Bargaining Theory with Applications

The first unified and systematic treatment of the modern theory of bargain-
ing, presented together with many examples of how that theory is applied
in a variety of bargaining situations.
Abhinay Muthoo provides a masterful synthesis of the fundamental re-
sults and insights obtained from the wide-ranging and diverse (game theo-
retic) bargaining theory literature. Furthermore, he develops new analyses
and results, especially on the relative impacts of two or more forces on the
bargaining outcome. Many topics — such as inside options, commitment
tactics and repeated bargaining situations — receive their most extensive
treatment to date. In the concluding chapter, he offers pointers towards
future research.
Bargaining Theory with Applications is a textbook for graduate students
in economic theory and other social sciences and a research resource for
scholars interested in bargaining situations.

Abhinay Muthoo is Professor of Economics at the University of Essex. He

was educated at the London School of Economics and the University of Cam-
bridge. Professor Muthoo has published papers on bargaining theory and
game theory, among other topics, in journals such as Review of Economic
Studies, Journal of Economic Theory, Games and Economic Behavior and
Economic Journal.
Bargaining Theory
with Applications

ABHINAY MUTHOO
University of Essex

CAMBRIDGE
UNIVERSITY PRESS
PUBLISHED BY THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge, United Kingdom
CAMBRIDGE UNIVERSITY PRESS
The Edinburgh Building, Cambridge CB2 2RU, UK
40 West 20th Street, New York, NY 10011-4211, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
Ruiz de Alarcon 13, 28014 Madrid, Spain
Dock House, The Waterfront, Cape Town 8001, South Africa
http://www.cambridge.org

© Abhinay Muthoo 1999

This book is in copyright. Subject to statutory exception
and to the provision of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.

First Published 1999

Reprinted 2002

Typeface Computer Modern llpt. System ET^X2e [Typeset by the author]

A catalogue record for this book is available from the British Library

Library of Congress Cataloguing in Publication data applied for

ISBN 0 521 572258 hardback

ISBN 0 521 576474 paperback

Transferred to digital printing 2004

To my parents
Contents

Preface xiii

1 Preliminaries 1
1.1 Bargaining Situations and Bargaining 1
1.2 Outline of the Book 3
1.3 The Role of Game Theory 6
1.4 Further Remarks 6

2 The Nash Bargaining Solution 9

2.1 Introduction 9
2.2 Bargaining over the Partition of a Cake 10
2.2.1 Characterization 12
2.2.2 Examples 15
2.3 Applications 16
2.3.1 Bribery and the Control of Crime 16
2.3.2 Optimal Asset Ownership 17
2.4 A General Definition 22
2.4.1 Characterization 24
2.5 Applications 25
2.5.1 Union-Firm Negotiations 25
2.5.2 Moral Hazard in Teams 27
viii Contents

2.5.3 Bribery and the Control of Crime: An Extension 29

2.6 Axiomatic Foundation 30
2.7 An Interpretation 33
2.8 Asymmetric Nash Bargaining Solutions 35
2.9 Appendix: Proofs 37
2.10 Notes 39

3 The Rubinstein Model 41

3.1 Introduction 41
3.2 The Basic Alternating-Offers Model 42
3.2.1 The Unique Subgame Perfect Equilibrium 43
3.2.2 Proof of Theorem 3.1 47
3.2.3 Properties of the Equilibrium 50
3.2.4 The Value and Interpretation of the Alternating-Offers
Model 53
3.3 An Application to Bilateral Monopoly 55
3.4 A General Model 59
3.4.1 The Subgame Perfect Equilibria 60
3.4.2 Small Time Intervals 64
3.4.3 Relationship with Nash's Bargaining Solution 65
3.4.4 Proof of Theorems 3.2 and 3.3 67
3.5 An Application to a Two-Person Exchange Economy 69
3.6 Notes 71

4 Risk of Breakdown 73
4.1 Introduction 73
4.2 A Model with a Risk of Breakdown 74
4.2.1 The Unique SPE when both Players are Risk Neutral 75
4.2.2 The Unique SPE with Risk Averse Players 77
4.3 An Application to Corruption in Tax Collection 81
4.4 The Effect of Discounting 85
4.4.1 Small Time Intervals 87
4.4.2 Risk Neutral Players: Split-The-Difference Rule 89
4.5 An Application to Price Determination 91
4.6 A Generalization 95
4.7 Notes 96
Contents ix

5 Outside Options 99
5.1 Introduction 99
5.2 A Model with Outside Options 100
5.2.1 Relationship with Nash's Bargaining Solution 104
5.3 Applications 105
5.3.1 Relationship-Specific Investments 105
5.3.2 Sovereign Debt Negotiations 107
5.3.3 Bribery and the Control of Crime Revisited 109
5.4 The Effect of a Risk of Breakdown 110
5.4.1 The Unique Subgame Perfect Equilibrium 111
5.4.2 Relationship with Nash's Bargaining Solution 113
5.4.3 The Impact of The Manner of Disagreement 114
5.4.4 A Generalization 115
5.5 Searching for Outside Options 116
5.5.1 Searching on the Streets 118
5.5.2 Searching while Bargaining 121
5.6 The Role of the Communication Technology 124
5.6.1 Equilibria in the Telephone Game 125
5.6.2 An Application to Relationship-Specific Investments 130
5.6.3 Rubinstein Bargaining with Quit Options 131
5.7 Appendix: Proofs 133
5.8 Notes 135

6 Inside Options 137

6.1 Introduction 137
6.2 A Model with Inside Options 138
6.3 Applications 143
6.3.1 Takeovers in a Duopolistic Market 143
6.3.2 Sovereign Debt Renegotiations 144
6.4 The Effect of Outside Options 146
6.4.1 And a Risk of Breakdown 149
6.4.2 Relationship with Nash's Bargaining Solution 151
6.4.3 A Generalization 152
6.5 An Application to Intrafamily Allocation 154
6.6 Endogenously Determined Inside Options 158
6.6.1 Stationary Equilibria 159
x Contents

6.6.2 Markov Equilibria 160

6.6.3 Uniqueness of SPE and Non-Markov Equilibria 165
6.7 An Application to Wage Renegotiations 170
6.7.1 Multiple Pareto-Efficient Equilibria 171
6.7.2 Equilibria with Strikes 173
6.8 Appendix: Proofs 174
6.9 Notes 185

7 Procedures 187
7.1 Introduction 187
7.2 Who Makes Offers and When 188
7.2.1 The Ultimatum Game 189
7.2.2 Repeated Offers 190
7.2.3 Simultaneous Offers 191
7.2.4 Random Proposers 192
7.2.5 Alternating-Offers with Different Response Times 193
7.3 The Effect of Retractable Offers 194
7.3.1 A Subgame Perfect Equilibrium 195
7.3.2 On the Uniqueness of the Equilibrium 195
7.3.3 Multiple Equilibria and Delay 197
7.3.4 Discussion and Interpretation 198
7.4 Burning Money: A Tactical Move 200
7.4.1 A Subgame Perfect Equilibrium 201
7.4.2 On the Uniqueness of the Equilibrium 201
7.4.3 Multiple Equilibria 202
7.4.4 Equilibrium Delay 204
7.4.5 Discussion 208
7.4.6 An Application to Surplus Destruction 208
7.5 Notes 209

8 Commitment Tactics 211

8.1 Introduction 211
8.2 The Basic Model 214
8.2.1 The Formal Structure 214
8.2.2 Interpretation 217
8.2.3 The Equilibrium 218
Contents xi

8.3 Discussion 222

8.3.1 Properties of the Equilibrium 222
8.3.2 Relationship with Nash's Bargaining Solution 223
8.3.3 Comparison with the Nash Demand Game 224
8.3.4 Robustness 225
8.3.5 A Generalization 227
8.4 An Application to Delegation 230
8.5 Uncertainty and Simultaneous Concessions 232
8.5.1 A Model with Simultaneous Concessions 233
8.5.2 An Example: Two Bargaining Positions 234
8.5.3 A Generalization 236
8.6 Uncertainty and Wars of Attrition 240
8.6.1 A Model with Wars of Attrition 241
8.6.2 Equilibrium in the War of Attrition Subgames 242
8.6.3 Equilibrium Partial Commitments 246
8.7 Notes 248

9 Asymmetric Information 251

9.1 Introduction 251
9.2 Efficiency under One-Sided Uncertainty 253
9.2.1 The Case of Private Values 254
9.2.2 The Case of Correlated Values 256
9.3 Applications 262
9.3.1 Efficient Wage Agreements 262
9.3.2 Litigation or Out-of-Court Settlement 263
9.4 Efficiency under Two-Sided Uncertainty 265
9.4.1 The Case of Private Values 266
9.4.2 The Case of Correlated Values 268
9.5 Applications 270
9.5.1 Indefinite Strikes 270
9.5.2 Litigation or Out-of-Court Settlement Revisited 271
9.6 Bargaining Power and Uncertainty 271
9.6.1 An Example of a Screening Equilibrium 273
9.6.2 General Results 280
9.6.3 The Effect of Retractable Offers 285
9.7 An Application to Wage-Quality Contracts 289
xii Contents

9.7.1 The Commitment Equilibrium 290

9.7.2 The Unique Perfect Bayesian Equilibrium 291
9.8 Notes 292

10 Repeated Bargaining Situations 295

10.1 Introduction 295
10.2 A Basic Repeated Bargaining Model 297
10.2.1 The Unique Stationary Subgame Perfect Equilibrium 300
10.2.2 Small Time Intervals Between Consecutive Offers 304
10.2.3 Comparison with a Long-Term Contract 307
10.2.4 Non-Stationary Subgame Perfect Equilibria 309
10.3 An Application to Dynamic Capital Investment 312
10.4 The Role of Outside Options 316
10.4.1 An Application to Firm Provided General Training 319
10.5 The Role of Long-Term Contracts 321
10.5.1 Equilibrium Without a Long-Term Contract 322
10.5.2 Equilibrium With a Complete Long-Term Contract 323
10.5.3 Equilibrium With an Incomplete Long-Term Contract 324
10.6 Reputation Effects 327
10.6.1 A Perfect Bayesian Equilibrium in a Simple Model 328
10.6.2 Further Remarks 329
10.7 Notes 330

11 Envoi 333
11.1 Introduction 333
11.2 Omissions 334
11.2.1 Non-Stationary and Stochastic Environments 335
11.2.2 Multilateral and Coalitional Bargaining 336
11.2.3 Arbitration and Mediation 338
11.2.4 Multiple Issues and the Agenda 339
11.2.5 Enforceability of Agreements 339
11.3 Thorny Issues 340
11.4 On the Role of Experiments 341

References 345

Index 354
Preface

Ariel Rubinstein's contribution to bargaining theory in Rubinstein (1982)

captured the imagination of the economics profession. The origins of most
papers in the wide-ranging and diverse literature that has since developed
can be traced back to that seminal paper; even those papers that cannot
have probably been inspired by the literature that has. At the same time
as the development of the theory of bargaining, applied economic theorists
have used models from this literature to construct models of a variety of
economic phenomena that had hitherto not been studied at all, or not been
studied properly. There is now a large literature that contains applications
of that bargaining theory.
With the exception of John Nash's path-breaking contributions to bar-
gaining theory in Nash (1950, 1953), much of the material in this book is
based upon and/or inspired by the literature (theoretical and applied) that
has developed since 1982.
I have written this book with two main objectives in mind. Firstly, from
a theoretical perspective, I synthesize, and organize into a coherent and
unified picture, the main fundamental results and insights obtained from
the bargaining theory literature. The chapters are organized around the
main forces that determine the bargaining outcome. I not only analyse the
impact on the bargaining outcome of each force, but I also often analyse
the relative impacts of two or more forces. And, secondly, from an applied
perspective, I show how the theory can be fruitfully applied to a variety of
xiv Preface

economic phenomena.
In order to achieve the first of the two objectives stated above, I have
had to take stock of, and reflect upon, the bargaining theory literature. In
the process of doing so, it has been necessary to conduct some new analyses
(not contained in the literature) — especially in order to develop an un-
derstanding of the relative impacts of two or more forces on the bargaining
outcome.
Since this book provides a unified treatment of bargaining theory and
contains new results, it is part textbook and part research monograph. As
such this book should be useful not only to graduate students and pro-
fessional applied economic and political theorists interested in bargaining
situations (that arise in many areas of economics and politics), but also to
bargaining and game theorists. This book can be used to learn bargaining
theory and to improve one's understanding of it. Furthermore, it should
help researchers apply that theory and/or construct their own models of the
specific real-life bargaining situations that interest them.
Chapter 1 introduces some basic issues and provides an outline of the
book. The theory and application of bargaining are developed in Chapters
2-10. The final (concluding) chapter, Chapter 11, draws attention to some
of the main omissions and weaknesses of the theory developed in this book,
and identifies specific avenues and topics for future research in the further
development of the theory and application of bargaining.

Acknowledgements
I began working on this book in October 1995, after much encouragement
from Ariel Rubinstein. I owe my greatest debt to him for several reasons,
besides that. Firstly, of course, because of his written contributions, which
have not only made this book possible, but have influenced my own work
and thinking on the subject. Secondly, because I have had the privilege of
discussing bargaining theory with him ever since I was a graduate student
at the University of Cambridge. I have learned a great deal and obtained
much insight from our discussions.
Ken Binmore has also played an important role in my work and thinking,
both through his important and insightful contributions and through our
discussions. He also provided encouragement to write this book.
While writing the book, I have accumulated many debts. Many friends
Preface xv

and colleagues made very detailed comments on several chapters that led
me to substantially revise them. They include Roy Bailey (Chapters 1, 2-4
and 11), Craig Brett (Chapters 1, 5, 6 and 11), Vince Crawford (Chapter 8),
Martin Cripps (Chapters 2 and 3), Leonardo Felli (Chapter 10), Shinsuke
Kambe (Chapters 7-9), Ben Lockwood (Chapters 5 and 6), Martin Osborne
(Chapters 3, 4, 7 and 8) and Anders Poulsen (Chapters 3 and 7).
Craig Brett, Roy Bailey, Ken Binmore, Vince Crawford, Osvaldo Fe-
instein, Drew Fudenberg, Oliver Hart, Jim Malcomson, Martin Osborne,
Alvin Roth and Ariel Rubinstein provided me with some helpful advice
and/or comments on some aspects of the book.
Without access to I^TgX, I think I would not have written this book. I
owe much gratitude to Roy Bailey, who (a few years earlier) introduced me
to this amazing software package, helped me learn it, and, while writing the
book, helped me when I had queries on how to do this or that. Craig Brett
also kindly provided some help in this latter respect.
Patrick McCartan was the economics editor at CUP for much of the time
while I was writing the book, before he moved on to head the CUP journals
department. Patrick's enthusiasm for the book and general support gave
me much encouragement. The anonymous referees (organized by Patrick)
provided me with some very useful specific and general comments. Ashwin
Rattan, who succeeded Patrick as the CUP economics editor, very kindly
and efficiently steered the completed typescript through the various (editing
and production) stages at the Press. It has been a pleasure to work with
the staff at CUP.
Last, but certainly not least, my family provided me with much encour-
agement and support.

Abhinay Muthoo
University of Essex
 Preliminaries

1.1 Bargaining Situations and Bargaining

Consider the following situation. Individual S owns a house that she values
at £50,000 (which is the minimum price at which she would sell it). In-
dividual B values this house at £70,000 (which is the maximum price at
which she would buy it). If trade occurs — that is, if individual S sells the
house to individual B — at a price that lies between £50,000 and £70,000,
then both the seller (individual S) and the buyer (individual B) would be-
come better off. This means that in this situation the two individuals have
a common interest to trade. But, at the same time, they have conflicting
interests over the price at which to trade: the seller would like to trade at a
high price, while the buyer would like to trade at a low price. Any exchange
situation, such as the one just described, in which a pair of individuals (or,
organizations) can engage in mutually beneficial trade but have conflicting
interests over the terms of trade is a bargaining situation.
Stated in general and broad terms, a bargaining situation is a situation
in which two players1 have a common interest to co-operate, but have con-
flicting interests over exactly how to co-operate. To put it differently, the
players can mutually benefit from reaching agreement on an outcome from
a set of possible outcomes (that contains two or more elements), but have
1
A 'player' can be either an individual, or an organization (such as a firm, or a country).
2 Preliminaries

conflicting interests over the set of outcomes.

There are two main reasons for studying bargaining situations. The first,
practical, reason is that many important and interesting human (economic,
social and political) interactions are bargaining situations. As mentioned
above, exchange situations (which characterize much of human economic
interaction) are bargaining situations. In the arena of social interaction,
a married couple, for example, are involved in many bargaining situations
throughout their relationship. In the political arena, a bargaining situation
exists, for example, when no single party on its own can form a government
(such as when there is a hung parliament); the party that has obtained
the most votes will typically find itself in a bargaining situation with one
of the other parties. The second, theoretical, reason for studying bargain-
ing situations is that understanding such situations is fundamental to the
development of the economic theory of markets.
The main issue that confronts the players in a bargaining situation is
the need to reach agreement over exactly how to co-operate — before they
actually co-operate (and obtain the fruits of that co-operation). On the one
hand, each player would like to reach some agreement rather than disagree
and not reach any agreement. But, on the other hand, each player would
like to reach an agreement that is as favourable to her as possible. It is thus
conceivable that the players will strike an agreement only after some costly
delay, or indeed fail to reach any agreement — as is witnessed by the history
of disagreements and costly delayed agreements in many real-life bargaining
situations (as exemplified by the occurrences of trade wars, military wars,
strikes and divorce).
Bargaining is any process through which the players on their own try to
reach an agreement. This process is typically time consuming and involves
the players making offers and counteroffers to each other. If the players
get a third party to help them determine the agreement, then this means
that agreement is not reached via bargaining (but, for example, via some
arbitration process). The theory developed in this book concerns bargaining
situations in which the outcome is determined entirely via some bargaining
process. The role of arbitrators and mediators in helping the players reach
agreement is briefly discussed in the final chapter.
A main focus of any theory of bargaining is on the efficiency and distri-
bution properties of the outcome of bargaining. The former property relates
1.2 Outline of the Book 3

to the possibility that the bargaining outcome is not Pareto efficient. As

indicated above, this could arise, for example, either because the players fail
to reach an agreement, or because they reach an agreement after some costly
delay. Examples of costly delayed agreements include: when a wage agree-
ment is reached after lost production due to a long strike, and when a peace
settlement is negotiated after the loss of life through war. The distribution
property, on the other hand, relates to the issue of exactly how the fruits
of co-operation are divided between the players (or, to put it differently,
how the gains from trade are divided). The theory developed in this book
determines the roles of various forces on the bargaining outcome (and, in
particular, on these two properties). As such it addresses the issue of what
determines a player's bargaining power.

1.2 Outline of the Book

A basic, intuitive, observation is that if the bargaining process is frictionless
— by which I mean, in particular, that neither player incurs any cost dur-
ing the bargaining process — then each player may continuously demand
(without incurring any cost) that agreement be struck on terms that are
most favourable to her. For example, in the exchange situation described at
the beginning of Section 1.1, the seller may continuously demand that trade
take place at the price of £69,000, while the buyer may continuously demand
that trade take place at the price of £51,000. It may therefore be argued
that the outcome of a frictionless bargaining process is indeterminate, since
the players may have no incentive to compromise and reach an agreement.
Consequently, it would seem hopeless to construct a theory of bargaining —
that determines the outcome of bargaining in terms of the primitives of the
bargaining situation (such as the set of possible agreements and the players'
preferences over this set) — based on frictionless bargaining processes.
Fortunately, in most real-life bargaining situations the bargaining process
is not frictionless. A basic source of the cost incurred by a player while
bargaining — that provides some friction in the bargaining process — comes
from the twin facts that bargaining is time consuming and time is valuable
to the player. Rubinstein's bargaining model — which is the subject of
study in Chapter 3 — is a formal exploration of the role of the players'
discount rates (that represent their values for time) in a time-consuming,
4 Preliminaries

offer-counteroffer process. It is shown that, indeed, the players will reach

an (immediate) agreement if and only if time is valuable to at least one
of the two players. Furthermore, a number of other fundamental results
and insights are obtained from the study of this model, including those
concerning the role of the relative magnitude of the players' discount rates
on the terms of the agreement. It is worth pointing out that these results
(and many others derived in the book) — although, in hindsight, are rather
intuitive — were not obtainable without formally modelling the bargaining
process. I should also emphasize here (although this will become clear in
Chapter 3) that Rubinstein's bargaining model provides the basic framework
that is extended and/or adapted (in several later chapters) to address the
roles of various other forces.
Another basic source of the cost incurred by a player while bargaining
comes from the possibility that the negotiations might randomly and ex-
ogenously breakdown in disagreement. Even if the probability of such an
occurrence is small, it nevertheless provides some friction in the bargaining
process — and as such may provide appropriate incentives to the players to
compromise and reach an agreement. The role of such an exogenous risk
of breakdown is studied in Chapter 4. I also explore the interplay of this
force with the players' discount rates, and study their relative impacts on
the bargaining outcome.
In many bargaining situations the players may have access to outside
options and/or inside options. For example, in the exchange situation de-
scribed above the seller may have a non-negotiable (fixed) price offer on the
house from a different buyer, and she may derive some utility while she lives
in the house. The former is her outside option, while the latter her inside
option. I should emphasize that when, and if, the seller exercises her out-
side option, the negotiations between individuals B and S terminate forever
in disagreement. In contrast, the seller's inside option describes her (flow)
utility while she temporarily disagrees with individual B over the price at
which to trade. The role of outside options is studied in Chapter 5, while
the role of inside options in Chapter 6. I also study the interplay amongst
the players' discount rates, outside options, inside options and an exogenous
risk of breakdown.
An important set of questions addressed in Chapters 3-6 are why, when
and how to apply Nash's bargaining solution, where the latter is described
1.2 Outline of the Book 5

and studied in Chapter 2. It is shown that under some circumstances, when

appropriately applied, Nash's bargaining solution describes the outcome of
a variety of bargaining situations. These results are especially important
and useful in applications, since it is often convenient for applied economic
theorists to describe the outcome of a bargaining situation — which may
be one of many ingredients of their economic models — in a simple (and
tractable) manner.
The procedure of bargaining constitutes the rules of the bargaining pro-
cess, and includes matters such as who makes offers and when. It seems
self-evident that the bargaining outcome will depend on the bargaining pro-
cedure. I study the impact that various specific procedural features have on
the bargaining outcome in Chapter 7. However, the thorny issue of what
or who determines the procedure is left unanswered. In the final chapter, I
return to this difficult issue.
The role of bargaining tactics is taken up in Chapter 8, where the focus
is on a particular type of tactic, known as the commitment tactic. The
basic model studied here establishes a number of fundamental results and
insights. In particular, it formalizes the notion that (in bargaining situa-
tions) weakness can often be a source of strength. An important aspect of
many bargaining processes is the making of claims followed by concessions,
which leads me to also study wars of attrition based bargaining models.
In Chapter 9 I study the role of asymmetric information. In particular,
I explore whether or not the presence of asymmetric information necessarily
implies that the bargaining outcome is inefficient. Furthermore, I explore
the impact that asymmetric information has on the players' respective bar-
gaining powers.
In the preceding chapters the focus is on 'one-shot' bargaining situations.
In Chapter 10 I study 'repeated' bargaining situations in which the play-
ers have the opportunity to be involved in a sequence of (possibly different
and/or interdependent) bargaining situations. After studying repeated bar-
gaining models, I explore whether or not the players might wish to commit
themselves to a long-term relationship by writing a long-term contract. I
then explore the notion that in such repeated bargaining situations a player
might build a reputation for being a particular type of bargainer.
In Chapters 2-10 I develop a theory of bargaining, and apply that theory
to a variety of bargaining situations. I should emphasize that the focus is
6 Preliminaries

on fundamentals. Furthermore, the models studied are particularly simple,

so as to bring out the main fundamental results and insights in a simple
but rigorous manner. I conclude in Chapter 11, where I describe some of
the main omissions and weaknesses of the theory developed in this book.
In particular, I identify potential avenues and topics for future research.
Furthermore, I offer some comments on the role of bargaining experiments.

1.3 The Role of Game Theory

A bargaining situation is a game situation in the sense that the outcome of
bargaining depends on both players' bargaining strategies: whether or not
an agreement is struck, and the terms of the agreement (if one is struck),
depends on both players' actions during the bargaining process. It is there-
fore natural to study bargaining situations using the methodology of game
theory. Indeed, almost all of the bargaining models studied in this book
are game-theoretic models. In particular, a bargaining situation is modelled
as an extensive-form game. When there is no asymmetric information, I
characterize its Nash equilibria if the bargaining game is static and its sub-
game perfect equilibria if it is dynamic. On the other hand, when there is
asymmetric information, I characterize its Bayesian Nash equilibria if it is
static and its perfect Bayesian equilibria if it is dynamic.
Although — as I briefly discuss in the final chapter — there are several
important weaknesses with the game-theoretic methodology, as it currently
stands, its strengths are considerable. In particular, it is currently the best
available tool with which one can formalize the phenomena under consider-
ation, and conduct a deep, insightful and rigorous investigation of the role
of various forces on the bargaining outcome.

1.4 Further Remarks

Most of the models studied in this book are dynamic games with perfect
information. As indicated above, I use the subgame perfect equilibrium
concept to analyse these models. Although, therefore, readers should have
some basic knowledge of the subgame perfect equilibrium concept, it is not
necessary to have taken a course in game theory in order to understand
1.4 Further Remarks 7

much of the material in this book.2

In order to make the theory as widely accessible as possible, and so as to
develop a relatively deeper understanding of it, I adopt several simplifying
assumptions, and focus attention primarily on bargaining situations that
can be represented as involving the partition of a cake (or, 'surplus') of
fixed size. When it is deemed worthwhile to do so, I generalize the results.
Since a significant proportion of the material in this book is based upon
and/or inspired by the literature, it may perhaps be misleading for me to
ascribe the material directly to the authors concerned. Hence, I provide
appropriate acknowledgements to the relevant literature at the end of each
chapter, under the heading 'Notes'.

2
It might nevertheless be of interest for some readers to refer to a game theory text.
Fudenberg and Tirole (1991), Myerson (1991), van Damme (1991) and Osborne and Ru-
binstein (1994) are fairly formal, while Binmore (1992) and Gibbons (1992) are much less
so.
2 The Nash Bargaining Solution

2.1 Introduction

A bargaining solution may be interpreted as a formula that determines a

unique outcome for each bargaining situation in some class of bargaining
situations. In this chapter I study the bargaining solution created by John
Nash.1 The Nash bargaining solution is defined by a fairly simple formula,
and it is applicable to a large class of bargaining situations — these features
contribute to its attractiveness in applications. However, the most important
of reasons for studying and applying the Nash bargaining solution is that
it possesses sound strategic foundations: several plausible (game-theoretic)
models of bargaining vindicate its use. These strategic bargaining models
will be studied in later chapters where I shall address the issues of why,
when and how to use the Nash bargaining solution.
A prime objective of the current chapter, on the other hand, is to develop
a thorough understanding of the definition of the Nash bargaining solution,
which should, in particular, facilitate its characterization and use in any
application.
In the next section I define and characterize the Nash bargaining solution
of a specific bargaining situation in which two players bargain over the par-

bargaining solution and the concept of a Nash equilibrium are unrelated con-
cepts, other than the fact that both concepts are the creations of the same individual.
10 The Nash Bargaining Solution

tit ion of a cake (or 'surplus') of fixed size. Although this type of bargaining
situation is not uncommon, a main purpose of this section is to introduce
— in a relatively simple and concrete context — some of the main concepts
involved in defining Nash's bargaining solution. Section 2.3 contains two
applications of the Nash bargaining solution — one is to bribery and the
control of crime, and the other to optimal asset ownership.
Having mastered the concepts and results in Section 2.2, it should prove
relatively easier to understand Section 2.4, where I define and characterize
the Nash bargaining solution in its general form, which is somewhat abstract.
Section 2.5 contains three further applications of the Nash bargaining so-
lution — one is to union-firm negotiations, the second to team production
under moral hazard, and the third extends the application to bribery and
the control of crime studied in Section 2.3.1.
Section 2.6 shows that the Nash bargaining solution is the only possible
bargaining solution that satisfies four properties. Although these properties
are commonly referred to as axioms, one could debate whether or not any
of these properties are axiomatic. In any case, this 'axiomatic' foundation
is interesting and provides some insights into Nash's bargaining solution. A
key insight is that the Nash bargaining solution may be influenced by the
players' attitudes towards risk.
It is argued in Section 2.7 that the definition of the Nash bargaining
solution stated in Sections 2.2 and 2.4 fails to provide it with a natural
interpretation. An alternative (but equivalent) definition is stated in Section
2.7 which suggests that the Nash bargaining solution may be interpreted as
a stable bargaining convention.
Section 2.8 defines and characterizes the asymmetric Nash bargaining
solutions. These generalizations of the Nash bargaining solution possess a
facility to take into account additional factors of a bargaining situation that
may be deemed relevant for the bargaining outcome.

2.2 Bargaining over the Partition of a Cake

Two players, A and I?, bargain over the partition of a cake of size TT, where
7T > 0. The set of possible agreements is X = {(XA,XB) - 0 < XA <
7t and XB = TT — x ^ } , where X{ is the share of the cake to player i (i = A, B).
For each X{ G [O,vr], Ui(xi) is player i's utility from obtaining a share X{ of
2.2 Bargaining over the Partition of a Cake 11

the cake, where player z's utility function Ui : [0, TT] —> 5R is strictly increasing
and concave. If the players fail to reach agreement, then player i obtains a
utility of di, where di > Ui(0). There exists an agreement x G X such that
UA{X) > dA and UB(X) > d#, which ensures that there exists a mutually
beneficial agreement.
The utility pair d = (dA^ds) is called the disagreement point In order
to define the Nash bargaining solution of this bargaining situation, it is
useful to first define the set Q of possible utility pairs obtainable through
agreement. For the bargaining situation described above, Q = {{UA,UB) •
there exists x G l such that UA(XA) = UA and UB(XB) — UB}-
Fix an arbitrary utility UA to player A, where UA G [UA{0), UA{TT)}. From
the strict monotonicity of Ui, there exists a unique share XA G [0, TT] such
that UA(XA) — UA] i-e., XA = U^1(UA)-> where U^1 denotes the inverse of
UA-2 Hence

is the utility player B obtains when player A obtains the utility UA- It
immediately follows that Q — {(UA,V>B) •UA(0) < UA < UA{K) and UB —
g(uA)}] that is, 17 is the graph of the function g : [C/^(0), C/A(TT)] —> 5ft.
The A/'as/i bargaining solution (NBS) of the bargaining situation de-
scribed above is the unique pair of utilities, denoted by ( i ^ , i ^ ) , that solves
the following maximization problem

max

where © = {(UA,UB) G fi : UA > dA and UB > ds} = { ( ^ A , ^ B ) : UA(0) <

UA < UA(K),UB = g(uA),UA > dA and uB > dB}>
The maximization problem stated above has a unique solution, because
the maximand {UA — dA)(uB — ds) — which is referred to as the Nash
product — is continuous and strictly quasiconcave, g is strictly decreasing
and concave (as stated below in Lemma 2.1), and the set © is non-empty.3
Figure 2.1 illustrates the NBS. Since u^ > dA and u^ > dB, in the NBS
the players reach agreement on (x^^x1^) = (C/^1(ix^), UQ1(U^)).
2
It should be noted that the inverse UAX is a strictly increasing and convex function,
whose domain is the closed interval [C/A(0), UA(^)] and range is the closed interval [O,TT].
3
In fact, there exists a continuum of utility pairs {UA, UB) £ B such that UA > dA and
12 The Nash Bargaining Solution

L e m m a 2 . 1 . g is strictly decreasing and concave.

Proof. In the Appendix. •

— constant

F i g u r e 2.1: uN is the Nash bargaining solution of the bargaining situation in which

the set £1 of possible utility pairs obtainable through agreement is the graph of g, and d
is the disagreement point.

2.2.1 Characterization
The following result provides a characterization of the NBS of the bargaining
situation described above, when g is differentiate.

Proposition 2.1. In the bargaining situation described above, if g is dif-

ferentiable, then the Nash bargaining solution is the unique solution to the
following pair of equations

-g and
uA -dA

where g' denotes the derivative of g.

2.2 Bargaining over the Partition of a Cake 13

Proof. Since the NBS is such that u^ > CIA and u1^ > d#, it may be
characterized by finding the value of UA that maximizes (UA — GU)(#(?M) —
(IB)- The proposition follows immediately from the first-order condition. •

LN

F i g u r e 2.2: When g is differentiable, the NBS is the unique point on the graph of g
where the slope of the line LN is equal to the absolute value of the slope of the unique
tangent TN.

It is instructive, and useful in some applications, to note the following

geometric characterization of the NBS — which is valid when g is differen-
tiable and follows from Proposition 2.1. The NBS is the unique point uN
on the graph of g with the property that the slope of the line joining the
points uN and d is equal to the absolute value of the slope of the unique
tangent to the graph of g at uN. This is illustrated in Figure 2.2. Consider
any point u on the graph of g to the left of uN. The slope of the line L
joining points d and u has increased relative to the slope of the line LN,
while the absolute value of the slope of the tangent T to the graph of g at u
has decreased relative to the absolute value of the slope of the tangent T N.
Therefore, the slope of L is strictly greater than the absolute value of the
slope of T. By a symmetric argument, it follows that the slope of the line
joining the point d with a point on the graph of g to the right of the NBS is
14 The Nash Bargaining Solution

strictly less than the absolute value of the slope of the tangent to the graph
of g at that point.
The result contained in the following corollary to Proposition 2.1 may
be useful in applications.

Corollary 2.1. In the bargaining situation described above, if g is differ-

entiable, then the share x^ of the cake obtained by player A in the Nash
bargaining solution is the unique solution to the equation
-xA) -dB
U'A{xA) " U'B(ir-xA) '

and player B 's share in the NBS is x% = TT — x^ .

Proof. The result follows immediately from Proposition 2.1 after differ-
entiating g (with respect to uA) and noting that Ui(xi) — U{ and x\ —
Ur\Ui). •
I now provide a characterization of the NBS when g is not assumed to be
differentiable. However, since g is concave, it is differentiate 'almost every-
where'. But, it is possible that the NBS is precisely at a point where g is not
differentiate. 4 Since g is concave, its left-hand and right-hand derivatives
exist. Let g'{uA—) and g f(uA-\-) respectively denote the left-hand and right-
hand derivatives of g at uA. Since g is concave, g\uA~) > g'(uA+). The
following result is straightforward to establish, and is illustrated in Figure
2.3.

Proposition 2.2. In the bargaining situation described above, if g is not

differentiable at the Nash bargaining solution, then there exists a number k,
where gf(u^—) > k > g'(u^+), such that the Nash bargaining solution is
the unique solution to the following pair of equations
u B - dB , , .
—k — and uB = g[uA).
uA -dA

As is illustrated in Figure 2.3, the NBS is the unique point uN on the

graph of g with the property that the slope of the line LN joining uN and
d is equal to the absolute value (namely, —k) of the slope of some tangent
to the graph of g at uN.
Chapter 8 studies a model of bargaining in which this is the case.
2.2 Bargaining over the Partition of a Cake 15

UB

dB

uA

F i g u r e 2.3: When g is not differentiable, the NBS is the unique point on the graph
of g where the slope of the line LN is equal to the absolute value of the slope of some
tangent TN.

Remark 2.1 (Comparative-Statics). The following results may be es-

tablished by using the geometric characterizations of the NBS, as illustrated
in Figures 2.2 and 2.3. Since the NBS of the bargaining situation described
above depends upon the disagreement point, I emphasize this by writing the
NBS as (uf(d),uf(d)). Let d and d1 denote two alternative disagreement
points such that d[ > d\ and d'- = dj (j ^ i). If g is differentiate at u^(d),
then uf (dr) > uf^(d) and u^(df) < u^(d). If, on the other hand, g is not
differentiate at u%(d), then uf (df) > uf(d) and uf (df) < uf(d).

2.2.2 Examples
Example 2.1 (Split-The-Difference Rule). Suppose UA{XA) — %A for
all XA G [0, TT] and UB{%B) — %B forall XB £ [0, TT]. This means that for each
UA £ [0,TT], #(IML) = TT — ^^, and G^ > 0 (i = A,B). Applying Proposition
2.1, it follows that

= ly i71 — and = - ((vr - cU +

U
A +
16 The Nash Bargaining Solution

Thus

%A = dA + x (^ " dA ~ dBJ and x = dB + - (K - dA - dBj,

which may be given the following interpretation. The players agree first of
all to give player i (i = A, B) a share d{ of the cake (which gives her a utility
equal to the utility she obtains from not reaching agreement), and then they
split equally the remaining cake TT — dA — dB. Notice that player i's share
xf is strictly increasing in d{ and strictly decreasing in dj (j ^ i).

Example 2.2 (Risk Aversion). Suppose UA(xA) — x\ f° r a ^ XA £ [O?71"]?

where 0 < 7 < 1, UB(xB) = xB for all xB G [0,TT] and dA = dB = 0. This
means that for each uA £ [0, TT], g(uA) — ir — u^ . Applying Corollary 2.1,
it follows that
iV 771" A N n

xA = —•— and xB =
1+7 1+7
As 7 decreases, x^ decreases and x1^ increases. In the limit, as 7 —> 0,
x^ -^ 0 and x^ —» 1. Player S may be considered risk neutral (since her
utility function is linear), while player A risk averse (since her utility function
is strictly concave), where the degree of her risk aversion is decreasing in
7. Given this interpretation of the utility functions, it has been shown that
player A's share of the cake decreases as she becomes more risk averse.

2.3 Applications
2.3.1 Bribery and the Control of Crime
An individual C decides whether or not to steal a fixed amount of money
7T, where TT > 0. If she steals the money, then with probability £ she is
caught by a policeman P. The policeman is corruptible, and bargains with
the criminal over the amount of bribe b that C gives P in return for not
reporting her to the authorities.
The set of possible agreements is the set of possible divisions of the stolen
money, which (assuming money is perfectly divisible) is {(TT — 6, b) : 0 < b <
TT}. The policeman reports the criminal to the authorities if and only if they
fail to reach agreement. In that eventuality, the criminal pays a monetary
2.3 Applications 17

fine. The disagreement point (dc, dp) = (TT(1 — v\ 0), where v E (0,1] is the
penalty rate. The utility to each player from obtaining x units of money is
x.
The bargaining situation described here is a special case of Example 2.1,
and thus it immediately follows that the NBS is u^ = TT[1 — (V/2)] and
UN = TTV/2. The bribe associated with the NBS is bN = TTZ//2. Notice
that, although the penalty is never paid to the authorities, the penalty
rate influences the amount of bribe that the criminal pays the corruptible
policeman.
Given this outcome of the bargaining situation, I now address the issue
of whether or not the criminal commits the crime. The expected utility to
the criminal from stealing the money is (TT[1 — (is/2)] + (1 — £)TT, because with
probability £ she is caught by the policeman (in which case her utility is UQ)
and with probability 1 — ( she is not caught by the policeman (in which case
she keeps all of the stolen money). Since her utility from not stealing the
money is zero, the crime is not committed if and only if TT[1 — (£z//2)] < 0.
That is, since n > 0, the crime is not committed if and only if (V > 2. Since
C < 1 and 0 < v < 1 implies that (V < 1, for any penalty rate v G (0,1]
and any probability ( < 1 of being caught, the crime is committed. This
analysis thus vindicates the conventional wisdom that if penalties are evaded
through bribery, then they have no role in preventing crime.5

2.3.2 Optimal Asset Ownership

Consider a situation with two managers, A and B, and two physical assets
a A and as- Manager i knows only how to use asset c^. There are three
possible ownership structures: (i) manager A owns asset a A and manager B
owns asset a # , which is referred to as non-integration, (ii) manager A owns
both assets, which is referred to as type-A integration, and (iii) manager
B owns both assets, which is referred to as type-S integration. Denote by
Ti the set of assets that manager i owns. Thus, 1^ E {{c^}? {&A, OLB}^ {0}}?
where {0} means that manager i owns neither of the two assets. The analysis
below determines the optimal ownership structure. 6
5
The application studied here will be taken up in Section 2.5.3.
6
The analysis is based on the idea that the ownership structure affects the disagreement
point in the bargaining situation that the managers find themselves in. This, in turn,
influences the respective levels of asset-specific investments in human capital made by
18 The Nash Bargaining Solution

Given an ownership structure, the managers play the following two-stage

game. At the first stage, asset-specific investments in human capital are
simultaneously made by the managers. Since manager i knows only how
to use asset o^, and may (at the second stage) work with this asset, her
investment may be thought of as improving her knowledge of this asset.7
Let E{ > 0 denote the level of such investment made by manager i. The
cost of such investment Ci(Ei) is incurred by manager i at this stage.
The second stage involves determining whether or not the managers co-
operate (by using the two assets and their respective human capital) in
the creation of a cake. If they agree to co-operate, then the two assets are
combined with their respective human capital in the most productive manner
to generate a cake whose size UA(EA) + ^-B(EB) depends on the levels of the
investments made at the first stage. The set of possible agreements is the set
of partitions of this cake. If and only if the managers fail to reach agreement,
each manager goes her own way taking with her the assets that she owns.
The payoff d{ > 0 that manager i obtains in that eventuality depends on
Ei and 1^, which I emphasize by writing it as di(Ei;Ti). The utility to a
manager from obtaining a share x of the cake is equal to x. Assume that for
any ownership structure and for any investment pair, HA(EA) + T1B(EB) >
<1A{EA', TA) + CIB(EB'<> r#), which ensures that gains from co-operation exist.

The Bargaining Outcome

For any ownership structure and any investment pair, the bargaining situa-
tion described here is a special case of Example 2.1. Hence, it follows that
the NBS is

- dB(EB; TB) + dA(EA; TA)]

=l- [uA(EA) + UB(EB) - dA{EA- YA) + dB(EB; TB)].

the managers, and consequently, the size of the 'gains from co-operation'. The optimal
ownership structure is one that maximizes such gains from co-operation.
7
It is implicitly being assumed that at this stage, whatever the ownership structure
might be, manager i has the opportunity to improve her knowledge of asset oti.
2.3 Applications 19

Investments
In order to determine the investment levels, I adopt the following assump-
tions. For each i, 11^ is twice continuously differentiable, strictly increasing
and strictly concave, C\ is twice continuously differentiable, strictly increas-
ing and convex, d{ is twice continuously differentiable, increasing and con-
cave in Eu II£(O) > 2C[(G) and Tl^Ei) - C[{Ei) converges to a strictly
negative number as Ei tends to plus infinity.
Given an ownership structure, the utility to manager i if she chooses
Ei and manager j (j ^ i) chooses Ej is Pi(Ei,Ej) = uf — Ci(Ei). Par-
tially differentiating Pi with respect to Ei, it follows that dPi(Ei, Ej)/dEi =
[Ufi(Ei) + d'^Ei; Ti)]/2 - C[(Ei). Given the assumptions stated above, it fol-
lows that manager i's investment level E* is the unique solution8 to [II^(I^) +
d'^-Ti)}^ = C[{Ei). Letting (E%,E%), {Ei,E£) and (E*,E§), respec-
tively, denote the pairs of investment levels under non-integration, type-A
integration and type-B integration, it follows that for each i

\ [ n j ( ^ ) + d\{Ef •{a%})\ = C'{E?) (2.1)

4(E]; {aA, aB})] = C^E\) (2.2)

c?(£?), (2.3)

where j ^ i.

The Optimal Ownership Structure

For any pair of investment levels E — {EA, EB), the surplus is S(EA,
UA(EA) + UB(EB) - CA{EA) - CB(EB). The first best investment levels
(E^jEg) maximize the surplus, and hence, they constitute the unique so-
lution to the first-order conditions WA(E%) = C'A(E%) and I I ^ ^ ) -
C'B(Eg). I shall compare the four pairs of investment levels under the
following assumption: for each i and for any Ei

d'^Ef, {aA, aB}) > d ^ ; {«;}) > <(£<; {0}). (2.4)

Notice that E* is manager z's strictly dominant investment level: i.e., for any Ei and
20 The Nash Bargaining Solution

A key result can now be put forth: for each i

(2.5)

where the first (resp., second) weak inequality is strict if the first (resp.,
second) weak inequality in (2.4) is strict. A formal proof of (2.5) is rather
trivial to write down, and hence I omit it. Instead, it is far more illuminating
to illustrate the result using Figure 2.4.9

(.; {aA,aB})]/2

El El Ef
F i g u r e 2.4: Manager z's first best investment level is strictly greater than her invest-
ment level when she owns both assets, which, in turn, is greater than her investment level
when she owns only asset a*, which, in turn, is greater than her investment level when
she owns neither of the two assets.

Under any ownership structure there is under-investment relative to the

first best levels of investment. The intuition for this result is straightfor-
ward: for any ownership structure manager i obtains strictly less than the
full marginal benefit 11^(Ei) from her investment, and, hence, she invests
strictly less than her first best investment level. Furthermore, relative to
non-integration, type-A integration increases manager A's investment level,
but decreases manager 5's investment level. Symmetrically, relative to non-
integration, type-B integration increases manager B's investment level, but
9
The shapes and relative positions of the various curves shown in Figure 2.4 follow
from the assumptions made above.
2.3 Applications 21

decreases manager A's investment level. Hence, integration of either type

has a benefit and a cost. The optimal ownership structure balances such
costs and benefits.
The surpluses generated under the three ownership structures are S(EN),
S(EA) and S(EB). The optimal ownership structure is the one that pro-
duces the largest surplus. It has been noted above that this maximized
surplus is strictly less than the first best surplus S(EF) generated by the
first best investment levels. I now determine the nature of the optimal own-
ership structure in two potentially interesting scenarios.
First consider the case when the two assets are 'independent' in the fol-
lowing sense: the assets a A and a# are said to be independent if and only
if for each i and for any 2^, di(Ei; {a^, as}) — d[(Ei] {a^}). It follows from
(2.1) and (2.2) that for each i, E? = E\. Hence, since (by (2.5)) Ef > E)
(j z£ i)5 it follows that the surplus under non-integration is greater than
or equal to the surplus under type-i integration (i = A, I?).10 Therefore,
manager i should own asset c^. The intuition behind this result is straight-
forward. Since ownership of asset ay does not affect manager i's investment
level, but may instead decrease manager j ' s investment level, surplus may
therefore decrease if manager i owns asset ay. Consequently, the optimality
of the non-integration ownership structure when aA and a# are indepen-
dent.
Now consider the case when the two assets are 'strictly complementary'
in the following sense: the assets a A and ajg are said to be strictly com-
plementary if and only if for some i (i — A or i = B) and for any E{,
d'iiEnicti}) = <CE7i;{0}). It follows from (2.1) and (2.3) that E? = Ej
(j 7^ i). Hence, since (by (2.5)) E3- > E1^, it follows that the surplus
under type-j integration is greater than or equal to the surplus under non-
integration. The intuition behind this result is straightforward. Since own-
ership by manager i of asset OL{ on its own does not affect manager i's
investment level, surplus could therefore be increased by transferring the
ownership of asset oti to manager j (as this would then induce manager j
to increase her investment level). Consequently, the non-optimality of the
non-integration ownership structure under strict complementarity. In gen-
10
Notice that I am appealing to the fact that S is strictly increasing over [0, E^] x [0, ER\,
and that the investment levels under any ownership structure are strictly below the first
best investment levels.
22 The Nash Bargaining Solution

eral, however, it is not possible to determine which of the two integration

type ownership structures (type-A or type-i?) is optimal.

2.4 A General Definition

A bargaining problem is a pair (fi,d), where Q C 3ft2 and d G 3ft2. I in-
terpret ft as a set of possible utility pairs obtainable through agreement,
and the disagreement point d — (c^, ds) as the utility pair obtainable if the
players fail to reach agreement. 11 Attention will be restricted to bargaining
problems which satisfy the conditions stated below in Assumptions 2.1 and
2.2.

Assumption 2.1. The Pareto frontier £le of the set Q is the graph of a
concave function, denoted by /i, whose domain is a closed interval IA ^ 3ft.
Furthermore, there exists UA G IA such that UA > dA and h(ujCj > ds-12

Assumption 2.2. The set ftw of weakly Pareto efficient utility pairs is
closed.13

Notice that (by the definition of the Pareto frontier) h is strictly decreas-
ing. The set of all bargaining problems which satisfy Assumptions 2.1 and
2.2 is denoted by E. That is, £ = {(ft,d) : Q C 3ft2, d G 3ft2 and the pair
(fi,d) satisfies Assumptions 2.1 and 2.2 }.

Definition 2.1. The Nash bargaining solution (NBS) is a function fN :

£ —•» 3ft2, defined as follows. For each bargaining problem (fi, d) that satisfies
Assumptions 2.1 and 2.2, the NBS fN(Q,d) = (/^(fi,d),/^(fi,rf)) is the
unique solution to the following maximization problem

max (UA - (1A)(UB -

where 9 = {(UA,UB) E Cte : UA > dA and UB > ds} = {(V>A,V>B) •

A > dA and ix#
11
If (UAJUB) ^ ^ , then this means that there exists an agreement which gives player i
(i = A,B) a utility m G 5R.
12
A utility pair (UA,UB) G ^ e if and only if (IXA,^S) G ^ and there does not exist
another utility pair (u'A,uB) G ^ such that uA > UA, U'B > UB and for some i^u'i> u%.
13
A utility pair (UA,UB) G QW if and only if (UA,UB) G ^7 and there does not exist
another utility pair (uA,uB) G ^ such that uA > uA and u'B > uB. Notice that Qe C fi^.
2.4 A General Definition 23

The maximization problem stated above has a unique solution, because

the maximand (UA — GU)(^B — dB) — which is referred to as the Nash
product — is continuous and strictly quasiconcave, and because Assumption
2.1 implies that h is strictly decreasing and concave, and the set O is non-
empty. It should be noted that the NBS has the property that / ^ ( f i , d) > d\
(i = A,B).
Fix an arbitrary bargaining problem (Q,d) G S. The NBS of this bar-
gaining problem will lie on the graph of h. Let I A = [uA^ ^A] 5 where UA > HA-
The range of h is h(IA) — { UB G ^ • there exists UA G / A such that
us — h(uA)}> It follows from Assumption 2.1 that H(IA) = [u B,uB], where
h(u.A) ~ UB > V±B ~ h(uA)- Furthermore, Assumption 2.1 implies that
dA < UA and ds < UB> However, the possibility that for some i (i = A or
i = B or i = A, B) d{ < u{ is not ruled out by Assumption 2.1.
If G^4 G / A and d^ G H{IA) — which (from the above discussion) means
that di > Ui {i = A, B) — then the NBS is illustrated in Figure 2.1 with g
replaced by h.14 In particular, the NBS lies in the interior of the graph of h;
that is, / ^ ( O , d ) G (UA^A) and fg(Q,d) G (UB^B)- However, if for some
i (i = A or i = B or i = A, B) di < Ui, then it is possible (but not necessary)
that the NBS is at one of the two corners of the graph of /i; that is, the NBS
fN(Q, d) may equal either (uA, UB) or (UA,UB) — as is illustrated in Figure
2.5.

R e m a r k 2.2. A bargaining problem (fi,d) — upon which the NBS is de-

fined — is an abstract concept. Although this is valuable in some respects
and it enhances the applicability of the NBS, it is nevertheless helpful to
interpret the concept of a bargaining problem in terms of the following ba-
sic elements of a bargaining situation: (i) the set X of possible physical
agreements, (ii) the 'disagreement' outcome D — which is the outcome,
or event, that occurs if the players fail to reach agreement, and (iii) the
players' utility functions UA : X U {D} - • 3ft and UB : X U {D} -> 3ft. A bar-
gaining problem (fi, d) may then be derived from these elements as follows:
ft — {(UA-) UB) : there exists x G X such that UA{X) — UA and UB(X) = UB}
andd=(UA(D),UB(D)).
14
It should be noted that in the specific bargaining situation studied in Section 2.2 the
Pareto frontier Qe = £7, the set of possible utility pairs obtainable through agreement
— and, hence, £T is the graph of g. In contrast, in an arbitrary bargaining problem
(Q,d) G E the Pareto frontier D e C Q — that is, it need not equal Q.
24 The Nash Bargaining Solution

TV
u
— dA){iiB ~ ds) = constant

dB

uA
dA = u uA

F i g u r e 2.5: If ds < M#, then the NBS may be at the right-hand corner of the graph
of h — that is, uN = (UA,UB)-

2 A.I Characterization
It is straightforward to extend Proposition 2.1 to any bargaining problem
(0, d) G £ such that h is differentiate and d\ > u_% (i — A, B) — as is done
in the following proposition. 15

Proposition 2.3. For any bargaining problem (fi,d) G S si^c/i £/ia£ /i is

differentiate and d{ > u{ {% = A,B) 7 t/ie iVBS' is the unique solution to the
following pair of equations
U
B -
—a \uA) = ana

As is also discussed in the context of the specific bargaining situation

studied in Section 2.2 (cf. Figure 2.2), Proposition 2.3 implies that, for any
bargaining problem specified in the proposition, the NBS is the unique point
N
u on the graph of h with the property that the slope of the line joining
15
It should be noted that this proposition is not valid if for some i (i = A or i — B or
% = A, B) di < Ui, because (as discussed above and illustrated in Figure 2.5) it is possible
that the NBS may then be at one of the two corners of the graph of h.
2.5 Applications 25

the points uN and d is equal to the absolute value of the slope of the unique
tangent to the graph of h at uN — which is illustrated in Figure 2.2, but
with g replaced by h.
A similar geometric characterization applies to the NBS of any bargain-
ing problem (fi, d) G S, even if it does not satisfy the additional hypotheses
stated in Proposition 2.3. Fix an arbitrary bargaining problem (fi,d) G S,
and let ( ^ , M # ) be the NBS, which lies on the graph of h. The NBS has
the following geometric property. The slope of the line joining the points
u and d — which equals (u^ — dB)/(u^ — djC) — is greater than or equal
to —h f(u^—) ifv,j[ > uA and is less than or equal to — h'(u I^+) ifu^ < UA>
Notice that this geometric property is similar to that stated in Proposition
2.2.
In some applications the bargaining problem (fi, d) is such that the
Pareto frontier Qe of the set Q of possible utility pairs obtainable through
agreement is the graph of the linear function h{ujC) = s — UA, where s > 0.
The NBS of such a bargaining problem may be derived from Proposition
2.3, and is stated in the following corollary.

Corollary 2.2 (Split-The-Difference Rule). For any (fi,d) E £ such

that h{uA) = s — UA, where s > 0; and di > U{ (i — A,B), the NBS
[uAluB) is

-is - dA — dB) and u% = dB + -is - dA - dB\

Zi \ / Zi \ /

Corollary 2.2 may be given the following interpretation. The players are
bargaining over the partition of s units of (transferable) utility, and they
agree first of all to give each other the utilities (dA and dB) that they would,
respectively, obtain from not reaching agreement, and then they split equally
the remaining utility s — dA — dB.

2.5 Applications
2.5.1 Union-Firm Negotiations
A firm and its union bargain over the wage rate w and the employment
level L. The set of possible agreements is the set of wage-employment pairs
(w,L) such that w > wUl R(L) — wL > 0 and L < LQ, where wu > 0
26 The Nash Bargaining Solution

is the rate of unemployment benefit, R(L) is the revenue obtained by the

firm if it employs L workers, and Lo is the size of the union. R(0) = 0
and R is strictly increasing and strictly concave. The constraint w > wu
captures the fact that no worker works at a wage rate that lies below the
rate of unemployment benefit, while the constraint R(L) — wL > 0 captures
the fact that the firm prefers to close down rather than receive a negative
profit, where its profit from a pair (w, L) is R(L) — wL. It is assumed that
the firm cannot employ more than LQ workers. Thus, the set of possible
agreements is X = {(w,L) : w > wUiL < Lo and R(L) — wL > 0}. If
the players fail to reach agreement, then the firm shuts down and the Lo
workers become unemployed. If agreement is reached on (w,L) G X, then
the profit to the firm is Tl(w, L) = R(L) — wL, and the union's utility is
U(w,L) = wL + (Lo — L)wUi which constitutes the total income received
by its members. Since R(0) = 0, the profit to the firm if the parties fail
to reach agreement is zero. The union's utility in that eventuality is u>wLo,
since its LQ members become unemployed. Hence, the disagreement point
d = (wuLo,O).
The Pareto frontier Qe of the set of possible utility pairs obtainable
through agreement may be derived by solving the following maximization
problem: m a x ^ ^ ^ j II(K;, L) subject to U(w, L) > fZ, where u is some con-
stant greater than or equal to wuLo. At the unique solution to this problem
L = L*, where L* is the first best employment level, namely R'{L*) = wu.w
Thus, a utility pair (U,TT) G Vte only if the employment level L = L*. The
Pareto frontier Qe is therefore the graph of the function h defined as fol-
lows. For each utility level of the union u G [m^Lo, s], h(u) = s — u, where
s = R(L*) + (Lo - L*)wu.
Applying Corollary 2.2, it follows that the NBS is TTN = (s - wuL0)/2
and uN = m^Lo + (s - wuLo)/2. The wage-employment pair (wN,LN)
associated with the NBS is now derived. It has been shown above that at
the NBS the employment level LN = L*. The wage rate wN may be derived
from TTN = R(L*) — wNL*. After substituting for nN and 5, it follows that
wN = [wu + (i?(L*)/L*)]/2. The wage rate is therefore equal to the average
of the rate of unemployment benefit and the average revenue. However,
since R'(L*) = wu, the wage rate is equal to the average of the marginal
and average revenues.
s
It is assumed that L* < Lo and R(L*) - wuL* > 0.
2.5 Applications 27

2.5.2 Moral Hazard in Teams

The output produced by a team of two players, A and i?, depends on their
1/2 1/2

respective individual effort levels: output Q = 2eA eB , where e^ > 0 is

player z's effort level. The cost to player i of effort level ei is C{ = o^e^/2,
where ai > 0.
The effort levels are not verifiable, and, thus, cannot be contracted upon
— the players can only contract upon the output level, which is verifiable.
Before the players simultaneously choose their respective effort levels they
bargain over the output sharing rule. The set of possible agreements is
X = (0,1), where x G X is the share of the output obtained by player A
and 1 — x is the share of the output obtained by player B. If the players
reach agreement on x G I , then they simultaneously choose their respective
effort levels e^ > 0 and eB > 0, and, consequently, player i obtains a profit
of XiQ — Ci, where x\ — x if i = A and X{ = 1 — x if i = B. If the players
fail to reach agreement, then player i receives no output and incurs no cost
— and, hence, the disagreement point d = (0,0).
I first derive the Nash equilibrium of the simultaneous-move game in
effort levels, for each possible agreement x G X. Fix an arbitrary x G X.
Since XiQ — C{ is strictly concave in e^, the unique Nash equilibrium e*A and
e*B is the unique solution to the following first-order conditions
- 1 / 2 1/2 , /-, x 1/2 - 1 / 2
xeA e^ = OLA^A and (1 — x)e e =

Thus, after solving for e^ and e#, it follows that

e
A = a3 /M
4 11 M
/4 and
a n d
eBB ~
e
= 1/4 3/4
3
a
B
After substituting for these Nash equilibrium values of e^ and e#, it follows
that the players' (equilibrium) profits if agreement is reached o n x G l are

UA(x) = ^x3/2(l-x)1/2 and UB(x) = (3xl'2{l - xf'2

where f3 = 3/2(aAaB)1/2.
Notice that UA is strictly increasing on the open interval (0,3/4), achieves
a maximum at x = 3/4 and is strictly decreasing on the open interval
(3/4,1), and UB is strictly increasing on the open interval (0,1/4), achieves a
maximum at x = 1/4 and is strictly decreasing on the open interval (1/4,1).
28 The Nash Bargaining Solution

This implies t h a t the Pareto frontier of the set of possible utilities obtainable
through agreement is Qe = {{UA^B) '• there exists a n x G [1/4,3/4] such
that UA(X) = UA and UB(X) = UB}.
The bargaining problem (fi,cf) described here satisfies Assumptions 2.1
and 2.2, and hence I use Definition 2.1 to derive the NBS of this bargain-
ing problem.17 Definition 2.1 implies that the agreement xN obtained in
the NBS is the unique solution to the following maximization problem:
max
a;G[i/4,3/4] UA(X)UB(X). It immediately follows from the first-order con-
dition to this problem that xN = 1/2. Hence, the NBS — which is shown
in Figure 2.6 - is « , < ) = (UA(l/2), UB(l/2)).

UN

U U
A B — constant

F i g u r e 2.6: The NBS of the bargaining problem associated with the bargaining sit-
uation that the two team members find themselves in when bargaining over the output
sharing rule under moral hazard.

For any values of a A and a#, in the NBS the output is split equally
between the two players. This means that for any values of a A and a#,
) ~ CB{Z*B)'> which, in turn, implies that the two players obtain iden-
17
Notice that the domain of the graph oih — which is ft e — is I A = [C/A(1/4), E/A(3/4)].
Hence, since di < u^ (i — A, B) — where u A = UA (1/4) and uB = L^B(3/4) — Proposition
2.3 is inapplicable.
2.5 Applications 29

tical profits. Thus, even if a^ > OLB (which implies that player A's effort is
relatively more costly, and e*A < e^), the costs incurred by the players are
identical and their respective profits are identical.

2.5.3 Bribery and the Control of Crime: An Extension

An implicit assumption underlying the application studied in Section 2.3.1
is that individual C (the criminal) has limited liability in the sense that
the maximal possible bribe which the policeman P can obtain equals the
amount n of stolen money (i.e., b < TT), and the maximal possible penalty
that the authorities can impose equals the amount n of stolen money (i.e.,
v < 1). It is thus perhaps not surprising that — as is shown in Section 2.3.1
— for any penalty rate v < 1 and any probability ( < 1 of being caught, C
finds it profitable to commit the crime.
I now remove this limited liability assumption, by only requiring that
the bribe b > 0 and the penalty rate v > 0 — hence, I now allow for
the possibility that the bribe and the penalty exceed the amount TT of stolen
money.18 If agreement is reached on 6, then C's payoff is TT — b and P's payoff
is b. Hence, the Pareto frontier of the set of possible utility pairs is the graph
of a function h defined as follows: for each uc < TT, up = h(uc) = TT — uc-
The disagreement point (dc,dp) — (TT(1 — z^),0).
Applying Corollary 2.2 — notice that uc = — oc and uP = 0 — it follows
that the NBS is u% = TT[1 - (i//2)] and u$ = nv/2. The bribe is bN = 7rz//2.
Although this is also the NBS obtained in Section 2.3.1, there is now no
upper bound on v — which means that if v > 2 then C's payoff is negative.
From the arguments used in Section 2.3.1, it follows that the crime is not
committed if and only if (is > 2. Hence, for any ( < 1, if the penalty
rate is sufficiently large — in particular, if v > 2/£ — then the crime is
not committed. Thus, in contrast to the conclusion arrived at in Section
2.3.1, this analysis successfully challenges the conventional wisdom that if
penalities are evaded through bribery then they have no role in preventing
1Q

crime.
18
If the bribe b > TT, then the difference b — n may be financed by C in several ways,
including from her current (and possibly future) wealth and by payment in kind. If the
penalty TTU > TT, then the difference TT(Z/ —1) may be interpreted as the monetary equivalent
of a prison sentence.
19
The analysis here seems relatively more plausible compared to the analysis of Section
30 The Nash Bargaining Solution

2.6 Axiomatic Foundation

It is now shown that the NBS is the only bargaining solution that satisfies the
four properties (or, axioms) stated below.20 This axiomatization provides a
justification for using the NBS. However, whether or not such a justification
is convincing depends on the plausibility or otherwise of these axioms.

Risk: A Key Idea

As indicated in Remark 2.2 above, the elements of a bargaining situation
that may be relevant in defining a bargaining problem (and hence the NBS)
are the set of possible agreements, the disagreement outcome and the players'
utility functions. While bargaining the players may perceive that there is
some risk that the negotiations may break down in a random manner. Hence
the players' attitudes towards risk is another element that may be relevant
for the outcome of bargaining. For example, it seems intuitive that if player
A is averse to risk while player B is not, then player A may be willing
to reach an agreement that is relatively more favourable to player B (cf.
Example 2.2).
The axiomatization of the NBS is based on this additional element,
namely, the players' attitudes towards risk. As is well known, von Neumann
and Morgenstern's theory of expected utility deals with situations in which a
player has to act under conditions of risk. In order to appeal to that theory,
a player's utility function is interpreted as her von Neumann-Morgenstern
utility function. Her attitude towards risk is captured, or expressed, by the
shape of her utility function — if it is strictly concave (resp., strictly convex)
then she is risk averse (resp., risk loving), and if it is linear then she is risk
neutral.

The Axiomatization
Fix a bargaining problem (Q,d) G S. As in Remark 2.2, one may interpret
the two elements ft and d of this problem as being derived from the set X of
possible agreements, the disagreement outcome D and the utility functions
2.3.1, because it may be unreasonable to assume that the criminal has limited liability.
The application studied here will be taken up in Section 5.3.3.
20
A bargaining solution is a function / : E —•3ft2 such that for each (Q,d) G S,
f(£l,d) G Q|J{<i}. Nash's bargaining solution fN is stated in Definition 2.1.
2.6 Axiomatic Foundation 31

UA and £/#. NOW consider another bargaining situation with the same set
X of agreements and the same disagreement outcome D but with different
utility functions UA and UB, where Ui = otiUi + fli for some ai > 0 and
fa G 5ft (z = A, S ) . The disagreement point d' and the set fi' of possible
utility pairs obtainable through agreement of this (new) bargaining situation
are as follows

d' = (aAdA + (3A, aBdB + 0B) (2.6)

;
fi = {{aAuA + /?A, <*BUB + PB) : (UA, ^ B ) e ft}. (2.7)

Since (ft,d) satisfies Assumptions 2.1 and 2.2, (Q',df) satisfies Assumptions
2.1 and 2.2, and, hence, (J}\df) G S. This new bargaining situation is, in
effect, identical to the one originally specified, since (by construction) both
Ui and Ui represent player i's preferences.21 The first axiom emphasizes
this point by describing how, given any bargaining solution / : £ —> 5ft2, the
solutions to these two bargaining problems should be related.

Axiom 2.1 (Invariance to Equivalent Utility Representations). Fix

(ft, d) G I! and a bargaining solution / : E —• 5ft2. Now consider (ftr, d;) G S,
where d! and ft7 are respectively defined by (2.6) and (2.7), with ai > 0 and
A G 5ft. Then, for i = A,B, / i (ft / , d;) = c^(ft, d) + A-
Axiom 2.1 is motivated by the viewpoint that it is the players' prefer-
ences, and not the particular utility functions which represent them, that
are 'basic'. The agreements associated with the bargaining solution in these
two (related) bargaining problems should be identical. Thus, although the
bargaining solution of these two problems may differ, they should be re-
lated in the manner specified. This axiom seems reasonable. However, the
next axiom, which requires that the bargaining solution for each bargaining
problem be Pareto efficient, is not as easy to justify. In many bargaining
situations players fail to reach agreement. For example, strikes and trade
wars are examples of phenomena that contribute to the inefficiency of the
bargaining outcome. However, in some bargaining situations, this may be a
plausible axiom.
21
Since a player's preferences satisfy the von Neumann-Morgenstern expected utility
theory, her von Neumann-Morgenstern utility function is unique only up to a positive
affine transformation. For an excellent account of this theory, see Luce and Raiffa (1957,
Chapter 2).
32 The Nash Bargaining Solution

Axiom 2.2 (Pareto Efficiency). Fix (fi,d) G E and a bargaining solu-

tion / : S —• 9?2. There does not exist a utility pair (UA,UB) E fi|J{d}
such that UA > ./U(fi,d), ^ B > /s(fi,d) and for some i (i = A or i = B),

The next axiom seems easy to justify (£2, of) is said to be symmetric if
GU = d # ( = d), and (x, y) E fHf and only if (y, x) E f2.

Axiom 2.3 (Symmetry). Fix (fi,d) G S and a bargaining solution / :

E —> !R2. If (fi, d) is symmetric, as defined above, then /^(fi, d) = /s(f2, d).

Of the four axioms, the last is the most problematic. I discuss the axiom
after formally stating it.

Axiom 2.4 (Independence of Irrelevant Alternatives). Fix a bargain-

ing solution / : E —» 3?2, (fii,di) G S and (£^2^2) E S such that di = d2,
O2 Cfti a n d / ( f i i , d i ) G fi2- Then, for i = A, B, / i ( ^ i , d i ) = /,(O 2 , d 2 ).
This axiom considers two bargaining problems in which the disagreement
points are identical, the set of possible utilities in one problem is strictly con-
tained in the set of possible utilities of the other problem, and the bargaining
solution of the latter problem is an element of the former set. It states that
the bargaining solutions to such related problems should be identical. The
motivation for this axiom can be put forth in the following manner. Sup-
pose the bargainers agree on the element x\ when the set X of agreements
consists of three elements, namely xi, X2 and £3, and when the disagree-
ment outcome is D. Now consider another bargaining situation in which
the players have to agree on an element from a subset Y of this set X, that
contains the element x\ agreed to in the preceding bargaining situation. For
example, suppose Y — {x\,X2}. And, moreover, the disagreement outcome
is the same, namely, D. The argument is that since x\ was agreed to over
X2 and £3 in the original bargaining situation, and since the disagreement
outcome is the same in the new bargaining situation, x\ should be agreed
to (again) over £2, despite the fact that £3 is no longer available. Indeed,
£3 is 'irrelevant'. At an intuitive level, this may seem persuasive. However,
it can be interpreted as an axiom about the process of negotiation, which is
not modelled here. In some negotiation processes the outcome may be influ-
enced by such apparently 'irrelevant' alternatives. For example, an outcome
based on some compromise may be influenced by such alternatives. Hence,
2.7 An Interpretation 33

as is the case with the other axioms, especially Axiom 2.2, one needs to
study plausible game-theoretic models of bargaining in order to assess the
circumstances under which this axiom is and is not plausible.
Notice that Axioms 2.2 and 2.3 are concerned with the outcome of a
particular bargaining problem, while Axioms 2.1 and 2.4, on the other hand,
are not about the outcome of any particular bargaining problem, but concern
the relationship between the outcomes of two somewhat related bargaining
problems. I now establish the remarkable result that the Nash bargaining
solution is the unique bargaining solution that satisfies Axioms 2.1 to 2.4.

Proposition 2.4. A bargaining solution f : £ —> 3ft2 satisfies Axioms 2.1

to 2.4 if and only if f = fN, Nash's bargaining solution.

Proof. In the Appendix. •

Although the four 'axioms' may not be axiomatic, and although Axiom
2.4 in particular is somewhat problematic, this proposition provides the NBS
with an interesting justification.22

2.7 An Interpretation
Although the definition of the NBS — in terms of a maximization problem —
is convenient in applications, its interpretation is unclear. How should the
maximization of the product of the players' utilities be interpreted? The
axiomatization discussed in the preceding section provides a justification
for the maximization of the Nash product: it follows from Axioms 2.1-2A
by logical deduction. Thus, an indirect way to interpret Definition 2.1 is
to interpret the four axioms. In this section I state an alternative (but
equivalent) definition of the NBS that has a natural interpretation. This
alternative definition, however, is relatively more complex, and, hence, it is
not as attractive for applications.
Unlike Definition 2.1, which is in terms of the players' utilities, the fol-
lowing definition of the NBS is in terms of the 'physical' agreement struck.
Furthermore, it is based explicitly on the players' attitudes towards risk.
22
It can be shown that none of the four axioms is superfluous: by dropping any one
of these axioms, there is an alternative bargaining solution that satisfies the other three
axioms (cf. Osborne and Rubinstein (1990, pp. 20-23)).
34 The Nash Bargaining Solution

Definition 2.2. For any bargaining situation as defined by X, D, UA and

UB — where X is the set of physical agreements, D the disagreement out-
come, and UA and UB the players' von Neumann-Morgenstern utility func-
tions — such that the induced bargaining problem (fi, d) (as defined in
Remark 2.2 above) satisfies Assumptions 2.1 and 2.2, the agreement x* G X
is an ordinal NBS if for any i = A,B, any x G X and any probability
p e [0, l]

+pUi(D) > Ui{x*) => (l-p)Uj(x*) +pUj(D) > Uj(x),

where j ^ i.
In Lemma 2.2 below, it is shown that x* is an ordinal NBS if and only
if the utility pair (C7^(x*), £/#(#*)) is the NBS. This means that Definitions
2.1 and 2.2 are equivalent. It also implies that an ordinal NBS exists for any
bargaining situation that satisfies Assumptions 2.1 and 2.2. Furthermore,
the ordinal NBS is unique if and only if there exists a unique x G X such

I now discuss the interpretation of the NBS prompted by this alterna-

tive definition. It involves interpreting the NBS as a stable (bargaining)
convention.
In any bargaining situation the players often put forward arguments
of various kinds supporting the implementation of specific agreements. The
ordinal NBS is interpreted as that agreement which is immune to a particular
class of arguments. In that sense it may be interpreted as a conventional
agreement, where the convention is stable in a particular manner.
Suppose that x* is the conventional agreement, but player i puts forth
an argument supporting the implementation of some alternative agreement
x. More importantly, the argument involves a 'risk of breakdown': with
probability p this argument forces bargaining to terminate, in which case the
disagreement outcome occurs. However, with probability 1—p the argument
'wins': that is, the alternative agreement x is implemented. It is assumed
that

(l-p)Ui(x)+pUi(D)>Ui(x*),

for otherwise player i would not put forth this argument. Now suppose that
player j (j ^ i) puts forth a counterargument insisting on implementing the
2.8 Asymmetric Nash Bargaining Solutions 35

conventional agreement x*. However, in putting forth this counterargument,

it is perceived that with probability p bargaining terminates in disagreement
and with probability 1 — p this counterargument wins. If

(l-p)Uj(x*)+PUj(D)>UJ(x),

then player j puts forth this counterargument to player i's initial argument
against the convention. Hence, if the conventional agreement satisfies Defi-
nition 2.2, then the convention is protected from arguments (or, objections)
put forth by player i.

Lemma 2.2. Fix a bargaining situation as defined by X, D, UA and UB

such that Assumptions 2.1 and 2.2 are satisfied. Without loss of generality,
assume that UA(D) = UB{D) = 0. 23 The agreement x* E X is an ordinal
NBS if and only if for all x G X

UA(x*)UB(x*) > UA(x)UB(x).

Proof. In the Appendix. •

2.8 Asymmetric Nash Bargaining Solutions

The NBS depends upon the set Q of possible utility pairs and the disagree-
ment point d. However, the outcome of a bargaining situation may be in-
fluenced by other forces (or, variables), such as the tactics employed by the
bargainers, the procedure through which negotiations are conducted, the in-
formation structure and the players' discount rates. However, none of these
forces seem to affect the two objects upon which the NBS is defined, and yet
it seems reasonable not to rule out the possibility that such forces may have
a significant impact on the bargaining outcome. I now state generalizations
of the NBS which possess a facility to take into account additional factors
that may be deemed relevant for the bargaining outcome.

Definition 2.3. For each r G (0,1), an asymmetric (or, generalized) Nash

bargaining solution is a function f^ : £ —> Sft2, denned as follows. For each
23
Since JJ% is a von Neumann-Morgenstern utility function, one can always normalize
utilities so that the disagreement point is at the origin.
36 The Nash Bargaining Solution

(Jl, rf) E S, f^(Q,d) is the unique solution to the following maximization

problem

max (UA — dA)T(uB — dB)1~r,

where 0 is stated in Definition 2.1.

For each r E (0,1), an asymmetric NBS f^ satisfies Axioms 2.1, 2.2

and 2.4. Furthermore, any bargaining solution that satisfies Axioms 2.1,
2.2 and 2.4 is an asymmetric NBS for some value of r. Unless r = 1/2,
an asymmetric NBS does not satisfy Axiom 2.3. If, on the other hand,
r = 1/2, then the asymmetric NBS is identical to the NBS. 24 I now state
two useful characterizations of any asymmetric NBS, which follow trivially
from Proposition 2.3 and Corollary 2.2.

Proposition 2.5. For any r E (0,1) and any bargaining problem (£), d) E S
such that h is differentiate and d\ >u.i (i = A, B), the asymmetric NBS is
the unique solution to the following pair of equations

, // x / T \ \uB — a B\
-h (uA) = z T~ and uB = h(uA).
\1-T J yaA ~dA\

Corollary 2.3 (Split-The-Difference Rule). For any r E (0,1) and any

(ft, d) E £ such that 1I(UA) — s — UA, where s > 0 ; and d{ >ui(i = A, B),
the asymmetric NBS (u^^u1^) is

N 1 y ( 7 7^\ 7 iV J . /1 \( J 1 \
U T
A — ®A + [s — MA — dB ] a n a uB = a B + (1 — r ) \ s — CLA — ctB 1.

Notice that as r increases, player A's utility increases while player i?'s
utility decreases. Corollary 2.3 may be given the following interpretation.
The players are bargaining over the partition of s units of (transferable)
utility, and they agree first of all to give each other the utilities (dA and d B)
that they would, respectively, obtain from not reaching agreement, and then
players A and 5 , respectively, obtain a fraction r and 1 — r of the remaining
utility s — dA — d B.
24
Strategic models of bargaining studied in later chapters provide guidance on what
elements of a bargaining situation determine the value of r.
2.9 Appendix: Proofs 37

2.9 Appendix: Proofs

Proof of L e m m a 2.1
First I establish that g is strictly decreasing. Fix an arbitrary pair u\ and u\
of utilities to player A such that u\> u\, where u\ G [£/A(0), UA(TT)] and
u\ G [C/A(0),C/A(TT)]. Since C/A1 is strictly increasing, UAl{u\) > UAl(u2A).
This implies that TT — Uj^iu1^) < re — f/^ 1 (^), and hence, since UB is
strictly increasing, g(u\) > g(u\). I now establish that g is concave. Fix an
arbitrary pair u\ and u\ of utilities to player A such that u\ > u\, where
u\ G [UA(0),UA(K)] and i ^ G [[/^(O), C/A(TT)], and fix an arbitrary number
a G [0,1]. Since [/# is concave

UB(x%) > aUB{xlB) + (1 - a)C/ B (x|), (2.8)

where xxB — TT —C/^ 1 (^), x\ — TT — UA1(UA) and x^ = a[7r — C

a)[7r —C/^ 1 (^)]. Furthermore, since t/^ 1 is convex, t r ^ 1 ( ^ )
(1 — a)C/^1(ix^), where u\ — aulA + (1 — a)i^. Hence, it follows that

(2.9)

Consequently, since UB is strictly increasing, it follows from (2.8) and (2.9)

that g(u\) > ag(u\) + (1 — a)g(uA), which establishes that g is concave.

Proof of Propostion 2.4

I first show that the NBS satisfies Axioms 2.1-2.4. Consider Axiom 2.1. By
definition, for any u G ©, (uA—dA)(u^—dB) > (UA~ dA)(uB—ds)> Which is
if and only if for any u G ©, ^ ^ ( ^ - r f A ) ^ - ^ ) > CXAOLB{UA—dA)(uB —
ds)- Which, in turn, is if and only if for any v G ©, (vA — dfA)(vg — d'B) >
(VA - dfA)(vB - d'B), where © = { ( V A , V B ) : 3 W G 0 s.t. Vi = a\Ui + A }
and vf — aiuf + Pi. Hence, the NBS satisfies Axiom 2.1. By definition,
the NBS satisfies Axiom 2.2. Now consider Axiom 2.3, and a bargaining
problem that is symmetric. By the definition of symmetry, (uA, h(uA)) G ©
implies that (h(uA),uA) G 0. Since the maximand (UA — d)(uB — d) is a
symmetric function, (h(uA), uA) maximizes (uA — d)(uB~d) over the set ©.
But since the maximizer is unique, uA — h(uA). Hence, the NBS satisfies
Axiom 2.3. Finally, consider Axiom 2.4. Since the disagreement points
in the two related bargaining problems are identical, the Nash products of
38 The Nash Bargaining Solution

these problems are identical. Hence, since fN(Qi,di) G Q2 and (^2 C fii
implies) 02 C Oi, fN(VLi,d\) maximizes (UA — (1A)(UB — (1B) over the set
©1 only if fN(Qi,di) maximizes (UA — (1A)(UB — (1B) over the set O2.
I now establish that no other bargaining solution satisfies Axioms 2.1-
2.4. Suppose / is a bargaining solution that satisfies Axioms 2.1-2A. I shall
show that / = fN. Fix (fi, d) G S. It needs to be shown that for i = A,B,
fi(n,d) = fl*(n,d).
Since uf > di, define for each z, ai = l/2(uf — di) and Pi = —di/2(uf —
di). Given (fi, d) and these values for 0:^4, o;^, /3^ and /?#, define (£y,cf),
where </ and fi' are respectively defined by (2.6) and (2.7). It is easy to
verify that for each i, d!{ = 0 and aiuf + (3i = 1/2 — the values of ai
and /?i are constructed to ensure these two conclusions (i.e., in order to
shift the disagreement point d to the origin and the NBS to the point
(1/2,1/2)). Since both / and fN satisfy Axiom 2.1, and since ai > 0,
it follows that /i(fi',0) = atfi{n,d) + & and yf (fi;,0) = (tiff*(to,d) + &.
Hence, /(ft,d) = fN(tt,d) if and only if /(ft ; ,0) = /^(fi',0). Thus, since
fN(fL',O) = (1/2,1/2), it needs to be shown that /(ft ; ,0) = (1/2,1/2).
Since fN(ttf,0) = (1/2,1/2), and since ftf satisfies Assumptions 2.1 and
2.2, it follows that 1/2 G IA and h(l/2) = 1/2. I now argue that for any
UA £ IA, h{uA) < ^ — UA- Suppose, to the contrary, there exists u'A G IA
such that h(ufA) > 1 — u'A. Since h is concave, any point (UA, JT>(UA)) on the
graph of h between points (1/2,1/2) and (u'A, h(ufA)) has the property that
h(uA) > 1—UA- Hence, there exists a point (u"A, h(ufA)) such that u'Ah(ufA) >
1/4. Since u"A > 0 and h(u'A)) > 0, this contradicts the established fact that
fN(tf,0) = (1/2,1/2).
This result implies that Q — {(UA, UB) : UA+UB < 1} contains the set fi'.
By Axioms 2.2 and 2.3, it follows that /(Q,0) = (1/2,1/2). Consequently,
by Axiom 2.4, it follows that /(ft',0) = (1/2,1/2).

Proof of Lemma 2.2

I first establish the i/part. Suppose, to the contrary, that for all x,

UA(x*)UB(x*) > UA(x)UB(x),

but that there exists an i, x and p such that (1 — p)Ui(x) > U{(x*) ==>
Uj(x) > (1 -p)Uj(x*). Since UA(D) = UB(D) = 0, there exists an x
2.10 Notes 39

such that UA(X) > 0 and UB(X) > 0, which implies that UA(X*) > 0 and
UB(x*) > 0. Hence, (l-p)Uz(x) > Ui(x*) implies that (l-p)Ui(x)Uj(x*) >
Ui(x*)Uj(x*). This, in turn, implies (from my supposition) that (l—p)Uj(x*) >
Uj(x), which is a contradiction. I now establish the only i/part. Suppose
that x* is such that for any z, x and p, (1 — p)Ui(x) > Ui(x*) =^ (1 —
p)Uj(x*) > Uj{x). Consider an arbitrary x such that UA{X) > 0, UB(X) > 0
and for some z, Ui(x) > Ui(x*).25 This means that for any (1 — p) >
Ui(x*)/Ui(x), (1 -p) > Uj(x)/Uj(x*), which implies that Ui(x*)/Ui(x) >
Uj(x)/Uj(x*), and, hence, crossmultiplication gives the desired conclusion.

