This publication was co-financed by the European Union from the European Social Fund.

Lecture notes on
Microeconomics

by

Lukasz Woźny

Warsaw School of Economics

2016
This publication is the result of the project Mlodzi projektuja֒ zarzadzanie
֒ co-financed
by the European Social Fund within Human Capital Operational Programme, Priority
IV ”Higher education and science,” Measure 4.1 ”Strengthening and development of
didactic potential of universities and increasing the number of graduates from faculties
of key importance for knowledge-based economy,” Sub-measure 4.1.1 ”Strengthening
and development of didactic potential of universities.”

Publisher:
Szkola Glówna Handlowa w Warszawie (SGH)

Reviewer: Marek Garbicz

First Edition

c Copyright by Szkola Glówna Handlowa w Warszawie (SGH)

al. Niepodleglości 162
02-554 Warszawa, Polska

ISBN: 978-83-65416-11-7

Circulation: 200 copies

Cover design: Monika Trypuz

Typesetting and printing

Agencja Reklamowa TOP
87-800 Wloclawek, ul. Toruńska 148
tel. 54 423 20 40
fax 54 423 20 80
e-mail: [email protected]
Preface

These notes are prepared for the Microeconomic courses I teach at the Warsaw School
of Economics. They are aimed to serve as a supplementary material for Microeco-
nomic course at the introductory or intermediate level. The material covers canonical
first level microeconomic topics including: consumer and producer choice, as well as
competitive and monopolistic (partial) equilibrium analysis. If time allows (and it
usually did during 15 meetings, hour and a half each) I also recommend to cover ad-
ditional topics including: choice under uncertainty, introduction to non-cooperative
games, selected issues from industrial organization or externalities (including analysis
of public goods). Finally to introduce the reader to more advanced microeconomic
topics I have prepared two short chapters on general equilibrium analysis as well as
economics of asymmetric information. These are, however, only sketched here. The
selection of material covered can depend also on major taught, and can vary be-
tween economics, finance or management. Each chapter includes a separate section
with (subjectively selected) references to some important further readings. Finally,
although material presented here is usually more than enough to cover during a stan-
dard one semester course, some important economic topics / disciplines are missing
here (including public choice, mechanism design, cooperative game theory to mention
just a few).
Clearly, the notes are far from being complete and cannot compensate for reading
a full textbook on Microeconomics. One reason is that some (important) details are
missing here. Firstly, whenever not restrictive to present the main argument I use
standard tools from constrained optimization for differentiable objectives and con-
straining functions, hence ”non-smooth”/discrete case is not covered here. Secondly,
as the exposition is mainly aimed to show the basic trade-offs but not solve all the
problems, I only occasionally discuss the second order optimality conditions. Thirdly,
when presenting some theorems or statements I miss their proofs but give a reference
for such or sketch an argument when necessary. I tried to keep the exposition clear,
though.
Writing these lectures I used Besanko and Braeutigam (2011), Varian (1992),
Mas-Colell, Whinston, and Green (1995) textbooks and which I recommend for a

3
4 REFERENCES

more detailed treatment of topics covered here (for introductory, intermediate and
advanced level respectively). I also recommend a textbook by Nicholson and Snyder
(2012) that presents intuitively and exemplifies many concepts covered in these notes.
Finally, I want to thank Pawel Dziewulski for reading an early draft of these notes
as well as the Department of Economics, University of Oxford for hosting during the
writing of these notes.

References
Besanko, D., and R. Braeutigam (2011): Microeconomics. International student
version. Wiley.

Mas-Colell, A., M. D. Whinston, and J. R. Green (1995): Microeconomic

theory. Oxford University Press.

Nicholson, W., and C. M. Snyder (2012): Microeconomic theory: basic principles

and extensions. South-Western, Cengage Learning, Mason OH.

Varian, H. (1992): Microeconomic analysis. W.W. Norton & Company, New York.
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1 Introduction to economic methods . . . . . . . . . . . . . . . . . . 7

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Consumer theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Demand and consumer surplus . . . . . . . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Producer theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1 Technology and output . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4 Perfect competition . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.1 Company at the perfectly competitive market . . . . . . . . . . . . . . 33
4.2 Competitive equilibrium and welfare . . . . . . . . . . . . . . . . . . . 34
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5 Monopoly and pricing . . . . . . . . . . . . . . . . . . . . . . . . . . 37

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

6 Risk and ambiguity . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

6.1 Expected utility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2 Evaluating risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.3 Subjective probability and state dependent utility . . . . . . . . . . . 49
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

5
6 CONTENTS

7 Game theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.1 Strategic form games . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.2 Extensive form games . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

8 Oligopoly and industrial organization . . . . . . . . . . . . . . . . 61

8.1 Cournot model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.2 Stackelberg model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.3 Bertrand model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.4 Monopolistic competition . . . . . . . . . . . . . . . . . . . . . . . . . 66
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

9 General equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
9.1 Exchange economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
9.2 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

10 Asymmetric information . . . . . . . . . . . . . . . . . . . . . . . . 81
10.1 Adverse selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
10.2 Principal-agent problem . . . . . . . . . . . . . . . . . . . . . . . . . . 83
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

11 Externalities and public goods . . . . . . . . . . . . . . . . . . . . . 87

11.1 Public goods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
11.2 Externalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Chapter 1

Introduction to economic
methods

A traditional definition of economics, advocated by Lionel Robbins, says that Eco-

nomics is the science which studies human behavior as a relationship between ends and
scarce means which have alternative uses. Resources are typically limited but needs
need not be, hence economists analyze trade-offs how to allocate their constrained re-
sources towards alternative uses. The typical questions that can be addressed within
this framework are: what goods and services to produce, how to produce them effi-
ciently, how to allocate goods and services among consumers, how to allocate gains
from production/trade among consumers etc. These questions can be analyzed at
micro or macro level. From perspective of economic theory this distinction is, to a
large degree, irrelevant but for applied economics it is important, as microeconomics
studies behavior of individuals (like consumers, producers, firms, managers etc.), while
macroeconomics studies behavior of aggregate variables (like employment, gross do-
mestic product, inflation etc.). Traditionally economic questions (or more precisely
answers) are divided into positive and normative. Positive economics explains
how economy works or predicts some future trends, while normative economics is
more concerned in social welfare and policy recommendations.
Some more modern definitions of economics stress that economics deals with in-
centives, or as Steven Landsburg puts it: Most of economics can be summarized in
four words: people respond to incentives. The rest is commentary. This means that
the fundamental question in economics is the analysis of incentives that govern in-
dividual choices but also how to design or manipulate incentives so that responding
individuals will behave in a desired way.
To a large extend economics is an operational science, i.e. economists try to solve
real life/economy problems. This does not mean that economists do not use formal

7
8

or theoretical tools. On the contrary much of economic research is based on abstract

models. This requires a comment. Real life economic problems are typically very
complicated and it is difficult to analyze them in their full complexity. For this reason
economists create models that are supposed to represent the main / important trade-
offs and problems of interest, but still abstract from other (non critical) elements.
Hence, by construction models are not realistic and based on questionable assump-
tions. For this reason economists say that: All models are unreal but some are useful,
stressing applicability of model’s results. This argument has been taken to extreme
by Milton Friedman in his ”as if” methodological proposal. Friedman stressed that
until model’s results fits the real data we can say that individual or economy behaves
“as if” it was generated by the model. In such case we say that model represents
reality. This is indeed appropriate in some applications, but still one shall be careful,
when going with this hypothesis too far. More on methodology of economics can be
found in a book by Blaug (1992).
There are three typical methods economists use: constrained optimization, equi-
librium analysis and comparative statics. We now briefly describe each of them.
Example of a constrained optimization problem is:

max f (x, θ),

x∈X
s.t. g(x, θ) ≤ m,

where X is a domain, f : X × Θ → R is an objective function, g : X × Θ → Rn

is a constraint function and m, θ are parameters (vectors in Rn and Θ). Variable
x is called endogenous as it is chosen/determined within the problem, while m, θ
are exogenous (parameters). Set {x ∈ X : g(x, θ) ≤ m} is called a feasible set. A
typical example of an objective is a profit of the company, where X, g represent tech-
nological constraints. Another example of an objective is a utility of the consumer,
X his consumption set, while g(x, θ) ≤ m represents her budget constraint. Finally,
we can think of f as representing some social preferences, and the decision problem
is to find the socially optimal outcome. Note that, these are all maximization prob-
lems, but economists sometimes analyze minimization problems, e.g. minimize costs
of producing at least x units of output. All in all, constrained optimization is part of
a decision theory and is one of the most typical economic tools. As we mentioned be-
fore, economists use constrained optimization problems for various reasons. They (i)
solve ’real life’ constrained optimization problems to find the best feasible solution; or
(ii) assuming a family of optimization problems (for various θ) is given they estimate
parameters θ from the observed data, such that solutions to the optimization problem
are similar to the one observed in the real data. Finally, they also (iii) ’invent’ con-
strained optimization problems, whose solutions coincide with observed choices. In
(i) they often create a recommendation for a decision maker, forecast some economic
variables or explain incentive and trade-offs that decision maker must be aware of; in
1 INTRODUCTION TO ECONOMIC METHODS 9

Figure 1.1: Examples of equilibrium analysis.

p p S(p)

bc b p∗
S(p)
D(p) D(p)
b

x x∗ x
p S(p) p S(p)
b
p∗3

p∗2
b p∗

b
p∗1 D(p) D(p)

x∗1 x∗2 x∗3 x x∗ x

(ii) they provide information about parameters θ that can be used for other economic
considerations (comparisons, forecasts), or provide interpretation of identified θ in
behavioral terms; finally in (iii) economists must invent X, f, g so that the solutions
to constrained optimization problem represent (usually uniquely) an observed real life
data. This is linked to revealed preference argument, as decision maker reveals his
objective and constraints via actual choices.
The second tool economists often use, is an equilibrium analysis. At this level
we use two general notions of equilibrium: (i) competitive equilibrium of an economy
in their partial or general incarnation and (ii) Nash equilibrium of a game. Both
notions of equilibrium can be understood as a fixed point, i.e. a point x∗ ∈ X such that
f (x∗ ) = x∗ for some f : X → X. In the competitive (partial) equilibrium example, an
equilibrium is a pair of price and quantity (p∗ , x∗ ), such that supply equals demand
S(p∗ ) = x∗ = D(p∗ ), which could be represented by a fixed point of some function f .
Figure 1.1 indicates (for the competitive equilibrium) equilibria may be non-existent,
unique, multiple or even form a continuum. A notion of general equilibrium is more
complicated and we leave it to be defined later in chapter 9. Similarly in a game, where
some players play against each other we would like to find a stable solution such that
no player has an incentive to change its own decision assuming that other will also
do not change theirs. In such case a (Nash) equilibrium is a fixed point of so called
best response map f . Having established equilibrium existence economist conduct
10 REFERENCES

parameter estimation, forecasts or just explain observed phenomenas. Also revealed

preference arguments are used. Finally, to mention, both supply and demand (or best
responses in the game equilibrium context) are functions/correspondences derived
from constrained optimization problems.
Finally for both types of tools: constrained optimization and equilibrium analysis
economists conduct comparative statics exercises1 . That is, they analyze how
a solution of a constrained optimization problem or equilibrium changes, if some
parameters θ of the primitives are changed. This includes forecasting or explaining
possible reactions to changes in policy parameters or tastes, for example.

References
Amir, R. (2005): “Supermodularity and complementarity in economics: an elemen-
tary survey,” Southern Economic Journal, 71(3), 636–660.

Blaug, M. (1992): The methodology of economics or how economists explain. Cam-

bridge University Press.

Milgrom, P., and C. Shannon (1994): “Monotone comparative statics,” Econo-

metrica, 62(1), 157–180.

Quah, J. K. H. (2007): “The comparative statics of constrained optimization prob-

lems,” Econometrica, 75(2), 401–431.

Samuelson, P. A. (1947): Foundations of economic analysis. Harvard University

Press.

Topkis, D. M. (1998): Supermodularity and complementarity, Frontiers of economic

research. Princeton University Press.

1 Historically
the classic ”first order” comparative statics was formalized by Samuelson (1947).
See Milgrom and Shannon (1994), Topkis (1998), Amir (2005) or Quah (2007) for some modern
developments in comparative statics.
Chapter 2

Consumer theory

In this chapter we will analyze consumer choices. This is important to understand

both (i) the demand for goods and services that consumers consume as well as (ii)
supply of endowments the consumers (initially) possess (e.g. capital or labor). Our
approach to understand consumer choices could be summarized in the following points:

1. observe real data about the pairs of consumption bundles chosen and prices of
goods and services in that bundle,

2. construct preferences that represent the observed choices under observed prices,

3. represent constructed preferences by some utility function,

4. analyze theoretically (for the purpose of forecasting or past behavior expla-

nation) the choices of consumers with derived utility function under various
incomes and market prices.

The following two sections will address all four points. At this level, however, we
start from the 2nd one rather than the 1st, i.e. we assume that the preferences of
consumers are given and then conduct the analysis of points number 3 and 4. At the
end we will go back to the 1st point.

2.1 Preferences
Consider a consumer faced with choices from a given set X. This is a set of all
consumption bundles consumer may think to consume and is called a consumption
set. At this level we will assume that X ⊂ RK+ , where K stands for the number of
goods. That is, we will consider consumption of nonnegative amounts of perfectly
divisible goods, and assume that number of goods is finite.

11
12 2.1 Preferences

One should note that K could be very large, though. Specifically, when differen-
tiating goods economists use at least these four criteria:

• physical characteristics (clearly apple is different from orange, by its color, size,
taste, flavor, etc.),

• location (clearly for a consumer located in Warsaw the goods delivered on Sahara
desert give much less satisfaction, than on place),

• time (clearly the goods promised to be delivered the next quarter give lower
satisfaction, than goods consumed today),

• state of the world (clearly an umbrella gives higher utility, when it is raining
then not).

One could perhaps think of some more criteria that differentiate goods, and the key
insight, when doing so could be reasoned from so called law of one price: goods
which prices differ, should be regarded in principal as distinct. That is, consumers
need a good reason to purchase similar goods at different prices. All in all, one can
easily see that in reality the number of goods can be infinite. At this level we assume
that K is finite1 , though, could be very large.
The consumer is assumed to have preferences denoted by  over bundles in X. If
for x, x′ ∈ X we write x′  x we mean that x′ is weakly preferred to x. That is, con-
sumer ranks his satisfaction from consuming x′ weakly higher, than from consuming
x, assuming both are available at no costs. Specifically, preferences describe consumer
wants, but not what she wants from bundles she can afford. In what follows we need
the following assumptions:

1. preferences are complete, i.e. for any x′ , x ∈ X, x′  x, or x  x′ . Hence

x′ could be preferred to x, or x could be preferred to x′ or both, meaning that
consumer is indifferent between x′ and x. In such a case we write x′ ∼ x,

2. preferences are transitive, i.e. if x′′  x′ and x′  x, then we require that

x′′  x. This assumptions means that consumer makes consistent choices,

3. weak monotonicity, i.e. if x′ ≥ x, then2 x′  x. This assumption states that

more is better: if bundle x′ has weakly more units of goods than x, than x′
must be preferred to x. However, if x′  x, i.e. bundle x′ has more food but
1 The analysis of consumer choice when the number of goods is infinite is important at the macroe-

conomic level, as economists analyze the dynamic economies with the infinite time horizon. Every
period the same physical good is consumed / traded but as the time horizon is infinite the number
of goods is infinite as well. For a discussion see Stokey, Lucas, and Prescott (1989) chapter 15 or
Aliprantis, Brown, and Burkinshaw (1990).
2 By ≥ in RK we mean a standard order on RK , i.e. x′ ≥ x if all coordinates of x′ are ≥ the x.
2 CONSUMER THEORY 13

less clothing than x, then without additional information we cannot say which
bundle is preferred. Weak monotonicity is sometimes strengthen to strong
monotonicity, i.e. if x′ ≥ x and x′ 6= x, then x′ ≻ x meaning that, if x′ has
strictly more units of some good and weekly more of the others, then x′ must
be strictly preferred3 to x.

Preferences that are complete and transitive are called rational.

Strict monotonicity can be restrictive as it means that goods are good, so it may
not apply to garbage or pollution. But if one interprets goods as less garbage or less
pollution, the strict monotonicity may be often assumed.
There are two other important assumptions concerning continuity of preferences
and their (strict) convexity. Continuity is technical and we will not discuss it at
this level. Convexity (resp. strict convexity) assumption means that if x′′ , x′ , x ∈ X
are such that x′′  x, x′  x (and x′′ 6= x′ resp.) then for any t ∈ (0, 1) we have
tx′′ + (1 − t)x′  x (resp. tx′′ + (1 − t)x′ ≻ x). Hence, convexity means that consumer
prefers to mix extreme bundles, rather than consume one of them. Strict convexity
is not always required nor satisfied.
In our discussion on preferences so far we have used the ordinal ranking. That is,
we know how consumer ranks / prefers bundle x′ to x, but we do not know how much
more he likes x′ to x. To analyze this we will switch to cardinal ranking. By cardinal
ranking we mean a ranking that can tell us, that consumer prefers e.g. x′ twice as much
as x. For this reason and also because operations on ordinal rankings are cumbersome
economists introduce the concept of a utility function. Utility function u : X → R
is a function that assigns numbers to every basket. We say that utility function u
represents preferences , if for any x′ , x ∈ X such that x′  x we have u(x′ ) ≥ u(x).
It can be shown4 that, if preferences are complete, transitive and continuous then there
exists a continuous utility function that represents them. Observe that the value of
utility does not have any particular interpretation, as any monotone transformation
of utility function u represents the same preferences5 .
Having such representation of preferences we can concentrate on the analysis of
utility u(x1 , x2 , . . . , xK ), where x = (x1 , . . . , xK ) denotes a typical bundle in X ⊂ RK .
We start by introducing some important concepts. First we define marginal utility.
△u
The marginal utility of good i is denoted by M Ui , defined by M Ui = △x i
and
interpreted as a rate at which utility changes as the level of consumption of good
i raises, holding consumption of other goods constant. For the differentiable utility
∂u
function M Ui = ∂x i
.
3 By this we mean that x′  x and x  x′ .
4 See Debreu (1964) theorem.
5 Let f : R → R be a strictly increasing function (monotone transformation). If utility function u

represents preferences  for any x′ , x ∈ X, then whenever x′  x, u(x′ ) ≥ u(x). At the same time
f (u(x′ )) ≥ f (u(x)). Hence, f ◦ u is also a utility function representing .
14 2.1 Preferences

Another important concept is an indifference curve. This is a set of all bundles

in X with equal utility. That is, a set of all bundles that consumer is indifferent
between (see figure 2.1).
Suppose now we increase consumption of good i, how much we have to reduce
the amount of good j consumed to keep the utility constant? This is measured by a
marginal rate of substitution of xi for xj . It is denoted by M RSi,j and calculated
dx
as M RSi,j = − dxji |u=const. . It can be shown6 that for differentiable utility the follow-
M Ui
ing useful formula holds: M RSi,j = M Uj . Graphically marginal rate of substitution
is a tangent to the indifference curve and hence serves as a local approximation of the
rate at which consumer is willing to trade-off goods in his basket along the indifference
curve. It can be shown that marginal rate of substitution does not depend of utility
representing particular preferences.
Let us now summarize the few basic properties of rational, strongly monotone
preferences:

• When M Ui is positive for all goods i, then indifference curves have a negative
slope. Why?

• Indifference curves for two different utility levels cannot intersect. Why?

• Every bundle lies on only one indifference curve. Why?

• Indifference curves are not thick. Why?

Now in a few examples for K = 2, we discuss important special cases of preferences

and analyze their properties using concepts defined at the moment.

Example 2.1 (Cobb-Douglas) A typical example of a utility function is a Cobb-

β
Douglas function u(x1 , x2 ) = xα1 x2 for α, β ≥ 0. The marginal utilities are M U1 =
α−1 β β−1
αx1 x2 and M U2 = βxα 1 x2 . Observe that M U1 is positive but x → M U1 is
decreasing for α < 1. The last property is called diminishing marginal utility and
denotes preferences for which each next unit of good 1 gives less and less utility for a
consumer. The marginal rate of substitution: M RS1,2 = αxβx1 and again x1 → M RS1,2
2

is decreasing due to convexity of preferences.

Example 2.2 (Prefect substitutes) Consider u(x1 , x2 ) = αx1 +βx2 . The M U1 =

α, M U2 = β and are constant rather then decreasing. The M RS1,2 = αβ . This means
that consumer has a constant rate at which she is willing to exchange goods at the
indifference curve, as both goods are perfectly substitutable.

6 By dxj M Ui
construction changes dxi and dxj satisfy M Ui dxi + M Uj dyj = 0. Hence dxi
= −M U
.
j
2 CONSUMER THEORY 15

Figure 2.1: Examples of indifference curves for convex preferences: perfect comple-
ments (right, top panel); perfect substitutes (right, bottom panel); quasilinear (left,
bottom panel).

x2 x2

x1 x1
x2 x2

x1 x1

Example 2.3 (Perfect complements) Consider u(x1 , x2 ) = min{αx1 , βx2 }. This

means that consumer’s utility is always limited by the minimal of the two numbers:
consumption of αx1 or βx2 . It means that consumer needs to consume these goods in
β
fixed proportion α . Observe that such indifference curve is not differentiable at any
point (x1 , x2 ), such that αx1 = βx2 .

Example 2.4 (Quasilinear preferences) Let u(x1 , x2 ) = v(x1 ) + x2 . Such prefer-

ences are called quasi-linear in x2 , as the utility level depends linearly in x2 , but not
1
necessarily in x1 . The M U2 = 1 and M RS2,1 = v′ (x 1)
and is constant in x2 .

