sales quantities as well as the manu-
facturer’s profit. This change of price elasticity motivates the manufacturer to set a
lower wholesale price in (S3) than in (S1). In short, the signaling role of the whole-
sale price under Bertrand competition alleviates the double marginalization in the
supply chain, which benefits the retailers and the supply chain but hurts the man-
ufacturer. This argument does not hold under Cournot competition with imperfect
substitutes, because the signaling role of w encourages the increase of w as it would
induce the retailers to increase order quantities which in turn benefits the manufac-
turer. That would only aggravate the double marginalization in the supply chain.
Comparing the scenarios, the paper shows that given the set of participating re-
tailers K, a higher degree of confidentiality results in a lower equilibrium wholesale
price, which harms the manufacturer and benefits all retailers (Propositions 5 and 6).
15 Information Leakage in Supply Chains 323

Thus, the manufacturer prefers full leakage, the scenario (S1), and the retailers pre-
fer full confidentiality, (S3). However, under (S1), the only equilibrium outcome is
no information sharing (K = 0), / which from the manufacturer’s perspective is worse
than any arrangement K ⊂ N under (S3). Therefore, the manufacturer and retailers
should try to realize (S3) through a binding confidentiality agreement.

15.2.3 Jain et al. (2011)

This paper tries to fill a gap left by Li (2002), Zhang (2002), Li and Zhang (2008).
It first shows or confirms the following negative results under Cournot competition
(and a single wholesale price): no information sharing is the unique equilibrium
in scenarios (S1) and (S3); in scenario (S2), no information sharing is always an
equilibrium, while full information sharing is an equilibrium only for a special range
of s and σ ; furthermore, truth-telling by all retailers is not an equilibrium.
The paper then demonstrates that by charging differential wholesale prices, full
information sharing can be achieved. The first stage in the sequence of events is
modified to:
1. M announces the pricing scheme {wi }i∈N , and each retailer decides whether to
participate in information sharing, which defines the set K ⊂ N.
The paper focuses on affine pricing schemes of the form: wki = Ak1 + Bk1 ∑ j∈K Y j −
DkYi , i ∈ K, or Ak2 + Bk2 ∑ j∈K Y j , i ∈ N\K, for constants Ak1 , Bk1 , Ak2 , Bk2 , and Dk . The
model assumes that the aggregate information ∑ j∈K Y j is publicly verifiable, which
is in effect scenario (S1). The signal Yi in the expression of wki for i ∈ K is reported
by retailer i. Thus, the wholesale price wki plays the role of screening a participating
retailer’s information. It is shown that (Proposition 2): The optimal wholesale prices
that sustain information sharing arrangement K ⊂ N and induce truth-telling are
determined by Ak1 = Ak2 = a/2, Bk2 = 1/(2(k + s)), Bk1 = Bk2 + (1/(k + s))Dk , and
a certain positive constant Dk ; under such wholesale prices, K = N is the unique
dominant strategy equilibrium.
The negative term in wki , for i ∈ K, reduces a participating retailer’s wholesale
price if he reports a higher signal Yi . This “good news bonus” offsets the retailer’s
incentive to distort his signal downward under Cournot competition and hence in-
duces truth-telling.
Although differential wholesale pricing is able to induce full and truthful infor-
mation sharing, it is unable to coordinate the supply chain. The paper proposes a
more flexible pricing scheme with a fixed charge to participating retailers on top of
differential wholesale prices. Such differential two-part tariffs are able to coordinate
the supply chain when β = 1 (with perfect substitutes) or achieve near full efficiency
(more than 99 %) when 0 < β < 1 (with imperfect substitutes), while inducing full
and truthful information sharing. The manufacturer is able to extract all supply chain
surplus in the former case, but not so in the latter.
324 G. Kong et al.

15.2.4 Gal-Or et al. (2008)

This paper assumes Bertrand competition between two retailers, as studied by some
other papers, but it generalizes the information structure as follows: the manufac-
turer observes a signal x0 (of the demand shock θ ), while retailers observe x1 and x2 ,
respectively. The main part of the paper focuses on “one-sided information sharing”
where only the manufacturer attempts to disclose her private information and as-
sumes that the manufacturer will do it truthfully.
If a retailer is left outside the information sharing club, he may infer x0 from
the wholesale price. Thus, the model is comparable to the information sharing sce-
nario (S2), with the aggregate retailer information ∑ j∈K Y j replaced by the manufac-
turer’s signal x0 . With two retailers, there are only three arrangements: K = 0, / i.e.,
no information sharing (NS); K = {i}, i.e., partial sharing with retailer i (PSi); and
K = N, i.e., full information sharing (FS). Within the class of affine wholesale prices
w = α0 + α x0 , the following results are shown (Corollary 1 and Proposition 1): The
manufacturer’s optimal wholesale prices satisfy w∗FS > w∗PSi > w∗NS , and her opti-
mal profits satisfy E(ΠFS ) > E(ΠPSi ) > E(ΠNS ); In addition, E(ΠPS1 ) > E(ΠPS2 )
if s1 > s2 , i.e., the manufacturer prefers to share x0 with the retailer who possesses
less accurate information.
Notice that the (FS) and (NS) arrangements can also be viewed as special cases of
scenarios (S1) and (S3), respectively. Thus, the optimal wholesale prices and manu-
facturer profits follow the same orders as in Li and Zhang (2008). The driving force
is the same signaling (or inference) effect, which alleviates double marginalization
under Bertrand competition when direct leakage can be (partially) prevented. In this
model, the manufacturer dictates the information sharing arrangement. In practice,
she must weigh the incremental benefits and costs from adding an (additional) re-
tailer to the information sharing arrangement. If only one retailer is to be chosen, she
should pick the less-informed one. That retailer’s need for the manufacturer’s infor-
mation is more acute, so if he had to infer it instead of receiving it directly from the
manufacturer, the signaling effect (the pressure to lower the wholesale price) would
be stronger.
The paper also studies “two-sided information sharing,” which is closer to the
model of Li and Zhang (2008). However, the paper only shows the following results
for a very special case (Proposition 3): Under two-sided communication with s1 =
0 and s2 = ∞ (i.e., retailer 1 has perfect information and retailer 2 has none), to
communicate with only one retailer, the manufacturer will choose the uninformed
retailer (retailer 2) if the competition is sufficiently weak (product differentiation
high) or the manufacturer’s own information is sufficiently accurate; and he will
choose the perfectly informed retailer (retailer 1) otherwise.
The intuition lies at the trade-offs between communicating with one of the re-
tailers. On the one hand, the signaling effect favors the uninformed retailer. On the
other hand, the value of information to the manufacturer favors the fully informed
retailer. In a setting with weak competition or well-informed manufacturer, the for-
mer effect dominates the latter. We note that the retailers’ incentives, i.e., voluntary
participation and truth telling, are not considered in this study.
15 Information Leakage in Supply Chains 325

15.2.5 Jain and Sohoni (2015)

This paper studies a supply chain with two retailers (R1 and R2) under imperfect
Cournot competition. Unlike previous models, the retailers order sequentially under
differentiated wholesale prices, as described by the sequence of events below:
1. M and R1 reach an agreement whether or not to disclose the information {w1 , q1 }
to R2 later (in stage 4);
2. M announces wholesale price w1 to R1;
3. R1 observes signal Y1 and orders quantity q1 ;
4. M announces wholesale price w2 to R2;
5. R2 observes signal Y2 and orders quantity q2 ;
6. Ordered units are sold and the market is cleared at the retail price p.
Although Y1 is not directly disclosed by R1, it can be inferred by M from q1 in
the equilibrium, so after stage 3 the model coincides with the standard model with
information sharing arrangement K = {1}. There are two leakage scenarios with
respect to R2: in the full disclosure (FD) case, {w1 , q1 } is leaked to R2 in stage 4,
which agrees with scenario (S1); and in the no disclosure (ND) case, {w1 , q1 } is not
leaked directly although R2 can infer q1 from w2 indirectly, which is consistent with
scenario (S2) or (S3) (they are identical when K = {1}).3
To sustain nonleakage (ND), it must be beneficial to both M and R1. The pa-
per shows the following: It is harder to persuade M to protect the information than
R1; When the competition is intense, M prefers (ND) to (FD) when the relative in-
formativeness of R2 (measured by x = [1/E[Var(Y2 |θ )]]/[1/E[Var(Y1 |θ )] + 1/σ 2 ])
is in a medium range; When the competition is weak, M prefers (ND) when x is
above some threshold; And a two-part tariff between M and R1 will make it easier
to achieve nonleakage.
These results are driven by the interplay of several effects. Nonleakage makes
it harder for R2 to learn R1’s demand information, but it also diminishes R1’s first
mover advantage (by using q1 as a means of deterrence). From R1’s perspective, the
former effect is positive and the latter is negative. The effects are less clear cut from
M’s perspective because she has the instrument w2 to fine tune the implications of
the information arrangement.

15.2.6 Shin and Tunca (2010)

This paper studies a model similar to that of Li (2002) but with two main differ-
ences. First, the retailers’ demand signals do not come for free. The cost for demand
forecasting, C(vi ), is increasing and convex in the precision of the acquired infor-
mation, vi = σ 2 /E(Var(Yi |θ )). Second, the sequence of events is as follows (where
q−i = (q1 , · · · , qi−1 , qi+1 , · · · , qn )):
3 It is shown in the paper that the functions q1 (w1 ,Y1 ), w2 (w1 , q1 ), and q2 (w1 , q1 , w2 ,Y2 ) (in FD)
or q2 (w2 ,Y2 ) (in ND) are all affine functions.
326 G. Kong et al.

1. M announces price scheme {P(qi , q−i )};

2. Retailer i makes investment vi ;
3. Retailer i observes signal Yi , and orders qi from M;
4. Retailer i pays M the total price P(qi , q−i );
5. Ordered units are received and sold, and the market is cleared at the retail price p.
Each retailer shares his private information with M indirectly through his order
quantity qi . As all retailers order at the same time, information leakage has no
impact on the outcome. More specifically, even though other retailers’ order quanti-
ties (or the total quantity) may be revealed to retailer i from the payment P(qi , q−i )
charged by M, he has no opportunity to utilize that information. He may regret
later the quantity he ordered, which is an issue related to the robustness of the
equilibrium.
The paper first considers the case when retailers’ information investments cannot
be observed by others. It shows that (Propositions 1 and 2): Under the wholesale
price contract P(qi , q−i ) = wqi , or two-part tariff P(qi , q−i ) = w0 + w1 qi , retailers
over-invest in demand forecasting compared to the first-best levels, and the loss in
supply chain profit can be substantial. The paper introduces the so-called market-
based or index-based contracts P(qi , q−i ) = w0 + p(q)qi −wd q2i , where p(q) = w1 +
w2 ∑nj=1 q j is the index price and wd > 0 corresponds to quantity discounting. It
shows that (Proposition 4) market-based contracts can fully coordinate the supply
chain, with respect to both information investments and order quantities.
These results are consistent with the findings in the literature, i.e., the more the
manufacturer can commit to nonleakage, the easier it is to induce information shar-
ing and to coordinate the supply chain. To achieve coordination, the price must be
adapted to retailers’ information, e.g., ∑nj=1 q j , which makes the model comparable
to the standard model in scenario (S3) with K = N. However, the setting is even
more stringent in this paper because at the time they place orders, the retailers can-
not infer anything about others’ signals, and thus direct and indirect leakage is fully
blocked in advance.4
The paper also considers the case when retailers’ investments can be observed
by competitors, in stage 2 of the sequence of events. In this situation, each re-
tailer benefits from a higher signal accuracy (known to all retailers) as it makes
his competitors more responsive to his order strategy, which creates stronger incen-
tives for over-investment and prevents full coordination even by the market-based
contract. As a remedy, the paper shows that a more complex mechanism, the so-
called uniform-price auction, can achieve full coordination. This mechanism allows
the manufacturer to announce a supply function in addition to a price scheme (in
stage 1) and the retailers to submit demand functions rather than fixed order quan-
tities (in stage 3). The equilibrium order quantities are determined by balancing the
total supply and total demand. In the equilibrium, the order quantity of a retailer is

4 Indirect leakage occurs only after the orders are made, through the index price p(q). However, as
shown in the paper (Proposition 5), the retailers’ equilibrium strategy is regret-free, i.e., they have
no incentive to alter their order quantities after learning other retailers’ quantities.
15 Information Leakage in Supply Chains 327

now related to the wholesale price he is facing, which in turn is related to the agg-
regate demand information. This essentially turns the model into scenario (S3) of
the standard model with K = N, where the retailers can infer the aggregate demand
information from the manufacturer’s wholesale price before ordering.

15.2.7 Shamir (2012)

Shamir (2012) provides a different perspective to the issues raised in prior works by
suggesting that retailers may in fact have an incentive to share information with
their competitors and the manufacturer. The paper considers a model similar to
Li and Zhang (2008) with one manufacturer supplying N (≥ 2) retailers in price
(Bertrand) competition. There are two scenarios: the retailers may share information
only with other retailers (horizontal information sharing) or may share information
with the manufacturer too in addition to their competitors (public information shar-
ing). Note that the scenario with public information sharing is similar to scenario
(S1) in Table 15.2 but it is the retailers here who initiate the sharing of information
rather than the manufacturer.
The paper shows the following results: when the information can be verified,
every retailer is better off by sharing his private information with other retailers
(horizontal) and a retailer is better off as more retailers share their private informa-
tion (Proposition 1); a retailer is worse off by sharing information with the man-
ufacturer, i.e. in the “public sharing” setting (Proposition 2). The first result is a
natural consequence of the Bertrand model. The second result follows from the fact
that the manufacturer will extract rent and make the retailer worse off under vertical
information sharing, similar to the insight in other papers.
Next, a scenario is considered where a retailers’ information cannot be verified
so he can engage in cheap talk. In this case, it is shown that: retailers have no incen-
tive to share information truthfully and accurately in either the horizontal or public
information sharing scenarios (Propositions 3, 4, and 5).
Finally, the paper considers a setting where the retailers can design a mechanism
to signal their private information truthfully while maximizing their profits. It shows
that in some situations the retailers prefer sharing information publicly to horizon-
tally, i.e., it is beneficial to invite the manufacturer into the information sharing club
(Proposition 12). It also shows that retailers incur a higher cost for reporting a high
signal (good news) in the horizontal information sharing setting and a high cost for
reporting low demand (or bad news) in the public sharing case. This is consistent
with the insight that retailers have a natural incentive to inflate demand in the hori-
zontal sharing setting (so as to keep retail prices high) and to deflate demand in the
public sharing case (so as to keep wholesale prices low).
328 G. Kong et al.

15.3 Ex-Post Information Sharing Arrangement

In the previous section, information sharing arrangement between the manufacturer

and retailers is pre-determined at the beginning, after which, the parties do not con-
cern themselves with the question of whether or not to share (or leak) information.
In addition, when the (aggregate) information is inferred instead of directly leaked,
it can be inferred perfectly, i.e., only separating equilibria are considered. A stream
of papers, starting with Anand and Goyal (2009), deviates from these assumptions.

15.3.1 Anand and Goyal (2009)

This paper studies a supply chain with one manufacturer (or supplier) and two ret-
ailers, an incumbent and an entrant, engaging in Cournot competition. The inverse
demand function is given by p = ã − (qi + qe ), where qi and qe are the order quanti-
ties of the incumbent and the entrant, respectively, and ã can be aH with probability
ρ or aL (< aH ) with probability 1 − ρ . Only the incumbent can observe the exact
ã because of his familiarity with the market. We refer to the incumbent as the high
(or low) type when ã = aH (or aL ). The sequence of events is as follows:
1. M announces wholesale price w;
2. Incumbent observes ã and places an order qi with M;
3. M decides whether to leak the information qi to the entrant;
4. Entrant places an order qe with M;
5. Retailers receive and sell ordered units, and the market is cleared at the retail
price p.
In this model, M does not make any formal arrangement with the incumbent on
information sharing and the incumbent does not share the observed ã directly. Keep-
ing the incentives of M and the entrant in mind, the incumbent plays the Stackelberg
leader in a signaling game. He must determine whether or not to let the supplier infer
the correct ã through his order qi and foresee whether M will leak that information
to the entrant. He has two options. First, he can order different quantities given
different ã, which will reveal the true ã to M and is called a separating strategy.
Second, the incumbent can order the same amount regardless of ã, which will pre-
vent M from inferring the demand information and is called a pooling strategy. In
both cases, M needs to decide whether or not to leak qi to the entrant.
Given the wholesale price w, define aH = aH − w, aL = aL − w, and μ  = μ − w,
where μ = ρ aH + (1 − ρ )aL . Let θ = aH /aL be a proxy for demand uncertainty.
The paper shows the following results (Propositions 1, 2, & 3): (1) If θ ≥ 3, the
incumbent orders q∗iH = aH /2 or q∗iL = aL /2, contingent on ã; M leaks the in-
cumbent’s order quantity to the entrant; and the entrant orders q∗eH = aH /4 or
q∗eL = aL /4 accordingly; (2) If Θ (ρ ) < θ < 3, the incumbent orders q∗iH = aH /2
or q∗iL = (2aH − aL − 3(aH )2 − 4aH aL + (aL )2 )/2, depending on ã; M leaks; and
15 Information Leakage in Supply Chains 329

the entrant orders q∗eH = aH /4 or q∗eL = [3aL − 2aH + (aH − aL )(3aH − aL )]/4 ac-
cordingly; (3) If 1 < θ ≤ Θ (ρ ), the incumbent orders q∗i = aL − μ  /2 regardless
of ã; M leaks; and the entrant orders q∗e = (3μ  − 2aL )/4. The threshold Θ (ρ ) above
is a decreasing function of ρ with Θ (0) = 3 and Θ (1) = 1.
The incumbent plays the separating strategy in the first two cases and pooling in
the third. The incumbent would always want the entrant to believe that the demand
is low (ã = aL ) so that he should order less. Thus, the low-type incumbent would
prefer M to leak the information and the high-type incumbent would try to mimic
the low type. In case (1), the demand uncertainty (gap between the two states) is so
significant that the low type can simply choose his optimal quantity under public
information, knowing that the high type cannot afford to imitate. The demand infor-
mation is truthfully revealed and no quantity distortion is exercised. In case (2), the
demand gap shrinks to a level that the low type needs to distort his order quantity
downward to be able to escape from the high type. In case (3), the gap becomes so
small that it would be too costly for the low type to separate out.
Why would M always leak the incumbent’s quantity? As the wholesale price
is fixed, M prefers larger quantities from the retailers. Suppose that the incumbent
plays a separating strategy. Without leakage, the entrant will assume an average
demand and place a moderate order. If the demand is actually high (revealed by a
large order from the incumbent), M will be better off by leaking that information
and attracting a larger order from the entrant. Thus, M will leak whenever she infers
a high demand, which is as good as leaking in both demand states, because if M does
not leak in the low demand state the entrant can infer the (low) demand correctly.
Second, suppose that the incumbent plays a pooling strategy. When the incumbent’s
order is relatively small, the supplier benefits from leaking that information and
encouraging the entrant to order more.

