Partition-form Cooperative Games in Two-Echelon Supply Chains

Gurkirat Wadhwa, Tushar Shankar Walunj and Veeraruna Kavitha

IEOR, Indian Institute of Technology, Bombay, India
{gurkirat, tusharwalunj, vkavitha}@iitb.ac.in

Keywords: Coalition formation game, Worth of coalition, Stackelberg game, Stability and blocking by a coalition.

Abstract: Competition and cooperation are inherent features of any multi-echelon supply chain. The interactions among
the agents across the same echelon and that across various echelons influence the percolation of market demand
arXiv:2401.04939v1 [econ.TH] 10 Jan 2024

across echelons. The agents may want to collaborate with others in pursuit of attracting higher demand and
thereby improving their own revenue. We consider one supplier (at a higher echelon) and two manufacturers (at
a lower echelon and facing the customers) and study the collaborations that are ‘stable’; the main differentiator
from the existing studies in supply chain literature is the consideration of the following crucial aspect – the
revenue of any collaborative unit also depends upon the way the opponents collaborate. Such competitive
scenarios can be modeled using what is known as partition form games.
Our study reveals that the grand coalition is not stable when the product is essential and the customers buy
it from any of the manufacturers without a preference. The supplier prefers to collaborate with only one
manufacturer, the one stronger in terms of market power; further, such collaboration is stable only when the
stronger manufacturer is significantly stronger. Interestingly, no stable collaborative arrangements exist when
the two manufacturers are nearly equal in market power.

1 Introduction etc. The manufacturers compete with each other to at-

tract ‘good’ amount of customer base at ‘good’ prices
Supply chains are complex systems that involve mul-
and rely on the supplier for the raw material. The sup-
tiple agents at multiple echelons. These agents com-
plier at the upper echelon quotes a per-unit price for
pete and/or collaborate with each other to acquire
raw materials to the manufacturers, and, the latter re-
the maximum possible market share at ‘good’ prices.
spond by either quoting a price of final product to the
The agents look for collaborative opportunities to pro-
customers or by deciding not to operate; the choice
vide better quality service, thereby attracting more
of manufacturers also depends upon the production
customers, resulting in enhanced individual perfor-
costs, demand response of the customers, etc.
mance, while others compete with each other if they
This paper aims to find ‘optimal’ pricing and col-
find it beneficial (e.g., in 2016, Walmart teamed with
laborative strategies of the agents using sophisticated
JD.com to compete with Amazon and Alibaba in
cooperative game theoretic tools. Majority of these
China).
games (e.g., in (Li et al., 2023; Zheng et al., 2021)) fo-
Handbooks in Operations Research, (Chen, 2003),
cus on the stability of the grand coalition and further
discusses the importance of coordination on the effec-
on scenarios where the worth of a coalition depends
tiveness of the supply chain (SC). Cooperative game
just upon its members. But many times, the grand
theory facilitates a systematic study of these interac-
coalition may not be stable, and further, the worth of
tions among the agents of SC (e.g., (Arshinder et al.,
the cooperating agents may depend upon the arrange-
2011; Thun, 2005; Nagarajan and Sošić, 2008)).
ment of agents outside the coalition. Such games are
We examine the interplay between cooperation
referred as partition form games, and a recent thesis,
and competition in a two-echelon SC, with two manu-
(Singhal, 2023), provides a comprehensive summary
facturers at the lower echelon directly facing the cus-
of these games (see also (Aumann and Dreze, 1974)).
tomers, and a single supplier at the upper echelon.
In any real-world SC, the revenue or the worth
Customers choose to buy (or not buy) the product
generated (for example) by a supplier, when all the
from one of the two manufacturers based on factors
manufacturers collaborate (i.e., operate as a single
like the quoted price, the reputation of the entities in-
unit) will obviously be different from that in a sce-
volved, the importance of the product (essentialness),
nario where the manufacturers also compete among
themselves. Thus partition-form based study is es- a single supplier; they neglect the partition-form as-
sential to capture the frictions in SC. pects by defining the worth of a coalition to be the
The first main contribution of this paper is to cap- pessimal worth, the minimum (anticipated worth) that
ture the above realistic aspects in an SC by model- the said coalition can generate irrespective of the ar-
ing it as a partition form game and deriving the in- rangement of the left-over agents. However, if a coali-
gredients of the same – to the best of our knowledge, tion (not currently operating) has to block/oppose an
none of the papers in SC literature consider this. We operating configuration (the set of operating coali-
further consider that the agents in any coalition op- tions and revenues/shares of all the agents of SC),
erate together as a single unit by pooling the best re- the coalition should anticipate to derive a better rev-
sources from each partner; furthermore, the possibili- enue than the sum total revenue that its members
ties of vertical and/or horizontal cooperation are also are currently deriving. In other words, the anticipa-
explored. tion is required only for estimating the worth of fu-
The exhaustive partition-form game based study ture (or blocking) coalition (as upfront it is not sure
resulted in some interesting insights. When the prod- of the retaliatory actions of the others), and not for
uct is essential, and when the customers are (almost) the worth currently derived (as considered in (Zheng
indifferent to the manufacturers, the grand coalition et al., 2021)). We consider pessimal worth as the an-
is not stable. It is actually the vertical cooperation ticipated worth of blocking-coalition, while the (cur-
between the supplier and one of the manufacturers rent) worth(s) in any operating configuration is de-
that results in a stable configuration. More interest- rived by solving an appropriate game or optimization
ingly, only the collaboration with the stronger manu- problem.
facturer (strong in terms of market power) is stable – Another recent study in (Li et al., 2023) consid-
no other attribute of the manufacturers makes a differ- ers two assemblers (like manufacturers in our study)
ence (when their reputation among the customers is and many irreplaceable suppliers, where the second
almost the same); the weaker manufacturer operates assembler only competes for customer base and has
alone and competes with the collaborating pair. Even its own set of suppliers – hence this study is not di-
more interestingly, no collaboration is stable when the rectly comparable to ours. However, the study again
manufacturers are of comparable market strengths. neglects the partition form nature of the game – the
When the supplier leads by quoting a price, there worth of any coalition is defined just based on its size.
exists a Stackelberg equilibrium (SBE) at which the As already argued, when one neglects the inherent
agents operate, in contrast, a scenario where all the partition form nature, the results could be misleading
agents make a simultaneous move results in a Nash – it would be interesting to analyze the SC of (Li et al.,
equilibrium at which none of the agents operate. In 2023), after incorporating partition form aspects.
fact, majority of the literature in SC seems to under- In (Nagarajan and Sošić, 2009), authors study
stand this at some level and considers the Stackelberg coalitional stability considering partition-form as-
(SB) framework (e.g., (Li et al., 2023; Zheng et al., pects. However, as is mentioned in the same paper,
2021)). The SB framework significantly favours the they do not consider the worth of the coalition (based
supplier – the supplier enjoys a huge fraction of the on partition), rather assume that all the players in the
revenue generated, which becomes even higher with coalition to agree to quote a common (best) price.
the competition at the lower echelon. This (rather restricted) assumption facilitates in the
The model is described in Section 2, the partition derivation of the revenue generated by a single agent
form games are in Section 4, and the SC is analysed in any partition and thereby study the stability aspects.
in Section 5. All the proofs are in Appendices. In our study, we derive the utility of any coalition de-
pending upon the partition and then consider stability
Literature Survey: There is a vast literature that
aspects based on the division of that worth and the
studies the scope of SC coordination. Almost all the
anticipated utility of the ‘opposing coalition’.
studies consider contract based cooperation (e.g., (Ca-
chon, 2003) and subsequent papers). There are few
strands of literature that study coalition formation 2 Model
ideas, where the agents are bound without any such We consider a two-echelon supply chain (SC), with
enforcement, because they find it beneficial to do so. two manufacturers at the lower echelon and a single
Important and relevant papers in this category are supplier at the upper echelon. The customers pur-
(Nagarajan and Sošić, 2009), (Zheng et al., 2021) and chase the final product from the manufacturers de-
(Li et al., 2023) etc. pending upon various factors (price and the essential-
In (Zheng et al., 2021), authors study a two- ness of the product, reputation of the manufacturer,
echelon sustainable SC with two manufacturers and etc.); while manufacturers obtain the required raw
materials from the supplier depending upon their own The partition PG = {G} where all the agents oper-
customer demand and the price quoted by the sup- ate together is referred to as the GC partition. While
plier, production cost, etc. we have an ALC partition PA = {S, M1 , M2 }, when
Any manufacturer can operate alone, or can col- all the agents operate alone. We also have VC (ver-
laborate with the other manufacturer, or with the sup- tical cooperation) partition PV i = {Vi , M−i } and HC
plier, or with both of them – when both the manufac- (horizontal cooperation) partition PH = {S, M}.
turers operate together, they choose the best among The worth, the revenue generated by any coalition
them for any aspect (e.g., influence, reputation, pro- must be shared appropriately among its members and
duction capacity), while the supplier and manufac- this payoff division also influences the stability as-
turer pair quote one price directly to the customers. pects (Aumann and Dreze, 1974; Singhal et al., 2021;
We examine the impact of the interplay between Singhal, 2023). Further, departing from a majority
cooperation and competition in the above SC using a of the literature (Li et al., 2023; Zheng et al., 2021),
cooperative game-theoretic framework; in particular, the worth of any coalition depends upon the operat-
our research aims to explore the potential for horizon- ing partition leading to a partition form cooperative
tal (within the same echelon) and vertical cooperation game (Singhal et al., 2021; Singhal, 2023); as already
(across echelons) in an SC. We now describe the in- mentioned, the analysis of such games is significantly
gredients of this study in detail. complicated, and the results obtained by omission of
this dependency can be misleading.
2.1 Coalitions and Partitions In all, as a result of the choices made by various
All the agents or a subset of them can operate to- agents in the system, each agent derives some rev-
gether by forming coalitions. Basically, the agents enue/share. The agents are selfish and aim to max-
within a coalition make joint decisions to generate imize their individual revenue/share, which drives
a common revenue while facing competition from their choices, including their collaboration attempts;
other coalitions or agents. One may have more than the paper precisely works in identifying the ‘stable
one coalition operating in the system. Any partition, configurations’ – the partitions and the correspond-
say P = {C1 , · · · , Ck }, represents the operating ar- ing payoff divisions.
rangement of agents into distinct coalitions and sat-
isfies the following:
2.2 Market Segmentation
In an SC, the manufacturers satisfy customers’ de-
∪m Cm = {M1 , M2 , S}, and Cm ∩ Cl = 0/ if m ̸= l. mands and rely upon suppliers for raw materials or in-
The goal of this paper is to study the interactions be- termediate products. Customers choose one manufac-
tween these coalitions and predict the emergence of turer (or none) based on the price, reputation, loyalty,
stable partition(s) (if any). Prior to this, we need to and other factors. The demand segmentation is also
understand the criteria for declaring a partition stable. influenced by the essentialness of the product, which
Even prior to this, we need to derive the revenues gen- we capture using a parameter γ and a cross-linking
erated by various coalitions in each partition – we re- factor ε that also captures the customers’ affinity to
fer to these revenues as the worths of the coalitions, a switch loyalties. When the manufacturers do not op-
term commonly used in the cooperative game theory erate together, the market is segmented between them
literature (Singhal et al., 2021; Singhal, 2023; Au- based on their selling prices pi and essentialness fac-
mann and Dreze, 1974). The stability concepts are tors (γ, ε) as in equation (1) given below. This is in-
discussed in Section 4, while the worths related to spired by the models commonly used in SC literature
various partitions are derived in various sections. For (see, e.g., (Zheng et al., 2021; Li et al., 2023). With
now, we discuss important and interesting partitions y+ = max{0, y}, the demand derived by Mi equals:
+
and coalitions.