Example 2.5 (Homothetic preferences) An important class of preferences are so

called homothetic preferences. Formally homothetic preferences are defined by the
following implication: if u(x) = u(x′ ) for two bundles x, x′ ∈ X, then u(tx) = u(tx′ )
for any t ∈ R+ . It means that, if the consumer is indifferent between two baskets,
then he will be still indifferent, if we multiple two baskets by a scalar.

The next figure 2.1 exemplifies indifference curves representing preferences dis-
cussed so far.
16 2.2 Choice

2.2 Choice
Here we will analyze a consumer’s choice. Let X ⊂ RK + and denote a typical bundle
x = (x1 , x2 , . . . , xK ). We start by defining a budget set. Given income I and prices
p = (p1 , p2 , . . . , pK ) of all goods, a budget set is a subset of X that consumer can
afford. A typical example of a budget set is B(p, I) = {(x1 , x2 , . . . , xK ) ∈ X ⊂ RK + :
PK
x p
i=1 i i ≤ I}. Note, that for any positive prices p, budget set is convex, which will
PK
play an important role in our analysis7 . The relation i=1 xi pi = I defines a budget
constraint. If one considers nonlinear prices, e.g. some price discounts, if consumer
buys at least a number of particular goods, then the budget set could be given by
more than one inequality and need not be convex.

Example 2.6 For a two goods case x1 , x2 the budget constraint can be written as
x1 p1 + x2 p2 ≤ I and the slope of the budget line is given by − pp12 . An increase in
income shifts the budget parallelly, further from the origin. An increase in the price
of a particular good changes the slope of budget line − pp21 accordingly. The slope of
the budget line gives the ratio at which one can exchange good 1 for 2 at the market.
This is illustrated in figure 2.2.

Figure 2.2: A typical budget set.

x2

I
(0, p2
) b

− pp21
b

x1
( pI1 , 0)

Having that, the problem of a consumer is to maximize his utility subject to budget
constraint:

max u(x1 , x2 , . . . , xK ), (2.1)

(x1 ,x2 ,...,xK )∈X
K
X
st. xi pi ≤ I.
i=1
7 Set
A is convex, if for any two elements in a, a′ ∈ A, and a scalar t ∈ (0, 1), ta + (1 − t)a′ ∈ A.
PK
Now take any x, x′ ∈ B(p, I) and a scalar t ∈ (0, 1). Observe that ′
i=1 [txi + (1 − t)xi ]pi =
PK PK
t i=1 xi pi + (1 − t) i=1 x′i pi ≤ tI + (1 − t)I = I. Hence tx + (1 − t)x′ ∈ B(p, I).
2 CONSUMER THEORY 17

We now analyze solutions to such problem.

First, one needs conditions guaranteeing that such a solution exists. For that, it
is enough that u is continuous and prices positive. Additionally if preferences are
strictly convex (i.e. utility function is strictly quasi-concave8 ) the solution is unique.
Second, observe that solution to the utility maximization problem does not depend on
particular u representing preferences. Third, if we multiply the income and prices by
the same constant the budget constraint and hence solution remains unaffected. Hence
we may also normalize prices such that p1 = 1. And finally, for strongly monotone
preferences a bundle that costs strictly less then income is never optimal, hence we
PK
can alternatively express the inequality constraint as equality: i=1 xi pi = I.
Formally the solution to problem (2.1) is denoted by x∗ (p, I). If x∗ (p, I) is unique
we call x∗ a (Marshallian) demand function. The maximal utility, that can be ob-
tained, is denoted by v(p, I) = u(x∗ (p, I)) and called a value or indirect utility
function. The indirect utility function can be used to construct direct money-metric
utilities9 .
If the utility function is differentiable, than any optimal, interior bundle must
satisfy10 :
M Ui M Uj
(∀i, j) = . (2.2)
pi pj
This says that at the optimal, interior bundle the marginal rate of substitution
must be equal to the ratio of prices for any two goods. The condition is intuitive as
it requires that the rate of which consumer is willing to exchange any two goods is
equal to the rate at which market does so.
This condition is also sufficient if the second-order conditions are satisfied. Can
you state and interpret them? The condition (2.2) is necessary for interior solutions
but not for corner solutions. By corner solution we mean optimizing bundles with
xj = 0 for at least one j. To find the global solution (interior or corner) one have
to compare utility levels at the interior solution using (2.2) and compare it to the
8 To see that observe the following. Let V = {x ∈ X : x  y} be the preferred set. Then by

convexity of preferences we must have that for any t ∈ (0, 1) and any x, x′ ∈ V we have tx+(1−t)x′ ∈
V . As a result convex preferences have convex sets V for any y. Now for a utility u representing 
we must have V = {x ∈ X : u(x) ≥ u(y)}. By definition, for any quasi-concave functions its upper
contour set V is convex.
9 By this we mean a utility defined over the consumption set of pairs: price and income. Such

function can be interpreted as a utility of ’consuming‘ pairs p, I directly. P

10 Consider a corresponding Lagrange function L(x, λ) = u(x) + λ(I −
i xi pi ). In our case
the Slater constraints qualification is satisfied and solution to maxx L(x, λ∗ ) equals the solution of
problem (2.1). Alternatively a pair (λ∗ , x∗ ) solves minλ∈R+ maxx∈X L(x, λ). Calculating derivatives
∂u
with respect to xi and equating them to zero (for interior solution) ∂x (x∗ ) − λ∗ pi = 0 we obtain the
i
desired equality. The envelope theorem assures that under certain conditions the value of Lagrange
multiplier λ∗ = ∂v
∂I
(p, I) and can be interpreted as the marginal utility of income. More on Lagrange
or Kuhn-Tucker methods for convex, constrained optimization problems can be found in Rockafellar
(1997) or Clarke (1983), Rockafellar (1981) for nonsmooth case.
18 2.2 Choice

Figure 2.3: Consumer choice. Unique solution (left panels) and multiple solutions
(right panels).
x2 x2

b x∗
x∗
b

x∗
x1 x1
x2 x2

x∗

x1 x1

utility levels for all possible corner solutions. Figure 2.3 presents example of a unique
interior, and corner solutions, as well as multiple solutions to consumer maximization
problem.
Remark 2.1 The utility maximization problem is dual to the following expenditure
minimization problem:
X
e(p, v) = min x i pi ,
x∈X
i
st. u(x) ≥ v,
that is a problem of minimizing the total cost of obtaining the utility level v. The
solution to this problem is denoted by h∗ (p, v) and called the Hicksian demand (func-
tion). Clearly e(p, u∗ ) = I, where u∗ = v(p, I); also h∗ (p, v(p, I)) = x∗ (p, I) and
h∗ (p, v) = x∗ (p, e(p, v)).
Example 2.7 (Derive demand for Cobb-Douglas utility) We consider a util-
β
ity function from two goods u(x1 , x2 ) = xα
1 x2 as before. Let income I > 0 and prices
p1 , p2 > 0 by given. Then the optimal bundle (x∗1 , x∗2 ) satisfies pp21 = M RS1,2 = M
M U1
U2 =
αx∗
2
βx∗ as well as the budget constraint x∗1 p1 +x∗2 p2 = I. Solving this system of two equali-
1
α I β I
ties we obtain that the optimal bundle is given by (x∗1 , x∗2 ) = ( α+β p1 , α+β p2 ). Observe
that demand increases with income and decreases with price of a good.
2 CONSUMER THEORY 19

Example 2.8 (Derive demand for CES utility) Consider a CES utility function
1
u(x1 , x2 ) = [xρ1 + xρ2 ] ρ , where ρ is a parameter. Then the optimal bundle (x∗1 , x∗2 )
ρxρ−1
satisfies pp21 = M RS1,2 = M
M U1 ∗ ∗
U2 = ρxρ−1 as well as the budget constraint x1 p1 + x2 p2 =
1

2
∗
I. Solving this system of two equalities we obtain that the optimal x1 is given by
1
p1ρ−1 I
ρ ρ and similarly for x∗2 .
p1ρ−1 +p2ρ−1

Example 2.9 (Labor supply) Consider a household choosing a consumption level

c and a number of hours worked l. The time is normalized to one and the hours spend
not working (leisure) are denote by h = 1 − l. Consumer has wealth T ≥ 0, and
the hourly wage is w. If the price of a consumption good p is normalized to 1, then
the corresponding budget constraint is given by: c ≤ wl + T . Consider preferences
over consumption and leisure given by u(c, 1 − l). The optimal (consumption, leisure)
u′ (wl∗ +T,1−l∗ )
basket satisfies u′2 (wl∗ +T,1−l∗ ) = w and says that the marginal rate of substitution
1
between leisure and consumption must be equal to the ratio of prices w1 . Our analysis
derives the demand function for consumption and leisure but also individual supply of
labor hours.

Example 2.10 (Borrowing/lending) Consider a household choosing a consump-

tion level in the first period c1 and the second period c2 . Let preferences be given
by u(c1 ) + βu(c2 ), where β ∈ (0, 1) is a discount factor. The consumer earns w1
in the first period and w2 in the second. Consumer can spend income w1 on con-
sumption c1 or savings s (that could be positive or negative). Savings give interest
r the next period. The budget constraints for the consumer are: p1 c1 + s ≤ w1 and
1 1
p2 c2 ≤ w2 + s(1 + r). Rearranging: p1 c1 + 1+r p2 c2 ≤ w1 + 1+r w2 which states that
the present value of consumption stream cannot exceed the present value of income.
u′ (c∗ )
The interior, optimal consumption basket satisfies: βu′ (c1∗ ) = 1p1p . Interpreting the
2 1+r 2

marginal rate of substitution between the periods must be equal to the ratio of prices
between the periods. Observe that our analysis derives demand for consumption in
both periods but also demand/supply of borrowing/lending.

The demand function derived from the consumer maximization problem can have
various properties as a function of income and price. We can now use them to present
a bacis categorization of goods.

• If demand is increasing as a function of income, we say that good is normal.

• If demand is decreasing as a function of income, we say that good is inferior.

• If income elasticity11 of demand exceeds one, than the good is called luxury.
f ′ (x)x
11 For a differentiable function f : R → R we define its elasticity at point x by εfx := f (x)
. A
20 2.2 Choice

A typical example of a normal good is a case of unitary income elasticity of demand.

This holds for homothetic preferences. In such a case the consumer always spends
the same fraction of income on a given good. Similarly we can consider a demand as
a function of price. Typically demand should be a decreasing function of price but in
the case of a Giffen (e.g. the case of demand for potato in Ireland during the Great
Famine) or Veblen (snob effect) good, where demand is increasing with price, is also
possible. For some new developments in comparative statics in consumer theory see
Quah (2003), Antoniadou (2006) and Mirman and Ruble (2008) along with references
contained within.
It happens that the demand change, as a result of a price change, can be decom-
posed into two effects: substitution and income. If the price decreases, the former
effect means that the consumer can now substitute his consumption towards relatively
cheaper goods. The latter one means that as the price of some goods has decreased
the consumer can purchase more as he has relatively more money. Formally speaking,
consider the demand for good xj , then the following Slutsky equation summarize such
decomposition:

∂xj (p, I) ∂hj (p, v(p, I)) ∂xj (p, I)

△xj ≈ △pi = △pi − xi (p, I)△pi .
∂pi ∂pi ∂I

Its derivation is omitted. The first term in the Slutsky equation is a substitution
effect (changes to the Hicksian demand keeping utility constant), while the latter is
the income effect. The following figure 2.4 illustrates this construction.
Another way to measure, how the changes in prices influence demand and utility in
monetary terms, are compensating and equivalent variation. The compensating
variation uses the new prices as a base, and answers what income change in current
prices can compensate (i.e. gives the same utility) for the price change. Similarly
equivalent variation uses the old prices as a base, and answers what income change
in current prices is equivalent (i.e. gives the same utility) to the price change.
The analysis we did so far assumed that we know prices and consumer preferences
or its representation by a utility function. As mentioned in the introduction to this
chapter the analysis should start, however, with the observed choices. The revealed
preference analysis allows us to construct the preferences that rationalize observed
△d
price elasticity of demand function d is denoted by εdp = d
△p , where △d = d2 − d1 and similarly,
p
′
d (p)p
△p = p2 − p1 . For differentiable demand function d, εdp = d(p) . Hence price elasticity of demand
measures % change in demand for a one % change in price. Similarly one introduces a concept of
△s
price elasticity of supply εsp = s
△p , where s is a supply function. Other useful elasticities can also
p
be defined including: income elasticity of demand, measuring relative changes in demand to relative
changes in income; or cross-price elasticity of demand, measuring relative changes in demand for
good 1 to relative changes of price of good 2.
2 CONSUMER THEORY 21

Figure 2.4: Hicksian decomposition into substitution and income effects. Price de-
crease from p1 to p′1 and demand changes from x to x′ . Change b − x reflects substi-
tution effect (utility is the same for both baskets), while x′ − b reflects income effect
(parallel shift in the budget line).
x2

b x x′
b

b b

p1 p′1 x1

choices. The idea behind revealed preferences is simple. Suppose we observe con-
sumer’s choice of bundle x under prices p. Then for any other bundle x′ whose price
p · x′ ≤ p · x it must be that x′  x, as the consumer could afford bundle x′ but did
not choose it. In such case we say that x is directly weakly revealed preferred to x′ .
Similarly for any bundle x′′ whose price was p·x′′ < p·x we must have x′′ ≺ x, and we
say that x is directly strictly revealed preferred to x′′ . Similarly we consider indirect
revealed preferences via transitivity. Having enough observation, and assuming con-
sumer preferences do not change, we can construct the utility function that consumer
has revealed to us by its choices.
Formally, this can be summarized in the following axiom (GARP): if basket x is
(directly or indirectly) weakly revealed preferred to x′ , then x′ cannot be (directly or
indirectly) strictly revealed preferred to x. Having that we can state the celebrated
Afriat’s theorem.
Theorem 2.1 (Afriat (1967)) The set of choices (xt , pt )Tt=0 satisfies GARP if and
only if there exists a continuous, monotone, concave utility function, that rationalize
these choice (as outcomes from utility maximization).

2.3 Demand and consumer surplus

The demand derived for each individual consumer should be aggregated to obtain
the market demand for a particular good. This is done by summing n individual
Pn
demands {x∗i (·)}ni=1 and denoted by: D(p, I1 , . . . , In ) = i=1 x∗i (p, Ii ). Hence, es-
sentially the aggregate demand function inherits all the properties of the individual
demand functions. However, if the number of consumers is large and they are suf-
ficiently diversified, then aggregate demand function may be continuous even if the
22 REFERENCES

individual demands are not12 . Apart from continuity (and homogeneity) the aggre-
gate demand function will not possess other nice properties unless all the individual
demands do.
The interesting case is that, under certain conditions, the aggregate demand func-
tion looks “as if” it was derived from the individual, representative consumer. The
necessary and sufficient conditions for these (see Gorman, 1953) are that the indirect
utility function takes the form vi (p, Ii ) = ai (p) + b(p)Ii . Then the aggregate indirect
Pn Pn Pn
utility function is simply V (p, i=1 Ii ) = i=1 ai (p) + b(p) i=1 Ii .
The final topic considers consumer welfare. The classic tool to measure it is a
consumer’s surplus. If the demand for a given good is as a function of price is
Rp
given by D(p), then the consumer surplus is simply CS = p01 D(p)dp. It happens
that if the utility function is quasilinear the CS is the exact measure of consumer
welfare. In such a case the compensating and equivalent variation are equal to the
change in consumers surplus.
We finish this chapter with references to Deaton and Muellbauer (1980) and
Deaton (1992) with serious extensions and applications of the basic models treated
here.

References
Afriat, S. (1967): “The construction of a utility function from demand data,” In-
ternational Economic Review, 8, 67–77.

Aliprantis, C. D., D. J. Brown, and O. Burkinshaw (1990): Existence and

optimality of competitive equilibria. Springer-Verlag, Berlin.

Antoniadou, E. (2006): “Comparative statics for the consumer problem,” Economic

Theory, Exposita Note.

Clarke, F. H. (1983): Optimization and nonsmooth analysis. John Wiley and Sons,
New York.

Deaton, A. (1992): Understanding consumption. Clarendon lectures in economics.

Clarendon Press, Oxford.

Deaton, A., and J. Muellbauer (1980): Economics and consumer behavior. Cam-
bridge University Press, New York.

Debreu, G. (1964): “Continuity properties of Paretian utility,” International Eco-

nomic Review, 5, 285–293.

Gorman, W. M. (1953): “Community preference fields,” Econometrica, 21, 63–80.

12 We refer the reader to the book of Trockel (1984) for formal argument.
REFERENCES 23

Mirman, L. J., and R. Ruble (2008): “Lattice theory and the consumer’s problem,”
Mathematics of Operations Research, 33(2), 301–314.

Quah, J. K. H. (2003): “Market demand and comparative statics when goods are
normal,” Journal of Mathematical Economics, 39(3-4), 317–333.

Rockafellar, T. R. (1981): The theory of subgradients and its applications to

problems of optimization. Convex and nonconvex functions, vol. 1 of Research and
education in mathematics. Heldermann Verlag, Berlin.

(1997): Convex analysis. Princeton University Press.

Stokey, N., R. Lucas, and E. Prescott (1989): Recursive methods in economic

dynamics. Harward University Press.

Trockel, W. (1984): Market demand. An analysis of large economies with non-

convex preferences, vol. 223 of Lecture Notes in Economics and Mathematical Sys-
tems. Springer, Heidelberg.
Chapter 3

Producer theory

In this chapter we analyze producer’s behavior. This is important to understand (i)

demand for input factors (like capital or labor), (ii) firm’s cost functions, as well as
(iii) technological progress. We start by introducing a standard way of describing
technology using production function, discuss important aspects of technology and
then define firm’s cost and study a cost minimization problem. All of that will be
necessary to analyze firm’s profit maximization in the following chapters.

3.1 Technology and output

Technology transforms inputs (or production factors) into outputs. Examples of
inputs include: labor services, physical capital, land, raw materials or intermediate
goods, etc. Inputs are goods, so again we shall speak about specific inputs, differenti-
ated not only with respect to physical characteristics, but also location, time or state
of the world. Production usually takes time, but if one differentiates goods according
to time, our description would be general enough to incorporate timing implicitly.
Typically output and inputs are multidimensional: to produce a given good one usu-
ally needs many inputs and similarly usually one produces many outputs at the same
time (e.g. energy and pollution). At this level we will focus on the examples with a
single output and two inputs. Technology is typically summarized using a produc-
tion function that transforms inputs into a maximal level of output that can be
produced from a given combination of inputs. An example of a production function
is given by f with values:
q = f (k, l),

where k stands for capital, l for labor and q for output. The set of inputs and outputs
that are possible to produce using technology f is called a production possibilities

25
26 3.1 Technology and output

set and given by: Y = {(q, −k, −l, ) : q ≤ f (k, l), k ≥, l ≥ 0}. The minus signs are
supposed to capture that k, l are (net) inputs while q is an (net) output. The term net
refers to a situation if some factor is used to produce a.o. itself (e.g. power plant uses
usually a.o. electric energy to produce electric energy). Technology can change over
time due to e.g. technological progress. For the time being we assume that technology
is constant but later we will shortly discuss technological progress.
We now introduce some important concepts of average and marginal productiv-
ity. Average productivity of, say labor, is simply APl = f (k,l) l , while marginal
△f (k,l) f (k,l+δ)−f (k,l) f (k,l+δ)−f (k,l)
productivity M Pl = △l = (l+δ)−l = δ and measure produc-
tivity of additional δ units of labor. Average productivity is hence global (i.e. depends
only on the amount of inputs used), while marginal is a local measure (depends on
the amount, as well as the change δ in the use of inputs). Again, for differentiable
production function M Pl = ∂f ∂l .
Typical assumptions on the production function or production possibilities set
include:

• monotonicity: f is a nondecreasing function of inputs,

• convexity: f is quasiconcave1 or equivalently, Y is convex,

• regularity: V (q) = {(k, l) : q ≤ f (k, l)} is closed and nonempty for all q.

Figure 3.1: Convex (left panel) and nonconvex technology set (right panel).

y′ f (k) y′ f (k)
b b

b b

ty + (1 − t)y ′ ty + (1 − t)y ′
b b b

k y k FC y

Monotonicity implies that more inputs allow to produce more outputs. This is
true, if free disposal of inputs is possible. The regularity condition is technical and
innocent in most applications. The strongest assumption concerns convexity. It means
that, if any two combinations of production are possible, then the mixed combination
is also possible. It is restrictive as it e.g. rules out technologies with fixed costs and
0 = f (0, 0) feasible not mentioning concave production sets. This is illustrated in
1 For convex X, we say a function f : X → R is quasiconcave iff (∀x , x ∈ X) and ∀α ∈ [0, 1] we
1 2
have f (αx1 + (1 − α)x2 ) ≥ min{f (x1 ), f (x2 )}.
3 PRODUCER THEORY 27

figure 3.1 for the example of single input concave production function f without and
with fixed costs (F C). We will sometimes dispense with convexity assumption in our
analysis.
The set of factors that allows to produce exactly q is called an isoquant. Isoquant
is supposed to show that level q is possible to be produced by different combinations
of inputs, e.g. substituting capital for more labor or vice versa. The measure of
such substitability is called marginal rate of technical substitution and denoted
dl
by M RT Sk,l = − dk |q=const. . Interpreting, M RT Sk,l says: by how much labor input
should be increased if capital is to be decreased by a unit to keep the level of production
constant. If isoquant curve is differentiable the MRTS is simply the tangent to the
isoquant curve at a given point, and hence a local approximation of trade-off (or
substitution) between inputs on the isoquant. It happens that, if the production
function is differentiable, then2 M RT Sk,l = M Pk
M Pl .
The MRTS measure the slope of the isoquant, while the elasticity of substitu-
tion measures the curvature of the isoquant. Specifically, elasticity of substitution
measures the percentage change in the factor ratio over percentage change of MRTS
along the isoquant:
△k/l
k/l
ES = △M RSTl,k
.
M RSTl,k

Typically ES is calculated using the so called logarithmic derivative: ES = d lnd|M

ln k/l
RT S| .
Finally we introduce an important (global) concept of returns to scale. We say
that technology exhibits increasing (resp. decreasing, constant) returns to scale if

(∀k > 0, l > 0)(∀A > 1) f (Ak, Al) > Af (k, l) (resp. <, =).