15.3.2 Kong et al. (2013)

This paper studies a model similar to that in Anand and Goyal (2009) except that it
considers a revenue sharing contract between the manufacturer and retailers instead
of a wholesale price contract. In a revenue sharing contract, the supplier sells the
product to the two retailers at a possibly lower wholesale price, say w, and instead
receives a share α of the retail revenue. The sequence of events is identical to that
in Anand and Goyal (2009) and the intercept of the inverse demand function is
ã = aH or aL , with mean μ = ρ aH + (1 − ρ )aL . It is assumed that the supplier can
communicate to the incumbent whether or not she intends to leak the incumbent’s
order quantity to the entrant, which is credible if the supplier can make a higher
profit with the intended action.
First, consider the scenario where the parameters of the revenue sharing contract
(w, α ) are fixed (announced in stage 1). Let θ = [aH − w/(1 − α )]/[aL − w/(1 − α )].
The following result in the paper establishes necessary and sufficient conditions for
330 G. Kong et al.
√
a nonleakage equilibrium (Theorem 1): “ Assume that θ ≥ (1 − ρ )/[3(1 − 2/2) −
ρ ] ≥ 0 and w/μ ≤ (1/2)(3aL /μ − 1)(1 − α ). A nonleakage equilibrium exists if
∗ ∗
qNiH ≥ q̄i and qSiL ≤ qi , and only if w/μ ≤ (3aH /μ + 2)(α (1 − α ))/(12 + 5α ) and
∗ ∗ ∗
qSiL ≤ qi .” The quantities qNiH and qSiL are, respectively, the incumbent’s optimal or-
der quantities under the nonleakage and separating leakage equilibria. The threshold
limits qi and q̄i are functions of w, α and μ ; if the incumbent’s order quantity falls
within the interval [qi , q̄i ], the supplier prefers leakage to nonleakage. Under leak-
age, the downstream retailers together may underorder when the demand is low and
overorder when the demand is high, compared with the supplier’s first-best quantity.
This type of quantity distortion may be mitigated in both demand states simultane-
ously if the supplier does not pass the demand information to the entrant so that the
entrant has to order an intermediate quantity, aimed at the average demand. Thus,
the supplier may benefit from nonleakage in both demand states. The conditions in
the theorem ensure that the incumbent retailer is also better off under nonleakage.
This is in contrast to a wholesale price contract, where the supplier always benefits
from leaking the incumbent’s order quantity; because a larger order translates into
higher profit for the supplier, the supplier would always like to inform the entrant
when the demand is high. This is no longer true under a revenue sharing contract,
where a larger order is not always better for the supplier.
The paper also shows that there exists a set of (w, α ) pairs that support a nonleak-
age equilibrium, referred to as the nonleakage region. The range of wholesale prices
that support nonleakage is relatively wide when α lies in the middle of the interval
[0, 1] and it shrinks as α moves toward 0 or 1. The case α = 0 is equivalent to the
wholesale price contract and the result is the same as in Anand and Goyal (2009)
that the supplier always leaks. When α increases, the supplier’s profit is more in
line with the supply chain profit and she is more willing to control the total quan-
tity in the channel by hiding the demand information from the entrant. However,
as α approaches 1, the feasible range of w that induces the retailers participation
diminishes, resulting in a narrow nonleakage band. The nonleakage region expands
when the ratio aH /aL is higher. In this case, there is greater demand variation and
the incumbent’s information advantage exacerbates the quantity distortion from the
supplier’s perspective and motivates the supplier to prevent information leakage.
The paper shows that the total supply chain profit may increase under nonleak-
age. More interestingly, not only do the supplier and incumbent benefit from non-
leakage, but sometimes even the entrant can be better off under nonleakage. This
is because while nonleakage prevents the entrant from adjusting the order quantity
based on better demand information, the entrant benefits from being able to place an
order simultaneously with the incumbent under nonleakage rather than sequentially
under leakage.
The paper shows that the results are robust even if the wholesale price is en-
dogenous, given a revenue sharing rate. Specifically, there exists a threshold on the
revenue sharing rate α above which the supplier’s optimal wholesale price lies in
the nonleakage region. Moreover, this threshold decreases as the ratio aH /aL in-
creases. This suggests that as the demand states become more distinguishable, a
smaller share of revenue is needed to persuade the supplier not to leak.
15 Information Leakage in Supply Chains 331

The paper also shows that the revenue sharing contract continues to be attractive
in terms of preventing information leakage when some of the model assumptions
are relaxed or altered. It is shown that the nonleakage region will be larger when:
(1) the incumbent could place a larger order and hold back (i.e., not sell) some
units to achieve a higher retail price, (2) the entrant may choose to ignore infor-
mation provided by the supplier. Finally, it is shown that there exists a substantial
nonleakage region even if the incumbent does not have a first mover advantage and
the incumbent and entrant play a simultaneous (rather than sequential) game after
the supplier has leaked the information to the entrant. Overall, the incentives of the
supplier and retailers are better aligned under revenue sharing and the supplier is
not simply trying to push product under all circumstances as in a wholesale price
contract.

15.3.3 Shamir (2015)

Shamir (2015) considers a framework with one manufacturer supplying N retailers

who compete on price. Unlike other papers with a Bertrand model, such as Shamir
(2012) and Li and Zhang (2008), this paper considers a Bertrand game with per-
fect substitutes (or homogeneous products). It also assumes that the retailers share
information after observing their private signals. While prior works (both ex-ante
and ex-post models) focus on the negative effects of information leakage and how it
might motivate retailers to not share information, this paper takes a counter-intuitive
perspective, like Shamir (2012). In particular, it considers the possibility that a re-
tailer may want to share information with the manufacturer expecting that it will
be disclosed through the wholesale price to other retailers, with the wholesale price
acting as a collusion device.
Each retailer i gets a private signal Yi ∈ {H, φ } about the market condition, which
can be high or low. If actual demand is high (H), the retailer has a probability ρ of
learning that it is H (informative signal). If actual demand is low (L), then the re-
tailer only observes the non-informative signal φ . Upon observing the signal Yi , re-
tailer i updates the probability that the market condition is H in a Bayesian fashion.
The paper considers an infinite horizon repeated game where demand and private
signals in each period are independent and identically distributed and the entire his-
tory of wholesale and retail prices is observable by all entities. The paper explores
three different information sharing scenarios: retailers share their private demand
information horizontally with other retailers and collude (I1); they do not share any
demand information (I2); they share their demand information solely with the man-
ufacturer (I3).
In setting (I1), it is shown that retailers (1) can collude and set a monopoly
price and (2) do not have an incentive to share information with the manufacturer
(Lemma 1). In scenario (I2), where they cannot share information and collude, two
possible settings are considered: retailers follow a rigid price that they all follow
each period or retailers can set a price each period based on their observed private
332 G. Kong et al.

signals. They obtain the following result: as the number of retailers increases above
a threshold, the retailers will gravitate towards using a rigid pricing scheme rather
than variable prices based on their private signals each period (Proposition 3).
This is because the likelihood that at least one retailer receives a non-informative
signal increases as the number of retailers increases. A retailer receiving a non-
informative signal will set a low price and the retailers who have an informative
signal will get zero profits. Thus, the retailers are not able to coordinate and collude
using variable pricing and the cartel prefers to use a rigid pricing scheme and ignore
the private information of its members.
In setting (I3), where retailers share private demand information with the man-
ufacturer, it is shown that the manufacturer will set one of two wholesale prices
wH or wφ depending, respectively, on whether they receive an informative or non-
informative signal (Proposition 4). The retailers infer, on observing wH , that at least
one retailer has received an informative signal and infer that all the retailers have re-
ceived the non-informative signal when the manufacturer chooses wφ . The manufac-
turer has to distort the wholesale price down in the non-informative case if demand
uncertainty is not high to send a credible signal and achieve a separating equilib-
rium. This distortion is similar to the distortion in the wholesale price in Gal-Or
et al. (2008). It is shown that (Propositions 5 and 6): when (1) the number of retail-
ers is large enough so that a rigid pricing scheme is preferred in (I2) and (2) demand
uncertainty is high enough, setting (I3) is preferred to (I2) by both the manufacturer
and retailers.
As a result, information is shared vertically with the manufacturer who uses it
to determine the wholesale price. Overall, vertical information sharing through the
manufacturer is preferred as a means to collude when the number of retailers is
high or demand uncertainty is high. Further, it is shown that consumer surplus may
actually be lower when the retailers collude by sharing information through the
manufacturer instead of directly with each other. The manufacturer benefits from
the information sharing and this in turn increases the manufacturer’s profit at the
expense of consumer surplus.

15.4 Other Dimensions

15.4.1 Uncertain Costs

Li (2002) also analyzes the case where the uncertainty (and private information)
is about the retailers’ marginal costs Ci . The retailers engage in perfect Cournot
competition. Stage 2 in the basic sequence of events is modified to:
2. Each retailer i observes cost Ci and, if i ∈ K, shares it with M.
The following assumptions are made: (1) Ci ’s are identically distributed with
(normalized) mean 0 and variance σ 2 ; (2) E(Ci |C−i ) = αii + ∑ j=i α ijC j , where
15 Information Leakage in Supply Chains 333

C−i = (C1 , · · · ,Ci−1 ,Ci+1 , · · · ,Cn ) and α ij ≥ 0 for all i and j. Thus, Ci ’s are posi-
tively correlated. The above assumptions are satisfied by the multivariate normal
distribution. Similar to the uncertain demand case, these assumptions lead to the
convenient property that E(Ci |C j , j ∈ K) = (1/(k + s)) ∑ j∈K C j , i ∈ N\K, where
s = (1 − ρ )/ρ and ρ = Cov(Ci ,C j )/σ 2 .
The paper shows: given any K ⊂ N and realized (C j ) j∈K , in a symmetric equilib-
rium, the sales quantities are given by
 
1
q∗i (Ci , w∗ ) = a − w∗ + Ak1 ∑ C j − Ak2Ci , i ∈ K, (15.8)
n+1 j∈K
 
1
∗ ∗
qi (Ci , w ) = a − w + B1 ∑ C j − B2Ci , i ∈ N\K,
∗ k k
(15.9)
n+1 j∈K

for some positive constants Ak1 , Ak2 , Bk1 , and Bk2 . It is then shown that (Proposition 9):
Given any information sharing arrangement K, the manufacturer is better off by ac-
quiring information from more retailers, and each retailer is better off by disclosing
information to the manufacture if ρ < 2(n2 − n − 1)/(2n2 − n − 1); Thus, when this
condition is met, complete information sharing is the unique equilibrium; Other-
wise, the equilibrium can be either complete information sharing or no information
sharing.
This result is markedly different from the uncertain demand case. The driving
force is the indirect (leakage) effect—retailers now benefit from leaking their cost
information to competitors! As evident from Eq. (15.9), it benefits a retailer j ∈ K to
spread the word when his cost is low, i.e., C j < 0, which more than compensates for
the loss from sharing the information when his cost is high. This positive leakage
effect dominates the negative direct effect (of loss of profit to the manufacturer).
When the correlation among the costs is relatively small or the number of retailers
is relatively large (so that ρ < 2(n2 − n − 1)/(2n2 − n − 1)), it is the unique equi-
librium. When ρ is above the threshold, complete information sharing is not the
only equilibrium. However, in the uncertain cost case, the manufacturer can always
purchase information from the retailers and make every party better off (than not
sharing information).

15.4.2 Auction

Chen and Vulcano (2009) consider a supply chain with one supplier and two res-
ellers in a two-stage game. The two resellers engage in Cournot competition, with
the maximum possible demand θ = θ0 + s1 + s2 , where s1 and s2 are independent
and uniformly distributed random variables and are privately and individually obs-
erved by the resellers at the beginning. In the first stage, the supplier auctions her
capacity as a bundle to the resellers. Each reseller bids for the capacity based on their
own demand signal. The supplier announces the winner and the bid. The winner’s
334 G. Kong et al.

bid is disclosed under a first-price auction, whereas the loser’s bid is disclosed under
a second-price auction. In either case, one of the resellers’ information is revealed
through the auction and the other’s is hidden. In other words, the winner has infor-
mation advantage under a second-price auction and the loser is more informed under
a first-price auction. In the second stage, the two resellers compete in the consumer
market. The winner has first-mover advantage in the second stage. The competition
game can be decried as follows:
Under a second-price auction, the winner has full information of the two signals
(sw from the winner and sl from the loser) and his objective can be represented
by: maxqw (θ − qw − ql )qw − c(qw −C)+ , where qw and ql are the winner and loser’s
order quantities, respectively, C is the auctioned capacity, and c is the unit cost in the
spot market. The loser’s decision is only based on his own signal sl and his objective
is: maxql Esw [(θ − qw − ql − c)ql | sw > sl , sl ]. In contrast, under a first-price auction,
the winner and loser’s objectives are: maxqw Esl [(θ − qw − ql )qw − c(qw −C)+ | sw >
sl , sw ] and maxql (θ − qw − ql − c)ql , respectively.
The paper finds that the possibility of revealing the bidders’ private information
leads to lower bids in equilibrium than under the conventional auction without re-
sale, regardless of the auction form. However, the form of auction affects the total
quantity in the consumer market, contingent on the difference in the resellers’ sig-
nals: if the signals are far apart, the first-price auction helps the loser to get access
to the high demand signal from the winner and hence increase his order quantity;
while if the signals are close, the second-price auction helps to maintain a high
order quantity. In addition, as the first-price auction reveals the winner’s private
information and thus decreases his willingness to pay, the supplier gains a higher
payoff under the second-price auction. The second-price auction also improves both
resellers’ payoffs by aligning the winner’s first-mover and information advantages
and reducing the downstream competition.

15.4.3 Competing Supply Chains

Ha et al. (2011) consider two competing supply chains each with one manufac-
turer and one retailer. The retailers have private information about demand and may
choose to share or not share demand with their respective manufacturers. The man-
ufacturer has production diseconomies, i.e. increasing marginal cost. The sequence
of events is as follows in each of the two supply chains: the manufacturer may pay
the retailer for sharing information and the retailer can decide whether to share in-
formation; next, demand is revealed to the retailer which they will share (truthfully)
or not share depending on the decisions at the first stage; the manufacturer then
sets the wholesale price followed by the retailer either deciding quantity (Cournot)
or price (Bertrand)—both models of retail competition are considered. The cost of
information sharing and wholesale price in one supply chain are not available to
the competing one. The paper shows that information sharing in one supply chain
triggers a competitive reaction from the other chain which is damaging to the first
15 Information Leakage in Supply Chains 335

chain in the Cournot model but may be beneficial in a Bertrand model. Information
sharing benefits a supply chain if production diseconomy is large and competition is
less intense in both models. Moreover, a supply chain may be worse off by improv-
ing information accuracy or reducing production diseconomy if it results in the rival
chain not sharing information within it. In the Bertrand model, the manufacturer
may be worse off by receiving information which does not occur in the Cournot
model.
Shamir (2015) considers a structure similar to Ha et al. (2011) with two compet-
ing supply chains comprising of one manufacturer and one retailer each. One chain
is an incumbent with the retailer in that chain having a private signal of demand
information not available to the other chain. Unlike in Ha et al. (2011), it is not
assumed that the retailer will share it truthfully with the manufacturer. The manufac-
turer makes capacity decisions based on the retailer’s information and the wholesale
prices are exogenous. The retailers compete in quantities (Cournot). The sequence
of events is as follows: the incumbent retailer observes a signal; he may (or may not)
share information with only his manufacturer (scenario I1) or publicly (I2); the ent-
rant firms may or may not enter the market; the incumbent or both manufacturers set
capacity levels; the incumbent or both retailers observe market demand and then or-
der quantities. When information is shared only within the incumbent supply chain
(I1), the incumbent retailer has an incentive to manipulate the shared information in
order to secure sufficient capacity. However, when the information is shared with the
competitor as well (I2), the incumbent retailer considers the trade-off between the
benefits of obtaining sufficient capacity in the high demand scenario and the cost
of more intense competition with the entrant. A key result in the paper is that by
making information available to the competitor, it is possible to achieve separation
between a retailer observing a high demand and a low demand. The retailer benefits
sufficiently in the high demand scenario from the increased capacity although this
comes with increased competition. In the low demand case, the retailer benefits by
truthfully revealing his demand signal as it weakens retail competition. The paper
also shows that the incumbent retailer may prefer to share information publicly rel-
ative to committing to a minimum purchase quantity as part of an advance purchase
contract.

15.5 Discussion and Future Research

15.5.1 Discussion of the Existing Literature

A number of interesting and common insights can be gleaned from the literature
about information sharing and leakage. As discussed in the introduction, leakage of
information (or at least the fear that there will be leakage) often occurs in a supply
chain when a retailer shares information vertically with a manufacturer. The liter-
ature, focusing primarily on demand information, has shown that vertical sharing
336 G. Kong et al.

of information by retailers with manufacturers always results in a negative effect

on the retailers as the manufacturer extracts some of the surplus—this is true in
both Cournot and Bertrand models of competition at the retailer level. However,
the effects of the manufacturer leaking this information to retailers (horizontal shar-
ing) can have a positive or negative effect, respectively, depending on whether the
competition among retailers is on price or quantity and whether their products are
substitutes or complements. The literature has shown that the negative effects of
vertical sharing are dominant enough that no information sharing is the equilib-
rium outcome in the majority of scenarios. This is shown to be true independent
of whether information sharing arrangements are made before or after private in-
formation is revealed. The level of accuracy of private information among different
entities also plays an important role. In general, when retailers have similar levels of
accuracy or the accuracy is low, information sharing is more likely and leakage less
likely. When information accuracy among retailers is asymmetric, the manufacturer
is more likely to leak to the less informed retailer. Finally, the results also depend on
whether two competing retailers move simultaneously or sequentially and the level
of demand uncertainty.
To incentivize truthful information sharing despite the potential for leakage and
its negative effects, the literature has come up with a variety of solutions: side pay-
ments by manufacturers to retailers (Li 2002), different wholesale prices charged to
different retailers (Jain et al. 2011), revenue sharing contracts (Kong et al. 2013),
market-based contracts (Shin and Tunca 2010) and costly actions (signals) by re-
tailers (Shamir 2012). In addition, retailers can enter into a binding confidentiality
agreement to prevent leakage which in turn facilitates information sharing. Some
of these solutions can coordinate the supply chain and sometimes benefit all the
entities, including ones that do not have private information. A few recent papers
also suggest that retailers may have less to fear and may even benefit from leakage
of information to their competitors and such sharing may even serve as a collusion
mechanism (Shamir 2012, 2015). However, these results are true only when the
retailers compete on prices and only under certain conditions.
While anecdotal evidence from industry as discussed in the introduction sug-
gests that firms primarily fear direct leakage of information, the academic literature
has shown that indirect leakage of information can have as significant an effect on
the incentives to share information as direct leakage. Furthermore, while real-world
firms primarily worry about the negative effects of information being leaked to their
competitors (horizontal sharing), academic research suggests that vertical informa-
tion sharing always has negative effects while horizontal sharing and leakage may
have a positive or negative effect. However, this is primarily because the academic
literature reviewed here has focused on the negative effects of vertical information
sharing and not considered the benefits. In the real world, some of the solutions
offered in the literature such as side payments by manufacturers, revenue sharing
contracts and confidentiality agreements are adopted to address the issue of infor-
mation leakage as well as other potential negative effects of information sharing.
15 Information Leakage in Supply Chains 337

15.5.2 Other Supply Chain Structures

The current literature on information sharing and leakage has focused on the supply
chain structure with a single manufacturer (supplier) and multiple retailers who have
private information on the market demand, as illustrated in Fig. 15.1. In reality, a
retailer often buys a type of product from multiple manufacturers (suppliers) and the
manufacturers, as well as the retailer, have private knowledge about certain aspects
of the consumer market. Thus, the supply chain structure illustrated in Fig. 15.2 is
also commonplace, and a natural extension to the current literature is to study this
alternative supply chain structure in which the retailer is located at a pivotal junction
of the information network.