2.10 Notes
The Nash bargaining solution was proposed in John Nash's seminal paper,
Nash (1950), where he provided the axiomatization discussed in Section
2.6. Binmore (1992, pp. 180-195) discusses additional properties of the
asymmetric NBS, and contains a proof of the result, mentioned in Section
2.8, that any bargaining solution which satisfies Axioms 2.1, 2.2 and 2.4 is an
asymmetric NBS for some value of r. Both Osborne and Rubinstein (1990,
pp. 9-27) and Binmore (1992, pp. 165-195) contain excellent expositions of
the Nash bargaining solution.
For further elaboration on the applications in Sections 2.3.1/2.5.3 and
2.5.1 respectively, see Basu, Bhattacharya and Mishra (1992) and McDonald
and Solow (1981). The application in Section 2.3.2 is based upon Hart (1995,
pp. 34-49), which is a simple exposition of the ideas of Grossman and Hart
(1986) and Hart and Moore (1990) — concerning the theory of the firm.
For an alternative analysis of the phenomenon studied in Section 2.3.2 —
based upon the outside option principle (which I describe in Chapter 5)
— see de Meza and Lockwood (1998). The application in Section 2.5.2 is
based upon Nandeibam (1996), who studies a general version of this problem
of team production under moral hazard, but uses Rubinstein's bargaining
model — which is studied in the next chapter — rather than the NBS.
The interpretation of the NBS discussed in Section 2.7 is due to Rubin-
stein, Safra and Thomson (1992).

25
For any other x it is trivially the case that UA(X*)UB(X*) > UA(X)UB(X).
 The Rubinstein Model

3.1 Introduction

In this chapter I study Rubinstein's model of bargaining. A key feature of

this model is that it specifies a rather attractive procedure of bargaining: the
players take turns to make offers to each other until agreement is secured.
This model has much intuitive appeal, since making offers and counteroffers
lies at the heart of many real-life negotiations.
Rubinstein's model provides several insights about bargaining situations.
One insight is that frictionless bargaining processes are indeterminate. A
bargaining process may be considered 'frictionless' if the players do not incur
any costs by haggling (i.e., by making offers and counteroffers) — in which
case there is nothing to prevent them from haggling for as long as they
wish. It seems intuitive that for the players to have some incentive to reach
agreement they should find it costly to haggle. Another insight is that a
player's bargaining power depends on the relative magnitude of the players'
respective costs of haggling, with the absolute magnitudes of these costs
being irrelevant to the bargaining outcome.
An important reason for the immense influence that Rubinstein's model
has had, and continues to have, is that it provides a basic framework, which
can be adapted, extended and modified for the purposes of application. This
will become evident in several later chapters of this book.
42 The Rubinstein Model

In the next section I describe and analyse a simple version of Rubin-

stein's model in which two players are bargaining over the partition of a
cake (or 'surplus') of fixed size. The analysis involves characterizing the
unique subgame perfect equilibrium of this game-theoretic model. I study
a fairly simple version of the model, because my objective in this section is
to present the main arguments in a simple, but rigorous, manner. A discus-
sion of the properties of the unique subgame perfect equilibrium and of the
value and interpretation of the model respectively are contained in Sections
3.2.3 and 3.2.4. An application to a bilateral monopoly market is studied in
Section 3.3.
In some applications of Rubinstein's model the set of possible agreements
and the players' utility functions may be relatively more complex than what
is assumed in Section 3.2. Hence, in Section 3.4, I study a general version of
Rubinstein's model. The method of analysing this generalization is similar
to that of the simple version. Section 3.5 contains an application to a two-
person exchange economy.
Under a (plausible) condition, the unique subgame perfect equilibrium
payoff pair of Rubinstein's model is identical to an asymmetric Nash bar-
gaining solution of an appropriately defined bargaining problem. This re-
markable result, which is the subject of discussion in Sections 3.2.3 and 3.4.3,
provides a compelling justification for the use of Nash's bargaining solution.

3.2 The Basic Alternating-Offers Model

Two players, A and i?, bargain over the partition of a cake of size n (where
7T > 0) according to the following, alternating-offers, procedure. At time 0
player A makes an offer to player B. An offer is a proposal of a partition
of the cake. If player B accepts the offer, then agreement is struck and
the players divide the cake according to the accepted offer. On the other
hand, if player B rejects the offer, then she makes a counteroffer at time
A > 0. If this counteroffer is accepted by player A, then agreement is
struck. Otherwise, player A makes a counter-counteroffer at time 2A. This
process of making offers and counteroffers continues until a player accepts
an offer.
A precise description of this bargaining procedure now follows. Offers are
made at discrete points in time: namely, at times 0, A, 2A, 3 A , . . . , t A , . . . ,
3.2 The Basic Alternating-Offers Model 43

where A > 0. An offer is a number greater than or equal to zero and less
than or equal to TT. I adopt the convention that an offer is the share of the
cake to the proposer, and, therefore, TT minus the offer is the share to the
responder. At time £A when t is even (i.e., t = 0, 2,4,...) player A makes
an offer to player B. If player B accepts the offer, then the negotiations end
with agreement. On the other hand, if player B rejects the offer, then A
time units later, at time (t + 1)A, player B makes an offer to player A. If
player A accepts the offer, then the negotiations end with agreement. On
the other hand, if player A rejects the offer, then A time units later, at time
(t + 2)A, player A makes an offer to player S, and so on. The negotiations
end if and only if a player accepts an offer.
The payoffs are as follows. If the players reach agreement at time tA
(t = 0 , 1 , 2 , . . . ) on a partition that gives player i (i = A, B) a share xi
(0 < Xi < TT) of the cake, then player i's payoff is xi exp(—ntA), where ri > 0
is player i's discount rate. On the other hand, if the players perpetually
disagree (i.e., each player always rejects any offer made to her), then each
player's payoff is zero.
This completes the description of the basic alternating-offers game. For
notational convenience, define 6i = exp(—r^A), where 6i is player i's discount
factor. Notice that 0 < Si < 1.
The subgame perfect equilibrium (SPE) concept will be employed to
characterize the outcome of this game. In particular, answers to the follow-
ing questions will be sought. In equilibrium, do the players reach agreement
or do they perpetually disagree? In the former case, what is the agreed
partition and at what time is agreement struck?

3.2.1 The Unique Subgame Perfect Equilibrium

How should one proceed to characterize the subgame perfect equilibria of
this game? Notice that, since this is an infinite horizon game, one cannot use
the backwards induction method. I shall proceed as follows. First, I shall
characterize the unique SPE that satisfies the two properties stated below.
Then, I shall show that it is the unique SPE of this game, which means that
there does not exist a SPE which fails to satisfy these two properties.
Consider a SPE that satisfies the following two properties: 1
x
At this point I do not claim that such an equilibrium exists. The argument that
44 The Rubinstein Model

Property 3.1 (No Delay). Whenever a player has to make an offer, her
equilibrium offer is accepted by the other player.

Property 3.2 (Stationarity). In equilibrium, a player makes the same

offer whenever she has to make an offer.

Given Property 3.2, let x* denote the equilibrium offer that player i
makes whenever she has to make an offer. Consider an arbitrary point in
time at which player A has to make an offer to player B. It follows from
Properties 3.1 and 3.2 that player S's equilibrium payoff from rejecting any
offer is 8BX*B. This is because, by Property 3.2, she offers x*B after rejecting
any offer, which, by Property 3.1, is accepted by player A. Perfection re-
quires that player B accept any offer XA such that n — XA > ^B^B-> an<^ r e J e c t
any offer XA such that TT—XA < ^B^B- Furthermore, it follows from Property
3.1 that TT — x*A > 6B%%. However, n — X*A ^ 6BX*B'<, otherwise player A could
increase her payoff by instead offering x'A such that TT — X*A > ix — x'A > 6BX*B.
Hence

ir-x*A = 6Bx*B. (3.1)

Equation 3.1 states that player B is indifferent between accepting and re-
jecting player A's equilibrium offer. By a symmetric argument (with the
roles of A and B reversed), it follows that player A is indifferent between
accepting and rejecting player B's equilibrium offer. That is

(3.2)

Equations 3.1 and 3.2 have a unique solution, namely

and x*B — HB^, where (3.3)

and
VB = z T--7-. (3.4)
1 — 0A0B
The uniqueness of the solution to equations 3.1 and 3.2 means that there
exists at most one SPE satisfying Properties 3.1 and 3.2. In that SPE,
player A always offers x*A and always accepts an offer XB if and only if
7T — %B ^ $AX% and player B always offers x*B and always accepts an offer
follows derives some further properties of such an equilibrium. I shall subsequently be in
a position to claim the existence of such an equilibrium.
3.2 The Basic Alternating-Offers Model 45

XA if and only if IT — XA > £>B%*BI where x*A and x*B are defined in (3.3). It
is straightforward to verify, as is done below in the proof of Proposition 3.1,
that this pair of strategies is a subgame perfect equilibrium.

Proposition 3.1. The following pair of strategies is a subgame perfect equi-

librium of the basic alternating-offers game:
• player A always offers x*A and always accepts an offer xB if and only if
%B < X*B,
• player B always offers x*B and always accepts an offer XA if and only if
xA < x*A,
where x\ and x*B are defined in (3.3).

Proof. I shall show that player A's strategy — as defined in the proposition
— is optimal at any point in the game, given that player B uses the strategy
described in the proposition. Consider any point in time tA (where t is an
arbitrary even number) when player A has to make an offer to player B.
If she uses the strategy described in the proposition, then her payoff is x*A.
Consider any alternative strategy for player A, where x\ denotes the offer
she makes at time tA. If x\ < x*A, then, since player B accepts such an
offer, deviation to this alternative strategy is not profitable. Now suppose
that xlA > x*A. In this case player B rejects the offer made at time tA. Since
player B always rejects any offer that gives player A a share greater than
x*A and always offers x*B, player A's payoff from this alternative strategy is
less than or equal to max{<5^(7r — x^),<5^x^}. Using (3.2), it follows that
deviation to this alternative strategy is not profitable.
Now consider any point in time tA (where t is an arbitrary odd num-
ber) when player A has to respond to an offer made by player B. I have
established above that at time (t + 1) A it is optimal for player A to use the
strategy described in the proposition. It therefore follows that it is optimal
for her to accept an offer xB if and only if TT — xB > 8AX\.
By a symmetric argument (with the roles of A and B reversed), it follows
that player fi's strategy — as defined in the proposition — is optimal at
any point in the game, given that player A uses the strategy described in
the proposition. •

Proposition 3.1 characterizes the unique SPE that satisfies Properties

3.1 and 3.2. The following theorem states that this SPE is the unique SPE
46 The Rubinstein Model

of the basic alternating-offers game. This means that there does not exist a
SPE which fails to satisfy Properties 3.1 and 3.2.

Theorem 3.1. The subgame perfect equilibrium described in Proposition

3.1 is the unique subgame perfect equilibrium of the basic alternating-offers
game.

Proof. In Section 3.2.2 below. •

In the unique SPE, agreement is reached at time 0, and the SPE is Pareto
efficient. Since it is player A who makes the offer at time 0, the shares of
the cake obtained by players A and B in the unique SPE are x*A and n — x\,
respectively, where x*A = \IA^ and TT — x*A = SBUB^-
The equilibrium share to each player depends on both players' discount
factors. In particular, the equilibrium share obtained by a player is strictly
increasing in her discount factor, and strictly decreasing in her opponent's
discount factor. Notice that if the players' discount rates are identical (i.e.,
^A — ^B — r > 0), then player A's equilibrium share TT/(1 + 6) is strictly
greater than player £Ts equilibrium share TT6/(1 + <5), where 6 = exp(—rA).
This result suggests that there exists a 'first-mover' advantage, since if TA =
rs then the only asymmetry in the game is that player A makes the first
offer, at time 0. However, note that this first-mover advantage disappears
in the limit as A —> 0: each player obtains one-half of the cake.
As is evident in the following corollary, the properties of the equilibrium
shares (when TA ^ TB) are relatively more transparent in the limit as the
time interval A between two consecutive offers tends to zero.

Corollary 3.1. In the limit, as A —* 0 ; the shares obtained by players A

and B respectively in the unique SPE converge to TJA^T and T\B^', where
TB , rA
r]A = and T]B = •
T'A ~\~ 1"B YA ~\~ rB

Proof. For any A > 0

1 — exp(—7

Since when A > 0 but small, exp(—r^A) = 1 — r^A, it follows that when
A > 0 but small, /J^A = ^ A / ( r ^ + r#)A — that is, /i^ = TB/{TA + re)-
The corollary now follows immediately. •
3.2 The Basic Alternating-Offers Model 47

In the limit, as A —> 0, the relative magnitude of the players' discount

rates critically influence the equilibrium partition of the cake: the equilib-
rium share obtained by a player depends on the ratio VA/TB- Notice that
even in the limit, as both TA and TB tend to zero, the equilibrium partition
depends on the ratio TA/TB- The following metaphor nicely illustrates the
message contained in Corollary 3.1. In a boxing match, the winner is the
relatively stronger of the two boxers; the absolute strengths of the boxers
are irrelevant to the outcome.

3.2.2 Proof of Theorem 3.1

The strategy of the proof of Theorem 3.1 is as follows. I shall first estab-
lish that the payoffs to each player in any two subgame perfect equilibria
are identical. Notice that this result does not rule out the possibility that
there exists more than one subgame perfect equilibrium. However, given
this result, I shall then establish Theorem 3.1. The central argument rests
on exploiting the stationary structure that underlies the alternating-offers
game: any two subgames that begin with player i's offer are 'strategically
equivalent', in the sense that the strategic structures of such subgames are
identical. This means that the sets of subgame perfect equilibria in such
subgames are identical. Hence, the sets of SPE payoffs to player i in such
subgames are identical. Let Gi denote the set of SPE payoffs to player i
in any subgame beginning with player i's offer. Formally, Gi = {ui : there
exists a SPE in any subgame beginning with player z's offer that gives player
i a payoff equal to ui}. I shall derive the result (that the payoffs to each
player in any two SPE are identical) by showing that the maximum and
minimum values of Gi are identical. Let Mi and rrii respectively denote the
maximum and minimum payoffs that player i obtains in any SPE of any
subgame beginning with her offer.

Remark 3.1. To be precise, Mi denotes the supremum of G^, and rrii its
infimum. This is because, at this point, I do not rule out the possibility
that Gi is an open set; that is, there may not exist a SPE (in any subgame
beginning with player i's offer) that gives player i a payoff exactly equal to
Mj. However, since Gi is bounded, there will exist a SPE that does give
player i a payoff which is arbitrarily close to Mi\ hence Mi denotes the
48 The Rubinstein Model

supremum, rather than the maximum. 2 Similar remarks apply to m^.

Lemmas 3.1, 3.3 and 3.4 (stated below) establish several relationships
amongst the four unknowns, namely, amongst MA, VTLA, MB and TUB- Lemma
3.2 is an intermediate result.

L e m m a 3.1. (i) TXIA > TT — 6 BMB and (ii) TUB > TT — 6AMA-

Proof. In any SPE player B accepts any offer XA such that TT — XA> 6BMB-
Hence, there does not exist UA G GA such that UA < TT — 6BMB', otherwise
player A could increase her payoff by offering x'A such that UA < x'A <
TT — 6BMB- Therefore, for any UA G GA, UA > TT — 6BMB- Hence, TTIA >
? •
L e m m a 3.2. (i) For any UB G GB, TT — 6BUB G GA and (ii) for any
UA G GA, TT - 6AUA G GB-

Proof. Fix an arbitrary UB G GB', let a denote the SPE that supports the
payoff UB to player JB, and let player A's payoff in this SPE be denoted
by VA- Now fix an arbitrary subgame beginning with player A's offer, and
consider the following pair of strategies. Player A begins by making the
offer xrA = 7T — 6BUB- Player B accepts an offer XA if and only if XA < x'A.
If any offer is rejected, then play proceeds according to a. I claim that this
pair of strategies is a SPE. Player B's response behaviour is optimal and
credible. Player A's initial offer is optimal provided she does not benefit by
making an offer greater than x'A. This is true provided n — 6BUB > 6AVA,
which holds since UB + VA < TT- Hence, it follows that TT — 6BUB G GA- D

Lemma 3.3. (i) VTLA < TT — 6BMB and MA > TT — 6BTTIB, and (ii) TUB <
7T — 6AMA and MB > TT —

Proof. If rriA > TT — 6BMB, then there exists UB G GB such that

TT — 6BUB, which contradicts Lemma 3.2(i). Similarly, if MA < TT —
then there exists UB G GB such that MA < TT — 6BUB, which contradicts
Lemma3.2(i). •
2
It should be noted that Proposition 3.1 implies that d is non-empty.
3
By a symmetric argument (with the roles of A and B reversed) part (ii) of the lemma
can be proven. For the same reason, in the proofs of Lemmas 3.2-3.4 below I shall only
establish the first parts of each of these lemmas.
3.2 The Basic Alternating-Offers Model 49

L e m m a 3.4. (i) MA < TT — 6Bras and (ii) MB < TT — 6 AW* A-

Proof. Fix an arbitrary subgame beginning with player ^4's offer. The set of
SPE in this subgame can be partitioned into types: (i) equilibria in which
player A's initial offer is accepted, and (ii) equilibria in which player A's
initial offer is rejected. Note that, in any SPE, player B rejects any offer
XA > 7T - 8STUB > Hence, in any type (i) SPE, player A's payoff UA <
TT — 8BrnB- Now consider any type (ii) SPE. Once, in such an equilibrium,
player A's offer is rejected the game begins with player B's offer. In any
SPE in such subgames, the sum of the equilibrium payoffs VA and UB to the
players is less than or equal to TT. This implies that VA < TT — TUB- Hence, in
any type (ii) SPE, player A's payoff UA — $AVA < 8A{TT — m^). In summary,
for any UA G GA, UA < maxJTr — SBTTIB^ $A{TT — ^IB)}- Part (i) now follows
since max{7r — ^ m ^ , <5U(TT — TUB)} — TT — SBTTIB' •

Combining Lemmas 3.1, 3.3 and 3.4, it follows that

MA = 7T - 6BmB (3.5)
mA = 7T - 6BMB (3.6)
MB = TT - ^ A ^ A (3.7)
= 7T - 5 A M A . (3.8)

Solving these equations for M^, TTT,^, M^ and m^, it follows that the payoffs
to each player in any two SPE are identical:

Proposition 3.2. MA = TUA — I^A^ and MB = TUB = HB^, where \IA and
IiB are defined in (3.4)-

Notice that MA — VTLA = x*A and MB — TUB = x*B, where x*A and x*B are
defined in (3.3). Theorem 3.1 follows immediately once it is shown (using
the result contained in Proposition 3.2) that in any SPE Property 3.1 is
satisfied. The argument is by contradiction. Thus, suppose there exists a
SPE in which some player's equilibrium offer, say player A's, is rejected.
From Proposition 3.2, the SPE payoff to player A in this subgame is x\.
Now, given my supposition, x*A < 8A(K — %%)] this is because after player
A's offer is rejected her SPE payoff in the subgame beginning with player
i?'s offer is not greater than TT — x*B. Since n — x*B = 6A%% it follows that
%*A ^ b\x*A-> which is a contradiction. This result, combined with Proposition
50 The Rubinstein Model

3.2, implies that in any SPE Property 3.2 is satisfied. In particular, the offer
that player i (i = A, B) always makes is x*. Hence, Theorem 3.1 is proven.

3.2.3 Properties of the Equilibrium

Uniqueness and Efficiency
It has been shown that if r A > 0 and TB > 0, then the basic alternating-offers
game has a unique SPE, which is Pareto efficient. As I now show, if, on the
other hand, r A — ^B — 0, then there exists many (indeed, a continuum) of
subgame perfect equilibria, including equilibria which are Pareto inefficient.
Fix an arbitrary number x* G [O,TT], and consider the following pair of
strategies:
• PlayerA always offers XA = ^* and always accepts an offer XB if and only
if %B < TV — X*.
• PlayerB always offers XB = TT — x* and always accepts the offer XA if and
only if XA < #*.
It is easy to verify that if TA — TB = 0, then this pair of strategies
is a subgame perfect equilibrium. Hence, if TA — TB = 0, then any par-
tition of the cake agreed to at time 0 can be supported in a SPE. 4 It is
instructive to see why the pair of strategies described above is not a sub-
game perfect equilibrium if at least one player has a strictly positive discount
rate. Suppose, for example, TB > 0. Consider a subgame that begins with
player B having to decide whether or not to accept an offer XA such that
<$B(TT — x*) < 7T — XA < 7T — x*. Notice that such an offer exists since TB > 0
implies that 6B < 1. According to her strategy described above, player B
rejects such an offer, and her payoff is 6B(TT — x*). This means that she can
profitably deviate to an alternative strategy in which she accepts this offer,
because her payoff from such an alternative strategy is n — XA> I have thus
shown that player I?'s plan to reject the offer XA is not credible.5
4
It is straightforward to construct an SPE in which agreement is reached on an arbitrary
partition at an arbitrary time. The main ingredients of such a construction are as follows.
Along the path of play before agreement is secured, say at time £A (where t > 1), each
player insists on receiving the whole cake. If neither player deviates from this path of play,
then at time tA play proceeds according to the SPE stated above. But if a player deviates
from the path of play at any time before tA, then immediately play proceeds according
to the SPE (described above) in which she receives no share of the cake.
5
Notice that this argument is valid even if TA — 0. It is, in fact, easy to verify that
3.2 The Basic Alternating-Offers Model 51

If TA = TB = 0, then neither player cares about the time at which

agreement is struck. This means that the players do not incur any costs
by haggling (i.e., by making offers and counteroffers) — which character-
izes, what may be called, a 'frictionless' bargaining process. One important
message, therefore, of the basic alternating-offers model is that frictionless
bargaining processes are indeterminate. This seems intuitive, because, if the
players do not care about the time at which agreement is struck, then there
is nothing to prevent them from haggling for as long as they wish. On the
other hand, frictions in the bargaining process may provide the players with
some incentive to reach agreement.
The importance of the concept of perfection — the requirement that each
player's actions should always be credible — is illustrated by the following
result. For any TA > 0 and TB > 0, the pair of strategies described above is a
Nash equilibrium.6 Hence, even if TA > 0 and TB > 0, the alternating-offers
game has many (indeed, a continuum) of Nash equilibria.

Bargaining Power

It seems reasonable to assume that the share of the cake obtained by a

player in the unique SPE reflects her 'bargaining power'. Thus, a player's
bargaining power is decreasing in her discount rate, and increasing in her
opponent's discount rate. Why does being relatively more patient confer
greater bargaining power? To obtain some insight into this issue, I now
identify the cost of haggling to each player, because it is intuitive that a
player's bargaining power is decreasing in her cost of haggling, and increasing
in her opponent's cost of haggling.
In the alternating-offers game, if a player does not wish to accept any
particular offer and, instead, would like to make a counteroffer, then she is
free to do so, but she has to incur a 'cost': this is the cost to her of waiting
A time units. The smaller is her discount rate, the smaller is this cost. That
is why being relatively more patient confers greater bargaining power.
Notice that even if both players' costs of haggling become arbitrarily
small in absolute terms (for example, as A —» 0), the equilibrium partition

Proposition 3.1 and Theorem 3.1 are valid for any TA > 0 and r# > 0 such that rA+re > 0.
6
This is trivial to verify. For example, given player S's strategy, deviation by player A
to any alternative strategy does not give her a payoff greater than x*.
52 The Rubinstein Model

depends on the relative magnitude of these costs (as captured by the ratio

Relationship with Nash's Bargaining Solution

It is straightforward to verify that the limiting, as A —» 0, SPE payoff pair
(T]A7T^T]B7T) — as stated in Corollary 3.1 — is identical to the asymmetric
Nash bargaining solution of the bargaining problem (fi,d) with r = TJA,
where ft = {(UA,UB) ' 0 < UA < TT and UB = TT — UA\ and d — (0,0) —
where the asymmetric Nash bargaining solution is stated in Definition 2.3. 7
This remarkable result — which is generalized in Corollary 3.4 — pro-
vides a strategic justification for Nash's bargaining solution. In particular,
it provides answers to the questions of why, when and how to use Nash's
bargaining solution. The asymmetric Nash bargaining solution is applicable
because the bargaining outcome that it generates is identical to the (limiting)
bargaining outcome that is generated by the basic alternating-offers model.
However, it should only be used when A is arbitrarily small, which may be
interpreted as follows: it should be used in those bargaining situations in
which the absolute magnitudes of the frictions in the bargaining process are
small. Furthermore, it should be defined on the bargaining problem (fi,d),
where Q is the set of instantaneous utility pairs obtainable through agree-
ment, and d is the payoff pair obtainable through perpetual disagreement.
The following definition is useful.

Definition 3.1 (Impasse Point). The payoff pair obtainable through per-
petual disagreement — that is, the payoffs obtained by the players if each
player always rejects any offer made to her — is called the impasse point,
which is denoted generically by (Z

7
As has been noted in Section 2.8 (immediately after Definition 2.3), for any bargaining
problem (f2, d) £ E, if r = 1/2 then the asymmetric Nash bargaining solution of (Q,d)
is identical to the Nash bargaining solution of (fi, d), where the Nash bargaining solution
is stated in Definition 2.1. Hence, it follows from the above result that, in the limit, as
A —> 0, the unique SPE payoff pair of the basic alternating-offers model converges to
the Nash bargaining solution of the bargaining problem (fi, d) if and only if the players'
discount rates are identical (i.e., TA = re). This result makes sense, because (unlike an
asymmetric Nash bargaining solution) the Nash bargaining solution embodies a symmetry
axiom that would not be satisfied if the players have different discount rates.
3.2 The Basic Alternating-Offers Model 53

In the basic alternating-offers model studied above, the impasse point

(TA,%B) — (0,0). It has been shown here that the disagreement point in
Nash's bargaining solution should be identified with the impasse point of
the basic alternating-offers model.8 Finally, it should be noted that since
the parameter r in Nash's set-up reflects the 'bargaining power' of player
A relative to that of player i?, and 1 — r reflects the 'bargaining power'
of player B relative to that of player A, it is plausible that T — T]A and
1 - T = VB-

3.2.4 The Value and Interpretation of the Alternating-Offers

Model
The basic alternating-offers game is a stylized representation of the following
two features that lie at the heart of most real-life negotiations:
• Players attemptto reach agreement by making offers and counteroffers.
• Bargaining imposes costson both players.
As such the game is useful because it provides a basic framework upon
which one can build richer models of bargaining — this is shown in later
chapters where, for example, I shall study extensions of this game that
incorporate other important features of real-life negotiations. Furthermore,
since this game is relatively plausible and tractable, it is attractive to embed
it (or, some extension of it) in larger economic models. An important value,
therefore, of the basic alternating-offers game is in terms of the insights that
its extensions deliver about bargaining situations, and also in terms of its
usefulness in applications.
Another important contribution of this model is that it provides a justi-
fication for the use of Nash's bargaining solution. Furthermore, as is shown
in later chapters, its extensions can guide the application of the Nash bar-
gaining solution in relatively more complex bargaining situations.
The basic alternating-offers model incorporates several specific assump-
tions that I now interpret. First, I shall interpret the (infinite horizon)
assumption that the players may make offers and counteroffers forever. Is
this assumption plausible, especially since players have finite lives? I now
argue that, when properly interpreted, this modelling assumption is com-
8
As will be shown in Corollary 3.4 and in later chapters, this result turns out to be
fairly general and robust.
54 The Rubinstein Model

pelling. The assumption is motivated by the observation that a player can

always make a counteroffer immediately after rejecting an offer. This obser-
vation points towards the infinite horizon assumption, and against the finite
horizon assumption.9 Suppose, for example, a seller and a buyer are bar-
gaining over the price of some object, and have to reach agreement within
a single day. The players' strategic reasoning (and hence their bargaining
behaviour) is typically influenced by their perception that after any offer
is rejected there is room for at least one more offer. This suggests that
the infinite horizon assumption is an appropriate modelling assumption —
notwithstanding the descriptive reality that bargaining ends in finite time.
The time interval A between two consecutive offers is a parameter of the
game, and it is assumed that A > 0. I now argue that attention should, in
general, focus on arbitrarily small values of A. The argument rests on the
observation that since waiting to make a counteroffer is costly, a player will
wish to make her counteroffer immediately after rejecting her opponent's
offer. Unless there is some good reason that prevents her from doing so, the
model with A arbitrarily small is the most compelling and least artificial.
Why not then set A = 0? There are two reasons for not doing so. The
straightforward reason is that it does take some time, albeit very small, to
reject an offer and make a counteroffer. The subtle reason is that such a
model fails to capture any friction in the bargaining process. The model
with A > 0, on the other hand, does contain frictions — as captured by
the players' positive costs of haggling. To further illustrate the difference
between the model with A = 0 and the model with A strictly positive but
arbitrarily small, notice the following. In the limit as A —> 0 the relative
magnitude of the players' costs of haggling is well defined, but if A = 0
then this relative magnitude is undetermined (since zero divided by zero
is meaningless). This difference is of considerable significance, since, as I
argued above, it is intuitive that the equilibrium partition depends critically
on the players' relative bargaining powers — as captured by the relative
magnitude of the players' costs of haggling — and not so much on absolute
9
The finite horizon version of the alternating-offers game begs the following question.
Why should the players stop bargaining after some exogenously given number of offers
have been rejected? Since they would prefer to continue bargaining, somehow they have
to be prevented from doing so. What, or who, prevents them from continuing to attempt
to reach agreement? Unless a convincing answer is provided, the finite horizon assumption
is implausible.
3.3 An Application to Bilateral Monopoly 55

magnitudes.10
The discount factor may be interpreted more broadly as reflecting the
costs of haggling — it need not be given a literal interpretation. For example,
set TA = TB = r. The literal interpretation is that r is the players' common
discount rate. An alternative interpretation is that r is the rate at which
the cake (or, 'gains from trade') shrinks. For example, when bargaining over
the partition of an ice cream, discounting future utilities is an insignificant
factor. The friction in this bargaining process is that the ice cream is melting,
where the rate r at which the ice cream is melting captures the magnitude
of this friction.

3.3 An Application to Bilateral Monopoly

Consider a market for some intermediate good with a single seller S and a
single buyer B. If the buyer buys a quantity q at a price p, then the profits
to the players are as follows

UB(p,q) = R(q)-pq (3.9)

Us(p,q)=pq-C(q), (3.10)

where R{q) is the revenue obtained by the buyer by transforming the quan-
tity q of the input into some output and then selling the output on some
final good market, and C{q) is the cost to the seller of producing the quan-
tity q of the input. Assume that R(0) = 0, R'(q) > 0, R"(q) < 0, C(0) = 0,
C'(q) > 0, C"{q) > 0 and R'(0) > C"(0).
The players bargain over the price and quantity of trade according to
the alternating-offers procedure. An offer is a pair (p, g), where p > 0 and
q > 0. If the seller and the buyer reach agreement at time £A on a pair
(p, q)1 then player i's (i = B, S) payoff is H(p, q) exp(—r^tA). On the other
hand, if the players perpetually disagree, then each player's payoff is zero.

Bilateral Monopoly Equilibrium

Consider a SPE that satisfies Properties 3.1 and 3.2. Let (p*, q*) denote the
equilibrium offer that player i makes whenever she has to make an offer.
10
In the limit as A —> 0 the first-mover advantage disappears — hence this is an addi-
tional reason for focusing on arbitrarily small values of A.
56 The Rubinstein Model

Furthermore, let Vf = Ili(p*,q*).

It follows from Properties 3.1 and 3.2 that the buyer's payoff from re-
jecting any offer is 6BV£. Perfection requires that the buyer accept any
offer (ps,qs) such that UB{ps,Qs) > ^BVB, and reject any offer (ps,qs)
such that H-B(ps,qs) < $BVB- Furthermore, Property 3.1 implies that
^B{Ps,qs) > SBVJB. However, UB{PS^QS) ^ ^BVB\ for otherwise, the seller
could increase her profit by offering a slightly higher price (i.e., by instead
offering a pair (pfSl q$) where pfs is slightly greater than p*s). Hence

nB(p*s,q*s)=6BV£. (3.11)

Furthermore, optimality requires that the seller's equilibrium offer (p*s,q$)

maximize her profit Tls(ps,qs) subject to lisips^qs) — ^BV^.11 This im-
plies that

q*s = <f, (3.12)

where qe is the unique solution to R'(q) — C'(q). Using (3.11), it follows

that

Vg = 7T - 6BVB, where (3.13)

e e
7r = R(q )-C(q ). (3.14)

By a symmetric argument (with the roles of B and S reversed), it follows

that

(3.15)
(3-16)
(3.17)

11
Suppose, to the contrary, that there exists a pair (p's,q's) s u c n that
Us(p*s, qs) a n ( l ns(P5, q's) = SBVB- Notice that the buyer is indifferent between accepting
and rejecting the offer (ps^q's)- If5 i n equilibrium, the buyer accepts the offer (ps,qfs)i
then the seller can increase her profit by instead offering the pair (p's^q's)- Suppose, on
the other hand, that, in equilibrium, the buyer rejects the offer (ps^q's)- ^n ^na^ case> the
seller can increase her profit by instead offering the pair (p'g, qfs) where p's is slightly lower
than p's (i.e., p's is such that ns(ps,q's) > us(p*s,q*s) and nB(ps,q's) > &BVB).
3.3 An Application to Bilateral Monopoly 57

Equations 3.13 and 3.17 have a unique solution, namely

VB
~
Vs = ^f. (3-19)
1 6o
Hence, since V* = TU{p*,q*), it follows (using (3.9), (3.10), (3.12), (3.14)
and (3.16)) that

(3 20)
'

I have established that there exists at most one SPE that satisfies Properties
3.1 and 3.2. In this SPE:
• The buyer always offers the pair (p*B,q%) and always accepts an offer
(PS, Qs) if and only if UB(ps, qs) > foVg, where p*B, q*B and VB are respec-
tively defined in (3.20), (3.16) and (3.18).
• The seller always offers the pair (p*s^q^) and always accepts an offer
(PBIQB) if and only if I I ^ ^ ^ , ^ ) > ^V^, where p^, q^ and V$ are re-
spectively defined in (3.21), (3.12) and (3.19).
It is easy to verify (by a straightforward adaptation of the proof of
Proposition 3.1) that this pair of strategies is a subgame perfect equilib-
rium. Here I shall verify that the buyer's strategy is optimal at any point
in time tA when she has to make an offer, given the seller's strategy. If she
uses the strategy stated above, then her payoff is VB. Consider any alter-
native strategy for the buyer, where (p^, qfB) denotes the offer she makes at
time tA. If this offer is such that IIs(p^, qfB) > 6sVg, then, since the seller
accepts such an offer and since (p*B,q%) maximizes I I ^ p ^ , ^ ) subject to
^S(PBIQB) > $sVsi deviation to this alternative strategy is not profitable.
Now suppose that the offer is such that Hsip^^q^) < dsVs- I n this case
the seller rejects the offer made at time tA. Since the seller always rejects
any offer (PB,QB) such that Hs(PB,qB) < ^S^s a n d always offers (p^,^),
the buyer's payoff from this alternative strategy is less than or equal to
^),5^Vg}, where VB is the maximum value of
58 The Rubinstein Model

subject to US(PB, QB) > feVg •Using (3.11), it follows that deviation to this
alternative strategy is not profitable.
Through a straightforward adaptation of the arguments in Section 3.2.2,
it can be proven that this SPE is the unique subgame perfect equilibrium
of the bilateral monopoly model (cf. Theorem 3.2, below).

Properties of the Bilateral Monopoly Outcome

In the unique SPE, the equilibrium quantity traded qe maximizes the sur-
plus (or, gains from trade) R(q) — C(q). This suggests that the bilateral
monopoly equilibrium is Pareto efficient — that is, there does not exist an-
other outcome which makes at least one player strictly better off without
making any player worse off. I now show that an outcome is Pareto effi-
cient if and only if agreement is reached at time 0 on a pair (p, q) such that
q — qe. First, notice that an outcome in which agreement is reached at time
£A where t > 1 on a pair (p,q) is not Pareto efficient, because (due to dis-
counting) both players are better off with the outcome in which agreement
is reached at time 0 on the same pair (p, q). The set of Pareto efficient pairs
(p, q) is then derived by maximizing Hs(p, q) subject to HB(P, Q) > ®-> where
a is some arbitrary number. At the optimum, the constraint binds. Thus,
after substitution, it immediately follows that a pair (p, q) is a solution to
this constrained maximization problem if and only if q = qe.
The equilibrium trade price is p*B if the buyer makes the offer at time 0,
and p*s if the seller makes the offer at time 0. In the limit as A —> 0 it does
not matter who makes the offer at time 0, since, as A —> 0, both p*B and p*s
converge to the same price p*, where

» =

Notice that the equilibrium trade price is a convex combination of the equi-
librium average revenue and average cost. Thus, the equilibrium trade price
is unrelated to the equilibrium marginal cost C'(qe). The intuition behind
this result is straightforward: the seller and the buyer set quantity to the
level which maximizes the gains from trade R(q) — C(q), and use the price
as an instrument to divide the generated surplus R(qe) — C(qe). Thus, for
example, if the ratio TB/TS is close to zero (which means that the buyer is
3.4 A General Model 59

significantly more patient than the seller), then p* is close to the equilibrium
average cost, which implies that the seller's equilibrium profit is close to zero
while the buyer receives most of the generated surplus.
Comparing (3.18)-(3.19) with (3.3)-(3.4), one may interpret the equilib-
rium payoffs in this bilateral monopoly model as the equilibrium shares the
players obtain when bargaining over a cake of size R(qe) — C(qe).

3.4 A General Model

I now study a general version of Rubinstein's model. Let X denote the
set of possible agreements.12 Two players, A and I?, bargain according to
the alternating-offers procedure in which an offer is an element from the
set X. If the players reach agreement at time tA on x G I , then player
i's (i = A,B) payoff is Ui(x) exp(-r^A), where U\ : X —> 3ft is player i's
utility function. For each x E X, Ui(x) is the instantaneous utility player
i obtains from agreement x. If the players perpetually disagree (i.e., each
player always rejects any offer made to her), then each player's payoff is
zero.
The set of possible utility pairs Q = {(UA,V>B) •there exists x E X such
that UA{X) — UA and UB(X) = UB} is the set of instantaneous utility pairs
obtainable through agreement. Thus, the set of possible utility pairs obtain-
able through agreement at time tA is fit = {(UAS^UBS^) : (UA,V>B) E fi}.
It should be noted that fi° = £1
The Pareto frontier Qe of the set Q is a key concept in the analysis of
the subgame perfect equilibria. A utility pair (UA->UB) E £le if and only if
[UA-, UB) E ft and there does not exist another utility pair (ufA, urB) E Q such
that u'A > UAI U'B > UB and for some z, v!i > U{. In order to study the
subgame perfect equilibria, I do not need to make any assumptions directly
about X, UA and Us- The assumptions I make (stated below in Assumption
3.1) are about the Pareto frontier of ft. However, since this concept is derived
from X, UA and UB, Assumption 3.1 does indirectly restrict the nature of
these three objects.13
12
The set of agreements in the basic alternating-offers model is the set of partitions of
the cake, namely, {(XA,XB) •0 < XA < TT and XA+ %B — TT}, and the set of agreements
in the bilateral monopoly model is {(p, q) : p > 0 and q > 0}.
13
For example, Assumption 3.1 implies that X has a continuum of elements — thus, X
60 The Rubinstein Model

Assumption 3.1. The Pareto frontier Qe of the set Q is the graph of a

concave function, denoted by </>, whose domain is an interval IA ^ 3? and
range an interval IB C !ft, with 0 G IA , 0 G IB and 0(0) > 0.

It should be noted that Assumption 3.1 implies that Vte — {(UA,UB) :

UA £ IA and ?X£ = 0(^^)}. Notice that (by the definition of the Pareto
frontier) 0 is a strictly decreasing function. Denote by (j)~1 the inverse of (f>.
By Assumption 3.1, it follows that (j)~1 is a strictly decreasing and concave
function from IB to IA , with (j)~1(0) > 0. It should be noted that for any
UA £ IA, (/>(UA) is the maximum utility player B receives subject to player A
receiving a utility of at least UA- Similarly, for any UB E IB, 4)~1{UB) is the
maximum utility player A receives subject to player B receiving a utility of
at least UB-
It can be verified that Assumption 3.1 is satisfied in the basic alternating-
offers model and in the bilateral monopoly model. In the former model UB =
<K^A) = TT — UA where I A = [0, TT] and IB = [O,TT], and in the latter model
us = (/)(uB) = 7T - uB where IB = (-oc, R(qe)} and Is = [-C(qe), +oo).

3.4.1 The Subgame Perfect Equilibria

Consider a SPE that satisfies Properties 3.1 and 3.2. Let x* denote the
equilibrium offer that player i makes whenever she has to make an offer.
Furthermore, let !/•*= Ui{x\) and W* = Ui{x)) (J + i).
It follows from Properties 3.1 and 3.2 that player 5 ' s payoff from re-
jecting any offer is 6BV^. Perfection requires that player B accept any
offer XA such that UB(XA) > ^BV^ and reject any offer XA such that
UB(XA) < SBVQ. Furthermore, Property 3.1 implies that WB > SBVB-
I now establish that player B is indifferent between accepting and rejecting
player A's equilibrium offer.14 That is,

W*B = 6BV^. (3.22)

Suppose, to the contrary, that WB > SBV^. Since 8BVB > 0 and 0(0) >
W^, 15 Assumption 3.1 ensures the existence of an agreement xfA G X such
is not a finite or a countably infinite set.
14
The formal arguments presented below — which establish (3.22)-(3.24) — may be
illustrated using a diagram that depicts the graph of c/)^1.
15
VB ^ 0? because player B can guarantee a payoff of zero by always offering x'B G
3.4 A General Model 61

that UB(xfA) = u'B and UA(X'A) = (j)~l(u'B), where WB > u'B > 6BVB.
Player B accepts x'A. Since (f)~1 is strictly decreasing, (f)~l(urB) > 4>~1(WB).
Hence (j)~l{ufB) > V|, because (p~1(WB) > VA. A contradiction is thus
obtained, because player A can increase her payoff by instead offering x'A.
I now establish that player A's equilibrium offer x*A maximizes her utility
UA(%A) subject to UB(XA) = $BVB. That is

VA > UA(xA) for all xAeX such that UB(xA) = $BVB. (3.23)

Suppose, to the contrary, there exists an x'A E X such that CTA(X^) > VA
and UB(x'A) = 6BV£. Since V^ > 0 and ^(SBV^) > UA(x'A), Assumption
3.1 ensures the existence of an agreement x"A E X such that UA(X;A) =
uA and UB(xfA) = (f)(u'A), where CTA(^A) > UA > ^4* Since 0 is strictly
decreasing, (j)(UA{x'j^)) < (/>(ufA). Hence <p(ufA) > 6BVB, because UA(X'A) <
(/)~1(SBVB) implies (/)(UA(X'A)) > SBVB. Therefore, player B accepts xA. A
contradiction is thus obtained, because player A can increase her payoff by
instead offering x"A.
It immediately follows from (3.23) that

VX = r\SBV^). (3.24)
By a symmetric argument (with the roles of A and B reversed), it follows
that

W*A = 8AVX (3.25)

VJB > UB(xB) for all xB € X such that UA{xB) = SAVA (3.26)
(3.27)

I now show that equations 3.24 and 3.27 have a unique solution in VA
and Vg. Furthermore, I shall show that this unique solution is such that
0<VX< ^ ( O ) and 0 < Vg < 0(0). Define, for each VA € IA

T(VA) = <f>(VA) - SB4>(8AVA). (3.28)

X such that (UA{X'B),UB{X'B)) — (0 1(0),0) and always rejecting any offer XA € X.
Similarly, VX > 0, W% > 0 and WB > 0. Now suppose, to the contrary, that WB >
0(0). Using Assumption 3.1, this implies that WB G IB and 4>~1(WB) < 0. Since
VX < ^(WB), it follows that VX < 0, which is a contradiction. Similarly, VB < 0(0),
VX < ^ ( 0 ) and WX < ^ ( O ) .
62 The Rubinstein Model

Assumption 3.1 implies that V is strictly decreasing and continuous, F(0) > 0
and r(0~ 1 (O)) < 0. The desired conclusions follow immediately.
The uniqueness of the solution to equations 3.24 and 3.27 means that
in any SPE that satisfies Properties 3.1 and 3.2 the players' equilibrium
payoffs are uniquely defined. Let Z{ denote the set of maximizers of Ui(xi)
subject to Uj(xi) = 6jV* (j ^ i). Assumption 3.1 ensures that Z^, which
is a subset of X, is non-empty. However, Assumption 3.1 does not rule
out the possibility that Z{ contains more than one element.16 The following
proposition characterizes the set of all subgame perfect equilibria that satisfy
Properties 3.1 and 3.2.

Proposition 3.3. For any x*A G ZA and x*B G ZB, the following pair of
strategies is a subgame perfect equilibrium of the general Rubinstein model:
• Player A always offers x*A and always accepts an offer XB if and only if

Player B always offers x*B and always accepts an offer XA if and only if

Proof. The proof involves a minor adaptation of the proof of Proposition 3.1.
Here I verify that player B's strategy is optimal at any point in time tA when
she has to make an offer, given player A's strategy. If she uses the strategy
stated above, then her payoff is VB. Consider any alternative strategy for
player B, where xlB G X denotes the offer she makes at time tA. If this offer
is such that UA^Q) > 6AVX-> then, since player A accepts such an offer and
since x*B maximizes UB(XB) subject to UA{XB) > SAVXI deviation to this
alternative strategy is not profitable. Now suppose that the offer is such
that UA^B) < SAVX- In this case player A rejects the offer made at time
tA. Since player A always rejects any offer XB such that UA(%B) < $AVX
and always offers x*A, player I?'s payoff from this alternative strategy is less
than or equal to maxj^VF^, 82BVB}, where VB is the maximum value of
UB(XB) subject to UA(XB) ^ SAVX- Using (3.22), it follows that deviation
to this alternative strategy is not profitable. •

If ZA contains a unique element and ZB contains a unique element,

then the SPE described in Proposition 3.3 is the unique SPE that satisfies
16
Given Assumption 3.1, a sufficient condition which ensures that Zi contains a unique
element is as follows: for any UA G IA and any utility pair (UA,4>(UA)) G ^ e there exists
a unique element x of X such that UA(X) — UA and UB(X) — <J)(UA)-
3.4 A General Model 63

Properties 3.1 and 3.2.17 If, on the other hand, for some i (i = A or i = B)
Zi contains more than one element, then there exists more than one SPE
satisfying these two properties — although in any such SPE each player
receives the same payoff.
I now establish the main results about the set of subgame perfect equi-
libria. It is useful (especially for future reference) to state the results in two
theorems.

Theorem 3.2. If ZA contains a unique element and ZB contains a unique

element, then the unique subgame perfect equilibrium described in Proposi-
tion 3.3 is the unique subgame perfect equilibrium of the general Rubinstein
model.

Proof. The proof involves a straightforward adaptation of the proof of The-

orem 3.1 — and is contained in Section 3.4.4 below. •

Theorem 3.3. If for some i (i = A or i = B) Zi contains more than one

element, then there exists more than one subgame perfect equilibrium in the
general Rubinstein model, which may be characterized as follows. Whenever
player i (i = A, B) has to make an offer, she offers an element from the set
Zi, and she accepts an offer Xj (j / i) if and only if Ui(xj) > 6iV*.

Proof. The proof is in Section 3.4.4, below.18 •

In any SPE, agreement is reached at time 0. Since it is player A who

makes the offer at time 0, her equilibrium payoff is V^ and player U's equi-
librium payoff is </>(V^). The equilibrium payoff pair is Pareto efficient.19
The result stated in the following corollary implies that each player's unique
SPE payoff is strictly increasing in her discount factor, and strictly decreas-
ing in her opponent's discount factor. Thus, the more patient a player is
relative to her opponent, the higher is her unique SPE payoff.
17
ZA contains a unique element if and only if there exists a unique element x A of X
such that UA(X*A) = VX a n c l UB(%A) — <KVA)- ^B contains a unique element if and only
if there exists a unique element xB of X such that UB(XB) = VB and UA(XB) = 4>~1(VB)-
18
It should be noted that the differences between any two subgame perfect equilibria
are insignificant, because any two elements of the set ZA (resp., ZB) generate the same
payoff pair, namely, (VZ,<f>(VX)) (resp., ( ^ ( V g ) , ^ ) ) .
19
If player B makes the first offer at time 0, then her equilibrium payoff is VB and player
A's equilibrium payoff is 4>~1(VB)-
64 The Rubinstein Model

Corollary 3.2. V£ is strictly increasing in 6A and strictly decreasing in fo.

Proof. For any pair of discount factors 6A and fo, V^ is the unique number
at which the function F — defined in (3.28) — is equal to zero. It can
be shown, using Assumption 3.1, that for any VA E IA, T(VA) is strictly
increasing in 6A and strictly decreasing in 6B- The desired conclusions follow
immediately. •

3.4.2 Small Time Intervals

I argued in Section 3.2.4 that, in general, attention should be focused on
arbitrarily small values of A. Hence, I now characterize the unique SPE
payoffs in the limit as A —> 0.20

Corollary 3.3. In the limit, as A —>> 0 ; the payoff pairs (V^, (t>{YX))
((/)~1(1/^), Vg) converge to the unique solution of the following maximization
problem

m8x(uA)riA(uB)riB

subject to (UA,UB) G O e ; UA > 0 and UB > 0, where T\A and T\B are defined
in Corollary 3.1.

Proof. For notational convenience, define for each UA > 0 and UB > 0

where rJA = Info/(In£4 + Info) and ffs = 1 — fjA- Using (3.24) and
(3.27), and noting that ffAln6A = felnfo, it follows that N(VX, (t>(VA)) =
N{())-l(V£),V£). Furthermore, (3.24) implies that VJ > (j)~l{V£). Thus,
as is illustrated in Figure 3.1, the two distinct utility pairs (V^, </>(V^)) and
lie o n t h e
(^(V&'VB) graPh of t h e
Unction N(uA,uB) = N{VXA{VX))
and on the graph of the function <j>. From (3.24) and (3.27), it follows that
as A -> 0, VX ~ <t>~l{V^) -> 0 and Vg - ^(VJ) - • 0. Hence, since, as
A -> 0, 7M ^ r/A and ffB -+ i)B, the pairs (VX,(/>(VX)) and ( ^ ( V g ) , Vg)
converge to the unique solution of the maximization problem denned in the
corollary. •
3.4 A General Model 65
uB

UA
o "
F i g u r e 3.1: An illustration of the proof of Corollary 3.3.

Let (u*A,u*B) denote the unique solution to the maximization problem

stated in Corollary 3.3, and x* G X be such that Ui(x*) = u* (i = A,B).
Although such a limiting (as A —> 0) SPE agreement exists, it should be
noted that Assumption 3.1 does not rule out the possibility that there exists
more than one such limiting SPE agreement. Of course, any limiting SPE
agreement generates the payoff pair (u*A,u*B).
It is useful to note — and this follows immediately from Corollary 3.3
— that x* is a solution to the following maximization problem

3.4.3 Relationship with Nash's Bargaining Solution

I now show that the limiting, as A —> 0, SPE payoff pair (u*A, u*B) is identi-
cal to an asymmetric Nash bargaining solution of an appropriately defined
20
As is shown in Corollary 3.3, in the limit as A —>• 0 a player's equilibrium payoff does
not depend on which player makes the first offer at time 0; in this limit, a first-mover
advantage does not exist.
66 The Rubinstein Model

bargaining problem — the latter bargaining solution is stated in Definition

2.3. I first state this remarkable result, which is an immediate consequence
of Corollary 3.3, and then I shall discuss it.
Corollary 3.4 (Relationship with Nash's Bargaining Solution). In
the limit, as A —» 0; the unique SPE payoff pair in Rubinstein's model con-
verges to the asymmetric Nash bargaining solution of the bargaining problem
(Q,d) with r = 7]A, where Q = {(UA,UB) : there exists x E X such that
UA{X) = UA and UB(X) = UB}, d = (0,0) and TJA is defined in Corollary
3.1.
Proof. First note that since the Pareto frontier Qe of the set Q (where Q is
as stated in the corollary) satisfies Assumption 3.1, the bargaining problem
(Q,d) (where d = (0,0)) satisfies Assumptions 2.1 and 2.2. The desired
result follows immediately by noting from Definition 2.3 that the asymmetric
NBS of the bargaining problem (fi, d) with r = T\A is identical to the unique
solution of the maximization problem stated in Corollary 3.3. •
As has been noted in Section 2.8 (immediately after Definition 2.3), for
any bargaining problem ($7, d) £ S, if r = 1/2 then the asymmetric Nash
bargaining solution of (fi, d) is identical to the Nash bargaining solution of
(f2, d)1 where the Nash bargaining solution is stated in Definition 2.1. Hence,
it follows from Corollary 3.4 that in the limit, as A —» 0, the unique SPE
payoff pair in Rubinstein's model converges to the Nash bargaining solution
of the bargaining problem (fi, d) if and only if the players' discount rates
are identical (i.e., r^ = r#), where (fi,d) is as defined in Corollary 3.4.
This result makes sense, because (unlike an asymmetric Nash bargaining
solution) the Nash bargaining solution embodies a symmetry axiom that
would not be satisfied if the players have different discount rates.
The remarkable result contained in Corollary 3.4, which generalizes the
result discussed in Section 3.2.3, provides a strategic justification for Nash's
bargaining solution. In particular, it provides answers to the questions of
why, when and how to use Nash's bargaining solution. The asymmetric
Nash bargaining solution is applicable because the bargaining outcome that
it generates is identical to the (limiting) bargaining outcome that is gen-
erated by Rubinstein's model. However, it should only be used when A is
arbitrarily small, which may be interpreted as follows: it should be used in
those bargaining situations in which the absolute magnitudes of the frictions
3.4 A General Model 67

in the bargaining process are small. Furthermore, it should be defined on

the bargaining problem ($1, d), where tt is the set of instantaneous utility
pairs obtainable through agreement and d = (0,0).
In Rubinstein's model studied here, the impasse point (1A, 1B) = (0,0),
where the impasse point is the payoff pair obtainable through perpetual
disagreement (cf. Definition 3.1). It has been shown here that the dis-
agreement point in Nash's bargaining solution should be identified with the
impasse point of Rubinstein's model. Finally, it should be noted that since
the parameter r in Nash's set-up reflects the 'bargaining power' of player
A relative to that of player i?, and 1 — r reflects the 'bargaining power'
of player B relative to that of player A, it is plausible that r = T\A and
1 - r = 7]B-

3.4.4 Proof of Theorems 3.2 and 3.3

The strategy of the proof is similar to that of Theorem 3.1. Let G{ denote
the set of SPE payoffs to player i in any subgame beginning with player i's
offer, and let Mi and rrii respectively be the supremum and infimum of G{.
The following lemma establishes a preliminary result.

L e m m a 3.5. For any UA £ GA, 0 < UA < 0~1(O); and for any UB £
0<uB<

Proof. Since player A can guarantee a payoff of zero by always offering

x'A E X such that (UA(X'A), UB(X'A)) = (0,0(0)) and always rejecting any
offer XB £ X, it follows that UA > 0 (for all UA £ GA)- By a symmetric
argument, it can be shown that UB > 0 (for all UB £ GB)- NOW suppose, to
the contrary, that for some UA £ GA, ^A > 0 1 (0). Using Assumption 3.1,
this implies that UA £ I A a n d <J)~1(UA) < 0. Let VB denote the equilibrium
payoff to player B in the SPE (of the subgame beginning with player A's
offer) that supports the payoff UA to player A. Since VB < 4>~1(UA)I it
follows that VB < 0. A contradiction is thus obtained, because player B can
guarantee a payoff of zero. By a symmetric argument, it can be shown that
uB < 0(0) (for all uB £ GB). •

Lemma 3.6. (i) TTIA > (J)~1(8BMB) and (ii) TUB > (J)(8AMA)-
68 The Rubinstein Model

Proof. In any SPE player B accepts any offer XA such that UB(%A) >
6BMB21 Hence, there does not exist UA G GA such that UA < (J)~1(8BMB)\
otherwise player A could increase her payoff by offering x'A such that UA{%'A) =
urA and UB{X'J^) — c/)(u'A), where UA < u'A< (/)~1(8BMB)- Therefore, for any
UA £ GA-> UA > 4>~1{8BMB)- Hence, TTIA > 4>~1(^BMB)22 •

L e m m a 3.7. (i) For any UB G GB, ^~1{^BUB) G GA and (ii) for any
UA G GA, ^(SAUA) G G B-

Proof. Fix an arbitrary UB G GB] let a denote the SPE that supports the
payoff UB to player B, and let player A's payoff in this SPE be denoted
by VA- NOW fix an arbitrary subgame beginning with player A's offer, and
consider the following pair of strategies. Player A begins by making the
offer x'A such that UB{X'A) = ^BUB and UA(X'A) = (/)'~1(6BUB)-23 Player B
accepts an offer XA if and only if UB(X'A) — ^BUB- If any offer is rejected,
then play proceeds according to a. I claim that this pair of strategies is
a SPE. Player i?'s response behaviour is optimal and credible. Player A's
initial offer is optimal provided she does not benefit by making an offer XA
such that UB(XA) < £>BUB> This is true provided (J)~1(6BUB) > SAVA, which
holds since VA < 4>~1{UB)' Hence, it follows that C/)~1(6BUB) G GA- Q

L e m m a 3.8. (i) TUA < 4>~1(6BMB) and MA > 4>~l

and MB > <t>{f>A™>A)-

Proof. If rriA > <^>~1(SBMB)1 then there exists UB G GB such that

1
(J)~1{SBUB)-) which contradicts Lemma 3.7(i). Similarly, if MA <
then there exists UB G GB such that MA < 4>~1(^BUB)^ which contradicts
Lemma3.7(i). •

Lemma 3.9. (i) MA < 0~1(^B^JB) and (ii) MB <

Proof Fix an arbitrary subgame beginning with player A's offer. The set of
SPE in this subgame can be partitioned into types: (i) equilibria in which
21
Notice that Lemma 3.5 implies 0 < 6BMB < <^(0). This, in turn, implies 6BMB G IB
and 0 < 4)-1(6BMB) < (t>~\0).
22
By a symmetric argument (with the roles of A and B reversed) part (ii) of the lemma
can be proven. For the same reason, in the proofs of Lemmas 3.7-3.9 below I shall only
establish the first parts of each of these lemmas.
23
Lemma 3.5 implies that 6BUB E i s , and hence such an offer does exist.
3.5 An Application to a Two-Person Exchange Economy 69

player A's initial offer is accepted, and (ii) equilibria in which player A's
initial offer is rejected. Note that, in any SPE, player B rejects any offer
XA such that UB(XA) < 8BmB. Hence, in any type (i) SPE, player A's
payoff UA < 4 ~ (SB'^B)-
> 1
NOW consider any type (ii) SPE. Once, in such
an equilibrium, player A's offer is rejected the game begins with player B's
offer. In any SPE in such subgames the equilibrium payoffs VA and uB to
the players are such that VA < (p~1(uB). This implies that VA < 4>~l{rnB)'
Hence, in any type (ii) SPE, player A's payoff UA — SAVA < ^A<t)~l{rr^B)' In
summary, for any UA £ GA, UA < m a x l ^ ^ ^ m ^ ) , SA4>~1(TTIB)}' Part (i)
now follows since max{0~1(5jB?7i^), ^ 0 ~ 1 ( m ^ ) } = (f)~1(6BmB). •

Combining Lemmas 3.6, 3.7 and 3.9, it follows that

\ (3.29)
(3.30)
MB = 0(8AmA) (3.31)
(3.32)
It immediately follows from the result that equations 3.24 and 3.27 have a
unique solution that

MA = mA = VX and MB = mB = Vg. (3.33)

Using (3.33), I now establish that in any SPE Property 3.1 is satisfied.
Suppose, to the contrary, there exists a SPE in which some player's equi-
librium offer, say player A's, is rejected. The SPE payoff to player A in
this subgame is F | . Now, given my supposition, V£ < SA^1^^)] this
is because after player A's offer is rejected her SPE payoff in the subgame
beginning with player .B's offer is not greater than </>~1(Vg). It follows that
y% < ^ A ^ H ^ B ^ A ) ? w hich is a contradiction. This result, combined with
(3.33), implies that whenever player i (i = A, B) has to make an offer, she
offers an element from Zj. Theorems 3.2 and 3.3 are thus proven.

3.5 An Application to a Two-Person Exchange

Economy
Consider an exchange economy with two individuals (or, players), A and
B, and two (perfectly divisible) goods, namely, bread and wine. Player A
70 The Rubinstein Model

owns one unit of bread and player B owns one unit of wine. The set of
possible agreements is X = {(##, ys) •0 < XB < 1 and 0 < ys < 1}, where
XB and ys are respectively the quantities of bread and wine obtained by
player 5 , and 1 — XB and 1 — ys are respectively the quantities of bread
and wine obtained by player A. If agreement is reached on x G X at time
£A, then player A's payoff is UA(%A, VA) exp(—rtA), where XA = 1 — %B and
yA = 1 — yB, and player I?'s payoff is UB^E^E) exp(—r£A), where r > 0
denotes the players' common discount rate. Player i's instantaneous utility
function Ui : [0,1] x [0,1] —> 5R is concave, and is strictly increasing in each
of its arguments. Furthermore, Ui(xi,yi) — 0 if either xi — 0 or yi = 0.
Finally, if the players perpetually disagree, then player z's payoff is zero.
The Pareto frontier fte of the set Q of instantaneous utility pairs obtain-
able through agreement may be derived by maximizing one player's utility
subject to the other player receiving a minimum utility level. Define for each
U
UA ^ 0, (/)(UA) = max[/#(£) subject to x G X and £7A(1 — XB-> 1 — VB) > A-
Since Ui is concave, it follows from the 'Theorem of the Maximum under
Convexity restrictions' (cf. Sundaram (1996, Theorem 9.17)) that 0 is a
concave function of UA, and hence, (since Ui is strictly increasing and there
exists x G X such that Ui(x) = 0), Assumption 3.1 is satisfied — with
IA = [0, UA(1,1)], IB = [0, UB(1,1)] and 0(0) = UB(l, 1) > 0. This means
that the analysis and results contained in Section 3.4 may be applied to char-
acterize the SPE of the alternating-offers bargaining game between players
A and B.
I shall characterize the SPE in the limit, as A —» 0. It follows from
Corollary 3.3 that in the limit, as A —•0, a SPE agreement (or, resource
allocation) x* G X is a solution to the following maximization problem

Assuming Ui is differentiate in its arguments, the first-order conditions

characterize the solution. Thus, (x*B,y^) is the unique solution to the fol-
lowing pair of equations

3 U B
TT( \ 9 U A
^ = UB(x B,yB)^
dxB dxA

dyA
3.6 Notes 71

Notice that these equations imply that at (x*B,yg)

dUB/dxB _ dUA/dxA
dUB/dyB dUA/dyA'
That is, in the limiting Rubinsteinian bargaining equilibrium, the players'
marginal rates of substitution between the two goods are equal. This, of
course, is the condition that characterizes a Pareto-efficient resource alloca-
tion.

3.6 Notes
The model studied in this chapter is due to Rubinstein (1982). It has been
assumed that each player discounts future utilities according to a constant
discount rate. For an analysis of the model when the players' time pref-
erences are allowed to take other structures, see Osborne and Rubinstein
(1990, Chapter 3). Despite the many expositions of Rubinstein's model, the
original Rubinstein (1982) paper is still a delight to read. This classic paper
in economic theory contains ideas which have yet to be fully explored.
The relationship with Nash's bargaining solution (discussed in Sections
3.2.3 and 3.4.3) was first discovered in Binmore (1987), which contains a
number of interesting extensions and generalizations of Rubinstein's model.
Some of these extensions are only briefly explored by him and await a fuller
investigation.
The idea of addressing the issue of the uniqueness of the SPE payoffs
by characterizing the suprema and infima of the sets of SPE payoffs to each
of the two players is due to Shaked and Sutton (1984), who adapt and
simplify the original proof contained in Rubinstein (1982). I should empha-
size that this 'method of proof should be applied with care in any exten-
sion/application of Rubinstein's model. Although its use in Rubinstein's
model turns out to be rather straightforward (especially since the method
generates four equations in four unknowns), it should be noted that the
method may only generate upper bounds on the suprema and lower bounds
on the infima — this point is illustrated in several of the later chapters.
 Risk of Breakdown

4.1 Introduction

While bargaining the players may perceive that the negotiations might break
down in a random manner for one reason or another. A potential cause for
such a risk of breakdown is that the players may get fed up as negotiations
become protracted, and thus walk away from the negotiating table. This
type of human behaviour is random, in the sense that the exact time at
which a player walks away for such reasons is random. Another possible
cause for the existence of a risk of breakdown is that 'intervention' by a
third party results in the disappearance of the 'gains from co-operation'
that exists between the two players. For example, while two firms bargain
over how to divide the returns from the exploitation of a new technology, an
outside firm may discover a superior technology that makes their technology
obsolete. Another example is as follows: while a corruptible policeman and
a criminal are bargaining over the bribe, an honest policeman turns up and
reports the criminal to the authorities.
In the next section I analyse a simple modification of the basic alternating-
offers model — studied in Section 3.2 — in which the risk of breakdown is a
main force that influences the bargaining outcome. It will be shown that the
players' payoffs when (and if) negotiations break down and their respective
degrees of risk aversion are crucial determinants of the bargaining outcome.
74 Risk of Breakdown

Section 4.3 contains an application to the problem of tax collection when

the tax inspector is corruptible.
Section 4.4 extends the model studied in Section 4.2 in order to inves-
tigate the relative impacts of discounting and a risk of breakdown on the
bargaining outcome. An application to price determination while searching
is contained in Section 4.5.
The results in Sections 4.2 and 4.4 are derived in the context of a bar-
gaining situation in which the players are bargaining over the partition of
a cake (or 'surplus') of fixed size. In Section 4.6 I study a generalization of
the model analysed in Section 4.4 to a bargaining situation in which the set
X of possible agreements can take any form, subject to satisfying a slightly
modified version of Assumption 3.1.
The issue of whether or not Nash's bargaining solution can be justified
in bargaining situations with a risk of breakdown is addressed in Section
4.2.2, while the same issue in bargaining situations with a risk of breakdown
and discounting is addressed in Section 4.4.1 — see also Section 4.6. Fur-
thermore, the issues of when and how to apply Nash's bargaining solution
are also addressed.

4.2 A Model with a Risk of Breakdown

Two players, A and S , bargain over the partition of a cake of size TT (where
7T > 0) according to the alternating-offers procedure, but with the following
modification: immediately after any player rejects any offer at any time
£A, with probability p (where 0 < p < 1) the negotiations break down in
disagreement, and with probability 1—p the game proceeds to time (t +1) A
— where the player makes her counteroffer.
The payoffs are as follows. If the players reach agreement at time tA
(t = 0 , 1 , 2 , 3 , . . . , and A > 0) on a partition that gives player i a share Xi
(0 < Xi < TT) of the cake, then her payoff is Ui(xi), where Ui : [0, TT] —> 5ft is
her (von Neumann-Morgenstern) utility function. 1 It is assumed that Ui is
1
Unlike in the model studied in Section 3.2, in the above described bargaining procedure
the players have to make decisions in the face of risk. Hence, I interpret each player's utility
function as her von Neumann-Morgenstern utility function. Furthermore, I shall assume
that the players do not discount future utilities, because my objective here is to focus
upon the role of a risk of breakdown on the bargaining outcome. In Section 4.4 I shall
4.2 A Model with a Risk of Breakdown 75

strictly increasing and concave. If negotiations break down in disagreement

at time £A, then player i obtains a payoff of bi, where Ui(0) < bi < C/^TT).
The payoff pair (pA, bs) is called the breakdown point.
Since bi G [£^(0), [^(TT)), there exists a number Z{ G [O,TT) such that
Ui(zi) = 6^; that is, Z{ = U^l(bi), where U~l denotes the inverse of Ui.
Assume that ZA + ZB < TT, which ensures that there exist mutually beneficial
partitions of the cake.
If the players perpetually disagree (i.e., each player always rejects any
offer made to her), then player i's payoff is

t=0
which equals bi. Thus, it follows from Definition 3.1 that the impasse point
(1A,%B) — (bA^bs)- The following lemma is an immediate consequence of
the observation that player i can guarantee a payoff of bi by always asking
for a share Zi and always rejecting all offers.

L e m m a 4 . 1 . In any subgame perfect equilibrium of any subgame of the

model with a risk of breakdown, player i's (i = A, B) payoff is greater than
or equal to bi.

It is assumed that as the time interval A between two consecutive offers

decreases, the probability of breakdown p between two consecutive offers
decreases, and that p —> 0 as A —>> 0.

4.2.1 The Unique SPE when both Players are Risk Neutral
The following proposition characterizes the unique subgame perfect equi-
librium on the assumption that UA(XA) — XA for all XA G [0, TT] and
r
UB{%B) — %B fc> all XB G [0,TT]. It should be noted that this implies
that zi = bi {% = A,B).
Proposition 4.1 (Risk Neutral Players). If for any i = A,B and any
Xi G [0,TT], Ui{xi) = Xi, then the unique subgame perfect equilibrium of the
model with a risk of breakdown is as follows:
• player A always offers x*A and always accepts an offer XB if and only if
xB < x*B,
study the interplay of discounting and a risk of breakdown on the bargaining outcome.
76 Risk of Breakdown

• player B always offers x*B and always accepts an offer xA if and only if
xA < x\, where

xA = b A + (TT -bA-bB)
2 — p\ )
X*B = bB + ( 7T — bA-bB)'
2— p \ /
Proof The proof involves a straightforward adaptation of the arguments in
Sections 3.2.1 and 3.2.2. In particular, in any SPE that satisfies Properties
3.1 and 3.2 player i is indifferent between accepting and rejecting player j ' s
(j^i) equilibrium offer. That is

TV — x*A= pbB + (1 - p)x*B and

7T - XB = pbA + (1 - p)x\.

The unique solution to these two equations is stated in the proposition. This
means that there exists at most a unique SPE that satisfies Properties 3.1
and 3.2, which is described in the proposition. Using an argument similar
to that contained in the proof of Proposition 3.1, it can be verified that the
pair of strategies described in the proposition is a subgame perfect equilib-
rium. Through a slight modification of the arguments presented in Section
3.2.2, it can be shown that there does not exist another subgame perfect
equilibrium.2 •

In the unique SPE, agreement is reached at time 0, and the bargaining

outcome is Pareto efficient: in equilibrium, the negotiations do not break
down in disagreement. Since player A makes the first offer, at time 0, the
unique SPE share to player A is x*A and to player B is n — x*A, where x*A is
defined in Proposition 4.1. Notice that player i's share is strictly increasing
in &i, and strictly decreasing in bj. If bA = bBl then x*A > TT — x*A, which
suggests that player A has a first-mover advantage. However, this first-
mover advantage disappears in the limit as A —» 0: each player obtains
one-half of the cake (since p —> 0 as A —> 0, and bA = bB).
2
A key modification involves replacing the terms 8iMi and dirra respectively with
pbi + (1 — p)Mi and pbi + (1 — p)rrii, which are respectively the maximum and minimum
SPE payoffs that player i obtains if she rejects any offer — since, after rejecting any offer,
with probability p the negotiations break down, and since the players do not discount
future utilities.
4.2 A Model with a Risk of Breakdown 77

As I argued in Section 3.2.4, attention should in general be focused upon

arbitrarily small values of A. I therefore obtain the following corollary to
Proposition 4.1.

Corollary 4.1 (Split-The-Difference Rule). In the limit, as A —> 0,

the unique SPE shares of the cake to players A and B respectively converge
to

bA + -(TT ~bA-bB) and bB + - (TT - bA - bB).

Zi \ / Zi \ /

The friction in the bargaining process underlying the above described

game arises from the risk that negotiations break down between two con-
secutive offers, which is captured by the probability p. As the absolute
magnitude of this friction becomes arbitrarily small — and thus the com-
mon cost of haggling to the players becomes arbitrarily small — the limiting
equilibrium partition of the cake, which is independent of who makes the
first offer, may be interpreted as follows. The players agree first of all to
give player i (i — A, B) a share b{ of the cake (which gives her a payoff equal
to the payoff she obtains when, and if, negotiations break down), and then
they split equally the remaining cake IT — bA — bB.

4.2.2 The Unique SPE with Risk Averse Players

I now study the model on the assumption that Ui is a strictly increasing and
concave function. In any SPE that satisfies Properties 3.1 and 3.2 player i
is indifferent between accepting and rejecting player j ' s (j ^ i) equilibrium
offer. That is

UB(K - XA) = pbB + (1 -p)UB(x*B) and (4.1)

UA(K ~ xB) = pbA + (l -P)UA(X*A). (4.2)

ires that

x\ > zA (4.3)
x*B > zB (4.4)
TT — X*A > ZB (4.5)
K - XB > ZA- (4.6)
78 Risk of Breakdown

I now show that there exists a unique pair (x^, x*B) that satisfies (4.1)-(4.6),
and that it has the property that

ZA < x*A < TV — ZB and ZB < x*B < TT — zA. (4-7)

Using (4.1) and (4.2), define for each i

fi(xi) = UJ(TT - Xi) - pbj -

(1 - P)U3 (IT - Ur1 (pbx + (1 - p)Ui{xi)) j , (U

where j ^ i. It is easy to verify that Ti(zi) > 0, F^TT) < 0 and F^ is strictly
decreasing and continuous in X{ on the open interval (^,TT). Hence, since
(4.1)-(4.4) imply (4.5)-(4.6), the desired conclusions follow immediately.
The following proposition characterizes the unique SPE.

Proposition 4.2. Let {x*A,x*B) be the unique pair that satisfies (4.1)-(4-6)-
The unique subgame perfect equilibrium of the model with a risk of breakdown
is as follows:
• player A always offers x*A and always accepts an offer XB if and only if
XB < x*B.
• player B always offers x*B and always accepts an offer xA if and only if
xA < x*A.

Proof. Since there exists a unique pair that satisfies (4.1)-(4.6), there exists
at most a unique SPE that satisfies Properties 3.1 and 3.2, which is described
in the proposition. It is easy to verify — using an argument similar to
that contained in the proof of Proposition 3.1, and (4.7) — that the pair
of strategies described in the proposition is a subgame perfect equilibrium.
Through a slight modification of the arguments presented in Section 3.4.2, it
can be shown that there does not exist another subgame perfect equilibrium.
•
In the unique SPE, agreement is reached at time 0, and the bargaining
outcome is Pareto efficient: in equilibrium, the negotiations do not break
down in disagreement. The unique SPE share to player A is x\ and to
player B is n — x*A, where x\ is defined in Proposition 4.2. Since, for each
x
A-> ^A(XA) — a s defined in (4.8) — is strictly increasing in 6^ and strictly
4.2 A Model with a Risk of Breakdown 79

decreasing in bs, it follows that player i's SPE share of the cake is strictly
increasing in 6^, and strictly decreasing in bj.
If UA = UB and bA = bsj then, since F ^ = Fj3, #^ = x*B. This implies,
using (4.1) and (4.7), that x\ > ix — x^ 5 which suggests that player A has
a first-mover advantage. However, this first-mover advantage disappears in
the limit as A —> 0: each player obtains one-half of the cake.
The following corollary to Proposition 4.2 characterizes the SPE in the
limit as A -> 0. 3

Corollary 4.2. In the limit, as A —> 0, the payoff pairs (UA{X*A)1UB('7T —

x*^)) and (UA(TT — x*B)^ UB(%%)) converge to the unique solution of the fol-
lowing maximization problem

max (uA - bA)(uB -

where Q — {(UAJUB) '- UA{0) < UA < UA(TT),UB = 9(UA)^A ^ ^A and

, and where for each UA £ [UA(0), UA(K)\, g{uA) = UB{K — U^1{UA

Proof. The proof is similar to that of Corollary 3.3. For notational conve-
nience, define for each UA > bA and UB > bs

N(UA, uB) = (uA - bA){uB - 6B).

Using (4.1) and (4.2), it can be shown that N(UA(X\),UB(TT - x*A)) =

N(UA{TT — X*B),UB(X*B))- Furthermore, from (4.2) and (4.7) it follows that
A
UA(%*A) > UA(K — %%)- AS is illustrated in Figure 4.1, it follows that the
two distinct utility pairs (UA{X*A), UB(K - x\)) and (UA(K - x*B), UB(X*B))
lie on the graph of the function N(UA, UB) = -/V([/A(£^), UB(TT — x\)), and
on the graph of the function g defined in the corollary — where the set O
of possible utility pairs obtainable through agreement is the graph of the
function g (cf. Section 2.2, where it is shown — in Lemma 2.1 — that g
is strictly decreasing and concave). From (4.1) and (4.2), it follows that
as A -> 0, UA(x*A) - UA(TT - x*B) -» 0 and UB(x%) - UB(n - x*A) -> 0.
3
As is shown in Corollary 4.2, in the limit, as A —> 0, a player's SPE payoff does not
depend on which player makes the first offer at time 0; in this limit a first-mover advantage
does not exist.
4
Suppose, to the contrary, that UA{%*A) < UA(^ — x%). Using (4.2), it follows that
x\ < ZA^ which contradicts (4.7).
80 Risk of Breakdown

Hence, both payoff pairs (C/^(x^), UB{ix — x*A)) and (UA(TT — £#), UB(x*B))
converge to the unique solution of the maximization problem defined in the
corollary. •

uB

N(uA, uB) = N(UA(x*A), UB(TT - x*A))

(UA(ir-x*B),UB(x*B))

(UA(x*A),UB(n-x*A))

bB

uA

F i g u r e 4.1: An illustration of the proof of Corollary 4.2.

It follows immediately from Corollary 4.2 and the definition of the Nash
bargaining solution (as stated in Section 2.2) that in the limit, as A —>
0 (and, hence, as p —> 0), the unique SPE payoff pair converges to the
Nash bargaining solution of the bargaining problem (fi, d), where the set of
possible utility pairs ft = {(uA, uB) • UA(0) <UA < UA{K) and uB = g(uA)}
and the disagreement point d — (bA,b B). Thus, the disagreement point in
Nash's set-up should be identified with the impasse point of the model with a
risk of breakdown. Furthermore, the Nash bargaining solution is applicable
when the friction in the bargaining process (namely, the risk of breakdown)
is arbitrarily small.
4.3 An Application to Corruption in Tax Collection 81

4.3 An Application to Corruption in Tax Collec-

tion
A corruptible tax inspector / and a citizen C bargain over the income to
be reported and the bribe, given that the citizen's true income is y G [0,1].
While bargaining there is a possibility — considered a random event — that
an honest superior of the tax inspector turns up and asks her to immediately
report the citizen's income.
An agreement between the tax inspector and the citizen is a pair (ra, 6),
with the interpretation that / certifies to the government that the true in-
come of C is ra, and C pays / a bribe of b. Immediately after the inspector
submits an income report ra, / and C face two possible outcomes. With
probability 1 — 7 the report is not audited, and / and C respectively ob-
tain uo + X(m)T(m) + b and y — T(ra) — b amounts of money, where uo is
the inspector's wage and A(ra)T(ra) is her commission on the reported tax
liability T(ra). On the other hand, with probability 7 the report is audited
(which reveals to the government the citizen's true income), and / and C
respectively obtain uo + X(y)T(y) + b — //(m, y) and y - T(y) — b — fc(m, y)
amounts of money, where fi(m,y) is the monetary fine paid by player i
(i = C, /) if the true income is y and the reported income is ra.
Assume that 0 < 7 < 1, the wage uo > 0, the tax liability T(x) G [0,x]
and the commission rate X(x) G [0,1] (for any income x), the penalties are
monotonic, namely, for any y G [0,1], fi(mi,y) > fi{rri21y) for all mi <
m2 < y and for all mi > m2 > y, with fi(y,y) = fc(y,y) = 0, and
for any y and any ra, fi(m,y) < uo + X(y)T(y) and fc{rn,y) < y - T(y).
Furthermore, assume that the players are risk neutral: the utility to each
player from receiving x amount of money is x.
Define for each true income y G [0,1] and any reported income x G [0,1]

ai{x) =uo + jX(y)T(y) + (1 - 7)A(x)T(x) - 7 //(x, y)

ac{x) = y- jT(y) - (1 - -f)T(x) - jfc(x, y).
Thus, the expected amounts of money obtained by the two players from
agreement on a pair (ra, b) are xj = aj(m) + b and XQ = ac(rn) — b.
If negotiations break down, which occurs with probability p after any
offer is rejected — since with this probability an honest superior of the
inspector turns up and asks her to immediately submit an income report
82 Risk of Breakdown

— then the inspector (unilaterally) chooses an income report n G [0,1] that

maximizes her expected amount of money aj(n) subject to ac(ri) > 0.
Assume that this maximization problem has a solution, which is denoted
by n*. Thus, the expected amounts of money obtained by the two players
when (and if) negotiations break down are 6/ = aj(n*) and be = ac{n").

The Bargaining Outcome

For each true income y, define 5(ra, y) = a/(ra) + ac(m). Assume that
there exists an income report which maximizes S(m,y), and denote it by
ra*. It is easy to verify that an agreement (ra, b) is Pareto efficient only
if 7Ti = 77i*. To simplify the argument, assume that at the beginning of
the bargaining process the players agree that the inspector reports ra* if
they reach agreement on the bribe. 5 To ensure that the expected money
obtained by each player is non-negative, it is assumed that the bribe b <G
[—a/(ra*), ac(ra*)]. One may now model the negotiations using the model
with a risk of breakdown, where the risk neutral players bargain over the
partition of S{m*,y) amount of money with be and bj defined above.6 If
the inspector obtains a share xj of this pot of money, then this means that
the bribe b = xj — aj{m*).
I now characterize the unique SPE bribe 6* in limit, as A —» 0 (i.e., when
the risk of an honest superior turning up is arbitrarily small). Applying
Corollary 4.1, it follows that

- 7 ) [[1 + A(n*)]T(n*) - [1 + A(m*)]T(m*)] +

(m*)-/c(m*) + /c(n*)-/ 7 (n*)]

5
This assumption is without loss of generality, because it can be shown by extending
the arguments presented in Section 4.2.1 — as is done in Section 4.6 — that in the unique
SPE of a general version of the model with a risk of breakdown, the players agree on a
Pareto-efficient agreement. It may be noted that this result follows essentially from the
result that the unique SPE in Rubinstein's model is Pareto efficient.
6
It follows, by definition, that S(m*,y) > S(n*,y), which implies that S(m*,y) >
be -\-bi. Furthermore (by assumption) be > 0 and (since the inspector has the option to
report the truth) 6/ > 0.
4.3 An Application to Corruption in Tax Collection 83

In equilibrium, agreement is reached at time 0. The equilibrium income re-

ported by the inspector is m* and the equilibrium bribe paid by the citizen
to the inspector is 6*. The income report n* (which is submitted if nego-
tiations break down) critically influences the level of the equilibrium bribe.
It does not, however, influence the income reported in equilibrium, because
that is chosen to maximize the gains from collusion. Since n* influences the
breakdown point, it influences the equilibrium partition of the gains from
collusion through its influence on the equilibrium bribe.
If, for example, the tax schedule T is strictly increasing, then m* < y
means that taxes are evaded. Corruption, on the other hand, takes place if
some bribe is paid. From the expression above for the equilibrium bribe, it
is clear that penalties may influence the level of corruption. However, notice
that if 77i* < y < n*, then an increase in the penalty on / for under-reporting
increases the bribe, while an increase in the penalty on / for over-reporting
decreases the bribe. This suggests that it is over-reporting (or extortion)
that needs to be penalized heavily to discourage corruption.

Optimal Policy
I now characterize the optimal government policy — namely, the tax sched-
ule T, reward system uo and A, and penalty functions fc and / / that maxi-
mize the government's expected revenue. Attention is restricted to policies
that induce truthful reporting: for any true income y G [0,1], the income
reported in the unique SPE m* — y. Any such evasion-proof policy satisfies
the following (evasion-proofness) condition7

For all y, S(y, y) > S(m, y) for all m.