The increasing returns to scale assumption states that if we increase proportionally all
the inputs in the production process than we could produce more than a proportional
increase in output. That is, a characteristic of a production process for which a larger
production is more efficient. Few typical sources of increasing returns to scale include:

• spreading fixed costs. Clearly the higher inputs the larger output to spread the
fixed costs,

• indivisibilities in a production process,

• physical characteristics of a production process, e.g. the cube-square rule,

• risk sharing. Larger firm can spread small risks of a production more efficiently,
2 Recallthe definition of the marginal rate of substitution, presented in chapter 2.1. Observe that
the two definition are very much alike. In fact, this observation will be useful, when discussing the
theory of general equilibrium.
28 3.1 Technology and output

• using large scale technologies, i.e. although the single technology may posses
constant or decreasing returns to scale at certain production levels, if the scale
increases the company may switch to more efficient large scale technologies and
actual production data would suggest an increasing returns to scale.

• few other including: economies of scale in purchases (higher discounts for higher
orders) or marketing (higher hit ratio of marketing campaigns of larger firms).

Capital intensive technologies usually have increasing returns, while labor intensive
constant or decreasing returns. Clearly increasing returns to scale violate convexity
assumption. Can you verify that?
Replication argument would suggest that, in the worst case scenario the company
may e.g. double its production by building a second factory aside, so the constant or
increasing returns to scale assumption should be natural. Put differently, we shall not
observe decreasing returns to scale in reality. However, there are examples of decreas-
ing returns to scale technologies. The reason is simple, the returns to scale analysis
assumes that all factors can be multiplied proportionally, however, sometimes it is
not possible as some factors may be constrained or even fixed (e.g. natural resources
available in a give region, or specific job supply in an area, or even communication
possibilities, etc.). So typically, when we speak of decreasing returns one should think
of fixed (production) factors.
Some factors may be fixed in the short run but become flexible in the long run.
Specifically, we talk about a short run technology / production function, if one of the
factors (like capital) is fixed: q = f (k̄, l), where k̄ is some predefined constant.
Efficiency requires that for industries with increasing returns to scale we should
observe a small number of large firms, while with decreasing returns a large number
of small firms. If both small and large companies may coexists at the same market we
could imply that scale does not matter and hence industry exhibits constant returns.
The following examples discuss the introduced concepts. See also Douglas (1948)
for a seminal development on production functions.

Example 3.1 (Leontief production function) Consider a production function f

with q = f (k, l) = min{αk, βl}, where α, β are positive. This is an example of a
perfectly complementary inputs as the production is always limited by the minimum of
both (perhaps multiplied) inputs. The isoquant curve is L-shaped and not differentiable
at the point where αk = βl. The function exhibits constant returns to scale. The
elasticity of substitution is zero as there is no possibility for substitution.

Example 3.2 (Linear production function) Consider a linear production func-

tion f given by f (k, l) = αk + βl, where α, β > 0. This is an example of a perfectly
substituable inputs and the isoquant is linear. The M RT Sk,l = α β . The function
exhibits constant returns to scale and the elasticity of substitution is ∞.
3 PRODUCER THEORY 29

Example 3.3 (Cobb-Douglas production function) Let f (k, l) = k α lβ , with pa-

rameters α, β > 0. Then M Pk = αk α−1 lβ and M Pl = βk α lβ−1 . Parameter α mea-
α−1 β
sures the elasticity of production with respect to k, i.e. εfk = kMfPk = αkkkα lβ l = α.
αl
The M RT Sk,l = βk and the elasticity of substitution is 1. The function exhibits the
constant returns to scale if α + β = 1, while increasing (decreasing), if α + β > (<)1.
σ−1 σ−1 σ
Example 3.4 (CES production function) Let f (k, l) = [αk σ + βl σ ] σ−1 for
α, β, σ positive. Observe that for σ = 0 the production function becomes Leon-
tief, if σ = 1 then production function becomes Cobb-Douglas, while for σ = ∞
−1
αk σ k
production function becomes linear. The M RT Sk,l = −1 . Rearranging: l =
βl σ
αM RT Sl,k σ
( β ) and ln kl = σ(ln α
β + ln |M RT Sl,k |). Hence the elasticity of substitution
ln kl
ES = ln |M RT Sl,k | = σ.

Observe that in the examples above marginal productivity of labor (capital) is

increasing (or nondecreasing) with the amount of capital (labor) used. This means
that capital and labor are complementary in the production process, but this feature
could be generalized to many inputs example. Complementarities (or “strategic
fit”) are now considered to be a very important factor in shaping the production
process and understanding modern manufacturing (see Kremer (1993) and Milgrom
and Roberts (1990,1995)).
Finally, we speak about a technological progress, if a change in the technology /
production function allows to produce more from a given number of inputs. In such
a case the isoquant is shifted towards the origin of the graph with inputs as variables.
If the shift leaves MRTS unchanged we speak of a neutral technological progress.
But if the M RT Sk,l decrease after a technological change, we say that technological
progress is capital saving. Intuitively, if M RT Sk,l = M Pk
M Pl decrease it means that
M Pl increases faster than M Pk , hence labor becomes relatively more productive and
proportionally some amount of capital could be saved. Similarly, if M RT Sk,l increases
after a technological change, we say that technological progress is labor saving.

3.2 Costs
Before we analyze the production cost minimization problem we introduce some basic
costs concepts. The economic costs are equal to accounting costs plus the al-
ternative costs. The alternative costs are implicit and account for the opportunity
cost (of working time or alternative capital use etc.), that is the value of the next best
alternative that could be done, if the current activity is surpassed.
Economists also differentiate between sunk and nonsunk costs. Sunk costs are
nonavoidable while nonsunk are avoidable. Usually the sunk costs are irrelevant for
economic analysis as they are nonavoidable and hence will not influence one’s decision.
30 3.2 Costs

Consider now a problem of minimizing the costs of producing output q given

technology f . Assume that price of a unit of labor is w (e.g. hourly wage of unskilled
worker) while renting price of capital is r. The total (minimum) cost of producing q
is hence:

T C(q) = min wl + rk,

k,l≥0
s.t. f (k, l) ≥ q.

The solutions to this minimization problem are called input demand functions and
are denoted by l∗ (q, w, r), k ∗ (q, w, r). For quasiconcave and differentiable f , optimal
(and interior) solution to this problem is characterized by:

M Pk M Pl
= ,
r w
or equivalently: M RT Sk,l = wr , hence the rate at which the firm wants to exchange
inputs along the isoquant must be equal to the rate at which the market can exchange
both inputs3 . For corner solutions the condition may not hold. However, since there
are only two inputs considered in the optimization problem, there are only two corner
solutions, which need to be verified, i.e. (k = 0, l > 0) or (k > 0, l = 0). Hence, one
need to compare the costs of wl with rk such that q = f (0, l) = f (k, 0).
If the ratio of prices changes, typically firms’ change their input employment. We
can analyze this by observing how the optimal l∗ (q, w, r), k ∗ (q, w, r) vary with prices.
Consider the following example.

Example 3.5 (Cost function for a Cobb-Douglas technology) Let prices w, r

2
be given and consider f (k, l) = k .5 l.5 . Solving q = k .5 l.5 for k gives isoquant k = ql .
2
Putting that to the optimization problem T C(q) = minl≥0 wl + r ql , where the first
2
order condition for interior optimal l∗ is w = r (lq∗ )2 and hence l∗ (q, w, r) = q wr ,
p

that is decreasing in w, increasing in r, and linear in q. The total costs function is

√
hence T C(q) = 2q rw.
T C(q)
Having derived the cost function we define average costs as AC = q and
△T C(q) T C(q+δ)−T C(q) T C(q+δ)−T C(q)
marginal costs M C = =
△q =
(q+δ)−q ,
for small increase
δ
′
δ in production, or simply M C(q) = T C (q) for differentiable total costs function. If
some part of the total costs does not change with q (i.e. F C = T C(0)), we call this
a fixed costs and write T C(q) = F C + V C(q) or AC(q) = FqC + V C(q)q . Notice that
fixed costs can be either sunk or not. Can you think of some examples?
3 Recall
the condition for optimal consumption bundle in the consumer choice problem, discussed
in chapter 2.2. Again, the similarity of the two conditions will be important in the general equilibrium
analysis.
3 PRODUCER THEORY 31

Consider the total costs function T C(q; w, r) = wl∗ (q, w, r) + rk ∗ (q, w, r). What
is the change of the total costs if factor prices (e.g. w) change? Observing the above
formula suggest, that we shall account for the direct effect of a price change, as well
as for an indirect effect via w → l∗ (q, w, r) and w → r∗ (q, w, r). Shephard’s lemma
assures that for differentiable factor demand functions the indirect effects cancel out,
and:
∂T C
(q; w, r) = l∗ (q, w, r).
∂w
This follows from envelope theorem and imply that a rate of change of the total
cost function with respect to input price is equal to the corresponding input demand
function. See Shephard (1978) for a formal treatment.
Interestingly4 the AC is increasing if M C > AC and decreasing if AC > M C.
Similarly M C = AC at the minimal AC. We call level of q for which AC is min-
imized a minimal efficient scale. If AC is decreasing (increasing) we speak of
economies (diseconomies) of scale. Clearly increasing returns of scale are equiv-
alent to economies of scale, while decreasing returns are equivalent to diseconomies
of scale. See Stigler (1951) for a classic reference.
If all inputs are variable, we say that T C(q) is the long run total cost and
denote it by T CLR , while if some factor is fixed (usually capital k̄), then the T C(q) =
wl + k̄r is the short run cost curve, with q = f (k̄, l) and notation T CSR . Clearly
T CSR ≥ T CLR as the long-run cost function is an envelope of the short-run one.
Similarly we can define short- or long-run average/marginal costs.
Two important concepts related to economies of scale are economies of scope and
economies of experience (also dynamic economies of scale or learning-by-doing effect).
If we consider production process producing two outputs q1 , q2 and derive a cost
function T C(q1 , q2 ), we speak of economies of scope, if T C(q1 , q2 ) ≤ T C(q1 , 0) +
T C(0, q2 ), that is, it is cheaper to produce both outputs at the same time than
separately5 . It implies complementarities in production process and a typical example
of such a pair is energy and pollution.
Economies of experience capture dynamic learning effect. Specifically, if av-
P
erage cost AC(Qt , qt ) is nonincreasing with Qt = k<t qk , which measures the total
output produced till period t, we say that technology exhibits economies of experi-
ence, as it is cheaper to produce the next unit, then the previous one. Observe that
qt → AC(Qt , qt ) may be increasing or decreasing, hence economies of experience are
not related to economies of scale (see Arrow, 1962).
We refer the reader to the textbook by Fuss and McFadden (1980) with extensions
and application of the basic models covered in this chapter.
4 Trivially T C(q) qM C(q)−T C(q)
AC(q) = q
hence AC ′ (q) = q2
and the desired inequalities follow if
we setAC ′ (q) S 0.
5 See also Topkis (1995), who analyze comparative statics of a firm, for some examples.
32 REFERENCES

References
Arrow, K. (1962): “The economic implications of learning by doing,” Review of
Economic Studies, 29, 155–173.

Douglas, P. H. (1948): “Are there laws of production?,” American Economic Re-

view, 38, 1–41.

Fuss, M., and D. McFadden (1980): Production economics: a dual approach to

theory and application. North-Holland, Amsterdam.

Kremer, M. (1993): “The O-Ring theory of economic developent,” Quarterly Journal

of Economics, 108(3), 551–575.

Milgrom, P., and J. Roberts (1990): “The economics of modern manufacturing:

technology, strategy, and organization,” American Economic Review, 80(3), 511–
528.

(1995): “Complementarites and fit. Strategy, structure and organizational

change in manufacturing,” Journal of Accounting and Economics, 19, 179–208.

Shephard, R. W. (1978): Theory of cost and production functions. Princeton Uni-

versity Press, Princeton.

Stigler, G. J. (1951): “The division of labor is limited by the extent of the market,”
Journal of Political Economy, 59, 185–193.

Topkis, D. M. (1995): “Comparative statics of the firm,” Journal of Economic The-

ory, 67, 370–401.
Chapter 4

Perfect competition

4.1 Company at the perfectly competitive market

Here we start with the analysis of a firm’s behavior at the perfectly competitive
market. Firm is taking a market price p as given and beyond its control (see Robinson
(1934) and Stigler (1957) for a seminal discussion of the basic assumptions underlying
perfect competition and its methodological development). The demand it faces is
hence given by: 
 0 if p > p,
D(p) = any amount if p = p,
∞ if p < p.


The firm can set any price it wants but if the price is higher that the market one p̄ it
will sell nothing, while if the price is lower than p̄ firm has an infinite demand. Still
the profits would be higher, if the price is actually equal to p̄, as then a firm may still
have as many clients as it wants. The profit maximization problem is hence simple
and requires to find the appropriate level of production to maximize revenues minus
total costs at the market price, say p:

max pq − T C(q),
q≥0

where function T C was derived in the previous chapter. Assuming T C is differentiable

and the second order condition T C ′′ (q ∗ ) ≥ 0 is satisfied, we have the necessary and
sufficient condition for interior, profit maximization quantity level q ∗ is:

p = T C ′ (q ∗ ) = M C(q ∗ ).

Observe that in some cases, like with fixed costs, the interior solution may not be
optimal. To see, when a corner solution is chosen, consider T C(q) = V C(q) + F C

33
34 4.2 Competitive equilibrium and welfare

with F C > 0 that are sunk. Then condition ’price equal to marginal costs’ is still
necessary, but not sufficient unless π ∗ = pq ∗ − V C(q ∗ ) − F C ≥ −F C. This says
that profit must exceed profit of producing zero (−F C). And that yields condition:
∗
p ≥ V C(q
q∗
)
= AV C(q ∗ ). The price level at which p = AV Cmin is called a shutdown
level as the company would be better off closed. For the case, when the fixed costs are
nonsunk the condition becomes π ∗ = pq ∗ − V C(q ∗ ) − F C ≥ 0 or p ≥ AT C. That is,
as the nonsunk costs are avoidable, then firm would continue production until profits
are positive.
Some costs are sunk in the short run, but in the long run most costs become
nonsunk. Hence p ≥ AV C(q ∗ ) is a condition for firm’s positive operations in the
short run, while p ≥ AT C(q ∗ ) in the long run. The intermediate cases, where some
fixed costs are sunk, while others not, can also be considered.
Observe that condition p = M C(q ∗ ) implicitly defines a supply function S(p) by
p = M C(S(p)) with T C ′′ (S(p)) ≥ 0 as long as (short or log run) operating conditions
are satisfied. Hence the inverse of marginal costs defines a (short or long run) firm’s
supply curve.
In the analysis so far we have assumed that T C function was derived from company
costs minimization problem. Still, a direct analysis is possible. We study it now.
Consider a profit maximization problem of a firm with production function f taking
prices p of output and inputs w, r as given:

max pf (k, l) − rk − wl.

k,l≥0

If the production function is strictly concave and differentiable then necessary and
sufficient conditions for interior profit maximization inputs (k ∗ , l∗ ) are f1′ (k ∗ , l∗ ) =
M Pk = pr and f2′ (k ∗ , l∗ ) = M Pl = wp . It says that the marginal productivity of
each factor is equal to its real price. Also, observe, that this condition determines
both: cost minimizing inputs k ∗ , l∗ (as it imply MrPk = MwPl ), as well as the optimal
production level q ∗ = f (k ∗ , l∗ ) given price p.

Remark 4.1 We have shown that, at the perfectly competitive market, price equals
marginal cost. One may argue that it is because of a price taking assumption, which
is justified with a large number of competitors. In chapter 8, however, we discuss an
example where this result still holds for m = 2 firms that are price setters.

4.2 Competitive equilibrium and welfare

In the previous section we have derived a supply curve S of a firm at the perfectly
competitive market. If we sum supplies of all m firms we obtain a market or ag-
Pm
gregate supply: j=1 Sj (p). Similarly in the previous chapter we obtained an
4 PERFECT COMPETITION 35

Pn
aggregate (or market) demand i=1 Di (p) of some n consumers. A pair (q ∗ , p∗ ) is
called a competitive (partial) equilibrium if:
m
X n
X
q∗ ∈ Sj (p∗ ) and q ∗ ∈ Di (p∗ ).
j=1 i=1

In such a case there is a price p for which quantitiy q ∗ is supplied and demanded. If
∗

demand and supply are functions this condition simply says:

m
X n
X
Sj (p∗ ) = q ∗ = Di (p∗ ),
j=1 i=1

which equates supply to demand. This is the condition that determines a market
price. Observe that in equilibrium all m firms may have positive or zero profits.
For a given demand and supply a natural question is, whether there exists a
price such that market clears. We postpone discussing answers to this questions till
chapter 9. Here we only mention that indeed the market equilibrium may or may not
exists. Moreover, if it exists its uniqueness is not guaranteed, even if demand and
supply are functions (see figure 1.1).
The long run perfectly competitive analysis endogenize m. That is: in the long
run perfectly competitive equilibrium m∗ is such that
∗
m
X n
X
Sj (p∗ ) = Di (p∗ ),
j=1 i=1

and
(∀j) AC(Sj (p∗ )) = M C(Sj (p∗ )) = p∗ .
Interpreting: in the long run competitive equilibrium the profits of all companies
are zero. This must hold since, as if they were positive, some new companies could
appear, enter the market and reduce them.
We now move to welfare analysis of competitive equilibrium. For this reason
assume there is a representative consumer with quasilinear, differentiable utility func-
tion u(x) + y, where y is simply money left for purchase of other goods, and it is
the market for good x that is under study. We already know that demand for x is
given implicitly by u′ (D(p)) = p, where p is a price of x. Assume also that a total
cost function of producing x is given by T C with T C ′′ (·) > 0 and T C(0) = 0. Then
supply of x is given implicitly by T C ′ (S(p)) = p. In the competitive equilibrium we
hence have:
u′ (D(p∗ )) = p∗ = M C(S(p∗ )).
Now consider a welfare maximization problem of:
max u(x) + y,
x,y≥0
s.t. y = ey − T C(x),
36 REFERENCES

where ey is an initial endowment in good y. This problem maximizes utility of a repre-

sentative consumer under the feasibility constraint, and yields a first order condition
for interior solution:
u′ (x∗ ) = M C(x∗ ),
that is also sufficient under our assumptions. Hence the competitive equilibrium
allocation maximizes total welfare as the price p∗ equates values of the two functions.
Rx
To see it from a different perspective, observe T C(x) − T C(0) = 0 M C(q)dq.
Similarly observe that u(x) is simply an area under the inverse demand function.
Recall that consumer surplus is CS = u(x) − px and producer’s surplus is P S =
px − T C(x). As a result the competitive equilibrium allocation also maximizes a
total surplus: P S + CS, i.e.

max P S + CS = max u(x) − px + px − T C(x).

x≥0 x≥0

These properties of a competitive equilibrium may not hold, however, when prices
are influenced by some taxes / subsidies or quotas. Finally we will discuss competitive
equilibrium in a general equilibrium setting in chapter 9.

References
Robinson, J. (1934): “What is perfect competition?,” Quarterly Journal of Eco-
nomics, 49, 104–120.

Stigler, G. J. (1957): “Perfect competition, historically contemplated,” Journal of

Political Economy, 65, 1–17.
Chapter 5

Monopoly and pricing

A monopoly is a firm with market power, i.e. it observes impact of own quantity
change on the market price. Hence we say that monopoly is a price maker. Observe
that in order to determine, whether a company is a monopolist or not, one needs an
appropriate definition / boundaries of a market. Monopolists face two constraints:
technological summarized by a cost function T C, and market summarized by demand
D. The problem of monopolists is to choose price and quantity to maximize profits
under these two constraints:

max pq − T C(q),
p,q≥0
s.t. q ≤ D(p).

In many cases it requires that D(p) = q (when it is not?) and hence the problem
reduces to:

max pD(p) − T C(D(p)).

p

Also, if D is invertable (with inverse demand function P ) we can alternatively express

the monopolist problem as:

max P (q)q − T C(q).

q≥0

For differentiable P and T C functions, the first order conditions for interior solution
imply:
P ′ (q ∗ )q ∗ + P (q ∗ ) = T C ′ (q ∗ ),
which reads that marginal revenue is equal to marginal costs. Inspecting this
equation we see than the monopolist increasing output balances cost change effect
(M C) with increased revenue of selling more at the current price (P ) and (typically)

37
38

decrease in revenue, as to sell more he needs to decrease the price of every unit offered
(P ′ (q)q). The second order condition is:

2P ′ (q ∗ ) + q ∗ P ′′ (q ∗ ) − T C ′′ (q ∗ ) ≤ 0.

This first order condition can be rearranged to give:

P ′ (q)
P (q)[1 + q] = T C ′ (q),
P (q)
or
P (q) − T C ′ (q) P ′ (q) 1
=− q = − D.
P (q) P (q) ǫp
The right hand side term is a (minus) inverse of price elasticity of demand and gives
the simple rule for monopoly pricing. The left hand side is sometimes referred to as
the Lerner index, measuring the % of price that finances markup (difference between
price and marginal costs). Also, it follows that at the optimal level of quantity the
price elasticity of demand must be greater than one in absolute values.
To challenge your thinking try to analyze the following case. Suppose marginal
costs of production increase as a result of technological change, or market regulations.
When is it a case that the monopoly will increase its absolute markup? How about
its relative markup?
We now present few examples or deriving the optimal monopoly price:

Example 5.1 (Linear demand and costs) Let T C(q) = cq and assume linear de-
mand with inverse P (q) = a − bq. Then the optimal production satisfies −bq ∗ + a −
bq ∗ = c or q ∗ = a−c ∗ a+c
2b with P (q ) = 2 .

