GAMES AND ECONOMIC BEHAVIOR 14, 90–123 (1996)

ARTICLE NO. 0043

Sequential Formation of Coalitions in Games with

Externalities and Fixed Payoff Division∗
Francis Bloch

Department of Economics, Brown University, Providence, Rhode Island 02912; and Department of
Finance and Economics, Groupe HEC, 78351 Jouy-en-Josas, France.

Received May 19, 1993

This paper analyzes a sequential game of coalition formation when the division of
the coalitional surplus is fixed and the payoffs are defined relative to the whole coalition
structure. Gains from cooperation are represented by a valuation which maps coalition
structures into payoff vectors. I show that any core stable coalition structure can be attained
as a stationary perfect equilibrium of the game. If stationary perfect equilibria may fail to
exist in general games, a simple condition is provided under which they exist in symmetric
games. Furthermore, symmetric stationary perfect equilibria of symmetric games generate a
coalition structure which is generically unique up to a permutation of the players. A general
method for the characterization of equilibria in symmetric games is proposed and applied to
the formation of cartels in oligopolies and coalitions in symmetric majority games. Journal
of Economic Literature Classification Numbers : C78, C71. © 1996 Academic Press, Inc.

1. INTRODUCTION

Since the publication of Theory of Games and Economic Behavior, the study
of endogenous formation of coalitions has been one of the most intriguing and
challenging problems open to game theorists. Many solution concepts such as
Von Neumann and Morgenstern’s stable sets (Von Neumann and Morgenstern,
1944) and Aumann and Maschler’s bargaining set (Aumann and Maschler, 1964)

∗ This paper is based on the second chapter of my Ph.D. dissertation at the University of Pennsylvania.

I am grateful to my advisor, Beth Allen, for her help and encouragement. Discussions with Elaine
Bennett, Ken Binmore, Roger Lagunoff, George Mailath, Andrew Postlewaite, Ariel Rubinstein and
Sang-Seung Yi were extremely valuable. I am particularly indebted to Debraj Ray and Rajiv Vohra for
pointing out an error in an earlier version of this paper. I am also grateful to participants at seminars
at Boston University, Brown, Columbia, Stony Brook and the European Meeting of the Econometric
Society in Brussels for their comments. Detailed comments by an associate editor greatly improved
the quality of the paper. This work was partly supported by a Sloan Foundation Doctoral Dissertation
Fellowship.

90
0899-8256/96 $18.00
Copyright © 1996 by Academic Press, Inc.
All rights of reproduction in any form reserved.
 SEQUENTIAL FORMATION OF COALITIONS 91

were in fact primarily designed as ways to solve the problem of joint determina-
tion of a coalition structure and the allocation of the coalitional surplus among
coalition members. While these approaches proved fruitful in the study of many
situations of cooperation, they mostly rely on the assumption that gains from
cooperation can be defined independently of the coalitions formed by external
players.1 Using the terminology introduced by Shubik (1982), cooperative game
theory has focused on games with orthogonal coalitions which are well-suited
to situations of pure competition but fail to capture the effects of externalities
among coalitions. The objective of this paper is to propose a model of forma-
tion of coalitions in nonorthogonal games where payoffs depend on the whole
coalition structure.
The presence of externalities among coalitions introduces a new difficulty in
the study of endogenous coalition formation. When players decide to form a
coalition, they must take into account the reaction of external players to the for-
mation of the coalition. The sequential model analyzed in this paper addresses
this problem by explictly describing a procedure in which individual players,
when deciding to form a coalition, consider the consequences of their actions
on the behavior of the other players. However, to keep the analysis tractable and
concentrate on the role played by externalities on the formation of the coalition
structure, I do not model the allocation of the coalitional surplus among mem-
bers of a coalition, and assume instead that the coalitional worth is distributed
according to a fixed sharing rule. Gains from cooperation are then represented
by a valuation which maps coalition structures into vectors of individual payoffs.
Arguably, the assumption that payoffs are determined by a fixed rule is very
restrictive and may seem a high price to pay for allowing externalities among
coalitions. But valuations arise naturally in two distinct categories of economic
models and the study of coalition formation in games represented by a valuation
may appear fruitful in the resolution of these models.
First, valuations are considered in the models of coalition formation studied
by Myerson (1978), Shenoy (1979), Hart and Kurz (1983) and Aumann and
Myerson (1988). In these models, the formation of coalitions is viewed as a
two-stage process where players form coalition in the first stage and decide
on the allocation of the coalitional surplus, given a fixed coalition structure,
in the second stage. Hence, at the time coalitions are formed, players evaluate
the payoffs they receive in each coalition structure according to a fixed rule.
The exact characterization of the rule employed in the second stage depends on
the situations considered in the different models. In Myerson (1978)’s threats
and settlement game, the fair settlement function assigns to each collection
of coalitions (not necessarily a coalition structure) a unique vector of payoffs.
Shenoy (1979) uses as an evaluation rule Aumann and Drèze (1974)’s extension

1 Two important exceptions are Thrall and Lucas (1963)’s study of games in partition function form

and Aumann and Drèze (1974)’s analysis of games with fixed coalition structures.
92 FRANCIS BLOCH

of the Shapley Value to games with fixed coalition structures. In Hart and Kurz
(1983)’s analysis, players evaluate coalition structures according to a different
extension of the Shapley Value first analyzed by Owen (1977). In Aumann and
Myerson (1988)’s study of formation of links among players, the valuation used
is Myerson (1977)’s extension of the Shapley Value to games with cooperation
graphs of players.2
Second, valuations emerge in various applications of Game Theory to In-
dustrial Organization and Public Economics involving competing coalitions of
economic agents. The study of the formation of cartels in oligopolies leads to a
natural definition of a valuation representing, for each cartel structure, the payoffs
obtained by the firms belonging to the different cartels.3 Similarly, the formation
of associations of firms which agree to share some common resource but behave
as competitors on the market can be analyzed with the use of a valuation.4 The
analysis of the provision of local public goods in a spatial setting where members
of a community can benefit from the public goods provided in neighboring com-
munities also requires the use of a valuation.5 As a final example, the formation
of customs unions allowing national firms to compete in a market characterized
by the existence of different customs unions also leads to the definition of a
valuation.
Cooperative solution concepts for games represented by a valuation were
introduced by Shenoy (1979) and Hart and Kurz (1983) in their models of en-
dogenous coalition formation.6 To predict which coalitions will be formed, they
propose different definitions of stability of coalition structures.7 The variety of
stability concepts accounts for the fact that, in games described by a valuation,
the payoffs obtained by members of a blocking coalition depend on the reac-
tion of the external players. The solution concepts range from the core stability
concept, which supposes a very optimistic conjecture about the reaction of the
external players since players deviate if there exists a coalition structure in which
they are better off to the α stability concept which is based on pessimistic con-
jectures since a coalition only deviates when it is guaranteed to obtain a higher

2 In Myerson (1978) and Hart and Kurz (1983), the emphasis is put on the axiomatic derivation of a
reasonable valuation rather than on the first stage game of coalition formation. This paper, by contrast,
focuses on the game of coalition formation.
3 Salant et al. (1983) were the first to point out in a simple model the problems of cartel formation in
oligopolies. Yi and Shin (1995) contains a very complete description of the derivation of the valuation
in the cartel problem.
4 The study of associations of firms, which can be interpreted as Research Joint Ventures or stan-

dardization committees, is taken up in a distinct paper (Bloch, 1995).

5 Guesnerie and Oddou (1981) analyze the provision of local public goods in a model with orthogonal

coalitions but discuss the role of externalities among communities.

6 Hart and Kurz (1983) analyze strong equilibria of a noncooperative game where players simulta-

neously announce coalitions.

7 Other concepts of stability of coalition structures are surveyed in Greenberg (1995).
 SEQUENTIAL FORMATION OF COALITIONS 93

payoff independently of the reaction of the other players. The study of stable
coalition structures raises three important difficulties. First, the definitions of
stability rely on ad hoc assumptions on the behavior of the other players after a
coalition has deviated. Second, all definitions of stability assume that external
players react to the formation of a coalition in a myopic way. Hence, when a
coalition forms, its members do not take into account the final result of their
decisions but only the immediate reaction of the other players. Finally, even the
less restrictive definition of stability (α stability) may not be useful, since α stable
coalition structures fail to exist in situations which are not easily characterized.
(Hart and Kurz (1984) give an example of a game without stable structure which
is otherwise well-behaved.)
By contrast, in this paper, I explicitly model the formation of coalitions as a
noncooperative sequential process in the spirit of Rubinstein (1982)’s alternating
offers bargaining game and its extensions to n players by Selten (1981) and
Chatterjee et al. (1993). Players are ranked according to an exogenous rule of
order. The first player starts the game by proposing the formation of a coalition.
If all prospective members accept the proposal, the coalition is formed. If one
player rejects the proposal, she becomes the initiator in the next round. The
important feature of the game is that, once a coalition is formed, the game is
only played among the remaining players and that established coalitions may
not seek to attract new members nor break apart. Hence, by agreeing to group
in a coalition, players commit to stay in that coalition.
I restrict my attention to stationary strategies and establish the following prop-
erties of stationary perfect equilibria. I first show that, if the game always admits
a subgame perfect equilibrium, stationary perfect equilibria may fail to exist. A
sufficient condition for the game to admit a stationary perfect equilibrium is that
the valuation and all its restrictions to smaller sets of players admit core stable
structures. Furthermore, any core stable coalition structure can be reached as a
stationary perfect equilibrium of the extensive form game of coalition formation,
provided that the set of stationary perfect equilibria is nonempty. I then study
the restricted class of symmetric games where all players are ex ante identical.
In this class of games, using a result due to Ray and Vohra (1995), I provide a
simple condition under which symmetric stationary perfect equilibria exist, and
I show that the coalition structures they generate are generically unique up to
a permutation of the players. Furthermore, I provide a general method for the
characterization of the coalition structures generated by symmetric stationary
perfect equilibria in symmetric games. This method is used to derive equilib-
rium coalition structures in two situations: the formation of cartels in a symmetric
oligopoly and the symmetric majority games discussed by Hart and Kurz (1984).
The game analyzed here is similar to games of coalition formation proposed by
Selten (1981), Chatterjee et al. (1993), Moldovanu (1992) and Winter (1993) in
the context of games in coalitional form. The games they analyze have the same
sequence of moves as the one described above. The crucial difference between
94 FRANCIS BLOCH