For any evasion-proof policy, the government's expected revenue is the ex-
pected value of [1 - X(y)]T(y) - CJ.8
I now maximize the government's expected revenue subject to the evasion-
proofness condition. It is optimal to set the penalities, for any m ^ y,
to their maximal admissible levels. Thus, for any y and m such that
7
This restriction is without loss of generality, because Hindriks, Keen and Muthoo
(1998), who study this model, show that for any policy such that for some y, m* / y,
there exists another policy which is evasion-proof and revenue-equivalent.
8
Assume that the distribution of income in society is such that the expectation of
[1 — X(y)]T(y) — uo w.r.t y is well defined.
84 Risk of Breakdown

m ^ y, fi(m,y) = u + X(y)T(y) and fc{m,y) = y - T(y). This im-

plies that the evasion-proofness condition becomes the requirement that
for all y, (1 - 7)[1 - A(ra)]T(ra) > [1 - A(y)]T(y) - 7(0; + y) for all m.
Since T(m)[l - A(m)] > 0 (for all m) and T(0)[l - A(0)] = 0, this implies
7(0; + y) > [1 — A(y)]T(y). Maximizing the government's expected revenue
then requires setting the RHS of this latter inequality to its maximum level
— that is, 7(0; + y) = [1 — A(y)]T(y). The government's expected revenue
then becomes the expected value of "yy — (1 — 7)0;, which, since u > 0, is
maximized by setting uo = 0.
It has thus been established that any policy which has the following
features is evasion-proof and maximizes the government's expected revenue:
for all y, 72/ = [1 — A(y)]T(y), u — 0, and for all y and ra such that m ^ y,
//(ra, 2/) = \{y)T{y) and /c(ra, y) — y — T(y). The maximized expected
revenue to the government is 7ye, where ye denotes the expectation of y.
The most striking aspect of the optimum is that revenue-maximization
places very few restrictions on either the tax schedule or the commission
structure. Nevertheless, the result implies that the average tax rate T(y)/y
be no less than the probability of audit 7, and the commission rate no
greater than 1 — 7. What is very substantially restricted, however, is the
relationship between T and A. It is perfectly possible, for example, for the
government to raise the maximum revenue by implementing a tax schedule
that is progressive in the sense that the average tax rate rises with income;
but in order to do so it must also implement a reward structure such that
the commission rate A also increases with the income reported. By the
same token, a constant commission rate is optimal only if tax liabilities are
proportional to income.
An optimal policy — with the above stated features — is corruption-
proof if and only if for all y, the equilibrium bribe 6* = 0. From the ex-
pression above for the equilibrium bribe 6*, it thus follows that an optimal
policy is corruption-proof if and only if for all y, n* = y.
I now show that there exists an optimal policy which is not corruption-
proof. Assume that the probability of audit 7 < 1/2, and consider the
optimal policy in which T(y) = ty, where t G (7,1 — 7), and A(y) = 1 — 7/t.
For any true income y > 0, aj(n = y/t) > aj(n < y) and ac(n — y/t) = 0.
Hence, for any y > 0, n* > y. The impact of the possibility of extortion is
now self-evident: if there is some possibility that the citizen is subjected to
4.4 The Effect of Discounting 85

extortion, then an optimal policy need not eliminate corruption.

It is intuitive that the inspector's incentive to over-report comes from
the commission aspect of the policy. Suppose, therefore, that incentive is
eliminated by choosing the optimal policy in which for all y, A(y) = 0. This,
however, implies that T(y) = jy. With this optimal policy, n* = y (for
all y). This conclusion is striking: a government that seeks to maximize
expected revenue but is averse to evasion and corruption can do no better
than use one of the simplest of all policies, involving a proportional tax
(at a rate equal to the probability of audit) with a fixed wage contract for
inspectors.

4.4 The Effect of Discounting

I now extend the model studied in Section 4.2 by assuming that the players
discount future utilities. As in Section 3.2, for notational convenience, define
6i = exp(—r^A), where Vi > 0 is player i's discount rate. To simplify the
notation, assume that Ui(0) = 0.9
If the players perpetually disagree (i.e., each player always rejects any
offer made to her), then player i's payoff is
oo ,
- pfdj, which is equal to fa = —

Since fa G [0,^), there exists a number x\ G [0, &i) such that Ui{x\) — fa\
that is, x\ = U^l{fa). Notice that, since ZA + zB < TT, x\ + xbB < TT. The
following lemma is an immediate consequence of the observation that player
i can guarantee a payoff of fa by always asking for a share x\ and always
rejecting all offers.

Lemma 4.2. In any subgame perfect equilibrium of any subgame of the

model with a risk of breakdown and discounting, player i's (i = A, B) payoff
is greater than or equal to fa.
9
This assumption is without loss of generality, because U% is a von Neumann-
Morgenstern utility function. Thus, if Ui(0) ^ 0, then one may rescale utilities, and define
another von Neumann-Morgenstern utility function Ui such that for each xi G [0, TT],
Ui(xi) = Ui(xi) — Ui(0), which also represents player i's preferences.
86 Risk of Breakdown

In any SPE that satisfies Properties 3.1 and 3.2 player i is indifferent
between accepting and rejecting player j ' s equilibrium offer. That is

UB(7r-x*A)=pbB + {l-p)6BUB(x*B) and (4.9)

17A(TT - x*B) = pbA + (1 - p)6AUA(x*A). (4.10)

In addition, Lemma 4.2 requires that

xA > x\ (4.11)
xB > xbB (4.12)
7T — x* A > x b B (4-13)
b
TV-X*B >x A. (4.14)

I now show that there exists a unique pair (x^, x*B) that satisfies (4.9)-(4.14),
and that it has the property that

xbA < x\ < TV — xbB a n d xbB < x*B < TV — x bA. (4-15)

Using (4.9) and (4.10), define for each i

\{i) j { i ) p j

(1 - p)SjUj (ir - Ur1 [pb% + (1 - p)SiUi(xi)>j J, (4.16)

where j ^ i. It is easy to verify that Y\{x\) > 0, T*(7r) < 0 and T* is strictly
decreasing and continuous in Xi on the open interval (xb,7v). Hence, since
(4.9)-(4.12) imply (4.13)-(4.14), the desired conclusions follow immediately.
The following proposition characterizes the unique SPE.

Proposition 4.3. Let(x*Aix*B) be the unique pair that satisfies (4-9)-(4-14)-

The unique subgame perfect equilibrium of the model with a risk of breakdown
and discounting is as follows:
• player A always offers x\ and always accepts an offer xB if and only if
%B < X*B-
• player B always offers x*B and always accepts an offer xA if and only if
xA < x*A.

Proof. Similar to the proof of Proposition 4.2. •

4.4 The Effect of Discounting 87

In the unique SPE, agreement is reached at time 0, and the bargaining

outcome is Pareto efficient: in equilibrium, the negotiations do not break
down in disagreement. The unique SPE share to player A is x\ and to
player B is n — x*Al where x\ is defined in Proposition 4.3. Since, for each
XA, ^\{XA) — as defined in (4.16) — is strictly increasing in 6^ and r#,
and strictly decreasing in bs and r^, it follows that player i's SPE share of
the cake is strictly increasing in bi and r^ and strictly decreasing in bj and

If UA = UB, t>A = bs and TA = TB, then, since F*A = P g , x*A = x*B.

This implies, using (4.9) and (4.15), that x*A > TT — x*A, which suggests that
player A has a first-mover advantage. However, this first-mover advantage
disappears as A —> 0: each player obtains one-half of the cake.

4.4.1 Small Time Intervals

I now characterize the SPE in the limit as A —> 0. Unlike in the model
without discounting, in order for this limit to exist it needs to be assumed
that the limit of the ratio p/A as A —> 0 exists, in addition to the assumption
that p —> 0 as A —> 0. Assume that when A is small (infinitesimal), pjA = A,
where A > 0.10 Thus, it follows from the expression above for (5i that when
A > 0 but small the payoff to player i from perpetual diagreement is

n +A-
since, when A > 0 but small, the discount factor 8i = exp(—r^A) = 1 - r ^ A .
Hence, in the limit, as A —> 0, $ converges to Xbi/(ri + A). Thus, the
(limiting, as A —>• 0) impasse point

Corollary 4.3. In the limit, as A —> 0, the SPE payoff pairs (UA(X*A), UB(TT-
x*A)) and (UA(TT — x^), UB(%*B)) converge to the unique solution of the fol-
lowing maximization problem

max (UA-1AYA{UB-1BYB,

10
Notice that this assumption is implied if, for example, negotiations break down ac-
cording to a Poisson process at a rate A > 0.
88 Risk of Breakdown

where © = {(UA,V>B) ' 0 < UA < UA(K),UB = g{uA),UA > XA and UB >
TB\, with g defined in Corollary j±.2, the (limiting) impasse point (IA,XB)
is defined above,

TB + A TA + X
a A = TTTT T T a n anaa <JB =
<JB : •
2A + r + r
2A + rA + rB 2\ + rA + rB
Proof. The proof is similar to the proofs of Corollaries 3.3 and 4.2. One
defines a function N such that for each UA > PA and UB > PB-> N(UA, UB) —
(uA-PAY*A{UB ~PBY*, where a\ = [ln(l-p)+ln« s ]/[21n(l-p)+ln5A +
In 6B] and a^ = 1—a\. The subsequent argument is similar to the arguments
in the proofs of Corollaries 3.3 and 4.2. n •

The following corollary is an immediate consequence of Corollary 4.3 and

Definition 2.3.

Corollary 4.4 (Relationship with Nash's Bargaining Solution). In

the limit, as A —> 0, the unique SPE payoff pair of the model with a risk
of breakdown and discounting converges to the asymmetric Nash bargaining
solution of the bargaining problem (£2, d) with r = [TB + A)/(2A + r^ + rs),
where Q = {(UA,V>B) '• 0 < UA < UA{K) and UB = UB(TT — U^1{UA))} and
d=(IA,lB), withIi = \bi/{n + \) (i = A,B).

When A is arbitrarily small, both the probability that negotiations break

down in any finite time and the cost of waiting to make a counteroffer are
arbitrarily small, and, thus, the friction in the bargaining process is arbi-
trarily small. It has thus been established that the model with a risk of
breakdown and discounting provides a strategic justification for the asym-
metric Nash bargaining solution when the friction in the bargaining process
is small. The Nash bargaining solution is justified if, in addition, the players'
discount rates are identical.
I now emphasize the manner in which Nash's bargaining solution should
be applied in bargaining situations when there is a risk of breakdown and
the players discount future utilities. With discounting, the breakdown point
{pA-, bs) — which is the payoff pair obtainable if negotiations break down at
any point in time — is different from the (limiting) impasse point (XA->XB) =
+ A), Xhs/i^B + A)) — which is the payoff pair obtainable (in the
1
Note that, as A —> 0, a*A —>• a A and, as shown above, /3i
4.4 The Effect of Discounting 89

limit, as A —» 0) if each player always rejects any offer made to her. I

have established in Corollary 4.4 that the disagreement point in Nash's set-
up should be identified with the (limiting, as A —> 0) impasse point of the
model with a risk of breakdown and discounting, and not with the breakdown
point. This is rather intuitive, because in any finite time the probability of
breakdown is arbitrarily small (in the limit as A —> 0).
The impasse point depends on the breakdown point and on the players'
expected discount rates VA/X and rs/X. Only if these expected discount
rates are arbitrarily small should the breakdown point — which, in this
case, is identical to the impasse point — be identified with the disagreement
point. For example, if discounting is considered an insignificant force relative
to the force of a risk of breakdown, then it is appropriate to identify the
breakdown point with the disagreement point. On the other hand, if the
risk of breakdown is considered insignificant relative to discounting — and
thus the model approximates the basic alternating-offers model — then the
impasse point (which converges to (0, 0) as X/ri —> 0) should be identified
with the disagreement point.
Example 4.1 (Differentiable Utility Functions). Suppose that the util-
ity functions, UA and £/#, are differentiable. Applying Proposition 2.5 to
the result stated in Corollary 4.4, it follows that the unique (limiting, as
A —> 0) SPE payoff pair is the unique solution to the following first-order
conditions

and uB=g(uA),
uA --LA
where ( Z A , 2 # ) is defined in Corollary 4.4.

4.4.2 Risk Neutral Players: Split-The-Difference Rule

Suppose that UA{XA) — XA for all XA £ [0, TT] and UB{%B) — %Bforall
XB £ [0,TT]. Applying Example 4.1, and noting that (with linear utilities)
9(UA) — K — UA, it follows that the unique limiting SPE payoffs (and shares)
to players A and B respectively are

and (4.17)
2A + TA + rB
( ) (4.18)
90 Risk of Breakdown

where (XA^B) is defined in Corollary 4.4. This limiting SPE partition may
be interpreted as follows. The players agree first of all to give player i
(i = A, B) a share Xi of the cake (which gives her a payoff equal to her
payoff from perpetual disagreement), and then player i (i = A, B) obtains
a fraction (rj + A)/(2 A + VA + re) of the remaining cake TT — I A — ^B- If
^A — ^Bi then they split equally the remaining cake; otherwise the relatively
more patient player obtains a relatively bigger share of the remaining cake.
It is easy to verify that player i's limiting SPE share is strictly increasing
in rj and strictly decreasing in r^. The intuition for these results is as follows.
An increase in rj decreases Xj, has no effect on Xi and increases the fraction
(rj + A)/(2A + VA + TB) of any remaining cake obtained by player i. Hence,
player i's limiting SPE share increases, because she obtains a bigger slice
of a bigger remaining cake. On the other hand, an increase in V{ decreases
Zi, has no effect on Xj and decreases the fraction (rj + A)/(2A + TA + re)
of any remaining cake obtained by player i. Hence, player i's limiting SPE
share decreases, because her limiting payoff from perpetual disagreement
decreases and she obtains a smaller slice of a smaller remaining cake.
The effect on player z's limiting SPE share as the rate A at which ne-
gotiations break down increases is relatively more complex. In particular,
as I show below, it depends on the difference b{ — bj. First notice that as
A increases, both players' limiting payoffs from perpetual disagreement in-
creases. The intuition for this is as follows: an increase in A decreases the
expected time to breakdown, and, hence, the expected discounted value of
bi increases. To simplify the discussion below, set rA — ^B — r-
If bA = bs, then player i's limiting SPE share is TT/2. Thus, r and A have
no effect on the limiting SPE partition, which makes sense because under
these conditions the players have equal bargaining power.
Now suppose that bA ^ &#• In this case player z's limiting SPE share is

r/X) + 1

which depends upon the difference bi — bj and the expected discount rate
r/X. It follows that player z's limiting SPE share is strictly decreasing in
r/X if bi > bj, and strictly increasing in r/X if bi < bj. This means that
the players have conflicting interests over the value of r/X. If bi > bj then
player i prefers r small and/or A large, while player j prefers the opposite.
4.5 An Application to Price Determination 91

The intuition is as follows. If b{ > bj, then break down of negotiations is a

relatively more desirable event for player i. Thus, as its impact is increased
(by decreasing the ratio r/A), player i's share of the cake increases (and
player j ' s decreases).

4.5 An Application to Price Determination

A buyer B and a seller S are bargaining over the price p at which to trade a
unit of some commodity. The von Neumann-Morgenstern utility functions
to the players are Us(p) = p — c and UB(P) = v — p, where v > c > 0. The
seller S has placed an advert in the local newspaper asking a non-negotiable
price qs for the unit of the commodity that she owns. Thus, while bargaining
with I?, different buyers telephone S according to a Poisson process at a rate
As, where As > 0. Similarly, the buyer B has placed an advert in the local
newspaper asking to buy a unit of this commodity at a non-negotiable price
qs- Thus, while bargaining with 5, different sellers telephone B according
to an independent Poisson process at a rate A#, where A# > 0. When
either a different seller telephones B o r a different buyer telephones 5, the
negotiations between B and S break down in disagreement.
In order to study this bargaining situation, I extend (in a simple man-
ner) the model with a risk of breakdown and discounting. Before I describe
that extension, two simplifying assumptions are made: the players discount
future utilities at a common rate r > 0, and A is strictly positive but suf-
ficiently small so that A#A (resp., As A) is (approximately) the probability
that a different seller (resp., buyer) telephones B (resp., S) in the small time
interval A.
The extension concerns the manner in which negotiations break down.
Immediately after any player rejects any offer at any time tA, there are four
possible events: (i) with probability A^AsA2 a different seller telephones B
and a different buyer telephones S, (ii) with probability A#A(1 — AsA) a
different seller telephones £?, (iii) with probability AsA(l — A#A) a different
buyer telephones £, and (iv) with probability (1 — AsA)(l — A#A) the game
proceeds to time (t + 1)A, when the player makes her counteroffer.
If either of the first three events, (i)-(iii), occurs, then the negotiations
between B and S break down in disagreement. If both players receive tele-
phone calls (event (i)), then they trade at the prices that they respectively
92 Risk of Breakdown

advertised. Their payoffs are bs = v — qs and bs = qs — c. If, on the other

hand, negotiations break down because exactly one player receives a phone
call (events (ii) and (in)), then the other player waits to receive a phone call.
If player i receives a phone call then her payoff is 6^, and player j ' s payoff
— who is left waiting for a phone call — is Xjbj/(r + Xj).
Indeed, as A —> 0, the payoff to player i from perpetual disgreement
converges to \ib{/(r + A^). Hence, the (limiting, as A —> 0) impasse point

Assume that IA+IB < V — C, for otherwise, gains from trade do not exist
between B and S. Furthermore, without loss of generality, assume that the
advertised prices qs and qs are greater than or equal to c and less than or
equal to v. It should be noted that in the limit, as A —> 0, the payoff to
player i in any SPE is greater than or equal to I{.

The Bargaining Outcome

Using arguments similar to those presented in Section 4.4, it can be estab-
lished that the unique SPE price is p*s if S makes the first offer and p*B if B
makes the first offer, where the pair (P*S,P*B) is the unique solution to the
following pair of equations (which are similar to equations 4.9 and 4.10)

v - p*s = \BAbB + A 5 A(1 - XBA)XB + (1 - A 5 A)(1 - A B A)(1 - rA)(v - p*B)

P%-c = XsAbs + A B A(1 - XSA)XS + (1 - A B A)(1 - A 5 A)(1 - rA)(p£ - c),

where (XA,XB) is defined in (4.19). It is easy to show that, as A —> 0, the

unique solution to these equations p*s and p*B converges to p*, where

v -p* =1B + -(v -c-XB -Is)-

This means that

p* - c = Xs + - (v - c - IB - Is) •

Hence, in the unique limiting SPE (as A —> 0) agreement is reached at

time 0 on the price p*. Notice that the limiting SPE outcome reflects the
4.5 An Application to Price Determination 93

Split-The-Difference Rule, as illustrated in Section 4.4.2. The players are,

in effect, bargaining over the partition of a cake of size v — c: they agree first
of all to give each other the limiting expected payoffs (1A and 1B) that they
would, respectively, obtain from perpetual disagreement, and then they split
equally the remaining cake.

Equilibrium Advertised Prices

I now determine whether or not the players place such adverts in the local
newspaper, and if player i chooses to do so, her advertised price q{. The
cost of placing such an advert is e > 0. The analysis focuses on arbitrarily
small values of e. Assume, as seems reasonable, that the rate A# at which
different sellers call B is strictly increasing in qs, and the rate \$ at which
different buyers call S is strictly decreasing in q$. Furthermore, assume that
Xi (i = B, S) is differentiate and concave, XB(QB) — 0 for any qg < c, and
^s(Qs) = 0 for any qs > v.
Before B and S bargain over the price at which to trade — which they
do if and only if gains to trade exist — they simultaneously choose whether
or not to place an advert in the local newspaper, and, if player i chooses to
do so, the price qi G [c, v]. If both players do not advertise then each player's
payoff is (v — c)/2 — there is then no risk of breakdown, and the bargaining
game boils down to the basic alternating-offers game (studied in Section
3.2). If player i does not advertise, while player j (j ^ i) does advertise and
sets a price q^ then (since qj G [c, v]) 1 A + 1 B < V — C (where, of course,
li — 0). This means that gains to trade exist between B and S, and hence,
from the bargaining outcome derived above, it follows that players i's payoff
is (v — c — lj)/2 and player j ' s payoff is (v — c + X/)/2 — e. Finally, if both
players advertise then player i's payoff is 1{ + max{(i> — C — 1B —TS)/2, 0}. 1 2
Let g* denote the unique value of qi that maximizes li — which is char-
acterized below. It is easy to verify that provided the cost e of placing
an advert in the local newspaper is sufficiently small, the above described
simultaneous-move game has a unique Nash equilibrium in which player %
places an advert in the local newspaper with price qi = q*. Since e is suf-
12
For some pairs (qs, qs) of advertised prices, XA + ZB < v — c, while, for other pairs,
XA + XB > V — C. In the former case, the payoffs follow from the bargaining outcome
derived above, while for the latter case there do not exist any gains to trade and hence
player i's payoff is Xi.
94 Risk of Breakdown

ficiently small, it is perhaps not surprising that in equilibrium both players

place adverts. I now characterize this unique pair of advertised prices q*B
and q*s in order to explore their properties. For example, it is interesting
to investigate whether or not the equilibrium prices are such that gains to
trade do exist between B and S.
Differentiating Xi w.r.t. qi1 it follows that

Since Xi is concave in q^ and since

dXAc) 1 dXA
^ ^ > 0 and -

the price q* set by player i is the unique solution to the first-order condition

Hence, the equilibrium advertised prices q*B and q*s respectively are the
unique solutions to the the following equations

= (v - qB)rXfB(qB)
= ~(qs ~ c)r\'s(qs).

These prices have the property that c < (fB < v and c < q*s < v. The
unique solution (q%,qg) to these equations may, in general, be such that
either XA+XB < V — C OY IA + XB > V — C. This means that, in general,
trade may or may not take place between B and S. I now show that in the
limit, as r —> 0, trade does not take place between B and S.
It is easy to verify that qB is strictly increasing in r, and q$ is strictly
decreasing in r. In the limit, as r —> 0, q B —» c, q* s —» v and rj\ —> 0.
Hence, as r —» 0, Xi -^ b{. These results are rather intuitive, because, as
r —> 0, each player is almost indifferent to the timing at which she receives a
phone call from a different trader, and, hence, she sets her most favourable
price (qB —> c and q^ —> v). Although this means that the arrival rate is
close to zero, this price maximizes her utility since r is infinitesimal relative
to Xi (and hence her expected discount rate is zero) and since in infinite time
4.6 A Generalization 95

she receives a phone call for sure. Therefore, since, in the limit as r —> 0,
XB +IS > v ~ ci there do not exist gains to trade between B and S. Hence,
each player i waits for a phone call from a different trader who is willing to
trade at the advertised (or, posted) price qi.

4.6 A Generalization
I now extend the model studied in Section 4.4 by assuming — as in Section
3.4 — that the players are bargaining over a set X of possible agreements,
where Ui : X —• 3ft is player i's (instantaneous) von Neumann-Morgenstern
utility function. As in Section 4.4, immediately after any player rejects any
offer at any time tA, with some probability negotiations break down: assume
that player i believes this probability to be pi, where pi = 1 — exp(—A^A)
with A^ > 0. Thus, I allow for the possibility that the players have different
beliefs about the rate at which negotiations break down in disagreement.
The payoff to player i from perpetual disagreement is
a* = PA
P
* l-(l-Pi)V
Similar to Assumption 3.1, assume that the Pareto frontier fie of the set
Q of instantaneous utility pairs obtainable through agreement is the graph
of a concave function, denoted by (/>, whose domain is an interval I A Q 5ft
and range an interval / # C 5ft, with j3\ G IA, PB e ^B and (f)((3\) > fi*B'
A minor adaptation of the analysis of Sections 3.4.1 and 3.4.4 charac-
terizes the unique SPE of this model. In particular, it follows that MA —
mA = VX and MB = TUB = Vg, where (V^,Vg) is the unique pair that
satisfies (4.20)-(4.25), stated below.13

+ (1-PB)SBV£) (4.20)
V£ = <t>{pAbA + (1 - PA)6AVX) (4.21)
VI > P*A (4.22)
V g > I3*B (4.23)
> PB (4-24)
> PA- (4-25)
13
Recall that Mi and rrii are respectively the supremum and infimum of the set of SPE
payoffs to player i in any subgame beginning with player z's offer.
96 Risk of Breakdown

In the following corollary — which can be proven along the lines of the
proofs of Corollaries 3.3 and 4.3 — I characterize the limit of the unique
SPE payoff pair, as A —> 0.
Corollary 4.5. In the limit, as A —» 0 ; the unique SPE payoff pair of the
general model with a risk of breakdown and discounting converges to the
unique solution of the following maximization problem

max (uA ~

where @ = {(UA,UB) : UA G IA^B — 4>{UA)^A > %A and UB > XB}, and

for each i = A,B

J J
and o{ = - — (j ^ i).
\A + ^B + rA + rB
It follows immediately from Corollary 4.5 and Definition 2.3 that the
unique SPE payoff pair converges, as A —» 0, to the asymmetric Nash bar-
gaining solution of the bargaining problem (£"2, d) with r = a A , where ft
is the set of instantaneous utility pairs obtainable through agreement and
d — (XA,XB) with Xi (i = A,B) and a A defined in Corollary 4.5. Thus, as
has been shown before (in Sections 3.2.3 and 3.4.3), the disagreement point
in Nash's set-up should be identified with the (limiting, as A —> 0) impasse
point, and not with the breakdown point (&A 5 ^B)-
It may be noted that the Nash bargaining solution is applicable if and
only if rA + A^ = r# + A# (since then r = 1/2). Furthermore, notice that
if \ A = \B = A, then the result stated in Corollary 4.5 is similar to that
stated in Corollary 4.3.
Suppose discounting is an insignificant force relative to the force of a
risk of breakdown. More precisely, consider the limiting SPE, as r^ —> 0
and rs —* 0. This means that the limiting impasse point converges to the
breakdown point (6^)^)? and Gi converges to XJ/{XA + XB)- In this limit,
therefore, player z's SPE payoff is decreasing in A^ and increasing in Aj,
which may be interpreted as follows: it pays to be optimistic, in the sense
of believing that the rate at which negotiations break down is small.

4.7 Notes
Binmore, Rubinstein and Wolinsky (1986) introduced the idea of a risk of
4.7 Notes 97

breakdown in the alternating-offers model, and investigated the relationship

with Nash's bargaining solution. The analyses in Sections 4.2 and 4.4 are
inspired by that paper.
The application studied in Section 4.3 is due to Hindriks, Keen and
Muthoo (1998). An interesting application — somewhat similar to that in
Section 4.5 — is contained in Rubinstein and Wolinsky (1985), who embed
the model studied in Section 4.5 in a model of a decentralized market. Un-
like in the application studied in Section 4.5, in Rubinstein and Wolinsky's
model the prices qs and qs respectively are determined by bargaining with
a different buyer and a different seller; hence the outcome in any one bar-
gaining situation is determined as part of a 'market equilibrium'. For an
excellent exposition of Rubinstein and Wolinsky (1985), see Osborne and
Rubinstein (1990, Chapter 6).
5 Outside Options

5.1 Introduction

Consider a situation in which University A and academic economist B bar-

gain over the wage. Moreover, suppose that the academic economist has
been offered a job at some alternative (but similar) university at a fixed,
non-negotiable, wage WB- A main objective of this chapter is to investi-
gate the impact of such an 'outside option' on the outcome of bargaining
between A and B. Although the academic economist B has to receive at
least a wage of WB if she is to work at University A, it is far from clear, for
example, whether the negotiated wage will equal WB, or strictly exceed WB-
In the next section I study the role of outside options through a simple ex-
tension of the basic alternating-offers model studied in Section 3.2. Section
5.3 contains applications to relationship-specific investments, sovereign debt
negotiations and bribery and the control of crime.
In contrast to the models studied in Chapter 4, in the model studied
in Section 5.2 there is no exogenous risk of breakdown. A player's decision
to take up her outside option (and thus, to terminate the negotiations in
disagreement) is a strategic decision: no random event forces her to opt out.
In Section 5.4 I extend the model studied in Section 5.2 by also allowing
for a risk of breakdown — and thus I study the relative impacts on the
100 Outside Options

bargaining outcome of the breakdown point and the outside option point1
In particular, in Section 5.4.4 I study a general model with outside options
and a risk of breakdown in which the set X of possible agreements can take
any form, subject to satisfying a modified version of Assumption 3.1.
The model studied in Section 5.2 may be interpreted as follows: at least
one of the two players has an outside option, which she has to accept or
reject within a short interval of time, and thus, the players bargain on the
expectation that within a short interval of time either agreement is struck
or outside options are taken up.2 The two models studied in Section 5.5,
on the other hand, may be interpreted as follows: when negotiations begin
neither player has an outside option, but while bargaining they may search
for an outside option, and when a player obtains an outside option she has
to immediately accept or reject it.
In Section 5.61 study a model that differs from the model studied in Sec-
tion 5.2 with respect to the points in the alternating-offers process at which
a player can choose to opt out. It will be suggested that the model studied in
Section 5.2 is suited to situations in which the players are making offers and
counteroffers face-to-face, while the model studied in Section 5.6 is suited
to situations in which the players are bargaining via a telephone/computer.
The issues of why, when and how to use Nash's bargaining solution in
bargaining situations with outside options is addressed in Section 5.2, while
the same issues are addressed in Section 5.4 in the context of bargaining
situations with outside options and a risk of breakdown.

5.2 A Model with Outside Options

Two players, A and I?, bargain over the partition of a cake of size n (where
7T > 0) according to the alternating-offers procedure, but with the following
modification. Whenever any player has to respond to any offer, she has
three choices, namely: (i) accept the offer, (ii) reject the offer and make
lr
The breakdown point is the payoff pair (bA,fr#)obtained through an exogenous risk
of breakdown, and the outside option point is the payoff pair (WA,WB) obtained when a
player strategically opts out (and the players take up their respective outside options).
2
Although the game is expected to end in a short interval of time, for reasons discussed
in Section 3.2.4 the infinite-horizon assumption better captures the players' strategic rea-
soning.
5.2 A Model with Outside Options 101

a counteroffer A time units later, and (iii) reject the offer and opt out, in
which case negotiations terminate in disagreement.
The payoffs are as follows. If the players reach agreement at time tA
(t = 0,1, 2 , 3 , . . . , and A > 0) on a partition that gives player i {i = A, B)
a share X{ (0 < Xi < TT) of the cake, then her payoff is x^exp(—r^tA). If,
on the other hand, the players do not reach agreement because a player
opts out at time tA, then player i takes up her outside option, and obtains
a payoff of t^exp(—r^tA), where WA < TT, WB < TT and WA + WB < TT.
The outside option point is the payoff pair (WA,WB), where Wi is player
i's outside option. 3 Finally, if the players perpetually disagree (i.e., each
player always rejects any offer made to her, and never opts out), then each
player's payoff is zero. It follows from Definition 3.1 that the impasse point

As in Section 3.2, for notational convenience, define Si = exp(—r^A). The

following proposition characterizes the unique subgame perfect equilibrium
of this model.

Proposition 5.1. The unique subgame perfect equilibrium of the model with
outside options and discounting is as follows:
• player A always offers x*A, always accepts an offer XB if and only if XB <
x*B, and always opts out after receiving an offer XB > x*B if &nd only if
6A%A ^ WA,
• player B always offers x*B, always accepts an offer XA if and only if XA <
x*A, and always opts out after receiving an offer XA > x\ if an^ onty if
3BX*B < WB, where

HATT if WA ^ 8AHATT and WB ^ ^^rj

TT — WB if WA < $A(TT — WB) and WB >

8BWA + (1 ~~ SB)TT if WA > SA/^ATT and WB ^ SB{TT — w /

TT — WB if WA > £>A(TT — WB) and WB > 8B(TT

3
It should be noted that I allow for the possibility that player i's outside option is
negative. It is only required that WA < TT, MB < TT and WA +WB < TT; for, otherwise, there
would not exist mutually beneficial partitions of the cake.
102 Outside Options

and

< $AHAK and WB <

< $A(K — ivB) and WB

TT — if WA > $AHAK and WB < 8B(TT

TT — WA if WA > SA{TT — WB) and WB > <5B(TT

with 11 A = (1 - 8B)/(I - SA^B) and \iB = (1 - <$A)/(1 - 8A8B)-

Proof. Since the proof involves a minor and straightforward adaptation of

the arguments in Sections 3.2.1 and 3.2.2 (which establish Theorem 3.1), I
shall only indicate the main changes to those arguments. In any SPE that
satisfies Properties 3.1 and 3.2 player i is indifferent between accepting and
not accepting player j ' s (j ^ i) equilibrium offer. That is
71
~~ X*A ~ max{<5^x^, WB} and
TT — x*B —

The unique solution to these equations is stated in the proposition. This

means that there exists at most a unique SPE that satisfies Properties 3.1
and 3.2, which is described in the proposition.4 Using an argument similar to
that contained in the proof of Proposition 3.1, it can be verified that the pair
of strategies described in the proposition is a subgame perfect equilibrium.
Finally, through a slight modification of the arguments presented in Section
3.2.2, it can be shown that there does not exist another subgame perfect
equilibrium.5 •

In the unique SPE, agreement is reached at time 0, and the bargaining

outcome is Pareto efficient: in equilibrium, the players do not take up their
4
If Wi = Six*, then player i is indifferent between opting out and making a counteroffer
when she receives an offer Xj > x*. In order to facilitate the statement of the proposition
(but without having any affect on the unique SPE path), I implicitly assume that she
breaks this indifference in favour of opting out. Without this (tie-breaking) assumption, if
Wi = Six* then there exists other subgame perfect equilibria, which, however, differ only
as to how player i breaks the above stated indifference; the SPE path is uniquely defined.
5
A key modification involves replacing the terms Si Mi and SiTfii respectively with
max{<5;Mi, w%) and maxl^ra;,!^}, which respectively are the maximum and minimum
SPE payoffs that player % obtains if she does not accept an offer (since, by not accepting
an offer, player i can either take up her outside option or make a counteroffer A time units
later).
5.2 A Model with Outside Options 103

respective outside options.6 However, the presence of the outside options

do influence the equilibrium partition of the cake, and this is relatively
transparent in the limit as A —> 0. The following corollary is an immediate
consequence of Proposition 5.1.7

Corollary 5.1 (The Outside Option Principle). In the limit, as A —>

0 ; the unique SPE share x\ obtained by player A converges to

{ 7T — WB
if WA < TJAK and WB <

if WA < T]ATT and WB >

an
where T\A = re/fjA + re) and T\B — ^A/^A + ^B)J d the unique SPE share
ifwA
obtained by player B converges to ix —>x**
VAK and wB <
A'
It is evident from Corollary 5.1 that if each player's outside option is less
than or equal to the share she receives in the (limiting) SPE of Rubinstein's
model (cf. Corollary 3.1), then the outside options have no influence on the
(limiting) SPE partition. On the other hand, if one player's outside option
strictly exceeds her (limiting) Rubinsteinian SPE share,8 then her (limiting)
SPE share is equal to her outside option.

Remark 5.1. Consider a modification of the model studied above in which

only a single player can choose whether or not to opt out. Thus, as in the
above model, a single player i (i = A or i = B) can choose whether or
not to opt out when responding to any offer, but player j (j ^ i) can only
choose between accepting and rejecting any offer. It is trivial to verify that
the unique SPE and limiting SPE respectively (of this modified version of
the model studied above) are as stated in Proposition 5.1 and Corollary 5.1
with Wj — 0.9
6
This makes sense, since otherwise the gains from trade, as captured by the 'surplus'
7T — WA — WB, are left unexploited.
7
It should be noted that in this limit the unique SPE partition does not depend on
which player makes the first offer, at time 0.
8
Notice that, since WA + ^ s < TT, both players' outside options do not exceed their
respective (limiting) Rubinsteinian SPE shares.
9
Notice that this means that the payoff player j receives when player % opts out has no
impact on the bargaining outcome; for further discussion on this point, see Remark 5.2.
104 Outside Options

As I now illustrate, the results derived above capture, in particular,

the notion that the players should not be influenced by threats which are
not credible. Suppose that University A and academic economist B are
bargaining over the wage w when neither player has any outside option.
The (instantaneous) utilities to A and B if they reach agreement on wage w
are 1 — w and w, respectively, and both players discount future utilities at
a common rate r > 0. Applying Corollary 3.1, it follows that the (limiting)
SPE wage w* = 0.5. Now suppose that B has an outside option WB > 0.
It seems intuitive that if £Ts outside option WB is less than or equal to 0.5,
then her threat to opt out is not credible. If she is receiving a wage of 0.5,
then getting an alternative job offer with a wage of 0.49, for example, should
be useless: University A should ignore any threats made by B to quit. If,
on the other hand, the alternative wage exceeds 0.5, then University A has
only to exactly match the outside wage offer; there is no need to give more.

Remark 5.2. In the model studied above I implicitly assume that the (in-
stantaneous) utility player i obtains when she opts out is equal to the (in-
stantaneous) utility she obtains when player j opts out (which is equal to
her outside option Wi). In some bargaining situations such an assumption
may not be valid. It is trivial to verify that the (instantaneous) utility player
i obtains when player j opts out has no impact whatsoever on the unique
SPE — which is rather intuitive, since player j ' s decision on whether or not
to opt out is not influenced by the utility player i obtains if she opts out.

5.2.1 Relationship with Nash's Bargaining Solution

I now show how to apply the asymmetric Nash bargaining solution in bar-
gaining situations with outside options. First, note that it follows from
Corollary 5.1 that in the limit, as A —> 0, the unique SPE payoff pair
(x^, 7T — x*A) converges to the unique solution of the following maximization
problem

uA,uBx

subject to (UA, UB) G O, M^ > 0 and UB > 0, where

: 0 < UA < K,UB = IT — UA, UA > WA and UB > WB}. (5.1)

5.3 Applications 105

The result stated in the following corollary is an immediate consequence of

this observation and Definition 2.3.

Corollary 5.2 (Relationship with Nash's Bargaining Solution). In

the limit, as A —• 0, the unique SPE payoff pair of the model with outside
options and discounting converges to the asymmetric Nash bargaining solu-
tion of the bargaining problem (il,d) with r = TBKTA + TB), where £1 is
defined in (5.1) and d = (0,0).

As has been shown in Chapters 3 and 4, Nash's bargaining solution is

applicable when the friction in the bargaining process is arbitrarily small
(i.e., A is arbitrarily close to zero). The important point to note here con-
cerns, in particular, how the outside option point should be mapped into
the objects upon which Nash's bargaining solution is defined. It should be
noted that in the bargaining situation considered above, there are two pos-
sible ways in which the players fail to reach agreement: through perpetual
disagreement and when a player opts out. As such there are two possi-
ble payoff pairs associated with a failure to reach agreement, namely, the
impasse point (IA,IB) = (0,0) and the outside option point (WA,WB)-
Corollary 5.2 states that the disagreement point in Nash's set-up should
be identified with the impasse point of the model with outside options and
discounting. The outside option point, on the other hand, constrains the
set Q of possible utility pairs on which Nash's bargaining solution should
be defined — by requiring that each utility pair (UA, UB) £ ^ be such that
Ui is at least as high as player i's outside option. I thus emphasize that the
outside option point does not affect the disagreement point.

5.3 Applications

5.3.1 Relationship-Specific Investments

At date 0 an investor (e.g., a Japanese firm) chooses the level of some in-
vestment (e.g., the size of a car plant in Birmingham), with the cost of
investment incurred at date 0. At a subsequent date 1, when the cost of in-
vestment is sunk, the investor and some second (non-investing) party (e.g.,
a union) bargain over the partition of a cake whose size depends on the
106 Outside Options

chosen investment level.10

More precisely, at date 0 player A chooses the level / of some investment,
where I > 0. The cost of investment I2 is incurred by player A at date 0.
Then, at date 1 player A and a second player B bargain over the partition
of a cake of size / according to the basic alternating-offers procedure. The
(instantaneous) utility player i obtains from a share X{ of the cake is x^,
and the players discount future utilities at a common rate r > 0. Applying
Corollary 3.1, it follows that the investment level chosen by player A in
the (limiting) SPE is 1/4. Therefore, there is under-investment relative to
the first best level of investment — where the first best investment level
maximizes the total 'surplus' I — I2. The intuition for this classic under-
investment result is as follows. For any investment level, in the (limiting)
Rubinsteinian SPE the cake is split equally between the two players. Hence,
since player A obtains only one-half of the marginal return on her date 0
investment, she under-invests relative to the first best level.
I now extend the model described above by giving player A the option
to sell her investment during the negotiation process with player B. Thus,
player A has an outside option: she can recover some fraction v (0 < v < 1)
of the cost of investment incurred at date 0, and, hence, her outside option
WA = vl2- The parameter v captures the degree to which the cost of
investment is sunk.
From Remark 5.1, one may apply Corollary 5.1 with WB = 0, and it
follows that for each investment level / chosen at date 0, the (limiting) SPE
payoff to player A at date 1 is11

\ul2 ifvI2>I/2.
Therefore, for each v E [0,1], the unique equilibrium investment level /*(^)
which maximizes P(/, v) — I2 is 1/4. This suggests that the degree of ineffi-
ciency of the equilibrium investment is independent of the degree of sunke-
10
It is assumed that it is impossible to write any kind of binding (long-term) contract
at date 0. In particular, the payment to be made by the investor to the union for the
latter's cooperation in generating the cake cannot be contracted upon at date 0.
n
I t should be noted that if WA < I, then Corollary 5.1 is applicable. However, if WA > I
(which is possible), then player A will certainly opt out — because if WA > I, then there
does not exist a mutually beneficial partition of the cake. Hence, if WA > /, then player
A's payoff is WA>
5.3 Applications 107

ness in the cost of investment. Contrary to conventional wisdom, the classic

under-investment result appears to have little to do with the sunkeness in
the cost of investment. Furthermore, this analysis suggests that the classic
under-investment result may not be robust to reasonable specifications of
the ex-post bargaining game when the investor may, in addition, be able to
strategically terminate the negotiations in order to sell her investment and
recover, if not all, at least some fraction of the cost of investment.

5.3.2 Sovereign Debt Negotiations

Country B1 whose wealth consists of one unit of some domestic commodity

/3, owes a large amount of some foreign commodity a to a foreign bank A.
By trading on international markets, B obtains P units of commodity a for
its one unit of commodity /?, where P > 1. The (instantaneous) utility to
B is the sum of the quantities of commodities o: and j3 that it consumes. In
the absence of any outside interference, B trades the unit of commodity (3
for P units of the foreign commodity a, and obtains a utility of P. However,
if B fails to reach agreement on some payment to A, then the bank seizes a
fraction v of the country's traded output.
The players bargain over the payment x that B makes to A. If agreement
is reached on x, then B trades the unit of its domestic commodity f3 for P
units of the foreign commodity a, and the (instantaneous) utilities to A and
B are x and P — x, respectively. The players discount future utilities at a
common rate r > 0.
While bargaining B can choose to opt out, and either consume the entire
unit of the domestic commodity (3 or trade without agreement. In the former
case her (instantaneous) utility is 1 and A's utility is zero, while in the latter
case her utility is (1 — v)P and A's utility is vP — since a fraction v of her
commodity is seized by the foreign bank. Hence, the outside option to B is
WB — max{l, (1 — v)P}. The foreign bank does not have an outside option:
it is only B who can choose whether or not to opt out immediately after
receiving any offer from A (cf. Remark 5.1, above).
From Remark 5.1, one may apply Corollary 5.1 with WA = 0, and it
follows that the (limiting) SPE payment made by the country to the foreign
108 Outside Options

bank is

{ P/2 if P > 2 and v > 1/2

P - 1 if P < 2 and v > 1 - (1/P)
i/P if z/ < 1/2 and v < 1 - (1/P).
Thus, if the gains from international trade P — 1 are sufficiently large and
the fraction v of traded output that can be seized is relatively high, then
£Ts threat to consume the unit of its domestic commodity (3 and her threat
to trade without agreement, respectively, are not credible. Hence, agree-
ment involves splitting the total (maximal) size of the cake. However, if the
gains from international trade are relatively small and the potential cost of
seizure vP relatively large, then country B's threat to consume the domes-
tic commodity is credible. Although, in equilibrium, agreement is reached
without this threat having to be carried out, it does affect the equilibrium
debt payment. Finally, if v < min{l/2,1 — (1/P)}, then B's threat to trade
in international markets without a debt agreement is credible. Hence, the
equilibrium debt payment equals vP.
A major insight obtained from this analysis is as follows. If v > min{l/2,
1 — (1/P)}, then v has no effect on the equilibrium debt payment. Thus, for
example, if v is currently above this critical value, then further international
trade sanctions (as captured by an increase in v) have no effect on debt
payments. In contrast, it is evident that an increase in the terms of country
£Ts international trade (as captured by an increase in P) always increases
debt payments.
It seems reasonable to assume that the value of v can be (strategically)
chosen by the foreign bank. Assuming that v is chosen before the negotia-
tions begin, and that the bank is committed to its choice, the bank will set
v to maximize the equilibrium payment x™.
If P < 2, then x*£ is strictly increasing in v over the closed interval
[0,1 - (1/P)], and equals P - 1 for any v in the interval (1 - (1/P),1].
Hence, if P < 2, then any v G [1 — (1/P),1] is optimal, and implies an
equilibrium payment of P — 1. If, on the other hand, P > 2, then x*£ is
strictly increasing in v over the closed interval [0,1/2], and equals P/2 for
any v in the interval (1/2,1]. Hence, if P > 2, then any v G [1/2,1] is
optimal, and implies an equilibrium payment of P/2.
Thus, for any P, it is not necessary for the foreign bank to seize all the
5.3 Applications 109

traded output in order to maximize the payment. However, the equilibrium

payment at the optimum does depend on the value of P. If P < 2, then at
the optimum the bank captures all of the gains from international trade —
it cannot, however, extract any greater amount of payment from Country
B. On the other hand, if P > 2, then they split the cake equally. This
implies that the equilibrium payment at the optimal value of v is strictly
increasing in P for all P > 1.

5.3.3 Bribery and the Control of Crime Revisited

I now reconsider the bargaining situation (studied in Section 2.5.3) between
the corruptible policeman P and the criminal C on the assumption that P
can choose to opt out and report C to the authorities, and C can similarly
choose to opt out and report herself to the authorities. The outside options
are wp — 0 and we = TT(1 — ^), and the players discount future utilities at
a common rate r > 0.
Applying Corollary 5.1, it follows that the equilibrium bribe paid by the
criminal to the policeman is

JTT/2 if iy > 1/2

\l/7T if ^ < 1/2.

The intuition for this equilibrium bribe is as follows. If v is relatively large,

then the criminal's outside option is relatively small, and thus (since the
policeman's outside option is zero), the outside options have no influence
on the equilibrium bribe. Notice that even if v is so large that the crim-
inal's outside option is negative, the policeman can only obtain a bribe of
TT/2. This is because the criminal's outside option does not determine the
credibility of the policeman's threat to opt out and report the criminal (cf.
Remark 5.2, above). Indeed, her threat to opt out and report the criminal
is never credible, since her outside option is zero. On the other hand, if
v is relatively small, then the criminal's outside option is relatively large,
and hence the amount of bribe that the policeman can extract is relatively
small: the criminal's equilibrium payoff TT — 6* is equal to her outside option
7T(1-I/).
Given this outcome of the bargaining situation, I reconsider the optimal
behaviour of the criminal and compare it with that derived in Section 2.5.3.
110 Outside Options

Her expected utility from stealing the money is £(TT — &*) + (1 — QTT. Since,
for any £ G (0,1) and v > 0, 6* < TT/2, her expected payoff from stealing the
money is greater than or equal to TT/2. Therefore, in equilibrium, for any
C E (0,1) and v > 0, the criminal steals the money.
This conclusion, that the instruments ( and v have no influence what-
soever on the determination of crime, appears to be rather striking — espe-
cially because I have not imposed an upper bound on either the penalty rate
v or the potential bribe b. Yet the result is rather intuitive. No matter how
large the penalty rate is, the criminal always evades paying the penalty by
giving the corruptible policeman a bribe, whose magnitude is at most equal
to half the amount of the money that she stole. Thus, even if the criminal
expects to be caught for sure, the crime is profitable. In sharp contrast to
the result derived in Section 2.5.3, the current result supports the conven-
tional wisdom that, since a penalty is evaded through bribery, society may
as well not institute the penalty.

5.4 The Effect of a Risk of Breakdown

I now extend the model studied in Section 5.2 by allowing for a risk of
breakdown — in order to explore how this force interacts with the force
of strategic opting out. In particular, I investigate the robustness of the
outside option principle (cf. Corollary 5.1) and the split-the-difference rule
(cf. Section 4.4.2).
The model analysed here involves the following extension of the model
studied in Section 5.2. At any point in time if a player chooses to reject any
offer and make a counteroffer (and thus, she chooses neither to accept the
offer nor to opt out), then immediately with probability p (where 0 < p < 1)
the negotiations break down in disagreement, while with probability 1 — p
the player gets to make her counteroffer A time units later.
For simplicity of exposition, I assume that the players are risk neutral: 12
the (instantaneous) utility to player i from obtaining a share xi G [0, n] of
the cake is Ui(xi) = X{. As in Sections 5.2 and 4.4, player i's discount rate is
T{ > 0. The outside option point (WA^B) is the instantaneous utility pair
obtainable if either player chooses to strategically opt out, and the breakdown
point (&A5&B) is the instantaneous utility pair obtainable if negotiations
This assumption is relaxed in Section 5.4.4.
5.4 The Effect of a Risk of Breakdown 111

break down randomly. Assume that WA < TT, WB < TT, WA + WB < TT,
bA G [O,TT), 6# G [O,TT) and 6A + fr# < TT — which ensure that mutually-
beneficial partitions of the cake exist.

5.4.1 The Unique Subgame Perfect Equilibrium

Through a straightforward adaptation of the arguments in Sections 4.4 and
5.2 (that lead to Propositions 4.3 and 5.1), it can be shown that the unique
SPE is as described in Proposition 5.1, but with the pair (x*A,x*B) different
from that stated in the proposition. In Proposition 5.2 below I characterize
the limit, as A —> 0, of the unique SPE. In deriving this limit I assume (as
in Section 4.4.1) that when A is small, p = AA, where A > 0 is the rate at
which negotiations randomly break down in disagreement.
Notice — see, for example, Section 4.4.1 — that the limiting (as A —» 0)
impasse point

Before proceeding, it should be noted that, unlike in the model with

only a risk of breakdown studied in Section 4.2, it is no longer true that
player i (i = A, B) can guarantee (in the limit as A —> 0) a payoff of X{ —
because player j (j ^ i) can permanently disagree by opting out, in which
case player i obtains her outside option W{. However, since player i can
guarantee a payoff (in the limit, as A —> 0) of W{ by opting out at the first
opportunity, the following counterpart to Lemma 4.2 is obtained.

Lemma 5.1. In any limiting (as A —» 0) subgame perfect equilibrium of

any subgame of the model with outside options, discounting and a risk of
breakdown, player i 's payoff is greater than or equal to W{.

Proposition 5.2. The unique limiting (as A —> 0) subgame perfect equilib-
rium of the model with outside options, discounting and a risk of breakdown
is as follows:
• player A always offers x\, always accepts an offer XB if and only if XB <
x*B, and always opts out after receiving an offer XB > x*B if and only if
112 Outside Options

• player B always offers x*B, always accepts an offer XA if and only if XA <
x\, and always opts out after receiving an offer XA > x\ if and only if
%*B < WB, where

(RA, RB) if WA < RA and WB < RB

(x%x*B) = ^ (TT - wB,wB) if WA < RA and WB > RB
(WA, TT — WA) if WA > RA and WB < RB

with RA and RB defined in (4-17) and (4-18).

The payoff pair (RA, RB) is the limiting SPE payoff pair when the players
do not have any outside options. Proposition 5.2 may thus be interpreted as
a generalization of the outside option principle (cf. Corollary 5.1) in which
the reference point is the pair (RA,RB) rather than the pair (TJA^,VB^),
where the latter is the limiting SPE payoff pair in the basic alternating-
offers model (cf. Corollary 3.1).
If both players' outside options are sufficiently unattractive (i.e., WA <
RA and WB < RB), then the equilibrium partition of the cake is determined
by the (limiting, as A —> 0) impasse point (XA,XB), which depends on
the breakdown point (bA,bs) — thus, in this case the outside options are
irrelevant to the bargaining outcome. On the other hand, if player z's outside
option is sufficiently attractive (i.e., w\ > i?i), then the equilibrium partition
of the cake is determined solely by her outside option Wi\ the breakdown
point has no influence whatsoever on the equilibrium partition, no matter
how attractive player j ' s payoff Xj from perpetual disagreement might be.
For example, if W{ = Xj = 0.9TT and Wj = X{ = 0, then players % and j
respectively obtain 0.9TT and O.ITT units of the cake. The message here is
that when some player's outside option is sufficiently attractive, then that
outside option has the decisive impact on the equilibrium partition of the
cake; the (limiting) impasse point is irrelevant to the bargaining outcome.
The intuition for this result follows from the fact that player i can opt out
immediately (either at t = 0 if i = 5 , or at t = A if i = A) and obtain her
attractive outside option, which thus leaves player j with her unattractive
outside option. Player j cannot obtain the attractive payoff of Xj, because
that payoff is obtained if and only if the players perpetually disagree.
It is interesting to note that if player i's outside option W{ > TT — Xj
(which implies that Wj < Xj), then player j ' s limiting SPE payoff TT — wi
5.4 The Effect of a Risk of Breakdown 113

is strictly less than Xj (but strictly greater than Wj). Thus, as indicated in
Lemma 5.1, player j cannot guarantee a payoff of Xj, because player i would
opt out if player j offers Xj > TT — Wi.

5.4.2 Relationship with Nash's Bargaining Solution

It follows from Proposition 5.2 that if WB < TT — I A and WA < TT — Z#,

where (XA,XB) is defined in (5.2), then the limiting SPE payoff pair - which
is (x*A,x*B) as stated in Proposition 5.2 — is the unique solution of the
following maximization problem

max (uA ~ dA)aA(uA - dB)l~(TA

uA,uB

subject to (UAIUB) G fi, UA > <$A and UB > d#, where

•0 < UA < 7T, UB — 7T — UA-, UA > U)A and UB > ^ B } (5.3)

(5.4)
= (rB + A)/(2A + r A + r B ). (5.5)

This observation implies the result — stated in Corollary 5.3 below — that if
WB < 7T—XA and WA < K—1B, then the limiting SPE payoff pair is identical
to the asymmetric Nash bargaining solution of the bargaining problem (fi, d)
with r = GA, where O, d and <JA are defined in (5.3)-(5.5). 13
On the other hand, if either WB > TT — XA or WA > TT — 2#, then the
bargaining problem (fi,d), where ft and d are respectively defined in (5.3)
and (5.4), does not satisfy Assumption 2.1 — since there does not exist a
utility pair (UA,V>B) £ ^ such that UA > dA and UB > dB> Hence, if either
UJB > TT — XA or WA > TT—XB 7 then Nash's bargaining solution is not defined
on this bargaining problem.

Corollary 5.3. Ifws < TT—XA andwA < TT — IB, where (XA,XB) is defined
in (5.2), then the unique SPE payoff pair of the model with outside options,
discounting and a risk of breakdown converges, as A —> 0, to the asymmetric
Nash bargaining solution of the bargaining problem (fi, d) with r — a A , where
ft, d and a A are defined in (5.3)-(5.5).
13
It should be noted that if WA > &A and WB >fr#,then WA < TT—XB and WB < TT —
114 Outside Options

Thus, if for each i = A,B, wi < TY — Xj (j ^ i) — which would be

satisfied if, for example, Wi > bi (i = A, B) — then Nash's bargaining
solution is applicable. The disagreement point in Nash's framework should
be identified with the (limiting, as A —> 0) impasse point, while the outside
option point appropriately constrains the set of possible utility pairs.
Corollary 5.3 combines the insights contained in Corollaries 4.4 and 5.2.
It should be noted that in the bargaining situation considered here, there
are three possible ways in which the players fail to reach agreement: through
perpetual disagreement, when a player opts out and when negotiations break
down in a random manner. As such there are three possible payoff pairs asso-
ciated with a failure to reach agreement, namely, the impasse point
the outside option point (WA,WB) and the breakdown point (6
Corollary 5.3 states that the disagreement point in Nash's set-up should
be identified with the impasse point. The outside option point affects the
set Q of possible utility pairs on which the Nash solution should be defined
— by requiring that each utility pair {UA,UB) G ri be such that U{ is at
least as high as the instantaneous utility player i obtains from her outside
option. The breakdown point, on the other hand, affects the disagreement
point through its impact on the impasse point.

5.4.3 The Impact of The Manner of Disagreement

I now focus on bargaining situations in which the breakdown point and
the outside option point are identical (i.e., 6^ = WA and bs — WB)- Such
bargaining situations may be interpreted as follows: there are two possible
manners in which negotiations can terminate in disagreement at any point in
time — either a player strategically opts out, or in a random manner when,
for example, a player gets fed up and 'walks away from the negotiating
table' — but the outcomes are independent of the particular manner in
which negotiations terminate in disagreement.
Straightforward computation show that in this case, for each i

Wi ^ Hi <^=^> \bi + o7- — 7T h f d

It thus follows that WA < RA and WB < RB if and only if both players'
expected discount rates TA/X and TB/X are sufficiently small. Hence, from
Proposition 5.2, it follows that:
5.4 The Effect of a Risk of Breakdown 115

Corollary 5.4 (Split-The-Difference Rule). Assume bA = WA andbs =

WB - For any bA and bs there exists numbers EA > 0 and CB > 0 such that if
TA/^ < ^A and r^/A < e#; then in the limit, as A —> 0; £/ie 5PE1 payoffs to
players A and B are respectively RA and RB, where RA and RB are defined
in (4.17) and (4.18).
Corollary 5.4 implies that if either the rate A at which negotiations break
down exogenously is sufficiently high or the players discount future utilities
sufficiently little (TA and r# are sufficiently small), then the limiting SPE
partition is captured by the split-the-difference rule. It can be verified that,
on the other hand, if the rate A at which negotiations break down exoge-
nously is sufficiently low, then the limiting SPE partition is captured by the
outside option principle. Therefore, the outside option principle (Corollary
5.1) is robust to a small risk of breakdown, while the split-the-difference
rule (cf. Section 4.4.2) is robust to strategic opting out if and only if both
players' expected discount rates (r^/A and rg/A) are sufficiently small.

5.4.4 A Generalization
I now extend the model studied in Section 4.6 by allowing for outside options
— player i (i = A, B) can choose whether or not to opt out after rejecting any
offer made by player j (j ^ i), where the outside option point is (WA,WB)-
In addition to the modified version of Assumption 3.1 specified in Section
4.6, assume that there exists an agreement x G X such that UA(X) > WA
and UB{X) > WB-
A minor adaptation of the analysis of Sections 3.4.1 and 3.4.4 charac-
terizes the unique SPE of this model. In particular, it follows that MA =
mA = V£ and MB = ms = Vg, where (V^,V^) is the unique pair that
satisfies the following two equations14

In the following proposition — which can be proven along the lines of the
proof of Proposition 5.2 — I characterize the limit of the unique SPE payoff
14
It should be noted that Lemma 5.1 is valid here, and thus, (4.22)-(4.25) are not
necessarily valid in this extension to the model studied in Section 4.6, in which players
also have access to outside options.
116 Outside Options

pair, as A —• 0.15

Proposition 5.3. In the limit, as A —• 0 ; the unique SPE payoff pair of

the general model with outside options, discounting and a risk of breakdown
converges to

(UA-,UB) if UJA < UA and WB <

< UA and WB > UB

> UA and WB < UB-,

where (UA,UB) is the unique solution to the maximization problem stated in

Corollary 4-5.

The following corollary to this proposition parallels the result contained

in Corollary 5.3.

Corollary 5.5. If WB < (J>(ZA) and WA < (J)~1{TB), then the unique sub-
game perfect equilibrium payoff pair in the generalized model with outside
options, discounting and a risk of breakdown converge, as A —> 0 ; to the
asymmetric Nash bargaining solution of the bargaining problem (fi, d) with
T = aA, where ft = {(UA,UB) •UA £ IA,UB — 4>(UA),UA > UJA and
UB > UJB} and d = (2^,2#) — with Ii {i = .A, B) and a A defined in
Corollary 4.5.16

5.5 Searching for Outside Options

The model studied in Section 5.2 may be interpreted as follows: at least
one of the two players has an outside option, which she has to accept or
reject within a short interval of time, and, thus, the players bargain on the
expectation that within a short interval of time either agreement is struck
or outside options are taken up. 17 The models studied in this section, on
the other hand, may be interpreted as follows: when negotiations begin
15
Recall that p -» 0 as A -> 0.
16
It should be noted that if WA > ^A and WB > bs, then WA < 4>~1(TB) and
17
Although the game is expected to end in a short interval of time, for reasons discussed
in Section 3.2.4 the infinite-horizon assumption better captures the players' strategic rea-
soning.
5.5 Searching for Outside Options 117

neither player has an outside option, but while bargaining they may search
for an outside option, and when a player obtains an outside option she has
to immediately accept or reject it.
The issue of how the bargaining process is interlaced with the search
process needs to be given careful consideration, as it may significantly affect
the outcome of bargaining. I consider the following two alternative manners
in which these two processes are interlaced. In the model studied in Section
5.5.1, a player can search for an outside option if and only if she physically
leaves the negotiating table — she cannot negotiate and search at the same
time. However, in the model studied in Section 5.5.2, a player receives
outside options while sitting at the negotiating table. The latter model
may be appropriate when, for example, a player has placed an advert in
the local newspaper and receives outside options via the telephone, while
the former model may be appropriate if a player has to go to an interview
before receiving an outside option.
In order to simplify the analysis, set TT = 1 and assume that only a
single player, player B, can search for outside options — which are located
according to a Poisson process at a rate A > 0. Furthermore, assume that
an outside option is a fixed share of a unit size cake. The magnitudes of
the outside options located by player B are independent and identically
distributed with a continuous cumulative probability distribution F, whose
support is the closed interval [0,1].
As a preliminary observation, it is useful to consider the case in which
player B searches for an outside option without ever negotiating with player
A. From Search Theory, it is known that her optimal sequential search strat-
egy has the reservation level property: that is, she keeps searching until she
locates an outside option whose magnitude is greater than or equal to some
predetermined value. Let y denote this predetermined value. Therefore, her
expected payoff from this search strategy is

C xdF(x)

It is straightforward to show that the optimal predetermined value y* satis-

fies

y* = P(v*). (5.7)
118 Outside Options

The following result states some standard properties about P.

L e m m a 5.2. The function P defined in (5.6) has the following properties:

(i) it has a unique fixed point, which is the unique global maximum, denoted
by y*, where 0 < y* < 1, and (ii) for any y G [0, y*)7 P(y) > y, and for any
ye{y*,i\,P{y)<y.
Proof. In the Appendix. •

5.5.1 Searching on the Streets

In the model considered here, player B cannot search and bargain at the
same time. The move-structure of the bargaining-search game, denoted by
G-, is as follows. At time 0, player A makes an offer to player B, who either
accepts the offer, or rejects the offer and makes a counteroffer ^RMC) A
time units later, or rejects the offer and withdraws from the negotiating table
in order to search for an outside option ('#£'). If player B chooses CRS\
then she keeps searching until she locates an outside option; during this time
neither player has any decisions to make. When player B locates an outside
option, she either accepts this outside option ('AOO1), or rejects the outside
option and continues searching ('C5'), or rejects the outside option, stops
searching and returns to the negotiating table (Cii/VT'), where it takes her
A time units to make an offer. If player B chooses 'CS\ then the move-
structure of the subgame that follows is identical to the move-structure of
the subgame that follows player 5 ' s decision to CRS\ Denote this move-
structure, of a subgame that begins with player B starting to search, by GN-
The move-structure of a subgame beginning with player z's offer, which is
denoted by Gi-, is independent of history. In GB-> player B makes an offer,
which player A either accepts or rejects.
The payoffs to the players are as follows. The game terminates if either
the players reach agreement on the partition of the unit size cake, or player
B accepts an outside option, or the players perpetually disagree, or player
B searches forever. In the latter two cases each player obtains a payoff of
zero. If player B accepts an outside option at time t > 0 whose magnitude
is y G [0,1], then her payoff is yexp(—rt), where r > 0 denotes the players'
common discount rate, and player A's payoff is zero. If the players reach
agreement at time t > 0 with player i obtaining a share X{ G [0,1], then
player i's payoff is a^exp(—rt).
5.5 Searching for Outside Options 119

The Unique Subgame Perfect Equilibrium

The analysis involves a minor adaptation of the arguments in Sections 3.2.1
and 3.2.2. In any SPE that satisfies Properties 3.1 and 3.2 player i is in-
different between accepting and not accepting player j ' s equilibrium offer.
That is

1 — x* B — 8x* A and (5.8)

I — x* A = maxjfe^, ZB}, (5.9)

where ZB is player S's SPE payoff when she begins to search for an outside
option, and it is the unique solution to the following equation

, ZB, y}dF(y) Xexp(-Xt)dt.

Using (5.6), (5.7) and Lemma 5.2, it follows that 18

_ IV if 2/* >6x*B
B
~\\[6x*BF(6x*B) +fix%ydF(y)]/(r + A) if y* < 6x*B.

Equations 5.8 and 5.9 have a unique solution, namely

\ i£y*> 6/(1 + 6).

It follows that there exists a unique SPE that satisfies Properties 3.1 and 3.2,
which is stated below in Proposition 5.4. In fact, I show in the proof to this
proposition that there does not exist another subgame perfect equilibrium.

Proposition 5.4. The unique subgame perfect equilibrium of the bargaining-

search game Q is stated in Tables 5.1 and 5.2, depending on whether y* <
6/(1 + 6) ory* > 6/(1 + 6).
18
The intuition for this is as follows. If y* > 6x*Bl then, in equilibrium, player B prefers
L
RS' to QRMC\ and 'CS" to lRNT\ On the other hand, if y* < 8x*Bi then the reverse
is the case: she prefers lRMCy to lRS\ and LRNT to lCS\ Hence, if y* < 6xBi then,
when she receives an outside option, she ' AOO' if its magnitude y > 6xB and LRNT' if
V
120 Outside Options

offer x*A = l/(l + 6)

Player A
accept xB < 1/(1 + 6)
offer x*B = l/{l + 6)
accept XA<l/(l + 6)
xA>l/(l + 6)
l
Player B RS' no
L
AOCT y> 6/(1 + 6)
no
L
RNT y<6/(l + 6)

Table 5.1: The unique SPE of the bargaining-search game Q if y* < 6/(1 + 6).

Proof. Using an argument similar to that contained in the proof of Propo-

sition 3.1, it can be shown that if y* < 6/(1 + 6) (resp., y* > 6/(1 + 8)),
then the pair of strategies stated in Table 5.1 (resp., Table 5.2) is a subgame
perfect equilibrium.19 Through a slight modification of the arguments in
Section 3.2.2, it can be shown — which is done in the Appendix — that
there does not exist another subgame perfect equilibrium. •

In the unique SPE there is immediate agreement: player B does not

engage in search. In the limit, as A —> 0, player S's SPE share 1 — x\
converges to

fl/2 if 2/* < 1/2

\y* if 2/* > 1/2.

The limiting SPE outcome is identical to the (limiting) SPE in the model
studied in Section 5.2 when player B has an outside option WB — y* (cf.
Remark 5.1). This means that the equilibrium partition in the bargaining-
search game Q is identical to the equilibrium partition in the modified version
of game Q in which player B does not have the option to return to the
negotiating table ('RNT') once she decides to leave it in order to search
Notice that this is a stationary subgame perfect equilibrium.
5.5 Searching for Outside Options 121

offer ^A — y
Player A
accept xB<l-6(l-y*)
offer xB = 1-6(1 -y*)
accept x A ^ J- — y

'RMC no
Player B <RS' x A> l-y*
<A00> y>y*
<CS' y<y*
L
RNT no

Table 5.2: The unique SPE of the bargaining-search game Q if y* > 8/(1 + 6).

for outside options — which suggests that the option to allow player B to
L
RNT is redundant.

5.5.2 Searching while Bargaining

The model studied here, in which player B can search while sitting at the
negotiating table, is a simple extension of the basic alternating-offers model
studied in Section 3.2. The extension is as follows. At any time £A (t =
0 , 2 , 4 , . . . and A > 0) immediately after player B rejects player ^4's offer,
with probability A A < 1 she receives an outside option y G [0,1], which
she either accepts ('AOO') or rejects ('RMC').20 With probability 1 - A A,
on the other hand, she does not receive an outside option, and, hence, she
makes a counteroffer at time (t + 1)A. Denote this bargaining-search game
by W. The focus is on arbitrarily small values of A.

20
Unlike in the otherwise similar model studied in Section 4.5, the negotiations do not
break down in a random manner. Although the opportunity to terminate negotiations is
a random event, player B strategically decides whether or not to terminate negotiations.
122 Outside Options

The Unique Subgame Perfect Equilibrium

The analysis involves a minor adaptation of the arguments in Sections 3.2.1
and 3.2.2. In any SPE that satisfies Properties 3.1 and 3.2

l-x*B = 6x*A and (5.10)

1 - x\ = (1 - XA)6x*B + XA [ max{54, y}dF{y). (5.11)

Jo
I now show that these equations have a unique solution. After substituting
for x\ in (5.10) using (5.11), and rearranging, it follows that

x*B (1 - 62) + <52AA [ l - F(6x*B)] - { 1 - 6 ) - 6XA [ ydF{y) = 0.

L J -*6XB

For each x*B and A, let L(x*B, A) be the LHS of this equation. For any r > 0
and A > 0, there exists a A > 0 such that for any A E (0, A), L(l, A) > 0.
Furthermore, L(0, A) < 0, the derivative of L w.r.t. x*B is strictly positive,
and L is continuous in x*B. Hence, for any A > 0 but sufficiently small,
(5.10) and (5.11) have a unique solution in x*A and x*B. This means that
there exists at most a unique SPE that satisfies Properties 3.1 and 3.2,
which is stated below in Proposition 5.5. Through a slight modification of
the arguments in the proof of Proposition 3.1 and Section 3.2.2, it can be
shown that this is the unique SPE. A key modification in the arguments
in Section 3.2.2 involves replacing the terms 6 STUB and 6BMB respectively
with

f1
(1 - XA)6rriB + A A / max{6mB,y}dF(y) and
Jo
(1-XA)6MB + XA [ max{6MB,y}dF(y)
Jo
— which respectively are the infimum and supremum of the set of SPE
payoffs to player B if she rejects any offer. It then follows from the unique-
ness of the solution to equations 5.10 and 5.11 that MA — m A = x*A and
MB = mB = x*B.

Proposition 5.5. Assume that A > 0 but sufficiently small, and let x*A and
x*B be the unique solution to (5.10) and (5.11). The unique subgame perfect
equilibrium of the bargaining-search game W is as follows:
5.5 Searching for Outside Options 123

• player A always offers x*A and always accepts an offer XB if and only if
xB < x*B,
• playerB always offers x*B, always accepts an offer XA if and only if XA <
x*A, and always CAOOJ y if and only if y > 6x*B.

In equilibrium agreement is reached immediately, at time 0. In order to

provide a relatively transparent comparison between player I?'s SPE shares
in the two bargaining-search games, Q and W, I first characterize player i?'s
limiting SPE share in game W as A —> 0.

Corollary 5.6. In the limit as A —> 0, the share 1 — x\ to player B in the

unique SPE of the bargaining-search game W converges to xB, where XB is
the unique solution to

„ _ !lB
XB
(2r/X) + [l-F(xB)Y
Proof For any A > 0 but sufficiently small, L(x^(A), A) = 0. Hence

llm
A
is derived by applying L'Hopital's Rule. Thus, after differentiating L w.r.t.
A and then taking the limit, it follows that, as A —> 0, x*B converges to xB,
where XB is stated in the corollary. The desired result follows immediately,
because in the limit as A —> 0, player £Ts SPE share 1 — x*A and x*B converge
to the same number. •

Unlike in the (limiting) SPE of the bargaining-search game £?, the pa-
rameters of the search process influence the (limiting) SPE partition of the
game W even when y* < 1/2. Indeed, player B's (limiting) SPE share XB
may be written as follows: XB = [7 + P(£#)]/(7 + 1 ) , where P is defined
in (5.6) and 7 = r/[r + A(l - F(xB))]- Since 7 > 0 and 0 < P(xB) < 1, it
follows by Lemma 5.2 that xB > max{y*, 1/2}. Hence, for any values of the
parameters, player B obtains a share that strictly exceeds both the expected
payoff y* that she obtains if she searches according to her optimal search
strategy and the (limiting) Rubinsteinian SPE share that she obtains in the
absence of any outside options — which is equal to one-half, because the
players have a common discount rate. Moreover, player S's (limiting) SPE
124 Outside Options

share in the bargaining-search game W is strictly greater than her (limiting)

SPE share in the bargaining-search game Q.
In contrast to game £?, in which player B has to physically leave the
negotiating table before any outside options are obtained, in game W outside
options arrive during the offer-counteroffer process, and the results above
suggest that the latter provides a relatively greater positive effect on player
B's bargaining power. The results obtained here may be illustrated with
reference to the academic job market. It is typically the case that if you
happen to be relatively young and/or not terribly good then you have to
apply for professorships and go to interviews in order to receive outside
offers, while if you are pretty desirable then you receive outside offers without
such efforts. The results derived above thus suggest an explanation for the
differences in academics' pay in terms of academic merit.

5.6 The Role of the Communication Technology

In the model studied in Section 5.2 player i can opt out after receiving player
j ' s offer, but not after her offer is rejected by player j . This is plausible if,
for example, the players are bargaining face-to-face, because player j can
always make an offer when player i is about to opt out, as the channel of
communication remains open until player i has physically left the negotiating
table. If, on the other hand, the players are bargaining via a telephone (or a
computer), then player i may opt out at any point in the negotiation process,
since the channel of communication is closed the moment the telephone is
hung (or the computer is switched off); in particular, player i can opt out
immediately after her offer is rejected.
In this section I study a model with outside options and in which bar-
gaining takes place via a telephone/computer. In order to derive the main
consequence of this alternative communication technology in a simple man-
ner, I adopt the following three assumptions: (i) player A cannot opt out,
(ii) player B can choose whether or not to opt out only immediately after
her offer is rejected, and (iii) player A's payoff when player B opts out is
zero.
The model, therefore, that I analyse below involves the following exten-
sion of the basic alternating-offers model studied in Section 3.2. Immediately
after player A rejects any offer made by player JB, player B has to decide
5.6 The Role of the Communication Technology 125

whether or not to opt out. If she opts out then negotiations terminate in
disagreement, and if she does not opt out then player A makes her coun-
teroffer A time units later. If player B opts out at time tA (where t is odd),
then the payoffs to players A and B respectively are 0 and WB exp(—r^tA),
where WB < ir.
In order to facilitate the discussion of the role of the communication
technology, I refer to the bargaining model with outside options analysed in
this section as the telephone game, and the model studied in Section 5.2 as
the face-to-face game.21

5.6.1 Equilibria in the Telephone Game

I begin by characterizing the subgame perfect equilibria that satisfy Proper-
ties 3.1 and 3.2 — the arguments involve a straightforward adaptation of the
arguments in Section 3.2.1. In any SPE that satisfies Properties 3.1 and 3.2
player i is indifferent between accepting and rejecting player j ' s equilibrium
offer. That is
71
~ X*A — 3
[O if wB >6B(K-X*A)
K ~ XB = \
[6AxA if WB < 6B{TT - xA).
Furthermore, since in any SPE that satisfies Properties 3.1 and 3.2 player
B either always opts out or always does not opt out, if WB = <$B(TT — x*A),
then 7T — x*B = 0 (resp., TT — X*B = 6AX*A) if player B always opts out (resp., if
player B always does not opt out). Solving these equations for x*A and x*B,
it follows that if either WB < 8%/^B^ or WB > S^TT (where \IB is defined in
Proposition 5.1), then these two equations have a unique solution in x*A and
x*B — which means that, for such values of WB, there exists at most a unique
SPE that satisfies Properties 3.1 and 3.2. However, if 82BJJLB^ < WB < ^B^I
then these two equations do not have a solution in x*A and x*B — which
21
It may be argued that whatever is the mode of communication (face-to-face or via
a telephone/computer), player i can only opt out after rejecting player j ' s offer — and
she cannot opt out after player j rejects her offer — because a 'rejection' can take the
form of a counteroffer: that is, after player B makes her offer, player A either accepts
it or makes a counteroffer. Although this kind of argument has merit, the players may
perceive otherwise: they may perceive that offers need to be rejected before counteroffers
are made. Hence, the telephone game may have some relevance.
126 Outside Options

means that, for such values of WB-> there does not exist a SPE that satisfies
Properties 3.1 and 3.2.
The following proposition characterizes the unique SPE when either
wB < $%VBK or wB > <5|?r.

Proposition 5.6. / / either WB < S%HBK or WB > $%K, then the unique
subgame perfect equilibrium of the telephone game is as follows:
• player A always offers x*A and always accepts an offer XB if and only if
xB < x*B,
• player B always offers x*B, always accepts an offer XA if and only if XA <
x^, and always opts out ifws > S^n and always does not opt out if WB <
^ where

, *
{{{1 -SBJTT.TT) lfwB

with iiA and fiB defined in Proposition 5.1.

Proof. Assume that either WB < 8%/^BK or WB > S^n. It has been shown
above that there exists at most a unique SPE that satisfies Properties 3.1
and 3.2, which is stated in the proposition. Through a minor modification
of the proof of Proposition 3.1, it can be verified that the pair of strategies
stated in the proposition is a subgame perfect equilibrium. I now establish,
through a straightforward adaptation of the arguments in Section 3.2.2, that
there does not exist another SPE.
First suppose that WB < SBTIBJ where UB is the infimum of the set of SPE
payoffs to player B in any subgame beginning with player A's offer. This
implies that in any SPE player B always does not opt out. It follows trivially
that Lemmas 3.1-3.4 are valid (because it is 'as if the players are bargaining
over the partition of the cake in the absence of any outside options). Hence,
the desired result is obtained, because UB = TT — MB = SB/^BK.
Now suppose WB > ^B^B-, where NB is the supremum of the set of SPE
payoffs to player B in any subgame beginning with player A's offer. This
implies that in any SPE player B always opts out, which, in turn, implies
that in any SPE player A always accepts any offer. Hence, TUB = MB = TT.
Since it is trivial to verify that part (i)'s of Lemmas 3.1-3.4 are valid, the
desired result is obtained, because NB = TT — TUB = SB^. D
5.6 The Role of the Communication Technology 127

I now show, by construction, that if 82BIIB^ < ^B < ^s71"? then there
exists many (indeed, a continuum) of subgame perfect equilibria in the tele-
phone game. The key idea involves the construction of two 'extremal' sub-
game perfect equilibria: these equilibria respectively give players A and B
their worst SPE payoffs.

state SA state SB
offer x\
Player A
accept T o <C T**
iiy ]-$ _^ Ou ~D
XB<X*B*

offer x*B x*B*

Player B accept XA < X\ XA < X%

opts out no yes

transitions switch to state SB switch to state SA

if player A makes if player B does
an offer XA > x\ not opt out

Table 5.3: A (non-stationary) SPE in the telephone game when 2

6 BHBK < WB <
an< X 7r
where xA = n — (WB/6B), XA* = (1 — <5#)7r, XB=TV — 6AX*A i *B* — -

Table 5.3 describes the two extremal SPE, depending on which of the
two states is the initial state. 22 If play begins in state SA, then the SPE
payoffs to players A and B (in the limit as A —> 0) are TT — WB and WB,
respectively — which gives player B her worst SPE payoff.23 On the other
hand, if play begins in state SB, then the limiting SPE payoffs to players
A and B are 0 and TT, respectively — which gives player A her worst SPE
payoff.
Using the One-Shot Deviation property, it can be verified that the strat-

22
It is convenient to describe (simple) non-stationary subgame perfect equilibria using
such a table. A player's equilibrium action at any point in the game depends on the state
that is prevailing at that point. Moreover, a transition rule dictates when, and if, the
state changes. For further explanation of this kind of table, see Osborne and Rubinstein
(1990, Section 3.5).
23
Since player B can opt out at time A after player A rejects her offer, her payoff in
any limiting SPE, as A —» 0, is greater than or equal to WB-
128 Outside Options

egy pair described in Table 5.3 is a subgame perfect equilibrium. 24 I shall

show here that in state SB it is optimal for player B to opt out. Consider
a decision node where player B decides whether or not to opt out, and
suppose that the state is SB- If player B plays according to the specified
strategy and opts out, then her payoff is WB> By the One-Shot Deviation
property, this behaviour is optimal if and only if she does not gain by de-
viating at that decision node and then conforming to the strategy specified
in Table 5.3 thereafter. Suppose she deviates and does not opt out, but
thereafter play proceeds according to the strategy pair described in Table
5.3. Since the state immediately switches to SA (before player A makes her
offer), player S's payoff from this one-shot deviation is <$B(TT — x*A) = WB-,
and thus this one-shot deviation is not profitable. I shall also verify that
whenever player A has to make an offer and the state is SA, she does not
gain from a one-shot deviation of making an offer XA > x*A. By making the
proposed equilibrium offer, her payoff is TT — (WB/SB)- Suppose she makes
a one-shot deviation by instead offering XA > X*A, and thereafter play pro-
ceeds according to the strategy pair described in Table 5.3. This deviation
induces the state to switch to SB- Since the hypothesis WB < S^TT implies
that TT — (WB/SB) > (1 — <5B)TT, player B rejects the offer XA > TT — (WB/$B),
and thus, player A's payoff from the proposed one-shot deviation is 0.
It is now straightforward to construct a continuum of subgame perfect
equilibria, based on the idea that a path of play is supported as a SPE path of
play by the credible threat that play moves to one of the extremal equilibria,
depending on which player needs to be 'punished' for having deviated from
this path. 25
For each 7, where (1 — <5B)TT < 7 < TT — {WB/^B)-, the pair of strategies
stated in Table 5.4 is a SPE, and the SPE payoffs to players A and B are
respectively 7 and TT — 7. The SPE described in Table 5.4 is based on the

24
The One-Shot Deviation property, which is also known by other terms, is essentially
the principle of optimality for discounted dynamic programming. A pair of strategies is
a SPE if and only if each player's strategy is immune to profitable one-shot (unilateral)
deviations. For a precise statement of this result, see, for example: Fudenberg and Tirole
(1991, Theorem 4.2) — who call it the 'one-stage deviation principle'; Osborne and Ru-
binstein (1994, Exercise 123.1) — who call it the 'one deviation property'; and Hendon,
Jacobsen and Sloth (1996) — who call it the 'one-shot-deviation principle'.
25
This idea also lies at the heart of the theory of infinitely repeated games; for an elegant
discussion, see Abreu (1988).
5.6 The Role of the Communication Technology 129

intial state 7
offer xA — 7
Player A
accept —

offer —
Player B accept XA <7
opts out —

transitions (i) if player A offers XA > 7,

then play switches to the SPE
stated in Table 5.3 beginning
in state SB-> and (ii) if player
B rejects an offer XA < 7,
then play switches to the SPE
stated in Table 5.3 beginning
in state SA

T a b l e 5.4: For each 7 such that (1 — <5B)TT < 7 < TT — (WB/8B), I state here a (non-
stationary) SPE in the telephone game when b\\iB^ < WB < ^B "-
71

simple idea mentioned above: if player i does not conform to the proposed
equilibrium path, then she is immediately punished by moving play to the
extremal SPE in which she obtains her worst SPE payoff. I leave it for the
reader to verify that neither player has any incentive to conduct a one-shot
deviation from any action when play is in the initial state 7. I have thus
established the following results.
Proposition 5.7. If S^/J,BTT < WB < ^B^, then, for any partition (7,TT — 7),
where (1 — 6B)TT < j < n — {WB/£>B), there exists a SPE in the telephone
game such that in equilibrium the players reach agreement at time 0 on this
partition.
Corollary 5.7 (Telephone Bargaining). In the limit, as A —» 0, i/0 <
W
B < VB^ (where T/B is defined in Corollary 3.1) then player B 's SPE payoff
is uniquely defined and equals r]B^, and if TJBTT < WB, then any number in
the closed interval [WB^TT] can be supported as a SPE payoff to player B.
A key message that emerges from the above analysis is that the impact
130 Outside Options

of a player's outside option when the players are bargaining via a telephone
may be relatively more potent than when they are bargaining face-to-face.
The intuition for this insight comes from the observation that, although
the credibility of the threat to opt out in the two models is ensured under
identical circumstances (i.e., only when the value of the outside option is
sufficiently large), the credible threat has a relatively more potent effect
in the telephone game, since (by being able to terminate communications
instantly) a player can effectively (in equilibrium) make a 'take-it-or-leave-
it' offer. Notice the other important insight obtained, which is robust to the
specification of the technology of communication, that the outside option is
useless if its value is sufficiently small (since the threat to opt out is, under
such conditions, not credible).