Example 5.2 (Constant elasticity of demand) Let T C(q) = cq and assume con-
stant price elasticity of demand function D(p) = Ap−b . In such case the monopoly
pricing rule gives: p∗ = 1−c 1 .
b

Example 5.3 (Monopsony) The analysis of the monopolistic behavior is not re-
stricted to the case of a company selling a good to customers. Similarly one can
analyze behavior of a company that is a single buyer of e.g. labor services at some
market. To see that formally let f be a production function from (a single input)
labor and w be an inverse labor supply curve. A single firm buying labor services on
that market that sees its impact on the wage offered is called a monopsony. Formally
a monopsony problem is to choose a labor input such that

max pf (l) − w(l)l,

l≥0

where we assume that the firm is price taker on the consumption good market. Simi-
larly as before, assuming differentiability of the objective, the first order condition for
5 MONOPOLY AND PRICING 39

∗
interior choice of l∗ is pf ′ (l∗ ) = w(l∗ ) + w′ (l∗ )l∗ or rewriting w(l w(l
)−pM PL
∗) = − ǫ1L ,
w
where ǫLw is a wage elasticity of labor supply. Observe that mixed cases are also possi-
ble allowing for company that has market power at both the input and output markets
for example.
Typically P ′ (q) < 0 and together with increasing marginal costs, a monopoly pric-
ing rule implies that monopolist would produce less than a perfectly competitive firm.
It also implies higher prices set by monopolist. This indicates that monopolistic solu-
tion is not socially efficient. To see that clearly, consider a single consumer economy
with quasilinear utility u(x)+y giving an inverse demand function p(x) = u′ (x). There
is also a monopolist with differentiable cost function T C. The social objective is to
maximize W (x) := u(x)−T C(x) which gives the first order condition u′ (x∗ ) = p(x∗ ) =
M C(x∗ ). On the other hand the monopolist chooses p(xm ) + p′ (xm )xm = M C(xm )
and hence W ′ (xm ) = u′ (xm ) − M C(xm ) = −p′ (xm )xm = −u′′ (xm )xm > 0 if u′′ is
negative. The inefficiency of a monopoly is sometimes called a deadweight loss.
The next question we consider is: why (generally inefficient) organizations as
monopolists are present in the market. The starting point concerns the so called
natural monopolies. That is the case, where a single firm is more efficient (has
lower average costs for all production levels at which inverse demand is higher than
average costs), than two or more firms operating separately. The typical example is
a company with decreasing average costs function that is characterized by economies
of scale for all relevant output levels (i.e. output levels such that inverse demand is
higher than average costs).
More generally there are barriers to entry that restrict new companies to enter
the market and to challenge the monopolist. Following a classification introduced by
Bain we consider structural and strategic barriers to entry. Structural barriers to
the entry exist, when incumbent firms have cost or demand advantages that would
make it unattractive for a new firm to enter the industry. One reason for that could
be some form of technological effects, like natural monopoly, but it may also include
some legal regulations making it very costly to enter. Alternatively strategic barriers
to entry include incumbent firm taking explicit steps to deter the entry. The examples
include limit pricing (decreasing own price to signal low markups and deter entry),
increasing production to move down the experience curve (learning-by-doing effect),
deliberately increasing customers’ switching costs or developing networking effects
via lowering costs of within network consumption. Although inefficient a monopoly
makes a higher profit, than a perfectly competitive firm, hence monopolistic firms
often engage in rent-seeking activities to protect their market power and increase
profit (fortunately not always at the costs of efficiency) (see Stigler (1971) for an early
introduction to the economics of regulation or Pepall, Richards, and Norman (2008)
for some more recent developments).
We have stated before that monopolistic production choice is socially inefficient.
40

One can think that this results from a market power assumption. However, even with
a market power the efficient solution is possible. This can be achieved or approximated
by various price discrimination techniques. This again suggests that inefficiency
may result also from an implicit assumption that monopolist claims only a single (and
linear) price.
The first degree (or perfect) price discrimination means that a seller charges
different prices for every unit of the good sold, such that the price for every unit
equals the maximal willingness to pay. The second degree price discrimination
(or nonlinear pricing) means that the seller asks a different price for each unit sold
but do not differentiate prices between customers. The examples include quantity
discounts. The third degree price discrimination means that different prices are
offered for separate groups (segments) of clients but each segment gets a linear price.
The examples include students/senior discounts but also charging different prices for
business and economy class airplane tickets. The distinction between all three price
discriminations is neither exhaustive nor mutually exclusive and moreover real life
examples are using many of them simultaneously.
Now in a series of examples (borrowed from Pepall, Richards, and Norman (2008))
we illustrate the role of various types of pricing strategies. In examples 5.4-5.6 we
consider an economy with two groups (segments) of clients with quasi-linear utilities
(young and old, denoted by y, o respectively), with inverse demand for some product
in each group given by po = 16 − qo , py = 12 − qy . The marginal costs are constant
and equal to average costs c = 4. In the example 5.4 we consider a benchmarking
case of perfect competition and (linear) monopolistic price.

Example 5.4 (Perfectly competitive and monopolistic prices) The total de-
mand for a product is given by q := qo + qy = 16 − p + 12 − p giving the inverse
total demand function: p = 14 − 2q . A perfectly competitive price is pP C = c = 4 and
perfectly competitive quantity q P C = 20 with zero profit.
Conversly, at the monopolistic market the optimality condition requires equating
marginal revenue with marginal costs giving: 14 − 2q − 2q = M R = c = 4 resulting in
pM = 9, q M = 10 with profits π M = 50. For further reference observe that 10 = q M =
qoM + qyM = 7 + 3.

In the next example we allow the monopolist to discriminate its price between two
groups of consumers and hence consider a case of the 3rd degree price discrimination.

Example 5.5 (Observable characteristics) Consider a monopolist choosing two

separate prices for each group of clients. We hence analyze the optimality conditions
equating costs and marginal revenue at each market separately. That is 4 = c =
M Ro = 16 − qo − qo and 4 = c = M Ry = 12 − qy − qy giving qo = 6 and qy = 4
with prices po = 10 and py = 8. The total profit of the monopolist is hence equal
5 MONOPOLY AND PRICING 41

to 36 + 16 = 52. The profits are higher than under the monopolistic solution in the
example 5.4, although the total output produced is the same.

As the next example illustrates the 3rd degree price discrimination proposed in
the example 5.5 is not socially optimal. That is, there exist a better tariff combining
the second and third degree price discrimination, simultaneously implementing the
perfect price discrimination solution.

Example 5.6 (Optimal two part tariff ) Consider the following two part tariff for
each group of clients. The unit price for each group is the same and equal marginal
costs py = po = 4 = c but on top of that the monopolist asks to pay a constant/fixed
fee t (not changing with the quantity consumed) to be allowed to consume this good.
Let to = 72 and ty = 32. Observe that under such tariff the old consumers consume
qo = 12 units but their surplus from such consumption is zero, i.e. to captures the
whole surplus from consumption of 12 units each at price 4. Similarly the young
consume qy = 8 and ty is chosen so that the surplus of the young is zero. The total
output produced in q = qy + qo = 20 and gives profit 32 + 72 + (4 − 4) × 20 = 104, which
is the highest possible in this examples. Hence two-part tariff can implement the first
degree price discrimination. Observe that in such a case the amount (denoted by M )
the consumer spends to buy a given number of goods q is nonlinear in q. Specifically:

0 if q = 0,
M (q) =
qp + t if q > 0.

In the next example we again consider the case with two customer groups, but now
(as opposed to age) the characteristics differentiating both groups are unobservable
(e.g. think of price discrimination based on income). So let ph = 16 − qh and pl =
12 − ql , where h, l stands for high and low income respectively.

Example 5.7 (Unobservable characteristics) Suppose we want to introduce an

optimal two part tariff from the example 5.6. Observe now that, if the company in-
troduces two separate tariffs, but cannot easily differentiate its consumers, these are
the customers that will choose a tariff that fits them best. Specifically, consider the
tariffs from the example 5.6: pl = 4 and tl = 32 and ph = 4 and th = 72. Now, low
income customers will choose the l tariff giving them zero surplus, but the high income
consumers instead of consuming the h tariff (giving them zero surplus) are better off
choosing tariff l, consuming 8 and giving the positive surplus of 64 − 32 = 32. In such
a case we say that tariff h does not satisfy the incentive compatibility constraints,
as the h clients can misreport their income, pretend they are from l group and be better
off. The optimal tariff for h, constrained by incentive compatibility, yields: ph = 4
and th = 40. In such a case h clients are indifferent between l, h tariffs and can choose
to take h. This gives (constrained) maximal profit of 40 + 32 = 72.
42

Similar consideration can be taken into account, when ones tries to discriminate
prices in time, i.e. set high prices in the beginning of sale of some good, and
decrease it in time to capture the consumers that can wait longer. Of course the
incentive compatibility must be taken into account (see Bulow, 1982).
There are other types of pricing strategies including: block pricing or bundling.
Block pricing is an example of the second degree price discrimination in which the
firm offers quantity discounts. An example with two blocks looks like this:

qp1 if q ≤ q̄,
M (q) =
q̄p1 + (q − q̄)p2 if q ≥ q̄,

with p1 > p2 . Bundling, or more generally tying, refers to situations with multiple
goods. The next example illustrates this. More on bundling can be found in (Adams
and Yellen, 1976).

Example 5.8 (Bundling) Suppose we have 4 groups of customers (each of equal

size 1) each with willingness to pay for goods A and C presented in the next table. Let
the cost of production be equal to 30.

Goods A C
1 20 100
2 40 80
3 80 40
4 100 20

If a company chooses prices of both goods separately, then the profit maximization
yields: pA = 80 and pC = 80 giving profits 2 × 80 + 2 × 80 − 4 × 30 = 200, as
only two groups of customers would buy each product. Now suppose that the company
sells bundles of goods A and C, with a price of such bundle pB = 120. Then each
group buys both products giving profit 4 × 120 − 8 × 30 = 240. Finally consider the
mixed bundling, in which there is a price of the bundle pB = 120 and one can also
buy each product at pA = pC = 99. Observe that in such a case consumers from
group 2 and 3 buy a bundle, while consumers from group 1 and 4 are better off buying
only one product and enjoying surplus of 1. In such a case the company’s profit is
2 × 99 + 2 × 120 − 6 × 30 = 258. Can you generalize intuition on, when bundling
increases profit of a company, and when mixed bundling increase it even further?

Tying is a more general form of bundling, and refers to the case, when the con-
sumption of one product requires consumption of the other (e.g. camera and appro-
priate lenses that fit).
REFERENCES 43

References
Adams, W. J., and J. L. Yellen (1976): “Commodity bundling and the burden of
monopoly,” Quarterly Journal of Economics, 90(3), 475–98.

Bulow, J. I. (1982): “Durable-goods monopolists,” Journal of Political Economy,

90(2), 314–32.

Pepall, L., D. Richards, and G. Norman (2008): Industrial organisation. Con-

temporary theory and empirical applications. Blackwell Publishing.

Stigler, G. J. (1971): “The theory of economic regulation,” Bell Journal of Eco-

nomics and Management Science, 2, 3–18.
Chapter 6

Risk and ambiguity

In this chapter we consider decisions under uncertainty. More specifically we will

analyze decisions over risky or ambiguous outcomes. As this may be useful to analyze
both decisions of a consumer or a firm we will generally talk on some decision makers.
Gilboa (2009) textbook is a great reference for topics discussed in this chapter.

6.1 Expected utility

In this section we consider decisions under uncertainty over the finite set of outcomes
X. By a lottery we mean1 a probability distribution on X. If X has cardinality n
we can think of f ∈ X as of vector of nonnegative numbers f1 , f2 , . . . , fn such that
Pn 2
i=1 fi = 1. By L denote the set of all lotteries on X. We now define a preference 
on L that we assume is rational (i.e. complete and transitive) and continuous. Here we
just mention that rationality assumption is the more restrictive the more complicated
the domain of choice is. Hence rationality assumption is sometimes questioned at
this level. However the critical assumption on preference  concerns the so called
independence. Specifically for all f, g, h ∈ L and all α ∈ [0, 1] we have

f  g iff αf + (1 − α)h  αg + (1 − α)h.

To understand the independence observe that αf + (1 − α)h can be seen as a mixed

lottery (with probabilities α) over f and h. Hence it is naturally to think that αf +
1 Alternatively, we can define a lottery as a function ℓ mapping states s ∈ S to outcomes x ∈ X,

i.e. ℓ : S → X. Assuming, that there exists some probability distribution p : S → [0, 1], each
outcome ℓ(s) occurs with probability p(s). Therefore, both representations are equivalent. However,
representing the lottery as a function might be useful, when we are willing to compare two different
lotteries, defined over the same state space. For example, lottery ℓ′ , such that ∀s ∈ S, ℓ′ (s) > ℓ(s)
has always higher outcomes, regardless of the state.
2 Therefore, consider X as a set of all vectors, which coordinates sum up to 1.

45
46 6.1 Expected utility

(1−α)h  αg +(1−α)h whenever f  g, as a result of consequentialism. That is, it

is either f or h that will happen/be consumed (never both), so the compound lottery
h should be irrelevant for preference between f and g. Note, that this in particular
means, that f  g if and only if αf + (1 − α)g  g, and f  αf + (1 − α)g. Therefore,
the axiom implies that the preferences relation is preserved for any linear combination
of the lotteries in question. Having that we can state the expected utility theorem.

Theorem 6.1 (von Neumann-Morgenstern) Suppose  is rational, continuous

preference relation satisfying independence. Then there exists numbers ui (unique up
to affine transformation) for each element of X such that
n
X n
X
(f1 , f2 , . . . , fn ) = f  g = (g1 , g2 , . . . , gn ) iff fi ui ≥ g i ui .
i=1 i=1

The von Neumann and Morgenstern (1944) theorem shows that the utility func-
tion we can use to evaluate (represent) lotteries take a very simple form of a linear
function, weighting utilities of outcomes ui with probabilities of these outcomes fi .
More generally, if we consider some random variable taking values x ∈ X with prob-
ability p(x), then the expected utility of choosing such a random variable is simply
P R
i u(xi )pi or X u(x)p(x)dx, where we set u(xi ) := ui . Observe that although the
expected utility is linear in probabilities it is not necessarily linear in outcomes, i.e.
x → u(x) need not be linear. That is to say that expected utility is not necessarily an
expected value of a random variable taking values in X. This observation is critical
to measure risk. Moreover although the utility u is defied on the set of outcomes X,
that in principle could be very general, and correspond to a consumption set, we often
focus on X as representing the wealth levels and u as an indirect utility function. As
we will mention later one needs to be cautious, though, when utilities are not defined
over monetary payoffs.

Example 6.1 (Demand for insurance) Consider a consumer with initial wealth
w facing a risk of loosing l with probability p. His expected utility from such a lottery
is pu(w − l) + (1 − p)u(w). There is also an insurance company selling policies to
cover a loss of q at price (premium) πq. To analyze the demand for such insurance
consider a maximization problem maxq≥0 pu(w − l + q − πq) + (1 − p)u(w − πq). The
fact that price −πq appears in both outcomes means that the premium must be paid
in advance of resolution of uncertainty, however, cover q is only present in case of a
loss. The first order condition for optimal cover q ∗ is then
pu′ (w − l + q ∗ − πq ∗ ) π
= ,
(1 − p)u′ (w − πq ∗ ) 1−π
equating marginal rate of substitution between states with the relative price of an
additional unit of a cover.
6 RISK AND AMBIGUITY 47

Now suppose also that an insurance company operates at the competitive market
such that its economic profit is zero: 0 = πq − pq. This implies that for given p
insurance premium π = p, which is sometimes called a fair insurance price.
If indeed π = p, then the consumer’s first order condition reduces to u′ (w − l +
q − πq ∗ ) = u′ (w − πq ∗ ), yielding q ∗ = l whenever u′ is strictly monotone. Hence
∗

under a fair insurance price a full coverage of a loss is optimal. The results of this
example must be analyzed with care, as we will show later that neither the zero profit
condition on an insurance market need not be satisfied nor the assumption that the
probabilities of events are independent of decision maker actions.

6.2 Evaluating risk

Consider a decision maker that prefers a sure event of obtaining an expected value
of some random variable taking values in R to an expected utility of choosing such
random variable i.e. X X
u( x i pi ) > u(xi )pi .
i i

We call such person risk averse. If the reverse holds, we say he is a risk lover. If
both expressions are equal we say she is a risk neutral. In the following discussion
we focus on risk aversion. Comparing both sides of the expression above suggest
that the appropriate risk measure should account for concavity of u. As this will be
demonstrated later generally, mean and variance of a random variable are not enough
to describe decision maker behavior nor his risk attitude. Indeed as showed by Arrow
and Pratt the appropriate notion of (absolute) risk aversion measure (for a twice
differentiable) utility function is

u′′ (x)
r(x) = − .
u′ (x)

By saying appropriate we mean that decision maker i would take more (small) gambles
than j, if and only if she has a higher (Arrow-Pratt) risk aversion measure. By small
we mean that the Arrow-Pratt measure is a local measure and may change with
respect to a ’wealth’ level x. To understand the concept of risk aversion consider a
notion of certainty equivalence of some random variable:
X
u(CE) = u(xi )pi .
i

That is, CE is a level of ’wealth’ that gives the same utility as expected utility of
a lottery. To see usefulness of this notion consider the decision maker faced with
a lottery. The difference between the expected value of a lottery and its certainty
equivalence RP = EV − CE measures how much (in maximum) a decision maker is
48 6.2 Evaluating risk

Figure 6.1: Utility over monetary outcomes for a risk averse consumer.

u(x)
b

b
u(EV )
b b
EU
b

x1 CE EV x2 x

willing to pay to sell risk associated with a lottery. This is called a risk premium.
All in all, Pratt’s theorem establishes that a decision maker with higher risk aversion
measure has ’more concave’ utility function or equivalently is willing to pay more
to sell a risk of a lottery allowing him to win or loose some amount with equal
probabilities. Figure 6.1 presents the example of a utility over monetary outcomes for
a risk averse consumer, and a lottery to get x1 or x2 with some probabilities.
A related concept of measuring risk concern the so called relative measure of
risk aversion defined by:
xu′′ (x)
ρ(x) = − ′ ,
u (x)
where word relative refers to a multiplication by x. See also Kihlstrom and Mirman
(1974) for a formal analysis of risk over multidimensional outcomes.
1−σ
Example 6.2 (CRRA) Consider a utility function u(x) = x1−σ . Its absolute risk
aversion is r(x) = σx and relative risk aversion is ρ(x) = σ, hence it is called a
constant relative risk aversion (CRRA) utility function and σ measures relative risk
aversion.

Example 6.3 (CARA) Consider a utility function u(x) = −e−σx . Its absolute risk
aversion is r(x) = σ and relative risk aversion is ρ(x) = xσ, hence it is called a
constant absolute risk aversion (CARA) utility function and σ measures absolute risk
aversion.

Example 6.4 (Mean-variance utility) We argued that the mean and variance are
not sufficient measures to capture the choice of a decision maker under uncertainty.
This is true, however, there exist a special class of utility function, where these two
measures are sufficient. To see that consider a utility function u(x) = −e−σx . Suppose
that outcomes x ∈ X are distributed according to density function p. Then the expected
2
σx
utility is equal Eu(x) = − X e−σx p(x)dx = −e−σ(x̄−σ 2 ) , where x̄ is the mean and
R

σx2 is a variance.
6 RISK AND AMBIGUITY 49

6.3 Subjective probability and state dependent util-

ity
Von Neuman-Morgerstern approach was criticized by the assumption that probabil-
ities are exogenously given and objective, i.e. all decision makers know some ’real’
probabilities of states they face. The next example highlights some of this argument.

Example 6.5 (’Allais paradox’) You are asked to choose between two lotteries:

A: 100% chance of receiving 1m.

B: 10% chance of receiving 5m, 89% of receiving 1m and 1% of receiving nothing.

Write down your choice. Then consider two other alternatives:

C: 11% chance of receiving 1m and 89% of receiving nothing,

D: 10% chance of receiving 5m, 90% of receiving nothing.

Again please write down your choice. Allais observed that many people prefer A over
B but D over C. Such choice violates expected utility theory. To see that observe that
utility function representing such choices must satisfy:

u(1m) ≥ .1u(5m) + 0.89u(1m) + .01u(0).

Rearranging: .11u(1m) ≥ .1u(5m) + .01u(0) and adding .89u(0) to each side:

.11u(1m) + .89u(0) ≥ .1u(5m) + 0.9u(0),

implying that C  D by an expected utility maximizer.

One of the explanations of Allais paradox indicated that although some objective prob-
abilities are given people have their own probability weights that are different from
objective ones. For example a very unlikely state but still with positive probability
may be considered by many as having (subjective) probability of zero. Moreover, in
many situations true or objective probabilities are unknown. As subjective probabil-
ities are unobservable, it is hard to measure them. The question is then, whether we
can construct both utility functions and subjective probabilities from some observed
choice data.
That was noticed first by Ramsey and de Finetti in the 1930/1931 that in or-
der to construct both objects of interest from the observed choice data over (simple)
lotteries the decision maker must be risk neutral. Later in their insightful theorem
Anscombe and Aumann (1963) showed that it is indeed possible to construct subjec-
tive probabilities of states by revealed preference argument, if one observes decisions
50 6.3 Subjective probability and state dependent utility

over a larger domain of acts (where acts, specify an objective (lottery over a set of
outcomes) for every state). A ’crowning glory’ of a decision theory was developed by
Savage (1954), however, who showed that if one observes choices over acts (specifying
outcomes for every state), by a revealed preference argument both subjective proba-
bilities and utility function can be constructed. This is summarized in the celebrated
subjective probability theorem.