When two manufacturers operate together, we DM = d¯M − αM pi + εαM−i p−i , (1)
i i i

have a coalition M = {M1 , M2 }, with horizontal co-

operation (HC) at the lower echelon. When the sup- with αM := α̃M (1 − γ), where,
i i

plier and a manufacturer operate together, we have a • d¯M is the dedicated market size of Mi ,
i

coalition with vertical cooperation (VC), e.g., Vi = • αM pi is the fraction of demand lost by Mi due to
i
{Mi , S}. When any agent operates alone, we have its price pi , sensitized by parameter αM , i
a coalition with a single player, e.g., Mi = {Mi } or
S = {S}. When all the agents operate together as in a • The essentialness factor γ dictates the sensitivity
centralized SC, we have a grand coalition (GC), rep- of price pi on demand – for example, when γ ≈ 1,
resented by G = {S, M1 , M2 }. the product is highly essential, and the customers
are insensitive to price,
 • εαM−i p−i is the fraction of customer base of M−i
First Echelon Supplier
that rejected M−i and shifted to Mi ,
q q
• The demand is positive as long as the term inside
(·)+ is positive; else, the demand is zero.
Second Echelon Manufacturers
The product is essential, either when γ ≈ 1 or
when ε ≈ 1 and then almost all the customers buy the p1 p2
product (for these parameters, observe DM + DM ≈
1 2

d¯M + d¯M , the total market size). When ε ≈ 1, the cus-

1 2 Market share of M1 Market share of M2
tomers buy the product from one or the other man-
ufacturer (need not be loyal); otherwise, they prefer Common Market
to buy from their own manufacturer (are loyal). Gen- Figure 1: System model, when all agents operate alone.
erally, the sum of demands of both manufacturers is
strictly less than the total market size, and the gap de- In a VC coalition Vi = {S, Mi }, the coalition sets
pends upon the essentialness parameters (γ, ε). a price q for supplying raw materials to the manufac-
HC Coalition: When both manufacturers operate to- turer outside the coalition, and sets a price pi directly
gether, as in M or G, they can potentially attract both to customers by jointly producing the final product. In
customer bases. Further, for manufacturing purposes, the grand coalition G = {S, M1 , M2 }, which involves
the coalition uses the methods of the manufacturer vertical and horizontal cooperation, the coalition di-
with the lowest manufacturing cost; thus, its per-unit rectly quotes a price p for customers and makes a
manufacturing cost is CM = minMi ∈C CM , where CM is combined effort to produce the final product. The re-
i i

the per-unit manufacturing cost of the manufacturer spective actions are represented by aV and aG and the
Mi . As the best of the two capabilities are utilized, action sets by AV and AG (defined as before).
and as the customers are aware of it, we assume the
2.3.1 Utilities in ALC Partition
reputation of the coalition equals that of the best. In
all, we assume the demand function of coalition with We begin with describing the utilities of various
horizontal cooperation to be: agents when all of them operate alone, i.e., when the
DM = d¯M − αM p, with αM := min{αM , αM }, partition is PA (see Figure 1). Let a := (aS , aM ) rep-
1 2
resent the actions of all the agents (the supplier, and
d¯M = d¯M + d¯M . (2)
1 2
both the manufacturers), where aM := (aM , aM ). 1 2

There is obviously no cross-linking (shift of cus- Manufacturers’ Utility: The utility of manufacturer
tomers from one manufacturer to the other), or ba- Mi is zero either if it chooses not to operate or if the
sically, the customers have no choice. In some cases, supplier does not operate. Otherwise, utility is the
this can be fatal to the system, as the customers can total profit gained minus the operating cost, where the
get discouraged by the unavailability of options. The former is the product of the demand attracted DM (1) i

product may not appear essential anymore, and the and the profit gained per-unit:
customers may find solace in other related products. UM (a) = (DM (aM )(pi −CM − q)FS − OM ) FM ,
i i i i i
(3)
We observe this phenomenon has significant influence
on stability results of sections 5.1-5.2. where CM is the per-unit production cost incurred
i

by Mi , OM is the fixed operating/setup cost, FC =

i

2.3 Costs, Actions and the Utilities 1{aC ̸=no } represents the flag that coalition C operates,
and q denotes the wholesale price quoted by supplier.
Any agent, supplier, manufacturer, or coalition Suppliers’ Utility: The demand for the supplier’s
has a fixed cost of operation. Therefore, if the raw materials (at higher echelon) percolates from the
agent/coalition does not generate sufficient profit, it lower echelon (manufacturers), based on the choices
incurs negative revenue and can choose not to oper- of the manufacturers. This dictates the utility of the
ate. Let no represent the choice of not operating. The supplier, which is non-zero only if the supplier and
utility of any agent/coalition is 0 when it chooses no . at least one of the manufacturers operate. In all, the
The supplier can decide not to operate, or can utility of the supplier S when it operates alone equals:
quote a price q ∈ [0, ∞). Thus, the action set of sup- ! !
plier when operating alone is AS := {no } ∪ [0, ∞), 2
US (a) = ∑D (aM )FM (q −CS ) − OS FS , (4)
and its action aS ∈ AS . Similarly, when manufacturer i=1
Mi i

Mi decides to operate alone, it quotes a selling price

pi ∈ [0, ∞). Thus, the action and the action set of man- where CS is the cost for procurement of a bundle of
ufacturer Mi is aM ∈ AM := {no } ∪ [0, ∞). raw material required for producing one unit of prod-
i i
uct and OS is the fixed operational cost of the supplier.
2.3.2 Utility in General Partition Recall the demand decreases as the price increases,
but the drop is reduced as the product becomes more
In a general partition, the utility of a coalition is de- and more essential ( when γ ≈ 1) and αM = α̃M (1−γ).
fined as the sum of the utilities of all the agents within We assume the following:
the coalition. As in equation (2) and as described in √
A.2 The total market size d¯M > αMCG + 2 αM O, with
the corresponding sub-section, the coalition utilizes
O := max{2OS , OG , 4OM }.
the best agent for each feature. Also, any VC-based
coalition directly quotes a price to the customers. Basically, the available market size has to be above a
Thus, for example, the action and utility of the grand certain threshold so that the profit from the attracted
coalition G are: demand surpasses the operating and manufacturing
costs – this ensures it is optimal for the agents to op-
aG ∈ {p ∈ [0, ∞)} ∪ {no }, and
erate (see for e.g., Lemma 4 of the Appendix). If it
UG (aG ) = (d¯M − αG p)(p −CG ) − OG FG , where,


is optimal for the agents not to operate, then there is
• OG is the combined operational cost, defined as nothing left to analyse.
OG = min{OM , OM } + OS .
1 2
3.1 ALC Partition
• CG is the combined production cost, defined as
CG = min{CM ,CM } +CS . In this partition, both agents operate independently
1 2
and aim to maximize their respective utilities. We
• αG is the price sensitivity of the grand coalition, consider a Stackelberg game framework, where the
defined as αG = min{αM , αM }.1 2 supplier first quotes a price q per bundle of raw ma-
It is important to note here that the agents in the terial to the manufacturer. The manufacturer then
grand coalition share the revenue generated (UG∗ = quotes a price aM = p (per unit of product) to the cus-
supaG UG ), and there is no price per item to be paid tomers; the customers, in turn, respond by generating
between any subset of them. The definitions of util- a demand DM (aM ), as in (5). Thus, the Stackelberg
ities for other partitions follow similar logic and will game between the supplier and the manufacturer is
be discussed in the respective sections. given by (with a = (aS , aM ), see (3)-(4)):
We conclude this section by making an important
US (a) = (DM (aM )FM (q −CS ) − OS ) FS , (6)
assumption (inspired by commonly made choices in
practical scenarios): UM (a) = (DM (aM )(p − q −CM )FS − OM ) FM . (7)
A.1 If any agent, either supplier or manufacturer, is We now derive the Stackelberg equilibrium (SBE)
indifferent between the action a = no and an a ̸= of the above game.
no , the agent prefers operating choices. Theorem 1. Assume A.1 and A.2. There exists an
We begin with studying an SC with a single man- SBE under which both the agents operate, which is
ufacturer and supplier, which provides the basis and given by a∗M = p∗ and a∗S = q∗ , where
benchmark for analysing the more generic SC (with 3d¯M + αM (CS +CM ) ∗ d¯M + αM (CS −CM )
two manufacturers) of section 5. The stability con- p∗ := , q := .
4αM 2αM
cepts of partition-form games are in section 4.
Further, the utilities at this SBE are given by
3 Single Manufacturer SC (UM∗ ,US∗ ) = (φ − OM , 2φ − OS ), (8)