5.6.2 An Application to Relationship-Specific Investments

I reconsider the problem studied in Section 5.3.1 by modelling the date 1
bargaining situation using the telephone game.
Applying Corollary 5.7, it follows that for each investment level / > 0
chosen at date 0, if 1/2 > vl2 then the unique (limiting) SPE payoff to
player A is 7/2, and if 1/2 < vl2 < I then for every k G [vI2,I] there
exists a (limiting) SPE such that the payoff to player A is k. Furthermore,
if / < vl2 then (since gains to trade do not exist) player A takes up her
outside option and her payoff is vl2.
In characterizing the equilibrium investment, I assume that for any / and
v such that 1/2 < vt1 < /, player A:s equilibrium payoff at date 1 bargaining
is equal to /. I select this particular SPE payoff because it gives player A
the full benefit from her investment, and thus, the equilibrium investment
level chosen at date 0 should possess the least amount of inefficiency.
Hence, for each v G [0,1], the unique equilibrium investment level I1(u)
maximizes P T ( / , v) — I2, where

{ 1/2 if vl2 < 1/2

I if 1/2 <vl2 <I
vl2 if vl2 > I.

It is straightforward to show that, for each v G [0,1], the unique limiting (as
5.6 The Role of the Communication Technology 131

A —» 0) SPE investment level is

T, N fl/4 if 0 < z / < 4 - v / 1 2

[ if 4 -

I now compare this equilibrium investment level with the first best invest-
ment level and with the investment level chosen when the date 1 bargaining
takes place according to the face-to-face game — the latter is IF{v) = 1/4
for any v G [0,1] (cf. Section 5.3.1) and the former is /* = 1/2.
If v < 4 - x/12 then IT(v) = IF(v) < P — for such values of i/, in
both games there is under-investment relative to the first best. However, if
4 - VT2 < v < 1 then IT(v) > I* > IF\v) — for such values of z/, there
is under-investment in the face-to-face game, and over-investment in the
telephone game. Finally, if v — 1 then IT{y) — /* > IF{y) — investment in
the telephone game is at the first best level, while in the face-to-face game
there is under-investment.
One main insight thus obtained here is that under some circumstances,
investment may be above (rather than below) the efficient level.

5.6.3 Rubinstein Bargaining with Quit Options

Consider the basic alternating-offers model studied in Section 3.2, in which
neither player has access to outside options. It is plausible to assume that
each player does have the option to quit bargaining at any point in the ne-
gotiation process (and, thus, give up any attempt to reach agreement). Of
course, in that eventuality each player's payoff is zero. I ignored this issue
in Section 3.2 on the implicit assumption that a rational player would never
quit bargaining; she would prefer to get some agreement to no agreement. If
the players are bargaining face-to-face, and a player can quit only after re-
ceiving an offer from her opponent, then this implicit assumption is justified.
However, as I now show, if the players are bargaining via a telephone (and
a player can quit immediately after her offer is rejected), then the option to
quit may have a significant impact on the set of subgame perfect equilibria.
Table 5.5 describes two extremal SPE: if play begins in state s^, then
(in the limit as A —» 0) player i obtains the whole cake. Notice that in
each of these extremal SPE the first offer (made by player A) is accepted.
In equilibrium, neither player actually quits. However, if each player has
132 Outside Options

state SA state SB
offer X*A = K *A = (1 " 6B)*
Player A accept XB < (1 - <$A)TT XB < 7T

opts out yes no

offer xB = (1- 6A)TT XB=7T

Player B accept XA<7T XA<{1~ SB)7T

opts out no yes

transitions switch to state SB switch to state SA

if player A does if player B does
not opt out not opt out

Table 5.5: Two extremal SPE in Rubinstein's model with Quit Options.

the option to quit immediately after her offer is rejected, then this pair of
strategies is a SPE. 26 Notice that in state s^ player i plans to quit and receive
a payoff of zero. She does not find it profitable not to quit, because if she
does deviate then the state immediately switches to Sj (j ^ z), where player
j immediately obtains the whole cake — which is because player j plans to
quit in state Sj for exactly the same reason. Thus, in the telephone game,
player A plans to quit if and only if player B plans to quit. Of course, since
a player plans to quit, she can obtain the whole cake; in equilibrium her offer
becomes a ctake-it-or-leave-it' offer. One may interpret these extremal SPE
as follows. A player quits because she knows (or correctly expects) that if
she does not quit, then next period her opponent gives her no share of the
cake as her opponent quits. Her opponent plans to quit for exactly the same
reason. Using these extremal SPE it is trivial to construct a continuum of
SPE (as I did above).

26
Notice that this pair of strategies is not a SPE in the face-to-face game, where each
player can only choose whether or not to quit immediately after receiving an offer.
5.7 Appendix: Proofs 133

5.7 Appendix: Proofs

Proof of L e m m a 5.2

Differentiating P w.r.t. y, it follows that the set of turning points of P is

equal to the set of fixed points of P. That is

S = {y: P'(y) = 0} = {y : P(y) = y}. (5.12)

Brouwer's fixed point theorem implies that S is non-empty. Since P(0) > 0
and P is continuous, s > 0, where s is the inflmum of S. I now establish
that

VyG[0,s),P'(y)>0. (5.13)

Suppose, to the contrary, that there exists a y G [0,s) such that P^y) < 0.
If Pr{y) = 0, then y G 5, which contradicts the hypothesis that y < s. If
P ; (y) < 0, then (since P'(0) > 0) there exists an x, where 0 < x < y, such
that P'[x) — 0, which is a contradiction, since x < s. I now establish that

Vy>s,y>P(y). (5.14)

Suppose, to the contrary, that there exists a y > s such that y < P(y). This
implies that there exists an x where s_ < x < y such that P(x) = x and
P'{x) > 1, which contradicts (5.12). I now establish that

s is a local maximum. (5.15)

Suppose, to the contrary, that s is a point of inflexion.27 Since P(s) > 0 and
P(l) = 0, there exists a y, where y > 8, such that y is a local maximum.
Hence, using (5.12), there exists an e > 0 such that for any x G (y — e, y),
P{x) > x. This contradicts (5.14) since y > s. I now establish that

Vy>s,P'(y)<0. (5.16)

I argue by contradiction. First, suppose that there exists a y > s such

that P'{y) = 0. From (5.15), it follows that there exists an x, where s <
x < y, such that P'{x) — 0 and P{x) ^ x, which contradicts (5.12). By
a similar argument a contradiction is obtained if there exists a y > s_ such
that P\y) > 0. The lemma follows immediately by noting that P(0) > 0
and P(l) = 0, and by using (5.12), (5.13), (5.15) and (5.16).
7
By (5.13), s cannot be a local minimum.
134 Outside Options

Proof of Proposition 5.4

Let HB and KB respectively be the supremum and the infimum of the set of
SPE payoffs to player B in any subgame with move-structure QN- Through
a straightforward adaptation of the arguments in Section 3.2.2, it follows
that

1 - MB = 8m A (5.17)
1 — mB = (5.18)
(5.19)
1 — rriA = max{<5M#, (5.20)
1
f°° r f i
HB = exp(-rt) m8Lx{8MB,HB,y}dF(y)\\exp(-\t)dt
Jo \_ Jo J
(5.21)
hB = exp(-rt) / , hB, y}dF{y) Aexp(-Ai)di.
Jo
(5.22)

I now show that these equations have a unique solution, in which MA =

rriA = x% MB = mB = x*B and HB = hB = ZB-
First suppose that HB > 6MB and KB > STUB- Thus, from (5.21), it
follows that

HB =
sk
HB
(r/A) + 1 - F(HB)

Hence, from Lemma 5.2, it follows that HB — y*> Similarly, KB — y*\ and
hence, from (5.19) and (5.20), it follows that MA = TTIA = 1 — y*, and thus,
from (5.17) and (5.18), it follows that MB = mB = 1 - S(l - y*). This is
the unique solution if the hypothesis holds — that is, if y* > 8/(1 + 6).
Now suppose that HB < SMB and KB < SmB. Thus, from (5.17)-(5.20),
it follows that MA = VTLA = MB — TUB = 1/(1 + 8). From (5.21), I thus
obtain that

A
HB = (3F{(3) ydF(y)
r+A
5.8 Notes 135

where (3 = (5/(1 + 6). Similarly, KB — HB- This is the unique solution if

the hypothesis holds — that is, if HB < /3. Since H(f3) < j3 if and only if
P(P) < /?, it follows from Lemma 5.2 (ii) that HB < 0 if and only if y* < f3.
Finally, it is easy to verify that contradictions are obtained from the
remaining possible hypotheses, namely, (i) HB > 6MB and KB < STUB, and
(ii) HB < 8MB and KB > STUB- Having thus established the uniqueness of
the SPE payoffs, it is straightforward to show that the SPE stated in the
proposition is the unique SPE.

5.8 Notes
The basic ideas and models studied in Sections 5.2 and 5.6 are respectively
due to Binmore (1985) and Shaked (1994). The specific observation made
in Section 5.6.3 is due to Ponsati and Sakovics (1998).
The application to relationship-specific investments studied in Sections
5.3.1 and 5.6.2 are simplified versions of that studied in Muthoo (1998),
while the application to sovereign debt negotiations studied in Section 5.3.2
is a simplified version of the model studied in Bulow and Rogoff (1989).
The models studied in Sections 5.5.1 and 5.5.2 are respectively based upon
Muthoo (1995c) and Wolinsky (1987).
6 Inside Options

6.1 Introduction

Consider the basic exchange situation in which a seller and a buyer are
bargaining over the price at which the seller sells an indivisible object (such
as a house) to the buyer. If agreement is reached on price p, then the seller's
payoff is p and the buyer's payoff is ix — p. Furthermore, the seller obtains
utility at rate g$ while the object is in her possession, where gs > 0; thus,
for A > 0 but small, she obtains a payoff of gs A if she owns the house for A
units of time. Given her discount rate rs > 0, this means that if she keeps
possession of the house forever, then her payoff is gs/^s, which is assumed
to be less than TT — for otherwise gains from trade do not exist. The payoff
that the seller obtains while the parties temporarily disagree is her inside
option — which equals gs[l — exp(—rsA)]/r# if they disagree for A units of
time. In contrast, her outside option is the payoff she obtains if she chooses
to permanently stop bargaining, and chooses not to reach agreement with
the buyer; for example, this could be the price p* (where p* > gs/^s) that
she obtains by selling the house to some other buyer.
A main objective of this chapter is to explore the role of inside options
on the bargaining outcome. In the next section I study a simple extension to
Rubinstein's bargaining model in which both players have inside options. It
is shown that a player's bargaining power is strictly increasing in her inside
138 Inside Options

option, and strictly decreasing in her opponent's inside option. Further-

more, it is shown that (under some conditions) the equilibrium outcome is
identical to an asymmetric Nash bargaining solution of an appropriately de-
fined bargaining problem, with the disagreement point in Nash's framework
being identified with the impasse point. It will be shown that the players'
inside options influence the impasse point. Section 6.3 contains two applica-
tions — one is to takeovers in a duopoly, and the other is to sovereign debt
renegotiations.
In Section 6.4 I extend the model studied in Section 6.2 in order to
explore the relative impacts of inside options, outside options and a risk of
breakdown. If the players' outside options are sufficiently unattractive, then
the equilibrium outcome is identical to the equilibrium outcome in the model
studied in Section 6.2 when the players do not have outside options — thus,
in this case the outside options are irrelevant to the equilibrium outcome.
On the other hand, if one player's outside option is sufficiently attractive,
then her equilibrium payoff equals her outside option, while her opponent
obtains the remaining surplus — thus, in this case the inside options are
irrelevant to the bargaining outcome. Section 6.5 contains an application to
intrafamily allocation.
In the models of Sections 6.2 and 6.4 the inside options are exogenously
given. However, in some bargaining situations the players' inside options
between times £A and {t + 1)A might be determined strategically after an
offer is rejected at time tA. In Section 6.6 I study a bargaining model in
which the players' inside options are endogenously determined. Section 6.7
contains an application to union-firm wage renegotiations in which, while
the parties temporarily disagree, the union decides whether or not to go on
strike.

6.2 A Model with Inside Options

Two players, A and i?, bargain over the partition of a cake of size TT (TT > 0)
according to the alternating-offers procedure (described in Section 3.2). The
payoffs are as follows. If the players reach agreement at time tA (t =
0,1, 2, 3 , . . . and A > 0) on a partition that gives player i a share X{ (0 <
6.2 A Model with Inside Options 139

< ?r) of the cake, then her payoff is

rtA
rt
/ exp(-r^A),
Jo
where T{ > 0 and gi > 0. The interpretation behind this payoff is as fol-
lows: the second term is her (discounted) utility from xi units of the cake
(where Ti is her discount rate), while the first term captures the notion that
until agreement is struck player i obtains a flow of utility at rate gi. After
integrating the first term, it follows that this payoff equals

If an offer is rejected at time £A, then in the time interval A — before

a counteroffer is made at time (t + 1)A — player i obtains a utility of
gi[l — exp(—riA)]/ri, which is her inside option.1 Notice that for A small,
her inside option is approximately equal to <^A. The pair (gA,9B) is called
the inside option point.
If the players perpetually disagree (i.e., each player always rejects any
offer made to her), then player i's payoff is gijri. Thus, it follows from
Definition 3.1 that the impasse point (XA^B) — {QA/TA^QBI^B)- Assume
that gA/rA+QB/rB < ?r; for otherwise, gains from co-operation do not exist.
The following lemma is an immediate consequence of the observation that
player i can guarantee a payoff of gi/vi by always asking for the whole cake
and always rejecting all offers.

L e m m a 6.1. In any subgame perfect equilibrium of any subgame of the

model with inside options and discounting, player i ;s payoff is greater than
or equal to gi/ri.

The following proposition characterizes the unique subgame perfect equi-

librium (SPE) of this model. For notational convenience, define 6i = exp(—r^A

Proposition 6.1. The unique subgame perfect equilibrium strategies of the

model with inside options and discounting is as follows:
• player A always offers x\ and always accepts an offer XB if and only if

x
It is the utility payoff that she obtains while the players temporarily disagree.
140 Inside Options

• player B always offers x*B and always accepts an offer xA if and only if
%A < %\, where

* 9A . 1 - SB ( QA 9B\ ,
xA = — + —— 7T and
rA 1 - 6A8B V rA rBJ
* _9B 1 -<Sg ( 9A 9B
r
rB 1 - 6A8B V A r5
Proof The proof involves a straightforward adaptation of the arguments in
Sections 3.2.1 and 3.2.2. In particular, in any SPE that satisfies Properties
3.1 and 3.2 player i is indifferent between accepting and rejecting player j ' s
equilibrium offer. That is

IT - xA = h oB x B and
' B

B
TA

The unique solution to these equations is stated in the proposition. This

means that there exists at most a unique SPE that satisfies Properties 3.1
and 3.2, which is described in the proposition. Using an argument similar
to that contained in the proof of Proposition 3.1, it can be verified that the
pair of strategies described in the proposition is a subgame perfect equilib-
rium. Through a slight modification of the arguments presented in Section
3.2.2, it can be shown that there does not exist another subgame perfect
equilibrium. 2 •

In the unique SPE, agreement is reached at time 0, and the bargaining

outcome is Pareto efficient. Although in equilibrium neither player ever
obtains her inside option, the players' inside options have a significant impact
on the equilibrium partition of the cake. Since player A makes the first
offer, at time 0, the unique SPE share to player A is x*A and to player B is
7T — x*A, where x*A is stated in Proposition 6.1. Notice that player z's share
is strictly increasing in g^ and strictly decreasing in gj. Thus, player i's
'bargaining power' is strictly increasing in her inside option, and strictly
decreasing in her opponent's inside option. If rA = rB and 9A~9B'I then
2
A key modification involves replacing the terms b%Mi and &mj respectively with
<7i(l — 6i)/ri + SiMi and g%(l — 6i)/ri + Sifrii, which are respectively the maximum and
minimum SPE payoffs that player % obtains if she rejects any offer.
6.2 A Model with Inside Options 141

x*A > 7T — x*A, which suggests that player A has a first-mover advantage.
However, this first-mover advantage disappears in the limit as A —> 0: each
player obtains one-half of the cake. As I argued in Section 3.2.4, attention
should in general be focused upon arbitrarily small values of A. I therefore
obtain the following corollary to Proposition 6.1.

Corollary 6.1 (Split-The-Difference Rule). In the limit, as A - • 0,

the unique subgame perfect equilibrium shares of the cake to players A and
B respectively converge to

9
QA = -±+T,A(*-9-±-9-?-
TA \ ^A TB

rB \ TA rB )

where T\A — rB/(rA + rB) and rjB = VA/(TA + rB).

The limiting equilibrium partition of the cake, which is independent of

who makes the first offer, may be interpreted as follows. The players agree
first of all to give each player i a share gi/ri of the cake — which gives her
a payoff equal to the payoff that she obtains from perpetual disagreement
— and then they split the remaining cake.
The limiting SPE payoff pair (QA,QB) is the unique solution of the
following maximization problem:

subject to {UA,UB) E fi, UA > cU and uB > dB, where

ft — {(uAi V>B) •0 < UA < ft and uB = TT — UA} (6-1)

(6.2)

This observation implies the result — stated in the following corollary —

that the limiting SPE payoff pair is identical to the asymmetric Nash bar-
gaining solution of the bargaining problem (fi, d) with r = rjA, where tt and
d are respectively defined in (6.1) and (6.2).

Corollary 6.2 (Relationship with Nash's Bargaining Solution). In the

limit, as A —> 0; the unique subgame perfect equilibrium payoff pair in the
142 Inside Options

model with inside options and discounting converges to the asymmetric Nash
bargaining solution of the bargaining problem (Q,d) with r — rjA, where ft
and d are respectively defined in (6.1) and (6.2), and T]A = TB/{TA + r ^ ) . 3

Corollary 6.2 shows how to incorporate the impact of the inside options
— as captured by the flow rates gA and gs — in Nash's bargaining solution:
they affect the disagreement point in Nash's bargaining solution. This is
another illustration of the insight — obtained also in other contexts (such
as when bargaining takes place with the possibility of a risk of breakdown
(cf. Corollary 4.4)) — that the disagreement point in Nash's framework
should be identified with the impasse point.

Remark 6.1 (Two Interpretations of the Model). The model studied

above has been motivated by the exchange situation between the seller and
the buyer of a house (as described at the beginning of Section 6.1). More
generally, the model may be interpreted as capturing the negotiations over
some fixed surplus between two players who do not interact with each other
after agreement is secured, and who obtain flow payoffs until agreement is
secured — where this inside option point may or may not be generated from
some kind of interaction between them before agreement is secured. The
application studied in Section 6.3.1 fits this interpretation. In contrast, the
application studied in Section 6.3.2 fits the following alternative interpre-
tation, in which players A and B interact repeatedly after an agreement
is secured. Before an agreement is secured, the players' per period payoffs
are captured by the inside option point (gAi9B)- If the players agree (in
some period tA) to co-operate with each other in some form, then in each
subsequent period they generate a cake of size ir' — and they do not obtain
their respective inside options. Since gA + 9B < ^' (where, assuming that
^A — TB — r, 7T = Tr'/r), such co-operation is mutually beneficial. Hence,
the players bargain over the partition of the per period cake of size nf. A
key assumption is that the players are committed to the agreed partition —
that is, they cannot renegotiate in any future period. 4 It should be noted
3
Notice that if TA—^B-, then the asymmetric Nash bargaining solution of this bar-
gaining problem is identical to the Nash bargaining solution of this bargaining problem
(cf. Definition 2.1).
4
In Chapter 10 I study bargaining in long-term relationships without such a commit-
ment assumption — that is, the parties are free to renegotiate in the future an agreement
struck in the past.
6.3 Applications 143

that the extensions to this model studied in Sections 6.4 and 6.6 may also
be interpreted in these two alternative ways.

6.3 Applications

6.3.1 Takeovers in a Duopolistic Market

Consider a market for a homogenous commodity with two firms, A and B.

If the market price is p > 0, then the total quantity demanded per unit
time is a — p, where a > 0. The constant marginal cost of production of
firms A and B are respectively c and zero, where a > 2c > 0. If qA and qB
are respectively the quantities produced by firms A and B per unit time, it
follows that the profits per unit time to firms A and B are respectively

UA = (a - qA - qB)qA and UB = (a - qA - qB)QB ~ cqB.

Assuming that the firms make their respective quantity decisions simulta-
neously (and that they do not subsequently change those decisions), the
Nash equilibrium quantities of this (Cournot) model — which are derived
by first differentiating Hi w.r.t. ^, and then solving the two equations for
qA and qB — are q^ = (a + c)/3 and q^ = (a — 2c)/3. This implies that
the Nash equilibrium profits per unit time to firms A and B are respectively
YlcA = (a + c)2/9 and n g = (a - 2c)2/9.
Firm B considers an amicable takeover of firm A. If she succeeds in doing
so, then she acquires the relatively more efficient production technology of
firm A and becomes the monopoly supplier of the commodity. This means
that she would then earn a monopoly profit per unit time, which (with
zero marginal cost of production) equals II m = a2/4. Assuming that the
two firms discount future profits at a common rate r > 0, when the two
firms bargain over the price at which firm B buys (or takes over) firm A,
their respective payoffs from perpetual disagreement are 11^/r and 11^jr.
Applying the bargaining model studied in Section 6.2, note that gA = 11^,
gB — n § and IT = II m /r. Since II m > 11^ + 11^, in this bargaining situation
there exist gains from trade: that is, gA/r + fl^/r < TT. Hence, applying
Corollary 6.1, it follows that firm B would buy firm A at time 0 and pay
144 Inside Options

firm A a price

Suppose, on the other hand, that firm A takes over firm B — not to buy
its relatively less efficient technology but to eliminate its presence from the
market. Applying Corollary 6.1 again, it follows that the price firm A would
pay firm B in order to buy it is
A rrg i/nm rr? £

Since 11^ < 11^, pA < pB. However, the equilibrium payoffs to firms A and
B do not depend on whether firm B buys firm A, or vice-versa — the payoffs
to firms A and B are respectively U^/r + (Um/r - U^/r - Il%/r)/2 and
n ^ / r + (II m /r - n ^ / r — IT§/r)/2, whoever ends up being the monopolist.
The initial duopolistic market need not, however, end being a monopoly;
although gains to trade exist between the two firms, wealth and borrowing
constraints may prevent either firm from buying the other — notice that
the prices pA and pB are quite large as they embody the present discounted
values of all future profits. This argument suggests that since pA < pB, it is
more likely that firm A takes over (or buys) firm B. Thus, one can say that
efficient firms will tend to buy inefficient firms.
Since both pA and pB are strictly increasing in a, this suggests that as
the market size increases, it is less likely that any takeover occurs, and thus,
more likely that the market remains a duopoly. Furthermore, since pA is
strictly decreasing in c and pB is strictly increasing in c, this suggests that as
the relative degree of inefficiency between the two firms increases, takeover
by the relatively more efficient firm becomes more likely.

6.3.2 Sovereign Debt Renegotiations

Country i?, who produces one unit of some domestic commodity j3 per unit
time, owes a large amount of some foreign commodity a to a foreign bank
A. By trading on international markets, B obtains P units of commodity a
for one unit of commodity /?, where P > 1. The utility per unit time to B
is the sum of the quantities of commodities a and j3 that it consumes. In
the absence of any outside interference, in each unit of time B would trade
6.3 Applications 145

the unit of commodity (3 for P units of the foreign commodity a, and obtain
a utility of P. However, if A and B fail to reach agreement on some debt
repayment scheme, then the bank seizes a fraction v of the country's traded
output.
The players bargain over the payment per unit time x that B makes to
A everafter. If agreement is reached on x at time iA, then in each future
unit of time B trades the unit of its domestic commodity j3 for P units
of the foreign commodity a — and thus, the payoffs to A and B (from
time £A onwards) are respectively x/r and (P — x)/r, where r > 0 is the
players' common discount rate. It is assumed that the amount that B owes
A exceeds P/r, and furthermore, the parties are committed to the agreed
debt repayment scheme. In the framework of the model studied in Section
6.2, the players are bargaining over the partition of a cake of size TT = P/r;
and if player A receives a share x\ of this cake, then this means that the per
unit time repayment x = rxi.
The inside options to the players are now derived. If any offer is rejected
at any time £A, then — before a counteroffer is made at time (t + 1)A — A
units of the domestic commodity is produced. Country B either consumes
all of it or trades without agreement. In the former case the inside options
of B and A are respectively A and zero, while in the latter case the inside
options of B and A are respectively (1 — z/)PA and vPA — e, where e denotes
an infinitesimal (small) cost of seizure. Hence, since B makes the decision
on whether to consume or trade, it follows (in the notation of Section 6.2)
that
((0,1) ifl>(l-!/)P
- 6, (1 - V)P) if 1 < (1 - U)P.
Noting that QAI^ + 9B/r < ft, one may apply Corollary 6.1 and obtain
that the players reach agreement immediately (at time 0) with Country B
agreeing to pay the foreign bank an amount x per unit time, where, in the
limit as e —> 0, x converges to

X =

If international trade sanctions (as captured by the value of v) are suffi-

ciently harsh, then Country I?'s inside option is derived from consuming the
146 Inside Options

domestic commodity — which implies that the equilibrium debt payment

per unit time equals half the gains from trade. On the other hand, if inter-
national trade sanctions are not too harsh, then Country U's inside option
is derived from trading the domestic commodity — which implies that the
equilibrium debt payment per unit time equals the quantity of traded good
seized.
A major insight obtained from this analysis is as follows. If v is suffi-
ciently high, then further international trade sanctions (as captured by an
increase in v) have no effect on debt payments. In contrast, an increase in
the terms of Country I?'s international trade (as captured by an increase in
P) always increases debt payments.
It seems reasonable to assume that the value of v can be (strategically)
chosen by the foreign bank. Assuming that v is chosen before the renego-
tiations begin, and that the bank is committed to its choice, the bank will
set v to maximize the equilibrium debt payment per unit time x*. For any
P, x* is strictly increasing in v over the closed interval [0,1 — (1/P)], and
equals (P — l)/2 for any v in the interval (1 — (1/P), 1]. Hence, since at
u = l - (1/P), x* = P - 1, the optimal value of v is v* = 1 - (1/P). Thus,
since 1 — (1/P) < 1, the optimal level of v is strictly less than one; that is,
it is not optimal for the foreign bank to seize all the traded output.5 The
optimal level of debt payment per unit time equals P — 1, the gains from
international trade per unit time. Thus, at the optimum, the bank captures
all of the gains from international trade — it cannot, however, extract any
greater amount of payment per unit time from Country B.

6.4 The Effect of Outside Options

I now extend the model studied in Section 6.2 by allowing for outside options
— in order to explore how they interact with inside options. In particular, I
investigate the robustness of the outside option principle (cf. Corollary 5.1)
and the generalized split-the-difference rule (cf. Corollary 6.1).
The model analysed here involves the following extension of the model
studied in Section 6.2. As in the model studied in Section 5.2, whenever any
player has to respond to any offer, she has three choices, namely: (i) accept
5
Notice that this result has been derived on the assumption that it is (virtually) costless
for the bank to seize any fraction of the traded output.
6.4 The Effect of Outside Options 147

the offer, (ii) reject the offer and make a counteroffer A time units later
and (iii) reject the offer and opt out, in which case negotiations terminate
in (permanent) disagreement. As in Section 6.2, between two consecutive
offers the players obtain their respective inside options — as embodied in
the flow rates QA and gs> As in Section 5.2, the outside option point is the
instantaneous utility pair (WA,WB) obtainable if either player opts out. In
addition to the assumption that g& > 0, QB > 0 and QAI^A + 9B/re < ^
assume, as in Section 5.2, that WA < TT, WB < TT and WA + WB < TT; for
otherwise gains from co-operation do not exist.
I emphasize that the outside option point is the payoff pair obtainable by
the players if a player decides to permanently stop bargaining — that is, she
decides to permanently disagree, while the inside option point — or, more
precisely (when A is small) the payoff pair (gAA, gs A) — is the payoff pair
obtainable by the players if a player decides to temporarily stop bargaining
(for A time units) — that is, she decides to temporarily disagree.
The following bargaining situation illustrates the distinction between
outside options and inside options. When a married couple are bargaining
over how to co-operate, their outside options are their respective payoffs from
divorce, while their inside options are their respective payoffs from staying
married but not co-operating with each other. Notice that it is possible, for
example, that the husband's outside option is much smaller than his payoff
from perpetual disagreement, because he is quite an unattractive man, while
his wife is quite an attractive woman — thus, the utility he obtains from
being married to her when they never co-operate is much higher than the
utility he obtains from being divorced from her.
As is illustrated above in the bargaining situation faced by the married
couple, in general there need not be any relationship between a player's
outside option and inside option; thus, Wi can be less than, equal to, or
greater than gi/ri.6
A straightforward adaptation of the arguments in Sections 5.2 and 6.2
(that lead respectively to Propositions 5.1 and 6.1) establishes the existence
6
In some bargaining situations, however, a player's outside option may equal the payoff
she obtains from perpetual disagreement. For example, when bargaining over the price of
the house, the seller can always stop bargaining and choose not to sell the house. If there
are no alternative buyers available, then, indeed, the seller's outside option ws equals her
payoff gs/rs obtained from keeping the house for herself.
148 Inside Options

of a unique SPE. It should be noted that unlike in the model with only inside
options studied in Section 6.2, it is no longer true that player i (i = A, B)
can guarantee a payoff of gi/ri — since player j (j ^ i) can permanently
disagree by opting out, in which case player i obtains her outside option W{.
However, since player i can guarantee a payoff of 8{Wi by opting out at the
first opportunity, the following counterpart to Lemma 6.1 is obtained.

L e m m a 6.2. In any subgame perfect equilibrium of any subgame of the

model with inside options, outside options and discounting, player i 's payoff
is greater than or equal to 6{Wi.

In the following proposition I characterize the limit, as A —> 0, of the

unique SPE of the model described above.

Proposition 6.2. The unique limiting (as A —> 0) subgame perfect equilib-
rium of the model with inside options, outside options and discounting is as
follows:
• player A always offers x*A, always accepts an offer XB if and only if XB <
x*B, and always opts out after receiving an offer XB > x*B if and only if
x\ - WA,
• player B always offers x*B, always accepts an offer XA if and only if XA <
x*A, and always opts out after receiving an offer XA > x\ if and only if
x*B < WB, where

{ (QA: QB)

(TT - wB,wB)
ifu)A < QA and wB < QB

if WA < QA and wB > QB

(WA, 7T - WA) if WA > QA and WB < QB
with QA and QB defined in Corollary 6.1.
The payoff pair (QA, QB) is the limiting SPE payoff pair when the players
do not have any outside options. Proposition 6.2 may thus be interpreted as
a generalization of the outside option principle (cf. Corollary 5.1), in which
the reference point is the pair (QAIQB) rather than the pair (T^TT,^Tr),
where the latter is the limiting SPE payoff pair in the basic alternating-
offers model (cf. Corollary 3.1).
If both players' outside options are sufficiently unattractive (i.e., WA <
QA and WB < QB), then the equilibrium partition of the cake is determined
6.4 The Effect of Outside Options 149

by the impasse point (which depends on the players' inside options) — thus,
in this case the outside options are irrelevant to the bargaining outcome.
On the other hand, if player i's outside option is sufficiently attractive (i.e.,
Wi > Qi), then the equilibrium partition of the cake is determined solely
by her outside option W{\ both players' inside options and player j ' s outside
option have no influence whatsoever on the equilibrium partition, no matter
how attractive player j ' s inside option might be. For example, if W{ —
gj/rj = 0.9TT and Wj = gi/ri = 0, then players i and j respectively obtain
0.9vr and O.lvr units of the cake. The message here is that when some player's
outside option is sufficiently attractive, then that outside option has the
decisive impact on the equilibrium partition of the cake; the inside options
are irrelevant to the bargaining outcome. The intuition for this result follows
from the fact that player i can opt out immediately (either at t — 0 if % — B,
or at t = A if i = A) and obtain her attractive outside option, which thus
leaves player j with her unattractive outside option. Player j cannot obtain
the attractive payoff of gj/rj, because that payoff is obtained if and only if
the players perpetually disagree. It is interesting to note that if player z's
outside option Wi > TT — (gj/rj) (which implies that Wj < gj/rj), then player
j's limiting SPE payoff n — W{ is strictly less than gj/rj (but strictly greater
than Wj). Thus, as indicated in Lemma 6.2, player j cannot guarantee a
payoff of gj/rj, because player i would opt out if player j offers Xj > TT — Wi.

6.4.1 And a Risk of Breakdown

I now extend the model with inside options and outside options studied
above by allowing for a risk of breakdown. As in Section 4.4, immediately
after a player rejects an offer (that is, she chooses neither to opt out nor
to accept the offer) with probability A A (where A A < 1) the negotiations
break down in (permanent) disagreement, and with probability 1 — A A the
player gets to make her counteroffer A time units later, where A > 0. The
breakdown point (6^, &#) is the instantaneous payoff pair obtainable in that
eventuality, where bA > 0, 6# > 0 and 6^ + bs < TT.
A straightforward adaptation of the arguments used to prove Propo-
sition 6.2 can be used to establish the existence of a unique SPE in this
extended model. Proposition 6.3 below characterizes the limit, as A —> 0,
150 Inside Options

of this unique SPE. 7 It is useful to first derive the impasse point in this bar-
gaining model with inside options, outside options, discounting and a risk
of breakdown.8 For any A > 0 but small, player i's payoff from perpetual
disagreement (i.e., when each player always rejects any offer made to her,
and never opts out) is

[XAbt + (1
71 = 0

Since (when A > 0 but small) 8{ = exp(—r^A) = 1 — r^A, it follows that

(after substituting for Si, and then simplifying) the above expression equals
Xbt + (1 - XA)9l
ri + X + nXA
Hence, the (limiting, as A —> 0) impasse point

y TA + A rB + X J
P r o p o s i t i o n 6.3. The unique limiting (as A —> 0) subgame perfect equi-
librium of the model with inside options, outside options, discounting and a
risk of breakdown is as follows:
• player A always offersx*A, always accepts an offer xB if and only if XB <
x*B, and always opts out after receiving an offer xB > x*B if and only if
x\ < wAi
• player B always offers x*B, always accepts an offer XA if and only if XA <
x*A, and always opts out after receiving an offer XA > x\ if an^ onty tf
x*B ^ WBJ where

{xAl xB) — ^ (TT — wBi wB) if WA ^ Q\ and wB > Q*B

if WA > Q*A and wB < Q*B

with Q\ = 1A + O-A[TT - 1A - TB] and Q*B = 1B + aB[n - 1A — 1B], where

), aB = 1 — (JA and {IA^IB) is defined in (6.3).
7
Notice that Lemma 6.2 is also valid in this extended model.
8
The impasse point is the payoff pair obtainable when the players perpetually disagree
— that is, each player always rejects any offer made to her, and never opts out (cf.
Definition 3.1).
6.4 The Effect of Outside Options 151

Comparing Propositions 6.2 and 6.3, it is evident that the limiting SPE in
this extended model is essentially the same as the limiting SPE in the model
without a risk of breakdown. The new aspects are: (i) that the impasse
point now depends on the breakdown point (bA^bs) and on the rate A at
which negotiations break down randomly, and (ii) that player i's 'bargaining
power', as reflected in the share a\ of the remaining cake ft — 1A—1B that
she obtains, depends on A. A main message here is that when some player's
outside option is sufficiently attractive (i.e., w\ > Q*), then that outside
option has the decisive impact on the equilibrium partition of the cake;
the inside options, the breakdown point, the rate at which negotiations
break down randomly and the players' discount rates are irrelevant to the
bargaining outcome.

6.4.2 Relationship with Nash's Bargaining Solution

It follows from Proposition 6.3 that if WB < ft — 1A and WA < ft — 1B,
where (1A,1B) is defined in (6.3), then the limiting SPE payoff pair - which
is {x*A,x*B) as stated in Proposition 6.3 — is the unique solution of the
following maximization problem:

subject to (UA,UB) £ f^, UA > oU and UB > ds, where

Ft = {{UA, UB) : 0 < UA < K,UB =ft— UA, UA > UJA and UB > WB} (6.4)
d=(lA,lB) (6.5)
aA = (rB + A)/(2A + rA + r B ). (6.6)

This observation implies the result — stated in Corollary 6.3 below — that if
WB < TT — XA and WA < TT — IB, then the limiting SPE payoff pair is identical
to the asymmetric Nash bargaining solution of the bargaining problem (fi, d)
with r = a A, where f2, d and a A are defined in (6.4)-(6.6).9
On the other hand, if either WB > TT — 1A OV WA > ft — 1B-> then the
bargaining problem (Jl, d), where Vt and d are respectively defined in (6.4)
and (6.5), does not satisfy Assumption 2.1 — since there does not exist a
utility pair (UA,UB) G ft such that UA > dA and UB > ds- Hence, if either
9
It should be noted that if WA > ^A and WB >frs,then WA < TT—1B and WB < TT —
152 Inside Options

WB > ft—I A or WA > TY—IB, then Nash's bargaining solution is not defined
on this bargaining problem.

Corollary 6.3. I/WB < TT — XA andwA < TT — IB, where (XA^B) is defined
in (6.3), then the unique SPE payoff pair of the bargaining model with inside
options, outside options, discounting and a risk of breakdown converges,
as A —» 0 ; to the asymmetric Nash bargaining solution of the bargaining
problem (ft,d) with r = a A , where Q, d and a A are defined in (6.4)-(6.6).
Thus, if for each i = A,B, Wi < n — Xj (j ^ i) — which would be
satisfied if, for example, W{ > bi (i = A, B) — then Nash's bargaining
solution is applicable. The disagreement point in Nash's framework should
be identified with the (limiting, as A —* 0) impasse point, while the outside
option point appropriately constrains the set of possible utility pairs.
Corollary 6.3 is a trivial generalization of Corollary 5.3. It should be
noted that in the bargaining situation considered here, there are three pos-
sible ways in which the players fail to reach agreement: through perpetual
disagreement, when a player opts out and when negotiations break down in
a random manner. As such there are three possible payoff pairs associated
with a failure to reach agreement, namely, the impasse point {XA,XB), the
outside option point (WA,WB) a n d the breakdown point (bA^bs). Corollary
6.3 states that the disagreement point in Nash's set-up should be identi-
fied with the impasse point. The outside option point affects the set f2 of
possible utility pairs on which the Nash solution should be defined — by
requiring that each utility pair (UA,V>B) E £1 be such that u\ is at least as
high as player i's outside option. The breakdown point and the inside op-
tion point, on the other hand, affect the disagreement point through their
respective impacts on the impasse point.

6.4.3 A Generalization
I now extend the model studied in Section 6.4.1 by assuming — as in Section
3.4 — that the players are bargaining over a set X of possible agreements,
where Ui : X —» 3ft is player i's (instantaneous) von Neumann-Morgenstern
utility function.
Similar to Assumption 3.1, assume that the Pareto-frontier Vte of the set
ft of instantaneous utility pairs obtainable through agreement is the graph
of a concave function, denoted by 0, whose domain is an interval I A ^ 3ft and
6.4 The Effect of Outside Options 153

range an interval IB ^ 3ft, with XA G I A and 2# G IB and (J>(TA) > ZB, where
the impasse point (IA,IB) is defined in (6.3). Furthermore, assume that
there exists an agreement x G X such that UA(x) > w>A and UB{X) > WB-
A minor adaptation of the analysis of Sections 3.4.1 and 3.4.4 charac-
terizes the unique SPE of this model. In particular, it follows that MA —
rriA = VA and MB = TUB — V^, where (V^Vg) is the unique pair that
satisfies the following two equations

(1 - 6B)/rB
- XA)\gA(l - 6A)/rA + SAVX)]}.

In the following proposition — which can be proven along the lines of Propo-
sition 6.3 — I characterize the limit of the unique SPE payoff pair, as A —> 0.

Proposition 6.4. In the limit, as A —> 0, the unique SPE payoff pair of
the general model with outside options, inside options, discounting and a
risk of breakdown converges to

where (UA->UB) is the unique solution to the maximization problem stated in

Corollary 4-5, but with (IA,IB) as defined in (6.3) and \ A = \B — A.

The following corollary to this proposition parallels the result contained

in Corollary 6.3.

Corollary 6.4. If WB < <\>^A) and WA < (/)~1(IB)^ then the unique sub-
game perfect equilibrium payoff pair in the generalized model with inside
options, outside options, discounting and a risk of breakdown converge, as
A —• 0 ; to the asymmetric Nash bargaining solution of the bargaining prob-
lem (Q,d) with r = a A, where ft = {(UA,UB) '•UA G IA,UB — (/)(u A),uA >
WA and UB > UJB}, d = (IA,TB) — as defined in (6.3) — and a A is as
defined in (6.6).10

0
It should be noted that if WA > &A and WB > &B, then WA < 4> 1(%B) and
154 Inside Options

6.5 An Application to Intrafamily Allocation

A husband H and a wife W bargain over the quantities of four variables,
namely, the per unit time consumption levels of a private good by the hus-
band and the wife (which are respectively denoted by XH and x\y) and the
per unit time quantities of two 'household' public goods (which are denoted
by q\ and q2). The income per unit time of player i (i = H, W) is Y{ > 0,
where Y# + Y\v — Y > 0. The price of the private good is normalized to
one, and the price (or, unit cost) of the public good qj~ [k — 1, 2) is p^. Thus,
the set of possible agreements (or, allocations) is

X = {{xH,xw,qi,q2) : xH + xw + piqi -\-p2q2 = Y}.

If and when agreement is struck, the parties are committed to the agreed
per unit time allocation; that is, they do not renegotiate it in the fu-
ture. Thus, if agreement is reached on x G X at time £A, then the pay-
offs to the husband and the wife (from time £A onwards) are respectively
UH(X) = VH(xH,qi,q2)/r and Uw(x) = Vw(xw,qllq2)/r, where r > 0 is
the players' common discount rate and V{ — which is strictly increasing in x^
q\ and q2 — is the utility player i obtains per unit time. Assume that the set
Q of instantaneous utility pairs obtainable from agreement satisfies Assump-
tion 3.1, where Q = {(UH,V>W) '•there exists x G X such that UH{X) =
un and Uw(x) — ^v^}-
While bargaining each player has the option to divorce — which entails
permanent disagreement. Thus, the instantaneous utility pair associated
with divorce is the outside option point. Player i's outside option Wi will
depend on Y{, amongst other parameters. Assume that the outside option
point (WH, WW) li es below the Pareto frontier of Q; for, otherwise, there are
no gains to being married — the parties would divorce at time 0.
If the players temporarily disagree over the allocation, they continue to
function and live as a married couple, but in the absence of co-operation and
coordination — which is formally modelled as follows. Based on established
gender roles, to which the parties are committed, the husband controls and
chooses the quantity of the household public good q\, and the wife controls
and chooses the quantity of the household public good q2. The players' inside
options are the Nash equilibrium payoffs of the following simultaneous-move
game. The husband chooses XH and q\ subject to his budget constraint
6.5 An Application to Intrafamily Allocation 155

X
H + P1Q1 — Yiti and simultaneously the wife chooses xw and q2 subject
to her budget constraint x\y + P2Q2 = Yw- The payoffs from these strat-
egy choices to the husband and the wife are respectively VH(XH, QI1Q2) and
Vw{xw-> QI1Q.2)' I assume that the Nash equilibrium payoffs in this game are
uniquely defined, and they define the inside option point (gH,9w)- Each
player's inside option depends on Y# and Yw, amongst other parameters.
Assume that (g#/r, #w'A*) lies below the Pareto frontier of $7; for otherwise,
there are no gains from co-operation within the marriage — the parties
would perpetually disagree (but not divorce, since I have assumed above
that the outside option point lies below the Pareto frontier of Q).
Assuming that there is no risk of breakdown in a random manner (i.e.,
A = 0), and letting the graph of uw = (J>{UH) denote the Pareto frontier of
fi, one may apply Proposition 6.4 and obtain that (in the limit as A —> 0)
the unique SPE payoff pair is 11

if WH < UH and ww < u\y

HWH < UH and w\y > uw
if WH > uH and ww <

where (UH, UW) is the unique solution to the following maximization problem

^max^tf - (gH/r))(uw ~ (gwM)

subject to {UH-)UW) G Q,UH > QHIT and u\y > 9w/r-

The Effects of Family Policies

I now consider the implications on the equilibrium intrafamily distribution

(I/JJ, Uyy) of three family policies aimed at affecting intrafamily distribution
of welfare.
First, consider the effect of a government policy that provides an al-
lowance T to the wife which is financed entirely by taxing the husband;
thus, 0 < T < YH — assume that YH > 0. This means that the husband's
post-tax income is Y'H = YH — T, and the wife's post-allowance income is
Y^Y = Yw + T. Since the total family income Y'H + Y^ is unaffected by this
11
Notice that in the limiting SPE agreement is reached immediately, at time 0; the
bargaining outcome is Pareto efficient.
156 Inside Options

policy (that is, Y'H + Y^ = YH + Yw), the set ft of instantaneous possible

utility pairs obtainable from agreement is unaffected. The players' out-
side options are also unaffected, because this policy is targeted to married
couples, not to divorcees. However, the husband's inside option decreases,
while the wife's inside option increases — because, although player i's in-
side option (which is a function of YH and Yw) is increasing in both YH and
Yw, the marginal effect of Y{ on player z's inside option is strictly greater
than the marginal effect of Yj on player i's inside option. Hence, it follows
(with reference to the pre-policy equilibrium outcome) that the post-policy
equilibrium payoff pair is

{ (u'H, ufw)

{(t)~ 1
{ww),ww)
if wH < u'H and ww < u'w

if wH < u'H and ww > u'w

where v!H < UH and u w > uw>12
f
ufH and ww < u'W)
If the pre-policy outcome (u*H, Uyy) — {UH-> UW) (which means that UH >
WH and uw > ww), then for any T G (0, YH] the wife's equilibrium payoff
increases and the husband's equilibrium payoff decreases relative to their
respective pre-policy equilibrium payoffs. On the other hand, if the pre-
policy outcome {u*H,Uyy) = (WH,<I>(WH)) (which means that UH < WH and
uw > ww)-) then the policy has no effect on intrafamily distribution: the
players' equilibrium payoffs in the pre-policy and post-policy situations are
identical. Finally, if the pre-policy outcome {u*H,Uyy) — {4 >~1(ww)i ww)
(which means that UH > WH and uw < ww), then if T is sufficiently small
the policy has no effect on intrafamily distribution, but if T is sufficiently
large then the policy makes the wife better off and the husband worse off
relative to the pre-policy situation. In summary, the key message is that
this policy may (but it need not) affect intrafamily distribution of welfare
(by making the wife better off and the husband worse off); it depends on
the nature of the pre-policy equilibrium outcome, which, in turn, depends
in particular on the couples' respective preferences, pre-policy incomes and
payoffs following divorce.
12
Since gfH < gu and g'w > gw, and since the set Q is unaffected by the policy (which
means that the function (p is unaffected), the solution (ufH,u'w) to the maximization
problem stated above is such that u'H < UH and uw > uw-
6.5 An Application to Intrafamily Allocation 157

I now consider the relative effects on intrafamily distribution of two

related government child allowance policies (or, schemes). In both schemes,
in the event of divorce it is the wife (mother) who gets the child allowance.
The two child allowance schemes differ according to which parent receives the
child allowance within marriage. Suppose that the child allowance is some
money C. Since total family income with either of these two child allowance
schemes is the same, namely, Y+C, it follows that the set ft of instantaneous
possible utility pairs obtainable through agreement is the same under either
scheme. Furthermore, the players' outside options are also the same under
either scheme — since in both schemes the child allowance is received by
the mother when the couple get divorced. It is the inside options which are
different under these two schemes. Under the scheme where the husband
receives the child allowance within marriage, his inside option increases and
the wife's inside option decreases; the reverse is the case under the scheme
where the wife receives the allowance within marriage. Thus, the equilibrium
distribution when the husband receives the child allowance within marriage
is

{ (x,yy) WH < Uft and

< u^ and ww > u^

and ww < Uyy ,

and the equilibrium distribution when the wife receives the child allowance

within marriage is

(tiff , u\)y) if WH < UJJ and w\y < ^ ^

(0~1(w;v^),^v^) if WH < S ^ and ww > ^W
{•>4>{WH)) if WH > u^ and

where u^ > u^ and u^ < u$r. The main key message from this analysis
— which is self-evident from these two equilibria — is that the two child
allowance policies may generate identical equilibrium payoffs. For some val-
ues of the parameters (which include the couples' respective preferences and
pre-policy incomes, and the level of the child allowance), these two alterna-
tive child allowance schemes do not have differing impacts on intrahousehold
distribution. When they do (which they will for some parameter values),
158 Inside Options

then the parent who receives the child allowance within marriage is better-
off relative to the alternative scheme in which her partner receives the child
allowance.

6.6 Endogenously Determined Inside Options

In the models studied in Sections 6.2 and 6.4 the players' inside options are
exogenously given. I now study a model in which the players' inside options
are endogenously determined. 13
Two players, A and S , bargain over the partition of a cake of size TT (TT >
0) according to a modified version of the basic alternating-offers process.
The modification is as follows. At any time tA (t = 0,1, 2 , . . . ) after any
offer is rejected, players A and B simultaneously choose 'actions' from their
respective sets of actions SA and SB- The players' inside options between
times tA and ( t + l ) A depend on the chosen actions: gi(s) denotes the rate at
which player i obtains utility when the chosen action profile is s — {SA-, SB)->
where gi(s) > 0. Thus, if at time tA (after an offer is rejected) the players'
chosen action profile is s, then player i's inside option between times tA
and (t + 1)A is (1 — 8i)gi(s)/ri) where r\ > 0 denotes her discount rate and
^ = exp(-r z A). 1 4
If agreement is reached at time tA (t = 0,1,2,3,...) on a partition
that gives player i a share X{ (0 < X{ < TT) of the cake, and if at time qA
(q = 0,1, 2 , . . . , t — 1) after the offer is rejected the players' chosen action
profile is sq, then player i's payoff is

t 1
~ (i _ g.\ .( ^

q=0

If, on the other hand, the players perpetually disagree (i.e., each player

13
There is neither a risk of breakdown nor do the players have outside options. The
focus in this section is entirely on the role of endogenously determined inside options (and
discounting) on the bargaining outcome.
14
A player's 'inside option' after an offer is rejected at time tA is the utility payoff
(discounted to time tA) that she obtains during the time interval [tA, (t + 1)A] while the
parties temporarily disagree.
6.6 Endogenously Determined Inside Options 159

always rejects any offer made to her), then player i's payoff is

(6.7)
n
q=o

where (sq) denotes the infinite sequence of action profiles chosen by the
players. Notice, therefore, that the impasse point (lA{(sq)),lB((sq))) —
which is the payoff pair obtainable from perpetual disagreement — depends
on the infinite sequence of chosen action profiles. In order to ensure that
there exist gains from co-operation, assume that the sum of the players'
payoffs from perpetual disagreement is less than the size of the cake. That
is:

Assumption 6.1. For any infinite sequence of action profiles (s 9),

)) < IT.

It should be noted that Assumption 6.1 implies that for any s = (SA, SB)
(where SA £ SA and SB £ SB), 9i(s) is bounded from above. It is helpful to
denote the simultaneous-move game with action sets SA and SB, and payoff
functions QA - S —> 5ft and QB •S —* 5ft, where S = SA X SB, by Q. I call this
game the 'disagreement' game, as it determines the players' payoffs while
they temporarily disagree. It will be assumed that the disagreement game
Q has at least one Nash equilibrium (NE).

6.6.1 Stationary Equilibria

In a stationary SPE each player's strategy is independent of history and
time: that is, each player always chooses the same action when playing the
disagreement game Q, always makes the same offer when she has to make
an offer and always responds to an offer in the same way when she has to
respond to that offer.
Consider an arbitrary stationary SPE, and let SA and SB respectively
denote the actions that players A and B always take when playing the
disagreement game Q. It is straightforward to see that the action profile
(SA,SB) is necessarily a Nash equilibrium of the disagreement game Q\ for
otherwise, some player can benefit from a unilateral deviation to an alterna-
tive action when playing Q. The set of stationary subgame perfect equilibria
are thus characterized in the following proposition.
160 Inside Options

Proposition 6.5. / / the disagreement game Q has N (where N > 1) Nash

equilibria, then the model with endogenously determined inside options and
discounting has N stationary subgame perfect equilibria, which are charac-
terized as follows. For each Nash equilibrium J of the disagreement game Q,
the unique subgame perfect equilibrium in which player i (i = A, B) always
chooses her Nash strategy s*i is as follows:
• player A always offers x\, always accepts an offer XB if and only if XB <
x*B, and always chooses the Nash strategy 'SA,
• player B always offers x*B, always accepts an offer XA if and only if XA <
x*A, and always chooses the Nash strategy s# ; where

7T — and
rA 1 V
9B 1-6B ( 9A 9B
r
rB 1 - OAOB V A rB

with'gi = gi(s).

Proof. The proposition follows from the argument that precedes it, and
Proposition 6.1. •

If the disagreement game Q has a unique Nash equilibrium, then there

exists a unique stationary SPE, which is identical to the unique SPE of
the model studied in Section 6.1 in which the inside option point is exoge-
nously specified. In the current model, however, the inside option point is
endogenously determined: it is (5,4(5),^(3)), where 's is the unique Nash
equilibrium of the disagreement game Q.

6.6.2 Markov Equilibria

In a Markov SPE each player's strategy is independent of history (but not

necessarily of time): that is, at each time tA (t — 0,1, 2 , . . . ) each player
chooses the same action when playing the disagreement game Q, and (de-
pending on whether she has to make or respond to an offer at time tA) she
makes the same offer and responds to any offer in the same way, whatever is
the history of play until time tA. Of course, any stationary SPE is a Markov
SPE, but the reverse is not true: there can exist a Markov SPE which is not
a stationary SPE.
6.6 Endogenously Determined Inside Options 161

Consider an arbitrary Markov SPE, and let sfA and slB respectively de-
note the actions that players A and B take at time tA. It is straightforward
to see that the action profile (s^,s#) is necessarily a Nash equilibrium of
the disagreement game Q\ for otherwise, some player can benefit from a uni-
lateral deviation to an alternative action when playing Q at time tA. This
implies that if Q has a unique Nash equilibrium, then there does not exist a
Markov SPE which is not a stationary SPE. However, if Q has two or more
Nash equilibria, then there exist many Markov SPE which are not stationary
SPE; in such a Markov SPE the players play different Nash equilibria of Q at
different times. The following proposition characterizes the set of Markov
subgame perfect equilibria that differ from the set of stationary subgame
perfect equilibria.

Proposition 6.6. (i) If the disagreement game Q has a unique Nash equi-
librium, then the model with endogenously determined inside options and
discounting does not have a Markov SPE that is different from the unique
stationary SPE.
(ii) If the disagreement game Q has N (where N > 2) Nash equilibria
and if rA = TB = r, then the model with endogenously determined inside
options and discounting has an infinite number of Markov SPE (that differ
from the N stationary SPE) and are characterized as follows. Let the set
of Nash equilibrium action profiles of the disagreement game Q be denoted
by J\f. For any (infinite) sequence of action profiles (s1) such that for each
t = 0,1, 2 , . . . , sf G N', and such that for some tf ^ t", s1 ^ sl , the unique
subgame perfect equilibrium in which the action profile chosen at time tA
(t = 0,1, 2 , . . . ) is sf is a Markov subgame perfect equilibrium. In this SPE
player A's equilibrium offer x\ at time 0 is accepted by player B, where

A
1+6

Proof. Proposition 6.6(i) follows from the argument that precedes it, and
Proposition 6.6(ii) is proven in the Appendix. 15 •
15
It should be noted that unlike Propositions 6.5 and 6.6(i), Proposition 6.6(ii) requires
that the players' discount rates are identical. The reason for this is explained in the
proof to Proposition 6.6(ii). It should also be noted that Proposition 6.6(ii) states that
the unique subgame perfect equilibrium in which the action profile chosen at time tA
162 Inside Options

Since N > 2, there exists an infinite number of (infinite) sequences of NE

action profiles (s1) as defined in Proposition 6.6(ii); and hence, there exists
an infinite number Markov SPE. Any pair of such Markov SPE differ from
each other, because there will exist a t1 such that the action profiles chosen
at time t'A in these two equilibria will differ. However, it should be noted
that some of these Markov SPE will be payoff equivalent, because player A's
equilibrium offer x®A at time 0 — which is accepted by player B — does not
depend on the entire (infinite) sequence of payoff pairs (gis1)) associated
with the infinite sequence of NE action profiles (s*); as is evident from the
expression for x^, x^ does not depend on player z's inside option when her
offer is rejected (that is, it does not depend on gj^{s2t) and gB(s2t+1) for all
£ = 0,1,2,...).
In any Markov SPE (when N > 2) the payoffs to players A and B are
respectively x\ and n — x\. An important thing to notice, as mentioned
above, is that x®A does not depend on the entire (infinite) sequence of payoff
pairs {g{st)) associated with the infinite sequence of NE action profiles (s1).
This means that player i's 'bargaining power' (as reflected in her share of the
cake in any Markov SPE) is: (i) strictly increasing in the inside options that
she obtains after she rejects offers made by player j (j ^ z), (ii) unaffected
by the inside options that she obtains after her offers are rejected by player
j (j ^z f)7 (Hi) strictly decreasing in the inside options that player j obtains
after she rejects the offers made by player i, and (iv) unaffected by the inside
options that player j obtains after player i rejects offers made by player j . 1 6

Example 6.1. Consider the bargaining situation in which the disagreement

(t = 0,1, 2,...) is s* is a Markov subgame perfect equilibrium — although in this section
it is only required to characterize the unique Markov SPE in which the action profile
chosen at time tA (t = 0,1, 2,...) is s*. However, this stronger result drops out naturally
from the required arguments. It may be noted that this stronger result implies that in any
non-Markov SPE (which are the subject of discussion in the next section), it is necessary
that at some time (and for some history) the action profile chosen in Q is not a Nash
equilibrium.
16
The intuition for these observations is straightforward, and further illustrates a key
insight about bargaining processes (as obtained in other contexts as well), which can be
put in the following general manner: a player's 'bargaining power' is independent of what
happens to her after her offers are rejected. This is because the decision to accept or
reject an offer is made by her opponent, who is only influenced by what happens to her
if she rejects an offer (and not by what happens to the proposer if she were to reject the
proposer's offer).
6.6 Endogenously Determined Inside Options 163

game Q is the 'Battle of Sexes' game: Romeo (player R) and Juliet (player
J) have two possible actions each, namely, 'Go to the Ballet' (b) and 'Go
to the Big Fight at the Colosseum' (/), and Table 6.1 depicts the payoff
pairs (##(., .),gj(.,.)) for each possible action profile. Assume that n = 6/r
and that the players have a common discount rate r > 0. Without loss
of generality, assume that Romeo makes offers at times £A where t is even
(i.e., t — 0, 2,4,...) and Juliet makes offers at times tA where t is odd (i.e.,

Juliet
b f

2,3 0,0

Romeo
0,0 3,2

Table 6.1: The Battle of Sexes game.

This disagreement game has two Nash equilibria in pure strategies: s(l) =
(6, b) and s(2) = (/, / ) . In the former Nash equilibrium g(l) = (2, 3), and in
the latter Nash equilibrium g{2) = (3,2). Hence, applying Proposition 6.5,
there are two pure-strategy stationary SPE, and in both of them agreement
is reached immediately, at time 0. In the stationary SPE in which the inside
option point is g(l) = (2,3), the limiting (as A —» 0) SPE payoffs to R
and J are respectively 5/2r and 7/2r, while in the other stationary SPE (in
which the inside option point is g(2) — (3, 2)), the limiting (as A —» 0) SPE
payoffs to R and J are respectively 7/2r and 5/2r.
The disagreement game also has a Nash equilibrium in mixed strategies:
Romeo chooses b with probability 2/5, and Juliet chooses b with probability
3/5. In this mixed NE the inside option point is (6/5, 6/5). Hence, it follows
from Proposition 6.5 that there exists a mixed-strategy stationary SPE, in
which the limiting (as A —> 0) SPE payoff to each player is 3/r.
I now characterize two pure-strategy Markov SPE, which are not these
164 Inside Options

stationary equilibria. Applying Proposition 6.6(ii), in the SPE in which at

times £A where £ is even the players play the Nash equilibrium s(k) (k =
1, 2) and at times £A where £ is odd the players play the Nash equilibrium
s(l) (/ =£fc),agreement is reached immediately at time 0 with player i?'s
equilibrium share being

o_ 6 SgR(l) _ _gj{k)
+ 6)r (1 + 6)r'

Hence, in the limit, as A —> 0, the SPE payoffs to R and J (namely x°R and
7T — x° R) respectively converge to

r6 _ gR(l) _ gj(k)i gj(k) 1 r6 _ gR(l) _ gj(k)i

2 lr
Lr r r \ r 2 Lr r r \

Hence, if k = 1 and 1 = 2 — which means that after J rejects offers

made by R the players' inside options are determined by the pair of flow
rates g(l) = (2, 3), while after R rejects offers made by J the players' inside
options are determined by the pair of flow rates g(2) = (3, 2) — then the
limiting (as A —> 0) SPE payoffs to R and J are identical and equal 3/r.
One the other hand, if k = 2 and 1 = 1 — which means that after J rejects
offers made by R the players' inside options are determined by the pair of
flow rates g{2) = (3, 2), while after R rejects offers made by J the players'
inside options are determined by the pair of flow rates g(l) = (2, 3) — then
the limiting (as A —> 0) SPE payoffs to R and J are identical and equal
3/r. Thus, although the players' limiting payoffs in these two Markov SPE
are identical, these two equilibria differ according to the Nash equilibrium
played at each time tA (£ = 0,1,2,...). Notice that relative to the players'
payoffs in each of the pure-strategy stationary SPE described above, these
two Markov SPE make one player better off and the other player worse
off. On the other hand, the players' payoffs in these two Markov SPE are
identical to their payoffs in the mixed-strategy stationary SPE. 17
17
It should be noted — as indicated in Proposition 6.6(ii) — that there exists infinitely
many other Markov SPE.
6.6 Endogenously Determined Inside Options 165

6.6.3 Uniqueness of SPE and Non-Markov Equilibria

In a non-Markov SPE a player's strategy is not necessarily independent of

history and time: that is, at each time tA (t = 0,1,2,...) each player's
action in the disagreement game Q, and (depending on whether she has
to make or respond to an offer at time tA) her offer and response to an
offer may depend on the history of play until time tA. In this section I
investigate the possible existence of non-Markov SPE. As has been noted
above, Proposition 6.6(ii) implies that any SPE in which the players' chosen
action profile when playing Q is always a Nash equilibrium of Q is necessarily
a Markov SPE. Hence, in a non-Markov SPE it is necessary that at some
time (and for some history) the action profile chosen in Q is not a Nash
equilibrium of Q.
In order to investigate the potential existence of such non-Markov SPE,
I first derive a useful property of the set of all SPE payoffs. This property
enables us to state conditions under which there exists an essentially unique
SPE — under such conditions, non-Markov SPE do not exist.

On the Uniqueness of the SPE

Given the underlying stationarity in the strategic structure of the bargaining

model, the sets of subgame perfect equilibria of any two subgames beginning
with player i's offer are identical. This implies that the set of SPE payoffs
to player i in any subgame beginning with her offer is uniquely defined18 —
which is denoted by G^.19 Let rrii and Mi respectively denote the infimum
and supremum of G{. The useful property referred to above, which is stated
in Lemma 6.3 below, is a lower bound on rrii and an upper bound on Mi. I
shall interpret these bounds after their formal statement in Lemma 6.3. It
should be noted that unlike Proposition 6.6(ii), this lemma is valid for any
TA and rB.

18
That is, this set of SPE payoffs depends neither on the time at which such a subgame
begins nor on the history of play that precedes such a subgame.
19
Formally, ui G Gi if and only if there exists a SPE in a subgame beginning with player
Vs offer that gives her a payoff of ui.
166 Inside Options

Lemma 6.3.

> =A i _ i ~ SB
TT —
^A VB
~ rA 1- rA rB

VA _ ^
TT —
~ rA 1 - 8 A6 B rA rB

. vB 1-8A VA _^B
TT —
m B > r-B + 1 - 8A6B rA rB

M <r VB , 1-6A (TT — VB

r5 1 - 0^0^ y rA rB

where vA= inf sup gA(s)

G5 y

£A = sup gA(s) - -yA sup 5B(SA, S'B) ~ 9B{S) and

ses\ \sfBeSB JJ
vB = sup gB(s) - j B sup
ses\ \s'AeSA

A = 8A(l-6B)rA/SB(l-6A)rB and^B = SB(1 -8A)rB/8A(l -8B)rA.

Proof. In the Appendix. •

By the definition of v_^ the worst possible payoff that player i can obtain
from the disagreement game between two consecutive offers in any SPE is
greater than or equal to (1 — ^ ) ^ / ^ - 2 0 In the interpretation to follow of the
bounds on mi and M^ I assume that {v_^Vj) is a Nash equilibrium payoff
pair of the disagreement game — that is, there exists a NE strategy pair s~
in Q such that gi^s) = y_{ and gj(s) = Vj. It follows from Proposition 6.5
that in the unique stationary SPE in which the action profile chosen in Q
is always 5, player i's SPE payoff in any subgame beginning with her offer
equals the lower bound on ra^, and player j ' s SPE payoff in any subgame
beginning with her offer equals the upper bound on Mj. Hence, for each
i (i = A,B), the lower bound on mi (resp., the upper bound on Mi) may
20
That is, in any SPE player j cannot force player i's inside option to be lower than
(1 — 8i)v_i/ri. It should be noted that v{ is player i's minimax payoff in Q.
6.6 Endogenously Determined Inside Options 167

be interpreted as player z's worst (resp., best) SPE payoff in any subgame
beginning with her offer.

Lemma 6.4. For any Nash equilibrium ^s of the disagreement game, VA >
9A(S) > V.A an& VB > gB^s) > v_B. Furthermore, for each i (i = A,B),

Proof. For each s G S, let

W(s) = gA{s) -~fA\ sup gB(sA,sB) -

where 7^ is defined in Lemma 6.3. Furthermore, let *s denote an arbitrary

NE of Q. Since, by the definition of a Nash equilibrium,

gB(s) = max gB(sA,sB)

it follows that W(s) = gA(s)- Hence, by the definition of VA — as stated in

Lemma 6.3 — it follows that VA > <M(S)- It is a standard result in game
theory — and, in fact, trivial to verify — that gA^S) is greater than or equal
to player A's minimax payoff, and thus, <M(S) > 21A- The above results
imply — in combination with the assumption that Q has at least one Nash
equilibrium — that VA > HA- A symmetric argument, with the roles of A
and B reversed, completes the proof. •

Lemmas 6.3 and 6.4 imply that if for each i = A, B, y_i = ^i-, then
Mi — rrii = x*, where
dA 9B\ A (a ox
7T and (6.8)
rA rBJ
* gB , 1 - $A / gA gB\ ,ark,
X = + 7T , (6.9)
rA rBJ
where (JJA^B) denotes the uniquely defined Nash equilibrium payoff pair of
the disagreement game. 21 Since Mi = rrii (i = A, B), it is straightforward to
argue that in any SPE of any subgame beginning with player i's offer, player
21
Lemma 6.4 implies that if for each i = A, B, v.% — v%^ then all Nash equilibria are
payoff-equivalent.
168 Inside Options

i's equilibrium offer is x*, and it is accepted by player j . This implies that
when playing the disagreement game at any time £A, the players always
choose some Nash equilibrium action profile of £/, which, if two or more
exist, are payoff equivalent. This, in turn, implies that player j accepts an
offer Xi if and only if TT — xi > (1 — 8j)dj/rj + fijx*j — that is, if and only if
%i < %*• I have thus established the following proposition.
Proposition 6.7. If for each i = A,B, v_i = V{, where v^ and V{ are defined
in Lemma 6.3, then the model with endogenously determined inside options
and discounting has (essentially) a unique subgame perfect equilibrium. In
any SPE player i (i — A, B) always offers x\ and always accepts an offer
x
j U ¥" 0 tf and onty if xj < x*ji where x*A and x*B are defined in (6.8)
and (6.9). Furthermore, when playing the disagreement game, the players7
chosen action profile is a Nash equilibrium of Q, where all Nash equilibria
of Q are pay off-equivalent.
Proof. The proposition follows from the argument that precedes it. •

It should be noted that any SPE described in Proposition 6.7 is a Markov

SPE; that is, under the hypothesis of Proposition 6.7 there does not exist a
non-Markov SPE. Furthermore, since all Markov SPE described in Proposi-
tion 6.7 are payoff-equivalent, it is valid for any TA and rs — unlike Propo-
sition 6.6(ii), which is valid if rA = ^B- The following example illustrates
this proposition.
Example 6.2. Consider the bargaining situation in which the disagreement
game Q is the 'Prisoners' Dilemma' game: player A and player B have two
possible actions each, namely, c and rf, and Table 6.2 depicts the payoff pairs
(gA("> -)J9B(-, •)) f°r each possible action profile. Assume that vr = 4/r and
that the players have a common discount rate r > 0. This disagreement
game has a unique Nash equilibrium, namely, 's = (d,d). In this unique
Nash equilibrium <? = (0,0). Hence, Proposition 6.5 implies that there is a
unique stationary SPE, in which the inside option point is <? = (0,0), and
the limiting (as A —> 0) SPE payoffs to players A and B are respectively
2/r and 2/r. Proposition 6.6(i) implies that there does not exist a Markov
SPE that is different from this unique stationary SPE. It is straightforward
to show that for each i = A,B, Vi = 0 and v_{ = 0. Hence, it follows from
Proposition 6.7, that the unique stationary SPE is the unique SPE of the
model — there does not exist a non-Markov SPE.
6.6 Endogenously Determined Inside Options 169

Player B
c d

0.4,0.4 -1.1,0.9
Player A
0.9,-1.1 0,0

Table 6.2: The Prisoners' Dilemma game.

Existence of non-Markov SPE

I now consider disagreement games that fail to satisfy the hypothesis of
Proposition 6.7; that is, for some i (i = A or i = i?), vi > v_{. A key result
— stated in Lemma 6.5 below — establishes the existence of two ('extremal')
SPE: for each i = A,B, the lower bound on rrii (as stated in Lemma 6.3)
is sustained as a SPE payoff to player i in any subgame beginning with her
offer. Thus, the lower bound on rrii is player i's worst SPE payoff in any
subgame beginning with her offer.

Lemma 6.5. Assume that TA = ^B — r. For each i (i = A, B) there exists

a Ai > 0 such that for any A < Ai there exists a SPE in any subgame
beginning with player i ;s offer such that player i 's equilibrium payoff equals

1 v_i
1 +6 r

Proof. In the Appendix. •

In the limit, as A —» 0, player i's worst SPE payoff in any subgame
beginning with her offer u^ converges to y., where y. is defined in Proposition
6.8 below. It should be noted that if for some i (i = A or i = B) v\ > v_{,
then VA + yB < /JT- I n o w provide an informal argument which shows that (in
the limit, as A —» 0) there exists a multiplicity of SPE. Consider any path
of play in which agreement is reached at time tA (t = 0,1, 2, 3,...) on a
partition (y^, ys) of the cake such that yi>y.(i = A,B), where yi denotes
170 Inside Options

player i's share. 22 Player i's payoff from this path of play in the limit as
A —» 0 is i/i. It immediately follows that in the limit as A —» 0, the proposed
path of play can be supported as a SPE path of play by reverting to one of
the two extremal SPE when a player unilaterally deviates from the proposed
path of play.23 This informal argument can be formalized, and shown to be
valid for A sufficiently small. Hence, I obtain the following result.

Proposition 6.8. Assume that TA = TB = r, and that for some i (i = A

or i = B), Vi > v_{. There exists a A such that for any A < A there exists a
continuum of SPE in the model with endogenously determined inside options
and discounting. For any partition (yA,VB) of the cake such that yi > y.
(i = A, B), where yi denotes player i 's share and

and any time tA (t = 0,1, 2, 3 , . . . ) , there exists a SPE such that in the limit
as A —> 0 agreement is reached at time tA on the partition (yAiUB)-

It should be noted that if the disagreement game satisfies the hypothesis

of Proposition 6.8, then (for any A sufficiently small) the bargaining model
with inside options and discounting has a multiplicity of SPE — notwith-
standing the possibility that the disagreement game has a unique Nash equi-
librium. Furthermore, there exists a continuum of Pareto-inefficient SPE.

6.7 An Application to Wage Renegotiations

A firm has discovered a new technology that will enable it to produce TT*
units of output per worker per unit time, where TT* > 1. With its existing
technology it produces one unit of output per worker per unit time. The
current wage per unit time is a, where 0 < a < 1, and the current profit per
22
Since, as shown above, yA + y < TT, there exists a continuum of such partitions.
23
If player i deviates from the proposed path of play, then play proceeds according to
the SPE that gives player i her worst SPE payoff in any subgame beginning with her offer.
Hence, in the limit as A —> 0, player i has no incentive to deviate from the proposed path
of play — because her (limiting) payoff from such a unilateral deviation is y., which (by
hypothesis) does not exceed her (limiting) payoff from the path of play, namely, yi.
6.7 An Application to Wage Renegotiations 171

worker per unit time is 1 — a. However, before the new technology will be
used, the firm F and its union U have to renegotiate the wage rate. Since
7T* > 1, it is mutually beneficial to use the new technology.
While bargaining over the new wage rate w, where w G [0,TT*], the
union can decide whether or not to go on strike. The renegotiation process
is modelled by the bargaining game described in Section 6.6 in which the
players have a common discount rate r, TT = vr*/r, and the disagreement
game is as follows. The union's set of actions is {0,1}, where 0 means that
the union goes on strike for A units of time, and 1 that it does not go on
strike. The firm has no action to take in the disagreement game. If the
union goes on strike for A units of time, then no output is produced and no
wage is paid, and hence, ^F(O) = <ft/(0) = 0. On the other hand, if the union
does not go on strike during the time interval A, then A units of output
is produced, and the payoffs to the firm and the union are respectively
(1 — <5)gi?(l)/r and (1 — 8)gu{l)/r, where # F ( 1 ) = 1 — a and <ft/(l) = a. 24
Thus, while the parties temporarily disagree and the union does not go on
strike, the old technology is used to produce output, and each worker is
employed at the old wage rate.

6.7.1 Multiple Pareto-Efficient Equilibria

The disagreement game — which is a single-person decision problem — has
a unique Nash equilibrium, in which the union does not go on strike. Hence,
the unique NE payoff pair of the disagreement game is 'g = (a, 1 — a).
It follows from Propositions 6.5 and 6.6(i) that there exists a unique
stationary SPE, which is the unique Markov SPE. In the unique (limiting,
as A —» 0) stationary SPE agreement is reached at time 0 on wage rate w*,
where

w* = a + -(TT* - 1).

The firm's profit per worker per unit time is

Notice that this unique (limiting) stationary SPE is Pareto efficient: agree-
ment is reached at time 0, and the union does not go on strike. Furthermore,
It is assumed that the union's objective is to maximize the wage rate.
172 Inside Options

the benefit of the new technology relative to the old technology — as cap-
tured by the increase in the output per worker per unit time, namely, TT* — 1
— is split equally between the firm and a worker.
I now consider the potential existence of Pareto efficient non-Markov
SPE. It is straightforward to show that the minimax payoffs to the firm and
the union in the disagreement game are respectively zero and a. In the no-
tation of Lemma 6.3, y_F = 0 and V_JJ = OL. Furthermore, it is straightforward
to show that vp = 1 — OL and VJJ = a. Hence, since vp > HF, Proposition 6.7
is not applicable. Indeed, it follows from Proposition 6.8 that there exists
a A such that for any A < A there exists a continuum of Pareto-efHcient
(non-Markov) SPE. In the limit, as A —> 0, any partition (yF,yu) °f the
cake agreed to at time 0 such that yi > y. {i = F, U) can be sustained by a
SPE, where

l/V a\ 1 l/V a (l-a)\

F
- 2 y r rJ -u 2\r r r J
Letting w_ denote the wage rate associated with the (limiting) SPE in which
the partition is such that yjj — y , it follows that

w_ = a + ~(TT* — 1).

Furthermore, letting w denote the wage rate associated with the (limiting)
SPE in which the partition is such that yp = yF, it follows that

W = -(TT +a).

Hence, in the limit as A —> 0, any wage rate w such that w_<w<w agreed
to at time 0 can be sustained in a SPE.
Notice that w* = w, and, hence, the unique (limiting) stationary SPE —
in which the union never goes on strike — sustains (in the limit as A —>• 0)
the worst (limiting) SPE wage (payoff) to the union. The pair of strategies
that sustain the worst SPE payoff to the firm are described in Table 6.3, with
the initial state being sp. Notice that in state sp the union always goes on
strike after the firm rejects any offer, and it does not ever go on strike after
it rejects any offer made by the firm. The idea is to maximize the 'cost of
rejection' to the firm and, at the same time, minimize the 'cost of rejection'
6.7 An Application to Wage Renegotiations 173

to the union. Since striking is costly to the union, it needs to be provided

with an incentive to go on strike. As shown in Table 6.3, if the union fails
to go on strike, then the state switches to su where the union obtains its
worst (limiting) SPE wage. Indeed, when A is small, the loss from going on
strike — which is approximately equal to aA — is outweighed by the gain
from going on strike — which equals (w — w)/r (i.e., (1 — a)/2r).

state sp state su
offer w UL
Firm
accept w<w w < yi

offer w ui

Union accept w>w w > ui

strike? 'yes' after the no
firm rejects any
offer; 'no' after
the union rejects
any offer
transitions switch to state su absorbing
if the union does
not go on strike af-
ter the firm rejects
an offer

Table 6.3: The two 'extremal' SPE of the wage renegotiations model; if play begins in
state Si (i = F,U), then player i obtains her worst SPE payoff.

6.7.2 Equilibria with Strikes

It is easy to construct SPE in which agreement is reached after the union
goes on strike for some time. Let w be any wage rate such that w_< w < w,
and T = tA where t = {1, 2, 3,4,... }. Consider the following path of play.
At any time t < T: (i) when a player has to make an offer, she asks for
the whole cake (that is, the union offers a wage rate of TT* and the firm
offers a wage rate of zero), (ii) when a player has to respond to an offer,
174 Inside Options

she rejects any offer, and (iii) the union always goes on strike. Then at
time T the wage rate offered is w and it is accepted. This path of play
can be supported (provided A is sufficiently small) as a SPE path of play by
reverting to one of the two extremal SPE when a player unilaterally deviates
from this path of play — and needs to be punished. Thus, if the firm (resp.,
the union) deviates from this path of play, then immediately play proceeds
according to the SPE described in Table 6.3 with the initial state being sp
(resp., su). Hence, in this model with perfect information it is possible to
rationalize the occurrence of strikes.

6.8 Appendix: Proofs

Proof of Proposition 6.6(ii)

The proof begins by not assuming that TA — ^B-, SO that I can show exactly
why one needs to make this assumption. Fix an arbitrary (infinite) sequence
of Nash equilibrium action profiles (s1) as specified in the proposition. The
strategy of the proof is as follows. Using an argument similar to that con-
tained in Section 3.2.1, I first establish that there exists a unique Markov
SPE in which the action profile chosen at time £A (t — 0,1, 2 , . . . ) is sl and
in which any equilibrium offer is accepted. I then, using an argument similar
to that contained in Section 3.2.2, show that there does not exist any other
SPE in which the action profile chosen at time £A (t = 0,1, 2 , . . . ) is sf.
This would then establish Proposition 6.6(ii).
Consider a Markov SPE in which the action profile chosen at time t A (t =
0,1, 2 , . . . ) is sf and in which any equilibrium offer is accepted. Let x2j[ de-
note the equilibrium offer of player A at time 2tA (t = 0,1, 2 , . . . ) and x2^+1
the equilibrium offer of player B at time (2t + 1) A (t = 0,1, 2 , . . . ) . Since
any equilibrium offer is accepted, it follows that at time 2tA (t = 0,1, 2 , . . . )
player B accepts any offer y2^ such that TT — y2^ > (1 —
? a n d rejects any offer y2^ such that TT - y% < (1 -
B', where the right-hand side of this inequality is player I?'s equilib-
rium payoff from rejecting any offer at time 2tA. Furthermore, since the
equilibrium offer is accepted, n — x2^ > (1 — $B)gB(s2t)/rB + SBX2^1. In
fact, TT — x2^ — (1 — 6B)gB(s2t)/rB + <5B2?£+1; for otherwise player A can
profitably deviate by offering y2^ > x2^. Since a symmetric argument estab-
6.8 Appendix: Proofs 175

lishes similar results for player A, it follows that the sequence (x

of equilibrium offers must be such that for each £ = 0 , 1 , 2 , . . . ,

and (6.10)

. (6.11)

Furthermore, since each player can always adopt the strategy in which she
asks for the whole cake and rejects all offers, it follows that for each k =
0,1,2,...

xf > ifUs*)) (6.12)

7r-4fc>/f«s4)) (6.13)
«*» (6.14)
t
, )), (6.15)

where ^((s1)) is player i's payoff (discounted to time kA) from time fcA
(fc = 0 , l , 2 , . . . ) onwards if the players perpetually disagree. That is

(6.16)
Ti
t=k

Notice that Assumption 6.1 implies that for any fc, if inequality 6.12 (resp.,
6.14) is satisfied, then inequality 6.13 (resp., 6.15) is also satisfied. Hence, it
follows that the Markov SPE being considered here exists only if there exists
a sequence ( a ^ t , a ^ + 1 ) £ 0 that satisfies (6.10), (6.11), (6.12) and (6.14).25
In Claim 6.1 below I show that there exists a unique sequence (a^, x2^1)^
that satisfies (6.10) and (6.11). This result is valid for any VA and TB- Then,
in Claim 6.2 I show that if TA = ^B, then this unique sequence also satisfies
(6.12) and (6.14). Hence, it is in establishing Claim 6.2 that I require the
assumption TA — TB-
Claims 6.1 and 6.2 thus imply that if r^ = r^ = T, then there exists at
most a unique Markov SPE in which the action profile chosen at time tA
(t = 0,1, 2 , . . . ) is s1 and in which any equilibrium offer is accepted. In this
25
It should be noted that any sequence that satisfies (6.10)-(6.15) is 'feasible': that is,
for any t = 0,1, 2, 3 , . . . , x™ G [0, n] and x ^ + 1 e [0, IT] — this is implied by (6.12)-(6.15),
because for any k = 0,1, 2, 3 , . . . , ^((s*)) > 0.
176 Inside Options

Markov SPE the sequence of equilibrium offers are defined by the sequence
{x2£, x2g+ )£^o> a n d when a player has to respond to any offer at any time tA
she accepts an offer if and only if it is less than or equal to the equilibrium
offer at time tA. It follows immediately from the arguments above that this
Markov SPE exists — that is, this pair of strategies is a subgame perfect
equilibrium.
I now show that there does not exist another SPE in which the action
profile chosen at time tA (t = 0,1,2,...) is sf. The argument is similar
to that contained in Section 3.2.2. Let M% and m2^ (t = 0,1,2,...) re-
spectively denote the supremum and infimum of the set of payoffs to player
A obtainable in any SPE in which the action profile chosen at time tA
(t — 0,1, 2 , . . . ) is sf of any subgame at time 2£A. Similarly, let M^+1 and
m
£? + 1 (^ = 0,1, 2 , . . . ) respectively denote the supremum and infimum of
the set of all payoffs to player B obtainable in any SPE in which the ac-
tion profile chosen at time tA (t = 0,1, 2 , . . . ) is s* of any subgame at time
(2t + 1)A. Through a slight modification to the arguments in Section 3.2.2,
it can be shown that for each t = 0,1, 2 , . . . , (Mjf, m g + 1 ) = (xf, x ^ + 1 ) and
( m ^ , M f + 1 ) = ( x ^ , x | + 1 ) , where the sequence ( x ^ , x | + 1 ) satisfies (6.10)
and (6.II). 26 Furthermore, for each k = 0 , 1 , 2 , . . . , m2jf = x2jf satisfies
(6.12) and (6.13), and mf+1 = x%+1 satisfies (6.14) and (6.15). From
Claims 6.1 and 6.2 it follows that for each t = 0,1, 2 , . . . , M2£ = m2£ and
M ^ + 1 = m2g+1. Hence, the payoffs to the players in any SPE in which the
action profile chosen at time tA (t = 0,1, 2 , . . . ) is sf are uniquely denned.
It is then trivial to show, using an argument similar to that used in Section
3.2.2, that in any such SPE the equilibrium offers are accepted and, more-
over, that there is only one such SPE — which is the unique Markov SPE
characterized above.