Theorem 6.2 (Savage) Consider a set of states S and finite outcomes set X. De-
fine a space of acts F = {f : S → X} and consider a preference order  on F .
Consider the following axioms:

1.  is complete, transitive, continuous3 and there exists f, g ∈ F such that f ≻ g,

2. the preference between f, g ∈ F should depend only on the values of f and g in

states that they differ,

3. preferences are independent of states,

4. subjective probabilities of states do not depend on outcomes.

Preference  satisfy the above axioms if and only if there exists a nonatomic probability
measure µ on S and a linear function u : X → R such that for every f, g ∈ F we have
Z Z
f  g iff u(f (s))dµ(s) ≥ u(g(s))dµ(s).
S S

This generalized the Von-Neumann Morgerstern analysis to a great extent and

allowed to rationalize much broader class of choices. However, not all decisions can
be again explained as the next example suggests.

Example 6.6 (Ellsberg ’paradox’) There are two urns: K,U with 100 balls each
(either white or black). The K urn contains 49 white and 51 black balls. The U urn
has unspecified amount of balls. Your are to drawn one ball from either K or U. There
are two choice situations

A: If you choose a black ball you get 1m and zero otherwise.

B: If you choose a white ball you get 1m and zero otherwise.

Ellsberg observed that many people in choice situation A prefer to choose from urn K,
and similarly in the choice situation B. But this violates subjective probability theory.
To see that, observe that to rationalize the first choice we must have:

.51u(1m) + .49u(0) ≥ pb u(1m) + (1 − pb )u(0),

3 Savage did not assume any topology on S but still assumed an axiom that played a similar role

as continuity.
REFERENCES 51

and hence pw = 1 − pb ≥ 0.49 assuming u(1m) ≥ u(0) ≥ 0. This immediately gives

that (pw − .49)u(1m) + u(0) ≥ (pw − .49)u(0) + u(0) implying

pw u(1m) + (1 − pw )u(0) ≥ .49u(1m) + (1 − .49)u(0).

And hence, in the choice situation B the subjective utility maximizer should choose to
drawn from U.

A typical explanation of Ellsberg ’paradox’ is to say that the decision maker prefers
a ’safe’ choice is a sense that, the probability distribution is known to him. This,
however, suggests that the Knight’s distinction between the risk (where the proba-
bility distribution is known) and uncertainty/ambiguity (where the probability dis-
tribution is unknown) may have some insight for decision theory. See also Gilboa
and Schmeidler (1989) and Klibanoff, Marinacci, and Mukerji (2005) for some recent
developments.
Finally an interesting generalization was proposed by Karni, Schmeidler, and Vind
(1983) concerning the so called state dependent utilities. They propose axiomati-
zation of such preferences and representation by a subjective probability and utility
function u(f (s), s), that depends on both outcomes and states directly. The unique-
ness of such representation may by problematic, however. State dependent utilities
are useful in considering the choice under uncertainty over acts, where Savage third
assumption is not satisfied, typically for non-monetary payoffs. A typical example
supporting using such preferences is that e.g. an umbrella may give different utility in
different states of nature. Although this example can be modeled using the Savage’s
model if the set of outcomes and states is rich enough, e.g. an umbrella is a different
good in every state, the Savage’s theory has problems in situations, where a decision
maker may change its preference relations and hence decisions over outcomes in dif-
ferent states of nature (e.g. serious illness of a relative). Then states dependent utility
assumption is more plausible. Finally, recently Karni (2011) provided derivation of
preferences over ’strategies’ (specifying actions and bets for every signal/information)
with action dependent probabilities.

References
Anscombe, F., and R. Aumann (1963): “A definition of subjective probability,”
The Annals of Mathematical Statistics, 34(1), 199–205.

Gilboa, I. (2009): Theory of decision under uncertainty. Cambridge University Press.

Gilboa, I., and D. Schmeidler (1989): “Maxmin expected utility with non-unique
prior,” Journal of Mathematical Economics, 18(2), 141–153.
52 REFERENCES

Karni, E. (2011): “A theory of Bayesian decision making with action-dependent

subjective probabilities,” Economic Theory, 48(1), 125–146.

Karni, E., D. Schmeidler, and K. Vind (1983): “On state eependent preferences
and subjective probabilities,” Econometrica, 51(4), 1021–31.

Kihlstrom, R. E., and L. J. Mirman (1974): “Risk aversion with many commodi-
ties,” Journal of Economic Theory, 8(3), 361–388.

Klibanoff, P., M. Marinacci, and S. Mukerji (2005): “A smooth model of

decision making under ambiguity,” Econometrica, 73(6), 1849–1892.

Savage, L. J. (1954): The foundations of statistics. Wiley, New York.

von Neumann, J., and O. Morgenstern (1944): Theory of games and economic
behavior. Princeton University Press.
Chapter 7

Game theory

Game theory is a branch of mathematics devoted to a study of strategic decision

making. It has found many applications in economics, political or social sciences
and biology, among others. In this chapter we introduce some basic concepts of
games, strategies and equilibria. They are interesting on their own but we use them
extensively in the next chapter 8 devoted to the analysis of oligopolies and industrial
organization. We start with strategic form games, that can be interpreted, as if
decisions of all players were taken simultaneously. Then in section 7.2 we consider
extensive form games useful for the analysis of dynamic or sequential games. At this
level we skip a branch of game theory devoted to the analysis of cooperative games,
indicating however, their usefulness in the analysis of general equilibrium concept.
More on introduction to game theory can be found in Dixit, Reiley, and Skeath
(2009). See also Osbourne and Rubinstein (1994), Myerson (1991) or Fudenberg and
Tirole (2002) for a more formal treatment.

7.1 Strategic form games

Formally a strategic form game Γ is a triple (N, (Ai , ui )i∈N ) such that N is a set
of players (with a slight abuse of notation we will denote cardinality of N by N as
well), Ai is a set of actions1 available to player i and ui : ×i∈N Ai → R is a payoff of
player i, when players choose action profile (a1 , a2 , . . . , aN ) ∈ ×i∈N Ai . We sometimes
write (ai , a−i ) to denote an action profile such that player i uses ai ∈ Ai and other
(other then i) players choose a−i ∈ ×j6=i Aj . Generally player’s payoff may depend
on the actions taken by all players. In the strategic form game actions are taken
1 Insuch simple, strategic form games we use actions / strategies terms equivalently. In extensive
form games or Bayesian games, when some information is revealed during the course of the game,
actions and strategies denote different concepts.

53
54 7.1 Strategic form games

’simultaneously’, i.e. when choosing its own action player does not know the other
players’ actions. To describe a game we adopt a matrix notation, as in the following
two players example.

Table 7.1: A dominant strategy game.

L R
U 1,2 2,1
D 3,1 5,0

This matrix describes a game between two players: 1 (row) and 2 (column), hence
N = {1, 2}. Actions available to player 1 are A1 = {U, D} and to player 2 are
A2 = {L, R}. When player 1 chooses D and player 2 chooses R the payoff of 1 is 5 and
payoff of 2 is 0, as summarized by the entry (5,0) in the row D and column R. Formally
u1 (U, L) = 1, u2 (U, L) = 2, u1 (U, R) = 2, u2 (U, R) = 1, u1 (D, L) = 3, u2 (D, L) = 1
and u1 (D, R) = 5, u2 (D, R) = 0.
We now proceed to the analysis of players’ behavior in the game. It is assumed
that each player knows the game he/she plays (i.e. the matrix) and aims to maximize
its own payoff. The unknown is the other player action as the game is a simultaneous
move one.
Observe that the game depicted in table 7.1 is relatively simple to analyze, i.e.
observe that player 1 has a strictly dominant strategy, i.e. whatever player two
chooses he is (strictly) better off by choosing D, than by choosing any other strategy.
Similarly observe that player 2 has also a strictly dominant strategy L. Hence we can
conclude that in this game players choose profiles (D,L) giving them payoffs (3,1).

Table 7.2: Battle of sexes (or Bach-Stravinsky game).

B S
B 3,2 0,0
S 0,0 2,3

In fact, dominant strategies are not typical in applied games. The next example 7.2
illustrates this. Indeed in game 7.2 it is better for player 1 to choose B, when 2 plays
B, while it is better to choose S, when 2 plays S. For this reason we need some other
’solution concept’ to analyze such games.
Definition 7.1 (Nash equilibrium) A pure strategy Nash equilibrium of the game
Γ is an action profile (a∗1 , a∗2 , . . . , a∗N ) ∈ ×i∈N Ai such that (∀i ∈ N ):
(∀ai ∈ Ai ) ui (a∗i , a∗−i ) ≥ ui (ai , a∗−i ).
7 GAME THEORY 55

Interpreting a profile a∗ is a pure strategy Nash equilibrium, if no player can strictly

increase its payoff by an unilateral deviation from a∗ .
In game 7.2 husband and wife wants to buy tickets for Bach or Stravinsky music
concert. Observe that such a game has two pure strategy Nash equilibria (B,B) and
(S,S). In any case unilateral deviations are not profitable as they result in zero payoff.
Hence, our first lesson is that a pure strategy Nash equilibrium does not need to be
unique. The second lesson says that, in some cases it may be non-existent (see game
in table 7.3).

Table 7.3: Matching pennies.

H T
H 1,-1 -1,1
T -1,1 1,-1

The game reflects a situation in which each player puts a penny on a table (choos-
ing heads or tails) so that the other does not see it. Then they simultaneously reveal
their actions. Indeed for payoffs in table 7.3 there is no pure strategy Nash equilib-
rium.
Let us now consider a coordination game depicted in table 7.4. There are two pure
strategy Nash equilibria (1,2) and (11,22) with respective payoffs (2,2) and (4,4). Our
third lesson says hence, that some Nash equilibrium payoffs may be preferred by
all players to the others. The name coordination reflects the fact that players must
coordinate to the one of two equilibria. Also observe that, when player 2 chooses 2,
the net benefit from the action increase by player 1 (that is action change from 1 to 11)
gives 0 − 2 = −2. The same move, when player 2 chooses 22, gives 4 − 1 = 3. Clearly
3 > −2 and hence in this game the higher the strategy of the opponent the higher
incentive to increase own strategy. In such a case, following Bulow, Geanakoplos,
Klemperer, we say that the game exhibits strategic complementarities. If the
reverse holds we say a game exhibits strategic substitutes2 .

Table 7.4: Coordination game.

2 22
1 2,2 1,0
11 0,1 4,4

2 It
happens that games of strategic complementarities (or supermodular games) are quite common
in economics (see Topkis, 1998). For a class of games with strategic substitutes we refer the reader
to a paper by Dubey, Haimanko, and Zapechelnyuk (2006).
56 7.1 Strategic form games

Finally we consider a celebrated prisoners’ dilemma game depicted in figure 7.5.

Table 7.5: Prisoners’ dilemma.

C D
C -5,-5 -1,-10
D -10,-1 -2,-2

Observe that in Prisoners’ dilemma game we have a unique pure strategy Nash
equilibrium (C,C) with equilibrium payoffs (-5,-5). Our forth lesson on Nash equi-
librium indicates that equilibrium payoff does not necessarily give players ’optimal’
payoffs. Specifically we mean that payoffs are not Pareto optimal3 , as there exists a
different action profile (D,D) such that both players are better off. Moreover, observe
that this Nash equilibrium is in strictly dominant strategies. Specifically, observer
that (D,D), although giving Pareto-optimal payoff profile, is not a Nash equilibrium
as every player has an incentive to deviate. This is in contrast to the coordination
game in figure 7.4.
The name prisoners’ dilemma corresponds to the following situation. Two men are
arrested, but the police do not possess enough information for a conviction. Men are
separated and each one of them is offered a proposal: if one testifies against the other
(confess), and the other remains silent (does not confess), the confessor gets 1 month
sentence and the other receives the 10 months sentence. If both remain silent, both
are sentenced to 2 months in jail for a minor charge. If both confess, each receives a
5 month sentence. Each prisoner must choose to confess or not and decisions must be
taken simultaneously.
Finally we introduce a useful concept of a best response. Formally we denote a
best response of player i by BRi : ×j6=i Aj → 2Ai , using the following maximization
problem:
BRi (a−i ) = arg max ui (ai , a−i ). (7.1)
ai ∈Ai

That is, the best response returns a set of all actions that are maximizing payer i pay-
off, when others use a−i . Observe that BRi is not necessarily a function, as for some
a−i there could be multiple maximizers. Also it may happen that for some a−i there
is not best response at all. We will see such cases, when analyzing Bertrand competi-
tion game in chapter 8. Importantly, there is relation between Nash equilibrium and
the joint best response correspondence BR, defined by

BR(a) = (BR1 (a−1 ), BR2 (a−2 ), . . . , BRN (a−N )),

3 See definition 9.2 in chapter 9 for a formal statement.
7 GAME THEORY 57

mapping any joint strategy profile, to a vector of best responses of all players. Indeed
a strategy profile a∗ is a Nash equilibrium if and only if a∗ ∈ BR(a∗ ), i.e. is a fixed
point of BR correspondence. We will use this observation in the applied games in the
following chapters.
We finish with some important extension. In the proceedings we allowed the
players to choose pure strategies, i.e. elements from Ai . Assume that Ai has a finite
number of elements ki . Suppose now we allow the players to choose mixed strategies,
i.e. probability distribution over elements of Ai . By △(Ai ) denote the set of all mixed
P ki
strategies, i.e. △(Ai ) = {σi : Ai → [0, 1] : j=1 σi (aj ) = 1}. We denote a typical
mixed strategy of player i by σi and denote an expected payoff of player i under the
strategy profile (σ1 , σ2 , . . . , σN ) as:

X N
Y
Ui (σi , σ−i ) = u(a1 , a2 , . . . , aN ) σj (aj ).
a∈×N
s=1 As
j=1

Definition 7.2 (Mixed strategy equilibrium) A mixed strategy Nash equilibrium

∗
of the game Γ is an action profile (σ1∗ , σ2∗ , . . . , σN ) ∈ ×i∈N △(Ai ) such that (∀i ∈ N ):

(∀σi ∈ △(Ai )) Ui (σi∗ , σ−i

∗ ∗
) ≥ Ui (σi , σ−i ).

Observe that a pure strategy Nash equlibrium is a mixed strategy Nash equilibrium
for degenerate lotteries.
Observe that the matching pennies (see table 7.3) have a unique mixed strategy
Nash equilibrium (σ1 , σ2 ) with σi (H) = 21 = σi (T ) for both players. Also the battle
of sexes has a mixed strategy Nash equilibrium (σ1 , σ2 ), that is not pure. This is a
case, when σ1 (S) = 25 = σ2 (B) and σ1 (B) = 35 = σ2 (S).

7.2 Extensive form games

In this section we consider extensive form games. At this level we do not introduce
a formal definition of a game and strategy, but start by considering the following
example depicted in figure 7.1. In this game there are two players N = {E, I}, firm
contemplating entry to a particular market, and an incumbent firm. The game has
two stages. In the first the entrant chooses to enter (In) or not (Out). If the firm
stays out the game finishes with payoffs (uE , uI ) =(0,2). If the firm enters, then at the
second stage, the incumbent decides to accommodate entry (A) or fight (F) with an
entrant. In each case the game ends and gives payoffs of (-1,-3) and (2,1) respectively.

This game differs from strategic form game as, among others, decisions are not
taken simultaneously but sequentially. Specifically, before choosing its strategy the
incumbent observes the move of the entrant. In such case the incumbent can condition
58 7.2 Extensive form games

Figure 7.1: Entry game.

Entrant
b

Out In

b b Incumbent
(0,2)

Fight Accomodate

b b

(-1,-3) (2,1)

its decision on the observed decision of the opponent. The second stage of the game in
which incumbent takes her decision is called a subgame (see part of the decision tree
in the dashed line rectangle). Hence the entrant has two strategies (In, Out) equivalent
to its decisions, but the incumbent has two strategies conditioned on observed moves:
(F if entry occurs, A if entry occurs).
Formally speaking a (pure) strategy is a function mapping player’s information
to his action set. There are two Nash equilibria of this game: (In, A if entry occurs)
and (Out, F if entry occurs). In both, players cannot increase their payoff by unilat-
eral deviation. Interestingly the second Nash equilibrium is somehow inconsistent as
it is based on the empty threat. Specifically the entrant stays out of the market
because he is afraid of getting -1, when the entry occurs and the incumbent fights.
Observe, however, that if the entry actually occurs the incumbent is better off by ac-
commodating. Hence, the Nash equilibrium (Out, F if entry occurs) is not a subgame
perfect Nash equilibrium, as the decision F, if the entry occurs, is not an equilibrium
of the subgame (which is a single player decision problem in our simple example).
More formally a subgame perfect Nash equilibrium (or SPNE for short) is the
Nash equilibrium, such that for all subgames, the part of a strategy profile restricted
to this subgame is a Nash equilibrium of this subgame. In finite games (i.e. games
with finite number of periods), there is a simple way to find SPNE, namely using
backward induction. Specifically consider the final subgame of our game and find the
Nash equilibrium profile. In the game 7.1 it is A. Then choose the Nash equilibrium of
the proceeding subgame, assuming that players know that in the following subgames
a Nash equilibrium profile will be played. One finds a SPNE repeating this procedure
to the initial node.
Extensive form games can also incorporate simultaneous moves. The example is
presented in figure 7.2. The dashed ellipse denotes an information set of incumbent
7 GAME THEORY 59

player, i.e. although incumbent observes if the entrant has entered or not, he does not
observe however, whether entrant has chosen small or large niche (two nodes in the
ellipse).

Figure 7.2: Niche choice game.

Entrant
b

Out In

b b Entrant
(0,2)

SN LN

b b
Incumbent
SN LN SN LN

b b b b

(-6,-6) (-1,1) (1,-1) (-3,-3)

The post entry subgame can be alternatively expressed using the matrix notation
in 7.6. Observe that a post entry subgame has two pure strategy Nash equilibria:
(SN,LN) and (LN,SN). In the former one entrant gets -1, while in the latter 1. As a
result we have two SPNE in the Niche choice game: (In, LN after entry, SN if firm E
enters) and (Out, SN after entry, LN if firm E enters).

Table 7.6: Post entry (sub)game.

SN LN
SN -6,-6 -1,1
LN 1,-1 -3,-3

To finish let us mention that a topic of strategic and extensive form games is
much broader than covered at this level. In includes equilibrium refinements as well
as analysis of auctions, Bayesian games, repeated strategic form games with celebrated
Folks’ theorem4 , or infinite horizon stochastic games.
4 Folks’ theorem states that if players are patient enough than any (convex combination of) indi-
vidually rational payoff vectors of a strategic form game can by supported by a Nash equilibrium of
infinitely repeated game.
60 REFERENCES

References
Dixit, A., D. Reiley, and S. Skeath (2009): Games of strategy. W.W. Norton &
Company.

Dubey, P., O. Haimanko, and A. Zapechelnyuk (2006): “Strategic complements

and substitutes, and potential games,” Games and Economic Behavior, 54(1), 77–
94.

Fudenberg, D., and J. Tirole (2002): Game theory. MIT Press, Cambridge.

Myerson, R. B. (1991): Game theory. Analysis of conflict. Harvard Universty Press,

London.

Osbourne, M. J., and A. Rubinstein (1994): A course in game theory. The MIT
Press, London.

Topkis, D. M. (1998): Supermodularity and complementarity, Frontiers of economic

research. Princeton University Press.
Chapter 8

Oligopoly and industrial

organization

In this chapter we consider oligopoly markets, i.e. markets with a small number of
competitors. The models of oligopolies describe a competition that is ’imperfect’, in
the sense that it may lead to an inefficient solution. One reason for such inefficient
outcomes comes from the fact that competitors are not price takers and hence see
an impact of own decisions (concerning prices or quantities produced e.g.) on other
competitors behavior. A fundamental tool of such an analysis is the game theory.
At the end of this chapter we also discuss a monopolistic competition model, that is
important for international trade, growth theory and industrial organization.
There are various other topics covered in the industrial organization literature
that we will not cover here. These include: formal analysis of horizontal and vertical
boundaries of firms, make-or-buy decisions, incomplete contracting, dynamic pricing
rivalry, mergers and acquisitions, entry and exit, research and development, and many
others. Here we refer an interested reader to some textbooks: Tirole (1988), Church
and Ware (2000), Besanko, Dranove, Shanley, and Schaefer (2007) or Pepall, Richards,
and Norman (2008).
Before proceeding we introduce two common measures of market concentration.
Suppose there are m firms in a market, each with a market share of si . Without
loss of generality suppose that firms names are ordered by its market share, with
Pn
i = 1 a market leader. Then CRn = i=1 si is the n-firm concentration ratio.
Typically one measures CR4 , that is a sum of market shares of 4 largest firms. The
second typical measure is a Herfindahl-Hirschman Index (HHI for short): HHI =
Pm 2
i=1 si , which is a sum of squared market shares of all companies on the market.
If there is a single company on the market HHI = 1. If there are m equal sized
1
companies on the market HHI = m . The closer the HHI or CRn to 1 the more

61
62 8.1 Cournot model

concentrated is the market.

We now proceed to present four canonical models of oligopolistic markets.

8.1 Cournot model

We start with a Cournot model of oligopolistic competition. Consider n = 2 firms
producing the homogeneous product each with constant marginal costs ci = c ≥ 0.
The inverse demand for the product is given by P (Q) = a − bQ, where a > c, b > 0.
Companies compete simultaneously choosing their production levels qi ∈ [0, ∞). The
total supply Q = q1 + q2 . The profit of a company choosing production level qi when
its competitor chooses q−i is given by:

πi (qi , q−i ) = (P (qi + q−i ) − c)qi = (a − b(qi + q−i ) − c)qi .