2
In a two-echelon SC with one supplier and one man- d¯M − αM (CS +CM )
ufacturer, the agents either operate together to form where φ := .
16αM
GC partition {G} or operate alone to form ALC par-
tition {S, M} (observe that M = {M}, G = {S, M} Proof. The utility of manufacturer (7) resembles that
are coalitions with one manufacturer in this section, in Lemma 4 and thus for any q, the optimizer of
while the same respectively represent M = {M1 , M2 } the manufacturer is operating √ (not equal to no ) only
and G = {S, M1 , M2 } in the rest of the paper). To when d¯M ≥ αM (CM + q) + 2 αM OM and then is given
completely understand the coalitional stability aspects by p∗ (q) = d¯M/2αM + (CM +q)/2.
of such a system, it is sufficient to analyze these two If one neglects the operating conditions, then after
partitions. Note that there is no competition among substituting p∗ (q) in (6) we have:
the agents in the same echelon in this case. In the ab- d¯M − αM (CM + q))(q −CS − 2OS


sence of this competition, the market demand (1) for US (q) = US (q, p∗ (q)) =
2
the manufacturer simplifies to (as in (2)):
which is again similar to that in Lemma 4, and
DM (aM ) = d¯M − αM p .


(5) then by the same lemma, the optimal q would have
been q∗ = (d¯M +αM (CS −CM ))/2αM – this is true when q∗ benefit by forming a coalition, as long as they agree
and p∗ := p∗ (q∗ ) = (3d¯M +αM (CS +CM ))/4αM both satisfy to share the extra profit (UG∗ − (UM∗ + US∗ )) in a way
the required operating conditions (i.e., respective ∆s that benefits both of them. Consider a configuration
are ≥ 0) – these are satisfied, as by A.2: (GC, (xM , xS )), where xi represents the payoff alloca-
tion of agent i and which satisfies:
αMCG + d¯M
αM (q∗ +CM ) =
p
< d¯M − 2 αM OM , and
2
p xM + xS = UG∗ , xM > UM∗ , and xS > US∗ . (14)
(d¯ − αMCM ) − αMCS − 2 2OS > 0.
When the profits are shared as above, none of the
By substituting (q∗ , p∗ (q∗ )) in (6) and (7), we derive agents prefer to operate alone. Now consider a config-
the optimal utilities. uration (GC, (xM , xS )) that does not satisfy (14). When
xM + xS < UG∗ , the generated revenue UG∗ is not com-
pletely shared; if share xM < UM∗ , the manufacturer
3.2 GC Partition would prefer to operate alone, as it would then derive
Both the agents operate together, and the system UM∗ ; similarly if xS < US∗ , the supplier would prefer
directly faces the customers and quotes a common to operate alone. Thus such configurations are ‘op-
price p. The per-unit cost CG = CS + CM of the sys- posed’ and hence are not stable. Before we study an
tem includes the procurement cost (of the raw mate- SC with two manufacturers, let us formally discuss
rials) and the production cost (see Section 2.3). Fur- the notions of stability in partition form games.
thermore, the system also has a fixed operating cost
OG = OS + OM , when it operates. Thus, the overall 4 Partition Form Games
utility of the system is:
A partition form game is described using the tuple
UG = DM (aM ) p −CG − OG FG .



(9) ⟨N, (wPC )⟩ where N is the set of players and wPC is
the worth of the coalition C under partition P and is
The utility of any coalition is defined as the optimal
defined only when C ∈ P. As mentioned in the in-
utility that it can derive. Therefore, we have the fol-
troduction, here the worth wPC also depends upon the
lowing simple optimization problem for deriving the ′
utility of the GC: partition P – basically wPC need not equal wPC for two
different partitions P ̸= P′ both of which contain C.
sup (DM (aM )(p −CG ) − OG ) FG . (10) Given a partition P and the worths {wPC } of each
aG ∈AG coalition in P, the next question is about a ‘pay-off’
This problem can be solved using basic (derivative- vector which defines the allocation to each agent in N.
based) methods, and the solution is as follows: A pay-off vector x = (x1 , · · · , xn ) is defined to be con-
sistent with respect to partition P if (see (Aumann and
Theorem 2. Assume A.1 and A.2. There exists an Dreze, 1974; Singhal et al., 2021; Singhal, 2023)):
optimizer at which the system operates, and the cor-
responding optimal price is a∗G = p∗G , where ∑ xi = wPj for all C j ∈ P.
C (15)
i∈C j
d¯M CG
p∗G = + , (11) The pair (P, x) is defined to be a configuration if the
2αM 2
latter is consistent with the former.
and the corresponding optimal utility is given by: The quest now is to study a ‘solution’ of the par-

2 tition form game. The ‘solution’ in this context de-
d¯M − αMCG
UG∗ = − OG . (12) scribes the configurations that are stable; in other
4αM words, it identifies the partitions and their companion
Proof. Again the utility function of GC (10), resem- consistent payoff vectors that can emerge or operate
bles that in Lemma 4. Furthermore, by assumption stably without being ‘opposed’.
A.2, ∆ > 0 for the GC, which implies that the GC will To study the stability aspects, one first needs to
operate. Thus the proof follows by Lemma 4. understand if a certain coalition which is not a part of
the partition can ‘block’ (or oppose) the given config-
Remarks: By Theorems 1 and 2 (observe here CG = uration – such a blocking is possible if the coalition
CS +CM as in subsection 2.3.2), has an ‘anticipation’ of the value it can achieve (ir-
UG∗ − (UM∗ +US∗ ) = φ . (13) respective of all scenarios that can result after coali-
tion blocks) and if the anticipated value is bigger than
Thus, the agents derive higher utility in GC than the what the members of the coalition are deriving in
combined utility that they derive when they operate the current configuration. Basically, if there exists at
alone (i.e., ALC). This means that both agents can least one division of this anticipated value among the
members of the blocking coalition that renders all the aM = no and equals infinity when aM ̸= no . Thus if
members to achieve more than that in the given payoff both the agents compete at the same level, the SC
vector, then the coalition has a tendency to oppose the would not operate and both of them derive 0 revenue.
current configuration. The above concepts are made On the other hand, when the supplier leads the
precise in the following definitions ((Singhal et al., market as in the SB game, by Theorem 1 the sys-
2021; Singhal, 2023)): tem operates resulting in positive revenues for both
Definition 1 (Blocking of a configuration by a coali- the agents. They derive even better utilities by op-
tion). A configuration, the tuple of partition and the erating together and hence GC is stable as shown in
consistent payoff vector, (P, x), is blocked by a coali- Lemma 1. However the supplier gets a much better
tion C ∈/ P, under the pessimal anticipation rule, if share; again from Lemma 1, the share of supplier xS
the coalition derives better than that in the original is at least 2φ − OS while that of the manufacturer is
configuration irrespective of the arrangement of op- at most 2φ − OM . These observations motivate us to
ponent players, i.e., if the pessimal anticipated utility analyse a more generic SC with competition at the
lower echelon. The aim in particular is to understand
′
wCpa := min
′
wP >
′ C ∑ xi . (16) the stable configurations, the profit shares, etc., in the
P :c∈P i∈C presence of lower echelon competition.
Definition 2 (Stability). A configuration (P, x) is said
to be stable if there exists no coalition C ∈ / P that 5 Two Manufacturer SC
blocks it. A partition P can be said to be stable if In this section, we explore the case of two echelon SC
there exists at least one configuration (P, x) involving consisting of a supplier and two manufacturers. As
P which is stable. explained in Section 2.2 and equation (1), the fraction
We now apply the above stability concepts to the of demand captured by each manufacturer is :
single manufacturer case study of the previous sec- DM (aM ) = (d¯M − αM pi + εαM p−i ) for all i. (17)
i i i −i