Claim 6.1. There exists a unique sequence (x^, x ^ + 1 ) ^ 0 that satisfies (6.10)
and (6.11).
Proof. After substituting for x2^+1 in (6.10) by using (6.11), and then rear-
ranging, it follows that x2l = /(2t, 2t +1) + 6A6Bx2l+2, where /(2t, 2t +1) =
26
A key modification involves replacing the terms 6B MB and feme respectively with
gB(s2t)(l-8B)/rB+6BMlt+1 and gB(s2t+1)(l-6B)/rB+6BmBt+1, which are respectively
the maximum (supremum, to be precise) and minimum (infimum, to be precise) payoffs
that player B obtains in any SPE in which the action profile chosen at time tA (t =
0,1, 2,...) is s* if she rejects any offer at time 2£A. Similarly, for player A.
6.8 Appendix: Proofs 177

TT(1 - 6B) + 6B(1 ~ ^A)9A{s2t+l)/rA - (1 - 6B)gB(s2t)/rB. It thus follows

that

After substituting for /(2z, 2z + 1) and simplifying, it follows that for each
t = 0,1,2,...

X2t = <l
1 -
oo
( 2 1
° (^)/r B ] • (6.17)

Finally, x ^ + 1 is derived after substituting for x2^2 in (6.11). •

Claim 6.2. / / TA = TB, ^ ^ n i/ie unique sequence that satisfies (6.10)-

(6.11) also satisfies (6.12) and (6.14)-
Proof. Before I provide an example that shows that this claim need not be
valid if TA^TB-, let us prove it when TA = TB = r. Since Assumption 6.1 im-
plies that Qi is bounded from above, the supremum of the sum gA(s)+9B{s)
over all s = (s^, sB) E SA X SB exists, which is denoted by g. Assumption
6.1 implies that n > g/r27 Using (6.16) and (6.17), it is straightforward to
show that if TA — ^B — T, then for each k = 0,1, 2, 3 , . . .

xf = lf((St)) + Y^r6- -^Y,S2(Z~k)i9A(s2z)+9B(s2z)}- (6-18)

z—k

By definition

l — o 2{z k)
~ \9A(s2z) + 9B(S2Z)]. (6.19)

The left-hand side of inequality 6.19 equals g/r(l + 6). Hence, since TT > g/r,
the right-hand side of inequality 6.19 is less than or equal to TT/(1 + <S), which
thus implies — using (6.18) — that inequality 6.12 is satisfied. Through a
symmetric argument it can be shown that inequality 6.14 is satisfied. 28 •
27
Suppose, to the contrary, that g > rix. This implies that there exists an s G SA X SB
such that g > gA{s) + QB(S) > rn, which contradicts Assumption 6.1.
28
I now provide an example to show that if TA ^ TB, then this claim need not be valid.
178 Inside Options

Proof of L e m m a 6.3

I shall establish the lower bound on TUA and the upper bound on MA]
through a symmetric argument, with the roles of A and B reversed, the
lower bound on TUB and the upper on MB can be established.
Let HB denote the set of all SPE payoffs to player B in any subgame
beginning with the play of the disagreement game Q, after player B has
rejected an offer made by A. Claim 6.3 below establishes that for any KB G
HB-> hs < supseS FB(S), where for each s G S

(1 6 9B
FB(s) = ^ ^ +6B(*- vA[l ~ 6A)/rA - 6AmA

sup gA(s'A, sB) - gA(s) ), (6.20)

s'AesA
with y_A as specified in the lemma. Hence, in any SPE of any subgame
beginning with player A's offer, player B accepts any offer XA such that
7T — xA > FB , where

FB = supFB(s) = SB(TT -VA(1 - 6A)/rA - <5A^A) + (1 -

v J
ses
(6.21)
with VB as specified in the lemma. This implies that TTIA > vr — FB- Thus,
after substituting for FB in this latter inequality, using (6.21), and then
rearranging, the lower bound for TTIA as stated in the lemma is obtained.
I now use this result to establish the upper bound on MA. I claim that
MA < 7T — (1 — SB)v_B/rB — SBTTIB- Suppose, to the contrary, that this
inequality is false. This implies that there exists an SPE in any subgame
beginning with player A's offer in which player A's payoff u*A is such that
MA > u*A > 7T — (1 — SB)V.B/TB — SBTUB- Since (by Assumption 6.1 and

Consider the model with the following disagreement game. The players' strategy sets are as
follows: 5A = {U, D} and SB = {L, R}. The payoffs are as follows: gA(U, L) = gB(U, L) =
1, gA(D,R) = gB(D,R) = 0, gA(D,L) = gB(U,R) = 0.2 and gA(U,R) = gB(D,L) = 0.
Finally, TT = 1/rA + l / ^ s . This disgreement game has two Nash equilibria, namely:
s (l) = (u, L) and s(2) = (D, R). Consider the Markov SPE in which at each time tA the
players play s(l) when t is even (t — 0, 2,4,...), and s(2) when t is odd. Applying (6.17),
it follows that if such a Markov SPE exists then in equilibrium player A offer's xA at time
0 which is accepted by player B, where xA = (1 — SB)/(1 — SA8B)rAj which — if rA > TB
— is strictly less than /£((«*» = 1/(1 + $A)rA.
6.8 Appendix: Proofs 179

by the fact that the cake is of size TT) U*A + u*B < n (where u*B denotes
player £Ts payoff in this SPE), it follows that (1 — SB)V.B/TB + ^B^B > v*^,
which is a contradiction, because the left-hand side of this inequality is the
worst possible payoff that player B can obtain by rejecting any offer made
by player A — since by the definition of y_B, the worst possible payoff that
player B can obtain from the disagreement game between two consecutive
offers in any SPE is greater than or equal to (1 — SB)V.B/TB' This then
establishes the upper bound on MA-

Claim 6.3. For any KB G HB, hs < supseS FB(s), where FB(s) is stated
in (6.20).

Proof. Suppose, to the contrary, that there exists an hs G HB such that

KB > supsE£ FB(s). Let s* G S denote the SPE action profile chosen the
first time the disagreement game Q is played in this subgame, and (u*Alu*B)
denotes the SPE payoff pair at the beginning of the subsequent subgame
— which begins with player i?'s turn to make an offer. This means that
KB = (1 — <5# )##(<§*)/re + SBU*B. I shall show that hB < FB(s*), which
contradicts my supposition that hB > supsG^ FB(s).
Since player A should not have an incentive to deviate at the beginning
of the subgame under consideration, it must necessarily be the case that

{1-SA)9A(S*) , , * ^1~8A( ,, * x V c
+ 6AuA > sup gA(sA, sB) + 8AnA,
r
A rA y , J
where nA denotes the infimum of the set of SPE payoffs to player A in
any subgame beginning with player S's offer. Since nA > iiA(l — $A)/rA +
SAmA — because, by the definition of y_A as specified in the lemma, the
worst possible payoff that player A can obtain from the disagreement game
between two consecutive offers in any SPE is greater than or equal to ^ ( 1 —
SA)/^A — it thus follows that

(l-6A)gA(s*) t K „ ^ 1 - 6A ( (, M
r : + 6AuA > sup gA{sA, sB)
A rA \'
(6.22)

Since u*A + u*B < vr (which implies that 6Au*A < 6A(ir — u*B)), it follows from
180 Inside Options

(6.22) that

which implies that

(l-6B)gB(s*)

the desired contradiction. D

The Proof of Lemma 6.5

I shall establish the existence (by construction) of a SPE that supports player
A's worst SPE payoff; the existence of a SPE that supports player S's worst
SPE payoff follows from a symmetric construction (with the roles of A and
B reversed). Let w — {WA^B) and b — (bA,bB) denote the strategy pairs
of the disagreement game Q such that V_A ~ 9A(W) and29

- sup gA{sfA) bB) -

Letting ? denote an arbitrary NE of Q, I denote by a the unique stationary

SPE in which the action profile chosen in Q is always s*. Suppose that
30
(V-AIVB) + (?A,?B), where % = g^s) (i = A,B). Furthermore, define x*B

29
In order to keep the construction relatively straightforward, assume the existence of
such strategy pairs.
30
If (V_A,VB) = (§A, <7#), then the required result follows immediately from Proposition
6.5.
6.8 Appendix: Proofs 181

and x*B as follows31

7T 6VB 8V_A

X*B=X*B sup gA(sfA,bB) - gA(b)

It will be shown below (using the One-Shot Deviation property) that there
exists a A ^4 > 0 such that for any A < A^ the pair of non-Markovian
strategies described in Table 6.4 is a SPE.
First consider player A's action in Q. If she chooses bA after player B
rejects any offer, then her payoff is VA(PA) = (1 — S)gA(b)/r + 8(n — x*B).
If she instead chooses an alternative action SA (and thereafter conforms to
the strategy specified in Table 6.4), then her payoff is

Hence, a one-shot deviation by player A from bA after player B rejects any

offer is not profitable for player A. If player A chooses WA after she rejects
any offer, then her payoff is (1 — 6)y_A/r + 8uA, which — by the definition
of y_A — is less than or equal to her payoff from a one-shot deviation to an
alternative action.
Now consider player A's responses in state 2. If she accepts an offer
%B < %*B, then her payoff is TT — xB. If she instead rejects such an offer, then
her payoff is (1 — 8)y_A/r + 8uA = TT — x*B. Since x*B < x*B, it follows from
the above argument that player A's responses in state 1 are also immune to
one-shot deviations.
If player A offers xA = uA, then her payoff is uA. If player A instead
offers xA > uA, then her payoff is V A ( 6 ^ ) , which (as shown above) equals

sup gA(sfA,bB)

31
It should be noted that Assumption 6.1 implies that uA (which is defined in the
lemma), x*B and x#* are elements of the closed interval [0,TT] — and, hence, they are
feasible offers.
182 Inside Options

state 1 state 2

offer XA = uA XA =UiA

Player A accept x B < x B* XB <XB

action in Q WA after player A WA after player A

rejects any offer, rejects any offer,
and bA after player and bA after player
B rejects any offer B rejects any offer

offer x B = x B* xB =XB

Player B accept XA < U.A XA <UA

action in Q WB after player A WB after player A

rejects any offer, rejects any offer,
and bs after player and bs after player
B rejects any offer B rejects any offer

transitions (i) switch to the sta- (i) switch to the sta-

tionary SPE a if tionary SPE a if
player B either of- player B either of-
fers XB > xB*, or re- fers XB > xBi or re-
jects any offer XA < jects any offer x^ <
uAi or chooses an uA, or chooses an
action in Q differ- action in Q differ-
ent from that speci- ent from that spec-
fied above, and (ii) ified above, and (ii)
switch to state 2 switch to state 1 if
if player A does player A chooses bA
not choose bA after after player B re-
player B rejects any jects any offer
offer

Table 6.4: The 'extremal' SPE strategies which support player A's worst SPE payoff.
6.8 Appendix: Proofs 183

which, after substituting for x*Bl equals

sup gA{s'A,bB)\ +6

which is strictly less than uA if and only if (after substituting for uA and
rearranging)

vB 1 9B{b)
7T > h ~ sup
r r
which follows from Assumption 6.1.
Hence, a one-shot deviation from the offer xA = uA to an offer xA > uA
is not profitable for player A. Since player B accepts any offer such that
%A < uA, a one-shot deviation from the offer xA = uA to an offer xA < uA
is also not profitable for player A.
Now consider Player B's action in Q. If she chooses the action specified
in Table 6.4 — which is either WB or bs — then her payoff is greater than
or equal to

(6-23)

where g denotes player B's minimum (infimum, to be precise) payoff in Q.

If she deviates to an alternative action, then (since play will subsequently
switch to the stationary SPE a) her payoff is at most equal to

(6.24)

where gB denotes player I?'s maximum (supremum, to be precise) payoff in

Q. There exists A' > 0 such that for any A < A', (6.23) is strictly greater
than (6.24).32 Therefore, for any A < A' a one-shot deviation from either
wB or bs is not profitable for player B.
32
In the limit as A —> 0, both and n — uA converge to TT/2 + VB/T — v_A/r, which
is strictly greater than TT/2 + gs #A/V — because (JJA^B) ^ {V_A,VB) implies (using
Lemma 6.4) that either VB > #B or 'CJA > V.A- Consequently, since (6.23) and (6.24) are
continuous in A, there exists A ; > 0 such that for any A < A' (6.23) is strictly greater
than (6.24).
184 Inside Options

Now consider player i3's responses. If she accepts the offer XA = UA->
then her payoff is TT — uA> If she rejects this offer, then (since play switches
to the stationary SPE a) her payoff is
/ \ i

(6.25)

Lemma 6.4 implies that TT — uA > V. This, in turn, implies that for any
offer XA < UA-> n ~ XA > V. Hence, it is optimal for player B to accept any
offer XA such that XA < ILA- ^ player B rejects an offer XA > UA-> then her
payoff is (1 — <5)g#(6)/r + 6x*B, which equals (after substituting for x*B and
simplifying)

1+ 6 (l + 6)r (l + 6)r'
If she instead accepts such an offer, then her payoff is TT — XA-> which is
strictly less than V7, because Vr > TT — uA. Hence, it is optimal for player
B to reject any offer XA > ILA-
Now consider player S's offer in state 2. She does not benefit by instead
making an offer XB < x*B. If she offers XB > x*B) then (since play switches
to the stationary SPE a) player A rejects the offer because her payoff from
rejecting equals

> TT — XB > 7T —

Consequently, player B's payoff by offering XB > x% is stated in (6.25),

which is strictly less than x*B (given Assumption 6.1, and using Lemma 6.4).
Now consider player BJs offer in state 1. She does not benefit by instead
making an offer XB < X*B- If she offers XB > £#*, then play switches to
the stationary SPE a. Since x^* < x^, it follows from the above argument
that player A rejects the offer XB > x*B. Consequently, player 5's payoff by
offering XB > x*B is stated in (6.25), which is strictly less than x*B. Hence,
there exists A" > 0 such that for any A < A" the payoff in (6.25) is strictly
less than x^*.33
The required result follows by setting A^ = minjA 7 , A"}.
33
In the limit, as A —> 0, x*g — x*B —>• 0. Hence (since for any A > 0, x*B > V) there
exists a A " > 0 such that for any A < A", x*B* > V.
6.9 Notes 185

6.9 Notes
The application in Section 6.3.2 on soverign debt renegotiations is based
upon Bulow and Rogoff (1989), while the application in Section 6.5 on in-
trafamily allocation is based upon Lundberg and Pollak (1993). The model
studied in Section 6.6 is due to Busch and Wen (1995) — although the for-
malization and the analysis are slightly different, and some of our results are
valid also when the players have different discount factors. The application
studied in Section 6.7 is related to the work of Haller and Holden (1990) and
Fernandez and Glazer (1991). Holden (1994) shows that a unique SPE is
obtained if the union can commit at any time to strike in any future period
— in the limit as A —» 0, the unique SPE wage equals w. Chang (1995)
studies a model in which two countries are bargaining over the gains from
monetary union. Although conceptually the bargaining model is similar
to that of Section 6.6 — with the countries' inside options being deter-
mined in the absence of monetary union — the disagreement game is not a
simultaneous-move game.
 Procedures

7.1 Introduction

The procedure of bargaining is the structure of moves of the bargaining

process — it defines the rules of the bargaining game. The alternating-
offers procedure — which underlies the models studied in Chapters 3-6 —
is an example of a bargaining procedure. In Section 7.2 I describe (and
evaluate) several alternative procedures of bargaining that differ from each
other on the key procedural matter of 'who makes offers and when'. A main
message of this section is that this aspect of procedure — who makes offers
and when — can have a significant impact on the bargaining outcome.
After an offer is accepted the proposer of that offer may have the option
to retract the offer, in which case the players resume bargaining. This proce-
dural feature is discussed in Section 7.3, where its impact on the bargaining
outcome is studied through a modified version of the basic alternating-offers
model. It is shown that this procedural feature may have a significant im-
pact on the bargaining outcome. In particular, under some conditions, there
exist subgame perfect equilibria that are Pareto inefficient.
While bargaining a player may have the option to temporarily close the
channel of communication (and thus, to temporarily stop the bargaining
process), which may impose costs on both players. For example, in the con-
text of union-firm negotiations, the union has the option to go on strike.
188 Procedures

Such an action may be considered a tactical move, and interpreted as 'burn-

ing money'. In Section 7.4 it is shown — through a modified version of the
basic alternating-offers model — that this procedural feature may have a
significant impact on the bargaining outcome. In particular, under some
conditions, there exist subgame perfect equilibria in which a player does
burn money.
In this chapter I show that procedures in general, and various specific
features of procedures in particular, may have a significant impact on the
outcome of bargaining. The question of what or who determines the bar-
gaining procedure, and its various features, is not touched upon. In the
language of game theory I show that the rules of the bargaining game mat-
ter, but I do not address the issue of what or who determines those rules.
In fact, this latter question has hardly been addressed in the economics and
game theory literature, because it is a very difficult question. I have noth-
ing substantial to say on this rather thorny issue, but to note that it is an
issue that deserves to be investigated in future research. It should also be
noted that in many real-life bargaining situations the procedures are, in fact,
somewhat ambiguous and/or not precisely well-specified.

7.2 Who Makes Offers and When

A basic characteristic of bargaining is making (and responding to) offers. If
neither player ever makes any offers, then agreement (if reached) is struck by
some procedure other than bargaining. Indeed, an agreement is determined
by bargaining if and only if at least one player makes at least one offer (that
is, proposes an agreement).
Rubinstein's model is based upon an alternating-offers procedure: the
players make offers alternately at equally spaced points in time, with the
responder reacting immediately to any offer. A main objective of this sec-
tion is to illustrate, by studying several alternative procedures, the insight
that the procedure of bargaining is an important determinant of the bar-
gaining outcome. A few of these alternative procedures may prove useful in
applications.
Throughout this section I consider the bargaining situation in which two
players, A and B, bargain over the partition of a (perfectly divisible) cake
of size 7T (where TT > 0). If agreement is reached that gives player i a share
7.2 Who Makes Offers and When 189

Xi of the cake, then her instantaneous utility is Xi.

7.2.1 The Ultimatum Game

This is the simplest of bargaining procedures. Player A makes an offer to

player B. If player B accepts the offer, then agreement is struck. Otherwise,
bargaining terminates in disagreement — and each player obtains a payoff of
zero. In the unique SPE — which can be derived by the backward induction
process — player A offers x*A = TT, and player B accepts any offer XA £ [0, TT].
It should be noted that there does not exist a SPE in which player B rejects
the offer XA — vr.
In equilibrium, player A obtains the whole cake. This suggests that
making offers confers bargaining power, while responding to offers gives no
bargaining power. The repeated version of this procedure studied in the
next section emphasizes this point. It should be noted that the ctake-it-or-
leave-it-offer' procedure implicitly assumes that the players are committed
not to continue bargaining if player B rejects player A's offer. In most real-
life negotiations making such commitments is rather difficult. As such this
procedure is not particularly plausible.

Random Determination of the Proposer

Consider the following extension to the take-it-or-leave-it-offer procedure:

with probability qA player A makes the single offer, and with probability qs
player B makes the single offer, where qA + qu — 1.
In the unique SPE the expected share of the cake obtained by player i
is qiir. This means (using Definition 2.3) that if qA > 0 and qs > 0 then the
expected SPE payoff pair (Q^TT, qB^) is identical to the asymmetric Nash
bargaining solution of the bargaining problem (Q,d) with r — qA-> where
Cl — {{UA-,UB) ' 0 < UA < 7T and UB = TT — UA} and d — (0,0).
Notice that the (limiting) SPE payoff pair in Rubinstein's model (cf.
Corollary 3.1) is identical to the expected SPE payoff pair in this model if
and only if qA = TB/(VA + re)- This result suggests that this extension
of the ultimatum game (with qA = TB/{TA + TB)) niay be interpreted as a
reduced-form of the more plausible, but more complex, Rubinstein model.
190 Procedures

7.2.2 Repeated Offers

Player A makes offers at each time £A (£ = 0,1, 2, 3 , . . . and A > 0), while
player B immediately responds to each such offer. Assume that player A's
discount rate TA > 0, and player B's discount rate TB > 0. If the players
perpetually disagree, then each player obtains a payoff of zero.
Through a minor adaption of the arguments in Sections 3.2.1 and 3.2.2
it is straightforward to show that in the unique SPE player A always offers
x*A = 7T, and player B always accepts any offer XA £ [O,^].1 Thus, in the
unique SPE player A obtains the whole cake at time 0. Notice that this
result holds even if player B is extremely patient relative to player A; that
is, it is valid for values of TA significantly larger than values of TB-
This result suggests that the procedure can have a far more potent im-
pact on the bargaining outcome than the discount rates. In fact, the discount
rates have no influence, whatsoever, in this repeated-offers game — unlike
in Rubinstein's alternating-offers game, where their relative magnitude sig-
nificantly influences the equilibrium partition.2
The repeated-offers bargaining game is a useful vehicle to emphasize the
point that making offers, rather than responding to them, confers a relatively
greater amount of bargaining power. However, it is unlikely that in real-life
bargaining situations the procedure of bargaining allows only one of the two
players to make offers. Player A has to be committed not to listen to any
offers made by player B in order for the repeated-offers procedure to have
much practical significance. But, since such a commitment is typically rather
difficult to sustain, the repeated-offers type procedure is not particularly
plausible.

1
First, one considers a SPE that satisfies Properties 3.1 and 3.2, and shows, in par-
ticular, that in such a SPE player B is indifferent between accepting and rejecting player
^4's offer; i.e., ix — xA — <5B(TT — xA)- Then, one adapts the arguments in Section 3.2.2
and obtains that UIA = TT — 6BNB, MA = TT — SBTIB, TLB = TT — MA and NB = TT — WIA,
where NB and n# are respectively the supremum and infimum of the set of SPE payoffs
to player B in any subgame beginning with player A's offer.
2
Furthermore, it is an illustration of the point that the impact of a particular force
(discount rates) on the bargaining outcome can depend on the nature of some other force
(the procedure).
7.2 Who Makes Offers and When 191

7.2.3 Simultaneous Offers

It may be quite natural when offers are communicated through certain kinds
of media (such as computers and the regular postal system) that the players
end up (strategically or otherwise) making 'simultaneous' offers. That is,
player A submits an offer to player i?, and before that offer reaches her, she
also submits an offer to player A. If the two offers are compatible, in a sense
made precise below, then a deal is struck. Otherwise, they simultaneously
submit fresh offers. This process may continue until a pair of compatible
offers are submitted. Below I present a bargaining model that incorporates
this repeated, simultaneous-offers, procedure. It is shown that such a model
is plagued by a multiplicity of subgame perfect equilibria, and, thus, fails
to resolve the fundamental indeterminacy that characterizes the bargaining
situation.
This analysis, therefore, further illustrates the potentially powerful im-
pact of the procedure of bargaining. Notice that the simultaneous-offers
aspect of the procedure considered here is partly tied in with the underlying
mode of communication. For example, if the players bargain face-to-face, it
seems unlikely that they would end up making such simultaneous offers.3
I consider the following simplest possible bargaining model that captures
the notion that the players make simultaneous offers. At each time £A
(t = 0,1,2,3,..., and A > 0), the players simultaneously submit their
respective offers, xlA and xfBi where I adopt the convention that an offer x\
submitted by player i at time tA denotes the share that player i would like
to have. Furthermore, I interpret the parameter A as constituting the time
that it takes for an offer submitted by player i to arrive at player j (which
is determined, in particular, by the efficiency of the postal system).
A pair of offers XA and XB are compatible if and only if the sum XA+XB <
7T. If at time tA the offers submitted are compatible, then the game ends
with an agreement that gives player i a share equal to her demand x\.
Otherwise the game moves to time {t + 1)A. The payoffs to the players are
as follows. If agreement is reached at time tA with player i's share being
3
The players could, of course, deliberately choose to simultaneously write down their
respective offers on pieces of paper, which are then exchanged. But then the procedure
would mimic communication via a computer, and thus undermine the spirit of bargaining
face-to-face, which is meant to be interpreted as vocal communication (making simulta-
neous use of player i's mouth and player j ' s ear).
192 Procedures

Xi, then player i's payoff is x^exp(—r^t), where n > 0 denotes her discount
rate. If the players perpetually disagree, then each player obtains a payoff
of zero.
Fix a pair x* = (x*A,x*B) G X, where X = {(XA,XB) •0 < x^ < TT and
X5 = 7T — x ^ } . It is trivial to verify that the following pair of strategies is
a subgame perfect equilibrium of the infinitely repeated simultaneous-offers
model: player A always asks for x*A, and player B always asks for x*B.
This means that this model has a multiplicity (indeed, a continuum) of
SPE. In each of the SPE described above, agreement is reached immedi-
ately at time 0. However, it is straightforward to construct a continuum of
SPE in which agreement is reached after some (considerable) delay.4 The
repeated simultaneous-offers bargaining game thus fails to resolve the basic
indeterminacy that characterizes bargaining situations. As is also known
from other game-theoretic contexts, it is the 'simultaneity' aspect of this
procedure that is responsible for this conclusion.

7.2.4 Random Proposers

Making offers confers relatively greater bargaining power than responding
to offers. An interesting bargaining procedure might thus determine the
proposer randomly. Although this might not be a descriptively persuasive
assumption, it may be a useful modelling assumption, especially because the
probabilities with which the players make offers parameterizes the players'
relative bargaining powers.
Consider the following bargaining model. At each time £A (t — 0,1, 2 , 3 , . .
and A > 0), with probability qA player A makes the offer to player B and
with probability qB player B makes the offer to player A, where qA + QB — 1-
If agreement is reached at time t A that gives player i a share X{ of the cake,
then her payoff is x^exp(—r^tA), where Ti > 0 denotes player i's discount
rate. On the other hand, if the players perpetually disagree, then each player
4
Fix an arbitrary agreement x* = (x\,x*B) (E X, and fix an arbitrary time T > 2A.
Now consider the following path of play: the players make incompatible demands until
time T — 2A, and then at time T — A the players' demands are defined by x*. This
path of play is sustained as a SPE path of play by the following off-the-equilibrium-path
strategies. If player i ever unilaterally deviates from this path of play, then immediately
play moves according to that SPE — described above — which gives player i a payoff of
7.2 Who Makes Offers and When 193

obtains a payoff of zero.

Through a straightforward adaptation of the arguments in Sections 3.2.1
and 3.2.2 it can be shown that this game has a unique SPE,5 in which player
i always offers x\ and always accepts an offer Xj if and only if Xj < x*, where

1 - Siqj - Sjqi

In the limit, as A —> 0, the SPE payoffs to players A and B are respec-
tively

Notice that if qA = qB — 1/2? then the limiting SPE payoff pair is

identical to the limiting Rubinsteinian SPE payoff (cf. Corollary 3.1). Fur-
thermore, notice that player z's limiting SPE share is strictly increasing in
qi and strictly decreasing in qj.

7.2.5 Alternating-Offers with Different Response Times

In the alternating-offers procedure that underlies Rubinstein's model, the
time interval between two consecutive offers is A > 0. This procedure
implicitly assumes that the amount of time it takes player A to make a
counteroffer after she rejects player B's offer is identical to the amount of
time it takes player B to make a counteroffer after she rejects player A's offer.
I now consider the modification to this procedure in which the amount of
time it takes player i to make a counteroffer after she rejects player j ' s offer
is Ai > 0. I thus allow for the possibility that A^ and AB are unequal. To
isolate the role of this difference, assume that the players discount future
utilities at a common rate r > 0. Define Si = exp(—rA^), which captures
the cost to player i of rejecting an offer.
It is trivial to verify that Theorem 3.1 is applicable, and it characterizes
the unique SPE of this model. In particular, when A^ and A# are small,
5
First, one considers a SPE that satisfies Properties 3.1 and 3.2, and shows, in partic-
ular, that in such a SPE player j is indifferent between accepting and rejecting player z's
offer; i.e., TT — x[ = 63\qi(TT — x*) -\-qjX*]. Then, one adapts the arguments in Section 3.2.2
and obtains that rrii = n — <5J[^(TT — rrii) + qjMj] and Mi = n — 8j[qi(7r — Mi) +
194 Procedures

the SPE shares (and payoffs) to players A and B are respectively

ABTT AATT
AA + AB and A A + A B'

The equilibrium partition depends on the ratio AA/AB. If this ratio equals
one, then the players split the cake equally. Otherwise, player i obtains a
bigger slice of the cake if and only if A^ < Aj. For example, if A^ = 1 sec-
ond and AB = 2 seconds, then player A obtains two-thirds of the cake while
player B obtains one-third of the cake. This is rather amazing, because a
small difference between A^ and AB has a significant impact on the SPE
partition. It reaffirms the basic message discussed in Section 3.2.3 that the
bargaining outcome depends critically upon the relative magnitude of the
players' costs of haggling, with the absolute magnitudes being irrelevant to
the bargaining outcome. Notice that player ^4's SPE share is strictly decreas-
ing in the ratio AA/AB, while player B's SPE share is strictly increasing in
the ratio AA/AB.

7.3 The Effect of Retractable Offers

I now study a simple extension to the basic alternating-offers model (studied
in Section 3.2) in which a proposer has the option to retract her offer after
it is accepted. Discussion of this procedural feature — the option to retract
accepted offers — is deferred to Section 7.3.4. I first, in Sections 7.3.1-7.3.3,
explore the impact of this procedural feature on the bargaining outcome.
Two players, A and I?, bargain over the partition of a cake of size n
(TT > 0) according to a modified version of the alternating-offers procedure.
The modification is as follows. At each time tA (t — 0,1,2,3,... and
A > 0), immediately after player j accepts an offer made by player i (where
(i,j) = (A,B) if t is even, and (i,j) = (B,A) if t is odd), player i has
to make a decision: either she retracts her offer, in which case the game
proceeds to time (t + 1)A where it is player jf's turn to make an offer, or
she does not retract her offer, in which case agreement is secured on the
accepted offer. The payoffs are as in the basic alternating-offers model (cf.
Section 3.2).
7.3 The Effect of Retractable Offers 195

7.3.1 A Subgame Perfect Equilibrium

Consider a SPE that satisfies Properties 3.1, 3.2 and the following Property
7.1.
Property 7.1 (No Retraction). A player does not retract her equilib-
rium offer.
It follows from a minor modification of the arguments in Section 3.2.1
that in equilibrium player i is indifferent between accepting and rejecting
player j ' s equilibrium offer. Hence, the equilibrium offers satisfy equations
3.1 and 3.2. Furthermore, in equilibrium, player i retracts an offer X{ if and
only if Xi < <^(TT — x*).6 The following lemma thus characterizes the unique
SPE that satisfies Properties 3.1, 3.2 and 7.1.

Lemma 7.1. The following pair of strategies is a subgame perfect equilib-

rium of the model with retractable offers:
• player A always offers x*A, always accepts an offer XB if and only if XB <
x*B, and always retracts an offer XA if and only if XA < 8\x\,
• player B always offers x*B, always accepts an offer XA if and only if XA <
x*A, and always retracts an offer XB if and only if XB < 82Bx*B,
where x\ — \IA^ and x*B = /i^Tr•, with \IA — (1 — <$B)/(1 — 8A8B) and

Notice that the SPE described in Lemma 7.1 is essentially the unique
Rubinsteinian SPE (cf. Theorem 3.1). Furthermore, by assumption, the
equilibrium offer is not retracted.

7.3.2 On the Uniqueness of the Equilibrium

I now address the issue of the existence of other SPE — by applying the
method of proof used in Section 3.2.2. As I show in Lemma 7.2 below, in
order to establish Lemma 3.1 I need to assume that 6A + 8B < I- 7
6
Without loss of generality, but in order to simplify the exposition, assume that if
player i is indifferent between retracting and not retracting an offer, then she does not
retract it.
7
The proof of Lemma 3.1 is based on the observation that in any SPE of Rubinstein's
model, player j accepts any offer x% such that TT — x% > SjMj. This observation is not
necessarily valid in Rubinstein's model with retractable offers, because in equilibrium an
accepted offer may be retracted.
196 Procedures

Lemma 7.2. If 8A + 8B < 1> then mi > TT — 8jMj (i ^ j).

Proof. Suppose, to the contrary, that rrii < TT — SJMJ. This implies that there
exists a SPE in the subgame beginning with player i's offer which gives her
a payoff ui such that mi < u\ < TT — SjMj. Now suppose player i deviates
and offers x\ — TT — 8jMj — e for some e > 0 such that x[> U{. Notice that
there exists e > 0 such that for any e G (0, e), x[ > U{. Player j accepts this
offer provided that player i does not subsequently retract it. The maximum
possible payoff to player i from retracting any offer is <^TT. Since Mj < TT
and (by assumption) 8A + £>B < 15 it follows that TT — SjMj > SITT. Hence,
there exists e* > 0 such that for any e G (0, e*), x\ > SITT. Thus, for any
6 G (0,min{e, e*}), x[ > SITT and x\ > ui\ that is, player i does not retract
the offer x\, and (by construction) the deviation is profitable for player i. •

In contrast to Lemma 7.2, which is valid provided 6A + 6B < 1, it is

straightforward to verify (through minor modifications to the proofs of Lem-
mas 3.2-3.4) that Lemmas 3.2-3.4 are valid for any values of the discount
factors. Hence, I obtain the following lemma.

Lemma 7.3. For any 6A and 8B, rrii < TT — 8jMj and Mi = TT — 8jvrt
j

Combining Lemmas 7.2 and 7.3, it follows that if 6A + 8B < 1 then MA =

mA = x*A and MB — TUB ~ x*Bi and if 8A + 8B > 1 then MA > x*A > TTIA and
MB > x*B > vriB-, where x\ and x*B are defined in Lemma 7.1. Consequently,
the following proposition can be proven (cf. Theorem 3.1).

Proposition 7.1. If 8A + 8B < 1, then the subgame perfect equilibrium

described in Lemma 7.1 is the unique subgame perfect equilibrium of Rubin-
stein's model with retractable offers.

It follows from Proposition 7.1 that if 8A + 8B < 1, then the option

to retract accepted offers has no influence on the bargaining outcome; the
unique Rubinsteinian equilibrium outcome is robust to this procedural fea-
ture. Notice that the condition 8A + 8B < 1 implies that A is sufficiently
large.
7.3 The Effect of Retractable Offers 197

7.3.3 Multiple Equilibria and Delay

I now establish (by construction) that if 6A + $B > 1> then there exists a
continuum of SPE. The key idea — which is similar to the construction of
the multiple equilibria in the telephone game (cf. Section 5.6.1) — involves
the construction of two 'extremal' SPE: these equilibria respectively give
players A and B their worst SPE payoffs.

L e m m a 7.4. If 6A+6B > 1, then Table 1.1 describes two SPE. If play begins
in state SA, then in equilibrium player A (who makes the first offer) offers
x*A = 7v, which is accepted by player B and subsequently is not retracted by
player A. If, on the other hand, play begins in state SB, then in equilibrium
player A offers x\ — 0 ; which is accepted by player B and subsequently is
not retracted by player A.

Proof. Using the One-Shot Deviation property, it is straightforward to verify

that if 6A + 6B > 1, then the pair of strategies described in Table 7.1 is a
subgame perfect equilibrium. However, in order to elucidate the role of the
condition 6A+6B > 1, I now establish the optimality of player A's behaviour,
in state SA, to reject any offer XB such that 6AK < TT — XB < TT. Suppose
that in state SA player A receives an offer XB such that 8 AIT < TT — XB < TT.
If she rejects this offer (as she is supposed to according to her strategy
described in Table 7.1), then the state does not change (see the transition
rule), and hence, her (discounted) payoff is 8ATT. Suppose she deviates from
the proposed behaviour and instead accepts such an offer (and thereafter
play conforms to the pair of strategies described in Table 7.1). Immediately
the state switches to SB (see the transition rule). Since 8ATT < TT — XB implies
(using the assumption 6A + 6B > 1) that XB < 8BTT, it follows from Table 7.1
that, in state SB, player B retracts such an accepted offer. Thus, player A's
payoff from deviating and accepting this offer is zero. Hence, this one-shot
deviation is not profitable for player A. •

The next result follows straightforwardly from this lemma.

Proposition 7.2. If 6A + 6B > 1, then for any partition x G {(XA,XB) '•

0 < XA < 7T and XB — TT — XA} and for any n G {0,1, 2 , 3 , 4 , . . . } there exists
a SPE such that along the equilibrium path the first n offers are rejected and
then the (n + l)th offer, which is the offer of the partition x, is accepted
198 Procedures

state SA state SB
offer xA = n x*A = 0

Player A accept XB=0 0 < XB < TX

retracts 0 < XA < SA^

offer xB=0 X*B=TT

Player B accept 0 < XA < 7T XA — 0

retracts 0< XB < SB^T

transitions switch to state SB if switch to state SA if

an offer XB > 0 is an offer XA > 0 is
accepted by player accepted by player
A

Table 7.1: Two extremal SPE of the model with retractable offers.

and subsequently is not retracted by the proposer. Hence, in equilibrium, the

partition x is implemented at time nA.
Proof. I establish the result by construction. Consider the following path of
play. At each time tA < (n —1)A, the proposer demands to receive the whole
cake and the responder rejects that demand. Then, at time nA, the proposer
offers the partition x, which the responder accepts and subsequently is not
retracted by the proposer. It is easy to verify (using the One-Shot Deviation
property) that this path of play can be supported as a SPE path of play
by the following off-the-equilibrium-path behaviour. If player i unilaterally
deviates from the proposed path of play, then immediately play proceeds
according to that extremal SPE described in Table 7.1 which begins in state
sj(j^i). a

7.3.4 Discussion and Interpretation

Fix the values of the players' discount rates r^ and rj5, and let A* denote
the unique value of A such that exp(—r^A) + exp(—r^A) = 1. Note that
A* > 0. I have shown (Proposition 7.1) that if A > A*, then the option
to retract accepted offers has no influence on the bargaining outcome; the
7.3 The Effect of Retractable Offers 199

unique Rubinsteinian SPE outcome is robust to this procedural feature. In

contrast, if A < A*, then I have shown (Proposition 7.2) that this proce-
dural feature can have a significant impact on the bargaining outcome. In
particular, the bargaining outcome can be Pareto inefficient. Since, as I ar-
gued in Section 3.2.4, interest in Rubinstein-type bargaining models centres
on the limiting case, as A —> 0, the case of A < A* is most persuasive.
Notice that in each of the continuum of SPE constructed above, along
the equilibrium path accepted offers are not retracted — which is consistent
with the potential observation that in real-life negotiations accepted offers
are not retracted. However, the results obtained above suggest: (i) that this
does not necessarily mean that the option to retract offers is unavailable in
real-life negotiations, and (ii) that the option to retract accepted offers (even
if it is not exercised) may have a potent impact on the bargaining outcome.8
If one adopts the 'classical' interpretation of the game form,9 then the
bargaining game ought to incorporate the option to retract accepted offers,
because it is typically physically feasible to exercise such an option. Making
irrevocable commitments not to retract accepted offers may be difficult in
real-life bargaining situations — especially because offers are typically made
verbally, and there is often a time lag between the acceptance of an offer
and its implementation.
If, however, one adopts the 'perceptive' interpretation, then, since the
players may perceive that accepted offers are not retractable (or, simply do
not even entertain the possibility of such an option), the Rubinstein model
without retractable offers may be vindicated. On the other hand, if the
players perceive that accepted offers are retractable, then the multiplicity of
equilibria obtained above becomes relevant.

8
As is well known in game theory, in general the equilibrium path of any game (i.e.,
the observed behaviour) is very much influenced and determined by what can and cannot
happen off the equilibrium path (i.e., by the non-observed behaviours that are admissible
and/or feasible). Analogously, the dog that did not bark provided a valuable clue to
Sherlock Holmes.
9
There are two ways of interpreting the game form (i.e., the rules of the game). The
'classical' interpretation is that the rules of a game should incorporate all actions which are
physically feasible in the situation that the game is meant to represent. In contrast, the
'perceptive' interpretation is that the rules of a game should incorporate only those actions
which are perceived by the players to be feasible and relevant. For further, insightful,
discussion of these interpretations, see Rubinstein (1991).
200 Procedures

7.4 Burning Money: A Tactical Move

While bargaining a player may have the option to take some action after an
offer is rejected and before a counteroffer is made that imposes costs on both
players. An example of such an action in the context of union-firm negotia-
tions is the union's option to go on strike. I now study a simple extension to
the basic alternating-offers model in which a player has the option to take
such an action — such an action may be interpreted as 'burning money'. It
is shown that this procedural feature may have a significant impact on the
bargaining outcome.
Two players, A and I?, bargain over the partition of a cake of size IT
(TT > 0) according to the following modified version of the alternating-offers
procedure. At time 0 player A makes an offer to player 5 , who either ac-
cepts or rejects the offer. In the former case agreement is secured and the
accepted offer is implemented, while in the latter case the game proceeds
to time A where it is player i?'s turn to make an offer. If player A accepts
player J3's offer then the accepted offer is implemented, but if she rejects the
offer then immediately player B has to choose between two actions: either
she allows player A to make her counteroffer at the earliest possible point
in time, namely at time 2 A, or she makes player A wait a further $ units
of time, and, thus, player A makes her counteroffer at time 2A + $. The
latter action, to be denoted by LMB\ is interpreted as 'burning money'.
The former action is denoted by CD\ Both A and $ take strictly positive
values. The parameter A is interpreted as the minimal time required to
formulate a counteroffer, while the parameter $ is the amount of time for
which communication between the two players is strategically closed. Fol-
lowing either action ('MS' or '£>'), the structures of the subgames at times
2A and 2A + $ are identical to the game at time 0. Notice, therefore, that
the structure of a subgame beginning with player A's offer is independent
of whether or not player B burns money.
The payoffs are as in the basic alternating-offers model. If agreement is
secured at time t > 0 with player i obtaining a share xi £ [0, TT] of the cake,
then her payoff is Xiexp(—rt), where r > 0 denotes her discount rate. If
the players perpetually disagree, then each player obtains a payoff of zero.
For notational convenience, define 6 = exp(—rA) and a = exp(—r(A + $)).
Notice that 0 < a < 6 < 1. Furthermore, notice that, as A —> 0, 6 —> 1 and
7.4 Burning Money: A Tactical Move 201

a —> exp(—r$).

7.4.1 A Subgame Perfect Equilibrium

Consider a SPE that satisfies Properties 3.1, 3.2 and the following Property
7.2.

Property 7.2 (No Money Burning). Whenever a player has to decide

whether or not to burn money, she chooses not to burn money.

From the arguments in Section 3.2.1, I obtain the following lemma which
characterizes the unique SPE that satisfies Properties 3.1, 3.2 and 7.2.

Lemma 7.5. The following pair of strategies is a subgame perfect equilib-

rium of the model with money burning:
• player A always offers x\ and always accepts an offer XB if and only if
xB < x*B,
• player B always offers x*B, always accepts an offer XA if and only if XA <
x*A, and always chooses not to burn money,
where x\ = x*B = TT/(1 + 8).

Notice that the SPE described in Lemma 7.5 is essentially the unique
Rubinsteinian SPE (cf. Theorem 3.1). Furthermore, by construction, in
equilibrium player B never burns money.

7.4.2 On the Uniqueness of the Equilibrium

I now address the issue of the existence of other SPE — by applying the
method of proof used in Section 3.2.2. Unlike in the model with retractable
offers, the arguments underlying the proofs to Lemmas 3.1-3.4 are applicable
to the model with money burning. I thus obtain that TTIA — TT —
TUB — TT — 6MA, MA = TT — STUB, MB > TT — SrriA and MB < TT —
Combining these equations and inequalities, it follows that

7T (1— a W , ( 1 — <5)TT 7T
J J
mB = —^ <MB< ^ — and \ — < mA < —-j: = M A.
1+6 1 — a6 1 — a6 1+6
This means that the SPE described in Lemma 7.5 gives players A and B
respectively their best and worst equilibrium payoffs. The following propo-
202 Procedures

sit ion establishes that if a(l + 6) < 1, then the SPE described in Lemma 7.5
is the unique SPE of the model with money burning.10
Proposition 7.3. // a(l + 6) < 1, then the subgame perfect equilibrium
described in Lemma 7.5 is the unique subgame perfect equilibrium of Rubin-
stein's model with money burning.
Proof. I first argue that if a(l + 6) < 1, then there does not exist a SPE
in which player B ever chooses to burn money. Let HB denote the set of
all SPE payoffs to player B in any subgame beginning with player A's offer,
and let NB and UB respectively denote the supremum and infimum of HB-
Since NB = 8MB and n# = brxiB-, it follows that NB < S(l — a)7r/(l — aS)
and VLB = Sir/(I + 6). Now consider any point in the game at which player
B has to decide between 'MB' and LD\ The best possible SPE payoff that
player B obtains by burning money is bounded above by aNs, and the
worst possible SPE payoff that player B obtains by not burning money is
bounded below by SUB- Using the hypothesis a(l + 6) < 1, it follows that
SriB > &NB' Hence, if a(l + 6) < 1 then there does not exist a SPE in
which player B ever chooses to burn money; that is, in any SPE player B
always chooses the action CD\ The proposition follows trivially from this
result. •

It follows from Proposition 7.3 that if a(l + 6) < 1, then the option to
burn money has no influence on the bargaining outcome; the unique Rubin-
steinian equilibrium outcome is robust to this procedural feature. Notice
that the condition a(l + 6) < 1 means that $ is sufficiently large. In the
limit, as A —> 0, a(l + 6) < 1 if and only if $ > In2/r.

7.4.3 Multiple Equilibria

I now establish (by construction) that if a{\ + 6) > 1, then there exists a
continuum of SPE. As in Section 7.3.3, I first characterize the 'extremal'
SPE that gives player A her worst equilibrium payoff.11
10
The intuition for this result is as follows. If a is sufficiently small relative to 6 (i.e.,
$ is sufficiently large), then it is not optimal for player B to burn money in any SPE. To
see this in a rather stark form, just consider the limiting case of $ —>• oo, given a finite
value of A.
11
Lemma 7.5 characterizes the 'extremal' SPE that gives player B her worst equilibrium
payoff.
7.4 Burning Money: A Tactical Move 203

L e m m a 7.6. / / and only if a(l + 6) > 1, the pair of strategies described in

Table 7.2 is a subgame perfect equilibrium. In equilibrium the game begins
in state s and player A offers XA = (1 — S)TT/(1 — ad), which is accepted by
player B. Furthermore, if and only if a(l + 6) > 1, VTLA — (1 — 6)TT/(1 — aS)
and MB = (1 - «)TT/(1 - a6).

Proof. It is easy to verify (using the One-Shot Deviation property) that the
pair of strategies described in Table 7.2 is a subgame perfect equilibrium.
However, in order to demonstrate the role played by the condition a(l + 6) >
1,1 show that it is optimal for player B to burn money when play is in state s.
Consider any point in the game where player B has to decide between CMB'
and CD\ and suppose play is in state s. According to her strategy described
in Table 7.2 she is supposed to burn money. Her payoff from conforming
to this behaviour is a(ir — XA)- Suppose she instead deviates and does not
burn money (and thereafter she conforms to her strategy described in Table
7.2). Since the state immediately switches to Ǥ*, her payoff from this one-
shot deviation is 6(TT — x*A). The latter payoff from the one-shot deviation
is less than or equal to her former payoff from conforming if and only if
a(l + 6)>l. •

state s state s*
offer xA x\
Player A
accept xB < x B XB < xB
offer xB xB
Player B accept XA < xA XA < X\

'MB' or '£>' 'MB'

transitions switch to state s* absorbing
if player B ever
chooses 'D'

Table 7.2: A SPE of the model with money burning, where x\ = x*B = TT/(1 + 6),
xA = (l- <5)TT/(1 - a6) and xB = (1 - a)?r/(l - a6).

The intuition behind the SPE described in Table 7.2 is as follows. Player
J?'s plan (or, threat) to burn money is credible, because if she fails to take
204 Procedures

that action then play immediately moves to the 'extremal' SPE that gives
player B her worst possible equilibrium payoff. Thus, in equilibrium, players
A and B have to wait respectively for A + $ and A time units before being
able to make their respective counteroffers. Therefore, it is 'as if the players
are playing Rubinstein's game (in which player B does not have the option to
burn money), where player fi's discount factor is 6 and player A's discount
factor is a.
Using the results contained in Lemmas 7.5 and 7.6 — which characterize
the two 'extremal' equilibria — it is now straightforward to construct a
continuum of SPE.

Proposition 7.4. If a(l + S) > 1, then for any partition x G {(XA^XB) •

0 < XA < 7T and XB = vr — XA} of the cake such that

there exists a SPE in which agreement is reached at time 0 on the partition

x.

Proof Fix an arbitrary partition x satisfying (7.1), and consider the follow-
ing pair of strategies. At time 0 player A offers XA and player B accepts that
offer. If player A deviates and instead offers x'A^ XA-, then immediately
(before player B decides whether to accept or reject the offer) play proceeds
according to the SPE described in Lemma 7.6. Furthermore, if player B
rejects the offer XA-, then immediately play proceeds according to the SPE
described in Lemma 7.5. It is straightforward to verify, using the One-Shot
Deviation property (and given that XA satisfies inequality 7.1), that neither
player can benefit from a unilateral deviation at time 0. •

Notice that each SPE described in Proposition 7.4 is Pareto efficient, in

that (in equilibrium) agreement is reached immediately without any delay.

7.4.4 Equilibrium Delay

I now construct a set of equilibria such that in each such SPE agreement is
reached on some partition of the cake at some time T; and moreover, along
the equilibrium path, player B burns money whenever she has to decide
whether or not to do so. Of course, assume that a ( l + <$)>!.
7.4 Burning Money: A Tactical Move 205

Fix an arbitrary non-negative integer k (i.e., k = 0,1, 2, 3 , . . . ) and define

T = fc(A + $). Furthermore, let x = (XA, XB) denote an arbitrary partition
of the cake such that XA satisfies (7.1).
It is convenient to specify the proposed SPE by first describing the pro-
posed SPE path of play, and then specifying how play proceeds if a player
deviates from this path. Consider the following path of play. At any time
t < T, whenever player i has to make an offer, she offers X{ = TV and player
j (j z^z i^j rejects that offer. Furthermore, at any time t < T, whenever
player B has to decide between iMB'> and iD\ she chooses CMB\ Given
this behaviour at any time t < T it follows (by the definition of T) that at
time T it is player A's turn to make an offer. Assume that she offers the
partition specified above, namely x, which player B accepts. Now I describe
how play proceeds if at any time t < T a player deviates from this path of
play. If player A (resp., player B) ever deviates from this path of play, then
immediately play proceeds according to the SPE described in Lemma 7.6
(resp., Lemma 7.5).
I now proceed to characterize the values (if any) of k and x for which
the pair of strategies thus described is a SPE. Letting Vi denote the payoff
to player i in this proposed SPE, I note that Vi = x^exp(—rT). Since, by
construction, unilateral deviations from the proposed path of play moves
play to a SPE, I need to only address the issue of whether or not a player
has an incentive to deviate from the path of play. That neither player has
an incentive to unilaterally deviate at time T follows from Proposition 7.4.
Now I consider any point of time t < T at which some player has to make
a decision.
First consider any point along the path of play at which a player, say
player j , has to reject the offer X{ — TT. Since Vj > 0, it follows that player
j does not benefit by instead accepting that offer.
Now consider any point of time t < T at which player A has to make an
offer. If she conforms to the proposed behaviour and offers XA = ft, then her
payoff is VA- Suppose she considers a one-shot deviation and instead offers
x'A < 7T. Then, immediately play moves to the SPE described in Lemma
7.6. If x'A < XA, where XA is defined in Lemma 7.6, then (as can be seen
from Table 7.2) player B accepts that offer, and, thus, the payoff V^{xfAlt)
to player A from this one-shot deviation is xfAex.p(—rt). For any 0 < t < T
and any 0 < x'A < XA, such a one-shot deviation is not beneficial (i.e.,
206 Procedures

VA > VA(x'A, t)) if and only if

xAexp(-rT)> {]~S)* (7.2)

1 — ao
On the other hand, if XA < xfA < 1, then (as can be seen from Table 7.2)
player B rejects that offer, and, thus, the payoff to player A from such a
one-shot deviation is ax A exp(—r(£ +A)) which is less than VA if %A satisfies
(7.2). In summary, (7.2) is a necessary and sufficient condition to ensure
that player A never has any incentive to unilaterally deviate from the path
of play at any point in time t < T.
Now consider any point of time t < T at which player B has to make an
offer. If she conforms to the proposed behaviour and offers XB — TT, then her
payoff is VB- Suppose she considers a one-shot deviation and instead offers
x'B < 7T. Then, immediately play moves according to the SPE described in
Lemma 7.5. By a similar argument to that provided above, it follows that
such a one-shot deviation is not beneficial if and only if

r T ) > ^ . (7.3)

Now consider any point of time t < T at which player B has to decide
between ''MB1 and iD\ It is easy to verify that (7.3) is a sufficient condition
to ensure that she does not benefit from a one-shot deviation. In summary,
(7.3) is a necessary and sufficient condition to ensure that player B never
has any incentive to unilaterally deviate from the path of play at point in
time t < T.
Since XB = TT — XA, (7.3) is equivalent to

xAexp(-rT) < Trexp(-rT) - -. (7.4)

Notice that if XA satisfies (7.2) and (7.4), then XA satisfies (7.1). Hence, the
pair of strategies described above is a subgame perfect equilibrium if and
only if k and XA satisfy

<< (7.5)
(1 - aS) exp(-rT) ~ ~ (l + S) exp(-rT)
Define the set W = {(/c,x^) : k is a non-negative integer, XA G [0, TT] and
(k,XA) satisfy (7.5)}. It is easy to verify that W is non-empty, and that it
7.4 Burning Money: A Tactical Move 207

is illustrated in Figure 7.1, where k* denotes the real number at which point
the two curves intersect and k denotes the largest integer smaller than k*.
Hence, I summarize the main insight thus derived in Proposition 7.5.
xA

F i g u r e 7.1: An illustration of the set W, which characterizes the set of SPE with delay
in the model with money burning — as stated in Proposition 7.5.

Proposition 7.5. If a(l + 6) > 1, then for any (fc,^) G W there exists a
SPE of the model with money burning such that in equilibrium agreement is
reached at time T — k(A + $) on the partition that gives player A a share
XA (and player B a share TT — XA), and in equilibrium player B burns money
k times.

Notice (from Figure 7.1, for example) that in any such SPE, agreement
is reached at some time T < T, where T = fc(A + $). Thus, T denotes the
maximal possible delay consistent with equilibrium behaviour. Assuming
for simplicity that /c* is an integer — in which case k is that value of k
where the two curves intersect — in the limit, as A —> 0, k —* I n 2 / r $ and
T —» In2/r. Notice that, in this limit, the maximal possible equilibrium
delay is unaffected by the value of $.
208 Procedures

7.4.5 Discussion

The basic insight (on the effect that money burning can have on the bargain-
ing outcome) established above can be generalized in many ways, and can
be applied to a variety of contexts. For example, one could also allow player
A the option to burn money immediately after her offer is rejected. Notice,
however, that I precluded player B from taking such 'money burning' ac-
tion immediately after she rejects player A*s offers. The reason for doing
so comes from the result (which is straightforward to verify) that player B
would never, in any SPE, burn money after rejecting offers made by player
A. The intuition behind this result is as follows. By burning money at such
points in the game would, by increasing player i?'s cost of rejecting offers,
decrease the share that player A would need to offer her. Thus, player B
would certainly not burn money after rejecting any offer made by player A.
On the contrary, she would wish to make her counteroffer as soon as possi-
ble. In contrast, by burning money after player A rejects player B's offers,
player B increases the cost to player A of rejecting her offers, and, thus, she
is able to offer player A a relatively smaller share. Indeed, the reason for
why 'money burning' by player i, after player j (j ^ i) rejects offers made
by player i, increases player z's 'bargaining power' is precisely because such
an action increases the cost to player j of rejecting player i's offers.

7.4.6 An Application to Surplus Destruction

Consider a bargaining situation in which a player can take an action (im-

mediately after her offer is rejected and before her opponent gets to make a
counteroffer) that results in the immediate and literal destruction of some
fraction of the (remaining) cake. This action, which does not affect the times
at which offers are made, is different from the particular 'money burning'
action modelled in the preceding section. However, since the strategic struc-
tures are identical, the analysis of the preceding section is applicable.
At each time tA (where t = 1,3,5,...), immediately after player A
rejects an offer made by player 5 , player B decides between 'MB' and
C
D\ The former action involves the destruction of a fraction 1 — 7 of the
(remaining) cake available at time tA. Thus, if nt denotes the size of the
cake at time tA and if player B chooses LMB\ then the size of the cake at
time (t + 1)A is iit+i = 7 ^ - On the other hand, if she chooses '£>', then
7.5 Notes 209

m+i — TTf. Note that 7 G [0,1] and TTO = TT. AS before, r > 0 denotes the
common discount rate. For notational convenience and for comparability
sake, let 8 = exp(—rA) and a = 7<5.12 For simplicity, I do not give player
A the option to 'burn money'. Thus, at each time tA (t = 0 , 2 , 4 , . . . ) ,
immediately after player A makes her offer, player B either accepts the
offer (in which case the game ends) or she rejects the offer (in which case
bargaining continues with player B making her offer at time (t + 1)A).
Propositions 7.3-7.5 are applicable. 13 This means that if 7 < 1/8(1 + 6),
then there exists a unique SPE in which the cake is never destroyed, and
the equilibrium partition agreed to at time 0 is the Rubinsteinian equilib-
rium partition. On the other hand, if 7 > 1/8(1 + <5), then there exists a
multiplicity of equilibria. Let us focus on the set of equilibria in the limit as
A —• 0. In this limit, if 7 < 1/2, then in the unique SPE the cake is never
destroyed. Thus, if the 'MB' action involves destroying more than 1/2 of
the (remaining) cake, then player B never takes this action. On the other
hand, if 7 > 1/2, then there exists a continuum of SPE. In particular, in
any SPE the cake is destroyed at most k times, where (assuming for sim-
plicity that k* is an integer) j k = TT/2. Hence, the maximal amount of cake
destroyed consistent with equilibrium is TT/2.

7.5 Notes
For further analysis of the infinitely repeated simultaneous-offers model see
Binmore (1987) and Chatterjee and Samuelson (1990). Perry and Reny
(1993) study a fairly complex continuous time bargaining model in which
a player can make an offer at almost any point in time. They show that
under some conditions the subgame perfect equilibria of such a model ap-
proximates the unique SPE of Rubinstein's (alternating-offers) model. Thus,
their analysis provides some further support for Rubinstein's model.
The model with retractable offers discussed in Section 7.3 is based upon
Muthoo (1990). Similar results hold in several other kinds of bargaining
12
Indeed, 7 is 'equivalent' to exp(—r<I>)
13
It may be convenient — when verifying that these results are applicable — to interpret
an offer x\ at time tA by player i as constituting the fraction of the cake of size n t
(available at time tA) that player i obtains; thus, players i and j respectively obtain n tx\
and 7Tt(l — x\) amounts of the cake.
210 Procedures

models which allow for retractable offers (see, for example, Chaudhuri (1992,
Chapter 6) and Muthoo (1995a)). Furthermore, it is easy to establish similar
results to those established in Section 7.3, if, for example, a responder has
the option to retract her acceptance of an offer after the proposer decides
not to retract her offer. Chaudhuri (1994) contains an application of the
model with retractable offers that relates a firm's pricing policy to its size.
Avery and Zemsky (1994) coin the term 'money burning'. Their pa-
per generalizes the results described in Section 7.4 in some useful ways,
and, furthermore, contains examples of this type of action. The applica-
tion discussed in Section 7.4.6 is motivated by the work of Busch, Shi and
Wen (1998) and Manzini (1999), who derive some further results about this
model.
In some bargaining situations the set of possible agreements is finite. For
example, the set of agreements is finite when bargaining over the partition
of one dollar, because of the existence of a smallest unit of money, namely,
one cent. Since this feature affects the set of actions (offers) available to a
proposer, it affects the rules of the bargaining game, van Damme, Selten and
Winter (1990) and Muthoo (1991) show that (under some conditions) Ru-
binstein's model with a finite number of agreements possesses a multiplicity
of subgame perfect equilibria.
8 Commitment Tactics

8.1 Introduction
In many bargaining situations the players often take actions prior to and/or
during the negotiation process that partially commit them to some strate-
gically chosen bargaining positions. Such commitments are partial in that
they are revocable, but revoking a partial commitment can be costly. The
following two extracts from Schelling (1960) illustrate this 'commitment'
tactic.

it has not been uncommon for union officials to stir up excitement

and determination on the part of the membership during or prior
to a wage negotiation. If the union is going to insist on $2 and
expects the management to counter with $1.60, an effort is made
to persuade the membership not only that the management could
pay $2 but even perhaps that the negotiators themselves are
incompetent if they fail to obtain close to $2. The purpose ...
is to make clear to the management that the negotiators could
not accept less than $2 even if they wished to because they no
longer control the members or because they would lose their own
positions if they tried. In other words, the negotiators reduce the
scope of their own authority and confront the management with
the threat of a strike that the union itself cannot avert, even
212 Commitment Tactics

though it was the union's own action that eliminated its power
to prevent strike. Schelling (1960, p. 27)
When national representatives go to international negotiations
knowing that there is a wide range of potential agreement within
which the outcome will depend on bargaining, they seem often
to create a bargaining position by public statements, statements
calculated to arouse a public opinion that permits no concessions
to be made. Schelling (1960, p. 28)

The main objective of this chapter is to investigate the role of such com-
mitment tactics on the outcome of bargaining. I conduct this investigation
by studying models based on the following ideas:
• Before embarking on some negotiation process (such as an offer-counteroffer
process), each player takes actions that partially commit her to some bar-
gaining position.
• Revoking such partial commitments can be costly.
• After such partial commitments have been made the players engage in
some negotiation process in order to strike a deal. Each player will wish to
minimize the extent to which she revokes her partial commitment.
In many bargaining situations there often exist a variety of 'mechanisms'
through which a player can achieve partial commitment to some bargaining
position. The two extracts from Schelling (1960) cited above suggest some
such mechanisms. It seems intuitive that the higher is the player's cost of
revoking a partial commitment, the greater is the degree to which she is
committed to her chosen bargaining position. For example, if this cost of
revoking is zero, then it is 'as if she is not committed at all. At the other
extreme, if this cost of revoking is sufficiently high (perhaps infinite), then
it is 'as if the bargainer is irrevocably committed to her chosen bargaining
position. Of course, different mechanisms induce different costs of revoking,
and, hence, different degrees to which commitment is achieved.
The models studied in this chapter do not formalize the mechanism (or
mechanisms) through which a bargainer may achieve partial commitment —
I assume that some such mechanism exists. The models, however, formalize
the key features of such mechanisms, which are the induced costs of revoking
such partial commitments. Much insight on the role of commitment tactics is
obtained from the analyses of such abstract (reduced-form) models. This is
8.1 Introduction 213

especially the case since, by parameterizing the costs of revoking, the models
encompass a wide variety of mechanisms that may be used to achieve partial
commitment, from those which induce small costs of revoking to those that
induce large costs of revoking.
In addition to exploring the effect that such commitment tactics may
have on the players' 'bargaining powers', I also explore the idea that the
deployment of such tactics may increase the likelihood of disagreements in
bargaining. Since it is possible that the players end up locking themselves
into incompatible bargaining positions, they may fail to reach agreement if
each player's cost of revoking is sufficiently high.
The models that I study are two-stage games (or, can be interpreted
as such). At the first stage the two players choose their respective partial
commitments. This stage can be interpreted as taking place outside the
formal negotiating process. After such partial commitments become known,
at the second stage the players enter the formal negotiating process, where
they attempt to reach agreement. This second stage could, for example, be
either a Rubinstein-type alternating-offers game or a concession-type game
(that may resemble a war of attrition). Broadly put, then, such bargaining
models can be interpreted as models of 'claims' and 'concessions'.
A main focus of interest is the nature of the equilibrium at the first
stage of such models, as it is at that stage that partial commitments are
strategically chosen. For example, as mentioned above, an important issue
from a welfare point of view is to study the circumstances under which, in
equilibrium, the players make incompatible partial commitments. Hence,
although the equilibrium at the first stage is influenced by the second stage
game, it is the first stage equilibrium actions that are the focus of interest
in this chapter.
The chapter is organized as follows. In the next section I formulate,
interpret and analyse the basic model designed to explore the role of com-
mitment tactics. A detailed discussion of the results obtained from this
model are contained in Section 8.3, where I also explore the robustness of
the main result to two extensions of this model. Section 8.4 contains an
application of this model to the role of delegation in bargaining. A crucial
assumption underlying Sections 8.2-8.4 is that the players' costs of revoking
their respective partial commitments are known to both players. Partly as a
result of this assumption, the unique equilibrium is always Pareto efficient.
214 Commitment Tactics

In particular, in equilibrium the players do not make incompatible partial

commitments. In Sections 8.5 and 8.6 I analyse two models in which this
assumption is dropped. More precisely, when taking actions that partially
commit the players to strategically chosen bargaining positions, both players
do not know either player's costs of revoking. Each player's costs become
known only to her after the 'claims' stage but before the second 'conces-
sions' stage. A main objective of these two sections is to explore whether or
not (as a result of the symmetric uncertainty at the first stage and asym-
metric information at the second stage) the players, in equilibrium, make
incompatible partial commitments.

8.2 The Basic Model

Although it will become evident that the model described here can be in-
terpreted as a two-stage game (in which at the second stage the players are
engaged in Rubinstein's alternating-offers game), the formal structure of
the model constitutes a one-shot, static, game. That is, I will not formally
model the second ('concessions') stage. Instead, I use the Nash bargaining
solution to characterize the outcome of the second ('concessions') stage. 1
This modelling assumption simplifies the analysis, and allows attention to
be focused on the first ('claims') stage.

8.2.1 The Formal Structure

Two players, A and I?, bargain over the partition of a cake of size TT (where
7T > 0) according to the following static game. They simultaneously and
independently choose numbers from the closed interval [0, TT]. Let Zi denote
the number chosen by player i (i = A,B). The interpretation is that player
i takes 'actions' which partially commit her not to accept a share strictly
less than Z{. A partial commitment can later be revoked at some cost to the
player.
The utility Ui(x{, Zi) to player i from obtaining a share X{ G [0, TT] of the
1
This can be justified by appealing to the result that the unique subgame perfect
equilibrium payoff pair in Rubinstein's model converges (as the time interval between two
consecutive offers converges to zero) to Nash's bargaining solution (cf. Corollary 3.4).
8.2 The Basic Model 215

cake, given that she partially committed herself to Zi G [0, TT], is

Ui(xhzi) = Xi-Ci(xi,Zi), (8.1)

where C{(xi, zi) denotes the cost to player i of revoking her partial commit-
ment Zi and obtaining a share x\. It is assumed that Ci(xi, zi) = 0 if xi > z^
and that Ci{x^ z\) > 0 if Xi < Z{.
More specifically

Cfe,,,) = {° **** (8.2)

[ki(zi -Xi) if Xi < Zi,
where hi > 0. Notice that this cost of revoking function captures the notion
that the cost of revoking a partial commitment is strictly increasing in the
extent to which it is actually revoked.2
I now turn to a description of the players' payoffs, where I denote player
z's payoff from a strategy pair z = (ZA, %B) by Pi(z).
First I describe the payoffs when the chosen partial commitments ZA and
ZB are such that ZA + ZB < TT- I n this case, neither player revokes her partial
commitment: the share Xi of the cake obtained by player i is such that Xi >
Zi. Specifically, the share obtained by player i is given by \{z), where XA
and XB are any functions such that XA(Z) > ZA and \B{Z) = TY-XA(Z) > ZB-
For example, it may be assumed that Xi(z) = Zi + (TT — ZA — ZB)/2- Hence,
if ZA + ZB < 7T, then player i's payoff Pi(z) — Xi(z).
If, when ZA + ZB > TT, agreement over the partition of the cake is struck,
then at least one of the players must have revoked her partial commitment.
The set Q(z) of possible utility pairs that can be the outcome of bargaining is
constructed using the set X of possible partitions of the cake and the utility
functions UA and [/#, where X = {(XA, XB) •0 < X^ < TT and XB = TT — XA}-
That is, the set ft(z) is the union of all pairs (UA(XA,ZA),UB(XB,ZB)) f° r
(XA,XB) £ X. Indeed, for each pair z £ [0,TT]2 such that ZA + ZB > TT, the
set Q(z) is the graph of the function #(.; z) defined by

g(uA;z) = UB(TT - U^1{UA]ZA)]ZB),

2
In Section 8.3.5 I study a more general version of the model studied here in which
(i) I do not impose any specific functional form on the cost of revoking functions, but I
instead restrict the class of such functions by a set of assumptions, and (ii) I assume that
the utility to player i from obtaining x% units of the cake is given by a strictly increasing
and concave function.
216 Commitment Tactics

where the domain and range of g(.;z) are the closed intervals [—kAZA^]
and [—fcBZB,7r], respectively (cf. Section 2.2). Notice that g(.;z) is concave
and strictly decreasing in UA (cf. Lemma 2.1).
If the players do not reach agreement, then each player obtains a payoff
of zero.
It is possible (but not necessary) that there exists values of ZA and ZB
such that g(0; z) < 0; in this case, PA{Z) = PB(Z) — 0. However, if #(0; z) >
0, then the payoff pair (PA(Z),PB(Z)) is defined as the Nash bargaining
solution of the bargaining problem (O(z), d), where the disagreement point
d = (0,0) (cf. Definition 2.1). That is, (PA(Z),PB(Z)) is the unique solution
to the following maximization problem

max
U
UAUB subject to UB = g{uA] Z),UA > 0 and UB > 0.
A:UB

This completes the formal description of the simultaneous-move game.

Figure 8.1 illustrates the utility possibility set Q(z) for an arbitrary pair
z = [ZA-> ^B) such that ZA + ZB > TT. The line segment RS depicts the set
X of possible partitions of the cake, while the graph TNQM depicts the
utility possibility set fl(z). If the agreed partition lies on the line segment
RV (resp., WS), then it is only player A (resp., B) who is revoking her
partial commitment. But, if the agreed partition lies in the interior of the
line segment VW, then both players are revoking their respective partial
commitments. Consider, for example, the partition (XA,XB) depicted by
the point a. The utility XA — ^A{ZA — XA) to player A is the 'x-coordinate'
of point c, while the utility XB — ^B^ZB — XB) to player B is the '^-coordinate'
of point b. Hence, this defines the utility pair (point d) associated with such
a partition.
It is straightforward to verify that the line segments TW, VM and NQ
are defined by equations 8.3, 8.4 and 8.5, respectively: 3

UA + (1 + kA)uB =7T + kA(7T - ZA) (8.3)

uB + (l + kB)uA = TT + kB(7T - zB) (8.4)
3
Let (UA,UB) be a utility pair on the line segment NQ. This means that the partition
(XA,XB) of the cake associated with this utility pair is such that XA < ZA and XB < ZB-
Hence, UA — XA — ^A{ZA — XA) and UB — XB — ^B[ZB — XB)- That is, UA/(1 + ^A) =
XA — &AZA/(1 + &A) ^ B / ( 1 + &B) — XB — kBZB/(l + kB)- Equation 8.5 follows from adding
the latter two equations, and then noting that XA + XB — TT.
8.2 The Basic Model 217

"B

F i g u r e 8.1: TNQM is an illustration of the graph of the function #(.; z) — which is

the set £l(z) of possible utility pairs.

(1 + + (1 + = 7T +

TT - zB) (8.5)

I note here that the slopes of these equations (and, in particular, the fact that
they are independent of ZA and ZB) play an important role in my geometric
analysis of the Nash equilibria of this model.

8.2.2 Interpretation
The model described above is designed to explore the role of commitment
tactics on the outcome of bargaining. It is a fairly simple and abstract
model. As I argue below, the model captures (in that it explicitly formalizes)
some salient aspects of such tactics, but omits (from its formal structure) a
significant amount of the richness that characterizes the deployment of such
218 Commitment Tactics

tactics. As is the case in the construction of any theoretical model, what is

included in the formal structure is chosen deliberately so that the analysis
is tractable and, at the same time, provides some insights on the impact of
such tactics on the bargaining outcome. Once this model is understood, it
may then become clear as to how it could fruitfully be enriched.
The mechanism through which a bargainer achieves partial commitment
to her chosen bargaining position is not specified in the formal structure of
the model. Moreover, the events that follow if the agreement involves the
bargainer revoking her chosen partial commitment are also not incorporated
in the formal structure of the model. Both of these crucial features of the
commitment tactic are represented in the model by her cost-of-revoking
function. It seems intuitive that the impact, if any, on the outcome of
bargaining is due to the costs of revoking partial commitments. Hence, it
seems reasonable to adopt this 'reduced-form' modelling approach.
In the model described above, I have imposed a particular structure
(functional form) on these cost-of-revoking functions, because the analysis
of the Nash equilibria is relatively straightforward, and it can be conducted
geometrically — which provides a relatively more intuitive understanding
of the arguments. However, in Section 8.3.5, I show that the unique Nash
equilibrium is robust to alternative forms of the costs of revoking functions.
Notice that the cost-of-revoking function stated in equation 8.2 captures
some rather intuitive features (such as the notion that the cost is increasing
in the extent to which a bargainer revokes her partial commitment).

8.2.3 The Equilibrium

I now derive the unique Nash equilibrium (NE) of the model described above.
The first result below establishes that any pair of more-than-compatible par-
tial commitments is not a NE.

Lemma 8.1. // ZA + ZB < it, then the pair z = (ZA,ZB) is not a Nash
equilibrium.

Proof. The proof is straightforward, because there exists an i (i = A or

i = B) such that n — z3 > Xi(z) (j ^ i); for otherwise, XA(Z) > TT — ZB and
XB{Z) ^ TT — ZA imply that ZA + ZB > TT. Hence, player i can benefit from a
unilateral deviation to z[ = TT — Zj. •
8.2 The Basic Model 219

The next result establishes that any pair of incompatible partial commit-
ments is not a NE.
Lemma 8.2. // ZA + %B > TT, then the pair z = (ZA,ZB) is not a Nash
equilibrium.
For notational convenience, define 7 = (1 + + &A)-

F i g u r e 8.2: The geometry behind the proof of Lemma 8.2.

Proof. First, assume that g(0]z) > 0. The argument, by contradiction, is

as follows (allowing for the roles of A and B to be reversed).4 Suppose
Z
PB(Z) < B (i.e., in Figure 8.1, the payoff pair (PA(Z),PB(Z)) lies either in
the interior of the line segment NQ or on the line segment QM). I now
4
A central part of my argument is based on the geometric characterization of the Nash
bargaining solution, that it is the unique point (UA, UB) on the graph of the function g(.; z)
where some tangent has slope —UB/UA (cf. Figures 2.2 and 2.3).
220 Commitment Tactics

argue that player B can benefit from a decrease in her partial commitment
to z'B = ZB — £ for some e such that 0 < e < ZB- Consider Figure 8.2 in
which the graphs TNQM and TN'Q'M1 respectively illustrate the utility
possibility sets Q(z) and fi(z'), where z1 — (zA,z fB). Note that the line
segment N'Q' has the same slope (namely, —7) as the line segment NQ (cf.
equation 8.5). Hence, if the pair (PA(Z), PB{Z)) lies in the interior of the line
segment NQ, such as at point D, then there exists an e G (0, ZB) such that
the payoff pair (PA{Z'),PB(Z')) is at the point Df that lies on the ray OD.
By a similar argument, if the payoff pair (PA(Z),PB(Z)) lies in the interior
of the line segment QM, then player B can (also) benefit from a decrease
in her partial commitment.
Now suppose that the payoff pair (PA(Z),PB(Z)) is at point Q, where
there is a kink in the graph TNQM. It the ratio —PB{Z)/PA{Z) is equal to
the slope of the line segment QM (namely, — (1 + fc#)), then (by a similar
argument to that given above) it follows that player B can benefit from a
decrease in her partial commitment. Now consider —PB(Z)/PA(Z) > — (1 +
ks)- Since the slope of the line segment Q'Mr is (also) equal to — (1 + &#),
it follows by continuity that there exists an e G (0, ZB) such that the slope
of the ray OQf is (also) strictly less than 1 + ks- Consequently, the payoff
pair (PA(Z'),PB{Z')) lies either at point Qr or to the left of point Q'\ thus,
PB(Z') > PB{z).
Now assume that g(0;z) < 0, in which case PA(Z) = PB(Z) — 0. If
ZA + ZB < 2TT: then there exists an i (i = A or i = B) such that n — Z{ >
0. And, hence, player j (j 7^ i) can benefit by unilaterally deviating to
Zj = 7r — Z{. Finally, consider the case z — (7r,7r). Let z'{e) = (TT, e) where
0 < e < 7T. Since g(0; zf(0)) > 0 and since g(0; z'(e)) is continuous in e, there
exists an ef G (O,TT) such that g(0;zf(ef)) > 0. And, hence, player B can
benefit by unilaterally deviating to z'B — e1. •
The following result establishes that any pair of exactly-compatible par-
tial commitments that does not satisfy a particular condition is not a NE.
Lemma 8.3. // ZA + ZB = TT and ZB/ZA ^ 7, where 7 = (1 + &B)/(1 + kA)}
then the pair z = (z^, ZB) is not a Nash equilibrium.
Proof. By contradiction. First suppose that there exists a pair (ZA,ZB)
such that ZA + ZB — TT and ZB/ZA > 7 which is a Nash equilibrium. I now
show that player A can benefit from an increase in her partial commitment
8.2 The Basic Model 221

to z'A = ZA + £ for some e such that 0 < e < ix — ZA- Consider Figure
8.3. The graph TNQM illustrates the utility possibility set £l(zf), where
zf = {z'A, ZB)- Since (by hypothesis) the slope of the ray OV (which equals
ZB/ZA) strictly exceeds 7, it follows (by continuity) that there exists an
e G (0,TT — ZA) such that the slope of the ray OQ is (also) strictly greater
than 7. Hence, since the slope of the line segment NQ is equal to —7
(cf. equation 8.5), the Nash bargaining solution of the bargaining problem
(Q(zf), d) — where d = (0,0) — lies on the line segment QM. Consequently,
PA(Z') > z'A, which implies that PA{Z') > ZA (a contradiction). A symmetric
argument establishes a contradiction when ZB/ZA < 7- •

F i g u r e 8.3: The geometry behind the proofs of Lemma 8.3 and Proposition 8.1.

I now use Lemmas 8.1-8.3 to characterize the unique NE.

222 Commitment Tactics

Proposition 8.1. The basic model with commitment tactics has a unique
Nash equilibrium, namely
z\ = and 4 = ,
A B
1+7 1+7
where 7 = (1 + kB)/{\ + kA)-
Proof. Lemmas 8.1-8.3 imply that if a pair z = (ZA,ZB) is a NE, then
ZA + ZB = TV and ZB/ZA — 7; that is, the pair is as defined in the proposition.
I now establish that this pair is a NE. First, suppose player A unilaterally
deviates to zA = z\ + e, where e is such that 0 < e < TV — zA. The graph
TNQM in Figure 8.3 illustrates the utility possibility set Cl(z'), where z' =
(z'A,z*B). Since the slope of the line segment NQ is equal to —7, it follows
that the payoff pair (PA{Z'),PB(Z')) is at the point G that lies on the ray OV.
Hence, the deviation is not profitable. Second, it is straightforward to show
that player A does not benefit from a unilateral deviation to any z'A such
that 0 < zA < zA. Finally, by a symmetric argument it follows that player
B does not benefit from a unilateral deviation to a partial commitment
z'B / z*B. •

8.3 Discussion
The model formulated above is explicitly built upon Nash's bargaining so-
lution. However, it can also be interpreted as an extension to Rubinstein's
model, in which the bargainers make partial commitments before engaging
in the offer-counteroffer process. It should be noted that this model is one
example of the framework of two-stage games of 'claims' and 'concessions'.
One could, for example, formulate second-stage 'concession' games that are
not based on Nash's bargaining solution or Rubinstein's model — as is done
in Sections 8.5 and 8.6.