The Cournot model can be analyzed using a game theoretical language, with each firm
being a player with payoff function πi and strategy set Ai = [0, ∞). We now proceed to
describe the Nash equilibrium of the Cournot game. Recall that the Nash equilibrium
in this case is a pair of production levels q1∗ , q2∗ such that ∀i and ∀qi ∈ [0, ∞) we have:

πi (qi∗ , q−i
∗ ∗
) ≥ πi (qi , q−i ),

that is the production profile such that no company can strictly increase its profits by
unilateral (production) deviation. Observe that the first order conditions for interior
production level qiBR solve ∂π BR
∂qi (qi , q−i ) = 0. With our assumption about marginal
i

costs and demand we hence have:

a − b(qiBR + q−i ) − c − bqiBR = 0,

which yields qiBR (q−i ) = a−c−bq

2b
−i
. This is the best response production level of
company i to the production level of company −i. Observe that the best response
is a decreasing function of q−i , i.e. the more competing firm is producing the less
company i produce as a best response. In such a case we say that Cournot game
exhibits strategic substitutes, i.e. company strategic choices are substitutes. To
find the Nash equilibrium we solve a system of equations:
 BR ∗
q1 (q2 ) = q1∗ ,
q2BR (q1∗ ) = q2∗ .

Which under our assumptions gives a (unique) Nash equilibrium (q1∗ , q2∗ ) = ( a−c a−c
3b , 3b ).
The equilibrium total output
 level is 2 a−c a+2c
3b , while equilibrium price is 3 . The Nash
2 2
equilibrium profits are (a−c)
9b ,
(a−c)
9b . Figure 8.1 presents the two best response
curves and Nash equilibrium.
8 OLIGOPOLY AND INDUSTRIAL ORGANIZATION 63

Figure 8.1: Nash Equilibrium (NE) and Pareto optimal (PO) allocation in a Cournot
game.
q2
a−c
b
BR1 (q2 )

a−c
2b
a−c b
NE
3b b
a−c
4b PO BR2 (q1 )
q1

Interpreting, both companies produce the same level of output and none of them
wants to unilaterally deviate from this output level. Moreover Cournot original anal-
ysis stressed stability of the Nash equilibrium, i.e. an iterative process (qit , q−i
t ∞
)t=0 of
t+1 BR t
quantity level updating qi = qi (q−i ) is convergent to the Nash equilibrium.
Interestingly the Nash equilibrium allocation is not Pareto optimal. To see that
m m a−c a−c
 off producing (q1 , q2 ) = ( 4b , 4b ).
observe that both companies would be better
(a−c)2 (a−c)2 a−c a−c
Such allocation gives profits 8b , 8b . The optimal total output 2b = 4b +
a−c
4b can be found by solving for optimal monopoly production level maxQ (a−bQ−c)Q.
Observe, however, that (q1m , q2m ) is not a Nash equilibrium, i.e. even if companies
coordinate to choose such production levels (e.g. creating a cartel) both of them have
an incentive to deviate to qiBR ( a−c
4b ). It should be mentioned, however, that if the
Cournot game is repeated infinitely many periods and companies are patient enough,
the cooperative (cartel) solution can be supported as the Nash equilibrium of such
dynamic game using appropriate punishment strategies.
The Cournot model analyzed in this section is the example as we assumed m = 2,
homogeneous product, liner demand function P and constant (and equal across firms)
marginal costs c. Generalizations are possible and analysis described in this section
can be repeated similarly. However, one will not expect all of the properties of our
example to remain valid. Specifically, when the inverse demand function is sufficiently
convex, or when (differentiated) goods are complements then Cournot game may
exhibit strategic complementarities, i.e. higher production of a competitor may lead
a quantity increase in the best response of the other player. Moreover, the game
may have multiple Nash equilibria, some of them stable/unstable (see Amir, 1996).
Limiting properties of the Nash equilibrium allocation are also studied letting n → ∞.
Under some assumptions the Nash equilibrium price converge to marginal costs c, but
it is not generally true. Also, it is not generally true that any Nash equilibrium price
pn is decreasing as a function of number of companies (see Amir and Lambson, 2000).
64 8.2 Stackelberg model

Figure 8.2: A subgame perfect Nash equilibrum (SPNE) of a Stackelberg game.

ql
a−c
b
BRf (ql )

a−c b
2b SP N E

a−c qf
4b

8.2 Stackelberg model

Another model of interest is a model of quantity leadership or Stackelberg. So consider
again two firms producing a homogeneous product and deciding on own output qi ∈
[0, ∞). Both have constant marginal costs c but now one firm (market leader l)
makes its choice first. Then (in the second stage) the follower f observes the output
decision of the leader and then chooses its own output. Profits of each company
are given by πi (ql , qf ) = (a − b(ql + qf ) − c)qi . We now analyze subgame perfect
Nash equilibrium of the Stackelberg game. For this reason we first consider the
follower choice for an observed choice of a leader ql . Similarly as in the Cournot game
we obtain the best response qfBR (ql ) = a−c−bq 2b
l
. This is the optimal choice in the
second stage, so let us now proceed to the first stage. Knowing the best response
function of the follower, the leader decides on its optimal quantity choice solving:
maxql ≥0 πl (ql , qfBR (ql )) = (a − b(ql + qfBR (ql )) − c)ql . Observe that this objective
depends only on ql , i.e. best response of the leader is a single point. Solving we
3(a−c)
obtain ql∗ = a−c ∗ a−c
2b and qf = 4b . Total output is 4b and price a+3c
4 . Leader obtains
2 2
profits: (a−c)
8b and follower (a−c)
16b . Interpreting, the leader produces more than the
follower and has higher profits. Also leader (resp. follower) produces more (resp. less)
and has higher (resp. lower) profits than in the Nash equilibrium of the Cournot game.
The Stackelberg output (resp. price) is higher (resp. lower) than in the Cournot game.
See figure 8.2 for illustration or Amir and Grilo (1999) for more general cases.

8.3 Bertrand model

We now analyze Bertrand model of oligopolistic competition. Consider m = 2 firms
producing a differentiated but substitutable products, each with constant marginal
costs ci . Demand for each product is given by Di (pi , p−i ) = a − pi + γp−i , where
a > 0 and 1 > γ > 0 reflecting that products are substitutable. Firms compete
8 OLIGOPOLY AND INDUSTRIAL ORGANIZATION 65

simultaneously choosing their prices pi ∈ [0, ∞). When a pair of prices is chosen the
profit of each firm is

πi (pi , p−i ) = (pi − ci )(a − pi + γp−i ).

We now find the Nash equilibrium of the Bertrand game. For this reason consider
the first order conditions for optimal, interior price: ∂π BR
∂pi (pi , p−i ) = 0 which gives:
i

a+ci +γp−i
pBR
i (p−i ) = 2 . Observe that as γ > 0 Bertrand model exhibits strategic
complementarities: higher price of the rivals increase the best response price of a
firm. The Nash equilibrium solves the system of equations:
 BR ∗
p1 (p2 ) = p∗1 ,
pBR ∗ ∗
2 (p1 ) = p2 .

Which under our assumptions gives a (unique) Nash equilibrium with prices
(2 + γ)a + 2ci + γc−i
p∗i = .
4 − γ2
Observe that the equilibrium price is increasing in own and rival’s costs. Similarly to
a Cournot model, it can be shown that Nash equilibrium allocation is inefficient. The
Pareto optimal level of prices is higher than the Nash equilibrium one, though, itself
is not a Nash equilibrium.
Again here we have focused on one example. For a detailed analysis of the Bertrand
model we refer the reader to a book by Vives (2000). The extensions include non-linear
demands, non-constant marginal costs, complementary products, productions con-
straints or use of mixed strategies (see Maskin (1986), Dasgupta and Maskin (1986),
Dastidar (1995), Baye and Morgan (1999)). Also price leadership models (similar to
Stackelberg) are considered (see Amir and Stepanova, 2006). We finish this section
with an ’extreme’ example of Bertnard competition with homogeneous products.

Example 8.1 (Bertrand with homogeneous products) Similarly as above con-

sider an example of m = 2 Bertrand competitors with constant marginal costs but
homogeneous product with demand D(p). Assume D is continuous, strictly decreasing
whenever positive and there exists a price p̄ such that (∀p ≥ p̄) D(p) = 0. If firms
choose a price pair pi , p−i demand for firm’s i output is:

 0
 if pi > p−i ,
D(p)
di (pi , p−i ) = 2 if pi = p−i ,

 D(p) if p < p .
i −i

Firms profits are hence: πi (pi , p−i ) = (pi − c)di (pi , p−i ). We claim there is a unique
Nash equilibrium of this game with (p1 , p2 ) = (c, c). That is the unique Nash equi-
librium prices are equal to marginal costs and companies have zero profits. To see
66 8.4 Monopolistic competition

that (c, c) is indeed the Nash equilibrium observe that (∀p) πi (pi , c) ≤ 0. To see that
this Nash equilibrium is unique, observe that c > pi ≥ pj cannot be an equilibrium as
j-company is better off if it increases price to pj = c. Similarly pi ≥ pj > c is also not
a Nash equilibrium as the i-company can set price pj − ǫ and increase its profits. To
see that observe that limǫ→0 πi (p − ǫ, p) = limǫ→0 (p − ǫ − c)D(p − ǫ) = (p − c)D(p) >
(p − c) D(p)
2 = πi (p, p) and hence there exists an ǫ > 0 such that πi (p − ǫ, p) > πi (p, p).
So there is a price undercutting strategy to increase own profits. Interestingly, there
is no least ǫ that accomplishes such a price undercut and hence firm i best response is
not defined for pj > c.
The uniqueness of Nash equilibrium with prices equal to marginal costs is known
as the Bertrand paradox, as there are only two companies in equilibrium and they set
prices equal to marginal costs. This example suggests that Bertrand or price undercut-
ting competition is severe, much more than Cournot. However, this is just a cooked
example. We have seen that if the companies have slightly differentiated products or
some production constraints (so that they cannot meet the whole demand on their
own) the Bertrand paradox will not hold.

Finally let us mention that at the first glance the differences between Cournot
and Bertrand may be misleading as in reality companies may choose both: either
simultaneously set prices and quantities or sequentially, first compete in production
possibilities (capital investment or scale of operations) (Cournot) and then, when pro-
duction possibilities are fixed, compete in prices (Bertrand). Also there are markets
in which in some business cycle phases companies compete a la Bertrand, while a la
Cournot in the others. These issues are analyzed in a few papers including Kreps and
Scheinkman (1983), Davidson and Deneckere (1986), Klemperer and Meyer (1989), or
Kovenock, Deneckere, Faith, and Allen (2000) and d’Aspremont and Dos Santos Fer-
reira (2009) for more recent developments.

8.4 Monopolistic competition

The final model we analyze in this chapter is a model of monopolistic competition (also
known as Chambarlain or Dixit and Stiglitz (1977) model). For this reason consider
m (think of m as large) firms producing differentiated but substitutable products
and (simultaneously) competing in prices. This is called a monopolistic competition
as each firm is a monopolist in producing its own product, but as goods are (to
some degree) substitutable firms must consider other prices as well. This is similar
to Bertrand with differentiated good example we analyzed so far, but not quite, as
with large m each firm will take a market price index P as given. Assume constant
marginal costs ci for each firm.
We consider an example of demand derived from CES utility function with n goods.
REFERENCES 67

1 ρ
Pm
Recall (see example 2.8) that in such case di (pi , P ) = P1 piρ−1 , where P = j=1 piρ−1 ,
0 < σ < 1 and we normalize consumer income I = 1. P is called a price index and as
m is large we will assume that, when choosing its optimal price, each firm will take
P as given.
For given CES demand functions di , a monopolistic competitive equilibrium
is a vector of prices (p∗i )ni=1 and index P ∗ such that:
1
piρ−1
1. (∀i) p∗i ∈ arg maxpi (pi − ci )( P∗ ),
Pn ρ
2. P ∗ = j=1 p∗i ρ−1 .

Observe that this is not the Nash equilibrium of the corresponding Bertrand game
as in the Bertrand game (even for large m) each firm should see its impact on the
price index P . Formally such game is called aggregative game as each player plays
against the aggregate and in equilibrium aggregate value is determined by joint actions
of all players.
The monopolistic competitive equilibrium is easily characterized for CES prefer-
ences as the optimal price company i does not vary with P (but generally it does not
need to be so). Specifically p∗i = cρi and observe price is, higher than marginal costs ci
as ρ < 1. Hence in equilibrium each firm has a margin resulting from its monopolistic
power.
Finally for an aggregative game, if one wants the Nash equilibrium being equal
to the aggregative equilibrium, one needs m to be large, usually a continuum. Hence
economists also analyze large games with various concepts of equilibria (see Mas-
Colell, 1984, Schmeidler, 1973, for seminal treatments).

References
Amir, R. (1996): “Cournot oligopoly and the theory of supermodular games,” Games
and Economic Behavior, 15(2), 132–148.

Amir, R., and I. Grilo (1999): “Stackelberg versus Cournot equilibrium,” Games
and Economic Behavior, 26(1), 1–21.

Amir, R., and V. E. Lambson (2000): “On the effects of entry in Cournot markets,”
Review of Economic Studies, 67(2), 235–54.

Amir, R., and A. Stepanova (2006): “Second-mover advantage and price leadership
in Bertrand duopoly,” Games and Economic Behavior, 55(1), 1–20.

Baye, M. R., and J. Morgan (1999): “A folk theorem for one-shot Bertrand games,”
Economics Letters, 65(1), 59–65.
68 REFERENCES

Besanko, D., D. Dranove, M. Shanley, and S. Schaefer (2007): Economics

of strategy. John Wiley & Sons, Hoboken, NJ.
Church, J., and R. Ware (2000): Industrial organization: a strategic approach.
Irwin McGraw-Hill.
Dasgupta, P., and E. Maskin (1986): “The existence of equilibrium in discontin-
uous economic games, II: applications,” Review of Economic Studies, 53(1), 27–41.
d’Aspremont, C., and R. Dos Santos Ferreira (2009): “Price-quantity compe-
tition with varying toughness,” Games and Economic Behavior, 65(1), 62–82.
Dastidar, K. G. (1995): “On the existence of pure strategy Bertrand bquilibrium,”
Economic Theory, 5(1), 19–32.
Davidson, C., and R. Deneckere (1986): “Long-run competition in capacity,
short-run competition in price, and the Cournot model,” RAND Journal of Eco-
nomics, 17(3), 404–415.
Dixit, A. K., and J. E. Stiglitz (1977): “Monopolistic competition and optimum
product diversity,” American Economic Review, 67(3), 297–308.
Klemperer, P. D., and M. A. Meyer (1989): “Supply function equilibria in
oligopoly under uncertainty,” Econometrica, 57(6), 1243–77.
Kovenock, D., R. Deneckere, T. Faith, and B. Allen (2000): “Capacity pre-
commitment as a barrier to entry: A Bertrand-Edgeworth approach,” Economic
Theory, 15(3), 501–530.
Kreps, D. M., and J. A. Scheinkman (1983): “Quantity precommitment and
Bertrand eompetition yield Cournot outcomes,” Bell Journal of Economics, 14(2),
326–337.
Mas-Colell, A. (1984): “On a theorem of Schmeidler,” Journal of Mathematical
Economics, 13(3), 201–206.
Maskin, E. (1986): “The existence of equilibrium with price-setting firms,” American
Economic Review, 76(2), 382–86.
Pepall, L., D. Richards, and G. Norman (2008): Industrial organisation. Con-
temporary theory and empirical applications. Blackwell Publishing.
Schmeidler, D. (1973): “Equilibrium points of nonatomic games,” Journal of
Statisitical Physics, 17(4), 295–300.
Tirole, J. (1988): The theory of industrial organization. MIT Press, Cambridge.
Vives, O. (2000): Oligopoly pricing. Old ideas and new tools. MIT Press, Cambridge.
Chapter 9

General equilibrium

In chapter 4 we have introduced a concept of partial equilibrium, i.e. a pair of price of

some commodity and quantity produced such that demand equals supply. We called
such an equilibrium partial as we have taken demand and supply of a commodity of
interest (say i) as given. Specifically, when deriving demand from a consumer max-
imization problem we have assumed that his/her income I is given and moreover,
when defining partial equilibrium, we have taken demand and supply as correspon-
dences mapping prices of this single commodity pi , i.e. treating all other prices pj6=i
as parameters. This approach is useful if one analyzes a single market that is small
relative to the others, i.e. when assumption, that situation on a given market is not
influencing consumer income nor prices on other markets, is justified.
In this chapter we analyze the general equilibrium, i.e. an equilibrium in all
markets simultaneously, where all markets clear and all prices are determined en-
dogenously. This allows to identify interactions between markets. Our analysis will
be organized around few questions:

• how to propose a meaningful definition of the general equilibrium, taking into

account interactions between markets?

• what are the conditions under which such general equilibrium exists, i.e. is this
likely that all markets will clear simultaneously?

• suppose by Ω we denote the set of all general equilibria of some economy. What
are the conditions such that Ω is a singleton (there is a unique equilibrium)? Is
the number of equilibria finite or infinite?

• what are the properties of a particular equilibrium? is equilibrium allocation

efficient in some sense?

69
 70 9.1 Exchange economy

• is general equilibrium empirically meaningful, i.e. is it general enough to ex-

plain many real markets’ situations? What testable restrictions an equilibrium
imposes on the observed set of data?

Economists have studied these questions for a long time including contributions
of Smith or Walras, but answered these questions relatively recently. The first formal
treatment of these questions have been pursued by Arrow, Debreu and McKenzie (see
Arrow and Debreu (1954) for example). The analysis conducted was positive in the
sense, that they discussed whether an economy can be in equilibrium in all markets
simultaneously, and what are the properties of such equilibrium, but were not aimed
to argue if some real economy actually is in equilibrium or not.
We start with a simplified model of an exchange economy, where there is no
production. It captures many phenomenas of our interest and allows to introduce the
full model (with production) smoothly.

9.1 Exchange economy

We start by defining an economy and then its equilibrium. There are n consumers
each with preferences over consumption set Xi ⊂ RK + represented by a utility function
ui : Xi → R. Each consumer has an initial endowment ei ∈ Xi . We summarize
economy by a list E = (Xi , ui , ei )ni=1 . Vector x such that x ∈ ×ni=1 Xi is called an
Pn Pn
allocation, while allocation satisfying i=1 xi = i=1 ei is called feasible.

Definition 9.1 (Competitive equilibrium) A competitive equilibrium of E is a

pair of (p∗ , x∗ ) such that p ∈ RK ∗ n
+ and x ∈ ×i=1 Xi and both consumer maximization
(CM), and market clearing condition (MC) hold:

(CM) (∀i) x∗i ∈ arg maxxi ∈Xi u(xi ) s.t. p∗ · xi ≤ p∗ · ei ,

Pn ∗
Pn
(MC) i=1 xi = i=1 ei .

Few comments are in order. First, note that an equilibrium is a pair of vectors,
i.e. notion of equilibrium captures the prices and quantity produced in all K-markets.
Here we have assumed that prices are nonnegative but this could be generalized
to some degree, if one wants to incorporate commodities that are ’bads’. Second,
PK
p∗ · xi is a scalar product of two vectors and p∗ · xi = k=1 p∗k xi,k . Third, consumer
maximization condition (CM for short) means that each vector x∗i belongs to the
demand/supply correspondence of a consumer i. Specifically it requires that taking a
vector of prices p∗ as given, each x∗i maximizes utility ui subject to a budget constraint.
This is the same maximization problem as we have considered in chapter 2, but now
consumer income I is endogenous and equal to the value of endowment sold at market
prices. Fourth, market clearing condition (MC for short) requires that demand equals
9 GENERAL EQUILIBRIUM 71

supply. Although prices do not appear in this condition, this is the one that determines
the equilibrium prices. MC is stated with equality sign but some authors state this
condition with ≤ sign. This is an important difference as here we do not allow any
endowment (think of garbage or pollution) not to be fully utilized1 . Fifth, suppose an
equilibrium p∗ , x∗ exists, then λp∗ , q ∗ is an equilibrium as well, for any λ > 0. This
is clear as the demand is homogeneous of degree zero in prices. This means that we
have some degree of freedom in choosing prices2 . Specifically we can normalize prices
PK
so that k=1 pk = 1 or such that p1 = 1, where commodity 1 is called a numéraire.
Also when we analyze number of equilibria, we mean number of normalized equilibria,
i.e. with prices determined up to a scalar multiplication.
Condition MC can be stated as 0 ∈ z(p∗ ), where 0 = (0, . . . , 0) ∈ RK+ and

n
X
z(p) = (x∗i (p, p · ei ) − ei ),
i=1

and x∗i is a Marshallian demand as introduced in chapter 2. Correspondence z is

called an excess demand as it expresses consumer demand net his/her endowment.
If demand is single valued, M C requires vector 0 = z(p∗ ). We will assume that z is
single valued for simplicity in our further analysis. An important property of z is so
called Warlas’ Law. Specifically, if consumers spend all their incomes, then for all
p ∈ RK+ we have p · z(p) = 0, that is the value of excess demand is zero. Warlas’ Law
follows from linearity of prices and assumption that all consumers spend their whole
budgets. It has important corollaries. It implies among other that, if K − 1 markets
clear, then the last market must clear as well. Alternatively, if one claims that some
market is not in equilibrium, he must point another market that is not in equilibrium
as well.
Exchange economy and its equilibrium for K = 2 and n = 2 can be nicely analyzed
using Edgeworth box. We now show a graphical example and then continue with
an algebraic one.