their games and mine stems from differences in the action spaces. By fixing the
division of the payoffs, I restrict the actions of the agents to announcements of
coalitions whereas they study a more general framework where players announce
both a coalition and the division of the coalitional worth. A further difference
is due to the underlying specification of gains from cooperation since they do
not allow for externalities among coalitions. Given these differences, the results
they obtain are not directly comparable to mine.
Different extensive form procedures of coalition formation in games repre-
sented by a valuation were proposed by Aumann and Myerson (1988) and Shin
and Yi (1995). The procedure in Aumann and Myerson (1988) is defined for
games where players evaluate cooperation graphs rather than coalition struc-
tures. The particular feature of cooperation graphs where coalition members
need not unanimously agree to admit new members leads them to define a game
where links can be formed at any stage. This approach cannot easily be applied
to situations where gains from cooperation accrue when coalitions are formed,
rather than bilateral links among players. Yi and Shin (1995) analyze games
based on a “matching procedure.” Players announce coalitions and coalitions
are formed whenever all its members have made identical announcements. In
general, the equilibria they obtain are very different from the equilibria of the
infinite horizon game analyzed in this paper.
The paper is organized as follows. The game of sequential formation of coali-
tions is introduced and the equilibrium concept defined in Section 2. In Section 3,
I analyze the relations between stationary perfect equilibria and stability con-
cepts for coalition structures in games described by a valuation. Section 4 is
devoted to the analysis of symmetric games. I present applications of the model
to the formation of cartels in oligopolies and of coalitions in symmetric major-
ity games in Section 5. My concluding remarks and some directions for future
research appear in Section 6.

2. SEQUENTIAL FORMATION OF COALITIONS

In this section, I introduce the sequential game of coalition formation and

the equilibrium concept that I will use. The set of players is denoted N , with
cardinality n. The index i will refer to the players. A coalition T is a nonempty
subset of players. A coalition structure π is a partition on the set N . The set
of all coalition structures is denoted by 5. For any subset K of N , the set of
partitions on K is denoted 5 K with typical element π K .
Gains from cooperation are described by a valuation v which maps the set
of coalition structures 5 into vectors of payoffs in <n . The component vi (π )
denotes the payoff obtained by player i if the coalition structure π is formed.
I assume that payoffs are normalized so that any player, by opting to leave the
game can get a strictly positive payoff. Formally, ∀i ∈ N , minπ ⊃{{i}} vi (π ) > 0.
 SEQUENTIAL FORMATION OF COALITIONS 95

A rule of order ρ is an ordering of the players, which is used to determine the

order of moves in the sequential game of coalition formation.
The sequential game of coalition formation is defined by the exogenous specifi-
cation of the valuation v and of the rule of order ρ. To emphasize this dependence,
I denote the game of coalition formation by 0(v, ρ).
The game 0(v, ρ) proceeds as follows. The first player according to the rule
of order ρ starts the game by proposing the formation of a coalition T to which
she belongs. Each prospective member responds to the proposal in the order
determined by ρ. If one of the player rejects the proposal, she must make a
counteroffer and propose a coalition T 0 to which she belongs. If all members
accept, the coalition is formed. All members of T then withdraw from the game,
and the first player in N \ T starts making a proposal.8
This game describes in the simplest way a procedure where coalitions are
formed in sequence. The main characteristic of the game is that, once a coalition
has been formed, the game is only played among the remaining players. The
extensive form thus embodies a high degree of commitment of the players. When
players agree to join a coalition, they are bound to remain in that coalition. They
can neither leave the coalition nor propose to change the coalition at later stages
of the game. Figure 1 depicts the extensive form of the game with three players.
A history h t at date t is a list of offers, acceptances and rejections up to period
t. At any point in the game 0(ρ, v), a history h t determines
• a set K̂ (h t ) of players who have already formed coalitions
• a coalition structure π K̂ (h t ) formed by the players in K̂ (h t )
• an ongoing proposal (if any) T̂ (h t )
• a set of players who have already accepted the proposal
• a player who moves at period t.
Player i is called active at history h t is it is her turn to move after the history
h . The set of histories at which player i is active is denoted Hi .
t

A strategy σi for player i is a mapping from Hi to her set of actions, namely

σi (h t ) ∈ {Yes, No} if T̂ (h t ) 6= ∅

σi (h t ) ∈ {T ⊂ N \ K̂ (h t ), i ∈ T } if T̂ (h t ) = ∅.

When T̂ (h t ) 6= ∅, player i is a respondent to the offer T̂ (h t ) and she can

choose to accept or reject it. If T̂ (h t ) = ∅, either a coalition has just formed and
player i is the first player in N \ K̂ (h t ) according to the rule of order ρ, or player

8 Each time a coalition T is proposed, the order of responses is fixed by ρ independently of the

history or the identity of the proposer. Hence, for example, if player 2 proposes the formation of a
coalition {1, 2, 3}, player 1 responds first and player 3 responds after player 1.
96 FRANCIS BLOCH

FIG. 1. The game 0.

i has just rejected an offer. In both cases, it is her turn to propose a new coalition
T which must be a subset of the remaining players to which she belongs.
I restrict my attention to strategies which only depend on the payoff-relevant
part of the history. For a player i active at history h t , the only payoff-relevant
features of the history are the set K of players who left the game, the partition
π K representing the coalitions they formed and the current offer T . In particular,
the set of players who have already accepted the offer T is uniquely determined
by the rule of order ρ.
A strategy σi is stationary if it only depends on the state s = (K , π K , T )
where K is a (possibly empty) subset of N , π K is a partition of K and T is a
(possibly empty) subset of N \ K . Formally, letting T (i, K ) define the collection
of subsets of N \ K to which player i belongs, a stationary strategy is a mapping
from the set of states at which player i is active, Si , to a set of actions, where

σi (K , π K , T ) ∈ {Yes, No} if T 6= ∅

σi (K , π K , ∅) ∈ T (i, K ).

Any strategy profile σ = {σi }i∈N determines an outcome (π(σ ), t (σ )) of the

game. If the game ends in a finite number of periods, π(σ ) is a coalition structure
on the set N , and t (σ ) is the period at which the agreement has been reached.
 SEQUENTIAL FORMATION OF COALITIONS 97

I assume that players do not discount the future. In the case of an infinite play
of the game, players who have not formed a coalition receive a payoff of zero.
More precisely, suppose that a subset N \ K of the players does not reach an
agreement in a finite number of periods. Payoffs are then given by
vi (π(σ )) = 0 for all players in N\K

vi (π(σ )) = maxπK ⊂π vi (π ) for all players in K.

DEFINITION 2.1. A subgame perfect equilibrium σ ∗ is a strategy profile such
that ∀i ∈ N , ∀h t ∈ Hi , ∀σi , vi (π(σi∗ (h t ), σ−i
∗ ∗
)) ≥ vi (π(σi (h t ), σ−i )).
DEFINITION 2.2. A stationary perfect equilibrium σ ∗ is a subgame perfect
equilibrium where ∀i ∈ N , σi∗ is a stationary strategy.
A coalition structure π generated by a subgame perfect equilibrium is called
an equilibrium coalition structure (ECS). Coalition structures generated by
stationary perfect equilibria are called stationary equilibrium coalition struc-
tures (SECS). The set of stationary equilibrium coalition structures is denoted
S EC S(v, ρ)
Remark 2.3. Since every player obtains a higher payoff by leaving the game
than by disagreeing forever, an infinite play of the game cannot be part of a
subgame perfect equilibrium. Hence, the concept of an equilibrium coalition
structure is well defined.
The payoffs of the game described above are not continuous at infinity. Hence
the existence of a subgame perfect equilibrium is not guaranteed. To circumvent
this difficulty, I first show that any subgame perfect equilibrium of the game
with sufficiently high discounting is a subgame perfect equilibrium of the game
0(v, ρ). To be more precise, let 0δ (v, ρ) denote the game where strategies and
moves are defined as above but payoffs are given by: vi (σ ) = δit (σ ) vi (π(σ )).
PROPOSITION 2.4. There exists δ ∈ (0, 1) such that, if ∀i, δi > δ, any sub-
game perfect equilibrium of 0δ (v, ρ) is a subgame perfect equilibrium of 0(v, ρ).
Proof. Observe first that, since 5 is finite, the set of payoffs of the game,
v(5) is finite. Hence, the set of possible coalition structures formed in 0δ (v, ρ)
is finite. In particular, this implies that, as δ varies continuously from 0 to 1,
the strategy profiles of the game can only lead to a finite number of coalition
structures. Hence, there exists a δ such that, for all δ, δ 0 > δ, if σ ∗ is a subgame
perfect equilibrium of 0δ (v, ρ), then σ ∗ is a subgame perfect equilibrium of
0δ0 (v, ρ).
Consider now δ 0 > δ, and let σ ∗ be a subgame perfect equilibrium of 0δ0 (v, ρ).
Then, for any player i, any history h t in Hi , any strategy σi and any δ ∈ [δ 0 , 1),
t (σi∗ (h t ),σ−i
∗
) ∗
t (σi (h t ),σ−i )
δi vi (π(σi∗ , σ−i
∗
)) ≥ δi ∗
vi (π(σi , σ−i )).
98 FRANCIS BLOCH

Taking limits as δ goes to 1,

vi (π(σi∗ (h t ), σ−i
∗ ∗
)) ≥ vi (π(σi (h t ), σ−i )).

Hence, σ ∗ is a subgame perfect equilibrium of 0(v, ρ).

