Game Theory

GamE Theory
An Applied
Introduction
José Luis Ferreira

GAME THEORY
GAME THEORY
An Applied Introduction
JOSÉ LUIS FERREIRA

© José Luis Ferreira, under exclusive licence to Springer Nature Limited 2020
All rights reserved. No reproduction, copy or transmission of this publication may be

made without written permission.
No portion of this publication may be reproduced, copied or transmitted save with

written permission or in accordance with the provisions of the Copyright, Designs and
Patents Act 1988, or under the terms of any licence permitting limited copying issued
by the Copyright Licensing Agency, Saffron House, 6–10 Kirby Street, London EC1N 8TS.
Any person who does any unauthorized act in relation to this publication may be liable
to criminal prosecution and civil claims for damages.
The author has asserted his right to be identified as the author of this work in accord-
ance with the Copyright, Designs and Patents Act 1988.
First published 2020 by

RED GLOBE PRESS
Red Globe Press in the UK is an imprint of Springer Nature Limited, registered in E

ngland,
company number 785998, of 4 Crinan Street, London N1 9XW.
Red Globe Press® is a registered trademark in the United States, the United Kingdom,
Europe and other countries.
ISBN 978–1–352–00791–6
This book is printed on paper suitable for recycling and made from fully managed and
sustained forest sources. Logging, pulping and manufacturing processes are expected
to conform to the environmental regulations of the country of origin.
A catalogue record for this book is available from the British Library.
A catalog record for this book is available from the Library of Congress.
To Lourdes, Ariel, Ainara and Marina, who gave me a family
The game is afoot.
Sir Arthur Conan Doyle, The Return of Sherlock
Holmes
CONTENTS
List of Figures and Tables xi

Acknowledgementsxvi
About this book xvii
Introductionxix
1. The most famous games 1

1.1 The coordination game 1
1.2 Choice of standards 2
1.3 The battle of the sexes 4
1.4 The chicken game 5
1.5 The prisoners’ dilemma 7
1.6 Matching pennies 12
1.7 The ultimatum game 14
Exercises17
2. Building the theory for simultaneous games 20

2.1 The normal form game 20
2.2 Towards a solution 22
2.3 Some propositions on maximin strategies, rationalizable
strategies and Nash equilibria 27
2.4 Finding the Nash equilibria 29
2.5 Complications in finding the Nash equilibria 33
2.6 The payoffs of the game and the mixed strategies 36
Exercises38
3. Static games 41
3.1 Fiscal battles 41
3.2 The median voter 42
3.3 The advantage of being indifferent 45
3.4 The broken windows theory 48
3.5 The independence of Sylvania 50
3.6 Cournot oligopoly 53
3.7 Bertrand oligopoly 56
3.8 Keeping up with the Joneses 58
Exercises60
vii
CONTENTS
4. Dynamic games 62
4.1 The extensive form and backwards induction 62
4.2 The prisoners’ dilemma with a previous contract 64
4.3 Subgame perfect Nash equilibrium 66
4.4 How to be credible (1) Eliminate strategies: Odysseus and the Sirens 69
4.5 How to be credible (2) Acquire costly compromises: Who enters? 70
4.6 How to be credible (3) Give up control: Separation of powers 72
4.7 The consequences of not being credible: The healthcare game 73
4.8 The payment of the debt 75
Exercises77
5. Voting 81
5.1 Sincere and strategic voting 81
5.2 The manipulation of the agenda 82
5.3 Condorcet’s paradox 83
5.4 Referendum with a minimum participation 83
5.5 The Borda count 84
5.6 Arrow’s theorem 86
5.7 The theorems by Gibbard–Satterthwaite and May 87
5.8 The median voter theorem 87
5.9 I’ll scratch your back and you’ll scratch mine 90
5.10 How to know the truth: The Groves–Clarke mechanism 91
5.11 Do we know what the people want? 94
5.12 The discursive dilemma 95
5.13 A referendum in Catalonia 96
Exercises99
6. Negotiation games 104

6.1 The model of offers and counteroffers 105
6.2 Impatience 107
6.3 Risk aversion 109
6.4 Negotiating with fanatics 110
6.5 Some discussion 112
6.6 An actual case: The hijacking of the Alakrana 113
6.7 The Coase theorem 114
6.8 When not to apply the Coase theorem 116
Exercises118
7. Repeated games 120

7.1 The Christmas truce 120
7.2 A game repeated twice 121
7.3 Cooperation in the infinite and indefinite repetitions 123
7.4 Some technical details 126
viii
CONTENTS
7.5 Other strategies in the repeated game 128

7.6 The cooperation in the prisoners’ dilemma repeated finitely
many times 130
7.7 What experiments say 130
7.8 What the empirical data say 132
7.9 Altruism, reciprocity and evolution 133
7.10 Not a zero-sum game 134
7.11 Axelrod’s tournament 134
Exercises136
8. Agency problems: Adverse selection 138

8.1 The agency problem 138
8.2 The information sets 140
8.3 If you didn’t have anything to hide you’d show me your e-mails 143
8.4 Adverse selection in a first agency problem 146
8.5 Adverse selection and public health systems 147
8.6 Other examples of adverse selection 151
8.7 Other types of adverse selection 152
8.8 Competition reveals information: When the principal has
information about the agent 154
8.9 On Rawls’ original position and the ex ante criterion 155
Exercises156
9. Agency problems: Signaling and moral hazard 159

9.1 Signaling with a discrete variable 159
9.2 The empirical evidence of education as a signal 163
9.3 Signaling with a continuous variable and discrimination in the
labor market 164
9.4 Moral hazard: Fixed payment or payment by performance? 166
9.5 Moral hazard: Co-payment, yes or no? 170
9.6 Moral hazard: Work in teams and cooperatives 174
Exercises175
10. Seven applications of game theory 177

10.1 The battle of the Bismarck Sea 177
10.2 The nuclear war 178
10.3 You cannot use information without revealing it 183
10.4 You should bluff from time to time 184
10.5 There may not be weapons of mass destruction:
Should we still attack? 187
10.6 Is free trade a prisoners’ dilemma? 192
10.7 Negotiations between Greece and the Troika 194
Exercises198
ix
CONTENTS
11. Seven more applications 200

11.1 The minority language 200
11.2 Pascal’s wager 204
11.3 The surprise exam paradox 206
11.4 The sentence as deterrence 208
11.5 Solidarity versus charity 213
11.6 Single-round versus runoff elections 215
11.7 How to eliminate illegal parking 218
Exercises220
12. Dynamics 223

12.1 Evolutionary dynamics: The hawk-dove game 223
12.2 Imitation dynamics: A segregation model 226
12.3 Best-reply dynamics: The emergence of language 228
12.4 No weakly dominated strategies dynamics: Self-inflicted injuries 232
12.5 Adaptive dynamics: Voluntary contribution to the provision
of public goods 235
Exercises237
13. Limited rationality and behavioral economics 240

13.1 Preferences changing with time: Which ones deserve priority? 240
13.2 Time inconsistency and energy saving 242
13.3 Irrationality due to the complexity of the election 244
13.4 Irrationality due to overconfidence 245
13.5 The age of majority 247
13.6 Indoctrination 253
13.7 Nudging: When to let others influence you 254
13.8 On other irrationalities that are not so irrational 255
13.9 Towards a behavioral theory 256
Exercises257
14. Power indices 260

14.1 Cooperative and majority games 260
14.2 Power indices in majority games 262
14.3 Application of power indices to three parliaments 265
14.4 Games with many quotas 270
14.5 The distribution of power in the EU after Brexit 272
14.6 Power indices with abstention 275
Exercises276
References278
Index282
x
LIST OF FIGURES AND TABLES
Figures
1.1 Coordination game 1
1.2 Choice of standards 2
1.3 Battle of the sexes 4
1.5a The prisoners’ dilemma 8
1.5b Prisoners’ dilemma with empathy 8
1.5c Prisoners’ dilemma with punishment 9
1.6a Matching pennies 13
1.6b Nadal-Federer game 13
1.7 The ultimatum game 15
2.1 D-day game 22
2.2a Tax war 24
2.2b Tax war after elimination of dominated strategies 25
2.3 Choosing a president we all like 28
2.4a Cleaning the street game 29
2.4b Best replies in “cleaning the street” 31
2.4c Cleaning the street after elimination of dominated strategies 31
2.5a Nadal-Federer game 32
2.5b Nadal-Federer game with an additional strategy 35
2.6a Expected payoff 37
2.6b Expected utility 38
3.1a Fiscal battle 42
3.1b Fiscal battle with high job profits 42
3.2 Electoral platform choice 43
3.3a CP prefers the LP candidate over the RP one 45
3.3b Game in Figure 3.3a after eliminating weakly dominated strategies 46
3.3c CP is indifferent between LP and RP candidates 47
3.3d Game in Figure 3.3c after eliminating weakly dominated strategies 47
3.4 Sylvanian referendum 51
3.5 Sylvanian civil disobedience 51
3.6 Keeping up with the Joneses 60
4.1a Dynamic battle of the sexes 63
4.1b Backwards induction in the dynamic battle of the sexes 64
4.2a Send the dossier game 65
4.2b Send the dossier game with a previous contract 65
4.3a The game of signing a contract 65
4.3b Solving the game of signing a contract 66
4.4 The normal form of the dynamic battle of the sexes 67
4.5 A subgame in the dynamic battle of the sexes 68
4.6 The normal form of the subgame in Figure 4.5 68
xi
4.7a Odysseus-sane vs. Odysseus-insane 69

4.7b Odysseus is tied to the mast 69
4.7c Odysseus and the Sirens game 70
4.8a The entry game 70
4.8b The entry game after a costly compromise 71
4.8c The game of acquiring a costly compromise 71
4.8d Solution after one equilibrium in the first subgame 72
4.8e Solution after the other equilibrium in the first subgame 72
4.9 Separation of powers game 73
4.10a The healthcare game: the government plays first 74
4.10b The healthcare game: the individual plays first 75
4.11a Payment of the debt game 76
4.11b Solving the payment of the debt game 77
5.1 Single-peaked preferences in a graph 89
5.2 I’ll scratch your back and you’ll scratch mine 90
5.3 The referendum game with preferences as in Barcelona 1 97
5.4 The referendum game with preferences as in Barcelona 2 98
6.1 A final offer 105
6.2 An offer and a counteroffer 106
6.3 A probability of interrupting negotiations 110
6.4 Fanatics’ preferences 111
6.5 Negotiating with fanatics 111
6.6a Coase theorem: the baker makes the offer 115
6.6b Coase theorem: the physician makes the offer 116
7.1 Fiscal battle 122
7.2 A typical prisoners’ dilemma 124
8.1 The battle of the sexes in extensive form 140
8.2a Prisoner 1 does not know whether Prisoner 2 is empathic 141
8.2b Prisoner 1 knows whether Prisoner 2 is empathic 142
8.3 The normal form of Figure 8.2a 143
8.4a If you had nothing to hide 144
8.4b Solving the game “If you had nothing to hide” 145
8.5 The normal form of the game “If you had nothing to hide” 145
8.6a Hiring a worker of unknown productivity 146
8.6b Solving the game “Hiring a worker of unknown productivity” 147
8.7 Adverse selection in the insurance market 150
9.1a Education as a signal 161
9.1b Solving the game of education as a signal 161
9.1c The normal form of the “education as a signal” game 162
9.2a Fixed payment 167
9.2b Payment by results 168
9.3 Co-payment 171
9.4a Working in teams 174
9.4b Working in a cooperative 174
10.1a The battle of the Bismarck Sea 177
10.1b The battle of the Bismarck Sea after the weather report 178
10.2a War as a zero-sum game 178
10.2b Army A develops nuclear weapons 179
10.2c Arms race 179
xii
10.2d Using the nuclear weapons first 179

10.2e The doomsday machine 180
10.2f Mutually assured destruction 180
10.2g Dynamic mutually assured destruction 181
10.2h Building the doomsday machine 181
10.3 U-boat war 183
10.4 Frequencies in the Nash equilibrium of the U-boat war 183
10.5 The poker game 185
10.6a The normal form of the poker game 185
10.6b Solving the poker game 186
10.6c The empty fort game 186
10.7 Preemptive war 187
10.8a Inspection game 188
10.8b Simplified inspection game 188
10.9a The normal form of the simplified inspection game 188
10.9b The inspection game when being inspected is too costly 189
10.10 Preemptive war with intelligence 191
10.11a The normal form of the preemptive war with intelligence services 191
10.11b Solving the preemptive war with intelligence services 191
10.12a Free trade vs. protectionism as a prisoners’ dilemma 192
10.12b Free trade vs. protectionism as seen by economists 192
10.12c Trade wars 193
10.13a Greece and the Troika 195
10.13b Greece and the Troika with more realistic payoffs 195
10.13c Greece and the Troika with reputation costs 196
10.13d Greece commits not to concede 196
10.13e Payoffs that would make Greece’s commitment credible 197
11.1 The minority language game 201
11.2a The normal form of the minority language game 203
11.2b The number of bilinguals increases 203
11.2c The frustration cost decreases 204
11.3a Pascal’s wager 205
11.3b God also plays in Pascal’s wager 205
11.4a Day 2 in the surprise exam paradox 207
11.4b Day 1 in the surprise exam paradox 207
11.5a The basic game to analyze deterrence 209
11.5b A more realistic deterrence game 212
11.6 The solidarity game 215
11.7 Plurality vote 216
12.1 The hawk-dove game 224
12.2 Preferences for diversity 227
12.3a Choice of standards 228
12.3b The communication game 229
12.3c The simplified communication game 229
12.3d The normal form of the simplified communication game 230
12.3e Best reply dynamics starting at (YR, RL) 230
12.3f Best reply dynamics starting at (YR, RR) 231
12.3g No way out of equilibria 231
12.4a The battle of the sexes 232
xiii
12.4b Player 1 burns a unit of money 232

12.4c The game of burning the money 233
12.4d The normal form of the “burning the money” game 233
12.4e Elimination of weakly dominated strategies: first iteration 234
12.4f Elimination of weakly dominated strategies: second iteration 234
12.4g Elimination of weakly dominated strategies: third iteration 234
12.4h Elimination of weakly dominated strategies: fourth iteration 234
12.4i Elimination of weakly dominated strategies: fifth and final iteration 235
12.5 Adaptive dynamics 236
12.6 Cleaning the street game 236
13.1 The saving-spending game 240
13.2 Learning to appreciate modern music 241
13.3 Commitment to save for old age 241
13.4 A complicated election 244
13.5a The child does not like education 248
13.5b Parents want to educate the child 249
13.5c The case for mandatory education 249
13.6a Preferences and decisions when facing a medical intervention 250
13.6b Parents make the choice 251
13.6c Parents use preferences of she-16 251
13.6d Parents use preferences of she-18 252
13.6e Parents choose Yes 252
13.6f Parents choose No 253
Tables
3.1 Market demand 53
3.2 Best replies for the Cournot game 54
3.3 Best replies for the Bertrand game 57
3.4 Utilities from different work-leisure choices 59
3.5 Utilities if the other consumer works eight hours 59
3.6 Utilities if the other consumer works x1 hours 60
5.1 Voting one of three projects 81
5.2a Friends’ preferences 85
5.2b Friends’ preferences with frog legs included 85
5.3a Single-peaked preferences 88
5.3b Pairwise vote 89
5.4 Do we know what the people want? 94
5.5 The discursive dilemma 96
5.6 Preferences for independence 97
5.7 Outcome depending on preferences and on referendum design 99
6.1a Solving the negotiation game 108
6.1b Player 2 is more impatient 109
8.1 Agency problems 139
10.1 Countries’ preferences 187
xiv
11.1 Parties’ preferences 216

14.1a Spanish Parliament after 2016 elections 265
14.1b Spanish Parliament after 2015 elections 265
14.1c Spanish Parliament after 2008 elections 266
14.1d Spanish Parliament after 2004 elections 266
14.1e Spanish Parliament after 1996 elections 266
14.2 British Parliament after 2017 elections 268
14.3a Provisional count for the 2016 Basque Parliament with quota at 38 269
14.3b Provisional count for the 2016 Basque Parliament with quota at 29 269
14.3c Final count for the 2016 Basque Parliament with quota at 38 270
14.3d Final count for the 2016 Basque Parliament with quota at 29 270
14.4 Brexit and the power indices in the EU 273
xv
ACKNOWLEDGEMENTS
My first game theory professor during my undergraduate years at the University of

the Basque Country was Federico Valenciano. After being exposed to it, I knew which
part of economics was my favourite. In the postgraduate courses prior to arriving in
the United States, I had Salvador Barberà not only as a professor but also as director
of my master’s thesis. When he said to me, “these are your topics,” I started to believe
that I could dedicate myself to the beautiful theory. Thanks to him, I was able to enter
the doctoral program in managerial economics and decision science at Northwestern
University. There, my teacher in the game theory course was Roger Myerson, who a
few months later would ask some of his students to proofread his book. I learnt not
only game theory, but, having read the manuscript of the chapter that I was to correct
again and again, I appreciated the richness of a concise, precise and intelligible way of
writing. I remember my anguish at not being able to respond with more than just a
small typo after his request to find errors and make suggestions. One year later, I was
the TA in the course taught by Itzhak Gilboa, where I was in fact another student, since
there was so much to learn. I worked on my doctoral thesis under the supervision of
Ehud Kalai, whom I believe I have not yet thanked enough for his dedication, support
and trust. Those were the years during which he was working on the project for an aca-
demic journal that was destined to become the reference in the discipline: Games and
Economics Behavior. In spite of this, he was able to take many hours out of his schedule
to attend to my questions as well as to encourage my research progress. Having pub-
lished in that journal has been one of the greatest satisfactions of my professional life.
This book is dedicated to all of them.
My friend and colleague, Carmelo Núñez, encouraged me to write this book and
put me in contact with Laura Pacey, a commissioning editor for Palgrave Macmillan
who liked the idea. She directed me to her senior publisher, Kirsty Reade, who believed
in the project from the very beginning. I also would like to thank the many anonymous
referees who convinced the board of editors to get the book published. When she left
Palgrave (now Red Globe Press), Jon Finch took the job of editing the book with the
same enthusiasm. Different chapters have gone through several examiners, including
my supportive wife, Lourdes. Finally, my first cousin once removed (I think that is the
name of my relation with him), Eric Irizarry, had the laborious work of correcting the
English text.
All have my gratitude.
Many people have participated in the building and application of Game Theory. In this
book I use examples from three sources: the folklore of the theory, my own invention,
and other writers. If one example is not properly credited, please contact the author for
a deserved recognition.
xvi
ABOUT THIS BOOK
The idea for a game theory textbook dedicated to social science and philosophy was
not born out of thin air. At Universidad Carlos III de Madrid, where I work, being
curious about their disciplines I would meet with a philosopher and a lawyer who in
turn were curious about game theory. We met in my office to discuss diverse topics
every Friday. A colleague put me in contact with Jesús Zamora, another philosopher
with whom I have published a couple of articles related to both philosophy and eco-
nomics. I ended up inviting him to collaborate in teaching a methodology course as
part of our undergraduate courses. I later took charge of that course myself and began
gathering up teaching material on the methods for social sciences. At the same time, he
invited me to attend the workshops he organized at his university, the UNED, as well
as to participate in the academic committee for international congresses in philosophy.
During one of those seminars I met Luc Bovens, yet another philosopher and also edi-
tor of Economics and Philosophy at the time. We published an article that consequently
led to me receiving continuous requests to evaluate manuscripts on paradoxes involv-
ing how to update beliefs in the face of new information. A modest exposure in social
networks thanks to my likewise modest blog, as well as my participation in a rational
and sceptical organization, introduced me to the task of using and explaining concepts
from game theory outside the sphere of economics. In recent years I have designed and
taught a course on game theory for political science in which this book germinated,
and I have collaborated with several philosophers and sociologists in the interdisci-
plinary seminar within the PPE degree. This book is my offering to the teaching in
social sciences and philosophy, to the effort of all the colleagues in those disciplines
who sometimes thought they learned something from me (I, of course, have learned a
lot from them), and to all the students who suffered my teaching. I hope they all find
some merit in this book.
The use of this book is no mystery. I advise the reader to put pen to paper to repro-
duce the operations, both mathematical and logical, that are carried out on its pages.
They are never complicated when taken in isolation; though once accumulated, a good
map is needed to follow them and understand the analysis. Technical subjects, such as
mathematics or economic theory, often deceive the student. The exercises seem easy
once resolved, but, when faced with the exercise and a blank piece of paper, the task of
finding the solution is usually far from trivial. This deception is more extreme in game
theory. The student faces not only the problem of figuring out what mathematical
operations need be performed, but also the particular and often counterintuitive logic
of game theory.
Chapters 1 through 4 present the basics of game theory, while Chapters 5 through
7 show their most direct applications. Chapters 8 and 9 are more advanced. Chapters
10 through 14 offer a miscellany of topics covering the application of the simple models
onto more complex case studies, an introduction to current issues within game theory
such as dynamics and limited rationality, as well as the study of power indices within
cooperative games.
xvii
ABOUT THIS BOOK
A basic undergraduate course of 40 hours can cover Chapters 1 through 4 quite

exhaustively and then incorporate the less specialized parts of Chapters 5 through 7.
The course can be completed with some material from the remaining chapters accord-
ing to particular interest. For example, if the course is oriented towards political s cience,
Chapter 14 will be a good option. A course oriented towards other social sciences or
philosophy will find cases relevant to its discipline in Chapters 10 through 13. An
intermediate or advanced course can cover Chapters 7, 8, 12, 13 and some of the case
studies in Chapters 10 and 11, as well as the more specialized parts of the rest of the
chapters. Chapters 10 and 11 are especially recommended for the preparation of an
undergraduate or a master’s thesis that uses the tools of game theory.
A solutions manual is available to lecturers on the companion website at macmillanihe.

com/Ferreira-GameTheory.
xviii
INTRODUCTION
A breakdown in the network has left the city of Gotham without traffic lights and
the police cannot cope with the traffic. Chaos reigns at a particularly busy inter-
section. In the absence of competent authority, a pedestrian walks to the center
of the confusion and begins giving orders. She stops the traffic coming in one
direction while allowing that from another to pass. She then switches the direc-
tions and also gives orders to pedestrians. She is not a policewoman and has no
authority; no one will be fined if they disobey her orders, and yet everyone pays
attention to her. Is it because deep down we are good people? Is it because people
will look down on us if we do not obey? Or is it because drivers and pedestrians
fear that she is a policewoman? Actually, you do not have to assume any of that.
Each driver and pedestrian thinks that others will follow the orders. If this is the
case, it is in everyone’s interest to stop when the traffic coming from the other
direction is passing and to wait until they are instructed to proceed in order to
avoid an accident.
The object of the theory of games can be summarized very briefly in two words: stra-
tegic decision. What differentiates the strategic decision from the simple decision is the
interaction with other decision makers. When I decide on what to buy at the supermar-
ket I do not have to consider how other people will buy or sell. I am too small a part of
the economy for my actions to have any influence on fashion or prices. It is true that
both things depend on what we all collectively do as a society, but as far as my decision
is concerned, the only thing that interests me is the more immediate environment in
which I find myself, with prevailing prices and fashions. Given the environment, I will
make my decision. In regards to strategic decisions I am not insignificant. My actions
have a substantial influence on the environment and I must take into account how
others will react to them. Let’s look at some examples:
1. The different parties in the Italian Parliament must decide on a prime minister
among different candidates.
2. Every country surrounding the Caspian Sea must make a decision on how to reduce
the pollution that goes into it.
3. A Welshman must decide on which language, either Welsh or English, to speak
when addressing a stranger.
4. The OPEC countries must agree on production quotas and must later individually
decide on whether or not to honor that commitment.
5. Airbus and Boeing must decide their commercial policies.
6. Palestinians and Israelis must decide what actions to undertake when there is an
escalation of hostilities, as well as when there are peace talks.
xix
INTRODUCTION
In the introductory example of the traffic chaos in the city of Gotham there seems to
be many players; the game would be identical, however, even if there were just a few.
A single player can cause an accident, which will affect all drivers by hindering the
traffic through that intersection. In fact, a simple coordination game consisting of
only two players, such as the very first one we will be studying in the next chapter, will
contain the most relevant elements for its analysis.
In each of the previous examples, each decision maker must have an idea of what
the rest are doing or thinking about doing, while all others are in that same position.
For example, when it comes to setting a pricing policy and deciding on all kinds of
clauses for the sale of its airplanes, Airbus would want to know what Boeing’s plans
are; in turn, Boeing’s plans depend on what it believes Airbus to be doing. The reader
can elaborate upon this type of argument in each of the other examples. At first glance,
the problem of strategic decision making is that of a circular argument: A needs to
know what B does in order to make his best decision, but B needs to know what A
does in order to make hers. One of the merits of game theory is showing that, in many
cases, the circularity of the argument is only apparent. To achieve this, the theory will
describe each situation by means of a mathematical object called a game in which the
players make decisions, with the characteristic property that they are making the best
choice for themselves given the decisions that the others make. In our example, A and
B would each be making a decision so that A’s decision is the best he can do given B’s
decision, while at the same time B’s decision is the best she can do given the decision
of A.
It is important to note four details:
1. The evaluation of what “best” means is something that must be determined and
which will depend upon the behavior attributed to the players. The attribution of
a rational behavior will imply that “best” means taking the option that maximizes
the player’s well-being, though other behaviors are possible. For example, calcu-
lations for finding the best action can be too complicated and the player may be
content with finding the best within a model of limited rationality, or in which the
calculations do not exceed a certain level of complexity.
2. The fact that each player does what is best for them given what others do does not
guarantee that the result is good in any material or moral sense.
3. It is possible that there is more than one equilibrium situation and it is also possible
that there is none.
4. In principle, detecting an equilibrium situation does not mean making a recom-
mendation or a prediction; it is simply the first step in the analysis of the game.
We could say that the accurate analysis of a game gravitates around equilibria
and that, in this sense, their calculation is of paramount importance. Sometimes
we will accept the equilibrium strategy as a good recommendation or as a good
description of a game situation, though sometimes the equilibrium analysis will
bring us back to reality so as to change or to add details in the specification of the
game.
As the reader delves into the book, I hope that these four details become
entrenched. Forgetting them will result in many misunderstandings in the study
xx
INTRODUCTION
of game theory. For the time being, a couple of additional examples can help to
avoid some of them.
A bird finds some seeds and is preparing to eat them when another bird of the
same species and of similar size appears with the intention of stealing them. The
newcomer threatens to attack but avoids fighting as soon as the other displays
his intentions of defending what is his. Why does this happen? Both birds have
the same chances of getting hold of the seeds, but one of them concedes. Why
does the second bird do it? In fact, if the following day it is this second bird that
finds some food and it is the first one that approaches with the intention of
disputing it, the roles will change and the newcomer will pretend to fight only
to back out at the first sign of a real fight. It is easy to understand that a fight
between equals is costly and that a threat is only useful for testing the strength of
the other, resulting in an actual fight only in the case that a weakness becomes
apparent. This argument, however, only indicates that avoiding the fight is a sen-
sible course of action; it does not explain why the newcomer gives in. To resolve
the issue we must venture a little further. What consequences would it have for
the interactions of birds of this species if the conflict were resolved one way or
another? If the newcomer concedes, then the problem is over, but if the owner
of the resource in question is the one to give in, then the other bird immediately
becomes the owner and, according to the coordination rule, it should concede to
the other. This would place the birds in a loop situation without clear resolution.
In this example, birds do not act as rational beings that consciously perform some form
of calculations. They follow their strategies instinctively, according to their genetic
condition. In this case, the strategy is beneficial for each individual because it allows
them to avoid the use of force in situations where the expected costs are not compen-
sated by the possible reward. It is not always like this. There are many behaviors in the
animal kingdom in which an individual sacrifices himself for others – most frequently
others who are genetically related.
Humans and Vulcans (inhabitants of the planet Vulcan) are allies and have
decided upon founding a joint colony on a newly discovered planet. Their desire
is for this city to be made up of an equal number of Humans and Vulcans. The
settlers arrive attracted by the idea of a diverse society and all of them want to
have neighborhoods with 50% Humans and 50% Vulcans in their ideal world.
Over time, however, the city becomes segregated into Human and Vulcan neigh-
borhoods. What has happened? Despite the preferences towards diversity, the
city authorities have not taken into account that both Humans and Vulcans also
prefer not to be a minority. For example, if by chance Humans make up 40% of
a given neighborhood, some of them will want to be able to move into another
neighborhood where there are not so many Vulcans. A neighborhood with a
50–50 split between Humans and Vulcans would be ideal, but a neighborhood in
which 60% of the inhabitants are Humans will also work better for these Humans
willing to move. A neighborhood with 75% Humans will be worse, since diversity
is also valued. While some move for this reason, those who remain are left in an
xxi
INTRODUCTION
even smaller minority and will be willing to move into even less diverse neighbor-
hoods where Humans are not a minority. At the same time, neighborhoods with
a more balanced population will be more difficult to find. The result is a level of
segregation that nobody wants.
In this last example, the behavior of the inhabitants of the colony is rational but
myopic. Each inhabitant only looks at the short-term consequences of their actions,
without taking into account that these actions (along with the actions of others) will
trigger new actions on the part of other inhabitants. However, even if they had taken
the long-term consequences into account and had anticipated segregation as a result of
everyone’s actions, a single individual would have been able to do very little to avoid
it. Whether or not the others move into another neighborhood, each inhabitant will
be better off if she does. Only a coordination among all of the neighbors, as well as a
verifiable commitment that this coordination will be fulfilled, might prevent the segre-
gation of the colony, though this coordination will not happen spontaneously as it did
in the first example about the city of Gotham. Good intentions are not sufficient for
reaching desirable situations. The theory of games illustrates this fact in many situa-
tions and provides elements of analysis so as to make a correct diagnosis of the problem
and advance towards a possible solution.
xxii
1 THE MOST FAMOUS GAMES
Many strategic interactions one may encounter in any discipline can be seen as varia-
tions of a few basic games. It is then convenient to know them by their name and to
understand the questions they address.
Strategic interactions typically involve situations with room for both common inter-
ests and conflict among the various players. Our first examples will start with a game of
pure common interest in Section 1.1 and end up with some of pure conflict in Sections
1.6 and 1.7; in the middle, we will show different balances between these two aspects.
Along the way, we will also introduce a few other ideas, like the role of the model and
its interaction with reality, the implications of having multiplicity of equilibria, and the
assumptions of selfishness and rationality. All will be extensively discussed throughout
the rest of the book.
1.1 The coordination game

Let us begin with the simplest of games. It is so simple that the reader may say that we
do not need a theory to talk about it. That being true, the important thing is not what
the theory says in this particular example, but the fact that with its help we can intro-
duce some of the concepts that will be useful in more complicated cases.
This is the game. Lisa and Bart each rent an apartment in the same building. Both
apartments share a garage with a door wide enough to accommodate one car entering
and the other exiting at the same time. There is a strategic decision to make, though,
as every time one enters the garage they must decide which side to use to make room
for the other neighbor in case he or she is exiting at the same time. If they both drive
on their right or both drive on their left there will be no problem, but if they drive on
different sides they will have to maneuver to avoid collision when they use the door at
the same time. The situation can be represented as seen in Figure 1.1.
Bart
Left Right
Left 1, 1 0, 0
Lisa
Right 0, 0 1, 1
Figure 1.1 Coordination game
The way to understand the information in the figure is as follows. Lisa must choose
between “Left” and “Right,” which are the two rows of the table. Bart does the same
with the columns. If both adopt the same driving convention they both will be better off
compared to the case in which they do not. The first number in each cell is the payoff
for Lisa, and the second number is the payoff for Bart. We have assigned a value of 1 to
the outcome in which both coordinate in the same rule, and 0 to the case where each
1
GAME THEORY
one drives on a different side. The only thing that matters is that 1 is greater than 0,
indicating that coordinating is better than not coordinating. We could have written any
other numbers, like 100 and −16, and the game would have been the same. Also, we are
saying that both ways of being coordinated are equally good, and both ways of being
uncoordinated are equally bad.
If both Lisa and Bart could talk to each other, it is reasonable to assume that they
would easily reach an agreement. Neither the details of the story nor the analysis of the
game allow us to say which agreement that would be, beyond that they will decide on
one of the two possible ways of coordinating. We can safely say that they will not agree
on “you use your right and I will use my left” or “every time we coincide at the garage
door we flip a coin to decide whether to use the right or the left side.”
But say that they cannot talk to each other. For instance, they just moved in the day
before, never met, and today they have to use the garage. Then, there is not much that we
can say. The first time they meet at the garage door, either one of the four combinations
may occur (Left-Left, Left-Right, Right-Left, Right-Right). However, not all the possibil-
ities have the same properties. If, for whatever reason, they drove on their right, we have a
situation in which none of them regrets their choice, as they managed to coordinate. Even
if there is no communication between them in the future, it will not come as a surprise
to see them both driving on their right from there after. A similar thing will happen if
the first time they meet they drive on their left, but not if they drive on different sides. In
this latter case we do not know what will happen in the future, but we can say that going
uncoordinated is an unstable situation. Perhaps only Bart will change his choice of side
and they will both be coordinated in the future, or perhaps no one will change and they
will be uncoordinated also the next time they meet. It may even happen that they both
change and will be annoyed to see they are still uncoordinated the next time they meet.
The important thing to understand is that we can detect two stable situations in this
game and the analysis of the game goes around them, regardless of what could happen.
The previous two paragraphs show two ways to reach the coordination, but we can
say more. The analyst studying the game will find that there are two stable situations
and then will be able to start a research line suggested by this multiplicity in the the-
oretical game. For instance, he can gather information about whether this building is
located in a left or right driving country and propose as an hypothesis that the strategic
choice of a side to enter or exit the garage will be solved using the same rule.
1.2 Choice of standards
Our next game is a small complication of the coordination game: now, our characters
Lisa and Bart enjoy a higher payoff if they coordinate by driving on the right rather
than on the left (e.g., they have cars with the steering wheel on the left). In this case,
the game changes to be like the one in Figure 1.2.
Bart
Left Right
Left 1, 1 0, 0
Lisa
Right 0, 0 2, 2
Figure 1.2 Choice of standards
2
THE MOST FAMOUS GAMES
Again, the numbers 0, 1 and 2 only indicate the ranking of preferences over the
different combinations. It is better to coordinate in the preferred rule than in the other
one, and any coordination is preferred to being uncoordinated.
If, as we did in the original coordination game, we try to anticipate what the neigh-
bors will do if they talk to each other before choosing a strategy, we will not only
anticipate the agreement in one of the sides, but we will be able to be more specific and
say that it will be the right side. The fact that both of them agree on the ranking of the
options will help us in the prediction.
However, if the neighbors cannot communicate, several things may happen. One is
that they are able to identify the game and reason that since the best option for them is
to drive on the right, that is what each one should do when deciding individually, hop-
ing that the neighbor reasons the same way. This is a stable situation. As soon as they
cross into each other the first time, and see that their expectations are confirmed, no
one has an incentive to change their choice. Notice that, for this situation to happen,
both players need to know that the other player also prefers that they coordinate on the
right (for instance, they are aware of the type of car the other player owns after seeing
it parked in the garage). Unless we specify otherwise, in the rest of the book we will
assume that the players know the game they are playing.
To be sure, the result above is the most reasonable thing to expect given the way the
game was described, but it is not the only possibility. If the players have not analyzed
the game and their first coordination was on the left, that would also be a stable situ-
ation. Given that we began coordinating on the left, should I change next time once I
realize that right is a better option? How do I know that my neighbor has also noticed
this? Even if he noticed, how do I know that he will change, trusting that I change
too? When do we change the rule? These difficulties are easily resolved if Lisa and Bart
talk to each other, but if they cannot do it, they may find themselves in a non-optimal
situation from which it may not be so easy to leave.
The coordination in driving on the right side to enter and exit the garage may not
seem a big problem, but in the real word there are situations that share the same strate-
gic structure and that definitely are a big problem. For instance, imagine that different
users must choose one out of two computer systems that are incompatible with each
other, making it difficult to share files or programs. Even if the users agree that one of
them is better, if everyone else is using the other system it may be rational to buy the
worst one to avoid the problem of incompatibility. To change driving sides is a trivial
matter in the case of the neighbors, but it is quite costly in the case of a whole country.
In 1967 Sweden did just that to be on the same side as the rest of continental Europe
and to favor its automotive industry, that until that time had to manufacture two
versions of each car. They did it before cars were as numerous as today. If the B ritish
wanted to change their driving side today, the cost would be much greater. Other
examples of inefficient coordination include the choice of technology for the develop-
ment of colour TV, the different standards for video tapes, and different choices for
office automation. The coordination of a group of friends or members of a family in a
social network is a modern example.
There is a case of inefficient coordination that has some special interest. Say that
a country has both a right-wing party (RP) and a left-wing party (LP). Lately, the RP
has been involved in many corruption cases, and some citizens decide to start a new,
corruption-free right-wing party (CP). Now, all right-wing voters agree that CP is a
3
GAME THEORY
better option than RP. Will they all vote for it? If they prefer any right-wing party to
a left-wing one, they may not want to risk dividing the vote between RP and CP and,
thus, facilitating the possibility of a left-wing government. This may be enough for
many voters to continue voting the RP. In the real world, not all voters will have the
same motivations. Some right-wing voters may prefer a left-wing government to a cor-
rupt right-wing one, or may be willing to risk losing the elections to build support for
CP. What our analysis says is that, in addition to all these considerations, one should
not disregard the coordination issues, which allow a more complete diagnosis of the
political scene beyond just saying that the voters of the RP are clueless or just do not
mind the corruption.
1.3 The battle of the sexes

In the previous games there was no disagreement about the best way to coordinate
actions. Let us change this in our new game. Sylvie and Bruno are a couple in the time
before cell phones, and they go out every Friday evening. Today is such a day, and yes-
terday they talked about two alternative plans, going to the movies or to the football
game, but did not say anything definitive. This morning they left to their respective
jobs and now are unable to communicate during the day. After work each of them
has to decide whether to go to the movie theater or to the stadium. They prefer going
together to the same place, but Sylvie likes football better, while Bruno prefers movies.
If they mismatch and go to different places, their evening is ruined regardless of where
they go. Figure 1.3 shows the game.
Bruno
Football Movies
Football 2, 1 0, 0
Sylvie
Movies 0, 0 1, 2
Figure 1.3 Battle of the sexes
Like in the previous examples, there are two stable situations that, from now on,
will be called equilibria: they both go to see the football match or they go to the movie
theater. In either case, there will be no regrets. For instance, if they met in the movie
theater, Sylvie will be satisfied with her decision, as it allowed her to meet Bruno.
Of course, she prefers to watch the football game, but had she gone to the stadium she
would have been alone. Recall that Sylvie can only decide for herself, and she cannot
choose the equilibrium because she cannot decide for Bruno. We have a similar situa-
tion in case they went to the football game. We do not know which equilibrium, if any,
will prevail. Contrary to the other games we have seen so far, talking may not solve the
problem. We may anticipate that they will agree on one plan, but we cannot forget that
besides the coordination problem there is a conflict that was absent in the coordination
or in the choice of standards games.
The multiplicity of equilibria may be telling us that our game model is incom-
plete and that we might want to obtain more relevant data from the real-life situation.
The game can give us some hints about what to look for; we must find details which we
did not take into consideration at first glance, and that may determine the equilibrium
4
selection. We can check whether Sylvie and Bruno live in a patriarchal society where
women always do as men say or whether they live in a patriarchal society where women
are treated as girls that need to be pleased. If the society is patriarchal of the first type,
they will go to the movies; if it is patriarchal of the second type, they will go to the foot-
ball game. There are other possible solutions. If the couple lives in an egalitarian society,
or if this couple in particular is that way and alternates plans every week to satisfy the
preferences of both of them, and, further, if we know that the previous week they went
to Bruno’s most preferred place, now we can safely predict that they will go to Sylvie’s.
It is important to distinguish between the model and the real-life situation.
The game model in Figure 1.3 has two equilibria, and we cannot put one above the
other. The real world, as we saw in the previous paragraph, may be more compli-
cated and at the same time may give us clues about how to resolve the equilibrium
selection. The analysis should not end here; if there are some features of the relation
between Sylvie and Bruno that may be relevant to their strategic decision, they should
be included in the description of the game. Doing so, we will end up with a more com-
plicated game compared to the one in Figure 1.3, and as such it should be analyzed.
In the rest of the book we will see how to add more details and how to analyze them.
The players themselves may learn something from the analysis. The problem of
equilibrium selection has arisen because the decisions had to be made without the
knowledge of what the other player was doing. If one of them, say Sylvie, can decide
first where to go, she can get her way. If she already went to the stadium, or has made a
credible commitment to go, and Bruno knows that, he would have no choice but going
to the stadium as well. To accept this line of reasoning, it has to be the case that what
happens today will not affect the future. For instance, it cannot be the case that Bruno
wants to gain a reputation of not allowing Sylvie to win with this kind of stratagem.
This last consideration opens the door to more possibilities. If Sylvie can send a one-way
message to Bruno, she will say that she is going to the stadium. A phone call may work:
-Darling, we’ ll meet at the stadium gates.

-What? No. I prefer to go to the movies.
-Sorry, I can’t hear you. It seems we have a bad connection. Bring scarves and sand-
wiches. See you!
-What? No! I want the movies! Can you hear? Hello?
-…
1.4 The chicken game

In the Nicholas Ray film Rebel without a Cause, with James Dean playing the main
character, a group of bored, young people, not knowing what to do with their lives,
have an interesting way of having fun: a car race. This is not any car race; two drivers
take their respective stolen cars and speed up towards a cliff (that is why they needed
stolen cars). The first one to jump out of the car loses this “chicken game.” The last to
jump will be the winner who, full of testosterone and adrenaline, will enjoy a higher
status within the group. In the movie, one of the boys, Buzz, challenges Jim, the new-
comer played by James Dean, to play the game.
5
GAME THEORY
To make the game a little more interesting, let us change the rule of driving towards
a cliff with the rule of driving against each other on a collision course. Now, the first one
swerving is the chicken, and the one keeping his way will be the winner. If they both swerve
they are both chickens, but the stigma will not be as high as in the case one is the only
coward. The worst-case scenario occurs if they both keep going and crash into each other.
They would have been very brave, but also lost their lives. Figure 1.4 summarizes the game.
Buzz
Swerve Keep going
Swerve 2, 2 1, 4
Jim
Keep going 4, 1 0, 0
Figure 1.4 The chicken game
The face-saving outcome, where both drivers swerve, is not an equilibrium. If, for
whatever reason, they agree on doing that, any one will have an incentive not to swerve.
If Jim follows the agreement, Buzz would be better off by not honoring it. The agree-
ment is not stable. In other words, the assumption that the agreement is an option
that the players will be willing to follow leads to the contradiction that they will want
to do otherwise. One can object that if Jim thinks as Buzz does, and both break the
agreement, they will be both worse off. That is true, but this only illustrates that the
agreement is very unstable, and that the outcome after the agreement is unpredictable.
There are two equilibria, two stable situations from which no one wants to deviate:
one swerves and the other one keeps going. As we saw in the previous games, to know
which one will swerve we better take a closer look at the group. If one of the two drivers
already has a reputation of never swerving, it will be difficult for the other one to win
the game. Again, remember that a player only chooses his strategy, not the equilibrium.
In the film, Buzz has this reputation, while Jim’s reputation is unknown to everyone in
the group. In the film, as they speed toward the cliff, Buzz’s jacket gets caught on the
door’s handle, and he cannot jump out of the car. We see Jim jumping out, but we do
not know what would have happened if Buzz could have jumped.
As in the battle of the sexes, one of the players could try committing to a strategy,
but how?
-Darling, you better swerve.

-??!!
No. That is not going to work. We need something more dramatic, an action that
shows a real commitment not to swerve or one that shows that the payoffs in the table
are different. Here are two possible strategies: (i) arrive at the racing place conspicu-
ously drunk and (ii) in the middle of the race, pull out the steering wheel and throw
it out the window, also conspicuously. The key to the strategy is, it was easy to notice,
the word “conspicuously.” This is an example of how being, or pretending to be, irra-
tional puts a player in a winning situation, leading to the paradox of calling irrational
a behavior that makes you win.
Like in the other games, the equilibria in the chicken game can be viewed as a pair
of fulfilled beliefs. For instance, in the equilibrium (Swerve, Keep going), Jim swerves
6
because he believes that Buzz will keep going, and Buzz keeps going because he believes
Jim is going to swerve. As we argued before, why this equilibrium is selected and, there-
fore, why these beliefs prevail are questions that cannot be answered with the level of
detail of this particular mathematical game model as a representation of the real-life sit-
uation (although a more complicated game may incorporate more details that explain
the selection). A system of compatible beliefs can help to select an equilibrium, but the
source of those beliefs may be different at different moments. When players observe
the actions in the equilibrium they get confirmation of their beliefs that, at this point,
may become independent of the reason they had those beliefs to begin with. This is an
important point if one wants to understand some behaviors.
For instance, a young couple decides to live together and he has the expectation that
she will do most of the housekeeping. It may not be something she wants, but if she
knows that she is expected to do that, it will be very hard to fight against the assumption
of the classical roles. Say the expectations came from a social norm in the society where
they live and that later on they understand that they do not need to follow the norm. But
now, given that they are established in the equilibrium, neither of them continues in their
roles because of the social norm, or at least not only because of that. They have a strong
incentive to follow the equilibrium because their belief that she will do the housekeeping,
and that he will not, is confirmed by the actions of both of them and not dictated exter-
nally by a norm. She does the housekeeping because he does not, and he does not do it
because she does. This does not mean that they can do nothing to change it, it means that
they have to do more than just realizing that they do not need to obey the social norm.
The chicken game was made famous by many analysts at the time of the negotia-
tions between the Greek government and the Troika during the financial bailouts in
2015. While Greece demanded soft conditions to accept the bailout, the European
institutions and the IMF wanted to impose stronger ones. Each party thought that its
conditions were the best for Greece to overcome the crisis and pay the loans back. For
the strategic analysis, what matters is not who is right, but what each party thinks it
can win in the different scenarios. The strategies are “concede” or “hold.” The worst-
case scenario is that no one concedes, which constitutes a disaster for Greece (leave the
euro, devaluation, bankruptcy, etc.) and also for the EU (a backing out of the common
currency, disaffection for the European project, etc.). Each party thinks that the best
outcome is that the other one concedes. If both concede, there will be an intermediate
agreement that does not fully satisfy anyone but that is better than the lack of an agree-
ment. In Chapter 10 we analyze this game in detail.
1.5 The prisoners’ dilemma

Two suspects are arrested and accused of committing a crime. The police put them in
separate cells and tell them the following:
We know you committed the crime but we don’t have enough on you for a con-
viction. You know that too. But if you confess and your partner does not, your
partner takes all the blame (5 years in prison) and you go free for helping us. If
you both confess, you share the punishment and serve 4 years each. If no one
confesses, we will make sure that you still go to jail for 1 year, accused of a minor
crime (illegal possession of firearms, for instance).
7
GAME THEORY
This game might be better understood with the help of Figure 1.5a, which represents
the number of years to be spent in prison. Such an unpleasant prospect is prefaced with
a minus sign, with zero being greater than −1, indicating that zero years in prison is
better than one year.
2
Do not confess Confess
Do not confess −1, −1 −5, 0
1
Confess 0, −5 −4, −4
Figure 1.5a The prisoners’ dilemma
Now the prisoners must decide what to do. If they could talk, reach an agreement
and then make a joint decision, they would almost surely decide not to confess. But the
catch here is that they must decide individually, and each one will see that, no matter
what the accomplice does, it is best to confess: “If my partner confesses, I better confess
myself (four years in jail are better than five), and if my partner does not confess, I still
better confess (going free is better than one year in jail).” This way they both confess
and enjoy four years in the slammer. This is a stable situation, as no one can do any-
thing to improve his situation given what the other prisoner is doing.
When one is exposed to the prisoners’ dilemma for the first time, the natural reac-
tion is to reject the conclusion. How can it be that, knowing the consequences of
confessing, the two prisoners are unable to take an action that substantially reduces
their sentence time? Very often the reason is that we forget that the choice is unilateral
and individual; it is taken with no attachments to what the other person is doing. At
other times we might tend to think that the game ends differently. For instance, the
two prisoners may be friends that care about each other. This would mean that the
numbers in Figure 1.5a do not adequately reflect the consequences of the prisoners’
actions. Suppose that each prisoner suffers when his friend is in prison. Specifically, say
that every year spent by one prisoner in jail causes remorse in his friend equivalent to
half a year of jail. In this case, the numbers would be similar to those in Figure 1.5b.
For example, if no one confesses, each suspect not only suffers his own year of pain but
also the equivalent of another half year caused by remorse at his partner’s year in jail.
2
Do not confess −1.5, −1.5 −5, −2.5
1
Confess −2.5, −5 −6, −6
Figure 1.5b Prisoners’ dilemma with empathy
Now the only equilibrium situation is that no one confesses. If one of the prisoners
deviates from this situation, he does go free but his friend’s five years in jail will cause
this prisoner pain equivalent to two and a half years in jail, which is more pain than the
one and a half years if no one confesses.
There are other ways out of the prisoners’ dilemma. Even if the prisoners are
indifferent to the suffering of their accomplices, it may happen that they are
8
members of a criminal gang that punishes traitors with death. If one of the prisoners
betrays the other, his payoff will be zero years in jail plus certain death, a grimmer
prospect than five years in prison. The game would be something similar to the one
in Figure 1.5c.
2
Do not confess −1, −1 −5, −100
1
Confess −100, −5 −100, −100
Figure 1.5c Prisoners’ dilemma with punishment
The number −100 indicates something very bad (death). The choice of this particu-
lar number is not important; it is enough that it is worse than −5, the number corre-
sponding to five years in jail. Again, the option “Do not confess” is now more attractive.
There are many more ways to escape the logic of the prisoners’ dilemma, and one
may be tempted to think that any one of them would always be able to prevail. The fact
is that in real life there are indeed situations that may be represented by the game in
Figure 1.5a and which offer no easy way out or, at least, one that has not yet occurred.
The problem of pollution is just one of these situations: we are all better off if we
pollute less, but if I go green, my action will be hardly noticed and I will be paying a
cost. No matter what others do, it is better for me to avoid this cost and keep polluting.
The result is that we contaminate too much, which is something that we definitely see
in the real world. An advertising campaign to make people conscious of the problem
might reduce some types of contamination – the ones that are cheap to solve – but in
general it would not achieve optimal results. For this to occur, the polluter would also
need to pay the cost of polluting. Penalties, taxes per unit of contamination or markets
for emission permits are different ways to face the problem. In some cases, such as
when we want to have a clean communal park, both social pressure and education may
be enough; however, if we are talking about industrial pollution, transport emissions
or agricultural contamination, things are very different. When we face a prisoners’
dilemma, it is better to understand, rather than ignore, its logic. A correct diagnosis
will help to solve the problem better and in a shorter time.
Other real-life examples with the structure of a prisoners’ dilemma game are volun-
tary contributions to financing public goods, environmental degradation, the depletion
of natural resources, greenhouse gas emissions, traffic congestion in the inner city,
noise pollution, basic research and cartel formation, among many others. One example
dearest to the author of this book is the opera Tosca, by Puccini, which we explain here.
We are in Italy at the end of the 18th century when the Royalists and Republicans
are at war. The troops of the Kingdom of Naples occupy Rome, where both Cavara-
dossi, the church painter, and his beloved Tosca live. As always in opera, things are
fine for the two lovers but circumstances plot against them. In this case, Cavaradossi
is arrested after helping to hide a Republican sympathizer and, for this, he will be shot
the next morning. The chief of police, the evil Scarpia, calls Tosca to his quarters and
tells her about the situation. He can give the order to replace the real bullets with fake
ones to save Cavaradossi, but in return Tosca has to spend the night with him. Tosca
9
GAME THEORY
accepts and Scarpia leaves the room to give the order. While he is out, Tosca sees a knife
on the table and starts thinking:
-If Scarpia has indeed given the order, I can kill him in case he changes his mind and
also to avoid being with him. If he did not give the order I will be avenging my lover.
At this time, the reader may already guess Scarpia’s thoughts:
-If Tosca has already accepted my conditions, there is no point on me giving the order
and risking being caught by my superiors. She will know nothing until everything is over.
The terms of the agreement are reasonably good for both of them. Tosca keeps her lover
alive at the price of spending one terrible evening, while Scarpia enjoys a good time at
the price of saving Cavaradossi. Of course, Tosca prefers not to be with Scarpia (for
that she has to kill him) and Scarpia prefers not to get into trouble with his superiors
(and for that reason he has to let Cavaradossi die). But if they follow their rationality,
they both will face a bad outcome: the first one living without her lover, the second one
being killed. It is not my intention to spoil the opera for the reader. Suffice it to say that
the plot goes according to the logic of the prisoners’ dilemma.
Many smart people have had trouble in their first analysis of the prisoners’ dilemma
and have tried to find a better logic. Perhaps the following paragraphs by cognitive
science researcher Douglas Hofstadter in his book Metamagical Themas (Hofstadter,
1985) are the best illustration of this (to his honor, he later changed his mind after he
understood the game)1:
Now, if reasoning dictates an answer, then everyone should independently

come to that answer. . . . Once you realize this fact, then it dawns on you that
either all rational players will choose D or all rational players will choose C.
This is the crux. . . .
All you need ask now is, “Since we are all going to submit the same letter, which
one would be more logical? That is, which world is better for the individual
rational thinker: one with all C’s or one with all D’s?” The answer is immediate:
“I get $57 if we all cooperate, $19 if we all defect. Clearly I prefer $57, hence
cooperating is preferred by this particular rational thinker. Since I am typical,
cooperating must be preferred by all rational thinkers. So I’ll cooperate.”
All these attempts to convince us, through seemingly logical and rational arguments,
could go on for a long time and yet they would still miss the point. To deduce individual
incentives from what would be a good deal is a logical fallacy. If they are all in the same
situation, each player will arrive at the end of his or her reasoning process to the same
conclusions as the others – but only at the end of this process. The symmetry does not
need to be carried through the tentative and counterfactual arguments along the deduc-
tive process. During the process of reasoning, one prisoner contemplates his best course
1 Note that Hofstadter was discussing a prisoners’ dilemma with money rather than years in jail, and
with more than two players. Also, he uses C for cooperate and D for defeat. Just use “one year” and
“four years” instead of the number of dollars that Hofstadter used and replace “all” with “both” to
translate his version into ours.
10
of action for every possible thing the other may be doing. For the mathematically inclined
reader, in Section 3.6 we conduct a more formal discussion of these symmetry issues. The
equilibrium postulates one prisoner’s own best response to the other’s actions – the key to
the equilibrium concept that we will be using extensively in game theory. We will provide
a precise definition later and name it in honor of John Nash, who was awarded the Nobel
Prize in Economics and the Abel Medal in Mathematics (the highest distinctions in the
respective fields) and who proposed and investigated this equilibrium.
Being aware of this logical fallacy has implications for how we examine games.
These may be summed up in the following hypothetical dialogue:
-If everyone else chooses not to cooperate, I better not cooperate. In fact, if the others do
cooperate, it is also best for me if I do not cooperate.
-True, but if we all cooperate we will be better off compared to the situation in which
no one does. If we all think like you do we will be worse off.
-Yes, but the case is that if every one thinks like I do, I am not going to be the only fool
that does not think like myself.
The Golden Rule in ethics states that “one should treat others as one would like others
to treat oneself.” Regrettably, the rule may not be an equilibrium situation, and individ-
uals may have reasons not to follow it. Improvements of the rule, like Kant’s categorical
imperative to “act only according to that maxim whereby you can at the same time will
that it should become a universal law,”2 follow the same fate, even when individuals
have the same preferences and agree on what it means to be well treated or on what
should become a universal law.
Because of analysis such as this, game theory (and economics) is sometimes accused
of defending and promoting selfishness. However, notice that our analysis does not say
anything about the ethics of the equilibrium (or of the Golden Rule, for that matter). It
just says that in the prisoners’ dilemma the actions “Defeat” constitute an equilibrium
and the actions “Cooperate” do not, and this fact has consequences for understanding
some social interactions. In Section 1.7 we will dedicate some more time to discussing
the selfishness assumptions. Throughout this book we will see many more examples
and variations of prisoners’ dilemma games. One of the most interesting extensions of
this theory has to do with the repetition of the interactions and the emergence of coop-
eration in that context. Chapter 7 is entirely dedicated to this question with extensive
discussions on the theory, the historical and field data, and lab experiments.
To finish this first encounter with the prisoners’ dilemma, let us see what this game
can add to some classical philosophical views of society. The negative vision on the
nature of human beings, for example, culminates in Hobbes’ Leviathan (Hobbes, 1651).
The expression homo homini lupus, which he borrows from the Roman p laywright
Plautus, captures this perspective, best expressed in the famous quote in Chapter XIII
of his book:
Whatsoever therefore is consequent to a time of war, where every man is

enemy to every man, the same consequent to the time wherein men live
without other security than what their own strength and their own invention
2 Immanuel Kant (1785), Grounding for the Metaphysics of Morals.
11
GAME THEORY
shall furnish them withal. In such condition there is no place for industry,
because the fruit thereof is uncertain: and consequently no culture of the
earth; no navigation, nor use of the commodities that may be imported
by sea; no commodious building; no instruments of moving and removing
such things as require much force; no knowledge of the face of the earth;
no account of time; no arts; no letters; no society; and which is worst of
all, continual fear, and danger of violent death; and the life of man, solitary,
poor, nasty, brutish, and short.
After Hobbes, Locke and, mainly, Rousseau (1762) defended an opposite view, closer to
the myth of the Good Savage. Arguably, one of the best alternatives to Hobbes is Adam
Smith (1776) and his Invisible Hand, a metaphor to explain how selfishness does not
imply Hobbes’ view of a society in the absence of a Leviathan. The quote from his book
The Wealth of Nations is no less famous than Hobbes’:
It is not from the benevolence of the butcher, the brewer, or the baker
that we expect our dinner, but from their regard to their own interest.
We address ourselves, not to their humanity but to their self-love, and
never talk to them of our own necessities but of their advantages.
The general equilibrium theory in economics is the theoretical corpus that draws
upon ideas established by Adam Smith (Smith, 1776). The first welfare theorem states
that if a number of assumptions are satisfied then competitive markets are efficient,
which is a keystone of the theory. Needless to say, neither Adam Smith nor neoclassi-
cal economists thought that markets were always efficient, and that is why economists
emphasize the theorem’s assumptions. Game theory, on the other hand, presents a
theoretical framework to generalize the concept of competitive equilibrium to many
situations that do not fit under the assumptions of the first welfare theorem. Among
those situations there is one that appears to be the opposite archetype to the Invisi-
ble Hand, and this is the prisoners’ dilemma. Notice that it is really a simple game,
formed with literally four numbers, and yet it helps to understand relevant real-life
problems and to offer a non-trivial analysis. Risking simplification, one may say that
game theory along with the general equilibrium theory allow us to say under which
circumstances we can see the Invisible Hand at work and when a human being is wolf
to another human being.
1.6 Matching pennies
In the prisoners’ dilemma there was a strong strategic tension. Even if the conflict
prevailed, there was still a cooperative outcome; although, it does not constitute a an
equilibrium. In our next game, matching pennies, there is no room for cooperation,
equilibrium or not. What one player wins is the loss of the other player. Any game in
which this happens is called a zero-sum game – a game of pure conflict.
The game is played as follows. Two players decide casting lots to select a winner (e.g.,
who starts an activity or who keeps some money they bet). Each player has to show one
side of a coin (a penny). If the sides match, because they both showed heads or both
showed tails, Player 1 wins. Otherwise, Player 2 wins. Figure 1.6a shows the game.
12
We immediately observe that this game has no equilibrium. If both players played
“heads” or both played “tails,” Player 2 regrets her choice. If they chose different sides, it
is Player 1 who has regrets. There is no stable situation: Player 1 wants the sides to match,
while Player 2 wants them to mismatch. However, as anyone who has played this game
or a similar one knows, there is a way to play: you need to be unpredictable by choos-
ing “heads” sometimes and “tails” some other times. Any deterministic rule to choose
between the two strategies is likely to be detected by the opponent and used in her favor.
For instance, if Player 2 alternates between “heads” and “tails,” Player 1 will soon react by
playing the same choice made by Player 2 in the previous turn. The best way to be unpre-
dictable is to choose one option at random and, in this precise game, with odds 50–50.
2
Heads Tails
Heads 1, −1 −1, 1
1
Tails −1, 1 1, −1
Figure 1.6a Matching pennies
It is not always the case that the best way to be unpredictable is playing the different
options with equal probabilities. This is shown in the next game, that strategically is just
a version of the matching pennies, played between the two best tennis players ever accord-
ing to the rankings. Say that, at some point in a Grand Slam match, Nadal serves and
Federer returns. Nadal has to decide whether to aim to the right or to the left, and Federer
has to decide whether to anticipate the ball coming to the right or to the left. Right and
left are always defined from the perspective of the server. It is important to understand
that Federer cannot just wait and see where the ball is coming, as its speed is so high that
the time it takes to cross the court is less than the reaction time of any human being.
Thus, to all effects, the decisions of both Nadal and Federer are simultaneous.
Let us write some numbers. For instance, if Nadal serves to the left and Federer mis-
takenly anticipates that the ball is going to the right, then Nadal has a great advantage
to win the point. Say that in this circumstance he wins the point 90 percent of the time.
Also, if Nadal serves to the right and Federer anticipates the left, then Nadal wins 70
percent of the time. Finally, if Federer anticipates the side correctly, the odds of Nadal
winning are 50–50. In this example, we have made up the numbers, but we can use
real numbers studying the historical averages in the matches between the two players.
Figure 1.6b summarizes all this.
Federer
Left Right
Left 50, 50 90, 10
Nadal
Right 70, 30 50, 50
Figure 1.6b Nadal-Federer game
What is the best course of action for Nadal? And for Federer? Notice that Federer
wants to match Nadal’s choice, while Nadal wants the opposite. They are in exactly
the same situation as the players in the matching pennies game. Should they try to be
unpredictable choosing their strategies with probabilities 50–50? Let us see.
13
GAME THEORY
Consider that Federer observes that Nadal serves with those probabilities. If Federer
returns from the left, he will win the point 50% of the time when Nadal plays “Left” and
30% of the time when Nadal plays “Right.” Since Nadal is playing “Left” and “Right”
50% of the time, on average, Federer will win 1 / 2 × 50% + 1 / 2 × 30% = 80 / 2 = 40
percent of the time. Similarly, if Federer returns from the right, he will win
1 / 2 × 10% + 1 / 2 × 50% = 60 / 2 = 30 percent of the time. Thus, Federer is better off if
he always returns from the left, and therefore he does not want to play randomly. Of
course, if Federer reacts this way, now Nadal will reply by playing always “Right”; how-
ever, going beyond this is not important at this point. We have just seen that if Nadal
chooses sides with probabilities 50–50, this cannot be a stable situation.
What is then the best way to be unpredictable? We will show that Nadal has to
choose “Left” one third of the time. If Federer returns from the right, he will win
1 / 3 × 50% + 2 / 3 × 30% = 110 / 3 = 36.66 percent of the points; if he returns from the
left, he will win 1 / 3 × 10% + 2 / 3 × 50% = 110 / 3 = 36.66 percent of the time. In other
words, Federer does not have a preferred side to return. Notice that, for Nadal, 33.66%
is better than the 40% we found before with the 50–50 play (he wins the points that
Federer does not).
Now we will show that the best way for Federer to randomize is playing “Left” two
thirds of the time. To see this is indeed the case, notice that if Nadal serves to the left he
will win the point 2 / 3 × 50% + 1 / 3 × 90% = 190 / 3 = 63.33 percent of the time, and
if he serves to the right he will win 2 / 3 × 70% + 1 / 3 × 50% = 190 / 3 = 63.33 percent
of the time.
To summarize: Nadal has a strategy that guarantees that he wins at least 63.33 per-
cent of the points, while Federer has a strategy that guarantees that Nadal wins at most
63.33 percent of the points. Since both numbers coincide, no one has an incentive to
do otherwise, and the situation is stable. Nadal cannot do better with any other strategy
as long as Federer plays this way, and vice versa.
The analysis was a little hard, but we have learnt that we can find an equilibrium
using random strategies and that this randomization does not need to put equal proba-
bilities in the different options. The reader can repeat the analysis to prove that, in the
original matching pennies game, the randomization 50–50 is indeed an equilibrium.
Recall that we just learned to check whether a given randomization is an equilibrium, but
we do not know yet how to find the equilibrium. We leave this task for the next chapter.
It seems then that, game theory in hand, we can advise tennis players to opti-
mize their play. Well, not really. According to some studies,3 professional players have
already learned the best strategies for serving and returning against different rivals.
Most likely they did it by trial and error, and not by using game theory. Nevertheless,
it is remarkable that we can model this equilibrium behavior.
1.7 The ultimatum game

Our last game is also a zero-sum game. We include it in the list for two reasons. First,
to provide one example in which players play sequentially rather than simultaneously
and, second and more importantly, because the game has become famous for its use
3 An example is Walker and Wooders (2001).
14
in the literature of behavioral economics in the study of the hypotheses of selfish and
altruistic behaviors, among others.
A benefactor gives Lisa and Bart the possibility of winning €100 between the two
of them. They only need to agree on how to share the money. The benefactor sets the
following rules: Lisa will propose how to divide up the €100 and then Bart must decide
whether or not to accept. In case he does, they share the money according to Lisa’s
proposal, but if he does not accept, they both get nothing and that will be the end of
the game.
If Lisa and Bart only care about money, Lisa will propose to keep almost all of it
and leave Bart with just one euro. As one euro is better than nothing, Bart should
accept. Before discussing the realism of this solution, let us dedicate some time to the
formalities of the game. The first thing to notice is that, because choices are not made
simultaneously, we will need to change the way we represent the game. Instead of using
a table, we will draw a tree as seen in Figure 1.7.
100
99 98 0
B B B …… B
a r a r a r a r
100 0 99 0 98 0 0 0
0 0 1 0 2 0 100 0
Figure 1.7 The ultimatum game
Lisa plays first and is represented with the letter L. She can propose to keep for her-
self any quantity between 100 and zero euros (for simplicity we assume that no offers
with cents can be made), a fact that is represented with all the branches that depart
from the point where Lisa decides. Bart is represented with the letter B and he can
accept (a) or reject (r) each offer. The numbers at the end are the payoffs; the first one
corresponds to Lisa and the second to Bart. For instance, if Lisa proposes to keep €98
for herself and Bart accepts, the payoffs are €98 and €2, respectively; if Bart rejects,
they both get zero.
In the tree, we can repeat the informal analysis above. In every possible con-
tingency, Bart has to decide one of two branches, and the one labeled “a” is always
better, except when he is offered to get zero, in which case he is indifferent between
accepting and rejecting the offer. In this latter case, if Bart rejects zero, Lisa will
propose to keep €99 for herself, Bart will accept €1 and this would be an equilibrium
of the game. There is another equilibrium, though, in which Lisa proposes to keep
the €100 and Bart accepts all offers, including this one (see that he does not gain
anything by rejecting zero). We could discuss how plausible it is that Bart accepts
nothing in this second equilibrium, but the discussion would be quite irrelevant:
both equilibria say practically the same thing, namely, that Lisa has a huge advantage
with these sharing rules.
This simple game, known as the ultimatum game, is of great interest in under-
standing the status of game theory, as many lab experiments have shown that what
15
GAME THEORY
usually happens is that the “Lisas” of the games offer to keep around 60 percent of the
money, and the “Barts” tend to reject offers that give them less than 40 percent. It is
clear that in this game the selfishness assumption is not satisfactory. However, before
discarding it, it is worth noting that there are some other situations in which it is a good
assumption.
Take the case of oligopolistic markets, normally studied with the tools of game
theory. If firms are selfish, these markets are not efficient and the firms will produce
too little at a too high price, and further still, their benefits will be disproportionate
when compared to those of consumers. This coincides with what we actually observe.
Firms could be less selfish and sell more at a lower price, as if they were in a perfectly
competitive market. However, if we want to write a theory of oligopolies to explain real
markets and simulate the consequences of different types of regulation, should we rely
on the assumption that firms are altruistic or selfish? We have every reason to believe
we would obtain a better description with the second assumption.
We thus have an array of different scenarios, each with its own behavior by the
players. What does all this imply for the selfishness assumption? There are at least two
options here:
1. Substitute the selfishness postulate with another one that explains all the cases stud-
ied using game theory, and not only some of them (e.g., oligopolies).
2. Keep the selfishness postulate in the models where it works better and replace it
with another one when it does not work.
The first of these options would be the best if only we had this other postulate that is
more general than that of selfishness and that improves the existing models. Unfortu-
nately, we are far from having such a thing and must keep our expectations to some-
thing closer to the second option.
So does this mean that we keep the selfishness postulate in the theory of oligopolies
but give it up in the theory that studies the ultimatum game? Not quite. Even if exper-
imental subjects are not selfish, the study of the selfish equilibrium is still important,
as it establishes a baseline with which we can compare observed behavior. For instance,
we can define a measure of altruism depending on how much the observed behavior
departs from the selfish equilibrium. It is also important because in some variations of
the game, subjects get increasingly closer to the equilibrium. When experimental play-
ers play against a computer, or when they play in teams, they tend to accept small offers
(see the works by van’t Wout, Kahn, Sanfey and Aleman, 2006, and by R obert and
Carnevale, 1997). Furthermore, the observed behavior gets even closer to the selfish
equilibrium when the situation is described as an anonymous interaction in a mar-
ket and when the experiment is conducted using a double blind so that the subjects
have warranties that no one, not even the persons conducting the experiments, will
know their actions (Hoffman, McCabe and Smith, 1996). Finally, comparing observed
behavior and the selfish postulate across various types of games may help establish
whether those two behaviors differ because of altruism, reciprocity or spitefulness,
among other possibilities.
Thus, the problem with the selfishness postulate is not whether it is true in general
(it is not), but whether we use it properly in the adequate models. To say that game
theory, or economics for that matter, is wrong because it uses this postulate is a bad
16
criticism. To point out that here and there some practitioners abuse the postulate and
use it where they should not is a good criticism.
Exercises for Chapter 1

1.1 Two car companies are planning to launch an electric car to the market at the
same time. Each of them is considering whether it should offer credit to the buy-
ers in order to reach a larger share of customers. However, offering credit would
imply incurring some costs. Both companies prefer not to offer credit, but they
are afraid that the other one will do so and will, therefore, attract more clients.
Suppose that the expected benefits for both companies are the following: If both
offer credit, each gets 40 million euros. If none of them offers credit, they get 60
million each. If one offers credit and the other one does not, the first one will
earn 80 million while the other will obtain 30.
(a) Represent the game in the normal form; that is, represent the players, strate-
gies and payoffs in a table, following the examples in the chapter.
(b) Which game in the chapter is closer to this one?
(c) Which equilibrium situations can you find?
1.2 Two states compete for a plan that will offer tax reductions in an effort to attract
firms. If both offer the same reductions, the tax rate and the tax revenues decrease
with no guarantees of new firms establishing in the state. In such a case, the states
would have preferred higher taxes. The idea is to attract firms even if it implies tax
reductions.
(a) Represent the game in the normal form. Use numbers that fit the case as
explained in the text.
1.3 The city council has to decide whether to build a high school or a nursery in a
particular zone of the city. It does not have budget to carry out both projects. The
person in charge of managing these matters has spoken with two indispensable com-
panies that can make any of these two projects: one construction company and
one carpentry company. Due to the composition of the population, the high school
would be bigger than the nursery (it requires more construction), but the nursery
will need a children’s play park (it requires more carpentry). In addition, each one
of the companies is interested more in participating in a certain project that in the
other (the construction company prefers the high school and the carpentry company
prefers the nursery), but both prefer signing the same contract to signing different
contracts, since the city council would not carry out any project if different contracts
are signed. The city council asks them to present a project. As none of the companies
has sufficient personnel available to process both projects, they must concentrate on
just one of them without knowing the project chosen by the other company.
17
GAME THEORY
1.4 Freedonia is a country that needs financial aid, and it can only obtain that aid
from the Pangean Union (PU), an economic union of countries to which it
belongs. As a condition for the aid, the PU requires that Freedonia make some
deep fiscal reforms. The government of Freedonia makes a counterproposal
with a different package of conditions (basically, minor reforms). Each side
believes that its proposal is the best and that the other one will imply a disas-
ter. They can hold their position or make some concessions to the other party.
If they both make concessions, negotiations will end up with a result that is
unsatisfactory for either of them but still better than if no one makes conces-
sions and holds their position (which would be the worst possible outcome for
both of them). Of course, each one would prefer to hold while the other party
is the only one making concessions.
1.5 Two electors, A and B, must vote among three alternatives, X, Y and Z. If both
vote for the same one, that alternative is chosen. In case of a tie between X and
any other alternative, X is chosen, while in case there is a tie between Y and Z, Y
is chosen. The preferences of the electors are ranked according to the following
table:
A B
X Z
Y X
Z Y
(a) Represent the game. Hint: Notice that now the players have three strategies,
rather than just two as in all previous games.
(b) Find the equilibria.
1.6 Consider the prisoners’ dilemma game in Figure 1.5a.
(a) Change the payoffs to reflect the case in which one year in prison suffered by
the partner causes as much pain as half a year suffered by oneself, but only if
the partner did not confess.
(b) Find the equilibria.
(c) How much pain should a year in prison by the partner cause (always only in
case he did not confess) for the game to have only an equilibrium in which
no one confesses?
(d) Repeat (a) and (b) for a case in which only Player 1 has the altruistic pref-
erences described in (a), while Player 2 has the original preferences in the
prisoners’ dilemma.
1.7 Consider the ultimatum game in Figure 1.7, but now with the players having
only €10 to share. As in the text, proposals cannot contain fractions of euros.
18
Now consider also that Lisa’s preferences show empathy for Bart. In particular,
one euro that Bart earns is worth half a euro to Lisa.
(a) Represent the game tree. Include all the branches.
(b) What offers will Bart accept? What will Lisa offer?
(c) Repeat the exercise for the alternative case in which one euro that Bart earns
is worth one euro to Lisa.
1.8 Consider the ultimatum game in Exercise 1.7, but now make Lisa’s preferences
regarding the well-being of Bart a bit more complicated. If Bart earns an amount
between 6 and 10 euros less than Lisa, each of these euros gives Lisa as much
satisfaction as 2 of her own euros. If Bart earns between 1 and 5 euros less than
Lisa, each of the first 2 euros Bart earns gives Lisa as much satisfaction as 2 euros
of her own, and each one of the next euros will give Lisa as much satisfaction as
one and a half euros. Lisa gets no satisfaction for Bart’s money if Bart earns the
same as or more than she does.
(a) Represent the game tree. Include all the branches.
(b) What offers will Bart accept? What will Lisa offer?
1.9 The popular game rock-paper-scissors is played between two players. Each player
has to choose one of the three objects, and they win according to these rules: rock
beats scissors, scissors beats paper and paper beats rock. In case they choose the
same object, they are tied. Write the normal form of the game.
1.10 In a penalty kick, the player can aim to the left or to the right. The goalie can
move to the left, stay put, or move to the right, but she must make the decision
before waiting to see the ball coming, as there is not enough reaction time. Rep-
resent the situation as a game and explain the payoffs you use.
1.11 Represent a coordination game between three players, where each one has to
choose between “Left” and “Right.” If only two players coordinate, the payoff
of these two players is higher than the payoff of the player left uncoordinated.
Payoffs are the highest if the three of them coordinate. Hint: You will need two
tables, each one with two rows and two columns. Player 1 chooses the row (the
same in both tables), Player 2 chooses the column (also, the same in both tables)
and Player 3 chooses the table.
1.12 Two firms compete in a market. Now each one is making one million euros of
profits. Firm A is considering whether to contract an advertisement campaign
that will increase its profits to €1.2 million (after subtracting all the costs of
advertising), while the profits of the rival, Firm B, will decrease to €650,000.
The problem is that Firm B is also considering its own campaign, with similar
consequences if it is the only one advertising. If they both advertise, each one will
have €850,000 profits.
(a) Represent the game in the normal form.
19
2 BUILDING THE THEORY FOR
SIMULTANEOUS GAMES
The purpose of game theory is to study situations of strategic decision making, and
to that end it develops a formal model of the situation and proposes various possi-
ble solution concepts. One must not confuse the real-life situation with the model.
Once the model and the equilibrium are defined, they will deliver a theoretical
result. To find the equilibrium in the game means just to find the combination of
strategies (one strategy for each player or, in some cases, one set of strategies for
each player) that satisfies the definition. After proceeding with this exercise, the
modeler may find the analysis unsatisfactory. If this is the case, one can alter the
model, incorporating some aspects from the real world that have been overlooked
in the previous version, or try a different equilibrium concept. The reasons the
analysis may be unsatisfactory are plentiful and could depend on the purpose of
the analysis. If the purpose is normative, in the sense that it proposes a course of
action to the players, the analysis will be unsatisfactory if the players do not accept
the advice. If the purpose is descriptive, in the sense that it tries to reproduce what
players do in the real world, then the model will not be appropriate if it does not
fit the empirical data.
In this chapter, we introduce some formal definitions that the reader may want
to postpone to a second or a third reading of the text. All of them are marked with
expressions such as “formal definition,” “formally” or “formal analysis” and are accom-
panied by less formal and more intuitive explanations that can be read more easily.
First, in Section 2.1 we present the definition of a normal form game, that includes
all the games in the previous chapter except the last one. In Section 2.2 we explore
different ways of thinking about what constitutes a solution of a game, and we will end
up proposing the Nash equilibrium as the most important concept for the purposes of
this book. Section 2.3 studies some relations among the different concepts presented in
the previous section and can be skipped on a first reading. Section 2.4 should be read
with special attention, as it provides a guide to look for equilibria, something that we
will be doing throughout the entire book. Section 2.5 completes this guide with some
special difficult cases. Finally, Section 2.6 completes the chapter with a technical detail
on strategy domination.
2.1 The normal form game

For those situations where the decisions by the different players can be considered
simultaneous, either because players physically play at the same time or because they
make their decisions without the knowledge of the actions taken by the other players,
game theory proposes the mathematical model of the normal form game.
20
BUILDING THE THEORY FOR SIMULTANEOUS GAMES
The definition of the normal form game

Informally, we have a normal form game whenever we have defined the following three
elements: a set of players, a set of strategies for each player, and a payoff for each player
after all possible combinations of strategies. The reader may want to skip the formal
definition next and go directly to the illustration of these elements in the battle of the
sexes game, after Note 3.
More formally, we write the elements this way:
• A set of players: = {1, 2,…,n}.
• A set of strategies for each player: Si = {si1 , si2 ,…, siki }.
• A payoff function for each player: this function is of the form ui : S1 × S2 ×…× Sn → R .
Another way to express the payoff function is to write ui ( s1 , s2 ,…, sn ) ∈ R ,
where each player i is playing his strategy si from his set of strategies Si .
Note 1: The number of strategies for Player i may be different from the number of
strategies of other players, which is why we have written ki as the number of strategies
for this player.
Note 2: The payoff function assigns a real number (a number in R) to every combination
of strategies. That is, if Player 1 plays s1, Player 2 plays s2, and so on until the last player,
then Player i enjoys the payoff in R assigned by ui . This payoff can be understood as
utility, satisfaction, profit or any other interpretation depending on the particular game.
Note 3: If we are unable to specify a payoff function for some player and for some com-
bination of strategies, then we do not have a w ell-defined game.
In the battle of the sexes game introduced in Chapter 1, the elements of the game are
as follows:
• The set of players is N = {1, 2}, where 1 is Sylvie and 2 is Bruno. The name or label
of the player is irrelevant, we can write N = {1, 2}, N = {Sylvie, Bruno}, or any other
labels. We only need to remember which label corresponds to which player when it
comes time to shed light on the real-life situation following the analysis of the game.
• The sets of strategies are S1 = {s11 , s12 }, S2 = {s21 , s22 }, where s11 is “Football,” s12 is “ Movies,”
s21 is “Football” and s22 is “Movies.” Again, the names of the strategies are irrelevant
for the normal form game. We could have written the strategy sets as SS = {F , M },
SB = {F , M }, where S is “Sylvie,” B is “Bruno,” F is “Football” and M is “Movies.”
• Each player has two strategies to choose; therefore, there are four possible combina-
tions of strategies that correspond to the possible outcomes of the play. These combi-
nations are (F, F), (F, M), (M, F) and (M, M), where the first letter indicates Sylvie’s
choice and the second letter indicates Bruno’s. For each of these four possibilities,
the payoff function of each player must assign a numerical value:
The payoff function for Sylvie is uS ( F , F ) = 2, uS ( F , M ) = 0, uS ( M , F ) = 0,

-
uS ( M , M ) = 1.
-The payoff function for Bruno is uB ( F , F ) = 1, uB ( F , M ) = 0, u B ( M , F ) = 0,
uB ( M , M ) = 2.
21
GAME THEORY
Figure 1.3 in Chapter 1 collects all these elements in a more appealing way. In games
with many players or in games with other types of complications in the description,
the use of a matrix or table to represent it may not be adequate, and we will need to
remember the generic formal definition described in this section.
2.2 Towards a solution
In this section we will explore two different ways in which players may choose their
strategies. The first is a prudent behavior that we then argue may be suitable for cer-
tain kinds of games called zero-sum games. The second is the elimination of certain
particularly bad strategies that, along with forward-looking reasoning, may simplify
a game and even solve it in some special cases. Our plan is to (i) describe the prudent
behavior and formalize it with the concept of maximin strategy, (ii) define zero-sum
games, (iii) formalize the “bad” strategies with the concept of dominated strategy and
(iv) describe and formalize the forward-looking reasoning with the concept of ration-
alizable strategy. The discussion of these introductory, behavioral aspects will pave the
way towards the main equilibrium concept in this book, the Nash equilibrium.
Maximin strategies
In 1972, William Drakert published an article in the journal Moves (Drakert, 1972),
dedicated to the simulations and analysis of conflicts, where he presented six possi-
ble plans for D-Day, the Normandy landing, that the Allies could be considering,
along with six plans for the Germans to defend their positions. Both armies must make
their decision without knowing the decision of the enemy. Next, Drakert simulated the
consequences of each of the 36 combinations, as summarized in Figure 2.1. We have
changed some numbers to make this exposition simpler.
We interpret the contents of the table as usual: the higher the number, the better
the advantage for the allied forces and, consequently, the worse for the Germans. This
table contains all the elements of a normal form game if we consider the numbers as
payoffs for the Allies and the negative of those numbers as the respective payoffs for the
Germans. For instance, in the cell 4E we find the number 21. To complete the game,
in that cell we should also have the number −21 as the payoff of the Germans. Keep
in mind that what matters most is whether a number is higher or lower than another
number, rather than its magnitude.
Germans
A B C D E F
1 13 29 8 12 16 23
2 18 22 20 22 29 31
3 18 22 15 31 27 37
Allies
4 11 22 12 21 21 26
5 18 16 19 14 19 28
6 23 22 21 23 30 34
Figure 2.1 D-day game
22
Let us assume that the Allies show a prudent attitude in the conflict, which means
that they first look for the worst-case scenario in each of the six plans they are consid-
ering, and they then choose the plan whose worst-case scenario is the least devastating:
• If the Allies act according to Plan 1, the worst thing that can happen to them is that
the Germans defend using their Plan C. This gives a payoff of 8 for the Allies.
• If the Allies choose Plan 2, the worst situation is that the Germans choose Plan A,
with a payoff of 18.
• In this fashion, we can complete the worst payoffs from Plans 3 through 6, giving
payoffs 15, 11, 14 and 21, respectively.
If the Allies decide according to the prudent criterion, and therefore use Plan 6, they
will not obtain a payoff worse than 21, the best guarantee that they can have with
any plan.
Now we can do the same exercise for the Germans. We need only to remember
that the higher the number, the worse for them. If they defend using their Plan A,
their worst result comes from the Allies using their Plan 6, with a payoff of −23 for the
Germans. For Plans B through F the respective worst-case scenarios give payoffs −29,
−21, −31, −30 and −37. The defensive plan that contains the least devastating result is
Plan C, with a payoff of −21.
After this analysis, we find that the value that the Allies can guarantee for them-
selves (21) coincides with the negative of the value that the Germans can guarantee
(−21). Thus, if the Allies choose Plan 6 and the Germans choose Plan C, both armies
will have the most they can be guaranteed to have, and neither can obtain more by
changing their plan. The Allies cannot have more than 21 because the Germans are
guaranteed to not have less than −21. The reader can easily check this fact in Figure 2.1.
In this game, we have found an equilibrium (6, C), and a behavior that leads to it
(the use of prudent strategies). In other games, such as in matching pennies or in the
game between Nadal and Federer, to find such equilibrium we need to allow players
to randomize among their strategies. The reader may also note that the equilibria in
those games result from the players using prudent strategies. As long as there are only
two players and the payoffs of these players add up to zero (like the Normandy landing
in Figure 2.1 or matching pennies in Figure 1.6a) or to a constant (like the Nadal vs.
Federer game in Figure 1.6b, where the chances always add up to 100%), the use of
prudent strategies will lead to an equilibrium, a situation from which a player cannot
win by changing his or her strategy. In other games this needs not be the case. For
instance, in the chicken game in Figure 1.4 the prudent strategy for each player is to
swerve, but as we saw in Section 1.4, this behavior does not lead to an equilibrium:
either player can win by changing his strategy to “Keep going.”
Formal definition of a maximin strategy

A maximin strategy is defined in three steps. We present here the definition for two-per-
son games, and just for Player 1 (for Player 2 it is entirely analogous):
1. For every strategy of Player 1, find the worst possible payoff. Formally, for all s1l ∈ S1
define min1 ( s1l ) = min{u1 ( s1l , s21 ) , u1 ( s1l , s22 ) ,…, u1 ( s1l , s2k2 )}.
23
GAME THEORY
2. Next, take the maximum of those values:

{ }
maxmin1 ( s1 , s2 ) = max min1 ( s11 ) , min1 ( s12 ) ,…, min1 ( s1k1 ) .
3. Finally, the maximin strategy is the one whose minimum value (worst payoff) coin-
cides with the maximum of the values found in step 2: s1* ∈ S1 is a maximin strategy
for Player 1 if min1 ( s1* ) = maxmin1 ( s1 , s2 ) .
Formal definition of a two-person, zero-sum game, and of its solution

A two-person, zero-sum game is a game such that N = {1, 2} and u1 ( s1 , s2 ) = −u2 ( s1 , s2 ).
In a two-person, zero-sum game, if the payoff that Player 1 can guarantee for
himself, using his maximin strategy, is the negative of the payoff that Player 2 can
guarantee for herself, again using her maximin strategy, then the game is said to have
a solution, which consists of both players using their respective maximin strategy. For-
mally, if maxmin1 ( s1 , s2 ) = −maxmin2 ( s1 , s2 ), then the game has a solution ( s1* , s2* ) where
min1 ( s1* ) = maxmin1 ( s1 , s2 ) and min 2 ( s2* ) = −maxmin1 ( s1 , s2 ).1
The elimination of dominated strategies

Two cities compete to attract 100 firms. Each city has 26 firms in their respective ter-
ritories, and those will remain there regardless of their policy on municipal taxes. The
remaining 48 firms will move to the city with the lowest taxes. In the case that taxes
are the same, they will randomly decide where to go. By law, the taxes a city can impose
are €1000, €2000 or €2500 by firm. Each city’s interest is the maximization of its tax
revenues. What tax will the cities impose? Figure 2.2a shows the game, with payoffs
expressed in thousands of euros. For instance, if both cities impose the €1000 tax, each
gets 50 firms, which will pay 50,000 in taxes. If City 1 chooses the €2000 tax and City 2
chooses €2500, City 2 keeps its 26 firms, but the remaining 74 will be placed in City 1.
City 2
€1000 €2000 €2500
€1000 50, 50 74, 52 74, 65
City 1 €2000 52, 74 100, 100 148, 65
€2500 65, 74 65, 148 125, 125
Figure 2.2a Tax war
In the table for this game we can observe that the strategy of taxing every firm with
€1000 is not a good idea, as a tax of €2000 is always better, regardless of what the other
city decides (we say that strategy €1000 is dominated by strategy €2000). To see this,
just compare the payoffs for City 1 in the first row (corresponding to a tax of €1000)
with the payoffs in the second row (a €2000 tax) cell by cell. The game is symmetric,
1 John von Neumann (mathematician) and Oskar Morgenstern (economist) presented the theory
of zero-sum games in their book The Theory of Games and Economic Behavior, the publication of
which is considered to have been the beginning of game theory as a discipline (von Neumann and
Morgenstern, 1944).
24
and the same can be said for City 2. Once we discard the €1000 tax, the game is sim-
plified to the game in Figure 2.2b.
City 2
€2000 €2500
€2000 100, 100 148, 65
City 1
€2500 65, 148 125, 125
Figure 2.2b Tax war after elimination of dominated strategies
In the new game, we observe that imposing a €2500 tax is always worse than imposing
€2000. This was not the case in the original game in Figure 2.2a, because there was a cir-
cumstance in which €2500 was better than €2000, namely, when the other firm imposed
€1000. But once the €1000 tax is out of the picture, there is no reason to use the strategy
of a €2500 tax. Thus, now, unlike before, we can delete the strategy €2500 (and, again, we
say that in this smaller game strategy €2500 is dominated by strategy €2000). After this
elimination, each city has just one option left: to impose a €2000 tax. Each one will collect
€100,000 in tax revenues and will not have any incentive to do otherwise. The strategies
that survive the iterative elimination of dominated strategies will be called rationalizable
strategies. The definition of these concepts is given in the following paragraphs.
In the game illustrated in Figure 2.1 we were able to find an equilibrium point
(6, C) and a behavioral story of how players got to play their respective equilibrium
strategies (the use of the prudent, maximin strategies). Here we have found something
similar, but with a different behavioral story, the iterative elimination of dominated
strategies. This procedure may help analyzing some games, but it will not find a solu-
tion in most cases. Exercises 2.1 and 2.2 illustrate these facts.
Definition of a dominated strategy

We say that a strategy si of Player i is dominated if there exists another strategy si′ of
that same player such that, for any choice of strategy by the other players, the utility of
Player i is always higher with si′ than with si .
Formally, for a two-person game the strategy s1 ∈ S1 is dominated if there exists
another strategy s1′ ∈ S1 such that for all s2 ∈ S2, u1 ( s1 , s2 ) < u1 ( s1′ , s2 ) . Dominated strate-
gies for Player 2 are defined similarly.
There is a weak version called weakly dominated strategy, defined similarly except
that the strategy that is better than the dominated one gives a utility higher than or
equal to the dominated strategy. It should be clear that a dominated strategy is also a
weakly dominated strategy, but that the converse is not true in general.
Formally, strategy s1 ∈ S1 is weakly dominated if there exists another strategy s1′ ∈ S1
such that for all s2 ∈ S2 , u1 ( s1 , s2 ) ≤ u1 ( s1′ , s2 ) , with u1 ( s1 , s2 ) < u1 ( s1′ , s2 ) for some strat-
egy s2 ∈ S2 . Weakly dominated strategies for Player 2 are defined similarly.
Definition of rationalizable strategy

The set of rationalizable strategies in a game is defined as the set of strategies that sur-
vive the iterative elimination of dominated strategies, which is done as follows:
25
GAME THEORY
1. Detect and delete from the game all dominated strategies.