Example 9.1 By 1, 2 denote two goods and by A, B two consumers. Let endowment
(eA , eB ) = (eA,1 , eA,2 , eB,1 , eB,2 ) and preferences A , B be given. Consider a graph
in figure 9.1. We put consumer A in the left, bottom corner while consumer B on the
top, right. Each consumers’ preferences are convex and increase towards the middle of
the box (as indicated by arrows). The length of the horizontal axis is simply the total
endowment of good 1, i.e. eA,1 + eB,1 . Similarly the length of the vertical axis is the
total endowment of good 2, and given by eA,2 + eB,2 . A point (eA , eB ) denotes, how
1 To see importance of this assumption refer to the paper by Cornet, Topuzu, and Yildiz (2003).
2 Recall the (CM) condition. Note, that prices enter the optimization problem only via the budget
constraint. However, for any λ > 0, p∗ · xi ≤ p∗ · ei if and only if λp∗ · xi ≤ λp∗ · ei . Hence the value
of prices does not matter as long as their relation is unchanged.
72 9.1 Exchange economy

Figure 9.1: Example of an Edgeworth box.

x2 B
B

b (x∗A , x∗B )

b
A (eA , eB )

A x1

total endowment is divided between consumers in the economy. Allocation (eA , eB ) is

not Pareto optimal, however. Observe that we can find an allocation that gives both
consumers higher utility: (x∗A , x∗B ). To reach this allocation from an initial endowment
(eA , eB ) it is sufficient that each consumer optimizes his utility subject to a budget
constraint given by the dashed line. A pair of prices (given by the slope of the dashed
line) and allocation (x∗A , x∗B ) constitutes a competitive equilibrium in this economy.

Example 9.2 Consider two consumers i = A, B with preferences over two goods
1−α β 1−β
k = 1, 2. Let uA (xA,1 , xA,2 ) = xα
A,1 xA,2 and uB (xB,1 , xB,2 ) = xB,1 xB,2 . Also let
eA = (1, 0) and eB = (0, 1). In such a case demand is given by (recall example 2.7):
" α(1p +0p ) # " β(0p +1p ) #
1 2 1 2

x∗A = p1
(1−α)(1p1 +0p2 ) , x∗B = p1
(1−β)(0p1 +1p2 ) .
p2 p2

Equating aggregate demand to supply we have:

βp∗
" # " # 
x∗A,1 + x∗B,1 α + p∗2
  
∗ ∗ 1+0 1
xA + xB = = ∗
(1−α)p1
1
= = .
x∗A,2 + x∗B,2 p∗ +1−β 0+1 1
2

p∗ 1−α
Which gives3 p2∗ = β .
1

Having defined an equilibrium the first question that checks consistency of this
concept is, under what conditions the equilibrium exists. We will not address this
question in details here, but we stress that to prove the equilibrium existence it is
sufficient to find a vector p∗ such that 0 ∈ z(p∗ ). Equivalently one can fixed a fixed
point of a map F mapping set △ = {p : RK
P
+ : k pk = 1} into itself, and defined by:

pk + max{0, z̃k (p)}

Fk (p) = P ,
1 + s max{0, z̃s (p)}
3 Observe again, that equilibrium prices are determined via their relation. Clearly, for any λ > 0,
λp∗ p∗
λp∗1 , λp∗2 is also equilibrium price, since λp∗2 = p∗2 = 1−α
β
.
1 1
9 GENERAL EQUILIBRIUM 73

Figure 9.2: Brower fixed point argument.

F (x) F (x)
b

b
b

bc
b

x x
F (x) F (x)
b

x x

where z̃ is some function selected from z, or simply z if single valued. Recall that a
fixed point of F is p∗ = F (p∗ ). To see that any fixed point of F is an equilibrium price
P
vector observe, that p = F (p) assures for all k: max{0, z̃k (p)} = pk s max{0, z̃s (p)}.
Multiplying by z̃k and summing over all k we have:
X X X
z̃k max{0, z̃k (p)} = max{0, z̃s (p)} zk pk = 0,
k s k
P
where the last equality follows by Walras’ law. Hence we have: k z̃k max{0, z̃k (p)} =
0, where each term in the sum is nonnegative. But as the sum is equal to zero hence
z̃k max{0, z̃k (p)} = 0 for all k implying z̃k (p) ≤ 0. To assure equality we must assume
some additional property like desirability of goods whose prices equal zero. This can
be done.
As a result to prove the equilibrium existence, it is sufficient to find a fixed point of
F for some selection z̃. A Brower fixed point theorem assures that a continuous
function mapping nonempty, compact and convex set into itself has a fixed point (see
figure 9.2 to get some intuition, why each of these conditions is necessary). Hence
it suffices to find a continuous function selected from z and appeal to this theorem.
Summing up, conditions guaranteeing the equilibrium existence (for economies with
finite number of commodities) are known and can be obtained under quite general
settings. This is summarized here.
74 9.1 Exchange economy

Theorem 9.1 Assume each Xi = RK + , ui is strictly increasing, continuous and quasi-

concave with (∀i, k) ei,k > 0. Then there exists a competitive equilibrium.

Generally equilibria are not unique. However, interestingly almost all economies
have a finite and odd number of them. Having established equilibrium existence we
analyze its welfare properties. We start with a definition.

Definition 9.2 (Pareto optimality) A feasible allocation x is Pareto optimal if

there does not exist another feasible allocation x̂, such that:

• (∀i) ui (x̂i ) ≥ ui (xi ) and

• (∃i) ui (x̂i ) > ui (xi ).

Few comments. First, Pareto optimality requires efficiency, i.e. there are no wasted
resources. Second, one can expect there are many Pareto optimal allocations. Third,
Pareto optimality will not be confused with any sort of ’fairness’. Having that we can
state a celebrated first welfare theorem.

Theorem 9.2 (First welfare theorem) Assume each Xi = RK + and ui is strictly

∗ ∗ ∗
increasing. If p , x is a competitive equilibrium, then x is Pareto efficient.

The first welfare theorem states that equilibrium allocation is not wasting re-
sources, or that equilibrium allocations are on the Pareto frontier. The second wel-
fare theorem asks if any allocation from a Pareto frontier can be supported as the
equilibrium allocation.

Theorem 9.3 (Second welfare theorem) Let x∗ be a Pareto optimal allocation

such that (∀i, k) x∗i,k > 0. Assume that each ui is continuous, quasi-concave and
strictly increasing. Then there exists a price vector p∗ ∈ Rk+ , p∗ 6= 0 such that x∗ , p∗
is a competitive equilibrium of an economy with endowments ei = x∗i .

Both welfare theorem nicely separate efficiency and welfare / redistribution problems.
Observe that the second welfare theorem is not a direct converse to the first, as
it requires preferences to be convex. Indeed for non-convex preferences it may be
impossible to define linear prices, under which markets could clear. However even if
preferences are non-convex but the number of consumers large, then one can conclude
that any interior Pareto optimal allocation can be supported by liner prices (see
Anderson, 1978).
To understand distinction between first and second welfare theorem, and also
clarify the role of prices in equilibrium we consider the following (welfare) maxi-
9 GENERAL EQUILIBRIUM 75

mization problem:
n
X
max
n
αi ui (xi ), (9.1)
x∈×i=1 Xi
i=1
n
X n
X
s.t. xi = ei ,
i=1 i=1

for some vector of (positive) consumers’ weights (αi )ni=1 . We now state a very im-
portant link between the set of Pareto optimal allocations, solutions to the welfare
maximization problem and competitive equilibrium allocations.

Theorem 9.4 (Negishi (1960)) Assume that each ui is continuous, concave and
strictly increasing.

1. Let x∗ be a Pareto optimal allocation. Then there exists a vector of weights

(αi∗ ) ≥ 0 α∗ 6= 0, such that x∗ solves the problem (9.1).

2. Let x∗ solve the problem (9.1) for some weights (αi∗ ) such that each αi∗ > 0.
Then x∗ is Pareto optimal.

3. If x∗ , p∗ is a competitive equilibrium such that each p∗ ·x∗i > 0, then x∗ solves the
problem (9.1) for αi = λ1i where λi is consumer’s i marginal utility of income.

4. If (∀i, k) ei,k > 0 and x∗ solves problem (9.1), then there exists a price vector
p∗ such that x∗ , p∗ is a competitive equilibrium. In this equilibrium αi = λ1i is
consumer’s i marginal utility of income.

The prices in this theorem are simply Lagrange multipliers associated with feasibility
constraints. More on this interpretation of prices can be found in Bewley (2007)
textbook.
Finally we consider the question, whether the general equilibrium analysis restricts
observed data to some extent. This is addressed in the theorem of Brown and Matzkin
(1996). So consider a scenario that an observer makes several observations of an
economy. Each observation t consists of a price vector pt , aggregate endowment et
and income distribution {I1t , . . . , Int }. We say that this set of data for t = 0, . . . , T
is Walrasian rationalizable, if there exists an increasing utility functions ui , i =
1, . . . , n generating demand functions x∗i (pt , Iit ), such that i x∗i (pt , Iit ) = et for all
P

t = 0, . . . , T .

Theorem 9.5 (Brown and Matzkin (1996)) Suppose such a set of observations
is given. Then, there is an algorithm that could determine (in a finite number of
steps), whether or not the set of observations is Walrasian rationalizable.

Note: that not all observations are Walrasian rationalizable.

 76 9.2 Extensions

9.2 Extensions
Much of the analysis from the previous section can be easily generalized to allow for
many extensions of the basic exchange economy model. We briefly present such three
here: production economy, dynamic economy and economy with assets. More exam-
ples of general equilibrium analysis can be found in textbooks by Ellickson (1993),
McKenzie (2002) and Bewley (2007).

Example 9.3 (Production economy) Consider m firms each with production set
Yj ⊂ RK . Recall positive entries of a vector y ∈ Yj denote outputs, while negative
denote inputs. By πj we denote a profit of firm j. By θi,j ∈ [0, 1] we denote a fraction
of company j owned by a consumer i. Each consumer has utility ui : Xi → R+ ,
endowment ei ∈ Xi and owns fraction of each firms profit θi,1 , . . . , θi,J . A competitive
equilibrium is defined as a triple x∗ , y ∗ , p∗ and profits π ∗ such that ∀i, j x∗i ∈ Xi ,
yj∗ ∈ Yj and p∗ ∈ RK+ such that:
Pm
(CM) (∀i) x∗i ∈ arg maxxi ∈Xi u(xi ) s.t. p∗ · xi ≤ p∗ · ei + j=1 θi,j πj∗ ,

(FM) (∀j)p∗ · yj∗ ≥ p∗ · yj for all yj ∈ Yj and we set πj∗ = p∗ · yj∗ ,

Pn ∗
Pn Pm ∗
(MC) i=1 xi = i=1 ei + j=1 yj .

The new thing in this definition is the the second condition (firm maximization or FM
for short). This requires that each firm chooses a bundle that is a profit maximizing.
In equilibrium each consumer takes both prices and firms’ profit as given. Existence
of competitive equilibrium for a production economy can be similarly established for
closed, convex Yj containing 0. Can competitive equilibrium exist with increasing
returns to scale production function? Moreover definition of Pareto optimality is
simply extended: a feasible allocation x, y is Pareto optimal, if there is no other feasible
allocation x̂, ŷ such that condition (9.2) holds. Also both the first and the second
welfare theorem hold, the latter for convex Yj .

The general equilibrium analysis can be also easily extended to allow specific
dynamic context. Specifically, observe that in the analysis so far we can interpret
goods, as goods in separate (finite number of) moments in time. However, more insight
about dynamic economies can be obtained, when one incorporates time explicitly into
preferences and production possibilities. This is done in the next example.

Example 9.4 (Equilibrium in time) Consider a T +1-period, production economy

with one consumer with preferences over consumption streams c0 , c1 . . . , cT given by:
PT t
t=0 β u(ct ) where 0 < β < 1. Each period consumer owns a unit of time and in
period t = 0 has also an endowment of capital k0 > 0. Consumer can accumulate own
capital using transition: kt+1 = (1 − δ)kt + it , where it ∈ R stands for investment
 9 GENERAL EQUILIBRIUM 77

and δ ∈ [0, 1] for depreciation. There is a single firm with constant returns to scale
production function f : R+ ×[0, 1] → R+ , producing yt = f (kt , lt ). A feasible allocation
in this economy is a list of {ct , lt , kt , it , yt }Tt=0 such that ct + it = yt = f (kt , lt ) and
kt+1 = (1 − δ)kt + it , kt ≥ 0, lt ∈ [0, 1]. Note that once a sequence of {kt , lt } have
been chosen sequences {yt } and {it } are determined. Hence we can simplify notation
dropping these two sequences.
A competitive equilibrium of such economy is a list {c∗t , lt∗ , kt∗ , ltf , ktf }Tt=0 and prices
{pt , wt∗ , rt∗ }Tt=0 such that:
∗

(CM) {c∗t , lt∗ , kt∗ }Tt=0 solves

T
X
max β t u(ct ) s.t.
{ct ,lt ,kt }T
t=0 t=0
T
X T
X
p∗t (ct + kt+1 − kt (1 − δ)) ≤ (wt∗ lt + rt∗ kt ),
t=0 t=0
(∀t) kt ≥ 0, lt ∈ [0, 1], and k0 is given,

(FM) {ltf , ktf }Tt=0 solves:

T
X
max (p∗t f (kt , lt ) − wt∗ lt − rt∗ kt ),
{kt ,lt }T
t=0 t=0

(CM) (∀t) c∗t + kt+1

∗
− (1 − δ)kt∗ = f (kt∗ , lt∗ ), lt∗ = ltf , kt∗ = ktf .

A few comments concerning this definition. First, the consumer has a single budget
constraint. He earns selling his labor services and renting capital. The amount earned
is used to cover consumption and investment spendings. Second, the firm’s problem is
dynamic but there is not explicit discounting. This means that equilibrium prices p∗t
must incorporate time preferences. Third, as (in our case) production function does
not depend on past decisions and also as the firm does not own any capital, we can
alternatively write FM as:

(∀t) ktf , ltf solves max p∗t f (kt , lt ) − wt∗ lt − rt∗ kt .

kt ,lt ≥0

Fourth, as the technology is constant returns to scale (and implies zero profit condi-
tion) we do not need to incorporate profits into household budget constraint. Fifth,
markets clear every period, i.e. for every good separately. As a result the consumption
in every period has a separate price p∗t . Sixth, as the consumer has a single budget
constraint he can (potentially) freely transfer the income between periods. This, ac-
companied with other comments, leads us to interpret markets in this definition as
future markets (for date-contingent claims) being opened only in period t = 0. As a
 78 9.2 Extensions

result this definition is sometimes referred to as Arrow-Debreu competitive equi-

librium, while in the case, where consumer has a sequence of budget constraints, we
say about sequential competitive equilibrium.

More on equilibrium in time can be found in Stokey, Lucas, and Prescott (1989),
chapter 15. Finally we discuss an example of an asset economy.

Example 9.5 (Asset economy) Consider a two period economy, with K goods and
S states of the world. There is no production. Consumers consume only in the
PS
second period and have preferences s=1 πi,s ui (xs ), where πi,s is a vector of (perhaps
subjective) probabilities and ui : RK + → R. The timing is the following: in the first
period s in unknown and trade (exchange) takes place, in the second period state
s is revealed to all consumers, contracts are fulfilled and consumption takes place.
Consumers can condition their consumption choices on the realized state of the world,
hence the consumption set is Xi = RKS + . Consumers have endowments ei ∈ Xi
denoting a vector of state dependent endowments of physical commodities in RK + . An
∗ n
Pn Pn
allocation x = (x1 , . . . , xn ) ∈ ×i=1 Xi is feasible if i=1 xi = i=1 ei . Observe this
requires that each coordinate of both vectors is equal, i.e. markets clear in every state
for every physical commodity. An (Arrow-Debreu) competitive equilibrium is a
pair x∗ , p∗ such that, x∗ ∈ ×ni=1 Xi and p∗ ∈ RKS and:
PS
(CM) (∀i) x∗i ∈ arg maxxi ∈Xi s=1 πi,s ui (xi,s ) s.t. p∗ · xi ≤ p∗ · ei ,
Pn ∗
Pn
(MC) i=1 xi = i=1 ei .

Few comments concerning this definition. First observe that consumers maximize
expected utility choosing state contingent consumption vectors. Again there is a single
budget constraint, indicating that trade is in future (state-contingent) assets. Such
assets are called Arrow securities, and they give a vector of physical goods in state
s and zero otherwise. We implicitly assume that there are markets for all Arrow
securities (this implies that markets are complete). There is an alternative definition
of a competitive equilibrium (called Radner equilibrium) explicitly incorporating
asset (other than Arrow securities) markets and their structures. If assets do not
allow to span the whole uncertainty space, we say that markets are incomplete.

Detailed economic analysis of equilibrium with asset markets and uncertainty includ-
ing complete and incomplete market case can be found in excellent textbooks by
Werner and LeRoy (2001) and Magill and Quinzii (1996).
We finish with mentioning some current research in the general equilibrium. Apart
from the already mentioned (dynamic economies, incomplete markets or both) this
includes work on equilibrum with infinitely many goods or consumers (Aliprantis,
Brown, and Burkinshaw, 1990), general equilibrium with adverse selection (Rustichini
and Siconolfi, 2008), moral hazard (Jerez, 2005), as well as endogenous risk (Magill and
REFERENCES 79

Quinzii, 2009) and extensive work on recursive competitive equilibrium (Ljungqvist

and Sargent, 2000, part III). See also Ginsburgh and Keyzer (1997) for an exposition
of computable and applied general equilibrium theory.

References
Aliprantis, C. D., D. J. Brown, and O. Burkinshaw (1990): Existence and
optimality of competitive equilibria. Springer-Verlag, Berlin.

Anderson, R. M. (1978): “An elementary core equivalence theorem,” Econometrica,

46(6), 1483–87.

Arrow, K. J., and G. Debreu (1954): “Existence of an equilibrium for a compet-

itive economy,” Econometrica, 22(3), 265–290.

Bewley, T. F. (2007): General equilibrium, overlapping generations models, and

optimal growth theory. Harvard University press.

Brown, D. J., and R. L. Matzkin (1996): “Testable restrictions on the equilibrium

manifold,” Econometrica, 64(6), 1249–62.

Cornet, B., M. Topuzu, and A. Yildiz (2003): “Equilibrium theory with a mea-
sure space of possibly satiated consumers,” Journal of Mathematical Economics,
39(3-4), 175–196.

Ellickson, B. (1993): Competitive equilibrium: theory and application. Cambridge

University Press.

Ginsburgh, V., and M. Keyzer (1997): The structure of applied general equilibrium
models. MIT Press, Cambridge.

Jerez, B. (2005): “Incentive compatibility and pricing under moral hazard,” Review
of Economic Dynamics, 8(1), 28–47.

Ljungqvist, L., and T. Sargent (2000): Recursive macroeconomic theory. The

MIT Press.

Magill, M., and M. Quinzii (1996): Theory of incomplete markets. MIT Press.

Magill, M., and M. Quinzii (2009): “The probability approach to general equilib-
rium with production,” Economic Theory, 39(1), 1–41.

McKenzie, L. W. (2002): Classical general equilibrium theory. The MIT Press.

Negishi, T. (1960): “Welfare economics and the existence of equilibrium in a com-

petitive economy,” Metroeconomica, 23, 92–97.
80 REFERENCES

Rustichini, A., and P. Siconolfi (2008): “General equilibrium in economies with

adverse selection,” Economic Theory, 37(1), 1–29.

Stokey, N., R. Lucas, and E. Prescott (1989): Recursive methods in economic

dynamics. Harward University Press.

Werner, J., and S. LeRoy (2001): Principles of financial economics. Cambridge

University Press.
Chapter 10

Asymmetric information

In this chapter we sketch two important problems of informational economics, that

of: adverse selection and principal agent models. Both incorporate assumption that
information is asymmetrically distributed between the parties. Adverse selection con-
cerns the situation, when information is asymmetric before signing of the contract,
while principal-agent models concern the situation, where asymmetric information de-
velops subsequently during the course of interactions. Such asymmetric information
can result for example from actions taken by one side.

10.1 Adverse selection

Consider three examples:

• on the car insurance market, a driver typically knows more about its driving
skills than an insurance company,

• on the second-hand sale market, a seller usually knows more about a product
quality than a potential purchaser,

• a worker usually knows more about its personal skills than its employer.

In all these cases information is asymmetric and can adversely impact the uninformed
side. Hence the uninformed side should take these considerations into account. We
now (following Mas-Colell, Whinston, and Green (1995)) consider the third example
but the same logic applies to other.
Consider the labor market with many firms each with a constant returns to scale
production function with a single input (labor). Firms are risk neutral and price
takers. Price of a product is normalized to 1. Firms hire workers. Workers differ in
terms of their productivity θ ∈ [θ, θ] ⊂ R. The distribution of productivities is given

81
82 10.1 Adverse selection

by f . Workers obtain salary w, if work but r(θ) of their reservation salary, if stay
at home. If firms cannot observe individual workers’ productivities, the competitive
equilibrium is a pair w∗ , Θ∗ such that:

w∗ = E[θ : θ ∈ Θ∗ ],
Θ∗ = {θ ∈ [θ, θ] : r(θ) ≤ w∗ }.