tion. In this case, N = {S, M} and the only possible If there is no cross-linking (i.e., if ε = 0), the demand
partitions are PG = {N} and PA = {{S}, {M}}. functions get decoupled, and each of them resemble
Clearly, the worth of grand coalition wPG G = UG∗ to that of the single manufacturer SC (see (5)).
given in (11), and that of manufacturer and sup- One needs to derive the worths {wPC } for all possi-
plier, while operating alone, are respectively given by ble coalitions C and partitions P to study the stability
wPMA = UM∗ and wPS A = US∗ of Theorem 1. These com- aspects. The worth wPC can be defined as the ‘best’
plete the definition of the partition form game. Also utility (the maximum sum utility) that the members
observe that any pay-off vector x = (xM , xS ) is consis- of C can derive, while facing the competition from
tent with GC partition if and only if xM + xS = UG∗ = agents outside the coalition arranged as in partition P.
wPG G . On the other hand, the only payoff vector con- The competition between various coalitions is
sistent with the ALC partition is x = (wPMA , wPS A ). The captured via a Stackelberg game (as in Subsec-
Theorems 1-2 immediately imply the following sta- tion 3.1), when at least one manufacturer is not collab-
bility result: orating with the supplier – the partitions of this type
are, ALC partition PA = {S, M1 , M2 }, HC partition
Lemma 1. In the single manufacturer SC: i) ALC PH = {S, M} and the VC partition PV i = {Vi , M−i }.
partition PA is blocked by grand coalition; and ii) In all these cases, the leader is the coalition CL with
The GC-core, the set of consistent pay-off vectors that the supplier. The coalitions with only manufactures
form stable configurations with PG ( see (8), is: form the followers - the followers respond optimally
{x : xS > 2φ − OS , xM > φ − OM and xS + xM = 4φ − OG } . for any given action aL (the quoted prices or no ) of
the leader. The solution of the followers is either an
Proof. From (16), the pessimal anticipated utilities optimizer (when all manufacturers form a coalition)
with |N| = 2, clearly equal wCpa = wPC G for any C. Thus or an NE. Let a∗M (aL ) represent this solution in either
by Theorems 1-2, wGpa = wPG G > wPMA + wPS A and hence case. The leader coalition is aware of this optimal
part (i); part (ii) follows by direct verification. choice, i.e., a∗M (aL ) for every aL is a common knowl-
edge. Thus, the optimal choice of the leader is,
Remarks: From (3)- (4), if the supplier and the man- a∗L ∈ arg max ∑ U j (aL , a∗M (aL ))) ,
ufacturer participate in a strategic form game (i.e., aL
j∈CL
when they make choices simultaneously), the resul- ∗ ∗ ∗
and then (aL , aM (aL )) represents the SBE. We then de-
tant Nash Equilibrium (NE) is (no , no ) – the best re- fine the worth of the leader coalition by:
sponse of the manufacturer is no for any aS = q > CM
or when aS = no , while that of the supplier is no when wPCL = ∑ U j (a∗L , a∗M (a∗L )) .
j∈CL
The worth of the rest of the coalitions of P can be partitions are significantly complex and hence we be-
defined similarly using the SBE (a∗L , a∗M (a∗L )). gin with a specific yet an important asymptotic case
We are just left with the GC partition PG = {G}, study in this conference paper – while the complete
which can be analysed exactly as in Subsection 3.2 generality would be considered in future. We con-
and is considered in the immediate next – we once sider an asymptotic regime near (ε, γ) ≈ (1, 1); as
again assume ‘operating-conditions’ assumption A.2 mentioned previously, here the customers are willing
(with terms like CG etc., accordingly changed); with- to switch the loyalties towards their manufacturers
out loss of generality, we consider d¯M ≥ d¯M .1 2
and hence we call such a regime as Essential and
Substitutable-Manufacturer (ESM) regime. We also
5.1 GC Partition consider manufacturers of equal reputation, i.e., with
In GC partition PG , the two manufacturers and the α̃M = α̃M . Towards obtaining the asymptotic study
1 2

supplier operate together as explained in Subsection we consider the following procedure.

2.3.2. The optimization problem is similar to that in ESM Regime: For any partition-coalition (P, C),
(10), hence we have the following with proof exactly consider the function (γ, ε) 7→ (1 − γ)(1 − ε)wPC .
as in that of Theorem 2: From all the expressions derived in this paper, i.e.,
for all (P, C), these functions are continuous. Hence
Corollary 1. Assume A.1 and A.2. The worth of PG
the following limits exist (for each (P, C)) and can be
defined using the optimizer is given by
rewritten as below:

2
PG ∗ d¯M − αMCG fCP := lim (1 − γ)(1 − ε)wPC
wG = UG = . □ (γ,ε)→(1,1)
4αM
= lim lim (1 − γ)(1 − ε)wPC . (18)
ε→1 γ→1
5.2 HC Partition
We refer to the above as worth-limits, with slight
We now consider the HC partition PH , where both the abuse of notation. Similarly define { fCpa } using antic-
manufacturers M = {M1 , M2 } operate together. The ipated worths {wCpa }. We will also require the limits
coalition of manufacturers M quotes a selling price p of the following derivatives
to the customers, and the leader (supplier) S quotes a
price q to M. Recall any of them may decide not to (1),P d w̃PC
fC :=lim with w̃PC := (1 − ε) lim (1 − γ)wPC , (19)
operate (choose no ). The SBE a∗ := (a∗M , a∗S ) of the ε→1 dε γ→1
Stackelberg game satisfies the following as before: (1),pa
∗ ∗ ∗ ∗
and also that of the anticipated worths, { fC }. The
aM = aM (aS ), and aS∗ ∈ arg max US (aS , aM (aS )), idea is to derive the stability results by comparing
aS ∈AS
the worth-limits instead of the actual worths {wPC },
where the utilities and the optimizers are given by: and further using the derivative limits (19) when the
US (a) := (DM (aM )(q −CS )FM − OS ) FS worth-limits are equal. We claim that such stability
results are applicable for all (ε, γ) in a neighbourhood
a∗M (aS ) := arg max UM (aM , aS ) with
aM ∈AM of (1, 1) because of the following reasons and proce-
UM (a) := (DM (aM )(p −CM − q)FS − OM ) FM , dure.
From (16) a configuration (P, x) is stable if the
with DM (aM ) defined in (2). This game is similar to following set of inequalities are satisfied:
that considered Subsection 3.1, and hence the follow-
ing result using the proof of Theorem 1. ∑ xi ≥ w pa for all C ∈/ P.
C (20)
i∈C
Corollary 2. Assume A.1 and A.2, the worths of the
agents in PH (defined using operating SBE) equal: (we identify only the configurations that satisfy the
above with strict inequality, a more complete study is
{wPS H , wPMH } = {2φ − OS , φ − OM }, again a part of the future work, and the reasons for
(d¯M −αMCG )
2 this omission is evident in the immediate next).
where φ = 16αM . □ If the inequalities in (20) are satisfied in a strict
Using Corollaries 1-2, as in Lemma 1, it is easy manner by some vector y and for some partition P
to conclude that the HC partition is blocked by grand using limits { fCpa } in place of wCpa , then by continuity
coalition G and hence is not stable. there exists γ̄ and ε̄ such that the above inequalities
(finitely many) are satisfied for all γ > γ̄ and ε > ε̄ —
5.3 Worth-Limits this implies that for all those (γ, ε), the configuration
For further analysis, one needs to study the ALC and 1
VC partitions. However the expressions for these two (P, β y) , with, β := ,
(1 − ε)(1 − γ)
is stable; thus one can obtain stability results near
ESM regime using the worth-limits { fCP , fCpa } (when
a∗M = a∗M (a∗Vi ), a∗M (aVi ) := arg max UM (aM , aV ), and
strict inequalities are considered in (20)). −i −i −i
aM−i −i −i

During blocking by mergers, i.e., say when block- a∗Vi ∈ arg max UVi (aV , a∗M (aV )), −i
ing coalition C = C1 ∪ C2 , then recall by consistency aV ∈AVi

in (15), the inequality (20) modifies to

and defines the worths, wPViV = UVi (a∗ ) and wPMV = −i
P pa
∑ xi = w 1 + w 2 ≥ w
C
P
C C . (21) UM (a∗ ). As already mentioned, it is complicated to
−i

i∈C analyze this game theoretically, we instead obtain the

And if now the worth-limits of both right and left hand ESM limits in the following:
sides are equal, then the comparison in neighbour- Lemma 2. Assume αM1 = αM2 = α. The worth-limits
hood is possible only by considering the derivatives. P P (1),P (1),P
( fViV i , fM V i ) and the derivative limits ( fVi V i , fM V i )
−i −i
This is because for such limits, by Taylors series ex- for ESM regime are respectively in Tables 1a and 1b.
pansion, near ε ≈ 1 we have (see (19)):
pa Proof. Refer to Appendix 9 for the proof.
w̃PC1 + w̃PC2 − w̃C (22)
pa

d w̃PC1 + w̃PC2 − w̃C 5.5 ALC Partition
= (ε − 1) + o((1 − ε)2 )
dε
 ε→1
 Recall the partition, PA = {S, M1 , M2 }, where all the
(1),P (1),P (1),pa
= (ε − 1) fC1 + fC2 − fC + o((1 − ε)2 ), agents operate alone and compete with each other. In
this partition, we have a SB game, where supplier
(1),pa (1),P S is the leader, quoting its price via the action aS ;
where the limits { fC , fC } are defined in (19).
Thus the required stability results can be established if the competing manufacturers {M1 , M2 } are follow-
now the derivative limits satisfy the required inequal- ers in the lower echelon, who respond to aS via a non-
ities – and then there exists an ε̄ < 1 such that the cooperative strategic form game (inner game between
stability results are true in a neighbourhood as below: the manudacturers), parametrized by fixed action of
supplier aS . As a leader of SB game, the supplier is
{(ε, γ) : ε ≥ ε̄ and γ ≥ γ̄ε } , (23) aware of the Nash Equilibrium (NE) a∗M (aS ) of the in-
where γ̄ε < 1 is a lower bound depending upon ε. ner game a∗M (aS ) for every aS . The utility of each of
Further to ensure that the coalitions under consid- these agents are given in (3)-(4). The utility of the
eration are operating, one would require conditions supplier S and a manufacturer Mi is ,
like that in A.2. However these conditions are triv-
! !
2
∗
ially satisfied in the limits γ → 1, once d¯M > 0 for all i
US (aS ) = ∑D Mi (aM (aS ))FM i
(q −CS ) − OS FS ,
i; furthermore, the conditions will also be satisfied in i=1
(26)
the neighbourhood of (1, 1) (if required by shrinking
UM (aM ; aS ) = ((DM (aM (aS ))(pi −CM − q)FS − OM ) FM .
the neighbourhood further) due to similar reasons. i i