COROLLARY 2.5. For any valuation v and any rule of order ρ, there exists a
subgame perfect equilibrium of the game 0(v, ρ).

Proof. Fix a δ > δ. The game 0δ (v, ρ) is a finite action game of perfect
information and is continuous at infinity. Hence, by a result of Fudenberg and
Levine (1983) (Corollary 4.2, p. 262), the game 0δ (v, ρ) has a subgame perfect
equilibrium. From Proposition 2.4, any subgame perfect equilibrium of 0δ (v, ρ)
is a subgame perfect equilibrium of 0(v, ρ).

By imposing stationarity, I require that strategies only depend on the payoff-

relevant part of the history. In the framework analyzed here, the payoff-relevant
part of the history is summarized by the state s characterizing the coalition
structure formed by the previous players and the ongoing offer. Chatterjee et al.
(1993) and Moldovanu (1992) show that, when players bargain over the division
of the coalitional worth, the set of nonstationary perfect equilibria may be very
large, and stationarity is a useful restriction to refine the set of subgame perfect
equilibria. A striking aspect of the game analyzed here is that stationary perfect
equilibria may fail to exist. This point is illustrated by the following example.
EXAMPLE 2.6. N = {a, b, c}, and ρ defines a < b < c.
π va (π ) vb (π ) vc (π )
a|b|c 1 1 1

ab|c 3 2 1

ac|b 2 1 3

a|bc 1 3 2

abc 1 1 1
In this example, player a wants to form a coalition with player b, player b
with player c, and player c with player a.
To show that the game 0(v, ρ) does not admit any stationary equilibrium
coalition structure, observe first that the three coalition structures {{a, b, c}},
{{a}, {b}, {c}} and {{a}, {b, c}} cannot be supported by any equilibrium since
player a would benefit from deviating and offering the formation of the coali-
tion {a, c} which player c would accept. The two other coalition structures
 SEQUENTIAL FORMATION OF COALITIONS 99

{{a, b}, {c}} and {{a, c}, {b}} can be supported by equilibria in nonstationary
strategies but not by a stationary perfect equilibrium. For {{a, b}, {c}} to be sup-
ported by a stationary perfect equilibrium, it must be that player c rejects the
offer {b, c}. But, in equilibrium, player c will only reject the offer {b, c} if
player a accepts the offer {a, c}. By stationarity, player b accepts the offer {a, b}
irrespective of the history of rejections which have preceded it. Hence, since
player b always accepts the offer {a, b}, player a cannot accept the offer {a, c}.
Similarly, the coalition structure {{a, c}, {b}} is only supported by a strategy
prescribing that player b rejects the offer {a, b}, implying that player c accepts
the offer {b, c}. Since, by stationarity, player a always accepts the offer {a, c},
player c should reject the offer {b, c}. Hence, the game 0(v, ρ) does not admit
any stationary perfect equilibrium.
However, the coalition structures {{a, b}, {c}} and {{a, c}, {b}} can be sup-
ported by equilibria in nonstationary strategies.9 To support these coalition struc-
tures as equilibria, one only needs to allow players to condition their actions on
the number of times they have received an offer. Consider first the coalition
structure {{a, b}, {c}} and the following strategies. Player a always accepts the
offer {a, b} and proposes {a, b}. She rejects {a, b, c} and accepts {a, c} when, in
the history h t , she has made the offer {a, b} to player b an odd number of times.
Player b accepts {b, c} and proposes {b, c}. She rejects {a, b, c} and only accepts
{a, b} if, in the history h t , the offer {a, b} has been made by player a an odd
number of times. Player c accepts {a, c} and proposes {a, c}. She rejects {a, b, c}
and only accepts {b, c} if, in the history h t , player a has made the offer {a, b} an
even number of times. These strategies form a subgame perfect equilibrium of
the game (in nonstationary strategies), and are depicted in Figure 2. A strategy
profile supporting the coalition structure {{a, c}, {b}} can be constructed in a
similar way.
In Example 2.6, the three players play a symmetric role. Hence, no change in
the rule of order can guarantee the existence of a stationary perfect equilibrium.
Moreover, Example 2.6 is generic, since the nonexistence of a stationary perfect
equilibrium is robust to small variations of the valuation. Nonexistence of sta-
tionary perfect equilibria is thus a robust phenomenon in games with more than
three players.
Note however that the nonexistence of a stationary perfect equilibrium in
pure strategies in Example 2.6 is linked to the fixed sharing rule. If players
were allowed to bargain freely over the worth of the coalition in a game with
transferable utility, the nonexistence result would disappear.
The central feature of Example 2.6 is the disagreement among players over
the coalitions which should be formed. A similar problem was noted by Shenoy
(1979) in Apex games, where a single big player faces a number of small play-

9 These strategies are closely related to strategies constructed by Shaked to support any division of

the payoffs in a three-person bargaining game (Sutton (1986)).

100 FRANCIS BLOCH

FIG. 2. Nonstationary equilibrium strategies supporting the coalition structure {{a, b}, {c}}.
 SEQUENTIAL FORMATION OF COALITIONS 101

ers (Example 7.5, p. 150). The preferred coalition for the big player is the grand
coalition, since it offers her the possibility of diluting the power of the small play-
ers. Small players, on the other hand, would rather form a two-member coalition
with the big player. This disagreement among players about the coalition which
should be formed leads, as in Example 2.6, to the nonexistence of a stationary
perfect equilibrium. This suggests that a sufficient condition for the existence
of an equilibrium coalition structure is high degree of unanimity among players
about the coalitions they wish to form. While this point is not pursued here, the
class of symmetric games analyzed in Section 4 provides an example of games
where players unanimously agree on the coalitions they want to belong to.

3. STABLE COALITION STRUCTURES

In this section, I compare the equilibrium coalition structures with coalition

structures satisfying cooperative concepts of stability. Concepts of stability in
games with externalities require a specification of the reaction of external play-
ers to the formation of a coalition, and different assumptions on the behavior
of external players give rise to different definitions of stability. Kurz (1988)
distinguishes five models of reaction of the external players. The core stability
concept, first introduced by Shenoy (1979), is based on the following dominance
relation. A coalition structure π dominates a coalition structure π 0 if there exists
a coalition in π whose members receive strictly higher payoffs than in π 0 . A
coalition structure is called core stable if it belongs to the core of the dominance
relation. In effect, this definition of stability is very restrictive, since it assumes
that, when a group of players deviate, they consider that external players react
in such a way as to maximize the payoff of deviating players.
Hart and Kurz (1983) propose four models of reaction of the external players.
In the γ model, coalitions which are left by some members dissolve. In the δ
model, members of coalitions which lose members remain together and form
smaller coalitions. The last two stability concepts are based on the β and the
α cores.10 In the β model, a group K of players deviates if, for any possible
reaction of the external players, namely any coalition structure π N \K of N \ K ,
there exists a coalition structure of K , π K , such that all members of K are better
off in the new coalition structure π = π N \K ∪ π K . In the α definition, a group
K of players deviates if there exists a coalition structure π K such that, whatever
the reaction of the external players, members of K are better off forming the
coalition structure π K .
Letting, for any fixed valuation v, the sets of Core stable, γ stable, δ stable,
β stable and α stable coalition structures be denoted by CC(v), Cγ (v), Cδ(v),

10 See Aumann (1967) for a complete description of the α and β core concepts.
102 FRANCIS BLOCH

Cβ(v) and Cα(v), the following lemma is easily established.11

LEMMA 3.1. For any valuation v,CC(v) ⊂ (Cγ (v) ∪ Cδ(v)) ⊂ Cβ(v) ⊂
Cα(v).
I will focus here on the two extreme concepts of core and α stability.12 For-
mally, a coalition structure π is core stable if there does not exist a coalition
K and a coalition structure π 0 such that K ∈ π 0 and ∀i ∈ K , vi (π 0 ) > vi (π ).
A coalition structure π is α stable if there does not exist a coalition K and a
partition π K0 on K such that, ∀i ∈ K , ∀π N \K ∈ 5 N \K , vi (π K0 ∪ π N \K ) > vi (π ).
The next proposition shows that, when the set of stationary equilibrium coali-
tion structures is nonempty, it contains the set of core stable structures.
PROPOSITION 3.2. Assume that there exists a rule of order ρ such that
S E SC(v, ρ) 6= ∅. Then CC(v) ⊂ S EC S(v, ρ).

Proof. Let ρ̃ denote one rule of order for which S EC S(v, ρ̃) 6= ∅. Let π̃
denote a coalition structure in CC(v). To prove the proposition, I construct a
stationary perfect equilibrium ρ̃ of the game 0(v, ρ̃) such that π(σ̃ ) = π̃. I
denote by T (i) the coalition to which player i belongs in the coalition structure
π̃ . A partition π K of a subset K of the players is called a subpartition of π̃ if
it is formed by the union of elements of π̃. The set of all subpartitions of π̃ is
denoted Sub(π̃ ). Pick a stationary perfect equilibrium σ̃˜ of the game 0(v, ρ̃).
A stationary strategy σ̃i for player i is then constructed as follows.
Assume that a subset K of players, where i 6∈ K , has already formed a
coalition structure π K .

If π K 6∈ Sub(π̃ ), σ̃i (K , π K , ·) = σ̃˜ i (K , π K , ·)

If π K ∈ Sub(π̃ ), σ̃i (K , π K , φ) = T (i)

σ̃i (K , π K , T (i)) = Yes

σ̃i (K , π K , T 0 ) = Yes if vi (π(T 0 )) > vi (π̃)

σ̃i (K , π K , T 0 ) = No if vi (π(T 0 )) ≤ vi (π̃),

where π(T 0 ) is the coalition structure generated by σ̃˜ after the coalition T 0 has
been formed.
11 Hart and Kurz (1983) derive the last three inclusions of the Lemma. The first inclusion is imme-

diate, once one reinterprets the core stability concept in terms of reaction of the external players to the
deviation of a group of players.
12 The absence of coincidence between α stable structures and equilibrium coalition structures can

be extended to the intermediate concepts of β, γ , and δ stability.