Fig. 15.2 Another common supply chain structure, with one retailer and multiple manufacturers

The first paper to study information flow in such a supply chain is Shang et al.
(2015). They consider two manufacturers engaging in a Bertrand competition who
supply to one retailer with private information about market demand. The retailer
decides whether to share the information with each manufacturer. Their model mir-
rors the “one-sided information sharing” model of Gal-Or et al. (2008) (consisting
of one informed manufacturer and two uninformed retailers) but with notable dif-
ferences in the game being played, which leads to different insights. Both models
capture information disclosure by the informed party to the uninformed ones, and,
strictly speaking, are about information sharing rather than leakage. The “two-sided
information sharing” model of Gal-Or et al. (2008) seems to be a mirror image of
the new model we are proposing. However, they only assume two retailers, one of
which has perfect information and the other has no information. More general set-
tings are worthy of investigation.
We note that the models with mirrored supply chain structures are not really the
mirror images of each other. As the wholesale prices are often set (by manufacturers)
before the retail prices are set (by retailers), the change of ownership of private
information results in a change of the sequence of events in terms of informed and
uninformed parties, which may lead to different conclusions as evident from Shang
et al. (2015) and Gal-Or et al. (2008).
338 G. Kong et al.

The new supply chain structure introduces new research challenges and oppor-
tunities regarding the type of private information. In the existing literature, mar-
ket demand is the predominant information under consideration, except Li (2002)
who also considers private marginal costs of the informed parties (retailers). In the
alternative supply chain, the manufacturers’ private production costs pose more int-
eresting questions. Similarly, production capacities that affect manufacturers’ com-
petitiveness and profitability are important, yet often private, information. Qian et al.
(2012) introduce this dimension to the original supply chain structure and show
that if the manufacturer has a capacity constraint, full information sharing can be
induced by a discriminative supply rule, i.e., allocating a significantly larger quan-
tity to retailers participating in information sharing when the total demand exceeds
capacity. In the alternative supply chain, the capacities of competing manufacturers
are valuable information to share or leak.
A more general supply chain structure consists of multiple retailers and multiple
manufacturers, as illustrated in Fig. 15.3. This is representative of the real world
where, for instance, manufacturers such as Procter and Gamble (P&G) and Colgate
Palmolive compete with each other but also supply to retailers such as Target and
Wal-Mart who in turn compete for consumers. It will be interesting to study the
possibility of information sharing among some members of the supply chain given
the possibility of leakage in such a setting, as it raises a host of new questions. For
example, as part of initiatives such as CPFR, Target may collaborate with P&G to
forecast demand and plan replenishment quantities. Suppose Colgate is planning a
promotion in the near future which may impact P&G’s sales at Target negatively.
This puts Target in a bind as it has to collaborate with P&G to plan demand and
order quantities but cannot share Colgate’s promotional plans. Conversely, the same
type of issue can also arise if Target is planning to promote P&G’s products which
may impact demand for P&G’s products at Wal-Mart but P&G may be unable to
share this valuable information with Wal-Mart. In this case, information sharing
is impossible without leakage and confidentiality agreements will not resolve the
problem. There are many interesting issues of this nature that require further study
in such multiple manufacturer, multiple retailer networks.

15.5.3 Other Types of Contracts

The existing information sharing and leakage literature has focused on the wholesale
price contract between manufacturers and retailers, partly due to its prevalence in
the real world and partly because of its analytical tractability. However, as evident
from Kong et al. (2013), the use of different types of contracts such as the rev-
enue sharing contract may change the results completely, e.g., from leaking infor-
mation to not leaking. Revenue sharing contracts are popular in some industries
such as entertainment, sports leagues, and software (see e.g., Dana and Spier 2001).
Other types of contracts, such as buy back and quantity discount, are also adopted
in the real world (Cachon 2003). As shown by Li and Zhang (2008) and others,
15 Information Leakage in Supply Chains 339

Fig. 15.3 A more general supply chain structure, with multiple manufacturers and retailers

if the intended receiver of some private information can commit not to leak that
information to a nonparticipating party, the owner of that information will be more
willing to share it. The findings of Kong et al. (2013) reinforce this insight. The rev-
enue sharing contract can align the incentives of supply chain partners better than
the wholesale price contract, so the former is more likely to motivate collaboration
between the receiver and sender of the information to keep it confidential, which
will facilitate information sharing in the first place. We conjecture that other types
of contracts or arrangements that help align the incentives of supply chain partners
can achieve similar outcomes, which is worthy of future research.

15.5.4 Long-Term Relationship

All papers discussed in this chapter except Shamir (2015) assume one opportunity
for each retailer to order from the manufacturer and sell in the consumer market.
That is, there is only a single period, albeit multiple stages of interactions. Under
such a model, unless a binding confidentiality agreement is in effect, the manu-
facturer faces no explicit consequence for leaking the retailers’ private informa-
tion. In reality, supply chain members tend to maintain a long-term relationship,
and leaking partners’ information without their consent will damage their relation-
ship and threaten future business opportunities. Faced with possible retaliation from
the retailers in the future, from refusing information sharing to ceasing the busi-
ness partnership, the manufacturer will be more conservative about leaking retail-
ers’ information. Thus, concerns about long-term relationships and reputation may
provide sufficiently strong incentives for the manufacturer to protect the retailers’
information voluntarily, which can substitute for a legal confidentiality agreement.
On the other hand, under some circumstances, the manufacturer may find it even
more tempting to leak some retailers’ information to others, especially when such
340 G. Kong et al.

information has significant value to other retailers in the future. Long-term relation-
ships and repeated interactions add reality to the model, but introduce new chal-
lenges to the analysis as well. They represent another interesting future research
direction.

References

Adewole A (2005) Developing a strategic framework for efficient and effective optimization of
information in the supply chains of the UK clothing manufacture industry. Supply Chain Manag
Int J 10(5):357–366
Anand KS, Goyal M (2009) Strategic information management under leakage in a supply chain.
Manag Sci 55(3):438–452
Cachon G (2003) Supply chain coordination with contracts, Chap. 6. In: de Kok AG, Graves SC
(eds) Handbooks in operations research and management science: supply chain management.
North-Holland, Amsterdam, pp 229–340
Cachon G, Lariviere M (2005) Supply chain coordination with revenue-sharing contracts: strengths
and limitations. Manag Sci 51(1):30–44
Chen F (2003) Information sharing and supply chain coordination, Chap. 7. In: de Kok AG, Graves
SC (eds) Handbooks in operations research and management science: supply chain manage-
ment. North-Holland, Amsterdam, pp 341–413
Chen Y, Vulcano G (2009) Effects of information disclosure under first- and second-price auctions
in a supply chain setting. Manuf Serv Oper Manag 11(2):299–316
Dana JD Jr, Spier KE (2001) Revenue sharing and vertical control in the video rental industry. J Ind
Econ 49(3):223–245
Deshpande V, Schwarz L, Atallah M, Blanton M, Frikken K, Li J (2010) Secure-computations
for collaborative planning, forecasting and replenishment (SCPFR). Working paper, Purdue
University
Douglas M (2004) Trust me! The human side of collaboration. Inbound Logistics (Jan). Accessed
21 Jun 2015
Gal-Or E, Geylani T, Dukes AJ (2008) Information sharing in a channel with partially informed
retailers. Mark Sci 27(4):642–658
Ha A, Tong S, Zhang H (2011) Sharing demand information in competing supply chains with
production diseconomies. Manag Sci 57(3):566–581
Jain A, Seshadri S, Sohoni M (2011) Differential pricing for information sharing under competi-
tion. Prod Oper Manag 20(2):235–252
Jain A, Sohoni M (2015) Should firms conceal information when dealing with common suppliers?
Nav Res Logist 62(1):1–15
Kong G, Rajagopalan S, Zhang H (2013) Revenue sharing and information leakage in a supply
chain. Manag Sci 59(3):556–572
Kurtulus M, Toktay B (2008) Category captainship practices in the retail industry, Chap. 7. In:
Agrawal N, Smith SA (eds) Retail supply chain management: quantitative models and empirical
studies. Springer, New York, pp 147–174
Lee HL, Whang S (2000) Information sharing in a supply chain. Int J Technol Manag 20:373–387
Li L (1985) Cournot oligopoly with information sharing. Rand J Econ 16(4):521–536
Li L (2002) Information sharing in a supply chain with horizontal competition. Manag Sci
48(9):1196–1212
Li L, Zhang H (2008) Confidentiality and information sharing in supply chain coordination. Manag
Sci 54(8):1467–1481
Mello JP Jr (2012) Hackers attack Foxconn for the laughs. Macworld (9 Feb). Accessed 21 Jul
2015
15 Information Leakage in Supply Chains 341

Qian Y, Chen J, Miao L, Zhang J (2012) Information sharing in a competitive supply chain with
capacity constraint. Flex Serv Manuf J 24(4):549–574
Shamir N (2012) Strategic information sharing between competing retailers in a supply chain with
endogenous wholesale price. Int J Prod Econ 136(2):352–365
Shamir N (2015) Cartel formation through strategic information leakage in a distribution channel.
SSRN working paper. http://ssrn.com/abstract=2292410
Shamir N, Shin H (2015) Public forecast information sharing in a market with competing supply
chains. Management Sci. Published online, Dec 18, 2015. http://dx.doi.org/10.1287/mnsc.
2015.2261
Shang W, Ha AY, Tong S (2015) Information sharing in a supply chain with a common retailer.
Manag Sci (June 5) articles in advance. doi: 10.1287/mnsc.2014.2127
Shin H, Tunca T (2010) Do firms invest in forecasting efficiently? The effect of competition on
demand forecast investments and supply chain coordination. Oper Res 58(6):1592–1610
Ye Q, Duenyas I, Kapuscinski R (2013) Should competing firms reveal their capacity? Nav Res
Logist 60(1):64–86
Zhang H (2002) Vertical information exchange in a supply chain with duopoly. Prod Oper Manag
11(4):531–546
Chapter 16
Bilateral Information Sharing and Pricing
Incentives in a Retail Channel

Anthony Dukes, Esther Gal-Or, and Tansev Geylani

Abstract This chapter evaluates the impact of sharing information on wholesale and
retail pricing incentives as well as on the distribution of economic rents. We consider
a model in which the manufacturer distributes its product to one or more retailers.
Each firm receives a private signal as an estimate of stochastic consumer demand.
We show that, in the absence of information sharing, the retailer is able to use the
wholesale price to infer the manufacturer’s private signal. This creates a pricing dis-
tortion which benefits the retailer. Downward sharing of the manufacturer’s private
signal eliminates this distortion. In contrast, when the retailer shares its private sig-
nal upstream, the manufacturer is able to set price closer to retailer’s value, thus
capturing downstream consumer surplus. In general, the manufacturer benefits from
more information sharing at the loss of downstream retailers and consumers. Hence,
information sharing arrangements in equilibrium require side payments and/or suf-
ficient cost savings (e.g., reduced inventory costs).

16.1 Introduction

While retailers typically have superior knowledge about local demand trends, they
are often not privy to national trends or category specific events. Manufacturers, on
the other hand, have information about demand not possessed by other members
of the retail channel. For instance, large CPG manufacturers have category-specific

A. Dukes ()
University of Southern California, Marshall School of Business, Los Angeles, CA 90089, USA
e-mail: [email protected]
E. Gal-Or • T. Geylani
University of Pittsburgh, Katz Graduate School of Business, Pittsburgh, PA 15260, USA
e-mail: [email protected]; [email protected]

© Springer International Publishing Switzerland 2017 343

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 16
344 A. Dukes et al.

knowledge or engage in marketing research at a national level to gauge the demand

for their products. This chapter explores the implications for sharing demand infor-
mation when both the retailer and the manufacturer have private information about
the state of demand.
Because the operational benefits (e.g., reduction in supply chain costs) associated
with information sharing in a retail channel are well-known (cf., Lee et al. 1997,
2000; Cachon and Fisher 2000), we focus on the benefits of information sharing
derived strictly from pricing incentives. Abstracting away from operational benefits,
this chapter examines who stands to gain from sharing private demand information
and the conditions necessary for an information sharing arrangement to arise. We
first consider the case in which information transmission is only from the manufac-
turer to the retailer. In this case the retailer has a better forecast of consumer demand
and is better able to “right-price” to consumers. However, in absence of this signal,
the retailer may still be able to infer the manufacturer’s signal through the wholesale
price. The manufacturer’s optimal pricing rule will account for this inference, which
as we show, has benefits for the retailer. Our results demonstrate that the retailer will
only share information if such sharing leads to sufficient operational benefits (e.g.,
due to a reduction of required inventory or order costs).
Consider further the case in which the retailer and the manufacturer bilaterally
exchange signals so that both channel members have improved forecasts about the
state of consumer demand. In this case both firms can set prices closer to buyer value
but the manufacturer is the sole beneficiary since he moves first and can fully ant-
icipate the retailer’s pricing reaction. No retailer, therefore, would rationally agree
to reveal her signal. However, if total channel surplus increases, then it may be
mutually beneficial for the manufacturer to broker a deal with the retailer for the inf-
ormation exchange. We provide conditions under which this is possible.
In our modeling framework, consumer demand is subject to a stochastic shock
which is privately and imperfectly observed by the manufacturer and its retailer(s).
That is, each firm receives a private and noisy signal concerning a common stochas-
tic demand parameter. Before observing signals, firms can enter into an arrange-
ment that binds one or both of them to truthfully share the signal realization. In
one such agreement, which we call downward sharing (DS), the manufacturer com-
mits to unilaterally revealing his private signal to the retailer(s). In another agree-
ment, which we call bilateral, or full, sharing (FS), private signals are bilaterally
exchanged between the manufacturer and retailer(s).
In absence of cost savings, information sharing implies two effects on pricing
behavior. These two effects unambiguously serve to benefit the upstream firm. One
is due to a pricing distortion, which is known as the “inference effect” (Gal-Or et al.
2008). The inference effect reflects a retailer’s use of the manufacturer’s wholesale
price to infer the manufacturer’s private signal when there is no commitment to ex-
change signals. This causes the manufacturer to distort wholesale price downward,
to the benefit of the retailer and consumers. When the manufacturer shares this infor-
mation with the retailer, the inference effect is alleviated, allowing the manufacturer
to set a higher wholesale price. Alleviating the inference effect comes at the loss of
retailer profit and consumer surplus.
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 345

The second effect relates to the ability of firms to price closer to the economic
value of their respective buyer. In particular, when a retailer has a better estimate of
demand (by pooling its own signal with that of the manufacturer) it is able to set
retail prices more reflective of true demand conditions. Similarly, the manufacturer,
with the two signals, sets wholesale price more reflective of the retailer’s value. That
is, with full information sharing, wholesale prices fluctuate more synchronously
with retail prices and retailer prices fluctuate more synchronously with demand—an
effect (shown in Dukes et al. 2011) known as the “congruence effect” that benefits
the manufacturer.
Because information exchange benefits the manufacturer at the expense of the
retailer and consumers, the retailer must be compensated for sharing its signal up-
stream. We present conditions when this might occur. In particular, if the retailer is
not compensated with lower operational and logistic costs, then there is the question
of whether the manufacturer is willing to compensate the retailer with a payment.
However, Sect. 16.3.3 of this chapter shows that this is feasible only when there is a
bilateral exchange of information. With downward sharing only, inefficiencies due
to double marginalization are exacerbated and, despite the benefits to the manufac-
turer, overall channel profits decrease. Therefore, in absence of costs savings, such
a sharing arrangement is not feasible.
In Sect. 16.4, we extend the model in order to examine how the magnitude of
these two effects is altered by retail competition. The extension indicates that the
inference effect is fundamentally connected to economic inefficiencies associated
with double-marginalization, which is alleviated with competition. The congruence
effect, however, remains present even when competition is perfect.
There is an established literature in economics which examines the implications
of pricing incentives resulting from the sharing demand information. Much of the
early work focused on exchange between horizontal competitors (e.g., Gal-Or 1985;
Li 1985; Novshek and Schonneschein 1982; Vives 1984; Raith 1996), which typi-
cally assumes simultaneous pricing decisions amongst rival firms. A key distinction
in a retail channel is the sequential nature of pricing decisions. In particular, strategic
choices of the manufacturer can reveal information to the retailer about the manu-
facturer’s private signal.
The operations and marketing literatures, in contrast, have studied the incentives
for information sharing within the supply chain (e.g., Li 2002; Niraj et al. 2007;
Ha and Tong 2008; Mittendorf et al. 2013). Li (2002), in particular, examines how
retailers’ private demand (and cost) information affects prices and the incentive to
share. However, most of this literature has not delved into the pricing incentives
when the manufacturer also has private information on the state of demand, which it
may share. One exception is He et al. (2006). They investigate information sharing
when the manufacturer has private information, but their model does not allow the
retailers to infer the manufacturer’s private information from the wholesale price
and are thus unable to detect the pricing distortions caused by the inference effect.1
1 Another related paper is that of Guo and Iyer (2010), who examine sequential consumer pref-
erence information acquisition by a manufacturer and the subsequent sharing of this information
with the downstream retailer.
346 A. Dukes et al.

Relating to the inference effect is a large signaling literature in economics (e.g.,

Spence 1973; Milgrom and Roberts 1980; Gal-Or 1987). This literature implies
distortions in strategic decisions when parties are asymmetrically informed. This is
the case with the inference effect as well, whereby the manufacturer distorts its price
relative to the full information case in order to account for inferences made by the
retailer.
The remainder of this chapter proceeds as follows. In Sect. 16.2 we present the
modeling framework that is used throughout the analysis. Section 16.3 develops
the main results through a single manufacturer-retailer model. Section 16.4 ex-
tends the framework to competing retailers. We conclude in Sect. 16.5, which is
followed by the appendix, which contains the proofs of all propositions and corol-
laries.