(27)
i i

5.4 VC Partition The equilibrium of the SB game satisfies

Recall in the partition with vertical cooperation, PV i a∗S = arg max US (aS , a∗M (aS ))) ,
the supplier collaborates with one of the manufactur- aS

ers Mi and competes with the other. The Stackel- aM = arg max UM (aM , aM ; a∗S ) for all i.
∗
i i i −i
aMi
berg game is between the coalition Vi as leader and
the manufacturer M−i as follower. The manufacturer The worth of supplier S is, wPS A = US (a∗S ), and the
M−i (when it operates) obtains raw material from Vi , worth of manufacturer Mi is, wPMA = UM (a∗M ). As al- i i

quotes p−i and the demand DM attracted by M−i also −i ready mentioned, it is complicated to analyze this
contributes towards the revenue of Vi ; the VC coali- game theoretically, we instead obtain the ESM lim-
tion Vi also derives utility due to its own demand its in the following:
DM (recall here a direct price pi is quoted to the cus-
i
Lemma 3. Assume αM = αM = α. The worth- 1 2
tomers). Thus the utilities of the two coalitions are:
limits ( fSPA , fMPA , fMPA ) and the derivative limits ( fS(1),PA ,
i −i

UVi = DM (aM )(pi −CM −CS ) + FM DM (aM )(q −CS ) (1),P (1),P

i i −i −i fM A fM A ) for ESM regime are respectively in Ta-
i −i

− OS − OM FVi , bles 1a and 1b.


(24)
i

UM = DM (aM )(p−i − q −CM )FVi − OM FM . (25)

−i −i −i −i −i
Proof. Refer to Appendix 10 for the proof.
The SBE (a∗ Vi , a∗
M−i
) (when exists) satisfies:
  
fGPG = 0 fSPH , fMPH = (0, 0) (1),PV i 2 −d¯2
2d¯Mi d¯M−i +d¯M (1),PV i d¯2 −i
f Vi = 16α̃
−i Mi
, fM −i
= − 16Mα̃
2
   ¯2     ¯2 
fVPi V i , fMPV i = 8dα̃M , 0 fSPA , fMPA , fMPA = 8dα̃M , 0, 0 (1),PA −(5d¯Mi +d¯M−i ) (1),PA d¯M
2
−i 1 2 fM i
= 144α̃ ∀i, fS = 8α̃

(a) Worth-limits (b) Derivative-limits

Table 1: ESM regime near (γ, ε) ≈ (1, 1)

6 Stability Results – ESM regime that a coalition blocks it. Then the said configuration
remains stable (or is blocked by the said coalition) for
We have derived the worths {wPC } and the worth-
all (γ, ε) around (1, 1) and in a set as in (23) of sub-
limits { fCP } in the previous section and now aim to
section 5.3. We begin with ALC partition.
identify the partitions and configurations that are sta-
ble in ESM regime. We have worth-limits only for ALC partition is not stable: Note that there is
VC and ALC partitions, and thus for the compari- only one scaled configuration at limit involving ALC
son purposes (see (16), (20) and (21)), compute the (PA , y) with yS = fSPA , yM = fMPA and yM = fMPA , be-
1 1 2 2

same for GC and HC partitions. By Corollaries 1-2, cause of consistency. The coalitions that can possibly
fGPG = fMPH = fSPH = 0; these are also tabulated in Ta- block ALC partition are the merger coalitions such as
ble 1. G, M, Vi . Consider a merger Vi which gets the same
From Table 1, the immediate result is that, the GC utility as yS + yM of ALC at limit. Thus we compare
i

and HC partitions (irrespective of the pay-off vectors) using the derivative limits of Table 1b ,
are both blocked by coalition Vi . Thus none among
2
these two partitions are stable. As discussed at the (1),PA (1),PA d¯M2 5d¯M + d¯M
i −i
fS + fM i
=− +
end of Subsection 2.2, the customers may no longer 8α̃ 144α̃
feel the product is essential in the absence of choices, d¯M2 − 2d¯M d¯M − d¯M2
i −i (1),P −i
< i
= fVi V , (31)
and this may be the reason for non-stability of the GC 16α̃
and HC partitions.1 and hence the merger Vi blocks PA (see (22) and ob-
We now identify the stable configurations. To- serve (ε − 1) is negative) and this is true with a strict
wards this, the pessimal anticipatory worth-limits (see inequality at limit. Therefore, PA is not a stable parti-
equation 16), using Table 1, are: tion in the ESM regime.
fSpa = min{ fSPH , fSPA } = 0 Continuing in this manner we obtain the following
result (the remaining proof is in the Appendix):
fMpa = fMPH = 0 (28)
Theorem 3. [ESM regime] Consider any
pa PV −i pa
α̃, d¯M , d¯M , {CC } and {OC } with d¯M ≥ d¯M . There ex-
PA
fMi
= min{ fM i
, fM } = 0, fG = 0, and
i 1 2 1 2

pa PV i d¯2 ists a ε̄ < 1 such that for every ε ∈ (ε̄, 1), there exists
fVi = fVi = M. (29) a γ̄ε < 1 and for any system with above parameters
8α̃
and with (ε, γ) ∈ {ε ≥ ε̄, γ ≥ γ̄ε }, the following are
In the following we prove that: i) the partitions PA and true:
PV 2 are not stable, while PV 1 is stable when √
√ i) When d¯M ≤ d¯M < ( 2 + 1)d¯M , none of the par-
2 1 2

d¯M > ( 2 + 1)d¯M ;

1 2
(30) titions are stable.
√
ii) When d¯M > ( 2 + 1)d¯M , then only PV 1 partition
and ii) none of the partitions are stable when the above 1 2

is stable. Further the configuration (PV 1 , x) is

inequality (30) is negated strictly. Towards this we es-
stable if
tablish few (strict) inequalities at ESM limit for each  
configuration – the set of inequalities are strict and xM ∈ wMpa − wPMV 1 , wPV1V 1 − wPV2V 2 + wPMV 1 , (32)
1 2 2
demonstrate either that the configuration is stable or
1 Observe
and the above interval is non-empty. □
when a worth-limit is non-zero,the corre-
sponding worth/optimal-revenue increases to infinity as Remarks: When there is a huge disparity between
(ε, γ) → (1, 1) and this is true only for worths related to the two manufacturers in terms of the dedicated mar-
VC and ALC partitions in ESM regime; by Corollaries 1- 2 ket demands, the SC has stable configurations. The
the worths for GC and HC partitions also increase to infinity supplier prefers to collaborate with stronger manufac-
with γ → 1, but not with ε → 1 and hence the corresponding turer (PV 1 is stable) and the weaker manufacturer has
worth-limits are zero.
no choice (no partner finds it beneficial to oppose PV 1 Further, the competition at the lower echelon sig-
by collaborating with weaker M2 ). nificantly favors the higher level supplier – the sup-
However the share of the manufacturer that col- plier enjoys a huge fraction of the revenue generated,
laborates with supplier is negligible in comparison to while the manufacturers draw a negligible fraction ir-
that of the supplier (the upper bound on scaled pay- respective of their market powers and irrespective of
off (1 − ε)(1 − γ)xM to 0 as seen from (32) and Table
1
whether they collaborate with the supplier or not.
1, while the lower bound on that of the supplier con- This research has opened up many new questions
verges to d¯M2 /8α̃ ); in fact the scaled share of the non- – which configurations are stable in an SC that sup-
collaborating manufacturer M2 also converges to 0 plies luxury goods (non-essential) or in a SC with
(from Table 1 fMPV 1 = 0). Thus in the essential and
2
loyal customers? More interesting questions are about
substitutable manufacturer regime, the supplier has the stable configurations with competition at both the
even higher advantage than that in the single man- echelons along with vertical competition.
ufacturer SC – the competition at the lower echelon
added with substitutability significantly improved the REFERENCES
benefits and the position of the supplier.
Another interesting aspect is that the supplier Arshinder, K., Kanda, A., and Deshmukh, S. (2011).
prefers to collaborate with stronger manufacturer (PV 1 A review on supply chain coordination: coor-
is stable, but not PV 2 ) – in the prelimit, the supplier dination mechanisms, managing uncertainty and
derives better coalitional utility when it collaborates research directions. Supply chain coordination
with stronger M1 . under uncertainty.
On the other hand, when the manufacturers are of
Aumann, R. J. and Dreze, J. H. (1974). Cooperative
comparable strengths, as in part i of Theorem 3, the
games with coalition structures. International
SC has no stable configuration. The system probably
Journal of game theory.
keeps switching configurations (any operating config-
uration is blocked by one or the other coalition). Cachon, G. P. (2003). Supply chain coordination with
contracts. Handbooks in operations research
and management science.
7 Conclusions
Chen, F. (2003). Information sharing and supply
The main takeaway of this work is that it establishes chain coordination. Handbooks in operations re-
the possible stability of vertical mergers (the collabo- search and management science.
ration between the supplier and a manufacturer) and Li, C., Cao, B., Zhou, Y.-W., and Cheng, T. E. (2023).
instability of centralised supply chain (all the mem- Pricing, coalition stability, and profit allocation
bers of SC); this is true for the SC that supplies essen- in the pull assembly supply chains under compe-
tial goods and that caters to not-so loyal customers. tition. OR Spectrum.
This is in contrast to the current literature which usu-
Nagarajan, M. and Sošić, G. (2008). Game-theoretic
ally establishes the stability of centralized SC. This
analysis of cooperation among supply chain
contrast is the consequence of the realistic consider-
agents: Review and extensions. European jour-
ation of the partition-form aspects – where the worth
nal of operational research.
of a coalition depends upon the arrangement of the
agents outside the coalition. When the manufacturers Nagarajan, M. and Sošić, G. (2009). Coalition stabil-
are significantly different in terms of market power, ity in assembly models. Operations Research.
the vertical cooperation (or merger) between the sup- Singhal, S. (2023). Navigating Resource Conflicts:
plier and the stronger manufacturer is stable, while Co-opetition and Fairness. PhD thesis.
the weaker manufacturer is left-out to compete with Singhal, S., Kavitha, V., and Nair, J. (2021). Coalition
the collaborating pair. Surprisingly, no collaboration formation in constant sum queueing games. In
is stable when the manufacturers are of comparable 60th IEEE Conference on Decision and Control
market strengths. (CDC).
In reality, the worth of any coalition depends upon Thun, J.-H. (2005). The potential of cooperative game
the arrangement of opponents in the market space; for theory for supply chain management. Research
example, the revenue generated by the supplier with Methodologies in Supply Chain Management: In
the two manufacturers operating together is different Collaboration with Magnus Westhaus.
from that when the two manufacturers operate inde- Zheng, X.-X., Li, D.-F., Liu, Z., Jia, F., and Lev, B.
pendently. This realistic aspect is captured by study- (2021). Willingness-to-cede behaviour in sus-
ing the SC using partition-form games, which paved tainable supply chain coordination. Interna-
way to the above mentioned contrasting results. tional Journal of Production Economics.
APPENDIX We get the above as w̃Mpa = w̃PMH and then refer to Ta-
ble 1b and equation (22). Similarly, to ensure the con-
In this appendix, we consider the following optimiza-
figuration (PV 1 , x) is not blocked by singletons S and
tion problem and derive its solution:
M2 , we require
U(a) = (d¯ − α p)+ (p − c) − Oc 1{a̸=no } .