2. In the game that is defined after step 1, delete all dominated strategies.
3. Continue eliminating dominated strategies until there are no new dominated
strategies to delete.
4. The strategies in the last game are called rationalizable strategies.
We can also define a procedure of iterative elimination of weakly dominated strategies,
although the surviving strategies do not have a special name.
The Nash2 equilibrium

The previous sections illustrated the difficulties of finding the best theory, one that
provides a definition of an equilibrium and also a behavioral model to explain how
that equilibrium is reached. Given this difficulty, perhaps an impossibility, the theory
has emphasized the developing of the equilibrium concept. We will discuss the reasons
after introducing the most important equilibrium concept in game theory.
We define a strategy profile as the election of one strategy for each player. A strategy
profile is a Nash equilibrium if, given what the other players are choosing in this pro-
file, no player has a strategy that gives him or her a higher utility than the one he or she
is obtaining in the profile. In other words, a Nash equilibrium has the property that no
player wants to deviate from it unilaterally.
The formal definition of the Nash equilibrium

A strategy profile s * = ( s1* , s2* ,…, sn* ) ∈ S = S1 × S2 ×…× Sn is a Nash equilibrium if
u1 ( s1* , s2* ,…, sn* ) ≥ u1 ( s1 , s2* ,…, sn* ) for any other strategy s1 ∈ S1 ,
u2 ( s1* , s2* ,…, sn* ) ≥ u2 ( s1* , s2 ,…, sn* ) for any other strategy s2 ∈ S2 ,
…
un ( s1* , s2* ,…, sn* ) ≥ un ( s1* , s2* ,…, sn ) for any other strategy sn ∈ Sn.
In this book, we will mostly present the rational side of game theory. This means
that we will try to answer the question of what situations will prevail if individuals
are rational. This is not the only aspect of the theory, as some other equilibrium con-
cepts and dynamics have been explored. A non-exhaustive list of behaviors that depart
from rationality includes partial rationality, learning, copying, herding behavior, social
dynamics, evolutionary determination of strategies and cognitive biases. Throughout
this book, we will also introduce some of these alternatives, but the rational behavior
will be our starting point. We will explore the rationality assumption in examples and
applications where it provides an insightful, realistic analysis of the problem at hand.
The building of a theory of games can be thought as the objective of writing The
Great Book of Game Theory that would contain all possible games. On each page there
would be a game, along with its solution in the form of a recommendation of what
every player should play. Such a fantastic book, of which every player would have a
2 John J. Nash presented his equilibrium theory as an extension of the theory by von Neumann and
Morgenstern for zero-sum games to more general games (Nash, 1950).
26
copy, must have a fundamental characteristic, namely, that the players would be will-
ing to carry out the recommendations. This means that one player in particular would
accept that the other players would choose as the book says and that, taking that as a
given, his own calculations about what is best for him would coincide with what the
book recommends. In other words, the solution must satisfy the definition of Nash
equilibrium. This definition may not be sufficient, and we may demand more from the
solutions in the book, but being a Nash equilibrium is a necessary condition.
Until now, we have justified the actions in the equilibrium as strategies chosen by
players and inferences made about other players’ choices. It is possible to explain the
notion of Nash equilibrium in a different way – as beliefs that sustain and enforce
one another. If each player believes that the others will play according to the equilib-
rium, then they maximize their utility by also playing as the equilibrium says. This
way, the Nash equilibrium is a set of beliefs that is self-sustained: if I believe that you
will play s2 , and that implies that I better play s1, and if similarly you believe that
I will play s1 implying that you prefer s2 , then, the pair of beliefs ( s1 , s2 ) is a Nash
equilibrium.
2.3 Some propositions on maximin strategies, rationalizable

strategies and Nash equilibria
We now explore some relations between the concepts defined in the previous sec-
tion. This will be done in Propositions 1 through 4. Some of these relations are better
expressed if we allow players to choose their strategies randomly, like in the matching
pennies game and in the game between Nadal and Federer analyzed in Chapter 1.
Thus, before stating and discussing the propositions, we first formalize the concept of
randomizing strategies.
Definition of a mixed strategy

When we define a game, each player has a definite set of strategies to use. Let us
call them pure strategies. When it is reasonable to assume that the player can choose
his strategy at random, like when trying to be unpredictable in the matching pennies
game, then the set of strategies is enlarged to accommodate this possibility. Once we
allow that, any way of randomly choosing a strategy among any subset of pure strate-
gies will also be a strategy, and we will call it a mixed strategy.
Here is the formal definition of a mixed strategy. First, we will provide the defini-
tion for the case of a set of four pure strategies as an example, and then we will give
the general definition. If the set of pure strategies of Player i is Si = {si1 , si2 , si3 , si4 }, then
the set of mixed strategies of this player is {(p1, p2, p3, p4), such that p1 ≥ 0,p2 ≥ 0,p3 ≥ 0
and p4 ≥ 0, with p1 + p2 + p3 + p4 =1}, where pk is the probability with which strategy sik is
played. If we have n pure strategies, the definition is analogous: If the set of pure strat-
egies of P layer i is Si = {si1 , si2 ,…, sin }, then the set of mixed strategies of this player is
{( p1, p2 ,…, pn ) , such that p1 ≥ 0, p2 ≥ 0,…, pn ≥ 0, with p1 + p2 + …+ pn = 1}, where pk
is the probability with which strategy sik is played.
27
GAME THEORY
Four propositions
Next we state, without proof, four useful propositions in game theory. They will
facilitate the analysis of some games. In particular, they will allow us to make some sim-
plifications, like eliminating dominated strategies knowing that we will not eliminate
any Nash equilibrium in doing so.
Proposition 1. If mixed strategies are allowed, in every two-person and zero-sum game,
the Nash equilibrium is obtained if both players use a maximin strategy. This result is
known as the minimax theorem. It is called minimax, and not maximin, to stress the
fact that by playing his maximin strategy, Player i is actually minimizing the maximum
payoff (hence the minimax) for Player j.
Proposition 2. The iterative elimination of dominated strategies does not eliminate any
maximin strategy.
Proposition 3. The iterative elimination of dominated strategies does not eliminate any
Nash equilibrium.
Proposition 4. All games with a finite number of players and of pure strategies have at
least one Nash equilibrium if mixed strategies are allowed.
Let us see an example in which Proposition 3 is not satisfied if we eliminate weakly
dominated strategies. Two neighbors must decide whether to support a candidate as
president of the community. They both like the candidate, but she will not be elected
unless she gets the support of both of them. If giving support is costless, the game can
be represented as in Figure 2.3.
Neighbor 2
Support No support
Support 1, 1 0, 0
Neighbor 1
No support 0, 0 0, 0
Figure 2.3 Choosing a president we all like
In the game we detect two Nash equilibria in pure strategies: (Support, Support)
and (No support, No support). At first glance, the second equilibrium looks improb-
able, and it may quite be so, but remember that, once the game and the equilibrium
concept are defined, we only need to apply the definitions. Any adjustments of the
model to better resemble the reality will come later. In this case, if both neighbors
make the decision not to support the candidate, neither can win by unilaterally chang-
ing his strategy, and that is enough to show that (No support, No support) is a Nash
equilibrium.
We also detect that for each neighbor “No support” is a weakly dominated strategy
(it is weakly dominated by “Support”), but not a dominated strategy. “Support” is not
always better than “No support”: if the other neighbor chooses “No support,” both
strategies give the same payoff. The elimination of dominated strategies does not sim-
plify the game; however, if we eliminate weakly dominated strategies, we will be left
with just the strategy “Support” for each player. In this new, trivial game with only one
strategy for each player, the Nash equilibrium will be ( Support, Support). This way we
get rid of the improbable equilibrium (No support, No support).
28
However, it is not always advisable to eliminate a Nash equilibrium in which play-

ers play weakly dominated strategies, as it may be the only Nash equilibrium in pure
strategies. In all the examples in this book nothing serious will happen if we eliminate
weakly dominated strategies, and we will very often allow ourselves these eliminations.
2.4 Finding the Nash equilibria

There are two houses on a street, and the two neighbors are considering spending some
time the following Sunday to clean their street. Each neighbor has two hours of free
time during that day that they can use to actually help clean the street or watch TV.
Let us denote by ci the number of hours that Neighbor i dedicates to cleaning the street
(and then she will spend 2 − ci hours watching TV). To simplify matters, say that each
neighbor can spend zero, one or two hours cleaning. Finally, each neighbor cares about
the cleaning of the street, but also about watching TV. More specifically, their payoff
functions (also called utility functions) are
U1 ( c1 , c 2 ) = (1 + c1 + c 2 ) ( 2 − c1 )
U 2 ( c1 , c 2 ) = (1 + c1 + c 2 ) ( 2 − c 2 ) .
The game is described in Figure 2.4a. Strategies 0, 1 and 2 are the v alues of ci , the number of
hours dedicated to cleaning the street. If Neighbor 1 spends 1 hour cleaning the street and
Neighbor 2 spends 2 hours, the utility for Neighbor 1 is U1 (1, 2 ) = (1 + 1 + 2 ) ( 2 − 1) = 4
and for Neighbor 2 it is U 2 (1, 2 ) = (1 + 1 + 2 ) ( 2 − 2 ) = 0 :
Neighbor 2
0 1 2
0 2, 2 4, 2 6, 0
Neighbor 1 1 2, 4 3, 3 4, 0
2 0, 6 0, 4 0, 0
Figure 2.4a Cleaning the street game
Next, we will illustrate how to find the Nash equilibrium in two different ways. First,
applying the definition directly and, second, through the examination of the best replies.
Using the definition of Nash equilibrium

The definition of Nash equilibrium (NE) states the conditions for a strategy profile
to be satisfactory. Thus, a direct application of the definition to find the equilibria
requires one to consider all possible strategy combinations and to check one by one if
they satisfy said definition.
The strategy combination (0, 0), which means that both neighbors choose 0 hours
of cleaning time, is an NE: payoffs are (2, 2), and no one wins by deviating. Fix Neigh-
bor 2’s cleaning time to 0 hours, as indicated in the strategy. If Neighbor 1 changes
her strategy to cleaning during 1 hour, her payoff will also be 2 (see her payoff in cell
29
GAME THEORY
(1, 0)), so her utility does not increase. Finally, she will have a utility of 0 if she dedicates
2 hours to cleaning the street (see her payoff in cell (2, 0)). The analysis for deviations
by Neighbor 2 is similar.
Here is the analysis for the rest of the strategy combinations:
• (0, 1) is an NE.
• (0, 2) is not an NE: Neighbor 2 would deviate to cleaning for just 1 or 0 hours.
• (1, 0) is an NE.
• (1, 1) is not an NE: either neighbor would deviate to clean for 0 hours.
• (2, 0) is not an NE: Neighbor 1 would deviate to cleaning for just 1 or 0 hours.
Thus, we find three Nash equilibria: (0, 0), (0, 1) and (1, 0).
Using the best reply

Say that there are n players in a game and that Players 2 through n have already chosen
their strategies, then we define the best reply or best response for Player 1 as the strategy
that gives her the highest payoff given the choices of the other players. Of course, we
define the best replies for the rest of the players similarly. This concept is going to be
extremely useful to find the set of Nash equilibria in a game, as the reader will discover
throughout this book. This is so because, as we can check, in the definition of Nash
equilibrium each player is opting for his or her best strategy (or one of the best, if there are
more than one) given what the others are doing. We can use this fact and look for the best
reply to any choice by the other players to find the Nash equilibrium. In our example, as
there are only two players, for each of them “the others” are just the other neighbor.
Best replies for Neighbor 1:
• If Neighbor 2 chooses 0, Neighbor 1 has two best replies: strategies 0 and 1.
• If Neighbor 2 chooses 1, the best reply for Neighbor 1 is 0.
Best replies for Neighbor 2:
• If Neighbor 1 chooses 0, Neighbor 2 has two best replies: strategies 0 and 1.
Let us underline the payoffs corresponding to the best replies of each neighbor in each
case. The result is shown in Figure 2.4b. Those strategy combinations (cells) where we
have underlined both payoffs satisfy the condition that each neighbor is playing her
best reply against the choice of the other one. These combinations are (0, 0), (0, 1) and
(1, 0), and they are the Nash equilibria of the game found by applying the definition
directly. Making use of this technique of looking for the best replies we just had to
30
check six cases (the six best replies) as opposed to the nine cases we needed to check
in the application of the definition we did before (the nine strategy combinations).
In more complicated games, the gain from using the technique of finding best replies
may be very high. Think of a two-person game where Player 1 has 10 strategies and
Player 2 has 20. We need to check 200 cells to apply the definition of Nash equilib-
rium, but only 30 best replies if we follow this second approach.
Neighbor 2
0 1 2
0 2, 2 4, 2 6, 0
Neighbor 1 1 2, 4 3, 3 4, 0
2 0, 6 0, 4 0, 0
Figure 2.4b Best replies in “cleaning the street”
In this particular game, we could have used some of the properties stated in Sec-
tion 2.3 to facilitate the search of the Nash equilibria. Specifically, for each player
the strategy of cleaning for 2 hours is dominated by any of the other two strategies,
and we could have eliminated it without losing any of the equilibria (Proposition 3).
The resulting game would have been the one in Figure 2.4c.
Neighbor 2
0 1
0 2, 2 4, 2
Neighbor 1
1 2, 4 3, 3
Figure 2.4c Cleaning the street after elimination of dominated strategies
In this reduced game we observe that the strategy of cleaning for 1 hour is weakly
dominated by the strategy of cleaning for 0 hours. If we eliminate it, we would elimi-
nate two of the three Nash equilibria.
Equilibria in mixed strategies

In the Nadal vs. Federer game in Figure 1.6b we said that the best way for Nadal to be
unpredictable was to use his “Left” strategy one third of the time (and his “Right” strat-
egy the remaining two thirds) and that the best way for Federer was to play his “Left”
strategy two thirds of the time. We showed that this was a stable election of strategies.
In fact, it is a Nash equilibrium of the game in mixed strategies. However, we did not
show how we found it. Here we present a technique for computing Nash equilibria
in mixed strategies for games of two players when they each have two strategies. We
reproduce the game in Figure 2.5a for convenience.
The first thing to notice is that for a player to be willing to randomize between two
strategies he must be indifferent between them, otherwise he will choose the one with
31
GAME THEORY
Federer
Left Right
Left 50, 50 90, 10
Nadal
Right 70, 30 50, 50
Figure 2.5a Nadal-Federer game
the highest payoff. Denote by q the probability with which Federer is randomizing
between “Left” and “Right.” The following expression gives the percentage of points
that Nadal wins if he plays “Left” as a function of q, where U N ( Left, q ) should be read
as “the payoff for Nadal when he plays ‘Left’ and Federer chooses ‘Left’ with probabil-
ity q”:
U N ( Left, q ) = 50 × q + 90 × (1 − q ) = 90 − 40q .
If, on the other hand, Nadal uses his “Right” strategy, his payoff in terms of percent-
age of winning points is
U N ( Right, q ) = 70 × q + 50 × (1 − q ) = 50 + 20q .
The two percentages must be the same for Nadal to be willing to randomize:
U N ( Left, q ) = U N ( Right, q ), or
90 − 40q = 50 + 20q .
Solving the last equation, we find

40 = 60q ,
40 2
which gives =
q = as the solution. This means that Federer must choose “Left”
60 3
with probability 2/3 for Nadal to be willing to randomize between his strategies.
Now let us denote by p the probability with which Nadal chooses “Left,” we can
similarly compute the percentage of Federer’s winning points when he plays “Left”
and when he plays “Right”:
U F ( p, L e ft ) = 50 × p + 30 × (1 − p ) = 30 + 20 p,
U F ( p, Right ) = 10 × p + 50 × (1 − p ) = 50 − 40 p .
Equating the two expressions we get the equation

30 × 20 p + 50 − 40 p ,
20 1
with solution =
p = . This means that Nadal must choose “Left” with probability
60 3
1/3 for Federer to be willing to randomize between his strategies.
32
The fact that the pair of individual strategies  1 , 2  is a Nash equilibrium was
3 3 1
shown in Section 1.6. Recall that the basic argument is if Nadal plays p = , then
3
Federer is indifferent between playing “Left,” “Right” or any way of choosing between
the two, which in particular means he would not gain if he deviates from choosing
2
q = , his best way to be unpredictable. The same argument works for Nadal if Federer
3
2
chooses q = .
3
2.5 Complications in finding the Nash equilibria

Nash equilibria in mixed strategies
It is relatively easy to find Nash equilibria in two-player games when each player has just
two pure strategies. However, in more complicated games, finding them may become
quite difficult. In this book, we will deal with just simple cases, either because we deal
with two-player, two-strategy games or because the games have some other particular
characteristic that makes the analysis easy. In all cases, the technique explained along
with the Nadal vs. Federer game will be useful to find the equilibria in mixed strategies.
All in all, the technique has to be used with care. If one applies it to the prisoners’
dilemma game, one will find that it leads to an equation with no solution. The reason
is that one strategy is dominated by the other and, thus, the player can never be indif-
ferent between them, which means there is no equilibrium in mixed strategies. In some
other cases one finds solutions with probabilities than are not in the [0, 1] interval.
Again, this means that there is no equilibrium in mixed strategies.
Nash equilibria with infinitely many strategies

This subsection uses very basic differential calculus. A reader not familiar with this
area of mathematics may disregard it. Very few parts of the book deal with examples
like this.
Let us go back to the game in Section 2.4 between the two neighbors that had to
decide how much time to dedicate to cleaning the street. Now we will allow the neigh-
bors to choose any amount of time between zero and two hours for the task of cleaning
the street, and not just an integer number of hours. There are infinitely many ways to
divide the two hours between cleaning and watching TV, so that each neighbor has
infinitely many strategies to choose from. This poses a difficulty at the time of finding
the Nash equilibria of the game with either method: checking whether the definition
is satisfied in each strategy combination or finding the best response to every strategy
by the opponent. We can try discretizing the two hours’ time. For instance, we can be
satisfied with analyzing the game when time cannot be divided beyond the minute or
the second. In the first case we have 121 strategies for each neighbor, and in the second
case, 7,201 strategies (3,600 seconds in one hour plus the strategy of zero seconds).
With a great deal of patience, we could write the normal form of the game (the table)
and find the equilibria, but there is a better way.
Remember that to find the best reply from Neighbor 1 against a strategy by Neigh-
bor 2 one must use Neighbor 1’s utility function, which is U1 ( c1 , c 2 ) = (1 + c1 + c 2 ) ( 2 − c1 ) .
33
GAME THEORY
For every value of c 2 we want to find the value of c1 that maximizes Neighbor 1’s util-
ity. This is the same thing we would do when looking for best replies in the tables that
represent the games. However we cannot use the table now, as it would have infinitely
many rows and columns. Fortunately, there are other techniques, and the reader famil-
iar with calculus will notice that the problem of finding the best reply for Neighbor 1
implies solving the problem maxU1 ( c1 , c 2 ) :
c1
max (1 + c1 + c 2 ) ( 2 − c1 ) .
c1
The first derivative equated to zero will provide us with the first-order condition for a
maximum:
dU1
= 1 − 2c1 − c 2 = 0.
dc1
From that equation we find the value of c1 that maximizes the utility function depend-
ing upon the value c 2 :
1 − c2
c1 = .
2
This expression is the best reply (or best response) function for Neighbor 1, and it gives
us the same information that we had when looking for best replies directly on the table of
the normal form of the game. For instance, if Neighbor 2 chooses c 2 = 0, 5, the expression
1 − 0, 5
tells us that the best reply for Neighbor 1 is c1 = = 0, 25, or a quarter of an hour.
2
We find the best reply function for Neighbor 2 the same way:
1 − c1
c2 = .
2
Now, if we take the two best reply functions as a system of two equations and two
unknowns, its solution will satisfy the definition of a Nash equilibrium: each neighbor
will be choosing the best reply against what the other one is choosing.
1
In our case the solution c=
1 c=
2 , a third of an hour or 20 minutes.
3
There are three complications that we have left unattended in the middle of the
exposition. They do not alter the solution of this example, but they may be of con-
sequence in some others. First, setting the first derivative equal to zero is a necessary
condition for a maximum, but it is not sufficient. We can say that it finds a candidate
for a maximum. For a sufficient condition, we need the second derivative to be negative
when evaluated at the candidate point, as in the case
d 2U1
= −2 < 0.
dc12
1 − c2
Second, the function c1 = does not completely define the best reply. If Neigh-
2
bor 2 chooses to clean for 1.5 hours, the function would say that Neighbor 1 must ded-
1 − 1, 5
icate c1 = = −0.25 hours. As time cannot be negative, the function is not giving
2
the correct best reply for this and other values of c 2 . When this happens, one has to
specify that Neighbor 1’s best reply is not a negative amount of time, but zero.
34
Third, after solving the system of equations, we must check that the solution is
feasible, that is that its value lies within the limits of what is specified in the problem.
In the example, this would be that the time dedicated to clean is a number between
zero and two hours. Since this is the case, the solution is valid. If it were a number
greater than 2 or smaller than zero, we would need to explore all the possible combina-
tions of extreme values (0 and 2) for both players.
A pure strategy may be dominated by a mixed strategy

Consider again the Nadal vs. Federer game in Figure 2.5a, and add a third strategy for
Nadal: he can now serve aiming to the middle. Federer, however, has the same strate-
gies as before. Say that Nadal wins the point 60% of the time when he plays “Middle”
and Federer plays “Left,” and 66% of the time when Federer plays “Right.” The new
game is depicted in Figure 2.5b.
Federer
Left Right
Left 50, 50 90, 10
Nadal Middle 60, 40 66, 34
Right 70, 30 50, 50
Figure 2.5b Nadal-Federer game with an additional strategy
In this game no player has dominated strategies if we consider that a strategy can
only be dominated by a pure strategy. Observe that, for Nadal, his payoffs in his strat-
egy “Middle” are midway between the payoffs in strategies “Left” and “Right.” This
creates the possibility that by appropriately choosing a mixed strategy between “Left”
and “Right,” Nadal can get higher payoffs than by using “Middle.” To see if this is
indeed the case, next we write the conditions that must be satisfied. We will first find
the expression for Nadal’s utility for a mixed strategy using “Left” and “Right” with
probabilities p and 1 − p, respectively, and then his utility from choosing “Middle.”
First, we do it if Federer chooses “Left”:
uN (( p [Left ], (1 − p )[Right ]), Left ) = 50 p + 70 (1 − p ),

uN ( Middle, Left ) = 60.
Since we want to see if it is possible that the mixed strategy dominates the “Middle”
strategy, we write the condition as
50 p + 70 (1 − p ) > 60,
from which we obtain

1
p< .
2
35
GAME THEORY
Next, we do the same if Federer returns from the “Right”:
uN (( p [Left ], (1 − p )[Right ]), Right ) = 90 p + 50 (1 − p ),

uN ( Middle, Right ) = 66.
Just like before, we also want to write the condition for the mixed strategy to dominate
“Middle”:
90 p + 50 (1 − p ) > 66,
from there
2
p> .
5
Combining the two conditions we see that the strategy “Middle” is dominated by
any mixed strategy using “Left” and “Right” that gives ¨Left” a probability between
2 and 1 .
5 2
Non-existence of Nash equilibria

There are games without Nash equilibria, even if we allow for the use of mixed strate-
gies. The following is perhaps the simplest example to show it. Two players, A and B,
must each choose a number between 0 and 1, but cannot choose the number 1. The
player choosing the highest number wins, but if they both choose the same number,
then no one wins. For example, the strategy combination in which both choose the
number 0.99 is not a Nash equilibrium, as any of them can pick a higher number,
like 0.999. There is no such thing as the highest real number that is smaller than
one, thus the game has no Nash equilibrium. Allowing for mixed strategies does not
help, as one can always choose a number higher than the average given by the mixed
strategy.
In the above example, the lack of equilibrium may not seem important. After all,
the best reply logic leads the players to choose something very close to (1, 1), which will
then be a good approximation to what the players actually may do.
If we modify the game so that players can choose any number (not just between
zero and one), then not only is there no equilibrium, but we also cannot approx-
imate the players’ actual actions – at least, not beyond just saying that they will
choose very large numbers. In a finite real world, this kind of game is not frequent,
and we will be within the limits of Proposition 4, which ensures the existence of a
Nash equilibrium.
2.6 The payoffs of the game and the mixed strategies

In the first pages of Chapter 1, where we explained how to understand the elements
of a game, we insisted that the actual numbers used to represent the utility or pay-
offs of the players did not matter much, as long as they preserved the ranking of the
36
preferences over the results after the different strategy combinations. This is true if
we are only concerned with pure strategies, whether because this restriction is a real-
istic one or because the Nash equilibria only use pure strategies. However, as soon as
we use mixed strategies we need to be more precise with the numbers representing
payoffs. The reader can easily check that the Nash equilibrium in mixed strategies
in the battle of the sexes game in Section 1.3 is different if we change the payoffs
in the cell (Football, Football) from (2, 1) to (3, 1), while the pure strategy Nash
equilibria remain the same.
In some games, like Nadal vs. Federer in Section 1.6, the payoffs are determined
in an objective manner. The same is true for other games in which the payoffs are
amounts of money. However, even if players only care about the money, they may
have different attitudes towards the risk. Let us see with an example what the problem
induced by these attitudes is and how we can deal with it.
Say that Player A is playing against Player B and that she is considering her best
reply against a mixed strategy by Player B. In particular, suppose that Player A must
choose between F and G, and that Player B is mixing between her strategies Y and Z
with equal probabilities. Payoffs in euros are shown in Figure 2.6a. The payoffs for
Player B are not important now and are not shown in the table.
B
Y Z
F 25 25
A
G 0 50
Figure 2.6a Expected payoff
If Player A chooses F, she will get €25 no matter what Player B does. If she
1 1
chooses G, she will get an expected payoff of × 0 + × 50 = €25. If these are the
2 2
relevant payoffs, we will say that both strategies give the same utility. However, in
the first case Player A earns €25 with total certainty, while in the second case the
€25 is an average number. If Player A presents some degree of aversion towards
risk-taking, she may prefer the sure €25 as opposed to the 50–50 chance of getting
€0 or €50. To deal with these issues, game theory draws from decision theory and
represents preferences according to the risk attitude. The most successful way to do
it, although not always the most appropriate, is representation through an expected
utility function. Basically, it says that if the individual preferences over uncertain
outcomes satisfy some reasonable properties, then we can assign a value to every
outcome in such a way that the comparison among different uncertain situations
can be performed just by calculating the average of those values. The values reflect
the risk attitude and are specific to each person. If we know the values for Player
1, we should use them instead of the monetary payoffs in Figure 2.6a and obtain a
different table, like the one in Figure 2.6b.
The numbers in Figure 2.6b are just an example of what those values can be. But
if they happen to be the values representing the utility of the monetary payoffs for
Player A, then her expected utility of choosing strategy F is 5, while her expected
37
GAME THEORY
B
Y Z
F 5 5
A
G 0 7.07
Figure 2.6b Expected utility
1 1
utility of G is × 0 + × 7.07 = 3.535 (remember that Player B is choosing Y and Z
2 2
with equal probabilities). Now we can conclude that Player A prefers F rather than G,
and analyze the game using the values that reflect the utility instead of the monetary
payoffs. In decision theory there are several families of functions to substitute payoffs
for utilities according to the risk attitude of the individual, and these functions allow
the researcher to develop models and simulations. For instance, one such function is
to substitute the payoffs with their square root. This is the function that we have used
to change the payoffs in Figure 2.6a with the utilities in Figure 2.6b, and it is a func-
tion that represents a particular degree of risk aversion. Another function would be the
cubic root of the payoff, which represents an even higher aversion towards risk-taking.

2.1 Consider the games “coordination,” “choice of standards,” “battle of the sexes,”
“chicken” and “prisoners’ dilemma” discussed in Chapter 1. Find what strategies
the players would choose if they were to use either
(a) the maximin criterion or
(b) rationalizable strategies.
(c) In which games do you find a unique solution according to one criterion or
the other?
(Consider only pure strategies.)
2.2 The actual values in the simulation of the Normandy landing by William
Drakert are shown in the figure below (in Figure 2.1 we changed only a few
values in column C).
Germany
A B C D E F
1 13 29 8 12 16 23
2 18 22 21 22 29 31
3 18 22 31 31 27 37
Allies
4 11 22 12 21 21 26
5 18 16 19 14 19 28
6 23 22 19 23 30 34
38
(a) Find the maximin strategy for each player. Is the combination of the two
strategies a Nash equilibrium?
(b) Find the rationalizable strategies.
(c) In the game obtained with the rationalizable strategies, iteratively eliminate
weakly dominated strategies and show the game that is left after completing
this process.
2.3 Find the Nash equilibria in pure strategies in the games “coordination,” “choice
of standards,” “battle of the sexes,” “chicken” and “prisoners’ dilemma” discussed
in Chapter 1.
2.4 Show that there are no pure strategy Nash equilibria in the games in Exercises 1.9
and 1.10.
2.5 Consider the game of the two neighbors described in Section 2.4, but slightly
change the utilities to
U1 ( c1 , c 2 ) = ( c1 + c 2 ) ( 2 − c1 )
U 2 ( c1 , c 2 ) = ( c1 + c 2 ) ( 2 − c 2 )
(a) Write the table that shows the normal form of the game.
(b) Find the Nash equilibria.
(c) Provide an interpretation of the change in the utility functions and how it
affected the equilibrium.
(d) Repeat (a), (b) and (c) with the new utility functions
U1 ( c1 , c 2 ) = (1 + c1 + c 2 ) ( 2 − c1 )
U 2 ( c1 , c 2 ) = ( c1 + c 2 ) ( 2 − c 2 ) .
2.6 The treasurer and bookkeeper of a certain political party, Mr. Luis Barreras,
has to launder one million euros, the bounty of some illicit activities. He can do
it through a company with an opaque account in a tax haven or, alternatively,
through different friends and family as front men. The first option costs one
quarter of a million euros, while the second one costs half a million. The Tax
Agency has resources to inspect just one of the two possibilities. If it inspects
the option that Barreras did not use, he will keep his net benefits and the Tax
Agency will not collect a single euro. If the Tax Agency inspects the option that
Barreras has used, it will locate the money and will seize all that was left in
Barreras’ possession, and it will, in addition, impose a penalty of half a million
euros.
(a) Represent the normal form of the game.
(b) Calculate the Nash equilibria (both in pure and mixed strategies).
2.7 Find the Nash equilibria in mixed strategies of the games “coordination,” “choice
of standards,” “battle of the sexes,” “chicken” and “prisoners’ dilemma” discussed
in Chapter 1.
39
GAME THEORY
2.8 Consider the popular game of rock-paper-scissors in Exercise 1.9.

(a) Suppose that Player 2 plays “rock” and “paper” with probabilities ½ and ½.
Which is Player 1’s best reply?
 1 1 1  1 1 1 
(b) Show that   [R ], [P], [S] ,  [R ], [P], [S]  is a Nash equilibrium
 3 3 3   3 3 3 
in mixed strategies.
2.9 Repeat points (b) and (d) in Exercise 2.5 when the neighbors can choose any
amount of time between zero and two hours as in Section 2.5. (Some calculus is
required.)
2.10 Consider the game between Nadal and Federer in Figure 2.5a, and add a third
strategy for Federer: he can now return from the middle. Nadal, on the other
hand, has only his original two strategies. If Federer uses his new strategy, he
wins the point 30% of the time if Nadal serves towards the left and 35% if Nadal
serves towards the right.
(a) Represent the normal form of the new game.
(b) Show whether the new strategy “Middle” for Federer is dominated by some
pure or mixed strategy.
2.11 Two armies fight a war. Army 1 has 4 divisions, while Army 2 has only 3. The
war is decided on two battlefields, and each army has to decide how many divi-
sions to locate on each one, where the army with more divisions wins the battle.
If the number of divisions is the same, then no one wins. Consider that the deci-
sions are made simultaneously and that payoffs are calculated as the number of
won battles minus the number of lost battles.
(a) Which are the strategies for each army? Indicate them as a pair of numbers,
the first for the number of divisions on Battlefield 1, and the second for divi-
sions on Battlefield 2.
(b) Write the normal form of the game.
(c) Iteratively eliminate weakly dominated strategies.
(d) Calculate the equilibrium in the game obtained in (c).
40
3 STATIC GAMES
In this chapter, we present a sample of games in the areas of sociology, political science
and economics. Their purpose is to showcase the topics where game theory sheds light
into complex, and sometimes paradoxical, problems. The reader should pay special
attention to the way we go from a verbal description of a situation to the formal writing
of the game. The search for Nash equilibria is almost always conducted using the “best
reply” concept but showing different techniques to find it. There are also cases in which
we look for the Nash equilibria using the rudimentary technique of “trial and error.”
It is important to realize that in all cases we are using the same formal definitions of a
game and of equilibrium. Finally, some games can show how given behavioral dynam-
ics lead to the equilibrium. Even if some particular dynamics are not very realistic,
these examples should convince the reader that the study of these dynamics is possible
and that they constitute a research line within game theory. From now on, Nash equi-
libria will often be abbreviated as NE.
More specifically, the games in Sections 3.1, 3.6 and 3.7 are of special interest in
economics. Sections 3.2 and 3.3 show one of the best-known applications of game theory
in political science, and Sections 3.4 and 3.5 present study cases in sociology. The games
are ordered to illustrate different techniques, problems and complications. The game of
fiscal battles in Section 3.1 is a straightforward application of what we have already seen
in the previous chapters and shows an interesting paradox of the kind that abound in
game theory. In this case, we see how improving the payoffs of the game may have grim
consequences for the players. Our second game, the median voter, allows us to explain
how to deal with many strategies – up to infinitely many of them – thanks to the appreci-
ation of the best reply concept. The game in Section 3.3 adds the complication of having
three players. Sections 3.4 and 3.5 pay special attention to the modeling of a game. In the
first one, the broken windows theory, we will deal again with infinite strategies, but this
time without the help of a previous, simpler version. In the second one, the independence
of Sylvania, we pay special attention to the realistic and relevant aspects of the game.
The Cournot model in Section 3.6 shows yet other ways of looking for best replies and
equilibria. It shows also our first encounter with two dynamics that sometimes lead to an
NE. Finally, and just for the mathematically inclined reader, it introduces the use of basic
calculus to find the NE. In the last game, the Bertrand model, we discuss the realism of
the assumptions vs. the realism of the results.
3.1 Fiscal battles
Two neighboring countries must decide how to tax a firm that is expected to make 100
in profits. The firm will be located in the country with the lowest taxes. Should taxes
be the same, it will decide at random where it will go. According to international agree-
ments, these two countries can impose only a 0%, 10% or 20% tax rate. Furthermore,
the country where the firm settles will receive an additional benefit of 8 because of the
new jobs created in the country. The game is illustrated in Figure 3.1a. For instance, if
41
GAME THEORY
both countries set a 10% tax, the payoff for the country where the firm settles will be
0.1 × 100 + 8 = 18, but since this happens with probability ½, the expected payoff is 9.
The other values in the table are computed similarly.
Country 2
0% 10% 20%
0% 4, 4 8, 0 8, 0
Country 1 10% 0, 8 9, 9 18, 0
20% 0, 8 0, 18 14, 14
Figure 3.1a Fiscal battle
The reader can easily check that there are two pure strategies NE (Nash equilibria)
in the game: (0%, 0%) and (10%, 10%). Notice also that the strategy of taxing 20% is
a weakly dominated strategy (by the 10% strategy). In this case, eliminating it for each
player does not eliminate any NE.
If the profits for new jobs are not 8, but 12, the game changes to be the one depicted
in Figure 3.1b. Now the only NE is (0%, 0%), which can be found by iteratively elim-
inating weakly dominated strategies. For each country, we first eliminate the 20% tax
(weakly dominated by 10%) and then eliminate 10. A remarkable feature of these two
fiscal battle games is the fact that, when the potential profits for the countries increase
(the benefit of new jobs), only the worst equilibrium survives.
Country 2
0% 10% 20%
0% 6, 6 12, 0 12, 0
Country 1 10% 0, 12 11, 11 22, 0
20% 0, 12 0, 22 16, 16
Figure 3.1b Fiscal battle with high job profits
3.2 The median voter

There are two political parties in Freedonia, the Leftist Party (LP) and the Rightist
Party (RP). The country’s citizens are uniformly distributed along the left-right polit-
ical spectrum measured from the far left (0) to the far right (5). Each citizen votes for
the party with the election platform closest to their ideology. In the case where they see
no difference they will vote for either party randomly. If the only interest of the parties
is to maximize the number of votes, in which ideological point will they place their
platforms? We will address this question modeling the situation as a game, and we will
do it with two different assumptions for the reader to compare.
Model 1: citizen ideology can take any value between 0 and 5, but the electoral plat-
form’s ideology can only be measured in integer values.
The model is described in Figure 3.2, where payoff numbers are the percentage of
votes. To understand how they have been computed, take the example of the strategy
42
STATIC GAMES
profile (1, 3), where LP is located at the ideological point 1 and the RP is located at 3.
LP will take all the votes of citizens with an ideology positioned between 0 and 1 (20%
of citizens) and half the citizens between 1 and 3 (the half located between 1 and 2, the
midpoint, which makes another 20%), with a total of 40% of the votes. The remaining
60% will vote for the RP. Finding the NE of this game applying the definition means
checking each of the 36 strategy combinations. In games like this we appreciate the
technique of looking for the best replies, which we trace by using the technique of
underlining the corresponding payoff, as we saw in Section 2.4.
RP
0 1 2 3 4 5
0 50, 50 10, 90 20, 80 30, 70 40, 60 50, 50
1 90, 10 50, 50 30, 70 40, 60 50, 50 60, 40
2 80, 20 70, 30 50, 50 50, 50 60, 40 70, 30
LP
3 70, 30 60, 40 50, 50 50, 50 70, 30 80, 20
4 60, 40 50, 50 40, 60 30, 70 50, 50 90, 10
5 50, 50 40, 60 30, 70 20, 80 10, 90 50, 50
Figure 3.2 Electoral platform choice
We find four cells where we underlined all payoffs, which correspond to four
NE in pure strategies: (2, 2), (2, 3), (3, 2) and (3, 3). Because of the way the model
was specified, there is no way to propose one of the equilibria as the best in any
sense. If we go back to the real world of Freedonia we may find that, besides maxi-
mizing the number of votes, parties have some other goals, like showing particular
characteristics that separate them from other parties. We may also know that the
LP will never be located to the right of the RP. If these are reasonable assumptions,
we may find that (2, 3) is the most appropriate NE. Regardless of these later con-
siderations, the characteristic feature of all equilibria is that parties tend to offer
similar platforms.
If we repeat the exercise with more ideological points, between 0 and 10 for instance,
the only equilibrium will be (5, 5). To show this is indeed the case, we can form a
table with 11 strategies for each party and 121 combinations or we can try to find the
equilibria without writing such a large matrix. We will be able to do that by taking
advantage of what we already know about how the best reply works in this game.
The midpoint between 0 and 10 is point 5. If the LP places itself in any point between
0 and 4, we know that the best reply for the RP is to offer a platform in an ideological
position just one point to the right of the LP (i.e., the next number) to maximize the
number of voters. It will attract all the voters to the right of its chosen point plus half
of the voters ideologically placed in the range between the two parties. If the LP set-
tles on point 5, the RP will also place itself on that number. If the LP chooses a point
between 6 and 10, the RP will choose a number immediately to the left. As the game
is symmetric, the above discussion also defines, mutatis mutandis, the best reply of the
LP as a function of the position of the RP.
43
GAME THEORY
After understanding the best reply functions, the next objective is finding through
which strategy combination each party chooses its best reply against the other’s choice.
Let us see:
• If LP chooses 0, the best reply for RP is 1. However, if RP chooses 1, the best reply
for the LP is not 0, but 2. Thus, there is no equilibrium with LP choosing 0. We get
a similar conclusion for any other choice between 1 and 4 by the LP.
• If the position of LP is 5, then RP’s best reply is also 5, and if RP is at 5, LP’s best
reply is also 5. Therefore, (5, 5) is a Nash equilibrium.
• If the LP chooses 6, the best reply from the RP is to choose 5. However, if RP
chooses 5, the best reply for the LP is not 6, but 5. Therefore, there is no equilibrium
in which LP chooses 6. For numbers between 7 and 10 we obtain a similar result.
The best way to conclude that the parties will offer similar platforms is to understand that
if they offer different platforms, each party will tend to get closer to the other. For instance,
if the RP is located to the right of the LP and gets a little closer to the LP, it will keep the
voters from its right and will gain some voters from its left. Being separated cannot be a
Nash equilibrium. The parties cannot be together in any other point that is not in the
middle, as one party can gain voters by moving a little to the place with the most voters.
The reason we had several equilibria with ideological points between 0 and 5 and
just one when points were between 0 and 10 is because in the first case there is an even
number of points, with two points in the middle, whereas in the second case there is
an odd number of points, with only one in the middle. With two points in the middle,
any placement in those points will constitute an NE.
Model 2: the ideology of the voters and the electoral platforms can take any value be-
tween 0 and 5.
Now we have infinitely many places for each party to place itself. To even attempt a
representation of the game using a table is impossible. The analysis, however, is rela-
tively simple:
• In an NE, the parties cannot be in different positions. If they were, the party most to
the left would gain voters by moving to the right. The party most to the right would
also want to move towards the other party, but it is enough to find one party that
wants to move to show that the position is not an NE.
• In an NE, the parties cannot be together in a place other than the middle point
(2.5). Say they are both located at the ideological point 2.4, each having 50% of
the votes. By moving a little to the right, say to 2.41, one party can get a little more
than 50% of the votes. A similar argument works if they are both together, but to
the right of point 2.5.
• Both located at 2.5 is an NE. At this point they each obtain 50% of the votes. If
either were to move they would lose some votes. Thus, the only NE is (2.5, 2.5).
As the reader may have appreciated, the model with infinitely many strategies is
not more complicated than its finite counterpart. In fact, once we understand and are
familiar with the best reply function, the infinite version is easier to work with. Since
the results in both models are the same (a tendency of the parties to meet in the politi-
cal center), the researchers prefer the infinite version of the model.
44
STATIC GAMES
Both models are unrealistic. Most certainly parties have other interests besides max-
imizing votes, some voters may be discouraged from voting if the parties show similar,
centered ideologies, or there may be more than two parties, among other possibilities.
All in all, these simple models show that, in addition to other considerations that can
enrich the analysis, there is a tendency towards moderate, centered positions that can
be explained by strategic and not ideological arguments.
The choice of an electoral platform is just one example of a location game. In
economics, similar models explain why there are two gas stations close together in the
middle of a road, rather than in a more dispersed layout. The reason the primary TV
channels have similar programs at the same time can also be explained by the incen-
tives to attract the median viewer.
3.3 The advantage of being indifferent

The seats in some parliament are divided between three parties. There is a left-wing
party (LP) and a center party (CP) each with 30 seats. There is also a right-wing party
(RP) with 40 seats. They must vote to choose the prime minister (PM) of the gov-
ernment, and each party presents a candidate. Parties prefer that the PM be as close
as possible to their ideology, and CP, in addition, prefers a left-wing PM rather than
a right-wing one. The leader of each party must decide which candidate to vote for
independently of the choices made by the other parties. All members of parliament are
disciplined and will follow their leader’s decision. The candidate with the most votes is
elected PM. Which are the NE?
The game is represented in Figure 3.3a. The LP must choose a row, which must be
the same in all three tables. Similarly, the CP must choose a column, also the same on
all tables. Finally, the RP must choose a table. The strategies L, C and R refer to voting
for the candidate of the LP, CP and RP, respectively. Each party gets a payoff of 2 if
their most preferred candidate is elected, of 1 if their second-most preferred is elected,
and of 0 if their least preferred candidate is elected. As an example, note that if all three
parties vote for their candidate (i.e., the vote is (L, C, R), where the components are the
votes by LP, CP and RP in that order), we will be in the first row of the second column
of the third table. Payoffs are (0, 0, 2), as RP sees how its preferred candidate is elected,
while for both LP and CP that candidate is their least preferred.
CP CP CP
L C R L C R L C R
L 2, 1, 0 2, 1, 0 2, 1, 0 2, 1, 0 1, 2, 1 0, 0, 2 2, 1, 0 0, 0, 2 0, 0, 2
LP C 2, 1, 0 1, 2, 1 0, 0, 2 1, 2, 1 1, 2, 1 1, 2, 1 0, 0, 2 1, 2, 1 0, 0, 2
R 2, 1, 0 0, 0, 2 0, 0, 2 0, 0, 2 1, 2, 1 0, 0, 2 0, 0, 2 0, 0, 2 0, 0, 2
L C R
RP
Figure 3.3a CP prefers the LP candidate over the RP one
At first glance, this game looks complicated. To avoid being overwhelmed by com-
plicated games, try different strategy combinations and determine the incentives for
45
GAME THEORY
players to deviate from them. If we find none, we were lucky and the strategy is an NE.
Most likely we find that one or more players want to deviate from it and, thus, that the
chosen strategy is not an NE. The good news is that even in this case we are learning
something about the best replies and about the strategic nuances of the game. With
that learning, our next guess for an equilibrium will probably provide better results, or
at least we will be able to analyze more rapidly the rest of the combinations to find the
equilibrium.
In voting games, it is normally good to start with the strategy combination in which
each party votes for its preferred alternative. However, in this game we will proceed
differently to illustrate a couple of important details of simultaneous majority voting
games that will prove helpful in their analysis: (i) if no player has a majority of votes,
all players voting for the same candidate is an NE and (ii) voting for the least preferred
alternative is a weakly dominated strategy.
Let us begin, then, by showing that the strategy profile in which everyone votes
for the RP candidate, (R, R, R) – third row, third column, third table – is indeed a
Nash equilibrium. The RP candidate is elected, and no unilateral deviation changes
that. There is something strange, though, in this equilibrium, as it requires that both
LP and CP vote for their least preferred candidate. The profile (L, L, L) is also a Nash
equilibrium for the same reason, and presents a similar problem, with RP voting for
its least preferred candidate. This intuition of things going strange in those equilibria
is translated into the fact that voting for the least preferred candidate is a weakly dom-
inated strategy. Note that all payoffs for LP in its strategy R, the first number in the
third row in all tables, are never higher than its corresponding payoffs in its strategy L.
Something similar occurs for CP’s payoffs in its strategy R, third column in all tables,
dominated by payoffs of strategy C, in all its second columns. Finally, RP’s payoffs in
its strategy L, the first table, are dominated by its corresponding payoffs in strategy R,
the third table. If we eliminate those weakly dominated strategies, the game is much
simpler, as we see in Figure 3.3b.
CP CP
L C L C
L 2, 1, 0 1, 2, 1 2, 1, 0 0, 0, 2
LP
C 1, 2, 1 1, 2, 1 0, 0, 2 1, 2, 1
C R
RP
Figure 3.3b Game in Figure 3.3a after eliminating weakly dominated strategies
In the new game, voting for the least preferred candidate between those that
remain as options is not a weakly dominated strategy in general. For LP, for instance,
voting for C is a better option in case CP votes for C and RP votes for R. However,
for the RP to vote for C is a weakly dominated strategy, and consequently we could
eliminate the first table if we continue using our criterion of not playing weakly
dominated strategies. We are not going to do it, however, as the game is already very
simple and because we want to practice finding best replies and Nash equilibria in
46
STATIC GAMES
games with three players (and two tables). Let us start with all strategy combinations
to check whether they are NE:
• (L, L, C) is not an NE because CP would deviate to C.
• (L, L, R) is an NE.
• (L, C, C) is not an NE because RP would deviate to R.
• (L, C, R) is not an NE because LP would deviate to C or CP would deviate to L.
• (C, L, C) is not an NE because LP would deviate L or RP would deviate to R.
• (C, L, R) is not an NE because LP would deviate C or CP would deviate to L.
• (C, C, C) is an NE, but RP is playing a weakly dominated strategy.
• (C, C, R) is an NE.
We find two Nash equilibria: (L, L, R), where the left-wing candidate is elected, and
(C, C, R), where the center candidate is elected. There is a third equilibrium, (C, C, C)
in which the center candidate is also elected, but the RP would be playing a weakly
dominated strategy, making this equilibrium unlikely.
Let us change the exercise to make CP indifferent between a left-wing candidate
and a right-wing one. The new game is depicted in Figure 3.3c.
CP CP CP
L C R L C R L C R
L 2, 1, 0 2, 1, 0 2, 1, 0 2, 1, 0 1, 2, 1 0, 1, 2 2, 1, 0 0, 1, 2 0, 1, 2
LP C 2, 1, 0 1, 2, 1 0, 1, 2 1, 2, 1 1, 2, 1 1, 2, 1 0, 1, 2 1, 2, 1 0, 1, 2
R 2, 1, 0 0, 1, 2 0, 1, 2 0, 1, 2 1, 2, 1 0, 1, 2 0, 1, 2 0, 1, 2 0, 1, 2
L C R
RP
Figure 3.3c CP is indifferent between LP and RP candidates
Notice that for CP both L and R are weakly dominated by C. For LP, strategy R is
weakly dominated by L, and for RP strategy L is weakly dominated by R. After elimi-
nating these strategies, the game is the one in Figure 3.3d.
CP CP
C C
L 1, 2, 1 0, 0, 2
LP
C 1, 2, 1 1, 2, 1
C R
RP
Figure 3.3d Game in Figure 3.3c after eliminating weakly dominated strategies
47
GAME THEORY
In this game, the Nash equilibria are (C, C, R) and (C, C, C). In either of them
CP’s candidate is elected. In (C, C, C), the RP is playing a weakly dominated strategy,
and so we could discard it. Observe how the fact that CP is indifferent between the
candidates of the LP and the RP, rather than preferring the LP’s, changes the situation
from one with two equilibria, and two possible winners, to another with only one equi-
librium in which CP has its candidate elected.
3.4 The broken windows theory

In a given neighborhood, 1000 people share and enjoy a park. They all prefer a clean
park, but taking the trouble to find a trash can is costly. The benefits for enjoying a
clean park have a value of 1000 for each neighbor if all neighbors are tidy. For every
neighbor that shows a careless attitude towards trash disposal, the park loses one unit
of value. Being a tidy citizen is not free: it costs each tidy citizen five units (think of
it as the cost of being careful). On the other hand, being a careless citizen is not free
either: if a neighbor decides to be careless, he or she will suffer a moral disapproval
m
proportional to the tidiness of the park, given by 20 × , where m is the number of
1000
clean neighbors. This disapproval is a cost to the careless neighbors and must be subtracted
from her or his valuation.
This game is simply too big to even try a representation using tables and matrices.
Instead, we may learn about the strategic issues by considering different situations.
Since all neighbors feel the same about the enjoyment of the park and about the
physical and moral costs, let us begin by considering the cases in which they all do
the same. Afterwards, we will consider the case in which they make different choices.
All neighbors choose to be clean

In this situation, each neighbor has a benefit of 995, the result of a valuation of 1000
for having a 100% clean park and subtracting the cost of being tidy. No moral disap-
provals are suffered, and therefore there is nothing else to subtract. For this situation
to be an equilibrium, no neighbor should fall into the temptation of being careless and
stop making the effort to use the trash cans. If one does, she or he will enjoy a park
that is valued in 999 (recall that the value of the park is reduced in one unit for every
999
non-cooperative neighbor) and will suffer a moral disapproval of 20 × = 19.98 .
1000
The net benefit will be 999 − 19.98 = 979.02, less than the 995 if she or he were to keep
being clean. Therefore, we conclude that this situation is a Nash equilibrium.
All neighbors choose to be careless

The park is worth zero to the neighbors, no one pays a cost for being clean and no
0
one suffers a moral reprobation (20 × = 0 ). In this situation, the benefit to each
1000
neighbor is zero. If one now decides to become clean, he or she will enjoy a park that is
worth one unit, will suffer a cost of zero and no moral reprobation. The net benefit will
be 1 − 5 = −4, smaller than the zero benefit of being unclean. This second situation is
also a Nash equilibrium.
48
STATIC GAMES
Some are clean, some are not

Let us have n clean neighbors and 1000 − n unclean. For this situation to be an equilib-
rium, we need for a clean person to not want to be careless, and vice versa. Let us begin with
the clean neighbors. Each of them gets a benefit of BC = (1000 − (1000 − n ) ) − 5 = n − 5.
If one of them changes his or her mind to be unclean, the new benefit to this neighbor
n − 1 49
will be BU ′ = (1000 − (1000 − ( n − 1) ) ) − 20 × = ( n − 1). For this deviation not
1000 50
49
to be beneficial we need n − 5 ≥ ( n − 1), which means n ≥ 201.
50
Consider now the unclean neighbors. Each gets a benefit given by BU = (1000 −
n 49
(1000− n)) − 20 × = n. If one of them chooses to be a good citizen, her or
1000 50
his new benefit will be BC ′ = (1000 − (1000 − ( n + 1) ) ) − 5 = n − 4. This deviation does
not pay if 49 n ≥ n − 4 , or n ≤ 200. Since there is no way to satisfy the two conditions
50
n ≥ 201 and n ≤ 200 at the same time, we conclude that there is no equilibrium in
which some are clean and some are not.
Another way to interpret this analysis is as follows. If there are 201 or more clean
neighbors, in the end all of them will be clean, as the cost that one saves for being care-
less does not compensate for the reprobation of being unclean. In a similar vein, if there
are 200 or less clean neighbors, one by one all of them will stop being clean, as the cost
does not compensate the benefit of keeping the park clean.
There are two equilibria: in the first one, everyone is a good citizen and gets a
benefit of 995; in the second, all are careless and get a benefit of zero. Which
equilibrium prevails in the neighborhood will depend on circumstances not included
in the description of the game. If, historically, the park has always been dirty (because
the previous neighbors had other priorities or because there were no maintenance
services, to simply name a couple of possible causes), the park will continue to be in
poor condition unless something happens that brings neighbors together to coordinate
in the good equilibrium. Similarly, if the park is well kept (perhaps there was a cleaning
campaign), the first neighbors to consider being careless and not use the trash cans will
know that they will suffer a high degree of reprobation by everyone else.
Reality will always be more complicated than the example. Not all people have
the same attitudes; some will not feel the cost of moral disapproval by fellow neigh-
bors, while others will consciously decide to be clean no matter what, and so on.
Still, the game shows another example of a society living in an efficient equilibrium
due to causes unrelated to the inhabitants’ preferences. With all the differences
between the examples, the ban on smoking in public spaces can be defended as one
way to progress from an inefficient situation to a more efficient one. When that
ban is ambiguous and leaves wide room for interpretation, as was the case in Spain
with the 2006 non-smoking law, it is not useful to make the equilibrium change.
When the rules are clear, as was the case with the new 2011 law, the change can
occur almost immediately and with little monitoring, as everyone has reasons to
expect that every other citizen will abide by the law. Fines for infringement are part
of the cost for deviating, but the fact that there are few deviators – much fewer than
49
GAME THEORY
when compared to the situation with the ambiguous law – means that the coordi-
nation problem has been solved in a satisfactory manner. Very few smokers want to
go back to the previous situation.
The broken windows theory was first introduced by social scientists James W ilson
and George Kelling in an article published in The Atlantic magazine (Wilson and Kel-
ling, 1982), and it was later further developed in the book Fixing Broken Windows:
Restoring Order and Reducing Crime in Our Communities by George Kelling and Cath-
erine Coles (Kelling and Coles, 1996). The expression “broken windows” comes from a
famous paragraph in Kelling and Wilson’s article, which first asks readers to “Consider
a building with a few broken windows.” It goes on to explain the lack of supervision peo-
ple infer from the visible damage and then the escalation of vandalism which follows.
This theory has been applied in cities such as New York, where the authorities pro-
ceeded to clean all the trash and garbage from the streets and all the graffiti in the subway
system. It is still debated to this day whether these policies were actually responsible for the
drop in crime that subsequently took place. What matters for the purposes of this book is
the manner in which game theory allows a formal representation of the broken windows
theory and shows its internal logical consistency. Whether this theory (or hypothesis, to
use a more proper expression) is proven valid or not is an empirical question.
3.5 The independence of Sylvania

Freedonia is a federal country. One of its states, Sylvania, is governed by a party that
seeks its political independence and which is considering the following two ways to
advance its agenda:
(i) Call for a referendum with the question: Do you favor the independence of Sylvania?
(ii) Call for a political campaign for civil disobedience, through which citizens would
pay their taxes to the newly created Sylvanian Tax Agency.
Sylvanians are divided in three groups of the same size according to their preferences on
the independence issue. Each citizen, however, will be free to make their own choices.
Group A: People in this group do not approve of the independence of Sylvania. They do
not believe that a positive result in the referendum would result in its independence, for
it would not be recognized by Freedonia; however, they do believe that it would help
negotiate better fiscal benefits for Sylvania, something that they do want. They are also
against civil disobedience, for they believe it would make the situation worse.
Group B: In this group, people believe that a favorable result in the referendum would
ultimately force Freedonia to recognize the independence of Sylvania, which they want.
They also believe that if more than half of the population in Sylvania participates in the
disobedience campaign, the state would also achieve independence. They are aware that
anyone participating in the disobedience campaign would suffer a personal cost, as they
may also be forced to pay the taxes, plus penalties, to the Freedonian Tax Agency. Even
if Sylvania were to eventually become independent, Freedonia could continue to force
tax collection from its citizens for a while until that happens.
Group C: This is the staunch pro-independence group. They are committed to vote “Yes”
in the referendum or to participate in the disobedience campaign regardless of costs and
benefits.
50
STATIC GAMES
Case 1: The referendum

In Figure 3.4 we represent the referendum game if this option were ultimately adopted
by the Sylvanian local government. Here the payoffs reflect that all groups prefer a
“Yes” outcome rather than a “No,” even though their preferences result from different
motives. Recall that people in the first group believe that a Yes result is just a way to
force the federal government to grant better fiscal benefits to Sylvania.
There are five Nash equilibria: (Yes, Yes, Yes), (Yes, Yes, No), (Yes, No, Yes), (No,
Yes, Yes) and (No, No, No), out of which the last four include at least one group voting
for its least preferred option, a weakly dominated strategy. If we discard those equilib-
ria, we are left with the one in which all groups vote for independence.
B B
Yes No Yes No
Yes 1, 1, 1 1, 1, 1 1, 1, 1 0, 0, 0
A
No 1, 1, 1 0, 0, 0 0, 0, 0 0, 0, 0
Yes No
C
Figure 3.4 Sylvanian referendum
Case 2: Civil disobedience

This is a more complex problem. A first approximation could be the game in Fig-
ure 3.5, where each group has to decide whether to participate or not in the campaign
for civil disobedience.
B B
Participate No Participate No
Participate 0, 1, 1 0, 1, 1 0, 1, 1 1, 0, 0
A
No 0, 1, 1 1, 0, 0 1, 0, 0 1, 0, 0
Participate No
C
Figure 3.5 Sylvanian civil disobedience
In this game, the only Nash equilibrium in non-weakly dominated strategies is

(No, Participate, Participate), with the result that a majority of the population adheres
to the disobedience campaign. However, this analysis does not take into account that it
is not the group that makes the decision whether to participate, but rather the individ-
uals. We know that individuals in Group A do not want independence, and therefore,
they have no reason to participate in the disobedience. We also know that people in
Group C will participate no matter what. These two groups can be modeled as if they
were one single decision maker.
51
GAME THEORY
However, an individual in Group B faces a dilemma: she wants the campaign to

be successful, but she may have to pay a personal cost for that. Say that in this group
there are one million individuals, then it would be enough that 500,001 participate
in the civil disobedience campaign (remember that all groups are of equal size, that
in Group C all participate, and that it is enough that over half of the population
participates to achieve independence). Let us consider three possible strategies within
Group B.
All persons in Group B participate in the campaign: This cannot be a Nash equilibrium, as
any person in the group can avoid the penalty for participating in the campaign with-
out being of consequence to the outcome. She can avoid the cost and still benefit from
everyone else’s participation.
No one participates: This situation is a Nash equilibrium. There is no independence and
there would still be no independence even if one person did participate. Therefore,
anyone changing their mind from not participating to actually participating would
suffer the cost without seeing any benefit.
Some participate, some others do not: The only possibility of a Nash equilibrium would
be if exactly 500,001 were to participate and if 499,999 were to not. Since they
would be a majority, the campaign would be successful and they would achieve
independence. If any of the individuals that participated decided not to, there would
no longer be a majority and the campaign would fail. If the personal benefit derived
from living in an independent country were higher than the costs of participating in
the campaign, then no individual currently willing to participate would change their
mind. However, this is a very fragile equilibrium. It relies on the very small chance
that just that exact number participate in the campaign, so that every single partici-
pant is indispensable to its success. If the number of participants were to be different,
then each individual would decide to change their mind. If the number were to be
higher, then any individual would be able to retire their support without being of
consequence to the benefits, while also saving their own costs. If the number were
to be lower, then there would only be costs, and there would be no benefits from
participating.
There is another factor that must make us doubt whether this is an interesting equilib-
rium. Recall that the game is a stylized approximation to reality. In the real world, it
is very unlikely that a campaign is considered successful if 500,001 people participate,
but a failure if only 500,000 do so. Further, it is also unlikely that individuals are aware
that participation happens to be precisely at this delicate balance. A more realistic
scenario is that the probability of success increases with participation. Thus, when the
number of participants gets closer to a number that makes success likely, one particular
individual may optimistically think that his participation can make a difference by
slightly increasing the probability of success. In this case, the individual would be will-
ing to pay the cost in exchange for a small increase in the probability of success. Again,
for most realistic calculations, the cost will be greater, and there will be no equilibrium
in which individuals in Group B participate.
This example should make us aware of the importance of distinguishing between
the interests of the group and personal interests.
52
STATIC GAMES
3.6 Cournot oligopoly
Two firms, A and B, compete in a market with a demand shown in Table 3.1, where we
can see the quantities that consumers will buy at different market prices. For example,
if the price is €4 per unit of product, firms can expect to sell 60 units. Another way to
read the table is by starting with a quantity and reading the corresponding price as the
one that will be fixed in the market for all that quantity to be sold. The table shows that
a quantity of 100 or higher cannot be sold at a positive price.
Table 3.1 Market demand
Price 0 1 2 3 4 5 6 7 8 9 10
Quantity 100 90 80 70 60 50 40 30 20 10 0
The oligopolistic model of competition developed in Cournot (1838) assumes that

firms decide how much to produce without knowing the decisions of the other firms. To
make the problem simpler, we will assume that quantities can only be produced in mul-
tiples of 10 units. Once firms decide on their production and deliver it to the market, the
selling price will be determined by the total quantity according to the table. To finish the
description of the problem, assume that each firm incurs a cost of €1 for each unit of prod-
uct. Thus, if Firm A produces 20 units and Firm B produces 10, the total production will
be 30, which will be sold at a unit price of €7. Since each firm has to pay a cost of €1 per
unit, they will be making a €6 profit on each unit. Firm A will have profits of €120, and
Firm B will make €60. The normal form of the game requires a table with 11 rows (quan-
tities from 0 to 100 in multiples of 10) as strategies for Firm A, as well as 11 columns for
Firm B. There will be 121 strategy combinations to check for Nash equilibria. Instead of
inspecting so many cells in such a large table, we instead proceed to find the best replies.
• Firm B produces 90 or 100 units: the best reply for Firm A is to not produce at all,
as any quantity will make the total production greater than 90 units, a number that
can only be sold at a lower-than-cost price.
• Firm B produces 80 units: the least that Firm A can produce is 10 units, and earn
zero, the same as if it does not produce at all. Both are best replies. Producing more
than 10 results in losses.
• Firm B produces 70 units: the best reply for Firm A is to produce 10. Total market
units will be 80, which can be sold for €2 per unit. Firm A will make €10 in profits.
More units being sold results in a price equal to or lower than the cost.
• Firm B produces 60 units: if A produces 10 units, the market price will be €3, and
A will earn €20 after costs. If A produces 20 units, it will also earn €20. Both are
best replies.
• Firm B produces 50: if A produces 10, it will earn €30; if it produces 20, it will earn
€40; and if it produces 30, it will earn €30. The best reply is to produce 20 units.
We continue doing the same for all remaining quantities. In Table 3.2 we show all best
replies of Firm A for any production by Firm B and, calculated in a similar fashion, the
best reply for Firm B after the different production decisions by Firm A.
53
GAME THEORY
Table 3.2 Best replies for the Cournot game

B’s production 90 80 70 60 50 40 30 20 10 0
A’s best reply 0 0/10 10 10/20 20 20/30 30 30/40 40 40/50
A’s production 90 80 70 60 50 40 30 20 10 0
B’s best reply 0 0/10 10 10/20 20 20/30 30 30/40 40 40/50
To find the Nash equilibria we need to find a pair of quantities, (X, Y), where X is
the quantity produced by Firm A and Y the quantity produced by Firm B, such that
Y is the best reply against X and X is the best reply against Y. Let us start with Firm B
producing 80. A’s best reply is 0 or 10. But neither of the best replies by Firm B against
these quantities is 80. Therefore, we conclude that there are no Nash equilibria in
which Firm B produces 80. After analyzing the tables, we find three Nash equilibria:
(30, 30), (40, 20) and (20, 40). In all of them the market price is €4 and net profits per
unit sold is €3. In the first equilibrium, each firm makes €90 in profits. In the second
one, Firm A earns €120 and Firm B earns €60; while in the last equilibrium, it is the
other way around.
An interesting feature in this game, common to most Cournot models, is that we
can find at least one equilibrium if we start with an arbitrary production by one firm
and then take successive best replies. For instance, if we start with Firm A’s decision
to produce 20 units, the best reply by B will be to produce 30 or 40 units. Take any of
those numbers, say 30. Now A’s best reply against 30 is 30, and B’s best reply against
30 is also 30. Thus, we have already found a Nash equilibrium. The reader can repeat
the process starting in any other quantity to find an equilibrium also. This iterative
process is called “adaptive dynamics.” This is another example in which we have both
an equilibrium as well as a behavior that leads to that equilibrium. Suppose that the
firms play the game for several rounds and that they choose their quantities by turns,
then each firm’s behavior according to this adaptive dynamics is to produce the best
reply against what the other firm produced in the last round. This behavior is not very
realistic, since firms act myopically. However, it shows a way to look for a behavioral
theory through which to reach the equilibrium.
There is another dynamics that leads to the equilibrium in this game. In Table 3.2
we can observe that, among the firms’ best replies, there are no quantities between 60
and 90. This means that no matter what the other firm does, one firm will never want
to produce these quantities. We can now go one step further: for instance, knowing
that Firm B will never produce any quantity between 60 and 90, Firm A will never
produce the quantities 0 or 10, as they are best replies against those quantities only.
The same analysis works for Firm B, who will not produce quantities 0 or 10 either.
We can keep applying this procedure. Knowing that quantities 0 and 10 (along with 60
to 90) will not be produced, no firm will produce 50, a quantity that is only a best reply
against 0. We are left with quantities 20, 30 and 40, that correspond to the quantities
in the three Nash equilibria. The process does not select an equilibrium, but comes very
close. This dynamics can be called “iterative elimination of strategies that are never a
best reply.”
If there were only one monopolistic firm to serve the demand, it would produce
40 or 50 units, which would be sold at €6 or €5, respectively, and the firm would earn
54
STATIC GAMES
€300 in either case. The reader can confirm that no other quantity would give the
monopolist higher profits. The price is higher than in the Cournot game with two
firms. As new firms enter the market, the price would get closer to the production
costs, €1, and the total quantity would be €90. The Cournot game shows through a
very simple model that more firms imply more competition and that oligopoly is an
intermediate case between perfect competition and monopoly. In future examples, we
will explore the possibility in which firms in an oligopoly form a cartel to extract all
monopolistic profits from the market.
The Cournot oligopoly where quantities may take any value. An analysis
using calculus
Here we again present the Cournot model for the mathematically inclined reader. Just
a little calculus can help extend the model for the case in which firms’ quantities are
not restricted to integers or multiples of 10. The market demand in Figure 3.6 can be
obtained after the function P = 100 − Q , that indicates at which price P can be sold
10
the quantity Q. Firms’ profits are given by the expressions
 100 − q A − qB 
B A ( q A , qB ) =   q A − 1 × q A,
 10 
 100 − q A − qB 
BB ( q A , qB ) =   qB − 1 × qB.
 10 
In those functions, the term  100 − q A − qB  qi = Pqi is sales revenues, and the term
 10 
 
1× qi is the production costs, the unitary cost of €1 multiplied by the produced
quantity.
Each firm maximizes profits by taking as the decision variable the quantity it wants
to sell in the market:
 100 − q A − qB 
max 
10  q A − 1 × q A.
qA
 
The first-order conditions for a maximum require that the first derivative be made
equal to zero:
90 − 2q A − qB = 0,
from there we find

90 − qB ,
qA =
2
which is the function that gives the best choice for Firm A as a function of the quantity
chosen by Firm B. By symmetry, Firm B’s best reply function is
90 − q A .
qB =
2
55
GAME THEORY
Solving the system of two equations formed by the two best reply functions, we obtain
the values in the Nash equilibrium: q= A q=
B 3 0, which coincide with one of the equi-
libria we found without calculus in the finite version above. The other two equilibria
resulted from firms being forced to produce in 10-unit packages and, in consequence,
being unable to be more precise in choosing their best reply.
In this version of the Cournot game, we can better see the failure in Douglas Hof-
stadter’s argument in Section 1.5. When solving the Cournot game, at the time of find-
ing the derivative of one firm’s profit function, the quantity of the other firm is held
constant. Only at the end of the process of finding the solution are the quantities for
both firms equal, as a feature of the solution, not before. Hofstadter’s argument would
mean that we set the two quantities to be the same before calculating the derivatives, as if
one firm were able to control its rival’s quantity. That would be the monopolist’s solution
or the case in which the two firms form a cartel, but it is not the solution to the oligopoly.
3.7 Bertrand oligopoly
Consider the same problem of two firms competing in a market, as in the previous
section, but now let the firms compete in choosing a price instead of a quantity. This
is the main characteristic of the Bertrand model of oligopoly.1 We will make the sim-
plifying assumption that prices can only take integer values of euros, without cents.
The demand and costs are the same as before. The use of price as the decision variable
presents some complications that did not appear with the quantities. If one firm pro-
duces X and the other firm produces Y, the market demand will specify a price corre-
sponding to the quantity X + Y . However, if one firm posts a price P and the other one
posts a different price R, we cannot enter the two prices in the demand function to find
a quantity; instead, we must specify the demand for each firm at those prices. Thus, we
will add the assumption that when prices are different, all consumers buy from the firm
that offers the lowest price, and when they offer the same price, consumers are divided
to buy from each firm equally.
Once we have specified how to use the demand function when the choice variable
is the price, we can look for the firms’ best reply functions. First, observe that if a firm
chooses a “reasonable” price, the interesting thing for the other firm to do is either to
match it, and share the market, or to set a price just a little lower and retain the whole
market, whichever is more profitable. If the rival puts a price below the production cost,
it does not pay to compete, and that is why we proposed a “reasonable” price. The rival’s
price may also be too high. For instance, if Firm A sets a price of €10, then the optimal
thing for Firm B to do is not to post €9, for it will sell very little. The best thing will be
to post the monopolist’s price, calculated in the previous section, since that is the price
that maximizes profits. Putting together all these considerations, we can construct the
best reply functions in Table 3.3.
• If Firm B fixes a price of €1, Firm A can only sell something if it also posts that price
or one lower. At a lower price, however, it will suffer a loss, and at a price of €1 it will
make zero in profits, the same as if it sets a higher price and sells nothing. The best
reply then is any price between €1 and €10.
1 Joseph Louis François Bertrand published his model of oligopolistic competition in 1883 in a review
of Cournot’s book (Bertrand, 1883).
56
STATIC GAMES
Table 3.3 Best replies for the Bertrand game
B’s price 1 2 3 4 5 6 7 8 9 10
A’s best reply 1–10 2 2 3 4 5 5/6 5/6 5/6 5/6
A’ price 1 2 3 4 5 6 7 8 9 10
B’s best reply 1–10 2 2 3 4 5 5/6 5/6 5/6 5/6
• B posts €2: if A also posts €2, they will sell 40 units each, and A will earn €40. If A
wants to take in the whole market, it must post €1, but it will then make no profits.
Therefore, €2 is the best reply.
• B posts €3: if A posts €3, they will sell 70 in total, which means that one sells 30 and
the other sells 40. Thus, Firm A will make €60 or €80. But if it posts €2, it will take
the entire market, sell 80 units and make €80. Let us consider €2 as the best reply.
• B posts €4: if Firm A also sets €4, they will sell 30 units each and make €90. If Firm
A sets a €3 price, it will sell 70 and make €140.
• If B posts a price between €4 and €6, follow analogous arguments to find €3, €4 and
€5 as the best replies, respectively.
• If Firm B posts a price between €7 and €10, then Firm A’s best reply is to set the
monopoly price, €5 or €6, to sell 50 or 40 and make €200 either way.
By inspecting the best replies, we find the following Nash equilibria: (1, 1) and (2, 2).
Setting a price of €2 is the best reply against the other firm also setting €2. Also, setting
a price of €1 is one of the best replies against €1. In the Nash equilibrium, prices in the
Bertrand model are lower than those in the Cournot model. In fact, they are very close
to the unit cost, which would correspond to the price in perfect competition. If prices
could be posted up to a cent, the reader can easily confirm that the Nash equilibria
would be (1, 1) and (1.01, 1.01), which are almost identical. The dynamics we analyzed
in the Cournot oligopoly to find the Nash equilibria also work in this case, but we leave
this as an exercise for the reader.
A characteristic of the Bertrand model of oligopoly is that it produces competitive
results with only two firms, which is very unrealistic. This is known as the Bertrand
paradox. On the one hand, the assumption that firms choose price as the strategic vari-
able seems more realistic than the Cournot alternative of choosing quantities. However,
in real markets we do not observe competitive results with just two firms, and in this
regard, the Cournot model looks more realistic. In other words, Bertrand’s model is
more realistic in the assumptions but less realistic in the results, just the opposite of
Cournot’s. In situations like this, the analyst prefers the model that obtains better
results and then sees its assumptions as just approximations to the problem. In the case
of the oligopoly theory, new models have been developed to reconcile the realism of
both assumptions and results. For example, if firms first choose their capacity (e.g., the
size of the factory) and then the price, the result is the same as in the Cournot model,
thus giving a reason to understand the assumption of quantity choice as an approxi-
mation of the more realistic, but also more complicated, assumption of capacity and
price choices. There are other models that keep the competition in the prices but add a
57
GAME THEORY
certain differentiation of products so that the demand does not all go to the firm with
just a slightly lower price. In these models, the equilibrium with two firms does not
get as close to perfect competition and, therefore, is also realistic in both assumptions
and results.
If we repeat the exercise for the case where prices can take any value between €0
and €10, and quantities any value between 0 and 100, we come to a problem. The
drastic way in which the profits change when a firm alters its price to be just a bit
above, equal or below the price of the rival makes it impossible to write its profit
function with just a formula. This means that we cannot use calculus techniques
to find the best reply functions and, afterwards, the Nash equilibria, as we did in
the Cournot model. However, the logic used to find the Nash equilibria in the
discrete version of this section allows us to find the Nash equilibria in the infinite
version also. Through trial and error, we can see that there cannot be an equilib-
rium with the firms setting different prices. We can then see also that there is no
equilibrium if they post the same price, though different from the cost. Finally,
we can confirm that the last possibility we are left with, namely that the prices are
the same and equal to the cost, is indeed a Nash equilibrium. Exercise 3.10 asks
for the details.
In the models of the median voter in Section 3.2 and in the oligopoly models of
Cournot and Bertrand oligopolistic competition, we have seen that players choose the
same strategies in the Nash equilibrium. A common mistake after observing this fact
is the belief that competition does not matter in these cases. Why should anyone be
concerned over which party wins or which firm sells if in the end they are all the same?
It is true that the two parties end up offering a similar, centered electoral platform, but
in the absence of competition between them, with only one party on the scene, there
would be either a right-wing or a left-wing dictatorship. Also, had the firms not been
competing, we would have had a monopolist firm setting a much higher price. Com-
petition among many firms is better than between a few, and a few are better than a
monopoly.
3.8 Keeping up with the Joneses