8.3.1 Properties of the Equilibrium

I now turn to a discussion of some key properties of the equilibrium of the
model analysed above. I begin by emphasizing the conclusion that the model
has a unique Nash equilibrium. This is a remarkable property of the model;
it demonstrates that commitment tactics can be a decisive force in resolving
the fundamental indeterminacy that characterizes the bargaining situation.
8.3 Discussion 223

In the unique Nash equilibrium, neither player revokes her partial com-
mitment; the unique NE is Pareto efficient. In particular, the model does
not predict any disagreement in equilibrium.
Player z's equilibrium partial commitment — which is identical to the
equilibrium share of the cake she obtains — is strictly increasing in ki, and
strictly decreasing in kj (j / i). Thus, if a high ki is interpreted as a weak
bargaining position for player i (as one's intuition may suggest), then this
model supports the viewpoint that, in bargaining, weakness can often be a
source of strength. 5
The result that player i's 'bargaining power' increases as her marginal
cost ki of revoking a partial commitment increases seems counterintuitive. 6
As player i's marginal cost of revoking increases ceteris paribus (that is,
relative to player j ' s marginal cost of revoking), it is relatively less costly, at
the margin, for player j to revoke her partial commitment. It is 'as if player
i's hands are tied behind her back in a relatively tighter manner. Thus, if,
for example, the players make incompatible commitments, then it is player j
who is more 'able' (and thus, more likely) to revoke her partial commitment
— and, hence, player i's payoff increases as her marginal cost of revoking
increases.7

8.3.2 Relationship with Nash's Bargaining Solution

Since the unique NE partial commitments are exactly compatible, it means
that the equilibrium payoff pair (PA(Z*),PB(Z*)) = (z*A,z*B). I now show
that the equilibrium payoff pair (z*A, z^) is identical to the asymmetric Nash
bargaining solution of an appropriately defined bargaining problem (O, d)
with an appropriate value for the parameter r. 8
5
This basic insight about bargaining situations is informally, but persuasively, articu-
lated — and discussed at some great length with many examples — by Thomas Schelling
in his classic essay on bargaining (cf. Schelling (I960)).
6
In the context of non-strategic decision making environments, the higher the cost the
lower typically is the decision maker's payoff. However, as is often the case, what is valid in
such non-interactive decision situations is invalid in strategic settings, such as bargaining
situations.
7
The insight is most transparent when, for example, ki is arbitrarily close to infinity
and kj is arbitrarily close to zero.
8
It should be noted that this is quite a remarkable result, notwithstanding the fact
that Nash's bargaining solution is used to determine the players' payoffs when they make
224 Commitment Tactics

It is natural to define Q as the set of possible utility pairs available

to the players before they make their partial commitments, which, since
player z's utility from a partition of the cake equals the share that she
obtains, means that ft = {(UA,UB) : 0 < ^ < TT and UB — TT — UA}-
Furthermore, since if the players fail to reach agreement then each player
obtains a payoff of zero, it is appropriate to define the disagreement point
d = (0,0). Applying Corollary 2.3, it follows that the unique NE utility
pair (z*A, Zg) is identical to the asymmetric Nash bargaining solution of the
bargaining problem (Q, d) with r = (1 + kj\)/{2 + k,A + &B). Notice that
r (resp., 1 — r ) is strictly increasing (resp., decreasing) in UA and strictly
decreasing (resp., increasing) in ks — which makes sense since r and 1 — r
respectively reflect the bargaining powers of players A and B. If kA = &s,
then r = 1 — r — 1/2, and the players split the cake equally. Indeed, if
kA — ks-, then the payoff pair (2:^,2^) is identical to the Nash bargaining
solution of (fi, d) — which makes sense since the Nash bargaining solution
(cf. Definition 2.1) embodies a symmetry axiom that would not be satisfied
by the model with commitment tactics unless kA = ks-

8.3.3 Comparison with the Nash Demand Game

In the Nash demand game, two players A and B simultaneously and inde-
pendently choose numbers from the closed interval [0,TT]. Let Z{ denote the
number chosen by player i. If the players' demands are incompatible (i.e.,
ZA + %B > TT)5 then the players fail to reach agreement over the partition
of the cake, and each player obtains a payoff of zero. The interpretation is
that player i has taken actions which irrevocably commit her not to accept
a share strictly less than Z{. If, on the other hand, the players' demands
are compatible (i.e., ZA + ZB < TT), then player % obtains a payoff equal to
Z{. It is straightforward to verify that the Nash demand game possesses a
continuum of Nash equilibria. In particular, any pair of exactly compatible
demands is a Nash equilibrium.
At the heart of the Nash demand game lies the assumption that the

incompatible partial commitments. In fact, letting Ui(xi) denote the utility that player
i obtains from xi units of the cake — where 11^ is linear (i.e, Ui(xi) — x%) in the basic
model studied here (cf. Equation 8.1) — I shall show in Section 8.3.5 that this result does
not hold if for some i (i = A or i = B), Hi is strictly concave (and, of course, strictly
increasing).
8.3 Discussion 225

bargainers can make irrevocable commitments. The model studied in Section

8.2 can be interpreted as a perturbation to the 'commitment' structure of
the Nash demand game, where this perturbation becomes arbitrarily small
in the limit as both kA and ks diverge to plus infinity. It follows from
Proposition 8.1 that in the limit as kA —> oo and /c# —> oo, keeping the
ratio kA/ks constant at some positive value 7*, the unique NE equilibrium
partial commitments z\ and z*B converge respectively to TT7*/(1 + 7*) and
TT/(1 + 7*). This result, besides selecting a unique equilibrium in the Nash
demand game, provides an interpretation of the indeterminacy (multiplicity
of equilibria) in the Nash demand game: the ratio kA/ks is an important
determinant of the equilibrium demands, but is undetermined in the Nash
demand game because, in this game, it is 'as if kA = +00 and ks = +00.
It is interesting to note that these results are somewhat analogous to
the results obtained in Rubinstein's bargaining model (cf. Chapter 3). If
the players' discount rates, VA and r#, are such that rA — ^B — 0, then
Rubinstein's model has a continuum of subgame perfect equilibria. But if
rA > 0 and r# > 0 then there exists a unique subgame perfect equilibrium,
and moreover, even in the limit as TA —> 0 and r^ —» 0 the unique equilibrium
partition depends on the ratio VAI^B (cf. Corollary 3.1). By studying the
limiting model (k,A —> 00 and ks —> 00) rather than the model at the limit
[kA — ks = +00), one discovers the significant role played by the ratio
kA/ks- In fact, it is rather intuitive that in order to generate a unique
equilibrium partition, one needs to know how 'strong' player A is relative to
player B. Since this question cannot be addressed when kA = ku — +00,
it is not surprising therefore that every partition can be supported as an
equilibrium in the Nash demand game. The metaphor of a boxing match
— used in the context of Rubinstein's model (cf. The discussion at the
end of Section 2.2.1) — is once again useful: who wins the match depends
crucially on which player is relatively stronger. The absolute magnitudes of
the players' strengths are relevant to the final outcome only in so far as they
determine the relative magnitude of their strengths.

8.3.4 Robustness

I now show that the unique NE is robust to two extensions of the model.
226 Commitment Tactics

Sequential Moves
How would the results obtained here change if one of the players, say player
A, were to choose her partial commitment ZA after observing the partial
commitment ZB chosen by player B. Would player B benefit from being
the 'first-mover'? I now sketch an argument which establishes that the out-
come is unaffected: that is, the unique subgame perfect equilibrium partial
commitments in the sequential-move game are identical to the unique Nash
equilibrium partial commitments in the simultaneous-move game. The ar-
gument exploits the property of the unique Nash equilibrium that the ratio
=
Z Z
B/ A 7 (°f- Proposition 8.1).
If ZB < z*B, then ZB/{TT — ZB) < 7- Using geometric arguments similar
to those contained in the proofs of Lemmas 8.2 and 8.3, it follows that
player A's unique best response is to set ZA = TT — ZB- NOW suppose that
player B chooses ZB > z%- This implies that ZB/{TT — ZB) > 7- It can be
verified, again using such geometric arguments, that player A's unique best
response is to set ZA equal to the unique solution of the following equation:
g[zA\ Z)/ZA = 7) where g(zA] z) = (1 + kB){^ - zA) - kBzB- Hence, player
A's unique best response to a ZB > zB is: ZA = [TT + /^(TT - ZB)]/J(2 + kjC).
For each ZB > ^ , player I?'s payoff equals g(zA] z) with ZA being player
A's unique best response to ZB- Straightforward computation, therefore,
establishes that player £Ts optimal choice is z^.9

On the Payoff Rule

I have assumed that if the players' partial commitments are compatible
(i.e., ZA + ZB < TT), then neither player revokes their respective partial
commitment. It may, however, be reasonable (in some contexts) not to
make this assumption. In that case one can define the players' payoffs using
Nash's bargaining solution, as I did if ZA + ZB > TT. With such a change in
the rule that defines the payoffs, the results are unaffected: Proposition 8.1
continues to describe the unique Nash equilibrium.
9
The intuition for this result, that the sequential-move game and the simultaneous-
move game have the same equilibrium outcome, comes partly from the observation that
each player's indifference curves are kinked at certain key points. For example, player
i?'s indifference curve through (z*A,z*B) is kinked at that point. This observation can be
obtained, at least informally, through the geometric arguments.
8.3 Discussion 227

8.3.5 A Generalization

In the model studied in Section 8.2 each player's cost of revoking function is
defined in equation 8.2; the cost to player i of revoking a partial commitment
Zi and obtaining a share Xi < Z{ is proportional to the difference zi — X{.
I now study a generalized version of the model in which player i's cost of
revoking function satisfies the conditions stated below in Assumption 8.1 —
which allow it to be non-linear.

Assumption 8.1. For each i, the cost of revoking function d : [0,1] x

[0,1] —> $ft is continuous. Furthermore, (i) for any pair {x^Zi) such that
Xi > Zi, Ci(xi,Zi) = 0; (ii) for each z\ G (0,1], d is strictly decreasing,
convex and twice continuously differentiate in x\ on the interval (0, zi)\ (iii)
for each X{ G [0,1), d is strictly increasing, convex and twice continuously
differentiable in z\ on the interval {x{, 1); and (iv) for each z^ G (0,1], the
left-hand derivative of d w.r.t. Xi, evaluated at X{ — z^ is strictly negative
and is independent of the value of Zi (that is, there exists a constant k{ > 0
such that for any Z{ G (0,1], C\{zi — \ Zi) = —ki).

The continuity and monotonicity assumptions on d capture the no-

tion that the cost to player i of revoking her partial commitment Z{ and
obtaining a share X{ < Z{ depends on the amount by which she revokes (de-
viates from) her partial commitment. The assumption that the marginal
cost —C\(zi — \Zi) of revoking a partial commitment is strictly positive is
quite reasonable. However, the assumption that this marginal cost is a con-
stant and independent of Z{ is made for simplicity. Having said this, it is
not clear whether it is reasonable to assume that it is relatively easier (less
costly) to revoke, at the margin, moderate (or extreme) partial commit-
ments. Below I shall briefly discuss whether and how the results obtained
here would be affected by any alternative assumption describing the depen-
dence of the marginal cost —C\(zi—\Zi) on Z{. An example of a cost of
revoking function that satisfies Assumption 8.1 is: d = 0 if x\ > Z{ and
d = ki{zi — Xi) + (qi/2)(zi — Xi)2 if Xi < z^ where ki > 0 and qi > 0.
I also generalize the model studied in Section 8.2 by assuming that the
utility Ui(xi,Zi) to player i from obtaining a share x\ G [O,TT] of the cake,
given that she partially committed herself to z^ G [0, TT], is
228 Commitment Tactics

where I now allow the utility from consuming a share Xi to be Hi(xi). It is

assumed that H satisfies the conditions stated below in Assumption 8.2.
Assumption 8.2. For each z, the function Hi is twice continuously dif-
ferentiable, strictly increasing and concave. Furthermore, without loss of
generality, I normalize and set 11^(0) = 0 and II^TT) = 1.
The result stated below establishes the uniqueness of the Nash equilib-
rium, and provides a characterization of this equilibrium.
Proposition 8.2. The generalized model with commitment tactics described
here in which Assumptions 8.1 and 8.2 are satisfied has at most a unique
Nash equilibrium. In this equilibrium the players? partial commitments z\
and Zg are the unique solution to zA + ZB =ftand

U'B(zB) + kB , ,
{ }
= k A' '
uA(zA) n
Proof. See Muthoo (1996, Proposition 2). •
This unique NE is similar to that derived in the basic model studied in
Section 8.2 (cf. Proposition 8.1). In fact, if n^ is linear (that is, Ui(xi) — x\
for i = A, B) and C{ (i = A,B) satisfies Assumption 8.1, then the equi-
librium is exactly the same. Indeed, a key (and rather interesting) feature
of the equilibrium partial commitments is that they depend only upon the
players' marginal costs, kA and &#, of revoking their respective partial com-
mitments. The equilibrium partial commitments are independent of other
features of the cost of revoking functions, including, for example, the degrees
of convexity of these functions.
Before I turn to the issue of the existence of this unique Nash equilibrium,
I briefly discuss the properties of this unique NE when for some i (i = A
or i = B), Hi is non-linear — which, given Assumption 8.2, means that
H is strictly concave — and d (i = A,B) satisfies Assumption 8.1. As
in the linear case (i.e., when for i = A,B, Hi(xi) = x\ and C% satisfies
Assumption 8.1), the unique equilibrium partial commitments are exactly
compatible, and, hence, the equilibrium is Pareto efficient. Furthermore, as
in the linear case, player i's equilibrium partial commitment z* — which
is identical to the share of the cake she obtains — is strictly increasing
in ki and strictly decreasing in kj (j ^ i). However, unlike in the linear
8.3 Discussion 229

case, the equilibrium payoff pair ( n ^ ( ^ ) , HB(Z^)) is not identical to any

asymmetric Nash bargaining solution defined on the appropriate bargaining
problem (fi,d), where Q is the set of possible utility pairs available to the
players before any partial commitments are made,10 and d = (0,0). n
I now turn to the issue of the existence of this unique Nash equilibrium.
Since, given Assumptions 8.1 and 8.2, the payoff function Pi need not be
quasi-concave in Z{ (although it is continuous in z and although the strategy
sets are compact), it is not possible to appeal to the standard theorems on
the existence of (pure-strategy) NE. Indeed, it turns out that the degree of
convexity of C{ is important to establish the existence of the unique NE.12
The next result, stated below, establishes the existence of this unique NE
under some additional set of conditions (stated below in Assumption 8.3)
and under the assumption that the degree of convexity of d is sufficiently
small.
Assumption 8.3. For each i and for any pair (a^, zi) such that xi < Z{\ (i)
-C}{xuZi) = C?(xi,Zi) and (ii)
This set of assumptions embodies the notion that X{ and Zi are to be
treated symmetrically in the cost of revoking function C{. For example,
if the cost of revoking Z{ and obtaining X{ < Z{ is a (possibly non-linear)
function of the difference Z{ — Xi, then Assumption 8.3 is satisfied.
Proposition 8.3. // Assumptions 8.1-8.3 are satisfied, and if for each i
and any pair (xi,Zi) such that x% < zi, C}l{xi,Zi) < kf/Ili(z*), then the
unique Nash equilibrium described in Proposition 8.2 does exist.
Proof. Muthoo (1996, Proposition 3). •
Therefore, if the 'degree of convexity' C}1 of the function C{ is strictly
less than fc?/H(z*), then the unique NE exists. Consider, for example, the
10
That is, ft — {(UA,UB) '• there exists a partition of the cake (XA,XB) such that
UA{XA) = UA and UB(XB) = uB}.
11
It follows from Definition 2.3 that the partition (xjJjXg) of the cake associated
with the asymmetric Nash bargaining solution defined on the bargaining problem (Q,d)
with parameter r is the unique solution to: XA + XB = TT and UB(XB)/^A(XA) =
(1 — r)U.fB(xB)/rU.fA(xA)- The desired conclusion follows because there does not ex-
ist a r such that for any exactly compatible partition, (1 — T)HB(XB)/TUA(XA) =
(n' B (x fl ) + kB)/(nA(xA) + kA).
12
In Muthoo (1996), an example is constructed to show that this unique NE fails to
exist if the degree of convexity of d is sufficiently large.
230 Commitment Tactics

following function, which satisfies Assumptions 8.1 and 8.3: C{ = 0 if X{ > Z{,
and \ix{ < Z{ then d — ki(zi — Xi) + (qi/2)(zi — Xi)2, where ki > 0 and q\ > 0.
Furthermore, assume that Hi(xi) = X{. Thus, if qi < /C?(2 + A:^ + A:JB)/(1 + ^ ) ,
then the unique NE exists. Notice that this bound on qi is strictly increasing
in kA (and in £;#), and it converges to plus infinity as either kA or ks or
both kA and ks tend to plus infinity. Clearly, therefore, for a large class of
convex cost of revoking functions satisfying Assumptions 8.1 and 8.3, the
unique NE exists.
I now briefly discuss whether and how the results contained in Proposi-
tions 8.2 and 8.3 are affected by any alternative assumption describing the
dependence of the marginal cost —C\[zi—\Zi) on z\. The characterization
of an equilibrium is essentially unaffected: ZA + ZB = TT and (8.6) are to be
satisfied, but with ki replaced by —C\{zi — \ Zi). Continuity of the marginal
cost in Zi ensures that the model possesses at most an odd number of Nash
equilibria. If in addition, for example, the marginal cost —C\(zi — \Zi) is
non-increasing in Zi, then uniqueness is (re)obtained. The necessary and
sufficient conditions for the existence of an equilibrium — stated in Muthoo
(1996, Lemma 4A) — are unaffected; however, whether they would be sat-
isfied depends on how the marginal cost varies with Z{.

8.4 An Application to Delegation

A simple extension of the model studied in Section 8.2 can provide some
insight on why players in a bargaining situation often delegate to some
other persons the job of negotiating on their behalfs. The extension involves
endogenously determining the values of the marginal costs kA and ks-
Formally, before the game described in Section 8.2 is played, the two
players, A and I?, simultaneously and independently choose numbers kA > 0
and ks > 0, respectively.
The interpretation is as follows. There is a large number of heterogeneous
professional negotiators who are up for hire. Each negotiator is characterized
by a number k > 0. If player i chooses ki = 0, then this means that she
does not hire a professional negotiator, but instead plays the bargaining
game herself. If, on the other hand, she chooses ki > 0, then she hires a
professional negotiator with characteristic ki to play the bargaining game on
her behalf. The point is that for each player revoking a partial commitment
8.4 An Application to Delegation 231

is not costly. But for a professional negotiator with characteristic ki the cost
of revoking a partial commitment is described by equation 8.2. The wage
of a professional negotiator with characteristic k is fixed and is denoted by
W(k), where W is a differentiate, strictly increasing and convex function,
with W(0) = 0.
From Proposition 8.1, it follows that the payoff to player i if she chooses
ki and player j (j / i) chooses kj is 13

Differentiating Zi w i t h respect t o ki, it follows t h a t

dZ% (l + fc> Jir,,,^

Hence, since Z{ is strictly concave in ki, it follows that if kj is such that

W'(0) > TT(1 + kj)/{2 + kj)2 then player i's unique best response is ki = 0;
and if kj is such that W'(0) < TT(1 + kj)/{2 + kj)2 then player i's unique
best response is the unique solution to

vlJ
(2 + ki + kj)*

Since TT(1 + kj)/{2 + fcj)2 is strictly decreasing in fcj, it follows that if

W'(0) > TT/4, then for any kj > 0, player i's unique best response is ki = 0.
Hence, if W'(0) > TT/4, then the unique Nash equilibrium is kA = ks = 0.
If Wf(Q) < TT/4, on the other hand, then in any Nash equilibrium kA > 0
and ks > 0. In particular, any solution to the following first-order conditions
is a Nash equilibrium
/-I i 7 \

- W'(kA) = 0

- W'(kB) = 0.

13
Although Proposition 8.1 is valid for kA > 0 and ks > 0 (cf. equation 8.2), it is trivial
to show that if for some i, ki = 0 then (although there may exist multiple Nash equilibria)
in any Nash equilibrium the payoffs to players A and B are respectively z\ and z*B — as
stated in Proposition 8.1.
232 Commitment Tactics

This system of equations has a unique solution in which k,A = ks = &*,

where A:* > 0 is such that TT/4 = W'(k*).u Thus if W'(0) < TT/4, then in
the unique Nash equilibrium both players hire identical negotiators, which
implies that each player ends up receiving one-half of the cake; and, since
each player has to also pay a wage of W(fc*), each player's equilibrium
payoff is strictly less than TT/2 — which is the payoff to each player if both
players do not hire any negotiators, but instead bargain themselves. This
means that the unique Nash equilibrium is Pareto inefficient. Although
equilibrium delegation is thus Pareto inefficient, for each player hiring a
costly negotiator strictly dominates bargaining by herself. Delegation occurs
not because the players are made better off, but because it is consistent with
Nash equilibrium behaviour.15

8.5 Uncertainty and Simultaneous Concessions

The model studied in Section 8.2 has shown that the deployment of com-
mitment tactics influences the equilibrium distribution of the 'gains from
trade' in a bargaining situation. The unique equilibrium outcome, however,
is Pareto efficient: the equilibrium partial commitments are not incompat-
ible. As such the deployment of such tactics do not generate disagreement
in a bargaining situation.
In this section I inject some uncertainty into the commitment process:
when a player chooses the bargaining position to which she partially com-
mits herself, she is uncertain about the costs of revoking such partial com-
mitments. The key message of this section is that with uncertainty in the
commitment process the equilibrium partial commitments are incompatible,
and with strictly positive probability the players fail to reach agreement.
14
Suppose there exists a solution to the first-order conditions in which UA > ks- This
implies that (1 + kB)7r/(2 + kA + kB)2 < (1 + &U)7T/(2 + /CA + kB)2. Hence, using the first-
order conditions, this in turn implies that W (kA) < W(kB), which is a contradiction.
15
This result may be interpreted in terms of a Prisoners' Dilemma type game. Assume
that each player has to choose between two values of k: k = k* (hire a particular type of
professional negotiator) and k = 0 (do not hire the professional negotiator). It thus follows
from the analysis above that for each player delegation is a strictly dominant strategy,
which leads to a Pareto inefficient outcome: as in the classic Prisoners' Dilemma game,
individual rationality conflicts with collective rationality.
8.5 Uncertainty and Simultaneous Concessions 233

8.5.1 A Model with Simultaneous Concessions

I study a model of bargaining based on commitment tactics with the fea-

ture that when the players simultaneously and independently choose their
respective partial commitments, both players do not know either players'
costs of revoking such partial commitments. This is a plausible feature of
many bargaining situations, because the costs to a player of revoking a par-
tial commitment may only become known to her as/after she actually takes
actions that partially commit her to some bargaining position.
In order to bring out the main insight in as transparent and simple a
way as possible, I assume that the cost to each player of revoking any partial
commitment is either zero or infinite. I model this assumption as follows.
Player i's cost of revoking function is defined in equation 8.2. But, at the
first stage, both players are uncertain about the exact values of both k^ and
ks] with probability p^, ki = 0, and with probability 1 —pi, ki — +oo, where
0<Pi < 1.
At the first stage, given the players' uncertainty about the values of kA
and &;#, the players simultaneously and independently choose their respec-
tive partial commitments ZA and ZB- Immediately after the partial commit-
ments have been made, they become known to both players. Furthermore,
the values of kA and ks are then drawn (randomly and independently), and
player i learns the value of ki but does not learn the value of kj.
If the partial commitments ZA and ZB are compatible (i.e., ZA + ZB < TT),
then — as in the model studied in Section 8.2 — neither player revokes her
partial commitment: the share X{ of the cake obtained by player % is such that
Xi > Zi. Specifically, the share obtained by player i is given by A^(z), where
XA and XB are any functions such that XA{Z) > ZA and XB{Z) = TT — XA{Z) >
ZB- For example, it may be assumed that Xi(z) = ^+(TT—ZA —ZB)/2- Hence,
if ZA + ZB < 7T, then player z's payoff is Xi(z).
If, on the other hand, the partial commitments ZA and ZB are incompat-
ible (i.e., ZA + ZB > TT), then play moves to the second stage of the game:
the players simultaneously and independently decide whether or not to back
down from their respective partial commitments. If both players choose to
back down, then each player obtains one-half of the cake. If player i only
backs down, then her share Xi — TT — Zj and player j ' s share Xj = Zj. And
if neither player backs down, then the players fail to reach agreement (and
234 Commitment Tactics

each player obtains no share of the cake).

The utility payoff to player i if she obtains a share X{ of the cake, given
that she chose Z{, is defined by equation 8.1. This completes the description
of the two-stage game. Notice that — unlike the model studied in Sec-
tion 8.2 — the second stage is not based on Nash's bargaining solution or
Rubinstein's model: it is a simultaneous, one-shot, concession game.

8.5.2 An Example: Two Bargaining Positions

In this section I analyse the above model with the assumption that for
each i, zi can take only two values, namely, 0 and 2TT/3. This additional
assumption considerably simplifies the analysis of the model. In particular,
it proves straightforward to derive the main result that for a large class
of values of the probabilities PA and PB-> in the unique Perfect Bayesian
Equilibrium (PBE) ZA = %B — 2TT/3, and with probability (1 — P A ) ( 1 — PB)
the players fail to reach agreement. 16
Suppose that at the first stage player A chooses ZA = 2TT/3 and player B
chooses ZB = 2TT/3. This means that the game proceeds to the second stage,
where player i knows the value of ki but does not know the value of kj. In
any Bayesian Nash Equilibrium (BNE) of the second stage game, if k{ = +oo
then player i does not back down. Letting qi denote the probability with
which player i backs down if k\ = 0, the (conditional) expected payoff to
player i if ki = 0 from backing down, E{[BD\ and from not backing down,
\ are respectively

= Pj[qj(ir/2) + (1 - <&)(T/3)] + (1 - PJ)(TT/3) and

Ei(D)=pjqj(2ir/3).

It is trivial to verify that if pj < 2/3, then for any qj G [0,1],

Ei(D). That is, \ipj < 2/3, then backing down is a strictly dominant action
for player i if ki = 0. 17 Hence, if PA < 2/3 and PB < 2/3, then the second
stage subgame (after the players choose ZA = 2TT/3 and ZB = 2TT/3) has a
unique BNE: each player i backs down if ki = 0 and does not back down if
16
In the next section I extend this kind of result to the general case when for each i, zi
can take any value in the interval [0, TT].
17
This result is fairly intuitive. If the maximum (unconditional) probability with which
player j backs down — which equals pj — is not too large, then it is optimal for player i
to back down if ki = 0.
8.5 Uncertainty and Simultaneous Concessions 235

ki = +00. In this unique BNE the (conditional) expected payoffs to player

i if ki = 0 and if hi = +00 are respectively

E{(ki = 0) =Pj(n/2) + (l-Pj)(Tr/3) and Ei(h = +00) = PJ(2TT/3).

I now proceed backwards to characterize the Nash equilibria of the first

stage of the game. First, notice that (ZA-, ZB) — (0, 0) is not a Nash equilib-
rium, because player i can benefit from unilaterally deviating to Z{ = 2TT/3
— since A^(z^) > 2TT/3, where VA = (2TT/3, 0) and vB — (0, 2TT/3). Given the
unique BNE of the second stage game, the (unconditional) expected payoff
to player i at the first stage if ZA = 2TT/3 and ZB = 2TT/3 is

£ Z (2TT/3, 2TT/3) = piE^h = 0) + (1 - pl)El(kl = +00),

where Ei(ki = 0) and Ei(ki = +oc) are defined above. Hence, (ZA,ZB) =
(2TT/3, 2TT/3) is the unique Nash equilibrium if and only if

EA(27T/3,2TT/3) > \A(VB) and EB(2TT/3,27T/3) > \B(VA).

Since A^(^) > 2TT/3 implies that A^(^) < TT/3, (ZA,ZB) = (2TT/3, 2TT/3)
is the unique Nash equilibrium if and only if £ ^ ( 2 T T / 3 , 2TT/3) > TT/3 and
£;£(27r/3,27r/3) > TT/3. That is, (zA,zB) = (2TT/3, 2TT/3) is the unique Nash
equilibrium if and only if the following two inequalities hold

2pA + ^PB ~ 3PAPB > 2 (8.7)

2pB + ApA - 3pAPB > 2. (8.8)

It is straightforward to show that there exists a large set of values of PA

and PB that satisfy inequalities 8.7 and 8.8 — which I denote by J, and
which is shown in Figure 8.4. Proposition 8.4 summarizes the main result
obtained here.

Proposition 8.4. If PA and pB satisfy inequalities 8.1 and 8.8 — that is,
the pair (PA^PB) belongs to the set J shown in Figure 8.4 — then in the
unique perfect Bayesian equilibrium of the example studied here, ZA = ZB =
2TT/3 ; and with probability (1—PA){1—PB) the players fail to reach agreement.

This proposition illustrates the main message of this section: if the val-
ues of PA and pB are neither too high nor too low, then the uncertainty in
 Commitment Tactics

pB = (2 - 4p A )/(2 -

-PA
1/3 P 1/2 2/3 1

F i g u r e 8.4: The area between pA = 2/3, pB = 2/3, pB = (2 - 2pA)/(4 - 3pA) and

PB = (2 — 4:PA)/(2 — 3pA) — where the latter two functions are strictly decreasing and
concave — is the set J\ that is, the set of all pairs (PA,PB) which satisfy inequalities 8.7
and 8.8. The point where the latter two functions intersect is (p,p), where p = 1 — l/\/3 ~
0.42.

the commitment process necessarily implies that players' equilibrium par-

tial commitments are incompatible, and with strictly positive probability the
players fail to reach agreement. The intuition behind this result is straight-
forward. If pj is not too high, then in the second stage subgame it is optimal
for player i to back down if k{ = 0. Consequently, if pj is not too small, then
it is optimal for player i to choose z\ — 2ir/3, because with a sufficiently
large probability player j backs down when k{ — 0.

8.5.3 A Generalization

I now generalize the result derived above to the case when, for each i, Z{ G
[0, TT]. In particular, I show that there exists a large class of values of PA and
8.5 Uncertainty and Simultaneous Concessions 237

PB such that in any PBE, the players' partial commitments are incompatible,
and with strictly positive probability the players fail to reach agreement.
Suppose that at the first stage the players' partial commitments ZA and
ZB are incompatible (i.e., ZA~\-ZB > TT). This means that the game proceeds
to the second stage, where player i knows the value of ki but does not know
the value of kj. In any BNE of the second stage game, if k\ — +00 then
player i does not back down. Letting qi denote the probability with which
player i backs down if ki = 0, the (conditional) expected payoff to player i
if ki = 0 from backing down, Ei(BD), and from not backing down, Ei(D),
are respectively

-zj) and (8.9)

Ei{D)=pJqjzi. (8.10)

I now derive the following proposition.

Proposition 8.5. //1/2 < pA < 2/3, 1/2 < pB < 2/3 and
PAPB
>msx{l-pA,l-pB} (8.11)
2[l - {1 - pA)pB - (I - PB)PA]
then in any PBE the equilibrium partial commitments z*A and z*B are in-
compatible (that is, z\ + Zg > TT), and with strictly positive probability the
players fail to reach agreement.

Since inequality 8.11 is satisfied when PA — PB — 2/3, and since the

left-hand side and the right-hand side of this inequality are continuous in
PA and PB, there exists pA < 2/3 and ps < 2/3 such that the inequality
is satisfied for any pair (PA,PB) £ (PA52/3) x (j>#,2/3). This establishes
that there exists a set of values of pA and PB under which the conclusion
contained in this proposition is valid.

Proof. Suppose, to the contrary, that there exists a PBE in which the
equilibrium partial commitments z\ and z^ are exactly compatible (i.e.,
Z
A + ZB = 7 r )- 1 8 This implies that there exists an i (where i = A or i — B)
such that z* < TT/2 — and, thus, z* > TT/2 (j ^ i).
First, suppose that z* < npi — which is possible since pi > 1/2. I show
that player j can benefit from a unilateral deviation to a partial commitment
18
Since Lemma 8.1 is valid in this model as well, there does not exist a PBE in which the
equilibrium partial commitments z\ and z*B are more than compatible (i.e., z\ + z% < TT).
238 Commitment Tactics

zfj such that z* < z'- < vr and piz'- > z*. Consider the second stage subgame
following this unilateral deviation. Since z\ < TT/2 and zr- < TT, it follows
from (8.9) and (8.10) that for any qj G [0,1], Ei(BD) > Ei{D) — which
implies that in the BNE of this subgame q*(zf) — 1, where zr = (z*,z');
that is, in equilibrium player i backs down with probability one if k{ = 0.
Hence, the (unconditional) expected payoff to player j from this unilateral
deviation equals
Ej(z')=pjm3x{pi(n/2) + (1 - PI)(TT - z[),p%z'3} + (\-p3)p%z].
Hence, Ej(z') > PiZ^ which implies that Ej(z') > z*, which is a contradic-
tion (since the equilibrium expected payoff to player j , Ej(z*A, z^) = z*).
Now suppose that z* > npi, which implies that z* < TT(1 — pi) — and
which in turn implies, since pi > 1/2, that z* < TT/2. I show that player i can
benefit from a unilateral deviation to a partial commitment z[ — z* + e for
some e E (0, TT/2 — z*) — which implies that z[ < TT/2. Consider the second
stage subgame following this unilateral deviation. From (8.9) and (8.10), it
follows that Ej(BD) = plqi(Tr/2) + (1 - ^ ) ( T T - ^ ) and E3{D) = piqiz*. It
is trivial to show that for any qi G [0,1], Ej(BD) is minimized at qi = 1 and
Ej(D) is maximized at qi = 1. The assumption 1/2 <p% < 2/3 implies that
the minimized value of Ej(BD) — which equals ^ ( T T / 2 ) + (1 — Pi)(7T — z[)
— strictly exceeds the maximized value of Ej(D) — which equals PiZ*.
This implies that in the BNE of this subgame, qj(z') = 1 — that is, in
equilibrium player j backs down with probability one if kj = 0. Given this
result, it follows that there exists an ei > 0 such that for any e G (0,ei),
the (conditional) expected payoff to player i from backing down if ki — 0
— which (using 8.9) equals PJ(TT/2) + (1 — Pj)z* — is strictly greater than
her (conditional) expected payoff from not backing down — which equals
Pj(zi + e ) - This means that in the unique BNE at this second stage subgame
q*{zf) = 1 as well. Hence, the (unconditional) expected payoff to player i
from this unilateral deviation equals

Ei(z') = PI\P3{K/2) + (1 - Pj)z*} + (1 - Pi)Pj(zt + e). (8.12)

Since z* < TT(1— pi), (8.11) ensures that there exists an e G (0, TT/2 — Z*) such
that (8.12) is strictly greater than z*, which constitutes a contradiction. •

It follows from Proposition 8.5 that provided PA and ps are neither

too large nor too small, uncertainty in the commitment process necessarily
8.5 Uncertainty and Simultaneous Concessions 239

implies that the players' equilibrium partial commitments are incompatible,

and with strictly positive probability the players fail to reach agreement.
Proposition 8.5 provides a characterization of any PBE. In the following
proposition I establish the existence of such a PBE.19

Proposition 8.6. If PA > 1/2 and ps > 1/2, then there exists a perfect
Bayesian equilibrium in which the partial commitments of players A and B
are respectively
. _ 2(5A{1 -pB) + i-pA
ZA
Z ZB
2(1 - pApB) ^ 2(1 - p

where $ = (1 -pi)/pi.20

Proof Unless otherwise specified below, fix an arbitrary BNE in each second
stage subgame that follows a pair of incompatible partial commitments.21
By construction, the pair (2^,2^) is the unique solution to the following
pair of equations

TT - zA) = PAZB

TT - zB) = PBZA-

Hence, in the second stage subgame following the (proposed) equilibrium

partial commitments, if ki = 0 then player i is indifferent between backing
down and not backing down, given that player j backs down with proba-
bility one if kj = 0 (cf. (8.9) and (8.10)). This implies that in this second
stage subgame there exists a BNE in which each player i backs down with
probability one if ki = 0. Given this BNE, the (unconditional) payoff to
player i in the proposed PBE is PjZ*.
I now show that the pair z* = (z\, z^) is a Nash equilibrium of the first
stage game. First, observe that since pjZ* > n — z*, player i does not profit
from a unilateral deviation to a partial commitment z[ = n — z*. Hence,
19
The proof is by construction for the following reason. Although there exists a BNE in
each second stage subgame (because such a subgame is finite), since the first stage game
is not finite and since the (unconditional) expected payoffs need not be continuous, it is
not possible to appeal to one of the standard theorems on the existence of Nash equilibria.
20
Notice that since PA > 1/2 and ps > 1/2, TT/2 < z* < TT — hence, these equilibrium
partial commitments are incompatible.
21
Notice that since each such subgame is finite, existence of BNE is ensured.
240 Commitment Tactics

it follows that she does not benefit from a unilateral deviation to a partial
commitment z[ < n — z*.
Now consider a unilateral deviation by player i to z[ such that n — z* <
Zi < z*. This implies that in the second stage subgame that follows the
choice of the pair zf — (z'^z*) there exists a BNE in which each player i
backs down with probability one if ki = 0. Therefore, given this BNE, the
unconditional expected payoff to player i from this deviation is strictly less
than PJZ\.
Finally, consider a unilateral deviation by player i to Z{ such that z* < z[,
and consider the second stage subgame that follows the choice of the pair
zr. Define, for each i — A^B

1 — z*

Notice that since pA > 1/2, pB > 1/2 and z\ > z\, 0 < q?(zf) < 1. This
probability is constructed to ensure that player i is indifferent between not
backing down and backing down if ki = 0. Hence, the pair (g^(V), q*B{z'))
constitutes a BNE of this subgame. It follows that player i's unconditional
expected payoff from this deviation is pjq"Az')z[ — which is strictly decreas-
ing in z[ on the interval (Z*,TT]. Since this payoff pjqUzf)z[ converges to
PjZ* as z[ tends to z* from above, it follows that deviating to z\ > z* is not
profitable. •

8.6 Uncertainty and Wars of Attrition

In the preceding section I have shown that with uncertainty in the commit-
ment process the players' equilibrium partial commitments are necessarily
incompatible, and with strictly positive probability the players fail to reach
agreement. As is shown in the current section, a crucial feature of the model
studied above that contributes to this result is that at the second stage the
players engage in a one-shot, simultaneous, concession game. In particu-
lar, if, when ZA + %B > ^ both players do not back down, then the game
ends in disagreement. This seems unpersuasive; it seems more reasonable to
allow (in that eventuality) the bargainers the option to play the one-shot,
simultaneous, concession game one more time. The criticism levelled here,
8.6 Uncertainty and Wars of Attrition 241

however, would continue to have force. Indeed, it seems much more plausible
to allow the players to engage in a so-called 'war of attrition' game, where
each player decides at any point in time whether or not to back down, and
in which the game ends if and only if some player backs down. This type
of infinite concession game captures the notion that the bargainers keep on
exploring the possibility of reaching agreement. A finite concession game,
on the other hand, embodies the implausible feature that the bargainers
are somewhat forced to terminate bargaining if neither player backs down
before the end of the exogenously specified finite horizon.
In this section I show that if, unlike in the model specified in Section
8.5, the second stage subgames constitute such wars of attrition, then in any
PBE of the two-stage game the players' partial commitments are exactly
compatible. Thus, from this analysis one may conclude that uncertainty
in the commitment process need not generate disagreement in a bargaining
situation.

8.6.1 A Model with Wars of Attrition

Although the model studied here differs from that studied in Section 8.5
mainly in the specification of the second stage game, I shall provide a full
description of the model to be analysed here. Player i's cost of revoking
function is defined in equation 8.2. But, at the first stage, both players are
uncertain about the exact values of both kA and ks] with probability pi,
ki = 0, and with probability 1 — p^ ki = +oc, where 0 < pi < 1.
At the first stage, given the players' uncertainty about the values of k&
and fc#, the players simultaneously and independently choose their respec-
tive partial commitments ZA and ZB, where 0 < Z{ < IT.22 Immediately
after the partial commitments have been made, they become known to both
players. Furthermore, the values of kA and ks are then drawn (randomly
and independently), and player i learns the value of ki but does not learn
the value of kj.
If the partial commitments ZA and ZB are compatible (i.e., ZA + ZB < TT),
then — as in the models studied in Sections 8.2 and 8.5 — neither player
revokes her partial commitment: the share Xi of the cake obtained by player
22
In this model I do not allow each player i to set z% = TT, because the analysis of the
war of attrition subgames presented in the next section applies only if Zi < TT.
242 Commitment Tactics

i is such that x\ > Z{. Specifically, the share obtained by player i is given
by \i(z), where XA and XB are any functions such that XA(Z) > ZA and
\B(z) = 7T — \A{Z) > ZB- For example, it may be assumed that Xi(z) =
Zi + (TT — ZA — ZB)/2- Hence, if ZA + ZB < TT, then player i's payoff is Xi(z).
If, on the other hand, the partial commitments ZA and ZB are incompat-
ible (i.e., ZA + ZB > TT), then play moves to the second stage of the game,
which is the following 'war of attrition' game. At any time t > 0, each
player decides whether or not to back down. The game ends at time t if
and only if at least one player backs down at time t. If the game ends with
only player i having backed down, then her share Xi — TT — Zj and player
j ' s share Xj = Zj. On the other hand, if the game ends with both players
simultaneously having backed down, then each player obtains one-half of
the cake. Finally, if neither player ever backs down, then the players fail to
reach agreement (and each player obtains no share of the cake).
The utility payoff to player i if she obtains a share X{ of the cake at time
£, given that she chose Zi, is Ui(xi, zi) exp(—r^), where the utility function
Ui is defined in (8.1) and T{ > 0 denotes player z's discount rate. This
completes the description of the model.

8.6.2 Equilibrium in the War of Attrition Subgames

Fix an arbitrary pair of partial commitments ZA and ZB which are incom-

patible (that is, ZA + ZB > TT). A pure strategy for each player i in the war
of attrition game involves deciding whether or not to back down at each
time t > 0, conditional on the game not having ended before time t and
conditional on whether k{ — 0 or ki = +oo. Define a mixed strategy for
each player i by a probability distribution function i7^, which is defined on
the closed interval [0,+oo], where for each t > 0, Fi(t) denotes the (un-
conditional) probability that player i has backed down by time t (inclusive).
Notice that, since player i never backs down if ki = +oo, for any finite t > 0,
Fiit) < Pi. Without loss of generality, define i^(+oo) = 1.
The first lemma establishes that in any PBE of this war of attrition game
player i does not back down with probability one at time 0 if ki = 0.

Lemma 8.4. In any perfect Bayesian equilibrium of the war of attrition

game, FA(0) < PA and FB(0) < ps-
8.6 Uncertainty and Wars of Attrition 243

Proof. Suppose, to the contrary, that there exists a PBE and an i such that
Fi(0) = pf, that is, player i backs down with probability one at time 0 if
ki — 0. Therefore, the (conditional) equilibrium expected payoffs to player
j if kj = 0 from backing down at t — 0 and from not backing down at t = 0
are respectively23

Pi(7r/2) + ( l - P i ) ( 7 r - ^ ) and (8.13)

PiZj + (1 -Pi)(7T - ^ ) e x p ( - r J e ) , (8.14)

where e > 0 but arbitrarily small. By comparing (8.13) and (8.14) — and
noting that e > 0 but arbitrarily small — it follows that in equilibrium,
Fj(0) = pj if Zj < TT/2 and Fj(0) = 0 if Zj > TT/2. Therefore, the (condi-
tional) equilibrium expected payoff to player i if ki = 0 is

\ if ZJ > TT/2.

Now suppose player i unilaterally deviates to an alternative strategy Fi

in which Fi(0) = 0 and i^(e) = Pi for some e > 0 but small. Player
z's (conditional) expected payoff from this unilateral deviation if ki = 0 is
greater than or equal to [pjZi + (1 —PJ)(TT — Zj)\ exp(—r^e). 24 A contradiction
follows immediately, because this payoff is strictly greater than player i's
(conditional) equilibrium expected payoff if ki = 0. •

The next lemma establishes some key properties of any PBE in the
war of attrition game. In particular, if ki = 0 then player i backs down
'continuously' before some time T*: that is, in any positive interval of time
within the interval (0, T*), she backs down with strictly positive probability.

Lemma 8.5. In any perfect Bayesian equilibrium of the war of attrition

game there exists a T* such that for each i (i = A, B), Fi(T*) = pi, Fi is
23
In deriving (8.14) I employ the following argument. With probability 1— pi, ki = +oo.
Hence, given player i's equilibrium strategy, if she does not back down at time 0, then
player j ' s equilibrium posterior probabilistic belief that ki = +oo equals one. Therefore,
if kj — 0 then (in equilibrium) player j back downs 'immediately after' t — 0, at t = e.
24
This is because, given player i's equilibrium strategy, with probability pj\ if Zj < n/2
then player j backs down at t = 0, and if Zj > TT/2 then player j backs down immediately
after t = 0. And with probability 1 — pj, she never backs down.
244 Commitment Tactics

differentiable on the interval (0,T*) and for each t E (0,T*) ;

Proof See Abreu and Gul (1999, Proposition 2). 25 •

An immediate consequence of Lemma 8.5 is that for any t < T*, Fi(t) =
1 — Qexp(—c^t), where Q is the constant of integration. Since for all t G
[0, +oo), Fi(i) < p^ this implies that T* is finite. The condition Fi(T*) = pi
implies that the constant

Lemma 8.4 implies that T* > 0, and, hence, since Fi is strictly increasing
on the open interval (0,T*), if ki = 0 then at any time t G (0,T*), player i
is indifferent between backing down and not backing down. If ki = 0, then
player i's equilibrium expected payoffs at any time t G (0,T*) from backing
down and from not backing down are respectively TT — Zj and [otjAzi + (1 —
otj A)(TT—ZJ)] exp(—riA) — because, if she does not back down at time t, then
after an infinitesimal time interval A player j backs down with probability
26
OLJA. Equating these two expressions, and solving for ay, it follows that

ri
ay = ^~y . (8.16)

The following result establishes that for at least some i (i = A or i = B),

Fi(0) = 0 (which implies that Q = 1).

Lemma 8.6. In any perfect Bayesian equilibrium of the war of attrition

game there exists an i (i = A or i = B) such that Fi(0) = 0.
25
Lemma 8.5 implies that at any time t G (0,T*), player i's (unconditional) instanta-
neous rate of backing down, c^, is strictly positive and independent of t. Notice that,
since player i never backs down if hi — +oo, in order to maintain this (unconditional)
constant and strictly positive instantaneous rate of backing down, player i backs down at
an increasing rate if hi = 0 .
26
As is standard in a mixed strategy equilibrium, player j ' s equilibrium instantaneous
rate of backing down ensures that player i is indifferent between backing down and not
backing down if fc = 0. Notice that, since ZA < TT, ZB < TT and ZA + ZB > TT, a A > 0 and
aB > 0.
8.6 Uncertainty and Wars of Attrition 245

Proof. Suppose, to the contrary, that FA(0) > 0 and Fg(O) > 0. Therefore,
the conditional equilibrium expected payoff to each player i if ki = 0 is
Fj(0)[Fi(0)(7r/2) + (l-Fi(0))zi\ + (l-Fj(0))(7r-zj), because her conditional
equilibrium expected payoff at any time t G (0,T*) is TT — Zj. On the other
hand, the conditional equilibrium expected payoff to each player i if ki = 0
from a unilateral deviation to an alternative strategy Fi in which i^(0) = 0
and Fi(e) = pi for some e > 0 but small is greater than or equal to Fj(0)zi +
(1 — FJ(0))(TT — Zj) exp(—Vie). Since there exists an i (i = A or i = B) such
that zi > TT/2, and since e > 0 can be chosen to be arbitrarily small, player
i benefits from such a unilateral deviation — which is a contradiction. •

Now define, for each i = A,B

T, = -ln(-^-V (8.17)

By construction, 1 — exp(—a^) = pi. That is, T{ denotes the time by

which player i will have backed down with probability ^ , given that she
backs down at time 0 with probability zero. Therefore, if TA > TB, then
CA < 1 (because if CA = 1 then T* = T4 > Tg, which is a contradiction).
Hence, by Lemma 8.6 it follows that CB = 1. Moreover, if T4 > Tg, then
T* = TB and the value of CA is determined by (8.15). Hence

r*=min{TA,rB} (8.18)

and Ci (i = A, B) is determined by (8.15). I have thus established the

following result.

Proposition 8.7. For each pair of partial commitments ZA and ZB such

that ZA + ZB > TT, ZA < TT and ZB < ^, the war of attrition game has a
unique PBE in which

z/T*<£<+oc
if t =+00,

where OL{ is defined in (8.16), T* in (8.18) — with Ti defined in (8.17)

and Ci in (8.15).
246 Commitment Tactics

The unique PBE has the following properties. If T{ > Tj, then player i
backs down at time 0 with probability 1—C{ and player j backs down at time 0
with probability zero. Furthermore, if TA = TB, then each player backs down
at time 0 with probability zero. Thereafter, on the open interval (0, T*), each
player i backs down at a constant, strictly positive, rate c^. At time T* it
becomes common knowledge that with probability one, kA — ^B — +oc,
and, thus, neither player ever backs down after this time.

8.6.3 Equilibrium Partial Commitments

Given the characterization of the unique PBE of each war of attrition sub-
game derived above, I now show in the following proposition that in any
Nash equilibrium at the first ('claims') stage of the game, the players' par-
tial commitments are exactly compatible, and, thus, in equilibrium, the
players reach immediate agreement without playing a war of attrition game.

Proposition 8.8. In any PBE of the entire two-stage game described in

Section 8.6.1, the players7 equilibrium partial commitments ZA and ZB are
exactly compatible (that is, ZA + ZB = TT).

Proof. Using Proposition 8.7, I first compute the expected payoffs to the
players if the chosen partial commitments are incompatible. Fix an arbitrary
pair of partial commitments ZA and ZB which are incompatible (i.e., ZA +
ZB > Tr)? and consider the war of attrition subgame. There exists an i (i = A
or i = B) such that T{ > Tj (j / i). Note that a < 1 if T{ > Tj and a = 1
if Ti = Tj. Moreover, note that Cj = 1 and that T* = Tj.
I first derive player j ' s conditional equilibrium expected payoff if kj = 0.
With probability 1 — Q, player i backs down at time 0 and (since player j
backs down at time 0 with probability zero) she obtains Zj. With probability
Q, therefore, her expected payoff equals n — Zj. This is because at each
time t G (0,T*), she is indifferent between backing down and not backing
down. Hence, player j ' s conditional equilibrium expected payoff if kj = 0 is
(1 -Ci)Zj + Ci(ir- Zi).
By definition, player j ' s conditional equilibrium expected payoff if kj =
+oc is

Jo
8.6 Uncertainty and Wars of Attrition 247

which equals (1 - a)zj + Q(TT - Zi) - (1 - pi)(ir - z{) exp(-r J -T*). 27

By similar arguments, but noting that player j backs down at time 0 with
probability zero, it follows that player i's conditional equilibrium expected
payoffs if k{ — 0 and if ki = +oc are respectively TT — Zj and (TT — Zj) — (1 —

Hence, if the pair of partial commitments ZA and ZB are incompatible

and such that T{ > Tj {i ^ j , i,j = A, B), then player j ' s (unconditional)
expected payoff is

PI){TT - z i )exp(-r J T*), (8.19)

and player i's (unconditional) expected payoff is

(TT - ZJ) - (1 - Pi)(l - Pi )(7r - zj) exp(-r,T*), (8.20)

where T* = 7) and Q is defined by (8.15).

I am now ready to establish the proposition. Suppose, to the contrary,
that ZA + ZB > 7T.28 There exists an i such that Ti > Tj, and, hence, player
z's expected payoff is given by (8.20). But, since Zj < TT and since T*(= Tj) is
finite, she can benefit from a unilateral deviation to the partial commitment
z[ = 7T — Zj, which is a contradiction. •
Thus, in contrast to the model studied in Section 8.5, in the current
model — where the bargainers engage in a war of attrition if their chosen
partial commitments are incompatible — the equilibrium partial commit-
ments are never incompatible. Thus, in any PBE agreement is reached with
probability one, without the players ever having to engage in any war of at-
trition. Indeed, this model establishes that uncertainty in the commitment
process need not generate disagreement in a bargaining situation.
Proposition 8.8 characterizes any Nash equilibrium of the first stage
game. In the following proposition I establish the existence (by construction)
of such an equilibrium.

Proposition 8.9. There exists a perfect Bayesian equilibrium in the entire

two-stage game described in Section 8.6.1 in which the equilibrium partial
27
Notice that this payoff is strictly less than j ' s conditional equilibrium expected payoff
if kj = 0.
28
Notice that since Lemma 8.1 applies here as well, the equilibrium partial commitments
are not more-than-compatible.
248 Commitment Tactics

commitments are

=
rAln(l-pA) + rBln(l-pB)
* _
B r lri(l -PA)
rA ln(l - ApA) + rB ln(l - pB)'
Proof. It should be noted that the pair of partial commitments z\ and z^
defined in the proposition is the unique pair of exactly compatible partial
commitments such that TA = TB. Since z\ + z^ = TT, the payoff to player
i in the (proposed) equilibrium is z\. Suppose player i unilaterally deviates
to a partial commitment z[ > £*.29 In the war of attrition subgame that
follows the pair (z\, z*) of partial commitments, T{ > Tj, and, hence, player
i's expected payoff from this unilateral deviation is given by (8.20) with
Zj = Zj. Since that payoff is strictly less than z*, the desired conclusion
follows immediately. D
In general, there may exist other pairs of exactly compatible partial
commitments that form part of a PBE. However, Kambe (1999) has shown
that in the limit as both PA and pB converge to one with the ratio TA ln(l —
PA)/rBln(l — PB) kept constant at some positive level 77, the model has a
unique PBE in which ZA = 1/1 + 7/ and zB = 77/1 + 77. Notice that these
partial commitments constitute the limiting values of z\ and z^ denned in
Proposition 8.9.

8.7 Notes
Thomas Schelling (in Schelling (I960)) was the first to informally explore
the role, feasibility and persuasiveness of commitment tactics in bargaining
situations. His paper contains brilliant discussions about the many issues
involved. In particular, he discusses at great length the details of several
mechanisms through which a bargainer may achieve partial commitment to
some bargaining position. The interested reader is urged to combine their
study of the models in this chapter with a reading of his classic essay.
The model studied in Section 8.2 is due to Muthoo (1996), while the
model studied in Section 8.5 is a simplified (and slightly different) version of
29
It is trivial to note that player i does not benefit from a unilateral deviation to a
partial commitment z[ < z*.
8.7 Notes 249

the model studied in Crawford (1982). The analysis and results in Section
8.5 are based upon Crawford (1982) — the latter paper contains thought-
provoking discussions of some of the issues involved in studying the role of
commitment tactics on the bargaining outcome. Section 8.6 is inspired by
the work of Kambe (1999). The analysis of the war of attrition subgames,
however, is based on the work by Abreu and Gul (1999). The analysis and
results of Section 8.6.3 is based on Kambe (1999).
In this chapter I have focused on the notion that a player makes a partial
commitment to the share of the cake (or, utility) that she would like to
have. The partial commitments affect the utility possiblity set, but do not
affect the players' payoffs if they fail to reach agreement. However, in some
bargaining situations the players make partial commitments to actions that
affect the 'disagreement' outcome. The Nash variable threat model (cf. Nash
(1953); alternatively, see Osborne and Rubinstein (1990)) is a model of a
situation in which the bargainers make irrevocable commitments to actions
that affect the payoffs from disagreement — thus, in this model, it is 'as if
the costs of revoking are infinite. Bolt and Houba (1997) also study a model
in which the bargainers take actions that affect the disagreement outcome,
but — in extreme contrast to Nash's variable threat model — their main
analysis assumes that the bargainers cannot make any partial commitments
— thus, in this model, it is 'as if the costs of revoking are zero.
In Section 8.3.3 I argued that the model studied in Section 8.2 can be
interpreted as a perturbation to the 'commitment' structure of the Nash
demand game. An alternative way to perturb the Nash demand game is
by introducing some shared (or, symmetric) uncertainty about the size of
the cake (or, more generally, about the Pareto frontier of the set of possible
utility pairs obtainable through agreement) — which implies that there is
some uncertainty as to whether a pair of demands are compatible or incom-
patible. This is a perturbation to the 'information' structure of the demand
game; it retains the assumption that the players are irrevocably committed
to their respective demands. This interesting idea is due to Nash (1953),
where he in fact introduced (what is now called) the Nash demand game.
John Nash showed, in Nash (1953), that in the limit, as the uncertainty
becomes arbitrarily small, there is a unique equilibrium. For detailed anal-
yses of this idea, see Binmore (1987), Osborne and Rubinstein (1990) and
van Damme (1991). Although the specific type of uncertainty introduced
250 Commitment Tactics

in Nash (1953) is interesting and generates a powerful result — in that it

selects a unique equilibrium — the plausibility of this type of perturbation is
questionable. Carlsson (1991) studies an alternative type of perturbation to
the information structure of the Nash demand game that is relatively more
plausible.
9 Asymmetric Information

9.1 Introduction
In some bargaining situations at least one of the players knows something
of relevance that the other player does not. For example, when bargaining
over the price of her second-hand car the seller knows its quality but the
buyer does not. In such a bargaining situation, the seller has private infor-
mation; and there exists an asymmetry in information between the players.
In this chapter I study the role of asymmetric information on the bargaining
outcome.
A player may in general have private information about a variety of
things that may be relevant for the bargaining outcome, such as her pref-
erences, outside option and inside option. However, in order to develop the
main fundamental insights in a simple manner attention is focused on the fol-
lowing archetypal bargaining situation. A seller and a buyer are bargaining
over the price at which to trade an indivisible object (such as a second-hand
car, or a unit of labour). The payoff to each player (from trading) depends
on the agreed price and on her reservation value.1 A key assumption is that
at least one player's reservation value is her private information.
I begin the study of this bargaining situation by addressing the normative
lr
The buyer's reservation value is the maximum price at which she is willing to buy.
Symmetrically, the seller's reservation value is the minimum price at which she is willing
to sell.
252 Asymmetric Information

question of whether or not the bargaining outcome can be ex-post efficient.2

Let me make this question a bit more precise. A bargaining procedure
combined with the players' information and preferences in the bargaining
situation under consideration defines a bargaining game with imperfect in-
formation. I shall say that a procedure induces a bargaining game. If there
exists a bargaining procedure — no matter how implausible or plausible
it may be — such that the induced bargaining game has a Bayesian Nash
Equilibrium (BNE) that generates an ex-post efficient outcome, then the
bargaining outcome can be ex-post efficient. On the other hand, if for any
bargaining procedure all of the BNE of the induced bargaining game gen-
erate ex-post inefficient outcomes, then the bargaining outcome cannot be
ex-post efficient.
The following argument illustrates the possibility that the bargaining
outcome cannot be ex-post efficient. A buyer and a seller are bargaining
over the price of a second-hand car, whose quality is the seller's private
information. If she owns a low quality car, then she has an incentive to
pretend to own a high quality car in order to obtain a relatively high price.
Since the buyer is aware of this 'incentive to lie', the maximum price that
she might be willing to pay may be strictly less than the high reservation
value of a seller owning a high quality car. Thus, if the seller actually owns
a high quality car, then mutually beneficial trade between the two parties
may fail to occur.3
Section 9.2 addresses the normative question stated above when exactly
one player's reservation value is her private information. The answer to
the normative question depends on whether or not the players' reservation
values are independent of each other. When the players' reservation values
are independent of each other, I say that the values are private; otherwise,
they are correlated. If the players' reservation values are private, then the
bargaining outcome can be ex-post efficient. But if the reservation values are
correlated, then (under a fairly general condition) the bargaining outcome
cannot be ex-post efficient. Two applications are studied in Section 9.3.

2
The bargaining outcome is ex-post efficient if and only if after all of the information is
revealed the players' payoffs associated with the bargaining outcome are Pareto-efficient.
The concept of ex-post efficiency is also known as full-information efficiency.
3
In his classic paper, Akerlof (1970), George Akerlof put forward this type of argument,
but in the context of competitive markets (with asymmetric information).
9.2 Efficiency under One-Sided Uncertainty 253

One application concerns the normative question of whether or not it is

possible for a firm and its unionized workforce to reach an agreement (over
the wage rate) without any costly delay. The other application concerns
the normative question of whether or not it is possible for a plantiff and a
defendant to settle their dispute out-of-court.
Section 9.4 addresses the normative question when each player's reserva-
tion is her private information. A main result obtained here is that (under
some fairly general conditions) the bargaining outcome cannot be ex-post
efficient, whether or not the players' reservation values are independent of
each other. Section 9.5 extends the two applications studied in Section 9.3.
Section 9.6 addresses some positive questions in the context of a bar-
gaining model in which exactly one player has private information, and the
other player makes repeated offers. The two main questions that motivate
the study of this model are as follows. Firstly, to what extent is the equilib-
rium payoff of the player who makes all the offers adversely affected when
her opponent has private information? And secondly, under what condi-
tions (if any) is the privately held information revealed through time via
the sequence of equilibrium offers? It will be shown that the answers to
these questions depend, in particular, on (i) whether or not offers are re-
tractable, (ii) whether or not the costs to the players of haggling are small,
and (iii) whether or not gains from trade are strictly positive with probabil-
ity one. An application to bargaining over a menu of wage-quality contracts
is studied in Section 9.7.

9.2 Efficiency under One-Sided Uncertainty

I study a bargaining situation in which player S owns (or, can produce)
an indivisible object that player B wants to buy. If agreement is reached
to trade at price p (p > 0), then the payoffs to the seller (player S) and
the buyer (player B) are respectively p — c and v — p, where c denotes the
seller's reservation value (or, cost of production) and v denotes the buyer's
reservation value (or, the maximum price at which she is willing to buy). If
the players do not reach an agreement to trade, then each player's payoff is
zero.
A key assumption is that exactly one player's reservation value is her
private information. Section 9.2.1 studies the case in which the players'
254 Asymmetric Information

reservation values are independent of each other, while Section 9.2.2 studies
the case in which the players' reservation values are correlated.
The outcome of this bargaining situation is ex-post efficient if and only
if when v > c the players reach an agreement to trade, and when v < c the
players do not reach an agreement to trade.
My objective is to address the normative question of whether or not the
outcome of the bargaining situation described above can be ex-post efficient.
As mentioned in Section 9.1, if there exists a bargaining procedure such that
the induced bargaining game has a BNE that generates an ex-post efficient
outcome, then the bargaining outcome can be ex-post efficient. On the
other hand, if for any bargaining procedure all of the BNE of the induced
bargaining game generate ex-post inefficient outcomes, then the bargaining
outcome cannot be ex-post efficient.

9.2.1 The Case of Private Values

In this section it is assumed that the players' reservation values are inde-
pendent of each other, and exactly one player has private information about
her reservation value. The other player's reservation value is known to both
players.
I begin by studying the case in which the buyer's reservation value is
her private information. This asymmetry in information is modelled as
follows. The buyer's reservation value is a random draw from the following
(binary) probability distribution: with probability a (where 0 < a < 1)
the buyer's reservation value is iiT, and with probability 1 — a the buyer's
reservation value is L, where H > L. The buyer knows the realization of
the random draw, but the seller does not. The seller only knows that the
buyer's reservation value is a random draw from this probability distribution.
The following lemma establishes that the bargaining outcome can be ex-post
efficient.4

Lemma 9.1. There exists a bargaining procedure such that the induced bar-
gaining game has an ex-post efficient BNE.
4
The lemma is valid whatever is the magnitude of c relative to H and L. It should be
noted that if c > H then gains from trade do not exist, but if L > c then gains from trade
exist with probability one. And if H > c > L then with probability a they exist, but with
probability 1 — a they do not.
9.2 Efficiency under One-Sided Uncertainty 255

Proof. Consider the following bargaining procedure. The buyer makes an

offer to the seller. If she accepts the offer, then agreement is struck and
the game ends. But if she rejects the offer, then the game ends with no
agreement. Letting p*H and p*L respectively denote the buyer's price offers
when v — H and v = L, the following pair of strategies is a BNE: p* H =
min{i7, c}, p*L — min{L,c}, and the seller accepts a price offer p if and
only if p > c. The lemma follows immediately, because this BNE is ex-post
efficient. •

It is now shown that Lemma 9.1 is also valid when the buyer's reser-
vation value can take more than two possible numbers. Let FB denote the
cumulative probability distribution from which the buyer's reservation value
is randomly drawn, where the support of FB contains two or more numbers.
The proof of Lemma 9.1 is still valid, but with the following modification
to the buyer's equilibrium strategy: p*(v) = min{i>, c} for all v, where p*(v)
denotes the buyer's price offer when her reservation value is v.
I now study the case in which the seller's reservation value is her private
information, and the buyer's reservation value is known to both players. As
above, I model this asymmetry in information by considering the seller's
reservation value to be a random draw from a probability distribution Fs —
whose support contains two or more numbers — the realisation of which is
only revealed to the seller. Consider the following bargaining procedure. The
seller makes an offer to the buyer. If she accepts the offer, then agreement
is struck and the game ends. But if she rejects the offer, then the game
ends with no agreement. The following pair of strategies is a BNE of the
induced bargaining game: p*(c) = max{^,c} for all c, where p*(c) denotes
the seller's price offer when her reservation value is c, and the buyer accepts
a price offer p if and only if p < v. Lemma 9.1 follows immediately, because
this BNE is ex-post efficient.
The following proposition summarizes the results obtained above.

Proposition 9.1 (One-Sided Uncertainty with Private Values). //

the players' reservation values are independent of each other, and exactly
one player's reservation value is her private information, then the bargaining
outcome can be ex-post efficient.
256 Asymmetric Information

9.2.2 The Case of Correlated Values

In this section I assume that the player's reservation values are correlated,
and exactly one player has private information about her reservation value.
This assumption is modelled as follows. There is a parameter 9 — which
is a real number — that determines both players' reservation values, and
furthermore, the value of 9 is the private information of exactly one player.
It is assumed that each player's reservation value is strictly increasing in 9.
Furthermore, for any #, the buyer's reservation value — which is denoted
by v{9) — is greater than or equal to the seller's reservation value — which
is denoted by c(#). 5
I begin by considering the case in which it is the seller who has private
information about 9. This asymmetry in information is modelled as follows.
The value of 9 is a random draw from the following (binary) probability
distribution: with probability a (where 0 < a < 1) the value of 9 is iJ, and
with probability 1 — a the value of 9 is L, where H > L. The seller knows
the realization of the random draw, but the buyer does not. The buyer only
knows that the value of 9 is a random draw from this probability distribution.
The following lemma establishes that the bargaining outcome can be ex-post
efficient if and only if ve > c(H), where ve = av(H) + (1 — a)v(L) is the
buyer's expected reservation value.

Lemma 9.2. (a) If ve > c(H), where ve = av(H) + (l — a)v(L), then there
exists a bargaining procedure such that the induced bargaining game has an
ex-post efficient BNE.
(b) If ve < c(H), then for any bargaining procedure the induced bargaining
game does not have an ex-post efficient BNE.
Proof. I first establish Lemma 9.2(a). Consider the following bargaining
procedure. The seller makes an offer to the buyer. If she accepts the offer,
then agreement is struck and the game ends. But if she rejects the offer,
then the game ends with no agreement. Since ve > c(H), the following pair
of strategies is a BNE: p*{H) = p*(L) = c(H) (where p*(H) and p*(L) are
respectively the seller's price offers when 9 = H and 9 = L), the buyer
accepts the price p = c{H) and rejects any price p ^ c(H). The desired
conclusion follows immediately, because this BNE is ex-post efficient. •
5
This implies that the outcome of this bargaining situation is ex-post efficient if and
only if the players reach an agreement to trade whatever value 0 takes.
9.2 Efficiency under One-Sided Uncertainty 257

I now proceed to prove Lemma 9.2(b). In order to do so I need to

consider the set of all possible bargaining procedures. However, I begin
by considering a particular subset of the set of all bargaining procedures,
which is called the set of all direct revelation procedures. In the context of
the bargaining situation under consideration, a direct revelation procedure
(DRP) is characterized by four numbers: A^, A#, PL and pn, where A^ G
[0,1] and ps > 0 (s = L,H). In a DRP the seller announces a possible
value of 9. If s denotes the announced value (where s G {£, i?}), then with
probability Xs trade occurs at price ps, and with probability 1 — Xs trade
does not occur.
Fix an arbitrary DRP, and consider the induced bargaining game (which
is a single-person decision problem). Let s(9) G {L,H} denote the seller's
announcement if the true (realized) value is 9 (9 = L,H). The DRP is
incentive-compatible if and only if in the induced bargaining game the seller
announces the truth — that is, s*(L) = L and s*(H) = H. Thus, the
DRP is incentive-compatible if and only if the following two inequalities are
satisfied

- c(L)) > \H(pH - c(L)) (9.1)

\H(PH ~ c(H)) > XL(PL - c(H)). (9.2)

Inequalities 9.1 and 9.2 are respectively known as the incentive-compatibility

constraints for the low-type seller and high-type seller.6 Inequality 9.1 states
that the expected payoff to the low-type seller by announcing the truth
is greater than or equal to her expected payoff by telling a lie. Similarly,
inequality 9.2 states that the expected payoff to the high-type seller by
announcing the truth is greater than or equal to her expected payoff by
telling a lie.
An incentive-compatible DRP is individually-rational if and only if in the
incentive-compatible DRP each type of seller and the buyer obtain an ex-
pected payoff that is not less than their respective payoff from disagreement
(which equals zero). That is, if and only if the following three inequalities

6
The seller is said to be of 'low-type' (respectively, 'high-type') if the true (realized)
value of 0 is L (respectively, H).
258 Asymmetric Information

are satisfied

XL(PL ~•c(L)) >o (9.3)

XH(PH ~ c(H)) >o (9.4)
aXH(v(H) -PH) + ( l-a)XL(v(L) -PL) >0. (9.5)

Since (by assumption) v(H) > c(H) and v(L) > c(L), a DRP is ex-post
efficient if and only if the buyer trades with the seller of either type with
probability one. That is, if and only if

XL = XH = I- (9.6)

I now state a rather powerful result — which is known as the Revelation

Principle — that allows me to establish Lemma 9.2(b) by considering only
the set of all incentive-compatible and individually-rational direct revelation
procedures.

Theorem 9.1 (The Revelation Principle). Fix an arbitrary bargaining

situation with asymmetric information and an arbitrary bargaining proce-
dure. For any BNE outcome of the induced bargaining game there exists
an incentive-compatible and individually-rational DRP that implements the
BNE outcome.

Proof. This result is proven and discussed in the context of bargaining sit-
uations in Myerson (1979). The proof can also be found in most advanced
microeconomics and game theory texts — see, for example, Kreps (1990a),
Fudenberg and Tirole (1991), Gibbons (1992) and Mas-Colell, Whinston
and Green (1995). •

It follows from the Revelation Principle that if there does not exist an
incentive-compatible and individually-rational DRP that is ex-post efficient,
then there does not exist a bargaining procedure whose induced bargaining
game has an ex-post efficient BNE. Lemma 9.2(b) is therefore an immediate
consequence of the following claim.

Claim 9.1. / / ve < c{H), where ve is defined in Lemma 9.2, then there
does not exist an incentive-compatible and individually-rational DRP that is
ex-post efficient.
9.2 Efficiency under One-Sided Uncertainty 259

Proof. Suppose, to the contrary, that there exists a DRP that satisfies (9.1)-
(9.6). Substituting (9.6) into (9.1) and (9.2), it follows that pL = pH.
Hence, after substituting (9.6) into (9.3)-(9.5), it follows from (9.3)-(9.5)
that ve > c(H), thus contradicting the hypothesis. •

A Generalization to More Than Two Types

The results obtained above may be generalized quite easily to the case when
9 takes more than two possible values. Assume that 9 is a random draw
from a probability distribution G — whose support J contains two or more
numbers — the realization of which is only revealed to the seller. Let 9_ and
9 respectively denote the minimum and maximum values of J.
A DRP is characterized by a pair of functions (A, p), where for each s G J,
A(s) and p(s) are respectively the probability with which trade occurs and
the price at which it occurs if the seller announces that the value of 9 is s.
A DRP is incentive-compatible if and only if

V# G J, A(0)[p(0) - c(9)} > X(s)[p(s) - c(0)], \/s G J. (9.7)

Inequality 9.7 states that for any possible true (realized) value of #, the
expected payoff to the #-type seller by announcing the truth is greater than
or equal to her expected payoff by telling a lie.7
An incentive-compatible DRP is individually-rational if and only if

\/9eJ, A(0)[p(0)-c(0)]>O, and (9.8)

Ee[\(9)[v(9)-p(9)]\ > 0. (9.9)

Inequalities 9.8 and 9.9 respectively state that the expected payoffs to the
0-type seller and the buyer in an incentive-compatible DRP are greater than
or equal to their respective payoffs from disagreement (which equal zero).
Since (by assumption) v{9) > c{9) for all 0, a DRP is ex-post efficient if and
only if

W G J, A(0) = 1. (9.10)

The following claim is a generalization of Claim 9.1.

7
The seller is said to be of '#-type' if the realization of the random draw from G is 0.
260 Asymmetric Information

Claim 9.2. If ve < c{9), where ve is the buyer's expected reservation value
— that is, v e = EQ\V{9)\ — then there does not exist an incentive-compatible
and individually-rational DRP that is ex-post efficient.

Proof. Suppose, to the contrary, that there exists a DRP that satisfies (9.7)—
(9.10). Substituting (9.10) into (9.7) implies that p(9) is constant for all 0 G
J. Letting p = p{6) for all 6 G J, it follows using (9.8) that p > c(0). Hence,
using (9.9) it follows that ve > c(0), which contradicts the hypothesis. •

It follows from the Revelation Principle that if ve < c(0), then for any
bargaining procedure the induced bargaining game does not have an ex-post
efficient BNE. Therefore, if ve < c(0), then the bargaining outcome cannot
be ex-post efficient. On the other hand, if ve > c(0), then there exists a
bargaining procedure such that the induced bargaining game has an ex-post
efficient BNE. 8

Remark 9.1 (The Buyer has Private Information). Now consider the
case in which it is the buyer (and not the seller) who has private informa-
tion about 9. Thus, the realization of the random draw from the probability
distribution G is only revealed to the buyer. It is straightforward to appro-
priately modify the above analysis and show that the bargaining outcome
can be ex-post efficient if and only if c e < v(8_), where ce is the seller's
expected reservation value — that is, c e = EQ[C{0)].

Hence, I have established the following proposition.

Proposition 9.2 (One-Sided Uncertainty with Correlated Values).

(a) When the seller has private information about 6 the bargaining outcome
can be ex-post efficient if and only if the buyer's expected reservation value
ve > c{9), the seller's maximum possible reservation value.
(b) When the buyer has private information about 9 the bargaining outcome
can be ex-post efficient if and only if the seller's expected reservation value
ce < v(0_), the buyer's minimum possible reservation value.
Consider the following bargaining procedure. The seller makes an offer to the buyer.
If she accepts the offer, then agreement is struck and the game ends. But if she rejects
the offer, then the game ends with no agreement. If ve > c(0), then the following pair of
strategies is a BNE: p*(0) — c(6) (where p*(9) is the seller's price offer when the realization
of the random draw from G is 9), the buyer accepts the price p — c(9) and rejects any
price p -^ c(9).
9.2 Efficiency under One-Sided Uncertainty 261

An Example in which Trade Never Occurs

Suppose that J is equal to the closed interval [0,1], G is a uniform distribu-

tion (i.e., G{9) = 0), v(0) = 39 and c(9) = 29. Assume that the seller has
private information about 9. Since ve = 1.5 and c(l) = 2, it follows from
Proposition 9.2(a) that the bargaining outcome cannot be ex-post efficient.
In fact, it is shown below that in any incentive-compatible and individually-
rational DRP, A(0) = 0 for all 9 e [0,1]. This striking result implies (by
appealing to the Revelation Principle) that for any bargaining procedure
and any BNE of the induced bargaining game trade occurs with probability
zero.
Fix an arbitrary incentive-compatible and individually-rational DRP. Let
UB and Us{9) respectively denote the expected payoffs (in the induced bar-
gaining game) to the buyer and the 0-type seller. Inequality 9.7 implies that
for each 9 e [0,1]

Us(9) = X(9)[p(9) - 29} = max X(s)\p(s) - 29}.

se[o,i]
From the Envelope Theorem, it follows that Us(-) is differentiate almost
everywhere with derivative U's{9) — —2\{9). This implies that

f 9dUs(9) = -2 [ 9\{9)d9. (9.11)

Jo Jo
After integrating by parts the LHS of equation 9.11, and then simplifying,
it follows that

/ Us(9)d9 = Us(l) + 2 [ 9X(9)d9. (9.12)

Jo Jo
Now consider the expected payoff UB to the buyer, which is the LHS of
inequality 9.9. After substituting for X(9)p(9) using the expression for Us(9),
it follows that

UB = f [9X(9) - Us(9)}d9,
Jo
which (using (9.12)) implies that

UB = -Us(l) - [ 9\(9)d9.
Jo
262 Asymmetric Information

Hence, it follows from inequalities 9.8 and 9.9 that

/ 0A(0)d0<O,
Jo
which implies that A(0) = 0 for all 9 G [0,1].

9.3 Applications

9.3.1 Efficient Wage Agreements

Consider a firm whose workforce is represented by a union. The firm and the
union are bargaining over the wage rate on the assumption that there will be
no firing and hiring. In order to simplify the notation, normalize the mass of
workers employed at this firm to unity. Assuming that the union's payoff is
the same as a worker's payoff, if agreement is reached on wage rate w, then
the (average) payoffs to the firm and the union are respectively R — w and
w, where R (R > 0) is the value of the (average) output generated by the
firm's workforce. If the players do not reach agreement on a wage rate, then
the entire workforce goes on indefinite strike. In this eventuality the firm
shuts down and obtains a payoff of zero, while each worker has recourse to
the union's strike fund and obtains an average payoff of a, where a > 0. The
bargaining outcome is ex-post efficient if and only if when R> a the players
reach a wage agreement, and when R < a the union goes on indefinite strike
and the firm shuts down.
It is helpful to normalize the union's payoffs so that its average payoff
from disagreement is zero. This implies that its (normalized) average payoff
from agreement on wage rate w is w — a. Indeed, the firm and the union
are bargaining over the price at which the union will sell afixedamount of
labour to the firm, where the firm's and the union's reservation values are
respectively R and a.
It follows from Proposition 9.1 that if either the firm has private informa-
tion about i?, or the union has private information about a (but not both),
then the bargaining outcome can be ex-post efficient. That is, there exists a
bargaining procedure whose induced bargaining game has a Bayesian Nash
equilibrium with the following property: when R > a a wage agreement
is reached, and when R < a the union goes on indefinite strike. It should
be noted that it is not unreasonable that the value of the (average) output
9.3 Applications 263

generated by the entire workforce is the firm's private information, and that
the magnitude of the union's strike fund is its private information.9

9.3.2 Litigation or Out-of-Court Settlement

Individual D has injured individual P, and they are bargaining over the
amount of monetary compensation that D will give P. The individuals can
either reach an agreement (and thus settle out-of-court), or litigate. If they
agree to settle out-of-court with individual D — the defendant — paying
individual P — the plantiff — an amount p (p > 0), then the payoffs to
the defendant and the plantiff are respectively — p and p. If, on the other
hand, they litigate, then with probability 7 (0 < 7 < 1) the court will find
the defendant guilty of the crime, in which case she will be required to pay
an amount x (x > 0) to the plantiff. However, with probability 1 — 7 the
defendant will not be found guilty, in which case she pays nothing to the
plantiff. Litigation will cost each individual an amount / (/ > 0).10 Hence,
the (expected) payoffs from litigation to the defendant and the plantiff are
respectively —7X — / and jx — f.
It is helpful to normalize the players' payoffs so that each player's payoff
from litigation is zero. This implies that the defendant's and the plantiff's
(normalized) payoffs from an out-of-court settlement on price p are respec-
tively v — p and p — c, where v = jx + / and c — ^x — f. Furthermore, each
player's payoff from disagreement — that is, from litigation — is zero. In-
deed, the defendant and the plantiff are bargaining over the price at which
the plantiff will sell her claim to the lawsuit, where the defendant's and
the plantiff's reservation values are respectively v and c. Notice that their
reservation values are correlated.
I apply Proposition 9.2. First, consider the case in which the plantiff
(the seller) has private information about x — which is not an unreasonable
assumption, because it is possible that the plantiff only knows the exact
extent of her injury. In the current bargaining situation the condition ve >
c(x) can be written as / > j(x — xe)/2, where xe is the expected value
of x. Hence, the bargaining outcome can be ex-post efficient if and only
9
In Section 9.5.1 it is shown that if the firm has private information about R and the
union has private information about a, then the bargaining outcome cannot be ex-post
efficient.
10
This is partly because litigation involves hiring a lawyer.
264 Asymmetric Information

if the cost of litigation to each player / > j(x — xe)/2. In particular, if

/ < 7(x — xe)/2, then for any bargaining procedure and any BNE of the
induced (pre-trial) bargaining game with a strictly positive probability the
plantiff and the defendant will proceed to litigation.
Now consider the case in which the value of x is known to both players,
but the defendant (the buyer) has private information about 7 — which is
not an unreasonable assumption, because it is possible that the defendant
only knows the exact extent to which her crime is provable in a court of
law.11 In the current bargaining situation the condition ce < v(j) can be
written as / > x(^e — 7)/2, where j e is the expected value of 7. Hence, the
bargaining outcome can be ex-post efficient if and only if the cost of litigation
to each player / > x(je — 7)/2. In particular, if / < x{^e — 7)/2, then for
any bargaining procedure and any BNE of the induced (pre-trial) bargaining
game with a strictly positive probability the plantiff and the defendant will
proceed to litigation.

The Effect of a Fee-Shifting Rule

In the analysis above it is (implicitly) assumed that each player bears her
cost of litigation. This is known as the American rule, because it is typical
in the USA. In England, on the other hand, it is typically the case that the
loser bears the winner's cost of litigation — hence, this is called the English
rule. I now consider the normative question in the context of the English
rule.
If the players proceed to litigation, then with probability 7 the court will
find the defendant guilty of the crime. Hence, with probability 7 the plantiff
wins the lawsuit, while with probability 1 — 7 the defendant is the winner.
This implies that the (expected) payoffs from litigation to the defendant and
the plantiff are respectively — jx — 27/ and 72; — 2(1 —7)/. Notice that these
disagreement payoffs differ from those under the American rule. As above,
normalize the players' payoffs so that each player's payoff from litigation
is zero. This implies that the defendant's and the plantiff's (normalized)
payoffs from an out-of-court settlement on price p are respectively v — p and
p — c, where v = ^x + 27/ and c = jx — 2(1 — 7)/.
11
In Section 9.5.1, I analyse the (pre-trial) bargaining situation under the assumption
that the plantiff has private information about x and the defendant has private information
about 7.
9.4 Efficiency under Two-Sided Uncertainty 265

I apply Proposition 9.2. First, consider the case in which the plantiff
has private information about x. It is straightforward to show that the
conclusion is the same as under the American rule. Now consider the case in
which the value of x is known to both players, but the defendant (the buyer)
has private information about 7. In this case it follows that the bargaining
outcome can be ex-post efficient if and only if the cost of litigation to each
player / > x{^e — 7)/2(l — 7 e + 7). Since 7 e > 7, it follows that if

x(le ~ 7) f %{le ~ 7)
2(1 -r + 7)'
then under the American rule the bargaining outcome can be ex-post effi-
cient, while under the English rule it cannot be ex-post efficient.
A main message contained in the results obtained above is as follows. If
and only if the probability with which the plantiff wins at trial is the defen-
dant's private information, then the disputants are more likely to proceed
to litigation under the English rule than under the American Rule.

9.4 Efficiency under Two-Sided Uncertainty

This section extends the analysis of Section 9.2 to the case when each player
has some private information. As in Section 9.2, I study the bargaining
situation in which player S owns (or, can produce) an indivisible object
that player B wants to buy. If agreement is reached to trade at price p
(p > 0), then the payoffs to the seller (player S) and the buyer (player B)
are respectively p — c and v—p, where c denotes the seller's reservation value
(or, cost of production) and v denotes the buyer's reservation value (or, the
maximum price at which she is willing to buy). If the players do not reach
an agreement to trade, then each player's payoff is zero.
Section 9.4.1 studies the case in which the players' reservation values are
independent of each other, and each player's reservation value is her private
information. Section 9.4.2 studies the case in which the players' reservation
values are correlated, and each player has some relevant private information.
As in Section 9.2, my objective is to address the normative question of
whether or not the bargaining outcome can be ex-post efficient.12 As in
2
The outcome of this bargaining situation is ex-post efficient if and only if when v > c
266 Asymmetric Information

Section 9.2.2, the analysis involves studying the set of incentive-compatible

and individually-rational direct revelation procedures.

9.4.1 The Case of Private Values

In this section the players' reservation values are independent of each other,
the seller's reservation value is her private information, and the buyer's
reservation value is her private information. This asymmetry in information
is modelled as follows. The buyer's reservation value is a random draw
from a probability distribution Fg, and the seller's reservation value is an
independent and random draw from a probability distribution Fs- The
buyer knows the realization of the draw from F#, but the seller does not:
she only knows that the buyer's reservation value is an independent and
random draw from FB- Symmetrically, the seller knows the realization of
the draw from Fs, but the buyer does not: she only knows that the seller's
reservation value is an independent and random draw from Fs-
Letting I{ {% = B,S) denote the support of i7^, denote the minimum
and maximum values of IB respectively by y_ and v, and the minimum and
maximum values of Is respectively by c and c.
In the context of the bargaining situation under consideration, a direct
revelation procedure (DRP) is characterized by a pair of functions (A,p). In
a DRP the seller and the buyer simultaneously announce their respective
reservation values. If the seller's announced value is c (c E Is) and the
buyer's announced value is v (v G /#), then with probability X(c,v) trade
occurs at price p(c, v), and with probability 1 — A(c, v) trade does not occur.
A DRP is incentive-compatible if and only if the following inequalities
are satisfied

Us(c) = Ev[x(c,v)[p(ciV) - c]\ >

Ev[\(c',v)[p(c',v)-c}\ (9.13)

Ec[\(c,v')[v-p(c,v')]\. (9.14)

the players reach an agreement to trade, and when v < c the players do not reach an
agreement to trade.
9.4 Efficiency under Two-Sided Uncertainty 267

An incentive-compatible DRP is individually-rational if and only if

VcG/s, Us(c)>0 and (9.15)

W G I B, UB(v) > 0. (9.16)

A DRP is ex-post efficient if and only if the buyer and the seller trade when
it is mutually beneficial to do so, but not otherwise. That is, if and only if
for each c E Is and v G IB

(1 if v > c
\(c,v) = { - (9.17)
[0 iiv<c.
In Proposition 9.3(a) below it is shown that \iy_>c — which implies that
gains from trade exist with probability one — then the bargaining outcome
can be ex-post efficient. In contrast, in Proposition 9.3(b) it is shown that
under some conditions on the distributions — which imply that there is
uncertainty over whether or not gains from trade exist — the bargaining
outcome cannot be ex-post efficient.