16.2 Modeling Framework

Our modeling framework borrows from Dukes et al. (2011). There is a manufac-
turer (M) who distributes his2 product through a single retailer (R).3 For any given
wholesale transaction there are fixed costs incurred by both parties. Let Fi , i = M, R
denote costs associated with placing and delivering an order. We assume that the
marginal costs for the manufacturer and the retailer are zero.
Both the retailer and the manufacturer have private, but imprecise, signals about
the state of demand. By sharing these signals, the manufacturer and the retailer may
reduce their costs Fi . We explore the implications of sharing the realizations of these
signals on firms pricing decisions. The manufacturer offers a uniform wholesale
pricew to the retailer. This is a-take-it-or-leave-it offer in the sense that there is no
negotiation of wholesale terms.4
Following the manufacturer’s wholesale pricing decision, the retailer sets its
price to consumers. The retailer R faces a linear stochastic demand:

q = a − br + u, a > b > 0, (16.1)

where u ∼ N(0, σ ). Both firms observe noisy signals of u. The signal observed by
the manufacturer is xM and by the retailer xR . Specifically, we assume xi = u + εi ,
i = M, R, and (u, εR , εM ) ∼ N(0, diag(σ , s, s)). Notice that in the above formula-
tion, the noise in a signal is captured by s, where higher s indicates more noise.
We assume s and σ to be known by both firms. It is also possible to consider the
2 For convenience, we use the gender pronouns “he” and “she” to refer to the manufacturer and

retailer, respectively.
3 Later in Sect. 16.4, we consider the case of two competing retailers.
4 Since the existence of double marginalization is the main reason for providing incentives to share

information in our model, we do not consider nonlinear pricing (two part tariffs, for instance),
which eliminates double marginalization altogether. Nonlinear pricing may not be feasible when
the manufacturer cannot prevent secondary, resale markets for his products or when signing long
term contracts between the manufacturer and retailer is not possible.
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 347

case when firms differ in the precision of their signals sM = sR (see Gal-Or et al.
2008). However, for simplicity we assume here identical noise levels. The model’s
information structure is illustrated in Fig. 16.1.

Fig. 16.1 Information

structure Manufacturer: xM
xM = u + ε M , ε M ~ N (0, s )

w
Retailer: xR
xR = u + ε R , ε R ~ N (0, s )

r
Consumers
q = a − br + u

cov(ε M , ε R ) = cov(ε M , u ) = cov(ε R , u ) = 0

The sequence of events and decisions are as follows. First, M and R decide
whether to share information. Subsequent to the information sharing decision, na-
ture determines the demand signals xM and xR and channel members observe a sub-
set of these realizations. Specifically, the manufacturer and retailers observe their
own private signals and any signals which they are privy to as determined by the
information sharing arrangement made in the first stage. Third, after the observation
of signal realizations, the manufacturer sets the wholesale price. Last, the retailer
sets her price.
We distinguish between two types of information sharing arrangements: down-
ward and bilateral sharing. With downward sharing, information is transmitted only
from the manufacturer to the retailer. Specifically, downward information sharing
implies that a retailer is informed of the manufacturer’s private signal xM . In
Sect. 16.3.1, we develop the model in the case of downward sharing and compare
it to the no-sharing case. Subsequently, in Sect. 16.3.2, we extend the analysis to
bilateral information sharing. In this case, information is transmitted not only from
the manufacturer to the retailer but also from the retailer to the manufacturer. Thus,
in the full sharing case, a decision to share information with retailer R implies that
the manufacturer and retailer both observe xM and xR in stage 2.
Several assumptions regarding the timing of the game require further explana-
tion. First, the manufacturer can adjust his wholesale price wcontingent upon the
demand signal xM that he observes. Since w can be chosen as a function of xM , it
follows that a retailer can draw inferences about the realization of the demand sig-
nal based upon the observation of w even when the manufacturer chooses not to
communicate directly with the retailer.
348 A. Dukes et al.

The second assumption that is implied by the timing is that the manufacturer and
retailer commit to the information sharing policy before observing his or her own
signal. This assumption presumes that the installation of the infrastructure necessary
to support the communication is a long term decision variable that predates the
actual realization of the demand signals. In particular, once the information sharing
arrangement has been established, both the manufacturer and retailer are committed
to delivering truthfully any signal realizations that they may receive.
Each agent in the model bases its pricing decision on his or her own observed
signal, those learned through information sharing, and in the case of the retailers, on
the manufacturer’s wholesale price. We thus define a decision rule for each agent
as a mapping fi , i = M, R from the space of possible signal values and, in the case
of retailers, possible wholesale prices. Specifically, the manufacturer’s decision rule
specifies the wholesale price using w = fM (xM ), where xM = (xM , . . .) is a vec-
tor of observed signals, which always includes the signal xM . In the case of full
information sharing, xM includes the signal of the retailer and, in Sect. 16.4, that
of a second retailer. The retailer decision rule specifies a retail price r = fR (w, xR ),
where xR = (xR , . . .) is a vector of observed signals, which always includes the signal
xR and in the case of downward or bilateral information sharing the manufacturer’s
signal xM .
The joint normal distribution of the random variables and quadratic payoffs in
our model leads to a unique Bayesian Nash equilibrium with all parties following
linear decision rules.5 Using this result, we restrict attention in our analysis to linear
decision rules fi (·) followed by each agent. The exact form of this linear function is
determined by matching it with equilibrium conditions of the particular game under
consideration.

16.3 Single Retailer

In this section we first explore the implication of information sharing when the
manufacturer, M, shares his signal with the retailer R. We do so by comparing two
information regimes—No Sharing (NS) and Downward Sharing (DS). Later in this
section, we compare these cases to that of Bilateral (Full) Sharing (FS) in which
signals are exchanged in both directions.

16.3.1 No Sharing and Downward Sharing

The benchmark situation is when there is no sharing (NS) arrangement between

firms. In this case, the manufacturer bases his wholesale price w solely on the real-
ization of his own signal xM . Denote his decision rule for w by

5 See Gal-Or et al. (2008) for more details.

16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 349

w = fMNS (xM ) = μ0 + μ1 xM , (16.2)

where coefficients μ0 and μ1 are to be determined by equilibrium conditions. The

retailer, on the other hand, in her choice of retail price r has access to her own signal
observation xR but also to the strategic action of the manufacturer, namely w. As
such, we can write her decision rule for r as

r = fRNS (w, xR ) = ρ0 + ρ1 w + ρ2 xR . (16.3)

In stage 3, the retailer chooses r to maximize expected profit, conditional on observ-

ing w and xR :

E(ΠR |w, xR ) = (r − w)E(q|w, xR )

= (r − w)[a − br + E(u|w, xR )]. (16.4)

Note that when the manufacturer does not commit to sharing its information, the
retailer can infer M’s signal, xM , by observing w. In general, if both firms’ signals
are known, then the expectation of u conditional on the pair of signals is

σ (xR + xM )
E(u|xR , xM ) = , (16.5)
2σ + s
which follows from the normality stochastic terms (see DeGroot 1970). Note that
as the signal precision improves (variance s ↓ 0), the conditional expectation of u
converges to the simple average of the signals. Otherwise, given noise in the signals
(s > 0), the retailer’s conditional expectation is reduced to some degree.
Despite the fact that the retailer is not privy to the realization xM , as long as
μ1 = 0, the she can invert (16.2) and solve for xM = (w − μ0 )/μ1 . Then the expec-
tation in (16.5), expressed in terms of w, is written

σ (xR + (w − μ0 )/μ1 )
E(u|w, xR ) = , (16.6)
2σ + s
which indicates how the retailer’s conditional expectation of u depends on M’s
choice of w. Because agents are rational, they are able to deduce optimal decision
rules in equilibrium. In particular, the retailer who knows μ0 and μ1 can fully infer
the realization of the manufacturer’s signal, xM .
The fact that (16.6) depends on w implies that R, when choosing r, will react
to w not only as a marginal cost of selling, but also as an inference devise. The
manufacturer must account both effects in his optimal choice of w in stage 2. To
evaluate this trade-off, the manufacturer uses his expectations of (16.3)

E(r |xM ) = ρ0 + ρ1 w + ρ2 E(xR |xM ),

when maximizing

E (ΠM |xM ) = w [a − bE(r |xM ) + E(u|xM )] . (16.7)

350 A. Dukes et al.

The conditional expectations of u and xR in (16.7) are derived as

σ
E (xR |xM ) = E (u + εR |xM ) = E (u|xM ) = xM , (16.8)
σ +s
where the last equality follows again from normality (DeGroot 1970). Note again,
that as signal precision improves, the manufacturer’s conditional expectation of the
stochastic demand parameter approaches his own signal.
To derive the equilibrium decision rules, we use the optimizations of (16.4) and
(16.7) and match coefficients with (16.2) and (16.3). Specifically, because the opti-
mal r must maximize (16.4), it must satisfy the first order condition, which implies
   
1 σ μ0 1 σ 1 1 σ
r(w, xR ) = a− + b+ w+ xR , (16.9)
2b 2σ + s μ1 2b 2σ + s μ1 2b 2σ + s

where ρ0 , ρ1 , and ρ2 correspond to the first term, the coefficient of w and the coef-
ficient of xR , respectively. Similarly, using the first order condition of the manufac-
turer, we have the optimal w:

a − bρ0 σ 1 − bρ2
w= + xM , (16.10)
2bρ1 σ + s 2bρ1
where μ0 and μ1 correspond to the first term and the coefficient of xM , respectively.
Matching coefficients using (16.9) and (16.10) imply a system of five equations,
which can be solved to derive the five coefficients characterizing the equilibrium
decision rules with no information sharing:

a σ σ2
wNS = fMNS (xM ) = + xM (16.11)
2b 2σ + s 2b(σ + s)(2σ + s)
a σ 3σ + 2s NS σ
rNS = fRNS (w, xR ) = + w + xR , (16.12)
2b 2σ + s 2σ 2b(2σ + s)

where the superscripts denote equilibrium values in the no-sharing (NS) regime.
Central to the main result, is the fact that M must account for the fact that R will
use her observation of the wholesale price w to draw an inference about xM . This
distorts the pricing incentive of the manufacturer relative to full information. This
can be seen in R’s equilibrium reaction to wholesale price. With full information, the
coefficient of w would be 1/2. But in (16.12), the coefficient of wNS exceeds 1/2.
Hence, when there is no information sharing, the retailer is more sensitive to whole-
sale price than otherwise.
This added sensitivity to wholesale price has adverse consequences to the man-
ufacturer. The retailer infers high states of demand with higher wholesale prices.
This exacerbates channel inefficiencies attributed to double-marginalization. Con-
sequently, the manufacturer must distort his wholesale price downward, relative
to the full information case. We refer to this phenomenon as the inference effect
(Gal-Or et al. 2008).
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 351

If the manufacturer, however, communicates his signal to the retailer before set-
ting wholesale price, then he avoids the adverse consequence of the inference effect.
We compare the NS regime discussed above with one in which the manufacturer in-
forms the retailer of his signal xM . In this case, information is exchanged in one
direction only—downward. Hence, we refer to this case as downward sharing, ab-
breviated DS. Later, in Sect. 16.3.2, we compare these regimes to a full exchange of
information in which both firms fully share their information.
To analyze the DS regime, we consider decision rules

w = fMDS (xM ) = μ0 + μ1 xM , (16.13)

r= fRDS (w, xR , xM ) = ρ0 + ρ1 w + ρ2 xR + ρ3 xM , (16.14)

where the retailer now reacts to xM directly, rather than through an inference. The
retailer maximizes expected profits, which in the DS regime are expressed as:

E (ΠR |xR , xM ) = (r − w)E (q|xR , xM )

= (r − w) [a − br + E (u|xR , xM )] . (16.15)

Then the conditional expectation in (16.15) is computed directly from realiza-

tions (xR , xM ) as stated in (16.5). The manufacturer, on the other hand, has the
same information as before, so that his expected profits are as given in (16.7).
The distinction in this case, however, is his expectation of retail price incorpo-
rates the fact that the retailer reacts directly to xM by using (16.14): E(r |xM ) =
ρ0 − ρ1 w + ρ2 E(xR |xM ) + ρ3 xM .
Determining the equilibrium decisions rules, as before, can be deduced by using
the two firms’ first order conditions and matching coefficients to those in (16.13)
and (16.14). This leads to a system of six equations, which can be solved for the six
coefficients.
a σ
wDS = fMDS (xM ) = + xM (16.16)
2b 2b(σ + s)
a 1 σ
rDS = fRDS (w, xR , xM ) = + wDS + (xR + xM ). (16.17)
2b 2 2b(2σ + s)

The absence of the inference effect in the DS case can be seen in (16.17) by the fact
that the coefficient on wDS is equal to 1/2. Proposition 1 offers a comparison of these
two regimes.

Proposition 1. Relative to no information sharing, downward sharing of informa-

tion leads to
(i) Higher expected prices: E(wNS ) < E(wDS ) and E(rNS ) < E(rDS ); and
(ii) Higher expected profits for the manufacturer: E(ΠM DS ) > E(Π NS ) and, gross of
M
any cost savings, lower expected profits for the retailer: E(ΠRDS ) < E(ΠRNS ).

Proposition 1 confirms that the inference effect harms the manufacturer in the
form of lower wholesale prices, w, but benefits the retailer. These conflicting
352 A. Dukes et al.

incentives imply that the retailer will accept to receive demand information only
if her cost saving from improved information is significantly high:
5σ + 3s
FRNS − FRDS > E(ΠRNS − ΠRDS ) = [a2 (σ + s) + σ 2 ] > 0 (16.18)
16b(2σ + s)2

Note that a larger market size (as measured by the parameter a) raises the threshold
for R to be willing to accept M’s private signal. This is a consequence of the assump-
tion on linear wholesale prices. Because the inference effect keeps wholesale prices
lower in the NS regime, it affects all units of the sale. The larger the market, the
larger R’s benefit from the inference effect. From the manufacturer’s perspective, a
larger consumer market means more benefit from sharing its signal downward.
Because any potential cost savings [left side of (16.18)] of downward sharing
is fixed with respect to quantity, it has no impact on retail prices and, therefore,
none on consumer surplus. As a result, any effect on consumer surplus is strictly an
implication of information sharing on pricing incentives. To investigate this effect
on consumers, denote by CSk as the total consumer surplus in equilibrium in regime
k = NS, DS. Then  r̄
CSk = (a − br + u) dr, (16.19)
rk
where r̄ = sup{r > 0|q = a − br + u > 0}. In the following corollary, we establish
that higher prices in NS relative to DS lead to lower expected consumer surplus.

Corollary 1. Relative to no information sharing, the downward sharing of informa-

tion leads to lower expected consumer surplus: E(CSDS −CSNS ) < 0.

The corollary implies that, in sharing arrangements in which a retailer is privy to

manufacturers’ information about the state of demand (and the manufacturer knows
the retailer possesses this information) consumers will face higher prices on average
and incur less expected consumer surplus. Therefore, under the decision criterion of
(16.18), all social benefits from (downward) information sharing are entirely inter-
nal to the two firms, R and M and leave consumers with unambiguously less surplus.
It is important to draw the distinction between the impact of the inference and the
situation in which uncertainty about a buyer’s valuation delivers informational rents
to high value buyers. In particular, the inference arises out of a firm’s reaction (the
retailer’s) to previously taken action (the manufacturer’s) when there is asymmetric
information. As identified in previous literature (Spence 1973; Milgrom and Roberts
1980; Gal-Or 1987), relative to full information, asymmetric information can lead
to distortions in strategic behavior in sequential move settings. The inference effect
is exactly such a distortion which happens to benefit consumers.
Note however, that we examined the case when the information is shared unilat-
erally downward. When, in Sect. 16.3.2, we consider bilateral information sharing,
the issue of shifting informational rents becomes present.
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 353

16.3.2 Bilateral (Full) Sharing

In the bilateral sharing (FS) scenario, both firms agree (and commit) to share their
private signals upon realization. Therefore, both firms make use of both signals xR
and xM in their pricing decisions:

w = fMFS (xM , xR ) = μ0 + μ1 xM + μ2 xR (16.20)

r= fRFS (w, xR , xM ) = ρ0 + ρ1 w + ρ2 xR + ρ3 xM . (16.21)

Furthermore, as in the DS case R makes no inferences from observing the strategic

decision of w by M. The fact that there is no inference effect means that the man-
ufacturer will not suffer the adverse pricing distortions incurred in the NS regime.
However, the question remains as to whether knowledge of the retailer’s private sig-
nal favor the manufacturer even further, or work against the benefits of the inference
effect.
It can be shown using the similar analysis as used in the DS regime, that the
equilibrium decision rules are
a σ
wFS = fMFS (xM , xR ) = + (xM + xR ) (16.22)
2b 2b(2σ + s)
a 1 σ
rFS = fRFS (w, xR , xM ) = + wFS + (xR + xM ). (16.23)
2b 2 2b(2σ + s)

Again, the coefficient on wFS in (16.23) reflects the absence of the retailer’s infer-
ence of xM through wholesale price. Using these equilibrium expressions, we can
compute and compare expected prices and profits to the two firms.

Proposition 2. Relative to downward sharing, bilateral information sharing leads to

(i) No changes in expected prices: E(wFS ) = E(wDS ) and E(rFS ) = E(rDS ); and
(ii) Higher expected profits for the manufacturer: E(ΠM FS ) > E(Π DS ) and, gross of
M
any cost savings, lower expected profits for the retailer: E(ΠRFS ) < E(ΠRDS ).

Proposition 2 indicates that possession of the additional signal xR benefits the

manufacturer at the harm of the retailer despite the fact that, on average, prices are
not higher or lower. This implies that the benefit accruing the manufacturer is a
result of a high order moment of prices. In particular, if knowledge of R’s private
information enables M set prices more congruently with R, then M is able to extract
information rents that would otherwise accrue to R given higher uncertainty about
demand. As a measure of this congruence, we can use (16.22) and (16.23), to com-
pare the covariance of wholesale price with retail in the DS and FS regimes:

3σ 2 σ 2 (5σ + 3s)
cov(rFS , wFS ) = > = cov(rDS , wDS ),
4b2 (2σ + s) 8b2 (σ + s)
354 A. Dukes et al.

where the inequality holds for all σ , s > 0. We refer to the fact that information
sharing aligns fluctuations in (upstream and downstream) prices as the congruence
effect.
Despite the adverse consequence of the congruence effect on the retailer, she may
be willing to share her private signal xR with an upstream supplier if there are cost
savings. The retailer might, for example, save on ordering or inventory handling
costs if the manufacturer can better anticipate the state of demand. Proposition 2
states, however, that these savings must compensate the retailer beyond the losses
associated with the congruence effect. Formally, R will agree to share its signal if
and only if cost savings satisfy

3σ 2 s
FRDS − FRFS > E(ΠRDS − ΠRFS ) = > 0. (16.24)
16b(2σ + s)(σ + s)

Note from (16.24) that the only demand parameter that affects the sharing decision
is b, the sensitivity of demand to retail price r. Because mean prices do not change
with FS relative to DS [Proposition 2(i)], the mean size of the market, a, does not
affect cost savings threshold given in (16.24).
If the condition in (16.24) holds then full sharing of information is beneficial to
both firms as the result of the adverse consequence of congruence effect on con-
sumers. Indeed, as one might expect, retail price fluctuates more consistently with
consumer demand:
6σ 2 σ 2 (5σ + 4s)
cov(rFS , u) = > = cov(rDS , u).
4b(2σ + s) 4b(σ + s)

Thus, full information sharing enables both firms to set prices closer to the true
economic value of their buyers and consumers suffer, as a result.