Among the possible external influences that affect a consumers’ decision and upset the
perfectly competitive markets assumptions is the presence of preferences that depend
on the relative consumption, and not just the absolute. Let us see this in an example.
As a baseline, start with the case of preferences depending only on the absolute level of
consumption. Say that there are two consumers, 1 and 2, with preferences given by the
utility function ui = xi (16 − l i ), where xi is the consumption level of Consumer i and
l i is the amount of time dedicated to work, measured in hours. Subtracting 8 hours to
sleep, there are then 16 − l i hours for leisure in the day. To keep the example simple,
say that the wage per hour is €10 and that the price of the consumption good is also
€10 per unit. Table 3.4 shows the utility for the consumer for different choices of hours
to work. To understand the numbers in the table, consider that the consumer works for
7 hours ( l i = 7 ). Then her leisure time will be 16 − 7 = 9 , she will buy 7 units of the
consumption good, and her utility will be ui = 7 × 9 = 63. Those are the numbers in
the third column. The other columns are obtained similarly.
58
STATIC GAMES
Table 3.4 Utilities from different work-leisure choices
li 6 7 8 9 10 11 12
16 − li 10 9 8 7 6 5 4
xi 6 7 8 9 10 11 12
ui = xi (16 − li ) 60 63 64 63 60 55 48
The highest utility is achieved if the consumer works for 8 hours, which implies a
salary of €80, with which she buys 8 units of consumption and gets a utility level of 64.
Notice that this is an individual problem. Each consumer makes their decision without
taking into consideration the choice of the other.
Let us now introduce a new term in the utility function of the consumers to show
preferences for the relative consumption. In particular consider preferences given by
ui = xi (16 − l i ) + 4 ( xi − x j ), where the new term 4 ( xi − x j ) is positive if xi > x j ,
that is if the own consumption is greater than the consumption of the other consumer,
and is negative in the opposite case. Now the utility of one consumer is affected by
the consumption level of the other, and so is her decision to work. To start solving the
game, consider that Consumer 1 works 8 hours, as in the previous case without the
consumption externality. The best reply by Consumer 2 can be found in Table 3.5.
Table 3.5 Utilities if the other consumer works eight hours
l2 7 8 9 10 11
16 − l2 9 8 7 6 5
x2 7 8 9 10 11
x2 − 8 −1 0 1 2 3
u2 = x2 (16 − l2 ) + 4 ( x2 − 8 ) 59 64 67 68 67
The best reply by Consumer 2 is to work 10 hours. Table 3.6 shows the result of
repeating the exercise for the cases in which Consumer 1 works x1 hours.
We can observe that the respective best choice by Consumer 2 is always 10, no
matter how many hours Consumer 1 works. Since Consumer 1 has the same utility
function, she will also choose to work 10 hours. The result is that they will buy 10 units
of the consumption good but enjoy fewer hours of leisure time, with a utility level of
60 each instead of the 64 they could enjoy if they each worked 8 hours. However, this
is not an equilibrium. Given the choice between 8 and 10 hours, the Nash equilibrium
is to work 10 hours each.
59
GAME THEORY
Table 3.6 Utilities if other consumer works x1 hours
l2 7 8 9 10 11
16 − l2 9 8 7 6 5
x2 7 8 9 10 11
x2 − x1 7 − x1 8 − x1 9 − x1 10 − x1 11− x1
u2 = x2 (16 − l2 ) +4 ( x2 − x1 ) 91− 4 x1 96 − 4 x1 99 − 4 x1 100 − 4 x1 99 − 4 x1
The game can be summarized as in Figure 3.6, showing just two strategies for each
player: the optimal and the equilibrium choices.
Consumer 2
l2 = 8 l2 = 10
l1 = 8 64, 64 56, 68
Consumer 1
l1 = 10 68, 56 60, 60
Figure 3.6 Keeping up with the Joneses

3.1 Repeat the analysis of the game in Section 3.1, but now instead of having an
extra benefit from jobs, the country that succeeds in attracting the firm suffers
an environmental cost equivalent to 1.
3.2 Consider the example of voters uniformly distributed between 0 and 10 in the
political spectrum and that parties can be located at any point in the spectrum.
Now, suppose there are three parties.
(a) Show that there is no Nash equilibrium in pure strategies.
(b) Consider now that there are four parties. Show that the only Nash equilib-
rium requires that two parties are located at point 2.5 (the middle of the
left side of the spectrum), while the other two are located at point 7.5 (the
middle of the right).
3.3 Find the Nash equilibria of the game in Figure 3.3a using the technique of
underlining the payoffs that correspond to the best reply of a player against the
different strategy choices by the other players.
3.4 Repeat the analysis of the game in Section 3.3 for the case in which the CP
prefers a candidate of the RP over one of the LP.
60
STATIC GAMES
3.5 Repeat the analysis of the game in Section 3.4 with the following change: the
neighbors do not feel any reprobation for not being clean, but being unclean is
penalized with 20. The probability of being penalized decreases with the num-
ber of unclean neighbors and is given by the expression 1000 − s , where s is the
number of unclean neighbors. 1000
3.6 Consider the game about the independence of Sylvania. How would things
change if all citizens belonged to Group B?
3.7 Consider the Cournot oligopoly in Section 3.6, but change the costs. Firm A’s
costs are still €1 per unit, but Firm B’s change to €2. Find the Nash equilibria
in the following two cases.
(a) Firms can only choose quantities in multiples of 10.
(b) Firms can choose any quantity.
(c) Compare the equilibrium quantities, price and firms’ profits in (a) and (b)
with the ones found in the case seen in Section 3.6.
3.8 Consider the Bertrand oligopoly in Section 3.7, where firms set prices using
euros in integer numbers.
(a) Show that the “best reply” or adaptive dynamics leads to a Nash equilibrium.
(b) Show that the “iterative elimination of strategies that are never a best reply”
dynamics leads to a Nash equilibrium if we eliminate first the price of €1
from the set of strategies.
3.9 Consider the Bertrand oligopoly in Section 3.7, where firms post prices using
euros in integer numbers. Firm A still produces at the cost of €1 per unit, but
Firm B’s costs are now €2 per unit. Find the Nash equilibria in this new case.
3.10 Consider the Bertrand oligopoly in Section 3.7, but now assume that firms
can post any price between 0 and 10 euros, and not just integer numbers. The
unitary costs for both firms are still €1 per unit. Let pA be the price posted by
Firm A and pB the one by Firm B:
(a) Show that pA > pB > 1 cannot be a Nash equilibrium. (For the same reason
pB > pA > 1 cannot be a Nash equilibrium either.)
(b) Show that in a Nash equilibrium it cannot be that pA < 1 or pB < 1.
(c) Show that if pA = pB > 1, the situation cannot be a Nash equilibrium.
(d) Show that the only possibility left, p=
A p=
B 1, is a Nash equilibrium.
3.11 Three countries (A, B and C) must simultaneously decide whether to declare
war on country D. If two or three declare war, they will win easily, and the
countries that declared war will share a bounty of 100 divided into equal parts.
If only one declares war, it will lose and will suffer a cost estimated at 30. Coun-
tries that do not declare war do not win or lose anything. Find the pure strategy
Nash equilibria.
61
4 DYNAMIC GAMES
In this chapter we introduce the representation and study of dynamic games – namely,
games in which the play is not simultaneous. Players may choose their strategies at
different times. Perhaps the simplest examples are those of some board games, where
players take rounds at playing and, further, know the story of moves at every moment,
like in tic-tac-toe and chess. Not all games share these characteristics. There may be
more than two players, the order of play may be more intricate, there may be some
chance involved (like in Parcheesi) or the players may have only a partial view of the
story (like in poker, where each player has some private information of what is going
on in the game).
We will begin with the simplest case of complete information and no chance moves
in Section 4.1, and we will continue with games that include both sequential and
simultaneous moves in Section 4.2. Further complications will be addressed in later
chapters. We will first notice that we need not only a new game model but also a new
solution. We will introduce two equilibrium concepts: backwards induction (simple
but not useful in all games) and subgame perfect Nash equilibrium (more complicated
but always applicable). The subgame perfect Nash equilibrium, defined carefully in
Section 4.3, can be understood as a generalization of both backwards induction and
Nash equilibrium. In fact, by analyzing the different equilibria we will find that, very
often, the differences among them rely on how they address questions like the use of
empty threats and of false promises. Consequently, after the presentation of the first
concepts we will dedicate the remaining sections to discuss the issue of credibility in
dynamic games, which will be illustrated with important social, political and economic
applications.
4.1 The extensive form and backwards induction

In Section 1.3 we presented the battle of the sexes game. In addition to its standard
version, where Sylvie and Bruno make their decisions simultaneously without knowing
each other’s plans, we also examined the case where one chooses first and then the other
chooses after observing the previous choice. This is an example of a simple dynamic
game in which we can introduce and discuss the first concepts we need for this type of
game. The first thing to take into account is that the normal form becomes inadequate
for our purposes. Instead, we will use the extensive form.
In simple dynamic games the extensive form means that we display all the informa-
tion about the game in a tree-like graph. The extensive form must start with a single
node, the root, from which some branches grow, and from those branches new ones
open, and so forth. The branches correspond to choices made by the player that plays
at the point from which they emerge. It is better to look at all of this through some
examples. Take, for this purpose, the battle of the sexes game in which Sylvie first
chooses her strategy and, knowing what she did, Bruno then chooses his. The game is
represented in Figure 4.1a, where we can find its elements.
62
DYNAMIC GAMES
Sylvie
F M
Bruno Bruno
F M F M
2 0 0 1
1 0 0 2
Figure 4.1a Dynamic battle of the sexes
The extensive form

The elements of the extensive form representation are
• the tree of the game (branches with their nodes at the beginning and at the end),
• the distribution of non-final nodes among the players, and
• the payoffs the players get at the final nodes (in the figure we understand that the
first number corresponds to Sylvie’s payoff and the second to Bruno’s).
With these elements defined we can read Figure 4.1a: First, Sylvie chooses between F
and M (football or movies). In the case that she chooses F, it is then Bruno’s turn to
choose between F and M as well. If Sylvie were to choose M, the situation remains the
same; Bruno has to choose between alternatives F and M.
As we said in the introduction of this chapter, there is an easy way and also a com-
plicated way to solve this game. The bad news about the easy way is that it will not be
valid for all dynamic games, while the good news about the complicated way is that it
will provide us with a general method to solve all dynamic games. Furthermore, this
general method is an adaptation of the concept of Nash equilibrium.
Backwards induction
We will begin our study of dynamic games with the easy way; we will leave the more
complicated, general way for a later section.
At the beginning, Sylvie does not know which would be the best choice for her to
make. After either choice, she can either win something or nothing, depending on what
Bruno does afterwards. An interesting observation is that she can have an idea about
what Bruno will do. If Sylvie plays F, Bruno will see this and know that his only chance
to get some positive payoff is to play F as well. If Sylvia plays M, Bruno will play M
for that same reason. Having been able to foresee Bruno’s choice in every contingency,
Sylvie will play F, the action which maximizes her payoff. Observe that in order to solve
the game we had to begin by solving its last part. First, we anticipated what the last
player would play and, at that point, we moved backwards to solve for the first player.
This process is called “backwards induction,” and we will use it every time that we can.
Figure 4.1b illustrates this situation, with the arrows indicating the solution.
Consequently, we will say that the backwards induction equilibrium is (F, (F, M)),
where the first F is Sylvie’s election and the pair (F, M) is Bruno’s: F at his first node,
the one which comes after Sylvie plays F, and M at his second.
63
GAME THEORY
Sylvie
F M
Bruno Bruno
F M F M
2 0 0 1
1 0 0 2
Figure 4.1b Backwards induction in the dynamic battle of the sexes
It is necessary to highlight the importance of saying that (F, M) is Bruno’s

choice, and not just F. Of course, if Sylvie chooses F, Bruno will choose F as well,
and the game will end there, but the equilibrium must say more than just what can
be observed. To examine whether players are playing according to the logic of the
equilibrium we need to know their entire plan, so that we can check whether play-
ers are discarding the choices that we cannot observe for the proper reasons. This is
important because we cannot tell whether the observed play is played for the good
reasons until we know these other non-observed parts of the plan. In this example,
Sylvie prefers to play F because she knows that Bruno will imitate whatever choice
she makes and because F is better for her. But let the reader imagine Sylvie’s payoffs
after she chooses M and Bruno chooses F were to be (3, 0) instead of (0, 0). In that
case, the best outcome for Sylvie is for her to choose M and for Bruno to choose
F. This is, however, something that will not happen, since it is still true that after
Sylvie chooses M, Bruno will also play M, and the solution by backwards induction
will therefore still be (F, (F, M)). If we only say that the equilibrium is simply that
Sylvie moves F and Bruno moves F as well, we will miss the reason Sylvie chooses
F as opposed to M.
Keep in mind that by saying that Bruno’s election is (F, M), we mean only that
this is his plan, not that he is doing every choice of strategy that is listed. As with all
plans, the choice of strategy has to cover all contingencies, not only the best one. For
instance, my plan for tomorrow is to take my umbrella if it rains, and to not take it if
it does not rain. The plan foresees two contingencies, of which only one will actually
occur.
4.2 The prisoners’ dilemma with a previous contract

For some situations that share the strategic configurations of the prisoners’ dilemma
there exists the possibility of signing a binding contract before actually playing the
game. Think of two firms that agree to exchange technological knowledge. On day
D they both must send a dossier to the other part with engineering details. Each firm
wins if it has access to the other’s technology, but each firm would also prefer not to
disclose its information. Say that having access to the other’s technology is worth 10
to a firm, but that sharing its own represents some costs estimated at 5. Thus, if both
share the technology, each firm wins 10 − 5 = 5 , and if only one shares, it loses 5 while
the other wins 10. Figure 4.2a shows the game.
64
DYNAMIC GAMES
Firm 2
Send dossier Do not send
Send dossier 5, 5 −5, 10
Firm 1
Do not send 10, −5 0, 0
Figure 4.2a Send the dossier game
The game is, in effect, a prisoners’ dilemma. The only Nash equilibrium is (Do not
send, Do not send), even though there is another strategy profile that gives both players
better payoffs (Send dossier, Send dossier). Consider now that firms can sign a contract
to ratify their verbal agreement on sharing the technology and the contract includes
a clause stipulating that, should a firm not send the information, said firm will pay a
penalty of 8 to the other firm. Also consider that the contract is legal and, in case one
firm does not comply, a court can enforce it. For the contract to be valid, both firms
must sign it. If they do, then the game is the one in Figure 4.2b.
Firm 2
Send dossier Do not send
Send dossier 5, 5 3, 2
Firm 1
Do not send 2, 3 0, 0
Figure 4.2b Send the dossier game with a previous contract
In the new game the only Nash equilibrium is (Send dossier, Send dossier). A sche-
matic representation of the whole game, where the contract is presented first to Firm
1 and, then, to Firm 2, can be seen in Figure 4.3a. In this figure, S stands for “sign
the contract” and NS for “do not sign the contract.” Game 4.2a means that at that
node the game is played according to Figure 4.2a, and Game 4.2b is to be interpreted
analogously. The whole display of these games at those nodes would make a more com-
plicated figure without providing additional clarity.
Firm 1
S
NS
Firm 2 Game
S NS 4.2a
Game Game
4.2b 4.2a
Figure 4.3a The game of signing a contract
In the game in Figure 4.2a we cannot use the backwards induction technique, as in

any of the final games there is no last player from which to start the inductive process.
However, we know the equilibrium in each of those games, and we can use this infor-
mation to replace the game with the payoffs in the respective equilibrium. This way we
65
GAME THEORY
can solve the part of the game which corresponds to the signing of the contract. This
is done in Figure 4.3b.
Firm 1
S NS
Firm 2
0
S NS 0
5 0
5 0
Figure 4.3b Solving the game of signing a contract
In the new, simplified game we can apply backwards induction to conclude that
both firms will sign the contract. The complete description of the solution must include
what each firm does in every part of the game where it has to make a choice. Thus, we
will write ((S, Send dossier, Do not send, Do not send), (S, Send dossier, Do not send,
Do not send)). Note that within the outermost pair of parentheses there are two sets
of actions, each between new parentheses. The first set of actions corresponds to the
choices by Firm 1, and must be read like so: Sign the contract, Send the dossier in case
both firms sign the contract, Do not send the dossier if Firm 1 signed but Firm 2 did
not sign, Do not send the dossier if Firm 1 did not sign the contract. The second set is
the actions of Firm 2 and must be read in a similar way.
Signing a contract to solve the strategic problems of a prisoners’ dilemma game is
not always an option. This becomes the case, for example, if the activity in which the
players are involved is an illegal one, like the actual prisoners of the original game.
The formation of cartels for a market-sharing agreement is another example. In some
cases, even if the activity is legal, there is no authority that can enforce the contract if
some party breaks it, as often is the case in international agreements among countries.
This does not mean that in those cases the players cannot escape the logic of the pris-
oners’ dilemma, but that the way in which they do it, if they find such a way, must be
different from the contract discussed here.
4.3 Subgame perfect Nash equilibrium

The Nash equilibrium is a concept defined in the normal form. If we want to use it in
dynamic games first we need a way to define a normal form after the extensive form.
Let us look at how to do it. Go back to the dynamic battle of the sexes game in Fig-
ure 4.1a. Recall that the normal form needs to define a set of players, a set of strategies
for each player and a payoff function that assigns a payoff for each player following any
combination of strategies. The part about the players is easy: we have Sylvie and Bruno.
Sylvie’s strategies are also straightforward: F and M. Determining the set of strategies
for Bruno requires some attention. Bruno plays in two different nodes and has two
options in each one. Therefore, the different contingent plans for Bruno are the combi-
nations of the actions he can choose from at his two nodes: (F, F), (F, M), (M, F) and
(M, M). In each of these four combinations the first component is the part of the plan
for the first node, after Sylvie plays F, and the second component is the action to take
66
DYNAMIC GAMES
in the second node, after Sylvie plays M. After we define that these contingent plans are
Bruno’s strategies, we can define the normal form in Figure 4.4.
Bruno
(F, F) (F, M) (M, F) (M, M)
F 2, 1 2, 1 0, 0 0, 0
Sylvie
M 0, 0 1, 2 0, 0 1, 2
Figure 4.4 The normal form of the dynamic battle of the sexes
To make sure we understand this normal form, take as an example the strategy
combination (F, (M, F)), where Sylvie chooses F and Bruno chooses M if he is in the
node after Sylvie chooses F, and Bruno chooses F if he is in the node after she chooses
M. What are the payoffs for these choices? Since Sylvie chooses F, the part of the
plan (M, F) chosen by Bruno that is activated is the first one, M. Thus, if each player
chooses as indicated in their plans, the final node of the extensive form that will be
reached is the one coming after Sylvie chooses F and Bruno chooses M, with zero
payoffs for both.
In Figure 4.4, we also marked the best replies according to the technique intro-
duced in Chapter 2 (Section 2.4) which underlines the payoffs that correspond to the
best replies. We find three Nash equilibria in pure strategies: (F, (F, F)), (F, (F, M))
and (M, (M, M)). Only the second one coincides with the solution found by back-
wards induction. What is wrong with the other two? Take the third one, (M, (M, M)),
which implies that both players go to the movies, and which is contrary to the con-
clusion drawn from using backwards induction. This is a Nash equilibrium because if
Bruno chooses his plan (M, M), through which he goes to the movies no matter what,
the best reply for Sylvie is to also go to the movies. Bruno is also choosing his best
reply against this choice by Sylvie. However, the players do not make their decisions
simultaneously. The game clearly specifies that Bruno must decide after knowing
what S ylvie does. The definition of Nash equilibrium considered unilateral deviations
because it was intended for simultaneous games; however, in this game, Sylvie cannot
think that her deviation will be the only one. If she changes her mind and plays F
instead of M, Bruno will observe Sylvie’s deviation and will deviate himself by playing
F as opposed to M. The Nash equilibrium does not take these induced deviations into
consideration.
Another way to interpret the strategy profile (M, (M, M)) is as if Bruno were
threatening Sylvie: “I will play M no matter what, which means that if you play F then
you will earn zero, so you better play M.” The problem is that this is an empty threat.
When it comes time for Bruno to carry out his threat, it will be against his interests
to do so.
All in all, the definition of Nash equilibrium can still be used to fit in dynamic
games and solve those problems of dynamic inconsistence. We will do it by insisting
that the strategy profile be not only a Nash equilibrium of the game as a whole, but also
a Nash equilibrium in every part of the game. In the game tree of Figure 4.1a we can
detect two smaller games contained within the whole game: one that begins after S ylvie
chooses F, and a second which begins after she chooses M. These are trivial games,
67
GAME THEORY
having only one player each, but they are games just the same. Figure 4.5 represents the
first. The second one is analogously represented.
Bruno
F M
2 0
1 0
Figure 4.5 A subgame in the dynamic battle of the sexes
The normal form of this little game is readily displayed in Figure 4.6.
Bruno
F M
2, 1 0, 0
Figure 4.6 The normal form of the subgame in Figure 4.5
In this game, the only Nash equilibrium is F. If we do the same with the other
small game, we will see that its only Nash equilibrium is M. Out of the three Nash
equilibria of the whole game (F, (F, F)), (F, (F, M)) and (M, (M, M)), the only one that
contains the Nash equilibria of the smaller games is the second one – the one that does
not include non-credible threats and that coincides with the solution resulting from
backwards induction.
The interesting property of this adaptation of the concept of Nash equilibrium,
called subgame perfect Nash equilibrium, is that it can be used in games in which
backwards induction cannot. This will happen every time we have both simultaneous
and sequential choices mixed in the same game.
According to the definition of subgame perfect Nash equilibrium we must first
find all the Nash equilibria of the game and then check one by one whether they
are still Nash equilibria in each of the smaller games. This procedure is tedious but,
fortunately, there is a better way. It will be much easier to start by finding the Nash
equilibria of the small games and build up the subgame perfect Nash equilibrium after
them. This is a sort of backwards induction; however, rather than taking individual
actions one at a time, we instead take Nash equilibria. This is in fact precisely what we
did in Section 4.2.
We have not carefully defined what a small game or subgame is. In the dynamic
games in the following chapters it will be made clear what they are. When we reach
a point in which we need more complicated dynamic games with not-so-easy-to-spot
subgames, we will provide a formal definition. In the meantime, it is enough to think
of subgames as smaller games within a game and as parts of the whole game in which
players play simultaneously. Later, they will be defined as parts of the game that pre-
serve a certain informational structure of the game. Next, we present several ways in
which the concept of subgame perfect Nash equilibrium is the natural definition by
which to talk about the credibility of the strategies.
68
DYNAMIC GAMES
4.4 How to be credible (1) Eliminate strategies: Odysseus

and the Sirens
Among the many adventures that Odysseus (Ulysses) endured during ten years of
sailing back to Ithaca after the Trojan War, his encounter with the Sirens stands out
(Homer, 8th century B.C.). Circe, a goddess seeking to protect Odysseus, warns him
that he will be passing near the place inhabited by the Sirens, monsters that pretend to
be beautiful women and whose beguiling songs lure any man into jumping overboard
to be killed by them. Odysseus wants to hear the songs, but he does not want to throw
himself into the sea. He is also aware that his mad self would be willing to do just that.
This episode is a game between Odysseus and himself, as represented in Figure 4.7a.
Odysseus-sane
Listen to the Sirens Do not listen
Odysseus-insane
0
Jump Do not jump 0
–10 10
10 –10
Figure 4.7a Odysseus-sane vs. Odysseus-insane
The first number is the payoff of the sane Odysseus, and the second, the payoff of
the insane Odysseus. The only subgame perfect Nash equilibrium, and also the only
backwards induction solution, is (Do not listen, Jump). This means that, knowing
that his insane self will jump, the sane Odysseus decides not to listen to the Sirens.
The actual legend has a different plot; Odysseus understands the game and changes it.
He wants the best of both worlds, to hear the Sirens’ songs and also not throw himself
overboard. He orders his men to tie him to the mast and put wax in their ears. This
way he eliminates one of the strategies available to his insane future self. The new game
is depicted in Figure 4.7b.
Odysseus-sane
Odysseus-insane
0
Do not jump 0
10
–10
Figure 4.7b Odysseus is tied to the mast
In this game, we could get rid of the insane Odysseus and his “Do not jump” strat-
egy as it is not really a strategy since he has no choice. Still, we will keep the two ele-
ments for the sake of completeness. Now the equilibrium implies that Odysseus listens
69
GAME THEORY
to the Sirens without being killed by them. We can write the complete game which
includes his decision to be tied to the mast. At this point, it will not contribute more
to the analysis, but it will help to understand the whole story as one single game with
its different subgames. This is the purpose of Figure 4.7c, in which we also include the
equilibrium marked with arrows.
Odysseus-sane
Odysseus-sane
Tie to the mast Do not tie 0
0
Odysseus-insane
Odysseus-insane
Do not jump Jump Do not jump
10 –10 10
–10 10 –10
Figure 4.7c Odysseus and the Sirens game
The equilibrium in the whole game is described as ((Listen to the Sirens, Tie to the
mast), (Do not jump, Jump)). The first part, (Listen to the Sirens, Tie to the mast),
describes the two actions in sane Odysseus’s strategy, while the second part, (Do not
jump, Jump), is the strategy of insane Odysseus.
Had sane Odysseus promised himself to not jump overboard after listening to the
Sirens, he would have made a non-credible promise. The only way to make it credible
is by eliminating it from his set of strategies.
4.5 How to be credible (2) Acquire costly compromises: Who

enters?
Two firms are simultaneously considering whether to enter a new market from which
they can extract profits worth ten million to be divided between them if they both
enter. The cost of entering is seven million. If a firm chooses not to enter, it will not
win or lose anything. Figure 4.8a shows the normal form game. If they both enter, each
will obtain half of the ten million after paying the seven million cost, meaning that
they will lose two million each.
Firm 2
Enter Do not enter
Firm 1 Enter −2, −2 3, 0
Do not enter 0, 3 0, 0
Figure 4.8a The entry game
70
DYNAMIC GAMES
Firms face a chicken game, in which the two Nash equilibria in pure strategies are
(Do not enter, Enter) and (Enter, Do not enter). Firm 1’s CEO analyzes the situation
and advises in favor of immediately investing a big part of the seven million necessary
to enter the market, even before either firm makes the actual decision to enter or not.
The investment will be lost if the firm ultimately decides to not enter. Specifically, he
proposes investing three million while making sure that the other firm observes how
the money is spent. Does it make sense to risk the investment? To answer the question,
let us see how the game is after the initial investment by Firm 1 (Figure 4.8b).
Firm 2
Enter Do not enter
Enter −2, −2 3, 0
Firm 1
Do not enter −3, 3 −3, 0
Figure 4.8b The entry game after a costly compromise
The game is like the previous one except that now, in case Firm 1 decides not to
enter, it will lose its investment of three million. The only Nash equilibrium in the new
game favors Firm 1. The complete game is laid out in detail in Figure 4.8c, where E
stands for “Enter” and NE for “Do not enter.”
Firm 1
Invest 3 Do not invest
Firm2 Firm2
E NE E NE
E −2, −2 3, 0 E −2, −2 3, 0
Firm 1 Firm 1
NE −3, 3 −3, 0 NE 0, 3 0, 0
Figure 4.8c The game of acquiring a costly compromise
We cannot use backwards induction, but we can look for subgame perfect Nash
equilibria. Any selection in the first step by Firm 1 leads to a subgame we have already
solved. However, keep in mind that in the subgame to the left, in which Firm 1 does
not invest the three million, we have found two Nash equilibria. This means that there
are two possibilities to consider when looking for the subgame perfect Nash equilibria.
Let us proceed just as we did in Section 4.2 and replace subgames with their payoffs.
Case 1: In the subgame following “Do not invest” the Nash equilibrium (Enter, Do not
enter) prevails. Replace the game with the payoffs in the equilibrium. Also replace the
game following “Invest 3” with only its equilibrium payoffs. This leads us to the game
in Figure 4.8d.
Firm 1 is indifferent between “Invest 3” and “Do not invest.” Therefore, there are two sub-
game perfect Nash equilibria in pure strategies in this case: ((Invest 3, E, E), (NE, NE))
71
GAME THEORY
and ((Do not invest, E, E), (NE, NE)). In either of them, Firm 1 is the only one entering
and keeping the market for itself.
Firm 1
3 3
0 0
Figure 4.8d Solution after one equilibrium in the first subgame
Case 2: In the subgame following “Do not invest” the Nash equilibrium most favorable
to Firm 2 prevails, (NE, E). If we replace the subgames with the equilibria we get the
game in Figure 4.8e.
Firm 1
3 0
0 3
Figure 4.8e Solution after the other equilibrium in the first subgame
Now Firm 1 will choose “Invest 3.” The subgame perfect Nash equilibrium is now
((Invest 3, E, NE), (NE, E)). Once again, the result is that Firm 1 is the only one
entering the market. Thanks to the three million investment, Firm 1 ensures that it
alone will be profiting from the new market. The costly investment is, therefore, a
credible compromise to enter.
4.6 How to be credible (3) Give up control: Separation

of powers
An absolute monarch faces mounting pressure as the high nobility start demanding
more privileges. Every time one of its members presents his demands, a game of chicken
begins, with the two known equilibria in pure strategies. Sometimes the king resists,
while sometimes the noble gets his way. One of the king’s advisors suggests a plan to
organize the state’s affairs differently. Instead of limiting all the power to the king,
they would have a judicial system with separation of powers. There would be laws that
judges would obey, and there would be judges whose careers would depend on how
good they are at ruling according to the law. When a noble had a petition, he would
be directed to a judge; if that petition were to be a privilege, then the judge would rule
against it.
Let us then consider the situation in which a noble wants a new privilege. If he
is certain that the court will rule against him, then he will not bother to present his
petition, as it is costly to go to court. The complete game is depicted in Figure 4.9.
The first numbers are the payoff for the king, while the second, the payoff for the
noble.
72
DYNAMIC GAMES
King
Separation of powers No separation
Noble Noble
Petition No petition Concede Hold
2, −0.1 2, 0 Concede 1, 1 0, 2
King
Hold 2, 0 −1, −1
Figure 4.9 Separation of powers game
On the left-hand side of the game, following the separation of powers, the noble
decides to not carry out his petition (he will lose 0.1 if he does). On the right-hand side,
without that separation of powers, we have the typical two equilibria in the game of
chicken. Thus, there are two combinations of Nash equilibria in the subgames:
1. ((No petition), (Concede, Hold))
2. ((No petition), (Hold, Concede))
The first part is the Nash equilibrium in the subgame after the separation of powers,
while the second part is the Nash equilibrium after the non-separation of powers. If the
equilibrium that prevails in the subgame on the right-hand side is (Concede, Hold),
then the king is better off choosing the separation of powers in the first place. If (Hold,
Concede) prevails, then the king is indifferent as to choosing between separation of
powers or not. There are then three subgame perfect Nash equilibria:
1. ((Separation of powers), (No petition), (Concede, Hold))
2. ((No separation), (No petition), (Hold, Concede))
3. ((Separation of powers), (No petition), (Hold, Concede))
Notice that this time we presented the actions grouped by subgames, not by players,
like in the previous games. In any of the equilibria the outcome is favorable to the king.
If, however, he is not sure as to which equilibrium will prevail in the case that there is
no separation of powers or if, as previously stated, he knows that sometimes the noble
wins, then the king will choose to have the separation of powers. In this game, the
only consequence of the separation of powers is that it prevents the high nobility from
assuming too many privileges (and with them, more power at the expense of the king).
Needless to say, in real life there would be more consequences; however, the game illus-
trates this one reason in favor of separation of powers from the king’s point of view even
if there are other reasons against it.
4.7 The consequences of not being credible: The

healthcare game
An individual values her health at 100. There is a ¼ probability that her health dete-
riorates to a value level of zero. The individual can buy a full insurance policy that
costs 20. If she were to buy it, should her health deteriorate she would receive care and
her health would recover. Aside from this private insurance that the individual can
73
GAME THEORY
obtain, the government is considering offering help to individuals in need; this would
be just basic help and the individual would only recover her health up to a level of 40.
The expected cost of providing this service is 5 per individual, but the service will be
offered for free. The goal of the government is to maximize the health of the individual
minus its own cost.
Case 1: First the government decides on whether to offer the basic help or not. Then,
the individual decides whether or not to apply for the private insurance. The game is
depicted in Figure 4.10a. The payoffs are computed as indicated in the description of
the game. If the government does not offer the help, the individual who buys the private
insurance will value her situation at 100, since she will have her health no matter what,
minus the 20 that the insurance policy costs. If the individual does not buy private
insurance, she then values her situation at 3 100 + 1 0 = 75 . Now, if the government
4 4
offers the free partial help, the situation remains the same if she buys insurance; without
that insurance, however, her situation will be worth 3 100 + 1 40 = 85 . The payoffs for
4 4
the government are the health level of the individual minus the cost of 5.
Government
Offers help Does not offer help
Individual Individual
Insurance
Insurance No No
95 80 100 75
80 85 80 75
Figure 4.10a The healthcare game: the government plays first
The subgame perfect Nash equilibrium is indicated with the arrows: (Does not offer
help, (No, Insurance)). The consequence of offering partial help is that the individual
will not obtain private insurance that offers full coverage. Thus, in the equilibrium the
government does not offer the free help, so the individual will get the private insurance.
Case 2: First the individual makes her decisions, and then the government makes its
move. This situation is illustrated in Figure 4.10b, where the first payoff now corre-
sponds to the individual and the second corresponds to the government.
The equilibrium in this version of the game is (No insurance, (Does not offer help,
Offers help)). Now the individual does not apply for private insurance because she
knows that, in case she needs healthcare, she will receive help from the government.
Even if this is in the form of partial assistance, because it is free, it is preferable to the
full coverage of private insurance at a high cost.
It is often the case, at least in the majority of developed countries, that the govern-
ment cannot legitimately leave people in need unattended (i.e., for these countries the
situation is closer to Case 2 than to Case 1). The result then is that many might prefer
not to obtain more complete health coverage. Many analysts have seen this fact as one
of the arguments in favor of a fully funded healthcare system.
74
DYNAMIC GAMES
Individual
Insurance No insurance
Government Government
Does not offer help
Offers help Does not offer help
Offers help
80 80 85 75
95 100 80 75
Figure 4.10b The healthcare game: the individual plays first
With this example we do not pretend to have incorporated all the problems of
health provision. For instance, the reader may have missed a reference to the possibility
that the individual does not have enough money to buy insurance. This is certainly an
important problem, but we are not addressing it because it has to do with the desirabil-
ity of a mandatory universal healthcare system rather than some strategic aspect that
we want to study using game theory. Actually, our game suggests that, even if everyone
could afford health insurance (a seemingly strong argument in favor of a private sys-
tem) there are reasons to have a public one. Each one of us would have their own way
to ponder all the reasons and consequences to prefer one or the other.
Another point of interest that may cause uneasiness to the reader is the fact that we
are adding money and health, as if they were measured in the same units. Let us clarify
this. For instance, say that an individual is willing to buy health insurance at the cost of
100, but not at the cost 1000. This individual is implicitly saying that she can compare
the benefits of the insurance (better health) and its costs (the price of buying it), mean-
ing that she can weight these two different magnitudes. In our example we used a very
simplified way to weight them, just adding the numbers and multiplying them by the
relevant probabilities. Real cases will be more complicated, but for our simple game to
show the credibility problem faced by the state, a more realistic utility function would
not have changed the bottom line.
4.8 The payment of the debt

Freedonia needs a loan of 100 to put its public finances in order. With a high proba-
bility (0.8), the money will be spent well and will create a wealth of 170 in the future.
With a low probability (0.2), that money will only produce 50. In the case of success,
the government will be able to pay the loan (100) as well as the interest (I), and the
country will enjoy the difference. If the plan fails, everything will depend on what the
legislation in Freedonia states about who is to be paid first. If the payment of debt is a
priority, the 50 will all go towards paying that debt, and then Freedonia will declare
bankruptcy. If the payment is not a priority, then 30 out of the 50 will be used to
finance some public spending, and the remaining 20 to pay off creditors. The bank that
can loan the money to Freedonia can also alternatively invest it in a safe asset that will
result in a 20 percent gain. The extensive form of the game is shown in Figure 4.11a,
where we see that Freedonia first decides its legislation regarding the priorities for the
use of the money and then the bank chooses whether to loan Freedonia the 100 with
75
GAME THEORY
interests given by I. Finally, they will all know whether the wealth produced by the loan
is 170 or 50. We will keep in mind that financial markets are competitive, and the
interests I will be such that the bank earns as much by giving Freedonia the loan as
through investing in the safe asset.
Let us review the computations to make sure we understand the game as laid out.
If the bank does not hand Freedonia the loan, it will gain 20% (that is, it will have a
total of 100 plus 20% of 100, or 100 × (1 + 0.2 ) = 120 ) while Freedonia will get 0. If
the bank grants the loan, it will recover 100 plus its interest, for a total of 100 × (1 + I )
if things go well, while Freedonia will retain the difference up to 170. If the loan is
badly used, and in the case that paying back the loan is a priority, the bank will get 50.
Otherwise it will only get 20, while Freedonia will keep 30.
For the bank to be willing to loan the money, it is required that the expected payoffs
after the loan are equal to the expected gain by means of the alternative. In the case that
paying off the debt is a priority, this would mean
0.2 × 50 + 0.8 × 100 (1 + I ) = 120,
where solving for I gives I = 0.375 or, as a percentage, a 37.5% interest rate. If there is
no priority in the payment of debt, then the interest that makes the loan attractive to
the bank must satisfy the expression
0.2 × 20 + 0.8 × 100 (1 + I ) = 120,
which gives the value I = 0.45 or 45% in interests.
Freedonia
Priority No priority
Bank Bank
Loan No loan
Loan No loan
0 0
0.2 0.8 120 0.2 0.8 120
0 170 − 100(1+ I) 30 170 − 100(1+ I)

50 100(1+ I) 20 100(1+ I)
Figure 4.11a Payment of the debt game
Once we have found the interest rates for both cases, we can compute the payoffs
for the players. The game is depicted in Figure 4.11b.
The expected payoff for Freedonia, in the case that it decides that debt payment
is a priority and borrows the money from the bank, is 0.2 × 0 + 0.8 × 32.5 = 26.
If it borrows after deciding that debt payment is not a priority then the payoff is
76
DYNAMIC GAMES
0.2 × 30 + 0.8 × 25 = 26. We have found that, in regards to expected payoffs, it is

exactly the same for Freedonia regardless of whether or not it gives priority to debt
payment. The higher interests will compensate exactly for the part of the payment that
will not be paid back to the bank.
The key to the analysis is that the interest rate can be perfectly adjusted to the con-
ditions of the loan, including terms of payment priority. Problems begin when those
conditions are not met. An example of this would be if the loan were given at a low
interest rate under the condition that the payment is a priority and if the country were
to change this condition unilaterally afterwards. The country can gain more in this
particular game, but it would lose credibility and all future loans would require an even
higher interest rate.
Freedonia
Priority No priority
Bank Bank
Loan No loan Loan No loan
0 0
0.2 0.8 120 0.2 0.8 120
0 32.5 30 25
50 137.5 20 145
Figure 4.11b Solving the payment of the debt game
There is another interesting fact. Even if there was no risk of the country changing
the conditions of the payment, the dispersion of payoffs is higher for Freedonia if the
payment is given priority compared to the case in which no priority is given. For the
bank it is just the opposite. This means that the bank risks more if no priority is given
to the payment, in which case it may impose an even higher interest rate if it shows
some risk aversion (all our computations assumed both the bank and Freedonia to be
risk neutral). Whether Freedonia prefers to give priority or not to debt payment now
depends on its own attitudes towards risk; however, the expected payoffs in the case
that the payment is not a priority will be higher.

4.1 Imagine that the vacuum cleaner market is dominated by a firm called Fast-
clean and that a new firm, Newclean, is considering entry into said market. If
Newclean enters, Fastclean has two choices: either accommodate the entry of
Newclean, accepting a decrease in market share, or fight the entry by starting
a price war. Suppose that if Fastclean decides to accommodate the entry of
77
GAME THEORY
Newclean, the latter would have a profit of 10 million euros; but if Fastclean
opts for a price war, Newclean would lose 20 million euros. Obviously, if New-
clean does not enter the market, its profits are zero. Additionally, suppose that
Fastclean can obtain profits of 30 million euros as a monopoly, that sharing the
market with its competitor will reduce its profits to 10 million, and that a price
war will cost the firm 10 million.
(a) Draw the extensive form game. Find also the normal form as well as all the
Nash equilibria in pure strategies.
(b) Find the solution using backwards induction.
(c) Do the answers in (a) and (b) coincide? Explain why or why not.
4.2 Sylvie and Bruno have to decide where to go on vacation. They have three
options: Amsterdam (A), Barcelona (B) and Copenhagen (C), but they do not
reach an agreement on where to go. In order to make a d ecision, they use the
following mechanism. First, Sylvie vetoes one of the three places. Then, Bruno,
after observing Sylvie’s veto, vetoes a second place. They go to the place that has
not been vetoed. Sylvie prefers A to B and B to C; Bruno prefers C to B and B
to A. Assuming that each player assigns a utility of 3 to their favored place, a
utility of 2 to the second-best alternative and a utility of 1 to the last, and that
both players want to go together on vacation, answer the following questions:
(a) Represent the game in extensive and normal forms.
(b) Find the Nash equilibria in pure strategies.
(c) Which of the Nash equilibria previously found are obtained using back-
wards induction? Explain your answer. Where will Sylvie and Bruno go on
vacation?
4.3 Two firms share the dairy market of a city. One of them, called OBESA, only
sells products with all their fat. The other firm, called LISA, only sells non-fat
products. It is well known that people in the city are not concerned with being
thin; therefore, if LISA does not launch an aggressive advertising campaign
about the risks of being overweight, LISA and OBESA profits would be 1 and 6
million euros, respectively. On the contrary, if LISA does launch its campaign,
OBESA has the choice of fighting back with the publication of a dossier warn-
ing consumers about the lack of vitamins in the non-fat products of its rival.
In this case, LISA has another option, which is launching a public message
about the lack of sanitation and health regulations at the production facilities of
OBESA. The marketing department of both firms forecast that if LISA were to
launch its campaign against excess weight and if OBESA were to not respond
with the dossier, profits would be 4 million euros for LISA and 3 million for
OBESA. On the contrary, if OBESA were to respond by publishing the dossier
after LISA launches its campaign, and if LISA were to not react to this action,
profits would be 2 million euros for LISA and 4 million for OBESA. However,
if LISA reacts to the publication of the dossier with a public message about the
lack of sanitation and health regulations in the production facilities of OBESA,
profits would be 3 million euros for LISA and only 1 million for OBESA.
(a) Represent the normal and extensive forms of the game. Obtain all the Nash
equilibria in pure strategies.
(b) Which of the Nash equilibria are obtained using backwards induction?
78
DYNAMIC GAMES
4.4 Describe an alternative version to the Odysseus and the Sirens game in Fig-
ure 4.7c in which the decision to be tied to the mast is taken before deciding
whether or not to listen to the Sirens. Does it change the equilibrium?
4.5 In the spring of 1519, Hernán Cortés left Cuba for the Mexican coast with 508
men, 16 horses and a few pieces of artillery. They disembarked and founded the
city of Veracruz. There they knew about the existence of Moctezuma, the Aztec
emperor in a splendid city in the interior who commanded a grand army. They
also knew that many cities under Aztec rule disliked the Aztecs and that they
might serve as allies. They decided to try and conquer the Aztecs. Moctezuma
knows about the Spaniards and must decide whether to be ready to fight or be
friendly to the strangers. If Moctezuma responds with hostility, the Spaniards
have the option to fight or flee. In the first case, losses will be large on both
sides (−15, −15). In the second case, neither will win nor lose anything (0, 0). If
Moctezuma did not respond with hostility, Hernán Cortés and his men would
gain 10, while Moctezuma would lose 10.
(a) Represent the extensive form of the game.
(b) Find the solutions by backwards induction.
Hernán Cortés is considering changing the game by adding a new action before
going on the conquest of Mexico: he can order his men to sink the ships. If he
does this, the possibility of fleeing is taken away. The news about the ships will
certainly reach Moctezuma.
(c) Represent the game after Hernán Cortés sinks the ships and find its equilib-
rium by backwards induction.
(d) Represent the entire game, including the decision whether or not to sink the
ships as a first move by Hernán Cortés.
4.6 Three countries (A, B and C) must decide whether to declare war on D. If two
or three do, they will win without any problem and will be rid of a common
enemy, something that they value at 100. Winning the war costs 120, a quantity
that will be divided equally among the countries that declare war. If only one
declares war, nothing will happen.
(a) Find the Nash equilibria in pure strategies if countries decide simultaneously.
(b) Find the subgame perfect Nash equilibria if the decision is sequential, first
A, then B and C, in that order, and if each one knows what the previous
country has decided.
4.7 Two fishing boats work in a lake. Each one of them can be greedy or conscious.
If both are conscious, each can capture fish worth 500. If both are greedy, they
will exhaust the resource and captures will be worth 100 for each ship. If one is
greedy and the other one is conscious, their captures will be worth 600 and 50,
respectively. The decision about how much to fish is done simultaneously and
independently.
(a) Find the Nash equilibria. Are they efficient?
Being aware of the bad results in the equilibrium, they resolve to sign a con-
tract stipulating that the greedy ship will pay a penalty of 300. Thus, in a first
moment, the owners of the ships must simultaneously decide between signing
79
GAME THEORY
the contract or not. If they both sign, the contract is valid and can be lawfully
enforced. If either of them does not sign, the contract is not valid and the game
will be as in (a).
(b) Find the subgame perfect Nash equilibria in pure strategies.
(c) Repeat the exercise in case they make the decision about signing the con-
tract sequentially, with the second ship knowing what the first one did.
4.8 Two firms compete in prices within a market with the demand detailed in the table.
Firm 1 decides first and announces a price, and then Firm 2 decides on a price also,
having observed Firm 1’s decision. Consumers will buy from the firm that offers
the lowest price according to the demand. If both firms choose the same price, they
will share the market equally. Both firms incur a unitary production cost of 3.
P 1 2 3 4 5 6 7 8 9 10
Q 90 80 70 60 50 40 30 20 10 0
(a) Find the subgame perfect Nash equilibria.

A CEO in Firm 1 proposes a new marketing strategy. In addition to announc-
ing its price, Firm 1 will publish a compromise by which it will match the rival’s
price in case this one has a lower price. According to the country’s law, this kind
of public compromise is binding, so if the firm does not honor its compromise,
it will pay a penalty of 100. The consumers’ association states that this strategy
is contrary to competition. The firm argues that it guarantees a low price and
therefore favors competition.
(b) Who is right? To answer this question, find the new subgame perfect Nash
equilibrium and compare the result with the equilibrium in (a).
4.9 Three neighbors must agree to repair a common zone. The best budget implies
a cost of 210 and each neighbor values the project at 100.
(a) Assume that each neighbor decides voluntarily how much to pay and that the
decisions are made simultaneously. The project is undertaken if the voluntary
contributions are enough to pay for the project. Which are the Nash equilibria?
(b) Same as before, but now the decisions are made sequentially; also, when
a neighbor makes his decision, he is aware of the decisions made by those
before him. Which are the subgame perfect Nash equilibria?
(c) Now assume that the neighbors simply vote either Yes or No. The project
goes ahead if all say Yes. Also, they know that if they go forth with the pro-
ject they will pay the cost in equal shares. Find the Nash equilibria if the
decision is simultaneous. Also, find the subgame perfect Nash equilibria if
the decision is sequential.
4.10 In the game of the debt payment, repeat the analysis after changing the follow-
ing data. Make one change at a time, and do not consider all of them at the
same time. In case of indifference, assume that the bank gives the loan.
(a) The probability that the money is not well spent is 0.1 instead of 0.2.
(b) In case the payment of the debt is not a priority, the amount of 50 will be
used by Freedonia and the bank will obtain nothing.
80
5 VOTING
In this chapter we present a brief summary of some of the most important findings in
the theory of voting. The seemingly technical analysis of the properties of different
voting systems leads to deep insights about the meaning of democracy and about the
necessary imperfection of any institution to make collective decisions. Next, we detail
the map of this chapter to clarify the purpose of each section for the understanding of
the general ideas in the theory of voting.
We start with the concept of sincere voting to immediately present the strategic
vote (often also called useful vote) as the appropriate concept to study voting mecha-
nism from the game theoretical perspective (Section 5.1). Throughout the chapter it
will become clear that strategic voting goes beyond compromising and voting for a
second-best candidate to minimize the chances of a least preferred one being elected.
In Sections 5.2 through 5.5, we present a series of examples that illustrate the kind
of problems one may encounter in a voting mechanism. They all have to do with the
fact that, when the voting mechanism translates the declared individual preferences
(strategic or sincere) into an aggregated “social preference,” some consistence of the
preferences may be lost. The next natural question is, then, whether a voting mecha-
nism exists that does not present those problems. The Arrow theorem shows that the
answer to this question is negative: all voting systems present some problems (Sec-
tion 5.6). The negative theorem is compounded by yet another negative result that says
that there is no voting mechanism in which individuals always have an incentive to
always vote sincerely (Section 5.7).
After the negative results, an important research line studies the properties of the
different voting systems, while another one tries to find particular cases in which the
general negative results can be avoided. We provide some examples of both in Sections
5.7 through 5.10. In Section 5.11 we discuss the meaning of the expression “what
the people want.” Section 5.12 relates the theory of voting to an important problem
in philosophy called the discursive dilemma. The last section presents a case study.
5.1 Sincere and strategic voting

The 100 inhabitants of a certain village are divided in three groups according to their pref-
erences in regards to three different projects: X, Y and Z. The groups are labeled A, B and
C and have 40, 35 and 25 supporters, respectively. Preferences are displayed in Table 5.1.
Table 5.1 Voting one of three projects
A B C
X Y Z
Y Z X
Z X Y
81
GAME THEORY
The city council can only finance one of the three projects and calls on citizens to
vote. The voting system will be sequential. First, there will be a poll to decide between
projects X and Y, and the winner will then move forward to a second round against
project Z to reach the final decision. If everyone votes sincerely, the votes by group
between X and Y will be (X, Y, X), where the results are written in the order A, B, C.
That is, Groups A and C will vote for X, while Group B will vote for Y. Project X gets
the majority of votes and so continues to the second round. Then, between X and Z the
votes will be (X, Z, Z). Thus, project Z will be undertaken.
Let us see what happens if we regard the above scenario as a game among the
three groups in which each group acts as one single player. If the reader thinks this is
too much to assume, think of the groups as parties, the 100 voters as representatives
in a parliament, and assume that all representatives follow their party’s discipline.
As we discussed in Chapter 4, we analyze this dynamic game starting at the end and
then proceeding backwards. There are two possibilities – two different subgames –
after the first round of voting: subgames following X’s win and subgames following
Y’s win.
If X and Z are the projects to vote for in the second round, the Nash equilibrium is
(X, Z, Z), indicating that Group A votes for X, while Groups B and C vote for Z. As is
almost always the case in voting games, there are more equilibria. In this case, (X, X,
X) or (Z, Z, Z), but the reader can confirm that at least one group is using a weakly
dominated strategy: Groups B and C in the equilibrium (X, X, X), as well as Group A
in (Z, Z, Z). In a similar fashion, if the vote in the second round is between Y and Z,
the Nash equilibrium in non-weakly dominated strategies is (Y, Y, Z), and project Y
will be selected.
Now we can solve the voting between X and Y in the first round. All groups know
that if X is elected in this first round, then Z will win the second. They also know that
if Y is elected in the first round, it will also be elected in the second and last round.
In other words, to vote for X in the first round is the same as voting for Z in the second,
while voting for Y in the first round means favoring Y in both rounds. Taking this into
account, in the first round the equilibrium is (Y, Y, X), and Y will win in the first and
second rounds.
We can observe a disparity between sincere and strategic voting. In the first round,
Group A knows that their preferred project, X, will never be elected in the second
round and will, consequently, vote for its second-most preferred project, Y, in order to
prevent project Z, their least preferred candidate, from being ultimately chosen.
5.2 The manipulation of the agenda

In Section 5.1 the order of the votes was fixed and exogenous to the voting mechanism.
Let us introduce a new player in the game, the city mayor, who has the prerogative
to decide on the order of the votes. There are three possibilities, determined by the
projects in the first round: X vs. Y, X vs. Z and Y vs. Z. Which one the mayor chooses
will depend on his personal preferences in regards to the projects. We have already
established that X vs. Y in the first round means that Y will win in the end. Therefore,
a mayor belonging to Group B will find this particular order attractive.
If the mayor is a member of Group A, which order will she choose? Let us see if any
of the other orders results in the selection of a project more attractive than Y.
82
VOTING
First, consider that the vote starts with X vs. Z. As in the previous section, we
proceed to analyze the two possibilities in the second round. If X were to win the first
round, in the second round, X vs. Y, the Nash equilibrium in non-weakly dominated
strategies is (X, Y, X), and X would win. On the other hand, if Z were to win in the
first round, then the vote in the second would be (Y, Y, Z), and Y would win overall.
In this case, a vote for Z is actually a vote for Y. Anticipating all of this, players would
vote in the first stage between X and Z as if it were a vote between X and Y, and the
Nash equilibrium would be (X, Z, X), with the result that X would win both the first
and the second rounds. Note that now it is Group C that votes strategically for X, their
second choice, instead of voting for Z, to prevent it from losing against Y in the second
round. Therefore, this will be the preferred order should the mayor belong to Group A.
In Exercise 5.1 the reader can confirm that a mayor from Group C will choose the
order that starts out by placing Y vs. Z.
5.3 Condorcet’s paradox
A voting mechanism is just a way of aggregating individual preferences in order to
determine a social preference. In principle, one wants to ensure that the aggregation
satisfies a minimum of good, reasonable properties. For instance, it seems like a good
idea that the social aggregation takes into account votes from all citizens, and not just
the vote of a dictator.
Another interesting property is that the aggregation maintains some coherence.
Say that social preferences (the voting system) choose X rather than Y, and Y rather
than Z, then we should expect a coherent voting system to choose X before Z in order
to preserve the property of transitivity. However, with the individual preferences in
Figure 5.1, we can observe that a very standard procedure to aggregate preferences, the
majority rule, will determine that X is socially preferred to Y (Groups A and C prefer
X over Y), and that Y is socially preferred over Z (Groups A and C). However, it will
also determine that, contrary to what transitivity dictates, Z is preferred to X (with
Groups B and C in favor of Z). The possibility that a voting system as natural as the
majority rule may be unable to select a socially most preferred alternative due to lack-
ing the transitivity property is called the Condorcet’s paradox. In other words, even if
individuals have transitive preferences, for some aggregation mechanisms, the society
as a whole may not.
5.4 Referendum with a minimum participation

Here we present another reasonable property to demand from a mechanism of prefer-
ence aggregation. Say all electors show their preferences, and that a voting mechanism,
using those preferences as an input, chooses alternative K. Suppose now that we have
made a mistake reading the preferences of one voter; we had thought that she voted for
H, when in fact she voted for K. Given that K had been elected before, then it would
make sense that K is still selected when we find that support for it is even bigger. Alter-
natively, we can say that if H had not been elected before, it should not be elected now
with even weaker support. This is the idea behind the monotonicity property.
It may come as a surprise to know that there are important decisions taken by a
voting mechanism that does not satisfy this property. For instance, it is often the case
83
GAME THEORY
that a referendum requires a minimum participation for the result to be acceptable.

Say that there is a vote to decide YES or NO, where YES means a change and NO
means maintaining the status quo. If a minimum participation of 50% of the census
is required for the winning option to be accepted, then the two options will be in
an asymmetric position. The NO option has an advantage. Imagine that 45% of the
people prefer YES, while 30% prefer NO and the remaining 25% do not have a clear
position and will therefore not vote. If all supporters of the YES and NO positions vote,
the referendum will give a result of 60–40 with a 75% participation (the proportion of
YES-NO voters is 3 to 2). This is a clear result in favor of the YES position. However, if
the voters that favor NO stay home and do not go to the polls, the result then is 100–0
in favor of the YES position, but with a participation of 45%. Since a minimum partic-
ipation of 50% is required, then the referendum bears no consequence and the status
quo continues, which is what the NO position favors. In other words, if the NO posi-
tion loses voters, it will win, contradicting the monotonicity principle.
This is exactly what happened in Hungary on October 2, 2016, when 98% of voters
chose to support a law against applying the European norms of refugees. Participation
was only 40%, less than the required 50% for the referendum to be considered valid, so
Hungary continued applying the European law, at least for the time being.
The referendum on the independence of Montenegro, celebrated on May 21, 2006,
also required a minimum participation of 50% of the census, and support for inde-
pendence of at least 55% of the votes. Participation was 86%, and the p ro-independence
ballot obtained 55.5% of the votes. Independence was therefore gained. Had the
anti-independence voters stayed home, then they would have won instead. In Exercise
5.2 we ask the reader to formulate this case as a coordination game.
To impose a minimum participation in a referendum is not an inconsistent require-
ment, but organizers should note its potential distorting element. The minimum par-
ticipation could be required with some qualifications, perhaps making it conditional to
a difference in votes between the two options.
5.5 The Borda count

There are voting mechanisms in which the elector gives a different number of points to
different options and then, after all points are counted, the alternative with the most
points wins. This is the system used in the Eurovision Song Contest in order to select
the winning song. It is also used in some parliamentary elections and in some elections
which decide who gets a high ranked position in some professional or academic insti-
tutions. A simple example goes like this: there are ten options and each voter is to give
ten points to her favourite, nine points to her second favourite and so on. This system
is known as the Borda count.
The idea of this system is that if individuals only vote for their most preferred alter-
native, the intensities of those preferences will not be taken into consideration by the
mechanism. If candidates A and B get almost the same number of votes, but with A get-
ting slightly more, there is still a reason to elect B over A if A’s supporters hold B in high
esteem, while B’s supporters regard A as their worst candidate. The Borda count is also
defended for cases in which more than one candidate or alternative is to be elected. If vot-
ers only vote for one, then the order of the candidates according to the number of received
votes may not be a good indicator of the second- and third-most preferred options.
84
VOTING
However, this rule has one problem: it does not satisfy another property of inter-
nal coherence. This time it is the property of independence of irrelevant alternatives.
The following example shows what this property is about. A client arrives at a restau-
rant, asks the waiter for the menu and the following dialogue takes place:
-Today we have two dishes, fish and lasagna.

-I will have the fish, please.
-Forgive me. I forgot to mention that we also have frog legs.
-In that case, I will change my selection. I will have the lasagna.
This strange change of choice is prevented by the property of independence of irrele-

vant alternatives. More precisely, we can enunciate the property as follows: an election
mechanism satisfies the principle of independence of irrelevant alternatives if the aggre-
gate order between alternatives X and Y depends only on the individual preferences over
these two alternatives. In our example, the client is the only voter and even then the
principle is not satisfied. Her preferences on “fish” and “lasagna” depend on the exist-
ence of a third seemingly irrelevant choice, a piece of information that should not be of
consequence according to the principle. Let us examine how the Borda count may fail
to satisfy this principle. Suppose that now five friends go together to the restaurant and
that they all must choose the same option (they get a special price if they do so). Their
preferences over fish and lasagna are displayed in Table 5.2a.
Table 5.2a Friends’ preferences
Friend 1 Friend 2 Friend 3 Friend 4 Friend 5

Fish Fish Lasagna Fish Lasagna
Lasagna Lasagna Fish Lasagna Fish
They make the decision using the Borda count, so each friend gives two points to
their most preferred option and one point to the least. If they vote sincerely, the “fish”
option will have eight points, and the “lasagna” just seven. So, fish it is.
When the waiter tells them that, in addition, they have frog legs, the preferences
are like the ones in Table 5.2b. Note that each friend now maintains the same order
between the options “fish” and “lasagna.” Each one, however, places the new option in
a different place in the ranking.
Table 5.2b Friends’ preferences with frog legs included
Friend 1 Friend 2 Friend 3 Friend 4 Friend 5

Fish Fish Lasagna Frog legs Lasagna
Lasagna Lasagna Frog legs Fish Frog legs
Frog legs Frog legs Fish Lasagna Fish
The friends must now give three points to the best option, two to the second and
one to the last. This way, the “fish” option gets 10 points, “lasagna” gets 11 and finally,
85
GAME THEORY
“frog legs” has 9. Now they will eat lasagna, violating the principle of independence of
irrelevant alternatives. The choice between fish and lasagna depends not only on the
individuals’ preferences between the two dishes, but also on the existence of the third
alternative and the way it is ranked by the friends.
5.6 Arrow’s theorem
In the previous three sections, we have seen examples of election rules that lack different
coherence properties. It would be great if we could devise a voting system that is always
coherent, regardless of how individual preferences rank the different options. Kenneth
Arrow proved mathematically that the only system that satisfies the properties of transi-
tivity, monotonicity and independence of irrelevant alternatives for all possible individual
preferences is the dictatorial system, where only the preferences of one pre-specified indi-
vidual are taken into account (Arrow, 1950). This important result in the theory of social
choice is known, in his honor, as Arrow’s impossibility theorem. He won the Nobel Prize
in Economics in 1972 for his contributions to the study of the problems of aggregation
of preferences. Of course, neither Arrow nor anyone else would defend dictatorship as
an adequate voting system. What the theorem states is that any sensible system to make
societal decisions will lack some coherence, and that we must live with imperfection not
because humans are imperfect, but because a perfect method to aggregate individual
preferences does not exist. The theorem also helps us to understand the attraction some
people have for dictatorships. To understand is not to justify but to have more elements
to make a correct diagnosis of a problem and to be in a better position to resolve some
social situations that may lead to undesired consequences – such as dictatorships.
The temptation of not weighting

Next, we propose another territory to apply Arrow’s theorem. Say that instead of deal-
ing with N voters, we have N ends or ideals that one would like to see fulfilled in a
society (liberty, equality, fraternity, security, responsibility, solidarity, justice, identity,
etc.) and that rather than choosing among strategies, the alternatives are the different
kinds of societies that one has to evaluate. This makes sense if, when judging a society,
one focuses on how well it is ranked in the different values (liberty, equality and the
rest) and tries to find a coherent way of weighting each one of them. It is as if inside
one’s mind, liberty, equality, fraternity and the other values were voting to give every
society a global evaluation in order to draw a ranking of all societies.
One must now see how much of each of the ideals is satisfied for every society. Sup-
pose that a given society has a lot of liberty, a little security, is average in its provision of
justice, and that only these three values matter. For a person that values liberty much
more than the rest of those values, this society will rank high on her preferences. A person
that values security the most will have that same society in the lower end of his ranking. If
both liberty and equality are equally weighted, or if justice is the most valued, the society
will rank as an average one. Arrow’s impossibility theorem implies that there is no way to
weight all the ideals in a coherent way that is not dictatorial, something that in this con-
text means that the individual judges a society paying attention to just one of the ideals.
The same way that one can understand the temptation of a dictatorship for those
minds unable to cope with the imperfections of democracy, we can also understand the
86
VOTING
temptation to restrict our moral positions and identify ourselves with just one ideal.
This way everything is simpler. If one identifies liberty as the supreme good, one only
needs to determine how a society performs in this dimension to compare it with other
societies, real or proposed. This simplicity in the analysis will be regarded as deep and
superior to others, having found a coherent method of evaluation. It will be coherent,
yet dramatically incomplete. The same happens if one chooses equality, the glory of the
motherland or any other ideal as the supreme good.
Instead of this attitude there is a better way, one that meets more goals and opts for
a moral democracy, where all those goals include a vote to decide on what is morally
acceptable. This is better in spite of its imperfections and incoherence, and despite the
complexity it adds to the analysis. Or perhaps, it is better not in spite of all that, but
precisely because of that, since we will force ourselves to be much more vigilant, which
is an attitude that one should demand in questions of moral choice.
5.7 The theorems by Gibbard–Satterthwaite and May

When we analyzed the problems with transitivity, monotonicity and independence of
irrelevant alternatives, and enunciated Arrow’s theorem, we took individuals’ prefer-
ences as given. These preferences can be the real ones or the ones that individuals
declare strategically. Philosopher Allan Gibbard and economist Mark Satterthwaite
asked about the incentive of individuals to declare their true preferences when par-
ticipating in a mechanism to aggregate preferences, such as a voting system (Gibbard,
1973; Satterthwaite, 1975). The result is another impossibility theorem, very much like
Arrow’s. The Gibbard–Satterthwaite theorem states that any voting system that is not
dictatorial and that does not veto any option is susceptible to strategic voting by some
voter, meaning that he will not vote according to his real preferences, like in the exam-
ples of Section 5.1.
Among so many negative theorems, there are some ways out. For instance, if the
voters’ preferences are very homogeneous it is sometimes possible to find a coherent
voting mechanism. Another way is to decide the winning option randomly, where the
probability for an option to win is proportional to the votes it obtains. One of the best
known positive results is May’s theorem, which states that for decisions between only
two alternatives, the majority rule is the only one that satisfies (i) neutrality, (ii) ano-
nymity and (iii) positive response (May, 1952). Neutrality is defined as follows. Say that
for some preferences A is selected over B. Now, if we exchange places between A and
B in the ranking for all individuals, then B should be selected. Anonymity means that
the identity of the voters is irrelevant for the outcome. Finally, positive response means
that if we are in a tied situation and a voter changes her preferences, then she decides the
winner. There are similar theorems that identify and define the properties that charac-
terize other voting rules and are specific to them, like the Borda count and two-round
elections.
5.8 The median voter theorem

In Section 3.2 we saw that in a society with two political parties, their electoral plat-
forms converged to the preferences of the median voter. In the case studied, the voters
were automata and the parties were the players that played strategically when choosing
87
GAME THEORY
their platforms. Since there were only two alternatives, there was no problem with stra-
tegic voting, and thus the behavior of the voters could be easily foreseen.
When there are more than two alternatives there is room for strategic voting. How-
ever, there is one instance in which voters will vote sincerely when using the majority
rule: when voters’ preferences are single-peaked. Furthermore, the elected option is the
one most preferred by the median voter. Let us illustrate these ideas with one example.
Single-peaked preferences
Let us order the alternatives over which individuals will vote. For example, if alternatives are
different budgets, we can order them from smallest to largest or according to the distance
from any given number. We say that a voter has single-peaked preferences over a particular
ordering of the alternatives if the voter has a best alternative and the other alternatives are
lower in the preferences as they are further away from the best one in that ordering.
Let us go back to the example in Figure 5.1. If we order the options alphabetically (X,
Y, Z), we will see that voter A has X as her best option, and that the others are getting
worse as they are further from X: Y is worse than X, and Z is even worse. Voter B has Y as
his best option and both X and Z are worse than Y (whether X is better or worse than Z
is irrelevant). Finally, voter C’s best option is Z, immediately before in the order is option
Y, which is indeed worse than Z, but still further we find X which is not worse, but
better than Y. This voter has no single-peaked preferences with respect to this ordering.
An illustration of the median voter theorem

Three voters (A, B and C) must choose between five projects to celebrate the festivities
of the city, with budgets 100, 150, 200, 250 and 300, respectively. Voters only care
about the cost of the project. Voter A is austere, and given any two choices, he prefers
the cheapest. Voter B is moderate, and C is in favor of a good celebration, although
she reckons 300 is a little too much and has 250 as her optimal budget. Preferences are
detailed in Table 5.3a.
Table 5.3a Single-peaked preferences
A B C
100 200 250
150 150 300
200 250 200
250 100 150
300 300 100
Figure 5.1 illustrates the preferences in a graph, where we can appreciate the peaks.
Each voter has a single peak from which alternatives further apart are ranked lower.
For convenience, we have given a value of five to the most preferred budget, four to the
second best, and so on. The precise values are irrelevant, as long as they preserve the
order. In the figure, A’s preferences are represented with a dashed line, while for B’s and
C’s we use a solid and a dotted line, respectively.
88
VOTING
A’s peak B’s peak C’s peak

5
100 150 200 250 300
Figure 5.1 Single-peaked preferences in a graph
To choose the project, several pairwise votes will be conducted, and the project that
wins the most times will be elected. Thus, in the vote between projects 100 and 150,
the second wins. Table 5.3b contains all pairings, votes and winners.
Table 5.3b Pairwise vote
Alternatives A’s vote B’s vote C’s vote Winner

100 vs. 150 100 150 150 150
100 vs. 200 100 200 200 200
100 vs. 250 100 250 250 250
100 vs. 300 100 100 300 100
150 vs. 200 150 200 200 200
150 vs. 250 150 150 250 150
150 vs. 300 150 150 300 150
200 vs. 250 200 200 250 200
200 vs. 300 200 200 300 200
250 vs. 300 250 250 250 250
We observe two interesting facts. First, the winning project in each pairing is
always the project most preferred by the median voter, which is B. Second, and as a
consequence of that first fact, the project most preferred by the median voter, the 200
budget, wins all its pairings and will be the chosen one. An alternative that wins all
its pairings is called a “Condorcet winner.” With all of these elements, we can state a
version of the median voter theorem that extends to more general cases than we have
seen in one example.
Median voter theorem: When all voters have single-peaked preferences over a set of alter-
natives ordered in a particular way, the alternative most preferred by the median voter is
a Condorcet winner. Furthermore, voters have no incentive to behave strategically and
will vote sincerely in majoritarian election systems.
In Exercise 5.6 we ask the reader to confirm the strategic vote in this example when the
voting mechanism is not majoritarian.
89
GAME THEORY
The empirical research has found that public expenditure is affected by the position
of the median voter. What is more, the work of Kang and Powell (2010) shows that this
effect is similar in different systems: in majority rule with one seat per district and in
proportional districts with multiple seats.
5.9 I’ll scratch your back and you’ll scratch mine

In a city with 2,000 citizens, parties A, B and C obtained 800, 700 and 500 votes,
respectively, in the last election, awarding the parties 8, 7 and 5 seats in the city coun-
cil. Each party has a favourite project, which is valued at 250 by each one of its voters;
any other project is worth zero. The cost of any project is 200,000, to be financed by
a special tax paid equally by all citizens. Knowing that each project benefits only one
party, the vote will be conducted in the following way. Each party will vote for one of
the three combinations (alliances) of projects: AB, AC and BC. If the combination AB
wins a majority of the votes, projects A and B will both go ahead. The same is true for
the other combinations. If no combination obtains a majority, then no project will be
carried out. If we discard a vote for the least popular combination – the one not con-
taining the favourite project – since it is a weakly dominated strategy, the game is the
one displayed in Figure 5.2.
B B
AB BC AB BC
AB 50, 50, −200 0, 0, 0 50, 50, −200 −200, 50, 50
A
AC 50, −200, 50 50, −200, 50 0, 0, 0 −200, 50, 50
AC BC
C
Figure 5.2 I’ll scratch your back and you’ll scratch mine
If a project is undertaken, this will mean a cost of 100 per citizen. If a given com-
bination of projects is selected, each of the voters that favors one of the projects in
the combination wins 250 − 100 − 100 = 50 , while each of the voters whose favourite
project is not approved loses 200. Despite the fact that passing a project that is not the
favourite implies only losses, it may be sensible to vote for it as part of a pact of the “I’ll
scratch your back and you’ll scratch mine” kind implicit in the vote of the combina-
tions of projects.
There are six Nash equilibria: {(AB, AB, AC), (AB, AB, BC), (AC, AB, AC), (AC,
BC, AC), (AB, BC, BC), (AC, BC, BC)}. Two parties coordinate in each of them to
vote for their projects. This kind of pact allows two parties to benefit at the expense of a
net social loss. Projects A, B and C give total benefits of 200,000, 175,000 and 125,000,
respectively. As each one costs 200,000, none of them deserves to be carried out.
This example warns us about the possibility that majorities can benefit while society
as a whole is worse off. A well-designed legislation must include provisions to prevent
this kind of action. The protection of minorities can be understood as one of these
provisions. When passing a legislation, it helps that the different political groups do
90
VOTING
not know in which position they will find themselves in the future. In the example in
this section, parties A, B and C would agree to have a legislation in order to avoid this
exchange of favors, also known as logrolling, before they know which of the equilibria
will prevail.
5.10 How to know the truth: The Groves–Clarke mechanism

Throughout the current chapter, we have seen how collective decisions face many
problems, including the hiding of own preferences for strategic reasons. We have seen
this in voting systems, but it may happen in other decision-making mechanisms as
well. For instance, say that the three neighbors, A, B and C, in a condominium must
decide whether to install an elevator in the building. The neighbors value the elevator
at 10, 20 and 30, respectively, while the total cost of the project is 42. The installation
is an economically efficient decision, since the total benefit (60, the sum of the valua-
tions) is higher than the cost. The fact that it is efficient implies that there is one way
everyone wins. For example, if they pay 4, 14 and 24, each neighbor wins 6 with the
installation.
The problem is that in order to make the decision and share costs in this way, they
need to know each other’s preferences, but these are private information. If the neigh-
bors decide to vote on whether to install the elevator, the vote will depend on how they
anticipate it will affect their pockets. We will look at different mechanisms. The first
four will serve to have an overview of the problems in these types of collective decisions,
while the fifth will show a way of obtaining truthful information about the preferences,
even if it implies some other kinds of problems.
Four mechanisms that induce false declarations

1. The elevator is financed through a public program by the government and will cost
nothing to the neighbors. It is enough that a majority of the neighbors votes in favor
in order to install it.
All neighbors will vote in favor and the elevator will be installed, which in this
case is the efficient thing to do. However, they would still have voted in favor even
if their valuations were lower than the cost, and the installation would then have
been a social loss.
2. Once again, the elevator is financed through a public program by the government
and will cost nothing to the neighbors. However, instead of voting, the neighbors
will now be asked to declare their valuations. The elevator will be installed if the
sum of the declared valuations is higher than the cost.
Now the best action for each neighbor is to exaggerate their valuations so that
city hall believes that the installation is efficient. As in the first case, the elevator
would have been installed even if the true valuations were not higher than the cost.
3. The elevator is installed if a majority votes in favor, and the cost will be divided
among the neighbors equally.
If the elevator is installed, each neighbor pays 14, or one third of 42. Neighbor
A will vote against it, and the others will vote in favor. Again, we would obtain
this result if valuations were 0, 15, 17, which make the installation economically
91
GAME THEORY
inefficient, since it carries more costs than benefits. This is the same problem with
“I’ll scratch your back and you’ll scratch mine” analyzed in Section 5.9.
4. The neighbors are asked to declare their valuation, and the elevator is installed if
the sum of the declarations is higher than or equal to the cost. If it is installed, each
neighbor will pay a part of the cost proportional to his declaration over the sum of
the declarations.
This case is more complicated. Suppose that all declare their true valuations: 10, 20
and 30. Neighbor A will have to pay 7, the proportion 10 of the cost of 42. Neighbor
60
20 30
B will pay × 42 = 14, and Neighbor C will pay × 42 = 21. The elevator will be
60 60
installed and the well-being of all neighbors will improve. This is not, however, a Nash
equilibrium: each neighbor will win if they under-report their valuation. For instance,
since the declarations of Neighbors B and C are already enough to make the decision
of installing the elevator, then A will be better off by declaring a valuation of zero. She
will enjoy the elevator and pay nothing.
There are many equilibria; in some the elevator is installed, while in others it is
not. Take the declarations (0, 16, 26). Note that the sum of the declarations is exactly
the cost of installation. Therefore, each neighbor will pay exactly their declaration,
and they will win 10, 4 and 4, respectively. Neighbor A does not want to deviate, as
this will only increase her payment. Neighbor B does not want to declare more, as he
will pay more, but he also does not want to declare less, for this will imply that the
elevator is not installed and then he will win nothing instead of 4. The same goes for
Neighbor C.
Declarations (2, 4, 8) also constitute a Nash equilibrium, but in this case the con-
sequence is that the elevator is not installed. If Neighbor C considers a deviation to get
the elevator installed, he must declare at least 36, the difference between 42 and the
sum of the declarations of the other two neighbors. But then he will also have to pay at
least 36, which is more than his own valuation. The reader can confirm that neither of
the other neighbors wants to deviate either.
In a more realistic interpretation of the game, with more neighbors and without
knowing each other’s valuations, the equilibria in which the elevator is not installed will
be more likely. Notice that the Nash equilibria in which the elevator is installed require
that the sum of the declarations is exactly equal to the cost, a highly unlikely event.
A truth-revealing mechanism
5. We ask the neighbors to declare their valuations and then we proceed as follows.
First, we calculate the sum of the valuations with and without the neighbor. If both
sums are at least as high as the cost, then the elevator is installed and the neighbor
pays nothing. If they are both lower than the cost, then the elevator is not installed
and again the neighbor pays nothing. If the sum of the valuations including the
neighbor is higher than the cost, but the sum without her is lower than the cost,
then the elevator is installed and the neighbor pays the difference between the cost
and the sum of the valuations without her. This is called the Groves–Clarke mech-
anism (Clarke, 1971; Groves, 1973).
92
VOTING
Before analyzing the equilibrium of the game, let us examine an example in order to
understand the mechanism. Say that neighbors declare (5, 17, 30). Since the sum is 52 –
more than the cost of 42 – the elevator is installed. Next, we compute the payments as
we explain the logic behind them:
Neighbor A: without her, the sum of the declarations is 17 + 30 = 47, so that the eleva-
tor would also be installed without Neighbor A. Thus, she pays nothing.
Neighbor B: without her the sum of declarations is 5 + 30 = 35, less than the cost. With
the declaration of Neighbor B included, the social decision changes from not installing
the elevator to installing it. She pays the difference between the cost and the others’
valuations, 42 − 35 = 7 .
Neighbor C: without him, declarations add up 5 + 17 = 22 so the elevator is not
installed. As was the case with Neighbor B, he pays the difference up to the cost,
42 − 22 = 20.
Having understood the calculations in the mechanism we can go on and see that to
declare true valuations is a Nash equilibrium and, what is more, that being truthful is
a weakly dominant strategy. So assume that the neighbors declare (10, 20, 30) and the
elevator is installed. Let us examine each neighbor’s best reply one at a time.
Neighbor A: neighbors B and C declare 20 and 30, respectively. What is A’s best reply?
The first thing to notice is that with these declarations the elevator will be installed,
regardless of what Neighbor A declares. The consequence of that is that A will pay
nothing, also regardless of what she declares. Thus, any declaration is a best reply by A.
Neighbor B: neighbors A and C declare 10 and 30, which means that without B the sum
is 40, less than the cost, so the elevator will not be installed. If Neighbor B declares 2 or
more, the elevator is installed and she pays 42 − 40 = 2, with a benefit of 20 − 2 = 18 for
her. She will not pay anything if she declares less than 2, but will not receive any benefit
either. For her, the only thing that matters is whether or not the sum is of at least 42
with her declaration, just how much more does not matter. So, telling the truth (20) is
a best reply along with any other number greater than or equal to 2.
Neighbor C: similar to B.
We have seen that to declare the truth is a Nash equilibrium. Let us now see that it is
also a weakly dominant strategy for each player. For example, for Neighbor A, could
there ever be a better strategy than telling the truth? Suppose that the sum of declara-
tions of B and C add up to X, a number smaller than 42.
1. If the difference between X and 42 is greater than 10, Neighbor A is not interested
in over-declaring her valuation to install the elevator, because she will have to pay
this difference. Under-reporting her valuation is not worth it either, as nothing is
accomplished with that.
2. If the difference between X and 42 is smaller than 10, Neighbor A wants to declare
a valuation that gets the elevator installed, since she will be paying less than her true
valuation regardless of her declaration. She is not interested in under-reporting it as
lower than the difference between X and 42, as that would result in the elevator not
being installed.
93
GAME THEORY
3. If the difference between X and 42 is exactly 10, then she is indifferent among all
her strategies.
The cases for the other neighbors are analyzed similarly. In fact, with this decision
mechanism, telling the truth is a weakly dominant strategy regardless of the number of
players, their valuations or the cost of the project.
The problem with the mechanism is that the sum of what the individuals pay may
not be enough to finance the cost of the project. In our example, the neighbors will be
paying 0, 7 and 27, respectively, making a total of 34, or 8 short of covering the cost
of the elevator. This is a general problem, and not just specific to the Groves–Clarke
mechanism. There is another impossibility theorem that states that there is no decision
mechanism that satisfies (i) the players reveal the true valuation and (ii) meets the
budget.
Even after the negative result, it is possible to reduce the problem by requiring
that the budget constraint is satisfied in average over all uncertain possibilities, or by
requiring that the additional necessary resources come from an account from which the
players do not expect any cost or profit.
5.11 Do we know what the people want?

In this section we provide an example of the difficulties that arise when trying to make
sense of the phrase “what the people want.” Suppose that in a society there are 100
people, divided in 6 parties (TP, UP, VP, XP, YP and ZP). They must decide among
five alternatives, for which they have preferences as shown in Table 5.4. The number in
parentheses shows the size of each group. In total, there are 100 people that may be the
members of parliament, while the groups may be political parties and the alternatives
may be different candidates for prime minister.
Which candidate should be elected? This is a tricky question. There is no such
thing as a “candidate that should be elected” (a normative statement) without a refer-
ence to a norm, and the norm may be one of many, without any of them being clearly
the fairest or the best. Consider five different voting systems (norms) and let us see who
wins if we respect the preferences of the individuals (i.e., if they vote sincerely).
Table 5.4 Do we know what the people want?
Party (# voters) TP (33) UP (16) VP (3) XP (8) YP (18) ZP (22)

A B C C D E
B D D E E C
Ranking
C C B B C B
D E A D B D
E A E A A A
• Plurality rule (relative majority): Each party votes for its most preferred candidate
and the one with the most votes wins.
A wins with 33 votes, compared to B’s 16, C’s 11, D’s 18 and E’s 22.
• Borda count: Each party assigns four points to its best candidate, three to the second
best, then two, one and zero. The candidate with the most points wins.
94
VOTING
B wins with 33 × 3 + 16 × 4 + ( 3 + 8 + 22 ) × 2 + 18 × 1 = 247 points, more than any

other candidate. For example, A gets 33 × 4 + 3 × 1 = 135 . The reader can check that
C, D and E get 244, 192 and 182 points, respectively.
• Condorcet winner: We make a list of all possible pairings of candidates, and the
candidate that wins all its pairings is selected. (Note: A Condorcet winner does not
always exist.)
C wins: When faced against A, C has 77 votes and A has 33. Against B, D and E,
Candidate C has 51, 66 and 60, respectively, a majority of the 100 voters.
• Single transferable vote: Individuals vote during a first round, and the candidate
with the least votes is eliminated. A second round is conducted with the remaining
candidates and, once again, the one with the least votes is eliminated. We proceed
this way until one candidate remains, and that candidate will be elected.
D wins: In the first round C is eliminated. Out of the 11 votes of this candidate, 3
go to D and 8 go to E in the second round (hence the term transferable vote) and B
will be eliminated. In the third round, B’s 16 votes go to D, and we have A with 33
votes, D with 37 and E with 30, and Candidate E is eliminated. In the last round D
wins against A with 77 votes.
• Double round: The two candidates with the most votes in the first round compete
in a second round in which the winner is elected.
E wins: In the first round, A and E are first and second, respectively. In the second
round, A gets 36 against the 64 that go to Candidate E.
The impossibility theorems, the theorems that show that different election rules satisfy
some properties but not others, and, finally, this example provide a pessimistic view
of the problems we can expect in collective decision making. All these are theoretical
results, and it may be that in practice things are not so negative. In fact, when real-life
elections are studied and researchers simulate what would have happened if another
voting system had been used, oftentimes the results would not have changed much.
Another lesson to learn is that the reasons to replace a voting mechanism with a differ-
ent one should be grounded in arguments that go beyond the fact that at one point it
produced a bad result, as this is something that could happen with any system.
5.12 The discursive dilemma

Here is an example of the discursive dilemma or doctrinal paradox. There are three
members of a jury that must decide by majority whether a candidate should be accepted
into a group. The rules specify that the candidate must fulfill two requirements, A and
B, in order to be accepted. The first member of the jury believes that the candidate
fulfills the two requirements, the second member believes she only meets A, whereas
the third member believes she only meets B. The paradox arises when one compares
two different methods to decide on the admission.
The first method requires that the jury votes first if the candidate meets requirement
A and then if she meets B. If the candidate obtains a majority in both votes, she will be
accepted. In our example this is indeed the case. Two jurors out of three (the first and
the second) think that she fulfills requirement A, and two jurors (the first and the third)
also think she fulfills B. According to this voting method, the candidate is accepted.
95
GAME THEORY
The second method requires that each jury votes directly whether they think the
candidate should be accepted. Now, only the first juror believes so, and the candidate
will therefore not be accepted. Table 5.5 summarizes the two methods.
Table 5.5 The discursive dilemma
Fulfills A? Fulfills B? Accepted?