(33) wMpa ≤ xM ≤ wPV1V 1 − wSpa . (37)
1 1
√
Lemma 4. Define ∆ = d¯ − αc − 2 αOc . The maxi- Finally, for stability against blocking by GC, we re-
mizer and the maximum value of (33) is given by: quire
d¯ + cα wPV1V 1 + wPMV 1 ≥ wPG G . (38)
a∗ = p∗ 1{∆>0} + no 1{∆<0} , where p∗ = , 2

2α In view of the above three required inequali-

 ¯
(d − αc)2

U(a∗ ) = − Oc 1{∆>0} . ties (35), (37) and (38), we require (if possible) an
4α ε̄ < 1 such that the following inequalities are satisfied
When ∆ = 0, we have two optimizers, p∗ and no . for each ε ≥ ε̄:
w̃PV1V 1 − w̃PV2V 2 + w̃PMV 1 − (w̃Mpa − w̃PMV 1 ) > 0,
2 2
Proof. Towards solving (33), we first consider opti- PV 1 PV 1
mizing the interior objective, more precisely, w(p) = w̃V1 + w̃M − w̃PG G 2
> 0, and
(d¯ − α p)(p − c) only w.r.t. p. The solution to this op- PV 1
w̃V1 − w̃S − w̃M pa pa
> 0. (39)
1
timization problem (using derivative techniques) is,
If the above inequalities are satisfied, then one can
∗ d¯ c (d¯ − αc)2 choose by continuity a γ̄ε < 1 for each ε ≥ ε̄ such that
p = + , and w∗ = w(p∗ ) = . (34)
2α 2 4α the strict inequalities in (39) are now satisfied with
Returning to the original problem (33), if ∆ > 0, then, {wPC } in place of {w̃PC }, for all γ ≥ γ̄ε and for each ε ≥
ε̄. Clearly, for such (ε, γ), the configuration (PV 1 , x)
d¯ αc is stable when:
α p∗ =
p
+ < d¯ − αOc < d, ¯  
2 2 xM ∈ wMpa − wMPV 1 , wPV1V 1 − wPV2V 2 + wPMV 1 , (40)
1 2 2

and hence, (d¯−α p∗ )+ = d¯−α p∗ , w∗ −Oc > 0. Thus, and the above interval is non-empty.
p∗ is also the maximizer of (33) with U ∗ = w∗ − Oc . There exists an ε̄ < 1 such that the last two in-
If ∆ < 0 then no is the optimizer, as U ∗ = U(no ) = 0 > equalities of (39) are satisfied (see Table 1 and Ap-
w∗ − Oc . The last sentence now follows trivially. pendices 9 and 10 for the proof of the values in the
Proof continued, Theorem 3: Without loss of gen- table, which follows in similar lines as (31)).
erality, consider stability of PV 1 . To ensure (PV 1 , x) is Thus from (36), all three strict inequalities of (39)
stable, it should not be blocked by HC coalition M, as are definitely satisfied (if required for a larger ε̄),
well as VC coalition V2 . In other words, we require, when (30) is satisfied.
Instability: On the other hand, say (30) is negated
wMpa − wPMV 1 ≤ xM ≤ wPV1V 1 − wPV2V 2 + wPMV 1 ,
2 1 2
(35) with strict inequality. Then from (36) there exists an
ε̄ < 1 such that for all ε ≥ ε̄ the following is satisfied:
and this is because for any pay-off vector consistent
with PV 1 , we have xM = wPMV 1 and xM + xS = wPV1V 1
2 2 1
w̃PV1V 1 − w̃VP2V 2 + w̃PMV 1 < (w̃Mpa − w̃PMV 1 )
2 2

and wVpa2 = wVP2V 2 . Towards this, consider the limit2 As before there exists γ̄ε for each ε ≥ ε̄, and then for
any γ ≥ γ̄ε and ε we have:
P P P pa P
w̃V1V 1 − w̃V2V 2 + w̃MV2 1 − (w̃M − w̃MV2 1 )
wPV1V 1 − wPV2V 2 + wPMV 1 < (wMpa − wPMV 1 ).
1−ε 2 2

! ! For all such (ε, γ) there exists no pay-off division, to

−2dM1 dM2 + d¯M2 1 − d¯M2 2
¯ ¯ −2d¯M1 d¯M2 + d¯M2 2 − d¯M2 1
= − be more precise, no xM such that (35) is satisfied.
1
16α̃ 16α̃
Thus PV 1 is blocked for any configuration either by
d¯M2 2 d¯2 V2 or by HC.
+ o((1 − ε)) + 2 − M
16α̃ 16α̃ Now assume without loss of generality, d¯M ≥ d¯M . 1 2

d¯2 − d¯M2 2 d¯M2 2 d¯2 Then (30) can never be satisfied when roles of M2 and
= M1 + − M + o((1 − ε)) M1 interchanged and thus the partition PV 2 is never
8α̃ 8α̃ 16α̃
d¯M2 1 − d¯M2 2 − 2d¯M1 d¯M2 stable in ESM regime. However, as proved above,
= + o((1 − ε)) (36) PV 1 is stable with payoff vectors additionally satisfy-
16α̃
ing (40) when (30) is satisfied and unstable when (30)
2 It
is easy to observe that the derivative limit in case of is negated with strict inequality.
HC from Corollary 2 is equal to fM1,PH = −d¯M2 /16α̃ . Thus we have proved the theorem. □
8 Appendix – Generic Game
In this appendix, we solve the game between the manufacturers, where the supplier announces a price q and
the manufacturers respond via the game ⟨{M1 , M2 }, (W1 ,W2 ), (AM , AM )⟩. The result of this game will be used
1 2

extensively in the further proofs. Here, the utility of manufacturer i is,


+ 
Wi (ai , a−i ) = d¯i − αi pi + εα−i p−i (pi − ci ) − Oci Fai . (41)
We define the following function ∆i , which is a bound on demand, and will be
∆i (p) := d¯i + εα−i p − αi ci − 2 αi Oci as in Lemma 4.
p
(42)
Lemma 5. If a−i = no , then set p−i = 0. The best response (BR) of the agent i in Game (41) is unique when
∆i (p−i ) ̸= 0 and is given by:
a∗i (a−i ) = p∗i (p−i )1{∆i (p−i )>0} + no 1{∆i (p−i )<0} .