 SEQUENTIAL FORMATION OF COALITIONS 103

The strategy σ̃ prescribes that player i follows her part of a stationary perfect
equilibrium σ̃˜ if a coalition structure π K off the equilibrium path has been formed,
and that she forms the coalition T (i) otherwise.
It remains to check that σ̃ is a subgame perfect equilibrium of the game
0(v, ρ̃). Observe first that, since σ̃˜ is a stationary perfect equilibrium profile,
the strategy profile σ̃ is a subgame perfect equilibrium if a coalition structure off
the equilibrium path has been formed. Suppose now that the previous players
have formed a coalition structure π K in Sub(π̃). To check that σ̃ is a subgame
perfect equilibrium on the equilibrium path, consider the possible deviations for
player i.
Player i can deviate by announcing a coalition structure T 0 6= T (i) when it
is her turn to announce a coalition. However, since π̃ is a core stable structure,
there exists a player j in T 0 such that v j (π(T 0 )) ≤ v j (π̃). Hence, any coalition
T 0 different from T (i) will be rejected.
If now player i receives an offer T (i), any deviation will lead to the formation
of the coalition T (i), since any different offer by player i will be rejected by
some player.
Finally, suppose that player i receives an offer T 0 6= T (i). If vi (π(T 0 )) ≤
vi (π̃ ), she cannot benefit from accepting the offer. If all other members of
T 0 accept the offer, the coalition T 0 is formed and player i obtains a payoff
vi (π(T 0 )), whereas, by rejecting the offer, player i obtains the payoff vi (π̃). If
vi (π(T 0 )) > vi (π̃ ), player i should accept the offer, since her rejection would
lead to the formation of the structure π̃, whereas her acceptance may either
secure the formation of π(T 0 ), if no player following player i rejects the offer
T 0 , or yield the formation of T (i), if some player following player i rejects the
offer T 0 .
Since player i has no incentive to deviate from her strategy σ̃i , the strategy
profile σ̃ forms a subgame perfect equilibrium of the game 0(v, ρ̃). Furthermore,
by construction, π(σ̃ ) = π̃ . Hence, CC(v) ⊂ S EC S(v, ρ̃).

In the statement of Proposition 3.2, I require the set of stationary perfect equi-
libria to be nonempty. This assumption is needed to show that, once a coalition
structure is formed off the equilibrium path, the game still admits a stationary
perfect equilibrium. The following example shows that the assumption cannot
be relaxed.

EXAMPLE 3.3. N = {a, b, c, d}.

104 FRANCIS BLOCH

π va (π ) vb (π ) vc (π ) vd (π )
abcd 5 5 5 5

a|bc|d 1 3 2 1

a|b|cd 1 1 3 2

a|bd|c 1 2 1 3

Others 1 1 1 1
The game of Example 3.3 admits a unique core stable structure, the grand
coalition which Pareto dominates any other coalition structure. However, the
subgame following the formation of the coalition {a} is identical to the game in
Example 2.6 and does not admit any stationary perfect equilibrium.
The difficulty illustrated by Example 3.3 can be alleviated by assuming that,
in addition to the valuation v, all restrictions of the valuation to subsets of
the players admit a core stable structure.13 Since payoffs depend on the whole
coalition structure, the restriction of the valuation v to a subset K of the players
must entail a description of the partition formed by the external players.
The restriction of the valuation v to a subset K of the players relative to the
coalition structure π N \K is defined as follows. v(K , π N \K ): 5 K → <|K | , where
v(K , π N \K )i (π K ) = vi (π K ∪ π N \K ).
LEMMA 3.4. Let v be a valuation such that CC(v) 6= ∅, and, for any restric-
tion v 0 of v, CC(v 0 ) 6= ∅. Then, for any rule of order ρ, S EC S(v, ρ) 6= ∅.

Proof. Let ρ be a fixed rule of order. I construct a stationary perfect equi-

librium strategy profile. For any restriction v 0 of v to a subset K of the players,
relative to the coalition structure π N \K , pick a core stable structure. This core
stable structure is denoted by C S(π N \K ), and, for any player i in K , T (i, π N \K )
denotes the coalition player i belongs to in C S(π N \K ).
Construct a stationary strategy profile σ as follows.

σi (N \ K , π N \K , ∅) = T (i, π N \K )

σi (N \ K , π N \K , T (i, π N \K )) = Yes

σi (N \ K , π N \K , T 0 ) = Yes if vi (C S(π N \K ∪ T 0 )) > vi (C S(π N \K ))

σi (N \ K , π N \K , T 0 ) = No if vi (C S(π N \K ∪ T 0 )) ≤ vi (C S(π N \K )).

13 This requirement is very similar to the condition of total balancedness for games without

externalities.
 SEQUENTIAL FORMATION OF COALITIONS 105

To show that σ forms a subgame perfect equilibrium, consider all possible

deviations for player i.
If player i proposes a coalition T 0 6= T (i, π N \K ), one of the members of T 0
will reject the offer, since C S(π N \K ) is a core stable structure. Hence, player i
cannot benefit from announcing a coalition different from T (i, π N \K ). Similarly,
by rejecting the offer T (i, π N \K ), player i cannot obtain a higher payoff since
the only coalition she can announce is the coalition T (i, π N \K ).
Suppose now that player i receives an offer T 0 off the equilibrium path. By the
same argument as in Proposition 3.2, she should accept the offer only if the payoff
she receives in the final coalition structure is higher than the payoff she receives
in C S(π N \K ). The final coalition structure obtained after the formation of T 0 ,
given the construction of the strategies, is the coalition structure C S(π N \K ∪ T 0 ).
Hence, no deviation from the strategy σi can be profitable and the constructed
strategy profile σ is a stationary perfect equilibrium.
Proposition 3.2 and Lemma 3.4 immediately lead to the following corollary.
COROLLARY 3.5. Let v be a valuation such that CC(v) 6= ∅, and, for all
restrictions v 0 of v, CC(v 0 ) 6= ∅. Then, for any rule of order ρ, CC(v) ⊂
S EC S(v, ρ).
Lemma 3.4 provides a sufficient condition for the existence of an equilibrium
coalition structure. Corollary 3.5 shows that any core stable structure of a valua-
tion v whose restrictions also admit core stable structures can be reached as the
outcome of a stationary perfect equilibrium of the game of coalition formation.
In the case of α stability, no such result can be expected. The following example
shows that the sets of stationary equilibrium coalition structures and of α stable
structures may be nonempty and disjoint.
EXAMPLE 3.6. N = {a, b, c, d, e}.
π va (π ) vb (π ) vc (π ) vd (π ) ve (π )
ab|cd|e 4 4 3 3 1

a|bc|d|e 1 5 5 4 1

ae|bc|d 1 5 5 4 1

a|bc|de 1 1 1 5 5

ac|b|de 1 2 1 1 1

a|b|c|de 1 2 1 1 1

Others 1 1 1 1 1
106 FRANCIS BLOCH

The game admits three α stable structures {{a}, {b, c}, {d}, {e}}, {{ae}, {bc},
{d}} and {{a}, {bc}, {de}}. To check that the coalition structure {{a}, {b, c}, {d},
{e}} is α stable, observe that the only players who have an incentive to deviate
are players d and e, who may want to form a coalition. However, their deviation
is prevented by the formation of the coalition structure {{a, c}, {b}} by the three
other players. The coalition structure {{ae}, {bc}, {d}} is α stable for the same
reason. The structure {{a}, {bc}, {de}} is α stable because the only two profitable
deviations can be prevented by the external players. If players a and b form
the coalition {a, b}, the three other players can react by forming the structure
{{c}, {d}, {e}}, inducing a payoff of 1 for the two deviating players. If player b
decides to break the coalition with player c, the four external players can form
the coalition {a, b, c, d} which yields a payoff of 1 for player b.
These three coalition structures are the only α stable structures of the game.
The coalition structure {{a, b}, {c, d}, {e}} is not α stable since players b, c and
d can deviate and form the structure {{b, c}, {d}} in which they are guaranteed
to obtain higher payoffs. All other coalition structures are Pareto dominated by
the coalition structure {{a, b}, {c, d}, {e}} and hence are not α stable.
I now claim that the unique stationary equilibrium coalition structure of the
game, independently of the rule of order ρ, is the coalition structure {{a, b},
{c, d}, {e}}. Two cases must be distinguished, one where ρ assigns as the first
player a or b, one where c, d or e are chosen to start the game. If player a starts
the game, player a should offer the formation of a coalition {a, b}. This offer
will be accepted by player b, since, if player b were to form the coalition {b, c},
players d and e would form a coalition, inducing a payoff of 1 for player b.
Given that players a and b have formed a coalition, player c should offer to form
a coalition with player d, who will accept. Hence, in equilibrium, the coalition
structure {{a, b}, {c, d}, {e}} is formed. The same line of reasoning applies when
player b starts the game.
If now player c starts the game, she should offer the formation of the coalition
{c, d}, since the offer {b, c} will be rejected by player b. This offer will be
accepted by player d. In fact, player d has no incentive to form the coalition
{d, e} since this induces player b to form the coalition {b}. Once the coalition
{c, d} is formed, players a and b form the coalition {a, b}, yielding the coalition
structure {{a, b}, {c, d}, {e}}. A similar line of reasoning applies to the cases
where d and e start the game.
Example 3.6 is robust to small variations of the valuation. Hence there exists a
class of valuations v, such that S E SC(v, ρ) 6= ∅, Cα(v) 6= ∅ and S EC S(v, ρ)∩
Cα(v) = ∅.
The absence of coincidence between the sequential game of coalition forma-
tion and the model of α stability stems from two countervailing forces in the
definitions of deviations. On the one hand, deviations in the sequential model
are easier to obtain, because the external players who have already formed a
coalition cannot freely react to the deviation. This suggests that there may ex-
 SEQUENTIAL FORMATION OF COALITIONS 107

ist α stable structures which cannot be outcomes of subgame perfect equilibria

of the game. In Example 3.6, the coalition structures {{a}, {b, c}, {d}, {e}} and
{{ae}, {bc}, {d}} are not stationary equilibrium structures, because, once players
a, b and c have left the game, players d and e can deviate and form the coalition
{d, e}. Similarly, the coalition structure {{a}, {bc}, {de}} cannot be obtained in
a stationary perfect equilibrium, because b has an incentive to deviate after the
coalition {d, e} has been formed.
On the other hand, deviations in the sequential model are harder to obtain,
because group deviations are not allowed, and players look forward to the fi-
nal consequences of their deviations. Hence stationary equilibrium coalition
structures are not necessarily α stable. In Example 3.6, the coalition structure
{{a, b}, {c, d}, {e}} is not α stable, because players b, c and d may deviate jointly
and form the coalition structure {{a}, {b, c}, {d}, {e}}.