Juror 1 Yes Yes Yes
Juror 2 Yes No No
Juror 3 No Yes No
Majority Yes Yes No
These two voting methods can be understood as two different ways to aggregate
information. The paradox shows that the majority rule delivers inconsistent results
when we aggregate information over premises compared to the case when we aggregate
conclusions. In the example, the conclusion follows the conjunction of the premises,
but other logical operations can be used to show the paradox. Furthermore, this prob-
lem is not specific to the majority rule. In fact, it is impossible to find an aggregation
mechanism that gives logically consistent judgments – always giving the same result
regardless of whether we aggregate premises or results.
5.13 A referendum in Catalonia

During the 2010s the Catalan government asked to have a referendum where the Cat-
alans could vote for independence from Spain. The central government denied that
possibility on the grounds of it being unconstitutional, something the Constitutional
Court confirmed, as the sovereignty resides in the whole people of Spain. Such a ref-
erendum would imply that the sovereignty may be divided, and therefore it could only
be conducted if the constitution is changed first. This also implies that negotiations for
a change in the constitution and for the terms of a referendum might include provisions
for dividing the potential sovereignty of Catalonia among the four provinces. In that
way, even if there were a majority in favor of independence in Catalonia as a whole, if
a province does not back the independence, it would remain in Spain. As a matter of
fact, in the local elections in 2015 and 2017, the pro-independence parties got less than
48% of the votes (although they got a majority of the seats in the local parliament). By
provinces, pro-independence parties got a majority of the votes in Lleida and Girona
and a minority in Barcelona (by far, the most populated) and Tarragona, although in
this last one, the numbers were very close.
Let us say that, after the appropriate changes in the constitution, a referendum is
negotiated. If there is a sufficient majority in all provinces, there is nothing of interest
for us to study. Thus, assume that there is this majority in all provinces except in Bar-
celona, and let us see how the way the referendum is conducted alters the outcomes.
First notice that if provinces may not opt out, then there are only two results: inde-
pendence and no independence. May’s theorem applies and people will vote sincerely.
Barcelona will be part of the independent Catalonia even if they voted against it.
The interesting cases arise if a province may remain part of Spain if it does not back the
96
VOTING
independence. Now there are at least three possible results: independence with three
provinces, independence with four provinces and no independence. According to the
Gibbard–Satterthwaite theorem no decision mechanism is free from strategic voting.
Before analyzing the different possibilities, let us consider the preferences of the piv-
otal voters in each part of Catalonia. For simplicity, assume two possibilities, displayed
in Table 5.6. Pivotal Catalans outside Barcelona prefer no independence if that implies
leaving Barcelona as part of Spain. The two cases we are considering differ in the prefer-
ences of the pivotal voter in Barcelona. According to the first preferences (Barcelona 1),
the last thing they want is the provinces to be divided in two states, but they prefer to
be together as part of Spain. According to the second preferences (Barcelona 2), the last
thing they want is to be part of an independent state, even if that means being divided.
We will assume that most voters will vote sincerely but that some voters are willing
to vote strategically as a group. For the strategic vote to have an impact it must be the
case that they can affect the result. This means that they are against independence in
Barcelona and pro-independence in the rest of Catalonia.
Table 5.6 Preferences for independence
Catalonia except Barcelona Barcelona 1 Barcelona 2 Value

Independence all No independence No independence 2
No independence Independence all Independence part 1
Independence part Independence part Independence all 0
Now we will consider different arrangements for the referendum: a single referendum
and two possibilities for a sequential vote with a referendum first and then an additional
opportunity for reconsidering the position after observing the referendum’s results.
Case 1: A single referendum

The question in the referendum calls for a vote of YES or NO to independence. If
Barcelona votes NO and the rest vote YES (and assume that in this case there is a
majority of YES in the whole region), then Barcelona will not join the rest of the
provinces in the independent state. Figure 5.3 shows the game with the preferences
of Barcelona 1.
There are two Nash equilibria in this game: (YES, YES) and (NO, NO). Either all
vote for independence or all vote to remain. However, in the second equilibrium the
pivotal voters in Barcelona will be using a weakly dominated strategy. The most likely
outcome in this case would be the independence of the whole territory.
Barcelona 1
YES NO
YES 2, 1 0, 0
Rest of Catalonia
NO 1, 2 1, 2
Figure 5.3 The referendum game with preferences as in Barcelona 1
97
GAME THEORY
When we consider the preferences as in Barcelona 2, things are different. The new
game is displayed in Figure 5.4, with only one Nash equilibrium: (NO, NO). Now
voting NO is a weakly dominant strategy for the pivotal voters in Barcelona, and in the
rest of Catalonia they will vote NO in order not to divide the provinces.
Barcelona 2
YES NO
YES 2, 0 0, 1
Rest of Catalonia
NO 1, 2 1, 2
Figure 5.4 The referendum game with preferences as in Barcelona 2
Case 2: Barcelona has the last word

There is a referendum. If Catalonia as a whole votes YES, but the province of Barcelona
votes NO, then Barcelona has to decide whether to remain in Spain or to go with the
rest to form a new state. With the preferences as in Barcelona 1, the decision of the piv-
otal voters in Barcelona would be not to split the provinces and go with the rest in the
independent state. Anticipating this, the pro-independence citizens have no problem
voting according to their true preferences and the result would be independence after a
vote (YES, NO) in the referendum and a vote in Barcelona to go with the rest after it.
However, if the preferences are like in Barcelona 2, they will decide to remain in
Spain. Now, the pivotal voter in the rest of Catalonia will vote NO to avoid the division
of the region. The result will be a (NO, NO) vote in the referendum.
Case 3: The rest of Catalonia has the last word

Consider that in the referendum Barcelona votes NO and Catalonia as a whole votes
YES. Now the pivotal voter in the rest of Catalonia will prefer not to divide the region
and will prefer that all remain in Spain. Anticipating this, the pivotal voter in Barce-
lona will vote against independence regardless of whether their preferences are like in
Barcelona 1 or Barcelona 2. The result would be a (YES, NO) vote in the referendum
and a decision by the rest of Catalonia not to activate the independence in order not to
divide the provinces.
Discussion
Table 5.7 summarizes the results. According to our analysis, the Spanish government
should aim at a mechanism in which the rest of Catalonia has the last word, while a
pro-independence local government in Catalonia should push for a referendum only or,
at least, for a mechanism in which Barcelona has the last word to opt out. Also, pivotal
voters in Barcelona should try to convince the pivotal voter in the rest of Catalonia
that their preferences are like in Barcelona 2, even if that is not the case. The reader
may be wondering which mechanism better respects the preferences of all the voters
in Catalonia. Here is the bad news: there is no such thing as a correct decision. There
are different ways to organize the decision that lead to different results, and there are
different degrees of social acceptance.
98
VOTING
Table 5.7 Outcome depending on preferences and on referendum design
Referendum + opt out
Referendum only Barcelona has the Rest of Catalonia

last word has the last word
Preferences Independence all Independence all No independence
Barcelona 1
Preferences No independence No independence No independence
Barcelona 2

5.1 In the game of agenda manipulation in Section 5.2 show that if the mayor is a
member of Group C, she will decide that Y and Z are voted in the first round
and then the winner is faced against X.
5.2 Monteblanco is holding a referendum to decide on its independence from Bal-
cania. For the independence to go forward, a minimum participation of 50% as
well as 55% of votes in favor are both necessary. It is estimated that 15% of the
potential voters will not vote. Among the remaining 85%, the polls show that
the vote can be anywhere between 40% and 50% in favor of the YES option
and between 45% and 35% in favor of the NO option. Assume that all voters
in favor of independence will vote YES, and consider the game among citizens
against independence. In the following exercise we will assume that to vote is
costless and that these voters are only interested in maximizing the probability
that there will be no independence.
(a) Show that the strategy profile in which all voters against independence
choose NO is a Nash equilibrium.
(b) Show that the strategy profile in which all voters against independence
choose not to vote is also a Nash equilibrium.
(c) Which of the two equilibria works better for the NO supporters?
(d) Which game shown in Chapter 1 most closely resembles this game?
5.3 Consider the voting mechanism using the Borda count in Section 5.5 when
preferences are like the ones displayed in Table 5.2b. Consider also that for each
friend, the utility of their preferred option being elected is 4, the utility of the
second best is 2 and of the third, 1. In case two or more options get the same
points, they will pick one of them with equal probabilities. Show that to vote
sincerely is not a Nash equilibrium.
5.4 Consider a parliament where three parties (A, B and C) have all equal rep-
resentation and have preferences over four projects (W, X, Y and Z) as shown in
the table below. To select one project, they will follow the procedure described
in Section 5.2. First there will be a vote between two projects and the winner
will be paired in a second vote against a different project. The new winner will
then be paired against the fourth project.
99
GAME THEORY
A B C
X W Y
Z X W
Y Z X
W Y Z
The president of the parliament is corrupt and has a vested interest in favor of
project Z, despite the fact that X is unanimously preferred over it.
(a) How can she manipulate the agenda if parties vote sincerely? Hint: Since
all parties prefer X over Z, she should try to eliminate X as early as possible.
(b) Do the choices in (a) constitute a subgame perfect Nash equilibrium?
5.5 The 17 members of a committee are divided into 4 groups with preferences over
3 options (X, Y and Z) as shown in the table below. The members follow the
group’s discipline so that each group behaves like one player.
A B C D
6 votes 5 votes 4 votes 2 votes
X Y Z X
Y X Y Y
Z Z X Z
The winner among the options will be decided in double-round voting: the
two options that receive the most votes in the first round pass on to the second
round, where the winner is decided by a majority vote.
(a) Assume that the members vote sincerely. Which option will win?
(b) Show that voting sincerely is not a Nash equilibrium. Hint: Consider devi-
ations by Group D.
(c) Show that the strategic vote in D’s deviation from the sincere voting implies
that this double-round mechanism does not satisfy the monotonicity
property.
5.6 In the example that illustrates the median voter theorem in Section 5.8, con-
sider the following mechanism. First there is a vote between two options, and
the winner and a third option are paired in a second vote. Then the winner
is again paired against a fourth, and finally that winner is paired against the
remaining option.
(a) Show whether voting sincerely is a subgame perfect equilibrium when the
options are paired in an increasing order (first, options 100 and 150 are
paired, then the winner is paired against 200 and so on).
(b) Show whether voting sincerely is a subgame perfect equilibrium when the
options are paired in a decreasing order.
(Notice that the mechanism is not a majority mechanism in the first round, so
there is no contradiction with the median voter theorem.)
100
VOTING
5.7 Consider the example in Section 5.8 and add two more voters with single-peaked
preferences. (Avoid having two or more median voters.)
(a) Who is the median voter in the new game?
(b) Show that in all pairwise votes the winning alternative is the one that the
median voter chooses.
5.8 There are four parties in a parliament, A, B, C and D, with 114, 81, 65 and
40 seats, respectively. Each party is one single player, and they have to select
a prime minister. Each party proposes a candidate, and their preferences over
them are shown in the table below.
A B C D
A B C D
C D A B
B C B C
D A D A
(a) Consider the plural voting system: there is a vote on all candidates, and the
one with the most votes is elected. Who will be elected if parties vote sin-
cerely? Is this a Nash equilibrium?
(b) In the plural voting system, find three Nash equilibria satisfying all of the
following conditions: (i) in each equilibrium a different candidate is elected,
(ii) at least one party is voting for its most preferred candidate and (iii) no
party is voting for its least preferred candidate. Show also that there is no
Nash equilibrium satisfying (ii) and (iii) in which D is elected.
(c) Is there any Nash equilibrium in which no party votes for its best or its worst
candidates?
5.9 Consider again Exercise 5.8, but now use the double-round voting system: the
two candidates with the most votes in the first round go on to the second
round, in which the candidate with the most votes is elected. Given that the
second round is the last one and that the vote here is between two options, you
can assume that parties will vote sincerely in it. Consider now just the game in
the first round (where winners are determined automatically, and not strategi-
cally, by sincere vote in the second round).
(a) Find the winner if parties also vote sincerely in the first round.
(b) Find two Nash equilibria satisfying all of the following conditions: (i) in
each equilibrium a different candidate is elected, (ii) at least one party is
voting for its most preferred candidate and (iii) no party is voting for its least
preferred candidate. Show also that there is no Nash equilibrium satisfying
(ii) and (iii) in which D is elected.
(c) Assume that D is dropped from the list of candidates. Show that voting
for the most preferred candidate in the first round is not a subgame perfect
Nash equilibrium.
(d) Assume that no one votes for D or for their least preferred candidate. Find
all subgame perfect Nash equilibria.
101
GAME THEORY
5.10 Consider again Exercise 5.8, but now use the following sequential voting mech-
anism: first the candidate of the party with most seats (A) presents its candidate
for an approval vote, where parties must vote YES or NO. If he gets a majority
of YES votes, the vote is over and this candidate is elected prime minister.
If he gets a majority of NO votes, then the candidate of the party with the
second-largest number of seats is presented for an approval vote. As long as
candidates continue to not obtain a majority of YES votes, then the candidate
of the next largest party is presented for an approval vote. If no candidate gets a
majority of YES votes, the parliament is dissolved and new elections are called.
This prospect is only preferred over the election of the worst candidate for each
party. Avoid the use of weakly dominated strategies.
(a) Assume that candidates A, B and C have had a majority of NO votes in the
sequential voting. Now is the turn of the candidate from party D. Will she
be elected in this last round or will new parliamentary elections be called?
(b) Anticipating the result in (a), what will happen to Candidate C in the third
round?
(c) Solve also for Candidates B and A in the second and first rounds.
(d) Show the subgame perfect Nash equilibrium of the game.
5.11 The region of Uqbar, in the country of Southasia, enjoys some degree of polit-
ical autonomy. However, recent movements call for statehood within a federal
nation, and even for full independence. After negotiations with the government
of Southasia, they agree to have a vote to determine the status of Uqbar: Auton-
omous Region (A), Federal State (F) or Independent country (I). The people of
Uqbar are divided in three groups, the Autonomy Party (AP), the Federation
Party (FP) and the Independence Party (IP); their preferences are indicated by
the rankings in the table below. (For example, AP first prefers A, then F, and
last, I.)
Party AP FP IP
Number of supporters 35 25 40
Ranking A F I
F A F
I I A
(a) Show the results of sincere voting in each of the following voting mechanisms:
1. Plural vote.
2. Borda count (2 points for the most preferred option, 1 for the second
best and 0 for the last).
3. Double round (the vote goes to a second round only if no option won
more than 50% of the votes in the first).
(b) Show the Nash equilibria for the plural vote mechanism. Assume that all
voters behave like one player, obeying their leader’s instructions. Hint: You
may assume that no one will vote for their least preferred option. You may
also apply iterative elimination of weakly dominated strategies.
(c) Show that sincere voting is a Nash equilibrium for the Borda count. How
many pure strategies does each player have in this mechanism? How many
pure strategy combinations are there in the game?
102
VOTING
(d) Show the subgame perfect Nash equilibria for the double-round mecha-
nism. Hint: First solve the three different subgames in the second round (A
vs. F, A vs. I and F vs. I), then construct the normal form of the game in the
first round where, if no one wins in the first round, the result is the corre-
sponding equilibrium found for the second round. You may apply iterated
elimination of weakly dominated strategies.
5.12 Repeat Exercise 5.11 if preferences are like in the table below.
Party AP FP IP
Number of supporters 35 25 40
Ranking A F I
F I F
I A A
5.13 Consider the case study in Section 5.13, and change the preferences of the piv-
otal voter in the rest of Catalonia to be like the ones displayed in the table
below. Repeat the analysis in that section with the new preferences.
Catalonia except Barcelona Value

Independence all 2
Independence part 1
No independence 0
103
6 NEGOTIATION GAMES
The negotiation – or bargaining – problem is simple. Two or more players must agree
on how to divide a surplus. There are at least two ways to address the question. The first
uses a dynamic game in which negotiators alternate offers and counteroffers until they
reach an agreement. The second makes use of a mediator accepted by all parties with
the understanding that she will make the division according to certain normative crite-
ria that all parties consider reasonable. In this chapter, we will review the first approach,
which can be studied with the game theory tools developed in the previous chapters.
For the second approach, the so-called cooperative game theory has proved useful, but
it remains beyond the scope of the present book. Nevertheless, the advanced literature
finds a correspondence between the two approaches to the negotiation problem, so that
very often the results in one can be translated into results for the other.
The bargaining game is important to economic analysis because on many occa-
sions there is no mechanism that determines the economic exchanges. When there is a
well-established market for a commodity, be it competitive or monopolistic, the market
price determines how much of the surplus generated by the exchange goes to the buyer
and how much to the seller. Other mechanisms, like auctions or public allocations, also
serve to solve the problem. However, there are instances where such a reference does
not exist. Consider a buyer and a seller trying to determine a price for a unique good
for which a market does not exist, in this case it is entirely up to them to establish how
much each will profit from the exchange, and it will be up to the economic or game
theoretical analysis to understand how different variables influence the result. This is
precisely what we will be doing in this chapter. Other examples are
• a firm and a union must share the potential profits of dedicating some extra hours
to a new project;
• several firms have to decide how to divide the profits of a joint venture;
• two countries must divide the resources of a disputed territory; or
• several parties in a coalition must decide how to divide the power once in charge of
the government.
This chapter offers an introduction to bargaining models. We will deal with the
simplified case in which the joint benefit that the players obtain is of equal value
to all of them. This is not the most general case, but it is enough to begin finding
interesting results that are still valid in more complicated and realistic models. The
simplification implies that we can assimilate all bargaining problems to the problem
of dividing a unit of surplus. Bear in mind, as usual, that our main purpose is to
show how game theory is an appropriate tool to analyze the problem. We also want to
familiarize the reader with the basic elements of the bargaining theory, while recog-
nizing that many aspects and developments of the theory are left for more advanced
readings.
104
NEGOTIATION GAMES
Section 6.1 presents the simple but fruitful model of alternating offers in which
we explore the possible advantages of playing first or last. The following sections show
the influence of patience (Section 6.2), risk aversion (Section 6.3) and of having very
strong moral positions that meddle with the bargaining mechanism (Section 6.4). After
reviewing our findings (Section 6.5) we present a real case of negotiating with pirates
(Section 6.6). Finally, we dedicate some time to the Coase theorem, which is very often
invoked to resolve some problems like externalities (pollution, for example) by means of
negotiations. To understand its significance, it is important to know what the theorem
says (Section 6.7) as well as what it does not say (Section 6.8). Many confusions can be
avoided after understanding this.
6.1 The model of offers and counteroffers

The simplest model of negotiation is the ultimatum game that we saw in Section 1.7,
and that will be our building block for presenting the negotiation theory. However,
for reasons that will become clear further along this chapter, we will present a version
in which the offers can take any value, and not just multiples of euros without cents.
Remember that in this game Player 1 offered a way in which to divide the surplus and
then Player 2 decided only whether to accept or to reject it. Figure 6.1 represents this
game in a schematic way.
First, we have simplified all offers as numbers between zero and one. Since there are
infinitely many real numbers between zero and one we cannot draw all the branches
representing every possible offer. Instead, we represent only the end points of those
possible branches and draw only the offers corresponding to the numbers zero, one and
a third non-specified number x. By convention, the offers only indicate the amount of
the surplus that goes to Player 1. Player 2 gets 1 − x . The figure is schematic because
we only specify the continuation of the game for the indefinite offer x, but we must
not forget that Player 2 plays on every point of the horizontal line, corresponding to
the infinite ways of dividing the surplus. In other words, after any of those points there
ought to be two branches, one for accepting the offer in that point and a second for
rejecting it. We draw only one general case, the offer x.
Player 1
0 x 1
Player 2
Reject
Accept
0 x
0 1–x
Figure 6.1 A final offer
To find the subgame perfect equilibrium we must first pay attention to Player 2:
if he accepts the offer x he will earn 1 − x, and if he rejects it, he will get zero. Thus,
105
GAME THEORY
he will accept any offer made by Player 1. Knowing this, Player 1 will offer to keep
everything. When we presented the ultimatum game in its discrete version (with
only integer euros) we found another equilibrium in which Player 1 gives Player 2 a
small amount. Although this may be more realistic, the only difference between the
two equilibria is this small amount, and they can be considered the same equilib-
rium for all practical purposes. In this current case there is no limit to the division
of the surplus and, for this reason, the second equilibrium does not exist. No quan-
tity different from zero that Player 1 gives Player 2 can be part of an equilibrium,
as there is always a smaller quantity to give. If we understand the infinite division
of the surplus as an approximation of the finite division, we can understand the
equilibrium in which Player 1 gives zero to Player 2 as an approximation of the more
realistic situation in which Player 1 gives just a little. For this reason, next we will
assume that when a player is indifferent between accepting and rejecting an offer,
they accept.
More important is the objection that, in the ultimatum game, experiments have
shown that players are more generous than the Nash equilibrium with selfish prefer-
ences states. Nevertheless, we will continue our exposition based on the equilibrium
analysis of the ultimatum game for three reasons: (i) this game is used only to illustrate
and understand the elements of the bargaining game; (ii) as we argued in Section 1.7,
when players understand the context of the game as one closer to a market, their behav-
ior is closer to the theoretical equilibrium – this is what happens in the bargaining
game relative to the ultimatum game and (iii) the conclusions we find starting with the
theoretical equilibrium in the game in Figure 6.1 are similar to the conclusions we find
if we start with a less extreme behavior.
The next step is to add a new round of negotiations. If Player 2 accepts the offer, the
game ends, but if Player 2 rejects the offer, it is then his turn to make a counteroffer,
which Player 1 will accept or reject, thus ending the game. Figure 6.2 shows a sketch
of the new game.
x1
2
R A
2 x1
1–x1
x2
1
A
R
0 x2
0 1–x2
Figure 6.2 An offer and a counteroffer
106
NEGOTIATION GAMES
For convenience, players are labeled using only their number. “A” stands for “Accept”
1 2
and “R” for “Reject.” Offers x and x are still the amounts for Player 1, while the
superscript refers to the round in which the offer is made, as opposed to the player. The
game has two rounds; in the first one, Player 1 makes an offer and Player 2 accepts or
rejects. If Player 2 accepts the offer, then the game ends, but if he rejects it, the game
moves on to a second round. In that case, Player 2 makes an offer that Player 1 can
accept or reject.
We already know the equilibrium in the second round: Player 2 keeps everything.
Knowing this, in the first round this player will reject anything that does not give him
the entire surplus, as he is certain to get everything if the game advances to the last
round. It does not matter how many rounds are added to this game, the player that
makes the last offer has all the negotiation power, as they can always wait until the
last round to obtain everything. Using our notation, we will write the equilibrium as
follows:
• Stage 2: Player 1 accepts if x 2 ≥ 0. Player 2 offers x 2 = 0 . (Recall that this means
he keeps 1 − x 2 = 1.)
• Stage 1: Player 2 accepts if 1 − x 1 = 1, or, equivalently, x 1 = 0 . Player 1 offers any-
thing: if she offers x 1 = 0 , Player 2 accepts. If she offers x 1 < 0, he will reject. In any
case, Player 2 takes everything.
6.2 Impatience
The next step in creating an interesting game is to introduce a new element, impa-
tience. To obtain one euro today is better than one euro tomorrow, but just how much
better will depend on the player. As an example, and to continue our exposition of the
game, we will begin by assuming that both players value one euro in the next round as
much as 80 cents in the current round. This means that, given the choice between 81
cents now and one euro in the next round, a player prefers 81 cents now. If the choice
is 79 cents versus one euro, the player prefers to wait. When the amount is exactly 80
cents, players are indifferent.
Let us see how impatience changes the equilibrium. In the last round the equilib-
rium is the same, where Player 2 takes it all, but in the previous round things change.
Now Player 1 can win something: since for Player 2 one euro in the second and last
round is equivalent to 80 cents in the first round, Player 1 does not need to offer more
than that much for Player 2 to accept. If she offers less, Player 2 will reject, while offer-
ing more is unnecessary. Again, we can discuss if Player 1 should offer a little more
than the 80 cents to make sure Player 2 accepts, but we will proceed as before and
assume that in case of indifference, the player accepts. The factor for converting future
earnings into present ones is called the discount factor. In our example the discount
factor is 0.8.
Let us add one more negotiation round. To take advantage of the discussions before,
we will place it at the beginning rather than at the end. This means that, in the first
round, Player 2 makes an offer, while in the second round it is Player 1’s turn to make
an offer, and in the third round Player 2 makes the final offer. As always, the game
advances to the next round only if the current offer is rejected. In this new game,
the analysis of the last two rounds (2 and 3) is exactly as in the above paragraph.
107
GAME THEORY
Anticipating it, in the first round, Player 2 will propose giving Player 1 the amount she
expects to win if she rejects the offer, forcing the game into the second round, which is
1 − 0.8 = 0.2, but multiplied by the discount factor of 0.8. This means Player 2 offers
0.2 × 0.8 = 0.16 for Player 1 and he keeps 0.84.
We can add more and more rounds and the rule will remain the same: when mak-
ing an offer, give the other player the amount that they expect to win if the game moves
on to the next round, multiplied by the discount factor. With this rule, we can calcu-
late what will happen if there are, for example, twelve rounds. These are the contents
of Table 6.1a. Player 2 is the one making the last offer. We can observe the subgame
perfect Nash equilibrium of the three-rounds game in the paragraph above if we just
read rounds 10, 11 and 12.
In the table we observe how, as the number of rounds increases, the players’ payoffs
get closer to each other, although the player making the last offer continues to enjoy
an advantage. If we added even more rounds we would observe that the advantage
disappears for a sufficiently large number of rounds. In fact, after many rounds the
advantage goes to the player making the first offer, as they can take advantage of the
other player’s impatience. In the limit – as the number of rounds approaches infinite –
it is shown that the equilibrium payoff of the player making the first offer approaches
1 0.8
the amount = 0.556, leaving = 0.444 for the other player. In Exercise
1 + 0.8 1 + 0.8
6.1 the reader can confirm all of this.
Table 6.1a Solving the negotiation game
Round Proposing player Surplus for Player 1 Surplus for Player 2

12 Player 2 0 1
11 Player 1 0.2 0.8 (=1 × 0.8)
10 Player 2 0.16 (= 0.2 × 0.8) 0.84
9 Player 1 0.328 0.672 (= 0.84 + 0.8)
8 Player 2 0.262 (= 0.328 × 0.8) 0.738
7 Player 1 0.41 0.59 (= 0.7.38 × 0.8)
6 Player 2 0.328 (= 0.41 × 0.8) 0.672
5 Player 1 0.462 0.538 (= 0.672 × 0.8)
4 Player 2 0.37 (= 0.462 × 0.8) 0.63
3 Player 1 0.496 0.504 (= 0.63 × 0.8)
2 Player 2 0.397 (= 0.496 × 0.8) 0.603
1 Player 1 0.517 0.483 (= 0.603 × 0.8)
In the next exercise, we make Player 2 more impatient. This means that one euro
tomorrow is valued even less today. Let us suppose that now, for this player, one euro
tomorrow is worth 0.6 euros today. Table 6.1a changes to be Table 6.1b.
Compared to the previous situation, when his discount factor was 0.8, Player 2’s
payoffs are smaller. Given that this is a zero-sum game, this implies that the payoffs
108
NEGOTIATION GAMES
for Player 1 are larger. When the number of rounds goes to infinite and Player 1 is the
1 − 0.6
first to play, Player 1 gets = 0.769 while Player 2 gets 0.231, compared to
1 − 0.8 × 0.6
payoffs 0.556 and 0.444, respectively, in the previous case. If Player 2 plays first, he gets
1 − 0.8
= 0.385, while Player 1 gets 0.615, compared to payoffs 0.444 and 0.556,
1 − 0.8 × 0.6
also respectively, in the previous case. This is a general result: the more impatient the
player, the worse their negotiation power.
All of the analyses we have seen so far are based on the fact that, in the last round,
the player making the offer gets the whole surplus. If we start with a result that closer
resembles the lab experiments, with a more balanced share of the surplus in the last
round, we will arrive at similar conclusions about the payoffs when the number of
rounds get bigger. (See Exercise 6.2.)
Table 6.1b Player 2 is more impatient
Round Proposing player Surplus for Player 1 Surplus for Player 2

12 Player 2 0 1
11 Player 1 0.4 0.6 (= 1 × 0.6)
10 Player 2 0.32 (= 0.4 × 0.8) 0.68
9 Player 1 0.592 0.408 (= 0.68 × 0.6)
8 Player 2 0.474 (= 0.592 × 0.8) 0.526
7 Player 1 0.684 0.316 (= 0.526 × 0.6)
6 Player 2 0.547 (= 0.684 × 0.8) 0.453
5 Player 1 0.728 0.272 (= 0.453 × 0.6)
4 Player 2 0.583 (= 0.728 × 0.8) 0.417
3 Player 1 0.75 0.25 (= 0.417 × 0.6)
2 Player 2 0.6 (= 0.75 × 0.8) 0.4
1 Player 1 0.76 0.24 (= 0.4 × 0.6)
6.3 Risk aversion
Let us return to the game in Figure 6.2 and introduce an element of uncertainty: if
Player 2 rejects the offer, there is a probability that the game is over with no time left
for counteroffers. This possibility will give some advantage to Player 1 compared to the
original game; Player 2 will not be so prone to reject offers, since the game can finish,
leaving him and Player 1 with nothing. It is even more interesting to analyze how the
offers change according to his degree of risk aversion. To isolate this effect, we go back
to the case where there is no impatience on the part of either player. Figure 6.3 sketches
the new game.
If the game continues, Player 2 is the last player making an offer in the second
round and, as we already know, will offer to keep everything for himself with Player 1
accepting: 1 − x 2 = 1, and x 2 = 0 . If the game stops after the first round, both players
1 1
get zero. Therefore, Player 2 expects to win × 1 + × 0 = 0.5 if he rejects the offer in
2 2
the first round. If Player 2 is risk neutral, meaning that he only cares about the expected
109
GAME THEORY
earnings, he will need an offer of 0.5 or more to accept in the first round. Player 1 is
aware of this and will propose the minimum that Player 2 will accept, 0.5.
If Player 2 is risk averse and prefers 0.5 with certainty as opposed to a 50–50 lot-
tery between one and zero, then he will accept an offer smaller than 0.5 in order to
avoid moving on to the second round and risking a premature ending of the game.
If, for instance, Player 2 uses as his utility function the square root (we discussed this
at the end of Section 2.5), the utility of winning zero or one with equal probabilities
1 1
is 1+ 0 = 0.5. Denote by x the certain amount that gives that same utility to
2 2
Player 2. This quantity must satisfy x = 0.5; therefore,= x 0= .52 0.25. This means
that the minimum quantity that Player 2 will accept is 0.25, and that is the offer that
Player 1 will make. If Player 2 were even more risk averse, with the utility given by the
1 1
cubic root, the utility of rejecting would be 3 1 + 3 0 = 0.5 , and the x which gives
2 2
that same utility must satisfy 3 x = 0.5, and then= x 0= .53 0.125 . As we observe, the
more risk averse a player is, the weaker their bargaining position.
x1
2
R
A
Game is over x1
1 1–x1
p= Game
2
continues
1 2
1−p=
0 2
0
x2
1
R A
0 x2
0 1–x 2
Figure 6.3 A probability of interrupting negotiations
6.4 Negotiating with fanatics

A fanatical attitude toward negotiation may be “all or nothing” or “the laws of our
sacred book are not open for negotiations.” For the fanatics, any option that is not their
preferred option is bad. The utility of such options is high in their optimal point and
decreases rapidly after it. Let us see an example of a conservative country that has to
decide whether or not to pass a legislation to legalize divorce. To clarify ideas, think of
110
NEGOTIATION GAMES
the decision to be made following negotiations by the two main groups in the country,
the NO group and the YES group. Zero means a prohibition of divorce in any given
case, while one means a very permissive law, and in-between numbers mean different
degrees of permissibility. Say that the NO group feels very strongly about its cause and
has preferences represented by the utility function uNO ( x ) = (1 − x ) , where x measures
3
the permissibility of the legislation. This function is shown in Figure 6.4.
1.2
0.8
Utility
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
x
Figure 6.4 Fanatics’ preferences
The utility of the YES group is uYES ( x ) = x . Negotiations are conducted as follows:
first, the NO group makes an offer (a value of x), and then the YES group either accepts
(A) or rejects (R). If the offer is accepted, then it becomes the law. If it is rejected, there
will be social turmoil and confrontation that will end with an equal probability of vic-
tory for either of the two groups. In this case, the winner imposes its preferred legisla-
tion. The NO group does not care for social confrontation, while the YES group would
see its utility reduced by 0.2 if it were to happen. Thus, its utility will be uYES ( x ) = x
if there is no confrontation and uYES ( x ) = x − 0.2 if there is. The game is shown in
Figure 6.5.
NO group
YES group
R
A
(1 − x)3
YES wins x
1 NO wins
p=
2 1
1−p=
2
0 1
1 − 0.2 0 − 0.2
Figure 6.5 Negotiating with fanatics
111
GAME THEORY
If the YES group rejects the offer, then it expects a utility given by
1 1
uYES ( reject ) = 0.8 + ( −0.2 ) = 0.3 , which means that the NO group has to
2 2
offer at least x = 0 . 3 if they want their offer to be accepted. But then, if x = 0.3
is the offer of the NO group, it will be accepted and the NO group’s utility will be
uNO ( 0.3 ) = (1 − 0.3 ) = 0.34 , while in case of disagreement, its utility would be
3
1 1
uNO ( no agreement ) = 0 + 1 = 0.5. The NO group would be better off making an
2 2
offer that the YES group wants to reject. The subgame perfect Nash equilibrium is the
NO group offers any x < 0.3, the YES group accepts any offer x ≥ 0.3 and rejects any
offer x < 0.3. The result is no agreement and, therefore, social confrontation.
The same kind of thing will happen if it is the YES group making the first offer
and the NO group either accepting or rejecting it. This case is provided to the reader
as Exercise 6.6.
An experiment
Fieke Harinck and Carsten De Dreu, psychologists and social researchers, conducted
the following experiment regarding negotiations in different legal cases (Harinck and
De Dreu, 2004). Experimental subjects were divided into two groups, with each group
being further divided into pairs in which one person would take the role of a defense law-
yer and the other the role of the prosecutor. In the first group, subjects were told to reach
agreements on purely selfish terms, aware that their careers depended on their gains fol-
lowing those agreements. The second group was told to make decisions on moral terms,
such as rejecting offers that were deemed unfair. The results of the experiments showed
that the selfish group systematically obtained mutually beneficial outcomes more often
than the moral group. They were able to concede some points of the negotiations in
order to obtain a gain that gave them more profits in some of the other ones.
This experiment does not command us to be selfish or immoral; it only shows
that there are situations in which an overly intense prejudice, even a moral one, may
be preventing us from reaching social agreements with people that do not share such
prejudices. It is up to each individual to decide whether this is a good or a bad thing
according to their convictions, regardless of whether they are moral, ideological, reli-
gious or political in nature. The theory and experimental evidence only show what the
consequences of having these convictions are. Knowing these consequences may help
us to check for consistency in our moral preferences.
6.5 Some discussion
We can summarize the results in the above sections with these highlights:
• If the negotiations have few rounds, the player making the last offer has more bar-
gaining power. This is because he will be able to make a “take it or leave it” offer or
to say “I’ll make you an offer that you can’t refuse.”
• If players are impatient, as the number of rounds, and therefore the length of the
negotiations, increase, the bargaining power of the player that makes the last offer
112
NEGOTIATION GAMES
disappears. The player who plays first is in a slightly better position then, since she
can take advantage of the impatience of the other player.
• Being more impatient works against the bargaining power of that player.
• In uncertain negotiations, being more risk averse also works against the negotiation
power of that player.
• When one of the parties is a fanatic, negotiations will easily break off.
Any skilled negotiator knows all of the above points without relying on game theory.
The interest of the theory relies on its capacity to derive all those results based on a few
minimal principles. It seems that a lesson worth learning is that, when negotiating,
one must be patient and avoid being risk averse or reaching high levels of fanaticism.
However, both patience and attitude towards risk-taking are part of the traits of an
individual and, therefore, cannot be easily changed.
There is one fact present in many real-life negotiations that has not shown in our
analysis so far. In all of the examples in the previous sections, players anticipated the
result in future negotiation rounds and, as a consequence, reached an agreement in
the very first round. In the real world, however, we often find that negotiations take
time and that the final agreement is reached at the deadline. There is a way to improve
our models to find equilibria in which players delay the acceptance of the offers. This
is what happens if players do not know the utility function of the other players. For
instance, Player 1 may not know how impatient or risk averse Player 2 is. In this case,
Player 2 will find it in his best interests to pretend to be patient and with low risk
aversion; he can try to do this by rejecting some offers. The games that deal with these
issues are complicated and are beyond the introductory level of this book.
6.6 An actual case: The hijacking of the Alakrana

On October 3, 2009, a group of Somali pirates hijacked the Basque tuna fishing boat
Alakrana. The boat was released after 47 days, following the alleged payment of four
million dollars by the Spanish government.
The best result for the government was to free the hostages, send the pirates to
prison and not pay the ransom; this was, however, materially impossible. For the pur-
poses of this section, we will not argue about the convenience, legality or morality of
paying the ransom in order to save the lives of the hostages, or its repercussion to incite
further hijackings. We assume that using force to resolve the situation is out of the
question and that a payment will be made. Given these restrictions, how should the
government conduct the negotiations?
The families of the hostages understandably wanted a rapid resolution to the hijack-
ing, and the media were eager to publicize their petitions and demands. The govern-
ment could not avoid the payment due to the great and tragic cost of having the deaths
of the crew on its conscience and on the conscience of the citizens. At the same time,
however, it should aim for very cautious negotiations. There are two reasons for that.
First, the party most in a hurry loses leverage. As we saw in the sections above, the most
impatient player ends up accepting worse conditions. Second, even if the government
pays the same amount after 100 days of negotiations as it would have paid after just
one day, the cost is huge for the hijackers. Very fast resolutions imply that pirates can
113
GAME THEORY
hijack a boat every few days. If it takes months to get a payment, then there will be
fewer hijackings, and the business of piracy will be less lucrative.
There was an additional circumstance that complicated the case. When the govern-
ment received the news, it sent the frigate Canarias to aid, but it was unable to resolve
the situation by force. The government did, however, apprehend two of the pirates that
had at one point left the boat. These pirates were taken to Spain to be judged. Was it
a good idea?
Much of the media claimed that this resulted in negotiations being more difficult
and that there was a higher risk of a violent resolution. If the pirates were judged and
sent to prison in Spain, it would be difficult to send them back to Somalia, even under
the condition that they would serve their sentences there. In fact, this difficulty had
good consequences for the Spanish position: they had one additional item to negotiate
and it provided them with credibility that giving up the pirates was expensive and
difficult due to the separation of powers. Negotiations lasted longer to Spain’s benefit
because of this.
6.7 The Coase theorem

A baker and a physician both have their establishments in the same building. The odors,
fumes and noises from the bakery harm the practice of the physician, who has fewer
and unhappier clients. There are not any other tenants in the building. The way in
which the decisions of one economic agent affect the productivity of another one, like
in this example, is called a production externality. Let us put some numbers to it:
• If the baker leaves the building, the physician will earn 100.
• With the baker in the building, the physician earns only 50 if he stays in that build-
ing, but he can earn 80 elsewhere.
• The baker is currently making 100, and she will make 60 if she leaves to another
location.
What should be done? Should the baker be entitled to her activity or should it be pro-
hibited or limited in some way? If the baker is allowed to stay, she can negotiate with
the physician and ask him money in exchange of her leaving, though the most he is
willing to pay for the baker to leave is 100 − 80 = 20 . If he is asked to pay more, he is
better off leaving himself. However, the baker needs 100 − 60 = 40 in order to accept
taking her business somewhere else. This means that if the baker has the right to stay in
the building, there is no room for an agreement; the result will be that the baker stays,
earning 100, while the physician leaves to earn 80.
Let us consider now that the baker’s business is not allowed in the building unless
she has permission from the physician. Will she obtain it? The baker can give him, say,
30 to get his permission to stay. If the physician agrees, he takes the money and then
leaves, since, with the baker staying, he will make more money elsewhere. This way he
will be making 80 + 30 = 110 . The baker will end up with 100 − 30 = 70. If, alterna-
tively, the physician does not accept the baker’s money and instead forces her to leave,
he will earn 100 and the baker will earn 60. He prefers to take the money.
Notice that, regardless of who has the right to decide on whether the baker stays, the
decision about who stays and who leaves is the same: the baker stays and the physician
114
NEGOTIATION GAMES
leaves. Furthermore, observe that the total surplus that is generated is the same. The
joint earnings of both the baker and the physician are the same (180) in the two cases:
100 and 80 if the baker has the right to stay vs. 70 and 110 if the physician has the
right to evict the baker. The only difference is that the one who has more rights also
has more earnings.
If, on the other hand, the law were to prohibit the activity of the baker in the build-
ing, removing the possibility of her staying even if she did negotiate with the physician,
they would earn 60 and 100, respectively. This adds up to a total surplus of only 160,
and a less profitable situation for both the baker and the physician than if they were
allowed to negotiate.
This is in essence the Coase theorem. More generally, it can be stated as follows:
as long as the property rights are well defined and the negotiation costs are negligible,
private negotiations over externalities lead to efficient outcomes. The wealth or income
distribution will depend on how the property rights are distributed.
Ronald Coase studied these questions and received the Nobel Prize in Economics
(Coase, 1937, 1960). He highlighted the importance of the theorem when, in cases not
much unlike our example, he described many court rulings in which judges argued in
favor of either party over the other using social economic efficiency criteria. The theo-
rem proved that efficiency was not at stake. Thanks to this theorem we also know that
imposing norms or regulating via Pigouvian taxes (such as a tax on a polluter) may not
be the only method of solving an externality, which is what economists believed at the
time.
The bargaining game between the baker and the physician when the physician has
the right to prevent the baker from conducting her business in the building, and when
the baker makes a “take-it-or-leave-it” offer to the physician, is displayed in Figure 6.6a.
In the figure, x is the amount that the baker offers the physician in order to be allowed
to stay. The first number is the baker’s payoff, and the second is the physician’s. If the
physician rejects the offer, the baker leaves; if he accepts, the baker stays and will earn
100 − x, while the physician leaves and earns 80 + x.
Baker
0 x 40
Physician
Reject Accept
60 100 − x
100 80 + x
Figure 6.6a Coase theorem: the baker makes the offer
The subgame perfect Nash equilibrium is (i) the physician accepts any offer x such
that 80 + x ≥ 100 , or x ≥ 20, and (ii) the baker offers x = 20. Payoffs are 80 for the
baker and 100 for the physician.
If the baker is more generous, if the physician demands more than just enough to
make the same money if he leaves, or if, as we modeled the game earlier in the chapter,
115
GAME THEORY
the negotiations have more rounds, then the final value of x will be greater than 20,
yet always smaller than 40. Whether it will be closer to 20 or to 40 will depend, as we
saw, on the impatience of the players as well as on the number of rounds. However, for
most analyses the precise value is not as important as knowing that the physician leaves
after some compensation.
If the baker has the right to stay and the physician makes the “take-it-or-leave-it”
offer, the game is like in Figure 6.6b. We can instantly observe that there is no amount
of x that the physician can offer the baker for her to leave. The subgame perfect Nash
equilibrium is (i) the physician offers any quantity between zero and 20 and (ii) the baker
rejects. As in the previous game, the physician abandons the building to get 80 elsewhere,
but this time he receives no compensation for that, and the baker makes 100 by staying.
Physician
0 20
x
Baker
Reject Accept
80 100 − x
100 60 + x
Figure 6.6b Coase theorem: the physician makes the offer
6.8 When not to apply the Coase theorem

In the last section, we specified that the Coase theorem needed costless negotiations
in order for the theorem’s result to hold. When the externality is caused or suffered by
many agents, this condition may not be satisfied. Let us explore this idea.
The difficulties to coordinate

Imagine a lake where thousands of people fish. They will very easily suffer the prob-
lem known as “the tragedy of the commons” and will overfish, depleting the resource.
There are several ways to prevent this from happening. For instance, if one firm is the
owner of the lake’s fishing rights, it will be in its best interests to avoid overfishing. This
is because it will maximize the present discounted value of the flow of future profits
from fishing and, unless it believes there is no tomorrow, this implies keeping a stock of
fish to allow for maximum sustainable catches. Even if the firm grants fishing permis-
sions to others, it will impose conditions on how much they can catch.
If several firms can fish in the lake, it would be in their best interests to reach an
agreement in order to limit the captures and avoid the depletion of the fishery. A nego-
tiation among a few firms is simpler than among thousands of individuals – the costs
of gathering all of them together, finding a satisfactory agreement and imposing and
enforcing penalties upon the non-abiding fishers might make negotiations impossible,
and the Coase theorem will not apply.
Firms may be multinational companies, fisher’s cooperatives or public firms from
the different villages around the lake. From the efficiency perspective, any alternative
116
NEGOTIATION GAMES
will work. From the distributional perspective, each one will result in different profits
for different individuals, depending on how many rights over the resource every indi-
vidual has.
The negotiation costs will also be high if a group of moviegoers are to negotiate
whether or not to allow smokers to smoke inside the theater, perhaps following some
compensation. A lengthy negotiation prior to watching a movie every time one goes to
the theater does not appear to be a good mechanism to solve the externality problem
caused by the smoke. It is much simpler to make the decision that, among the ones that
can be easily implemented, is the most efficient. In this case, this is the norm of prohib-
iting smoking as the default option. The same thing happened in bars, restaurants and
other public spaces. For the same reason, whenever the smoking option is a reasonable
position by default, there should not be a problem having it as the norm. This already
happens in private clubs where smoking is allowed, where people attend knowing that
it is a smoking club and where there are plenty of non-smoking places around as alter-
natives for leisure time.
The lack of information about others’ preferences and costs

Aside from the number of individuals, another circumstance that makes negotiations
costly is the lack of knowledge in regards to others’ preferences and costs. Negotiations
in this case will, in general, be inefficient. Individuals can reach an agreement, but
research shows that it will be delayed, as each one may want to probe the others to
figure out their preferences, which in itself implies an added cost to the negotiation.
Rent seeking
Consider a new version of the problem between the baker and the physician. The prop-
erty rights belong to the physician, so the baker pays the physician to be allowed to
stay in the building, and the physician moves elsewhere. But as soon as he leaves, a
new physician moves into the building, attracted by the legislation in his favor and the
prospect of getting paid by the baker to leave. If the baker anticipates a series of new
physicians coming to the building to demand payments after the previous ones leave,
she will prefer abandoning the building herself. This situation would be inefficient.
We have found another case for inefficiency where the Coase theorem does not
apply. In a static world, there are no considerations for the entrance of new opportunis-
tic players in the game, and this was the realm for the application of the theorem. The
new dynamic aspect introduces the incentives for an inefficient activity of rent seeking.
The truth is that the Coase theorem could be applied if only the baker knew the
identity of all those potential physicians that could install themselves in the building as
soon as the previous one had left. In this case, she could make a deal with all of them,
for instance, by paying 30 to the incumbent physician and then just a small amount
to each of the remaining ones. In exchange, the incumbent would grant the baker per-
mission to operate in the building – and afterwards leaving – while the others would
commit not to bring their practice to the building. They all win: the baker and the first
physician are better off, as we have thoughtfully analyzed. Notice that, if the remaining
physicians do not accept the agreement, the baker, anticipating that they will come to
the building in order to demand payments, will simply leave the building and then the
physicians will obtain nothing. The problem is that, in any realistic scenario, it would
117
GAME THEORY
be almost impossible to identify these other potential entrants, and the conditions of
the Coase theorem would not be satisfied.
The role of the state

All of the difficulties mentioned above are reasons that, very often, private negotiations
are not enough to solve problems with externalities, since the Coase theorem cannot be
invoked. It is sometimes argued that the existence of an element of inefficiency implies
a potential profit if it is corrected and, thus, that there is an incentive for the individu-
als’ economic ingenuity to find a solution without state intervention. The error in this
kind of argument is that, oftentimes, this economic ingenuity is precisely the political
decision taken by a public administration that individuals agree to have.

6.1 Complete Table 6.1a for the case in which there are 20 rounds of negotiations
(with rounds 9 to 20 being identical to rounds 1 to 12 on the table). Keep in
mind that you must proceed backwards, from one round to the previous one.
Observe that the offers get closer to the one computed in the limit case in the
text.
6.2 Repeat Exercise 6.1 for the case in which, in round 20, Player 2 (the player mak-
ing the last proposal) offers 0.3 to Player 1 instead of zero. Compare the offers
in the first rounds for this case and in the previous exercise.
6.3 Consider the game in Figure 6.3. After Player 2 rejects Player 1’s offer, rather
than the game being over with probability ½, consider that the following hap-
pens: with probability ½ Player 2 makes his counteroffer, just like before, but
now with probability ½ it is Player 1 who makes a new offer. The game ends
after the other player accepts or rejects.
(a) What is the subgame perfect Nash equilibrium if Player 2 is risk neutral?
(b) What is the subgame perfect Nash equilibrium if Player 2 is risk averse with
preferences given by the utility function u2 ( x ) = x ?
6.4 A firm and a union must find a way to share the profits of their economic activ-
ity, estimated at two million euros. They establish a three-round negotiation
game. Offers and counteroffers are made sequentially following the order firm-
union-firm. The game goes on as usual: after an offer is made, the other player
can accept or reject it. If the offer is accepted, the game ends and they share
the profits as established in the accepted offer. If the offer is rejected, the game
moves on to the next round. If all offers are rejected, they both get a payoff of
zero.
(a) What is the equilibrium if both parts use the discount factor δ = 1 / 4 to
compare payoffs of consecutive rounds?
(b) What is the equilibrium if the firm uses discount factor δ F = 1 / 4 and the
union uses δU = 1 / 2 ?
(c) Compare your results in (a) and (b).
118
NEGOTIATION GAMES
6.5 (You will need calculus to solve this problem.) A firm and a union negotiate the
salary for the employees. The firm’s profit function is Π F = (10 − w ) , where w
2
is the salary. The objective of the union is to maximize ΠU = 10w − w − 5 . The

2
negotiation schedule is as follows: The firm makes an initial offer, which the union
may accept or reject. If it accepts the offer, then the game ends, but if it rejects it,
then the union makes a counteroffer. If the firm rejects the counteroffer, there is
no production and both get zero. The discount factor is 0.8 for both parties.
(a) Show the extensive form of the game.
(b) What is the salary the union will offer in the second round? What profits
does each one make in this second stage?
(c) Find the only subgame perfect Nash equilibrium of the game. What salary
is agreed on? What are the firm’s and union’s profits?
6.6 Consider the game in Section 6.4, but now the YES group is the one making
the offer, which the NO group may accept or reject. Find the new subgame
perfect Nash equilibrium. Remember that the payoff for the NO group is not
policy x, but the utility derived by this policy.
6.7 Consider the problem between the baker and the physician in Section 6.7, as
well as the case in which the physician has the right to prevent the baker’s activ-
ity in the building. Figure 6.6a presents the game in which the baker can make
a “take-it-or-leave-it” offer to the physician. Let us change that, and now let the
physician be the one making this kind of offer. In other words, the physician will
grant the baker permission to work in the building in exchange for some money.
(a) Draw a sketch of the game.
(b) Find the subgame perfect Nash equilibrium of the game.
(c) Compare this result and the one after the game in Figure 6.6a.
6.8 Consider the game in Figure 6.6b, where the baker has the right to stay and the
physician makes a “take-it-or-leave-it” offer to the baker. Let us change that,
and now let the baker be the one making this kind of offer. In other words, the
baker will ask the physician for money in exchange for her leaving the building.
(c) Compare this result and the one after the game in Figure 6.6b.
6.9 Sylvie and Bruno bargain over €100 that they can share if they reach an agreement.
There will be three rounds of alternating offers, starting with Sylvie. They are not
impatient and do not discount the future, but in order to participate in the second
round they each have to pay €15, and another €15 to participate in the third.
6.10 Consider the Exercise 6.9, but now, instead of having to pay in order to enter the
second and third rounds, the total amount of money shrinks to €70 if the agree-
ment is made in the second round and to €40 if it is made in the third round.
119
7 REPEATED GAMES
In the previous chapters we have stressed that games must be analyzed as if there were
no other interaction among the players. This does not mean that those other inter-
actions are irrelevant or should not matter in the analysis. What it means is that the
analysis must be a constant relation between the game model and reality. If, properly
analyzed, a game does not seem to reflect the real-life situation it is supposed to study,
then the game or the equilibrium concept used in the analysis must be changed, prob-
ably to include interactions that were missed in the first approach. In the past chapters
we have considered many cases that started as a simple game and were later complicated
to add new strategic decisions. One way of adding interesting interactions is consider-
ing the repetition of a game. When players must interact repeatedly, new features can
arise, like the building of a reputation or the developing of mutual trust.
In the present chapter we will focus on maybe the most important topic analyzed
with repeated games, namely the emergence of cooperation. We will use the prison-
ers’ dilemma game as the starting point. After a motivating example (Section 7.1), we
explore the issue in a very simple case (Section 7.2) to move to the more general case
(Section 7.3). After discussing some details in the next three sections we move to the
experimental and empirical evidence on the theory (Sections 7.7 and 7.8). In Sections
7.9 and 7.10, drawing from the findings in the previous sections, we consider some
hypotheses regarding the evolution and the future of humankind. The chapter ends
with a brief exposition of the results of Axelrod’s tournament, in which many kinds
of people were asked to participate in a contest to submit strategies for the repeated
prisoners’ dilemma.
7.1 The Christmas truce

In the course of trench warfare in the Western front during World War I, a remark-
able episode occurred. It was an act of peace, an example of cooperation in war (the
most uncooperative scenario one can imagine). If cooperation can emerge with some
stability in the battlefield among mortal enemies, and if it can be explained rationally
(appealing not to the abstract collective best interest, but rather to the individual’s best
interest), we may be on our way to gaining a better comprehension of the concept and
of the situations in which to expect it. This will not exclude or deny other explanations
that also help understanding cooperation (e.g., altruistic preferences), as it may be com-
plementary to them. Its merit will be to show the existence of explanations that do not
require ad hoc assumptions, which opens a research agenda to explain where those other
assumptions may come from (e.g., why we are altruists).
The episode is known as the Christmas truce. On December 24, 1914, the Ger-
man troops decorated their trenches and started singing carols. The French and British
responded by singing their own Christmas songs and, soon after, they were exchanging
small gifts with their enemies. The situation continued for some days and was mim-
icked in other places and at some other times along the trench lines.
120
REPEATED GAMES
Pope Benedict XV had called for a truce some time before, to no avail. Generals
and officers tried to discourage this fraternization with the enemy. What were the
circumstances then that allowed the emerging of cooperation? Soldiers have a very
different perspective of war than generals do. In trench warfare the forces on either side
were unable to break the stalemate. The battalions have two options: to fight or not to
fight. If both fight, there will be many casualties on both sides with very little chance of
capturing an enemy position. If neither one fights, there will be no casualties and life in
the trenches will be bearable. The problem is that if one fights and the other does not,
the side not fighting gives the enemy a chance to attack and score an easy victory. It is
a prisoners’ dilemma. It is war.
But this is not any prisoners’ dilemma. Soldiers face it day after day, week after
week, with no foreseeable end. A repeated game is very different from a game played
only once, especially if the game is not zero-sum as is the case here. Due to the repeti-
tion, new strategies and equilibria arise.
Consider the following strategy:
We, the soldiers on this side of the trenches, will not shoot and we will continue
to not shoot if you, the soldiers on the other side, do likewise. But as soon as we
hear a shot, we will use our guns also.
If both battalions follow this strategy, the situation will be an equilibrium. It will be in
neither one’s interest to start shooting. The gains which can be obtained with the first
shooting – using the element of surprise, taking a position and causing some casualties –
does not compensate for the unavoidable fighting which will follow.
They had to pretend to still be fighting to avoid catching the attention of their supe-
riors. It is said that the disguise took the form of shooting a cannon or a mortar, but
always at the same time of day, and aiming at the same irrelevant target. It is also said
that if someone shot by mistake, he would jump out of the trench exposing himself to
the enemy’s fire, making it believable that the shot was unintended. Officers and gen-
erals, in order to avoid this fraternization, commanded attacks with no definite tactical
interest and changed the positions of the battalions so that they met with unknown
soldiers on the other side, thereby preventing them from having a relation that could
be viewed as a repeated game.
In the next sections we will develop the model of repeated games in order to show
how cooperation may emerge in a strongly non-cooperative environment.
7.2 A game repeated twice

Think back to the first game in Section 3.1. Two neighboring countries must decide
how to tax the 100 in profits a firm expects to make after installing itself in one of
them. The firm will go to the country with the lowest tax, and if they both tax the
same, it will randomly decide where to go. The countries can impose a tax rate of 0%,
10% or 20%. In addition, the country that attracts the firm will enjoy an additional
benefit of 8 because of the new jobs created. The game was detailed in Figure 3.1a and
is reproduced again for convenience in Figure 7.1. There are two Nash equilibria in
pure strategies in the game: (0%, 0%) and (10%, 10%).
121
GAME THEORY
Suppose now that these two countries will face the same situation currently as well
as the following year, and that they have to make a decision about what tax rate to
apply currently and the following year as well. In other words, they play the game in
Figure 7.1 this year and again next year, with a new potential firm to settle in one of
the two countries.
Country 2
0% 10% 20%
0% 4, 4 8, 0 8, 0
Country 1 10% 0, 8 9, 9 18, 0
20% 0, 8 0, 18 14, 14
Figure 7.1 Fiscal battle
The repeated game can be depicted as a dynamic game in extensive form, but for
our purposes in this chapter we will conduct the analysis without that representation,
although we will still have to find subgames and subgame perfect Nash equilibria.
Additionally, in repeated games, the number of equilibria can be extremely large, so we
will not look for all of them; we will simply show how to find some interesting ones.
Let us begin by observing that, since there are two Nash equilibria in the game, their
four combinations are subgame perfect Nash equilibria in the game twice repeated.
• ((0% in the 1st repetition, 0% in the 2nd), (0% in the 1st repetition, 0% in the 2nd))
As usual, in the description of the equilibria we list first the actions corresponding
to the first player, and then we list the actions corresponding to the second player.
But we must insist on some precisions in these games. Recall that in the dynamic
games we must include actions in all possible contingencies. In this example, the
first time the game is played can end in nine different ways, corresponding to
the nine possible combinations of the tax rates chosen by the countries. Thus, when
the game continues the following year, it may carry nine different histories behind
it, each one defining a different subgame. The second repetition can be played after
the first one has resulted in (0%, 0%), in (0%, 10%) or in any of the other possi-
bilities. In the description of the four equilibria above we were implicitly assuming
that the actions for the second repetition were the same after any history. This needs
not be the case. We can prescribe a way of playing in the second repetition that
depends on how the first one plays out. This will allow us to find more equilibria.
For instance, take the following agreement:
The first year we both impose a 20% tax. The second year we set a 10% tax in the
case that in the first year we both kept the agreement and imposed the 20% tax.
Should anyone break the agreement, we will set a 0% tax the second year.
122
REPEATED GAMES
The strategy prescribes a Nash equilibrium in each of the subgames in the second year.
Let us then see what happens in the first year. The strategy requires that each country
chooses the 20% tax rate, an action that is not part of a Nash equilibrium of the orig-
inal non-repeated game. Still, we will see that no country has a profitable deviation.
To understand this, observe that if both follow the agreement, each one wins 14 the first
year and 9 the second, making a total of 23 each. If one country deviates in the first year
and imposes a 10% tax rate instead of the agreed 20% (the best deviation, as seen in the
table), it will win 18 during the first year, which certainly is better than the 14 if it keeps
the agreement; however, it will win only 4 the following year, making a total of 22,
which is less than the 23 it would earn if it keeps the agreement. It is important to see
that, in case anyone broke the agreement in the first year, in the second year the strategy
in the agreement requires playing (0%, 0%), a Nash equilibrium, and that the country
that deviated the first year will not deviate from this equilibrium the second year.
By adequately combining the two equilibria in the subgames of the second repeti-
tion we can sustain the best outcome (14, 14) during the first one. In the second repe-
tition players must play according to a Nash equilibrium (a requirement for a subgame
Nash equilibrium), and there are two of them: one gives better payoffs than the other.
In the agreement, they decide to play the actions that produce the best outcome the
first year. Then, if they do so, they will award themselves the better of the two Nash
equilibria in the second period. If someone did not follow the agreement in the first
period, they punish themselves with the bad Nash equilibrium in the second one. If the
difference between what one wins by deviating in the first period is less than what it
loses by inducing the bad equilibrium the second year, then the deviation does not pay,
and the agreement is a subgame perfect Nash equilibrium. That is what happened in
our example. The result is that we found a way to support cooperation without invok-
ing any altruistic preferences, just one’s own interest. If the game were to be repeated
three times, cooperation could be maintained during the first two years (see Exercise
7.1). In general, and in this game, with any number of repetitions the cooperation can
be sustained in all but the last one, which must always prescribe a Nash equilibrium of
the original game.
In games with a single Nash equilibrium we cannot use these kinds of strategies in
order to sustain payoffs higher than those in the Nash equilibrium, should such payoffs
exist. Since they cannot combine prizes and punishments in the second repetition, play-
ers know that, regardless of what they do in the first period, the Nash equilibrium will
follow in the second. This implies that if they can deviate and win in the first repetition,
they will do so, since it carries no future consequences. This, in turn, means that they
have to play the Nash equilibrium in the first repetition as well. Nothing will change
with three or more repetitions. Later in the chapter we will discuss whether these results
adequately reflect the reality. For the time being, and for our purposes in this section, it
is enough that we understand the logic of repeated games, since we need it in order to
advance to the more complicated games repeated infinitely or indefinitely many times.
7.3 Cooperation in the infinite and indefinite repetitions

Let us take a typical prisoners’ dilemma game, like the one represented in Figure 7.2. It may
be a game consisting of two fishermen in a common lake, two firms in an oligopoly, two
roommates taking care of shared spaces or any one of the examples we have thus far seen.
123
GAME THEORY
Now consider that the situation leading to the prisoners’ dilemma game is repeated
after some time. The fishermen go fishing every month, the firms decide on their
production plans every year or the roommates clean the shared space every week. It
is important that the number of repetitions is not known beforehand; otherwise, we
would be brought back to the model in the previous section.
2
Cooperate Defeat
Cooperate 5, 5 0, 6
1
Defeat 6, 0 1, 1
Figure 7.2 A typical prisoners’ dilemma
The game in Figure 7.2 will be the baseline or stage game for our analysis of the
infinite repetition. Later, we will show that some ways of modeling the indefinite repe-
tition are equivalent to the infinite repetition, thus justifying this seemingly unrealistic
model. An infinite game is too large to illustrate in a graph, even in a schematic one.
Instead, we will make use of the regularities of the repetition to make some simplifi-
cations. As in the case of finite repetitions, the number of equilibria will be too large;
therefore, we will just describe some interesting ones in order to arrive at some conclu-
sions after these games. For convenience we will sometimes change the way in which
we write the equilibria (as well as the strategies in general). In addition to the standard
way of listing the actions separated by players, when useful, we will separate the actions
by the different stages in the repetition of the game.
The most important detail we need to understand in a repeated game is how to
find the subgames. In the repeated games we have subgames at the beginning of each
repetition. At the beginning of the first repetition (the first time the game is played),
the subgame is the whole game. At the beginning of the second repetition, there are
four subgames, which correspond to the four possibilities of play in the first repeti-
tion: (Cooperate, Cooperate), (Cooperate, Defeat), (Defeat, Cooperate) and (Defeat,
Defeat). At the beginning of the third repetition we have 16 different subgames, since
rounds one and two contain 4 × 4 possible histories. For each of the four histories after
the first round there are another possible four in the second repetition. In each repeti-
tion, the number of subgames is multiplied by four.
Finally, it is also important to note that, regardless of what repetition we are in, the
subgame is exactly the same as the whole game. Players have an infinite repetition of
the prisoners’ dilemma game before them. This is something that did not happen in the
finite repetition, where, as the game is repeated, the number of repetitions that follow
is smaller, which means that subgames are different. Remember that, even if subgames
are all equal, the strategy may be indicating to do different things in each of them.
For example, the strategy may indicate to play one way after one particular history but
some other way after some other history.
Now we are in a position of finding some equilibria. One first subgame perfect
Nash equilibrium is the unconditional repetition of the Nash equilibrium in each stage
(Defeat, Defeat). By unconditional we mean that the static Nash equilibrium is played
in each repetition, regardless of the past history. In other words, in each subgame of all
124
REPEATED GAMES
future repetitions of the game the strategy is (Defeat, Defeat) in all stages. No player
wants to deviate to Cooperate in any period, since it would imply winning 0 as opposed
to 1 in every period in which the deviation takes place and no consequences for the
rest of the periods, where the payoff would still be 1. This is true in every subgame, so
the strategy meets the definition in order to be a subgame perfect Nash equilibrium.
A more interesting equilibrium is explained next. With a strategy similar to the one
described for the Christmas truce, we can sustain (Cooperate, Cooperate) in an equi-
librium. We will refer to it as a “trigger strategy.”
The strategy for each player is
• cooperate in the first period;
• cooperate in period T if in all previous periods players played (Cooperate, Cooper-
ate); and
• defeat in period T if in any previous period something besides (Cooperate, Cooper-
ate) was played.
The first thing to check is that the trigger strategy is indeed a strategy, meaning that
it specifies what to do in every contingency; namely, it tells the player what to do in
each period in any subgame for any possible past history. Notice that the trigger strategy
divides the non-initial periods in two: on the one hand are the subgames that are reached
after a history of full cooperation, and on the other hand are the rest. This is an exhaustive
division of the subgames, and in all of them the strategy defines a course of action to take.
In order to confirm that each player playing the trigger strategy is indeed a subgame
perfect Nash equilibrium we need to check three things:
1. The strategy is a Nash equilibrium of the whole game.
2. The strategy is a Nash equilibrium in the subgames after full past cooperation.
3. The strategy is a Nash equilibrium in the subgames after a history with some defeat.
Had the strategy divided subgames in a different way in order to define actions, our
plan to check for subgame perfection would have had to use the same division.
1. The trigger strategy is a Nash equilibrium of the whole game.
First, we will calculate how much a player wins if both use the trigger strategy. Then
we calculate how much one particular player wins if she deviates to some other strategy
while her rival keeps using the trigger strategy. If no deviation gives the player a higher
payoff, then the trigger strategy will, in effect, be a Nash equilibrium of the whole game.
The first part is simple: each player obtains a payoff of 5 in each period if they
follow the strategy. Computing payoffs after deviations is more difficult, since there
are many of them. A player can deviate from the trigger strategy at any subset of
the set of all subgames. Our plan will be to find the best deviation and compute the
payoff after it.
We will begin by noting what happens after a deviation which involves not abid-
ing by the trigger strategy for just one period. In this period the payoff for the
deviator is 6, which is more than the 5 she would win if she did not deviate. Accord-
ing to the trigger strategy, after that deviation, both players have to play “Defeat”
continuously afterwards. Thus, if the deviating player ceases to deviate, she has to
defeat – just as the rival must – and she will get a payoff of 1 in all future periods.
125
GAME THEORY
If the player is considering deviating in more than one period, the first period will
grant 6 as opposed to 5. Then she will win 1 in the following periods up until the
next deviation, in which she will win 0 rather than 1, with no future gains. Thus,
deviating in more than one period is worse than deviating in just one period. There-
fore, we need only consider one-period deviations.
To summarize, the trigger strategy grants 5 on all periods, while the best devi-
ation gives 6 in one period and then 1 in all of the periods afterwards. Only very
impatient players will prefer to win one extra unit today at the cost of losing four
units tomorrow and every day after that. We then conclude that for all players, save
for the extremely impatient, the trigger strategy is a Nash equilibrium.
2. The trigger strategy is a Nash equilibrium in subgames after full cooperation.
As we have seen before, all subgames are exactly the same as the whole game. In
addition, the trigger strategy prescribes to keep playing (Cooperate, Cooperate) in
subgames following full cooperation up until someone deviates, in which case the
strategy indicates to play (Defeat, Defeat) continuously. However, this is precisely
what the trigger strategy recommends for the whole game. Therefore, the argument
which showed that the trigger strategy is a Nash equilibrium for the whole game is
the same that shows it is also a Nash equilibrium in these subgames.
3. The trigger strategy is a Nash equilibrium in subgames after some defeat.
Again, these subgames are also the same infinitely repeated prisoners’ dilemma
game. Now the trigger strategy prescribes to play (Defeat, Defeat) in all periods
unconditionally. We argued earlier that this is indeed a Nash equilibrium of the
whole game, and therefore of these subgames. Actually, we found that it is some-
thing even stronger, a subgame perfect Nash equilibrium.
Once the three claims have been established, we conclude that the trigger strat-
egy is a subgame perfect Nash equilibrium of the prisoners’ dilemma repeated
infinitely many times. By doing that, we have also proved how cooperation can
emerge in a very uncooperative environment without appealing to altruistic moti-
vations. We only need there to be individual potential gains for the players and
a strategy that supports the cooperation by defining prizes and punishments
adequately.
7.4 Some technical details

We have been a little sloppy when calculating the payoffs in the repeated games. For
instance, we have said that using the trigger strategy gives a payoff of 5 in each period
to both players, also that the best unilateral deviation gives 6 in the first period and
then 1 in every period afterwards. To be more rigorous we should be able to add up all
the quantities in all periods. The sum of infinite many 5s is an infinite number,1 the
sum of infinitely many 1s is also infinite; however, having 5s in all periods is clearly
better than having 1s, so we must find an alternative way of comparing these two
clearly different payoffs. We will use a discount factor as we showed in Section 6.2 to
1 Actually, saying this is misleading. What really happens is that the infinite sum of 5s is a non-defined
operation, a meaningless concept.
126
REPEATED GAMES
represent time preferences that give more value to present payoffs. Thus, if the discount
factor is δ, Player 1’s utility when both players follow the trigger strategy is
u1 ( trigger strategy, trigger strategy ) = 5 + 5δ + 5δ 2 + 5δ 3 + …

= 5 (1 + δ 2 + δ 3 + …) .
Reviewing our mathematics for sums of convergent geometrical series we have
1
1+δ2 +δ3 +…= ,
1−δ
and after that
1 .
u1 ( trigger strategy, trigger strategy ) = 5 ×
1−δ
On the other hand, the utility of deviating while Player 2 keeps using the trigger strat-
egy is
δ .
u1 ( deviation, trigger strategy ) = 6 + δ + δ 2 + δ 3 + … = 6 +
1−δ
1 δ 1
Now, we can calculate that the deviation is profitable if 5 × ≥6+ or δ ≥ .
1−δ 1−δ 5
This means that as long as one euro next period is worth more than 20 cents today,
Player 1 will not want to deviate from the trigger strategy.
Throughout this chapter, we have insisted that infinite and indefinite repetitions
of the game were similar cases. Now it is time to make this statement clear. One of
the simplest ways of specifying an indefinite repetition of the game is to consider
that, after every time that the game is played, there is a probability p that the game
will be repeated again and, consequently, a probability 1 – p that the game will end.
This means that the game is played a second time with probability p, a third time
with probability p2 and so on. With there being a probability in each period that
the game will end, the probability that the game will never end is zero, which is a
more realistic scenario than the infinite repetition. If we now want to compute the
expected payoffs of this indefinitely repeated game with no discount, we have the
following:
1 ,
u1 ( trigger strategy, trigger strategy ) = 5 + 5 p + 5p 2 + 5 p 3 + … = 5 ×
1− p
p .
u1 ( deviation, trigger strategy ) = 6 + p + p 2 + p 3 + … = 6 +
1− p
We observe that these expressions are entirely similar to the ones obtained in
the infinitely repeated game with a discount rate, where the probability p plays the
127
GAME THEORY
mathematical role of the discount δ. Of course, we can have both a probability that the
game continues as well as the discount rate:
1 ,
u1 ( trigger strategy, trigger strategy ) = 5 + 5 pδ + 5 ( pδ ) + 5 ( pδ ) + … = 5 ×
2 3
1 − pδ
pδ .
u1 ( deviation, trigger strategy ) = 6 + pδ + ( pδ ) + ( pδ ) + … = 6 +
2 3
1 − pδ
In this case, both players playing the trigger strategy constitutes a subgame perfect
1
Nash equilibrium if pδ ≥ .
5
7.5 Other strategies in the repeated game

Different payoffs with other trigger strategies
Using the same structure as in the trigger strategy, we can sustain many other outcomes
as subgame perfect Nash equilibria. The next is such an example, which treats Player
1 better than Player 2:
• Period T = 1: Players start playing (Cooperate, Cooperate).
• In period T they play (Defeat, Cooperate) if T is an even number and (Cooperate,
Cooperate) if T is odd (i.e., they alternate between these two strategy profiles) and
keep alternating this way as long as they have alternated in the past.
• In period T they play (Defeat, Defeat) if in the past at least one player did not play
as indicated above.
By following this strategy, Player 1 receives 5 in odd periods and 6 in even periods,
while Player 2 receives 5 and 0 in those same periods. If Player 2 deviates in an even
period he will win 1 rather than 0, but from here on after he will get 1 in all periods.
The deviation will not pay if the player shows some patience. A similar thing happens
if he deviates in an odd period (Exercise 7.3).
Many other strategies, with different payoffs for the players, are also equilibria. In
fact, any combination of payoffs that gives players more in average than the payoffs
after (Defeat, Defeat), as long as they are feasible, can be achieved in a suitable trigger
strategy as a subgame perfect Nash equilibrium. For that purpose, define as a “normal”
way of playing one which gives the desired payoffs, and keep playing in the normal way
as long as they have played like that in the past. If they did not, go to (Defeat, Defeat)
from that moment on.
Less strict punishments

Trigger strategies have two drawbacks. First, they punish both the deviator and the
complier, and second, they punish them too harshly. Let us begin with this second
problem. Once there has been a deviation from the cooperative behavior, instead of
punishing with defeating forever, why not, at some point, let bygones be bygones and
try cooperation again? But for cooperation to be credible, the punishment needs to be
128
REPEATED GAMES
credible as well. If the players anticipate that the punishment will be forgotten, then the
cooperating strategy may not be credibly supported. There are several ways to solve this
problem. The easiest one is to make the punishment phase last for just a few periods,
enough for the deviations to not be profitable, and then make the play return to the
original strategy, with its cooperation and its corresponding punishments after new
deviations.
There are also ways of improving the trigger strategy while not punishing the player
that cooperates. If Player 2 deviates, let the strategy include a punishment phase in
which players play (Defeat, Cooperate) for a few periods. Note that this play has pay-
offs (6, 0) that benefit Player 1 and punish Player 2. If Player 2 does not accept his
punishment and deviates in the punishment phase, the punishment phase starts all
over again. Deviations by Player 1 are punished similarly. It is a little hard to define
this strategy formally, but it is possible to do it and also to prove that it constitutes a
subgame perfect Nash equilibrium.
Other versions of the repeated game

A small village has a population of 100. In their daily life, any two of the inhabitants may
have an interaction that is a kind of prisoners’ dilemma game. The interaction occurs
from time to time and involves different neighbors. Being a small town, everyone knows
what everyone else did during those interactions in the past. In this scenario one can also
prove that there are multiple equilibria, including versions of the ones seen in Section 7.3.
For instance, “always defeat” by everyone in every interaction is a subgame perfect Nash
equilibrium: no one will win anything by deviating to cooperate. The strategies that
sustain cooperation are more interesting. For this, it is necessary that the strategy keeps
track of who is a complier and who is a defeater. We can try something like this:
When two neighbors get involved in a prisoners’ dilemma situation, they

cooperate if both have cooperated in the past in their respective interactions,
and they defeat if at least one of them defeated at some point in the past
when they should have cooperated (i.e., outside a punishment phase). If it is
the first interaction for one of them, he or she is regarded as a cooperator.
It is as if each inhabitant were labeled in the eyes of the rest. Some are labeled as “coop-
erators” and some as “defeaters.” An individual is born with the “cooperator” label and
retains the label as long as she cooperates with cooperators and defeats against defeaters.
The first time she does not follow the above rule, her label changes to “defeater.” If the
probability that an individual is involved in a prisoners’ dilemma game is high enough,
and if the individual is also patient enough, it will be in her best interests to maintain
her reputation. This way of playing shares the same logic as the trigger strategy; we
need only adapt the discount factor to the expected time lapses between interactions.
As was the case with the trigger strategies, we can calibrate the strategy to have
less drastic punishment phases. For instance, a neighbor with the “defeater” label who
cooperates for several times when facing neighbors with a “cooperator” label may regain
the “cooperator” label. After vindicating himself this way he can recover the profits
from future cooperation. The number of times that he has to expiate his sins in order
to get back the “cooperator” label must be large enough to deter the defeating behavior.
129
GAME THEORY
7.6 The cooperation in the prisoners’ dilemma repeated

finitely many times
At the end of Section 7.2 we said that the only subgame perfect Nash equilibrium in
a finitely repeated prisoners’ dilemma was (Defeat, Defeat) in each round. When the
number of repetitions is large it seems difficult to accept that rational, selfish individu-
als would not see a way of cooperating during the first periods even though they know
that as the game advances and approaches the end they will stop cooperating. Game
theory has detected several ways in which something like this can actually happen.
Of course, some of the assumptions need to be relaxed.
Say that the prisoners’ dilemma is repeated 200 times, but Player 2 mistakenly
believes that the repetition is infinite (or indefinite) while Player 1 has a correct percep-
tion of the game and is also aware of Player 2’s mistake. Assume now that Player 2 plays
the trigger strategy. What is Player 1’s best reply? Since Player 2 will cooperate as long
as the past is a history of (Cooperate, Cooperate) actions in each period, Player 1’s best
reply will be to cooperate in all rounds except the last one. This way the cooperation is
sustained in 199 periods out of 200. Player 2 is not playing his best reply against Player 1,
as he should defeat in the last period, but he does not know this and may believe that
Player 1 is also playing the trigger strategy. In any case, Player 2 is playing a strategy
that is almost a best reply against Player 1’s actions. The formal model of games that
include a player’s beliefs about the rival is complicated; however, it can be demonstrated
that cooperation can indeed be achieved in most periods in an equilibrium following
our intuitive argument.
Player 2’s lack of rationality can be even weaker. In fact, we can have a perfectly
rational Player 2 that knows his game, but with the assumption of his irrationality
being only inside Player 1’s mind. In other words, Player 2 knows the game lasts 200
periods but Player 1 thinks that Player 2 believes the game to be infinite. In order not
to complicate the game much further, let us stop here and assume that Player 2 knows
that Player 1 is mistaken about his beliefs. If, before the play, Player 2 could talk Player
1 out of her error, they would not cooperate. If, on the other hand, he does nothing to
convince Player 1 that he knows the game, they both will cooperate for 199 periods and
will defeat in period 200.
There are many more ways of achieving cooperation with minimal changes in the
assumptions. Even if the probability that one player is irrational is very small, or, alterna-
tively, the probability that a player assigns to the rival being irrational is small, cooperation
can still be sustained. The irrationality may result from the belief that the game is infinite
when it in fact is finite, or from being a naïve player that always cooperates, or from a
player losing count of the number of repetitions already played, among other possibilities.
7.7 What experiments say

Along with the ultimatum game discussed in Section 1.7, the prisoners’ dilemma is
one of the most popular in experimental economics for studying the actual behavior
of individuals in economic interactions. Among the situations that result in a pris-
oners’ dilemma game, one of the most extensively employed is the voluntary con-
tribution towards the provision of a public good. In 1997, John Ledyard and, later
in 2015, Colin Camerer and Teck-Hua Ho published two articles in which they
130
REPEATED GAMES
summarized the experimental literature on these games until that moment (Led-
yard, 1997; Camerer and Ho, 2015). Next, we summarize their conclusions from the
surveys.
Ledyard presents six main conjectures and beliefs after writing his survey:
1. Subjects contribute more than the rational, selfish hypothesis predicts.
2. Still, selfishness plays a role in explaining the results, along with altruism, decisions
costs and fairness, among other possibilities. Experiments should focus on separat-
ing the different forces that contribute to subjects’ behavior.
3. Altruism cannot explain the data either. Self-interest appears to account for up to
50% of subjects’ behavior, and altruism is reduced to around 30–40% after subjects
get a better understanding of the experiment.
4. As the design of the experiment includes heterogeneity, anonymity (subjects know
their behavior is not observed by others, including the experimenter), experience,
low marginal profits from contributing and detailed information about all this,
self-interest motivates 90% of the subjects.
5. As the design of the experiment includes homogeneity, face-to-face discussions, no
experience, high marginal profits from contributing and little information, almost
all subjects contribute towards the group interest.
6. There seems to be three types of subjects: (i) 50% of Nash players play very much in
accordance with what the theory predicts with rational and selfish preferences, (ii)
40% of players follow the patterns in points 4 and 5 above and (iii) 10% of players
behave in ways that cannot be explained.
Camerer and Ho illustrate their findings with the subjects’ behavior in the
following game:
Four players are given 20 tokens each, which they may keep or use to contribute to a
public pool. The public pool is multiplied by a number M and then the total is divided
equally by the players. For instance, if players contribute (10, 5, 5, 20) and M = 2, the
privately kept tokens are (10, 15, 15, 0) and the public pool is 10 + 5 + 5 + 20 = 40,
that becomes 80, from which each player gets 20. Then the final number of tokens for
players is (30, 35, 35, 20). In this game, keeping all tokens is a dominant strategy and,
therefore, a Nash equilibrium. However, the total number of tokens is maximized if all
players contribute 20 to the common pool. This way there will be 80 tokens in the pool
that, after doubling, will give 40 to each player.
Camerer and Ho’s conclusions with these kinds of experiments are as follows:
1. Experiments typically start with two types of players, one contributing 100% and
the other contributing 0%. The two types seem to be equally abundant, so that
average contributions are 50%.
2. After a time, when subjects learn from experience, contribution erodes to 10% after
10 repetitions of the game. Contributions are higher if partners do not change
during the experiment and lower if they are strangers chosen randomly in every
repetition. If the whole experiment is conducted again, subjects go back to their
initial behavior in point 1.
131
GAME THEORY
3. There seems to be two types of players: (i) one group who are selfish and (ii) another
group who are “conditionally cooperative” (i.e., they start by being cooperative and
then choose to contribute only if the others contribute as well).
7.8 What the empirical data say

In social and economic interactions there are plenty of examples of situations with
the structure of a prisoners’ dilemma. In Section 1.5 we mentioned a few. The
instability of cartels and the legislation against them confirm the existence of an
underlying prisoners’ dilemma, with potential gains if firms reach an implicit or
explicit cooperation agreement, and with punishments in case someone breaks the
agreement. In the real world, strategies cannot be monitored as precisely as we did
in our models. They may not be readily observed and, even if the decrease in the
price can be shown as proof that some firm has defeated to justify the punishment
phase, there is always the possibility that the cause for the decrease in price was
exogenous to the actions of the members in the cartel. This is reflected in the efforts
by the OPEC to keep the price of oil high by disciplining not only its members but
other countries as well. These cartel practices among firms are not permitted in
most developed countries, but there is no international authority that can prohibit
this behavior among countries. International negotiations to reduce the emission
of greenhouse gases and of polluting particles show how the sharing of costs and
gains for cooperating is the major difficulty to reach an agreement, more so than
the consensus about its need.
More recently, the work by Nobel Prize-winning economist Elinor Ostrom has
gained some notoriety. She studied the tragedy of the commons such as the depletion
of a fishery in a lake due to overfishing, which constitutes another example of a prison-
ers’ dilemma game. Her work shows how, in many societies, populations which share a
common resource (a forest, a lake, a pasture, etc.) are often capable of regulating its use
in order to avoid overexploitation and to achieve sustainable use. The stability of the
relations among the populations is important to activate strategies that achieve cooper-
ation. In her celebrated book, Governing the Commons: The Evolution of Institutions for
Collective Action, Ostrom summarizes the eight keys for cooperation to be successful in
these situations (Ostrom, 1990):
1. Clearly identify the boundaries about who belongs in the community and about the
physical limit of the resource.
2. Show congruency between the rules for appropriation and provision as well as the
local conditions.
3. Introduce collective decision mechanisms. Affected individuals should be allowed
to participate in the modification of the rules.
4. Whoever is in charge of monitoring the compliancy of the norms must answer to
the users.
5. Penalties against wrongdoers must be proportional to the infraction.
6. Rapid availability of conflict-solving mechanisms.
132
REPEATED GAMES
7. Governmental authorities must recognize the struggle of communities to

self-organize.
8. Economic activity should be organized around nested firms, where the resource is
part of a bigger system.
7.9 Altruism, reciprocity and evolution

In Chapter 1, when we introduced the prisoners’ dilemma, we also explored different
ways out: signing contracts, punishments by a third party or altruism towards the part-
ner. With the repetition of the game, we can add a new one: reciprocity.
According to biology, there is a clear reason for showing altruist feelings towards
our close family: the genes we share with them. We are predisposed by our genetic code
to cooperate with our close relatives because in this way our gene pool is more likely to
survive in the population. The greater the cooperation, the higher the survival and pro-
creation rates – and it is through this propagation of the genes that evolutionary success
is measured. No teleology is involved; it just happens that genes that favor this behavior
are more likely to be found in the gene pool of the next generations. The genes induce
cooperative behavior within the family through somewhat refined feelings, depending
on the species. No doubt, however, that they include moral sentiments and strong emo-
tions for those species that have already developed complex enough nervous systems.
In some species, relatives are identified via a chemical signal, as is the case with ants,
while in some others it may depend on other factors. The females among mammals can
recognize their offspring easily. For males, however, the problem is more complex. If
they belong to a monogamous or a polygamous species in which relations are stable, the
male will accept the offspring of the females he mates with as his own. In other species
with less stable relations, being unsure of whether the offspring is his, the male can shy
away from his descendants.
Genes are not observable, since they are hidden in the DNA. The individuals can
only observe circumstances that are very highly correlated with the fact that genes are
indeed shared. For example, one individual will be easily inclined to cooperate with
those other individuals that are raised by her mother, under the very likely assumption
that they are her genetic siblings. All these explanations are very general, and the shar-
ing of genes is just one force among many that influence behavior. In fact, there are
species where siblings fight in the nest, each one competing against the others to be the
most fed by the parents, when resources are scarce.
Up to this point, we have a coherent way of explaining cooperation among relatives
via altruism (through genetics) and a second way of explaining cooperation among
strangers via reciprocity (through game theory). There are many details to fill in,
among them, the explanation as to why there are moral feelings towards non-relatives.
Some biologists and psychologists propose that the initial feelings towards family find a
way of becoming extended across wider circles, depending on the social circumstances
that characterize the species and on the cultural evolution that may follow. The altru-
ist emotion may begin linked to close family members, but once it exists and does
not rely on the direct observation of genes, it is malleable enough so as to be adapted
towards friends or the clan. The fact that cooperation via reciprocity gives good results
133
GAME THEORY
clears the room for a nervous system that makes the individual feel satisfaction when
reciprocating and when punishing individuals that do not show this trait. This way
reciprocity is reinforced in evolution. We can very well just refer to this satisfaction as
altruism. These ideas are explained in the books The Extended Phenotype, by Richard
Dawkins, and Evolution and the Theory of Games, by John Maynard Smith (Dawkins,
1982; Maynard Smith, 1982).
For the purposes of this book, defending one hypothesis in biology over another
does not matter. That is something for biologists to settle upon through the scientific
method. At this point we are interested in showing how some of these hypotheses can
be formulated and understood within the models of game theory and how this helps
with analyzing their internal consistency.
7.10 Not a zero-sum game

In zero-sum games, cooperation is not possible; what one player wins is the loss of the
rest. In these games, nothing is created because of cooperation, and nothing can be
gained from cooperating. In all other games, different strategy choices result in dif-
ferent gains. Sometimes the desirable outcomes are achieved in an equilibrium of the
baseline game, while some other times, like in the prisoners’ dilemma, we need some-
thing more elaborate, as we have seen in this chapter. In non-zero-sum games, trigger
strategies allow the emergence of cooperation in very general circumstances; they are,
however, relatively complex, not just because of their formulation but also due to the
need of registering and memorizing the history of who cooperated with whom and who
did not, as well as remembering who enforced a punishment and who did not. In order
to achieve the benefits of cooperation, individuals need a complex enough cognitive
system. In other words, individuals that develop a complex cognitive system to be used
in that manner have a great deal of profits to make.
Opposing the view of evolution as a totally unpredictable process, some authors cling
to the idea that the facts mentioned above are enough to anticipate at least one progres-
sion: there will be more species with strategies that allow them to better capture the
potential gains from cooperating, and thus more complex cognitive and social systems.
This does not mean that all species will evolve this way, nor that the evolutionary process
is predetermined to lead to something like human intelligence. It just claims that some
species will fill the niche of adaptation through cooperation, and this will require more
complexity. Robert Wright, a journalist and science writer, takes this idea to the extreme
in his controversial book Nonzero: The Logic of Human Destiny. He extends this pro-
gression towards human civilization and presents an optimistic view of the future that,
through necessity, will benefit increasingly from the gains of cooperation (Wright, 1999).
7.11 Axelrod’s tournament
In 1980, Robert Axelrod, a professor of political science at the University of Michigan,
conducted an experiment on cooperation that is already a classic and which he describes
and discusses in The Evolution of Cooperation (Axelrod, 2006). He programmed sev-
eral tournaments in order to play a repeated prisoners’ dilemma with the following
characteristics:
134
REPEATED GAMES
• Instead of having to make decisions at every round, players must send a pre-defined
strategy in the form of a computer program.
• Axelrod announced that he would include the strategy “random,” which selects
cooperate or defeat with equal probabilities in each period.
• Players would be colleagues from various departments of economics, mathematics,
political sciences, psychology and computer science. They would be invited to sub-
mit their strategies.
• Strategies would be paired for 200 repetitions several times.
• After the tournament, he ranked strategies according to accumulated payoffs.
• After this first tournament, he published the results and asked to submit new strat-
egies for a second tournament.
• There was a third tournament played in an evolutionary fashion. In each round the
strategies that were more successful in the first pairings entered in higher numbers in
the following pairings, while the less successful entered in lower numbers. This way,
some strategies went extinct and others became more abundant.
From the game theoretical perspective, there are two salient characteristics. First,
because the decision about which strategy to submit is made simultaneously, without
the knowledge of other strategies, the appropriate equilibrium concept is the Nash
equilibrium. Second, from the perspective of the submitter, the problem is not how
to play against another player but how to play in an environment with a variety of
strategies. Each submitter has to devise a strategy that is a best reply against what they
anticipate is the pool of strategies.
A great variety of strategies were submitted, both simple and complicated. Some
were kind, trying to promote cooperation. Others were predators, designed to take
advantage from cooperators. Among all of them, the winner was a strategy called Tit
For Tat, which starts by cooperating and then does what the rival has done in the
previous round. Even after this result following the first tournament was published,
it won the next ones also. No one could design a better strategy. In the evolutionary
tournament, some predatory strategies propagated during the first generations, but
then became extinct because of the lack of weak strategies on which to prey upon and
because they did not call for cooperation. Tit For Tat beat them all and survived all the
others. Its weapons are the following:
• Not being too competitive: in fact, Tit For Tat cannot win any other strategy in
pairwise confrontations, yet no other strategy can take advantage from it for more
than one period. Any strategy willing to beat it in another round must allow Tit For
Tat to win first. Its winning position comes from achieving cooperation more often
than the other strategies when they play among them.
• Being friendly: it starts by cooperating.
• Being forgiving: when the rival comes back to the cooperating mode, it initiates
another cycle of cooperation.
• Responsiveness and moderation: it strikes back immediately, but the punishment
lasts just one period.
135
GAME THEORY
• Clarity: the strategy is very simple and easy to identify. If the rival knows who they
are playing against, they can adapt more easily.
In the finite repetition of the prisoners’ dilemma, the only Nash equilibrium is always
defeat. Therefore, the use of the strategy Tit For Tat for both players cannot be a Nash
equilibrium, yet it is so in the infinite repetition (Exercise 7.9) and in scenarios of lim-
ited rationality, where it is possible to sustain cooperation for many periods.