The first condition equals wage with average productivity of all workers that (second
condition) accept such wage. This restriction is called the adverse selection restriction,
i.e. firms properly anticipate that, when salary is too small the high productive workers
will stay out of the market (and enjoy r(θ)).
This condition may lead to nonexistence of equilibrium, its multiplicity or even
the total collapse of the market, e.g. if the firms expects only the least productive
workers to stay at the market and offer corresponding wage. All of these can occur
even if r(θ) ≥ θ, i.e. it is efficient to employ all workers if information was public.
Hence the equilibrium allocation does not need to be Pareto optimal and hence
the conclusion of the first welfare theorem fails. Examining the assumptions of the
first welfare theorem we immediately see that the inefficiency results, as we have a
single market (for work) for different goods θ. For the efficiency of allocation, one
would require that each productivity θ is traded in a separate market. If this is not
possible (because of asymmetric information) agents are enforced to trade different
goods under an average price. As a result the most efficient agents are not willing to
work under such average price as this is to small to compensate their skills. They stay
out of the market and hence the average is reduced and new group of agents go out.
This process continue until equilibrium conditions are met. Adverse selection is
hence said to occur if the trading decision of one (informed) party adversely influence
the uninformed party.
Observe that, in the case of adverse selection, Pareto-optimality concept (used in
the first welfare theorem) is somehow inappropriate. Specifically, it is not appealing
to compare market allocations (with uninformed agents) to efficient allocations (al-
lowing for perfect information). Hence the concept of constrained Pareto-optimality
is introduced, where the Pareto-improvement is a subject to the same information
constraints as specified in the model. Importantly: some competitive equilibria under
adverse selection may be constrained Pareto-optimal.
Various solutions to the adverse selection problem (Akerlof, 1970) has been pro-
posed including signaling and screening, both formalized using game theoretic con-
cepts (see Rothschild and Stiglitz, 1976, Spence, 1973). Signaling is a class of games,
where the informed party moves first and signals its private information. Screen-
ing is a class of games, where the uninformed party moves first and tries to screen
the informed one for its private information. Equilibria in such games may be sep-
arating (separating workers with respect to their productivities) or pooling (not
10 ASYMMETRIC INFORMATION 83

separating). Separating equilibria allocations may be constrained Pareto optimal in

the case of signalling. In the case of screening usually pooling equilibria do not exists,
and separating may be non-existent as well.

10.2 Principal-agent problem

In this section we consider the problem of principal rewarding agent for his work.
Specifically an agent chooses an action a ∈ {a, a} that affects a probability distribution
f of realization of random output y ∈ {y1 , . . . , yn }. The principal owns (random)
production technology and contemplates an optimal reward contract for the agent.
Principal is risk neutral, while risk-averse agent has (quasilinear) utility of income
u(w) and cost of action c(a). Principal does not observe the effort (action) of the
agent but only the realization of y. The problem of the principal is to find the optimal
contract specifying wage as a function of income, or {wi }ni=1 for short. Consider a
problem of:
n
X
max
n ∗
(yi − wi )f (yi |a∗ ) s.t.
{wi }i=1 ,a
i=1

Pn
(PC) i=1 u(wi )f (yi |a∗ ) − c(a∗ ) ≥ ū,
Pn
(IC) a∗ ∈ arg maxa∈A i=1 u(wi )f (yi |a) − c(a).

The principal wants to maximize an expected value of output y minus salary w.

The first constraint is called participation constraint (or individual rationality)
and reflects the outside option ū of the agent, while the second is called incentive
compatibility and assures that agent in motivated to choose the action a∗ that prin-
cipal is trying to implement. The principal must take both of these into account when
choosing optimal w. The above problem can be solved in two steps, first the princi-
pal chooses the optimal contract {wi } implementing each action, and then compares
profits for all actions a ∈ A. The latter is simple, hence we will focus on the former.
Suppose that principal wants to implement a and he writes a Lagrangean:
n
X
(yi − wi )f (yi |a) +
i=1
n
" #
X
λ u(wi )f (yi |a) − c(a) − ū +
i=1
" n n
#
X X
µ u(wi )f (yi |a) − c(a) − u(wi )f (yi |a) + c(a) .
i=1 i=1
84 10.2 Principal-agent problem

Multiplier λ is associated with participation constraint and µ with incentive compat-

ibility. For differentiable primitives, the first order condition implies for all i:

−f (yi |a) + λu′ (wi )f (yi |a) + µu′ (wi )[f (yi |a) − f (yi |a)] = 0,

and hence:  
1 f (yi |a)
=λ+µ 1− .
u′ (wi ) f (yi |a)
This is a fundamental first order condition for optimal (interior) contract (wi )ni=1 . It
states that optimal contract should give inverse marginal utility equal to the sum of
Lagrange multiplier for participation constraint (corresponding to the ”fixed salary”)
and product of a Lagrange multiplier of incentive compatibility and one minus likeli-
hood ratio ff (y i |a)
(yi |a) (corresponding to the ”variable pay“ or ”motivating salary“). The
likelihood ratio ff (y i |a)
(yi |a) specifies how more/less likely it is to get a draw yi , if action a
is chosen relative to a.
Usually participation constraint is binding hence λ > 0, while incentive compati-
1
bility may be binding or not. In the case of µ = 0, we have ∀i u′ (w i)
= λ and hence
salary wi is independent of yi . This indicates that optimal salary is fixed and fully
insures an agent from the company’s random result. On the other hand, if µ > 0
then optimal salary is variable and follows the likelihood ratio ff (y i |a)
(yi |a) . Specifically if
yi → ff (y
(yi |a)
i |a)
is decreasing, then optimal contract wi is increasing in yi for decreasing
marginal utility.
Finally observe that the Lagrangean is not symmetric in c(a) and c(a). Specifically
c(a) appears in both participating constraint and incentive compatibility, while c(a)
is present only in the latter. This implies that it is weakly better for the principal to
motivate (decrease costs of preferred action), than punish (increase costs of alternative
action).
The basic model (Grossman and Hart, 1983, Holmstrom and Milgrom, 1987,
Rogerson, 1985) has been extended in both applied and theoretical research. These
include Holmstrom and Milgrom (1991) analysis of reward with multiple tasking,
optimal contract for a project work (Holmstrom, 1982), contracting under the ca-
reer concerns (Gibbons and Murphy, 1992) or optimality of tournaments (Lazear and
Rosen, 1981). More advanced studies include work with many principals or many
agents, as well as more recent work in dynamic contracting (Spear and Srivastava,
1987) applied e.g. to design the optimal unemployment insurance (Hopenhayn and
Nicolini, 1997) or model personal bankruptcy (see Ljungqvist and Sargent, 2000, part
V).
Hidden action (or moral hazard) problems are closely related to hidden infor-
mation problem, where an uninformed principal tries to extract information from
an informed agent. That may be different from the adverse selection models analyzed
before, where the information asymmetry was present even before contracting. Both
REFERENCES 85

kinds of problems are nicely analyzed in Salanie (1997) or Laffont and Martimort
(2001).

References
Akerlof, G. A. (1970): “The market for ’lemons’: quality uncertainty and the
market mechanism,” Quarterly Journal of Economics, 84(3), 488–500.

Gibbons, R., and K. J. Murphy (1992): “Optimal incentive contracts in the pres-
ence of career concerns: theory and evidence,” Journal of Political Economy, 100(3),
468–505.

Grossman, S. J., and O. D. Hart (1983): “An analysis of the principal-agent

problem,” Econometrica, 51(1), 7–45.

Holmstrom, B. (1982): “Moral hazard in teams,” Bell Journal of Economics, 13(2),

324–340.

Holmstrom, B., and P. Milgrom (1987): “Aggregation and linearity in the provi-
sion of intertemporal incentives,” Econometrica, 55(2), 303–28.

(1991): “Multitask principal-agent analyses: incentive contracts, asset own-

ership, and job design,” Journal of Law, Economics and Organization, 7(0), 24–52.

Hopenhayn, H. A., and J. P. Nicolini (1997): “Optimal unemployment insur-

ance,” Journal of Political Economy, 105(2), 412–38.

Laffont, J.-J., and D. Martimort (2001): The theory of incentives: the principal-
agent model. Princeton University Press.

Lazear, E. P., and S. Rosen (1981): “Rank-order tournaments as optimum labor

contracts,” Journal of Political Economy, 89(5), 841–64.

Ljungqvist, L., and T. Sargent (2000): Recursive macroeconomic theory. The

MIT Press.

Mas-Colell, A., M. D. Whinston, and J. R. Green (1995): Microeconomic

theory. Oxford University Press.

Rogerson, W. P. (1985): “The first-order approach to principal-agent problems,”

Econometrica, 53(6), 1357–67.

Rothschild, M., and J. E. Stiglitz (1976): “Equilibrium in competitive insurance

markets: an essay on the economics of imperfect information,” Quarterly Journal
of Economics, 90(4), 630–49.
86 REFERENCES

Salanie, B. (1997): The economics of contracts: a primer. MIT Press, Cambridge.

Spear, S. E., and S. Srivastava (1987): “On repeated moral hazard with discount-
ing,” Review of Economic Studies, 54(4), 599–617.

Spence, A. M. (1973): “Job market signaling,” Quarterly Journal of Economics,

87(3), 355–74.
Chapter 11

Externalities and public goods

In this chapter we study a class of public goods and so called externalities. We skip
exposition of many important social choice and political economy problems but give
reference to Olson (1965), Sen (1970) or Persson and Tabellini (2000) instead.

11.1 Public goods

Consider the clasification of goods based on two characteristics: excludability and
rivalry. One can, in principal, exclude some consumers from a consumption of a
excludable good. The rivalry means that consumption of a good by one person reduces
possibilities to consume this good by the others. Private goods are both excludable
and rival. Public goods are goods that are nonexcludable and nonrival, while club
goods are excludable but nonrival. The goods, we have considered so far, were
private. The example of a public good is a lighthouse light, radio signal or knowledge.
We now consider the problem of some public good provision. We start by char-
acterizing its efficient level and then move to the problem of its private provision.
Consider two consumers with preferences given by ui (ci , G) over private consumption
ci and a public good G. Assume each is endowed with income Ii , and that production
of a public good is linear G = g1 + g2 , where gi is consumer i provision to the public
good. Consider the following welfare maximization problem:
max α1 u1 (I1 − g1 , g1 + g2 ) + α2 u2 (I2 − g2 , g1 + g2 ).
g1 ,g2

Observe that the choice of g1 is non only influencing utility of consumer 1 but also
(directly) consumer 2. Such effect is called externality. Assuming differentiability, the
first order condition for interior gi gives:
∂u1 ∂u2 ∂u1 ∂u2
α1 (c1 , G) + α2 (c2 , G) = α1 (c1 , G) = α2 (c2 , G).
∂G ∂G ∂c1 ∂c2

87
88 11.1 Public goods

Rearranging:
2 ∂u1 ∂u2
∂G (c1 , G) ∂G (c2 , G)
X
M RSi = ∂u1
+ ∂u2
= 1 = M RT.
i=1 ∂c1 (c1 , G) ∂c2 (c2 , G)

Interpreting, the efficiency condition equates the marginal rate of transformation with
the sum of marginal rates of substitution for all consumers. The latter condition is new
and reflects the public character of a public good. Consider the following example:

Example 11.1 (Efficient provision of public good) Let ui (ci , G) = γi ln G + ci .

P 2 γi
Then efficiency condition requires: i=1 G = 1 giving G = γ1 + γ2 and feasibility
c1 + c2 + G = I1 + I2 .

As we will show in the moment, private provision of public goods is rarely efficient.
To analyze the private provision of a public good in an (partial) equilibrium framework
we must incorporate some strategic motives into a definition. To see that consider the
following strategic form game between two players each contributing gi to G = g1 +g2 .
We now aim to solve for Nash equilibrium of this game by calculating best responses.
So consider the problem of consumer 1:

max u1 (c1 , g1 + g2 ),
c1 ,g1
s.t. c1 + g1 = I1 ,
and g1 ≥ 0.

Substitute g1 = G − g2 and consider:

max u1 (c1 , G),

c1 ,G
s.t. c1 + G = I1 + g2 ,
and G ≥ g2 .

The second formulation indicates that the consumer 1 is effectively choosing the total
of public good G, for each level of g2 . Apart from the inequality constraint this is a
standard consumer maximization problem. So denote by d1 (I) the demand for G by
consumer 1 at income I, and add inquality constraint to see that: G = max{d1 (I1 +
g2 ), g2 } or equivalently that a Nash equilibrium profile (g1∗ , g2∗ ) satisfies:

g1∗ = max{d1 (I1 + g2∗ ) − g2∗ , 0},

g2∗ = max{d2 (I2 + g1∗ ) − g1∗ , 0}.

To see more continue example 11.1.

Example 11.2 (Private provision of public good) As before consider two con-
sumers with ui (ci , G) = γi ln G + ci and assume that γ1 > γ2 . Then di (I) = γi and
11 EXTERNALITIES AND PUBLIC GOODS 89

(typically for quasilinear utilities) is income independent. Now the condition for the
Nash equilibrium requires:
g1∗ = max{γ1 − g2∗ , 0},
g2∗ = max{γ2 − g1∗ , 0},
and summing:
max{γ1 , g2∗ } = G∗ = max{γ2 , g1∗ }.
With γ1 > γ2 it implies: g1∗ = γ1 with g2∗ = 0. It means that player 1 is contributing
the whole amount of G, while player 2 is free-riding, i.e. benefiting from a public
good, but not contributing. Level G∗ = γ1 is, of course, not efficient as we have seen
before.
Economists considered many mechanisms, that try to implement efficient allo-
cations in the presence of public goods. These include: voting, Lindhal (allocation
and prices), or Groves-Clarke mechanism among others. More on that can be found
in Cornes and Sandler (1996) and Moore (2007).

11.2 Externalities
The public goods analyzed in the previous chapter are only a special case of more
general phenomena called externelities. By externalities we mean that an action of
one decision maker influences directly (i.e. not via prices in the general equilibrium
context) objective function of the other. The examples include the consumption or
production externalities and both can cause positive or negative effects. The example
of negative consumption externalities is cigarette consumption, while negative pro-
duction externalities is pollution caused during the production process. A typical
example of a positive consumption externality is a public good.
In the presence of externalities the conclusion of the first welfare theorem generally
does not hold. The main reason in that, there are goods / services (namely external-
ities) that are not priced. Remember that the implicit assumption of the first welfare
theorem was that all goods / services influencing utilities were priced. Now using a
particular (production externality) example we address the problem and formulate
some solutions.
Example 11.3 (Private and external costs) Consider two firms: 1,2. One pro-
duces x using technology with (private) costs c(x). There is also an external cost
(pollution) of producing x namely e(x) that is perceived by firm 2. Assume that c, x
are monotone, convex and differentiable. The profits are:
π1 (x) = px − c(x),
π2 (x) = −e(x).
90 11.2 Externalities

The competitive equilibrium requires p = c′ (xP C ), which gives too large production
level as compared to the choice that maximizes the social welfare π1 (x) + π2 (x) =
px − c(x) − e(x). To see that consider the optimality condition p = c′ (xP O ) + e′ (xP O )
and observe that price should be equal to the total social costs including private and
external effects.

Economists proposed two main solutions to the above problem: taxes (Pigou) and
opening of the missing markets (Coase). We discuss them in the next two examples.
Example 11.4 (Pigovian Tax) Continue example 11.3 but now suppose that emis-
sion of external effects is taxed with a linear rate t. Now the problem of the first firm
is:
max px − c(x) − tx,
x

and gives the first order condition p = c′ (x) + t. Now setting t = e′ (x) would restore
efficiency of the allocation. Of course this solution is subject to the tax authority
knowing function e and optimal level xP O .

Example 11.5 (Missing markets) Continue example 11.3 but now suppose we in-
troduce a market for externalities, where companies can trade (buy and sell) externali-
ties (e.g. pollution), under the price q. In such a case, problems of the both companies
are

max px1 + qx1 − c(x1 ),

x1
max −qx2 − e(x2 ).
x2

where the company 1 sells externalities and the company 2 buys them. In equilibrium
we have −e′ (x2 ) = q = c′ (x2 ) − p. When the market clears x1 = x2 and hence
we obtain the efficient outcome. Observe that in equilibrium q < 0 indicating that
externalities (pollution) is a ’bad’.

The efficient solution obtained in 11.5 indicates that the problem of externalities is a
problem of missing markets. When introducing a new market one should also specify
endowments of property rights to the goods traded (of simply endowments). Coase
argued that if trading is costless, it does not matter for the efficiency, who owns the
property rights. In our example, indeed, it does not matter, if firm 1 has a right to
pollute or firm 2 has a right for a clean air. Such allocation of property rights matters
of course for a division of income / earnings among firms / consumers. Finally and
from a different perspective, the missing markets argument can also illustrate that
company two has an incentive to buy company 1 and coordinate production level to
the optimal one. Indeed, as the joint profit will exceed the sum of individual profits,
the company 2 will always have enough money to cover the market price (i.e. profit)
of purchase of firm 1.
REFERENCES 91

References
Cornes, R., and T. Sandler (1996): The theory of externalities, public goods, and
club goods. Cambridge University Press.

Moore, J. (2007): General equilibrium and welfare economics. Springer.

Olson, M. (1965): The logic of collective action. Harvard University Press, Cam-
bridge.

Persson, T., and G. Tabellini (2000): Political economics: explaining economic

policy. MIT Press, Cambridge.

Sen, A. K. (1970): Collective choice and social welfare. Holden-Day, San Francisco.
Index

allocation fixed, 27, 31, 34

feasible, 70, 77 long run, 31
Pareto optimal, 74 marginal, 30, 33
ambiguity, see risk vs. uncertainty minimization problem, 30
Arrow securities, 78 nonavoidable, 29
nonsunk, 29, 34
barriers to entry, 39 sunk, 29, 33
best response, 54, 63, 64, 88 total, 30
block pricing, 42 variable, 31
budget
constraint, 16 deadweight loss, 39
set, 16 demand, 17
bundling, 42 CES utility, 19
Cobb-Douglas utility, 18
certainty equivalence, 45 excess, 71
compensating variation, 20 for inputs, 30
complementarities for insurance, 44
consumption, 14 for leisure, 19
economies of scope, 31 Hicksian, 18
production, 29 inverse, 60
strategic, 53, 61, 63 linear, 38
complete markets, 78 Marshallian, 17, 71
concentration ratio, 59
consequentialism, 44 economics
consumer maximization problem, 16 definition, 7
borrowing/lending, 19 macro-, 7
labor supply, 19 micro-, 7
consumption set, 11 normative, 7
costs positive, 7
accounting, 29 economies
alternative, 29 of experience, 31
average, 30 of scale, 31
avoidable, 29 of scope, 31

92
INDEX 93

Edgeworth box, 71 niche choice, 57

elasticity Prisoners’ dilemma, 54
of demand, 38 Stackelberg, 62
cross-price, 20 strategic, 51
income, 20 sub-, 56, 62
price, 20 generalized axiom of revealed preferences,
of substitution, 19, 27 21
constant, 29 goods
of supply club, 87
price, 20 complements, 61
empty threat, 56 different, 12
equilibrium excludable, 87
Arrow-Debreu, 77, 78 inferior, 19
competitive, 35, 70 luxury, 19
general, 9, 69, 70 normal, 19
long run, 35 perfect substitutes, 60
Nash, 9, 52, 57, 60, 63, 88 private, 87
normalized, 71 public, 87
partial, 35, 69 rival, 87
pooling, 82 Groves-Clarke mechanism, 89
Radner, 78
separating, 82 hidden
subgame perfect, 56, 62 action, 84
equivalent variation, 20 information, 84
exchange economy, 70
expenditure minimization problem, 18 incentive compatibility, 41, 83
externality, 87, 89 incomplete markets, 78
index
fair insurance price, 45 Herfindahl-Hirschman, 59
fixed point, 72 Lerner, 38
indifference curve, 13
game information
aggregative, 65 asymmetric, 81, 84
battle of sexes, 52, 55 set, 56
Bertrand, 62–64 isoquant, 27
coordination, 53
Cournot, 60, 64 learning-by-doing, 31
entry, 56 Lindhal, 89
extensive form, 55 lottery, 43, 55
large, 65
matching pennies, 53, 55 marginal
94 INDEX

costs, 30, 33 minimization problem, 30

rate of substitution, 14 production function, 25
rate of technical substitution, 27 CES, 29
rate of transformation, 88 Cobb-Douglas, 29, 30
revenue, 37 Leontief, 28
utility, 13 linear, 28
model long run, 28
Bertrand, 62 short run, 28
Chambarlain, 64 production possibilities set, 26
Cournot, 60 productivity
monopolistic competition, 64 average, 26
Stackelberg, 62 marginal, 26
monopsony, 38
moral hazard, 84 rent-seeking, 39
representative consumer, 22
natural monopoly, 39 returns to scale, 27
nonlinear prices, 16, 41 constant, 28, 77
decreasing, 28, 31
paradox fixed factor, 28
Allais, 47 increasing, 27, 31
Bertrand, 63 risk
Ellsberg, 48 averse, 45, 83
Pareto optimality, 54, 74, 82 aversion
constrained, 82 absolute, 45
participation constraint, 83 relative, 46
Pigovian tax, 90 lover, 45
preferences, 12 neutral, 45, 47, 81, 83
GARP, 21 premium, 46
Gorman, 22 vs. uncertainty, 49
homothetic, 15
perfect complements, 14 scale
perfect substitutes, 14 diseconomies of, 31
quasi-linear, 15 economies of, 31
revealed, 20 minimal efficient, 31
price discrimination, 40 screening, 82
price taking, 33, 76 signaling, 82
Prisoners’ dilemma, 54 Slutsky equation, 20
probability strategic
subjective, 47, 78 complementarities, 53, 61, 63
producer substitutes, 53, 60
maximization problem, 33, 34 strategy, 51
INDEX 95

dominant, 52 state dependent, 49

mixed, 55, 63 von Neumann-Morgenstern, 44
pure, 52
surplus Warlas’ Law, 71
consumer, 22, 36, 41, 42 welfare maximization problem, 35, 74, 87
producer, 36
total, 36

technology, 25
progress, 29
capital saving, 29
labor saving, 29
neutral, 29
theorem
Afriat, 21
Anscombe and Aumann, 47
Brower, 73
Brown and Matzkin, 75
Debreu, 13
equilibrium existence, 73
first welfare, 74
folks’, 57
Negishi, 75
Savage, 48
second welfare, 74
Shephard’s lemma, 31
von Neumann-Morgenstern, 44
tools
comparative statics, 10
constrained optimization problem, 8
equilibrium analysis, 9
two part tariff, 41
tying, 42

utility function, 13
CARA, 46
CES, 19
Cobb-Douglas, 14
CRRA, 46
mean-variance, 46
quasi-linear, 15