4. SEQUENTIAL FORMATION OF COALITIONS IN SYMMETRIC

GAMES

In this section, I analyze the formation of coalitions in the restricted class

of symmetric games. Symmetric games are described by valuations where all
players are ex ante identical. Hence, the payoffs received by the players only
depend on coalition sizes and not on the identity of the coalition members.
Formally, let p denote a permutation of the players in N . For any coalition
structure π of N , let pπ denote the coalition structure obtained by permuting the
players according to p. A valuation v is symmetric if and only if ∀i ∈ N , vi (π ) =
v pi ( pπ ).
A symmetric game is a game described by a symmetric valuation. Observe
that in symmetric games all members of a coalition receive the same payoff
and payoffs only depend on the sizes of the coalitions. An important feature
of symmetric games is that two coalition structures which only differ by the
distribution on the players in the coalitions generate the same payoff distribution.
This leads to the notion of equivalence of coalition structures in symmetric
games.
Two coalition structures π and π 0 are called equivalent if there exists a per-
mutation p of the players in N such that π 0 = pπ. Two equivalent partitions
are said to be equal up to a permutation of the players. The equivalence class
of a coalition structure π is denoted by eq(π ). If the valuation v is symmetric,
two equivalent partitions generate the same distribution of payoffs. Hence, in
symmetric games, the study of coalitions can be restricted to the study of equiv-
alence classes of partitions. An equivalence class of partitions can be identified
with a list of coalition sizes, that is a sequence of positive integers adding up
to n. I assume that the rule of order ρ is fixed, and let the players be indexed
by the ordered set I = 1, 2, . . . , n. This can be done without loss of generality,
108 FRANCIS BLOCH

since any coalition structure emerging as an equilibrium of the game 0(v, ρ 0 ),

for ρ 0 6= ρ, is equivalent to a coalition structure generated by an equilibrium of
the game 0(v, ρ). Since the rule of order ρ is fixed, the game 0 will only be
indexed by the valuation v.
Since in a symmetric game, all players are ex ante identical, I restrict my
attention to symmetric equilibria where all players adopt similar strategies. For-
mally, a strategy profile σ = {σi }i∈N is called symmetric if and only if (i) at any
two states s = (K , π K , T ), s 0 = (K , π K , T 0 ) with |T | = |T 0 | 6= 0, for any two
players i ∈ T , j ∈ T 0 , σi (s) = σ j (s 0 ) and (ii) at any state s = (K , π K , ∅), for
any two players i, j ∈/ K |σi (s)| = |σ j (s)|. In words, a strategy profile is sym-
metric if, at any state, all responders adopt the same strategy and all proposers
announce coalitions of the same size. The set of coalition structures supported
by symmetric stationary perfect equilibria is denoted SS EC S(v).
I first show that, in a symmetric game, any symmetric stationary perfect equi-
librium coalition structure can be reached as the outcome of a finite game of
choice of coalition sizes. Furthermore, under a simple condition proposed by Ray
and Vohra (1995), any equilibrium outcome of the game of choice of coalition
sizes can be obtained as a symmetric stationary equilibrium coalition structure
of the sequential game of coalition formation. Using this equivalence, I derive a
sufficient condition under which a symmetric game admits a symmetric station-
ary equilibrium coalition structure and prove that this structure is generically
unique.
The game of choice of coalition sizes 1(v) is described as follows. Player 1
starts the game and chooses an integer k1 in the interval [1, n]. Player k1 + 1 then
moves and chooses an integer k2 in the set [1, n − k1 ]. Player k1 + k2 + 1 chooses
at the next stage an integer k3 in the set [1, n − k1 − k2 ]. The game P continues
until the sequence of integers (k1 , k2 , . . . , k j , . . . , k J ) satisfies k j = n. The
game for three players is depicted in Fig. 3.
A strategy τi for player i in the game 1(v) is a mapping from the set 5i−1
to the set of integers in the interval [1, n − i − 1]. In words, for any coalition
structure πi−1 of the first i − 1 players, player i chooses a coalition size τi (πi−1 ).
All players need not be called to announce coalition sizes in the game. Observe,
however, that, for any strategy profile τ, a single coalition structure π(τ ) is
formed. The payoffs received by the players are then given by vi (π(τ )).
A strategy profile τ ∗ is a subgame perfect equilibrium if and only if for
all players i, for all coalition structures πi−1 in 5i−1 and for all strategies
τi , vi (π(τi∗ (πi−1 ), τ−i
∗ ∗
)) ≥ vi (π(τi (πi−1 ), τ−i )). As before, a coalition structure
generated by a subgame perfect equilibrium τ ∗ is called an equilibrium coalition
structure of the game 1(v). The set of equilibrium structures of 1(v) is denoted
EC S 0 (v).

LEMMA 4.1. For any symmetric valuation v, EC S 0 (v) 6= ∅.

 SEQUENTIAL FORMATION OF COALITIONS 109

FIG. 3. The game 1.

Proof. The game 1(v) is a finite game of perfect information with perfect
recall. Hence it admits a subgame perfect equilibrium in pure strategies.

In the next proposition, I show that any symmetric stationary equilibrium

coalition structure of the game 0(v) can be reached as an equilibirum coalition
structure of the game 1(v), up to a permutation of the players.

PROPOSITION 4.2. For any coalition structure π in SS EC S(v) there exists a

coalition structure π 0 equivalent to π such that π 0 ∈ EC S 0 (v).

Proof. Let σ be the symmetric stationary perfect equilibrium of the game

0(v) supporting the coalition structure π. I first show that the strategy pro-
file σ cannot involve any delay and that all offers prescribed by σ are ac-
cepted. Suppose to the contrary that some player i rejects an offer T with
110 FRANCIS BLOCH

|T | ≥ 2 at some state s = (K , π K , T ). Since the strategy profile σ is sym-

metric, |σi (K , π K , ∅)| = |T | and for all players j ∈ σi (K , π K , ∅), we have
σ j (K , π K , σi (K , π K , ∅)) = σi (K , π K , T ) = No. Hence offers are continuously
rejected and the play of the game is infinite yielding a payoff of 0 to player i.
Since however minπ⊃{{i}} vi (π ) > 0, player i has an incentive to deviate and
leave the game. This shows that, at a symmetric equilibrium σ , all offers are
accepted. Hence the strategy σ can be described by a list of offers made by the
players at all states where they are proposers.
As a second step, I show that we can assume without loss of generality that, at
any two equivalent states s = (K , π K , ∅) and s 0 = (K 0 , π K0 , ∅) where |K | = |K 0 |
and the two coalition structures π K and π K 0 are equivalent, |σi (s)| = |σi (s 0 )|. To
see this first reorder the players according to a rule of order ρ̂ consistent with
the order in which the coalition structure π is formed. Now, for any set K with
i ∈/ K , let K̂ denote the first ρ̂-elements in N \ {i} and, for any partition π K
of K , let πˆK denote the equivalent partition of K̂ . Construct then the strategy
σ̂i as follows. At any state s = (K , π K , ∅), let σ̂i (s) be a subset of N \ K
containing i such that |σ̂i (K , π K , ∅)| = |σ ( K̂ , πˆK , ∅)|. In words, I select for
any state s = (K , πk , ∅) a particular representative of the equivalence class
eq(π K ) and σ̂i assigns the action chosen for this representative state to the entire
equivalence class. Clearly, the strategy σ̂ satisfies the condition that sets of the
same cardinality are chosen at two equivalent states. Furthermore, given the
particular order ρ̂ chosen, π(σ̂ ) = π(σ ). It remains to show that σ̂ forms a
subgame perfect equilibrium of the game 0(v). To see this, consider a state
s = (K , π K , ∅) and note that, since the strategy σ̂ is played, any action of player
i induces a unique partition of the set N \ K . Now suppose by contradiction
that σ̂i is not an optimal choice, i.e. that there exists a strategy σ̃i inducing a
partition π N˜\K such that vi (π K ∪ π N˜\K ) > vi (π K ∪ π Nˆ\K ) where π Nˆ\K is the
coalition induced by σ̂i . Next consider a permutation p̂ of the players such that
p̂K = K̂ . Since the game is symmetric, vi ( p(π K ∪ π N˜\K )) = vi (π K ∪ π N˜\K ) >
vi (π K ∪ π Nˆ\K ) = vi ( p(π K ∪ π Nˆ\K )), contradicting the fact that σi is an optimal
choice at ( K̂ , πˆK , ∅).
Since we may assume, by the preceding step, that the strategy σ assigns sets
of the same cardinality at any two equivalent states, we are ready to construct
a strategy profile τ in the game 1(v) as follows. For any player i and any
coalition structure πi−1 of the preceding players, let τi (πi−1 ) = |σi (K , π K , ∅)|.
To show that τ forms a subgame perfect equilibrium of the game 1(v), suppose
by contradiction that player i has a profitable deviation τi0 6= τi after the coalition
structure πi−1 is formed. I claim that this implies that player i has a profitable
deviation from σi in the game 0(v). To see this, suppose that a coalition structure
π K equivalent to πi−1 has been formed and let player i reject any offer T such
that |T | 6= τi0 and propose the formation of a coalition T 0 of size τi0 . Since τi0 is a
profitable deviation in the game 1(v) and letting π 0 denote the coalition structure
 SEQUENTIAL FORMATION OF COALITIONS 111

induced by the choice τi0 , we must have vi (π 0 ) > vi (π ). Now, by symmetry, for
all players j in T 0 , v j (π 0 ) = vi (π 0 ) > vi (π ) = v j (π ), so that player i’s offer is
accepted.