7.1 Find a subgame perfect equilibrium that sustains payoffs (14, 14) during the
first two years in the game in Figure 7.1 repeated three times.
7.2 Find the subgame perfect Nash equilibria in the prisoners’ dilemma repeated
twice. Is it possible to achieve cooperation in the first period?
7.3 Find the payoff for Player 2 if he deviates in an odd period with respect to the
strategies described at the beginning of Section 7.5.
7.4 Consider the game in Figure 7.1, but instead of playing simultaneously, coun-
tries decide sequentially. First, Country 1 chooses its strategy and then, after
observing Country 1’s choice, Country 2 decides. This dynamic game is
repeated twice. Show that the strategy like the one described in Section 7.2,
which allowed the result (14, 14) during the first repetition in a subgame perfect
Nash equilibrium, does not work for achieving cooperation in the sequential
version. (You need to adapt the strategy to the sequential version.)
7.5 Consider the prisoners’ dilemma game in Figure 7.2, but instead of playing
simultaneously, players decide sequentially. First, Player 1 chooses her strategy
and then, after observing Player 1’s choice, Player 2 decides. This dynamic game
is repeated infinitely many times. Show that a strategy like the one described in
Section 7.3, which allowed the result (5, 5) in each period as a subgame perfect
Nash equilibrium, still works in the sequential version. (You need to adapt the
strategy to the sequential version.)
7.6 Consider the chicken game detailed in the figure below repeated three times.
2
Swerve Keep going
Swerve 2, 2 1, 3
1
Also consider the following strategy, where S stands for Swerve and K stands for
Keep going:
• Play (S, S) in the first round.
• Play the mixed strategy Nash equilibrium in the second and third rounds if
(S, S) or (K, K) was played in the first round.
136
REPEATED GAMES
• Play the Nash equilibrium (S, K) in the second and third rounds if (K, S) was
played in the first round.
• Play the Nash equilibrium (K, S) in the second and third rounds if (S, K) was
played in the first round.
(a) Discuss the strategy and show that it is well defined (i.e., that it specifies
what to do in every contingency).
(b) Show that the strategy is a subgame perfect Nash equilibrium.
(c) Why can’t the payoffs (2, 2) be sustained if there are only two periods?
7.7 Consider the chicken game in Exercise 7.6 repeated infinitely many times. Find
a subgame perfect Nash equilibrium that sustains payoffs (2, 2) in each period.
Assume a discount rate as high as necessary (as long as it is smaller than 1).
7.8 Confirm that the strategy Tit For Tat played in the repeated prisoners’ dilemma
(finite or infinite) cannot beat any other strategy (i.e., cannot win more points
than the rival) in a pairwise confrontation.
7.9 Check that the strategy Tit For Tat is a Nash equilibrium in the infinitely
repeated prisoners’ dilemma, but not in the finitely repeated version of the game.
7.10 Confirm that the strategy Tit For Tat is not a subgame perfect Nash equilib-
rium in the infinite repetition of the prisoners’ dilemma game. Start by observ-
ing that, in order to know what the strategy dictates in a given period, we need
to know what happened in the period immediately before it, and that there are
three possibilities: both cooperated, both defeated, or one cooperated and one
defeated.
(a) What must be played in each case according to Tit For Tat?
(b) Is there any player that profits from a deviation in any of those cases?
7.11 Consider the public goods game in Section 7.7 used by Colin Camerer and
Teck-Hua Ho to illustrate their findings.
(a) Show that the strategy of contributing with zero is dominant.
Consider now a reduced version of the game with only three players, with
2
ui = 20 − ci + ( c1 + c 2 + c3 ) , and with only three strategies for each player: con-
3
tribute 0, 10 and 20 tokens.
(b) Show the normal form of the game and find the Nash equilibria.
(c) Argue that the game has the structure of a prisoners’ dilemma. Hint: You
can do it by reducing the game even further, with two players, two strategies
1.5
and ui = 20 − ci + ( c1 + c2 ).
2
137
8 AGENCY PROBLEMS:
ADVERSE SELECTION
In the next two chapters we present games in which different players have different
information. The information that one player has and the others lack is called private
information. Very often, the private information is about the preferences or the charac-
teristics of the player, but it can be more than that. For instance, one player may have
access to information about an external event that other players do not have. Examples
of this are privileged information and information acquired after conducting some
research. These new games are more difficult to analyze, but they allow the applica-
tion of game theory to many important and interesting aspects of social and economic
behavior, and they prompt discussion regarding some important philosophical issues.
First, in Section 8.1, we introduce a very general framework to study asymmet-
ric information: the agency problem. In Section 8.2 we add the information set as a
new element to model in the extensive form. The reader must pay special attention to
this new concept to understand Chapters 8 and 9 as well as some of the study cases
in Chapters 10 and 11. To facilitate the reading, Section 8.3 provides a first simple
application of these games. Sections 8.4 and 8.5 present one of the main topics in the
study of asymmetric information, the adverse selection problem, with an application
to public health systems. Other types of adverse selection are presented in Sections 8.6
and 8.7. Section 8.8 offers some discussion of information revelation via competition.
Finally, Section 8.9 presents a discussion on the Rawls’ concept of the veil of ignorance
in the language of the models examined in the chapter.
8.1 The agency problem

Many social relations consist of one person or organization that requires the services of
another party to achieve a goal which they cannot achieve alone. Using economics and
game theory jargon, we will refer to the party that establishes the objective as principal,
and the party whose services are required will be the agent. For this reason, the agency
problem is also known as the principal-agent problem.
The agency problem arises when the relation has the following characteristics:
• The agent has the information to achieve the goal and the principal does not.
• The principal does not observe the agent’s ability or effort.
• The attainment of the goal depends not only on the ability and effort employed by
the agent, but also upon random circumstances.
• Principal and agent have different objectives.
138
AGENCY PROBLEMS: ADVERSE SELECTION
Table 8.1 shows some examples of agency problems.

The question we will try to answer in this chapter is how to design a contract or a
relation between the two parties in the agency problem. There are, in principle, two
ways to address the question. In the first one, the principal specifies the multiple tasks
that she deems necessary for the achievement of the goal and then designs a contract
based on the realization of each one of those tasks. A second possibility is to give free-
dom of action to the agent and design a contract according to whether the goals set by
the principal are achieved or not. The principal has a goal and pays to obtain it, while
the agent has the information and decides on how to achieve that goal. Principal-agent
literature focuses on this second method, which is found to be more efficient in many
instances. To model these situations as games we need a way to deal with the informa-
tion or lack of it. In the next sections, we do just that.
Table 8.1 Agency problems
Principal’s Needs the Agent’s

Principal Agent Problem
objective agent for objective
Firm Maximize Production Workers Maximize Effort or
profits and sales salary and ability are not
minimize observable
effort
Organization Offer a Work on Departments Maximize The relevance
service different budget of of the
tasks power department’s
work is not
observable
Firm Maximize Get a loan Bank Maximize Information
own profits own profits over the
firm’s
investments
are not
observable
Client Maximize Risk Insurance Maximize The client’s
welfare insurance company own profits caution is not
observable
Citizen Maximize Medical State Universal Client’s risk
welfare insurance coverage at group is not
minimum observable
cost
Team Complete a Do each of Personnel Minimize Effort is not
task the parts in effort observable
the task
Buyer Minimize Buy a Seller Maximize The quality of
expenditure second-hand profits the car is not
car observable
139
GAME THEORY
8.2 The information sets

The simultaneous battle of the sexes in the extensive form
Up until now we have considered the normal and extensive form as two separate types
of games. In Section 4 we saw a way of going from the extensive form to the normal
form; the opposite is also possible. The static battle of the sexes game in Figure 1.3 can
be represented in the extensive form as seen in Figure 8.1, with F standing for Football
and M for Movies.
The discontinuous line that joins the two nodes where Bruno plays indicates that
he cannot tell them apart. In these two nodes Bruno has the same information (in this
case, its absence): if Sylvie played F, Bruno would not know whether he is in the node
after she played F or after she played M, just as if Sylvie played M. For this reason, we
placed the label “Bruno” only once, corresponding to the line that contains the two
nodes. If Bruno decides to play F, it will mean playing F in the two nodes; the same if he
decides on M. Bruno cannot condition his action on the node and play different things,
since he does not know where he stands. In the extensive form of the game, Sylvie plays
first and obviously cannot be made aware of Bruno’s choice, whereas Bruno, in spite
of playing second, does not know what Sylvie has done because he did not observe her
choice. For all purposes, it would be the same situation as if they played simultane-
ously. In other words, the situation that we represent in the normal form can also be
represented in the extensive form by joining the nodes in which the player has the same
information and by understanding the restrictions it imposes upon the players’ actions.
Observe the differences between this extensive form that represents a simultaneous play
and the extensive form in Figure 4.1a which follows an actual sequential game.
Sylvie
F M
Bruno
F M F M
2 0 0 1
1 0 0 2
Figure 8.1 The battle of the sexes in extensive form
The set of nodes that a player cannot distinguish are called information sets. To define
the players’ strategies we will have to consider their actions not in the nodes, but rather
in their information sets. In the game in Figure 8.1, Sylvie has two strategies (F and M)
while Bruno has those two as well (F and M). If we are given this as the original game in
order to obtain the normal form afterwards, we will get the game in Figure 1.3.
The prisoners’ dilemma with two types for Player 2

Let us complicate the prisoners’ dilemma game a little. Start with the game as in Fig-
ure 1.5a, which depicts the years in prison following the different strategy choices by
both players. Now add the possibility that Player 1 has empathic preferences for Player 2
in such a way that Player 1’s payoffs are his own years in prison plus half the years in
prison of Player 2, but only if Player 2 has empathic preferences also. This case was
140
taken into consideration in the game in Figure 1.5b. Player 2 may have empathic pref-
erences or be more of the selfish type, as in the original game. Player 2 is aware of what
type he is, which is unknown to Player 1; the latter, in turn, assigns the probabilities
¾ and ¼ that Player 2 is empathic and selfish, respectively. All this is known by both
players and can be represented in the extensive form of Figure 8.2a, where the reader is
asked to look closely at the information sets.
Let us first understand the representation of the game. Nature, N, starts by decid-
ing the type of Player 2 with the probabilities indicated above. To the left we have the
empathic type of Player 2, and to the right we have the selfish type. Observe that the
payoffs on the right side correspond to the original prisoners’ dilemma in Figure 1.5a.
The payoffs on the left correspond to the empathic preferences by both players, as
in Figure 1.5b. After Nature chooses the type, Player 1 has to choose between NC
(do not confess) and C (confess). He makes his choice unaware of what type Player 2
is, which is why his two nodes are included in one information set, indicated by the
discontinuous line containing them. Finally, Player 2 must choose what to play with-
out knowing Player 1’s choice, while aware of his own type. His nodes are divided
into two information sets. To the left is the information set that contains the nodes
in which Player 2 knows he is the empathic type, 2.E, but does not know Player 1’s
choice. To the right is the information set in which Player 2 knows he is the selfish
type, 2.S, but, again, does not know the actions of Player 1. Remember that each line
is an information set.
3/4 1/4
NC C NC C
2.E 2.S
NC C NC C NC C NC C
–1,5 –5 –2.5 –6 –1 –5 0 –4
–1.5 –2.5 –5 –6 –1 0 –5 –4
Figure 8.2a Prisoner 1 does not know whether Prisoner 2 is empathic
When faced with this kind of game for the first time it is tempting to solve each
part separately, but that would be a mistake. We will begin by examining why it would
be an error to solve each part separately. If we sought the Nash equilibrium of the game
when both players are empathic (the game in Figure 1.5b as if it were the left half of
the game in Figure 8.2a) we would find (NC, NC). If we were to do the same for the
right half with selfish preferences (as if the game in Figure 5.1a were the left half in
Figure 8.2a) we would have (C, C). The mistake is made evident once we realize that
doing this would imply that Player 1 plays NC in his first node and C in his second,
something that he cannot do. What Player 1 does in one node must be the equal to
what he does in other nodes within the same information set because he does not know
where he is.
141
GAME THEORY
The game in which Player 1 does know his position within the game (i.e., he knows
Player 2’s type) will be the one depicted in Figure 8.2b, where we can solve each part
independently.
If we recall the definition of a subgame perfect Nash equilibrium seen in Sec-
tion 4.3, its application will be as follows in the game in Figure 8.2b. Each node in
which Player 1 plays is the start of a subgame so, according to what we explained in
that section, we must find a Nash equilibrium in each one of those subgames in order
to build up the subgame perfect Nash equilibrium. It will be like this: ((C, NC), (C,
NC)). The first part of the equilibrium corresponds to the strategy chosen by Player 1:
cooperate in the first subgame (where players are empathic) and do not cooperate in the
second (when they are selfish). The second part corresponds to Player 2’s strategy and
is read similarly. But this is not the way to solve the game in Figure 8.2a. For that, we
need to clarify some things first.
3/4 1/4
1
1
NC C NC C
2 2
NC C NC C NC C NC C
–1,5 –5 –2,5 –6 –1 –5 0 –4
–1,5 –2,5 –5 –6 –1 0 –5 –4
Figure 8.2b Prisoner 1 knows whether Prisoner 2 is empathic
An improved definition of subgame and its application to the prisoners’

dilemma with two types
In the game in Figure 8.2a we cannot repeat what we did previously – when we
grouped Player 1’s two nodes within an information set, we eliminated the possibility
that the two parts of the game could be seen as separate subgames. In fact, the defi-
nition of subgame must be altered for games with information sets. Without writing
a rigorous mathematical definition, we can say that, in an extensive form game, a
subgame is any part of the game that starts in one node and contains all the elements
following that node as long as we do not “break” any information set. This is nec-
essary in order not to change the informational structure of players. As an example,
in Figure 8.2a we cannot take the first node by Player 1 and all posterior elements
because we would be breaking this player’s information set apart. In this case, the
only subgame is the whole game; therefore, the set of subgame perfect Nash equilibria
coincides with the set of Nash equilibria. To find them we will construct the normal
form, as drawn in Figure 8.3.
Player 1’s strategies are NC and C. Player 2’s strategies have two components,
the first one indicates what to do when he is the empathic type and the second indi-
cates what to do if he is selfish. The payoffs are calculated by taking the appropriate
142
Player 2
(NC, NC) (NC, C) (C, NC) (C, C)

Player 1 NC −1.375, −1.375 −2.375, −1.125 −4, −2.125 −5, −1.185
C −1.875, −5 −2.875, −4.75 −4.5, −5.75 −5.5, −5.5
Figure 8.3 The normal form of Figure 8.2a
averages. For instance, if Player 1 plays NC and Player 2 plays (C, NC), the payoff
3 1
( )
for Player 1 is calculated as follows: u1 NC, ( C, NC ) = ( −5 ) + ( −1) = −4. This
4 4
3
is because, according to this form of playing, with probability he will be playing
4
with the empathic type, which would be playing C (and then Player 1 would win −5),
1
while with probability he would be playing with the selfish type, who would be
4
playing NC (and the Player 1 would win −1). Player 2’s payoffs are calculated as the
expected payoff in the case that he is empathic (and wins −2.5) or selfish (and wins −1)
3 1
( )
u2 NC, ( NC, C ) = ( −2.5 ) + ( −1) = −2.125. Payoffs for all of the other strategy
4 4
combinations are calculated similarly.
In the normal form game, we have underlined the payoff corresponding to each
best reply in order to find combination (NC, (NC,C)) as the only Nash equilibrium.
Because there are no subgames besides the whole game, this is also the only subgame
perfect Nash equilibrium.
8.3 If you didn’t have anything to hide you’d show me your

e-mails
Let us take a look at another example of how to introduce the information using infor-
mation sets in the extensive form games. We will do it by exploring whether the sen-
tence “if you didn’t have anything to hide you’d show me your e-mails” makes any
sense.
Statistical research shows that 90% of males in a country are faithful to their wives,
while the remaining 10% are unfaithful. Individual A, a male, is coupled with B,
who has no additional information on his faithfulness aside from the male population
average. Aware of this, B thinks, “If I could access A’s e-mails, I would know for sure
whether he is faithful to me.” Then B argues, “If you didn’t have anything to hide you’d
show me your e-mails,” and threatens A with getting a divorce.
Let us represent the situation above as a game in which A’s choices of action are “let
B read my e-mail” and “do not let B read my e-mail,” while B’s choices are “divorce”
and “do not divorce.” A’s utility is higher (3) when he does not let B read his e-mail and
when B does not ask for a divorce. From there, divorce costs 2 to the faithful type and
1 to the unfaithful, while letting B read the e-mail costs 1 to the faithful and 2 to the
unfaithful. For her part, B prefers the divorce if A is the unfaithful type (with a utility
143
GAME THEORY
of 1) rather than not (0) and prefers not to get a divorce from the faithful (3) rather
than to go ahead with it (0). These numbers also reflect the fact that B’s utility is higher
when B is married to the faithful type. Figure 8.4 shows the game.
The game starts with Nature (N) defining A’s type. If he is faithful (A.F, with prob-
ability 0.9) he has to make a choice between giving A permission to read his e-mail (P)
or not granting that permission (NP). His choices will be the same if he is unfaithful
(A.U). Following A’s actions, it is B’s turn: she has to decide whether to get a divorce
from A (D) or not (ND). In the nodes after reading A’s e-mail, B will learn A’s type, but
if she does not read the e-mail she will not. This fact is shown in the discontinuous line
embracing the two nodes when that is the case, and that constitutes an information set.
When B reads A’s e-mail and A happens to be faithful, we are at the beginning of
a subgame with a simple solution: B chooses ND. Similarly, if A is unfaithful, B will
get a divorce from him after being allowed to read the e-mail. These two equilibrium
decisions are indicated with the arrows in the figure.
Faithful Unfaithful
0.9 0.1
A.F A.U
P NP NP P
B B B
D ND D ND D ND D ND
0 2 1 3 2 3 0 1
0 3 0 3 1 0 1 0
Figure 8.4a If you had nothing to hide
After having solved these two simple subgames, the game is like the one in Fig-
ure 8.4b. There we observe that there are no additional subgames aside from the whole
game. To construct the normal form game to find the Nash equilibria, note that A has
four strategies: (P, P), (P, NP), (NP, P) and (NP, NP). The first component in each
strategy indicates the action for the faithful type, while the second indicates the action
for the unfaithful type. B has two strategies, D and ND, as she only plays in one infor-
mation set and cannot distinguish between the two nodes in which she could be. If she
decides to get a divorce from A, it is a decision she will make without knowing for sure
whether A is unfaithful or not.
Once the strategy sets are identified, we can construct the normal form of the
game in Figure 8.5. Remember that we eliminated B from two places when we sub-
stituted the two simple subgames with their payoffs (the change from Figure 8.4a to
Figure 8.4b), so that we are left with B’s strategic choice when she is not granted per-
mission to read A’s e-mail. This means that sometimes, when A plays P, Player B does
not get to play at all. Let us stop for a moment to show how the utilities are calculated.
For instance, take the strategy combination ((P, NP), ND). A’s payoffs are calculated as
144
Faithful Unfaithful
0,9 0,1
A.F A.U
P NP NP P
B
2 0
3 1
D ND D ND
1 3 2 3
0 3 1 0
Figure 8.4b Solving the game “If you had nothing to hide”
follows. With probability 0.9 he is faithful and chooses P, which gives him a utility of
2 (B does not play, and her choice of (ND) is inconsequential). With probability 0.1 he
is unfaithful and chooses NP, B then chooses ND and A gets 3. A’s utility is then the
( )
average calculated as uA ( P, NP ) , ND = 0.9 × 2 + 0.1 × 3 = 2.1. B’s payoffs are 3 and
(
0, respectively, given an expected utility of uB ( P, NP ) , ND = 0.9 × 3 + 0.1 × 0 = 2.7. )
Other utilities are calculated similarly.
B
D ND
(P, P) 1.8, 2.8 1.8, 2.8
(P, NP) 2, 2.8 2.1, 2.7
A
(NP, P) 0.9, 0.1 2.7, 2.8
(NP, NP) 1.1, 0.1 3, 2.7
Figure 8.5 The normal form of the game “If you had nothing to hide”
In Figure 8.5 we have marked the best replies, from where we find two Nash equi-
libria in pure strategies: ((P, NP), D) and ((NP, NP), ND). In the first one, the faithful
type allows B to read his e-mail, whereas the unfaithful type does not. When B is not
allowed to read the e-mail, she can correctly deduce that A is unfaithful (otherwise, and
according to the equilibrium, he would have let her read it). Thus, if this is the case, B
gets a divorce from A. However, in the second equilibrium neither of A’s types gives B
permission to read the e-mail. In this case, when B is not allowed to read the e-mails, she
cannot deduce that A is unfaithful since she is not allowed to read them by the faithful
type either. Now B can only believe that A is unfaithful with probability 0.1, and that
he is faithful with probability 0.9. With these probabilities, she maximizes her utility by
not getting a divorce from A. Thus, the argument “if you didn’t have anything to hide
you’d show me your e-mails” only makes any sense in the first equilibrium.
145
GAME THEORY
8.4 Adverse selection in a first agency problem

We have a firm operating in a given sector and a potential worker that can be of
high or low skill (called Types W.H and W.L, respectively). If she is of Type W.H,
her productivity in the sector is 150, but if she is of Type W.L, her productivity
is only 60. The probabilities of Types W.H and W.L are 1/3 and 2/3, respectively
(e.g., the worker may be a random individual from a population of workers where
the proportions of high- and low-skill workers are 1/3 and 2/3, respectively). The
firm hires in perfect competition with other firms; therefore, the firm must pay
a salary equal to the productivity of the worker or she will go to a different firm
that will pay a higher salary. The reservation salary is 100 for type W.H and 40
for Type W.L. The reservation salary is what a worker expects to be paid if she
works at her best job opportunity in another sector. The firm is considering paying
the worker one of these three figures, 60, 150 and 90, which correspond to the
productivity of a W.L type, a W.H type, and a random worker, respectively. What
will the firm do?
We also assume that the worker cannot do anything to show her productivity and
that there are no different levels of effort that the worker can exert. In the next chap-
ter we will include these possibilities. The extensive form of the game is detailed in
Figure 8.6a.
N
1/3 2/3
Type W.H Type W.L
60 90 150 60 90 150
W.H W.H W.H W.L W.L W.L

A R A R A R A R A R A R
90 0 60 0 0 0 0 0 –30 0 –90 0
60 100 90 100 150 100 60 40 90 40 150 40
Figure 8.6a Hiring a worker of unknown productivity
First, Nature (N) defines the worker’s type. Keep in mind that this says nothing
about whether productivity is innate or acquired; it only says that, from the viewpoint
of the firm, the probabilities of the types are determined previous to its decision on
which salary to offer. Then, not knowing the type, the firm (F) must offer a salary (60,
90 or 150) that the worker must accept (A) or reject (R). If the worker rejects the salary,
the firm gets zero and the worker gets her reservation salary. If she accepts, the firm
obtains the productivity minus the salary and the worker obtains the accepted salary.
In the figure we observe six subgames in which the worker has to make her decision.
In each of them, she will accept as long as the salary offered by the firm is higher than
her reservation salary, as indicated with the arrows.
Once the subgames are solved, what remains is a game in which only the firm plays
to decide what salary to offer. This is represented in Figure 8.6b.
146
Now let us calculate what the firm expects to win for every salary it can offer. If the
firm offers 60, it will win zero regardless of the productivity of the worker. If it offers
1 2
90, it will win × 0 + × ( −30 ) = −20 and, finally, if it offers 150 it will expect to win
3 3
1 2
× 0 + × ( −90 ) = −60. Thus, the firm offers a salary of 60. It cannot offer the salary
3 3
corresponding to the high-productivity type, which is 150, because both high- and
low-productivity types will accept, and the firm will lose money. It cannot offer the
salary that reflects the average productivity, that being 90, thinking that both types
will accept, as this salary is too low for the high-productivity worker to accept and will
only attract low-productivity workers. The only possibility is to accept that the firms
can only attract the low-type workers if they do not want to lose money, and then pay
them accordingly.
Type H Type L
1/3 2/3
F
60
150 60
90 90 150
0 0 0 0 –30 –90
Figure 8.6b Solving the game “Hiring a worker of unknown productivity”
This phenomenon, by which the good workers cannot find an adequate contract, is
called adverse selection. More generally, it happens when high quality cannot be distin-
guished from low quality before the decision to buy or sign a contract. In this situation,
the most that the principal can pay is a price or salary that reflects the average quality
of the agent. If this is not enough to attract the high-quality type, the market will only
select the low quality, providing meaning to the expression adverse selection.
Throughout this problem we have assumed that if the worker is not hired by the
firm, she will get her reservation salary outside the sector. But why would she not
obtain a quantity equal to the salary by a rival firm in the same sector? There are two
reasons for this. First, the salary that any firm in the sector can afford to pay is precisely
what we have calculated. Second, once we calculate the equilibrium salary with one
firm, and given the perfect competition assumption, this is also the best salary any firm
can offer. No firm in the sector can improve upon it. In the end, it would be the same
if the worker had to choose between her reservation salary on one hand and the salary
of one firm, or of any firm in the sector, on the other hand.
8.5 Adverse selection and public health systems

The adverse selection problem has important consequences in the design of an
insurance system, be it private or public. The basic reason for buying insurance is to
reduce risks due to uncertain events. If I have a property worth 100, but I face a 2%
probability ( p = 0.02) that it is destroyed, the expected value of the property will be
V = 0.98 × 100 + 0.02 × 0 = 98 . With such probabilities, if I am a risk-averse person,
147
GAME THEORY
I am definitely going to prefer 98 (e.g., a property worth that amount) as opposed to a

lottery between zero and 100 with the stated probabilities. This means that there is a
margin to come under contract with an insurance policy with a company that is less risk
averse than me. For instance, a risk-neutral insurance company can offer me a policy
that provides complete indemnity in case my property is destroyed at the price of 2. The
company would break even in expected terms, and I would have a riskless property worth
98 (100 minus the price of the insurance policy). In real life, the company will charge
more than 2 or will set an indemnity below the total loss in order to make some profit. In
our examples, however, we will assume the most favorable case for the consumer to avoid
unnecessary complications, as the conclusions will not depend on this simplification.
Insurance companies can show a much smaller degree of risk aversion compared to
consumers or small firms because they can pool the risks of many clients with a very
high degree of accuracy. If an insurance company sells one million policies like the one in
the example, it will collect a revenue of two million (charging 2 per policy), just enough
to cover the expected claims of 20,000 clients that would lose their properties.
This general view is also valid for health insurance. Consider a society in which
individuals have a health of 100 as estimated by some appropriate index. If they become
ill their health level will be zero, which means that it costs 100 to restore the utility
level of 100 (i.e., the health index measures the cost to restore health to the 100 level).
Say that half of the individuals are high risk (call them group H) and have a probability
0.75 of becoming ill. The probability is 0.25 for the other half of individuals, the low-
risk individuals (L). All individuals are risk averse and have the same utility function
u ( x ) = x that measures their satisfaction for a health level x.
The best policy for each group

An insurance company that observes who belongs in each group and that offers the
best policy to each individual (because of a perfectly competitive insurance market and
because of the zero costs and no-profit simplifications) will offer the following policies:
• Group A’s policy: Full indemnity, 100, at a price (premium) of 75. This policy is
denoted by (100, 75).
• Group B’s policy: Full indemnity, 100, at a price of 25, denoted by (100, 25).
Note that the company makes zero profits with either group:
( )
P Policy (100, 75 ) = 0.75 × ( 75 − 100 ) + 0.25 × 75 = 0,
P ( Policy (100, 25 ) ) = 0.25 × ( 25 − 100 ) + 0.75 × 25 = 0 .
Now consider the individuals’ decision to buy the policies. If they get their health back
after becoming ill, which is worth 100, and pay the premium whether they are sick or
not, then for the individuals in group H, utilities without and with the policy are
uH ( No policy ) = 0.75 × 0 + 0.25 × 100 = 2.5 and
uH ( Policy (100, 75)) = 0.75 100 − 75 + 0.25 × 100 − 75 = 25 = 5 .
148
Similarly, for the low-risk group L, we have
uL ( No policy ) = 0.25 × 0 + 0.75 × 100 = 7.5 and
uL ( Policy (100, 25)) = 75 = 8.66 .
Both types prefer the insurance designed for them rather than being without insurance
and assuming the costs of becoming ill.
Adverse selection
When the insurance company is unable to identify the individual’s group we may have
an adverse selection problem, as was the case in the previous section. For this to occur,
it is necessary that individuals know their own types. Let us begin with an attempt by
the company to set a policy that offers full coverage at an average premium. Since both
groups are equally abundant, the premium would be exactly the average of the prices
1 1
calculated before: × 75 + × 25 = 50 . A simple computation shows that by charging
2 2
this price, the company makes zero profits if all individuals buy the policy:
1
( )
P Policy (100, 50 ) , all buy =
2
( 0.75 × (50 − 100 ) + 0.25 × 50 )
1
+ ( 0.25 × ( 50 − 100 ) + 0.75 × 50 ) = 0.
2
The utility of the different individuals with this policy (100, 50) are
uH ( Policy (100, 50)) = 0.75 100 − 50 + 0.25 × 100 − 50 = 50 = 7.07 and
uL ( Policy (100, 50)) = 50 = 7.07.
Type H individuals will buy the policy, but Type L ones will not. Remember that, as we
previously computed, without any policy their utilities were 2.5 and 7.5, respectively.
Therefore, low-risk individuals would not be interested in buying this policy designed
for the average citizen: it is too costly for them. They would be subsidizing the high-risk
individuals. Low-risk individuals leave the market, causing an adverse selection prob-
lem. The company, foreseeing this, would only offer a policy designed for the high-risk
individuals: (100, 75).
The extensive form of this game is similar to the game in Section 8.4, and is rep-
resented in Figure 8.7. First, Nature defines the type of the consumer. Then, without
knowing that type, the firm must offer an insurance policy, which we have reduced
to two options: full indemnity with a price of 75 designed for high-risk individuals
and full indemnity with a price of 50 designed for the average individual. The arrows
show the subgame perfect Nash equilibrium: ((Policy (100, 75)), (A, A), (R, R)). The
first part, (Policy (100, 75)) is the insurance company’s strategy, (A, A) is the strategy
of an individual from group H who accepts any insurance policy, and (R, R) is the
strategy of L type who rejects both. In the game, when we refer to the policies of the
company we write the price of a policy only, since indemnity is always 100.
149
GAME THEORY
Type H 1/2 1/2 Type L
Company
75 50 75 50
Type H Type H Type L Type L
A R A R A R A R
0 0 –25 0 50 0 25 0
5 2.5 7.07 2.5 5 7.5 7.07 7.5
Figure 8.7 Adverse selection in the insurance market
The occurrence of adverse selection should not be confused with other problems
that may be affecting the health insurance market. For instance, the possibility of
adverse selection does not contradict the fact that, when the company does know the
individuals’ types, it will charge a high price to the high-risk individuals who, con-
fronted with financial restrictions, may not be able to afford these policies and will
leave the market. Companies may know the type of some individuals but not others,
and both situations may coexist within the same market.
Two ways of preventing adverse selection

All the agency problems derive from the lack of information and could be potentially
solved if the information is revealed. How can this occur? In our insurance market
example, the low-risk individual would be happy to inform the company that she is
indeed a low-risk person, but there is no reason for the company to believe her since it
is in the interest of the high-risk individual to pretend to be a low risk. The company
needs a proof, maybe in the form of a medical test. If the low-risk person can provide
the proof, then she will get her low-price policy. If there is such a test to tell the groups
apart, the low-risk person will be willing to be tested. The high-risk individual will not,
but that will be just the same since the company will infer that she is a high risk. Once
the company has the information, it offers policy (100, 75) to the high-risk group and
(100, 25) to the low-risk group.
In a second solution, the government mandates that every citizen purchase the pol-
icy that reflects the average risk of the population, (100, 50). Obviously, individuals
from the H group would prefer the second solution, while those in the L group would
prefer the first, according to our calculations above.
There are, however, a couple of additional considerations in favor of the second
solution. If we have to decide between a system in which we must reveal our risk group
and a system in which we are obliged to buy the single policy, and if we have to make
that choice before we know our own risk group, which system would we prefer?
Let us consider our prospects in the two systems. Take the first one and remember
that we expect to be of either type with probability ½. This means that if we are of
Type H, we will be offered a policy (100, 75) and will have utility 5. If we are of Type L,
150
we will get policy (100, 25) and utility 8.66, as calculated previously. This means
1 1
we will obtain an expected utility of 5 + 8.66 = 6.83 in the case in which we are
2 2
required to take a test in order to learn and reveal our type before contracting an
insurance policy. In the second system, regardless of what our type is, we are fully
insured with the policy (100, 50) and get utility 7.07, which is higher than in the
previous case. That is, when individuals do not know their type, they prefer the single
mandatory policy rather than being sorted out by type. After all, being of Type H or
L is an uncertain outcome – a risk that individuals face. A risk-averse person would
willingly buy a fair insurance against that risk. This is what the single mandatory
policy provides.
The alert reader has probably made a connection between this ex ante election crite-
rion – the economic jargon for making choices before knowing the information – and
Rawls’ original position in which citizens are asked to make choices about their society
behind a veil of ignorance (i.e., without knowing their place (status, gender, etc.) in
that society). Even if decisions in society are made by voters that typically know many
of their own characteristics, and particularly their risk group, the mental exercise of
choosing as if they did not know is attractive if they care about the well-being of their
descendants, whose characteristics are not fully known.
A last consideration in favor of mandatory insurance has to do with the adverse
selection problem seen in Section 4.7 in the health game where the government cannot
make the commitment of refusing to provide medical care to a citizen in need.
The case of the genetic testing in the UK

In the mid-1990s, Europe witnessed a debate over the use of genetic testing by health
insurance companies before selling a policy. The public debate was stronger in the UK
than in other countries, and in the year 2000 the Genetics and Insurance Committee
accepted the use of a genetic test to detect a predisposition for Huntington’s disease
as a condition to purchase life insurance. However, in 2001, the British Parliament
reopened the debate and in October that year reached an agreement with the Asso-
ciation of British Insurers to recommend a five-year moratorium for this type of test.
The moratorium has been extended several times since that year and is still in effect at
the time of writing this book.
Among the arguments in favor of the test was the need to solve the adverse selection
problem and to provide the best policy for each individual. The arguments against
the test, besides its unreliability, were based on criteria of fairness and equity and were
divided into two groups. On the one hand, there was the non-discrimination solution
to the adverse selection problem we just saw; on the other hand we have the argument
that individuals may not take the test when it was needed and thus would lose a key
element in the diagnosis of the disease due to fear that the result would be used against
them if they wanted to buy an insurance policy.
8.6 Other examples of adverse selection

The problem of adverse selection may arise every time individuals cannot reward good
quality the way it deserves. Let us look at some more examples.
151
GAME THEORY
The market for second-hand goods

The seller knows if her car is a good car or a lemon, though the buyer cannot tell. Under these
circumstances, the buyer will not be willing to pay more than a price that reflects the value
of an average car. This price may be too low for the owner of a high-quality car; therefore,
she will not put her car up for sale and we will see only low-quality cars in the market. Actu-
ally, this was the first example studied with some rigor within the literature of e conomics
of information by George Akerlof in his famous article “The Market for L emons” (Akerlof,
1970). He later won the Nobel Prize for his contributions to this literature.
The credit market

A bank has to decide on whether to loan money to a firm, not knowing if that firm
has a good or a bad project to invest it in. If the project is good, the firm will be able
to return the loan plus interest, but if it is bad, the firm will lose the money and will
declare bankruptcy. Anticipating this possibility, the bank will be willing to loan the
money at a high interest to compensate for the possibility of bankruptcy. However, the
high interest may discourage a firm with a good project from asking for credit, and
the firm may try another way of seeking financing.
The case of the political party and the expert

A political party needs an expert within its ranks and is considering hiring one, but it
cannot tell a true expert from a fraud. The reader can fill the details to make an adverse
selection story in line of the previous examples. But let us make the game a little more
interesting by changing the roles of the agent and the principal.
An expert decides how much effort to dedicate to the political cause but cannot distin-
guish if the party he is thinking of joining is a good one or not. A good one will give him
an interesting career in politics, whereas a bad one, filled with hypocrites and careerists,
will not. Alternatively, instead of the whole party being good or bad, it may just be the sec-
tion in his city or his mentor. The expert is only interested in joining the party if he is paid
with the opportunity of a career in politics. Without knowing if he will be rewarded that
way, he can only commit himself with normal dedication to the party; however, the good
party will only be interested in someone with a commitment to excellence. Only a bad
party will settle for less than that. Knowing this, his dedication to the party will be low.
If, in addition, the bad experts can also offer a low dedication more easily than the good
ones, but cannot offer a high dedication, we will have bad experts advising the parties.
Depending on the particular adverse selection problem, there will be ways to solve it
or at least alleviate it. In the case of medical insurance, we looked at two such ways. In the
market for used goods, a system of guarantees specified in a contract or through a reliable
mediator may do the job. In all the cases, individuals need to pay a cost in order to acquire
information about the quality. In the next chapter, we will see yet another example.
8.7 Other types of adverse selection

In economics of information and in game theory, the concept of adverse selection is the
one described and studied in the previous sections. However, there are other circum-
stances in which a principal can end up with the worse agent for reasons of a different
kind than the ones already mentioned. Here are a couple of such examples.
152
The principal knows the agent’s type, but cannot keep her promises
Let us go back to the political party that wants to attract an expert, and consider the
case in which the principal is the party and the agent is the expert (who can be good or
bad). Now we will suppose that the party can tell the type of the agent. Under these cir-
cumstances there would be no problem of signing a contract beneficial to both parties:
one generous contract offering perspectives for a career in politics for the good expert,
and a lesser contract (or no contract at all) for the bad expert. However, the party may
have commitment problems. For instance, the leaders may change and the new ones
may not honor previous agreements or these may not be attractive to the expert, or
the party may not achieve any political success and may be unable to provide a career
to the expert. This may discourage the good expert from entering into a relation with
the party, providing another example in which good quality is expelled from a market.
This time, the problem is not the observability of the quality but the difficulties in
keeping the promises.
Choosing the bad type to win in the short run

A part of the population feels discriminated against. It may be an ethnic, linguistic or
religious minority, for example. There are demonstrations and riots causing a violent
conflict with the rest of the country almost to the point of civil war. After the conflict
come the peace negotiations. The minority must choose either a politician or a military
leader. Either of them will negotiate better conditions for that minority, as well as for
herself and her acolytes. The politician has an inferior negotiation position, but her
propensity towards corruption is limited, as it is controlled by the citizens. With armed
militia under her orders, the military leader can negotiate from a stronger position,
though she could be a corrupt leader also. The rebels have a dilemma; they must opt
between having a strong position for negotiations while risking corruption afterwards
or having a weaker position yet better future governance.
If the differences between the negotiation power between the politician and the
military leader are big enough – for instance, if the politician has no power at all– then
the rebels may choose the military leader. This may be the case even if the future ben-
efits of having an honest politician are greater than the expected benefits should the
military leader be in power, because it is highly likely that she will be corrupt. The rea-
son is that the rebel community cannot capitalize those future benefits and use them
to obtain concessions in the peace negotiations (for instance, committing to future
payments to the country in exchange for concessions now). Another possibility is that
the rebels do not do their math adequately, thus giving disproportionate weight to the
present gains relative to the future ones. Either way, by electing a military leader, they
will also be choosing the worse future leader with some probability. A third possibility
is that the leader is chosen not by all of the rebels but by her closer followers. Now the
military leader’s troops have more to gain personally and will dedicate more effort and
resources to make sure that their leader is in charge instead of the politician. They will
be choosing in a rational way according to their interests, but not according to the
minority’s.
According to social scientists Scott Gates and Kaare Strøm, this mechanism explains
why many conflicts in Africa and the Middle East end up in peace negotiations with
the large presence of military leaders (Gates and Strøm, 2012).
153
GAME THEORY
8.8 Competition reveals information: When the principal has

information about the agent
In all this chapter we assumed that the agent has private information that the principal
ignores. Let us dedicate some time to the opposite case in which the principal knows
more about the type of the agent than the agent herself.
Oftentimes you do not know the value of something in your possession. You may be
like Kino, the character in John Steinbeck’s novel who found the “Pearl of the World”
and tried to sell his valuable treasure to a pearl dealer. Or you may be looking for health
insurance knowing that the companies know your risk type more accurately than you.
After all, they have actuarial tables that can place you very precisely in a risk group
according to your age, sex, neighborhood, profession, marriage status, etc. Even more,
they can require you to conduct a test to know about your genetic predispositions.
So, unless you have relevant private information about yourself that is not the average
of your group, and unlike our assumptions in the previous sections, the insurance
company can know you better than you know yourself. In this situation, you go to an
insurance company and they tell you that you are a risky costumer and demand a high
price to sell you an insurance plan. Do you trust them? Is it not in their interest to cheat
you? Would they not be like the pearl dealers of the novel who confabulate to get higher
profits than they would in a competitive market?
Imagine that the actuarially fair price for your insurance policy is 100, but you
do not know that. If insurance companies compete for customers, in the only Nash
equilibrium they have to offer you the policy at 100. Any firm charging more than
that will be undercut by another company offering to charge less. Things become more
interesting when we realize that there is no need for all firms to know your risk type
to make you a good offer. If they see other firms offering you a low price, they know
that you must be a low-risk individual. This is just an illustration of a more general the-
orem in economics that states that perfectly competitive markets are informationally
efficient, in the sense that the price aggregates and reflects all the relevant information
that buyers and sellers have. Of course, it does not reflect any other information that
is not known to anyone in the market, and the theorem does not hold if there is not
perfect competition (e.g., the firms form a cartel). Still, it is a very powerful theorem
to analyze markets.
In regular markets, with no information dispersion, prices are posted and everyone
knows them. In these other markets, with private information and not knowing the
price that corresponds to your case, competition still works if it is the party that com-
petes who knows the price. This was not the case when the private information was
held by the agent, as the agent had monopoly power over that information.
There is a puzzling aspect to this information revelation. If the market indeed
reflects the competitive price, then there is no need for you or anyone else to ask around
to get the best offer. But if this is the case, then there is no need for the firms to give
you the best deal. Of course, this cannot be an equilibrium, as in that case it would pay
for you to do some search. Is there a solution to this puzzle? One easy way to justify the
competitive equilibrium in this situation is by considering that there is always a chance
that some people do search around for the best price even if the markets are compet-
itive. After all, the perfect competition is only perfect in a theoretical model, and in
the real world any imperfection will induce looking for the best price. That would be
enough for firms in competitive markets to lower their prices.
154
8.9 On Rawls’ original position and the ex ante criterion

At the end of Section 8.5 we mentioned the similarities between the ex ante criterion –
the choice of a mechanism before the agents have information about their types – and
Rawls’ veil of ignorance. In this section we discuss this in more detail. First, we present
a simplified version of Rawls’ criterion that may not do justice to the author. Never-
theless, it is a version that is discussed quite often and it will serve as a starting point
to discuss the more complex criterion. As usual, our main goal with this discussion is
not so much to convince the reader that a particular point of view is correct, but to
show how some points of the different theories can be expressed very naturally in the
language of game theory.
The simplified original position

The motivating principle is that a good justice criterion to compare two societies is
to detect which society would be preferred by an individual in the original position.
This means that the individual will live in the society of their choice, but without
knowing which position they will occupy in the society. They may be a woman or
a man, a slave or a cruel master, a peasant or a noble, an illiterate or a well-educated
person, a coward or a reckless individual, among many other things. In other words,
behind the veil of ignorance, without knowing which character you will incarnate
in Calderón de la Barca’s Great Theater of the World, in which play do you want
to act?
To make the example simple. Consider two possible societies, each with five
inhabitants. In Society A, all the individuals have an income of 400. In Society B, one
individual has an income of 300 while the other four have 1000 each. There are no
other differences among the individuals. Which society is preferred? The choice under
the simplified original position implies that individuals must choose according to the
maximin rule seen in Section 2.2. This means that the individual will first detect the
person in the least advantaged position in each society and then choose the society
where this person is better off. In our case, in Society A the lowest income for an
individual is 400, and in Society B it is 300; therefore, Society A is preferred. However,
we can easily observe that the consequence of the maximin rule does not follow from
the antecedent of the veil of ignorance. Many people will be willing to risk being the
person with income 300 rather than 400 if that means a very big chance to have an
income of 1000 instead of 400. Only very risk-averse individuals would prefer Society
A over B.
A more complex original position

Rawls’ argument was not that simple, and it departs from the above in at least three
important nuances:
• Before deciding behind the veil of ignorance, individuals may agree on some cir-
cumstances they would appreciate in a society.
• Rawls did not say that the maximin was to be applied in every case.
• In addition to not knowing their position in the society, individuals do not know
which utility function to use behind the veil of ignorance.
155
GAME THEORY
Let us address these questions. Say that the individuals agree that no one should have
an income smaller than 300. Then we can order societies using the maximin criterion,
starting with the worst ones depending on how far the least fortunate individual is from
this amount. For societies where everyone is above 300, then we could follow the ex ante
criterion using a utility function, rather than the maximin criterion. However, in this
case the maximin criterion may not be workable when individuals agree on more than
one minimum. For instance, say that the individuals agree that no one should have an
income less than 300 and also that no one should have less than 8 years of education.
Now, take these two societies: in Society C the least fortunate has 100 as income and
6 years of education, while in Society D the numbers are 200 and 4. The maximin
criterion does not offer an answer regarding which of the two societies is better.
The problem with the maximin criterion is that it can almost never be applied.
As we just saw, the moment we focus on more than one dimension (income and educa-
tion, in the above example) we may not even be able to define who is the least fortunate
person in the society. This occurs, for instance, when the two persons of the previous
example belong in the same society. We need to balance (100, 6) vs. (200, 4), which
calls for a utility function.
Finally, if individuals do not know which utility to apply, they have no way to eval-
uate the society using a particular degree of risk aversion. There are many utilities that
show a great deal of risk aversion, as there are many others showing other risk attitudes.
Choosing an average over all possible utility functions is just one possibility, if such a
thing could be precisely defined. Rawls maintains that, in these circumstances, the
maximin criterion is the reasonable thing to do. However, it is still not deduced from
the conditions of the thought experiment. Actually, nothing is deduced, and the choice
of a criterion to evaluate the societies cannot be done without a rhetorical appeal to
moral preferences. In the absence of data nothing can be said.
Out of the discussions on the Rawlsian approach based on the original position,
one of particular interest is the Rawls–Harsanyi dispute. Harsanyi won the Nobel Prize
in Economics for his contributions to the study of games with asymmetric informa-
tion, and he presented arguments in line with what we have discussed in this section.
The dispute is still being studied in philosophy. For instance, philosopher Michael
Moehler attempts to reconcile the two views (Moehler, 2015). His proposal, however,
lacks some of the imprecisions we just noted.

8.1 Sylvie and Bruno play the battle of the sexes game in Section 1.3 but with
an added complication: Bruno may be of two types, Normal and Depressed.
If Bruno is Normal, his payoffs are like those in the original game. If he is
Depressed, Sylvie’s payoffs do not change, but Bruno’s become zero if he coin-
cides with Sylvie in the same place. On the other hand, when they do not meet,
he gets 1 if he goes to the football game and 2 if he goes to the movies. Sylvie
does not know Bruno’s mood, but he knows his own mood. The probability
that Bruno is Depressed is 1/3, and they are both aware of that.
(a) Draw the extensive form of the new game. Hint: It is easier if Bruno goes
first in the tree.
(b) Find the subgame perfect Nash equilibria in pure strategies.
156
8.2 In the game in Section 8.3, change the probabilities of being faithful and
unfaithful for Player A so that now the probability of being faithful is 0.1.
(a) Find the subgame perfect Nash equilibria in pure strategies.
(b) What is the minimum probability of A being faithful so that the game has
an equilibrium in which no type shows their e-mail?
8.3 A worker may be one of three Types, A, B or C, depending on whether his
productivity is 50, 100 or 150, respectively. A firm is interested in hiring him
and, due to the competition from other firms, will have to pay according to
his expected productivity. The problem is that the firm cannot observe the
productivity. Assume that reservation prices for the worker are 0, 40 and 100,
respectively, and that Type A constitutes 40% of the population, while Types B
and C each constitute 30%.
(a) Calculate the salary that corresponds to the average productivity. Which
type will not accept this salary and leave the market?
(b) Calculate the salary that corresponds to the average productivity of the
types that did not leave the market in (a). Which types will accept it?
(c) Describe the subgame perfect Nash equilibrium in light of the results in (a)
and (b).
8.4 Consider the game in Figure 8.6a. Is there an adverse selection if we change
the following data? (Consider only one change at a time, without accumulating
them.)
(a) The probability of being Type W.H is 1/2 instead of 1/3, and salaries may be
60, 105 or 150. (Notice that 105 is the new average productivity.)
(b) The reservation salary for Type W.H is 85 instead of 100.
(c) Type W.L’s productivity is 90 instead of 60, and salaries may be 90, 110 and
150. (Notice that 90 corresponds to the new low-type productivity and 110
is the new average.)
8.5 Consider the game of health insurance in Section 8.5, but now with the clients
having a utility function given by u ( x ) = 3 x . The insurance company knows
the types of its clients.
(a) Calculate the utility of each type in case they buy the policy designed for
them.
Assume now that the insurance company cannot tell the types apart. (Like in
the game in Figure 8.7.)
(b) Calculate the utility of each type in case they buy a policy designed for the
average client.
(c) Is there a problem of adverse selection?
(d) What is the subgame perfect Nash equilibrium?
8.6 Consider the game of the health insurance in Section 8.5, but now the insur-
ance company sets a price 10 units higher than the prices calculated in the
section in order to cover the operation costs. The insurance company knows the
types of its clients.
(a) Calculate the utility of each type in case they buy the policy designed for
them.
157
GAME THEORY
Now assume that the insurance company cannot tell the types apart. (Like the
game in Figure 8.7.)
(b) Calculate the utility of each type in case they buy a policy designed for the
average client.
(c) Is there a problem of adverse selection?
(d) What is the subgame perfect Nash equilibrium?
8.7 Consider the game of health insurance in Section 8.5 as depicted in Figure 8.7.
Now assume that the probability of Type H is 1/3 instead of 1/2.
(a) Describe the insurance policy designed for the average type.
(b) Draw the extensive form of the game following the steps in Section 8.5.
(c) Find the two subgame perfect Nash equilibria of the game. Hint: In one of
them the equilibrium policy is (100, 75); in the other, the equilibrium policy
is the one found in (a).
(d) Which one is preferred by the client behind the veil of ignorance?
(e) Which equilibrium is more likely to prevail in the market? Hint: Think of a
strategy by a rival firm to capture clients from the first firm.
8.8 Represent the market for second-hand goods in Section 8.6 as an extensive
form game and show the subgame perfect Nash equilibria. To do that, choose
some appropriate numbers to represent payoffs and also the probabilities for
the adverse selection problem to arise (e.g., a good car may be worth 100 to the
seller and 130 to the buyer). Hint: Recall the games in Figures 8.6a and 8.7.
8.9 Repeat Exercise 8.8, only now for the case of the political party and the expert
in Section 8.6.
8.10 Repeat Exercise 8.8, now choosing the numbers so that the adverse selection
problem does not arise.
158
9 AGENCY PROBLEMS:
SIGNALING AND
MORAL HAZARD
In this second chapter dedicated to the study of asymmetric information we will explore
two other expansive topics regarding game theory and economics of information: sign-
aling and moral hazard. Even if a high-ability worker cannot show their ability directly,
they can often show something that is closely related to it. Frequently, that “something”
will serve as a signal of their ability if the same “something” cannot be shown by the
low-ability worker. In Section 9.1 we present the precise formulation of this idea and
its application to the study of education as a signal. Section 9.2 offers some empirical
findings that support the theoretical idea, and Section 9.3 uses this model to illustrate
some possible aspects of discrimination in the labor market. The moral hazard problem
differs from adverse selection in the way the types of the agent are defined. In adverse
selection (and in signaling) the types are defined exogenously, and the agent cannot do
anything about that. However, in moral hazard problems, the agent chooses the type –
generally, at a cost. The implications are explored in three cases. Section 9.4 defines and
illustrates the concept in the labor market, while Sections 9.5 and 9.6 show applications
to the health system and to working in teams, respectively.
9.1 Signaling with a discrete variable

One of the aspects in which modern economics differs most from the neoclassical
paradigm is the treatment of information, following the works of Akerlof, Spence and
Stiglitz, who won well-deserved Nobel Prizes for their contributions. Some theoretical
models were introduced to explain economic phenomena that did not fit well in the
old theory, such as Akerlof’s market for lemons, seen in Chapter 8. With other models,
the interaction between theory and observation was the opposite. Spence (1973) stud-
ied the consequences of asymmetric information in the job market and found some
unexpected theoretical results; for instance, in addition to its more standard uses (e.g.,
consumption and acquisition of human capital), education may serve to signal the
ability of a worker. The empirical validation of the theoretical hypothesis came later.
To understand how signaling works, it is better to isolate its effect in a simple model.
For that, Spence considered a situation in which the ability of workers was fixed and
could not be changed through education. This way education could not be used to
accumulate human capital and obtain higher salaries. In addition, Spence assumed that
education was costly and that individuals did not get any direct benefit from its con-
sumption. Under these conditions, it seems that education is not good for anything.
However, Spence found that if the worker is aware of her own ability, but the prospec-
tive employer is not, and if the cost of acquiring an education is lower for employees
159
GAME THEORY
with higher abilities, then an equilibrium may exist in which only high-ability workers
pay the cost to receive the education. They do it in order to signal that they are of high
ability. The low-ability workers do not invest in education because it is very expensive for
them to do so; the higher salary that they could get pretending to be high-ability workers
does not compensate for the elevated cost of education. If this is the case, employers can
be certain that the educated workers are also those with high ability and will pay them
accordingly, thus completing the conditions for the situation to be an equilibrium. Next,
we express all of this in the language of game theory using an example.
Signaling with a discrete variable

Nature selects, with equal probabilities, the ability (a) of a worker (high with a = 5 or
low with a = 1 ). The worker, who knows her own ability, chooses an education level e
( e = 0 or e = 2 ). The firm does not know the worker’s ability, but does observe her
education level, and must decide which salary w to offer ( w = 1, w = 3 or w = 5 ).
Finally, the worker must decide whether to accept or reject the offer. Reservation sal-
aries are wH = 3.5 and wLR = 0 for a high- and a low-ability worker, respectively. The
R
profits function for the firm is B ( a, w ) = a − w (observe that it does not depend on
e
the education level). The worker’s utility is uH ( e , w ) = w − if she is of high ability
2
and uL ( e , w ) = w − 3e if she is of low ability (also note that the education does not
change the ability of the worker and does not produce any satisfaction: it is only a cost).
The higher cost of education for low-ability workers can be understood as the greater
effort they exert in order to get the same degree as the high-ability worker (more years
of school, more hours of study or a higher opportunity cost). The salary w = 1 is the
maximum a firm is willing to pay a low-ability worker, whereas w = 5 is the maximum
to pay a high-ability worker, and w = 3 is the maximum to pay a worker that can be of
either type with equal probabilities.
To simplify the game, we will assume that the firm considers paying w = 1 or w = 3
if it observes that the worker has no education (i.e., the firm is discarding the case that
only high-ability workers decide not to get education) and w = 5 or w = 3 if the worker
is educated (it also discards the case in which only low-ability workers acquire educa-
tion). In other words, if the high-ability worker thinks education does not pay, so will
the low-ability worker with more reason. Workers and firms know all of this.
Figure 9.1a captures the extensive form of the game. H stands for the high-ability
type of worker (h = 5) and L for the low-ability worker (h = 1); F0 indicates that the
firm observes the education level e = 0 and F2 that it that observes e = 2. A means
“accept the contract,” while R means “reject the contract.”
Notice that nodes with different ability but equal education belong in the same
information set for the firm and are joined with a dotted line. Payoffs are calculated
according to the indications in the text describing the game. The first payoff is the
workers’, while the second is the firm’s. Let us see check the payoffs for a couple of cases.
1. The worker is of type H and chooses e = 2. The firm offers w = 3 to a worker with
that education level.
If the worker accepts: The worker has ability h = 5 and the firm is paying w = 3,
thus the value of the worker for the firm is B ( h = 5, w = 3 ) = 5 − 3 = 2. The worker’s
2
utility is uH ( e = 2, w = 3 ) = 3 − = 2 .
2
160
AGENCY PROBLEMS: SIGNALING AND MORAL HAZARD
If the worker rejects: The firm wins zero and the worker earns her reservation salary
2
minus what she has spent in education, uH ( e = 2, w = 3.5 ) = 3.5 − = 2.5.
2
2. The worker is of type L and chooses e = 0, and the firm offers the salary w = 3 to a
worker with that education level.
If the worker accepts: The worker has ability h = 1 and the firm pays her w = 3,
thus the firm’s profits are B ( h = 1, w = 3 ) = 1 − 3 = −2. The worker’s utility is
uB ( e = 0, w = 3 ) = 3 − 3 × 0 = 3.
If the worker rejects: The firm has zero profits and the worker earns her reserva-
tion salary minus the education costs, uB ( e = 0, w = 0 ) = 0 − 3 × 0 = 0.
N
h=5 h=1
1/2 1/2
H L
e=0 e=2 e=0 e=2
F0 F2
w=3
w=3 w=3 w=3
w=1 w=5
w=1 w=5
H H H H L L L L
A R A R A R A R A R A R A R A R
3 3.5 1 3.5 2 2.5 4 2.5 3 0 1 0 –3 –6 –1 –6

2 0 4 0 2 0 0 0 –2 0 0 0 –2 0 –4 0
Figure 9.1a Education as a signal
The rest of the payoffs are calculated similarly. The arrows indicate the acceptance
or rejection of the salary for the worker in the different circumstances and the equilib-
rium in eight subgames. After substituting the subgames for their equilibrium payoffs,
the game is reduced to the one illustrated in Figure 9.1b.
N
h=5 h=1
1/2 1/2
H L
e=0 e=2 e=0 e=2
F0 F2
w=3 w=3 w=3 w=3

w=5 w=1 w=5
w=1
3.5 3.5 2.5 4 3 1 –3 –1

0 0 0 0 –2 0 –2 –4
Figure 9.1b Solving the game of education as a signal
161
GAME THEORY
Drawing from the extensive form we can find the normal form in order to look for
Nash equilibria. The worker has four strategies: (0, 0), (0, 2), (2, 0) and (2, 2), where the
first component in each strategy indicates the education level chosen by the H type and
the second component indicates the level chosen by L. The firm also has four strategies:
(3, 3), (3, 5), (1, 3) and (1, 5), where the first component indicates the salary offered
to a worker with education level e = 0 and the second component indicates the salary
offered to a worker with education level e = 2. Figure 9.1c shows the normal form.
Again, let us calculate two examples of the payoffs to make sure we understand the
numbers in the table:
1. The worker chooses (2, 0) and the firm chooses (3, 3): this means that H chooses
e = 2 and L chooses e = 0.
The firm pays w = 3 regardless of the worker’s education level. With probability
½ the worker is H and her utility is 2.5 (see the third final node from the left in
Figure 9.1b), and with probability ½ she is L with utility 3 (fifth final node from
1 1
the left), which gives an expected utility of × 2.5 + × 3 = 2.75. The firm wins
1 1 2 2
× 0 + × ( −2 ) = −1.
2 2
2. The worker chooses (0, 2) and the firm chooses (1, 5): this means that H chooses
e = 0 and L chooses e = 2.
The firm pays w = 1 to a worker with education level e = 0 and w = 5 to a worker
with education level e = 2. With probability ½ the worker is of type H and her
utility is 3.5 (second final node), and with probability ½ she is L with utility −1
1 1
(last final node), with an expected utility 3.5 + × ( −1) = 1.25. The firm wins
1 1 2 2
× 0 + × ( −4 ) = −2.
2 2
The game can be simplified by eliminating weakly dominated strategies: (1, 5) is
weakly dominated by (1, 3) for the firm, while both (0, 2) and (2, 2) are dominated by
either (2, 0) or (0, 0) for the worker. However, it will not be necessary. In Figure 9.1c we
have underlined the best replies that show the two Nash equilibria in pure strategies:
((0, 0), (1, 3)) and ((2, 0), (1, 5)).
Firm
(3, 3) (3, 5) (1, 3) (1, 5)
(0, 0) 3.25, −1 3.25, −1 2.25, 0 2.25, 0
(0, 2) 0.25, −1 1.25, −2 0.25, −1 1.25, −2
Worker
(2, 0) 2.75, −1 3.5, −1 1.75, 0 2.5, 0
(2, 2) −0.25, −1 1.5, −2 −0.25, −1 1.5, −2
Figure 9.1c The normal form of the “education as a signal” game
In the first equilibrium both types of worker chose e = 0, and the firm offers salary
w = 1 to any worker showing education level e = 0 and w = 3 to a worker showing
e = 2. In order to see that this is in fact an equilibrium, one must notice that Type H
162
loses if she changes her education level. Its only use is to obtain a better salary, but the
firm is offering 3 at the most and her reservation salary is 3.5, so she gets no education
and leaves the market. Type L has a lower reservation salary. She would like to be paid
w = 3, but for that she needs to pay too high a cost in order to get the necessary edu-
cation. The cost of acquiring education level e = 2 is 3 × 2 = 6, which is much more
than the 3 that she will earn. To obtain w = 1, she does not require education. So, both
H and L are playing their best reply given the firm’s strategy. This way, the firm will
know that all workers requesting a job will be of type L and, therefore, will pay them
according to their productivity, w = 1. But what will happen if the firm offers a higher
salary to educated workers? Could the firm offer w = 5 to a worker with education
level e = 2 knowing that it will attract only H types? According to the game we have
specified, no. In order for that to work, it requires that the worker be able to change
her education choice after the firm deviates in this proposed way, and she is unable to
do that. In our game, the worker chooses first, and she cannot later react to a deviation
by the firm. Education is chosen before the salary offers are known. If the firm offers
a salary w = 5 to the educated worker, it will attract no one, because all workers have
already decided not to get education. Another way of interpreting this is that by offer-
ing a salary 3 to the educated worker it is implicitly assuming that this educated worker
may be of either type. Because the education is not chosen in equilibrium, the firm may
reasonably think that a worker only chooses it by mistake, and that this mistake can be
equally made by both types.
In the second equilibrium, the different types of workers choose different educa-
tion levels, and the firm will reward the education with a higher salary. In the end,
this means that although the firm cannot see the type directly, the firm pays workers
according to it, because the education reveals that type. The situation is an equilibrium
because the H type does not want to change and choose zero education, since that
would mean that her savings from not paying for education would not compensate for
the lower salary she would be receiving. In a similar vein, the L type does not want to
pretend to be an H type by acquiring education, for her education would be too costly
and would not compensate for the higher salary.
In the language of game theory, the equilibrium where all types choose the same
action is called a pooling equilibrium. If different types choose different actions, we talk
about a separating equilibrium. In the second equilibrium of our example, education
has allowed the firm to tell the types apart in a separating equilibrium. This happening
in real situations is the hypothesis of education as a signal.
9.2 The empirical evidence of education as a signal

One of the characteristics of the signaling use of education is that it implies a higher
investment in education relative to a situation in which the signaling plays no role.
To understand this, notice that, in our example, if ability were observable, no one
would get any level of education and the situation would be optimal. Thus, the strategy
to find empirical evidence for the signaling effect focuses on detecting overinvestments
in education and then checking to see whether they can be attributed to the signaling
effect. Alternatively, one expects that there will be a higher return to education in those
sectors where signaling has a higher potential relative to other sectors where productiv-
ity is observed more easily.
163
GAME THEORY
Using this research strategy, several works have found that, indeed, the hypoth-
esis of education as a signal explains some changes in education investment better
than alternative hypotheses. For instance, according to the signaling hypothesis, a legal
increase in the age of mandatory education implies that those not affected by the new
law – those that already received their education up to that age – also increase the
number of years in school. However, according to the human capital hypothesis, that
should not be the case (as researchers Lang and Kropp have shown). Between 1910 and
1970 in the United States, as well as in many other countries, the age of mandatory
education increased several times. Using data from this natural experiment, Lang and
Kropp (1986) show that there is evidence in favor of the signaling hypothesis.
Another economist, Bedard, studies a different natural experiment (Bedard, 2001).
Within the human capital model, local colleges in the United States provide afforda-
ble alternatives to higher education and, consequently, help to increase enrollment.
According to the signaling model, they may also result in an increase in high school
dropouts. If attending college becomes easier and the better students make this choice
in greater proportions, then the average skills of the population with only a high school
degree decrease; this way, the incentive of obtaining a high school diploma is also
reduced for those that do not have the perspective of attending college. This pro-
duces higher numbers of dropouts. In this line, Bedard studies the effects of a higher
presence of local colleges on the dropout rates for high school students and finds that
the theoretical prediction is supported by the data. In particular, the percentage of
dropouts is at least a 33% of the percentage of increase in college registration, some-
thing we should not observe with the alternative hypotheses. Other authors conduct
similar studies for the case of education of self-employed workers and for promotions
within the firm and find how decisions about education are explained by the signaling
hypothesis.
9.3 Signaling with a continuous variable and discrimination

in the labor market
Using the same model as in the beginning of this section, consider now that work-
ers can choose any education level and that the firm can choose any salary. Utility
and profit functions remain unchanged. Let us look for both pooling and separating
equilibria.
Pooling equilibrium: As in the example with discrete variable, if both types of
worker choose the same education e = 0 and the firm pays according to the average
productivity w = 3 with no regard for education, then that situation constitutes an
equilibrium. No worker wants to deviate to get some education, since that is costly and
will not affect the salary. Additionally, the firm will not offer a different salary because
education has already been decided upon and will not be affected. As in the discrete
case, we are implicitly assuming that if, for whatever reason, a worker comes to the
firm with a higher education level, she will be treated as being of either type with equal
probabilities as the proportions of types within the population.
Separating equilibrium: For an education level e to work as a signal, it must be such
that the H type prefers to acquire that level of education while the L type prefers not to,
so that the firm is willing to pay 5 to the workers showing that level and 1 to the ones
not showing it. Let us look at how to write these conditions.
164
Workers belonging to the H type: If the workers get educated at level e, they will
e
win uH ( e , w = 5 ) = 5 − ; if they do not get education at all, they will win 3.5 (their
2
reservation utility, as they will not accept working for a salary of 1). For them to prefer
e
the education, it must be that 5 − ≥ 3.5, or e ≤ 3.
2
Workers belonging to the L type: If the workers become educated at level e, they
will win uB ( e , w = 5 ) = 5 − 3e; if they do not get education at all, they will win 1. For
4
them not to prefer the education, it must be that 5 − 3e ≤ 1, or e ≥ = 1.33.
3
Any education level between 1.33 and 3 demanded by the firm in order to give the
salary 5 will work as a signal to separate the H and L types. The most efficient signal
is the cheapest one that can do the job, 1.33, but there is no guarantee that the equi-
librium signal will take that value. There is an equilibrium for every number between
the levels 1.33 and 3. As previously, we must consider what happens to the workers that
choose an education level different from zero and reach the required level e in order to
get the salary 5. A simple possibility (though not the only one) is that the firm treats
the workers that choose any education level below e as an L type, and all that choose an
education level equal to or greater than e are treated as an H type. Since education is
costly, all workers will choose the smallest value they need in order to be treated as their
true type. In the example with a discrete variable in Section 3.1 we chose the separating
value e = 2, but now we have seen that any other number between 1.33 and 3 could
have worked equally as well.
An application to discrimination in the labor market

Consider now that workers can be men or women. Either sex can be of two types, one of
high skill and the other of low skill. The firm can tell the sex of the worker, but not their
skill level. Each group, men and women, plays the game we previously saw with the firm,
and in each of the two games there will be an equilibrium level to separate the high- and
the low-skill types. As we saw above, this level is any arbitrary number between 1.33 and
3. In particular, this means that it is possible that in the men’s game the required level is,
for instance, e = 1.5, while the required level in the women’s game is e = 2.5. Of course,
the opposite may also occur; however, the historical inertia of assuming a lower skill for
women may have caused the present equilibrium to be worse for women than men. This
may be true even if at the present moment no one shows any prejudice against women.
No one is individually responsible for the equilibrium and no one can do anything to
change it even if they all agree that it should be changed. There are undoubtedly many
more causes for discrimination; this example just shows one additional possibility.
Some other examples of signaling

Advertising serves many purposes: it lets consumers know about your brand, it helps
consumers to differentiate your product from those of the competition, it may influ-
ence the consumers’ behavior and it can mark the quality of your product or brand.
Advertising is costly, and a company that does not have a good product or service
may find that incurring this cost may not be as profitable as for a firm that does offer
quality. In addition to the other uses of advertising, the firm may be saying: “I have so
165
GAME THEORY
much confidence that you will like and buy my product that I am willing to spend a
great deal of money for publicity.”
Ostensive expenditure by a firm shows that its financial situation is sound, so the
competition will think twice before beginning a price war and potential partners will
show a greater disposition towards becoming associates.
Some animal species have developed specific characteristics and traits that, at first
glance, look paradoxical. For instance, the tail of the male peacock is captivating and
beautiful, but it is also a burden to his mobility. A big, symmetrical tail in perfect
condition may reveal the good health and genetic material of the specimen – so good
that he can afford such an impediment. Similarly, some gazelles in a herd start jumping
conspicuously in the presence of a predator, thus attracting its attention. The paradox
can be explained if, through this signal, the predator is receiving the message that those
particular gazelles are formidable runners, so it would be a waste of time to try chasing
them. A weak, sick or injured gazelle will be unable to display such behavior.
According to the Biblical story, two women came to King Solomon claiming to be
the mother of the same newborn. Solomon decided to slice the child in two and give
one half to each mother. One accepted, but the second withdrew her claim over the
child and allowed the first mother to have the child intact and alive. Solomon, under-
standing that only the real mother would sacrifice herself this way in order to save her
child’s life, gave her the baby. This is a nice story, if it only would not have the false
mother behave so unintelligently. It would have sufficed for her to have also withdrawn
her claim in order to be in the same position as the real mother. One thing we have
learned in our signaling model is that for one type of player to not be willing to pre-
tend to be any other type, the signal that identifies that other type must be too costly
for the pretender. For instance, if Solomon had known that the true mother had some
information that the false mother did not, he could simply have asked about it in order
to identify each type of mother, since the false mother could not have answered (in our
model this means that the cost to give that signal is infinite for her). Any other signal
that is too costly for the false mother yet easy for the authentic one to provide could
also work. But the signal to which Solomon paid attention was of small cost to both
women – simply a small speech that either could utter with equal ease – and which
could not help to discriminate between the two mothers.
9.4 Moral hazard: Fixed payment or payment

by performance?
Up to this point we have examined examples of the agency problem in which the
agent’s characteristics are fixed and cannot be changed. This was true in the example
of the risk group of the client of the insurance company and also in the case of the
worker in the labor market. In the following examples, we will introduce cases in which
the characteristic of the agent changes with the effort. A careful owner of a property,
by exerting some effort, can reduce the risk of an accident ruining it. An individual
that quits smoking, eats healthy and exercises can make herself a low-risk type, and an
employee that works hard is a good type of worker for the firm. The agency problem
arises when the effort put in by the agent is not observed by the principal, who only
notices the results of the effort. If these results depend not only on the effort, but
also on other non-observed variables and on hazardous circumstances, we will have a
166
problem designing a proper contract, one which promotes effort and is profitable for
both parties. We will start with an example of an employer and an employee.
Fixed payment
Firm F hires an employee to perform a task. If the employee makes an effort, perfor-
mance will be “outstanding” with probability 2/3 and “normal” with probability 1/3.
If the employee does not put in any effort, performance will be “normal” with cer-
tainty. If performance is “outstanding,” the firm will gain 9, whereas if it is “normal,”
it will gain 6. The cost of effort for the employee is 1, who must decide whether to
accept the contract with the firm without knowing if performance will be “outstand-
ing” or “normal” in case he decides to exert some effort. Finally, the law enforces a
minimum salary of 1, and the Collective Labor Agreement prohibits any payment
based on results. The game is depicted in Figure 9.2a. First the firm (F) chooses a
salary to offer, which the employee (E) accepts or rejects (A or R). If he rejects the
offer, both get zero. If he accepts, he must then decide on his effort level (e = 0 means
no effort and e = 1 means to make an effort). Finally, Nature (N) determines perfor-
mance according to the probabilities described above. The payoff to the firm is the
result minus the salary it pays, while the payoff to the employee is the salary minus
the cost of the effort.
w≥ 1
E
A R
E
e=0 e=1 0
0
N N
p=0 p=1 p = 2/3 p = 1/3
9−w 6−w 9−w 6−w

w w w−1 w−1
Figure 9.2a Fixed payment
The employee earns w if he does not make an effort and w − 1 if he does. If the
salary is fixed, he will choose not to make the effort. Anticipating this, the firm will
choose the minimum possible salary, the legal minimum w = 1. The firm will make
6 − 1 = 5 and the employee will receive 1.
This situation illustrates the moral hazard problem. The employee would like to
commit to a higher effort in exchange for a higher salary (w ≥ 2, to compensate for the
higher effort). The firm would agree to such a deal if w ≤ 3. Keep in mind that the firm
2 1
wins ( 9 − w ) + ( 6 − w ) = 8 − w; for this expression to be larger than the 5 it expects
3 3
to win when the employee makes no effort and it pays 1, it must be that 8 − w ≥ 5 , or
167
GAME THEORY
w ≤ 3. In other words, a salary between 2 and 3 in exchange for a level of effort e = 1

leaves both the firm and the worker better off. The deal depends on whether the firm
can trust that the employee will keep his word to make the effort, but the firm cannot
be that naïve. It is aware that, once it hires the employee with a salary between 2 and 3
and has committed by contract to pay that salary, it would be in the employee’s interest
not to make any effort. Remember that the effort is not detected beforehand and the
contract cannot include a clause stating that payments will be made only if the effort
is made.
Payment by results
One way to solve the moral hazard problem is to pay based on results. Since the firm can
only observe the results, and not the effort, it can remunerate the employee by paying
him one salary if the result is normal, wN, and another salary if the result is outstanding
wO. To simplify the game, let us assume that the firm pays the minimum salary if the
performance is normal (wN = 1), so that the only problem is finding the equilibrium
salary in case performance is outstanding. Figure 9.2b shows the new game.
Wo
E
A R
E
e=0 e=1 0
0
N N
p=0 p=1 p = 2/3 p = 1/3
9 − Wo 6−1 9 − Wo 6−1
Wo 1 Wo− 1 1−1
Figure 9.2b Payment by results
This game can be solved using backwards induction, as there are no information sets
with more than one node. Let us begin by calculating the expected payoff for the emp
loyee after each level of effort. If he chooses e = 0 he will expect to win 0 × wO + 1 × 1 = 1,
2 1 2
whereas if he chooses e = 1 he will expect ( wO − 1) + (1 − 1) = ( wO − 1) . For him
3 3 3
2
to be willing to make the effort it must be that (wO − 1) ≥ 1, which means that
3
wO ≥ 2.5. Knowing this, if the firm wants the employee to make the effort, it will pay
168
the minimum salary that gives him the incentive to do so: wO = 2.5 . This way, the firm
2 1
can expect to win ( 9 − 2.5 ) + ( 6 − 1) = 6.
3 3
If the firm desists from giving the employee an incentive to make the effort, it will
pay him the minimum salary regardless of the result. This way the worker will not
make any effort: if he does, he wins 1 − 1 = 0, and if he does not, he will win just 1. In
this case, the firm wins 6 − 1 = 5 . It pays to motivate the high effort.
The subgame perfect Nash equilibrium (also, the backwards induction solution) is
this: the firm offers the contract wO = 2.5 (and wE = 1, not indicated in the extensive
form since this is a given, non-strategic choice) and the employee accepts any salary
(his reservation salary is zero). In addition, if the salary is wO ≥ 2.5 , he will make
the effort e = 1, otherwise he will choose e = 0 . As we have seen, the firm expects to
2 1
win 6, while the employee expects to earn 2.5 + 1 = 2. The firm obtains more than
3 3
the 5 it expected in the fixed payment case. The worker gets a higher salary, but his
utility is the same, since the salary only increases to compensate the higher effort. This
means that, in this example, the firm is able to appropriate the entire surplus created
by the effort. This will happen if the firm is the only one that can benefit from the
worker’s effort. If there are more firms, and with them more competition, the surplus
will be divided between the firm and the worker, and both will be strictly better off in
the situation with results-based payments.
Payment by results with a risk-averse employee

In our example, changing the fixed payment to payment based on results we have
been able to solve the moral hazard problem inherent to the relation between the
employer and the employee. However, results-based payments will not solve the prob-
lem completely, although they will alleviate it if the employee is risk averse. Let us
see why.
Say that the agent’s utility is not as simple as calculating the salary minus the effort,
but is instead given by the function ( w, e ) = w − e ; that, as we have seen on other
occasions, is an example of utility that represents preferences showing risk aversion. To
illustrate the new game, it is enough to substitute the worker’s payoffs with their square
roots in the final nodes of Figure 9.2b. We will now repeat the calculus from earlier,
but using the square roots.
Let us start by calculating the employee’s expected utility for each effort level. If he
chooses e = 0, he will expect utility 0 × wO + 1 × 1 = 1; if he chooses e = 1, he expects
2 1 2
wO − 1 + 1 − 1 = wO − 1 . For him to be willing to make the effort, it must be
3 3 3
2
that wO − 1 ≥ 1, which implies wO ≥ 3.25 .
3
Following this analysis, if the firm wants the agent to make an effort, the lowest salary
2 1
it must pay is wO = 3.25. This way, the firm expects to win ( 9 − 3.25 ) + ( 6 − 1) = 5.5 .
3 3
If the firm does not promote the effort, it will pay 1 and will win 6 − 1 = 5 . Thus, it is
better to pay in order to motivate effort.
169
GAME THEORY
Let us summarize the employee’s situation. He receives a salary of 3.25 if the result
is outstanding and 1 if it is normal, while he makes an effort that costs him 1. All this
2 1
gives him a utility 3.25 − 1 + 1 − 1 = 1. There is a cheaper way for the agent to
3 3
enjoy this same utility: give him a fixed salary 1 in exchange for him making the effort.
This salary must satisfy w − 1 = 1, or w = 2. If this is done, the firm will win 6, more
than in the previous situation. This means that if the firm gives the worker a salary
slightly higher than 1, then both will be better off. Again, the problem is that if the
worker receives a fixed salary, then he will not have any incentive to make the effort,
causing a moral hazard problem for this case.
With a fixed salary, the firm wins 5. With a salary based on results, it wins 5.5. The
best possible situation, when it detects the effort and contracts it to be e = 1, gives it 6.
The employee’s utility is 1 in all cases. The payment by results alleviates the problem
of inefficiency caused by moral hazard, but does not solve it completely. It was easier to
analyze the efficiency when keeping the agent’s utility fixed to 1, so that the different
levels of efficiency are simply the different profits for the principal. We could have done
the opposite by fixing the profits of the firm (e.g., if there are many firms that perfectly
compete for the worker) and measuring the levels of efficiency as levels of the worker’s
utility; we would have obtained the same result, namely, that the moral hazard problem
could not be solved completely.
9.5 Moral hazard: Co-payment, yes or no?