(43)
where,
εα−i p−i d¯i + ci αi
p∗i (p−i ) = pbi + with pbi = . (44)
2αi 2αi
Otherwise, a∗i (a−i ) = {p∗i (p−i ), no }, i.e., BR has two elements.
Proof. The utility function in (41) has a similar structure as the utility function given in (33), and thus the proof
follows from Lemma 4.
Definition 3 (Operating NE and Fully Operating NE). (i) A NE is called an Operating NE if the actions of at
least one of the agents in the NE differ from the non-operating action no .
(ii) A NE is called a Fully Operating NE if the actions of both the agents in the NE differ from no .
Lemma 6. [Inner Game between Manufacturers] We define the following action:
ε d¯−i + c−i α−i + 2 d¯i + ci αi



∗
pi := for any manufacturer Mi .
(4 − ε 2 )αi
Without loss of generality, consider ∆2 (p∗1 ) ≤ ∆1 (p∗2 ). Then the following is true for the game defined in (41).
(i) If ∆2 ((p∗1 ) ≥ 0, then (p∗1 , p∗2 ) is the unique fully operating NE. In fact, this is the unique NE if ∆2 (p∗1 ) > 0.
(ii) If ∆2 (p∗1 ) < 0 ≤ ∆1 (0), then (pb1 , no ) is the unique operating NE. In fact, this is the unique NE if ∆1 (0) > 0.
(iii) If ∆2 (p∗1 ) < 0 and ∆1 (0) < 0 then there is no operating NE. Here (no , no ) is the unique NE.
□
Proof. If there exists a fully operating NE ( p̃∗1 , p̃∗2 ), then by definition, the NE should simultaneously satisfy the
following two equations:
p̃∗i ( p̃∗−i ) = p̃∗i for both i = 1, 2.
That is, by Lemma 5, any such NE must satisfy
d¯i + εα−i p̃∗−i + ci αi
= p̃∗i for both i = 1, 2.
2αi
By linearity and by full rank condition, the above pair of equations have an unique solution given by (p∗1 , p∗2 )
(easy to verify by substitution). Thus there exists at maximum one fully operating NE – if there exists a fully
operating NE, then it has to be (p∗1 , p∗2 ) – no other fully operating pair of actions can form an NE.
Case 1: Consider the case, where ∆i (p∗−i ) ≥ 0 for both i = 1, 2. Using the above arguments and Lemma (5),
observe that (p∗1 , p∗2 ) is the unique fully operating NE.
Case 2: Consider the case, where ∆2 (p∗1 ) < 0 ≤ ∆1 (0). Thus by Lemma (5), the best response against p∗1 equals,
B2 (p∗1 ) = {no }. Thus, (p∗1 , p∗2 ) is not an NE. Again, by initial arguments of this Lemma, there exists no fully
operating NE.
By (42), the mapping p1 → ∆2 (p1 ) is strictly increasing. Thus, using ∆2 (p∗1 ) < 0 and Lemma 5,
B2 (p1 ) = {no } ∀p1 ≤ p∗1 . (45)
Similarly, as ∆1 (0) ≥ 0 and the mapping p2 7→ ∆1 (p2 ) is strictly increasing, we have ∆1 (p2 ) > 0 for all p2 > 0.
Thus, again by Lemma 5, we have
/ B1 (p2 ), and further, B1 (p2 ) = {p∗1 (p2 )} ∀ p2 > 0.
no ∈
Further, using similar logic and (44), for p2 > p∗2 , we have p∗1 (p2 ) > p∗1 = p∗1 (p∗2 ). Thus, by initial arguments of
this lemma, any pair (p1 , p2 ) ̸= (p∗1 , p∗2 ) with p1 > 0, p2 > 0 is not a NE. Now, consider the remaining case, where
p2 < p∗2 . For all such p2 , by strict monotonicity of p2 7→ p∗1 (p2 ), we have p∗1 (p2 ) < p∗1 , and thus by equation (45),
we have
B2 (p∗1 (p2 )) = {no }.
Thus, in all, any p2 > 0 i.e., any a2 ̸= no is not NE strategy for manufacturer 2.
As ∆1 (0) ≥ 0, we have
pb1 ∈ B1 (no ) and B2 (pb1 ) = {no } as pb1 < p∗1 .
Thus we have that (pb1 , no ) is an NE and this the only operating NE. Further, observe that when ∆1 (0) > 0, this is
the unique NE.
Case 3: Consider the case where ∆2 (p∗1 ) < 0 and ∆1 (0) < 0. By direct substitution (substituting p = 0 in (42))
(by monotonicity, we also have ∆2 (0) < 0) one can verify that (no , no ) is NE.
Similar to case 2 above, for all p1 ≤ p∗1 , we have ∆2 (p1 ) < 0, and hence the corresponding B2 (p1 ) = {no }.
Further, again similar to case 2 above, p1 > p∗1 cannot be a NE. However, observe that B1 (no ) = {no }, thus there
is no operating NE.

9 Appendix – Vertical Cooperation Partition

Proof of Lemma 2. Without loss of generality, we will prove the lemma for the partition PV 1 = {V1 , M2 }, the
case with PV 2 = {V2 , M2 } follows because of similar arguments. Firstly, we will simplify the utilities of different
agents/coalition in (24), (25), and then obtain the best response of M2 (i.e., p∗2 (p1 , q)) given the actions p1 , q of
coalition V1 . Then we obtain the optimizers, p∗1 and q∗ assuming αM = αM = α using derivative based argumets.
1 2

After obtaining these relevant quantities, and checking the operating condition for M2 and the coalition V1 , we
find the worth limits and derivative limits of coalition V1 and the agent M2 for the ESM regime. Now, we begin
with the detailed proof.
The utility of the vertical coalition V1 , as defined in (24), is modified as follows, by substituting the demand
DM (aM ) from (1),
i

 
UV1 (aV1 ; aM ) = DM (aM )(p1 −CM −CS ) + DM (aM )FM (q −CS ) − OS − OM FV1 ,
2 1 1 2 2
(46)1

 
= (d¯M − αM p1 + εαM p2 )(p1 −CM −CS ) + (d¯M − αM p2 + εαM p1 )FM (q −CS ) − OS − OM FV1 .
1 1 2 1 2 2 1 2 1

(47)
Using (25), the utility of the manufacturer M2 for a given aV is,
 
UM (aM ; aV ) = DM (aM )(p2 −CM − q)FV1 − OM FM .
2 2 2 2 2 2
(48)

Since M2 is follower, for any given (q, p1 ) the optimizer of UM using Lemma 4 is, 2

 ¯ 
∗ ∗ dM + εαM p1 CM + q
2 1 2
aM (p1 , q) = p2 (p1 , q) = 1{q≤θ2 (p1 )} + + no 1{q>θ2 (p1 )} , (49)
2
2αM 2 2

where √
d¯M + εαM p1 − αM CM − 2 αM OM
2 1 2 2 2 2
θ2 (p1 ) = . (50)
αM 2
For all the following calculations in this section, we assume αM = αM = α. Define 1 2

d¯M 2
CM 2
e1 := + ,
2α 2
 ¯
ε2
  
¯ dM CM CS εα
e2 := α 1 − (CM +CS ) + dM + εα 1
+ − , 1
2 2

2 2α 2 2
d¯M CM εα α
e3 := −α
2
− (CM +CS )
2
+CS .
2 2 2 2
If q < θ2 (p1 ) then from Lemma 4, we have p∗2 (p1 , q) = q/2 + p1 ε/2 + e1 . Since coalition V1 is the leader, and
M2 is the follower, as in proof of Theorem 1, substituting aM = p∗2 (p1 , q) into (46) (by neglecting the operating
2

conditions), consider
ε2
   
¯ εαq
UV1 (p1 , q) := UV1 (p1 , q; p2 (p1 , q)) = dM − α p1 1 − + εαe1 +
1
(p1 −CM −CS ) 1
2 2
 
α p1 ε − αq
+ d¯M − αe1 + (q −CS ) − OS − OM .
2 1
2
The first step of this analysis is to derive the optimizers of the function UV . Later we show that the operating
conditions are satisfied at these optimizers (as in proof of Theorem 1). Now, differentiating UV1 with respect to
p1 and q, we get
ε2 ε2
   
dUV1 ¯ εαq (q −CS )εα
= −α 1 − (p1 −CM −CS ) + dM − α p1 1 −
1
+ εαe1 + 1
+ ,
d p1 2 2 2 2
= −α(2 − ε 2 )p1 + αεq + e2 ,
dUV1 (p1 −CM −CS )εα α p1 ε − αq (q −CS )α
= 1
+ d¯M − αe1 + − 2
= αε p1 − αq + e3 .
dq 2 2 2
Observe that the Hessian matrix H(p1 , q) of second order partial derivatives of UV1 is always positive for any
(p1 , q), and the diagonal entries of the Hessian matrix are always negative for any (p1 , q). Solving the simultane-
ous equations dUV1/d p1 = 0 and dUV1/dq = 0, we get the unique extreme p∗1 and q∗ , where
e2 + e3 ε e3 + αε p∗1
p∗1 = and q ∗
= . (51)
α(2 − 2ε 2 ) α
Since these optimizers are unique, using the second derivative test, we conclude that p∗1 and q∗ are the global
optimizers.
Now consider the ESM regime, where (ε, γ) → (1, 1). Now consider the following limits, which will be used
in the further calculations,
d¯M d¯M d¯M
lim lim αe1 = 2
, lim lim e2 = + d¯M , and lim lim e3 = 2
(52)
1
2

ε→1 γ→1 2 ε→1 γ→1 2 ε→1 γ→1 2

The demand DM (aM ) of coalition V1 in the limit at SBE is (see (46)):
1

lim lim DM (aM ) = lim lim d¯M − α p∗1 + εα p∗2 (p∗1 , q∗ )

1 1
ε→1 γ→1 ε→1 γ→1
 εe3  d¯M d¯M
= lim lim d¯M − α p∗1 (1 − ε 2 ) + εαe1 +
1
= + . 1 2
(53)
ε→1 γ→1 2 2 4
The demand DM (aM ) of M2 in the limit at SBE is (see (48)):
2

lim lim DM (aM ) = lim lim d¯M − α p∗2 (p∗1 , q∗ ) + εα p∗1

2 1
ε→1 γ→1 ε→1 γ→1

q∗ d¯M
 
ε
= lim lim d¯M − αe1 + α p∗1 − α
2
= . 2
(54)
ε→1 γ→1 2 2 4
To ensure that at limit q∗ ≤ θ2 (p∗1 ), using (50), (51), and (52),
d¯M
lim lim θ2 (p∗1 ) − q∗ = d¯M + εα lim lim p∗1 − α lim lim q∗ = d¯M − e3 =
2 2
> 0.2