While Proposition 4.2 guarantees that any symmetric equilibrium can be ob-
tained as an equilibrium outcome of the game of choice of coalition sizes, it
does not imply that the equilibrium coalition structures of the game 1 form
symmetric stationary equilibrium outcomes of the sequential game of coalition
formation. In fact, as noted by Ray and Vohra (1995), a stronger condition is
needed for this assertion to hold : the coalitions formed in the game 1 must
have the property that the players’ payoffs are decreasing in the order in which
coalitions are formed.

PROPOSITION 4.3 (Ray and Vohra (1995)). Let π be an equilibrium coalition

structure of the game 1(v) with the property that players’ payoffs are decreasing
in the order in which coalitions are formed. Then there exists a coalition structure
π 0 equivalent to π such that π 0 ∈ SS EC S(v).

Proof. Let τ be the subgame perfect equilibrium supporting τ . Define a

strategy σi for player i in the game 0(v) as follows. At any state s = (K , π K , ∅)
let player i announce a subset T of N \ K with |T | = τ j (π j−1 ) for the coalition
structure π j−1 equivalent to π K . At any state s = (K , π K , T ) with T 6= ∅, let
σi (s) = Yes if |T | = τ j (π j−1 ) and σi (s) = No otherwise. This strategy profile
is symmetric and yields a coalition structure π(σ ) equivalent to π. It remains
to show that it forms a stationary perfect equilibrium of the game 0(v). First
consider player i’s possible deviation at a state s = (K , π K , ∅) when it is her
turn to make an offer. If she makes any offer T 0 such that |T 0 | 6= τ j (π j−1 ) and
|T 0 | ≥ 2, her offer will be rejected. Hence player i will belong to a coalition
formed later in the game and, by assumption, her payoff is lower than the one
she obtains in coalition T . By the same reasoning, player i has no incentive to
reject an offer T where |T | = τ j (π j−1 ). Finally, consider player i’s response to
an offer T 0 with |T 0 | 6= τ j (π j−1 ). By rejecting the proposal and offering to form a
coalition T of size |T | = τ j (π j−1 ), she can secure the formation of the coalition
structure π . Since τ is a subgame perfect equilibrium of the game of choice of
coalition sizes, vi (π) ≥ vi (π K ∪ π N˜\K ) for any other coalition structure π N˜\K
induced by the formation of a coalition T 0 at state s = (K , π K , ∅). Hence no
player has any incentive to deviate from the strategy prescribed by σ .

Propositions 4.2 and 4.3 provide a sufficient condition on the underlying valu-
ation v for the equivalence between the symmetric stationary perfect equilibrum
outcomes of the sequential game of coalition formation and the subgame perfect
equilibrium outcomes of the game of choice of coalition sizes. This result is
formally stated in the next corollary.
112 FRANCIS BLOCH

COROLLARY 4.4. Suppose that, in the game 1(v), players’ payoffs are de-
creasing in the order in which coalitions are formed. Then, for any coali-
tion structure π in SS EC S(v) and any coalition structure π 0 in EC S 0 (v),
eq(π) = eq(π 0 ).
Hence, under a simple condition, the game of choice of coalition sizes pro-
vides an easy method for the construction of equilibrium coalition structures in
symmetric games. The exact nature of the restriction that players’ payoffs are
decreasing in the order in which coalitions are formed is difficult to interpret.
Ray and Vohra (1995) provide an example where the condition is violated and
the subgame perfect equilibrium outcome of the game of choice of coalition
sizes does not form a symmetric stationary perfect equilibrium of the sequential
game. However, in most economic applications of coalitions with externalities,
including the formation of cartels and of coalitions in majority games discussed
in this paper, this condition is satisfied. The equivalence result of Corollary 4.4
can now be used to establish several important properties of equilibrium coalition
structures in symmetric games.
COROLLARY 4.5. Let v be a symmetric valuation such that, in the game 1(v),
players’ payoffs are decreasing in the order in which coalitions are formed. Then
S EC S(v) 6= ∅.
Proof. Follows from Lemma 4.1 and Corollary 4.4.
Corollary 4.4 also leads to a simple sufficient condition for the uniqueness
of symmetric stationary equilibrium coalition structures in symmetric games. A
valuation v is called strict if, for any player i, and for any two different partitions
π and π 0 , vi (π) 6= vi (π 0 ). In a game described by a strict valuation, every agent
receives different payoffs in different coalition structures. The next proposition
shows that the strictness condition is sufficient to guarantee the uniqueness of
the equilibrium coalition structure in the game 1(v).
PROPOSITION 4.6. Let v be a strict symmetric valuation. Then the game 1(v)
has a unique equilibrium coalition structure.
Proof. The proof is by induction on the number n of players. If n = 1, the
game 1(v) has a unique subgame perfect equilibrium. Suppose now that, for any
n 0 < n, the game admits a unique subgame perfect equilibrium, and consider the
first player’s choices in a game with n players. For any choice of an integer k,
the continuation game has less than n players, and thus admits a unique subgame
perfect equilibrium τ ∗ ({k}). Since the valuation is strict, there exists a unique
k ∗ , such that
v1 ({k ∗ } ∪ π(τ ∗ ({k ∗ })) > v1 ({k} ∪ π(τ ∗ ({k})) ∀k 6= k ∗ .
Hence the n player game admits a unique subgame perfect equilibrium.
 SEQUENTIAL FORMATION OF COALITIONS 113

Proposition 4.6 implies that, if the valuation is strict, all equilibrium coalition
structures of the game 0(v) are equivalent. Hence I obtain the following corollary.

COROLLARY 4.7. Let v be a strict symmetric valuation such that, in the game
1(v), players’ payoffs are decreasing in the order in which coalitions are formed.
Then the game 0(v) has a unique symmetric stationary equilibrium coalition
structure, up to a permutation of the players.

5. APPLICATIONS

In this section, I apply the sequential model of coalition formation to two

particular symmetric situations. I first analyze the formation of cartels in a sym-
metric Cournot oligopoly. The second application is based on Hart and Kurz
(1984)’s study of endogenous coalition formation in symmetric majority games.
In both applications, I derive the subgame perfect equilibrium of the game of
choice of coalition sizes. It is straightforward to check that players’ payoffs are
decreasing in the order in which coalitions are formed, so that the equivalence
result of Corollary 4.4 can be applied.

5.1. Cartels in a Symmetric Cournot Oligopoly

It has long been noted that the formation of cartels in oligopolies involves
a fundamental instability (See Stigler (1968)), since, once a cartel has been
formed, members of the cartel obtain a lower profit than outsiders, and hence
have an incentive to leave the cartel. Salant et al. (1983) analyze this instability
in a simple symmetric Cournot oligopoly with linear demand and homogeneous
goods, and show that there exists a minimal profitable size of the cartel which is
never lower than four fifths of the members of the industry. This cartel is however
(intuitively) unstable since members of the cartel would prefer to stay out and
let the other firms form a cartel. In the sequential model analyzed here, firms
have the power to commit to stay out of the cartel. Hence, the unique equilibrium
coalition structure predicts that firms choose to remain outside of the cartel, until
the remaining firms form the cartel of minimal profitable size.
More precisely, considerPa Cournot oligopoly where firms face a linear inverse
demand curve, D = α − qi , where qi is the quantity produced by each firm
i. All firms are assumed to have a constant marginal cost of λ. Suppose that
K cartels have formed on the market, and that the structure of cartels is given
by π = {T1 , T2 , . . . , Tk , . . . , TK }. Straightforward computations show that, in
equilibrium, each cartel will produce qi∗ (π ) = (α − λ)/(K + 1).14 Hence, firm

14 It is important to note that the equilibrium quantity produced by each cartel only depends on the

number of cartels on the market.

114 FRANCIS BLOCH

i in the cartel T (i) of size t (i) obtains a payoff of

(α − λ)2
Pi∗ (π ) = .
t (i)(K + 1)2
The problem of cartel formation can thus be summarized by the valuation defined
by vi (π) = Pi∗ (π).
PROPOSITION 5.1. Any equilibrium of the game of cartel formation is char-
acterized by π ∗ = (T1∗ , { j} j6∈T1∗ ) where t1∗ is the first integer following (2n +
√ √
3 − 4n + √5)/2. (If 4n + 5 is an integer, √ t1∗ can take on the two values
(2n + 3 − 4n + 5)/2 and (2n + 5 − 4n + 5)/2.)

Proof. See the Appendix.