Let us go back to the insurance model. In Section 8.5 we considered the problem
of adverse selection in this sector, where the risk group was given and did not
depend on the individual. We will change this now and consider that the risk the
individual faces can be altered by her actions. Needless to say, the most realistic
cases exist between these two extremes, but recall that we show best the analysis
and applications of game theory if we begin with the simplest cases. In addition,
if we can isolate the different questions, we will be in a better position to under-
stand more complex problems such as how to organize a health insurance sector
in a society.
In our example, we have an individual in an initial situation valued in 100 (e.g., her
wealth or her health indicators) that she can lose with probability p = 0.75 unless she
takes care of herself, in which case the probability of losing everything goes down to p
= 0.25. Taking care costs e = 2, while being careless costs nothing (e = 0). The individ-
ual is risk averse and has a utility function given by u ( x ) = x − e . We will pose the
question of what is the best policy that an insurance company can offer her. The reader
may think that the company is one among many in the market and is pressured by
fierce competition to attract clients, that this is a mutual insurance company where the
insured are the owners, or that it is a public insurance system. In any case, the company
must cover its costs which, to simplify the analysis, we will assume to be negligible.
This way, all the revenues coming from contracting insurance policies will be used
to pay the indemnities. This means that the price of the policy must be equal to the
expected loss, that is the probability of the loss times the amount lost. If the individual
is careless, the policy that offers indemnity I will cost 0.75 × I : the company will cash
170
0.75 × I when it sells the policy, and it will expect to pay this same amount, making the
expected profit zero.
The game between the company and the individual is similar to the one seen in Sec-
tion 9.4 between the firm and the worker, and it is represented in Figure 9.3. First the
insurance company (Co) chooses a policy, which means choosing to offer an indemnity
I and a price given by p × I , where p is the probability of a loss. Then, the individual or
client (C) must decide whether to accept or reject contracting the policy. After that, the
consumer has to again decide between being cautious or careless. Finally, Nature (N)
determines whether there is a loss or not.
To understand the moral hazard problem let us first see which are the best policies
in two cases: (i) if the company anticipates that the individual will be careless and (ii)
if the company anticipates that the individual will put effort e = 2 in order to be cau-
tious. In the first case, the policy will be (100, 75), and in the second (100, 25). The
first number is the indemnity while the second is the price. For a risk-averse individual,
total indemnity is the best situation, since she is guaranteed to enjoy the same utility
level in all events, loss or no loss.
Co
(I, pI)
C
A R
C
e=0 e=2 0
5.5
N N
1 3
p= p=
3 4 1
4 p= p=
4 4
pI − (1− p)I pI −(1−p)I
√ 100− pI √(1− p) I √100 − pI − 2 √(1− p)I−2
Figure 9.3 Co-payment
Before solving the game, note the following.

1. In the absence of insurance, the consumer prefers to be cautious:
1 3
u ( no insurance ( R ) , careless ( e = 0 ) ) = 100 + 0 = 2.5 ,
4 4
3 1
u ( no insurance ( R ) , careful ( e = 2 ) ) = 100 + 0 − 2 = 5.5 .
4 4
2. If the insurance company knows whether the consumer will be cautious or not,
it will offer the policy (100, 75) for the careless individual and (100, 25) for the
171
GAME THEORY
cautious one, so the consumer will prefer to be cautious. This is also better than the
situation with no insurance:
1 3
(
u accept (100, 75 ) , careless ( e = 0 ) = ) 4
100 − 75 +
4
100 − 75 = 5,
3 1
u ( no insurance ( R ) , careful ( e = 2 ) ) = 100 + 0 − 2 = 5.5,
4 4
3 3
(
u accept (100, 25 ) , careful ( e = 2 ) = ) 4
100 − 25 +
4
100 − 25 − 2 = 6.66 .
The company will break even in any of the three situations.

3. If the company cannot know whether the consumer will be cautious or not, and if
it offers the policy (100, 75), the consumer will not accept it because she will prefer
to be cautious and not get the insurance:
1 3
(
100 − 75 +
4
100 − 75 = 5 ,
3 1
u ( no insurance ( R ) , careful ( e = 2 ) ) = 100 + 0 − 2 = 5.5.
4 4
4. If the company cannot know whether the consumer will be cautious or not, and if it
offers the policy (100, 25), the consumer will accept it, but will be careless and the
company will therefore lose money:
3 1
u ( accept (100,25), careful (e =2)) = 100 − 25 + 100 − 25 − 2 = 6.66,
4 4
3 1
(
100 − 25 +
4
100 − 25 = 8.66,
1 3
(
Profits accept (100, 25 ) , careless ( e = 0 ) =
4
)
25 + ( 25 − 100 ) = −75 .
4
From the considerations above, we can see that the company cannot offer full indemnity
if it wants the individual to be cautious. The only other possibility is that it offers a
policy with partial indemnity. If this policy really promotes an effort to be cautious, the
price will be computed multiplying the indemnity times the probability ¼. To work as
1
desired, the policy ( I , p = 4 I ) that offers partial indemnity must satisfy the condition
  1   1 1 3 1
u  accept  I , I  , careful ( e = 2 )  = I− I + 100 − I − 2 ≥
  4   4 4 4 4
 1  3 1 1 1
u  accept (I , I ), careless  = I− I + 100 − I .
 4  4 4 4 4
172
The expression defines a second-degree inequation. Solving it we find, approximately,

that any indemnity I ≤ 40 will make the individual prefer to exert effort to be careful.
1
The best policy is I = 40 with the price 10 = 40 . The reader can verify that
4
( ) ( )
u accept ( 40,10 ) , careful ( e = 2 ) = 6.48 ≥ u accept ( 40,10 ) , careless ( e = 0 ) = 6.47 .
This utility is also higher than the utility of being uninsured. All of this means that
there is a way of offering partial insurance (in the example, a 40% indemnity) that
the individual will want to buy and with which she will be willing to be cautious.
The company understands that and treats the buyer of the policy as a cautious client.
The problem is that the policy does not fully insure the individual. Like before, she
would like to commit to being cautious in exchange for full insurance, and the com-
pany would like that to be true. In the absence of a guarantee, however, the most it can
offer is the partial insurance.
Partial indemnity may come under different specifications, like (i) a limit to the
covered expenses (in our example, 40), (ii) a coverage of a percentage of the loss (the
40% of our example), or (iii) a co-payment (in the example, the individual pays the first
60 units and the company pays the rest). In the example, all these interpretations are
adequate because there is only one scenario in which the indemnity is claimed (total
loss). In general, there may be more complicated cases with partial losses and these
three ways of offering partial indemnity will not be equivalent.
Incomplete indemnity in insurance policies

In the insurance sector, and particularly in health insurance, the normal situation is to
not have complete coverage. Even if the company reimburses the value of a lost car in
case of an accident, it will not compensate for all the associated inconveniences after the
accident. In the best-case scenario, the health insurance will compensate the patient for
the hospitalization costs to regain health. However, it is unlikely that she will be just as
well after having lived through her medical condition compared to her situation before
it. All costs associated with time and effort are hardly ever fully compensated, and there
is also always a chance that health recovery is not full. Whether these non-covered
costs are enough to discourage being careless or abusing the system, or whether they
are insufficient so the client must pay an even higher cost in order to find an attractive
policy, are matters that must be studied case by case. The convenience of adding or
removing a given co-payment is something that cannot be resolved aprioristically. The
relevant piece of information needed to solve the problem is whether the co-payment
reduces abuse and allows for a better use of the always-scarce sanitary resources. For
example, it could be that the co-payment means that the unnecessary visits to the ER
or the family doctor are reduced without consequence of medical neglect, allowing the
system to dedicate more resources to other services. On the other hand, it could be
that the co-payment causes some people to not seek the medical attention they need, a
harm that does not compensate for the reduction of abuse of the system. To appeal to
ideological consigns in order to make a decision one way or the other will not solve the
problem. However, studying the effects will get us closer to better management of the
resources, and understanding the moral hazard implications will improve the study.
173
GAME THEORY
9.6 Moral hazard: Work in teams and cooperatives

Two persons, A and B, form a team to complete a task that will result in some profits
given by the expression P = 3 ( e A + e B ), where e A is the effort exerted by A and e B
the effort exerted by B. Effort can take values 1 (low) and 2 (high). Once the task
is done, they will share equal parts of the profits. The utility of the players depends
on their share of the profits and on the effort they make, according to the functions
P P
uA = − 2e A and uB = − 2e B . What are the equilibrium levels of effort? Figure 9.4a
2 2
shows the normal form of this simultaneous game.
B
eB = 1 eB = 2
eA = 1 1, 1 2.5, 0.5
A
eA = 2 0.5, 2.5 2, 2
Figure 9.4a Working in teams
This is a prisoners’ dilemma game, where the players maximize profits if they both
exert a high level of effort, but each player also prefers a “free ride” on the effort of
the other player. This is also a moral hazard problem. Each player would like to trust
and be trusted in the agreement in order to make the high effort, but lacking any
commitment device, there would be no cooperation in equilibrium. The only Nash
equilibrium is ( e A = 1, e B = 1) .
In earlier chapters, we have seen different ways of getting out of the logic of the
prisoners’ dilemma. In this section, we will take a look at a new one. Consider that the
two players form a cooperative, and to become a member each one must pay a capital
contribution of 2. Once the cooperative is formed, they must decide on the effort level as
previously, except in this case the sharing rule includes not only the profits from the task
but also the initial contributions to the cooperative, according to the following rules:
• If profits add up to 12 or more, they share equally the profits plus the initial
contributions.
• If profits add up to less than 12, they share only the profits.
With this new sharing rule, the game becomes the one in Figure 9.4b.
B
eB = 1 eB = 2
eA = 1 1, 1 2.5, 0.5
A
eA = 2 0.5, 2.5 4, 4
Figure 9.4b Working in a cooperative
In the new game there are two Nash equilibria, as marked in the figure:
(e = 1, e B = 1) and ( e A = 2, e B = 2 ). The coordination in the second equilibrium, with
A
higher payoffs, should not be difficult. The biggest problem with this kind of solution
174
is the credibility of the rule of not sharing the initial contributions. After all, it is in the
players’ benefit to share them. This time inconsistency can be fixed if the members of
the cooperative contract the use of the contributions with a third party. Exercise 9.9
shows that this third party can be inside the cooperative itself.

9.1 Consider the signaling game in Figure 9.1a, and change the reservation salary
of the H type from 3.5 to 1.5.
(a) Describe the extensive form of the game.
(b) Find the corresponding normal form game.
(c) Find the Nash equilibria in pure strategies and discuss them.
9.2 Consider the signaling game in Figure 9.1a, and change the utility of the H type
worker from uH ( e , w ) = w − e to uH ( e , w ) = w − 2e .
2
(b) Find the corresponding normal form game.
(c) Find the Nash equilibria in pure strategies and discuss them.
9.3 Consider the signaling game with a continuous variable and change the reser-
vation salary of the H type from 3.5 to 1.5.
(a) Find, if possible, an equilibrium in which neither type of worker chooses to
get an education.
(b) Find, if possible, an equilibrium in which the H type gets education and the
L type does not.
9.4 Consider the signaling game with continuous variable and change the utility of
e
the H type worker from uH ( e , w ) = w − to uH ( e , w ) = w − 2e .
2
(a) Find, if possible, an equilibrium in which neither type of worker chooses to
get an education.
(b) Find, if possible, an equilibrium in which the H type gets education and the
L type does not.
9.5 Consider the signaling game represented in Figure 9.2b and change the cost for
the worker when she makes the effort from e = 1 to e = 2.8.
(a) What salary does the worker need in order to make the effort?
(b) Is this salary profitable for the firm?
(c) Describe the subgame perfect Nash equilibrium of this new game.
9.6 Consider the signaling game represented in Figure 9.2b and change the firm’s
incentives. Because of pressure of competition from other firms, it must offer
the best possible contract to the worker.
(a) What salary will the firm offer if it is not interested in the high level of effort
from the worker?
(b) What salary will it offer to promote the high effort?
(c) Which one of the two contracts is more profitable for the worker? Which is
more profitable for the firm?
175
GAME THEORY
9.7 Consider the game between the insurance company and the consumer in Sec-
tion 9.5. Find the precise indemnity (up to two decimal places) as well as the
price for the best insurance policy for the consumer that satisfies that (i) the
company breaks even and (ii) the company cannot observe whether the con-
sumer is cautious or careless.
9.8 Repeat the game between the insurance company and the consumer in Sec-
tion 9.5 for the case in which the probability of accident in case of being cau-
tious is zero.
9.9 Consider the game in Section 9.6, but with a four-player team, A, B, C and D,
instead of two. Players are paired into two groups, one with A and B, and the
second with C and D. Each group must complete a task independently of the
other group, with a sharing rule and effort costs identical to the ones that lead
to the payoffs in Figure 9.4a. Understanding the moral hazard problem, the
four players agree to contribute 2 prior to performing the task and to share the
contributions according to the following rules:
• If the profits in one group add up to less than 12, then they share only the
profits. The contributions of that group are then available to be shared by the
members of the other group.
• If the profits in one group add up to 12 or more, then they share the profits
and their members’ contributions, in addition to those of the opposite group
if the profits of that group do not add up to 12.
(a) Find the Nash equilibria of the new game.
(b) Discuss the time consistency of the sharing rule.
9.10 Two women come to King Solomon (S), both claiming to be the mother of the
same child. One of them is the truthful mother (T), while the other one is the false
mother (F). Each woman can insist upon being the mother (I) or refuse the claim (R)
and they must make the decision simultaneously. If both insist or refuse, Solomon
will slice the child in two and give one half to each woman. If one insists and the
other one refuses, Solomon has to decide whom to give the child without knowing
who the true mother is. The true mother has a utility of 2 if she receives the child,
1 if the other woman gets the child, and zero if the child is killed. The false mother
gets 2 if she gets the child, 1 if the child is cut in half, and zero if the other woman
gets him instead. Solomon gets 2 if he gives the child to the true mother, zero if he
gives him to the false mother, and 1 if he kills the child.
(a) Show the extensive form of the game. Hint: Start with one of the women,
then the other, and then Solomon, who has to make a decision only if one
mother chooses I while the other chooses R. Since he does not know who
the true mother is, those nodes belong in the same information set.
(b) Show that there are no Nash equilibria in pure strategies. Hint: Start by con-
sidering the possibility that Solomon chooses to give the child to the mother
who chose I, and show that this leads to no equilibrium. Then repeat for the
case in which he gives the child to the woman who chose R.
(c) Show that if both T and F choose I with probability 1/2 and S chooses to
give the child to the woman who chose I with probability 1/2, then the
strategy choices are a subgame perfect Nash equilibrium.
176
10 SEVEN APPLICATIONS OF
GAME THEORY
In this chapter we present several case studies dealing with different kinds of wars:
open conflict, cold war, preventive attacks, trade wars and even a diplomatic conflict.
Special attention is given to the role of information. In some cases, a player wants
to have private information like in the case in Section 10.1, in some others he wants
the opponent to have some information (Section 10.2). More complicated situations
arise in Section 10.3 where using private, advantageous information means revealing
it. In Section 10.4 we explore the need of bluffing, and in Section 10.5 we discuss
the use of imperfect information obtained by espionage. Trade wars are examined in
Section 10.6. Finally, in Section 10.7 we relate the historical events of the negotiations
between Greece and the Troika in the light of game theory.
10.1 The battle of the Bismarck Sea

In February of 1943 during World War II, the forces under the command of Admi-
ral Imamura received the order of reinforcing the Japanese positions of Papua New
Guinea from their base, three sailing days away. The plan was discovered by the U.S.
forces, and General Kenney received his own orders to intercept the convoy and cause
the greatest possible damage to the Japanese in a joint mission of the American and
Australian air forces. The convoy could follow either the northern or the southern
route around the island of New Britain. The Allies could also concentrate their recon-
naissance flights within one route or the other. If the route chosen by General Kenney
did not coincide with the Japanese route, the bombing time of the convoy would have
been reduced by one day, the time necessary to concentrate the attack on the other
route.
Under these circumstances, this game is similar to “matching pennies,” in which
one player prefers that the decisions coincide, while the second one prefers just the
opposite. Figure 10.1a shows the game, where payoffs are the number of days on which
the Japanese fleet is under attack and where the only Nash equilibrium is that both
players choose their strategies randomly with a 50–50 probability. The expected pay-
offs in the equilibrium are equivalent to a bombing during two and a half days.
Imamura
North South
North 3, −3 2, −2
Kenney
South 2, −2 3, −3
Figure 10.1a The battle of the Bismarck Sea
177
GAME THEORY
The weather reports add a new element in the game with the forecast of rain and
poor visibility in the northern route, but clear skies in the south. Poor visibility would
reduce the days of bombing by one if the Japanese were to use the northern route.
The new game is shown in Figure 10.1b.
Now, for the Japanese, choosing the southern route is a weakly dominated strategy.
Once eliminated, Kenney’s decision is to use the northern route also. The only Nash
equilibrium in pure strategies is (North, North), which can be reached by iteratively
eliminating weakly dominated strategies. This is in fact just what they did, and the
battle took place along the northern route, in the Bismarck Sea.
Imamura
North South
North 2, −2 2, −2
Kenney
South 1, −1 3, −3
Figure 10.1b The battle of the Bismarck Sea after the weather report
There are more Nash equilibria in mixed strategies, where the Japanese choose at
random between the two routes, while the Americans use the northern route. In fact,
all strategy combinations ((North), (North with probability p, South with probability
1 − p, with 1 ≤ p ≤ 1)) are Nash equilibria. Nevertheless, in all of them, the result is
2
two days of bombing.
The analysis of this battle using game theory was written and published by O. G.
Haywood, a U.S. Army officer and engineer in charge of designing plans for different
scenarios in a nuclear war (Haywood, 1954).
10.2 The nuclear war

War as a zero-sum game
Let us first present a war in which the army that attacks first has an advantage and wins
easily without large fighting costs. The victorious army obtains territory and resources at
the expense of the loser. The two armies have similar forces; if they both attack in an effort
to be first, it will result in a stalemate and the attacks will not prosper. We can interpret
this attack as a threat or display of military equipment in an arms race in the case that the
other army does the same, and we can interpret it as a real attack and victory if the other
army is not prepared. This is a zero-sum game where “Attack” is a dominant strategy, and
the result is not particularly lethal to the armies. The game is represented in Figure 10.2a.
Army B
Attack No
Attack 0, 0 1, −1
Army A
No −1, 1 0, 0
Figure 10.2a War as a zero-sum game
178
SEVEN APPLICATIONS OF GAME THEORY
The arms race is not zero sum

Now, Army A has nuclear weapons that cost 0.5 but which allow an easy victory in case
they decide to attack, regardless of the decision made by the other army. Army B does not
have the technology to develop nuclear weapons and can only threaten by means of a con-
ventional arms race. The new game is displayed in Figure 10.2b. The arms race (Attack,
Attack) is still an equilibrium. The scenario (Attack, No) is too, even if Army B is using
a weakly dominated strategy. If, by having attacked, Army B can alleviate its own defeat,
then (Attack, No) will no longer be an equilibrium. On the other hand, if by attacking
the defeat is instead aggravated, then (Attack, No) will be the only equilibrium.
B
Attack No
Attack 0,5, −1 0,5, −1
A
No −1, 1 0, 0
Figure 10.2b Army A develops nuclear weapons
If Army B can develop a nuclear arsenal as well, then the situation will be the follow-
ing. Now, in the case that they both escalate, then neither will win and the cost of the
arsenal will be wasted. The game has therefore become a prisoners’ dilemma. “Attack”
is a dominant strategy, and the equilibrium is (Attack, Attack); however, the payoffs
are now worse than those in the equilibrium in the first situation in Figure 10.2a, since
developing a nuclear arsenal is very expensive. This is shown in Figure 10.2c.
B
Attack No
Attack −0,5, −0,5 0,5, −1
A
No −1, 0,5 0, 0
Figure 10.2c Arms race
The arms race continues. Consider now that “Attack” means to make actual use
of the nuclear weapons. The result in the case that only one army attacks is better for
that army, since it prevents the other army from counterattacking. Now the situation
becomes a chicken game, represented in Figure 10.2d, with two equilibria in pure strat-
egies. In each equilibrium, one army attacks while the other does not. Having come to
this point, the loss in the case of a total nuclear war is big, and so is the loss in the case
of one army being attacked by surprise. The reader can write even larger numbers in
those cases, but the equilibria would not change.
B
Attack No
Attack −2, −2 1, −1,5
A
No −1,5, 1 0, 0
Figure 10.2d Using the nuclear weapons first
179
GAME THEORY
The doomsday machine

In Section 1.4 we saw different ways in which players could tip the equilibrium in
the chicken game in their favor. One of them was to appear as irrational, as a player
that will attack without thinking what the other party will do. This may be an urban
legend, and we may never truly know, but during the Cold War there were rumors
that Nixon wanted the CIA to spread information about him being much too fond of
drinking and that the KGB exaggerated medical reports indicating that Brezhnev was
developing senile dementia.
There is another way of obtaining an advantage in the chicken game: eliminate
the “No attack” strategy or at least make the strategy “Attack” an automatic reaction
in case the other army attacks, even if this is worse than surrendering once the attack
takes place. The logic is that, if such an automatism is possible, that moment will never
come, as the other army will never attack. A system that triggers an attack after receiv-
ing an attack, and which will cause catastrophic consequences to both sides, is called a
doomsday machine. If one side is able to build this machine, and this fact is known to
the opposing side, the game will be like the one displayed in Figure 10.2e, where Army
2 has built the machine and the only equilibrium is (No, Attack).
B
Attack No
Attack −2, −2 −2, −2
A
No −1, 1 0, 0
Figure 10.2e The doomsday machine
Mutually assured destruction

If both sides have a doomsday machine, the result is that the strategies (No, No) con-
stitute a Nash equilibrium. The strategy combination (Attack, Attack) is also a Nash
equilibrium, but players are using weakly dominated strategies, so not attacking is a
more robust equilibrium. This version of the nuclear game is known as mutually assured
destruction or MAD. It is shown in Figure 10.2f.
B
Attack No
Attack −2, −2 −2, −2
A
No −2, −2 0, 0
Figure 10.2f Mutually assured destruction
The MAD doctrine can also be seen in a dynamic version of the game, where one
side is more aggressive while the other plays more defensively. For instance, if Army A
chooses between “Attack” and “No” and Army B chooses between “Counterattack”
and “Surrender,” the game is as in Figure 10.2g, where the arrows indicate the only
equilibrium (both subgame perfect Nash and backwards induction) in which Army A
has an advantage.
180
No Attack
0 Counterattack Surrender
0
–2 2
–2 –1,5
Figure 10.2g Dynamic mutually assured destruction
Afraid of this being the game in the Cold War, Army B develops the doomsday
machine with the consequence that “Surrender” is no longer an option, since “Counter-
attack” is triggered automatically in the case that Army A attacks. The game has now
changed as shown in Figure 10.2h. Now, in equilibrium, Army B builds the machine
and Army A refrains from attacking. Army B builds the machine because otherwise
the equilibrium continuation would be that Army A attacks and Army B surrenders (as
seen in Figure 10.2g), which is much worse for Army B. Army A attacks if the machine
is not built; however, if the machine is built, Army A does not attack. The complete
subgame perfect Nash equilibrium is written as ((Machine, Counterattack, Surrender),
(No, Attack)), where the first part corresponds to Army A’s strategy and the second part
to Army B’s. The equilibrium is marked with arrows.
Machine No machine
A A
No Attack No Attack
B B
0
0
0
Counterattack 0 Counterattack Surrender
–2 –2 –2
–2 –2 –1,5
Figure 10.2h Building the doomsday machine
If we review how the equilibrium has changed from the game in Figure 10.2a, we
will observe that the presence of MAD has achieved the goal of peaceful coexistence.
Of course, this is a simplified analysis. To begin with, in the real world the game is not
played just once, but rather it is a continuously ongoing situation that can be modeled
as a repeated game. This means that if the confrontation has the form of a chicken
game or of prisoners’ dilemma as in Figures 10.2c and 10.2d, the possibility of an
equilibrium in which no one attacks first using the trigger strategies seen in Chapter 7
still exists even in the absence of a doomsday machine. Thus, the real change occurs
181
GAME THEORY
when the interpretation of the Cold War changes from being a zero-sum game – or
almost a zero-sum, where attacking first is the only equilibrium – to a non-zero-sum
game (chicken game, prisoners’ dilemma, MAD). All in all, the MAD policy makes
the non-aggressive equilibrium more likely and robust, since it is a weakly dominant
strategy of the baseline game. One problem that MAD does not solve is the possibility
that a mistake can result in an attack that triggers the catastrophe.
History
In his book Prisoner’s Dilemma: John von Neumann, Game Theory, and the Puzzle of the
Bomb, William Poundstone discusses these strategic arguments during the Cold War
(Poundstone, 1993). The mathematician John von Neumann is a key element of the story
because of his scientific contributions to the Manhattan Project and as a member of the
committee in charge of the strategic decisions relating to the nuclear arsenal. Addition-
ally, he and economist Oskar Morgenstern we the founders of game theory as a discipline
after their book The Theory of Games and Economic Behavior, in which they developed the
non-cooperative game theory for zero-sum games (as well as the cooperative game the-
ory). According to Poundstone, von Neumann’s position was the perception of the Cold
War as a zero-sum game, where the best strategy is to attack first. The development of
the Soviet nuclear arsenal, as well as the arms race, made him change his mind, becoming
one of the proponents of the mutually assured destruction policy. It seems that the term
MAD is due to him and his particular sense of humor in forming acronyms.
The historical doomsday machine to support the MAD policy consisted of under-
ground nuclear missile silos, strategic bombers in constant flight and strategic subma-
rines, all with nuclear weapons on the American side. They built a network with the
capacity to counterattack and with its own decentralized decision-making authority in
case of a nuclear conflict. The Soviets created their own strategic system.
Fiction
Whenever the MAD doctrine during the Cold War is discussed, it is customary to also
make a reference to Dr. Strangelove, a film by Stanley Kubrick, based on the strategies
in nuclear war. In the film, an American general obtains the codes for launching a
nuclear attack with strategic bombers. The president is surprised that the codes can be
used without his permission and learns that the system was designed that way to provide
someone with decision-making authority in the event that all of the higher command
are deceased. The Americans try to convince the Soviets that the attack was an unin-
tentional mistake and, after heated discussions, the Soviet ambassador reveals that they
have built a doomsday machine. This time it is the scientific advisor, Dr. Strangelove,
who is surprised that they have built the machine without informing about it since all of
its strategic value revolves upon the enemy’s knowledge of its existence. The ambassador
responds that the disclosure was due in a few days, coinciding with the birthday of the
Soviet president, who is fond of surprises. After assaulting the headquarters where the
rebel general was entrenched, they manage to obtain the codes for calling the bombers
back, all save one, who cannot receive the order. The film is based on the book Red Alert
by Peter George (George, 1958). According to Thomas Schelling, winner of the Nobel
Prize in Economics for his contributions to game theory, conversations he had about
Red Alert with the author and Kubrick led the famous director to make the film.
182
10.3 You cannot use information without revealing it

During World War II the Allies suffered important naval loses due to German U-boats
attacking their convoys. If the Allies could have known which convoy was going to be
attacked, they could have defended it efficiently. One way to obtain this information
would be decoding the encrypted communication system used by the Germans. How-
ever, once the Germans realized their system has been deciphered, they would change
it. Let us explore the strategic aspects of this game, starting with a simple situation in
which the Allies have two convoys, one twice as valuable as the other. The Germans
can attack only one, and the Allies can defend only one. The Allies send convoys like
that every week. The game is displayed in Figure 10.3.
Germans
Attack Convoy 1 Attack Convoy 2
Defend Convoy 1 2, −2 −1, 1
Allies
Defend Convoy 2 −2, 2 1, −1
Figure 10.3 U-boat war
The Nash equilibrium of this game is ((1/2, 1/2), (1/3, 2/3)), where the Allies protect
Convoy 1, the most valuable, with probability 1/2 and defend Convoy 2 with prob-
ability 1/2. The Germans attack Convoy 1 with probability 1/3 and Convoy 2 with
probability 2/3. The probabilities of the different outcomes are (1/6, 2/6, 1/6, 2/6), and
they are displayed in Figure 10.4.
Germans
1 2
Attack 1 with p = Attack 2 with p =
3 3
1 1/6 2/6
Defend 1 with p =
Allies 2
1 1/6 2/6
Defend 2 with p = 2
Figure 10.4 Frequencies in the Nash equilibrium of the U-boat war
Thus, everyone expects convoys to be successfully attacked half of the time. Among
them, 1/3 correspond to Convoy 1 and 2/3 to Convoy 2. Now, imagine that the Allies
are able to know in advance which convoy is going to be attacked. If the Allies defend
the correct convoy every time they have the information, the Germans will notice that
very soon. According to the probabilities before, after six weeks the most likely out-
come is that three convoys were attacked successfully. Two or four successful attacks
are also quite possible. However, the Germans would notice if none of the attacks were
successful. How likely is it that there would be no successful attacks? The probability
6
of six unsuccessful attacks in a row is  1  = 1 = 0.0156, or 1.56%. The Germans
 2  64
183
GAME THEORY
should suspect something is going on and change the codes. Even four failed consecu-
4
1 1
tive attacks are very unlikely, with a probability of   = = 0.0625, or 6.25%. Say
 2  16
that the Germans analyze the attacks every four weeks and change their encrypting
system if they suspect it has been hacked, and that they suspect they have been hacked
if the number of failed attacks drops below 10%. Given the numbers above, they will
change the codes after four consecutive weeks of failed attacks. This way the Allies
avoided only two extra attacks (they avoided four, but they expected to avoid two in the
equilibrium without espionage).
The Allies should allow some attack to be successful if they plan to send more
convoys. For instance, only one successful attack in four weeks is something to be
4
1 4
expected with probability 4 ×   = = 0.25, or 25% (successful and failed attacks
 2  16
both occur with probability ½ every week, there are four weeks, and the failed attack
may be any of the four). The Germans may attribute that to bad luck. If the Allies allow
one attack during the first four weeks and none during the next four, in their analysis
after eight weeks the Germans will notice something strange is happening. Four weeks
with only one success may be bad luck (something to be expected with probability ¼,
as we saw), but eight with only one hit has a much smaller chance.1 Under this new plan
by the Allies, the Germans change their system after eight weeks. During this time the
Allies defended seven convoys when they expected to defend only four in the equilib-
rium. Thus, they saved three convoys – more than with the plan before.
The only way in which the Germans never realize that they are being spied upon is
if the Allies keep the probability of a successful attack at ½, which is the same as not
using the information at all. Anything that departs from a ½ chance of success will
be detected in the long run. The detection will be made sooner if the probability of a
successful attack is higher to begin with. For instance, if there are more convoys and
still only one can be defended every time.
10.4 You should bluff from time to time

Thirty-Six Strategies, a collection of Chinese militaristic tactics, describes the empty
fort strategy in a chapter dedicated to desperate stratagems (Peters, 2013):
When the enemy is superior in numbers and your situation is such that you
expect to be overrun at any moment, then drop all pretence of military prepar-
edness and act calmly so that the enemy will think you are setting an ambush.
This stratagem has to be used sparingly and only after one has first developed a
reputation for military prowess. This also depends on having a clever opponent
who, in perceiving a trap, may over-think his reaction.
This tactic has been allegedly used a few times, although it seems we only receive a
retelling from the successful cases. It describes a bluff that can lead to a victory when
1 Actually, to compute the likelihood of few successful attacks the Germans should calculate the
probability of one or less successful attacks, but for our exposition we keep it simple.
184
one has a weak position. It works if the bluff is not called. So the important question
is whether it pays to bluff. Is it a bold strategy that only a reckless leader will follow?
Does the risk to be annihilated compensate the benefits? Under which circumstances?
Let us study first how bluffing works in the game famous for its bluffs, poker, and
let us do it in a very simplified way. Players have received their hands and it is Player
1’s turn to make a bet or pass. If she bets and Player 2 calls the bet, the loser pays 5 to
the winner. If Player 1 passes, she loses 1 to Player 2; if Player 2 folds after a bet, he
loses 1 to Player 1. Say that Player 1 may have a good (G) or a bad (B) hand with equal
probabilities. If she has a good hand and her bet is called, she wins for sure; if she has a
bad hand, she loses for sure. We could relax these assumptions, but the strategic aspects
would be very similar. The game is depicted in Figure 10.5.
1/2 1/2
G B
1 1
Pass Pass
Bet Bet
–1 –1
1 1
2
Call Fold Call Fold
5 1 –5 1
–5 –1 5 –1
Figure 10.5 The poker game
First Nature (N) gives a good (G) or a bad (B) hand to Player 1, who can Bet or
Pass. If she passes the game is over and she pays 1 to Player 2. If she bets, Player 2 must
call or fold, not knowing the hand of Player 1. There are no subgames, so the set of
subgame perfect Nash equilibria coincide with the set of Nash equilibria, that can be
found in the normal form. Player 1 has four strategies: (Bet, Bet), (Bet, Pass), (Pass,
Bet) and (Pass, Pass). Player 2 has only two strategies – Call, Fold – that come into play
only after Player 1 bets. The normal form is presented in Figure 10.6a.
Player 2
Call Fold
(Bet, Bet) 0, 0 1, −1
(Bet, Pass) 2, −2 0, 0
Player 1
(Pass, Bet) −3, 3 0, 0
(Pass, Pass) −1, 1 −1, 1
Figure 10.6a The normal form of the poker game
To understand the payoffs, consider, for instance, the cell in which Player 1 bets
with a good hand but passes with a bad one (Bet, Pass), and Player 2 calls if Player 1
1 1
bets (this means that Player 2 chooses Call). In this case, Player 1 wins 5 + ( −1) = 2
2 2
185
GAME THEORY
and, being a zero-sum game, Player 2 loses 2. The payoffs in the other cells are com-
puted similarly and are left as an exercise. The first thing to observe is that strategies
(Pass, Bet) and (Pass, Pass) are both dominated by (Bet, Bet). In other words, Player 1
never passes when she has a good hand. After deleting those strategies, we have a sim-
pler game, shown in Figure 10.6b.
Player 2
Call Fold
(Bet, Bet) 0, 0 1, −1
Player 1
(Bet, Pass) 2, −2 0, 0
Figure 10.6b Solving the poker game
There are no Nash equilibria in pure strategies. In mixed strategies we find the
equilibrium: ((2/3[Bet, Bet], 1/3[Bet, Pass]), (1/3[Call], 2/3[Fold])), which the reader
can compute in Exercise 10.5. This means that Player 1 always bets with a good hand
and bets 2/3 of the times she has a bad hand (when she is bluffing). Player 2 calls the
bet 1/3 of the times he has a chance. The expected payoffs in the Nash equilibrium are
2/3 for Player 1 and -2/3 for Player 2. Notice that if Player 1 never bluffs and plays (Bet,
Pass), the best reply for Player 2 is Fold and the payoff for Player 1 will be 0. Thus, it
pays to bluff from time to time.
Going back to the empty fort strategy, consider the same game as in Figure 10.5,
where now the game is not poker, but war, and where G means to have a good army and
B to have a small one. Both Bet and Call mean Attack, and both Pass and Fold mean
Retreat. Now Army 1 uses the empty fort strategy (attack with a small army) 2/3 of the
time. If we change the probabilities of having a good army to be 9/10, the new normal
form game after deleting dominated strategies (Retreat, Attack) and (Attack, Attack) is
shown in Figure 10.6c (the reader can check the payoffs).
Now the only equilibrium is ((Attack, Attack), Retreat), which indeed indicates that
the empty fort strategy works better when your enemy knows that you usually have a
large army. However, as we saw with the previous probabilities, the bluff should also be
used if the probabilities are not so high.
Army 2
Attack Retreat
(Attack, Attack) 4, −4 1, −1
Army 1
(Attack, Retreat) 4.4, −4.4 0.8, −0.8
Figure 10.6c The empty fort game
The interested reader can consult the work by Cotton and Liu (2011), who wrote a
more detailed study of the empty fort strategy used in the Chinese military legends of
Li Guang and his 100 horsemen and Zhuge Liang and the empty city.
186
10.5 There may not be weapons of mass destruction: Should

we still attack?
Our next case is about the decision to attack a country that is suspected to be develop-
ing weapons of mass destruction. Examples in recent history include Operation Opera,
the surprise Israeli attack to destroy a nuclear reactor in Iran (1981) and the second
Gulf War against Iraq (2003), when the United States accused the Hussein regime of
possessing weapons of mass destruction, mostly chemical. We will not discuss the eth-
ical or legal implications of the attacks, only the strategic issues presented by one coun-
try having the capability to build the weapons and another country, feeling threatened
by the potential peril, having the capability of launching a preemptive attack. In addi-
tion, we will add the possibility of allowing inspections and of having an intelligence
service that can report on the issue. Here we present the analysis by Jelmov, Tauman
and Zeckhauser (2017) using one of their examples.
The baseline case

Country 1 has to decide whether to build a bomb (or any kind of highly destructive
weapon system). Country 2 has to decide whether to attack Country 1’s installations to
prevent the development and manufacturing of the bomb. Each country must decide
without the knowledge of the other’s plans.
There are four possible strategy combinations: (Bomb, Attack), (Bomb, No attack),
(No bomb, Attack) and (No bomb, No attack). The natural preferences of the coun-
tries are shown in decreasing order in Table 10.1. Recall that these are preferences about
the hypothetical outcomes and that nothing is said about the actual decisions, as they
do not know what the other country is deciding. Figure 10.7 shows the game.
Table 10.1 Countries’ preferences
Country 1 Country 2
Bomb, No attack No bomb, No attack
No bomb, No attack Bomb, Attack
No bomb, Attack No bomb, Attack
Bomb, Attack Bomb, No attack
Country 2
No attack Attack
No bomb 0.15, 1 0.1, 0.2
Country 1
Bomb 1, 0 0, 0.7
Figure 10.7 Preemptive war
In a simple exercise left to the reader, one can easily check that there is no NE in
7 8
pure strategies, but that there is one in mixed strategies:  [No bomb] + [Bomb], 
 15 15 
 10 85  , with payoffs: u = 10 ≈ 0.105 , 7
 [No attack] + [Attack] 1 u2 = ≈ 0.467.
 95 95  95 15
187
GAME THEORY
Observe that, in the equilibrium, Country 2 is attacked with high probability, even if it
has no weapons, and that Country 1 decides to go ahead with the building of the bomb
with probability 8 ≈ 0.53.
15
Country 1 may allow inspections
Consider now that if Country 1 decides not to build the bomb (NB), it can also choose
between opening its facilities and allow Country 2 to inspect them to verify its peace-
ful intentions (O) and not opening its facilities (NO). Opening the facilities does not
make sense if the country decides to build the bomb (B), as it would be attacked imme-
diately. The game is shown in Figure 10.8a, where A stands for “Attack” and NA for
“No attack.” If Country 2 is not allowed to inspect the installations, it does not know
whether Country 1 decided to build the weapon.
1
NB
1 B
O NO
2 2
NA A NA A NA A
0.15 0.1 0.15 0.1 1 0

1 0.2 1 0.2 0 0.7
Figure 10.8a Inspection game
We can simplify the game considering that the two actions NB and O are just one,
NBO, and NB and NO are also one, NB. Also notice that Country 2 will choose NA in
case there is no bomb and it has been allowed to inspect. This is done in Figure 10.8b.
1
NBO B
NB
2
0.15 NA A NA A
1
0.15 0.1 1 0
1 0.2 0 0.7
Figure 10.8b Simplified inspection game
The corresponding normal form is depicted in Figure 10.9a.
Country 2
NA A
NBO 0.15, 1 0.15, 1
Country 1 NB 0.15, 1 0.1, 0.2
B 1, 0 0, 0.7
Figure 10.9a The normal form of the simplified inspection game
188
There is an NE in pure strategies in this game: ((NBO), A). It means that Country 1
does not build the bomb and allows inspectors from Country 2. On its part, Country 2
attacks if it is not allowed to inspect. Utilities are (0.15, 1), higher than in the case with
no inspections.
Country 1 suffers a cost from being inspected

The government of Country 1 may lose its political influence if it welcomes foreign
inspectors. If this cost is not too high, the situation is similar to the case we just ana-
lyzed, with payoffs that can still be better than the first case without inspections. Let us
see what happens if the cost of allowing the inspections is very high, for instance equal
to 0.2. Figure 10.9b shows the new game.
Country 2
NA A
Country 1 NBO −0.05, 1 −0.05, 1
NB 0.15, 1 0.1, 0.2
B 1, 0 0, 0.7
Figure 10.9b The inspection game when being inspected is too costly
Notice that NBO is now strictly dominated by either NB or B. After eliminat-

ing it, the game is exactly as the one in Figure 10.7, with an NE in mixed strate-
7 8 10 95
gies,  [No bomb]+ [Bomb], [No attack] +

[Attack] and with utilities
 15 15 105 105 
u1 ≈ 0.095, u2 ≈ 0.467. In other words, with a high cost of allowing inspections, Coun-
try 1 may be better off risking the attack rather than allowing foreign inspectors.
Country 2 does not know the type of Country 1

If Country 2 knows what kind of government Country 1 has, the equilibrium is as
follows. If Country 1 regards inspections as not too costly, it will not build the bomb, it
will allow inspections and Country 2 will not attack. If the cost of inspections is high,
Country 2 will not allow for inspection, and the random strategy choice examined
before will follow for both countries.
In this complication, consider that Country 2 does not know for sure the type of
Country 1. However, it does know that if Country 1 is of the weak type (for whom the
cost of inspection is low) it will allow the inspection. Thus, if Country 1 does not allow
inspection it can safely be assumed to be of the strong type (with high inspection costs).
Therefore, the equilibrium is the same as before.
Country 2 does not know the type of Country 1 but it has intelligence
services
Now Country 2 has an Intelligence System (IS) that can get information about Coun-
try 1’s plans. However, the intelligence is not 100% reliable and reports the true status
189
GAME THEORY
of Country 1 with probability p = 0.8. This means, for example, that if Country 1 is
building the bomb, the IS will report the message “They are building the bomb” with
probability p = 0.8, and the message “They are not building the bomb” with probabil-
ity 1 − p = 0.2. The case is similar if Country 1 is not building the bomb. Country 2
has an a priori belief (before IS reports) that Country 1 is of the type that has high costs
of being inspected with probability 0.9.
By introducing the IS, Country 2’s strategies can be conditioned on the message,
giving four possibilities for the case when Country 1 does not allow inspections:
• Do not attack regardless of the message (NA, NA).
• Do not attack if the message says “No bomb,” attack if it says “Bomb” (NA, A).
• Attack if the message says “No bomb,” do not attack if the message says “Bomb” (A,
NA). (This is a very unlikely strategy, but still possible.)
• Attack regardless of the message (A, A).
At this point the game becomes quite complicated, but we can find some shortcuts to
simplify the analysis. In the case without IS, we observed that if Country 1 has no costs
of being inspected, it will allow inspections. We can start here and ask whether in the
new situation with IS there is an equilibrium in which this is still the case. In other
words, we will first assume that if Country 1 is of the weak type, it will allow inspec-
tions and consider the game between Country 2 and the type of Country 1 with high
costs of inspection. After we do this analysis we have to check that, given Country 2’s
equilibrium strategy, the type of Country 1 with no inspection costs is not better off by
denying the entrance to inspectors.
The extensive form is as follows. First Nature selects the type of Country 1. With
probability 0.1 it is the type that suffers no costs from being inspected (denote it by
1.N) and then automatically chooses NBO, ending the game. With probability 0.9
Country 1 suffers high inspection costs, 1.H, and it has to choose among the three pos-
sibilities we have considered previously. Actually, the game starts now, after node 1.H.
We have included the branches before that for completeness. If 1.H chooses NBO the
game ends. If it chooses NB, then the IS will report “No bomb” (n) with probability
0.8, and “Bomb” (b) with probability 0.2. However, if 1.H chooses B, the report will
say “Bomb” (b) with probability 0.8, and “No bomb” (n) with probability 0.2. Country
2 has two information sets. In the first one, 2.n, it receives report “n,” and in the second
one it receives “b.” Thus, Country 2 does not observe directly whether Country 1 has
built the bomb, it only has the information from the report. However it cannot know
whether that information is true or false since the IS is not 100% reliable. The infor-
mation sets are shown with discontinuous lines. In each information set Country 2
can play either NA or A. Remember that choosing NA, for instance, means choosing
NA in the two nodes in the information set. The extensive form game is as shown in
Figure 10.10.
The two information sets are just saying that Country 2 can condition its action on
the report, and thus has the four strategies we described before. Once we have detected
the three strategies for Country 1.H and the four strategies for Country 2, we can con-
struct the normal form, which is shown in Figure 10.11a.
To understand the utilities in the game, consider the cell (NB, (NA, A)). After the play
NB, the message is “n” with probability 0.8, in which case Country 2 plays NA (in 2.n)
190
N
P(N)=0.1 P(H)=0.9
1.H
1.N
NBO B
NBO NB
IS IS
0.15 −0.05 P(n)=0.8 P(b)=0.2 P(n)=0.2 P(b)=0.8
1 1
2.n 2.b
NA A NA A
NA A NA A
0.15 0.1 0.15 0.1 1 0 1 0

1 0.2 1 0.2 0 0.7 0 0.7
Figure 10.10 Preemptive war with intelligence
Country 2
NA, NA NA, A A, NA A, A
NBO −0.05, 1 −0.05, 1 −0.05, 1 −0.05, 1
Country 1.H NB 0.15, 1 0.14, 0.84 0.11, 0.18 0.1, 0.2
B 1, 0 0.2, 0.56 0.8, 0.14 0, 0.7
Figure 10.11a The normal form of the preemptive war with intelligence services
giving a payoff 0.15 to Country 1.H. With probability 0.2 the message is “b” and 2.b
plays A, giving Country 1.H a utility of 0.1. Thus, the expected utility for Country
1.H after (NB, (NA, A)) is 0.8 × 0.15 + 0.2 × 0.1 = 0.14 . The rest of the utilities are
computed similarly. The reader is asked to do that in Exercise 10.8. Notice that NBO is
dominated for Country 1.H and that, once we eliminate this strategy, (A, NA) is dom-
inated for Country 2. The game after these eliminations is depicted in Figure 10.11b.
Country 2
NA, NA NA, A A, A
NB 0.15, 1 0.14, 0.84 0.1, 0.2
Country 1.H
B 1, 0 0.2, 0.56 0, 0.7
Figure 10.11b Solving the preemptive war with intelligence services
The NE in mixed strategies in this game is ( 0.18[ NB] + 0.82 [B], 0.625[ NA,A ] + )
( 0.375[ A,A ]), which gives utilities u 1.H
≈ 0.125, u1 ≈ 0.59 (see Exercise 10.9). Let us not
forget about Country 1 if it has no inspection costs. It does not build the bomb and
allows inspection with a utility of u1.N ≈ 0.135. It remains to show that 1.N does not
want to deviate. Given that Country 2 is playing ( 0.625 [ NA,A ] + 0.375 [ A,A ]) , Coun-
try 1.N’s utility if it plays NB or if it plays B is 0.125, lower than its utility of 0.135 if it
does not deviate. This completes the analysis.
191
GAME THEORY
We found one equilibrium. To show that there is no other equilibrium takes a little
longer, and we are not doing it here. The main result of this case is that even if Country
2 receives negative reports about the weapons of mass destruction, it will still attack
Country 1 with probability 0.375.
10.6 Is free trade a prisoners’ dilemma?

Some authors write about free trade as an example of a prisoners’ dilemma: Countries A
and B practice a protectionist policy and each win 6. If they both open their borders to
free trade, they will each win 8. However, if only one opens the borders while the other
continues being protectionist, the latter will win 9 and the former will win 4. The game
is depicted in Figure 10.12a. In this game, we can repeat all of the analysis we have con-
ducted with the prisoners’ dilemma throughout the book, where countries must sign
a binding agreement to trust the other country keeping its word or, alternatively, they
can play a trigger strategy in order to sustain cooperation in a repeated game.
Country B
Free trade Protectionism
Free trade 8, 8 4, 9
Country A
Protectionism 9, 4 6, 6
Figure 10.12a Free trade vs. protectionism as a prisoners’ dilemma
Both history and economic theory, however, indicate that the free trade option
among countries is another kind of game, except in very particular circumstances.
Regardless of the strategy taken by other countries, being open to trade with the rest of
the world is generally better. This increases the possibilities of commerce and of special-
ization. Gains would be even higher if other countries did the same, but even if they do
not, it is better not to harm oneself by refusing to trade as openly as possible. In other
words, every time a country restricts its trade opportunities, it causes itself harm and
harms the rest. If the other countries restrict trade as well, then the situation becomes
even worse for all of them. The purpose of this section is not to review the theories and
empirical facts about free trade and protectionism, but to present the strategic impli-
cations under the different assumptions. Figure 10.12b illustrates this case (notice the
differences with the game in Figure 10.12a). Now free trade is a dominant strategy and
if governments internalize its benefits, then free trade will prevail without the need of
signing any treaty or of using trigger strategies.
Country B
Country A
Figure 10.12b Free trade vs. protectionism as seen by economists
192
The term protectionism is misleading, but is still being used for historical reasons.
Both protectionism and free trade were competing ideologies and for a time there was
no clear understanding of their consequences. Nevertheless, the model of the situation
as a prisoners’ dilemma may not be erroneous if we study the game more closely. Even
if an open border policy is beneficial to a country as a whole, different sectors may
benefit disproportionally and others may lose in the process (e.g., the public sector may
see a reduction of revenues that come from import tariffs). If those sectors influence
the government, whose future may depend on their support, and if the situation is sim-
ilar in the other country, then those governments may find themselves in a prisoners’
dilemma. In this case, for the governments to internalize the gains from free trade, the
other countries also must become open to trade so that the additional gains from open
trade will compensate the initial losses. If these gains are still not sufficient, then that
country will remain protectionist.
Trade wars
If the government of Country B has protectionism as a dominant strategy, it will harm
not only its own population as a whole but that of Country A also. The governors of
Country A, who want free trade, can then try and punish Country B by imposing trade
restrictions as well, even if they are fully aware that they will be harming their own
country in the process. Figure 10.12c shows this new scenario. Observe that Country
A’s payoffs are like those in Figure 10.12b, while Country B’s payoffs are like the ones
in Figure 10.12a.
Country B
Country A
Figure 10.12c Trade wars
The only equilibrium occurs when they both use their dominant strategy (Free
trade, Protectionism). Things become interesting in the infinitely repeated game,
where one can find a strategy in which Country A disciplines Country B into accept-
ing free trade. It will be a more complicated strategy than the ones seen thus far, but
we can explore it.
The usual trigger strategy will not function to sustain (Free trade, Free trade)
because the punishment (Protectionism, Protectionism) endlessly will not work. This
is because it is not a subgame perfect Nash equilibrium and because the Nash equi-
librium (Free trade, Protectionism), whose repetition will be a subgame perfect Nash
equilibrium, will not punish Country B in case of a deviation from a free trade agree-
ment. To define a strategy that works, let us define several phases. Phase C, coopera-
tion, consists of playing (Free trade, Free trade), while phase P, punishment, consists of
playing (Protectionism, Protectionism). The strategy is defined as follows:
• Begin in phase C and continue in this phase as long as Country B does not deviate.
• If Country B deviates in phase C, move on to phase P.
193
GAME THEORY
• Phase P will last for one period as long as no one deviates in this phase.
• If a country deviates in phase P, then phase P starts again.
To check whether this strategy is indeed a subgame perfect Nash equilibrium, let us
confirm that no country wants to deviate in either phase.
• Country A does not have any incentive to deviate in phase C, since it is already play-
ing a dominant strategy and a deviation has no future consequences.
• Country B does not have any incentive to deviate in phase C either: it will win 9
instead of 8, but this gain of one is not compensated since the strategy will move on
to phase P, where it will win 6 instead of 8 during one period, so long as the country
is not excessively impatient.
• Country A does not have incentives to deviate in phase P: it will win 7 rather than
6 during one period, but it will then win 6 rather than 8 in the next period. Again,
this does not compensate unless the country is too impatient.
• Country B will not deviate in phase P since it is playing a dominant strategy, and the
deviation will not affect future payoffs.
The game has an additional twist if the government in Country B finds that its pro-
tectionist option is dominant, but this is due to political pressure rather than its own
conviction about the effects of free trade. In this case, it can give in to the temptation of
being obligated by the other country to choose free trade before even trying to convince
or negotiate with the pressure groups inside its own country who do not want that trade
strategy. This way, it blames the external conditions and minimizes inside accusations. In
addition, Country B may showcase the internal forces against free trade in order to show
that the option is too costly for them. It can then try to obtain additional advantages in a
free trade agreement, arguing that they will be necessary to appease internal opposition.
Alternatively, by allowing a trade war (or negotiations with the implicit threat of a
trade war) to be the method of achieving free trade, the government of Country B will be
losing leadership in the project of opening up the country and may irritate Country A.
These two circumstances may reduce its negotiation power. Whether internal oppo-
sition will serve to improve or to worsen the bargaining position of the country will
depend on circumstances such as the relative size of the country. If Country A is much
bigger than B, or is a set of countries, its position will be much stronger and it will not
give in easily to countries like B, which do not show a collaborative attitude. In these
cases, exercising a strong internal leadership in favor of free trade may be a better strat-
egy, unless the internal pressure groups have a very disproportionate level of strength
compared to other sectors in the country that will benefit from a free trade agreement.
If the opposite is true, and Country B is much bigger than A, then showcasing internal
pressure against free trade may be a good way of convincing Country A to accept fewer
advantages in exchange of trading with a big country.
10.7 Negotiations between Greece and the Troika 

When the Great Recession began in 2008, special circumstances in Greece meant that
its debt crisis was far deeper than that in other countries. Without any possible financ-
ing resources in the international markets, Greece was forced to seek help from its
194
European partners, represented by the Troika (European Commission, European Cen-

tral Bank and IMF). Help came in the form of two bailout loans with severe conditions
imposed by the Troika. Bailout negotiations are seen often as being similar to a game
of chicken. Each party has the options “concede” and “hold.” If both parties concede,
they reach an agreement that is halfway between the preferred option of each party. If
they both hold, then no agreement is reached. If one concedes while the other party
holds, the non-conceding side will impose its conditions. In the case of the Greek bail-
out negotiations, the Troika’s conditions for financing the Greek government with low
interest loans, exemptions, deductions, etc., included imposing structural reforms in
the labor market, pensions system, taxes, privatizations and the like, as well as auster-
ity measures for reducing public expenditure. The Greek government demanded even
better conditions on the loans (more money and better interests and deductions), less
demanding reforms and more laxity for public expending. In this analysis, we are not
concerned with whose measures were better for the Greek economy; instead, we accept
that each party thinks their proposal is better. If no agreement is made, Greece risks
going bankrupt and exiting the European monetary system, something no one wants.
The game can be described as the classic game of chicken shown in Figure 10.13a.
Troika
Concede Hold
Concede 2, 2 1, 3
Greece
Hold 3, 1 0, 0
Figure 10.13a Greece and the Troika
The situation, however, is not symmetrical, since one can safely assume that Greece
has the most to win or lose. Changing the payoffs to reflect this will not change the
fact that there are two equilibria in the game, (Concede, Hold) and (Hold, Concede),
as seen in Figure 10.13b.
Troika
Concede Hold
Concede 3, 2 1, 3
Greece
Hold 4, 1 −10, 0
Figure 10.13b Greece and the Troika with more realistic payoffs
One can argue that with the new payoffs the “Hold” position is very risky for
Greece, since bankruptcy and its exit from the euro will hurt them more than it will
hurt the Troika. This does not imply that (Hold, Concede) is not an equilibrium, but
it is a factor for concluding that it is an unlikely equilibrium. Another question to
take into account is that, for the Troika, the cost of conceding may be higher than we
anticipated – its reputation will be damaged if it has to negotiate bailouts with other
countries, such as Ireland, Spain or Italy (which was a real possibility at the time).
Greece does not have these reputation costs. In fact, if we add this cost, the Troika’s
195
GAME THEORY
payoffs will be lower than the ones in Figure 10.13a or 10.13b. If they are low enough
as in Figure 10.13c, the only equilibrium is (Concede, Hold), through which the Troika
gets its way.
Troika
Concede Hold
Concede 3, 1 1, 3
Greece
Hold 4, −1 −10, 0
Figure 10.13c Greece and the Troika with reputation costs
It did not come as a surprise then that the Troika imposed its conditions in the first
two bailouts. However, let us maintain the payoffs as in Figure 10.13b. Time goes by
and Syriza is brought into power after the 2014 elections as a result of popular discon-
tent with the austerity measures. This is a radical left-wing party that, in coalition with
a nationalist, rightist party, achieved a majority in parliament as well as the election
of its candidate as prime minister. These two parties were newcomers to the Greek
political scene, which was dominated up until that moment by ND (New Democracy,
the conservative party) and PASOK (the socialist party). The new government was very
much against the conditions imposed by the Troika and, supported by the electors,
proceeded to change the conditions for the third bailout. However, their victory in the
elections did not change the fact that their options were still “Concede” and “Hold,”
or that its payoffs were the same as before. Speeches by the members of the new gov-
ernment stressed how popular opinion tied their hands and how the government itself
would suffer a higher cost should they concede. In short, they wanted to show that
the new game was like the one in Figure 10.13d, in which “Concede” ceases to be an
option since it has been eliminated, or like the one in Figure 4.11e, in which the pay-
off following “Concede” is much worse (if not for Greece, at least for the government
making the decision). In both cases, the equilibrium is (Hold, Concede), favorable to
the Greek position.
Troika
Concede Hold
Greece Hold 4, 1 −10, 0
Figure 10.13d Greece commits not to concede
The game in Figure 10.13d is similar to the game of Odysseus and the Sirens
following the elimination of one of the strategies by Odysseus in Section 4.4. Despite
the rhetoric of the Greek government, the option “Concede” did not disappear, and
no one really believed that the game had changed in this way. What’s more interesting
is the argument that the new government had much more to lose; however, in Fig-
ure 10.13e we see that in order to avoid the equilibrium in which Greece concedes, the
payoff to the government in the (Concede, Hold) box has to be lower than in (Hold,
Hold), which is very unlikely. Again, no one believed that this was the game they were
playing.
196
Troika
Concede Hold
Concede 1, 2 −11, 3
Greece
Hold 4, 1 −10, 0
Figure 10.13e Payoffs that would make Greece’s commitment credible
The Greek government insisted on its position – that the strategy “Hold” was not
an option or that its payoffs in the (Concede, Hold) cell were lower than in the (Hold,
Hold) cell. For this purpose, after some months of failing to reach an agreement, and
three days before the deadline of June 30, 2015, the government called for a referen-
dum to take place on the following 5th of July, and the voters rejected the Troika’s
conditions. However, several factors put an end to all of the government’s chances for
success.
First, the option “Concede” was still a possibility and the referendum did not
change that. Second, perhaps it was true that the referendum’s outcome made it more
difficult for the government to concede. Taking an action against the explicit popular
vote, would imply paying a very high cost by the prime minister. Would it be enough
to be in the game in Figure 10.13e? Well, here is what happened immediately after
the results of the referendum were announced: almost all of the opposition parties,
who until this moment were opposed to Syriza’s approach to conducting negotiations,
declared that they would support Prime Minister Tsipras regardless of his decision.
Many thought that this statement by the opposition parties made Tsipras’ position
for choosing “Hold” even stronger, when in fact, their support meant that the costs
of conceding would be reduced (the opposition would not call for a campaign against
Tsipras in the case that he decided to concede). Any possibility of being in the game of
Figure 10.13e was reduced following this move by the opposition. The game went back
to being the one in Figure 10.13b.
Third, after calling for a referendum, but before it took place, PM Tsipras did
something foolish: he sent a proposal to the Troika that included almost all of the
conditions that he had previously rejected and asked for very small concessions
in exchange. The Troika refused to continue making new negotiations until after
the referendum, but this action by Tsipras already showed his hand and demon-
strated that he did not believe that his government had much chance in those
negotiations.
Finally, there were still other conditioning elements. After the referendum,
there was no time to negotiate (in fact, it took place after the deadline). As we
saw in Chapter 6, rushing negotiations is worse for a comparatively more impa-
tient player, as was the case with the Greek government which was in dire need of
money.
The results of the negotiations for the third Greek bailout are already history.
The Greek government accepted all of the conditions imposed by the Troika and
they were all ratified by the Greek Parliament three days later, with the support
of three opposition parties and with only some members of Syriza voting against
them.
197
GAME THEORY

10.1 In the battle of the Bismarck Sea, Kenney considers a third possibility: sending
half of the planes to the northern route and the other half to the south. Sending
half of the planes implies causing half the damage.
(a) Describe the game in normal form for each case, with no difference in vis-
ibility in the two routes, and with the difference explained in Section 10.1.
(b) Find the Nash equilibria in both cases.
10.2 In the battle of the Bismarck Sea consider that the northern route takes 3.5
days.
(a) Describe the game in normal form for each case, with no difference in vis-
ibility in the two routes, and with the difference explained in Section 10.1.
(b) Find the Nash equilibria in both cases.
10.3 Countries A and B, both with nuclear weapons, face a situation that could
easily escalate into a serious conflict. Country A can ignore Country B, with
zero payoffs for both, or respond to the small attacks. If Country A responds,
Country B can retire with payoffs (1, −1) or escalate and face the MAD game in
Figure 10.2g (subtracting 0.5 in the payoffs of the country that does not attack).
(b) Find the subgame perfect Nash equilibria.
10.4 Countries A and B, both with nuclear weapons, face a situation that could
easily escalate into a serious conflict. Country B has installed a missile base in
a territory near Country A. After this, the game is as follows: Country A can
establish a naval blockade against that territory or order an air attack. The air
raid gives payoffs (−1, 1), since Country B will attract the sympathy of the inter-
national community following a disproportionate reaction by Country A. After
the blockade, Country B may respond by dismantling the base or keeping it.
The first option leads to payoffs (0, 0). After the second option, Country A still
has the option of doing nothing, with payoffs (−1, 1), or ordering the air attack,
without fear of an adverse international reaction since it comes after Country B
did not withdraw after the blockade. More specifically, if Country B dismantles
the base after the air attack, the payoffs are (0, −1.5), and if it insists on keeping
the base, they will face the MAD game in Figure 10.2g.
10.5 Check the payoffs of the games in Figures 10.6a and 10.6c and also check that
((2/3[Bet, Bet], 1/3[Bet, Pass]), (1/3[Call], 2/3[Fold])) is indeed the Nash equi-
librium in the game depicted in Figure 10.6b.
10.6 Consider the game in Figure 10.5, and change the probability of having a big
army (a good hand) to a very low number (for instance, 1/50).
(a) Which is the new equilibrium?
(b) How often does Player 1 bluff?
(c) Which is the most likely outcome?
198
10.7 Find the Nash equilibria of the game in Figure 10.7. Find also the utilities in the
equilibrium.
10.8 Find the payoffs for the normal form in Figure 10.11a.
10.9 Find the Nash equilibria of the game in Figure 10.11b. Find also the utilities in
the equilibrium. To that end answer the following questions.
(a) Show there are no equilibria in pure strategies.
Notice that there are four possibilities for Country 2 to randomize. It can do it
between (NA, NA) and (NA, A), between (NA, NA) and (A, A), between (NA,
A) and (A, A) and among the three of them.
(b) Now consider that only (NA, NA) and (NA, A) are used by Country 2 and
compute a Nash equilibrium in mixed strategies as if the other strategy (A,
A) did not exist.
(c) Find Country 2’s utility in the equilibrium in (b).
(d) Compute now the utility of Country 2 if it plays (A, A) and Country 1.H
uses its equilibrium strategy in (b), and compare it with its utility in the
equilibrium in (b). If it is higher, then the equilibrium in (b) is not an equi-
librium and (A, A) is not erased. If it is lower, it is an equilibrium.
(e) Repeat (b), (c) and (d) if only (NA, NA) and (A, A) are used.
(f) Repeat (b), (c) and (d) if only (NA, A) and (A, A) are used.
(g) Show that Country 2 will never be indifferent between its three strategies.
It is enough to check that the conditions for indifference in (b) are different
for different pairs of strategies.
10.10 Extend the games in Figures 10.12a and 10.12b for three countries.
10.11 Show what strategies can sustain free trade in the infinite repetition of the
games with three countries in Exercise 10.10.
10.12 Consider the game between Greece and the Troika as depicted in Figure 10.13b.
Let us see a way of making sense of the idea that the equilibrium most favorable
to the Troika is also the most likely.
(a) Calculate the Nash equilibrium in mixed strategies. Denote by p the proba-
bility that Greece chooses “Concede,” and q the probability that the Troika
chooses “Concede.”
(b) Confirm that for every probability smaller than the q in the mixed Nash
equilibrium found in (a) the best reply for Greece is “Concede.”
(c) Confirm that for every probability smaller than the p in the mixed Nash
equilibrium found in (a) the best reply for the Troika is “Concede.”
(d) What is the relative size of the intervals for p and q that make “Concede” the
best reply for the other party?
10.13 In light of the results in Exercise 10.12, change the payoffs for the Troika in
the game in Figure 10.13b so that the strategy “Concede” by the Troika is
a best reply to more values of q (the probability with which Greece chooses
“Concede”) than the strategy “Concede” by Greece is to values of p.
199
11 SEVEN MORE APPLICATIONS
This chapter contains seven more applications of game theory. Here we will explore a
variety of topics. There are examples appealing to the different social sciences and to
philosophy; in some respects they can be seen as a continuation of Chapter 3 rather
than Chapter 11. In any case, the reader is asked to pay special attention to the way
a social interaction, and in the case of Pascal’s wager, a human-god relation, can be
translated into a game.
11.1 The minority language

Iriberri and Uriarte (2012) study the interactions among speakers of a minority
language. In this section, we show some of their conclusions through one example.
Imagine a society with a majority language, X, known to all, and a minority language,
Y, known only to a portion of the population. The interaction that these authors ana-
lyze is of which language to use when a conversation between two strangers begins.
If the bilingual individuals do not have preference for a particular language, we can
easily conclude that they will always use language X, with the consequence that the
minority language will never be used when addressing a stranger, even if they both
belong to the minority who use that language, at least in a first conversation. This
situation will not affect the well-being of those who are bilingual, since we have
assumed that they have no preference over which language to use. However, if the
bilingual individuals prefer language Y (e.g., this is their mother tongue), then this
situation will mean a lost opportunity to use their most preferred language, as well
as a welfare loss.
To solidify the ideas, the authors propose the following interaction. When two
citizens meet, 1 and 2, if the first one is bilingual (B) he must decide what language
to use when addressing the second. Then, if the second individual is also bilingual
she must decide in which language to answer. If she was addressed in language Y,
we can assume that she will answer back in that same language; however, if she was
addressed in language X, she may still try to speak language Y in case the first citizen
was bilingual but for some reason chose to use language X. Monolingual citizens do
not make any decisions, since they will always use language X. For bilingual citizens,
establishing a conversation in language Y gives a utility of 3, while doing so in lan-
guage X gives just 2. In addition, trying to establish a conversation in language Y to
end up speaking in language X implies a frustration cost of 1. The utility of mono-
lingual citizens (M) is not relevant for this analysis, but we can fix it at 2. The game
is depicted in Figure 11.1.
200
SEVEN MORE APPLICATIONS
N
1/3 2/3
1.B B M N
Y X
N N 1/3
2/3
1/3 B
1/3 B 2.B M
2/3
B
M
Y M
2.B
1.B X 2/3 Y X
Y
Y
3 1 3 2 2 2 2 2
3 2 3 2 2 1 2 2
Figure 11.1 The minority language game
First, Nature (N) defines the type of Citizen 1 (1), who can be either bilingual (B)
or monolingual (M), with respective probabilities 1/3 and 2/3. Then the game proceeds
as follows:
1. In case Citizen 1 is bilingual, he has to decide the language with which to address
Citizen 2 (2), in language X or in language Y.
1.1 If Citizen 1 uses language Y, Nature determines if Citizen 2 is bilingual (1.B)
or monolingual (1.M), also with probabilities 1/3 and 2/3. If Citizen 2 is bilin-
gual, she will respond in language Y, and the game ends in the leftmost final
node with payoffs (3, 3). If Citizen 2 is monolingual, she will answer in lan-
guage X (not shown in the figure, since really there is no choice). The utility
for Citizen 1 is 2 − 1 = 1, the utility 2 of speaking in language X minus the
frustration cost.
1.2 If Citizen 1 uses language X, Nature again determines whether Citizen 2 is
bilingual or monolingual with probabilities 1/3 and 2/3. If Citizen 2 is bilin-
gual (2.B), she must choose between answering in language X or in language
Y (this makes sense since there is a chance that Citizen 1 decided to speak in
language X in order to avoid the frustration cost). If she decides to use language
Y, Citizen 1 will switch to this language and both will enjoy a utility of 3. If
Citizen 2 is bilingual and answers in language X, they will continue in that
language and each will enjoy a utility of 2. If Citizen 2 is monolingual (2.M)
she will use language X (the action is not shown) and both will again have
utility 2.
2. If Citizen 1 is monolingual, he will use language X (not shown in the game), and
then Nature will determine whether Citizen 2 is bilingual or monolingual, with
probabilities 1/3 and 2/3.
2.1 If Citizen 2 is bilingual, she must decide which language to use, Y or X. The
rationale to use Y is in case Citizen 1 is bilingual but chose X to avoid frustra-
tion. Since we are in the case that Citizen 1 is monolingual, even if Citizen 2
chooses Y, the conversation will continue in language X, giving a utility of 2 to
Citizen 1 and of 1 to Citizen 2.
201
GAME THEORY
2.2 If Citizen 2 is monolingual, he will use language X (not shown in the game),
and both citizens will have utility 2.
The discontinuous line joins the two nodes where Citizen 2 knows that she has been
addressed in language X but does not know whether it is because Citizen 1 is monolin-
gual or because, being bilingual, he chose X to avoid frustration. Let us keep in mind
that the discontinuous line indicates that the player does not know in which node she
is and that her action must be the same in both nodes.
A bilingual Citizen 1 has two strategies: use Y or use X when he initiates a conversa-
tion. Other actions by this player are automatic and do not constitute a strategic choice:
if he is monolingual, he always chooses X; if he is bilingual and has been answered back
in language Y, he will choose Y as well. Citizen 2 also has two strategies in case she is
bilingual: answer back in X or in Y in case she was addressed in language X. If she was
addressed in Y, there is no strategic choice, since she answers in Y also. The normal
form of the game in Figure 11.1 is depicted in Figure 11.2a.
The payoffs are computed as follows. If both use language X, they get utility 2
regardless of their knowledge of language Y. If they both use Y in the places where there
is a strategy choice (they are bilingual and have to decide what language to use when
starting a conversation or answering back), the payoffs are
11 2  21 2  17
u1 (Y ,Y ) =  × 3 + × 1  +  × 2 + × 2  = ,
33 3  33 3  9
11 2  21 2  17
u2 (Y ,Y ) =  × 3 + × 2  +  × 1 + × 2  = .
33 3  33 3  9
If Citizen 1 uses X and Citizen 2 uses Y when she is bilingual, payoffs are
11 2  21 2  19
u1 ( X ,Y ) =  × 3 + × 2  +  × 2 + × 2  = ,
33 3  33 3  9
11 2  21 2  17
u2 ( X ,Y ) =  × 3 + × 2  +  × 1 + × 2  = .
33 3  33 3  9
If Citizen 1 is bilingual and uses Y and Citizen 2 is bilingual and chooses to answer
back in X, utilities are
11 2  2 17
u1 (Y , X ) =  × 3 + × 1  + × 2 = ,
33 3  3 9
11 2  2 19
u2 (Y , X ) =  × 3 + × 2  + × 2 = .
33 3  3 9
The game is depicted in Figure 11.2a and has a unique Nash equilibrium: (X, X). Both
players use X even when they are bilingual. The utility of speaking the preferred lan-
guage does not compensate for the possibility of a frustration cost.
202
Citizen 2
X Y
X 2, 2 19 17
,
9 9
Citizen 1
17 , 19 17 , 17
Y
9 9 9 9
Figure 11.2a The normal form of the minority language game
The number of bilinguals increases

From this point, we can analyze the game for different specifications of the parame-
ters. For instance, if the proportion of bilingual citizens were 2/3 rather than 1/3, the
normal form of the new game would be as illustrated in Figure 11.2b (Exercise 11.1).
In this game, there are two Nash equilibria in pure strategies: (X, Y) and (Y, X). In the
equilibrium (X, Y), a bilingual Citizen 1 chooses X to start a conversation. A bilingual
Citizen 2 will choose Y to answer back when she is addressed in language X, and the
conversation continues in language Y if they are both bilingual. In this equilibrium, the
situation of two random citizens conversing in language Y will occur with probability
2 2 4 , the probability of both of them being bilingual.
× =
3 3 9
Citizen 2
X Y
X 2, 2 22 20
,
9 9
Citizen 1
20 22 20 20
Y , ,
9 9 9 9
Figure 11.2b The number of bilinguals increases
In equilibrium (Y, X) Citizen 1 starts using Y when he is bilingual, and Citizen 2,

when being bilingual, responds in the same language that Citizen 1 uses (X if mono-
lingual, Y if bilingual). Again, the probability of two citizens meeting at random and
speaking language Y is 4/9. In both equilibria, bilingual citizens always speak language
Y among themselves, they only differ on who starts the conversation in that language:
the one who initiates the conversation or the one who responds.
The frustration cost decreases

Let us maintain the initial proportion of bilinguals as 1/3, but now change the frustra-
tion cost from 1 to 0.25. The new extensive form game is like the one in Figure 11.1,
except that, instead of payoff 1 ( = 2 − 1 ) after a frustrated use of language Y, we have
1.75 ( = 2 − 0.25 ). With this minor change, the normal form of the new game is the
one seen in Figure 11.2c (Exercise 11.1).
203
GAME THEORY
Citizen 2
X Y
X 2, 2 19 18.5
9 ’ 9
Citizen 1
18.5 19 18.5 18.5
Y , ,
9 9 9 9
Figure 11.2c The frustration cost decreases
Again, there are two equilibria in pure strategies – (X, Y) and (Y, X) – with the same
characteristics as in the previous case, where the proportion of bilinguals was higher.
In other words, for the purposes of inducing conversations between bilingual persons,
decreasing the frustration costs may be as important as increasing the number of bilin-
guals with preferences over language Y.
11.2 Pascal’s wager
In his book Thoughts, Blaise Pascal posed the following argument about the dilemma
of whether to believe in God (Pascal, 1670):
You have two things to lose: truth and good, and two things to compromise: your
reason and your will, your knowledge and bliss; and your nature has two things to
escape: error and misery. Your reason is no more hurt by choosing either of them,
since it is necessary to choose. This is a vain question. But your bliss? Weigh the
gain and the loss, calling out cross that God exists. Let’s consider these two cases:
if you win, you win everything; if you lose, you lose nothing. Bet He is, without
hesitation.
In short, Pascal is saying that

• if you do not believe in God, and if God does not in fact exist, nothing bad happens;
• if you do not believe in God and, as a matter of fact, God does exist, then you get
error and misery;
• if you believe in God and God does not exist, nothing bad happens; and
• if you believe in God and God does indeed exist, then you gain happiness.
The wager is a trivial one-person game, which can be described as in Figure 11.3a.
Payoffs are zero when nothing happens, 1 if something good happens, and −1 if
something terrible occurs. Let p denote the probability that God exists. If Pascal
chooses to believe, he will win p, and if he chooses not to believe, he will win − p .
Therefore, the rational thing to do is to believe, regardless of how slim the chance that
God exists is.
To better examine this wager, there are versions that make God an active player
who, in case of existing, prefers to punish the non-believer and reward the believer.
204
Pascal
Believe Do not believe
N N
Exists Exists
Does not exist Does not exist
p p
1−p 1−p
1 0 −1 0
Figure 11.3a Pascal’s wager
The game is like the one in Figure 11.3b. The result is the same. Payoffs 1, 0 and −1
can be changed with other payments that respect the ranking, and the result would
not change. In the case that God does not exist, we only include the payoff for Pascal.
Making God an active player allows for an explicit meaning of the hypotheses in
which Pascal’s wager is based. If God had other intentions (e.g., punishing anyone who
believes without evidence or punishing someone who only believes in order to win the
game), things would be very different (see Exercise 11.4). We can see this more clearly
if we take another wager using the Tooth Fairy, the fantasy figure who gives money to
children when they lose a tooth and leave it underneath their pillow:
• If you do not believe in the Tooth Fairy, and it does not exist, nothing bad happens.
• If you do not believe in the Tooth Fairy, and it does exist, there will be no money
under the pillow.
• If you believe in the Tooth Fairy, and it does not exist, again, nothing bad happens.
• If you believe in the Tooth Fairy, and it does exist, there will be money.
Pascal
Believe Do not believe
N N
Exists Exists
Does not exist Does not exist
p p
1−p 1−p
God God
0 0
Heaven Hell Heaven Hell
1 −1 1 −1
1 −1 −1 1
Figure 11.3b God also plays in Pascal’s wager
It is not easy to imagine that philosophers, from Voltaire to Bunge, who bothered to
look for inconsistencies in Pascal’s wager would have taken the Tooth Fairy’s dilemma
seriously, either to speculate whether the reward is temptation enough to have a solid
205
GAME THEORY
belief (Voltaire) or to entertain the idea about whether taking the existence of the Tooth
Fairy as something hazardous has any scientific meaning or is morally or philosophi-
cally confusing (Bunge). Most likely, the general position would be that the belief in
the existence of the Tooth Fairy cannot be based on speculating a conjecture in order
to enunciate this type of dilemma, but that it must be based on actual proof of its exist-
ence. The argument of God instead of the Tooth Fairy is exactly the same, regardless
of our personal attachment to it one way or the other. Needless to say, the above dis-
cussion says nothing about the existence of God or, for that matter, of the Tooth Fairy;
it only says that the argument in Pascal’s wager is erroneous from its very beginning.
11.3 The surprise exam paradox

A professor announces a surprise exam for the following week. The students reason that
it cannot take place on Friday, because if Friday comes and there has not been an exam
on any of the previous days, then they will know the exam is on Friday, thus removing
the element of surprise. Once Friday is discarded, the same argument goes for Thurs-
day and, by induction, for all other days. The students then conclude that the professor
cannot give a surprise exam that following week. Much to the surprise of everyone, the
professor gives them an exam on Wednesday.
The paradox has been solved long ago. Basically, it consists of proving that the
professor’s statement has various propositions that are mutually incompatible. We can
see it better if we think of a week with just two days. Under these circumstances, the
professor is saying
• if the exam is on day 1, the day before (or on day 1 before class) the students will not
know that the exam is on day 1;
• if the exam is on day 2, the students will know that it has not been on day 1, but will
not have known the day before that the exam would be on day 2; and
• the exam will take place on one of the two days.
Using the rules of propositional logic, one can show that the three statements cannot
be true at the same time, but we are not doing that here. All logicians agree with that.
What has led to confusion is the fact that, despite the inconsistency, the professor gets
her way and gives a surprise exam. It is at this point that the discussion has resulted in
dozens of articles appearing in serious journals.
Some authors define the degree of surprise as based on a definition of entropy
and look for a strategy for the professor that maximizes that entropy. Other authors
argue that the professor uses self-referential propositions and that nothing good can be
deduced from them. Still others define epistemological blind spots to try to shed some
light onto the issue without much success. Others think that it is important to distin-
guish between knowing a proposition one day and knowing it some other day. In the
midst of this confusion, Sober (1998) proposes a nice way of addressing the problem,
even if at the end he also makes some distinctions between prudential and evidential
propositions with no clear purpose.
In fact, the problem is simpler (once you know the solution, almost all problems
are). Let us begin with Sober’s approach using game theory, and then we will show that
all those abstruse attempts to solve the problem are unnecessary. Keep the two-day
206
week, as well as the professor who has to decide whether or not to give an exam at the
beginning of each day. If the professor gives an exam on day 1, there will be no exam
on day 2. At the beginning of each day, the students must write YES or NO on a piece
of paper. All students act as one player and all do the same thing. On a given day, if
there is an exam and they wrote YES, or if there is no exam and they wrote NO, we
conclude that there was no surprise that day. If the opposite was true, then there was
a surprise. Say that a surprise means a benefit of 1 to the professor and of −1 to the
students. The absence of a surprise changes the numbers to −1 and 1, respectively. Pay-
offs are arbitrary and could be changed with no effect on the essence of the analysis.
Remember that it is impossible to achieve these two things together: (i) to give a sur-
prise exam with probability 100% and (ii) that the exam is a surprise with probability
100%. For what follows we will assume that the professor may not give an exam with
probability 100%. That is, the professor may have lied about the certainty of giving
an exam and will try to maximize the surprise effect. In Exercise 11.6 we will assume
that the professor will give the exam but also that the exam may not be a surprise with
probability 100%.
If there was no exam on day 1, then on day 2 the professor and the students face the
game in Figure 11.4a. The only consistent way to play this game, and the only Nash
equilibrium, is for the professor to choose “Exam” and “No exam” with probabilities
(½, ½), and for the students to choose YES and NO with those probabilities also.
The expected profit for each player is zero.
Day 2 Students
YES NO
Exam −1, 1 1, −1
Professor
No exam 1, −1 −1, 1
Figure 11.4a Day 2 in the surprise exam paradox
Knowing this, the game is similar on day 1, with only a minor difference. If on day
1 the professor chooses to give the exam, the game ends that same day, regardless of the
students’ choice. If the professor chooses not to give the exam and the students choose
NO, then the game goes to day 2, with the payoffs in Figure 11.4a. If the professor
chooses not to give the exam on day 1 and the students choose YES, should the game
continue on to day 2? In our analysis we will not allow that, but the reader can try to
analyze the game allowing it to go on for one more day. We think that this is the most
natural scenario, allowing only one surprise. With this clarification, on day 1 the game
is like that indicated in Figure 11.4b.
Day 1 Students
YES NO
Exam −1, 1 1, −1
Professor
No exam 1, −1 0, 0
Figure 11.4b Day 1 in the surprise exam paradox
207
GAME THEORY
We have placed zeroes in the cell (No exam, NO), which is the expected payoff
when the game continues on to the second day. The cell (No exam, YES) can be inter-
preted as the game ending with payoffs (1, −1), or the game may go on but the result on
day 2 is irrelevant, since the students have already been surprised. Now, the only Nash
equilibrium is that the professor gives the exam with probability 1/3 and the students
choose YES with probability 1/3 also. Other variations of the game can be found in
Ferreira and Zamora (2008). In this analysis we can clearly see the two fundamental
facts of the paradox.
First, to give an exam and for the exam to be a surprise are two incompatible state-
ments. This means that the statements cannot both be true with probability one. In
2 1 1
our game, in equilibrium, there is a probability × = that there is no exam: the
3 2 3
probability that the professor does not give the exam the first day times the probability
that she does not give it on the second day either. There is also a positive probability
that, when the professor gives the exam, it constitutes no surprise to the students:
1 1 2 1 1 5
× + × × = . Had we insisted that the professor must give the exam one day
3 3 3 2 2 18
or the other, we would only need to change the game in the second day in order to reach
the conclusion that, again, the exam cannot be a surprise with probability one.
But the teacher manages to give the surprise exam! This second fact occurs because
the wording of the story tells us only one of its possible resolutions, in which the dice
roll in a way that the professor gives the exam and the students do not anticipate it.
According to our analysis, this can happen with positive probability, but does not nec-
essarily need to happen, since there is also a positive probability that the students cor-
rectly guess the day (or that the professor never gives the exam).
What happened is that neither the professor nor the students can reason without
giving thought to what they think the other player will do, or disregarding the value
of the surprise. This aspect places the paradox in the realm of game theory since prop-
ositional logic cannot address the interaction between actions and beliefs, at least not
before developing a theory of games.
11.4 The sentence as deterrence

What is the purpose of a penal system? Revenge? Deterrence? Rehabilitation? Restitu-
tion? A person takes €1000 from another person using violence and damaging his eye.
Revenge would imply that the victim causes the robber greater damage, like taking
even more money from the robber and hurting her more severely. Deterrence would
require punishing her harshly enough so that no one else, or very few people at least,
will be tempted to do something similar for fear of being punished likewise. Reha-
bilitation would consist in reeducating the perpetrator to make her understand that
stealing and harming people is immoral, perhaps through psychological therapy or by
giving her the necessary means of finding a decent job. Restitution would imply that
the thief returns the money and repairs the damage; if that is not possible, perhaps
the lex talionis is applied – the thief has to offer an eye for an eye. In addition to these
possible ends, the law usually covers many other considerations (e.g., the sentence has
to be proportional to the crime or the probability of punishing an innocent must be
208
small). The purpose of this section is not to elucidate which is the best penal system but
to show how some of these questions can be expressed in the language of game theory.
Consider the case for deterrence. Very often it happens that punishing a delin-
quent imposes a costly burden to the system, with the cost being sometimes even
larger than that of the property that the criminal can steal or damage if left free.
However, the cost to society is not only the damage that this particular thief can
cause, but rather the damage that all potential thieves can perpetrate if they are
encouraged by the lack of punishment. How does deterrence work? Let us start with
a simple game between a thief and the state. First, the thief decides whether to steal
or not. If she steals, the state must decide whether to track the thief down and punish
her or let her go. Say that the game is like the one in Figure 11.5a. Observe that the
absence of a punishment costs 1 to society, while the punishment costs 2, more than
doing nothing.
In this game, the only subgame perfect Nash equilibrium is the one marked with
the arrows: the thief steals, and the state lets her go without a punishment. The game
has another Nash equilibrium in which the thief does not steal and the state does
punish her in case she does steal; it is not subgame perfect, however, because the pun-
ishment is not a credible threat. A more realistic game would include a probability of
capturing the thief, though that would not alter substantially the game and its analysis.
In fact, we can interpret the payoffs after “Punish” as the expected payoffs after weigh-
ing the probability of having tracked the thief down.
Thief
Steal Not to steal
State
Punish Not to punish 0

0
−1 1
−2 −1
Figure 11.5a The basic game to analyze deterrence
To focus on the issue of deterrence, consider that at different moments there is a

potential thief that ponders whether to take advantage of an opportunity to steal. Say
that this happens once a day. Every time a potential thief steals, the state must decide
whether to let her go or punish her if caught. The new game is a sort of repeated game,
as the ones seen in Chapter 7, the difference being that while the state plays in every
repetition, it faces a different thief each time. If the number of repetitions is finite, then
the state will not punish and the thief will steal. In the last repetition, the state will not
punish, just like in the previous game in Figure 11.5a. Then, by backwards induction,
we will have the same behavior in the previous stages of the game.
Things change in infinite repetition (or the indefinite repetition, as seen in Chapter 7).
Now the state can sustain a reputation of punishing thieves in a subgame perfect Nash
equilibrium. To see this, we will use a strategy similar to the trigger strategy, but adapted
to the specificities of the present game. Let us call it the “deterrence trigger strategy.”
209
GAME THEORY
1. In the first period ( t = 0 ) players choose (Not to steal, Punish).