Corollary 2. Relative to downward sharing, the bilateral sharing of information

leads to lower expected consumer surplus: E(CSFS −CSDS ) < 0.

The implication from Corollary 2 is that since all benefits from information shar-
ing are enjoyed by firms, if (16.24) holds, then consumers unambiguously suffer
from information sharing (also shown in Dukes et al. 2011).
Summarizing the propositions and corollaries presented above we have a full
ordering on the distribution of expected economic rents (gross of cost savings) in all
three regimes:

Manufacturer M : E(ΠM
FS
) > E(ΠM
DS
) > E(ΠM
NS
)
Retailer R : E(ΠRFS ) < E(ΠRDS ) < E(ΠRNS )
Consumer: E(CSFS ) < E(CSDS ) < E(CSNS ).

Overall, information sharing unambiguously benefits the manufacturer at the ex-

pense of consumers.
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 355

16.3.3 The Sharing Decision

The results in Sect. 16.3.2 imply that the retailer will never enter an information
sharing arrangement unless it is compensated. As mentioned above, this compensa-
tion may take the form of cost savings due to, for example, logistic improvements.
It also possible, however, that the manufacturer is willing to pay the retailer to es-
tablish a sharing arrangement.
To investigate the feasibility of this option we examine the impact of the three
informational regimes on total channel surplus ΠC = ΠM + ΠR . Direct computation
implies the following ordering:

−(3σ + s)[a2 (σ + s) + σ 2 ]
E(ΠCDS − ΠCNS ) = < 0. (16.25)
16b(2σ + s)2

This inequality echoes the fact that downward sharing exacerbates the economic
inefficiencies of double-marginalization. Furthermore, it implies that without
improvements in supply chain cost efficiencies due to information sharing, there
is no chance of a DS arrangement in equilibrium even with side payments or bar-
gaining between channel members.
In contrast, however, bilateral information sharing improves channel profits rela-
tive to DS:
σ2 6σ 2 + 4σ s − s2
E(ΠCFS − ΠCDS ) = >0 (16.26)
16b(2σ + s) 2 σ +s

for σ > s. This reflects the channel’s improved ability to extract informational rents
from consumers. This restores the possibility that the manufacturer and retailer can
make a mutually beneficial information sharing arrangement in absence of cost sav-
ings. Such a collusive arrangement is possible if and only if the sum of (16.25)
and (16.26)

σ2 3σ 2 − 2s2 a2 (3σ + s)(σ + s)

E(ΠCFS − ΠCNS ) = −
16b(2σ + s)2 σ + s 16b (2σ + s)2

is positive. This condition holds when market size parameter a is sufficiently small
and σ > s. Otherwise, information sharing motivated by purely collusive incentives
is not possible.

16.4 Competing Retailers

In this section we examine the implication of retailer competition on the effects

identified in the one retailer case. In particular, we are interested in the implication
of competition on the inference and congruence effects.
356 A. Dukes et al.

The basic set up remains as described in Sect. 16.2, except where noted. As
before, the manufacturer sets a take-it-or leave it wholesale price w to each retailer.
We assume that retailer-specific wholesale prices are not permissible.
Following the manufacturer’s wholesale pricing decision, the retailers i = 1, 2
compete on retail prices ri . Each retailer faces a linear stochastic demand:
a bri dr j
qi = − 2 + 2 + u, i = j; a > 0 and b > d > 0, (16.27)
b+d b −d 2 b − d2
where u is normally distributed with mean 0 and variance σ . Retailers are indexed
by i = 1, 2 and receive private signals xi = u + εi , where u is common to both ret-
ailers.6 The manufacturer receives the signal xM = u + εM , as above. All random
variables are independent and the signal-noise terms are drawn from the common
distribution. Specifically, (u, εM , ε1 , ε2 ) ∼ N(0, diag(σ , s, s, s)). Figure 16.2 clarifies
the information structure in the setting with competition.
The derivation process in the competing retailer case is identical to that discussed
above in the single retailer case. Therefore, we omit the details in the main text, but
state the main result.

Proposition 3.
(i) With competing, but differentiated retailers (b > d), relative to no informa-
tion sharing, downward sharing of information leads to higher expected prices:
E(wNS ) < E(wDS ) and E(rNS ) < E(rDS ).
(ii) As competition becomes less differentiated (d ↑ b) the effects of information
sharing disappear. That is, limd↑b E(rNS − rDS ) = 0 and limd↑b E(wNS −
wDS ) = 0.

Proposition 3(i) confirms that the inference effect identified in Sect. 16.3.1 is
present with competing, yet differentiated retailers (b > d). But part (ii) of the
proposition illustrates how the inference effect is connected to competitive forces.
Recall that with no-sharing, economic inefficiencies associated with double-
marginalization cause the retailer to distort price downward. However, as compe-
tition between retailers reduces these inefficiencies, the need for wholesale pricing
distortions decreases as well.
Finally, we turn to the bilateral information sharing case and investigate the imp-
lication of retail competition on the congruence effect. It is assumed that with full
sharing, all realized signals are available to all firms. We relegate the full analysis to
the appendix and arrive at the following result.

Proposition 4.
(i) With competing, but differentiated retailers (b > d), downward sharing of inf-
ormation and bilateral information sharing lead to the same expected prices:
E(wDS ) = E(wFS ) and E(rDS ) = E(rFS ).
6 We maintain a common u for simplicity. It is possible, however, to consider retailer-specific
stochastic demand terms. To the extent that there is positive correlation across retailers’ terms, we
expect the basic quantitative effects to survive.
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 357

Manufacturer: xM
xM = u + ε M , ε M ~ N (0, s )

w w

Retailer 1: x1 Retailer 2: x2
x1 = u + ε 1, ε1 ~ N (0, s ) x2 = u + ε 2 , ε 2 ~ N (0, s )

r1 r2
Consumers Consumers

q1 = a
b+d
− b
r+
b2 − d 2 1
d
b2 − d 2
r2 + u q2 = a
b+d
− b2 −bd 2 r2 + b2 −d d 2 r1 + u

cov(ε M , ε1 ) = cov(ε1 , ε 2 ) = cov(ε M , ε 2 ) = 0

cov(ε M , u ) = cov(ε1 , u ) = cov(ε 2 , u ) = 0
Fig. 16.2 Information structure with competing retailers

(ii) With competing retailers, relative to downward sharing of information, bilat-

eral information sharing leads to strictly greater congruence between prices
and demand fluctuations for all d ∈ (d, b], for some d < b: cov(rFS , wFS ) >
cov(rDS , wDS ) > 0 and cov(rFS , u) > cov(rDS , u) > 0 for all d ∈ (d, b].

As in Sect. 16.3.2, Proposition 4 indicates that the congruence effect is a second

moment effect. In particular, expected prices are unchanged but the covariance of
prices increases when information is fully exchanged. Finally note that, in the case
of competing retailers, we do not examine the shift in the distribution of profits
as a result of information sharing. Our interest in this section is to determine how
competition influences the inference and congruence effects. As long as competition
is imperfect (b > d), both of these effects are present and will tend to shift rents
in the direction indicated in Sect. 16.2. However, competition erodes the inference
effect and, in the perfectly competitive limit, eliminates it. In this case, retailers
earn zero profits in all regimes. In contrast, bilateral sharing offers more congruent
pricing even in the perfectly competitive limit.
Note that in our model retailers are symmetric with respect to the precision of
their information about demand. That is, the noise in their signals (i.e., s) is the
same. Gal-Or et al. (2008) investigate information sharing between a manufacturer
and two retailers that have asymmetric signals. They show that the inference effect,
which adversely affects the manufacturer, is more profound in the channel whose
retailer has a noisier signal. The retailer that is less informed about demand relies
358 A. Dukes et al.

to a greater extent on the manufacturer’s wholesale price for inferences about the
level of demand. Hence, sharing information with such a retailer alleviates the inf-
erence effect to a greater degree than the other retailer. This result also suggests that
when there is a cost associated with information transmission, the manufacturer may
choose to enter into an information sharing arrangement with only the less informed
retailer rather than with both.

16.5 Conclusions and Directions for Future Research

The analysis in this chapter indicates that, in the absence of operational bene-
fits, information sharing in a distribution channel benefits the upstream firm—the
manufacturer in our model—to the detriment of consumers. If a manufacturer has
private information about the state of demand, it is communicated through his
wholesale pricing decision even in absence of information sharing. As we showed,
inference through the strategic choice of wholesale price implies a distortion
(the inference effect) which benefits the retailer and consumers in the form of lower
prices. Therefore, when the manufacturer transmits his signal in a downward shar-
ing arrangement, this distortion is no longer necessary which implies that the retailer
must enjoy sufficient cost savings to partake in such sharing arrangement.
A full exchange of information, in which both firms know each other’s signal,
further harms the retailer and consumers to the benefit of the manufacturer. Despite
the fact that average prices remain unchanged, relative to downward sharing, bilat-
eral sharing improves firms’ ability to “right price” according to the buyer’s valua-
tion (the congruence effect). That is, firms are able to align prices more closely to
demand fluctuations thereby reducing consumer surplus.
The model reveals that downward sharing, in the absence of any cost savings,
exacerbates economic inefficiencies associated with double-marginalization, conse-
quently lowering total channel profits. This implies that, despite the fact that the
manufacturer’s profits increase from downward sharing, he would not be willing to
pay the retailer a sufficient amount to induce her to accept the manufacturer’s sig-
nal. In contrast, a full exchange of information may raise total channel profits (at the
expense of consumers). When this happens, a negotiated arrangement involving a
monetary transfer from the manufacturer to the retailer is possible to induce the
retailer to participate.
Our model and its results rely on several crucial assumptions. Relaxing these
assumptions can lead to interesting future research. For example, we assume that the
manufacturer dictates the wholesale price to the retailer. However, it is also possible
that a more powerful retail buyer can negotiate wholesale terms. The retailer and
the manufacturer can enter a bargaining process in which both make alternating
offers of wholesale price (Rubinstein 1982). Offers made by each channel member
reflect their private information. Specifically, the manufacturer’s offers can be used
by the retailer to infer xM , and the offers made by the retailer can be used by the
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 359

manufacturer to infer xR . Such inferences can change channel members’ incentive

to enter into information sharing arrangements that involve the exchange of their
demand signals.
Another fruitful avenue for future research is the sharing of demand information
about complementary products supplied by different suppliers. For instance, manu-
facturers may have information about what products are complements to their own.
Similarly, multiproduct retailers may have information on what their customers’
shopping baskets consist of. Because pricing complementary goods involves addi-
tional strategic considerations, such as bundling and product line pricing, sharing
such information may lead to different pricing incentives than found here.
It will also be interesting to consider the consequence of upstream competition.
Such competition can increase retailer power and change a variety of incentives in
the channel (Dukes and Geylani 2016). For example, it can put downward pressure
on wholesale prices, and consequently alter the incentives for information sharing.
The retailer will still be able to infer the manufacturer’s private signal in this case.
However, the manufacturer’s limited flexibility with the wholesale price due to com-
petition may decrease his benefit from the congruence effect and the alleviation of
the inference effect, and therefore may reduce the information sharing’s adverse
consequences on both the retailer and consumers.

Appendix

This appendix provides the proofs of all propositions and corollaries stated in the
main text.

Proof of Proposition 1

A comparison of the expected values of (16.11) and (16.12) with the expected values
of (16.16) and (16.17) gives the results of part (i). To deduce part (ii), we compute
realized profits to each firm in the two regimes.
NS: The realized quantity sold is

qNS = a − brNS + u
a 3σ + 2s σ (3σ + 2s) σ
= − xM − xR + u.
4 2σ + s 4(σ + s)(2σ + s) 2(2σ + s)

Computing profits at the retailer

ΠRNS = (rNS − wNS )qNS

 
Δ 3σ + s σ + 2s
= a + xM + xR × qNS ,
2b 2σ 2(σ + s)
360 A. Dukes et al.

and at the manufacturer

ΠMNS = wNS qNS

 
Δ σ σ2
= a + xM × qNS ,
2b 2σ + s (2σ + s)(σ + s)

where Δ ≡ σ /(2σ + s). Taking expectations over (u, xM , xR ) gives

   
Δ 2 2 3σ + 2s 2 9σ 2 + 20σ s + 8s2
E(ΠRNS ) = a + (16.28)
16b σ σ +s

and  
Δ 2 2 3σ + 2s σ (3σ + 2s)
E(ΠM
NS
)= a + , (16.29)
8b σ σ +s
which makes use the following facts:

E(xM xR ) = E(u2 + εM u + εR u + εM εR ) = σ
E(xi )2 = E(u2 + 2εi u + εi ) = σ + s, (16.30)
E(xi u) = E(u + uεi ) = σ ,
2

for i = M, R.
DS: The realized quantity sold is

qDS = a − brDS + u
a Δ Δ (4σ + 3s)
= − xR − xR + u.
4 2 4(σ + s)

Computing profits at the retailer

ΠRDS = (rDS − wDS )qDS

 
1 Δs
= a+ xM + 2Δ xR × qDS ,
4b σ +s

and at the manufacturer

 
a σ
ΠMDS =w qDS DS
= + xM × qDS .
2b 2b(σ + s)

Taking expectations over (u, xM , xR ) gives

 
1 σ Δ (2σ + 5s)
E(ΠR ) =
DS
a +
2
(16.31)
16b σ +s
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 361

and  
1 2 σ2
E(ΠM
DS
)= a + . (16.32)
8b σ +s
Using (16.28) and (16.31), the difference E(ΠRDS − ΠRNS ) is given by the expression
in (16.18). For the manufacturer, use (16.29) and (16.32) to compute
    2 
1 2 σ (3σ + 2s) σ σ 3 (3σ + 2s)
E(ΠM NS
−ΠM DS
)= a 1− + − < 0.
8b (2σ + s)2 σ + s (2σ + s)2 (σ + s)
( )* + ( )* +
− −

Q.E.D.

Proof of Corollary 1

The change in expected consumer surplus can be computed using (16.19) for
k = NS, DS. This leads to
b
E(CSNS −CSDS ) = [E(rNS )2 − E(rDS )2 ] − a[E(rNS ) − E(rDS )]
2
− [E(urNS )2 − E(urDS )2 ]
1
=
32b(2σ + s)2
 
2 σ
2
× a (5σ + 3s)(σ + s) + (7σ + 8σ s + 3s )
2 2
,
σ +s

which makes use of the facts in (16.30) to compute the expectations in the brackets
above and some amount of algebra. The expression above is obviously positive.
Q.E.D.

Proof of Proposition 2

The DS case is analyzed in the proof of Proposition 1. We derive here the results
for FS. Note that part (i) of the proposition follows from the expressions (16.25)
and (16.26). Part (ii) requires the computation of profits in FS regime.
FS: The realized quantity sold is

a 3Δ
qFS = a − brFS + u = − (xR + xR ) + u.
4 4
Computing profits at the retailer
1
ΠRFS = (rFS − wFS )qFS = [a + Δ (xM + xR )] × qFS ,
4b
362 A. Dukes et al.

and at the manufacturer

1
ΠMFS = wFS qFS = [a + Δ (xM + xR )] × qDS .
2b
Taking expectations over (u, xM , xR ) gives

1 2
E(ΠRFS ) = [a + Δ 2 (4σ + 2s)] (16.33)
16b
and
1 2
E(ΠM
FS
)= [a + Δ 2 (7σ + 2s)]. (16.34)
8b
Using (16.33) and (16.34), the difference E(ΠRFS − ΠRDS ) is given by the expression
in (16.24). For the manufacturer, use (16.33) and (16.34) to compute
 
Δ 2 3σ 2 + 5σ s + s2
E(ΠM FS
− ΠM
DS
)= > 0. Q.E.D.
8b σ +s

Proof of Corollary 2

Following the computation in Corollary 1 and using the corresponding derivations

for the FS regime we have

3sσ 2
E(CSDS −CSFS ) = > 0,
32b(σ + s)(2σ + s)

which makes use of the fact that E(rFS ) = E(rDS ). Q.E.D.

Proof of Proposition 3

NS: We start with the no sharing NS regime and specify decision rules for the three
firms as follows

w = fMNS (xM ) = μ0 + μ1 xM (16.35)

ri = fiNS (w, xi ) = ρ0i + ρ1i w + ρ2i xi , i = 1, 2. (16.36)

M sets a price w according to fMNS and retailers respond by simultaneously maxi-

mizing

E(Πi |w, xi ) = (ri − w)E(qi |w, xi )

 
a bri dE(r j |w, xi )
= (ri − w) − 2 + + E(u|w, xi ) . (16.37)
b + d b − d2 b2 − d 2
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 363

for i = 1, 2 and j = i. Retailer i’s conditional expectations of its rivals’ price is

facilitated by (16.36):

E(r j |w, xi ) = ρ0 j + ρ1 j w + ρ2 j E(x j |w, xi ), i = 1, 2 and j = i.

Again, perfect inference of xM is possible from observing wholesale price w and

inverting (16.35). Specifically, retailer i’s conditional expectations in (16.37) are

σ (xi + xM ) σ (xi + (w − μ0 )/μ1 )

E(x j |w, xi ) = E(u|w, xi ) = = ,
2σ + s 2σ + s
which were similarly derived in Sect. 16.2.
The manufacturer sets wholesale price w to maximize expected profits:

E(ΠM |xM ) = wE(q1 + q2 |xM )

 
2a (b − d)
=w − 2 [E(r 1 |xM ) + E(r 2 |xM )] + 2E(u|xM ) , (16.38)
b + d b − d2

where M’s conditional expectations are computed using (16.36) and (16.8) so that
σ
E(xi |xM ) = E(u|xM ) = xM .
σ +s
As before, maximizations of (16.37) for each i = 1, 2 and (16.38) yields three first
order conditions which may be used to match coefficients with (16.35) and (16.36),
respectively. This leads to a linear system of eight equations and eight unknown
characterizing the equilibrium decision rules. However, symmetry across retailers
reduces the solution by three variables leading to the following solution:
 
σ
wNS = Γ a + (b + d) xM
σ +s
riNS = μ0NS + μ1NS w + μ2NS xi ,

where
4bd(σ + s)2 + 2σ (σ + s)(2b2 + bd + d 2 ) + σ 2 (4b2 − d 2 )
Γ= ;
2[4b2 (2σ + s) − d 2 σ 2 ]
 
b−d 2ab(σ + s)
μ0NS = a− ;
2b − d 2b(2σ + s) − d σ
σ (b2 − d 2 )
μ1NS = ; and
2b(2σ + s) − d σ
 
b 2(b − d)(σ + s)
μ2NS = 1− .
2b − d Γ 2b(2σ + s) − d σ
364 A. Dukes et al.

Note that Γ has the following properties:

1 1
Γ< for all d < b and lim Γ = . (16.39)
2 b↓d 2

DS: In this regime both retailers have knowledge of the manufacturer’s realized
signal xM , but the manufacturer has no additional information relative to NS. This
information structure implies the following linear forms of decision rules:

w = fMDS (xM ) = μ0 + μ1 xM
ri = fiDS (w, xi , xM ) = ρ0i + ρ1i w + ρ2i xi + ρ3i xM ; i = 1, 2.