5.2. Coalitions in Symmetric Majority Games

In their study of endogenous coalition formation, Hart and Kurz (1983) advo-
cate a two-stage approach, where players evaluate their payoffs, in any coalition
structure, according to a fixed rule (Owen (1977)’s extension of the Shapley
Value to games with coalition structures), and play a game of coalition forma-
tion using the value as their expected payoff. Owen (1977)’s value differs from
Aumann and Drèze (1974)’s value in assuming that players bargain over the
worth of the grand coalition, as opposed to the worth of the coalition they belong
to in the coalition structure. The formation of a coalition is thus interpreted as
a way for the players to modify the environment in which they bargain over the
worth of the grand coalition.15
Owen (1977)’s value is computed, for any game in coalitional function form
w, any coalition structure π and any player i as

φi (w, π ) = E(w(P ∪ i) − w(P )),

where the expectation is taken over any random order which is consistent with
the coalition structure π (i.e. ranks consecutively members of any coalition in
the coalition structure) and P is the set of predecessors of i according to the
random order.
Hart and Kurz (1984) apply Owen (1977)’s value to analyze the formation of
coalitions in different types of games in coalitional function form. We consider
here only symmetric majority games.
DEFINITION 5.2. A symmetric majority game M(n, m) is defined as follows.
The number n is the total number of players, and the integer m (the majority) is
15 The axiomatic derivation of the two different values are given in Aumann and Drèze (1974) and

Hart and Kurz (1983). The differences are thouroughly discussed in Hart and Kurz (1983).
 SEQUENTIAL FORMATION OF COALITIONS 115

any integer in the interval [(n + 1)/2, n]. The coalitional function is given by
• w(T ) = 0 if t < m
• w(T ) = 1 for t ≥ m,
where T is any coalition, and t denotes the cardinality of coalition T .
To compute the Owen value in the symmetric majority game M(n, m), let me
consider a coalition structure π containing K coalitions, π = {T1 , T2 , . . . , Tk , . . .
TK }. The total number of random orders consistent with the coalition structure π
is K !t1 !t2 ! . . . tk ! . . . t K !, where tk denotes the number of elements of the coalition
Tk . It is then clear that for the incremental value of player i to be positive, it must
be that player i is ordered at position m in the random order. Denoting by T (i) the
coalition player i belongs to and letting ωi (π ) denote the number of orderings
of the coalitions in π such that a member of the coalition T (i) is at position m,
I obtain the following simple expression for the Owen value
ωi (π )
φi (π ) = .
t (i)K !
Hence I can now define the valuation vi (π ) = φi (π ). The characterization
of the equilibrium coalition structures is made difficult by the lack of structure
of the function ωi (π). In the absence of general characterization results, Table I
describes the equilibrium coalition structures of any symmetric majority game
with n ≤ 10.16
The equilibrium coalition structures of symmetric majority games are not
easily interpreted. When the majority required to win (m) is small, it appears
that the minimal winning coalition forms, members of the winning coalition all
obtain 1/m and external members, who obtain 0, organize themselves freely.
When the number of votes required to win increases, the share of any member
of the winning coalition decreases and it may become profitable to form smaller
coalitions. This effect explains why the minimal winning coalition does not
necessarily form in the symmetric majority games M(5, 4), M(6, 5), M(7, 6),
M(8, 6), M(9, 7), M(9, 8) and M(10, 8). However, if all votes are required to
win, the only equilibrium coalition structures are the grand coalition and the
coalition consisting of singletons. In fact, in that case, the probability to win is
independent of the size of the coalition, and players should always try to form
the smallest coalitions. Hence, the only possible equilibrium coalition structure
are the coalition consisting of singletons and the grand coalition which yield
the same payoff of 1/n to all players. Finally, it should be noted that Hart and
Kurz (1984) observed that the majority game M(10, 8) has no α stable coalition
structure. However, in my framework, an equilibrium coalition structure exists
for this game.
16 The computations leading to the characterization of the coalition structures are not reproduced

here and are available from the author.

116 FRANCIS BLOCH

TABLE I. Coalition Structures in Symmetric Majority Games

n=3
m=2 m=3

ab|c a|b|c
abc

n=4
m=3 m=4

abc|d a|b|c|d
abcd

n=5
m=3 m=4 m=5

abc|d|e ab|cd|e a|b|c|d|e

abc|de abcd|e abcde

n=6
m=4 m=5 m=6

abcd|e| f abc|de| f a|b|c|d|e| f

abcd|e f abcde f

n=7
m=4 m=5 m=6 m=7

abcd|e| f |g abcde| f |g abcde f |g a|b|c|d|e| f |g

abcd|e f |g abcde| f g ab|cd|e f |g abcde f g
abcd|e f g abc|de f |g

6. CONCLUSIONS

In this paper, I analyze a sequential noncooperative game of coalition forma-

tion when the rule of payoff division is fixed and payoffs depend on the whole
coalition structure. The extensive form of the game is closely related to the exten-
sive forms proposed by Selten (1981), Chatterjee et al. (1993) and Moldovanu
(1992) for games of coalitional bargaining. I show that any core stable structure
can be obtained as the outcome of a stationary perfect equilibrium, provided that
the set of stationary perfect equilibria is nonempty. I analyze games described by
symmetric valuations and provide a condition under which, when all the players
 SEQUENTIAL FORMATION OF COALITIONS 117
TABLE I. Continued.

n=8
m=5 m=6 m=7 m=8

abcde| f |g|h abcde f |g|h abcde f g|h a|b|c|d|e| f |g|h

abcde| f g|h abcde f |gh abcde f gh
abcde| f gh abc|de f |g|h
abc|de f |gh

n=9
m=5 m=6 m=7 m=8 m=9

abcde| f |g|h|i abcde f |g|h|i abcd|e f g|hi ab|cd|e f |gh|i a|b|c|d|e| f |g|h|i
abcde| f g|h|i abcde f |gh|i abcd|e f g|h|i abcde f gh|i abcde f ghi
abcde| f g|hi abcde f |ghi
abcde| f gh|i
abcde| f ghi

n = 10
m=6 m=7 m=8 m=9 m = 10

abcde f |g|h|i| j abcde f g|h|i| j abcde f gh|i| j abcde f ghi| j a|b|c|d|e| f |g|h|i| j
abcde f |gh|i| j abcde f g|hi| j abcde f gh|i j abcde f ghi j
abcde f |gh|i j abcde f g|hi j abcd|e f g|h|i| j
abcde f |ghi| j abcd|e f gh|i j
abcde f |ghi j

are identical ex ante, the game admits a symmetric equilibrium coalition struc-
ture which is generically unique up to a permutation of the players. I also provide
examples to show that stationary perfect equilibria may fail to exist in general
valuations and that the noncooperative approach followed here is unrelated to
standard cooperative game-theoretic solution concepts.
The determination of the equilibrium coalition structure in the sequential game
of coalition formation is driven by two basic features of the extensive form.
First, the exogenous rule of order imposes a fixed order of moves by players
in the game. Depending on the valuation, players may have an advantage in
moving first, second or in any other position in the game. The rule of order thus
creates an asymmetry among players which is determined outside the game.
An important direction for future research is to eliminate this asymmetry and
to explore conditions under which the equilibrium of the extensive form game
is independent of the rule of order. This line of research has been pursued by
Moldovanu and Winter (1995) in the context of games of coalitional bargaining.
They show that order independent equilibria only exist when the underlying
118 FRANCIS BLOCH

game in characteristic function form, as well as all its restrictions, have nonempty
cores.
The second important feature of the extensive form is the commitment power
of the players. I assume that, by accepting the offer to join a coalition, players are
bound to remain in that coalition whatever coalition structure the other players
may form. This implies that coalitions are formed one after another and that
coalitions may not compete to attract members. In fact, this sequential structure
of the process of coalition formation is the feature of the extensive form which
guarantees the existence of an equilibrium. Extensive form games where players
do not commit to stay in a coalition can easily be constructed. The existence and
characterization of equilibria in these games constitutes a difficult but important
area for future research.
Finally, the model analyzed in this paper assumes that the coalitional worth is
divided according to a fixed sharing rule. While this approach greatly simplifies
the analysis, it clearly restricts the applicability of the model. The study of ex-
tensive form procedures allowing players to bargain over the worth of coalitions
seems to me to be the foremost topic for future research.

APPENDIX: PROOF OF PROPOSITION 5.1

The proof consists in three steps. In the first two steps, I explicitly construct
the stationary perfect equilibria of the game. Observe first that the only payoff-
relevant part of any history of the game is the number of coalitions which have
already been formed. To fix notations, let K be the number of coalitions already
formed, and m be the number of remaining players in the game, after a given
history.
Step 1. After a given history, suppose that K coalitions have been formed,
and that m players remain in the game. Suppose furthermore that, if a coalition
of size µ is formed, the remaining m − µ players remain isolated. Then the
optimal choice of µ is given by:
µ∗ = 1 if m ≤ (K + 1)2

µ∗ = m if m ≥ (K + 1)2
Given that the remaining m − µ players form singletons, the optimal number
of players in a coalition, µ∗ , solves:
(α − λ)2
max F(µ) =
µ(K + m − µ + 2)2

subject to 1 ≤ µ ≤ m.
 SEQUENTIAL FORMATION OF COALITIONS 119

The function 1/µ(K + m − µ + 2)2 is strictly decreasing for 1 ≤ µ ≤

(K + m + 2)/3, and strictly increasing for (K + m + 2)/3 ≤ µ ≤ m. Hence, the
optimal choice µ∗ is either 1 or m. Now,

(α − λ)2
F(1) =
(K + m + 1)2

(α − λ)2
F(m) = .
m(K + 2)2
Solving the quadratic in m, I obtain:

F(1) ≤ F(m) if and only if m ≥ (K + 1)2 .

Step 2. The game admits two stationary perfect equilibria, given by

Strategy 1.

If m < (K + 1)2 choose µ = 1

If (K + 1)2 ≤ m < (K + 2)2 + 1 choose µ = m

If (K + 2)2 + 1 ≤ m choose µ = 1

Strategy 2.

If m ≤ (K + 1)2 choose µ = 1

If (K + 1)2 < m ≤ (K + 2)2 + 1 choose µ = m

If (K + 2)2 + 1 < m choose µ = 1.