Proposition 9.3 (Two-Sided Uncertainty with Private Values), (a)

If V.> c, then the bargaining outcome can be ex-post efficient,
(b) If IB = [v.i v]j Is = [c, c], Fi (i = B,S) has a continuous and strictly pos-
itive density, and the interiors of the intervals IB and Is have a non-empty
intersection, then the bargaining outcome cannot be ex-post efficient.

Proof. I first prove of part (a). Consider the following bargaining proce-
dure. The seller makes an offer to the buyer. If she accepts the offer, then
agreement is struck and the game ends. But if she rejects the offer, then
the game ends with no agreement. Since y_ > c, the following pair of strate-
gies is a BNE: p*(c) = c (where p*(c) is the seller's price offer when her
reservation value is c), the buyer accepts the price p = c and rejects any
price p ^ c whatever is her reservation value. The desired conclusion follows
immediately, because this BNE is ex-post efficient.
Since the proof of part (b) is a bit technical, I omit it, and instead refer
the interested reader to Myerson and Satterthwaite (1983) — the authors
of this result. However, as I now show, it is straightforward to establish
part (b) under the following specific assumptions: IB = Is = [0,1] and
268 Asymmetric Information

Fi (i = B,S) is uniform (i.e., Fi(x) = x for x G [0,1]).13 By the Reve-

lation Principle it suffices to show that there does not exist an incentive-
compatible and individually-rational DRP that is ex-post efficient. Suppose,
to the contrary, that there exists a DRP that satisfies (9.13)—(9.17). Inequal-
ity 9.13 implies (using the Envelope Theorem) that Us(c) is differentiable
almost everywhere with derivative U's(c) = — Ev[\(c,v)}. Since (9.17) im-
plies that Ev[\(c,v)] = 1 — c, it follows that the seller's (unconditional)
expected payoff C/| = [/^(l) + 1 / 6 . By a symmetric argument it follows that
the buyer's (unconditional) expected payoff Ug = UB{0) + 1/6. However,
by definition, the sum of the players' (unconditional) expected payoffs is
Ev,c[(v — c)A(c, i?)]. Substituting (9.17) into this expression, and integrat-
ing, it follows that Ev^c[(v — c)X(c,v)] = 1/6. This therefore implies that
Us(l) + UB(0) = - 1 / 6 , which contradicts (9.15)-(9.16). •

9.4.2 The Case of Correlated Values

In this section the player's reservation values are correlated, and each player
has some relevant private information. This assumption is modelled as fol-
lows. There are two parameters 9 and p — both of which are real numbers
— that determine both players' reservation values, and, furthermore, the
value of 9 is the private information of the seller and the value of p is the
private information of the buyer. The asymmetry in information is modelled
as follows. The values of 9 and p are independent and random draws from
the probability distributions G and if, respectively. The seller knows the
realization of the draw from G, but the buyer does not: she only knows that
the value of 9 is an independent and random draw from G. Symmetrically,
the buyer knows the realization of the draw from if, but the seller does not:
she only knows that the value of p is an independent and random draw from
H.
Letting J and K respectively denote the supports of G and if, denote
the minimum and maximum values of J respectively by 9_ and 0, and the
minimum and maximum values of K respectively by p and p.
Both the seller's reservation value — denoted by c(0, p) — and the
buyer's reservation value — denoted by v{91p) — are strictly increasing
13
The conceptual argument in the proof of the general case in Myerson and Satterthwaite
(1983) is similar to that in the proof to follow in this special case.
9.4 Efficiency under Two-Sided Uncertainty 269

in 9 and in p. Furthermore, assume that for any 9 and p, v(Q,p) > c(9,p).
In the context of the bargaining situation under consideration, a direct
revelation procedure (DRP) is characterized by a pair of functions (A,p).
In a DRP the seller announces the value of 9 and simultaneously the buyer
announces the value of p. If the seller's announced value is 91 (9r E J) and
the buyer's announced value is pr (// E if), then with probability X(9/', //)
trade occurs at price p(9',pr), and with probability 1 — A(0', pf) trade does
not occur.
A DRP is incentive-compatible if and only if the following inequalities
are satisfied

y € J, Usip) = Ep[\(e,PMe,p)-c(d,p)]\ >

W € K, UB{p) = Ed[\(9,p)[v(e,p)-p(8,p)]\ >

Ee[x(e,p')[v(e,p)-p(e,p')]\. (9.19)

An incentive-compatible DRP is individually-rational if and only if

M9 E J, Us(0) > 0 and (9.20)

VptK, UB(p)>0. (9.21)

A DRP is ex-post efficient if and only if

V9eJ and Vp E if, A(0, p) = 1. (9.22)

The following proposition addresses the normative question of whether or
not the bargaining outcome can be ex-post efficient.

Proposition 9.4 (Two-Sided Uncertainty with Correlated Values).

The bargaining outcome can be ex-post efficient if and only if ve(p) > ce(9),
where ve(p) = Ee[v(9,p)} and ce(9) = Ep[c(0,p)].

Proof I first establish sufficiency. Consider the following bargaining proce-

dure. The buyer makes a price offer to the seller. If she accepts the offer,
then agreement is struck and the game ends. But if she rejects the offer,
270 Asymmetric Information

then the game ends with no agreement. If ve(p) > c e (#), then the following
pair of strategies is a BNE: for any value of p the buyer offers the price
p = c e (#), and for any value of 9 the seller accepts the price p — ce{9)
and rejects any price p ^ ce{9). The desired conclusion follows immedi-
ately, because this BNE is ex-post efficient. I now establish necessity. By
the Revelation Principle it suffices to show that if ve(p) < c e (#), then there
does not exist a DRP that satisfies (9.18)-(9.22). Suppose, to the con-
trary, that there exists such a DRP. After substituting the ex-post efficiency
condition (9.22) into the seller's incentive-compatibility condition (9.18), it
follows that the expectation of p(6, p) with respect to p is independent of 9.
Let it be denoted by pes. Similarly, after substituting the ex-post efficiency
condition (9.22) into the buyer's incentive-compatibility condition (9.19), it
follows that the expectation of p(91 p) with respect to 9 is independent of
p. Let it be denoted by peB. After substituting (9.22) into (9.20) it thus
follows from the seller's individual-rationality condition (9.20) that for any
9 e J, p%> Ep[c(0,p)]. Symmetrically, after substituting (9.22) into (9.21)
it follows from the buyer's individual-rationality condition (9.21) that for
any p e K, Ee[v(0,p)] > p%. This implies that pe > ce(9) and ve(p) > pe,
where pe is the expectation of p(0, p) with respect to 9 and p. Consequently,
ve(p) > c e (#), which contradicts the hypothesis. •

9.5 Applications
9.5.1 Indefinite Strikes
Consider the bargaining situation between the firm and its union as laid out
in Section 9.3.1, but with the assumption that the firm has private informa-
tion about R and the union has private information about a. Assume that
R and a are independent and random draws from two continuous probabil-
ity distributions with strictly positive densities, where R and a respectively
take values from the closed intervals [R, R] and [a, a].
It follows from Proposition 9.3(a) that if R > a, then the bargaining
outcome can be ex-post efficient. Thus, if with probability one it is mutually
beneficial for the union to sell its labour to the firm, then there exists a
bargaining procedure whose induced bargaining game has a BNE with the
following property: with probability one a wage agreement is reached.
9.6 Bargaining Power and Uncertainty 271

On the other hand, if a < R < a < R, then Proposition 9.3(b) implies
that the bargaining outcome cannot be ex-post efficient. Thus, if there is
uncertainty as to whether or not it is mutually beneficial for the union to
sell its labour to the firm, then any BNE of the bargaining game induced
by any bargaining procedure has the following property: when R> a with
a strictly positive probability the union goes on indefinite strike.

9.5.2 Litigation or Out-of-Court Settlement Revisited

Consider the bargaining situation between the defendant and the plantiff as
laid out in Section 9.3.2, but with the assumption that the defendant has
private information about 7 and the plantiff has private information about
x. Assume that x is a random draw from a probability distribution G, and
7 is an independent random draw from a probability distribution H.
It follows from Proposition 9.4 that the bargaining outcome can be ex-
post efficient if and only if ve{^) > ce(x). Under the American rule — in
which each party bears her cost of litigation — this implies that if / <
(7ex — 7Xe)/2, then the bargaining outcome cannot be ex-post efficient. On
the other hand, under the English rule — in which the loser at trial bears the
winner's cost of litigation — it follows that if / < (jex — 7x e )/2(l — 7 e + 7),
then the bargaining outcome cannot be ex-post efficient. Hence, since j e > 7,
the players are more likely to proceed to litigation under the English rule
than under the American rule.
Since 7ex — jxe is strictly greater than both j(x — xe) and x(je — 7), it
follows by comparing the results obtained here with those obtained in Section
9.3.2 that the players are more likely to proceed to litigation under two-sided
uncertainty than under one-sided uncertainty, whether they operate under
the American rule or under the English rule.

9.6 Bargaining Power and Uncertainty

This section studies a bargaining model in which exactly one player has pri-
vate information about her reservation value, while her opponent makes all
the offers. In Section 7.2.2 I studied the repeated-offers game with perfect
information, and established that (in the unique subgame perfect equilib-
rium) the player who makes all the offers obtains all of the gains from trade.
272 Asymmetric Information

This section explores the extent to which the equilibrium payoff of the player
who makes all the offers is adversely affected when her opponent has private
information. Another motivation for the model studied in this section is to
explore whether or not the uninformed player screens her opponent's private
information through time via the sequence of equilibrium price offers. The
existence of such a screening equilibrium would illustrate the notion that in
the presence of asymmetric information the bargaining procedure can be a
mechanism through which private information is revealed over time.
Two players, B and 5, bargain over the price at which to trade an
indivisible object — that is owned (or, can be produced) by player S —
according to the following procedure. At each time t = 0, A, 2 A , . . . , where
A > 0, the buyer (player B) makes a price offer p (p > 0) to the seller
(player S). If the seller accepts the price offer, then agreement is struck and
trade occurs. On the other hand, if the seller rejects the price offer, then
bargaining continues (and the game proceeds to time t + A). If agreement
is reached at time t on price p, then the payoffs to the seller and the buyer
are respectively (p — c) exp(—ri) and (v —p) exp(—rt), where c is the seller's
reservation value (or, cost of production), v the buyer's reservation value
and r (r > 0) the players' common discount rate. If the players perpetually
disagree (i.e., the seller rejects all the offers), then trade does not occur
and each player's payoff is zero. For notational convenience, define 6 =
exp(—rA). Notice that the discount factor 6 G (0,1).
The players' reservation values are independent of each other, and the
seller's reservation value is her private information. This asymmetry in
information is modelled as follows. The seller's reservation value is a random
draw from a probability distribution G. The seller knows the realization of
the random draw, but the buyer does not. The buyer only knows that the
seller's reservation value is a random draw from G. Hence, she believes that
the probability that the seller's reservation value is less than or equal to c is
G{c). Thus, G defines the buyer's prior beliefs, as they are her beliefs about
the seller's reservation value before bargaining begins. For short, I say that
G is the buyer's prior. The minimum value (or, infimum) and the maximum
value (or, supremum) of the support of G are respectively denoted by c and
c, where c > c > 0. That is, G(c) = 1, G(c) = 0 if c < c and G(c) > 0 if
c > c. When the seller's reservation value is c, it is convenient to call her
the c-type seller.
9.6 Bargaining Power and Uncertainty 273

An important feature of this bargaining game is that the buyer may

acquire some information about the seller's reservation value after any offer
is rejected. Thus, at each time t > A, before making her offer she will
update her beliefs about the seller's reservation value, and base her price
offer on her updated (or, posterior) beliefs.14 I employ the perfect Bayesian
equilibrium (PBE) concept to characterize the outcome of this bargaining
model, which ensures, in particular, that (in equilibrium) the buyer does
not update her beliefs in an arbitrary manner, but (whenever possible) she
updates her beliefs in accordance with Bayes' rule.
It is helpful to introduce the following notation to describe the buyer's
posterior beliefs about the seller's reservation value. Fix an arbitrary sub-
game beginning at time nA (n > 1). Denote by Gn the probability distri-
bution that defines the buyer's (posterior) beliefs at the beginning of this
subgame. Thus, she believes that with probability Gn(c) the seller's reserva-
tion value is less than or equal to c. Furthermore, denote by An the minimum
value (or, infimum) of the support of Gn — that is, G n(c) = 0 if c < An and
Gn(c) > 0 if c > Xn. Thus, An denotes the lowest possible reservation value
that the buyer believes the seller could have.

9.6.1 An Example of a Screening Equilibrium

In this section I derive some of the main insights in a simple manner. In
order to do so, however, it is convenient to assume that the buyer's prior G
is uniformly distributed on the closed interval [0,1] — that is, c = 1, c = 0
and G(c) = c (when c G [0,1]) — and that the buyer's reservation value
v = c.
Fix an arbitrary PBE in which the players use the following pair of
strategies.
• (Seller'sstrategy). Fix an arbitrary c G [0,1]. At time 0 the c-type seller
accepts price p if and only if p > a + (1 — a)c, where a > 0. Fix any time
t (t > A), and any history of price offers (po,Pi,... ,pt-A)- If the c-type
14
Although the buyer may also acquire some information about the seller's reservation
value after any offer is accepted, in this game that is irrelevant, since once a price offer is
accepted the game ends with trade taking place at the accepted price. In Section 9.6.3 I
shall, however, study the effect on the equilibria of this game of allowing the buyer the
option to retract her offer after it is accepted. The role of such an option on the bargaining
outcome in bargaining situations with perfect information has been studied in Section 7.3.
274 Asymmetric Information

seller rejected all these price offers, then at time t she accepts price p if and
only if p > a + (1 — a)c. 1 5
• (Buyer's strategy). If the buyer believes that the lowest possible reservation
value the seller could have is A (where A G [0,1]), then she offers price
p(\) = 0 + (1 - /?)A, where 1 > P > a.
Notice that the seller's strategy is stationary, in the sense that her de-
cision to accept or reject a price offer p at any time t (when she is still
around, which means that she did not accept any of the prices offered until
time t — A) depends only upon her reservation value c and the price offered
p — it does not depend upon the history of prices offered until time t — A.
Furthermore, notice that the buyer's strategy is Markovian, in the sense
that her price offer p at any time t depends only upon the lowest possible
reservation value that she believes at time t the seller could have — it does
not depend upon the history of prices offered until time t — A. 16 It may also
be noted that both strategies are 'linear', in the obvious sense.
This (stationary, linear) PBE is parameterized by two numbers: a and
P such that 1 > P > a > 0. Since there are many such values of a and
/?, there may exist many such equilibria. 17 The strategy of my analysis is
as follows. I first derive for an arbitrary such (stationary, linear) PBE the
buyer's equilibrium posterior beliefs at the beginning of any subgame. Then,
I show that there exists a unique pair (a*, /?*), where 1 > P* > a* > 0, such
that the strategies described above is a PBE if and only if (a, p) = (a*, /?*).
I conclude with a discussion of the main features of this unique (stationary,
linear) PBE.

Equilibrium Posterior Beliefs

Fix an arbitrary subgame beginning at time t = nA (n > 1). The buyer's
equilibrium posterior beliefs Gn at the beginning of this subgame will depend
on the seller's equilibrium strategy and the history of price offers. Let pn
15
Notice that I do not specify the c-type seller's behaviour — whether or not to accept
a price offer p — after any history of price offers (po,Pi, • •IPI-A)
• such that she accepted
one of these price offers. Another way to express the c-type seller's strategy is as follows.
She has a 'reservation' price — which equals a + (1 — a)c — and the instant a price greater
than or equal to it is offered, she accepts to trade at that price.
16
In general, and this will be the case in equilibrium, the history of prices determines
the buyer's beliefs at time t.
17
It should be noted that I have yet to establish the existence of such an equilibrium.
9.6 Bargaining Power and Uncertainty 275

denote the maximum price offered thus far — that is, p n is the maximum
of {poiPi,... ,pt-A}- And let An denote the minimum value (or, infimum)
of the support of Gn — it is the lowest possible reservation value the buyer
believes at time t the seller could have.
It follows from the seller's equilibrium strategy and Bayes' rule that
\ n — 0 if pn < a and Xn = c*(pn) if a < pn < 1, where

c (p) = l
1— a
and, moreover, Gn(c) = (c — A n )/(1 — Xn) if An < c < 1, and Gn{c) = 0 if
0 < c < A n . 18
Notice that Gn is a truncation of the (uniformly distributed) prior at
An. Hence, the buyer's equilibrium posterior beliefs at the beginning of
any subgame can be characterized by a unique number, namely, the lowest
possible reservation value that the buyer believes the seller could have.

Existence and Uniqueness

At time 0 the equilibrium price offered po = /?, which the seller accepts if
and only if her reservation value c < c*(/3). Hence, at time 1 the equilibrium
price offered pi = /3 + (1 - /?)c*(/?), since the lowest possible reservation
value that the buyer believes (at time 1) the seller could have is c*(/3). The
seller accepts p\ if her reservation value c is such that c*(/3) < c < c*(pi).
It thus follows that po — c*(/?) = 8[p\ — c*(/3)].19 After substituting for p0?
pi and c*(/?), and then simplifying, it follows that

a = PS. (9.23)
Hence, the pair of strategies stated above is a PBE only if a and (3 satisfy
equation 9.23. I now show that the seller's strategy is immune to profitable
one-shot (unilateral) deviations if and only if a and /3 satisfy equation 9.23.
18
Given the seller's and the buyer's equilibrium strategies, the buyer's posterior beliefs
Gn when pn > 1 are irrelevant — since the buyer never offers a price strictly greater than
one, and all types of sellers always accept any price that is greater than or equal to one.
19
Suppose, to the contrary, that po — c*(j3) > 6[pi — c* (/?)]. This implies that there
exists a c G (c*(/3),c*(pi)) such that po — c > 8{p\ — c), which contradicts the result
(stated above) that such a c-type seller rejects po, and then accepts p\. Now suppose, to
the contrary, that po-c*(/3) < 6\pi-c*(/3)]. This implies that there exists a c G (0,c*(/3))
such that po — c < 6(pi — c), which contradicts the result that such a c-type seller accepts
276 A s y m m e t r i c Information

Fix an arbitrary c G [0,1], a subgame in which the c-type seller is still

around, and a price offer p' > 0. First suppose that p' > a + (1 — a)c, which
implies that in equilibrium she accepts p1. Thus, her equilibrium payoff is
pr ~c. Now suppose she considers a one-shot deviation: that is, she rejects pf,
and then plays according to her equilibrium strategy. Her payoff from this
one-shot deviation is 6\p(c*(p')) - c], where p(c*(p')) = /? + (1 - /3)c*(p').20
If a and /3 satisfy equation 9.23, then p' — c> S\p(c*(pf)) — c], as required.
Now suppose that p' < a + (1 — a)c, which means that in equilibrium
the c-type seller rejects this price offer. Since (after rejecting p') she can
accept the equilibrium price offered A time units later — which I denote
by p" — it follows that her equilibrium payoff Us(c) > 8{p" — c). It is
straightforward to show, as required, that if a and /3 satisfy equation 9.23,
then 8{p" — c) > p' — c, where pr — c denotes her payoff from the one-shot
deviation in which she instead accepts p'?l
Let C/B(A) denote the buyer's equilibrium payoff at the beginning of any
subgame when the lowest possible reservation value the buyer believes the
seller could have is A. Fix any such subgame, and let WB(P, A) denote the
buyer's payoff if she offers price p at the beginning of this subgame and then
(if the price is rejected) she plays according to her equilibrium strategy. The
buyer's equilibrium strategy is immune to profitable one-shot (unilateral)
deviations if and only if

for all A E [0,1], WB(p(\), A) > WB(p, A) for all p > 0,

where p(X) =/3 + ( l - / ? ) A .

It is trivial to note that if A = 1, then WB(1, 1) = 0 > WB(p, A) for all
p > 0. Now suppose that A < 1. If p > 1 then WB(P, A) = 1 — p, and if p <
a + ( 1 - a ) A then WB(P, A) = 6UB(X). Furthermore, if 1 > p > a + ( l - a ) A ,
20
Since the c-type seller is still around, the lowest possible reservation value that the
buyer believes (when making the price offer p) the seller could have is less than or equal
to c. This implies, since p > a + (1 — a)c, that if p' is rejected, then the lowest possible
reservation value that the buyer believes the seller could have is c*(p). Hence, she offers
the equilibrium price p(c*(pf)). Since p > a + (1 — a)c, this equilibrium price offer would
be accepted by the c-type seller if she rejects the price p .
21
If p' < a + (1 — a)A;, where A; (A; < c) denotes the lowest possible reservation
value that the buyer believes (when making the price offer p') the seller could have, then
p" = p(\') = £ + ( ! _ p)\'. And if p' > a + (1 - a)A', then p" = p(c*(pf)).
9.6 Bargaining Power and Uncertainty 277

then
c*(p)-\ 1 - c*{p)
WB(p,\) = (I-P) + 8UB(c*(p)).
1-A 1-A

Since for any A < 1 and p $ (a + (1 — a)X, 1), WB(P, A) < 0, it follows that
for all A < 1

C
f/ B (A)= max \P) T (l-p)+ ?C Pj
6UB{c*(p)).
l>p>a-\-(l-a)X 1— A 1— A
(9.24)

Furthermore, it follows that the buyer's equilibrium strategy is immune to

one-shot deviations if and only if for all A < 1, a solution to the maximization
problem defined in 9.24 is p = p(A).
It is convenient to define VB(A) = (1 — A)UB(A), and rewrite the state-
ment in (9.24) as follows

VB(X) = max (c*(p) - A)(l - p) + 6VB(c*(p)). (9.25)

Let p* denote an arbitrary solution to the maximization problem defined in

(9.25). Applying the Envelope Theorem to (9.25), it follows that

(9.26)

The first-order condition, which p* must satisfy, is

(9.27)

Substitutingp* = p(A) into (9.27), it follows that p(X) satisfies the first-order
condition if and only if

(9.28)
1-a

After substitutingp* = p(A) into (9.26), it follows that V£(A) =

A). This implies that

(l-p)\-a
(9.29)
l-a
278 Asymmetric Information

Using (9.29) to substitute for Vg(.) in (9.28), it follows that for all A < 1,
p(X) satisfies the first-order condition if and only if

(1 - a ) ( l - 2(3 + a) - 6(1 - (3)2 = 0. (9.30)

If a and (3 satisfy equation 9.23, then the second-order condition is satis-

fied.22 Hence, since (9.26) implies that the first-order condition (9.27) has
at most a unique solution, it follows that for all A < 1, the unique solution
to the maximization problem defined in (9.24) is p — p(X) if and only if a
and (3 satisfy equations 9.23 and 9.30. This thus establishes that the buyer's
equilibrium strategy is immune to profitable one-shot (unilateral) deviations
if and only if a and (3 satisfy equations 9.23 and 9.30.
There exists a unique pair (a, (3) such that 1 > (3 > a > 0 which satisfies
equations 9.23 and 9.30, namely

(9.31)

Consequently, there exists a unique (stationary, linear) PBE in which the

players use the pair of strategies described above with a and (3 taking the
values stated in (9.31).

Main Insights
The sequence of prices (pn) and the sequence of the buyer's posteriors (An)
along the equilibrium path of the unique (stationary, linear) PBE are defined
by the following two difference equations: pn = (3 + (1 — (3)\n and A n+ i =
c*(pn), given that Ao = 0, where a and (3 are defined in (9.31). Solving these
equations, it follows that

pn = 1 - (1 - (3)(3n and An = 1 - (3n (n = 0,1, 2 , . . . ) .

Notice that 1 > p n +i > pn > 0 and 1 > A n+ i > An > 0 (for all n —
0,1,2,...)-
22
An arbitrary solution p* to the maximization problem defined in (9.25) satisfies the
second-order condition if and only if —2(1 —a)+<5V^((p* —a)/{I —a)) < 0. It follows from
(9.26) that VB(A) = dp*/d\. Hence, at p* = p(A), Vg (A) = 1-/3. Therefore, p* = p(A)
satisfies the second-order condition if and only if —2(1 — a) + 6(1 — /3) < 0. The latter
inequality is satisfied if a and j3 satisfy equation 9.23.
9.6 Bargaining Power and Uncertainty 279

Along the equilibrium path of this unique (stationary, linear) PBE, the
prices are strictly increasing, and trade occurs at each time with a strictly
positive probability. The time at which trade occurs is increasing in the
seller's cost of production (or, reservation value). A low cost seller trades
at an earlier time and at a lower price, since she is more eager to trade
and does not want to wait to trade at a higher price that is offered in the
future. Indeed, the buyer screens the seller's private information through the
equilibrium price path — the seller's privately held information is (partly)
revealed through time. It is cas if the buyer engages in intertemporal price
discrimination.
The equilibrium is inefficient, since trade occurs with positive probabil-
ity after time 0 — which is costly, since both players discount future payoffs.
Furthermore, trade does not occur with probability one in finite time. How-
ever, since pn —•» 1 (and An —> 1) as n —» oc, it follows that trade occurs
with probability one in infinite time.
I now discuss the properties of the PBE in the limit as A —> 0. In this
limit, <5 —» 1, which implies that the first price offered po —> 1. Hence, as
the time interval between two consecutive offers becomes arbitrarily small,
the buyer's equilibrium price offer at time 0 becomes arbitrarily close to
the 'reservation' price of the highest cost-type seller — which means that
(in this limit) all types of the seller accept the first price offered. Hence,
in this limit there is no screening of the seller's private information, and
trade occurs with probability one at time 0. Furthermore, in this limit, the
buyer's equilibrium payoff converges to zero. The intuition behind these
limiting properties is straightforward. In the limit as A becomes arbitrarily
small, the cost of rejecting any offer to a c-type seller (for any c G [0,1])
becomes arbitrarily small. Hence, she is willing to wait to obtain a high price
that might be offered in the future. Thus, the buyer effectively gives up any
attempt to screen (and engage in intertemporal price discrimination), and
consequently, she offers the price p = 1 at time 0.
It follows from the above discussion that the screening equilibrium sat-
isfies the following two properties — which are related to those of the equi-
librium pricing strategy of a durable-good monopolist (cf. Coase (1972)).

Property 9.1 (Coasian Dynamics). Prices are monotonic across time,

and the more eager a player the earlier she trades.
280 Asymmetric Information

Property 9.2 (Coase Conjecture). In the limit, as A —> 0, (i) all po-
tential gains from trade are realized without any costly delay, and (ii) the
profit of the uninformed player (who makes all the offers) is arbitrarily close
to zero — that is, she loses almost all the bargaining power that making
offers confers.
There are two main messages contained in the screening equilibrium de-
rived above. Firstly, the bargaining process may reveal none of the privately
held information if and only if the cost of haggling is arbitrarily small. And
secondly, the player who makes all the offers loses all her bargaining power
when her opponent has private information if and only if the cost of haggling
is arbitrarily small.23

9.6.2 General Results

I now analyse the set of all PBE of the model. Unless otherwise stated,
the formal results in this section impose no restrictions on the buyer's prior
G. However, for ease of interpretation, it is helpful to assume that G is
continuously distributed on the closed interval [c, c]. The first preliminary
result is Lemma 9.3 below, which implies that in any PBE if a high cost
seller accepts to trade at price p with a positive probability, then any lower
cost seller accepts to trade at this price for sure (with probability one). The
intuition for this result comes from the observation that a low cost seller is
more eager to trade than a high cost seller.

Lemma 9.3 (The Skimming Property). Fix an arbitrary PBE, a reser-

vation value c, a price p, and a subgame beginning with the buyer Js offer.
If at the beginning of this subgame the c-type seller accepts the price p with
a strictly positive probability, then, for any cr < c, at the beginning of this
subgame the c1 -type seller accepts the price p with probability one.

Proof. The hypothesis implies that p — c > £/#(c), where Us(c) is the PBE
payoff to the c-type seller if she rejects the price p at the beginning of
this subgame. I now show that for any d < c, p — d > Us(d). After
rejecting the price p, the c'-type seller will accept with positive probability
the equilibrium price pt+nA offered by the buyer at time t + nA for some
23
Notice therefore that the buyer's bargaining power is intimately connected with her
ability to screen the seller's private information across time.
9.6 Bargaining Power and Uncertainty 281

n > 1, where t denotes the time at which the subgame under consideration
begins. Hence, Us(d) = <5n(pt+nA — cr). Since the c-type seller can mimic
the c'-type seller's strategy, it follows that Us(c) > 6n(pt+nA — c). Hence,
U s ( c ) > Us(c') + 6n(cf - c ) . S i n c e p - c > U s ( c ) , i t f o l l o w s t h a t p - d >
Us(d) + (1 — S )(c — cr). The desired result follows immediately, since 6 < 1
n

and n > 1. •

As is the case in the screening equilibrium derived in the previous section,

it is now shown that in any PBE the buyer's posterior belief at the beginning
of any subgame is a truncation of her prior G. Fix an arbitrary PBE and a
price p > 0. Suppose that at time 0 the buyer offers the price p. Either it is
rejected with probability one, or it is accepted with positive probability. In
the latter case, it follows from Lemma 9.3 that there exists a Ai G [c, c] —
where Ai in general depends on p — such that the price p is accepted if and
only if the seller's reservation value c < \±. Letting G\ denote the buyer's
posterior at time A after the rejection of the price p, it follows from Bayes'
rule that Gi(c) = [G(c) - G(Ai)]/[l - G(Ai)] if Ai < c < c and Gi(c) = 0
if c < c < Ai. The following corollary, which generalizes this relationship,
follows immediately from the above argument.

Corollary 9.1. Fix an arbitrary PBE and a time nA (where n > 1). In
any subgame beginning at time nA there exists a \ n G [c, c] such that the
buyer's posterior Gn at the beginning of this subgame is a truncation of the
prior G at An. That is, Gn{c) = [G{c] - G(A n )]/[l - G(An)] if Xn < c < c
and Gn{c) = 0 if c< c < Xn.

Corollary 9.1 implies that (given the prior G) in any PBE the buyer's
posterior at any time nA can be characterized by a unique number, namely,
An (which is the lowest possible reservation value the buyer believes that
the seller could have). Lemma 9.4 below implies that in any PBE the buyer
never offers a price strictly greater than c, which, in turn, implies that for
any c G [c, c] the c-type seller always accepts the price p = c.

L e m m a 9.4. Fix an arbitrary PBE and time t. In any subgame begin-

ning at time t, the equilibrium price offered pt at time t is such that pt <
min{c, v}.
282 Asymmetric Information

Proof. By contradiction. Suppose that there exists a PBE in which the

maximum price ever offered by the buyer is p > min{c, v}.24 Fix a subgame
in which at the beginning of this subgame the buyer's equilibrium price offer
is p = p. Since this is the maximum price ever to be offered, trade occurs
with probability one, and hence, the buyer's PBE payoff in this subgame is
v — p. If min{c, v} = v, then a contradiction is obtained, since the buyer's
equilibrium payoff is strictly negative. Now suppose that min{c, v} = c, and
suppose the buyer considers the following one-shot deviation: she offers the
price pr = p — 6 where e > 0 and satisfies p' — c > 8{p — c) for all c G [c, c].
Since p > c, such a price exists, and, hence, it is accepted with probability
one. This implies that the buyer's payoff in this subgame from this one-shot
deviation strictly exceeds v — p, which is a contradiction. •

Lemmas 9.3-9.4 and Corollary 9.1 are valid whatever the magnitude of
the buyer's reservation value v relative to c. It is now shown that if v = c
then there exists a continuum of PBE, but if v > c then there exists a unique
PBE. When v > c, gains from trade exist with probability one and they are
bounded away from zero. However, when v = c, gains from trade need not
exist. It may be recalled that the screening equilibrium described in Section
9.6.1 is based on the assumption that v = c.

The Gap Case

Here it is assumed that v > c, which is called the gap case. It is instructive
to first of all understand why the (stationary, linear) strategies described in
Section 9.6.1 is not a PBE. Assume, as is done in Section 9.6.1, that the
buyer's prior G is uniformly distributed on the closed interval [0,1]. But,
in contrast to the assumption made in Section 9.6.1, assume that v > 1.
It follows from the buyer's strategy that for any A < 1, p(X) < 1. Thus,
even if the lowest possible reservation value the buyer believes the seller
could have is arbitrarily close to one, the buyer does not offer the price
equal to one. This is not optimal when v > 1. The intuition for this
result is straightforward, and runs as follows. When the buyer is sufficiently
'pessimistic' — that is, she believes A is sufficiently close to one — then
24
It should be noted that p cannot be infinite. For otherwise, the buyer would obtain
a negative payoff, which is impossible, since the buyer can guarantee a payoff of zero by
always offering the price p — 0.
9.6 Bargaining Power and Uncertainty 283

(since v > 1 and 6 < 1) her payoff from offering the price equal to one
strictly exceeds the maximum possible payoff she can obtain from offering a
price strictly less than one. 25
The above intuitive argument is, in fact, valid for any prior G and any
PBE. Indeed, in any PBE and at the beginning of any subgame in which
the buyer is sufficiently pessimistic — that is, the lowest possible reservation
value the buyer believes the seller could have is sufficiently close to c —
the buyer offers the price equal to c. It then follows that in any PBE
trade occurs with probability one in finite time, since (by Corollary 9.1)
at some finite point in time the buyer will become sufficiently pessimistic.
Hence, unlike the screening equilibrium derived in Section 9.6.1, when v >
c bargaining terminates in finite time. However, just like the screening
equilibrium derived in Section 9.6.1, when v > c in the unique PBE — which
is formally stated below in Proposition 9.5 — the equilibrium path prices are
strictly increasing, and the buyer screens (some of) the seller's privately held
information. Furthermore, the unique PBE satisfies the Coase conjecture
— which, as mentioned above, is also satisfied by the screening equilibrium
derived in Section 9.6.1.

Proposition 9.5 (Gap Case). If v > c, and either G{c) > 0 or G has a
strictly positive and continuous density at c, then there exists a (essentially)
unique PBE. In particular, the unique PBE path of play is characterized as
follows. There exists an N such that the price offered at time 0 is po > c,
and the price offered at time NA is PNA = c. Furthermore, the equilibrium
path prices are strictly increasing: pt+A > Pt for t = 0, A, 2 A , . . . , (TV — 1)A.
The seller accepts each of these TV + 1 price offers with a probability that is
strictly positive but strictly less than one. In the limit, as A —» 0, p*(0) —» c
and trade occurs with probability one at time 0.

Proof. The proof is based on a backward induction argument, since in any

PBE bargaining ends in finite time. However, the proof is rather complex.
Hence, I omit it and refer the reader to Fudenberg, Levine and Tirole (1985)
and Gul, Sonnenschein and Wilson (1986). •
25
Her payoff from offering the price equal to one is v — 1, since (by Lemma 9.4) all types
of sellers will accept this price offer. On the other hand, her maximum possible payoff
from offering a price strictly less than one is arbitrarily close to 6(v — 1) as A becomes
arbitrarily close to one. Indeed, since 6 < 1 and v > 1, there exists a A* < 1 such that if
A > A* then it is optimal to offer the price equal to one.
284 Asymmetric Information

Notice that the unique PBE possesses most of the key properties of the
screening equilibrium derived in Section 9.6.1. Indeed, this unique PBE is
also a screening equilibrium. The only main difference is that trade occurs in
finite time with probability one. Hence, with probability zero the players will
perpetually disagree and fail to reach a price agreement. The equilibrium
is nevertheless inefficient, because with a strictly positive probability trade
occurs after some costly delay. However, the delay to agreement vanishes in
the limit as the time interval between two consecutive offers tends to zero.
Furthermore, in this limit, the buyer's payoff converges to v — c, and the
lowest cost seller earns a payoff of c — c. It can be said that, in this limit, a
player looses much of the bargaining power that making offers confers when
her opponent (the responder) has private information.

The No Gap Case

Now assume that v — c, which is called the no gap case. In any PBE of the
no gap case, trade does not occur with probability one in finite time — as is
shown in the screening equilibrium derived in Section 9.6.1. This is because
in any PBE there is no incentive for the buyer to offer a price greater than
or equal to v at any finite point in time. This implies that, unlike in the gap
case, one cannot use a backward induction argument to characterize the set
of PBE. It turns out that there exists a multiplicity of PBE. In particular,
there exists a large number of PBE that do satisfy the Coase conjecture,
but there also exists a large number of PBE that do not satisfy the Coase
conjecture. Hence, even when the costs of haggling are arbitrarily small, it
is possible (in the no gap case) for the buyer to retain the bargaining power
that making offers confers even when the seller has private information.
Let UB denote the buyer's unique PBE payoff if she could make a take-
it-or-leave-it-offer to the seller. Thus, UB > (v — p)G(p) for all p > 0. In
any PBE of the repeated-offers model, the buyer's equilibrium payoff is less
than or equal to UB- The buyer would obtain a payoff equal to UB (in the
repeated-offers model) if she could commit to always offering the price p*,
where (v — p*)G(p*) > (v — p)G{p) for all p > 0. However, making such a
commitment stick is rather difficult. Suppose she offers price p* at time 0.
Any c-type seller such that c > p* will not accept this price offer. Hence, if
this price offer is rejected, then at time 1 the buyer will update her beliefs,
and consequently, she has an incentive to offer a higher price (and thus break
9.6 Bargaining Power and Uncertainty 285

her commitment). I call UB the buyer's commitment payoff.

The following proposition provides a characterization of the set of buyer
payoffs sustainable as a PBE.

Proposition 9.6 (No Gap Case). Assume that v = c, and that G is

continuously distributed on the closed interval [c, c]. Let UB be such that
UB > (v — p)G(p) for all p > 0. For any e > 0 there exists a A' > 0 such
that for any A < A' and any UB G [C,UB — e] there exists a PBE such that
the buyer's payoff equals UB-

Proof. The basic idea of the proof involves the construction of reputational
equilibria. In such PBE almost any path of play is supported as a PBE path
by the threat of reverting play to a PBE that satisfies the Coase conjecture
if the buyer deviates from any given path of play 26 However, the details
of the proof are complex. Hence, I omit the proof and refer the interested
reader to Ausubel and Deneckere (1989) — the authors of this result. •

Thus, when the time interval between two consecutive offers is arbitrarily
small, there exists a PBE in which the uninformed buyer's payoff is arbi-
trarily close to her commitment payoff UB- AS such this result overturns
the Coase conjecture that characterizes the unique PBE in the gap case (cf.
Proposition 9.5).

9.6.3 The Effect of Retractable Offers

This section extends the model studied above by allowing the buyer the
option to retract her offer after it is accepted (and then to make a new offer).
In Section 7.3 I have studied the role of such an option, but in the context
of the alternating-offers model with perfect information. The pros and cons
of this option are discussed in Section 7.3.4. In the context of the current
setting of asymmetric information, this option may have a significant effect
on the nature of the equilibria, since accepting an offer may reveal some
of the seller's privately held information. Indeed, it is shown below that
provided A is sufficiently small, a result similar to Proposition 9.6 is valid
in both the gap case and the no gap case.
26
In a PBE that satisfies the Coase conjecture, the buyer's payoff is arbitrarily close to
zero.
286 Asymmetric Information

Let me first clarify the nature of the extension. Whenever the seller
accepts a price offer p, the buyer then has to decide whether or not to
retract it. If she does not retract it, then trade occurs at price p. But if
she retracts the price offer, then (A time units later) she makes a new price
offer to the seller.
The following lemma contains the key result behind the analysis of the
set of PBE of this extended model. It describes a PBE — for values of
6 > 1/2 (i.e., for values of A sufficiently small) — in which trade occurs
with probability one at time 0 on price i>, and the buyer's equilibrium payoff
equals zero. 27

state s* state sf
offer p= v p= c

buyer retracts p>v p> v(l-6)+ 6c

beliefs G D

c-type seller accepts p— v p>c

transitions switch to state absorbing

s' if a price of-
fer p < v is ac-
cepted by the
seller

Table 9.1: A PBE in the repeated-offers model with asymmetric information and
retractable offers, where G is the buyer's prior beliefs and D is the (degenerate) distribution
function whose mass is concentrated at c = c (i.e., D(c) = 1 and D(c) = 0 for c < c).

L e m m a 9.5. If v > c and 6 > 1/2, then the pair of strategies and buyer
beliefs described in Table 9.1 is a PBE of the repeated-offers model with
asymmetric information and retractable offers. In equilibrium trade occurs
with probability one at time 0 on price v, and the buyer's equilibrium payoff
equals zero.
2
It should be noted that in the model studied above — when the buyer does not have
the option to retract offers — such a PBE does not exist. Furthermore, in the gap case,
when v > c, such a PBE does not exist even in the limit as A —» 0.
9.6 Bargaining Power and Uncertainty 287

Proof. First, note that in state sf the buyer believes that with probability one
the seller's reservation value c = c. Hence, since this is an absorbing state,
the actions of the players in this state is the unique SPE of the repeated-
offers model with perfect information. Now consider the initial state s*. It
is easy to verify using the One-Shot Deviation property that no player can
benefit from a one-shot (unilateral) deviation. For example, suppose a price
pf < v(l — 6) + 6 c is offered. In the proposed equilibrium the c-type seller
(for any c E [c, c]) rejects this price offer, and obtains a payoff of 6(v — c).
Suppose she considers a one-shot deviation, and instead accepts the offer. In
that case the state switches to s1, and the buyer does not retract this offer,
which implies that the c-type seller's payoff from this one-shot deviation is
pr — c. Since 6 > 1/2 — and I note that this is the only point in the proof at
which this condition is required — it follows that 6v + (l — 6)c > v(l — 6)+6c.
This inequality implies that pf < 6v + (1 — 6)c. That is, pf — c < 6{v — c).
Hence, the one-shot deviation is not profitable. •

Along the equilibrium path of the PBE described in Table 9.1, the buyer
offers price v1 which is accepted by all seller types, and the buyer does
not retract this price offer. In equilibrium, therefore, trade occurs with
probability one at time 0 on price v. The buyer's equilibrium payoff equals
zero.

Remark 9.2. In the PBE described in Lemma 9.5, if a price p < v is ac-
cepted by the seller in the initial state s*, then a 'zero-probability' event
has occurred, since (given the seller's strategy) no seller type accepts such
a price in the initial state. Thus, the buyer cannot use Bayes' rule to up-
date her beliefs about the seller's reservation value. In the PBE described
above, it is assumed that the buyer conjectures that the deviation (i.e., the
acceptance of the price p < v in the initial state) was made by the lowest
cost-type seller. This optimistic conjecture is not unreasonable, since the
type who has the strongest incentive to deviate is the seller with the lowest
cost. Thus, many refinements of the PBE concept — including, for example,
those proposed and used in Rubinstein (1985a), and Grossman and Perry
(1986) — would not eliminate this PBE. Having said this, in Muthoo (1994),
I construct an alternative PBE which also supports the outcome of the PBE
described in Lemma 9.5, but has the feature that the buyer makes only
reasonable pessimistic conjectures. I therefore suspect that the 'zero buyer
288 Asymmetric Information

profit' PBE outcome is robust to any refinement of the PBE concept that
only acts to constrain the buyer's off-the-equilibrium-path beliefs.

I call the PBE stated in Lemma 9.5 the zero buyer-profit equilibrium.
Given this 'extremal' equilibrium, it is straightforward to construct a con-
tinuum of other PBE (when v > c and 6 > 1/2). In particular, any price
path (pt) is supported as a PBE path as follows. If at any time t the buyer
does not offer pt, or retracts the price pt-> then immediately play proceeds
according to the zero buyer-profit equilibrium. The seller's strategy is a
best response to the buyer's strategy. I omit the details of the construc-
tion of such reputational equilibria, and simply state the main result in the
following proposition.

Proposition 9.7. If v > c and 6 > 1/2, then any buyer payoff between zero
and UB can be supported by a PBE of the repeated-offers model with asym-
metric information and retractable offers, where UB is defined in Proposition
9.6.

By comparing Propositions 9.5-9.7, it is clear that the impact of the

option to retract offers is non-trivial. When offers are retractable there is
no relevant distinction between the gap case and the no gap case: the set
of equilibria does not depend on the relative magnitude of v and c. This
seems like a more plausible conclusion, since the sharp 'discontinuity' in the
equilibrium set at v = c in the model without retractable offers is somewhat
problematic. Furthermore, with retractable offers the Coase conjecture does
not hold when v > c. This result is also more plausible, since it is unlikely
that the player who makes all the offers will lose all her bargaining power
when her opponent has private information. It should also be noted that
Proposition 9.7 is a much stronger result than Proposition 9.6 in the sense
that it is valid for any A G (0, A'), where A' is the unique solution to
exp(—rA) = 1/2, while the full force of Proposition 9.6 comes only in the
limit as A —> 0. As such with retractable offers there exists a PBE in which
the buyer obtains her commitment payoff UB even when A > 0. In contrast,
the same is true in the model without retractable offers if and only if v = c
and A is arbitrarily close to zero.
9.7 An Application to Wage-Quality Contracts 289

9.7 An Application to Wage-Quality Contracts

Consider a firm and a worker who are bargaining over the wage w and the
quality q of a unit of output, where w > 0 and q > 0. If agreement is
reached at time t (t = 0, A, 2A,...) on a wage-quality contract (w, q), then
the payoffs to the firm and the worker are respectively

y/q — w and w — 9q,

where 9 is the worker's private information. This asymmetry in information

is modelled in the usual way by considering the value of 9 to be a random
draw from some distribution, with its realization only revealed to the worker.
Assume that the distribution from which 9 is randomly drawn is as follows:
9 — 2 with probability a, and 9 = 1 with probability 1 — a, where 0 < a < 1.
I call the worker 'good-type' if 9 = 1, and 'bad-type' if 9 = 2.
At each point in time t the firm offers a menu of wage-quality contracts.
More precisely, an offer is a pair {m^ra^}, where rrii = (wi,qi) (i = b,g).
The worker then decides whether to accept one of the two contracts on offer,
or to reject both of them (in which case bargaining continues with the firm
proposing a new menu of contracts at time t + A). If the players perpetually
disagree, then each player obtains a payoff of zero. Thus, this bargaining
model is an extension to the one studied in Section 9.6. The key difference
is that the uninformed player now has two devices through which she can
attempt to screen the informed player's private information: through time
(as in the model above) and through the menu of contracts (at a single point
in time).
As I show below, since the players are bargaining over two variables (or,
two 'dimensions'), this additional screening device turns out to significantly
affect the nature of the PBE. However, it may be noted that when bargaining
over a single dimension (such as price), the set of PBE is independent of
whether or not the uninformed player — who makes all the offers — may
offer a menu of (price) contracts.
It will be shown below that for any A > 0, in the unique PBE agreement
is reached at time 0 with probability one, and, furthermore, the firm retains
all of its bargaining power. This striking result sharply contrasts with the
PBE derived in the model studied in the previous section in which the
players bargain over a single dimension. Before proceeding further, it may
290 Asymmetric Information

be noted that with perfect information about the value of #, the quality that
maximizes the 'surplus' (or, gains from trade) y/q — Oq is q* = 1/402, which
is called the first best quality level.

9.7.1 The Commitment Equilibrium

As a preliminary exercise, I derive the menu of contracts that maximizes

the firm's expected profit subject to appropriate incentive-compatibility and
individual-rationality constraints. That is

(^qb - wb) + (1 - a)(y/qg - wg)

subject t o wb — 2qb > wg — 2qg, wg — qg > wb — qbi wb — 2qb > 0 and

w
g ~ Qg > 0- The former two inequalities respectively are the incentive-
compatibility constraints for the bad-type worker and the good-type worker,
while the latter two inequalities are their respective individual-rationality
constraints. The solution to this maximization problem characterizes the
equilibrium of the model in which the firm makes a take-it-or-leave-it-offer.
As such I call it the commitment equilibrium, since it involves the firm being
able to commit itself to making a single offer. It may be noted that the
firm's payoff in any PBE of the repeated-offers model is less than or equal
to her payoff in the commitment equilibrium. 28
Let us now solve this maximization problem. It is straightforward to
show that the single-crossing property of the worker's payoff function im-
plies that at the optimum only the bad-type worker's individual-rationality
constraint and the good-type worker's incentive-compatible constraint bind.
Thus, at the optimum, wb = 2qb and wg = wb + qg — q^. Using these expres-
sions to substitute for Wb and wg in the maximand, it follows that I need to
now solve the following unconstrained maximization problem

- 2qb) + (1 - a)(y/q^ - qg - qb).

28
One may, alternatively, provide the following (normative) interpretation of the maxi-
mization problem stated above. Its solution characterizes the maximum expected profit to
the firm in any incentive-compatible and individually-rational direct revelation procedure.
Hence, the Revelation Principle (which is stated in Section 9.2.2) implies that its solution
characterizes the maximum expected profit to the firm in any BNE of any bargaining
game.
9.7 An Application to Wage-Quality Contracts 291

The first-order conditions imply that the solution is

Hence, it follows that

a2 , _ 2a2 + 2a + 1

Therefore, the commitment equilibrium — i.e., the solution to the con-

strained maximization problem stated above — is the following menu of
contracts: {m^, m^}, where fhi = (u^, %). It may be noted that in the com-
mitment equilibrium the good-type worker produces her first-best quality,
while the bad-type worker a suboptimal quality. Furthermore, the good-
type worker's payoff is strictly positive, while the bad-type worker's payoff
is zero.

9.7.2 The Unique Perfect Bayesian Equilibrium

In the following proposition I characterize the unique PBE of the repeated-
offers model described above (in which the firm cannot commit to making
a single offer).

Proposition 9.8. For any A > 0 there exists a unique PBE. In equilibrium
agreement is reached at time 0 with probability one. Thefirm'sequilibrium
offer at time 0 is the menu of contracts {fn^, fhg} that is offered in the com-
mitment equilibrium stated above. The bad-type worker accepts the contract
nib and the good-type worker accepts the contract mg.

Proof. Since the proof is rather lengthy, I omit it, and instead refer the
reader to Wang (1998) — the author of this result. However, I now sketch
an argument which shows that there exists a PBE that supports the out-
come stated in the proposition. If the proposed equilibrium offer at time 0
— namely, {m^ra^} — is rejected, then (since in the proposed equilibrium
both types accept one of the two contracts in the equilibrium menu) a 'zero-
probability' event has occurred. In the PBE the firm conjectures (following
this zero-probability event) that the worker is good-type with probability
one. Hence, in the subgame that begins at time 1, the players' strategies are
292 Asymmetric Information

the unique SPE strategies of the repeated-offers model with perfect infor-
mation (that the worker is good-type). In this equilibrium the firm always
offers a single contract, namely, (w,q) — (1/4,1/4). Such a contract gives
the bad-type worker a negative payoff and the good-type worker a payoff of
zero. Hence, both worker types will accept the appropriate contract from
the equilibrium menu offered at time 0. The existence of the desired PBE
follows immediately, since the firm has no incentive to offer at time 0 a
menu that differs from {ra^ra^} — because, by definition, this equilibrium
menu generates the maximum possible payoff that she can obtain in any
PBE. 29 •

This is a striking result, for many reasons. It shows that (in multi-
dimensional bargaining situations) the commitment equilibrium is supported
in the unique PBE of the repeated-offers model in which commitment to a
single offer is not assumed possible. Furthermore, the unique PBE does
not satisfy the Coase conjecture, since the firm retains all its bargaining
power. In particular, the firm obtains her commitment equilibrium payoff.
Moreover, agreement is reached without any costly delay. Of course, the
PBE outcome is not ex-post efficient, since the bad-type worker does not
produce the first best quality.

9.8 Notes
The normative question studied in Sections 9.2 and 9.4 was first studied by
Myerson and Satterthwaite (1983) in the context of private values, and by
Samuelson (1984) in the context of correlated values. Indeed, Proposition
9.3(b) is due to Myerson and Satterthwaite (1983). The focus of atten-
tion in these two sections is on whether or not the bargaining outcome can
be ex-post efficient. When the bargaining outcome cannot be ex-post effi-
cient, it might be interesting to derive the maximum expected gains from
trade consummated in any BNE of any bargaining game. By the Reve-
lation Principle this involves characterizing the incentive-compatible and
individually-rational DRP with maximal consummated expected gains from
29
The proof of uniqueness involves two key results. First, one shows that in any PBE
the bad-type worker earns a payoff of zero. And then, using this result, one shows that in
any PBE the firm's payoff equals her commitment equilibrium payoff.
9.8 Notes 293

trade. Myerson and Satterthwaite (1983) have addressed this normative is-
sue in the context of private values, while Samuelson (1984) does the same
with correlated values.
The application to litigation (in Sections 9.3.2 and 9.5.2) is inspired by
Spier (1994) and Hay and Spier (1998). Spier (1994) studies the effect on the
likelihood of an out-of-court settlement of various legal rules that allocate the
private costs of litigation between the plantiff and the defendant — see also
Bebchuk (1984) and Reinganum and Wilde (1986). Hay and Spier (1998)
provide a nice discussion of the many fascinating issues involved on this
topic, and a guide to the literature on litigation and (pre-trial) bargaining.
The repeated-offers bargaining model with asymmetric information stud-
ied in Section 9.6 was first studied by Fudenberg, Levine and Tirole (1985)
and Gul, Sonnenschein and Wilson (1986) — in the context of private values
and without retractable offers. Proposition 9.5 is due to them. However,
Proposition 9.6 is due to Ausubel and Deneckere (1989). For an alternative
exposition of the model, see Fudenberg and Tirole (1991) — who also show
how this model may be reinterpreted as a model of the pricing behaviour
of a durable-good monopolist. As such the model underlies the large liter-
ature on the Coase conjecture (put forward by Coase (1972)). The effect
of retractable offers was studied in Muthoo (1994). The repeated-offers
model with correlated values (but without retractable offers) has been stud-
ied in Evans (1989) and Vincent (1989). Not surprisingly (given Lemma
9.2), they show that (under some conditions) the (generically) unique PBE
of the model is not ex-post efficient. In particular, the Coase conjecture is
not satisfied by this equilibrium. The application studied in Section 9.7 is a
special case of the model in Wang (1998).
I have not discussed the kinds of positive questions addressed in Section
9.6 in the context of models in which a player with private information can
make offers. The reasons for this omission are two-fold: (i) such models
are plagued by a great multiplicity of perfect Bayesian equilibria, since in
such models an informed player has the opportunity to signal her private
information when making a price offer, and (ii) the analysis of the PBE
in such a model is quite complex. It is well known that signalling models
— such as Spence's classic job market model (cf. Spence (1974)) — are
characterized by a great multiplicity of perfect Bayesian equilibria. The root
cause of such multiplicity of equilibria are the 'zero-probability' events that
294 Asymmetric Information

pervade signalling models. Following such an event, Bayes' rule cannot be

used to update a player's beliefs. As such the player has to form new beliefs,
which are called conjectures. Rubinstein (1985b) contains a discussion and
analysis of several types of conjectures. For an overview of the literature
on bargaining models in which a player with private information can make
offers, see Fudenberg and Tirole (1991, Section 10.4).
10 Repeated Bargaining Situations

10.1 Introduction

In this chapter I study situations in which two players have the opportunity
to be involved in a sequence of (possibly different and/or interdependent)
bargaining situations. Such a situation will be called a 'repeated' bargaining
situation (RBS). Examples of repeated bargaining situations abound. For
instance: (i) in any marriage the wife and the husband are in a RBS, and (ii)
in most bilateral monopoly markets the seller and the buyer are in a RBS. To
illustrate the potential for interdependence amongst the bargaining situa-
tions in a RBS, consider a bilateral monopoly market for some input in which
the seller and the buyer have the opportunity to be involved in a sequence of
'one-shot' transactions, and suppose that the buyer's reservation value for
the input depends on the level of her capital stock. Since the price at which
trade occurs determines the buyer's profit — which, in turn, determines her
investment in capital stock — it follows that the buyer's reservation value
in any one transaction is determined by the outcomes of past transactions.
Hence, any pair of one-shot transactions are interdependent.
In Sections 10.2-10.4 I study repeated bargaining models, in which the
outcomes of any pair of bargaining situations are negotiated separately, and
moreover, the outcome of each bargaining situation is negotiated when (and
if) it materializes. Although in the formal structure of these repeated bar-
296 Repeated Bargaining Situations

gaining models there is no reference to contracts, it is possible to provide

the following contractual interpretation: the repeated bargaining models
embody the notion that the RBS is governed by a sequence of short-term
(or, limited-term) contracts.
Section 10.2 studies a model of a simple RBS in which two players se-
quentially bargain over the partition of an infinite number of cakes. Subject
to two important qualifications, the repeated bargaining model studied here
is an infinite repetition of Rubinstein's bargaining game (that is studied
in Section 3.2). The reasons for studying this simple model are two-fold:
(i) to lay down the basic structure of the models of relatively more com-
plex repeated bargaining situations that are studied in later sections, and
(ii) to focus attention on the role of the players' discount rates in repeated
bargaining situations.
It will be shown that the impact of the players' discount rates on the
outcome of the RBS depends on the frequency of the bargaining situations.
For example, if this frequency is large, then it is possible that a player's
share of each and every cake is increasing in her discount rate. In contrast,
the opposite is the case in Rubinstein's bargaining model — when the play-
ers bargain over the partition of the single available cake.1 Indeed, this
result (and the other results derived in Section 10.2) imply that a RBS is
fundamentally different from a single ('one-shot') bargaining situation.
Section 10.3 considers a RBS in which two players bargain in each pe-
riod over the amount of the current output to save and invest in the creation
of new capital, and over the partition of the remaining output for current
consumption. The sequence of bargaining situations in this RBS are inter-
dependent.
A key assumption of the repeated bargaining situations studied in Sec-
tions 10.2 and 10.3 is that the players do not have any outside options. This
means, in particular, that they are committed to the RBS, in the sense that
they will attempt to reach agreement in each and every bargaining situation.
As such the players are in a long-term relationship. Section 10.4 explores
the role of outside options in a simple RBS. An application to firm-provided
general training is contained in Section 10.4.1.
x
It will be shown that the repeated play of the unique subgame perfect equilibrium
(SPE) of Rubinstein's model (cf. Theorem 3.1) is not an SPE of the repeated bargaining
model studied in Section 10.2.
10.2 A Basic Repeated Bargaining Model 297

Section 10.5 studies the role of long-term contracts in repeated bargain-

ing situations. Such a contract specifies the outcome of each and every bar-
gaining situation to be encountered. A main message obtained is that long-
term contracts can have a beneficial role in those repeated bargaining situ-
ations in which the parties have the opportunity to engage in relationship-
specific investments that enhance the value of the RBS — notwithstanding
the possibility that such contracts may be highly incomplete.
Section 10.6 explores the notion that in a RBS a player has an incentive
to acquire a reputation for being a particular type of bargainer.

10.2 A Basic Repeated Bargaining Model

Consider a RBS in which two players have the opportunity to sequentially
produce an infinite number of 'cakes'. Subject to two important qualifi-
cations, the model of this RBS described below is an infinite repetition of
Rubinstein's bargaining game studied in Section 3.2. The first qualification
is that the players start bargaining over the partition of the (n + l)th cake
(where n = 1, 2,...) if and only if they reach agreement on the partition of
the nth cake. Thus, if the players perpetually disagree over the partition of
the nth cake, then their relationship is terminated. The second qualification
is that the time at which the players start bargaining over the partition of
the (n + l)th cake is determined by the time at which agreement is struck
over the partition of the nth cake. A description of the model now follows.
Two players, A and B, bargain over the partition of a cake of size vr (TT >
0) according to the alternating-offers procedure (as described in Section 3.2).
If agreement is reached at time ti, where t\ — 0, A, 2A,..., and A (A > 0)
is the time interval between two consecutive offers, then immediately the
players consume their respective (agreed) shares. Then r (r > 0) time units
later, at time t\ + r, the players bargain over the partition of a second cake
of size 7T according to the alternating-offers procedure. Agreement at time
£2, where £2 = t i + r , t i + r + A,ti + r + 2A,..., is followed immediately with
the players consuming their respective (agreed) shares. Then r time units
later, at time £2 + T, the players bargain over the partition of a third cake of
size 7T according to the alternating-offers procedure. This process continues
indefinitely, provided that the players always reach agreement. However, if
the players perpetually disagree over the partition of some cake, then there
298 Repeated Bargaining Situations

is no further bargaining over new cakes; the players have terminated their
relationship. 2 Without loss of generality, I assume that player i makes the
first offer when bargaining begins over the partition of the (n + l)th cake
(n = 1,2,...) if it was player j (j ^ i) whose offer over the partition of the
nth cake was accepted by player i. Furthermore, player A makes the offer
at time 0.
The payoffs to the players depend on the number N (where N = 0,1, 2 , . . . )
of cakes that they partition. If N = 0 — that is, they perpetually disagree
over the partition of the first cake — then each player's payoff is zero. If
1 < N < oc — that is, they partition N cakes and perpetually disagree over
the partition of the (N + l)th cake — then player z's (i = A, B) payoff is

n=l

where x™ (0 < xf < TT) is player i's share of the nth cake, tn is the time at
which agreement over the partition of the nth cake is struck, and v\ {ri > 0)
is player i's discount rate. Finally, if N = oc — that is, they partition all
the cakes — then player z's payoff is

n=l

Define for each i = A,B, Si = exp(—r^A) and ai = exp(—^r). The parame-

ters 8A and 8B capture the bargaining frictions: they respectively represent
the costs to players A and B of haggling over the partition of a cake. In
contrast, the parameters a A and as respectively represent the values to
players A and B of future bargaining situations.
One main result obtained below (in Section 10.2.1) is that if r > A,
then in the unique stationary subgame perfect equilibrium player i's share
of each and every cake is strictly increasing in <$$, but strictly decreasing
in ai. This result implies that a decrease in r^ has two opposite effects on
player i's equilibrium share of each and every cake — because a decrease
in T{ increases both 8{ and c^. It will be shown that under some plausible
2
It may be noted that, therefore, this dynamic game is not a standard infinitely re-
peated game. For an analysis of such games, see any game theory text (such as Fudenberg
and Tirole (1991), Gibbons (1992) and Osborne and Rubinstein (1994)).
10.2 A Basic Repeated Bargaining Model 299

conditions the effect through ai dominates that through <5^, thus implying
that as player i becomes more patient her equilibrium share of each and
every cake decreases. Before I turn to a derivation of these (and other)
results, the following remarks are in order.

Remark 10.1 (On the notion of long-term relationships). Loosely

stated, the players in a RBS are in a long-term relationship if they are some-
what committed to the RBS — that is, they expect to exploit most of the
bargaining situations in the RBS, since the cost to each player of refusing
to do so is sufficiently high. For example: (i) if the cost of divorce to ei-
ther partner is sufficiently high, then the married couple are in a long-term
relationship, and (ii) if the players have no outside options (such as in the
RBS modelled above), then the players are in a long-term relationship. On
the other hand, for example, if at least one player's outside option is suffi-
ciently attractive, then the players need not be in a long-term relationship
— since the player concerned could choose (with negligible cost) to exercise
her attractive outside option at any point during the RBS.

Remark 10.2 (Finite versus infinite number of bargaining situations).

In the repeated bargaining model described above it is assumed that the
players have the opportunity to sequentially produce an infinite number of
cakes — that is, there exists an infinite sequence of bargaining situations.
This is not an unreasonable assumption. Many repeated bargaining situa-
tions — such as those in a marriage (which can potentially last forever) —
are literally consistent with this assumption. Furthermore, an alternative
model in which it is assumed that there is a finite number of bargaining
situations is not particularly plausible, since such a model (implicitly) as-
sumes that the players know (from the start of their relationship) the exact
number of cakes that they can partition. In most repeated bargaining situ-
ations in which the players know that they will encounter a finite number of
bargaining situations, it is typically the case that the players are uncertain
about the exact number of bargaining situations that they will encounter. A
simple extension and/or re-interpretation of the repeated bargaining model
described above is the appropriate model of such a RBS.3
3
For example, oti may be interpreted as player i's probabilistic belief that they can
partition another cake after consuming the current cake.
300 Repeated Bargaining Situations

Remark 10.3 (Simplifying assumptions). The RBS modelled above is

very simple. For example, it is assumed that the bargaining situations in
the RBS are identical. This assumption has been deliberately chosen so as
to develop some of the fundamental insights (especially about the role of
the players' discount rates) in a simple manner. It is, however, conceptually
straightforward to extend the model to more complex (and more realistic)
repeated bargaining situations. In Section 10.3, for example, I consider a
RBS in which the size of the (n + l)th cake is determined by the outcome
of the nth cake.

Remark 10.4 (Two interpretations). One interpretation of the repeated

bargaining model described above is as follows. Two players have the oppor-
tunity to engage in an infinite sequence of 'one-shot' transactions, where TT
denotes the size of the surplus generated from each one-shot transaction and
r the frequency of such one-shot transactions. The outcomes of any pair of
one-shot transactions are negotiated separately, and, moreover, the outcome
of each one-shot transaction is negotiated when (and if) it materializes. Let-
ting a one-shot contract denote a contract that specifies the outcome of a sin-
gle one-shot transaction, the model embodies the notion that the long-term
relationship is governed by a sequence of one-shot contracts. An alternative
interpretation of the model is as follows. Two players have the opportunity
to generate (through some form of co-operation) a flow of money at rate
7?. They bargain over a contract that specifies the partition of the money
over the duration of the contract, where r is the duration of the contract.
Thus, the present discounted value of the total amount of money generated
over the duration of any single contract is TT = TT[1 — exp(—r mr)]/rm, where
?~m (^m > 0) is the market interest rate. The model embodies the notion
that the long-term relationship is governed by a sequence of limited-term
contracts (of duration r).

10.2.1 The Unique Stationary Subgame Perfect Equilibrium

Fix an arbitrary subgame perfect equilibrium (SPE) that satisfies Properties
3.1 and 3.2 (which are stated in Section 3.2.1). Property 3.1 states that
whenever a player has to make an offer, her equilibrium offer is accepted
by the other player, while Property 3.2 states that in equilibrium a player
makes the same offer whenever she has to make an offer.
10.2 A Basic Repeated Bargaining Model 301

Given Property 3.2, let x* (i = A, B) denote the equilibrium offer that

player i makes whenever she has to make an offer. I adopt the convention
that an offer is the share to the proposer. Furthermore, letting V* denote
player z's equilibrium payoff in any subgame beginning with her offer, it
follows from Properties 3.1 and 3.2 that V* = x\ + ^(TT - x*) + afV*
(j ^ i). Hence, it follows that

X n ( )
1 — aA 1 - aB
Consider an arbitrary point in time at which player i has to make an offer
to player j . By definition, player j ' s equilibrium payoff from rejecting any
offer is 6jV*. Therefore, perfection requires that player j accept any offer X{
(where 0 < X{ < TT) such that TT — xi + OLJVJ > SjV*, and reject any offer xi
such that 7T — xi + otjVj < SjVf. Hence, if A > r, then optimality implies
that x\ = x*B = 7T.
On the other hand, if A < r then player i is indifferent between accepting
and rejecting player j ' s equilibrium offer.4 That is

n - x*B + aAVX = SAVX and ir - x*A + aBV£ = 6BV£. (10.2)

After substituting for VA and VB in (10.2) using (10.1), and then solving for
x\ and x*B) it follows that

r* _ {l-6AaA)(l-6B)(l + aB)7r
A
~ (1 - SAaA)(l - 6BaB) - (SA - aA)(6B - aB)
(1 - 8BaB)(l - 6A)(1
* =
B
(1 - 8AaA)(l - 6BaB) - (6A - aA)(6B - aB)'
Hence, I have characterized the unique SPE that satisfies Properties 3.1 and
3.2, which is described in the following proposition.

Proposition 10.1. The unique subgame perfect equilibrium that satisfies

Properties 3.1 and 3.2 is as follows:
• player A always offers x*A and always accepts an offer xB if and only if
xB < x*B,
• player B always offers x*B and always accepts an offer xA if and only if
xA < x*A,
4
It should be noted that A < r implies that TT + otjV* > 6jV3*, because by definition
302 Repeated Bargaining Situations

where if A > r then x*A = x*B = TT, and if A < r then x*A and x*B are respec-
tively stated in equations 10.3 and 10.4- In equilibrium agreement is reached
immediately over the partition of each and every cake. The equilibrium par-
tition of the nth cake is (x*A, TT — x*A) if n is odd (i.e., n — 1, 3, 5 , . . . ) , and
(TT — x*B,x*B) if n is even (i.e., n = 2, 4, 6 , . . . ) .

Remark 10.5 (Stationary SPE). A stationary SPE is a subgame per-

fect equilibrium that satisfies Property 3.2 and the following property: in
equilibrium, for any offer xi G [0, TT] made by player i, player j always either
accepts it or rejects it. 5 Proposition 10.1 characterizes the unique stationary
SPE, because — as I now show — any stationary SPE must satisfy Prop-
erty 3.1. Suppose, to the contrary, that there exists a stationary SPE that
does not satisfy Property 3.1. This means that there exists an i {i = A or
i — B) such that player i's equilibrium offer x\ is always rejected by player
j . Let Vj and Wi respectively denote player j ' s and player i's equilibrium
payoffs in any subgame beginning with player j ' s offer. It must be the case
that TT — x\ + OLjVj < SjVj. If A > r, then a contradiction is obtained.
Hence, in what follows, assume that A < r. In this stationary SPE player
j accepts any offer xi such that TT — Xi + otjVj > SjVj. Fix an arbitrary
subgame beginning with player i's offer, and suppose player i considers the
following one-shot deviation: she offers x[ such that x[ < TT + (<x/ — Sj)Vj.
Any such offer is accepted by player j . Hence, if there exists such an x[ that
makes the one-shot deviation profitable for player i, then a contradiction is
obtained. This will be the case if there exists an x\ < TT + (aj — Sj)Vj such
that x\ + aiWi > 6iWi, which is implied if the following inequality holds:
TT+(aj-6j)Vj > (Si-aj)Wi — that is, TT > {Sj—aj)Vj + (Si-aj)Wi. Ifplayer
j ' s equilibrium offer is also always rejected by player i, then Vj = Wi = 0.
On the other hand, if player j ' s equilibrium offer x* is always accepted by
player z, then Vj = x*/(l - a?) and Wi = (1 - £*)/(l - a?). In either case,
the desired inequality holds.

Hence, if A > r, then in the unique stationary subgame perfect equilib-

rium (SSPE) player A obtains the whole of the nth cake when n is odd and
player B obtains the whole of the nth cake when n is even. The intuition
behind this result is that if A > r, then (in a SSPE) the proposer effectively
5
This property and Property 3.2 imply that each player's strategy is history-
independent.
10.2 A Basic Repeated Bargaining Model 303

makes a ctake-it-or-leave-it-offer\ Although this result is rather provocative,

it is not plausible that A > T. It is more likely that r > A — the time
interval between two consecutive offers during bargaining over the partition
of any cake is smaller than the time taken for the 'arrival' of a new cake.
Indeed, in terms of either interpretation described above in Remark 10.4, it
is implausible that A > r. Hence, from now on I assume that r > A.
Before proceeding further, however, I note that (in general) the unique
SSPE partition of each and every cake is different from the unique SPE
partition of the single available cake in Rubinstein's bargaining model (cf.
Theorem 3.1). The intuition for this difference — which is further developed
below — is based on the following observation. In Rubinstein's model the
cost to player i of rejecting an offer is captured by <5^; rejecting an offer
shrinks, from player i's perspective, the single available cake by a factor
of 6{. In contrast, in the model studied here the rejection of an offer not
only shrinks the current cake, but it also shrinks all the future cakes, thus
inducing a relatively higher cost of rejecting an offer.
As is evident from the expressions for x\ and x*B in (10.3) and (10.4),
when r > A the equilibrium partition of each and every cake depends on
the parameters TA, TB, A and r in a rather complex manner. However, one
of the main insights of the model can be obtained in a simple manner by
examining the impact of the derived parameters OLA, a # , 6 A and 6B on the
unique SSPE partitions. Notice that if r > A, then 6i > OL{ (i = A, B).

Corollary 10.1. For each i = A,B, x\ is (i) strictly increasing in 6i on

the set Z', (ii) strictly decreasing in ai on the set Z, (Hi) strictly decreasing
in 6j (j 7^ i) on the set Z and (iv) strictly increasing in aj on the set Z,
where

Z = {{aA) aB, 6A, 6B) : 6A> aA and 6B > aB}.

Proof. The corollary follows in a straightforward manner from the four

derivatives of x* (which is stated in (10.3)-(10.4)) — with respect to <5i,
a^, 6j and ay. •

It follows immediately from Corollary 10.1 and Proposition 10.1 that if

r > A, then player z's share of each and every cake in the unique SSPE is
strictly increasing in 6{, but strictly decreasing in c^. The former effect is
consistent with the insight obtained in the context of Rubinstein's bargaining
304 Repeated Bargaining Situations

model when the players bargain over the partition of the single available cake
(cf. Theorem 3.1; for intuition, see Section 3.2.3). The latter effect, however,
is novel, but the intuition behind it is straightforward. As ai increases,
the value to player i of future bargaining situations increases. Thus, when
bargaining over the partition of any cake, her desire to proceed to bargain
over the partition of the next cake has increased, which works to player j ' s
advantage.
Notice that Corollary 10.1 implies that a decrease in Ti has two opposite
effects on player i's equilibrium share of each and every cake — because a
decrease in ri increases both Si and c^. It will be shown below that if A
is arbitrarily small and (for each i = A,B) TIT > 0 but small, then the
effect through c^ dominates that through Si, thus implying that as player i
becomes more patient her equilibrium share of each and every cake decreases.

10.2.2 Small Time Intervals Between Consecutive Offers

In Corollary 10.2 below I characterize the unique SSPE in the limit, as

A —» 0. I focus attention on this limit because it is the most persuasive case
(cf. the discussion in Section 3.2.4). Besides, the expressions for x\ and x*B
in this limit are relatively more transparent compared to those in (10.3) and
(10.4).

Corollary 10.2. Fix any rA > 0, r # > 0 and r > 0. In the limit, as
A —> 0, x\ and x*B (as defined in Proposition 10.1) respectively converge to
* rB7T + rAir
zA = and zB — , where
TB + <t)ATA TA + (\>BTB

aA)(l-aB)

Furthermore, the payoffs to players A and B in the limiting (as A —> 0)

unique SSPE are respectively

?L and n
" l-aA l-aB
Proof. Fix r > A. When A > 0 but small, Si = 1 - r^A. Using this to
substitute for SA and SB in (10.3) and (10.4), it follows (after simplifying)
10.2 A Basic Repeated Bargaining Model 305

that when A > 0 but small

^ (laA + rAaAA)(l + aB)rBA7v
A
(1 aAa)(rA + ^ r ^ r A ^ + (a aA){r ~ rA)A
* (l-aB + rBaBA)(l
B
(1 rB - rArBA)A + (aA - aB)(rA - rB)A'

After dividing the numerator and the denominator of each of these expres-
sions by A, and then letting A tend to zero, it follows that x\ —> z*. Since
Z
A + ZB = ni ^ follows from Proposition 10.1 that the equilibrium payoffs
in this limit are as stated in the corollary. •

It follows from Corollaries 10.2 and 3.1 that unless (f)A = (\)B = 1, the
limiting (as A —> 0) unique SSPE partition of each and every cake in the
repeated bargaining model is different from the limiting (as A —» 0) unique
SPE partition of the single available cake in Rubinstein's bargaining model.
Notice that for any rA > 0, rB > 0 and r > 0, fa = 1 if and only if rA — rB.
Furthermore, since fa —^ 1 as r^r —» oc, it follows that when rA ^ rBi the
limiting (as A —> 0) unique SSPE partition of each and every cake will have
similar properties to the limiting (as A —•» 0) unique SPE partition of the
single available cake in Rubinstein's model if and only if the value to each
player of future bargaining situations is arbitrarily small.
The following corollary characterizes the properties of the limiting (as
A —> 0) unique SSPE when rA ^ rB and (for each i = A,B) TiT is small
(which means that the value to player i of future bargaining situations is
large).

Corollary 10.3. Assume that rA ^ rB, rAr > 0 but small and rBr > 0 but
small.
(i) z* (i — A,B) is strictly increasing in ri, and strictly decreasing in rj

(ii) Ifn > rj (i ^ j), then z* > z*.

(iii) Ifn > r3 (i + j), then V** < V**.
(iv) V** (i = A, B) is strictly decreasing in r^; and strictly decreasing in rj

(v) If ri > rj (i ^ j) then z* is strictly increasing in r, and if ri < rj then

z* is strictly decreasing in r.
(vi) In the limit as rAr —> 0 and rBT —> 07 z\ —> TT/2 and z^ —> TT/2.
306 Repeated Bargaining Situations

Proof. When TiT > 0 but small, oti = 1 — rir. Using this to substitute for
a A and aB in the expressions for (j)A and (j)B (stated in Corollary 10.2), it
follows that when (for each i = A,B) riT > 0 but small

( ) (2 - rAr)rB , , (2 - rBr)rA
t A = ^-^ —\ and cj)B = ^—- '-—.
rA(2-rBr) rB(2 - rAr)
This implies that when (for each i = A,B) V{T > 0 but small

„ (2 - rBr)7T „ (2 - rAr)n:
zA = -±-
4 - (rA +- ^TB)T
— and z B^ = 4 - (rA + r B ) r '

The results in the corollary are now straightforward to derive, given these
simple expressions for z*A and z*B. •

Corollary 10.3(i) states that when rA ^ rB and the value to each player
of future bargaining situations is large, player i's (i = A, B) share of each and
every cake in the limiting (as A —> 0) unique SSPE decreases as she becomes
more patient and/or her opponent becomes less patient. In contrast, in
Rubinstein's bargaining model a player's share of the single available cake
in the limiting (as A —> 0) unique SPE increases as she becomes more patient
and/or her opponent becomes less patient (cf. Corollary 3.1). Thus, this
fundamental insight of Rubinstein's model does not carry over to long-term
relationships when the players have the opportunity to bargain (sequentially)
over the partition of an infinite number of cakes and the value to each player
of future bargaining situations is large. The intuition behind this conclusion
is as follows.
In the repeated bargaining model studied here a player's discount rate
determines not only her cost of rejecting an offer, but also her value of fu-
ture bargaining situations. Suppose that one of the players — say, player
i — becomes more patient. This means that her cost of rejecting an offer
decreases. However, it also means that her value of future bargaining situ-
ations increases. When bargaining over the partition of a cake, the former
effect increases her bargaining power (as she is more willing to reject offers),
but the latter effect decreases her bargaining power — because she is more
willing to accept offers so that the players can proceed to bargain over the
partition of the next cake (cf. Corollary 10.1). When A is arbitrarily small
the former effect is negligible, and when TIT > 0 (but small) the latter effect
is non-negligible. Thus, the latter effect dominates the former effect.
10.2 A Basic Repeated Bargaining Model 307

The above argument also provides intuition for the result stated in Corol-
lary 10.3(ii) that the less patient of the two players receives a greater share
of each and every cake. Not surprisingly, however, the limiting (as A —> 0)
unique SSPE payoff of the less patient of the two players is smaller than her
opponent's limiting unique SSPE payoff (Corollary 10.3(iii)). This means
that the more patient player's overall bargaining power in the long-term
relationship is relatively higher. And Corollary 10.3(iv) shows that overall
bargaining power increases as she becomes even more patient, but decreases
as her opponent becomes less patient.
Corollary 10.3(v) implies that the more patient player's limiting (as A —>
0) SSPE share of each and every cake decreases as r increases. The intuition
behind this conclusion runs as follows. As r increases, the decrease in the
more patient player's value of future bargaining situations is smaller than
the decrease in her opponent's value of future bargaining situations, which
works to her opponent's advantage.
Notice that if rAT > 0 and rsT > 0, then the players' shares of each and
every cake (and payoffs) in the limiting (as A —> 0) unique SSPE depend
on the relative magnitude of the players' discount rates. However, Corol-
lary 10.3(vi) shows that in the limit as rAT —> 0 and rsT —> 0, this is no
longer true. This implies that (under these limiting conditions) the players'
bargaining powers are identical no matter how patient or impatient player
A is relative to player B. Hence, the insight of Rubinstein's bargaining
model — that the relative magnitude of the players' discount rates critically
determine the players' bargaining powers even when the time interval be-
tween two consecutive offers is arbitrarily small — does not carry over to
long-term relationships when the players have the opportunity to bargain
(sequentially) over an infinite number of cakes and the value to each player
of future bargaining situations is arbitrarily large.

10.2.3 Comparison with a Long-Term Contract

For any r^ > 0, rs > 0 and r > 0, the payoffs to players A and B in
the limiting (as A —> 0) unique SSPE are respectively F | * and Vg* as
stated in Corollary 10.2. With reference to the interpretations described
above in Remark 10.4, these are the players' (equilibrium) payoffs from their
long-term relationship when that relationship is governed by a sequence of
308 Repeated Bargaining Situations

short-term contracts. It is instructive to compare these payoffs with those

the players obtain if, instead, their relationship is governed by a long-term
contract, which is the unique SPE outcome of the following non-repeated
bargaining model.
The players bargain (according to the alternating-offers procedure) over
long-term contracts, where such a contract specifies the same partition of
each and every cake. Thus, a long-term contract is characterized by a single
number x G [O,TT], which denotes player A's share of each and every cake.
If agreement is reached at time t = 0, A, 2 A , . . . on a long-term contract
x G [O,TT], then bargaining ends and the payoffs to players A and B are
respectively

x exp(—VAt) . (TT — x) exp(—

±-aA and
1
If the players perpetually disagree, then each player's payoff is zero. This
model is very similar to Rubinstein's bargaining model studied in Section
3.2. Indeed, the unique SPE of this model is similar to the unique SPE of
Rubinstein's bargaining model. In particular, in the limiting (as A —» 0)
unique SPE player i (i = A,B) receives a share rjn/(rA + rB) (j ^ i) of
each and every cake. Hence, it follows that for any TA > 0, TB > 0 and
r > 0, the payoffs to players A and B in the limiting (as A —* 0) unique
SPE of this alternative model are respectively

ViA = 1 ^ . and V£ = 7 ^ r .
(l-aA)(rA + rB) (1 - aB)(rA + rB)
It is straightforward to establish the following corollary, which provides the
desired comparison.

Corollary 10.4. For any TA > 0, TB > 0 and r > 0

V** = V^T if and only if r\ — rj ( i , j — A , B and j ^ i ) .

Thus, if the players are equally patient/impatient (rA = rB), then each
player is indifferent between the equilibrium sequence of short-term con-
tracts and the equilibrium long-term contract. If, on the other hand, player
i is more impatient than player j {ri > rj), then player i prefers the equilib-
rium sequence of short-term contracts over the equilibrium long-term con-
tract, while player j has the opposite preference. The intuition for this
10.2 A Basic Repeated Bargaining Model 309

conclusion is based on the intuition provided for Corollary 10.3(i). In the

repeated bargaining model the more patient player is more keen to secure
agreement over a short-term contract — so as to consummate it, and then
proceed to the next set of negotiations over another short-term contract.
This works to the advantage of the less patient player. However, when bar-
gaining over long-term contracts, this advantage (to the less patient player)
is not present. That is why the more patient player prefers the equilibrium
long-term contract, while the less patient player prefers the equilibrium se-
quence of short-term contracts.