2. At any other period t > 0 , they play
(i) (Steal, Not to punish) if in the past some thief and the state chose (Steal, Not to
punish) at some point and
(ii) (Not to steal, Punish) if at no point a thief and the state chose (Steal, Not to
punish).
Before continuing, see that the strategy is well defined in the sense that it determines a
plan for every player under any circumstance. As always, to show that it is a subgame
perfect Nash equilibrium, we must check that it constitutes a Nash equilibrium in
every subgame. Observe that the stage game is a dynamic one and that we have sub-
games starting not only at the beginning of any repetition but also in the middle of a
repetition, after the potential thief has chosen “Steal.”
The deterrence trigger strategy is a Nash equilibrium of the whole game

Notice that if both the state and the thieves follow the strategy, no one will ever
steal and the state will not need to carry out any punishment. If the state follows
its part of the strategy, then no potential thief will have an incentive to change
hers, since she would win −1 instead of 0. If all thieves follow their part, the state’s
choice is irrelevant, since the nodes where it must take an action are never reached.
Therefore, the strategy is a Nash equilibrium. This is actually very similar to the
non-credible Nash equilibrium in the game without repetition. The difference being
that the repetition will make the threat credible, as it will be a Nash equilibrium in
the subgames.
The deterrence trigger strategy is a Nash equilibrium in the subgame

at the beginning of a period t > 0 after at some past period players
chose (Steal, Not to punish)
In these subgames, according to clause (i) the strategy prescribes playing (Steal, Not to
punish) now and in every future period. Since all of the thieves steal, the state suffers a
loss every time it changes to “Punish” (its payoff changes from −1 to −2) without alter-
ing any future behavior. Given that, the best reply for the thieves is to steal. This means
that the strategy is indeed a Nash equilibrium in these subgames. Notice that this part
of the strategy is the reason why the state has the incentive to punish any occasional
thief. If at some point the state does not punish, then its reputation would be lost and
all future potential thieves would steal.

at the beginning of a period t > 0 after no players chose (Steal, Not to
punish) in any past period
Remember that, at the beginning of any period, the continuation of the game is exactly
the same game as in the original infinite repetition. Since in these subgames the strat-
egy prescribes the same thing as in the game as a whole, the argument showing that the
strategy is a Nash equilibrium of the whole game applies here as well.
210

at the middle of a period (after the thief has chosen, but before the
state has chosen)
If the thief has chosen “Not to steal,” the state does not have a choice to make in
that period, and the game will move on to the next repetition, where we have already
shown that the strategy is a Nash equilibrium. So we need only analyze the case where
the thief has chosen “Steal.” If in this period the thief was supposed to play “Steal”
because the clause 2(i) of the deterrance trigger strategy applies, then the strategy pre-
scribes “No to punish,” and then a continuation (Steal, Not to punish) in every future
period. We have already seen that this is a Nash equilibrium. If the thief was supposed
to play “Not to steal” because either clause 1 or 2(ii) in the definition of the deterrence
trigger strategy apply, then the state is supposed to play “Punish” according the same
deterrence trigger strategy; then the players are prescribed to play as in clause 2(ii).
Now, if the state follows the strategy, it gets −2 now and zeroes in the future. However,
if it deviates, it will get −1 today and then −1 in all future periods. If the state is not
too impatient, it will prefer to stick with the strategy rather than to deviate. Thus, the
strategy is a Nash equilibrium in these subgames.
We have seen that the action of punishing the thieves is a credible strategy in the
repeated game, even if it is very costly and not credible in the stage game. This is not
the only possible scenario. In the analysis above we have already seen that to play (Steal,
Not to punish) in every stage is also a subgame perfect Nash equilibrium. Which equi-
librium will prevail depends on circumstances not made explicit in the game, or not
considered in the definition of the equilibrium we used. In fact, one can argue that
since the state is the only player present in each stage, it will have an advantage when
signaling its preferred equilibrium. Suppose that in a society the equilibrium (Steal,
Not to punish) prevails in every repetition and that the state deviates and starts punish-
ing the thieves. What should future thieves think? That the state has made a mistake
and will return to the equilibrium? Not necessarily. The state’s behavior can be ration-
alized as a way of building up a reputation towards its behavior in the equilibrium using
the deterrence trigger strategy, signaling that it is only considering that equilibrium and
that it expects the thieves to play accordingly. The study of the equilibrium refinements
or dynamics which define this deterrence trigger strategy as a more stable equilibrium
than the other goes beyond the scope of this book.
Variations of the game

The baseline game in Figure 11.5a can be altered and adapted in order to incorporate
new elements. For instance, if the state’s action is not directly to punish the thieves but
rather to search for suspects and put them on trial, the game can look like the one in
Figure 11.5b. Nature, N, determines whether or not the suspect is the real thief with
probabilities p and 1 − p.
The following features of the game are worth emphasizing:
• To punish an innocent is the worst result for the state.
• The thief has utility −1 only in the case that she steals and is punished. The first
number is always the utility of the actual thief. The utility of the wrongly accused
individual is not shown, since she is not a player of the game.
211
GAME THEORY
Thief
Steal Not to steal
State
Search for a suspect Not to search 0
0
N
1
Guilty Nonguilty −1
p 1−p
State
Not to punish Not to punish

Punish Punish
−1 1 1 1
−2 −1 −3 −1
Figure 11.5b A more realistic deterrence game
• The action of stealing may be profitable for the thief even if the state decides to
search and punish. This is clearly the case if the probability of capturing the wrong
suspect is very high (in the example, the reader can confirm that this probability
must be at least ½).
• Probability p may be different in each repetition. We will assume that both the state
and the thief know the probability.
• The game can become more complicated with more features, such as arresting more
suspects and the state incurring a cost for every arrest. At the end of the day, we
will still see that there is a probability that the wrong suspect is sentenced, which is
already covered in the present game.
• If the probability of arresting the wrong suspect is high, then the thief will steal in
any case. Thus, the state may decide that it should not punish the suspects.
The deterrence trigger strategy in the new repeated game is as follows:
• In the first period ( t = 0 ) players choose (Not to steal, (Search for a suspect, Punish)).
• At any other period t > 0 , they play
(i) (Steal, (Not to search, Not to punish)) if during all past periods players chose
(Steal, (Not to search, Not to punish)) or (Steal, (Search, Not to punish)) when
p was high or if it is high in the current period and
(ii) (Not to steal, (Search for a suspect, Do not punish)) if at some point in the
past (Steal, Do not search) or (Steal, (Search for a suspect, Do not punish)) was
played when p was not high or if, in the present moment, p is not high.
This strategy is a subgame perfect Nash equilibrium. Observe that we still need to define
the value of p lower than that beneath which the state considers the suspect must not be
punished. Each value will result in a different equilibrium, and in each of them there
will be thefts that would go unpunished as well as thefts that would have been prevented.
212
The state can devote more resources to investigating the thefts in order to have the high-
est probability of detaining the correct suspect; however, at the same time it must weigh
the cost of these resources with the additional quantity of thefts that would be prevented.
In this new complication we still see that the thief who is aware that she will not be
punished will steal (because the probability of not being arrested is high and because,
even if arrested, she will be found not guilty), and the thief who expects to be punished
will not steal. This means that we will never see a thief be condemned because, for the
cases for which the state will punish (p is high), deterrence works perfectly.
In real life, things will not be as clear, and there will be thieves who will believe
themselves part of the first group when they are actually part of the second. Those
thieves will end up stealing but will face a high probability of being arrested and pun-
ished. If this is the case, two problems may arise. On the one hand, the punishment
may not be proportional to the crime (until now that was of no concern, since no one
was punished due to deterrence working perfectly), on the other hand, there will be
a probability that an innocent person will be punished. These two problems are of
special concern in democratic regimes as compared to authoritarian ones. In the most
extreme case we find concentration camps and prisons, where punishments are severe
and where, if the true culprit is not found, the punishment of innocents is sometimes
explicitly included. Armies also resort to this form of imposing discipline. Dictator-
ships are often able to reduce crime rates by abusing the severity of punishments and
through the risk of condemning innocents without fearing opposition.
Of course, this kind of abuse on the part of authorities may have additional impli-
cations not covered in the game. Here we only point out how the rule of law may have
more problems for fighting against crime, at least in the short run. Whether this possi-
bility is a sensible hypothesis that explains real facts is something that will be discerned
with data, as usual. Remember that our purpose with these examples is to show how to
formulate different hypotheses using game theory, and, in doing so, to shed some light
on their significance and scope.
We can change the game to introduce the element of rehabilitation. Let us do so in the
game in Figure 11.5a, where the strategy “Punish” is now to be understood as “Reclusion
under a rehabilitation program.” Since rehabilitation will help the thief, the payoff in the
case of being punished will be greater than −1, but it must still be smaller than 0, the payoff
after “Not to steal,” for the game to make sense. Otherwise, the thief will always find it prof-
itable to steal. Once we have made these adjustments, the analysis is conducted as before.
If we introduce rehabilitation in the game in Figure 11.5b, the thief will find it
attractive to steal more often, since the expected payoff after stealing is now higher. If in
each stage the potential thieves are different people, the result is that there will be more
thefts and the state will have to put forth more resources in order to arrest the true thief
and thus prevent further thefts. If the prevented thefts that would have been committed
by the same now rehabilitated thief outnumber those that are induced by the incentive
for other thieves to steal more frequently, the total quantity of thefts will decrease.
11.5 Solidarity versus charity

In different examples throughout this book we have seen situations where voluntary
contributions to the provision of a public good result in a game in which the equi-
librium is not efficient. Usually, we understand a public good as that which can be
213
GAME THEORY
consumed by one individual without precluding its consumption by another. Thus, a

bottle of water with a label that states it being of public interest is not a public good,
for if I drink it, no one else gets to drink it. There are local public goods (street lights)
and global public goods (national defense), those that are subject to congestion (a park)
and those that are not (the standardization of systems). Finally, in the case of certain
public goods, the provider can exclude some individuals from their consumption (a
sports event inside a stadium), while in some other cases this cannot be done (an open
radio broadcast). The funding for the provision may be private or public, as might be
the provision itself. The canonical case is made up of global public goods not subject to
congestion and without exclusion.
Solidarity is also a public good. To simplify the question and crystallize the con-
cepts, we will focus on one aspect of solidarity, specifically, the preference towards
income equality among people. There are different indexes that measure this inequal-
ity. Suppose we agree that one of them, for instance, the Gini index, gives an accurate
enough measure of the state of inequality in a society. If this index moves one way or
the other following variations in inequality, the individuals who care about equality
will see their utility levels changed without excluding others from experiencing similar
changes. In other words, people’s utility depends on the value of the index, and this
value is the same for all. Therefore, this aspect of solidarity not only constitutes a public
good, it is also a canonical example of these goods. A similar argument can be made
about social segmentation or discrimination.
The importance of this good depends on the importance that individuals associate
with it, and this will depend on each person’s preferences as influenced by ideology,
morality, ethics, religion, philosophy, education or any other circumstance. Suppose that,
as a matter of fact, solidarity concerns all the people within the society. Being a public
good, individuals would freely accept a contract that makes them committed to dedicat-
ing part of their resources to favor solidarity. This situation is very different from ask-
ing individuals to provide it voluntarily. The first mechanism addresses the problem of
free-riding, while the second does not. It is true that such a contract is not actually signed,
but what matters for the sake of the argument is that, in this instance, the state works in
a way that its actions can be interpreted as if it were the product of such a contract.
All of the above can be expressed in the language of game theory. Suppose that a
society is composed of two rich individuals and one poor individual. Each rich per-
son has a wealth of 10, whereas the poor only has 1. Each of the rich individuals can
donate part of their wealth to charity, which will be given to the poor. The utility of
each rich individual is given by the function u = 10 × W × (10 − SD ), where W is one’s
own wealth while SD is the standard deviation of the wealth distribution in the society
(the factor 10 is used only to obtain nice round numbers in the game). The standard
deviation is a measure of dispersion of wealth and, even if it is not a good measure of
inequality, it will serve our purposes, since it is easy to calculate. The utility function
states that rich individuals are better off if their wealth increases, but also if the ine-
quality, as measured by the standard deviation of the wealth distribution, decreases.
To simplify the game, assume that donations can be of 0, 1, 2 or 3 units of wealth.
All of this defines a game between the two rich individuals, called Rich 1 and Rich 2,
depicted in Figure 11.6.
To understand the utilities in the figure, let us explain the numbers in one of the
cells, for example, cell (0, 2). If Rich 1 contributes zero and Rich 2 donates 2 units,
214
Rich 1’s wealth will be 10, while Rich 2’s will be 8 (his 10 minus the 2 that he donates).
The poor individual will have 3 (her 1 plus the 2 units that Rich 2 donates). The wealth
10 + 8 + 3
average in the society is = 7 , so the standard deviation is calculated as
3
(10 − 7 )2 + (8 − 7 )2 + ( 3 − 7 )2
SD (10, 8, 3 ) = = 2.94 ;
3
therefore, Rich 1’s utility is u1 (10, 8, 3 ) = 10 × 10 × (10 − 2.94 ) = 706 , while Rich 2’s is
u2 (10, 8, 3 ) = 10 × 8 × (10 − 2.94 ) = 564 .
The only Nash equilibrium of the game has every rich person donating one unit
of their wealth to the poor, with a final wealth distribution of (9, 9, 3), which gives a
utility of 645 to each of the rich. This situation is not optimal for them. If they each
donate 3, then the final wealth distribution will be (7, 7, 7), with the rich enjoying a
utility of 700. In other words, in the society we just have described, the rich individuals
have a strong preference for equality, but the voluntary contribution does not influence
them to be as generous as they wish to be as a society due to the free-rider problem.
Each of the rich prefers that the other contributes the most part, with the consequence
that they both end up contributing too little. A social contract to implement solidarity,
in which each commits to donating three units of wealth if they all do, would be signed
voluntarily by all of the rich.
Rich 2
0 1 2 3
0 576, 576 644, 580 706, 564 755, 529
1 580, 644 645, 645 706, 627 753, 586
Rich 1
2 564, 706 627, 706 687, 687 735, 643
3 529, 755 586, 753 643, 735 700, 700
Figure 11.6 The solidarity game
11.6 Single-round versus runoff elections

Consider a country with one left-wing party (LP) and two right-wing parties. One of
the right-wing parties is radical (RP) and the other is moderate (MP). Consider also
that the moderates attract more voters than the radicals. To fix ideas consider that
the percentage of supporters are 43, 35 and 22 from left to right. In this situation, a
single-round plurality vote to elect one candidate will result in the victory of the one
proposed by LP (43% of votes) if all voters vote for their most preferred candidate.
Knowing this, radical rightists may decide to vote for MP. The final result depends
on how radical voters balance their preferences for a candidate closer to their ideology
with a chance to win (MP) and the wish to support and give visibility to their most
preferred option (RP).
In a runoff system, where the two top candidates are selected to compete in a second
round, the radicals can show their support for their candidates in the first round, and
215
GAME THEORY
will most likely vote (rather than abstain) for the moderate in the second round, as the
radical option no longer has a candidate.
Let us consider the game of forming alliances between the two right-wing parties.
The way MP and RP supporters rank the feasible results (only MP or LP win) are like in
Table 11.1. In addition, RP has a value of 1.5 if they get visibility, which is obtained by run-
ning as a party in an election, but will not be obtained if they go in a coalition with MP.
Table 11.1 Parties’ preferences
MP RP Value
MP alone MP in coalition with RP 2
MP in coalition with RP MP alone 1
LP LP 0
Plurality vote
Consider the vote among one candidate of each of the three parties using the plurality
vote. The candidate with most votes wins. In this case, RP supporters will vote for the RP
candidate to get 1.5 because of visibility, which is more than the value of 1 they get if the
MP’s candidate is elected. As a result, the candidate presented by LP wins. Anticipating
this, the MP makes an offer to the RP, consisting of accepting some of the RP’s proposals
and forming an electoral coalition to support the MP’s candidate. Let us measure the
concession to RP as a number C between zero and one, with zero meaning no concession
and one meaning adopting the totality of RP’s platform. If MP concedes x = 0.6, its util-
ity in case of its candidate being elected is 2 − 0.6, while the utility of RP will be 1 + 0.6.
With the above specifications, define a game in which first MP makes an offer to
RP. If the offer is accepted, they form the coalition. If it is rejected, RP must decide
whether to vote for MP or for RP. This is illustrated in Figure 11.7. Starting from the
end, we see that RP will vote for its own candidate in case it rejects the proposal, mak-
ing LP win, and it will get a utility level of 1.5, the value of visibility. Going one step
backwards, this implies that RP will accept any concessions such that C ≥ 1.5, which
in turns implies that MP will offer just C = 1.5. A coalition is then formed with MP
making some concessions to the radicals.
MP
RP
A R
RP
2–C
C Vote MP Vote RP
2 0
1 1.5
Figure 11.7 Plurality vote
216
Runoff
In the runoff voting mechanism, the two candidates with the most votes go to a sec-
ond round. In this second round, there are only two options and, therefore, voting
sincerely is the best option. Also, if the candidates in the second round are the ones
belonging to LP and MP, the RP has no visibility to gain, and its supporters will vote
for the MP’s candidate. Thus, MP does not need to make any concessions to RP. In
the first round, each party will get the votes of its supporters. Supporters of LP and
MP want to guarantee that their candidates go to the second round, and supporters of
RP want to give their candidate visibility. MP and LP go to the second round, where
MP wins.
Discussion
After analyzing these strategic aspects in a more complete version of the game, Bor-
dignon, Nannicini and Tabellini (2016) conclude that
if the electorate is polarized and moderate parties are large, the runoff system
allows moderate candidates to pursue their own platform without being forced
to merge and compromise with the neighboring extreme candidates. Intuitively,
with single round and under sincere voting, the extremes can threaten to cause
the electoral defeat of the nearby moderate candidate if he refuses to strike an
alliance. Under runoff this threat is empty, provided that enough extremist voters
are willing to vote for the closest moderate in the second round.
The authors then find three testable predictions of the model:

(a) Runoff elections increase the number of political candidates compared to
single-round elections.
(b) Extremist parties are less likely to merge with moderate parties under
runoff.
(c) Policy moderation is higher – and policy volatility lower – under runoff.
To test these hypotheses, the authors take a natural experiment in Italy’s municipal
elections. Previous to 1993 all municipalities used the same electoral mechanism. Since
that date the system was changed to one of plurality rule, with a single round for
municipalities below 15,000 inhabitants, and with a runoff system above. The authors
then selected the municipalities around the threshold (between 10,000 and 20,000
inhabitants). These are very similar municipalities where the difference in the election
system does not correlate a priori with any other characteristic.
In general, before the reform, one could find that the number of candidates
increases with the size of the municipality, with no discontinuity for municipalities
around 15,000. However, after the reform, there is indeed a statistically significant
discontinuity, as the authors find that the number of candidates for mayor increases
by 29% under the runoff system, other things being equal. They also find that the
chances a radical candidate is elected as mayor is smaller in the runoff system and that
the changes in tax policy after a change in the mayor is greater in the single-round
system. All these data are at least a partial support for the hypothesis (a), (b) and (c)
suggested by the theoretical model.
217
GAME THEORY
11.7 How to eliminate illegal parking

Each day a city’s downtown area receives 2,000 private vehicles of drivers coming to
their jobs. Each of them must decide whether to use a parking lot that costs 10 or leave
their car illegally parked on the street (e.g., by double parking). The downtown area,
packed with illegally parked cars, becomes congested easily, which harms everyone
living, working or just visiting there. More specifically, let us say that the quality of
traffic is valued in 100 by each driver if the 2,000 vehicles use legal parking, and that
the valuation decreases by 0.05 for each illegally parked car.
Each of the 2,000 drivers is a player with two options. We cannot represent the game
with matrices, but we can represent the drivers’ utilities as a function of their strategies
with the following formulae, that depend on the number n of illegally parked cars:
u ( parking ) = 100 − 0.05n − 10 ,
u ( illegal ) = 100 − 0.05n .
The first expression shows the utility of a driver who parks legally, given that there are
n drivers that park illegally. The second expression does the same for the choice of park-
ing illegally. With them, we can try three configurations to find the Nash equilibria:
1. If all drivers use the legal parking and, then, n = 0, their utility will be
u ( parking ) = 100 − 0.05 × 0 − 10 = 90 . If one of them changes her mind and parks
illegally, her utility will be 100 − 0.05 × 1 = 99.95 (n = 1). Thus, all drivers using the
legal, costly parking is not a Nash equilibrium.
2. If they all park illegally, each driver will have a utility 100 − 0.05 × 2, 000 = 0.
If one changes her mind and decides to park legally, her utility becomes
100 − 0.05 × 1, 999 − 10 = −9.95 , which is lower. This situation constitutes a Nash
equilibrium.
3. In general, it does not matter how many cars are illegally parked; drivers who use
lawful parking will win if they stop being law-abiding citizens and park illegally.
Notice that they will go from winning u ( parking ) = 100 − 0.05n − 10 to winning
u ( illegal ) = 100 − 0.05 ( n + 1) = 100 − 0.05n − 0.05, which is bigger.
This means that the only equilibrium consists of all drivers parking illegally, as this
strategy is dominant.
Let us move forward. City hall wants to eliminate the illegal parking; therefore,
they allocate more police patrols to catch and fine all illegally parked cars. There are,
however, two restrictions. First, according to law, the fine for illegal parking is 100.
Second, patrols have only enough time to give 120 tickets to illegally parked cars in a
day. The new formulae for computing drivers’ utilities are
u ( parking ) = 100 − 0.05n − 10 ,
120
u ( illegal ) = 100 − 0.05n − × 100 if n > 120 , and
n
u ( illegal ) = 100 − 0.05n − 100 if n ≤ 120.
218
In the utilities in the case of illegal parking, we have subtracted the 100 fine multiplied
by the probability of actually being fined, which is 120 in the case that there are more
n
than 120 vehicles that are illegally parked (remember that patrols only have time to give
120 tickets), and 1 if there are 120 or less.
Let us calculate the Nash equilibria with fines.
1. If all drivers use the legal parking, the utility for each one is
u ( parking ) = 100 − 0.05 × 0 − 10 = 90 (n = 0 ). If one of them deviates from this
behavior and parks illegally, her utility will be 100 − 0.05 × 1 − 1 × 100 = −0.05
(n = 1). The deviation carries a fine with certainty, so all drivers choosing to use paid
parking is a Nash equilibrium.
2. If everyone parks illegally, each driver will have a utility given by
120
100 − 0.05 × 2, 000 − × 100 = −6.67. If one deviates and uses the parking, her
2000
utility will be 100 − 0.05 × 1, 999 − 10 = −9.95. Thus, everyone parking illegally and
causing congestion in the downtown area is also a Nash equilibrium.
3. In general, if there are n vehicles parked illegally, and one of the legal drivers con-
siders deviating, her utility will change from u ( parking ) = 100 − 0.05n − 10 to
120
u ( illegal ) = 100 − 0.05 ( n + 1) − × 100, which means that the deviation will
n +1
120
not be worthwhile if 0.05 + × 100 ≥ 10 , which implies n ≤ 1,205. If one of
n +1
the drivers that parked illegally considers going legal, her utility will change from
120
u ( illegal ) = 100 − 0.05n − × 100 to u ( parking ) = 100 − 0.05 ( n − 1) − 10. The
n
120
deviation does not pay if × 100 ≤ 9.95, which implies n ≥ 1,206. Since both
n
conditions cannot be met simultaneously, we conclude that there is no equilibrium
with drivers being divided in their choice of strategy.
To summarize, we have found two equilibria: either all drivers park legally or they
all park illegally. If the policy to fine illegal parking followed a situation in which
all parked illegally, the historical inertia would easily cause the bad equilibrium to
persist.
In view of the above, city hall decides to change its policy. They divide the down-
town into two areas, A and B, with each receiving half of the incoming cars, and
order the patrols to watch and give tickets in zone A only. Once they solve the parking
problems there, they will move on to zone B. The plan is made public for all drivers to
be aware of. Will the new plan work? To answer this question, let us calculate the new
equilibria.
1. If all drivers in zone A use the legal parking, their utility will be
u ( parking ) = 100 − 0.05 × 1000 − 10 = 40 (there will still be 1000 illegally parked
cars in zone B). If one driver in zone A decides to park illegally, she will certainly
219
GAME THEORY
get a ticket and her utility will change to 100 − 0.05 × 1001 − 1 × 100 = −50.05
(n = 1001). That is, if all drivers in zone A use the legal parking, then that situation
constitutes a Nash equilibrium.
2. If everyone parks illegally in zone A, each driver gets utility
120
100 − 0.05 × 2000 − × 100 = −12. If one deviates and uses paid parking, she
1000
will get 100 − 0.05 × 1, 999 − 10 = −10.05 , a worthwhile deviation. Therefore, this
situation is not a Nash equilibrium.
3. Computations similar to the previous ones show that there are no equilibria with
some drivers parking legally and some illegally. We will not repeat them here.
Once the good equilibrium prevails in zone A, the parking enforcement patrols can ded-
icate their efforts to controlling zone B. They will still have to devote some time to zone
A, but with all drivers being law-abiding citizens, they can count on giving very few tick-
ets to the occasional offender in that zone. Almost all of their time can be concentrated in
zone B with results similar to those in zone A. We work out the details in Exercise 11.12.

11.1 Show the calculations for the payoffs in Figures 11.2b and 11.2c.
11.2 Show that nothing changes in the analysis in Section 11.1 if we change the
utility of the monolinguals from 2 to 3, while the utility of the bilinguals stays
as it is in the game in Figure 11.1.
11.3 When a minority language is not normalized, bilingual speakers must find
ways of expressing new ideas. Suppose that new technologies introduce the
new idea “I.” The majority language can express this idea in many forms, but
through different media (the industry, mass media or educational institutions,
for instance) one of them becomes the normalized form. Let “N” be the nor-
malized form of the idea “I.” The minority language can also express the idea
“I” in different ways, but lacking normalization (through industry, mass media
or educational institutions), the speakers must choose which one to use. Let
“P,” “Q” and “R” be the possible ways of expressing idea “I” in the minority
language. Consider now the game between two speakers of the minority lan-
guage, who, in order to express the idea “I” may use the standardized way of the
majority language, “N,” or may use one of the different expressions from the
minority language. If they use the same expression in their preferred language
(the minority language), they get a utility of 3. If they coincide in the use of the
standard expression of the majority language, they get a utility of 2. Finally, if
they use different expressions, they get a utility of 1.
(a) Represent the game in normal form.
(c) Which of the equilibria is particularly conspicuous? (We are not asking for
the best, but for the one that, being different from the others, may attract
more attention.)
220
(d) Look up information about focal points in game theory and argue that the
equilibrium in (c) satisfies the definition.
11.4 Let us modify the game in Section 11.2 between Pascal and God. Pascal ques-
tions the existence of God and can take one of two actions: “accept the uncer-
tainty” or “pretend to believe.” Next, one of two of Nature’s events regarding
God’s existence takes place: he exists with probability p and does not exist with
probability 1 − p. If God exists he can send Pascal to heaven or to hell. Pascal’s
utilities are like those in the game in Figure 11.3b, but God’s utilities change:
he prefers sending Pascal to heaven in the case that he is sincere and to hell in
the case that he is not sincere.
(a) Represent the normal form of the game.
11.5 In Section 11.3, the game was divided between Figures 11.4a and 11.4b. Show
the complete extensive form game that includes all of the actions of the two
days.
11.6 Consider the surprise exam paradox game in Section 11.3 for the case in which
the professor is obliged to give an exam. In other words, if she did not give the
exam on day one, she has to give it on day two.
(a) Show the subgame that corresponds to day 2 and its equilibrium.
(b) Show the game in day 1. For the case that the players’ actions imply the
game continues on to the next day, simply put the utilities of the equilib-
rium in day 2 found in (a).
(c) Show the complete extensive form of the game, showing the actions in
both days.
(d) Find the subgame perfect Nash equilibrium. What is the probability of the
exam being a surprise?
11.7 Let us complicate the baseline game in Figure 11.5b. After deciding “Search for
a suspect,” Nature determines whether a suspect is found or not, with probabili-
ties q and 1 − q, respectively. In case the suspect is not found, payoffs are (1, −1),
the same as after the action “Not to search.” After finding a suspect, the game
continues as in Figure 11.5b, with Nature determining whether the suspect is
guilty or not.
(a) Show that in this game all subgame perfect Nash equilibria imply “Steal”
for the thief and that she will never be punished.
Consider the infinite repetition of the new game, with a different thief each
time.
(b) Show a subgame perfect Nash equilibrium in which there is a deterrence
effect. Does a thief ever get punished?
(c) Is there any substantial difference with the analysis of the game in Fig-
ure 11.5b repeated infinitely many times?
11.8 Let us complicate the baseline game in Figure 11.5b. After deciding “Not to
search,” Nature determines that even in that case there is still a probability
that the real thief is arrested and punished with probability q, with payoffs
221
GAME THEORY
(−1, 1). With probability 1 − q the thief is not arrested and payoffs are (1, −1).
In the case that the state decides “Search for a suspect,” the game continues as
in Figure 11.5b.
(a) Show a subgame perfect Nash equilibrium in which the thief chooses to
steal. What does the state choose in that equilibrium?
Consider the infinite repetition of the new game, with a different thief each time.
(b) Show a subgame perfect Nash equilibrium in which there is a deterrence
effect. Does a thief ever get punished?
(c) Is there any substantial difference with the analysis of the game in Fig-
ure 11.5b repeated infinitely many times?
11.9 Check the payoffs of the game in Figure 11.6.
11.10 Consider the game between Rich 1 and Rich 2 in Section 11.5, and assume that
only Rich 1 has preferences as indicated in the text, while Rich 2 only cares
about her wealth, given by the utility function u = 100 × W .
(a) Show the normal form of the new game.
11.11 In the game in Section 11.7, consider the case in which the patrols only give
tickets in zone A. Show that there are no Nash equilibria in pure strategies in
which some drivers use the parking lot while other drivers park illegally.
11.12 In the game in Section 11.7, consider that in zone A the Nash equilibrium is
all drivers use the parking lot, and the patrols dedicate enough time to give 5
tickets in that zone as well as 115 tickets in zone B. Find the Nash equilibria in
zone B.
222
12 DYNAMICS
In the first chapters of this book we introduced two problems that sooner or later
should be addressed within game theory. The first deals with the assumption of ration-
ality that underlies the notions of Nash equilibrium as well as of subgame perfect Nash
equilibrium. In regards to this problem, we stated that the game model should be capa-
ble of accommodating other equilibrium definitions in which the rationality assump-
tion was either relaxed or eliminated altogether. Once rationality is defined, there is
only one way of being rational, though many ways of being irrational, which makes the
task of including irrationality a difficult one. For this reason, we found it convenient
to begin the theory with the rationality assumption (in addition to rationality being
a useful assumption in many contexts). The second problem is the completion of the
modeling based on equilibria in order to include a behavioral theory of individuals that
explains how they reach the equilibrium. In Chapter 2 we saw some early approxima-
tions to this problem, with the use of maximin strategies for zero-sum games as well as
through the definition of rationalizable strategies for some other games.
In this chapter we will present different dynamics for advancing in those two direc-
tions. As the reader may have understood from the above paragraph, there will not be
a single answer to the problem of relaxing the rationality assumption, and the theory
will have to explore different alternatives, whose applicability should be studied case
by case. The problem of finding a dynamics that leads to the Nash equilibrium is a
well-defined research line, but there are still many difficulties due to the possibility of
having multiple equilibria in one game. The proposals for dynamics become rapidly
complicated and their rigorous definitions and analysis go far beyond the scope of this
book. The introduction of some of these dynamics and their application through some
examples will, however, illustrate how game theory can be applied to simple, yet inter-
esting situations without advanced mathematical tools.
In a dynamic setting, players react to a situation determined by the history of what
they played in the past. A dynamics is defined as the rules that govern these reactions
and implies a particular evolution on the strategy choice along time. We will see five
different dynamics and will present examples in which the problem of equilibrium
selection is resolved through their use. However, the examples are particular, since most
of these dynamics do not converge in general to an equilibrium or group of equilibria.
12.1 Evolutionary dynamics: The hawk-dove game

The game models and equilibrium definitions in terms of rational behavior can be
adapted to situations of complete lack of reflection, where strategies compete against
each other in an evolutionary environment and where the best adapted survive and
propagate. In this section we present an example.
Two individuals of the same species compete for a resource (a mate, food or shel-
ter, for example). Neither of the individuals is evidently bigger or stronger than the
223
GAME THEORY
other. What should they do? Be a hawk and fight? Or be a dove and threaten to fight,
but retreat as soon as the other attacks? Run away? If the individuals have been pro-
grammed to fight through evolution, one of the two will obtain the resource. If the cost
of fighting, however, is so high that the value of the resource does not compensate, then
fighting may prove to be a bad strategy unless the other individual does not fight. In a
world of fighters, an individual who does not fight will avoid incurring this high cost
and will enjoy greater success whenever there is an encounter with these characteristics,
while the rest spend their time and strength fighting each other. But in a world of doves,
a hawk will take advantage of the non-fighters in order to keep the resource without
cost in every confrontation, and will thus be the one enjoying the higher success.
To fix the ideas in our game, suppose that the value of the resource is 2 and that
the cost of fighting is 4 in the case that the individual loses. If they both fight, each
one has a chance of winning with probability ½ and will get an expected payoff given
1 1
by × 2 + ( −4 ) = −1. If one fights and the other does not, the one who fights keeps
2 2
the resource, while the other gets nothing. If neither one fights, but simply pretends to,
then either of them can get the resource with probability 1/2. The game is shown in
Figure 12.1.
2
Fight Pretend
Fight −1, −1 2, 0
1
Pretend 0, 2 1, 1
Figure 12.1 The hawk-dove game
In the simplest of models in evolutionary games a new assumption is usually added,

namely, that all individuals of the same species use the same strategy, as programmed
in their DNA. In complex species, this assumption is very restrictive, but we will use it
for this first illustration of the application of game theory to evolutionary games. Under
these conditions the Nash equilibria in pure strategies (Fight, Pretend) and (Pretend,
Fight) cannot be the result of a strategy selected through evolution, since they require a
different behavior from different individuals. A third equilibrium exists within mixed
strategies, which is a candidate to being selected. Therefore, our first task is to compute
this equilibrium. Remember that for this it is necessary that both fighting and pretend-
ing give the same expected payoffs:
u1 ( Fight, q ) = q × ( −1) + (1 − q ) × 2 = q + (1 − q ) × 1 = u1 ( Pretend, q ) .
Solving the equation, we find that Individual 2 must choose to fight with probability
1
q = for Individual 1 to then be indifferent. By the symmetry of the game, Individ-
2 1
ual 1 must choose to fight with probability p = also. Now the two individuals are
2
using the same strategy in a Nash equilibrium. To see that this strategy is evolution-
arily stable we need to show that no individual with a mutation that makes it use an
224
DYNAMICS
alternative strategy can perform better than the individuals in the Nash equilibrium.
Let us confirm this with an example. What would happen if some individuals had a
mutation that made them fight with probability, say, 1/3 instead of 1/2? Because of the
mutation, we have two types of individuals within the species, the normal ones and the
mutants. We have to compute the payoffs for each individual in four possible pairings,
normal-normal, normal-mutant, mutant-normal and mutant-mutant, in order to see
which type has the higher payoffs, the normal or the mutant. For simplicity we will
compute the payoff for the first individual.
1 1 1 1 1 1 1 1
u1 ( Normal, Normal ) = × × ( −1) + × × 2 + × × 0 + × × 1 = 0.5 ,
2 2 2 2 2 2 2 2
1 1 1 2 1 1 1 2
u1 ( Normal, Mutant ) = × × ( −1) + × × 2 + × × 0 + × × 1 = 0.83 ,
2 3 2 3 2 3 2 3
1 1 1 1 2 1 2 1
u1 ( Mutant, Normal ) = × × ( −1) + × × 2 + × × 0 + × × 1 = 0.5 ,
3 2 3 2 3 2 3 2
1 1 1 2 2 1 2 2
u1 ( Mutant, Mutant ) = × × ( −1) + × × 2 + × × 0 + × × 1 = 0.78 .
3 3 3 3 3 3 3 3
The payoffs reveal the following: When a normal individual faces another normal indi-
vidual, it gets the same payoff (0.5) as when a mutant faces a normal individual; in these
confrontations no one has an advantage over the other. When a normal individual faces
a mutant, however, it has a better payoff than when a mutant faces another mutant
(0.83 versus 0.78). In other words, mutation makes the mutants worse off, and there-
fore that mutation will not spread out. The reader can confirm that the same thing will
happen with any mutation that gives a probability of fighting different from ½.
We still need to verify that the Nash equilibrium in mixed strategies is the only
possibility that prevents an invasion by mutants. In order to do this, it suffices to see
that any other strategy will be invaded by mutants. The arguments in the previous par-
agraph also work for showing this. Take the strategy “Fight with probability 1/3,” and
assume that the population consists of individuals playing it. The numbers show that a
mutation that programs individuals to fight with probability 1/2 (the probability in the
mixed strategies Nash equilibrium) will invade since it performs better (it performs the
same when facing a normal individual, but better against a mutant).
Definition of evolutionarily stable strategy: This last condition tells us that an evolu-
tionary strategy must be a Nash equilibrium, and the calculations above tell us that,
in addition, it must have better payoffs than a mutant within a population that mixes
normal with mutant individuals.
There is another, more complicated way of finding a different evolutionarily stable
strategy, but it requires that individuals can identify themselves as Player 1 and Player 2
in every encounter. If both are aware of who is who, they can use more strategies, since
they can condition the play in their role. For instance, they can use the strategy that
says “Fight if you are Player 1, pretend if you are Player 2.” One way of identifying the
roles is by size: the bigger individual is Player 1, though that is not the only possibility –
besides, we insisted that the individuals be about the same size. Another possibility
is that the incumbent (whoever found the resource) is labeled as Player 1, so long as
225
GAME THEORY
there are no additional considerations such as strength or size. This is an interesting

situation, since it leads to a very stable situation that is, in fact, often found in the
natural world. However, looking at the individuals and their possibilities to obtain the
resource, there are no clear reasons this needs to be the case. They both have about
the same chances, yet one concedes. Biologists have performed experiments to test
the extent of this behavior and, through some tricks, have been able to convince each
individual involved in confrontation that they are the incumbent, with the result that
they indeed fight.
The coordination behavior by which the newcomer (labeled Player 2) fights and
the incumbent (labeled Player 1) concedes leads to an unstable situation, since after
the encounter, the newcomer gets the resource and becomes the incumbent and the
former incumbent becomes the newcomer, so it could get the resource by fighting. This
completes a game theoretical explanation of a biological behavior. Thus, game the-
ory describes an explicit model to help in understanding the equilibrium mechanism
behind it. Biologist John Maynard Smith applied game theory extensively in order
to explain many other behaviors and strategies selected by evolution. The hawk-dove
game above is owed to him and his co-author George Price (Maynard Smith and Price,
1973). His book Evolution and the Theory of Games (Maynard Smith, 1982) is his best-
known work, and it is a classic reference in the literature of evolutionary games theory.
Despite the dynamics behind the definition of the evolutionarily stable strategy,
the definition itself is static. A proper dynamics should specify how the proportions of
normal and mutant individuals change throughout time beginning with one strategy,
with mutants first entering in small numbers until the equilibrium is achieved. Game
theory studies different dynamics rigorously as well as different definitions of evolu-
tionary stability, their connections, properties and explanatory power.
12.2 Imitation dynamics: A segregation model

The next game shows how urban segregation can arise within a population in which
individuals prefer to live in a diverse city or neighborhood as opposed to a homogenous
one. It is a simple version of Thomas Schelling’s segregation model, better known as
the Schelling Tipping Model, published in his book Micromotives and Macrobehavior
(Schelling, 1978).
Suppose that there are two cities, A and B, each with 1000 people. In A there are
600 inhabitants of Type X and 400 of Type Y, while the exact opposite is true for B.
All individuals prefer to live as part of a diverse city rather than in a homogeneous
one; however, and this is crucial, they do not like to live in a city where their group is
a minority. In particular, assume that the preferences of an individual are given by the
function in Figure 12.2.
The graph shows that, as the proportion of inhabitants of one’s own type increases
from zero to 50 percent, the utility of the individual increases from zero to one.
The maximum utility is achieved when the two types are represented in equal propor-
tions in the city. After that, the utility decreases as far down as ½ when all of the inhab-
itants are of the same type as the individual. These preferences can hardly be regarded
as discriminatory, since the perfect mix is regarded as the best situation. The only bias
in favor of one’s own group is that, given the choice of being in the majority versus
being in the minority, individuals prefer to live where their group is in the majority.
226
DYNAMICS
Utility
1/2
Proportion of individuals
of the own type
0% 50% 100%
Figure 12.2 Preferences for diversity
Let us go back to the cities and consider that individuals will be prone to moving
from one city into another if they observe that individuals of their own type are better
off in the other city (so that they imitate their choice of city). To make things simpler,
say that all who expect to be better off by moving will try to do so, but that each city
has room for only 1100 people. This is what will happen:
1. Each one of the 400 inhabitants that live as a minority in their respective city
observe that their kind is better off in the other city and will therefore try to move
there (in the utility function of Figure 12.2, observe that being in the 60% majority
is better than being in the 40% minority).
2. Since there are 1000 people in each city and we have said that there is room for up
to 1100, only 100 will be able to move. Say that a group of 100 individuals of Type
Y move from City A to City B. Now there are 600 Type X and 300 Type Y inhab-
itants in City A, and 400 Type X and 700 Type Y in City B.
3. Now City B is full, but there is room for up to 200 additional people in City A,
where 200 Type X can move from B. After this move, there will be 800 Type X and
300 Type Y in City A, and 200 Type X and 700 Type Y in City B.
4. Now City A is full, and 200 Type Y individuals in this city will move to City B.
After that, the remaining 200 Type X individuals in City B will move to A, and
finally, the last 100 Type Y individuals in City A will move to B.
The result is complete segregation in spite of the preference towards diversity. It was
enough that individuals had a slight preference of not being a minority for segregation
to occur. In fact, preferences for not being a minority can be much milder than in our
example and segregation would still occur.
The original model by Schelling is not much more complicated, but it will take
longer to run a simulation verbally. It goes like this: Begin with a surface divided into
small squares, like a checkerboard. We have many tokens that come in two colour, and
we must place each token in a different square. The number of tokens is smaller than
the number of squares, so there is room to move the tokens after they have all been
placed somewhere. If one token is surrounded by more tokens of a different colour, we
say it is “unhappy.” If there is about an equal number of tokens of the different colour
around it as there are of the same colour, then that token is “happy.” Finally, if it has
more of its own colour around, it is “indifferent.” From any random initial distribution
227
GAME THEORY
of tokens throughout the board, move them according to this simple rule: take any
“unhappy” token and move it to any square where it will be “happy” or “indiffer-
ent,” then repeat. The final distribution will show a board with tokens segregated into
“neighborhoods.” This occurs regardless of the precise limit as to what is considered
“almost equal numbers” of tokens of different colour surrounding any given one, and
for almost all initial distributions. The reader can find computer programs and web
pages for playing this game under different specifications and confirm that the final
result is almost always segregation.
The fact that one can find a dynamics that leads to segregation in a society where
individuals not only do not have preferences for discrimination but actually prefer
diversity does not mean that all causes for segregation are due to this kind of dynamics.
However, this fact does add a new cause, a potentially powerful one, which may have
important implications in making a correct diagnosis of the problem and in proposing
policies for preventing segregation. Schelling himself gave the example of Oak Park, a
suburb in the Chicago area, where city hall prohibited showing For Sale signs outside
houses in order to avoid the contagious effect. Properties could be sold, of course; the
prohibition was limited to this particular form of advertisement. City hall also offered
owners insurance against the possible loss of property value due to a change in the
racial diversity of the city. Other possible interventions under this dynamics may try
to establish diversity from the very beginning (e.g., mixing students when sitting in a
classroom or imposing quotas to guarantee the presence of different groups as candi-
dates in job interviews or as CEOs in big corporations).
12.3 Best-reply dynamics: The emergence of language

To present this dynamics, again consider the choice of standards game as seen in Sec-
tion 1.2, which we reproduce here for convenience in Figure 12.3a. Recall that the game
represents a simple coordination situation: both players win if they coordinate their
actions to complete a task, but they win even more if they coordinate in the better action.
2
Left Right
Left 2, 2 0, 0
1
Right 0, 0 1, 1
Figure 12.3a Choice of standards
If the two players can communicate with each other, they can easily agree on playing
(Left, Left); but what if they do not speak the same language? What if the players are
animals that have not yet developed a communication system? Let us continue making
assumptions. Before making his decision between “Left” and “Right,” Player 1 can choose
X or Y, a choice that Player 2 observes. If we incorporate these new actions we obtain the
game illustrated in Figure 12.3b, where L stands for “Left” and R stands for “Right.” We
have written 2.X for Player 2 who observes X and 2.Y for Player 2 who observes Y.
Actions X and Y can be anything, such as two different sounds or gestures. In fact,
they do not even need to be recognized as signals by Player 2. To simplify our analysis,
228
DYNAMICS
X Y
1 1
L R L R
2.X 2.Y
L R L R L R L R
2 0 0 1 2 0 0 1
2 0 0 1 2 0 0 1
Figure 12.3b The communication game
we will assume that the choices between X and Y, and between L and R by Player 1 are
made simultaneously. This shall be the case even if the physical actions of doing X or Y
go first and these are then followed by the choice between L and R, which is simulta-
neous for both players. It is important to note that, although Player 2 can distinguish
between actions X and Y before making any choice, he does not know whether Player 1
has chosen L or R when he himself chooses between these two options. The new game
is depicted in Figure 12.3c.
X,L Y,R
X,R Y,L
2.X 2.Y
L R L R L R L R
2 0 0 1 2 0 0 1
2 0 0 1 2 0 0 1
Figure 12.3c The simplified communication game
Next, we provide the normal form of the game in Figure 12.3d. In this representa-
tion, we marked six Nash equilibria using the technique of underlining the utilities in
the best replies: (XL, LL), (XL, LR), (XR, RR), (YL, LL), (YL, RL) and (YR, RR).
In four of them players coordinate in (Left, Left), though the bad coordination (Right,
Right) prevails in two.
Remarkably, we can add some dynamics to this game in which only equilibria in
the good coordination are selected. This dynamics just states that the best reply against
a given strategy is played increasingly by more players. For this purpose, we begin at
any selection of strategy, (YR, RL) for instance, which happens to give zero to both
players. In this strategy, Player 1 first emits sound Y and then plays R, while Player 2
plays R if she hears sound X and plays L if she hears Y. If individuals within a popula-
tion obey these rules, then from time to time (e.g., for fun, by mistake or just to see)
229
GAME THEORY
Player 2
LL LR RL RR
XL 2, 2 2, 2 0, 0 0, 0
XR 0, 0 0, 0 1, 1 1, 1
Player 1
YL 2, 2 0, 0 2, 2 0, 0
YR 0, 0 1, 1 0, 0 1, 1
Figure 12.3d The normal form of the simplified communication game
those in the role of Player 2 (the Player 1 role corresponds to the one that makes sound
X or Y first) can try a different strategy: LL, LR or RR. Those changing to LL will see
that their payoffs remain unaffected, while those playing LR or RR will observe their
payoffs increase from zero to one (recall that Player 1 keeps playing YR). Suppose that
out of these last two options (LR or RR), option LR is just by chance used more often,
so that in the end it is the only strategy chosen by players in the role of Player 2. Now
players are in the combination (YR, LR). At this moment, it is time for individuals in
the role of Player 1 to try some other options, and those trying XL will see their payoffs
go from one to two. From that point on, (XL, LR), no player can win by changing
their strategy choice. We can illustrate these dynamics by drawing some arrows in the
normal form, indicating the move towards a higher utility, as seen in Figure 12.3e.
Player 2
LL LR RL RR
XL 2, 2 2, 2 0, 0 0, 0
XR 0, 0 0, 0 1, 1 1, 1
Player 1
YL 2, 2 0, 0 2, 2 0, 0
YR 0, 0 1, 1 0, 0 1, 1
Figure 12.3e Best reply dynamics starting at (YR, RL)
Only one small additional detail is needed to complete the dynamics. We have
considered that individuals can weigh out different actions and that more individuals
end up using the strategies that give better results, but we can say a little more; players’
actions can drift between strategies that give the same payoffs. When this happens, we
simply draw double-headed arrows. For instance, had we started with the profile (YR,
RR), we could have had arrows pointing like in Figure 12.3f, where we show a path
from (YR, RR) to (XL, LR).
If we start from (YR, LR) there will be a path from there to (XL, LR) and another
one to (YR, RR). But observe that there is no way back from (XL, LR), while there is,
as we saw, a way to escape from (YR, RR).
230
DYNAMICS
Player 2
LL LR RL RR
XL 2, 2 2, 2 0, 0 0, 0
XR 0, 0 0, 0 1, 1 1, 1
Player 1
YL 2, 2 0, 0 2, 2 0, 0
YR 0, 0 1, 1 0, 0 1, 1
Figure 12.3f Best reply dynamics starting at (YR, RR)
The reader will find that by following this dynamics, any combination of strategies
will always lead towards one of the good equilibria. The last thing to check is that once
the players find the good way of coordinating, they remain there. This is illustrated in
Figure 12.3g.
To summarize: From any situation, including Nash equilibria in the bad coordina-
tion, we can find a way that leads to the equilibria in the good coordination (following
the arrows). If we are in a good coordination equilibrium already, however, we can only
go on to another good equilibrium. In other words, the good coordination absorbs the
rest of the strategies according to these best-reply dynamics. Individuals do not need
to be rational, but need only give more weight to the strategy that is the best reply in
the current situation (and take a random walk between best replies if there is more
than one).
Player 2
LL LR RL RR
XL 2, 2 2, 2 0, 0 0, 0
XR 0, 0 0, 0 1, 1 1, 1
Player 1
YL
2, 2 0, 0 2, 2 0, 0
YR 0, 0 1, 1 0, 0 1, 1
Figure 12.3g No way out of equilibria
All in all, the most interesting lesson to learn from this example is the fact that in
two of the four good equilibria reached using these dynamics, the sound X winds up
being associated with Left and sound Y with Right. In the other two equilibria it is
the other way around. We can interpret either situation as two sounds acquiring the
meanings “I will be playing L” and “I will be playing R.”
231
GAME THEORY
The above dynamics and example are adapted from Gilboa and Matsui’s article
“Social Stability and Equilibrium” (Gilboa and Matsui, 1991). The idea of using evo-
lutionary games for showing the emergence of language has its origin in the work
Convention by David Lewis (Lewis, 1969), in which the author proposes some other
dynamics in order to provide meaning to messages in a sender-receiver game.
12.4 No weakly dominated strategies dynamics: Self-

inflicted injuries
In Chapter 2 we described the process of iteratively eliminating dominated strategies
while also advancing the possibility of a similar process with weakly dominated strategies.
Recall that a strategy si of Player i is weakly dominated if another strategy si′ exists for the
same player such that, for all possible choices of strategies by the other players, the utility
of Player i is always higher or equal with si′ than with si, while for some choices it is strictly
higher. The process of iteratively eliminating weakly dominated strategies is as follows:
1. Find and eliminate all weakly dominated strategies for all players from the game.
2. In the game defined after following step 1, find and eliminate all weakly dominated
strategies.
3. Continue until no additional weakly dominated strategies can be eliminated.
As we saw in Chapter 2, this process can stop at a very early stage in many games,
though in some others it can go very far. Let us see one such example, where the pro-
cess can be seen as a dynamics. Consider the battle of the sexes game, reproduced for
convenience in Figure 12.4a.
2
Football Movies
Football 3, 1 0, 0
1
Movies 0, 0 1, 3
Figure 12.4a The battle of the sexes
Now introduce a seemingly awkward twist. In addition to choosing between Foot-

ball and Movies, Player 1 can burn an amount of money equivalent to one unit of
utility. The normal form of the game after Player 1 burns the money is shown in Fig-
ure 12.4b. The reader can easily see that this game has the same pure strategies Nash
equilibria as the original one: (Football, Football) and (Movies, Movies).
2
Football Movies
Football 2, 1 −1, 0
1
Movies −1, 0 0, 3
Figure 12.4b Player 1 burns a unit of money
232
DYNAMICS
To simplify matters, let us consider that Player 1 makes her decisions simultane-
ously. Player 2 observes whether Player 1 burns the money or not, but not where she
chooses to go. The extensive form is displayed in Figure 12.4c.
B,F N,M
B,M N,F
2.B 2.N
F M F M F M F M
2 –1 –1 0 3 0 0 1
1 0 0 3 1 0 0 3
Figure 12.4c The game of burning the money
Player 1 has four strategies: to burn the money and go to the football game (B,
F), to burn the money and go to the movies (B, M), to not burn the money and
go to the football game (N, F) and to not burn the money and go the movies (N,
M). Player 2’s strategies are go to the football game regardless of what Player 1 did
with the money (F, F), go to the football game if Player 1 burns the money and go
to the movies if she does not burn the money (F, M), go to the movies if Player 1
burns the money and go to the football game if she does not burn the money (M,
F), and go to the movies regardless of anything (M, M). Figure 12.4d shows the
normal form of the game.
Player 2
FF FM MF MM
BF 2, 1 2, 1 −1, 0 −1, 0
BM −1, 0 −1, 0 0, 3 0, 3
Player 1
NF 3, 1 0, 0 3, 1 0, 0
NM 0, 0 1, 3 0, 0 1, 3
Figure 12.4d The normal form of the “burning the money” game
Observe that there are four Nash equilibria in pure strategies: (BF, FM), (NF, FF),
(NF, MF) and (NM, MM), as marked in the figure. The four equilibria correspond
to the four combinations of the two equilibria in each of the games in Figures 12.4a
and 12.4b. Let us see what happens when we iteratively eliminate weakly dominated
strategies.
First iteration: Strategy BM is weakly dominated by strategy NF for Player 1. There
are no further weakly dominated strategies in this game. After its elimination, the new
game is the one in Figure 12.4e.
233
GAME THEORY
Player 2
FF FM MF MM
Player 1 BF 2, 1 2, 1 −1, 0 −1, 0
NF 3, 1 0, 0 3, 1 0, 0
NM 0, 0 1, 3 0, 0 1, 3
Figure 12.4e Elimination of weakly dominated strategies: first iteration
Second iteration: Strategy MM is weakly dominated by FM, and strategy MF is

weakly dominated by FF for Player 2. After their elimination, the game becomes the
one in Figure 12.4f.
Player 2
FF FM
BF 2, 1 2, 1
Player 1 NF 3, 1 0, 0
NM 0, 0 1, 3
Figure 12.4f Elimination of weakly dominated strategies: second iteration
Third iteration: Strategy NM is weakly dominated by BF. Without it, the new game
is shown in Figure 12.4g.
Player 2
FF FM
BF 2, 1 2, 1
Player 1
NF 3, 1 0, 0
Figure 12.4g Elimination of weakly dominated strategies: third iteration
Fourth iteration: Strategy FM is weakly dominated by FF for Player 2 (Figure 12.4h).
Player 2
FF
BF 2, 1
Player 1
NF 3, 1
Figure 12.4h Elimination of weakly dominated strategies: fourth iteration
234
DYNAMICS
Fifth iteration: Strategy BF is weakly dominated by NF (Figure 12.4i).
Player 2
FF
Player 1 NF 3, 1
Figure 12.4i Elimination of weakly dominated strategies: fifth and final iteration
Having concluded the iterative process, only one combination of strategies remains:
(NF, FF), which corresponds to the equilibrium that is most favorable for Player 1 in
the original game. Just by adding the possibility of burning money, Player 1 gets her
preferred equilibrium selected, and she does not actually need to burn the money.
The dynamics starts by taking into account that to burn the money and to then
play “Movies” is nonsense (her worse course of action). From there, Player 2 playing
“Movies” if he observes that Player 1 has burnt the money does not make sense either.
Once going to the movies in the case that the money is burnt is out of the question,
Player 1 will not go to the movies when she does not burn the money, since going to the
football game is a better option. Contrary to the previous dynamics, this one requires
very rational players with a high capacity for foresight.
12.5 Adaptive dynamics: Voluntary contribution to

the provision of public goods
The simplest way of defining adaptive dynamics for a two-player game is to consider
that, for starters, one player decides and announces her strategy. Next, the second player
chooses and announces one of his strategies which is a best reply against the choice of
the first player. After he does that, then the first player changes her first choice to a best
reply against the second player’s choice. The dynamics continues this way, with the play-
ers alternating in choosing a best reply against the choice of the other. This dynamics
requires some rationality in order to calculate the best reply, but it is limited in the sense
that players do not foresee the consequences that their choices have on the future selec-
tion by the other player. The first choice is not strategic; it merely represents a starting
point. Typical analysis using this dynamics includes the study of its convergence proper-
ties, such as the way the process leads to final strategy choices depending on the starting
conditions as well as the relation to other dynamics and equilibrium points of the game.
In this section, we will apply the adaptive dynamics to a game with infinitely many
strategies, specifically the neighbors’ game in Section 2.4. Neighbors can choose any
amount of time between zero and two hours to clean the street. Under these conditions,
1 − c2 1 − c1
we found the best-reply functions of the neighbors to be c1 = and c 2 = .
2 2
Remember that the best-reply functions tell us which is the best response of one neigh-
bor against the time choice made by the other. Additionally, they are equivalent to the
best replies described in Section 2.4 for the finite version of this same game as well as
the best replies in Table 3.2 for the finite version of the Cournot oligopoly.
235
GAME THEORY
The next step is to plot the functions into a graph. We do this in Figure 12.5.
The functions intersect at a point that satisfies the definition of Nash equilibrium:
Neighbor 1’s decision is the best reply against Neighbor 2’s, and vice versa. This point
is (0.33, 0.33).
C2
1 1 − C2
C1 =
2
0.5
1 − C1
C2 =
2
0.1
C1
0.45 0.5 0.8 1
Figure 12.5 Adaptive dynamics
The arrows show the adaptive dynamics in action. We can take any choice by
Neighbor 1 as a starting point, for example c1 = 0.8. The best reply by Neighbor 2
1 − 0.8 1 − 0.1
is c 2 = = 0.1. Neighbor 1’s best reply against 0.1 is c1 = = 0.45. Next,
2 2
1 − 0.45 1 − 0.275 1 − 0.3625
we find c 2 = = 0.275 , c1 = = 0.3625 , c 2 = = 0.31875 ,…
2 2 2
This series converges to c= 1 c=
2 0.33, the Nash equilibrium.
The dynamics in the finite version of the game

In Section 2.4, we introduced the finite version of the game, which we reproduce in
Figure 12.6 for convenience, where the strategies are zero, one or two hours to dedicate
to watching TV (while the remaining hours, up to two, to cleaning the street). The
Nash equilibria are (0, 0), (0, 1) and (1, 0).
Neighbor 2
0 1 2
0 2, 2 4, 2 6, 0
Neighbor 1 1 2, 4 3, 3 4, 0
2 0, 6 0, 4 0, 0
Figure 12.6 Cleaning the street game
Consider the adaptive dynamic starting with strategy 0 by Neighbor 1. Then, the
best reply by Neighbor 2 is 0 or 1. Every time there are multiple best replies, we have
to consider all possibilities.
236
DYNAMICS
1. Neighbor 1 chooses 0  Neighbor 2 chooses 0  Neighbor 1 chooses 0…

2. Neighbor 1 chooses 0  Neighbor 2 chooses 1  Neighbor 1 chooses 0  Neigh-
bor 2 chooses 1…
bor 2 chooses 0…
The three different possibilities converge to (0, 0), (0, 1) and (1, 0), respectively,
which are the Nash equilibria of the game. There are other ways in which the
dynamics alternates among different equilibria:
bor 2 chooses 0  Neighbor 1 chooses 0  Neighbor 2 chooses 0  Neighbor 1
chooses 1 …
If we start with Neighbor 1 choosing 1, we will obtain similar results: it is enough
to follow the previous processes from the point in which Neighbor 1 chooses 1 at
any point. If we start with Neighbor 1 choosing 2, the dynamics gives
bor 2 chooses 0…
After that point, it will follow any of the previous paths. In any case, it will lead to
a Nash equilibrium or to alternating among them.
For other games, convergence favoring one equilibrium or alternating among
different ones is not satisfied. In Exercise 12.10 we present such an example.

12.1 Consider the game of election of standards shown in Figure 12.3a.
(a) Find the Nash equilibrium in mixed strategies.
Let us confirm that the strategy that players use in this equilibrium is not an
evolutionarily stable strategy since it can be invaded by mutants playing “Left.”
(b) Find the payoff of an individual in a population that plays just as in the
equilibrium in (a).
(c) Find the payoff that an individual using the strategy in (a) gets when facing
a mutant that plays “Left.”
(d) Find the payoffs of a mutant when playing against a normal individual (who
plays as in the Nash equilibrium in (a)) and when playing against another
mutant.
(e) From your answers above deduce that the strategy in the Nash equilibrium
in mixed strategies is not an evolutionarily stable strategy.
12.2 In the game in Exercise 12.1 show that the strategy in the Nash equilibrium in
mixed strategies can also be invaded by a mutant playing “Right.”
12.3 In the game in Exercise 12.1 show that the strategy of playing “Right” cannot
be invaded by a mutant that plays the strategy “Left.”
12.4 In Schelling’s segregation game in Section 12.2, assume now that preferences of
the individuals are given by the function in the figure below.
237
GAME THEORY
Utility
1/2
Proportion of inhabitants
of one’s own type
50% 100%
(a) Show how an imitation dynamics like the one in Section 12.2 works. Be
specific about all of the steps until no individual wants to move.
(b) What is the final situation of the types?
12.5 In Schelling’s segregation game in Section 12.2, assume now that preferences of
the individuals are given by the function in the figure below.
Utility
1/2
Proportion of inhabitants
of one’s own type
50% 100%
With these new preferences, notice that an individual is better off being a part
of a 40% minority than a 60% majority. In addition, consider that the initial
situation of the cities is complete segregation (each city has only individuals of
one type).
(a) Show how an imitation dynamics like in Section 12.2 works. Be specific
about all of the steps until no individual wants to move.
(b) What is the final situation of the types?
12.6 Consider the game in Figure 12.4d.
(a) Show which equilibria are selected according to the best-reply dynamics
shown in Section 12.3.
(b) Interpret the result.
12.7 When playing the battle of the sexes in Figure 12.4a, Player 1 has the oppor-
tunity to go to a party with her friends, something that will give her a utility
of 2, as well as a utility of zero for Player 2. Player 2 observes whether Player 1
238
DYNAMICS
chose going with her friend or not, but not what her plan would be (going to the
football game or to the movies) in the case that she did not go to the party.
(a) Show the extensive form of the game and its corresponding normal form.
(c) Show which equilibria are selected according to the best-reply dynamics
described in Section 12.3.
(d) Interpret the result.
12.8 Apply the adaptive dynamics described in Section 12.5 to the Cournot model
in Section 3.6.
12.9 Consider the game in Exercise 12.1 and the adaptive dynamics in which Player 1
plays first.
(a) For which initial mixed strategies by Player 1 does the dynamics converge
to (Left, Left)?
(b) For which ones does it converge to (Right, Right)?
(c) For which ones does it converge to the equilibrium in mixed strategies?
12.10 Consider the matching pennies game, reproduced in the figure below, and the
adaptive dynamics where Player 1 chooses first.
2
Heads Tails
Heads 1, −1 −1, 1
1
Tails −1, 1 1, −1
(a) For which initial mixed strategies by Player 1 do the dynamics converge to
the equilibrium in mixed strategies?
(b) For which ones do the dynamics not converge?
239
13 LIMITED RATIONALITY AND
BEHAVIORAL ECONOMICS
Here we continue the presentation of alternatives to rationality that started in Chap-

ter 12. Now we focus on the changes in the utility function (Sections 13.1 and 13.2)
and in the way the information is processed (Sections 13.3 and 13.4), and, as usual, we
illustrate these ideas with some applications of interest (Sections 13.5, 13.6 and 13.7).
We present a discussion on utilitarianism and rationality in Section 13.8, and we finish
with an illustration on how to advance a behavioral theory that departs from rationality
and accounts for experimental evidence.
13.1 Preferences changing with time: Which ones deserve

priority?
To present the problem posed by this type of departure from the standard assumption
of rationality, let us begin with a series of simple examples. In the first one, Young-me
does not identify with the preferences of Old-me, even if he knows those preferences.
He does not care for retirement money, though he knows that later he will; he wants
to spend now. We represent the situation in Figure 13.1, along with the equilibrium.
Young-me
Spend Save
Old-me Old-me
Spend Spend
Save Save
3 2 2 1
1 2 2 3
Figure 13.1 The saving-spending game
In the next example, Young-me does identify with the preferences of Old-me. He does
not like contemporary music now, but he grows curious through seeing so many people
enjoying it. He believes that if he undertakes a course on it, he will end up listening to and
enjoying it, a prospect that he accepts. Figure 13.2 shows the game and the equilibrium.
Let us complicate the first situation. Young-me wants a good retirement, but he
knows that as soon as he starts making money, Later-me will be tempted to consume
considerably (C) rather than keep some savings (S) to the side. He also knows that he
will come to regret that in the future, when he becomes Old-me. Young-me must now
decide whether to make a commitment to save (e.g., he accepts a job and sets part of
240
LIMITED RATIONALITY AND BEHAVIORAL ECONOMICS
Young-me
Course No
Old-me Old-me
Listen Listen
No No
3 1 0 2
3 1 0 2
Figure 13.2 Learning to appreciate modern music
the money aside for retirement or he supports legislation for mandatory pension plans).
Once the commitment is made, it will be very difficult to reverse the situation (for
the sake of the game, let us say impossible or, at least, considerably expensive), and the
game will continue as in Figure 13.3 between Later-me and Old-me. The payoffs of
Young-me are the same as the payoffs of Old-me. It is Later-me who thinks differently.
This kind of situation introduces an ethical and political problem: if Young-me
does not have a means of committing to save, should the government force him? Game
theory does not answer this question; it can only help in understanding the conse-
quences of deciding one way or the other. The consequences are in regard to (i) what
the payoffs for each self in the equilibrium are and (ii) problems related to the consist-
ency of whose preferences are respected. Should the government’s policies be based on
respect for the preferences of the first self or the last? And based on what? Notice the
different consequences for the games in Figures 13.1 and 13.3.
Young-me
No Commitment
commitment to save
Later-me Later-me
C S S
Old-me Old-me Old-me
C S C S S
1 2 2 3 3
3 2 2 1 1
1 2 2 3 3
Figure 13.3 Commitment to save for old age
Here is a sample of the diversity of the situations. Whose preferences must be safe-
guarded in each case?
• Young-me wants to try a drug but does not want to become an addict (although
this is something that will happen with some probability after trying the drug).
The preferences of Later-me who is already an addict are different, since he wants
241
GAME THEORY
to continue consuming it. With some probability, Old-me will have gone through
rehabilitation and regret having consumed it the first time.
• Young-me wants to be environmentallyfriendly, but Later-me must make decisions
about work, consumption and investments, in which being environmentally friendly
is too costly.
• Young-me does not care about the environment, but Old-me does.
An adequate criterion that guarantees that some preferences take priority over others
must be very clearly spelled out and agreed upon in society, and they must be of use for
a wide range of situations. In the next section we will try to clarify some of the terms
in the discussion by showing the consequences of different approaches and policies.
13.2 Time inconsistency and energy saving

The most common way in which economics and game theory model intertemporal
choices is based on a discount factor, as we have already seen in repeated games and in
negotiations. If an individual has a discount factor of, say, 0.9 annually, this means that
she is indifferent between having one euro the following year and having €0.9 the cur-
rent year. We can compound the factor in order to compare other periods. For instance,
one euro in two years will be equivalent to 0.9 × 0.9 = 0.92 = 0.81 euros today. In gen-
eral, one euro in T years will be equivalent to 0.9T today.
This kind of intertemporal preference shows the important property of temporal
consistency. In order to understand it, consider that an individual with a discount
factor of 0.9 is offered the choice between one euro in four years and 1.3 euros in six
years. The first option is equivalent to 0.94 = 0.656 euros today, and the second, to
1.3 × 0.96 = 0.69 euros, also today. She should choose the 1.3 euros in six years. After
four years we can ask her again in case she has changed her mind. Would she prefer
the one euro at that time or would she prefer waiting two more years in order to obtain
1.3? According to her preferences, waiting two more years for 1.3 euros is equivalent to
1.3 × 0.92 = 1.05 , so she prefers to wait. She does not change her mind and her prefer-
ences are time consistent. The reader may have observed that, using discount factors
comparing the equivalences now or later will always give the same results, since the
only difference in the two comparisons is that, by comparing later, we just simplify the
expression by removing the same quantity from both sides of the equation.
More formally, suppose we want to compare quantity M in T years to quantity
N in T + S years using discount factor δ . The first option is equivalent to M × δ T
euros today, and the second to N × δ T +S . The first is better than the second if
M × δ T > N × δ T +S , but this is true if M > N × δ , which is the comparison the indi-
S
vidual will be making in T years.

This time consistency is not satisfied by other possible temporal preferences. Here
is a simple example: given the choice between 10 euros today or 11 euros tomorrow, a
person declares that he prefers the 10 euros today. If this same person prefers 11 euros
in 366 days as opposed to 10 euros in 365 days, he will be showing time inconsistency,
since in 365 days he will change his mind and will take the 10 euros instead of waiting
one more day for one additional euro. According to some studies, we tend to show this
kind of inconsistent preference in regards to events happening in the far future. These
preferences can be explained if we assume that the discount rate decreases with time.
242
Preferences like these maybe explain why it sometimes happens that today we wish
things for the future but do not take the actions consistent with those desires when the
time comes. The next example shows some economic consequences of all this.
A consumer is thinking of buying a home appliance. The first option is a device
that costs €100 and that will consume €10 a year in electricity. The second device costs
€70, but its electricity consumption is €15 a year. Conscious about pollution and CO2
emissions, the government adds a 100% tax on energy. The appliance has a four-year
lifespan. Let us compare a consumer’s decision with both a constant discount factor as
well as a decreasing one.
Constant discount factor of 0.9:
• If she buys the more energy-efficient device, its present discounted cost will be
100 + 0.9 × 20 + 0.92 × 20 + 0.93 × 20 + 0.94 × 20 = 161.9 euros.
• If she buys the cheaper, yet less-efficient one, the present discounted cost will be
70 + 0.9 × 30 + 0.92 × 30 + 0.93 × 30 + 0.94 × 30 = 162.8 euros.
• She will buy the more expensive, more efficient appliance.
Decreasing discount factor: 0.9 the first year, 0.8 the next, 0.7 and 0.6 the third and
fourth.
• The present discounted cost of the efficient device will be 100 + 0.9 × 20 + 0.9
× 0.8 × 20 + 0.9 × 0.8 × 0.7 × 20 + 0.9 × 0.8 × 0.7 × 0.6 × 20 = 148.53 euros.
• The present discounted cost of the cheaper device will be 70 + 0.9 × 30 + 0.9 ×
0.8 × 30 + 0.9 × 0.8 × 0.7 × 30 + 0.9 × 0.8 × 0.7 × 0.6 × 30 = 142.79 euros.
• With these preferences, the consumer will buy the less-efficient appliance.
Observe that the decreasing discount factor effectively represents a time-inconsistent
consumer. Given the choice between €1 in three years and €1.5 in four years, she will
prefer the first option, with a discounted value of 0.9 × 0.8 × 0.7 ≈ 0.5, higher than
the discounted value of the second option, 1.5 × 0.9 × 0.8 × 0.7 × 0.6 ≈ 0.45 . However,
if she were to be asked again after three years, she would then change her mind and
would prefer waiting one more year, since €1.5 the following year is equivalent to
0.9 × 1.5 = 1.35 euros today, which is clearly more than €1 today. This is because one
euro the following year is equivalent to 0.9 euros today, while one euro in four years is
equivalent to just 0.6 euros in three years.
There is not much sense in saying that some preferences are more correct than
others. The most we can say is that the consumer will use the time preferences she
has at the time of making the decision and that this may imply that she will regret
her decision in the future, showing a time inconsistency. If the government decides
upon encouraging the consumer to buy the more efficient device, we can also say that
it would be going against the current preferences of the consumer; however, the con-
sumer would accept the government’s policy as time goes by. Game theory does not
state how time inconsistency must be resolved in this game between the consumer and
her later self, but it does clarify the terms of the debate. If consumers realize that they
are inconsistent and also accept that they do not want to be so, then they may approve
the passing of a law that promotes buying the efficient, but more expensive device. This
is the equivalent to the situation in the game in Figure 13.3.
243
GAME THEORY
A decreasing discount rate has consequences when finding the optimal taxation
for pollution (optimal from the perspective of the consistent consumer). For instance,
Tsvetanov and Segerson (2013) study the evaluation of policies for improving energy
efficiency, such as the use of standards, Pigouvian taxes (on pollution) and taxes and
subsidies for certain goods. The authors show that when the government knows the
proportions of occasional and frequent users of durable energy-consuming goods, a
combination of Pigouvian taxes (a tax on pollution-causing activities) and the estab-
lishment of standards is a better policy than just the Pigouvian tax, which is the optimal
policy when consumers are time consistent. Thus, these two instruments are comple-
mentary rather than substitutes for each other.
13.3 Irrationality due to the complexity of the election

A consumer faces the decision of paying for something that she is buying today either
now or the next year. The seller offers her a special promotion to postpone the pay-
ment: a very low interest rate and a small commission. Paying today has the benefit of a
discount price. Confused by all the information, the consumer does not know what is
best, as she does not know what calculations she needs to make in order to check which
option works better for her time preferences. To fix ideas, suppose that the price of the
good is €100, the interest rate is 1%, the commission for signing the contract to delay
the payment is €5 (to pay now), and the price discount for paying today is 3%. Suppose
also that the consumer’s discount factor is 0.9. The consumer’s decision is shown in
Figure 13.4.
Consumer
Pay now Postpone
€100 now €100 in one year

minus 3% plus 1% interest
plus €5 now
Figure 13.4 A complicated election
If the consumer does not know what to make of all of this data in order to find
the best option according to her own preferences, then the probability of making the
wrong decision will be high. The consumer could contract the services of a consultant
who would tell her something such as “If you pay now, you pay 100 − 0.03 × 100 = 97
euros. If you delay the payment for one year, however, then the present discounted
value according to your preferences is 5 + 0.9 × (100 + 0.01 × 100 ) = 95.9. Your best
choice is to postpone the payment and save the equivalent of €1.1 today.” However, the
consultant fees will most likely be higher than the cost of making the wrong decision.
In many countries, there are regulations under which it is mandatory that for
delayed payments all costs (commissions, interests, discounts, etc.) are summarized
in an implicit, equivalent rate. This rate is calculated as follows. First, we will see that
there is a connection between the discount factor and the interest rate. Suppose that
244
the interest rate is 5% (in mathematical terms the rate is written as 0.05). This means
that 1 euro in the bank gives 1.05 euros after one year. An individual with access to the
financial market will be indifferent as to 1 euro today or 1.05 euros the following year,
since he can convert one into the other simply by borrowing or lending money. If 1 euro
today is equivalent to 1.05 euros the following year, this means that the discount factor
1
is calculated to satisfy 1 = δ × 1.05, or δ = = 0.95. In general, an interest rate
1 1 + 0.05
r implies a discount factor δ = . Now, in our example, the seller can either charge
1+ r
97 euros today or 5 euros today plus 101 euros after one year. The implicit discount fac-
tor in the option to delay payment must satisfy 5 + δ × 101 = 97, which gives δ = 0.91.
This means that the implicit interest rate the seller is applying in his offer satisfies
1
0.91 = , or r = 0.098 (9.8%). The consumer can now compare this interest rate
1+ r
with those of other forms of payment or alternative offers by other sellers, as well as
1
with her own implicit interest rate after her discount factor: 0.9 = , giving r = 0.11,
1+ r
an 11% interest rate. Notice that the seller is applying a lower interest rate; therefore,
the buyer should be interested in the offer.
The consumer will welcome the sort of regulation that standardizes all con-
tracts in the sense that they must all give the information in the same manner; in
that way the consumer will be able to compare numbers that refer to exactly the
same concept in each contract. She will accept the regulation even if this implies
that the standardization may conceivably impede the addition of some favorable
clause. The benefits of the standardization will easily compensate for that potential
loss. Furthermore, even if the legislation allows for the signing of non-standard
contracts, it may still be of benefit to the consumer by requiring that a standard
one must be offered also.
13.4 Irrationality due to overconfidence

Every year I consider an offer to sign up for gym membership and accept it after over-
confidence makes me believe that I will attend regularly, even if in the end I do not.
In other words, I am overestimating the offered contract. The opposite is also possible.
For example, I can underestimate the value of car insurance because I underestimate
the possibility of having to file an accident report. Overconfidence can take several
forms. One of them is excessive optimism, such as when we overestimate our ability of
committing ourselves to one task (such as going to the gym) or of paying attention to
the numerous details of a contract (e.g., knowing the balance of our checking account
at all times so as to avoid overdraft charges). Even if we are not biased in estimating the
average use of a service, we may be biased in estimating its variance. This illusion of
precision is a second form of overconfidence.
Psychologists have documented overconfidence for decades, but here we will be
specifically interested in a consumer’s overconfidence in the market. Grubb (2015)
reviews field studies and laboratory experiments on this topic and finds that, in
effect, consumers consistently show excess confidence. Moreover, studies show that
this overconfidence does not disappear easily. Some works find that learning with the
245
GAME THEORY
appropriate suggestions can mitigate overconfidence, but also that companies have no
incentive for providing this type of learning. In this line, laboratory experiments show
that the suggestions clients receive are ineffective for all purposes, while field studies
show that learning is not sufficient so as to avoid mistakes (e.g., choices improve too
slowly or lessons received are forgotten). After these considerations, Grubb examines
the following three questions.
What will companies do to exploit customers’ overconfidence?

One strategy that companies use for taking advantage of customers’ overconfidence is to
reduce the initial fixed payments and increase the price per use when customers under-
estimate consumption. For example, suppose that a consumer pays a registration fee of
€2 and then €1 per unit of consumption. If she uses two units she will end up paying €4
(two from the fee and one for each unit), but if she makes the mistake of thinking that
she will only use one unit, she will mistakenly believe that she will end up paying €3.
Under these conditions, if the company changes its pricing policy to stop charging a fee
and instead raises the price of each unit to €3, the consumer will again expect to pay €3
when in reality she will end up paying €6. When the consumer overestimates the amount
of use, the incentive is then the opposite: raise the quota and decrease the price per use.
One of the most important aspects of this complex pricing strategy is that it stands
even in very competitive markets. To understand this, consider the following exam-
ple. A mobile phone service has a fixed cost of €40 and marginal costs of 10 cents per
minute. Consumers value the service at 45 cents per minute up to a satiation point
(which can be 100, 400 or 700 minutes, with equal probabilities); due to overcon-
fidence, however, they believe that the satiation point is 400. If prices are equal to
the costs (as would be the case in a competitive and therefore efficient market) each
consumer would expect from this service a benefit of 400 × ( 0.45 − 0.10 ) − 40 = 100
euros (the term 0.45 − 0.10 is the difference between the value of one minute for the
consumer and the cost). However, this situation cannot be an equilibrium: another
company can offer a contract with a fee of €60, 400 free minutes and 45 cents per
minute thereafter. This way the consumer expects a profit of 400 × 0.45 − 60 = 120
euros, when in reality she will only have 75 euros. To see this, notice that with prob-
ability one she will enjoy 100 minutes for free, with probability 2/3 she will enjoy
another 300 minutes also for free, and with probability 1/3 she will consume 300
minutes at 45 cents each. After subtracting the €60 fee, this amounts to a benefit
2 1
of 100 × 0.45 + × ( 400 − 100 ) × 0.45 + (700 − 400) × ( 0.45 − 0.45 ) − 60 = 75, while
3 3
the company will earn €25 more.
Which are the welfare consequences of overconfidence?

Consumers who derive a high utility from the contract will benefit from the compe-
tition, since lower prices will always imply that the same utility can be achieved with
the new prices. However, when consumers for whom the expected benefits are not very
high are overconfident, a higher price may imply that they will not sign the contract.
This might be a better option than signing only to then end up paying more than
expected due to that overconfidence. Thus, policies aimed at protecting consumers
from overconfidence can have little effect if competition makes them end up signing
246
inefficient contracts with lower, luring prices. It can be shown, however, that harm to
the consumer is limited when the impact rate is high (the fraction of the increase in cost
that is passed to the consumer in the form of a price increase) and when the elasticity
of demand in regards to the price is low (buyers do not react much to change in prices).
If there are rational consumers, those who show overconfidence may be subsidizing
them. An example can explain this: An overconfident consumer may be paying €100 in
the form of penalties for an overdraft in his checking account. If every client were like
him, the banking competition would charge a fee per checking account €100 below
its cost and would recover the money with overdraft penalties. If only 50% of clients
are like this and the rest are rational, then banks will reduce the current account fee
by €50 to all customers. This way, rational customers are being subsidized. Stango
and Zinman (2009, 2014) estimate that $32 billion are paid in the United States to
cover penalties and that these correspond to less than half of current account holders.
In addition, given that those who pay the penalties are mostly low-income individuals,
this price policy is regressive.
Which are the implications for public policy?