The two equilibria only differ in the rules chosen to break ties. In the first
equilibrium, if a player is indifferent between forming a cartel of size m or
forming a singleton, she chooses to form a cartel. In the second equilibrium, she
chooses to remain isolated. In the remainder of the proof, I focus on strategy 1,
and show that given that ties are broken according to the rule that indifferent
players choose to form coalitions, strategy 1 is the unique stationary perfect
equilibrium of the game.
The proof is by induction on the number of remaining players in the game.
If m = 2, the player before last chooses whether to form a cartel of size 2 or
to remain isolated, in which case the last player remains isolated as well. Since
K ≥ 0, 2 < (K + 2)2 + 1. Hence strategy 1 prescribes that a cartel is formed if
and only if 2 ≥ (K + 1)2 , and by Step 1, this is the unique optimal strategy for
the player before last.
120 FRANCIS BLOCH

Suppose now that, for any m 0 < m, strategy 1 is the unique equilibrium
strategy. Consider the different possibilities with m players remaining in the
game.
If m < (K + 1)2 , then ∀m 0 < m, m 0 < (K + 2)2 . Hence, whatever coalition
the player forms, all subsequent players choose to remain isolated. Then, by
Step 1, the unique optimal strategy is to choose to form a singleton.
If now (K + 1)2 ≤ m < (K + 2)2 + 1, similarly, ∀m 0 < m, m 0 < (K + 2)2 .
Hence, irrespective of the coalition formed by the player, the subsequent players
choose to remain isolated and, by Step 1, since m ≥ (K + 1)2 , the player should
choose to form a coalition of size m.
Finally, when m ≥ (K + 2)2 + 1, different possibilities have to be considered.
The player may either choose to form a coalition µ such that (m −µ) ≥ (K +2)2 ,
in which case the remaining players form a coalition, or a coalition µ such that
(m − µ) < (K + 2)2 , in which case the remaining players choose to remain
separate.
When the coalition size µ is such that (m −µ) < (K +2)2 , the player’s payoff
is given by:

(α − λ)2
F(µ) = .
µ(K + 2 + m − µ)2

From Step 1, since m > (K +1)2 , the optimal choice of coalition size is µ∗ = m.
In the case where µ is chosen small enough, other players form a coalition
later in the game. Given the specification of the strategy, after the formation of
the coalition of size µ, a group of players will choose to remain separate, and
the last players will form a single coalition. The number of players who choose
to remain isolated, ν, is the unique integer satisfying:

(K + 2 + ν)2 ≤ (m − µ − ν) < (K + 3 + ν)2 + 1.

A simple computation shows that ν is the first integer following:

√
∗ 9 + 4(K + m − µ) − (2K + 5)
ν = .
2
Hence, the payoff to a player who chooses a coalition of size µ where m − µ ≤
(K + 2)2 is given by:

(α − λ)2
G(µ) = ,
µ(K + 3 + ν ∗ )2
or
(α − λ)2
G(µ) = √ .
µ( 9 + 4(K + m − µ) + 1)2
 SEQUENTIAL FORMATION OF COALITIONS 121

The optimal value µ∗ is thus the minimum over the interval [1, m − (K + 2)2 ]
of the function
p
H (µ) = µ( 9 + 4(K + m − µ) + 1)2 .
Next consider the derivative H 0 of H ,
p p
H 0 (µ) = ( 9 + 4(K + m − µ) + 1)( 9 + 4(K + m − µ)
4µ
+1− √ .
9 + 4(K + m − µ)
A study of the sign of H 0 shows
√ that the function H is increasing up to the value
µ = (16K + 16m + 35 + 73 + 32K + 32m)/32, and decreasing thereafter.
Hence, the optimal choice of µ, µ∗ , is either µ∗ = 1, or µ∗ = m −(K +2)2 . Now,
a simple computation shows that the choice µ∗ = m − (K + 2)2 is dominated
by µ∗ = m.
To complete this step, it suffices to show that µ∗ = 1 is the optimal choice,
that is that H (1) ≤ m(K + 2)2 .
√
H (1) = 14 ( 9 + 4(K + m − 1) + 1)2
√
= 1
2
(3 + 2K + 2m + 9 + 4(K + m − 1)

< 1
2
(3 + 2K + 2m + 9 + 4(K + m − 1))

< (3K + 3m + 4).

Hence,
m(K + 2)2 − H (1) > m(K + 2)2 − (3K + 3m + 4)

> m(K 2 + 4K + 1) − 3K − 4

> (K + 2)2 (K 2 + 4K + 1) − 3K − 4

> 0.
Step 3. The coalition structure generated by the stationary perfect equilibria
corresponding to strategies 1 and 2 is given by π ∗ = (T1∗ , { j} j6∈T1∗ ) where t1∗ is
√ √
the first integer following (2n + 3 − √4n + 5)/2. (If 4n + 5 is √ an integer, t1∗
can take on the two values (2n + 3 − 4n + 5)/2 and (2n + 5 − 4n + 5)/2.)
When K = 0, strategy 1 prescribes that the first player forms a singleton. In
fact, singletons will continue to be formed as long as m ≥ (n − m + 2)2 + 1.
The unique coalition formed will comprise t ∗ members, where t ∗ is the unique
integer such that
(n − t ∗ + 1)2 ≤ t ∗ < (n − t ∗ + 2)2 + 1.
122 FRANCIS BLOCH
√
It is easy to show that t ∗ is the first integer following ((2n + 3) − 4n + 5)/2.
The only possible difficult arises when there exist two integers, t1∗ and t2∗ such
that t1∗ = (n − t1∗ + 1)2 and t2∗ = (n − t2∗ + 2)2 + 1. Then, strategy 1 prescribes
that a coalition of size t1∗ is formed whereas strategy 2 induces a coalition of size
t2∗ .

REFERENCES

Aumann, R. (1967), “A Survey of Games without Side Payments,” in Essays in Mathematical Eco-
nomics (M. Shubik, Ed.), pp. 3–27. Princeton, NJ: Princeton Univ. Press.
Aumann, R., and J. Drèze (1974). “Cooperative Games with Coalition Structures,” Int. J. Game Theory
3, 217–237.
Aumann, R., and M. Maschler (1964). “The Bargaining Set for Cooperative Games,” in Advances in
Game Theory (M. Dresher, L. Shapley, and A. Tucker, Eds.), pp. 443–447. Princeton, NJ: Princeton
Univ. Press.
Aumann, R., and R. Myerson (1988). “Endogenous Formation of Links Between Players and of Coali-
tions: An Application of the Shapley Value” in The Shapley Value: Essays in Honor of Lloyd Shapley
(A. Roth, Ed.), pp. 175–191. Cambridge, U.K.: Cambridge Univ. Press.
Bloch, F. (1995). “Endogenous Structures of Association in Oligopolies,” Rand J. Econ. 26, 537–556.
Chatterjee, K., B. Dutta, D. Ray, and K. Sengupta (1993). “A Noncooperative Theory of Coalitional
Bargaining,” Rev. Econ. Stud. 60, 463–477.
Fudenberg, D., and D. Levine (1983). “Subgame Perfect Equilibria of Finite and Infinite Horizon
Games,” J. Econ. Theory 31, 251–268.
Greenberg, J. (1995). “Coalition Structures,” in Handbook of Game Theory with Applications, Vol. II
(R. Aumann and S. Hart, Eds.), pp. 1305–1337. Amsterdam: North Holland.
Guesnerie, R., and C. Oddou (1981). “Second Best Taxation as a Game,” J. Econ. Theory 25, 67–91.
Hart, S., and M. Kurz (1983). “Endogenous Formation of Coalitions,” Econometrica 51, 1047–1064.
Hart, S., and M. Kurz (1984). “Stable Coalition Structures” in Coalitions and Collective Action
(M. Holler, Ed.), pp. 236–258. Vienna: Physica Verlag.
Kurz, M. (1988). “Coalitional Value,” in The Shapley Value: Essays in Honor of Lloyd Shapley (A. Roth
Ed.), pp. 155–173. Cambridge, U.K.: Cambridge Univ. Press.
Moldovanu, B. (1992). “Coalition-Proof Nash Equilibria and the Core in Three-Player Games,” Games
Econom. Behav. 4, 565-581.
Moldovanu, B., and E. Winter (1995). “Order Independent Equilibria,” Games Econom. Behav. 9,
21–34.
Myerson, R. (1977). “Graphs and Cooperation in Games,” Math. Oper. Res. 2, 225–229.
Myerson, R. (1978). “Threat Equilibria and Fair Settlements in Cooperative Games,” Math. Oper. Res.
3, 265–274.
Owen, G. (1977). “Value of Games With A Priori Unions” in Essays in Mathematical Economics and
Game Theory (R. Hein, and O. Moeschlin, Eds.), pp. 76–88. New York: Springer-Verlag.
Ray, D., and R. Vohra (1995). “Binding Agreements and Coalitional Bargaining,” mimeo, Department
of Economics, Boston University and Brown University.
Rubinstein, A. (1982).“Perfect Equilibrium in a Bargaining Model,” Econometrica 50, 97–109.
Salant, S., S. Switzer, and R. Reynolds (1983). “Losses from Horizontal Mergers: The Effects of
 SEQUENTIAL FORMATION OF COALITIONS 123

an Exogenous Change in Industry Structure on Cournot-Nash Equilibrium,” Quart. J. Econ. 98,

185–199.
Selten, R. (1981). “A Noncooperative Model of Characteristic Function Bargaining” in Essays in Game
Theory and Mathematical Economics in Honor of Oskar Morgenstern (V. Bohm, and H. Nachtkamp,
Eds.), pp. 131–151. Mannheim: Bibliographisches Institut Mannheim.
Shenoy, P. (1979). “On Coalition Formation: A Game Theoretical Approach,” Int. J. Game Theory 8,
133–164.
Shubik, M. (1982). Game Theory in the Social Sciences. Cambridge, MA: MIT Press.
Stigler, G. (1968). The Organization of Industry, Homewood, IL: Irwin.
Sutton, J. (1986). “Noncooperative Bargaining Theory: An Introduction,” Rev. Econ. Stud. 53, 709–724.
Thrall, R., and W. Lucas (1963). “N-Person Games in Partition Function Form,” Nav. Res. Logist.
Quart. 10, 281–298.
Von Neumann, J., and O. Morgenstern (1944). Theory of Games and Economic Behavior, Princeton,
NJ: Princeton Univ. Press.
Winter, E. (1993). “Mechanism Robustness in Multilateral Bargaining,” mimeo, Center for Rationality
and Interactive Decision Theory, The Hebrew University of Jerusalem.
Yi, S. S., and H. S. Shin (1995). “Endogenous Formation of Coalitions. Part I: Theory,” mimeo,
Department of Economics, Dartmouth College.