Following the same process as in the NS regime without the inference of xM leads
to the following equilibrium decision rules:
 
1 σ
wDS = a + (b + d) xM (16.40)
2 σ +s
riDS = μ0DS + μ1DS w + μ2DS xi + μ3DS xM , (16.41)

where
a(b − d) σ (b2 − d 2 ) b
μ0DS = ; μ1DS = ; μ2DS = ;
2b − d 2b(2σ + s) − d σ 2b − d
2bσ (b2 − d 2 )
and μ3DS = .
(2b − d)[2b(2σ + s) + d σ ]

The difference in expected wholesale prices: E(wNS − wDS ) = (Γ − 12 )a, which

from (39) is shown negative for d < b and equal zero in the limit d ↑ b.
Finally, the difference in retail prices
E(rNS − rDS ) = (ρ0NS − ρ0DS ) + a(Γ ρ2NS − 12 ρ2DS )
is seen negative for d < b by noting that ρ0NS < ρ0DS for d < b, ρ2NS = ρ2DS for d ≤ b
and reusing (16.39). Again, (16.39) implies this difference is zero in the limit d ↑ b.
Q.E.D.

Proof of Proposition 4

To prove this we derive the equilibrium outcome in the FS regime and compare it to
the outcome in the DS regime derived in the proof of Proposition 3. In the FS case,
we assume all signals x1 , x2 , xM are known by all firms. Hence, the assumed linear
form for the decision rules imply
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 365

w = fMFS (x1 , x2 , xM ) = μ0 + μ1 x1 + μ2 x2 + μ3 xM
ri = fiFS (w, xi , x j , xM )
= ρ0i + ρ1i xi + ρ2i x j + ρ3i w + ρ4i xM ; i = 1, 2; j = i. (16.42)
Expected profit to retailer Ri given x1 , x2 , xM and w is
E(Πi |w, x1 , x2 , xM )
 
a bri
= E (ri − w) −
b + d b2 − d 2

d
+ 2 E(r j |w, x , x
1 2 M, x ) + E(u|w, x , x
1 2 M, x ) (16.43)
b − d2
and to the manufacturer

E( ΠM | x1 , x2 , xM )
  
a bri dr j
=E w − 2 + + E(u|w, x1 , x2 , xM ) . (16.44)
b + d b − d 2 b2 − d 2

Following DeGroot (1970) we have that

σ
E(u|x1 , x2 , xM ) = (x1 + x2 + xM ).
3σ + s
Furthermore, because retailers i = 1, 2 in the FS regime are in all aspects symmetric,
we can exploit the fact that their decision rules will be identical in a symmetric equi-
librium. Using these facts and the first order condition for maximizing Ri ’s expected
profit (16.43), we can solve for the equilibrium coefficients of (16.42):
 
a b+d σ 3b − 2d
rFS = + (x1 + x2 + xM ) , (16.45)
2 2 3σ + s 2b − d

which has the expectation: E(rFS ) = (a/2)[(3b − 2d)/(2b − d)]. Using the retailers’
reaction, (16.42), the first order condition from the maximization of (16.44) permits
the solution
a b+d σ
wFS = + (x1 + x2 + xM ), (16.46)
2 2 3σ + s
which has the expectation E(wFS ) = a/2. Part (i) of the proposition follows from
taking expectations of wDS and rDS , given in Eqs. (16.40) and (16.41) derived in the
proof of Proposition 3.
To establish part (ii) of the proposition, note that the covariances can be computed
directly using (16.40), (16.41), (16.45), and (16.46) and making use of the facts
in (16.30):
366 A. Dukes et al.
   
b+d σ2 b2 − d 2 2b b(b + d)
cov(wDS , rDS ) = σ+ (σ + s) +
2 σ + s 2b(2σ + s) − d σ 2b − d 2(2b − d)
 2 
σ 2 (b − d )(4b − d) b(b + d)
2
cov(rDS , u) = +
2b − d 2b(2σ + s) − d σ 2(σ + s)
3σ (b + d)2 3b − 2d
cov(wFS , rFS ) =
4 2d − d
3σ (b + d) 3b − 2d
2
cov(rFS , u) = .
2(3σ + s) 2d − d

Evaluating the limits of these expressions for d ↑ b implies the orderings stated in
the proposition. This guarantees a d < b such that the ordering holds in the left
neighborhood of b. Q.E.D.

References

Cachon GP, Fisher M (2000) Supply chain inventory management and the value of shared infor-
mation. Manag Sci 46(7):1032–1048
DeGroot M (1970) Optimal statistical decisions. McGraw Hill, New York
Dukes A, Gal-Or E, Geylani T (2011) Who benefits from bilateral information exchange in a retail
channel. Econ Lett 112(2):210–212
Dukes A, Geylani T (2016) Dominant retailers and their impact on marketing channels. In: Emek
B (ed) The handbook of economics of retail and distribution. Edward Elgar.
Gal-Or E (1985) Information sharing in oligopoly. Econometrica 53(1):329–344
Gal-Or E (1987) First mover disadvantages with private information. Rev Econ Stud 54:279–292
Gal-Or E, Geylani T, Dukes A (2008) Information sharing in a channel with partially informed
retailers. Mark Sci 27(3):642–658
Guo L, Iyer G (2010) Information acquisition and sharing in a vertical relationship. Mark Sci
29(2):483–506
Ha AY, Tong S (2008) Contracting and information sharing under supply chain competition. Manag
Sci 54(3):701–715
He C, Marklund J, Vossen T (2006) Vertical information sharing in a volatile market. Mark Sci
27(2):513–530
Lee HL, Padmanabhan V, Whang S (1997) Information distortion in a supply chain: the bullwhip
effect. Manag Sci 43(3):546–558
Lee HL, So KC, Tang CS (2000) The value of information sharing in a two-level supply chain.
Manag Sci 46(4):626–643
Li L (1985) Cournot oligopoly with information sharing. RAND J Econ 16(3):521–536
Li L (2002) Information sharing in a supply chain with horizontal competition. Manag Sci
48(8):1196–1212
Milgrom P, Roberts J (1980) Limit pricing and entry under incomplete information: an equilibrium
analysis. Econometrica 50(2):443–459
Mittendorf B, Shin J, Yoon D-H (2013) Manufacturer marketing initiatives and retailer information
sharing. Quant Mark Econ 11(1):263–287
Niraj R, Iyer G, Narasimhan C (2007) Inventory and information in distribution channels. Manag
Sci 53(9):1551–1561
Novshek W, Schonneschein H (1982) Fulfilled expectations Cournot duopoly with information
acquisition and release. Bell J Econ 13(1):214–218
16 Bilateral Information Sharing and Pricing Incentives in a Retail Channel 367

Raith M (1996) A general model of information sharing in oligopoly. J Econ Theory 71(1):
260–288
Rubinstein A (1982) Perfect equilibrium in a bargaining model. Econometrica 50(1):97–110
Spence M (1973) Job market signaling. Q J Econ 87(3) 355–374
Vives X (1984) Duopoly information equilibrium: Cournot and Bertrand. J Econ Theory 34(1):
71–94
Chapter 17
Sharing Demand Information Under Simple
Wholesale Pricing

Albert Y. Ha and Hongtao Zhang

Abstract We investigate a retailer’s incentive in sharing private demand informa-

tion with a manufacturer under a linear wholesale price contract. We present a
summary of the analysis and the main results of several existing models for the
following manufacturer-retailer relationships: one-to-one, two competing chains,
one-to-many, and two-to-one. By synthesizing the major findings of these models,
we provide a common framework for understanding the impact of some key drivers
on the retailer’s information sharing decision. We also illustrate the basic method-
ology for analyzing related models.

17.1 Introduction

It is well known that collaboration between manufacturers and retailers could imp-
rove supply chain performance. According to a survey by SCM World, demand
information sharing is the most common activity involved in supply chain collabo-
ration (O’Marah 2013). Many initiatives such as quick response, CPFR (collabora-
tive planning, forecasting and replenishment) and vendor managed inventory rely on
information sharing to facilitate collaboration between supply chain partners. With
the advance of information technology, retailers routinely and efficiently acquire
rich market data related to point-of-sales, market basket (i.e., all the items purchased
by a consumer in a shopping trip), loyalty card program, consumer demographics

A.Y. Ha ()
Hong Kong University of Science and Technology, Kowloon, Hong Kong
e-mail: [email protected]
H. Zhang
School of Business and Management, Hong Kong University of Science and Technology,
Hong Kong, China
e-mail: [email protected]

© Springer International Publishing Switzerland 2017 369

A.Y. Ha, C.S. Tang (eds.), Handbook of Information Exchange in Supply Chain
Management, Springer Series in Supply Chain Management 5,
DOI 10.1007/978-3-319-32441-8 17
370 A.Y. Ha and H. Zhang

and trade promotion (Keifer 2010). Although such data can provide valuable in-
formation about future demand and improve supply chain decision-making, many
retailers are either reluctant to share any data or share it only selectively with some
manufacturers. For example, Beckett (2012) reports that it is more common for mass
merchant retailers, drug stores and supermarkets to offer data-sharing programs to
their suppliers, but less so for internet retailers and department stores. Keifer (2010)
observes that some retailers offer data-sharing programs to their suppliers for free,
while others charge an annual subscription fee for participation. Moreover, a re-
tailer may offer data-sharing program to only some, but not all, of the suppliers in
the same product category.
In this chapter, we investigate a retailer’s incentive in sharing private demand in-
formation with a manufacturer under a linear wholesale price contract. We present
a summary of the analysis and the main results in Li and Zhang (2008), Ha et al.
(2011), and Shang et al. (2016). By synthesizing the major findings in these mod-
els, we provide a common framework for understanding the impact of some key
drivers on the retailer’s information sharing decision. We also illustrate the basic
methodology for analyzing related models.
In Sect. 17.2, we consider the case of a single supply chain with a manufacturer
selling to a retailer. We focus on non-linear production cost and inventory as drivers
of the retailer’s information sharing decision. We show that information sharing ex-
acerbates double marginalization of linear wholesale pricing, which hurts the supply
chain. For the case of non-linear production cost, information sharing could lower
production cost because it allows the manufacturer to adjust wholesale price to in-
fluence the variability of the retailer’s orders. When the degree of production econ-
omy/diseconomy is high enough, the benefit of production cost saving outweighs
the cost of exacerbated double marginalization. As a result, information sharing
benefits the supply chain and the retailer will share information either voluntarily or
with a side payment. For the case when the manufacturer has the option of making
to stock, information sharing allows the manufacturer to make better inventory deci-
sion to lower the cost of mismatch between supply and demand. When the inventory
costs are high enough, the benefit of inventory cost saving outweighs the cost of ex-
acerbated double marginalization. Consequently, information sharing benefits the
supply chain and the retailer will share information with for a side payment.
In Sects. 17.3 and 17.4, we assume non-linear production cost and extend our
analysis to the case of two supply chains when they compete in price and quantity,
respectively. We distinguish between the direct and competitive effects of informa-
tion sharing. The first effect does not account for the competition reaction from the
rival chain, whereas the second effect is due to that reaction. We characterize these
two effects and show how they jointly determine the retailer’s information sharing
decision. The case of two competing supply chains with production economy has
not been explored in the literature. For price competition, our analysis implies that
when the degrees of production economy is high, the competitive effect is negative
but the retailer still has an incentive to share information as long as competition is
not too intense. For quantity competition, we show that the direct and competitive
effects have the same sign, and therefore the retailer’s information sharing decision
does not depend on competition.
17 Sharing Demand Information Under Simple Wholesale Pricing 371

In Sects. 17.5 and 17.6, drawing upon the insights from our analysis in the earlier
sections, we discuss the cases of one-to-many and two-to-one supply chains. For a
one-to-many supply chain, we highlight the role of the wholesale price in signaling
demand information when the information shared between two firms is confidential
and cannot be disclosed to other firms. For a two-to-one supply chain, we explain
how non-linear production cost and information contracting sequence (i.e., whether
the manufacturers simultaneously or sequentially offer a side payment for the in-
formation) influence the retailer’s decision in sharing information with none, one or
both of the manufacturers.

17.2 Single Supply Chain

We consider one supply chain in isolation, with a manufacturer in the upstream and a
retailer in the downstream. The demand function at the retail market is q = a+ θ − p,
where p is the retail price set by the retailer and q is the realized market demand (i.e,
sales quantity). The random variable θ , with zero mean and variance σ 2 , represents
demand uncertainty. The retailer has a constant marginal retailing cost, which we
normalize to zero. The retailer has access to a demand signal Y (an unbiased esti-
mator of θ ).
The sequence of events and decisions is as follows:
1. Before the retailer observes the demand signal Y , the manufacturer and the re-
tailer contract on whether Y will be shared. We say that supply chain is commu-
nicative if the retailer is to share Y with the manufacturer, or non-communicative
otherwise.
2. The retailer observes a signal Y . This signal will be truthfully disclosed to the
manufacturer if the supply chain is communicative. We say that the manufac-
turer in a communicative (non-communicative) supply chain is informed (unin-
formed).
3. The manufacturer determines a wholesale price w and then the retailer determines
a retail price p.
4. The uncertainty, θ , resolves and the manufacturer supplies the realized market
demand, q = a + θ − p, to the retailer. Finally, firms receive their payoffs.
For convenience, we refer to the manufacturer as she and a retailer as he.
Given any wholesale price w set by the manufacturer, the retailer maximizes his
expected profit, (p − w)(a + E[θ |Y ] − p), by setting the retail price to

p = (a + E[θ |Y ] + w)/2,

which results in retail sales

q = a + θ − (a + E[θ |Y ] + w)/2.
372 A.Y. Ha and H. Zhang

Anticipating the above, the manufacturer sets a wholesale price to maximize her
expected profit, conditional on the information that she has received from the
retailer, if any.
Note that, given p, the retail quantity is not completely determined by the demand
signal Y because the signal is imperfect and cannot precisely estimate θ . We call this
the residual demand uncertainty and measure its magnitude by

λ = E[Var[θ |Y ]].

We will see that the residual demand uncertainty has an impact on firms’ payoffs
only when the manufacturer’s production cost is nonlinear. Let

δ = E[(E[θ |Y ])2 ].

This measures the variability of the estimate of θ from Y . If the demand informa-
tion has a linear-expectation conjugate structure (including normal-normal, gamma-
Poisson, and beta-binomial), we will have

tσ 2 tσ 4 σ2
E[θ |Y ] = Y, δ= , λ= ,
(1 + t σ 2 ) 1 + tσ 2 1 + tσ 2

where t = 1/E[Var[Y |θ ]] is the signal accuracy. For more details on the linear-
expectation information structure, refer to Vives (1999, Sect. 2.7.2).
Cost Structures We consider four cost structures of the manufacturer’s produc-
tion: make-to-order with a constant unit cost, with diseconomies of scale, and with
economies of scale; and make-to-stock with linear costs and expedition.
Retailer’s Decision Choice After learning the wholesale price, the retailer could
decide either a retail price p, at which the sales will be a + θ − p, or a retail quan-
tity q, which will be sold at the market clearing price of a + θ − q. If the production
cost is linear, the residual demand uncertainty has no impact, and the two decision
choices, q or p, will result in the same equilibrium payoffs for the firms in the single
supply chain and in this sense are equivalent. But the two decision choices differ in
a key aspect if the production has economy or diseconomy of scale, as will be seen.

17.2.1 Linear Production Cost

The manufacturer has a constant marginal production cost, b. The cost structure is
common knowledge.
If the supply chain is communicative, the manufacturer knows Y and she maxi-
mizes wE[q|Y ] − bE[q|Y ] by setting the wholesale price to

a−b 1
w = b+ + E[θ |Y ].
2 2
17 Sharing Demand Information Under Simple Wholesale Pricing 373

Given this w, the retailer’s best decision is

3(a − b) 3E[θ |Y ]
p = b+ + .
4 4
If the supply chain is non-communicative, the manufacturer does not know Y and
she maximizes wE[q] − bE[q] by setting the wholesale price to

a−b
w = b+ .
2
Given this w, the retailer’s best decision is
3(a − b) E[θ |Y ]
p = b+ + .
4 2
Remark 1. The retail price is more responsive (i.e., sensitive) to the demand signal
(via E[θ |Y ]) if the supply chain is communicative than if it is non-communicative.
The informed manufacturer adjusts the wholesale price in the direction of the
demand signal and the retailer responds by adjusting his retail price in the direc-
tion of the wholesale price adjustment. As a result of these adjustments, double
marginalization is aggravated.

We can work out the firms’ and the supply chain’s ex ante payoffs:

δ δ 3δ
ΠRS = Π̄R + , ΠM
S
= Π̄M + , Π S = Π̄ + ,
16 8 16
δ δ
ΠRN = Π̄R + , ΠMN = Π̄M , Π N = Π̄ + ,
4 4
where the superscript, S or N, stands for sharing or not sharing information, respec-
tively, and Π̄R , Π̄M , and Π̄ are the deterministic payoffs (when there is no uncer-
tainty).1 It is easy to see that ΠRS < ΠRN , ΠM
S > ΠN, ΠS < ΠN.
M
As Li and Zhang (2002) pointed out, information sharing makes double marginal-
ization more pronounced, benefiting the manufacturer but hurting both the retailer
and the supply chain. In this case, the retailer has no incentive to share his infor-
mation on a voluntary basis, nor can the manufacturer offer a payment high enough
that the retailer would accept to sell his information.

Remark 2. Greater demand variability (larger σ 2 ) always makes the retailer better
off. It increases the manufacturer’s profit if she is informed or leaves it unchanged
if she is uninformed.

1 Specifically, Π̄R = (a − b)2 /16, Π̄M = (a − b)2 /8, Π̄ = 3(a − b)2 /16.
374 A.Y. Ha and H. Zhang

17.2.2 Production Diseconomies of Scale

The production cost to the manufacturer of producing q units is given by

bq + cq2 .

The production cost function consists of a linear term, bq, and a convex quadratic
term, cq2 , with c > 0, reflecting diseconomy of scale, where a larger c corresponds
to a greater production diseconomy. The production diseconomy can be thought of
as having increasingly more expensive production capacity or input (e.g. supplied
materials). We will see that due to convexity, all else being equal, variability in
production quantity hurts the manufacturer. We can show that, when σ and c2 are
small relative to a, it is optimal for the manufacturer, with a probability very close
to one, to fully meet the retailer’s order.
If the supply chain is communicative, the manufacturer maximizes wE[q|Y ] −
bE[q|Y ] − cE[q2 |Y ] by setting the wholesale price to

(1 + c)(a − b) 1 + c
w = b+ + E[θ |Y ].
2+c 2+c
Given this w, the retailer’s best decision is
(3 + 2c)(a − b) (3 + 2c)E[θ |Y ]
p = b+ + .
2(2 + c) 2(2 + c)

The resulting retail quantity, a + θ − p, equals

a−b (3 + 2c)E[θ |Y ]
qS = − +θ,
2(2 + c) 2(2 + c)

which has a variance given by

δ
Var[qS ] = +λ.
4 (2 + c)2

Remark 3. For a communicative supply chain, the wholesale price and the retail
price are more responsive to the demand signal if the production diseconomy is
larger. With a larger production diseconomy, the variability of the order quantity has
a stronger negative impact on the expected production cost. Therefore, the informed
manufacturer charges a more responsive wholesale price to make retail quantity less
variable. This can be seen from the fact that Var[qS ] is decreasing in c.