ε→1 γ→1 ε→1 γ→1 ε→1 γ→1 2

Therefore, there exists a neighbourhood of (γ, ε) near (1, 1) such that q∗ < θ2 (p∗1 ). Now, using (49), (51), (52),
we get (recall that α = α̃(1 − γ))
(1 − γ)(1 − ε)(−q∗ + p∗1 ε + 2e1 )
lim lim (1 − γ)(1 − ε)(p∗2 (p∗1 , q∗ ) − q∗ ) = lim lim = 0, and
ε→1 γ→1 ε→1 γ→1 2
(1 − γ)(1 − ε)(q∗ + p∗1 ε + 2e1 ) 1 ¯
lim lim (1 − γ)(1 − ε)p∗2 (p∗1 , q∗ ) = lim lim = (dM + d¯M ). 1 2
ε→1 γ→1 ε→1 γ→1 2 4α̃
Finally, substituting the values of p∗1 , q∗ , p∗2 (p∗1 , q∗ ) from (51), (49) into the utility of a coalition V1 (see (46))
and manufacturer M2 (see (48)) at SBE in limits are:
1 ¯
2 d¯2
fVP1V 1 = lim (1 − γ)(1 − ε)wPV V = dM + d¯M = M , 1 2
(ε,γ)→(1,1) 8α̃ 8α̃
fMPV 1
2
= lim (1 − γ)(1 − ε)wPMV = 0. 2
(ε,γ)→(1,1)

Again, by substituting the values of p∗1 , q∗ , p∗2 (p∗1 , q∗ ) from (51), (49) into the utility of a coalition V1 (see (46))
and manufacturer M2 (see (48)) at γ → 1, we get,
 ¯
ε d¯M
 ¯
dM + d¯M ε
  ¯  ¯
dM + ε d¯M

PV 1 dM dM
lim (1 − γ)wV1 = + + 1 2
(55) 1 2 2 2 1

γ→1 2 4 2α̃(1 − ε 2 ) 4 2(1 − ε 2 )α̃

d¯M2
lim (1 − γ)wPMV = 2
(56)
γ→1 2
16α̃
Thus after scaling the equations (55) and (56) by (1 − ε) and differentiating the scaled equations with respect to
ε and then taking the limit ε → 1, we get,
2d¯M d¯M + d¯M2 − d¯M2 (1),PV 1 1 2
f V1 . = 2 1

16α̃
(1),P d¯2
fM V 1 = − M . 2
2

16α̃
Along the similar lines, we can derive the expressions of worth-limits and the derivative-limits for the other
partition PV 2 which completes the proof.

10 Appendix – All Alone Case Partition

Proof of Lemma 3 . In this partition, the supplier quotes a price q, and the manufacturers respond to it via an
inner game between them. Initially, we find the inner game equilirium for given supplier price q, and then we
find the optimal action of the supplier by optimizing it’s utility. Towards this, we define the following quantities,
which will be extensively used later in the proof.
√
d¯M − αM CM − 2 αM OM
σia := i
, i i i i
(57)
αM i
√
b 4d¯M + 2ε d¯M + 2εαM CM + 2(ε 2 − 2)αM CM − 2(4 − ε 2 ) αM OM
−i −i −i
σi := i
, (58) i i i i

2((2 − ε 2 )αM − εαM ) i −i

d¯M CM + q
pbi (q) := + ,i i
(59)
2αM 2 i

ε d¯M + (CM + q)αM + 2 d¯M + (CM + q)αM

p∗i (q) :=
−i −i −i i i i
, (60)
(4 − ε 2 )αM i
p
∆M (p; q) := d¯M + εαM p − αM (CM + q) − 2 αM OM .
i i −i i i
(61) i i

Here, σia and σib

are different thresholds on action of supplier, based on which manufacturers will decide whether
to operate or not. Observe that the utilities of the inner game between the manufacturers, given by (27), are of
similar form as utilities of the generic game considered in (41). Therefore, Lemma 6 is applicable. Observe from
(57), (61) that ∆M (0; q) < 0 is equivalent to q > σia ; and from (58), (61) that ∆M (p∗i (q); q) < 0 is equivalent to
i i

q > σib . Thus by Lemma 6, one can derive the NE of the inner-game (p∗1 (q), p∗2 (q)) for any supplier action
aS = q (which is not equal to no ) when both the manufacturers operate and (pbi (q), no ) when only manufacturer i
operates; after substituting this NE into (26), we get the utility of supplier as below (assume w.l.o.g that σ1b ≥ σ2b ):


 0 if aS = no
−O if aS ̸= no and q > max{σ1a , σ2b }


 ¯ S

US (aS ) = dM −αM (CM +q) (62)
  2 (q −CS ) − OS
1 1 1
if aS ̸= no and σ2b < q ≤ σ1a
 (d¯M +d¯M +(ε−1)(αM CM +αM CM )−(1−ε)(αM +αM )q (q −CS ) − OS if aS ̸= no and q ≤ σ b .

 
 1 2 1 1 2 2 2 1
(2−ε) 2

Observe here that the case when only the first manufacturer operates is given by third row of (62) and the case
when both manufacturers operate is given by the fourth row of (62).
Initially, we need to find the optimizer q∗ of the utility function US , and then the optimal utilities of the supplier
and the manufacturers to compare with other cases. Specifically, we find these quantities in the ESM regime, as
(γ, ε) → (1, 1). As before, we assume that αM = αM = α and also recall α = α̃(1 − γ). Towards optimizing (62),
1 2

we first need to analyse {σib , σia }; by (57) and (58), the corresponding limits in the ESM regime are:
lim (1 − γ)(1 − ε)σia = 0, , (63)
(ε,γ)→(1,1)

2d¯M + d¯M
lim (1 − γ)(1 − ε)σib = i
. −i
(64)
(ε,γ)→(1,1) 3α̃
Observe that in the ESM regime, at limit, when d¯M ≥ d¯M , we have at the limit, 0 = σ1a < σ2b < σ1b . Now, using
1 2

this strict inequality and continuity of functions σ1a , σ1b , and σ2b with respect to (ε, γ), the same strict inequality
is satisfied for all (ε, γ) in a neighbourhood of (1, 1). For all such (ε, γ), after substituting the inner-game NE
(p∗1 (q), p∗2 (q)) from (60), the utility of supplier as given in (62) modifies as follows:

0
 if aS = no
−O
US (aS ) =  S if aS ̸= no and q > σ2b (65)
 d¯M +d¯M +(ε−1)α(CM +CM )−(1−ε)2αq (q −C ) − O if a ̸= n and q ≤ σ b .
 
1 2 1 2
(2−ε) S S S o 2

For any such (γ, ε), using Lemma 4, we get the optimizer of US as given below:
( )
d¯M + d¯M + (ε − 1)α(CM +CM )


∗ CS b
1 2 1 2
q = min + , σ2
4α(1 − ε) 2
Using simple algebra, and (64), observe that
!
d¯M + d¯M + (ε − 1)α(CM +CM ) d¯M + d¯M


1 2 CS 1 2
lim (1 − γ)(1 − ε) + = , and 1 2

(ε,γ)→1 4α(1 − ε) 2 4α̃

2d¯M + d¯M
lim (1 − γ)(1 − ε)σ2b = 2
. 1

(ε,γ)→(1,1) 3α̃
Thus as (d¯M +d¯M )/4α̃ < (2d¯M +d¯M )/3α̃ , we have that in the neighbhourhood of (ε, γ) → (1, 1), the optimizer of
1 2 2 1

function in (65) is,

d¯M + d¯M + (ε − 1)α(CM +CM )


∗ 1 CS 2 1 2
q = + (66)
4α(1 − ε) 2
Thus after substituting this q∗ in (65) and then taking the limits after scaling, we get
d¯2
lim (1 − γ)(1 − ε)US (a∗S ) = M (67)
(ε,γ)→(1,1) 8α̃
From (27), that the optimal utility of the manufacturer i is given by:
UM (a∗M , a∗S ) = (d¯M + εαM p∗−i (q∗ ) − αM p∗i (q∗ ))(p∗i (q∗ ) − q∗ −CM )FS − OM FM


i i −i i i i i
(68)
Also observe that after substituting q∗ from (66) in (60), and then taking the limits we get that
lim (1 − γ)(1 − ε)(p∗i (q∗ ) − q∗ ) → 0 ∀i
(ε,γ)→(1,1)

Thus we have,
lim (1 − γ)(1 − ε)UM (a∗M , a∗S ) = 0
i
∀i. (69)
(ε,γ)→(1,1)

Thus from (67)and (69), the ESM limit of worths of coalition ALC are,
d¯M2
fSPA = lim (1 − γ)(1 − ε)wPS A =
(ε,γ)→(1,1) 8α̃
PA
fM =
i
lim (1 − γ)(1 − ε)wPMA = 0 ∀i.
i
(ε,γ)→(1,1)

Observe that after substituting q∗ from m (66) in (60), and then substituting q∗ and the obtained p∗i (q∗ ) in (65)
and (68) we get
d¯M2 ((6 − ε)d¯M + (3ε − 2)d¯M )2
lim (1 − γ)wPS A = , and lim (1 − γ)wPMA = i −i
for all i . (70)
γ→1 8α̃(1 − ε)(2 − ε) γ→1 i
16(4 − ε 2 )2 α̃
Thus after scaling the worths in equation (70) by (1 − ε) and differentiating the scaled worths with respect to ε
and then taking the limit ε → 1, we get,

(1),PA d¯M2
fS = .
8α̃
(1),PA −(5d¯M + d¯M )2 −i
fM i
= i
.
144α̃
This completes the proof.
‘