10.2.4 Non-Stationary Subgame Perfect Equilibria

In this section I explore the potential existence of non-stationary SPE.6 The
following lemma establishes the existence (for some values of the parameters)
of two non-stationary SPE, which are described in Table 10.1. In one SPE
play begins in state SA, while in the other SPE play begins in state SB- In
both of these equilibria agreement is reached immediately on the partition of
each and every cake. In the SPE in which play begins in state Si (i — A, I?),
player i obtains, for all n (n = 1, 2, 3,...), the whole of the nth cake. Thus,
in this SPE players i and j (j ^ i) respectively obtain their best and worst
possible SPE payoffs.

Lemma 10.1. The pair of strategies described in Table 10.1 is a SPE if and
only if a A + 6A > 1 and OLB + <$B > 1- If play begins in state S{ (i = A, B),
then in equilibrium for all n (n = 1, 2, 3,...) agreement over the partition
of the nth cake is reached immdiately with player i obtaining the whole of
the nth cake.

Proof. Using the One-Shot Deviation property it is straightforward to verify

that the pair of strategies described in Table 10.1 is a SPE if and only if
OLA+^A > 1 and as+Ss > 1. Note that if and only if OLI + SI > 1 is it optimal
for player i in state Si to reject any offer Xj such that 0 < Xj < TT. •

Given the existence of these two ('extremal') SPE, it is straightforward

to show that if OLA+^A > 1 and (XB+8B > 1, then there exists a large number
(indeed, a continuum) of other non-stationary SPE. Indeed, any sequence of
In such a SPE a player's strategy is history-dependent.
310 Repeated Bargaining Situations

state SA state sB
offer x*A = n x*A=0
Player A accept XB=0 0 < xB < n
rejects 0 < xB < 7T

offer xB=0 X D == 7T

Player B accept 0<XA<7T

rejects 0 < XA < 7T

transitions switch to state SB switch to state SA

if an offer XB such if an offer XA such
that 0 < XB < 7T is that 0 < XA < 7T is
accepted by player accepted by player
A

Table 10.1: Two non-stationary SPE in the repeated bargaining model.

bargaining outcomes can be supported by a non-stationary SPE. 7 The basic

idea is as follows. Fix an arbitrary sequence of bargaining outcomes. Con-
sider the path of play that generates this sequence, and in which each player
otherwise always demands the whole cake. This path of play is supported
as a SPE path by the threat of reverting to an appropriate extremal SPE: if
player i deviates from the path of play, then the appropriate extremal SPE
is the one which gives player i her worst possible SPE payoff (which is a
payoff of zero). I state this result in the following proposition.

Proposition 10.2. IfaA+^A > 1 andas+^B > 1? then there exists a large
number (continuum) of non-stationary SPE. In particular, any sequence of
bargaining outcomes can be supported by a non-stationary SPE.
In the limit, as A —• 0, the hypothesis (namely, the two inequalities) of
Proposition 10.2 are valid, and thus, in this limit there exists a large number
of (non-stationary) SPE. It should be noted that this result is valid for any
7
In an arbitrary sequence of bargaining outcomes agreement is reached over the par-
tition of N cakes (where 0 < N < oo), and the players perpetually disagree over the
partition of the (JV + l)th cake. The sequence also specifies the times at which agreements
are struck, and the agreed partitions.
10.2 A Basic Repeated Bargaining Model 311

rA > 0, rB > 0 and r > 0 (i.e., for any OLA G (0,1) and aB G (0,1)). 8
The case of r —> oo, in particular, has the following striking interpretation.
The uniqueness property of Rubinstein's bargaining model (cf. Theorem
3.1) is not robust to small 'external effects', in the sense that the basic
indeterminacy of a bargaining situation is re-obtained if the players expect
to bargain, with an arbitrarily small probability, over the partition of another
cake each time they reach agreement over the partition of an existing cake. 9

Remark 10.6 (Static versus dynamic efficiency). The unique station-

ary SPE is efficient in the following 'static' sense: agreement over the par-
tition of each and every cake is reached without any costly delay. However,
as I now show, when TA ^ rB it is not efficient in the following 'dynamic'
sense: there exists an alternative sequence of partitions of the cakes that
makes both players strictly better-off (in terms of their payoffs). Suppose
that rB > TA, and consider the following alternative sequence of partitions
of the cakes: player B receives the whole of the first N cakes, and player A
the whole of the remaining (infinite number of) cakes. The payoffs to players
A and B respectively from this alternative sequence are a^7r/(l — a A) and
(1 — a^)7r/(l — QLB), where af = exp(—r^rN) (i = A, B). It is straightfor-
ward to verify that since rB > rA, there exists an TV such that each player's
payoff from this alternative sequence is strictly greater than her payoff in
the unique SSPE (which is stated in Corollary 10.2). In contrast, there exist
non-stationary SPE which satisfy neither notions of efficiency, and there also
exist non-stationary SPE that do satisfy both notions of efficiency.10
8
In contrast, in a standard infinitely repeated game it is typically the case that a
multiplicity of SPE is obtainable if and only if the discount factors (aA and OLB) are
sufficiently large — see any game theory text (such as Fudenberg and Tirole (1991),
Gibbons (1992) and Osborne and Rubinstein (1994)).
9
This interpretation makes sense since a, where a = a A = &B, may be interpreted as
the probability that the players (expect to) bargain over the partition of another cake after
reaching agreement over the partition of the currently available cake.
10
A sequence of bargaining outcomes is dynamically efficient when TB > TA — that is,
maximizes one player's payoff subject to her opponent receiving some fixed payoff level u
— only if there exists an N > 1 (which depends on u) such that the more patient player
A receives the whole of the nth cake for all n > iV, and if N > 1 then the less patient
player receives the whole of the nth cake for all n < N. It should be noted that a sequence
of bargaining outcomes satisfies dynamic efficiency only if it satisfies static efficiency, but
not vice-versa.
312 Repeated Bargaining Situations

10.3 An Application to Dynamic Capital Invest-

ment
Two players, A and JB, have access to a technology of production which
allows them to generate output by using capital as input. The output per
period is F(K) if the capital stock is K, where the production function F
is continuously differentiate, strictly increasing and strictly concave. Fur-
thermore

F(0) = 0, lim F'(K) = 0 and lim F'(K) = oc.

The initial capital stock is K\ > 0. Capital depreciates during the produc-
tion process at a constant rate 7 per period, where 0 < 7 < 1. It does not,
however, depreciate when it is not in use (such as during the bargaining
process).
At the beginning of each period the players bargain over the amount of
output to be saved and invested in the creation of new capital, and over the
partition of the remaining output for their current consumption. Letting
In denote the quantity of output saved and invested in period n (where
n = 1,2, 3,...), it follows that the capital stock in period n + 1 is

Kn+1 = (l->y)Kn + In.

The model of the repeated bargaining situation faced by the two players is
an extension of the repeated bargaining model studied in Section 10.2. In
particular, at time 0 the players bargain (according to the alternating-offers
procedure) over how much of the output F(Ki) to save and invest, and how
to partition the remaining output for their current consumption. Thus, an
offer may be described by a pair of numbers (x, /) where 0 < x, / < F(K\)
and x+I < F(K\), with the following interpretation: / denotes the quantity
of output saved for investment, while x and F(K\) — I — x respectively
denote the shares to the proposer and the responder of the remaining output
for their current consumption. If agreement is reached at time £1, then
immediately they produce the output, save and invest the agreed quantity
/1, and consume their respective (agreed) shares of the remaining output. 11
11
If the players perpetually disagree, then no output is produced — and furthermore,
the players' relationship is terminated and each player's payoff is zero.
10.3 An Application to Dynamic Capital Investment 313

Then, r time units later — that is, next period — the process starts again
with capital stock K2 = (1 — 7)^1 + Ii-> and so on.
Notice that the output level in period n + 1 is F(Kn+i). Hence, since
(for any n > 1) Kn+i depends on the investment level In agreed to in the
previous period, the output level in period n + 1 depends on the outcome
of bargaining in the previous period. As such the sequence of bargaining
situations in this RBS are interdependent.
One could now proceed to study the set of Markov subgame perfect equi-
libria of this dynamic model, in which each player adopts a Markov strategy.
In such a strategy a player's offer is conditioned only upon the current cap-
ital stock, and her response to any offer is conditioned only upon that offer
and the current capital stock.12 Thus, a Markov strategy for player i can be
defined by three functions x*, /* and i?*, with the following interpretation.
Whenever player i has to make an offer, she offers (x*(K),I*(K)) if the
current capital stock is K, and whenever she has to respond to any offer
(xj,Ij) made by player j (j 7^ i), her response is R*(XJ,IJ,K) G {Accept,
Reject}.
Although the analysis of the set of Markov SPE of this dynamic model
is conceptually straightforward, it is technically demanding, and hence I
leave it for future research. Instead, in the following section I characterize
the unique steady state SSPE of the model under the assumption that the
players do not bargain in any period over the investment level, which allows
for a straightforward application of Proposition 10.1.

Bargaining Equilibrium with an Exogenous Savings Rate

Assume that the players do not bargain (in any period) over the invest-
ment level. The amount of output that will be invested in period n (n =
1, 2, 3,...) is sF(Kn), where s (0 < s < 1) is an exogenously given savings
rate. This assumption implies that the sequence of bargaining situations are
not interdependent. Furthermore, the output level F(Kn+i) in period n + 1
(for any n > 1) is exogenously given, since the capital stock Kn+i in period
n + 1 is uniquely determined by the following recursive equation

12
History matters only to the extent that it determines the current capital stock and
the current offer.
314 Repeated Bargaining Situations

where K\ > 0 is exogenously given.

Given our assumptions on F, the equation sF(K) = ^K has a unique
(strictly positive) solution, which is denoted by K* — it is the unique
(strictly positive) steady state capital stock. If, for some N (N > 1),
KN = K*, then, for all n > N, Kn = K*. The intuition behind this
observation is straightforward. Suppose that in period TV the capital stock
is K*. By the definition of if*, it follows that in period N the amount of
output saved and invested in new capital sF(K*) is equal to the amount of
capital lost through depreciation 7K*. Hence, the capital stock in the next
period stays at K*.
Thus, if Ki = K*, then Kn = K* (for all n). But if K\ < K*, then the
infinite sequence of capital stocks (Kn) — defined by the above recursive
equation — is strictly increasing and converges to K* (as n tends to infinity).
On the other hand, if K\ > K*, then the infinite sequence of capital stocks
is strictly decreasing and also converges to K*.
The size of the cake in period n available for current consumption —
over which the two players will bargain — is 7vn — (1 — s)F(Kn). It follows
from the above properties of the infinite sequence of capital stocks that if
K\ = K*, then 7rn = (1 - s)F(K*) for all n > 1. But if, Kx < K*, then
the sequence of cakes is strictly increasing and converges to (1 — s)F(K*).
On the other hand, if K\ > K* then the sequence of the cakes is strictly
decreasing and also converges to (1 — TT)F(K*).

The Unique Steady State SSPE

It thus follows that if K\ — K*, then Proposition 10.1 — with TT = (1 —

s)F(K*) — characterizes the unique SSPE of this repeated bargaining model.
This is the unique steady state SSPE, since Kn = K* for all n > 1. The role
of the players' discount rates VA and VB in the limiting (as A —> 0) unique
steady state SSPE is as discussed in Section 10.2.2. Furthermore, it follows
from Corollary 10.2 that the payoff to player i (i = A, B) in the limiting (as
A —> 0) unique steady state SSPE is V* = ^TT, where TT = (1 - s)F(K*)
and & = rj/(l - a^rj + fan) (j ^ i).
10.3 An Application to Dynamic Capital Investment 315

The Optimal Unique Steady State SSPE

Consider any pair (V, s") of exogenously given savings rates, where 0 < sf <
s" < 1. Since K*(s') < K*(s"), it follows that TT' = (1 - s')F(K*(s')) is less
than, equal to, or greater than TT" = (1 — s")F(K*(s")). This implies that
V£(S') + VB(S') is less than, equal to, or greater than V£(S") + VB(S"), where
y|(s) + Vg(s) is the sum of the players' payoffs in the limiting (as A —>• 0)
unique steady state SSPE when the exogenously given savings rate is s. I
now derive the value of s G (0,1) that maximizes the sum V^(s) + Vg(s). It
follows from the expression above for V* that this value is identical to the
value of s G (0,1) that maximizes TT = (1 — s)F(K*(s)).
Differentiating TT w.r.t. s, it follows that dir/ds = (l—s)F'(K*)[dK*/ds\ —
F(K*). Since sF(K*) = 7K*, it follows that dK*/ds = -F(K*)/[sF'(K*)-
7]. Therefore, dn/ds = -F(K*)[F'(K*) - i]/[sF'(K*) - 7].
Now, given my assumptions on F, there exists a unique solution to
F'(K) = 7, which is denoted K<>. Furthermore, define s9 = jK9/F(K9).
It thus follows that TT is strictly increasing in s on the open interval (0, s9),
achieves a maximum at s = s9, and is strictly decreasing on the open interval
(s9,1). Indeed, the 'optimal' unique steady state SSPE — which maximizes
the sum of the players' equilibrium payoffs — is the one in which the exoge-
nously given savings rate is s9. Notice that when s = s9, the steady state
capital stock K9 has the property that F'(K9) = 7. That is, the marginal
product of capital equals the rate of capital depreciation.13

Bargaining over the Set of Unique Steady State SSPE

It follows from the properties (derived above) about the relationship between
TT and s that for any s G (0,1) such that s ^ s9, V£(s9) > V£(s) and
VB(S9) > VB(S)- Hence, if the players were to bargain at the beginning
of their relationship over the set of all unique steady state SSPE, and if
the bargaining outcome is Pareto efficient, then the players would agree to
implement the optimal unique steady state SSPE derived above in which
the savings rate is set at s9.
13
It may be noted that in the context of the one-sector Solow-Swan growth model, s 9 is
known as the golden rule savings rate, and Kg the golden rule steady state capital stock
(cf. Phelps (1961)).
316 Repeated Bargaining Situations

10.4 The Role of Outside Options

The role of outside options in one-shot bargaining situations has been stud-
ied in Chapter 5. In this section I explore their role in repeated bargaining
situations. In Section 10.2 it has been shown that the impact of the players'
discount rates in repeated bargaining situations can differ fundamentally to
that in one-shot bargaining situations. In contrast, it is shown below that
the impact of outside options in repeated bargaining situations is identical
to that in one-shot bargaining situations. In particular, the outside option
principle (which is stated in Corollary 5.1) carries over to repeated bargain-
ing situations.
A player's outside option in a RBS is defined as the payoff she obtains
when (and if) either player terminates the relationship that underlies the
RBS. For example, the payoffs from divorce are the players' outside options
in a marriage. Let E{ (where Ei > 0) denote player i's (i = A, B) outside
option. Thus, if some player terminates the relationship at time t (t > 0),
then player i's payoff is £^exp(—rib). Assume that

(1 - aA)EA + (1 - aB)EB < vr,

which ensures that it is mutually beneficial for the players never to exercise
their outside options. Notice that (1 — ai)Ei is player i's 'average' outside
option.
I explore the role of outside options in the context of a simple extension
to the repeated bargaining model studied in Section 10.2. The extension is
as follows. Whenever player j has to respond to an offer of a partition of
some cake made by player i (i ^ j), she has three options: (i) she can accept
the offer, (ii) she can reject the offer and (iii) she can 'opt out', in which
case the relationship between the players is terminated.
It is straightforward to adapt the arguments in Section 10.2.1 to charac-
terize the unique SSPE in this extended repeated bargaining model. In the
following proposition I characterize the unique SSPE in the limit as A —> 0.

Proposition 10.3 (Outside Option Principle in a RBS). The extended

repeated bargaining model with outside options has a unique SSPE. In equi-
librium, agreement is reached immediately on the partition of each and every
cake. In the limit, as A —» 0 ; the equilibrium shares of each and every cake
10.4 The Role of Outside Options 317

to players A and B are respectively z*A and z^, where z^ = vr — z*A and

Z
A if ZA - W
A and ZB -
Z W
A — \7i — WB if zA — A and zB < WB
1 if zA < WA and zB > WB,

with zA and zB defined in Corollary 10.2, WA = (1 — QLA)EA and WB =

(1 — (XB)EB- Furthermore, in this limit, the equilibrium payoffs to players A
and B are respectively VA* and VB*, where, for each i,j — A^B and i ^ j

W
if zA — A and zB > WB
y% — ^ y7i Wjj/yi (JL %j LJ z^ ^_ w% ana z- <^ Wj

if z\ < Wi and z* > Wj,

with VA and VB defined in Corollary 10.2.

Proof. Since the proof is a straightforward adaptation of the argument lead-

ing to Proposition 10.1, I only sketch out the main amendments to that
argument. The critical difference is that in a SSPE player j ' s payoff from
not accepting any offer is now max{£j, 6jV?}. It follows immediately that
if EA < 8AVA and EB < SBVBI then the unique SSPE is as stated in Propo-
sition 10.1. This means that in the limit, as A —> 0, the unique SSPE is as
stated in Corollary 10.2 provided that WA < zA and WB < zB.
Now suppose that EA < 8AVA and EB > $BVB. Assuming that A > 0
but small and in particular r > A — since I am going to focus on the
limiting case when A is arbitrarily small — it follows that the first equation
in (10.2) continues to be valid, while the second equation in (10.2) needs to
be slightly amended: the right-hand side is to be replaced by EB- These
two equations combined with the two equations in (10.1) — which continue
to be valid — have a unique solution, namely

1 - OLA^A + OLB{SA - OLA)

VB = TV + aBEB - (6A ~ OL
XA = TT - EB
xB = TT - (6A-
318 Repeated Bargaining Situations

Hence, this solution characterizes the unique SSPE provided that EA <
SAVA and EB > 8BVB — that is, provided that

™*

< SA^ and EB

- aB8B'

where x^ is stated in equation 10.4. In the limit, as A —+ 0, XA —> TT — wBi

xB —• wBi VA —* (TT — W B ) / ( 1 — OLA) and Vg —> £"B. Hence, since (from
Corollary 10.2) x*B —> 2:^ as A —• 0, it follows that in the limiting (as
A —> 0) unique SSPE, the shares of each and every cake to players A and B
are respectively n — wB and wB, provided that wB > z^.14 By a symmetric
argument, with the roles of A and B reversed, one may characterize the
limiting (as A —> 0) unique SSPE when WA > z\ and wB < z^.15 D

In Proposition 10.3 I have stated the impact of the players' outside op-
tions conditional on the relative magnitudes of z* and W{ (i = A,B), where
the former is the SSPE share of each and every cake to player i when neither
player has any outside options, while the latter is player i's 'average' outside
option. 16 However, it should be noted that since (for each i = A,B)

z* = Wi if and only if V* = E{,

the impact of the outside options may be stated conditional on the relative
magnitudes of V* and E{ {% = A,B).
Proposition 10.3 shows that the outside option principle, which was dis-
covered in the context of one-shot bargaining situations, is valid in the con-
text of repeated bargaining situations. In particular, the players' outside
options influence the SSPE if and only if some player's outside option is
strictly greater than her SSPE payoff.

Remark 10.7 (Opting out on a period-by-period basis). It is implic-

itly assumed above that if a player opts out in a RBS, then the players'
relationship is terminated forever. This makes sense in many repeated bar-
gaining situations. For example, divorce (typically) means that the couple
14
Notice that WB > zB implies that WA ^ zA.
15
Since WA + WB < TT and zA + zB — TT, it is not possible that WA > zA and WB > zB.
16
It is possible to interpret Wi as the share of some alternative cake that player i obtains
(every r time units) when and if the players terminate their relationship.
10.4 The Role of Outside Options 319

have terminated their marriage forever. However, in some repeated bar-

gaining situations it may be possible for the players to opt out on a 'period-
by-period' basis: that is, a player could opt out in any given period, but
subsequently the players continue their relationship (with the option to opt
out again for one period). For example, consider a bilateral monopoly mar-
ket for some input in which the seller and the buyer have the opportunity to
engage in a sequence of one-shot transactions. Furthermore, suppose that
the seller always has the option to sell the input to some alternative buyer
at some fixed price, and the buyer always has the option to buy the input
from some alternative seller at some fixed price. In this RBS the seller and
the buyer may choose to trade with these alternative traders in any given
period, but subsequently return to trade with each other. I now argue that
the impact of outside options in such a RBS is as stated in Proposition
10.3. Consider the following amendment to the extended repeated bargain-
ing model studied above (which formalizes this 'period-by-period' opting
out): when a player opts out (after the rejection of an offer of a partition of
some cake), player i immediately obtains a payoff of wi (i = A,B), and then
r time units later the two players return to bargain over the partition of the
next cake. Assuming that W{ > 0 and WA + WB < TT, it is straightforward to
show that Proposition 10.3 characterizes the unique SSPE of this amended
model.17

10.4.1 An Application to Firm Provided General Training

Consider a repeated bargaining situation in which, in each period, a firm
and a worker can produce a quantity of output /(/) after they reach agree-
ment over that period's wage, where / > 0 denotes the amount of general
training provided by the firm at the beginning of the RBS.18 Furthermore,
while bargaining over the wage rate (in any period) the worker has the op-
tion to quit working for this firm, and instead work for another firm at an
exogenously given wage rate v(I). If she exercises this option, then the firm
17
The argument is similar to that contained in the proof of Proposition 10.3, except
that it should be noted that in any SSPE player j ' s payoff from not accepting any offer is
now max{u>j + ctjV*, SjV*}.
18
The worker is assumed to be credit constrained, and, hence, cannot provide any such
training for herself.
320 Repeated Bargaining Situations

earns zero profit. Assume that the two players have a common discount rate
r > 0, and that / ( / ) > v(I)forall / > 0.
I apply Proposition 10.3. For any / > 0, the wage rate in the unique
(limiting, as A -> 0) SSPE is

j7(I)/2 if/(I)/2 >*(/)

w(I) = <
\v(I)

Hence, for any / > 0, the firm's equilibrium profit per period is

n(J) f/W/2 •df(I)/2>v(I)

\f(I)-v(I) if/(/)/2 <<;(/).

Letting c(I) denote the cost to the firm of providing I units of general
training before the RBS unfolds, it follows that the amount of firm provided
general training /* (where /* > 0) maximizes the firm's net total profit,
which is

where a = exp(—rr). Assuming that c'(0) = 0 and /7(0) > 0, it follows that
if /'(O) > v'(0) then II7(0) > 0. Thus, the firm will provide a strictly positive
amount of general training if the first unit of such general training increases
the marginal product of the worker more than it increases the worker's wage
rate at the other firm.19
As is well known (since Becker (1964)) this conclusion does not hold
in a competitive labour market — since in such a market the competitive
equilibrium wage rate is equal to the worker's marginal product. That is,
in such a market w(I) = v(I) = / ( / ) , and, hence, the firm in a competitive
labour market will provide no training in general skills.
The first best level of general training Ie maximizes / ( / ) — c(/). It is
clear that in general (with appropriate assumptions on / , v and c) /* < Ie.
Thus, although (when wages are determined through bargaining) the firm
will provide some training in general skills, she does not provide it at the
efficient level.
9
It should be noted that / ( / ) — and not / ' ( / ) — is the marginal product of the worker.
10.5 The Role of Long-Term Contracts 321

10.5 The Role of Long-Term Contracts

I now study the role of long-term contracts in the context of the archetypal
repeated bargaining situation in which a seller and a buyer engage in an
infinite sequence of one-shot transactions. Before the RBS unfolds, the
following sequence of events take place at 'dates' 1-3.
At date 1 the buyer and the seller have the opportunity to write a long-
term contract that specifies the terms of trade of each and every one-shot
transaction to be encountered. I shall elaborate on the details of such a
long-term contract in Sections 10.5.2 and 10.5.3. It is worth emphasizing
here that a long-term contract commits the parties to trading on the terms
specified in the contract — in the sense that when trade occurs either party
can choose to enforce that it take place on the terms specified in the contract.
However, trade can occur on terms that differ from those specified in the
contract if and only if both players agree to do so.
Whether or not a long-term contract is signed at date 1, at date 2 the
players simultaneously choose their respective investment levels IB and J^,
where Ii >0 (i = B,S). The cost to player i of investing an amount I\ is /^,
which she incurs at date 2. Then, at date 3 the state of nature is randomly
realized, where O denotes the finite set of possible states of nature, and 7(6*)
the probability that the state is 9 G O.
The buyer's reservation value v for a unit of the input depends on IB
and #, and the seller's unit cost of production c depends on Is and 9. They
are written as V(IB',0) and c(Is]9). It is assumed that for any 9 E O, v
is twice continuously differentiate, strictly increasing and strictly concave
in IB-> and that c is twice continuously differentiable, strictly decreasing
and strictly convex in 1$. The appropriate first-order and second-order
derivatives with respect to the investment levels are denoted by vf, v", c'
and c". For simplicity, but with some loss of generality, I assume that for
any /#, Is and 0, V(IB\0) > c(Is',9) — that is, gains from trade always
exist.
At date 4 the RBS begins to unfold. The buyer requires a single unit of
the input every r (r > 0) time units. As such the RBS involves a sequence
of one-shot transactions. The instantaneous profits to the buyer and the
seller respectively from trading a unit of the input at price p — given that
at date 2 the chosen investment levels are IB and Is, and at date 3 the
322 Repeated Bargaining Situations

realized state of nature is 9 — are V{IB\ 9) —p and p — c{Is\ 9). The players
have a common discount rate r > 0, where a = exp(—rr).
If no long-term contract is written at date 1, then the sequence of prices
at which trade occurs is determined via bargaining in the manner described
by the repeated bargaining model studied in Section 10.2. That is, the
players negotiate the price of the input when the buyer requires it. On the
other hand, if a long-term contract is written at date 1, then depending on
whether or not the contract is 'complete' — a notion that is discussed in
Section 10.5.2 — the players will either trade at the price specified in the
contract or renegotiate and trade at some other price.
In addressing the role of long-term contracts, attention will focus on the
extent to which such contracts affect the players' incentives to invest at date
2. As such it is useful to define the first best investment levels, which are
the investment levels Ig and J | that maximize the difference between the
expected present discounted value of the gains from trade in the RBS and
the total cost of investment. That is
~V{!B;6) - c(Is;0)
= arg max Ee - I B - IS-
IBJS 1 -a
Thus, (IB, Is) is a solution to the following pair of equations
Ee[v'(IB;9)] = 1 - a and E9[-J(IS\6)] = 1 - a. (10.5)

10.5.1 Equilibrium Without a Long-Term Contract

I first characterize the unique (limiting, as A —> 0) SSPE on the assumption
that the parties do not write a long-term contract at date 1. For any /#,
Is and 0, the unique (limiting) SSPE of the repeated bargaining model at
date 4 is characterized in Corollary 10.2 with n — V(IB\0) — c(Is\0) and
rA = TB = T. In this limiting equilibrium trade occurs immediately in
each and every bargaining situation at the same price, which is denoted by
P*(IB,IS,9)- Corollary 10.2 implies that the equilibrium profits from each
transaction to the buyer and the seller are respectively defined in (10.6) and
(10.7) stated below

v(IB;0) -p-(/fl>lSt9) = <IB;0)-C{IS;9) {1Q6)

P*(IB,Is,e) ~ c(Is;e) = * ( ^ ' ) - c ( J * ' ) .

10.5 The Role of Long-Term Contracts 323

Hence, the trade price

p (IB,IS,9) = . (10.8)
z
I now derive the Nash equilibrium investment levels chosen at date 2. For
any pair (/#, 1$), the expected equilibrium payoff to player i (i = B, S) is

Ee
[ 2(1-a)
Hence, a Nash equilibrium pair of investment levels (Jg, Ig) is a solution to
the following first-order conditions

E0[v'(IB',e)/2] =l - a a n d Ee[-c'(Is;9)/2} = 1 - a. (10.9)

It follows by comparing the equations in 10.9 with those in 10.5 that the
Nash equilibrium investment levels will be strictly less than the first best
investment levels — that is, /* < If (i = B,S). Each player under-invests
relative to her first best investment level because she only receives one-
half of the (average) expected marginal return from her investment. The
other half is negotiated away to her opponent at date 4. Thus, when the
long-term relationship between the buyer and the seller is governed by an
infinite sequence of short-term contracts, the parties have poor incentives to
undertake relationship-specific investments that enhance the expected gains
from trading within the relationship.

10.5.2 Equilibrium With a Complete Long-Term Contract

Now suppose that at date 1 the parties write a long-term contract that
specifies a price contingent on the realized state of nature. That is, the
long-term contract specifies the price p as a function of 6. Thus, if the
buyer chooses /#, the seller Is and the realized state of nature is #, then the
contract states that they trade the input at price p(9).
This type of contract can be enforced by a third party (such as the
courts) if and only if the realized state of nature is verifiable by the courts.
It is not sufficient (for the enforceability of the contract) that the realized
state of nature is observable by the buyer and the seller. When it is possible
for the courts to verify the values of all variables upon which the parties
324 Repeated Bargaining Situations

wish to condition the price, it is said that the parties are able to write a
complete long-term contract. In the next section I explore the role (if any)
of incomplete long-term contracts.
Fix an arbitrary complete long-term contract (p(0))eeQ such that for
any 9 G 9 , V(O;0) - p{0) > 0 and p(0) - c(0; 9) > 0. Many such complete
long-term contracts exist given my assumption that gains from trade always
exist. Since vf > 0 and c' < 0, it follows that for any 0 G O, IB > 0 and
Is > 0: V(IB'->0) > p{0) and p{0) > c(Is,0). Hence, given the complete
long-term contract, whatever the investment levels chosen at date 2 and
whatever the realized state of nature, at date 4 each player will prefer to
trade at the contracted price rather than not to trade. Consequently, for
any pair (/#, l^), at date 2 the expected payoffs to the buyer and the seller
are respectively

IB and I s.
I— a 1— a
Hence, a Nash equilibrium pair of investment levels (/#, 1$) is a solution to
the following first-order conditions

E0[v'(IB]0)] = l-a a n d E0[-cf(IS]0)] = l-a. (10.10)

It follows by comparing the equations in 10.10 with those in 10.5 that the
Nash equilibrium investment levels — given the complete long-term contract
— are equal to the first best investment levels. This is because each player
receives the full (average) expected marginal return from her investment.
Thus, when the long-term relationship is governed by a complete long-term
contract, the parties have excellent incentives to undertake relationship-
specific investments.

10.5.3 Equilibrium With an Incomplete Long-Term Contract

The assumption made in the preceding section that the realized state of
nature can be verified by third parties (such as the courts) is not particularly
realistic. For a variety of reasons — such as those due to the complexity and
multidimensionality of the state of nature — it is often too costly to write
a state-contingent contract in a sufficiently clear and unambiguous manner
that it can be enforced. In this section I therefore assume that the realized
10.5 The Role of Long-Term Contracts 325

state of nature is observable to the buyer and the seller, but not verifiable by
the courts. As such the parties cannot write a long-term contract in which
the price is conditioned on 9. For similar reasons, it is assumed that the
contracted price cannot be conditioned on the investment levels chosen at
date 2. It is said that such long-term contracts are incomplete.20
Suppose, nevertheless, that the parties do write an incomplete long-
term contract at date 1, which is a constant price p. Thus, for any chosen
investment levels and any realized state of nature, the contract specifies that
if trades occurs, then it occurs at price p. However, either player can choose
not to trade. The cost to a player of refusing to trade is zero, since the
courts cannot verify which party refused to trade if trade does not occur,
or alternatively, that the courts do not wish to enforce trade. For example,
in the absence of slavery, courts do not force an employee to work for an
employer.
Fix an arbitrary incomplete long-term contract p, investment levels IB
and /#, and state of nature 9. If the following two inequalities hold, then
trade occurs in each and every one-shot transaction at the contracted price
V
v(IB]0)-p>0 and p - c(IS] 9) > 0. (10.11)
These inequalities ensure that each party prefers to trade rather than not to
trade. Now suppose that one of these two inequalities fails to hold.21 This
implies that one of the two parties prefers not to trade at the contracted
price p. Since (by assumption) there are no penalities for not trading, she
will refuse to trade. In which case each player's profit per period is zero —
ignoring the costs of investment (which are sunk). But this is not ex-post
efficient, since V(IB', 0) > c(Is] 9). That is, if the parties were to renegotiate
the price and agree on a price p' such that V(IB]0) > p' > c(Is',0), then
each player would earn a strictly positive profit per period. Therefore, since
it is mutually beneficial for the parties to renegotiate the trade price, they
will indeed tear up the (incomplete) long-term contract and renegotiate.
Suppose that in this eventuality (when one of the two inequalities in 10.11
fails to hold) the parties do not write a new long-term contract, but now
20
For an excellent discussion on the reasons for contractual incompleteness, see Hart
and Holmstrom (1987).
21
It should be noted that since (by assumption) V(IB',0) — c(Is]0) > 0 for any /#, Is
and 6, at least one of these inequalities must hold.
326 Repeated Bargaining Situations

the sequence of trade prices are determined as in the repeated bargaining

model. Hence, the price (in the unique limiting SSPE) at which trade occurs
in each and every transaction is given by (10.8).
Hence, at date 4 the equilibrium expected payoffs to the buyer and the
seller are respectively

Uv(IB;0)-p)/(l-a) ifc(Is;e)<P<v(IB]0)
\(v(IB; 0) - c(Is; 0))/2(l - a) otherwise

and

f(p-c(Js;0))/(l-a) if c(Is;e)<p<v(IB] 6)
\(y(IB',0) - c(J5; 0))/2(l - a) otherwise.

For each pair I = {IBJS), define the set T(I) C 9 as follows: T(I) = {9 G
0 : c(Is\0) < p < V(IB]0)}. Hence, given the arbitrary incomplete long-
term contract p, the expected payoffs at date 2 to the buyer and the seller
respectively from an arbitrary pair {IB,IS) a r e

^ee/r(j)
l-a

^6>ee/r(/)
l-a

Consequently, given the incomplete long-term contract p, a Nash equilibrium

pair of investment levels (7^*,/^*) is a solution to the following first-order
conditions

Eoer{i)[vVBlO)]+E0ee/r{T)[v\IBiO)/2] - l - a (10.12)
Eeer(i)[-<!(Is\0)]+Eoee/r(i)[-<!(Is;0)ffl = I - a. (10.13)

It follows by comparing equations 10.12-10.13 with the equations in 10.5

that in general there is under-investment: /** < If (i = B,S). Thus, when
the players sign an incomplete long-term contract at date 1, each player's
equilibrium investment level is in general less than her first best investment
level. However, by comparing equations 10.12-10.13 with the equations
in 10.9 it follows that in general each player's equilibrium investment level
when an incomplete long-term contract is signed at date 1 is greater than her
10.6 Reputation Effects 327

equilibrium investment level when no long-term contract is signed at date

1. Therefore, although incomplete long-term contracts do not in general
achieve first best investments, the parties will prefer to sign such a contract
rather than not signing any long-term contract.
A main message that emerges here is that long-term contracts can have
a beneficial role in those repeated bargaining situations in which the par-
ties have the opportunity to engage in relationship-specific investments that
enhance the value of the RBS — notwithstanding the possibility that such
contracts will necessarily be highly incomplete.

10.6 Reputation Effects

A player's reputation for being a particular type of bargainer may have a
significant impact on the bargaining outcome. Consider, for example, a seller
whose unit cost of production is low, but who has a reputation for having a
high unit cost of production — by which I mean that the buyer attaches a
high probability to the seller's unit cost being high. With such a reputation
the negotiated price will tend to be relatively higher than it would be were
the seller not to possess such a reputation. As another example consider
a union that is weak in the sense that its members are unlikely to support
and endure a long lasting strike, but which has a reputation for being strong
— by which I mean that the firm attaches a high probability to the union
being strong. With such a reputation the negotiated wage and the other
terms of employment will tend to be relatively more attractive to the union
than they would be were the union not to possess such a reputation.
A host of questions arise concerning the role of reputations on the bar-
gaining outcome, including the following. What actions can a player take
in order to build (and maintain) a reputation for being a particular type of
bargainer? What are the costs of such actions, and what are the benefits
from acquiring such a reputation? Under what circumstances would a player
build a reputation?
Below I briefly consider such questions in the context of a simple repeated
bargaining model in which a seller and a buyer can trade a unit of some
input in each period, and the seller's unit cost of production is her private
information — it is either high or low.
328 Repeated Bargaining Situations

10.6.1 A Perfect Bayesian Equilibrium in a Simple Model

In each period t (where £ = 0 , l , 2 , . . . ) a seller and a buyer can trade a unit

of some input. The seller's unit cost of production c is her private informa-
tion, while the buyer's reservation value v is known to both players. This
asymmetry in information is modelled as follows. Before their relationship
begins in period 0, the seller's unit cost of production is randomly drawn
from the following binary distribution: with probability a (0 < a < 1) the
unit cost of production c = 1, and with probability 1 — a the unit cost of
production c — 0. The seller knows the realization of the random draw, but
the buyer does not. 22 It is assumed that v > 1, which implies that gains
from trade exist with probability one.
In each period £, the buyer makes a price offer pt to the seller. If she
accepts it, then trade occurs, and the profits obtained by the buyer and the
seller in this period are respectively v — pt and pt — c. However, if the seller
rejects the price offer, then trade does not occur in this period, and each
player obtains no profit in this period. In either eventuality the game then
moves to period t + 1. The discount rates of the buyer and the seller are
respectively r # and r s , where r # > 0 and rs > 0.
I now construct a perfect Bayesian equilibrium (PBE) that illustrates
the main point of this model. Consider the following pair of strategies. The
seller accepts a price offer p in any period if and only if p > 1, whether her
unit cost of production is high (c = 1) or low (c = 0). Letting fit denote the
probability the buyer attaches in period t — before making her offer — to
the seller having a high unit cost of production, the buyer's pricing strategy
is as follows: if fit > 0 then she offers pt = 1, and if fit = 0 then she offers
Pt = 0.
The buyer's posterior beliefs consistent with Bayes' rule — given the
above described pair of strategies — are as follows: (i) if fit = 0 then
jjit+1 = 0, (ii) if ^t > 0 and pt > 1 is rejected/accepted then /x^+i = l^ti (iii)
if fit > 0 and pt < 1 is rejected, then fit+i = /it, (iv) if fit > 0 and pt < 1 is
accepted, then fit+i = 0.
I now verify that the above described pair of strategies is a PBE. The
high cost type of seller has no incentive to deviate from this strategy. Fur-
22
The buyer only knows that the seller's unit cost of production is a random draw from
this binary distribution.
10.6 Reputation Effects 329

thermore, given the seller's strategy, the buyer has no incentive to deviate
from her strategy. The key issue is whether or not the low cost type of seller
can benefit from a one-shot (unilateral) deviation in which she accepts (in
any period t) a price offer pt < 1. If she deviates and accepts such a price
offer, then her profit in period t is pt and her profit in any period after (and
including) period t + 1 is zero. On the other hand, her payoff by conforming
to her proposed strategy, and thus rejecting this price offer, is Ss/(l — 8s) >
where 6s = exp(—rs). Hence, if #s/(l - 8s) > 1, then she does not benefit
from such a one-shot deviation.
In summary, therefore, the above described pair of strategies and buyer
beliefs is a PBE if 8s > 0.5. Thus, provided the seller is sufficiently patient,
there exists a PBE such that along the equilibrium path the buyer's price
offer in period t (for all t = 0,1, 2,...) is pt = 1, which is accepted by the
seller. The buyer has no incentive to deviate from this equilibrium price
path — no matter how small a might be — because of the credible threat
made by the low cost type of seller to reject any price offer p < 1.
Notice that in this PBE the buyer's and the seller's per-period profits
are respectively v — 1 and 1 — c. In contrast, it may be noted that with
perfect information — that is, when the seller's unit cost c is known to the
buyer — there exists a unique subgame perfect equilibrium (SPE) in which
trade occurs in each period on price p — c. Thus, with perfect information,
all the surplus from trading in each period is obtained by the buyer. In par-
ticular, the seller (whatever is her cost) obtains no profit. This observation
illustrates the potentially significant role that reputation effects can have on
the bargaining outcome.

10.6.2 Further Remarks

It is easy to show (by construction) that the model studied above possesses
a multiplicity of other PBE.23 Clearly further research is required here to
narrow down the set of equilibria in this type of model. It may be noted that
Hart and Tirole (1988) and Schmidt (1993) have studied the finitely repeated
version of this model — that is, with the assumption that the relationship
between the seller and the buyer lasts a finite number of periods, say N.
23
The construction is based on the following idea. A price path is supported in a PBE
by the threat of reverting to the PBE described above were the buyer to deviate from it.
330 Repeated Bargaining Situations

They show that the model has a unique PBE which converges, as TV —» oo,
to the PBE characterized above.
It goes without saying that the model studied above is very simple in
many respects. There is considerable scope to extend and amend it to
better understand the role of reputations in repeated bargaining situations.
For example, an interesting extension would be to assume that the buyer
(also) has private information (about her reservation value).
I conclude by noting that there is much need for research into the fas-
cinating topic of the role of reputations in repeated bargaining situations.
More models need to be constructed and studied, both in the abstract and
motivated by specific real-life bargaining situations. For example, I would
argue that a union that goes on a costly strike during the current year's wage
negotiations does so partly to build a reputation for being strong, where
some of the benefit from such a reputation is obtained during next year's
wage negotiations. It remains a challenge to construct plausible models of
such phenomena.

10.7 Notes
The repeated bargaining model studied in Section 10.2 is based upon Muthoo
(1995b). For an alternative repeated bargaining model, see Groes (1996).
The application to dynamic capital investment considered in Section 10.3
needs much further work. In particular, the analysis of the Markov SPE of
the model awaits characterization.
Felli and Harris (1996) and Leach (1997) study repeated bargaining mod-
els in which the sequence of bargaining situations are interdependent. The
former paper considers a situation in which a worker and two competing
firms bargain in an environment in which the worker accumulates some spe-
cific (payoff-relevant) information. The latter paper, on the other hand,
considers a situation in which a firm and a union bargain in an environment
in which the firm chooses whether or not to accumulate inventories. How-
ever, in order to simplify the analysis, Felli and Harris (1996) adopt (respec-
tively, Leach (1997) adopts) the (strong) assumption that in each period the
two competing firms make (respectively, the union makes) take-it-or-leave-
it-offers to the worker (respectively, firm). Nonetheless, both papers derive
some interesting results.
10.7 Notes 331

The application to firm-provided general training is based upon the anal-

yses in Acemoglu and Pischke (1999).
The analysis in Section 10.5 is inspired by the work of Hart and Moore
(1988). For an excellent introduction to the subject matter of incomplete
contracts and their role in long-term relationships, see Hart and Holmstrom
(1987, Section 3) and Hart (1995). For further references to the large litera-
ture on the subject, see the survey by Malcomson (1997). For an analysis of
short-term (complete) contracting in long-term relationships, see Crawford
(1988).
The finitely repeated version of the model studied in Section 10.6 has
been studied by Hart and Tirole (1988) and Schmidt (1993) — the former
also address the role of long-term contracts in this setting. Although the
finite-horizon assumption is problematic and has been criticized in this book,
these papers contain some useful, and insightful, techniques of analyses and
messages. It may be noted that there is a large literature on reputation
effects, but — with the exception of Hart and Tirole (1988) and Schmidt
(1993) — not in the setting of repeated bargaining situations. Nevertheless,
the techniques and insights from that literature should prove useful in future
research on the role of reputations in repeated bargaining situations. For an
excellent introduction to that literature, see Fudenberg and Tirole (1991).
11 Envoi

11.1 Introduction

The theory developed in the preceding chapters contains some fundamental

results and insights concerning the role of some key forces on the bargaining
outcome. It cannot be overemphasized that the focus of this theory is on
the fundamentals. Indeed, in the first stage of the development of an un-
derstanding of any phenomenon that is precisely the kind of theory that is
required — one that cuts across a wide and rich variety of real-life scenarios
and focuses upon their common core elements.
This objective to uncover the fundamentals of bargaining has meant
that my study has centred on some basic, elementary, models. That is
how it should be. However, it is important that we now move beyond the
fundamentals, in order to develop a richer theory of bargaining. With the
theory just developed providing appropriate guidance and a firm foundation,
it should be possible to construct tractable and richer models that capture
more aspects of real-life bargaining situations. At the same time, future
research should continue to develop the fundamentals; not only should we
further study the roles of the forces studied in this book — especially the
study of models in which many such forces are present — but we should
also study the role of other forces that have not been addressed in this book
(some of which I mention in Section 11.2).
334 Envoi

It is worth acknowledging that the role played by the game-theoretic

methodology in the development of the theory described in this book has
been crucial. I would like to suggest that notwithstanding the justifiable
criticisms of this methodology — which I touch upon in Section 11.3 —
research in the further development of the theory and application of bar-
gaining should continue to employ these methods. This methodology still
provides (and it will continue to do so in the foreseeable future) the best
set of tools currently available to formalize the richness of real-life bargain-
ing situations, and generate (in a relatively tractable manner) some useful
and insightful results. However, as I briefly argue in Section 11.3, research
should also (at the same time) proceed on developing new types of models
of bargaining, based upon a different set of tools (which are currently under
development).
I conclude this envoi (in Section 11.4) by offering some comments on the
role that bargaining experiments might play in the further development of
bargaining theory.
Before proceeding, however, I would like to make a general point con-
cerning the modelling of specific real-life bargaining situations. It appears
to me that, in many contexts, applied theorists have not yet developed mod-
els of bargaining that adequately and persuasively capture the essentials of
the bargaining situations that interest them. 1 It should be understood that
bargaining theory is there to guide on how best to construct bargaining mod-
els of real-life bargaining situations, and not to supply readily made such
models; it is meant primarily to provide the fundamentals.

11.2 Omissions
I begin by identifying two topics that have been studied in this book, but
where there is room for much further productive research (using the game-
theoretic methods employed in this book).
The first topic concerns repeated bargaining situations. The analyses in
Chapter 10 have only just scratched the surface of the complexities that
1
For example, I believe that we still do not have a plausible and persuasive model of
union-firm wage and employment negotiations. Roberts (1994) is an improvement over
the models that one tends to find in the labour economics literature, which seem to omit
some key aspects of such negotiations.
11.2 Omissions 335

characterize repeated bargaining situations. This is a topic of some consid-

erable importance, and research should be motivated substantially by the
many real-life repeated bargaining situations. The various types of linkages
that often exist amongst the sequence of bargaining situations — but which,
with the exception of the application in Section 10.3, were not studied in
Chapter 10 — are key elements of real-life repeated bargaining situations.
They need to be emphasized, and their roles studied, in future research. The
role of reputations, and reputation building and maintenance, in repeated
bargaining situations (briefly touched upon in Section 10.6) is clearly an
important force that we have hardly any deep understanding of.
The second topic concerns bargaining tactics. In Chapter 8 I focused
entirely on one type of tactics employed by bargainers. There is room, no
doubt, for much further research on the role of such commitment tactics,
which may perhaps proceed by enriching the models studied in that chapter.
But, in order to develop a deeper understanding, we need, in particular, to
develop models in which the processes through which players attempt to
make commitments stick are made explicit. That is, of course, a more
challenging enterprise. And, of course, the study of other types of tactics
employed by bargainers needs to be put on to the research agenda.

11.2.1 Non-Stationary and Stochastic Environments

A weakness of the theory developed thus far is that the bargaining environ-
ments modelled are stationary and deterministic. Although these assump-
tions have been useful in order to develop some of the fundamental results
in a simple manner, we should now study models based on non-stationary
and/or stochastic environments that characterize many real-life bargaining
situations. For example, in the context of union-firm wage negotiations, a
firm's market share might significantly decrease after (and if) the union has
been on strike for, say, two weeks. This introduces a non-stationary element
in the structure of the wage bargaining process that may have a significant
impact on the bargaining outcome — for a preliminary analysis of this type
of consideration, see Hart (1989).
There is a small literature that has made some inroads on this issue from
a theoretical perspective, which includes Binmore (1987), Merlo and Wilson
(1995), Groes and Sloth (1996), Cripps (1998) and Coles and Muthoo (1999).
336 Envoi

However, as will become clear from reading these papers, a lot more research
needs to be done, especially from an applied perspective; we ought to study
bargaining models motivated by the specific types of non-stationarity and/or
randomness found in real-life bargaining situations — as is, for example,
done in Hart (1989).

11.2.2 Multilateral and Coalitional Bargaining

In this book I have focused entirely on bilateral bargaining situations — that

is, bargaining situations involving two players. However, many important
and interesting bargaining situations involve three or more players. The
reasons for this omission are two-fold. Firstly, because of the primacy of
bilateral bargaining situations — they are the more basic, and the more
prevalent in real life. And, secondly, and this is partly related to the first
reason, the literature on multilateral and coalitional bargaining that uses
the (strategic or non-cooperative) game-theoretic methodology employed in
this book is extremely small (albeit growing) and under-developed. Having
said so, this is clearly an important area for future research.
An important, and rather fundamental, new element that rears its head
in bargaining situations with three or more players — but which is non-
existent in bilateral bargaining situations — is that players might consider
forming coalitions. This element makes the modelling of such bargaining
situations somewhat tricky. A theory of such bargaining situations should
not only determine what each player gets individually, but also determine
which coalitions will and will not form. It should thus, in particular, model
(in the extensive-form) the processes through which the players consider
whether or not to form a coalition.
Consider, for example, an extension to the exchange situation described
at the beginning of Section 1.1 in which there is another buyer who also
values the house at £70,000. In modelling this three player (one seller-
two buyers) bargaining situation, one should allow for the possibility that
the two buyers might consider forming a coalition, and then bargain with
the seller over the sale of the house in some way that takes into account
the side-deal they have made. At the same time, perhaps while the buyers
are negotiating the terms and structure of their coalition, the seller may
have the opportunity to approach one of the two buyers and negotiate to
11.2 Omissions 337

sell the house to her — a strategy intended to prevent the formation of the
buyers' coalition. Clearly, such factors need to be considered when modelling
bargaining situations with three or more players.
There is indeed a rich set of issues that arise here which are absent
from bilateral bargaining situations making the construction (and study) of
tractable, yet plausible, models somewhat tricky. I suggest that the devel-
opment of a theory of such bargaining situations should be pursued hand-
in-hand with the development of applied models specifically tailored to fit
specific real-life bargaining situations. This may help the former, since real-
life bargaining situations often have, to some extent, established mechanisms
through which coalition-format ion takes place.
Consider the extension of Rubinstein's bargaining model (which was
studied in Chapter 3) in which N players (where N > 3) bargain over
the partition of a single cake of fixed size according to an alternating-offers
procedure. It has been shown by Avner Shaked that the model possesses a
multiplicity of subgame perfect equilibria, provided that the time interval
between two consecutive offers is sufficiently small — see, for example, Os-
borne and Rubinstein (1990, Section 3.13) for a precise description of this
result. This result — which concerns a basic multilateral bargaining situa-
tion (in the absence of any coalition-formation issue) — has been a source of
frustration in the development of a theory of multilateral bargaining. One
particular recent resolution is contained in Krishna and Serrano (1996). But,
their model and results depend on a somewhat strong assumption (which is
the subject of discussion in section 8 of their paper). There is, indeed, room
for much further research on such multilateral bargaining situations.
Chatterjee, Dutta, Ray and Sengupta (1993) study a model of coalitional
bargaining. It is a generalization of Rubinstein's alternating-offers model in
which N players bargain over both what coalitions to form and what each
individual gets in each such coalition. Although they derive some interest-
ing and useful results, much research on modelling coalitional bargaining is
desperately required.
Issues of coalition-formation naturally arise when, for example, a single
firm bargains with more than one worker. Shaked and Sutton (1984) study
a model of such a situation on the strong assumption that the firm only
requires a single worker — and, furthermore, disallow (by assumption) the
338 Envoi

workers from forming a coalition.2 We still await — especially for the sake of
labour economics — for a plausible extension/generalization of the Shaked
and Sutton model to the case in which the employment level is endogenously
determined. Issues of coalition formation ought to figure prominently in such
a model.3
Although it is convenient to disallow coalition-formation when modelling
bargaining situations involving three or more players — and this may be jus-
tified when studying some real-life bargaining situations — future research
ought to focus attention on this issue.
It may be noted that bargaining and matching models of large markets
also rule out (by assumption) coalitions from being formed. For an excellent
introduction to that literature, see Osborne and Rubinstein (1990, Part II).
Most of this literature assumes that sellers and buyers are randomly matched
in pairs to bargain over the price at which to trade. Hence, it is not possible
for a group of sellers, say, to organize and form a coalition.4

11.2.3 Arbitration and Mediation

The role of arbitrators and mediators on the bargaining outcome is another
area that needs much research. The distinction between them is important.
While arbitrators impose an agreement, mediators facilitate the reaching of
an agreement by the players. There is a large literature on arbitration —
see, for example, Crawford (1981) for a brief survey. But most models in
that literature do not give the players sufficient opportunity to negotiate
an agreement on their own. Models in which the players negotiate in the
shadow of an arbitrator are hard to come by — in such a model the players
would call upon the arbitrator when, and if, they jointly agree to do so. For
2
Hendon and Tranes (1991), Jehiel and Moldovano (1995), and de Fraja and Muthoo
(1999) also study a one seller-two buyers bargaining situation and also disallow (by as-
sumption) the buyers from forming a coalition. The first of these papers studies the effect
of differing buyer reservation values, the second the effect of externalities between the
buyers, and the third the effect of asymmetric information.
3
Things will obviously get complicated when one considers the interaction of such
imperfect labour markets with oligopolistic product markets. For an early attempt, see
Davidson (1988).
4
For a bargaining and matching model in which sellers and buyers are not matched
randomly, but in which coalitions are also ruled out (by assumption), see Coles and Muthoo
(1998).
11.2 Omissions 339

two such models, see Manzini and Mariotti (1997) and Compte and Jehiel
(1998).
Turning to mediation, it seems to me that mediators do play some role
in some real-life bargaining situations, such as when a married couple are
facing difficulty reaching agreement. For example, a mediator may help co-
ordinate the players' expectations when at least one of the players possesses
private information on some variable. Ponsati and Sakovics (1995) explore
the impact of mediation on the efficiency of the outcome in a bargaining
situation with asymmetric information.

11.2.4 Multiple Issues and the Agenda

When there is more than one issue over which the players may negotiate,
the agenda may be important. The agenda not only specifies the order
and details of negotiations, but also the issues that will and will not be
subject to negotiations. Exploring the role of the bargaining agenda, and
how players might manipulate it, is an interesting and important area of
research, especially given that many real-life negotiations (such as trade
negotiations) involve multiple issues. The literature on this topic is small
(albeit growing) — see, for example, Fershtman (1990), Bac and Raff (1996),
and Lang and Rosenthal (1998).

11.2.5 Enforceability of Agreements

The issue of the enforceability of the agreements struck via bargaining is an
important one — as has been implicitly raised when I introduced the notion
of the incompleteness of a long-term contract in Section 10.5.2. With the
exception of the analyses in Sections 10.5.2 and 10.5.3, I have implicitly
assumed throughout the book that the agreement struck is implemented, in
the sense that each player fulfils her part of the agreement. Presumably this
assumption may be defended by an appeal to a contract that is enforceable
via the courts. However, in many bargaining situations (for a variety of
reasons) this cannot be taken for granted — notwithstanding the presence
of third parties (such as the courts).
The point is that players in real-life bargaining situations will be (and,
indeed, are) influenced by the issue of which agreements can and cannot be
enforced (either through third parties, or in a self-enforcing manner). Hence,
340 Envoi

the enforceability of agreements ought to play some role in determining the

bargaining outcome. This should be explored in future research.

11.3 Thorny Issues

The theory of bargaining developed in this book is based upon the Nash equi-
librium concept and its refinements (namely, the subgame perfect, Bayesian
Nash, and perfect Bayesian Nash equilibrium concepts) — in the sense that
the outcome of a bargaining situation is an equilibrium of some bargaining
game. An implicit notion which underlies these equilibrium concepts is that
the players are somewhat rational. Especially since there are bounds on the
rationality of humans, it is important and interesting to define such bounds
and explore their role on the bargaining outcome. Although this ought to
be a main topic for future research in bargaining theory, it is not an easy
topic to study. The basic problem lies in the construction of (tractable
and persuasive) models of boundedly rational players that formalize such
bounds on the rationality of humans, and which can be applied to study
bargaining situations. Rubinstein (1998) contains many interesting ideas
and tools that should prove helpful in constructing appropriate bargaining
models with boundedly rational bargainers.5
An attractive (and fairly tractable) methodology that has already been
used to explore the role of boundedly rational players on the bargaining
outcome is provided by evolutionary game theory and learning models.6
The handful of papers include Young (1993), Ellingsen (1997), Binmore,
Piccione and Samuelson (1998) and Poulsen (1998). As will become evident
from reading these papers, there is considerable scope for the development of
this type of bargaining model. Such models should complement the models
based on the game-theoretic methodology used in this book. Furthermore,
the role of (bargaining) norms and emotions — and there would appear to
be a role for them in real-life bargaining situations7 — may be fruitfully
explored using this type of model.
5
For an interesting discussion of some of the issues that such models ought to address,
see Bazerman, Gibbons, Thompson and Valley (1998).
6
Recent textbooks on these methods include Weibull (1995), Samuelson (1997) and
Fudenberg and Levine (1998).
7
See, for example, Elster (1989).
11.4 On the Role of Experiments 341

In order to use the game-theoretic methodology — as I have done in

this book — the rules of the bargaining process have to be precisely and
unambiguously specified, since only then can one specify the extensive-form
that underlies the bargaining game. This can be problematic. Although
bargaining situations are games, it is often the case — as is briefly mentioned
in Chapter 7 — that these rules are either only partly specified and/or
ambiguous. Some type of methodology based upon and/or adapted from the
methods of evolutionary game theory and learning models could perhaps be
used to fruitfully model bargaining situations without the need to precisely
and/or unambiguously specify the rules of the bargaining process.8

11.4 On the Role of Experiments

I now offer some comments on the role that bargaining experiments might
play in the further development of the theory and application of bargaining.
It is instructive to first re-emphasize the nature of the theory developed in
this book. This theory concerns the fundamentals of bargaining; it is meant
to provide an understanding of the roles of the main forces (or, variables)
on the bargaining outcome. It should therefore be judged on such terms
— that is, on the extent to which it helps us to better understand the
roles of various forces on the bargaining outcome.9 Furthermore, from an
applied perspective, this theory should help in the construction and study of
models of economic and political phenomena that are (partly) characterized
by bargaining situations.
It cannot be over-emphasized that the theory developed in this book does
not purport to generate falsifiable predictions a la Popper — and indeed it
cannot do so, since some of its key ingredients (such as the rationality of
the players — which is implicitly embodied in the equilibrium concepts em-
ployed) are neither observable nor controllable. As such this theory cannot
be tested by an appeal to experimental or field data. Besides, as stated
above, that is not its purpose.
A primary role of bargaining experiments should be to make new dis-
coveries about bargaining — for example, to identify new forces that may
8
For a most stimulating discussion of the issues raised in this section, see Kreps (1990b).
9
For illuminating discussions of what constitutes better understanding, see Kreps
(1990a, pp. 7-10) and Rubinstein (1998, pp. 190-194).
342 Envoi

have a significant impact on the bargaining outcome. In that way bargaining

experiments can help in the further development of the theory of bargaining.
It is worth noting that since the subjects of such experiments are hu-
mans (whose unobservable reasoning processes may differ from what may
be implicitly implied by the use of game-theoretic equilibrium concepts),
experimental observations and discoveries should help in the construction
of descriptive models of bargaining. Such models should complement the
game-theoretic bargaining models studied in this book.
Most of the bargaining games experimented on so far have been finitely
repeated versions of Rubinstein's model — with much attention being fo-
cused upon the ultimatum game (which is described in Section 7.2).10 It
may be noted that too much discussion has centred on the observation that
subjects in many of these experiments tend to behave in ways that differ
from what is indicated by the theory developed in this book. In view of
what I have said above about the nature of this theory, I fail to appreci-
ate the significance of this observation (and much of the discussion of it).
On the other hand, some of the discoveries provided by such experiments
(such as on the effects of deadlines) are to be welcomed. Indeed, motivated
(partly) by the experimental studies on the effects of deadlines, Ma and
Manove (1993) construct and study a game-theoretic bargaining model that
explores, and provides some understanding of, the role of deadlines on the
bargaining outcome. This is an illustration of the potentially important role
that bargaining experiments can play; they can help stimulate the construc-
tion and study of new (and useful) game-theoretic models of bargaining.
This, in turn, would improve our understanding of the determinants of the
bargaining outcome.
A basic limitation of bargaining experiments comes from the difficulty
of controlling variables that relate to the subjects of the experiments. These
subjects are not atoms, or chemicals. These subjects are humans, who — no
matter what they might say in an experimental set-up — will bring into the
experiment their life experience, expectations, perceptions, norms, moral
principles, and — most important of all — their reasoning processes. As
such there is little hope (at least in the foreseeable future) of generating
falsifiable predictions from both game-theoretic and descriptive models of
bargaining. Therefore, the focus of bargaining experiments should be on
°For an excellent survey of the literature on bargaining experiments, see Roth (1995).
11.4 On the Role of Experiments 343

making new discoveries about bargaining, which would then feed into the
further development of the theory of bargaining.
References

Abreu, D. (1988), "On the Theory of Infinitely Repeated Games with Dis-
counting," Econometrica, 56: 383-396.
Abreu, D. and F. Gul (1999), "Reputation and Bargaining," Econometrica,
forthcoming.
Acemoglu, D. and J. Pischke (1999), "The Structure of Wages and Invest-
ment in General Training," Journal of Political Economy, forthcoming.
Akerlof, G. (1970), "The Market for Lemons: Qualitative Uncertainty and
the Market Mechanism," Quarterly Journal Of Economics, 89: 488-500.
Ausubel, L. and R. Deneckere (1989), "Reputation in Bargaining and
Durable Goods Monopoly," Econometrica, 57: 511-531.
Avery, C. and P. Zemsky (1994), "Money Burning and Multiple Equilibria
in Bargaining," Games and Economic Behavior, 7: 154-168.
Bac, M. and H. Raff (1996), "Issue-By-Issue Negotiations: The Role of
Information and Time Preferences," Games and Economic Behavior, 13:
125-134.
Basu, K., S. Bhattacharya and A. Mishra (1992), "Notes on Bribery and
the Control of Corruption," Journal of Public Economics, 48: 349-359.
Bazerman, M., R. Gibbons, L. Thompson and K. Valley (1998), "Can Ne-
gotiators OutPerform Game Theory," in J. Halpern and R. Stern (eds.),
Debating Rationality, Cornell University Press, Ithaca, New York.
Bebchuk, L. (1984), "Litigation and Settlement under Imperfect Informa-
tion," Rand Journal of Economics, 15: 404-415.
346 References

Becker, G. (1964), Human Capital, University of Chicago Press, Chicago.

Binmore, K. (1985), "Bargaining and Coalitions," in A. Roth (ed.), Game-
Theoretic Models of Bargaining, Cambridge University Press, Cam-
bridge.
Binmore, K. (1987), "Perfect Equilibria in Bargaining Models," in K. Bin-
more and P. Dasgupta (eds.), The Economics of Bargaining, Basil Black-
well, Oxford.
Binmore, K. (1992), Fun and Games, D.C. Heath and Company, Lexington,
Massachusetts.
Binmore, K., M. Piccione and L. Samuelson (1998), "Evolutionary Stability
in Alternating-Offers Bargaining Games," Journal of Economic Theory,
80: 257-291.
Binmore, K., A. Rubinstein and A. Wolinsky (1986), "The Nash Bargaining
Solution in Economic Modelling," Rand Journal of Economics, 17: 176-
188.
Bolt, W. and H. Houba (1997), "Strategic Bargaining in the Variable Threat
Game," Economic Theory, 11: 57-77.
Bulow, J. and K. Rogoff (1989), "A Constant Recontracting Model of
Sovereign Debt," Journal of Political Economy, 97: 155-178.
Busch, L., S. Shi and Q. Wen (1998), "Bargaining with Surplus Destruction,"
Canadian Journal of Economics, forthcoming.
Busch, L. and Q. Wen (1995), "Perfect Equilibria in a Negotiation Model,"
Econometrica, 63: 545-565.
Carlsson, H. (1991), "A Bargaining Model where Parties make Errors,"
Econometrica, 59: 1487-1496.
Chang, R. (1995), "Bargaining a Monetary Union," Journal of Economic
Theory, 66: 89-112.
Chatterjee, K., B. Dutta, D. Ray and K. Sengupta (1993), "A Non-
Cooperative Theory of Coalitional Bargaining," Review of Economic
Studies, 60: 463-477.
Chatterjee, K. and L. Samuelson (1990), "Perfect Equilibria in Simultaneous
Offer Bargaining," International Journal of Game Theory, 19: 237-267.
Chaudhuri, P. (1992), Joint Ventures and Bargaining, Ph.D. thesis, Indian
Statistical Institute.
Chaudhuri, P. (1994), "Firm Size and Pricing Policy," Jawaharlal Nehru
University, mimeo.
References 347

Coase, R. (1972), "Durability and Monopoly," Journal of Law and Eco-

nomics, 15: 143-149.
Coles, M. and A. Muthoo (1998), "Strategic Bargaining and Competitive
Bidding in a Dynamic Market Equilibrium," Review of Economic Stud-
ies, 65: 235-260.
Coles, M. and A. Muthoo (1999), "Bargaining Equilibrium in a Non-
Stationary Environment," University of Essex, mimeo.
Compte, O. and P. Jehiel (1998), "When Outside Options Force Concessions
to be Gradual," C.E.R.A.S., Paris.
Crawford, V. (1981), "Arbitration and Conflict Resolution in Labor-
Management Bargaining," American Economic Review Papers and Pro-
ceedings, 71: 205-210.
Crawford, V. (1982), "A Theory of Disagreement in Bargaining," Econo-
metrica, 50: 607-637.
Crawford, V. (1988), "Long-Term Relationships Governed by Short-Term
Contracts," American Economic Review, 78: 485-499.
Cripps, M. (1998), "Markov Bargaining Games," Journal of Economic Dy-
namics and Control, 22: 341-355.
Davidson, C. (1988), "Multi-Unit Bargaining in Oligopolistic Industries,"
Journal of Labor Economics, 6: 397-422.
de Fraja, G. and A. Muthoo (1999), "Equilibrium Partner Switching in
a Bargaining Model with Asymmetric Information," International Eco-
nomic Review, forthcoming.
de Meza, D. and B. Lockwood (1998), "Does Asset Ownership Always Mo-
tivate Managers?: Outside Options and the Property Rights Theory of
the Firm," Quarterly Journal of Economics, 113: 361-386.
Ellingsen, T. (1997), "The Evolution of Bargaining Behavior," Quarterly
Journal of Economics, 112: 581-602.
Elster, J. (1989), The Cement of Society, Cambridge University Press, Cam-
bridge.
Evans, R. (1989), "Sequential Bargaining with Correlated Values," Review
of Economic Studies, 56: 499-510.
Felli, L. and C. Harris (1996), "Learning, Wage Dynamics and Firm-Specific
Human Capital," Journal of Political Economy, 104: 838-868.
Fernandez, R. and J. Glazer (1991), "Striking for a Bargain Between Two
Completely Informed Agents," American Economic Review, 81: 240-252.
348 References

Fershtman, C. (1990), "The Importance of the Agenda in Bargaining,"

Games and Economic Behavior, 2: 224-238.
Fudenberg, D. and D. Levine (1998), The Theory of Learning in Games,
MIT Press, Cambridge, Massachusetts.
Fudenberg, D., D. Levine and J. Tirole (1985), "Infinite-Horizon Models of
Bargaining with One-Sided Incomplete Information," in A. Roth (ed.),
Game-Theoretic Models of Bargaining, Cambridge University Press,
Cambridge.
Fudenberg, D. and J. Tirole (1991), Game Theory, MIT Press, Cambridge,
Massachusetts.
Gibbons, R. (1992), Game Theory for Applied Economists, Princeton Uni-
versity Press, Princeton, NJ.
Groes, E. (1996), "Asymptotically Finite Repetition of Rubinstein's Bar-
gaining Game," Institute of Economics, University of Copenhagen,
mimeo.
Groes, E. and B. Sloth (1996), "Sequential Bargaining Theory with Decreas-
ing Discount Factors," Institute of Economics, University of Copenhagen,
mimeo.
Grossman, S. and O. Hart (1986), "The Costs and Benefits of Ownership: A
Theory of Vertical and Lateral Integration," Journal of Political Econ-
omy, 94: 691-719.
Grossman, S. and M. Perry (1986), "Sequential Bargaining under Asym-
metric Information," Journal of Economy Theory, 39: 120-154.
Gul, F., H. Sonnenschein and R. Wilson (1986), "Foundations of Dynamic
Monopoly and the Coase Conjecture," Journal of Economy Theory, 39:
155-190.
Haller, H. and S. Holden (1990), "A Letter to the Editor on Wage Bargain-
ing," Journal of Economy Theory, 52: 232-236.
Hart, O. (1989), "Bargaining and Strikes," Quarterly Journal of Economics,
104: 25-44.
Hart, O. (1995), Firms, Contracts, and Financial Structure, Oxford Univer-
sity Press, Oxford.
Hart, O. and B. Holmstrom (1987), "The Theory of Contracts," in T. Be-
wley (ed.), Advances in Economic Theory, Cambridge University Press,
Cambridge.
Hart, O. and J. Moore (1988), "Incomplete Contracts and Renegotiation,"
References 349

Econometrica, 56: 755-786.

Hart, O. and J. Moore (1990), "Property Rights and the Nature of the
Firm," Journal of Political Economy, 98: 1119-1158.
Hart, O. and J. Tirole (1988), "Contract Renegotiation and Coasian Dy-
namics," Review of Economic Studies, 55: 509-540.
Hay, B. and K. Spier (1998), "Settlement of Litigation," in P. Newman (ed.),
The New Palgrave Dictionary of Economics and the Law, Macmillan
Reference Limited.
Hendon, E., H. Jacobsen and B. Sloth (1996), "The One-Shot-Deviation
Principle for Sequential Rationality," Games and Economic Behavior,
12: 274-282.
Hendon, E. and T. Tranes (1991), "Sequential Bargaining in a Market with
One Seller and Two Different Buyers," Games and Economic Behavior,
3: 453-466.
Hindriks, J., M. Keen and A. Muthoo (1998), "Corruption, Extortion and
Evasion," University of Essex, mimeo.
Holden, S. (1994), "Bargaining and Commitment in a Permanent Relation-
ship," Games and Economic Behavior, 7: 169-176.
Jehiel, P. and B. Moldovano (1995), "Negative Externalities May Cause
Delay in Negotiations," Econometrica, 63: 1321-1335.
Kambe, S. (1999), "Bargaining with Imperfect Commitment," Games and
Economic Behavior, forthcoming.
Kreps, D. (1990a), A Course in Microeconomic Theory, Princeton University
Press, Princeton, NJ.
Kreps, D. (1990b), Game Theory and Economic Modelling, Oxford Univer-
sity Press, Oxford.
Krishna, V. and R. Serrano (1996), "Multilateral Bargaining," Review of
Economic Studies, 63: 61-80.
Lang, K. and R. Rosenthal (1998), "Multi-Issue Bargaining with Perfect
Information," Boston University, mimeo.
Leach, J. (1997), "Inventories and Wage Bargaining," Journal of Economic
Theory, 75: 433-463.
Luce, D. and H. Raiffa (1957), Games and Decisions, John Wiley and Sons,
New York.
Lundberg, S. and R. Pollak (1993), "Separate Spheres Bargaining and the
Marriage Market," Journal of Political Economy, 101: 988-1010.
350 References

Ma, C. and M. Manove (1993), "Bargaining with Deadlines and Imperfect

Player Control," Econometrica, 61: 1313-1339.
Malcomson, J. (1997), "Contracts, Hold-Up, and Labor Markets," Journal
of Economic Literature, 35: 1916-1957.
Manzini, P. (1999), "Strategic Bargaining with Destructive Power," Eco-
nomics Letters, forthcoming.
Manzini, P. and M. Mariotti (1997), "A Model of Bargaining with the Possi-
bility of Arbitration," Working Paper No. 374, Economics Dept., QMW
College, London.
Mas-Colell, A., M. Whinston and J. Green (1995), Microeconomic Theory,
Oxford University Press, Oxford.
McDonald, I. and R. Solow (1981), "Wage Bargaining and Employment,"
American Economic Review, 71: 896-908.
Merlo, A. and C. Wilson (1995), "A Stochastic Model of Sequential Bar-
gaining with Complete Information," Econometrica, 63: 371-399.
Muthoo, A. (1990), "Bargaining Without Commitment," Games and Eco-
nomic Behavior, 2: 291-297.
Muthoo, A. (1991), "A Note on Bargaining Over a Finite Number of Feasible
Agreements," Economic Theory, 1: 290-292.
Muthoo, A. (1994), "A Note on Repeated-Offers Bargaining With One-Sided
Incomplete Information," Economic Theory, 4: 295-301.
Muthoo, A. (1995a), "A Bargaining Model with Players' Perceptions on the
Retractability of Offers," Theory and Decision, 38: 85-98.
Muthoo, A. (1995b), "Bargaining in a Long-Term Relationship with En-
dogenous Termination," Journal of Economic Theory, 66: 590-598.
Muthoo, A. (1995c), "On the Strategic Role of Outside Options in Bilateral
Bargaining," Operations Research, 43: 292-297.
Muthoo, A. (1996), "A Bargaining Model Based on the Commitment Tac-
tic," Journal of Economic Theory, 69: 134-152.
Muthoo, A. (1998), "Sunk Costs and the Inefficiency of Relationship-Specific
Investment," Economica, 65: 97-106.
Myerson, R. (1979), "Incentive Compatibility and the Bargaining Problem,"
Econometrica, 47: 61-73.
Myerson, R. (1991), Game Theory: Analysis of Conflict, Harvard University
Press, Cambridge, Massachusetts.
Myerson, R. and M. Satterthwaite (1983), "Efficient Mechanisms for Bilat-
References 351

eral Trading," Journal of Economic Theory, 29: 265-281.

Nandeibam, S. (1996), "Bargaining in Partnerships," University of Birming-
ham, mimeo.
Nash, J. (1950), "The Bargaining Problem," Econometrica, 18: 155-162.
Nash, J. (1953), "Two-Person Cooperative Games," Econometrica, 21: 128-
140.
Osborne, M. and A. Rubinstein (1990), Bargaining and Markets, Academic
Press, San Diego, California.
Osborne, M. and A. Rubinstein (1994), A Course in Game Theory, MIT
Press, Cambridge, Massachusetts.
Perry, M. and P. Reny (1993), "A Non-Cooperative Bargaining Model with
Strategically Timed Offers," Journal of Economic Theory, 59: 55-77.
Phelps, E. (1961), "The Golden Rule of Accumulation: A Fable for Growth-
men," American Economic Review, 51: 631-643.
Ponsati, C. and J. Sakovics (1995), "Mediation is Necessary for Dynamic
Efficiency in Bargaining," UAB (Barcelona), mimeo.
Ponsati, C. and J. Sakovics (1998), "Rubinstein Bargaining with Two-Sided
Outside Options," Economic Theory, 11: 667-672.
Poulsen, A. (1998), "An Evolutionary Analysis of Bargaining," University
of Essex, mimeo.
Reinganum, J. and L. Wilde (1986), "Settlement, Litigation, and the Allo-
cation of Litigation Costs," Rand Journal of Economics, 17: 557-566.
Roberts, K. (1994), "Wage and Employment Determination through Non-
Cooperative Bargaining," STICERD Theoretical Economics Discussion
Paper No: TE/94/274, London School of Economics.
Roth, A. (1995), "Bargaining Experiments," in J. Kagel and A. Roth
(eds.), Handbook of Experimental Economics, Princeton University
Press, Princeton, NJ.
Rubinstein, A. (1982), "Perfect Equilibrium in a Bargaining Model," Econo-
metrica, 50: 97-110.
Rubinstein, A. (1985a), "A Bargaining Model with Incomplete Information
about Time Preferences," Econometrica, 53: 1151-1172.
Rubinstein, A. (1985b), "Choice of Conjectures in a Bargaining Game with
Incomplete Information," in A. Roth (ed.), Game-Theoretic Models of
Bargaining, Cambridge University Press, Cambridge.
Rubinstein, A. (1991), "Comments on the Interpretation of Game Theory,"
352 References

Econometrica, 59: 909-924.

Rubinstein, A. (1998), Modeling Bounded Rationality, MIT Press, Cam-
bridge, Massachusetts.
Rubinstein, A., Z. Safra and W. Thomson (1992), "On the Interpretation of
the Nash Bargaining Solution and Its Extension to Non-Expected Utility
Preferences," Econometrica, 60: 1171-1186.
Rubinstein, A. and A. Wolinsky (1985), "Equilibrium in a Market with
Sequential Bargaining," Econometrica, 53: 1131-1150.
Samuelson, L. (1997), Evolutionary Games and Equilibrium Selection, MIT
Press, Cambridge, Massachusetts.
Samuelson, W. (1984), "Bargaining under Asymmetric Information,"
Econometrica, 52: 992-1005.
Schelling, T. (1960), The Strategy of Conflict, Harvard University Press,
Cambridge, Massachusetts.
Schmidt, K. (1993), "Commitment through Incomplete Information in a
Simple Repeated Bargaining Model," Journal of Economic Theory, 60:
114-139.
Shaked, A. (1994), "Opting Out: Bazaars versus 'High Tech' Markets,"
Investigaciones Economicas, 18: 421-432.
Shaked, A. and J. Sutton (1984), "Involuntary Unemployment as a Perfect
Equilibrium in a Bargaining Model," Econometrica, 52: 1351-1364.
Spence, M. (1974), Market Signaling, Harvard University Press, Cambridge,
Massachusetts.
Spier, K. (1994), "Pretrial Bargaining and the Design of Fee-Shifting Rules,"
Rand Journal of Economics, 25: 197-214.
Sundaram, R. (1996), A First Course in Optimization Theory, Cambridge
University Press, Cambridge,
van Damme, E. (1991), Stability and Perfection of Nash Equilibria, Springer-
Verlag, Berlin, second edition.
van Damme, E., R. Selten and E. Winter (1990), "Alternating Bid Bargain-
ing with a Smallest Money Unit," Games and Economic Behavior, 2:
188-201.
Vincent, D. (1989), "Bargaining with Common Values," Journal of Eco-
nomic Theory, 48: 47-62.
Wang, G. (1998), "Bargaining over a Menu of Wage Contracts," Review of
Economic Studies, 65: 295-305.
References 353

Weibull, J. (1995), Evolutionary Game Theory, MIT Press, Cambridge, Mas-

sachusetts.
Wolinsky, A. (1987), "Matching, Search and Bargaining," Journal of Eco-
nomic Theory, 42: 311-333.
Young, P. (1993), "An Evolutionary Model of Bargaining," Journal of Eco-
nomic Theory, 59: 145-168.
Index

agenda, 339 two-sided uncertainty, 265-270

alternating-offers procedure, see bar- correlated values, 268-270
gaining procedures, alternating- private values, 266-268
offers asymmetric Nash bargaining solu-
arbitration, 338-339 tion, see Nash bargaining
asymmetric information, 251-292 solution, asymmetric
applications
pre-trial bargaining, 263-265, bargaining
271 informal definition, 2
wage negotiations, 262, 270- bargaining costs, see bargaining power,
271 haggling costs
wage-quality contracts, 288- bargaining frictions, 3-4, 50-51, 54-
292 55, 77, 88
bargaining power, 271-292 bargaining power, 2-3
efficiency, 253-271 discounting, 51-52, 63-64
one-sided uncertainty, 253-261 haggling costs, 51-52, 54-55
correlated values, 256-261 bargaining procedures, 187-209
private values, 254-255 alternating offers, 42-43
repeated offers model, 272-288 different response times, 193—
gap case, 282-284 194
no gap case, 284-285 random proposers, 192-193
retractable offers, 285-288 ambiguous, 340-341
Index 355

repeated offers, 190 two-sided uncertainty, 266-269

retractable offers, 194-199 discount factor, 43
simultaneous offers, 191-192 discount rate, 43
ultimatum game, 189 discounting
who makes offers and when, role of, 46-47, 90
188-194
enforceability of agreements, 339
bargaining situations
evolutionary bargaining models, 339-
importance of, 2
340
informal general definition, 1-
experiments, 341-342
2
bargaining tactics, see commitment finite set of agreements, 210
tactics frictionless bargaining processes, see
bounded rationality, 339-340 bargaining frictions
burning money, 200-209
application generalized Nash bargaining solu-
surplus destruction, 208-209 tion, see Nash bargaining
solution, asymmetric
coalitional bargaining, 336-338
Coase conjecture, 278-280 impasse point
Coasian dynamics, 278-280 definition, 52
commitment tactics, 211-248 inefficient outcomes, 2-3, 173-174,
application 197-198, 204-207, 209-210,
delegation, 230-232 235-236, 260-261, 267-270,
general model, 226-230 278-280, 292, 309-311
illustration, 211-212 inside options, 137-184
simple model, 214-226 applications
commitment tactics and uncertainty, intrafamily allocation, 154-
232-248 158
simultaneous concessions, 233- sovereign debt renegotiations,
240 144-146
wars of attrition, 240-248 takeovers, 143-144
continuous time bargaining, 209 wage renegotiations, 170-174
endogenously determined, 158-
deadlines, 341-342 170
delay, see inefficient outcomes simple model, 138-143
direct revelation procedures inside options and outside options
one-sided uncertainty, 257-260 simple model, 146-149
356 Index

inside options, outside options, risk risk of breakdown, 80

of breakdown and discount- risk of breakdown and dis-
ing counting, 88-89, 96
general model, 152-153 Nash demand game, 224-225
simple model, 149-151 Nash perturbed demand game, 249-
250
long-term relationships, 299
Nash variable threat game, 249
mediation, 338-339 norms, 340
multilateral bargaining, 336-338
multiple issues, 339 one-shot deviation property, 127-
128
Nash bargaining solution, 9-39 outside option principle, 103
applications generalization, 111-113, 148-
asset ownership, 17-22 149
corruption, 16-17, 29 outside options, 99-135
moral hazard in teams, 26-
applications
29 corruption, 109-110
union-firm negotiations, 25- relationship-specific investment,
26 105-107, 130-131
asymmetric sovereign debt negotiations,
definition, 35-36
107-109
axiomatization, 29-33
searching for, 116-124
definition, 11
simple model, 100-105
general definition, 22-23
telephone bargaining, 124-130
geometric characterization, 13-
outside options and risk of break-
14
down, 110-116
interpretation, 33-35
general model, 115-116
why, when and how to apply
simple model, 110-115
commitment tactics, 223-224
split-the-difference rule or out-
discounting, 52-53, 65-67
side option principle, 114-
inside options, 141-142
115
inside options, outside op-
tions, risk of breakdown procedures, see bargaining proce-
and discounting, 151-153 dures
outside options, 104-105
outside options and risk of repeated bargaining situations, 295-
breakdown, 113-114, 116 330
Index 357

applications when and how to apply

dynamic capital investment, simple version
312-315 description, 42-43
general training, 319-320 Nash equilibria, 51
long-term contracts, 321-327 unique subgame perfect equi-
outside options, 316-319 librium, 43-46
simple model, 297-311 small time intervals, 46-47
reputation effects, 327-330
revelation principle, 258 screening equilibrium, 273-280
skimming property, 280-281
risk aversion, 16, 30, 33, 77-80,
split-the-difference rule
95-96
inside options, 141
risk of breakdown, 73-96
Nash bargaining solution, 15-
application
16, 25, 36
corruption, 81-85
risk of breakdown, 77, 89-91
simple model, 74-80
risk of breakdown and outside
risk of breakdown and discount-
options, 114-115
ing, 85-96
stochastic environments, 335-336
application
strikes, see inefficient outcomes
price determination, 91-95
general model, 95-96 wars of attrition, see commitment
simple model, 85-91 tactics and uncertainty
Rubinstein's bargaining game, 41-
71
applications
bilateral monopoly, 55-59
exchange economy, 69-71
general version
description, 59-60
small time intervals, 64-65
unique subgame perfect equi
librium, 60-63
interpretation, 53-55
quit options, effect of, 131-132
relationship with Nash's bar-
gaining solution, see Nash
bargaining solution, why,