If the supply chain is non-communicative, the manufacturer maximizes wE[q] −

bE[q] − cE[q2 ] by setting the wholesale price to

(1 + c)(a − b)
w = b+ .
2+c
17 Sharing Demand Information Under Simple Wholesale Pricing 375

Given this w, the retailer’s best decision is

(3 + 2c)(a − b) E[θ |Y ]
p = b+ + .
2(2 + c) 2
The resulting retail quantity, a + θ − p, equals
a−b E[θ |Y ]
qN = − +θ,
2(2 + c) 2
which has a variance given by
δ
Var[qN ] = +λ.
4

Remark 4. Var[qS ] < Var[qN ], i.e., information sharing makes the order quantity less
variable for the manufacturer.

We can work out the firms’ and the supply chain’s ex ante payoffs:
δ δ (3 + c)δ
ΠRS = Π̄R + , ΠM
S
= Π̄M + − cλ , Π S = Π̄ + − cλ ,
4(2 + c)2 4(2 + c) 4(2 + c)2
δ cδ (1 − c)δ
ΠRN = Π̄R + , ΠMN = Π̄M − − cλ , Π N = Π̄ + − cλ ,
4 4 4
where Π̄R , Π̄M , and Π̄ are the deterministic payoffs.2 Because the demand signal Y
is not perfect, the residual uncertainty increases the variability of order (production)
quantity and leads to the additional cost cλ , regardless of whether the supply chain
is communicative or not.3
As noted earlier, if the marginal production cost is constant, information sharing
makes double marginalization more pronounced, hurting the supply chain. However,
if production exhibits diseconomy of scale, information sharing has another effect:
it makes the retailer’s order quantity less variable and reduces the average of the
convex quadratic production cost. From the supply chain’s perspective, this effect on
production cost generates significant benefit if the production diseconomy is large.
This explains the following result.

Proposition 1. Information sharing makes the retailer worse √ off and the manufac-
turer better off. It makes the supply chain better off if c > 2 − 1.

To examine the effect of greater demand variability, we assume linear-expectation

information structure. The effect of an increase in demand variability depends on the
magnitude of production diseconomy, described by three thresholds in the proposi-
tion below.4
2 Specifically, Π̄R = (a − b)2 /[4(2 + c)2 ], Π̄M = (a − b)2 /[4(2 + c)], Π̄ = (3 + c)(a − b)2 /
[4(2 + c)2 ].
3 If the signal Y were perfect, i.e., Y = θ , we would have Var[θ |Y ] = 0 and λ = 0.
4 LS is the positive value of z satisfying 4z (2 + z) = t σ 2 (2 + t σ 2 ), LS is the positive value of z
M
satisfying (16z + 16z2 + 4z3 )/(3 + z) = t σ 2 (2 +t σ 2 ), and LN = t σ 2 (2 +t σ 2 )/(8 + 4t σ 2 + 2t 2 σ 4 ).
376 A.Y. Ha and H. Zhang

Proposition 2. When demand becomes more variable (larger σ 2 ), (a) for a com-
municative supply chain, the retailer is better off, the manufacturer is better of if
S , and the supply chain is better off if c < LS , (b) for a non-communicative
c < LM
supply chain, the retailer is better off, the manufacturer is worse off, and the supply
chain is better off if c < LN .

For an informed manufacturer, greater demand variability allows her to bene-

fit more from responsive wholesale pricing, but it also increases her production
cost due to greater residual demand uncertainty. The net effect of greater demand
variability on the manufacturer, and on the whole supply chain, will be negative if
the production diseconomy is too large. Note that, if the demand signal is perfect
(t = ∞), then the residual demand uncertainty disappears (λ = 0); in this case, larger
σ 2 always benefits the supply chain and the informed manufacturer. 5

Remark 5 (Retailer’s Decision Choice). If, after learning the wholesale price, the
retailer decides a retail quantity q, which will be sold at price a + θ − q, the firms
will have the same payoffs except without the term cλ . Proposition 1 will hold
exactly as stated. Proposition 2 will take on the following form.

Proposition 3. Suppose the retailer’s decision choice is quantity. When the demand
becomes more variable (larger σ 2 ), (a) for a communicative supply chain, both the
retailer and the manufacturer are better off; (b) for a non-communicative supply
chain, the retailer is better off but the manufacturer is worse off, and the supply
chain is better off if c < 1.

17.2.3 Production Economies of Scale

The manufacturer’s production cost exhibits economies of scale and is modeled by

bq − cq2 ,

where c > 0. This is decreasing in q ∈ (0, b/(2c)). The production cost function con-
sists of a linear term, bq, and a concave quadratic term, −cq2 , reflecting economy
of scale, where a larger c corresponds to a greater production economy. We will see
that due to concavity, all else being equal, variability in production quantity helps
the manufacturer.
We assume that the system parameters are such that q < b/(2c) holds in equilib-
rium with a probability very close to one, i.e., the probability of q ≥ b/(2c) is so
small that we can ignore this event when calculating expected payoffs. The region
of parameters where interesting results are found is: b < a < 2b, σ is much smaller
relative to a − b, and c < min(3/2, 2b/a). Under this condition, all the formulas can
be carried over from the case of production diseconomy except replacing c with −c.

5 S , LS and LN become infinite if t = ∞.

LM
17 Sharing Demand Information Under Simple Wholesale Pricing 377

If the supply chain is communicative, the manufacturer maximizes wE[q|Y ] −

bE[q|Y ] + cE[q2 |Y ] by setting the wholesale price to

(1 − c)(a − b) 1 − c
w = b+ + E[θ |Y ].
2−c 2−c
Given this w, the retailer’s best decision is
(3 − 2c)(a − b) (3 − 2c)E[θ |Y ]
p = b+ + .
2(2 − c) 2(2 − c)

The resulting retail quantity, a + θ − p, equals

a−b (3 − 2c)E[θ |Y ]
qS = − +θ,
2(2 − c) 2(2 − c)

which has a variance given by

δ
Var[qS ] = +λ.
4 (2 − c)2

Remark 6. For the informed manufacturer, adjusting the wholesale price in the dir-
ection of the demand signal brings higher revenue, whereas adjusting the wholesale
price against the demand signal brings about lower costs by increasing variability in
production quantity. The trade-off depends on the degree of production economy, c.
If c > 1, the wholesale price is set lower (higher) for positive (negative) demand
signal.

If the supply chain is non-communicative, the manufacturer maximizes wE[q] −

bE[q] + cE[q2 ] by setting the wholesale price to

(1 − c)(a − b)
w = b+ .
2−c
Given this w, the retailer’s best decision is
(3 − 2c)(a − b) E[θ |Y ]
p = b+ + .
2(2 − c) 2

The resulting retail quantity, a + θ − p, equals

a−b E[θ |Y ]
qN = − +θ,
2(2 − c) 2

which has a variance given by

δ
Var[qN ] = +λ.
4
378 A.Y. Ha and H. Zhang

Remark 7. If the production economy is large, c > 1, we have Var[qS ] > Var[qN ], i.e.,
information sharing makes the order quantity more variable for the manufacturer.
Furthermore, if c > 1, information sharing makes the retail price less responsive to
the demand signal.

The payoffs can be expressed as

δ δ (3 − c)δ
ΠRS = Π̄R + , ΠM
S
= Π̄M + + cλ , Π S = Π̄ + + cλ ,
4(2 − c) 2 4(2 − c) 4(2 − c)2
δ cδ (1 + c)δ
ΠRN = Π̄R + , ΠMN = Π̄M + + cλ , Π N = Π̄ + + cλ ,
4 4 4
where Π̄R , Π̄M , and Π̄ are the deterministic payoffs.6

Proposition 4. Information sharing always makes the manufacturer better off. It

makes the retailer (or the supply chain) better off if and only if c > 1.

Remark 8. If c > 1, under information sharing, the wholesale price is adjusted

against the demand signal and this alleviates the double marginalization and reduces
the average of the concave quadratic production cost by making the sales quantity
more variable. The retailer will share information voluntarily if c > 1.

Proposition 5. When the demand becomes more variable (larger σ 2 ), both the
retailer and the manufacturer are better off, regardless of whether the supply chain
is communicative or not.

Remark 9 (Retailer’s Decision Choice). If, after learning the wholesale price, the
retailer decides a retail quantity q, which will be sold at price a + θ − q, the firms
will have the same payoffs except without the term cλ . Propositions 4 and 5 will
hold exactly as stated.

17.2.4 Make-to-Stock

The manufacturer now has the option of making to stock. She has an opportunity to
produce an initial lot Q at the same time she sets the wholesale price w. We assume
that the manufacturer is obligated to meet the demand from the retailer. This is a
reasonable assumption when the manufacturer wants to maintain her reputation and
resolves to always satisfy downstream orders. The marginal cost for producing the
initial Q units is b. If q > Q, then q − Q additional units are expedited at a cost of
b per unit, b > b. If q < Q, the leftover Q − q is sold at the salvage value of v per
unit, v < b.

6 Specifically, Π̄R = (a − b)2 /[4(2 − c)2 ], Π̄M = (a − b)2 /[4(2 − c)], Π̄ = (3 − c)(a − b)2 /
[4(2 − c)2 ].
17 Sharing Demand Information Under Simple Wholesale Pricing 379

The retailer’s retail price decision p in response to an announced wholesale

price w is the same as before since he is not concerned with Q, knowing that his
sales, whatever it is, will be precisely sourced from the manufacturer at the same
wholesale price w. The retailer sets the retail price to

p = (a + E[θ |Y ] + w)/2,

which results in retail sales

1 1
q = (a − w) + θ − E[θ |Y ].
2 2
If the supply chain is communicative, the manufacturer’s expected profit condi-
tional on Y , as a function of w and Q, can be written as

E[πM |Y ] = E[wq − bQ − b (q − Q)+ + v(Q − q)+ |Y ]

= (w − b)E[q|Y ] − E[(b − v)(Q − q)+ + (b − b)(q − Q)+ |Y ].

Note that the second term is the mismatch cost of newsvendor type. The manufac-
turer chooses w and Q to maximize her expected profit. The maximization can be
done in two logical steps,
 
max E[πM |Y ] = max max E[πM |Y ] .
w,Q w Q

Since E[q|Y ] = (a − w + E[θ |Y ])/2 does not depend on Q, the inner maximization
is equivalent to minimizing the mismatch cost for a given w,

Q∗ (w) = arg min E[(b − v)(Q − q)+ + (b − b)(q − Q)+ |Y ].

Q

This is a newsvendor problem with overage cost b − v and underage cost b − b. For
any given w, we can write the residual demand, the difference between the demand
realization and its expected value, as

q̃ = q − E[q|Y ] = θ − E[θ |Y ].

We see that q̃ is independent of the choice of w. We thus have Q∗ (w) = Q̃∗ + E[q|Y ]
where
Q̃∗ = arg min E[(b − v)(Q̃ − q̃)+ + (b − b)(q̃ − Q̃)+ |Y ]
Q̃

is independent of w. We can also express the minimized inventory cost as

T S = min E[(b − v)(Q − q)+ + (b − b)(q − Q)+ |Y ]

Q

= E[(b − v)(Q̃∗ − q̃)+ + (b − b)(q̃ − Q̃∗ )+ |Y ].

380 A.Y. Ha and H. Zhang

Note that T S does not depend on w. The optimal wholesale price can then be found
by ignoring the inventory cost,
  a−b 1
arg max (P − b)E[q|Y ] − T S = arg max(P − b)E[q|Y ] = b + + E[θ |Y ].
w w 2 2

Thus, the wholesale price w has the same expression as in the case of make-to-order
with linear cost.
Similarly, if the supply chain is non-communicative, the manufacturer’s optimal
wholesale price is
a−b
w = b+ ,
2
and
T N = min E[(b − v)(Q − q)+ + (b − b)(q − Q)+ ]
Q

does not depend on w. If the information structure is conjugate normal-normal, we

can show that T S < T N , i.e., information sharing reduces the manufacturer’s inven-
tory costs.
We can work out the firms’ and the supply chain’s ex ante payoffs:

δ δ 3δ
ΠRS = Π̄R + , ΠM
S
= Π̄M + − T S , Π S = Π̄ + − T S,
16 8 16
δ δ
ΠRN = Π̄R + , ΠMN = Π̄M − T N , Π N = Π̄ + − T N ,
4 4
where Π̄R , Π̄M , and Π̄ are the deterministic payoffs.
Voluntary information sharing benefits the manufacturer but hurts the retailer.
However, if overage and underage costs, b − v and b − b, are high enough, the
inventory savings from information sharing will be high, and the supply chain will
become better off from sharing information such that it will be possible to achieve
information sharing through a side payment from the manufacturer to the retailer.
As before, greater demand variability always makes the retailer better off. How-
ever, if overage and underage costs are too high, greater demand variability will hurt
the supply chain.

17.3 Two Supply Chains Competing in Price

Consider two supply chains, each consisting of one manufacturer and one retailer.
Each manufacturer supplies her own retailer only. The retailers compete in price
(Bertrand retail competition). The demand function for retailer i is

qi = a + θ − pi + β p j ,
17 Sharing Demand Information Under Simple Wholesale Pricing 381

where pk is the retail price set by retailer k, the parameter β ∈ (0, 1) indicates com-
petition intensity, and the random variable θ , with zero mean and variance σ 2 , rep-
resents demand uncertainty.
The cost to manufacturer i for producing q units is given by bi q + ci q2 , ci > 0. We
first consider production diseconomies of scale. At the end of this section, we will
discuss information sharing under linear costs and production economies of scale.
Each retailer i has access to a demand signal Yi , which is an unbiased estimator
of θ , and he may choose, before observing the actual value of Yi , to share it with
his manufacturer. We assume a linear-expectation information structure. Define the
signal accuracy as ti = 1/E[Var[Yi |θ ]]. The accuracy ti is proportional to the sample
size when Yi is a sample mean from independent sampling. It can be shown (Ericson
1969) that
ti σ 2
E[θ |Yi ] = E[Y j |Yi ] = Yi .
1 + ti σ 2
The cost structure and the information structure are common knowledge.
We consider a multi-stage game with a sequence of events as follows:
1. Each manufacturer i decides a payment mi , which is unobservable to supply
chain j, to buy information from retailer i. Retailer i decides whether to ac-
cept this payment and, if he does, to share Yi with manufacturer i. We say that
supply chain i is communicative if retailer i shares Yi with manufacturer i, or
non-communicative otherwise. The information sharing arrangement—whether
supply chain i is communicative or not—subsequently becomes known to supply
chain j.
2. Each retailer i observes a signal Yi . This signal will be truthfully disclosed to man-
ufacturer i if supply chain i is communicative. We say that the manufacturer in a
communicative (non-communicative) supply chain is informed (uninformed).
3. Each manufacturer i determines a wholesale price wi and then each retailer i
determines a retail price pi . The wholesale price wi is unobservable to supply
chain j.
4. Market demands, q1 and q2 realize and manufacturer i supplies qi to retailer i.
Finally, firms receive their payoffs.7
Manufacturer i and retailer i make an information sharing arrangement in the
first stage, which we denote by Xi = S (sharing) or N (not sharing). The information
sharing arrangement Xi in supply chain i will become known to supply chain j. This
will be the case when firms in supply chain j observe related activities (such as
the setting up of systems for information transfer) or learn from third parties (e.g.,
consultants, vendors and employees). Then, given (X1 , X2 ), the manufacturers and
the retailers make wholesale price and retail quantity/price decisions, respectively,
before the selling season to allow time for production. We assume the production
lead-time is short relative to the selling season so that each manufacturer produces
in the selling season to meet the realized demand, with negligible shortage or over-
production.
7 We can show that, when σ and c2i are small relative to a, it is optimal for manufacturer i, with a
probability very close to one, to fully meet retailer i’s order.
382 A.Y. Ha and H. Zhang

We solve the game backward by first solving for the equilibrium wholesale prices
and retail prices and, based on these, compute the ex ante payoffs of the firms for
different information sharing arrangements. The ex ante payoffs will then be used
to solve for the equilibrium information sharing decisions in the first stage.

Lemma 1. Information sharing in a supply chain makes the retail price more re-
sponsive and the demand more variable in both supply chains.

As in a single supply chain (see Remark 1), information sharing in supply chain i
makes pi more responsive to Yi . Because retail prices are strategic complements
and the two signals are positively correlated, a more responsive pi triggers a reac-
tion from supply chain j that makes p j more responsive to Y j . A more responsive
p j makes the demand intercept facing supply chain i, a + θ + β p j , more variable.
Similarly, a more responsive Yi makes the demand intercept facing supply chain j,
a + θ + β pi , more variable.
To examine the effect of information sharing, we first look at how information
sharing in a supply chain impacts the payoffs of the other supply chain. Since sup-
ply chain i’s information sharing makes supply chain j’s demand more variable
(Lemma 1), the impact of this on supply chain j is similar to that of increased de-
mand variability in the single supply chain setting (Proposition 2), also described in
terms of three thresholds, LMS , LS , and LN .
j j j

Proposition 6 (Spill-Over Effect). When information is shared in supply chain i,

(a) if supply chain j is communicative, retailer j is better off, manufacturer j is
better off if c j < LM
S or worse off if c > LS , and supply chain j is better off if
j j Mj
c j < LSj or worse off if c j > LSj ; (b) if supply chain j is non-communicative, retailer
j is better off, manufacturer j is worse off, and supply chain j is better off if c j < LNj
or worse off if c j > LNj .

We next look at how information sharing in supply chain i impacts its own pay-
offs. It can be shown that information sharing makes retailer i worse off, thus a
retailer never has incentive to voluntarily share information with his manufacturer.
However, different from the single supply chain case, the impact on manufacturer i
is not so straightforward. While information sharing allows her to better adjust the
wholesale price to fluctuations in demand, she has to deal with greater demand vari-
ability because, when information is shared in supply chain i, the competitive reac-
tion from supply chain j makes supply chain i’s demand more variable (Lemma 1),
and the greater demand variability hurts manufacturer i if her production disecon-
omy is large. In fact, for some parameter settings,8 the overall effect of information
sharing is negative to the manufacturer, i.e., she may become worse off by receiving
information from her retailer. This will not happen in the single supply chain case.

8 One such setting is when retail competition is intense (β close to one), supply chain j is non-
communicative (X j = N) and has accurate information (large t j ), supply chain i has large produc-
tion diseconomy (large ci ) and inaccurate information (small ti ).
17 Sharing Demand Information Under Simple Wholesale Pricing 383

Information sharing can be achieved through a payment from manufacturer i to

retailer i if it benefits the supply chain (and the manufacturer by implication). The
value of information sharing in supply chain i to itself can be decomposed into the
sum of the direct effect (without considering the competitive reaction from supply
chain j) and the competitive effect (due to the reaction
√ from supply chain j). By
Proposition 1, the direct effect is positive if ci > 2 − 1. The competitive reaction
from supply chain j makes supply chain i’s demand more variable (Lemma 1) and
increases its residual demand uncertainty. The impact of this on supply chain i de-
pends on the magnitude of its production diseconomy, similar to Proposition 2 for
the single supply chain setting.
√
Proposition 7. The direct effect of information sharing is positive if ci > 2−1; the
X X
competitive effect is positive if ci < Zi j or negative if ci > Zi j , for some threshold
X
Zi j > 0.

For small c1 and c2 , the positive competitive effect can dominate the negative
direct effect such that the overall value of information sharing is positive for the
supply chain. In general, however, since the competitive effect can be either positive
or negative, competition between the supply chains can either enhance or hamper
information sharing.
We examine the equilibrium outcome of information sharing decisions for the
case of very accurate demand signals (i.e., t1 and t2 are large). In this case, the
residual demand uncertainty is negligible and the competitive effect is positive. 9
√ The
overall effect will be positive if the direct effect is also positive (i.e., ci > 2 − 1);
in fact, the overall
√ effect can still be positive even if the direct effect is somewhat
negative (ci < 2 − 1).

Proposition 8. Suppose c1 ≤√c2 . There exists cN = cN (β ) and cS1 = cS1 (β , c2 ) with

the property that cS1 ≤ cN < 2 − 1 and cS1 is decreasing in c2 , such that, given c1
and c2 , for sufficiently large t1 and t2 , (a) (N, N) is the unique equilibrium if c1 < cS1
and c2 < cN , (b) (N, S) is the unique equilibrium if c1 < cS1 and c2 > cN , (c) (N, N)
and (S, S) are the (only) two equilibria if cS1 < c1 ≤ c2 < cN , (d) (S, S) is the unique
equilibrium if c1 > cS1 and c2 > cN .

The above proposition characterizes the equilibrium outcome when demand sig-
nals are accurate. This is an interesting case of practical importance, since firms
care more about sharing information when that information is accurate and thus has
higher potential value. A supply chain will share information if its production dis-
economy is above √ a certain hurdle. Note that this hurdle is below that for a single
supply chain, 2 − 1, because the competitive effect is positive.
Another interesting practical case is when production capacity constraint is rather
rigid, i.e., the production diseconomies, c1 and c2 , are large. In this case, the direct
effect is positive and the competitive effect is negative. It turns out that, for large
c1 and c2 , the positive direct effect dominates the negative competitive effect if
X
9 The threshold Zi j in Proposition 7 becomes arbitrarily high when t1 and t2 increases.
384 A.Y. Ha and H. Zhang

the competition intensity, β , is low or moderate. Specifically, we can show that,

for β < 1/2, if c1 and c2 are both large enough, (S, S) is the unique equilibrium
information sharing outcome. Intuitively, the competitive effect is not strong under
moderate retail competition.
In sum, information sharing benefits a supply chain when the production disec-
onomy is large, and either competition is less intense or information is accurate.
Linear Production Costs If production costs are linear, c1 = c2 = 0, the residual
uncertainty has no impact and the competitive effect of information sharing is pos-
itive. The direct effect, however, is negative. We can show that, if the competition
intensity, β , is high and the signals are not too inaccurate, the competitive effect
dominates the direct effect and both supply chains sharing information is an equi-
librium.
Production Economies of Scale We now examine the situation where the produc-
tion exhibits economies of scale and the cost to manufacturer i for producing q units
is given by bi q − ci q2 . We focus on the case of both c1 and c2 being greater than one,
such that the direct effect of information sharing is positive and that, competition
aside, each retailer would share information with his manufacturer voluntarily. We
argue, however, that in this case the competitive effect is negative. From the single
supply chain analysis under production economies of scale, information sharing in
supply chain i makes pi less responsive to Yi . This makes p j , and thus θ + β p j , less
variable. The reduction in its demand variability hurts supply chain i. Nonetheless,
the overall effect of information sharing is still positive if the competition intensity,
β , is small.

17.4 Two Supply Chains Competing in Quantity

The retailers compete in quantity (Cournot competition) with the inverse demand
function for retailer i given by

pi = a + θ − qi − γ q j ,

where the parameters γ ∈ (0, 1) indicates competition intensity. We first assume that
the production exhibits diseconomies of scale and, at the end of this section, we will
discuss information sharing under linear costs and production economies of scale.
The sequence of events and decisions is the same as in the case of Bertrand retail
competition except that each retailer i now decides qi .

Lemma 2. Information sharing in supply chain i makes its retail quantity qi less
responsive to Yi but the other supply chain’s retail quantity q j more responsive to Y j .

As in a single supply chain, information sharing makes qi less responsive to Yi

(due to the wholesale price adjustment by manufacturer i). Because retail quan-
tities are strategic substitutes and the two signals are positively correlated, a less
17 Sharing Demand Information Under Simple Wholesale Pricing 385

responsive qi makes q j more responsive. The greater responsiveness of q j to Y j in

turn further reduces the responsiveness of qi to Yi . A less responsive qi makes the
demand intercept facing supply chain j, a + θ − γ qi , more variable. A more respon-
sive q j makes the demand intercept facing supply chain i, a + θ − γ q j , less variable.
In other words, information sharing in a supply chain makes the demand for itself
less variable but the demand for the supply chain more variable.
To examine the effect of information sharing, we first look at the spill-over effect,
i.e., how information sharing in a supply chain impacts the payoffs of the other
supply chain. By Lemma 2, information sharing in supply chain i makes the demand
for supply chain j more variable. Proposition 3 then implies the following.

Proposition 9 (Spill-Over Effect). When information is shared in supply chain i,

(a) if supply chain j is communicative, both retailer j and manufacturer j are better
off; (b) if supply chain j is non-communicative, retailer j is better off but manufac-
turer j is worse off, and supply chain j is worse off iff c j > 1, i.e., the production
diseconomy in supply chain j is large.

We now look at how information sharing in supply chain i impacts its own pay-
offs. It is easy to show that information sharing makes manufacturer i better off and
retailer i worse off. Therefore, a retailer has no incentive to voluntarily share infor-
mation with his manufacturer. The manufacturer may use a side payment to induce
information sharing and that is possible if the supply chain’s total profit is increased.
The value of information sharing in supply chain i to itself can again be decom-
posed into the sum of the direct effect
√ and the competitive effect. By Proposition 1,
the direct effect is positive if ci > 2 − 1. The competitive reaction from supply
chain j makes supply chain i’s demand less variable (Lemma 2) and hurts the com-
municative supply chain i (Proposition 3).
√
Proposition 10. The direct effect of information sharing is positive if ci > 2 − 1;
the competitive effect of information sharing is negative.

Because of the negative competitive effect, competition between supply chains

hampers information sharing. However, the overall effect will positive if the pro-
duction S N
√ diseconomy is large enough. Specifically, there exist thresholds ci (c )
> 2 − 1 such that, when supply chain j is (non)communicative, information shar-
ing in supply chain i benefits itself if ci > cSi (cN ).
The next proposition shows the equilibrium outcome of information sharing de-
cisions.

Proposition 11. Suppose c1 ≤ c2 . (a) If c1 < cS1 and c2 < cN , (N, N) is the unique
equilibrium. (b) If c1 < cS1 and c2 > cN , (N, S) is the unique equilibrium. (c) If
cS1 < c1 ≤ c2 < cN , (N, N) and (S, S) are the (only) two equilibria. (d) If c1 > cS1 and
c2 > cN , (S, S) is the unique equilibrium.
386 A.Y. Ha and H. Zhang

We make several observations:

1. Supply chain 2 may be better off when it suffers a drawback in forecasting cap-
ability (lower t2 ) or in production diseconomy (larger c2 ), if the equilibrium
changes from (N, S) to (S, S). This is due to the spill-over effect that benefits
supply chain 2 when these parameter changes induce supply chain 1 to share
information (Proposition 9).
2. Supply chain 1 may be better off when it suffers a drawback in forecasting capa-
bility (lower t1 ) if the equilibrium changes from (N, N) to (N, S). This is due to
the spill-over effect on supply chain 1 when the parameter change induces supply
chain 2 to share information.
3. Under certain parametric conditions, (S, S) is the unique equilibrium but both
manufacturers would be better off at (N, N). This prisoners’ dilemma situation is
also observed in Ha and Tong (2008).
4. When t1 = t2 , b1 = b2 , and c1 < c2 , supply chain 1 always has a higher payoff than
supply chain 2, but manufacturer 1, who has a smaller production diseconomy,
may earn less profit than manufacturer 2. This happens when the equilibrium
is (N, S) where manufacturer 1, unlike manufacturer 2, cannot use responsive
wholesale pricing and therefore earns a smaller share of her supply chain’s profit.
Linear Production Costs If production costs are linear, c1 = c2 = 0, information
sharing always hurts a supply chain because both the direct and competitive effects
are negative, and thus no information sharing is the unique equilibrium.
Production Economies of Scale Production exhibits economies of scale and the
cost to manufacturer i for producing q units is given by bi q − ci q2 .
For ci > 1, the direct effect of information sharing is positive (Proposition 4). As
for the competitive effect, for ci > 1, information sharing in supply chain i makes
qi more responsive to Yi . By strategic substitutability, this makes q j less responsive
to Y j and in turn makes θ − γ q j more variable. The increase in its demand variability
benefits supply chain i. Therefore, as long as ci > 1, information sharing in supply
chain i benefits both retailer i and manufacturer i, regardless of whether supply
chain j is communicative or not.
For ci < 1, the direct effect of information sharing is negative (Proposition 4). As
for the competitive effect, for ci < 1, information sharing in supply chain i makes
qi less responsive to Yi . By strategic substitutability, this makes q j more responsive
to Y j and in turn makes θ − γ q j less variable. The decrease in its demand variability
harms supply chain i. Therefore, as long as ci < 1, information will not be shared in
supply chain i, regardless of whether supply chain j is communicative or not.
In short, under production economies of scale, Cournot competition between sup-
ply chains has little impact on the information sharing decisions in each individual
supply chain.
17 Sharing Demand Information Under Simple Wholesale Pricing 387

17.5 One-to-Many Supply Chain

Double marginalization, competition, and information asymmetry are often cited as

causes of supply chain inefficiency because they allegedly distort firms’ incentives
and drive individual decisions away from overall optimality. However, these three
factors do not always distort incentives in the same direction. On the contrary, their
interplay may sometimes balance out one another’s effect such that the supply chain
performs close to, or even exactly at, optimality.
Li and Zhang (2008) consider a two-echelon supply chain of one manufacturer
and n retailers. The manufacturer provides a common base product at a wholesale
price of w per unit and the retailers further process it to make similar but nonidenti-
cal products through different processes of customization. The retailers compete in
price in the end market for which the demand function is given by
φ
n−1 ∑
qi = a + θ − (1 + φ )pi + p j.
j=i

The products are imperfect substitutes and so φ > 0. A larger φ indicates a higher
degree of substitution and greater intensity of retail competition. Each retailer i has
access to a private signal Yi about θ and may decide to share it with the manufacturer
before Yi is observed. The manufacturer decides w after receiving shared signals.
The information exchange is said to be confidential if the manufacturer keeps the
received information to herself, or nonconfidential if she gives that information to
some or all retailers.
Confidential vertical information sharing centralizes each retailer’s dispersed
information at the manufacturer without direct disclosure to other retailers. The
wholesale price signals the information to retailers who, on the basis of it, form
their rational expectation on the market condition. This strategic signaling has a dra-
matic impact on the wholesale pricing. Without confidentiality, the wholesale price
is simply a unit cost to retailers and a unit revenue to the manufacturer. In the pres-
ence of confidentiality, it plays an additional role of signaling demand information.
A higher (lower) wholesale price signals to retailers a more (less) favorable market
condition, inducing them to set higher (lower) retail margins which in turn reduce
(increase) retailers’ orders from the manufacturer. Hence, the signaling effect ren-
ders the manufacturer’s demand more elastic to the wholesale price. It is this added
price elasticity for the manufacturer, when information is shared confidentially, that
prompts her to lower the wholesale price, alleviating the double marginalization
effect and improving the supply chain efficiency.
This intuitive argument seems to point to a condition under which confidential-
ity improves the supply chain efficiency; namely, the signaling effect of an increase
(decrease) in the wholesale price has to result in a decrease (increase) in retail quan-
tities. On the other hand, the efficiency result does not hold if the aforementioned
condition fails. For example, if the retailers compete in quantity, confidentiality
makes the wholesale price higher, impairing the supply chain efficiency.
388 A.Y. Ha and H. Zhang

Li and Zhang (2008) show: (1) Without confidentiality, information sharing

always harms the retailers. (2) With confidentiality, information sharing benefits
all parties if retail competition is intense. (3) When all retailers share their infor-
mation confidentially, truthful reporting by all retailers is an equilibrium and the
decentralized decisions achieve the centralized optimal solution. In other words,
confidentiality induces truth telling and coordinates the supply chain.

17.6 Two-to-One Supply Chain

Shang et al. (2016) consider a supply chain with two identical manufacturers selling
substitutable products to a retailer. Manufacturer i charges a wholesale price wi and
the retailer determines the retail prices pi and p j . The demand function is given by

qi = a + θ − (1 + φ )pi + φ p j .

Before the retailer observes a private demand signal, he makes concurrent and
identical offers to the manufacturers by charging each of them the same payment
for sharing the demand information. Consistent with the case of a single supply
chain, information sharing occurs when production diseconomy/economy is large.
However, unlike the case of two competing supply chains, information sharing oc-
curs when competition is intense. This can be explained as follows. With manufac-
turer’s production diseconomy, information sharing between a manufacturer and the
retailer makes double marginalization more pronounced (a negative effect) and re-
duces the variability of production quantity (a positive effect). More intense compe-
tition makes an informed manufacturer’s wholesale price less responsive to the de-
mand signal, which alleviates the negative double marginalization effect and makes
information sharing more valuable. With manufacturer’s production economy, from
Remark 6 in Sect. 17.2, more intense competition makes the revenue effect less
significant and an informed manufacturer will adjust her wholesale price against the
demand signal at a lower degree of production economy. Indeed, Shang et al. (2016)
show that this happens when c > 1/(1 + φ ).
Shang et al. (2016) demonstrate that partial information sharing (i.e., the retailer
shares information with only one manufacturer) can be an equilibrium under pro-
duction economy but not production diseconomy. With production economy, when
the retailer shares information with one of the two firms, say manufacturer i, wi re-
sponds to the demand signal and makes q j , manufacturer j’s production quantity,
more variable. The increase in q j ’s variability is higher if production economy is
larger. When the retailer also shares information with manufacturer j, both wi and w j
responds to the demand signal and they drive q j ’s variability in the opposite direc-
tions. If production economy is large enough, manufacturer j could be worse off by
receiving information from the retailer because q j becomes less variable when com-
pared with the case that only manufacturer i is informed. It can be shown that man-
ufacturer i and the retailer could also be worse off. Thus, sharing information with
17 Sharing Demand Information Under Simple Wholesale Pricing 389

the second manufacturer may not create a positive value to the supply chain. With
production diseconomy, sharing information with the second manufacturer benefits
both manufacturers. It turns out that the positive production variability effect domi-
nates the negative double marginalization effect, and therefore it benefits the supply
chain too.
Shang et al. (2016) also consider the case when the retailer makes sequential
offers to the manufacturers for sharing the demand information. They show that
partial information sharing now may occur under production diseconomy, and it
occurs over a larger parametric space (when compared with concurrent offering)
under production economy. When the incremental value of sharing information with
the second manufacturer is not high, the retailer can extract more profit by selling
information to only one manufacturer. This is because the manufacturer is willing to
pay a premium to gain an information advantage in competing with an uninformed
rival.

17.7 Concluding Remarks

Most of the studies on information sharing in supply chains assume a linear structure
in supply and demand—linear production costs and linear demand function. While
linearity assumptions facilitate analyses, it limits the application of insights derived.
This chapter provides a summary of a few recent results with various forms of non-
linear production costs, including economy and diseconomy of scale and make-to-
stock with expediting. This is but a first step in incorporating nonlinearity into the
study of supply chain information sharing.
Nonlinearity, inherent in many of key supply chain processes, may dramatically
alter the incentive for sharing information. It will be interesting to see how non-
linearity in the demand function changes the impact of information sharing. The
problem is challenging but can probably be tackled for some specific forms of non-
linear demand functions such as power and exponential functions, starting from a
single supply chain and incrementally progressing to competing supply chains.
To describe demand uncertainty for a demand function with multiple parameters,
one has to be specific as to which of these parameters are uncertain. This chapter
has focused on uncertainty about the intercept of the demand function. Malueg and
Tsutui (1996) study the duopoly information exchange when the slope of the de-
mand function is unknown and obtain very different results from those for the case
of intercept uncertainty. To our knowledge, no study has yet examined information
sharing in supply chains where the information is about the uncertain slope.
One key assumption we have made is that the manufacturer either makes to order
or makes to stock with expediting. This guarantees that there is no shortage in the
retail market. But in reality production delay does cause shortage. Li and Zhang
(2015) consider information sharing in a single supply chain where the manufacturer
390 A.Y. Ha and H. Zhang

has just one opportunity to build up an initial stock, without the chance of second
production. Generalizing their study to competing supply chains would be a worth-
while endeavor.
In sum, results about sharing demand information in supply chains reported in
this chapter are derived under some key assumptions. Therefore one must interpret
them with a measure of caution when applying them to specific practical situations
where these assumptions are questionable. This very limitation also calls for con-
tinued effort in this important and fruitful field as supply chains expand their global
reach and more information of various forms becomes available.

References

Beckett J (2012) A CG manufacturer’s guide to retail data gold mines. VMT and CGT White paper
Ericson WA (1969) A note on the posterior mean of a population mean. J R Stat Soc 31(2):332–334
Ha A, Tong S (2008) Contracting and information sharing under supply chain competition. Manag
Sci 54(4):701–715
Ha A, Tong S, Zhang H (2011) Sharing imperfect demand information in competing supply chains
with production diseconomies. Manag Sci 57(3):566–581
Keifer S (2010) Beyond point of sale data: Looking forward, not backwards for demand forecast-
ing. GXS White paper
Li L, Zhang H (2002) Supply chain information sharing in a competitive environment. In: Song
JS, Yao DD (eds) Supply chain structures: coordination, information and optimization. Kluwer
Academic Publishers, Norwell
Li L, Zhang H (2008) Confidentiality and information sharing in supply chain coordination. Manag
Sci 54(8):1467–1481
Li T, Zhang H (2015) Information sharing in a supply chain with a make-to-stock manufacturer.
Omega 50:115–125
Malueg D, Tsutui S (1996) Duopoly information exchange: The case of unknown slope. Int J Ind
Organ 14:119–136
O’Marah K (2013) Effective collaboration in trading partner relationships. SCM World research
report
Shang W, Ha A, Tong S (2016) Information sharing in a supply chain with a common retailer.
Manag Sci 62(1):245–263
Vives X (1999) Oligopoly pricing. The MIT Press, Cambridge