As we have seen, the transparency of terms in the contract is not enough. In cell phone
services, for example, three measures of transparency are added: (i) the seller can report
the average cost for customers having a certain plan; (ii) if there is already a previous
relationship, then the seller can give the buyer an estimate based on the previous use
and (iii) the seller can allow the buyer to share his use and payment history with a third
party in order to find the best plan in the market.
A second policy allows the use of information for helping make better use of the
plan. For example, in the United States and the EU, customers should be warned of
roaming fees when they cross borders, and U.S. customers are alerted when they come
close to reaching the limit of free or inexpensive units or when they exceed it. These
transparency policies can solve the attention problems. The extension of this kind of
policy onto banking services could provide substantial benefits also.
Finally, one may try some type of price regulation, but this policy requires a lot of
information in order to be practical. It is possible, however, to avoid certain practices,
such as excessive penalties for small overdrafts in a checking account. Other regula-
tions, such as prohibiting three-part tariffs, can have some undesired effects.
Anticipating business reaction is an important part of any policy. U.S. telephone
companies were forced to alert customers as they approached the limit of free minutes.
In the end, however, the response of the companies was to increase their monthly fees,
thus eliminating the advantage of this policy for consumers. On the other hand, when
a similar policy was adopted for the use of credit cards, the reduction for consumers was
$11 billion a year. The reason for this difference between sectors is due to the inelastic
demand for telephone services as well as its high rate of cost impact (close to 100%).
In the case of credit cards, the opposite occurs.
13.5 The age of majority

Child custody defines the legal status of the relations between parents and their chil-
dren until the age of majority. The legal system considers that a minor does not have
the capacity of making certain decisions and allows a parent to make them on their
247
GAME THEORY
behalf. Even if the age of majority is 18 years, as is the case in most countries, the law
usually grants a minor the right of making their own decisions, or at least the right to
have their opinions considered, in certain cases. The age of consent for having sexual
relations or getting married, for instance, is usually lower than the age of majority.
Other examples include the decision to undergo risky surgeries, terminate a pregnancy
or choose which parent they want to act as custodian.
The criterion to elucidate which decisions the minor can make must be decided upon
in the best interests of the minor. How can this be done? If we were to ask the minor or
the parent, each of them would most likely prefer making all of the decisions. The key
may be finding out whether it is the current preferences of the minor or those of the par-
ent that will most adequately represent the minor’s interests at the age of majority.
We can safely assume that a person at the age of 18 prefers having received an edu-
cation, been vaccinated and been provided healthcare. Had minors been asked whether
they wanted all of this at an early age, it is likely that they might have refused. In such
cases, it is better to leave these decisions to the parents; even an 18-year-old will be
happy that he or she was not allowed to make those decisions as a child.
In cases such as marriage and risky surgery, it may be more likely that the individ-
ual at the age of 16 (me-16) has preferences similar to that same individual at age 18
(me-18), and that this similarity compares favorably in relation to the preferences of
the parents. The experience gathered from research with teenagers, expert opinion and
other sources may help determine whether a given decision problem is indeed solved
better by letting the minor or the parent have the last word. Let us develop and clarify
these ideas with the use of games.
Education
A very young minor does not value education and therefore cannot grasp its importance
for her or his formation and future. They would probably prefer spending more time
with their parents and playing. When they turn 18, the age at which society respects
the preferences of the individual, they would come to regret not having been obligated
to study. The minor’s decision is a very simple single-player game, but with conse-
quences for two actors: me-minor and me-18. It is represented in Figure 13.5a, where
the first number is the payoff for the minor and the second is the payoff for me-18.
Minor
Education Playing
0 1
1 0
Figure 13.5a The child does not like education
If the parents, as opposed to the child, make the decision and if they identify
with the preferences they think the child will have at age 18, we have the game
in Figure 13.5b. From top to bottom, payoffs are for the parents, me-minor and
me-18.
248
Parents
Education No education
1 0
0 1
1 0
Figure 13.5b Parents want to educate the child
Not all parents internalize the well-being of the child and may therefore make the
wrong decision. In this case, the government can intervene and rule that the education
of the minor is mandatory. We see this in Figure 13.5c, where p is the probability that
the parents are responsible, while 1 − p is the probability that they are irresponsible and
do not care for the child’s education. There may be other possibilities, but these two
suffice for building a game. First the government decides on whether to make educa-
tion mandatory. If it does not, it is left up to the parents whether or not to educate the
child after Nature (N) dictates whether they are responsible or irresponsible. The order
of the payoffs is government, parents, me-minor and me-18. E stands for “Education”
and NE for “No education.”
Government
No mandatory Mandatory
education education
p 1−p
1
Responsible Irresponsible 1
parents parents 0
1
E NE E NE
1 0 1 0
1 0 0 1
0 1 0 1
1 0 1 0
Figure 13.5c The case for mandatory education
Against the possibility that parents may be irresponsible, the best action for the gov-
ernment is to make education mandatory, since this gives the government and me-18 a
utility of one as opposed to p. The subgame perfect Nash equilibrium is marked with
arrows.
The minor’s consent

Even if the age of majority is 18 (or any other age, depending on the country), the
law typically gives minors certain rights on decisions that affect them. A country may
249
GAME THEORY
respect the preferences of a 12-year-old minor when assigning child custody to one
of the parents in the case of a divorce. It may also require the consent of 15-year-olds
before having risky medical procedures, or it may allow 16-year-olds to get married and
have sexual relations but require them to be 18 in order to have an abortion. The justi-
fications for the different ages and decisions are many. In certain cases, legislators may
invoke the maturity of the minor, for example, a minor may be declared mature enough
to decide over his health but not mature enough to choose political representatives.
In other cases, legislation is based on moral principles (difficult decisions correspond
to the person who has custody of the minor) or appeals to a syllogism (if the minor is
mature enough to get married then she is also mature enough to decide over her preg-
nancies). In this section we propose an approach based not on a priori considerations,
but instead on the similarity between the preferences of me-minor and me-18 (18 years
being the age of majority). Consider a minor at the age of 16; who will have the more
similar preferences to her two years later? Me-18, the parents or the state? We will later
discuss how to elucidate the way preferences change between 16 and 18. In the present
discussion we are interested in clarifying the terms of the debate over giving the minor,
the parents or the state the right of making certain decisions that affect that minor.
Take the case of a major surgery or the termination of a pregnancy. Comparing the
preferences of she-16 and she-18 there are four possibilities: (i) both she-16 and she-18
prefer the intervention, (ii) she-16 prefers the intervention, but she-18 does not, (iii) she-
16 does not want the surgical intervention, but she-18 does want it and (iv) neither of
them wants the intervention. To simplify matters we will construct a particular exam-
ple in which the probabilities of the four cases are 0.4, 0.3, 0.2 and 0.1, respectively. If
the law grants the decision to the minor we have the situation shown in Figure 13.6a.
First, Nature determines which of the four cases prevails, and then the minor chooses.
The first payoff corresponds to the minor and the second to she-18. To simplify even
further we will consider only zero and one as possible payoffs, but we could use differ-
ent numbers for the different cases and different selves.
N
0.4 0.1
0.3 0.2
She-16 She-16 She-16 She-16
Yes No Yes No Y es No Yes No
1 0 1 0 0 1 0 1
1 0 0 1 1 0 0 1
Figure 13.6a Preferences and decisions when facing a medical intervention
With these probabilities and payoffs, she-16’s expected utility before knowing which
of the four possibilities applies is 0.4 × 1 + 0.3 × 1 + 0.2 × 1 + 0.1 × 1 = 1, while the utility
of she-18 is 0.4 × 1 + 0.3 × 0 + 0.2 × 0 + 0.1 × 1 = 0.5. Another way of reading these num-
bers is by saying that she-18 will agree with she-16 only 50 percent of the time.
Let us now take a look at the situation in which the state leaves the decision to the
parents. Again, we may have different kinds of parents: some will listen to their child,
others will try to put themselves in the shoes of she-18, some others will make the
250
decision one way or the other no matter what, and still others will consult a specialist or
a moral advisor, and so on. In our example, consider just four possibilities: (i) the par-
ents choose according to the preferences of she-16, (ii) the parents choose according to
the preferences of she-18 (that they can guess with no mistake), (iii) the parents choose
“Yes” and (iv) the parents choose “No.” Suppose that the probabilities of these cases are
0.3, 0.2, 0.2 and 0.3, respectively. This situation is shown in Figure 13.6b.
0.3 0.3
0.2 0.2
Parents use Parents use Parents Parents

preferences preferences choose choose
of she-16 of she-18 Yes No
Figure 13.6b Parents make the choice
If parents choose according to the preferences of she-16, the game continues as in

Figure 13.6c, where Nature (N) firsts determines the preferences for she-16 and she-
18. The parents then choose by using the preferences of she-16. Payoffs first show the
parents’ preferences (P = 16) that coincide with the preferences of she-16, then the
preferences of she-16 and finally, those of she-18.
0.4 0.1
0.3 0.2
P = 16 P = 16 P = 16 P = 16
Yes No Yes No Yes No Yes No
1 0 1 0 0 1 0 1
1 0 1 0 0 1 0 1
1 0 0 1 1 0 0 1
Figure 13.6c Parents use preferences of she-16
Parents will choose as indicated with the arrows. The expected utility of she-
16, the same as that of the parents, will be one, while the utility of she-18 will be
0.4 × 1 + 0.3 × 0 + 0.2 × 0 + 0.1 × 1 = 0.5 .
If the parents choose according to the preferences of she-18, the game contin-
ues as indicated in Figure 13.6d. The only difference with respect to Figure 13.6c
is that the parents (now P = 18) have the same preferences as she-18. Now the
utility of the parents and of she-18 will be one, while the utility of she-16 will be
0.4 × 1 + 0.3 × 0 + 0.2 × 0 + 0.1 × 1 = 0.5 .
If both parents prefer “Yes,” then the game continues as in Figure 13.6e. The par-
ents get utility one if they choose “Yes” and zero if they choose “No.” Now the parents’
utility is one, the utility of she-16 is 0.4 × 1 + 0.3 × 1 + 0.2 × 0 + 0.1 × 0 = 0.7 and the
utility of she-18 is 0.4 × 1 + 0.3 × 0 + 0.2 × 1 + 0.1 × 0 = 0.6.
251
GAME THEORY
0.4 0.1
0.2
0.3
P = 18 P = 18 P = 18 P = 18
1 0 0 1 1 0 0 1
1 0 1 0 0 1 0 1
1 0 0 1 1 0 0 1
Figure 13.6d Parents use preferences of she-18
0.4 0.1
0.3 0.2
P = Yes P = Yes P = Yes P = Yes
1 0 1 0 1 0 1 0
1 0 1 0 0 1 0 1
1 0 0 1 1 0 0 1
Figure 13.6e Parents choose Yes
Finally, if the parents prefer “No,” we have the game in Figure 13.6f. The parents
get utility one if they choose “No” and zero if they choose “Yes.” The expected utility
of she-16 is 0.4 × 0 + 0.3 × 0 + 0.2 × 1 + 0.1 × 1 = 0.3, while the expected utility of she-18
is 0.4 × 0 + 0.3 × 1 + 0.2 × 0 + 0.1 × 1 = 0.4.
We are now in position to answer the question of how government actions depend
upon its objectives. If the goal is to respect the parents’ wishes, then it simply legislates
accordingly. The same will happen if the government is only interested in respecting
the preferences of the minor. An interesting case occurs when the government wants
to respect the preferences of she-18. In the example, with those specific probabilities
and utilities, leaving the decision to she-16 gives she-18 utility 0.5, as we saw after
solving the game in Figure 13.6a. If the decision is left to the parents, she-18 will have
an expected utility that is the sum of the utilities found in games in Figures 13.6c,
13.6d, 13.6e and 13.6f, each of them multiplied by the corresponding probability
as indicated in Figure 13.6b: 0.3 × 0.5 + 0.2 × 1 + 0.2 × 0.6 + 0.3 × 0.4 = 0.39, which is
lower than for the case in which she-16 makes the decisions. If the probabilities or
utilities were different, the result might also be different.
We have saved the question of how to know the preferences of she-18 until the
end. For obvious reasons, this is a piece of information we do not have at the moment
of making the decisions. However, one can collect statistical information about the
circumstances and frequency of changing preferences between the ages of 16 and 18,
252
0.4 0.1
0.3 0.2
P = No P = No P = No P = No
0 1 0 1 0 1 0 1
1 0 1 0 0 1 0 1
1 0 0 1 1 0 0 1
Figure 13.6f Parents choose No
as well as the percentage of cases in which individuals come to regret or approve of the
decisions taken by them or on their behalf when they were young and the intensity of
that regret or approval. With this information we can estimate at what point the regret
is smaller – when the decision is taken by the parents or by the minor?
13.6 Indoctrination
When indoctrination occurs it is impossible to elucidate things as in the previous sec-
tion. If the decisions of the parents include choosing whether to indoctrinate the minor,
we can have the following situation:
• On one hand, the preferences of the child without indoctrination are closer to his
or her own preferences when the child reaches the age of majority than to the pref-
erences of the parents.
• On the other hand, the preferences of the indoctrinated 18-year-old are closer to
those of the indoctrinating parents than to those of the non-indoctrinated minor.
In the situation described, it is not possible to define what is best from the point of
view of the future person at the age of 18. The indoctrinated child prefers to have been
indoctrinated while the non-indoctrinated child prefers not to have been. Is it possible
to answer this question? Is it possible to rationally reject indoctrination? How can one
convince a society of indoctrinated people to stop indoctrinating their children?
Thus stated, the problem does not have a solution. The indoctrinated society will
believe that it is a mistake not to indoctrinate, while the non-indoctrinated society will
see the error in the other. The key is that societies are not exactly like that. It is likely
the case that indoctrination is never 100% effective. Elaborating on this circumstance,
we can easily think that, although the majority of those indoctrinated want indoctri-
nation, and that the majority of the non-indoctrinated prefer non-indoctrination, there
will surely be a higher percentage of indoctrinated who rebel than of non-indoctrinated
who voluntarily embrace indoctrination.
If that is the case, there will be a social dynamics that, in the medium or long term,
will erode the acceptance of indoctrination. Many circumstances can affect rebellion
against indoctrination. The comparison between ways of life can favor societies with
less indoctrination. In this case, access to this information will favor rebellion. If the
253
GAME THEORY
indoctrination goes against important aspects of human nature (e.g., one that separates
the children from their parents), makes claims that science contradicts, or produces a
stagnation of society, it will easily be the case that the indoctrinated society performs
worse than the non-indoctrinated on many levels.
Thus, the solution to the problem is not an a priori reasoning that allows establish-
ing the logical error of indoctrination, but empirical evidence that human beings live
better without it.
13.7 Nudging: When to let others influence you

That the postulate of rationality is unrealistic is something that economists have always
known. The problem was how to replace it with a more realistic one for situations in
which studying a particular economic problem under the rationality assumption is not
a good approximation. Once rationality is defined, there is one way of being rational,
but there are virtually infinite ways of being irrational. Oftentimes, criticism against
the rationality postulate or alternative models of economic analysis not based on ration-
ality is, in turn, based on a postulate of irrationality that is even more unrealistic. It
is very easy, and intellectually and scientifically useless, to postulate the irrationality
necessary in order for a model to reach the desired conclusion. The works on behav-
ioral economics initiated by psychologists Kahneman (winner of the Nobel Prize in
Economics for advances in this line of research) and Tversky managed to systematize
the irrationality inherent in human decisions. To say that we are often irrational is not
important, what matters is to show that we are systematically obeying a series of iden-
tifiable cognitive biases.
Based on these works, authors like economist Richard Thaler and legal scholar
Cass Sunstein propose some public interventions that delicately push the decisions of
citizens in the right direction, defined according to the citizen’s own preferences. Their
book, Nudge: Improving Decisions about Health, Wealth, and Happiness (Thaler and
Sunstein, 2008) is the first and best-known reference on the subject. A nudge may
consist of simple and unproblematic regulations such as proposing that, in restaurants
and s elf-services, salads are more visible than less healthy alternatives, since studies have
shown that this measure increases their consumption. Defenders of this idea argue that
it does not deny or disproportionately hinder the option of any dish; however, it pro-
motes the consumption of healthy food, which is something that the average consumer
generally wants but does not always achieve due to giving in to temptation.
There are examples in which the nudge carries more consequence. Take, for
instance, the policy that banks must offer a default mortgage that includes the clause of
dation in payment. This clause in a mortgage contract implies that, in case of default,
the bank can only claim the payment of the amount of the debt up to the limit of the
value of the property before the law. If the price of the property is lower, then the bank
loses the difference. This clause will typically increase the cost of the mortgage (includ-
ing interests and commissions), but the consumer will be insured against a decrease
in the value of the property. Including this clause in the contract by default would be
justified if the consumer were not able to calculate the risks at the time of making the
contract or did not understand the convenience of this clause despite the higher cost.
In any case, in the spirit of the nudge idea, the consumer would still be free to contract
the mortgage without that clause.
254
Therefore, in order for the nudge to make sense, it must be the case that the default
option is the most advantageous for the consumer and that the other options remain
available. The justification that the default position is the most advantageous must be
documented, in addition, by the preferences expressed by the consumers themselves,
who must be aware that this policy is being taken for the aforementioned reasons.
However, economists and philosophers continue to debate the morality and conven-
ience of this type of policy. Although we have been able to define when to allow the
nudges, things will not be so clear in practice and we could be justifying the manipu-
lation of behaviors.
13.8 On other irrationalities that are not so irrational

Used profusely in both economics and game theory, utilitarianism is often criticized as
irrational or, at least, as a very incomplete rule for analyzing moral problems. The men-
tal experiments of the trolley problem type serve as a discussion model.
A runaway trolley is about to roll over and kill a group of five people. The only
possibility of saving them is to divert the tram onto another track in which there is only
one person who will die. What should be done if one could press a button that would
make the trolley switch tracks?
This kind of problem shows us that our moral positions may not have a solid jus-
tification. Particularly, neither has the naïve utilitarian position, according to which
one should sacrifice the one person in order to save five. This is due to it being well
documented in experiment after experiment (mental, of course, no one is ever killed)
that the answers to the question depend on seemingly irrelevant variables. For instance,
the answer varies if the person to be sacrificed is a company worker who is repairing the
tracks or is a person who is simply wandering on the tracks irresponsibly. The outcome
would also vary if, instead of diverting the trolley, one could throw a fat person from a
bridge onto the tracks in front of the trolley, thus making it derail, killing the fat person
while saving the five lives on the tracks. Another variation involves making the trolley
follow the track of one’s choice. There are infinite variations of the theme.
Why should the answer change? What if those five people were citizens in a hospital
that would die unless they received a transplant while the person who could save them
by donating his or her organs and afterwards die is a mere regular citizen? In all cases,
the problem revolves around saving five lives by sacrificing one.
One of the problems with this analysis is that there is not a way of thinking about
any of these dilemmas based on their exact description. Taken at face value, they
require a unique solution. Taken into context, and everyone is free to read the problem
within a context, the same person may find a different solution to each version of the
trolley problem without falling into a contradiction.
A society that sacrifices one citizen at random in order to donate their organs is
unattractive and unfeasible. Questionably, the organs of the relatives of the elite or
of doctors will not be seen in the operating room. Citizens will also dedicate a great
deal of resources towards hiding from patrols seeking donors. A solution to the trolley
problem that requires the sacrifice of one person in order to save five would not be util-
itarian, since it would not be taking all of the benefits and costs into account. A small
group of castaways in a lifeboat or astronauts stranded in a spaceship will very likely
opt for saving the greater number of lives, knowing the very special situation in which
255
GAME THEORY
they are and that whatever they do there will have no consequences for society in case
they are rescued. Philosopher and psychologist Joshua Greene exposes these issues in an
eloquent and rigorous way in his book Moral Tribes (Greene, 2013).
13.9 Towards a behavioral theory

At the base of all behavioral theories lies the concept of reply: how individuals react to
a given environment that, in the case of game theory, includes the behavior of others.
In the rational model – the Nash equilibrium – players always reply by choosing the
best response given the actions of others. In this section we present some results of the
most recent literature regarding different responses.
Quantal best response

Authors McKelvey and Palfrey (1995) proposed that individuals choose better responses
with higher probabilities. Without entering into technical details, we can illustrate the
concept in the chicken game between Jim and Buzz that we saw in Chapter 1. For con-
venience, we have reproduced the game in Figure 13.7.
Buzz
Swerve Keep going
Swerve 2, 2 1, 4
Jim
Figure 13.7 The chicken game
Say that Buzz chooses his strategies with probabilities (2/3, 1/3). Jim’s utility if he
2 1 5 2 1 8
chooses “Swerve” is 2 + 1 = , while if he chooses “Keep going” he gets 4 + 0 = ,
3 3 3 3 3 3
which is higher. The rational best response is, then, “Keep going”; however, the quan-
tal best response just says that he has to choose “Keep going” with higher probability
than the alternative. How much more is specified by a formula that depends on the
relative utilities affected by an exponential function. The details are not important at
this time. Suffice it to say that the results give a quantal response by Jim according to
which he plays his strategies with the approximate probabilities (1/4, 3/4). A quantal
best response equilibrium is achieved when both players are playing their quantal best
response against the other.
Level-k model
According to this model (proposed by Costa-Gomes, Crawford and Broseta, 2001),
individuals are able to consider only k levels of strategic reasoning. A 0-level implies an
individual chooses her strategy randomly. A 1-level implies the individual chooses the
strategy that maximizes her utility if all other individuals choose theirs using a 0-level
strategic reasoning. By induction, a k-level individual chooses the strategy that maxi-
mizes her utility if all others use a (k-1)-level of strategic reasoning.
256
In our example, if Jim is a 0-level responder, he will choose his strategies with prob-
abilities (1/2, 1/2). If Buzz is a 1-level responder, he will choose thinking that Jim is a
1 1 3
0-level responder. Therefore, he thinks his utility will be 2 + 1 = if he swerves
1 1 2 2 2
and 4 + 0 = 2 if he keeps going, and he will therefore choose “Keep going.” The
2 2
reader can easily find that if any of them is a 2-level responder, he will choose “Swerve.”
Other models
We can complicate these non-rational behaviors. For instance, in games with more
players, one of them can think that different players respond at different levels. Also,
in games of two players, a 3-level responder may not have a clear belief of the reason-
ing level of the rival, and may think, for instance, that he is a 0-level responder with
probability ½, a 1-level responder with probability ¼, and a 2-level responder with
probability ¼. Other complications include that players make mistakes in their choices
and that the mistakes increase with the level of reasoning.
Comparing models
After building a data set with all the published experiments on the different models
of non-rational response for non-repeated games, Wright and Leyton-Brown (2017)
conduct a meta-analysis to find which one performs best as a descriptive theory. They
find that the best model by far is the quantum level-k model – a mix of the quantum
best response and the k-model. According to the authors, the good performance of
the model lies in its combination of bounded iterate reasoning and cost-proportional
errors. They believe that the iterative reasoning describes an actual cognitive process,
but that the same cannot be said about the cost-proportional errors. They may describe
human reasoning or they may simply be a closer approximation to human behavior
than the usual uniform error specification.

13.1 Ander has to get up early tomorrow, but he knows that when the alarm goes off,
there is a 1/3 chance he will turn it off and fall back asleep. To avoid this, he
decides to place the alarm clock away from his bed to force himself to get up.
Illustrate this situation as a game.
13.2 Consider the problem of purchasing a home appliance in Section 13.2. Find the
maximum price at which the inefficient device would have to be sold so that the
consumer would want to buy it with the decreasing discount factor of the example.
13.3 Consider the example in Section 13.3.
(a) Describe the game in extensive form corresponding to the example in Sec-
tion 13.3. To do this, take the following steps:
(i) Initially, the consumer decides on whether to support the government’s
policy of making it mandatory that firms detail the equivalent interest
rate of each payment.
257
GAME THEORY
(ii) If she supports the government, the policy will be approved, otherwise,
it will not be approved.
(iii) If the policy is approved, the terms of the payments appear clearly, as
detailed in the text, otherwise they appear in a confusing way, as shown
in Figure 13.4.
(iv) When the payments are clear, the consumer chooses what is best for
her.
(v) When the payments are not clear at the time of choosing, the consumer
1
will make a mistake and with probability p < will choose the option
2
she did not intend.
(b) Show the subgame perfect Nash equilibrium.
13.4 Consider the example of the minor’s consent in Section 13.5. Change the prob-
abilities of cases (i), (ii), (iii) and (iv) showing the possible combinations of pref-
erences of she-16 and she-18, so that it is better to grant decision-making rights
to the parents.
13.5 Consider the example of the minor’s consent in Section 13.5. Change the prob-
abilities of cases (i), (ii), (iii) and (iv) showing the possible combinations of pref-
erences of the parents, so that it is better to grant decision-making rights to the
parents.
13.6 The social services of two sociologically similar countries, A and B, have col-
lected the results of an extensive survey of 18-year-old girls who had unwanted
pregnancies at the age of 16 or 17. In Country A minors are granted autonomy
to make the decision of terminating the pregnancy, while in Country B it is the
parents who decide. The surveys yielded the following results:
• Country A: 50% of the girls terminated their pregnancy and of them, 80%
do not regret it, though the remaining 20% do.
Of those who did not termi-
nate the pregnancy, 50% do not show regret.
• Country B: 70% of the girls terminated their pregnancy and of them, 30%
show regret. Of those who did not terminate the pregnancy, 70% do not
show any regret.
Making the right decision (with no regrets) gives a utility of 1, while showing
regret gives a utility of zero. Show how to use that information to determine
which country best respects the decision of girls at age 18.
13.7 Repeat Exercise 13.6 changing the utilities. Regret after having terminated the
pregnancy now gives a utility of -1. The other utilities remain the same.
13.8 In Exercise 13.6, change the results of Country A’s survey so that the conclusion
is the opposite of the one found in that exercise.
13.9 Illustrate the case of indoctrination in the terms of Section 13.6 as an exten-
sive form game. Nature first chooses with equal probabilities whether the par-
ents are indoctrinators or not. Then the parents must choose whether or not to
indoctrinate the child. After this, the child must make decision A or B. Parents
who are indoctrinators prefer decision A, while those who are not indoctrinators
258
prefer the same decision as the non-indoctrinated child. The indoctrinated son
will have the preferences of the parents, the non-indoctrinated will prefer B.
13.10 Repeat Exercise 13.9, but now with indoctrination having only a 90% chance
of success. If the indoctrinated children will become indoctrinating parents
with 80% probability, and the non-indoctrinated children will be indoctrina-
tors with probability 5%, how many indoctrinating parents will there be in the
following generation?
259
14 POWER INDICES
Thus far we have dedicated ourselves to non-cooperative game theory, which receives
its name not because there is no cooperation in it, but rather because ultimate decisions
are in the hands of the individuals who must make their choices according to their
likewise individual criteria. In contrast, cooperative game theory gets its name because
it is dedicated to answering a different type of question – for example, which agreement
will prevail if individuals accept an a priori criterion on how to play and on how to
accept potential gains rather than because of a lack of disputes among players. What is
more important to this part of the theory is that once the criteria are agreed upon, the
result of the game is no longer in the hands of the players but rather proceeds according
to the accepted rules. Another way of looking at this is to consider that players have
given up the prerogative of choosing their strategies and have given a third party (an
arbitrator, for example) the power of providing the outcome of the game.
In this chapter we will only see one part of the cooperative game theory, that
which studies the so-called majority games and in which power indices will be the
object of study. Majority games are models to study the formation of majorities
among members of a collective (e.g., an assembly, a committee or a parliament), and
they are versatile enough to accommodate different requirements to achieve a deci-
sion by majority. Because typically the members of the collective belong to parties or
groups of interest, an important question is the analysis of the relative power of those
parties or groups given the number of members they each have and the requirements
to form a majority. It turns out that some of the sharing rules that are studied in
cooperative games can be interpreted as an indication of the power of the different
groups in majority games. We first define the majority games in Section 14.1 and
then the power indices in Section 14.2. Next, we present some applications to actual
parliaments in Section 14.3. In Section 14.4, we extend the model to deal with rules
that require majority in different dimensions (e.g., a number of countries in favor,
but with the condition that they have a majority of the population), and we illustrate
this extension in the case of the European Parliament. A last section shows the model
when abstentions are permitted.
14.1 Cooperative and majority games

Cooperative games
A cooperative game is defined formally according to the following three elements:
1. A set of players, N.
2. A set of possible coalitions of players, K.
3. A value for every coalition C included in K. We will denote the value of coalition C
as v(C).
260
POWER INDICES
This definition may seem too abstract, but the less mathematically inclined reader
should get a better understanding after the explanations and examples that come next.
Unless otherwise indicated, the set of possible coalitions will be the total number of
coalitions that can be formed. Mathematically, it is the set of all the subsets of N except
for the empty set. A coalition’s value v(C), can be interpreted as the total payment that
the coalition can obtain for distributing among its members, regardless of what the
other players in the game do. Because the set of all the players, N, is also a coalition,
a solution of the cooperative game will be a way of distributing v(N) among all the
players taking into account the power of the intermediate coalitions.
Majority games
A particular case among cooperative games is the set of majority games. In these games,
the value of each coalition is either zero or one. The value of one is given to those
coalitions that have a number of votes greater than or equal to a pre-specified given
number, called a quota. If the quota is half of the votes, then we would have a game
for achieving an absolute majority among members of a collectivity; if by contrast that
quota is greater than half of the votes, then the game becomes about forming a quali-
fied majority. In each game one must define the number of votes that each player has
(e.g., a player consists of a group or a political party). A player’s number of votes is often
called the weight of that player. The number of votes each coalition has will be the sum
of the votes of each player within it. Let us look at some examples.
Example 1
There are three groups represented in a committee, A, B and C, with 4, 3 and 2 mem-
bers each. Each group works as a single player. The proposals in the committee need
two thirds of the votes in order to be approved. One way of summarizing the informa-
tion of this game is by writing it as [6; 4, 3, 2], where the first number is the quota and
the other three numbers are the number of seats of the parties.
Given that information, the game is described by calculating the value for each
coalition, a value that equals one if the coalition has at least six votes, and zero otherwise:
v(A) = 0, v(B) = 0, v(C) = 0,
v(A, B) = 1, v(A, C) = 1, v(B, C) = 0,
v(A, B, C) = 1.
Example 2
Four parties are represented in a parliament, A, B, C and D with 7, 6, 3 and 1 seats,
respectively. Proposals need an absolute majority of at least nine votes. The game can
then be written as [9; 7, 6, 3, 1]. The values for the different coalitions are
v(A) = v(B) = v(C) = v(D) = 0,

v(A, B) = v(A, C) = v(B, C) = 1, v(A, D) = v(B, D) = v(C, D) = 0,
v(A, B, C) = v(A, B, D) = v(A, C, D) = v(B, C, D) = 1,
v(A, B, C, D) = 1.
261
GAME THEORY
Example 3
After the election on June 26, 2016, the Spanish Parliament was configured as follows:
PP, 137 seats; PSOE, 85; UP, 71; C’s, 32; ERC, 9; CDC, 8; PNV, 5; Bildu, 2 and CC, 1.
The parliament must elect a prime minister (referred to as the president of the government
in Spain). A candidate is elected if they receive at least 176 votes. This majority game is as
follows: v(C) = 0 if the seats of the members in coalition C are less than 176, and v(C) =
1 if the seats add up to 176 or more. We have not listed all of the possible coalitions in this
example in detail because there would be far too many of them, 511 to be precise.
14.2 Power indices in majority games

Power indices are, in general, an adaptation for majority games of some solution con-
cepts defined for more general cooperative games. The underlying idea of the most
important indices is that the power of a player is not as much related to the number of
seats as to how often it is needed in order to achieve majorities through the different
coalitions that can be formed.
The Shapley–Shubik power index

This index is the adaptation of the Shapley value as defined in cooperative games (Shap-
ley and Shubik, 1954). This value can be seen in two ways: first, by the properties which
characterize and define it and, second, by the formula for calculating it. For the purposes
of this chapter, the second will suffice. The calculation is done through several steps.
1. List all of the possible orderings in which to arrange the players.
2. For each ordering, calculate the sum of the players’ votes following the order of the
players until the quota is reached.
3. For every player, count the number of orderings in which she is pivotal (i.e., with
her, the quota is reached; without her, the quota is not reached).
4. A player’s Shapley–Shubik value is the proportion of orderings in which she is piv-
otal over the total number of orderings. The Shapley–Shubik value corresponding
to Player X is denoted by S−S (X ).
Let us follow these steps in Example 1:
1. There are six (3! = 6) possible orderings for Groups A, B and C: ABC, ACB, BAC,
BCA, CAB and CBA.
2. and 3. ABC: 4 + 3 = 7. B is pivotal (we have counted first the four seats of party A,
and since they do not reach the quota of six we added the seats of the next party in
the ordering – the three votes of B – and now we have reached the quota). Seven is
equal to or more than six, so B is pivotal.
ACB: 4 + 2 = 6. C is pivotal
BAC: 3 + 4 = 7. A is pivotal
BCA: 3 + 2 + 4 = 9. A is pivotal
CAB: 2 + 4 = 6. A is pivotal
CBA: 2 + 3 + 4 = 9. A is pivotal
4. A is pivotal in 4/6 of the orderings: S−S (A) = 0.66.
B is pivotal in 1/6 of the orderings: S−S (B) = 0.16.
C is pivotal in 1/6 of the orderings: S−S (C) = 0.16.
262
POWER INDICES
For Group A it is enough for it to be joined by either of the other two in order to obtain
a decisive majority. This does not happen to party B or C because a coalition of both of
them does not add up to the quota. This is why A has the highest power index.
Shapley–Shubik index in Example 2:
1. There are 24 (4! = 24) ways to arrange Groups A, B, C and D: ABCD, ABDC,
ACBD, ACDB, ADBC, ADCB, BACD, BADC, BCAD, BCDA, BDAC, BDCA,
CABD, CADB, CBAD, CBDA, CDAB, CDBA, DABC, DACB, DBAC, DBCA,
DCAB and DCBA.
2. and 3. ABCD: 7 + 6 = 14. B is pivotal
ABDC: 7 + 6 = 14. B is pivotal
ACBD: 7 + 3 = 10. C is pivotal
ACDB: 7 + 3 = 10. C is pivotal
ADBC: 7 + 1 + 6 = 14. B is pivotal
ADCB: 7 + 1 + 3 = 11. C is pivotal
BACD: 6 + 7 = 14. A is pivotal
BADC: 6 + 7 = 14. A is pivotal
BCAD: 6 + 3 = 9. C is pivotal
BCDA: 6 + 3 = 9. C is pivotal
BDAC: 6 + 1 + 7 = 14. A is pivotal
BDCA: 6 + 1 + 3 = 10. C is pivotal
CABD: 3 + 7 = 10. A is pivotal
CADB: 3 + 7 = 10. A is pivotal
CBAD: 3 + 6 = 9. B is pivotal
CBDA: 3 + 6 = 9. B is pivotal
CDAB: 3 + 1 + 7 = 11. A is pivotal
CDBA: 3 + 1 + 6 = 10. B is pivotal
DABC: 1 + 7 + 6 = 14. B is pivotal
DACB: 1 + 7 + 3 = 11. C is pivotal
DBAC: 1 + 6 + 7 = 14. A is pivotal
DBCA: 1 + 6 + 3 = 10. C is pivotal
DCAB: 1 + 3 + 7 = 11. A is pivotal
DCBA: 1 + 3 + 6 = 10. B is pivotal
2. A is pivotal in 8/24 of the orderings: S−S (A) = 0.33.
B is pivotal in 8/24 of the orderings: S−S (B) = 0.33.
C is pivotal in 8/24 of the orderings: S−S (C) = 0.33.
D is not pivotal in any ordering: S−S (C) = 0.
In this example, although party C has only three seats, it is necessary in the same num-
ber of coalitions as parties A and B, with the largest number of seats. Party D is not
necessary in any coalition and therefore its power is null.
Banzhaf power index

This index (Banzhaf, 1965) is very similar in concept to the Shapley–Shubik power
index. The only difference is that it does not consider orderings that differ because of
the position of the players that come before or after the pivotal player. In Example 1, the
BCA and CBA orderings have A as a pivotal, while B and C are arranged in two different
263
GAME THEORY
ways before Player A. The Banzhaf index, which we will denote by B (X), reduces these
two orderings to one. There are no further orderings in which this happens; therefore, in
the calculation of the Banzhaf index in Example 1, Player A will be pivotal in 3/5 of the
orderings and B and C in 1/5 each: B (A) = 0.6, B (B) = 0.2 y B (C) = 0.2.
Another way of looking at it is by considering the winning coalitions and, in them,
indicating which players are needed. The index will be the proportion of the times each
player is needed. In Example 1 the winning coalitions are AB, AC and ABC. Player A
is needed in coalitions AB, AC and ABC; Player B is needed only in coalition AB; and
Player C only in AC. That is, Player A is needed in three out of the five times someone
is needed, B in one out of five and C also in one out of five.
Example 2 requires a more complicated calculation. The following pairs of order-
ings are considered as one according to the Banzhaf power index. The pivotal player
has been marked in bold font. Note that in all pairs of orderings the players before and
after the pivotal player are the same.
Player A as pivotal:
BACD and BADC,
CABD and CADB,
CDAB and DCAB,
BDAC and DBAC.
Player B as pivotal:
ABCD and ABDC,
ADBC and DABC,
CBAD and CBDA,
CDBA and DCBA.
Player C as pivotal:
ACBD and ACDB,
BCAD and BCDA,
ADCB and DACB,
BDCA and DBCA.
According to the Banzhaf power index, Player A is pivotal in 4/12 of the orderings, the
same as B and C. Player D is never pivotal: B (A) = 0.33, B (B) = 0.33, B (C) = 0.33,
B (D) = 0.
In Example 3, the number of parties with representation in the Spanish Parliament
after the general election on June 26, 2016, is nine. Thus, the number of orderings is
9! = 362,880; therefore, computers become necessary to calculate the power indices.
There are several websites available for calculating them online.
Some empirical evidence

Geller, Mustard and Shahwan (2013) conduct a series of experiments to address the
question of how well power indices reflect human behavior. The different experiments
mimic different majority games. The experimental subjects must share a resource that
they can obtain if they form a winning coalition. They are thereby encouraged to reach
an agreement once they are informed of the number of votes and the value of the quota.
The authors experiment with different specifications and find that, on average, the way
264
POWER INDICES
the resource is divided among the players is very close to the proportions indicated by
the Shapley–Shubik index.
14.3 Application of power indices to three parliaments

The Spanish Parliament and the power of nationalist parties
In this section we will see a first application of how to use the power indices to interpret
election results. We will be particularly interested in seeing whether the nationalist par-
ties in the Spanish Parliament obtain power that is disproportionate to their size and,
if so, the reason for this. In Tables 14.1a to 14.1e we gather the composition of the par-
liament after the latest elections without any one party winning an absolute majority
(otherwise, all power would correspond to that majority party). One of the first things
that a newly formed parliament must do is elect the president of the government; there-
fore, it is of utmost importance that the different parties in a coalition agree upon a
Table 14.1a Spanish Parliament after 2016 elections
Elections June 26, 2016

Seats % votes % seats S-S index B index
PP 137 33 39.1 44.2 43.6
PSOE 85 22.7 24.3 19.1 18.1
UP 71 19.7 20.3 19.1 18.1
C’s 32 13.1 9.1 10.9 12.8
ERC 9 2.6 2.6 2.5 2.7
CDC 8 2 2.3 2.5 2.7
PNV 5 1.2 1.4 0.9 1.2
Bildu 2 0.8 0.6 0.6 0.7
CC 1 0.3 0.3 0.2 0.4
Table 14.1b Spanish Parliament after 2015 elections
Elections December 20, 2015
PP 123 28.7 35.1 40.2 37.7
PSOE 90 22 25.7 22 21
Podemos 69 20.7 19.7 22 21
C’s 40 13.9 11.4 6.9 8.4
ERC 9 2.39 2.6 3 4
CDC 8 2.3 2.3 2.5 3.3
PNV 6 1.2 1.7 2 2.6
IU 2 3.7 0.6 0.6 0.8
Bildu 2 0.9 0.6 0.6 0.8
CC 1 0.3 0.3 0.2 0.3
265
GAME THEORY
Table 14.1c Spanish Parliament after 2008 elections
Elections March 9, 2008

PSOE 169 43.9 48.3 53.8 56
PP 154 39.9 44.0 11.8 9.6
CiU 10 3 2.9 11.8 9.6
PNV 6 1.2 1.7 10.4 9.4
ERC 3 1.2 0.9 3.5 4.5
IU 2 3.8 0.6 2.1 2.7
BNG 2 0.8 0.6 2.1 2.7
CC 2 0.7 0.6 2.1 2.7
UPyD 1 1.2 0.3 1.2 1.4
NB 1 0.2 0.3 1.2 1.4
Table 14.1d Spanish Parliament after 2004 elections

PSOE 164 42.6 46.9 48.5 51.6
PP 148 37.8 42.3 13.6 10.6
CiU 10 3.2 2.9 10.5 10
ERC 8 2.5 2.3 7.7 8
PNV 7 1.6 2.0 6.4 6.7
IU 5 5 1.4 4.6 4.2
CC 3 0.9 0.9 3.4 3.5
BNG 2 0.8 0.6 2.1 2.3
ChA 1 0.4 0.3 1.1 1.1
EA 1 0.3 0.3 1.1 1.1
NB 1 0.2 0.3 1.1 1.1
Table 14.1e Spanish Parliament after 1996 elections

PP 156 38.8 44.6 45.2 45.4
PSOE 141 37.7 40.3 18.4 17.2
IU 21 10.6 6 18.4 17.2
CiU 16 4.6 4.6 11.9 14.1
PNV 5 1.3 1.4 1.7 1.5
CC 4 0.9 1.1 1.7 1.5
BNG 2 0.9 0.6 0.8 0.9
HB 2 0.7 0.6 0.8 0.9
ERC 1 0.7 0.3 0.4 0.4
EA 1 0.5 0.3 0.4 0.4
UV 1 0.4 0.3 0.4 0.4
266
POWER INDICES
candidate as this will condition the legislature. In the tables, the sum of the percentage
of votes does not add up to 100 because there are additional small parties that have not
obtained parliamentary representation.
Nationalist and local parties are highlighted in bold font. There are 350 seats and
a majority is reached with 176, which is the quota used to calculate the power indices.
To make the comparison between the columns more visually appealing, the power
indices are expressed in percentages rather than as numbers between zero and one, as
defined in Section 14.2.
The Spanish press often highlights the excessive dependence of governments on
their agreements with nationalist parties and, therefore, on the excessive weight they
have relative to their size. This overrepresentation is used as an argument in favor of
a possible change in the electoral law with the aim of limiting their representation.
We can, however, observe that, generally speaking, the percentage of seats obtained
by these parties is not far from the percentage of their votes, so there is no data to sup-
port their overrepresentation. If there is any excess of influence (beyond parliamentary
weight), this is due to their ability to form a majority with one of the larger parties when
none achieves an absolute majority. This is very well reflected in their power indices,
but only in the case of bipartisanship, with two large parties, as can be seen in the
composition of the parliament after the elections of 1996, 2004 and 2008. When there
are more than two large parties, their power is diluted – there are alternative majorities
made up of coalitions encompassing some of the large parties, as could be seen after the
elections of 2015 and 2016.
The British Parliament after the 2017 elections

Prior to the 2017 elections, the Conservatives held 331 seats in parliament, more
than half of the 650 seats. This meant that the Tories had a Shapley–Shubik power
index of 100%, while the rest of the parties had 0%. With the new election, the
Tories were still the first party, but their 318 seats kept them 8 seats short from
the 326 necessary for the majority. A coalition was needed in order to form a new
government. Things were a little easier since Sinn Féin, which obtained seven seats,
never sends its representatives to the British Parliament, meaning that only 322
votes were needed for a majority. The Tories still needed at least 4 additional votes
to add to their 318 seats. Table 14.2 shows the Shapley–Shubik power indices for
different scenarios.
First, notice that neither the Green Party nor the Independent seat (in a Northern
Ireland constituency) is necessary in any coalition, and thus both have zero power.
A priori, the Conservative Party has great power since it can form a majority with any
other party (except the two with zero power). However, a coalition with their first rival,
the Labour Party, or with the nationalist parties of Scotland or Wales (SNP and Plaid
Cymru) is extremely difficult, leaving only the options of the LD (Liberals) and the
DUP (Northern Ireland Unionists).
In the last three columns we show different coalitions. A blank box in one of these
columns indicates that the party does not participate in that coalition. In each column
the index takes into account only the orderings between parties in the coalition while
maintaining the quota at 322 in order to form majorities.
267
GAME THEORY
The column CON+DUP shows the sharing of power within the coalition that
was ultimately formed, as well as the Conservative Party’s significant dependence on
the DUP. A coalition with both the DUP and the LD would have been much better
for the Tories, but the Liberals (pro EU) would not agree to that. The last column
shows the almost impossible coalition of every other group against the Conservatives,
in which the biggest party, Labour, does not have more power than the rest since
they are all crucial to the coalition (save for the two small parties with no power).
Table 14.2 British Parliament after 2017 elections
Party % Vote Seats S-S index CON+DUP CON+LD+DUP All but CON

CON 42.4 318 66.6 50 66.6
LAB 40.0 262 6.6 20
SNP 3.0 35 6.6 20
LD 7.4 12 6.6 16.6 20
DUP 0.9 10 6.6 50 16.6 20
Plaid Cymru 0.5 4 6.6 20
Green Party 1.6 1 0 0
Indep. 0.05 1 0 0
The Basque Parliament after the 2016 elections

On September 25, 2016, elections to the Basque Parliament were held. One of the first
responsibilities of the new members of parliament is to elect the Lehendakari (the pres-
ident of the Basque government). Table 14.3a shows the composition of the parliament
following the day of elections. We can observe the great power of PNV, a nationalist
party that either alone or as part of a coalition has governed the Basque Autonomous
Community for most of its history. As we can see in column 2, if PNV were to rely
on the two parties of the radical left (Bildu, a pro-independence party, and EP, not
pro-independence but in favor of holding a referendum for consulting on the issue),
then it would have two thirds of the power according to the Shapley–Shubik index.
The same thing would happen if it relied on the two non-nationalist Spanish parties,
PSE and PP (column 3) or a combination of both (EP + PSE, since PSE will not join
forces with Bildu), as indicated in column 4. Column 5 shows a possible coalition with
PSE, the party that has shared the government with PNV the most times. The PNV is
in a very advantageous position and should be able to obtain support for its candidate
at a very low cost.
This analysis, with the quota set at 38 votes, is valid for all votes in which the
options are YES or NO. However, the election of the Lehendakari uses a plurality vot-
ing rule, through which each group can present its candidate and the one with the most
votes is elected. There is no voting against, as is the case in the Spanish Parliament;
therefore, the most realistic quota for the analysis will be the one which corresponds to
the party most voted for or to that of a plausible alternative coalition. In the example,
PNV has 29 votes which are only exceeded by some impossible coalition. Table 14.3b
shows the Shapley–Shubik indices for a quota equal to 29.
268
POWER INDICES
Table 14.3a Provisional count for the 2016 Basque Parliament with quota at 38
Basque Parliament 2016

(September, 25) 1 2 3 4 5
quota 38
PNV (29) 60 66.7 66.7 66.7 50
Bildu (17) 10 16.7
EP (11) 10 16.7 16.7
PSE (9) 10 16.7 16.7 50
PP (9) 10 16.7
Table 14.3b Provisional count for the 2016 Basque Parliament with quota at 29

(September, 25) 1 2 3 4 5
quota 29
PNV (29) 60 100 100 100 100
Bildu (17) 10 0
EP (11) 10 0 0
PSE (9) 10 0 0 0
PP (9) 10 0
Things changed a few days later when votes cast by mail were tallied, with PNV
losing one seat in favor of Bildu. The proportion of votes and seats hardly changed,
though it did mean a great difference for PNV, as seen in both its power index ex ante
as well as in a number of the coalitions (Table 14.3c, and in Table 14.3d with the quota
equal to the votes of a possible coalition of Bildu + EP). Compared to the previous
situation, Bildu and EP win ex ante, and likewise in the coalition in column 2 when
the quota is placed at 29. More interesting is what happens to the parties PSE and PP,
which both lose ex ante power with the 38 quota but gain slight power with it at 29;
with the 38 quota they also gain in the coalition in column 3. PSE loses power in the
hypothetical coalition with PNV when the quota is absolute majority, but gains with
the more realistic quota of 29. This is not the most interesting coalition for PNV in
terms of short-term power. However, historical evidence showing how PNV has been
able to negotiate with anyone in the past makes this possibility attractive, since it can
change partners easily.
In fact, the last seat that went from the PNV to Bildu after counting the mail vote
was being disputed during the evening of the election, and already it was foreseen to
have a high probability of finally falling on Bildu’s side. The comments made by PNV
leaders clearly indicated that they were aware of its importance. What is interesting
is not that these leaders were using the power indices as opposed to their experience
for understanding the implications of the last seat’s allocation – for they did not – but
rather the fact that power indices are able to formalize and illustrate all those implica-
tions in a systematic way.
269
GAME THEORY
Table 14.3c Final count for the 2016 Basque Parliament with quota at 38

(Final count) 1 2 3 4 5
quota 38
PNV (28) 40 66.7 33.3 50 0
Bildu (18) 23.3 16.7
EP (11) 23.3 16.7 50
PSE (9) 6.7 33.3 0 0
PP (9) 6.7 33.3
Table 14.3d Final count for the 2016 Basque Parliament with quota at 29

(Final count) 1 2 3 4 5
quota 29
PNV (28) 36.7 33.3 66.7 66.6 50
Bildu (18) 20 33.3
EP (11) 20 33.3 16.7
PSE (9) 11.7 16.7 16.7 50
PP (9) 11.7 16.7
14.4 Games with many quotas

Quite often, a decision body must satisfy a majority at two levels or more in order to
make their decision. For instance, an international committee must reach an agreement
that is supported by a majority of countries with the additional condition that it must
be backed by a majority of the population. Each of these majorities can be understood
as a regular majority game; however, when the two conditions must be satisfied, the
result will not always be a majority game, even though we may still find power indexes
in the same way we did before. Let us look at an example.
Four countries, A, B, C and D, with populations 4, 2, 2 and 1, respectively, make
decisions based on two criteria: a majority of countries as well as 2/3 of the population.
Considering the two criteria as two different majority games, we will write them as
follows: vM = [3; 1, 1, 1, 1] and vP = [6; 4, 2, 2, 1].
The first game, which involves forming a winning coalition of a majority of coun-
tries, can be rewritten in further detail as the following cooperative game:
vM (i) = 0,
vM (i, k) = 0,
vM (i, j, k) = 1,
vM (A, B, C, D) = 1.
270
POWER INDICES
Coalitions made up of one or two players do not reach the majority, whereas coalitions
of 3 or 4 players do. All countries are in the same position, so they will be pivotal in the
same number of orderings. Thus, the Shapley–Shubik index will give the same power
to each of the four countries:
1
S − S M ( A ) = S − S M ( B ) = S − S M (C ) = S − S M ( D ) =
.
4
The cooperative game of the majority-of-population game is a bit more complicated:
vP (i) = 0,
vP (A, B) = 1, vP (A, C) = 1, for all other pairs: vP (i, k) = 0,
vP (A, i, j) = 1, vP (B, C, D) = 0,
vP (A, B, C, D) = 1.
To compute the Shapley–Shubik index in this second majority game, notice that there
are 24 orderings (4!). Country A is pivotal in all six orderings in which it enters last
as well as in all six orderings in which it is third to enter. Country D is never pivotal.
In the remaining 12 orderings, countries A, B and C are in a symmetrical situation,
with each one being pivotal in 4 cases. This gives a total of 16 out of 24 orderings in
which A is pivotal, and 4 out of 24 in which either B or C are pivotal. Thus, the coun-
tries’ power according to the Shapley–Shubik index is
2
S − SP ( A ) = ,
3
1
S − S P ( B ) = S − S P (C ) = ,
6
S − SP ( D ) = 0 .
The game which requires the two majorities is
vM+P (i) = 0,
vM+P (i, j) = 0,
vM+P (A, i, j) = 1, vM+P (B, C, D) = 0,
vM+P (A, B, C, D) = 1.
Only the coalitions of three countries which include A, as well as the coalition of all four,
meet both quotas. In this game, A is pivotal in all six orderings that end with A. When
any other country is last, the remaining three are in a symmetric position: if B is last,
countries A, C and D are pivotal in two orderings each, whereas if C is last, then coun-
tries A, B and D are pivotal in two orderings each; finally, if C is last, then countries A,
271
GAME THEORY
B and C are pivotal in another two orderings each as well. In all, A is pivotal in 12 out
of 24 orderings, while the other three players are pivotal with the same frequency in the
remaining 12. This gives the following Shapley–Shubik power indices:
1,
S − SM + P ( A ) =
2
1
S − SM + P ( B ) = S − SM + P (C ) = S − SM + P ( D ) = .
6
Remember that a majority game is defined once the weight of each player and a quota
are established. After this specification, a cooperative game can be described with a
value for every coalition. However, this procedure is generally not reversible. Not all
cooperative games can be described as majority games, not even those which come
after decision rules which require that multiple quotas are met. For this reason, the
algorithms for computing power indices in majority games do not work when two or
more quotas must be reached in order to make a decision.
The reader can verify that the above example belongs to the exception and can
be represented as just one majority game (Exercise 14.10). Another example is the
United Nations Security Council. It is made up of 15 countries, 5 of which are per-
manent members (China, USA, Russia, France and UK), while the remaining 10 are
non-permanent and are elected among the rest of the UN countries for a limited time.
A resolution in this council requires (i) a total of at least nine countries and (ii) the
approval of all the permanent members. If we begin by listing the permanent members,
condition (i) can be written as
vvote = 9; 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
while condition (ii) can be written as
vveto = 5; 1,1,1,1,1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0  .
In Exercise 14.11 the reader can verify that the rule whereby both (i) and (ii) must be
satisfied can be written as
v = vvote ∩ vveto = 39; 7, 7, 7, 7, 7,1,1,1,1,1,1,1,1,1,1.
The Shapley–Shubik power of each permanent member is 19.63%, while the

non-permanent members have only 0.19% each.
14.5 The distribution of power in the EU after Brexit

Depending on the issue being discussed, the council of the European Union makes its
decisions by
• simple majority (15 member states vote in favor),
• qualified majority (55% of member states, representing at least 65% of the EU pop-
ulation, vote in favor), or
• unanimous vote (all votes are in favor).
272
POWER INDICES
In these rules, any state is in the same position as any other and will, therefore, have the
same power index, except in the rule that requires 65% of the population. Table 14.4
lists the Shapley–Shubik power index of the different countries according to this rule
with and without the United Kingdom.
The first thing we can observe is that the Shapley–Shubik power index is not very
different from the population percentage of each country. However, the small devia-
tions favor the four largest countries. The last two columns repeat the exercise, only
this time without the UK, so as to estimate the effects of Brexit in the distribution
of power among the remaining countries. As can be expected, all see an increase of
their power, though some benefit more than others. Particularly, the largest countries
Table 14.4 Brexit and the power indices in the EU
% Pop. S-S index

Population % Pop. S-S index
Country without without
in millions with UK with UK
UK UK
Germany 82.8 16.18 17.39 18.56 20.66
France 67.2 13.1 13.54 15.06 15.86
UK 65.8 12.86 13.20
Italy 60.6 11.84 11.97 13.58 14.02
Spain 46.5 9.09 8.85 10.43 10.35
Poland 38 7.42 7.17 8.51 8.09
Romania 19.6 3.84 3.73 4.40 4.05
Netherlands 17.1 3.34 3.21 3.82 3.52
Belgium 11.4 2.22 2.10 2.58 2.37
Greece 10.8 2.1 1.96 2.41 2.20
Czech Rep. 10.6 2.07 1.96 2.38 2.20
Portugal 10.3 2.01 1.90 2.31 2.12
Sweden 10 1.95 1.90 2.24 2.03
Hungary 9.8 1.91 1.82 2.20 2.03
Austria 8.8 1.71 1.61 1.97 1.81
Bulgaria 7.1 1.39 1.30 1.59 1.46
Denmark 5.7 1.12 1.04 1.29 1.18
Finland 5.5 1.08 1.00 1.23 1.11
Slovakia 5.4 1.06 1.00 1.22 1.11
Rep. of Ireland 4.8 0.93 0.86 1.07 0.97
Croatia 4.2 0.81 0.75 0.93 0.84
Lithuania 2.8 0.56 0.52 0.64 0.58
Slovenia 2.1 0.4 0.36 0.46 0.41
Latvia 1.9 0.38 0.36 0.44 0.41
Estonia 1.3 0.26 0.24 0.29 0.26
Cyprus 0.9 0.17 0.16 0.19 0.17
Luxembourg 0.6 0.12 0.11 0.13 0.12
Malta 0.4 0.09 0.08 0.10 0.09
273
GAME THEORY
(Germany, France, Italy and Spain) observe the steepest relative increase (17% increase
and greater), while the smallest countries (Malta, Luxembourg, Cyprus and Estonia) as
well as a few middle-sized countries (Romania, Netherlands and Sweden) benefit less
(an 11% increase or less). The rest of the countries show an increase of their power of
around 11% to 14%.
As we read in the description of the different ways of voting within the European
Council, the population rule is used only in the qualified majority vote and must be
applied along with the qualified majority of countries. Thus, a coalition of countries
has a value of 1 if it likewise has a value of 1 in each of the two following majority
games in the EU with 28 members (similar games can be written for the EU with 27
members following Brexit):
v = [16; 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
v* = [65; 16.18, 13.1, 12.86, 11.84, 9.09, 7.42, 3.84, 3.34, 2.22, 2.1, 2.07,
2.01, 1.95, 1.91, 1.71, 1.39, 1.12, 1.08, 1.06, 0.93, 0.81, 0.56, 0.4,
0.38, 0.26, 0.17, 0.12, 0.09].
Game v has a quota of 16, the minimum number of countries that reaches a 55% of
countries within the EU with 28 members. In this game, each country is just one
vote.
Game v* has a quota of 65, and the votes are simply the percentage of the population
that each country represents.
The game which requires majorities like those in both v and v* can be defined in
the same way as the example in Section 14.4. László Á. Kóczy (2016) does precisely this
and studies the changes in power indices in the qualified vote system following Brexit.
He obtains the same results in regards to the large and small countries that we have just
discussed with only the rule of the majority of the population.
In a similar vein, Bilbao, Fernández, Jiménez and López (2002) studied a previ-
ous voting rule in the EU which required a triple majority when making important
decisions: (i) a majority of votes (bigger countries had more votes, and 225 out of
345 were necessary), (ii) a majority of countries (14 out of 27) and (iii) 62% of the
population. In particular, these authors studied the effect of a proposal for rein-
forcing condition (ii) by requiring a majority of 2/3 of the countries. Using power
indices they show that, first, the power was very similar with just a majority of
votes (condition (i)) relative to the requirement of conditions (i) and (ii) and relative
to the requirement of the three conditions. They also showed that the reinforce-
ment of the majority in condition (ii) did not alter the previous power distribution.
One must be cautious, however, when interpreting this result. It does not mean
that the proposed reform did not matter: even if countries have the same power to
influence policies, the need for wider acceptance may change the kind of agree-
ments the countries will vote for (even if in these agreements some countries can
still exert a significant influence). Incidentally, the decision rule of the European
Council requiring (i), (ii) and (iii) is an example of a game with multiple quotas
that cannot be represented as just one majority game (one quota and one weight for
every player), as shown by Freixas (2004).
274
POWER INDICES
14.6 Power indices with abstention

One can extend the definitions and formulas of power indices so as to include the pos-
sibility that parties abstain. This introduces great complexity into the problem, not just
in the computation but also in the different ways in which abstention can be treated.
In this section we will explore a simple case.
Let’s have a parliament with three parties, X, Y and Z, with 100, 99 and 2 seats,
respectively. The quota for an absolute majority is 101 votes. According to the Shapley–
Shubik index, the three parties have 33.3% of power each (the small party is as nec-
essary as the big ones in any coalition). However, if we introduce the possibility of
abstention and a quota that is the majority of votes, things change. Let us proceed by
considering all the possibilities in order.
No party abstains
This is the standard case without abstention and with quota 101. There are six possible
orderings and each party is pivotal in two.
One party abstains

We can model this as if the party that abstains now has zero seats and the quota is the
majority of the seats of the other two parties.
X abstains: [52; 0, 99, 2], Y is pivotal in all 6 orderings.
Y abstains: [51; 100, 0, 2], X is pivotal in all 6 orderings.
Z abstains: [100; 100, 99, 0], X is pivotal in all 6 orderings.
Two parties abstain

X and Y abstain: [2; 0, 0, 2], Z is pivotal in all 6 orderings.
X and Z abstain: [50; 0, 99, 0], Y is pivotal in all 6 orderings.
Y and Z abstain: [51; 100, 0, 0], X is pivotal in all 6 orderings.
Three parties abstain

This is a trivial game: [0; 0, 0, 0], with each party being pivotal in 2 of the 6 orderings.
2+6+6+6+2
This way we have that X’s power is = 0.458 , Y’s power is
2+6+6+2 48
2+6+2
= 0.333 and Z’s power is = 0.208 .
48 48
Abstention by everyone is most certainly an impossible situation; therefore, we can
try an index without even considering it. The new index will give X, Y and Z the
2+6+6+6 2+6+6 2+6
following power indices: = 0.476 , = 0.333 and = 0.19 ,
42 42 42
respectively. If we were to go further and consider the abstention by two parties also
2+6+6
impossible, the respective powers would thereby be = 0.583 , 2 + 6 = 0.333
24 24
2
and = 0.083.
24
275
GAME THEORY
In any of these cases we can observe how the new indices which consider the pos-
sibility that parties abstain break the symmetry in the Shapley–Shubik index and give
more power to the bigger party.

14.1 Calculate the Banzhaf index in Example 2 of Section 14.2 using the winning
coalition method.
14.2 There are four parties in a parliament, A, B, C and D, with 10, 8, 2 and 1 seats,
respectively. Decisions will be taken by an absolute majority of 11 votes.
(a) Define the situation as a majority game.
(b) Using the formula, calculate the Shapley–Shubik and Banzhaf indices for
each party.
14.3 Repeat Exercise 14.2 for the case in which parties have 9, 8, 2 and 2 seats,
respectively.
14.4 A judicial committee has nine members, and a majority of five is necessary for
making decisions. It is well known that three of these members always vote as
a bloc. There is a proposal to change the majority requirement to either six or
seven members. Use the Shapley–Shubik power index to analyze the conse-
quences of passing any of the proposals.
14.5 Consider the Spanish Parliament after the 2016 election (Table 14.1a). UP
is actually the sum of different alliances in different constituencies. Specifi-
cally, Podemos-IU-Equo, En Comú Podem, Podemos-Compromís-EUPV and
Podemos-En Marea-Anova-EU obtained 45, 12, 9 and 5 seats, respectively.
(a) Calculate the Shapley–Shubik and Banzhaf power indices when each of the
groups behaves independently as a separate party.
(b) Compare these indices with the power index by UP as one party.
(c) Compare also the changes in the power of the other parties.
14.6 Consider the Spanish Parliament after the election of 2008 (Table 14.1c). Let
us distribute the seats as proportional to the votes (only for those parties repre-
sented in the parliament):
(a) Normalize the percent of votes as a percent of the votes of the parties in the
parliament. The sum of all the numbers in the % votes is 95.9. To find the
43.9
normalized percent of the PSOE just calculate × 100 = 45.78 . Do the
95.9
same for the other parties.
(b) Calculate the number of seats that represent the same percentage as the
normalized percent of votes (in general, the number of seats will not be an
integer). For example, given that PSOE had a normalized 45.78% of the
votes, it will get 160.22 seats, calculated as 45.78% of 350.
(c) First assign the seats that correspond to the whole numbers. For example,
PSOE will have 160 seats.
(d) After (b) there will still be some of all 350 seats left to assign. Assign these seats
to the decimal part of the seats assignment in part (a). Do this by starting with
276
POWER INDICES
the highest and continuing with the lowest until all 350 seats are assigned. For
example, the decimal part of PSOE’s number of seats is 0.22.
(e) Compare this hypothetical situation with the actual composition. Who
loses? Who wins?
Consider the proportional parliament estimated above.
(f) Calculate the power indices for each party.
(g) Compare these indices with the ones they have in the actual parliament.
14.7 Repeat Exercise 14.6 for the British Parliament after the 2017 election
(Table 14.2) with and without the Sinn Féin, who obtained 0.7% of the total
votes. (Include UKIP, since they received 1.8% of the votes and could have
obtained some seats through this method.)
14.8 In Exercise 14.7, recalculate the power indices for the case in which nationalist
parties (Sinn Féin, SNP and Plaid Cymru) act as a bloc.
14.9 Repeat the analysis in Section 14.3 for the British Parliament for the case in which
Sinn Féin decides to participate. In particular, address the following questions:
(a) Find the S-S power index of the parties.
(b) Find the S-S power index within the most probable winning coalitions.
(c) What is the effect of the absence of Sinn Féin in regards to the power of
their main adversaries, the DUP?
14.10 Find one majority game in which the coalitional values are the same as in the
game vM+P in Section 14.4.
14.11 Verify that the majority game describing the UN Security Council decision
making in Section 14.4,
v = vvote ∩ vveto = 39; 7, 7, 7, 7, 7,1,1,1,1,1,1,1,1,1,1 ,
has the same winning coalitions as the game in which the quotas in both vvote
and vveto must be met, as described in the text.
14.12 Let us have a parliament with three parties, X, Y and Z, with 49, 26 and 26
seats, respectively. Calculate the power index of each party considering the pos-
sibility of abstention according to the method of Section 14.6.
14.13 The Shapley–Shubik and Banzhaf indices satisfy the property of symmetry: if two
players are symmetrical they will have the same power. Two players are symmetrical
if the set of coalitions in which one of them is necessary is the same set as that in
which the other is also necessary. A player is necessary in a coalition if it is a winning
coalition and if that coalition would not reach the quota if that player were removed.
Note that two players can be symmetrical despite having a different number of seats.
(a) Show that PSOE and UP are symmetrical in the Spanish Parliament after
the election of 2016.
(b) What about ERC and CDC?
(c) Show that LAB, SNP, LD, DUP and Plaid Cymru are symmetrical in the
British Parliament after the 2017 election without Sinn Féin.
(d) What happens to the symmetry in (c) if Sinn Féin participates in the
parliament?
277
REFERENCES
Chapter 1
Hofstadter, D. “Dilemmas for superrational thinkers, leading up to a luring lottery.” In Metamagical
Themas: Questing for the Essence of Mind and Pattern (New York: Basic Books, 1985).
Kant, I. Grounding for the Metaphysics of Morals, 3rd edn translated by J.W. Ellington (Indianapolis/
Cambridge: Hackett Publishing Company, 1993 [1785]).
Hobbes, J.T. “Chapter XIII.” In Leviathan or The Matter, Forme and Power of a Common-Wealth Eccle-
siasticall and Civill (Oxford: Oxford University Press, 2014 [1651]).
Rousseau, J. On the Social Contract, or Principles of Political Law (Indianapolis: Hacket Pub. Co.,
1983 [1762]).
Smith, A. “Book 1, Chapter 2: Of the principle which gives occasion to the division of labour.”
In An Inquiry into the Nature and Causes of the Wealth of Nations (New York: Modern Library,
1994 [1776]).
Walker, M., and J. Wooders. “Minimax play at Wimbledon,” The American Economic Review 91, no. 5
(2001): 1521–1538.
van’t Wout, M., R.S. Kahn, A.G. Sanfey, and A. Aleman. “Affective state and decision-making in the
ultimatum game,” Experimental Brain Research 169, no. 4 (2006): 564–568.
Robert, C., and P.J. Carnevale. “Group choice in ultimatum bargaining,” Organizational Behavior and
Human Decision Processes 72, no. 2 (1997): 256–279.
Hoffman, E., K. McCabe, and V. Smith. “Social distance and other-regarding behavior in dictator
games: Reply,” American Economic Review 86, no. 3 (1996): 653–660.
Chapter 2
Drakert, W. “Normandy: Game and reality,” Moves: Conflict Simulation, Theory and Technique 6
(1972): 6–7.
von Neumann, J., and O. Morgenstern. The Theory of Games and Economic Behavior (Princeton:
Princeton University Press, 1944).
Nash, J. “Equilibrium points in n-person games,” Proceedings of the National Academy of Sciences 36,
no. 1 (1950): 48–49.
Chapter 3
Wilson, J., and G. Kelling. “Broken windows,” The Atlantic, March issue (1982).
Kelling, G., and C. Coles. Fixing Broken Windows: Restoring Order and Reducing Crime in Our Com-
munities (New York: Free Press, 1996).
Cournot, A.A. Researches on the Mathematical Principles of the Theory of Wealth (Norwood, MA: Nor-
wood Press, 1897 [1838]).
Bertrand, J. “Book review of theorie mathematique de la richesse sociale and of recherches sur les prin-
ciples mathematiques de la theorie des richesses,” Journal de Savants 67 (1883): 499–508.
Chapter 4
Homer. The Odyssey (8th century B.C.).
278
REFERENCES
Chapter 5
Arrow, K. “A difficulty in the concept of social welfare,” Journal of Political Economy 58, no. 4 (1950):
328–346.
Gibbard, A. “Manipulation of voting schemes: A general result,” Econometrica 41, no. 4 (1973):
587–601.
Satterthwaite, M.A. “Strategy-proofness and arrow’s conditions: Existence and correspondence the-
orems for voting procedures and social welfare functions,” Journal of Economic Theory 10, no. 2
(1975): 187–217.
May, K.O. “A set of independent necessary and sufficient conditions for simple majority decisions,”
Econometrica 20, no. 4 (1952): 680–684.
Kang, S.C., and G.B. Powell. “Representation and policy responsiveness: The median voter, election
rules, and redistributive welfare spending,” Journal of Politics 72, no. 4 (2010): 1014–1028.
Clarke, E. “Multipart pricing of public goods,” Public Choice 11, no. 1 (1971): 17–33.
Groves, T. “Incentives in teams,” Econometrica 41, no. 4 (1973): 617–631.
Chapter 6
Harinck, F., and C.K.W. De Dreu. “Negotiating interests or values and reaching integrative agree-
ments: The importance of time pressure and temporary impasses,” European Journal of Social Psy-
chology 34, no. 5 (2004): 595–611.
Coase, R.H. “The nature of the firm,” Economica 4, no. 16 (1937): 386–405.
Coase, R.H. “The problem of social cost,” Journal of Law and Economics 3, no. 1 (1960): 1–44.
Chapter 7
Ledyard, J.O. “Public goods: A survey of experimental research.” In J. Kagel and A. Roth (eds) The
Handbook of Experimental Economics (Princeton: Princeton University Press, 1997).
Camerer, C., and T-H. Ho. “Behavioral game theory experiments and modeling.” In Handbook of
Game Theory with Economic Applications, vol. 4 (Amsterdam: Elsevier, 2015): 517–573.
Ostrom, E. Governing the Commons: The Evolution of Institutions for Collective Action (New York: Cam-
bridge University Press, 1990).
Dawkins, R. The Extended Phenotype (Oxford: Oxford Landmark Science, 1982).
Maynard Smith, J. Evolution and the Theory of Games (Cambridge: Cambridge University Press, 1982).
Wright, R. Nonzero: The Logic of Human Destiny (New York: Pantheon Books, 1999).
Axelrod, R. The Evolution of Cooperation (Cambridge, MA: Revised edition, Perseus Books
Group, 2006).
Chapter 8
Akerlof, G. “The market for ‘Lemons’: Quality uncertainty and the market mechanism,” The Quarterly
Journal of Economics 84, no. 3 (1970): 488–500.
Gates, S., and K. Strøm. “Partisan versus military leadership and peace-building: Adverse selection and
civil conflict,” (Olso: Peace Research Institute Oslo, University of Oslo, mimeo, 2012).
Moehler, M. “The Rawls–Harsanyi dispute: A moral point of view,” Pacific Philosophical Quarterly 99,
no. 1 (2015): 82–99.
Chapter 9
Spence, M. “Job market signaling,” Quarterly Journal of Economics 87, no. 3 (1973): 355–374.
Lang, K., and D. Kropp. “Human capital versus sorting: The effects of compulsory attendance laws,”
The Quarterly Journal of Economics 101, no. 3 (1986): 609–624.
279
REFERENCES
Bedard, K. “Human capital versus signaling models: University access and high school dropouts,”
Journal of Political Economy 109 (2001): 749–775.
Chapter 10
Haywood, O.G. “Military decision and game theory,” Journal of the Operations Research Society of
America 4, no. 2 (1954): 365–385.
Poundstone, W. Prisoner’s Dilemma: John Von Neumann, Game Theory and the Puzzle of the Bomb (New
York: Anchor Books, 1993).
George, P. Red Alert (Cabin John, Maryland: Wildside Press, 2008 [1958]).
Cotton, C., and C. Liu. “100 horsemen and the empty city: A game theoretic examination of decep-
tion in Chinese military legend,” Journal of Peace Research 48, no. 2 (2011): 217–223.
Jelmov, A., Y. Tauman, and R. Zeckhauser. “Attacking the unknown weapons of a potential bomb
builder: The impact of intelligence on the strategic interaction,” Games and Economic Behavior 10
(2017): 177–189.
Taylor, P. The Thirty-Six Stratagems: A Modern-Day Interpretation of A Strategy Classic (Oxford: Infinite
Ideas, 2013).
Chapter 11
Iriberri, N., and J.R. Uriarte. “Minority language and the stability of Bilingual equilibria,” Rationality
and Society 2, no. 4 (2012).
Pascal, B. Pensées (1670). Translated by Krailsheimer, A., (London: Penguin, 1995).
Sober, E. “To give a surprise exam, use game theory,” Synthese 115 (1998): 355–373.
Ferreira, J.L., and J. Zamora. “The surprise exam paradox, rationality, and pragmatics: A simple game
theoretic analysis,” Journal of Economic Methodology 15, no. 3 (2008): 285–299.
Bordignon, M., T. Nannicini, and G. Tabellini. “Moderating political extremism: Single round versus
runoff elections under plurality rule.” American Economic Review 106, no. 8 (2016): 2349–2370.
Chapter 12
Maynard Smith, J., and G.R. Price. “The logic of animal conflict,” Nature 246 (1973): 15–18.
Maynard Smith, J. Evolution and the Theory of Games (Cambridge: Cambridge University Press, 1982).
Schelling, T. “Sorting and mixing: Race and sex.” In Micromotives and Macrobehavior (New York: W.
W. Norton and Co., 1978).
Gilboa, I., and A. Matsui. “Social stability and equilibrium,” Econometrica 59, no. 3 (1991): 859–867.
Lewis, D. Convention (Cambridge: Harvard University Press, 1969).
Chapter 13
Tsvetanov, T., and K. Segerson. “Re-evaluating the role of energy efficiency standards: A behavio-
ral economics approach,” Journal of Environmental Economics and Management 66, no. 2 (2013):
347–363.
Grubb, M. “Overconfident consumers in the marketplace,” Journal of Economic Perspectives 29, no. 4
(2015): 9–36.
Stango, V., and J. Zinman. “Exponential growth bias and household finance,” The Journal of Finance
64, no. 6 (2009): 2807–2849.
Stango, V., and J. Zinman. “Limited and varying consumer attention: Evidence from shocks to the
salience of bank overdraft fees,” The Review of Financial Studies 27, no. 4 (2014): 990–1030.
Thaler, R., and C. Sunstein. Nudge: Improving Decisions about Health, Wealth, and Happiness (New
Haven & London: Yale University Press, 2008).
280
REFERENCES
Greene, J. Moral Tribes: Emotion, Reason, and the Gap Between Us and Them (London: Penguin
Press, 2013).
McKelvey, R., and T. Palfrey. “Quantal response equilibria for normal form games,” Games and Eco-
nomic Behavior 10, no. 1 (1995): 6–38.
Costa-Gomes, M., V. Crawford, and B. Broseta. “Cognition and behavior in normal-form games: An
experimental study,” Econometrica 69, no. 5 (2001): 1193–1235.
Wright, J.R., and K. Leyton-Brown. “Predicting human behavior in unrepeated, simultaneous-move
games,” Games and Economic Behavior 106 (2017): 16–37.
Chapter 14
Shapley, L.S., and M. Shubik. “A method for evaluating the distribution of power in a committee
system,” American Political Science Review 48, no. 3 (1954): 787–792.
Banzhaf, J.F. “Weighted voting doesn’t work: A mathematical analysis,” Rutgers Law Review 19, no. 2
(1965): 317–343.
Geller, C.R., J. Mustard, and R. Shahwan. “Focused power: Experiments, the Shapley–Shubik power
index, and focal points,” Economics Discussion Paper No. 2013–42 (2013).
Kóczy, L.Á. “How Brexit affects European Union power distribution,” Working Paper Series 1601,
(2016) Óbuda University, Keleti Faculty of Business and Management.
Bilbao, J.M., J.R. Fernández, N. Jiménez, and J.J. López. “Voting power in the European Union
enlargement,” European Journal of Operational Research 143 (2002): 181–196.
Freixas, J. “The dimension for the European Union Council under the Nice rules,” European Journal
of Operational Research 156 (2004): 415–419.
281
INDEX
100 horsemen, 186 Ambush, 184

Anonymity, 87, 131
A Anonymous interaction, 16
Ability, 138–9, 159–163 Arbitrator, 260
overestimation of, 245 Arms race, 178, 179, 182
Abortion, 250 Arrow, Kenneth, 86
Absence of agreement, 112, 114, 217, 197 Arrow’s impossibility theorem, 81, 86–7
Absolute majority, 261, 265, 267, 269, 275–6 Auction, 104
Abstention, 260, 275, 277 Axelrod, Robert, 120, 134, 135
Ad hoc assumptions, 120 Axelrod’s tournament, 120, 134
Adaptation, 134
Addiction, 241 B
Adverse selection, 138, 146–159, 170 Backwards induction, 62–68
Advertising, 9, 19, 165 Bailout, 7, 195–197
Age of consent, 248 Bankruptcy, 7, 75, 152, 195
Age of majority, 247–250, 253 Banzhaf, 263
Agency problem, 138–176 Banzhaf ’s power index, 263–4
Agents, 138–9, 147, 152–155, 159, 166, 169–70 compare Shapley-Shubik power index
Aggregated preferences, 81, 83, 86–7 Barcelona, 96–99, 103
Agreement, 2–3, 6–8, 104, 116, 122–3, 132, 153, Bargaining see negotiation
195, 267 Basic research, 9
absence of, 112, 114, 217, 197 Basque Autonomous Community, 268
binding, 192, 260, 264, 270 Basque parliament, 268–270
collective labor, 167 The battle of the Bismarck sea, 177, 178
final, 113 Battle of the sexes, 4, 21, 38–9, 62, 140, 156,
free trade, 193–4 232, 238
international, 41, 66 dynamic, 62–67
market-sharing, 66 Bedard, K., 164
terms of, 10 Behavioral economics, 15, 240, 254
verbal, 65, 174; see also compromise Behavioral dynamics, 41
Aid, financial, 18 Behavioral theory, 26, 54, 223, 240, 256
Akerlof, George, 152, 159 Behavioral history, 25
Akerlof ’s market for lemons, 152, 159 Beliefs, 6–7, 27, 130–1, 208
Alakrana, 113 Benedict XV, Pope, 121
All or nothing, 110 Bertrand, Joseph Louis François, 56
Allocation, public, 104 Bertrand oligopoly, 41, 56, 57, 58, 61
Altruism, 16, 131, 133–4; see also altruistic pref- compare competition, competitive market,
erences; charity; empathic preferences; Cournot oligopoly, monopoly
solidarity Bertrand paradox, 57
compare selfish preferences, selfishness Best reply see best response
Altruistic preferences, 18, 120, 123; see also altruism; Best response, 11, 30, 33–37, 41–44, 53, 235
charity; empathic preferences; solidarity quantal, 256, 257
compare selfish preferences, selfishness Bet, 12, 185–6, 198
282
INDEX
Bias, cognitive, 26, 254; see also cognitive process; Cognitive,

cognitive science; c ognitive system bias, 26, 254
Bilbao, J.M., 274 process, 257
Bildu, 268–70 science, 10
Bilingual, 200–204, 220 system, 134
Binding agreement, 192, 260, 264, 270 Coherence, 83, 85–6
Biology, 133–4 Cold war 177, 180–182
Bipartisanship, 267 Coles, Catherine, 50
Bismarck sea, the battle of, 177–8 Collective labor agreement, 167
Blockade, 198 Combination of strategies, 20, 26, 29–30
Bluffing, 177, 184–186, 198 Committee, 100, 182, 260–1, 270, 276
Borda count, 84–7, 94, 99, 102 Genetics and Insurance, 151
Bordignon, M., 217 Commitment, 5–6, 151–153, 197, 240–1
Brezhnev, 180 Common zone, 80
Brexit, 272–274 Commons, tragedy of the, 116, 132
British conservative party, 267–8 Communicate, 3–4, 228
British parliament, 151, 267, 277 Communication, 2, 183, 228, 229, 230
Broken windows theory, 41, 48–50 game, 229–30
Burning the money game, 233 Community, 28, 132, 153, 198
Bunge, 205–6 international, 198
rebel, 153
C Companies, 116, 246–7
Calculus, 33–4, 40–1, 55–58, 119 insurance 139, 148–9, 154, 166, 170–1
Camerer, Colin, 130–1, 137 Competition, 53, 55–58, 138, 146–7, 154,
Capital, human, 159, 164 165–6, 169–70, 246–7
Car insurance, 245 banking, 247
Cartel, 9, 55–6, 66, 132, 154 perfect, 55, 57, 146, 147, 154; see also
Catalonia, 96–99, 103 monopoly; oligopolistic model
Categorical imperative, 11 Competitive market, 16, 19, 55, 57, 70, 104,
Change of equilibrium, 49 147, 154, 246
Charity, 213–4; see also altruism; altruistic compare monopoly, oligopolistic model
preferences; empathic preferences; Competitive price, 154; see also market price
solidarity compare monopoly price
compare selfish preferences, selfishness, Complete information, 62
Chicken game, 5–7, 71, 136–7, 179–181, Complexity, 87, 134, 244, 275
256 Compromise, 70–72, 80, 217; see also a
Child custody, 247, 250 greement
Christmas truce, 120, 125 Conditionally cooperative, 132
Choice of standards, 2, 4, 38–9, 228 Condorcet paradox, 83
CIA, 180 Condorcet winner, 89, 95
Civil disobedience, 50–52 Congestion, 9, 214, 219
Civil war, 153 Consent, 248–250, 258
Circe, 69 age of, 248
City council, 17, 82, 90 Consistency, 50, 112, 134
Clarke, E., 92 of preferences, 241
Cleaning the street game, 29–33, 236 time, 242
Coalition, 104, 216, 260–277 Consumption, 58–9, 159, 214, 242–3,
Winning coalition, 264, 270, 276–7 246
Coase, Ronald, 115 preferences on relative, 58–9
Coase theorem, 105, 114–118 Context, 11, 106, 133, 255
283
INDEX
Contract, 64–66, 133, 153, 167–170, 244–247 D-day, 22

design, 139 De Dreu, Carsten, 112
insurance, 148, 151, 171 Debt 58, 76–7, 194, 254
job, 147 the payment of the debt game, 76–80
mortgage, 254 Declaration, 91–93; see also message
prisoners’ dilemma with a previous, 64–66 Default option, 154, 255
social, 214–5 Defeater, 129
Contribution, 9, 80, 130–1, 174–176, 213, 215, 235 Demand, 247, 53–57
voluntary, 9, 80, 130, 213, 215, 235 Democracy, 81, 86–7
Convergence, 235, 237 Democratic Unionist Party, 267–8
Convoy, 177, 183–4 Desperate strategy, 184
Cooperation, 11–2, 120–136, 174, 192–3, 260 Deterrence, 208–213, 221–2
Cooperative, 116, 174–5 Dictatorship, 58, 86, 213
conditionally, 132 Dilemma,
Cooperative game, 104, 182, 260–262, 270–272 discursive, 81, 95–6
Cooperative game theory, 104, 182, 260 prisoners’, 7–12, 18, 33, 38–9, 140–1, 174,
compare non-cooperative game theory 179, 181–2, 192–3
Cooperator, 129, 135 Discount factor, 107–8, 118–9, 242–245
Coordination game, 1, 19, 84 Discount price, 244
Copayment, 170–173 Discrimination, 151, 159, 164–5, 214,
compare payment by performance 228
Cortés, Hernán, 79 preferences for, 228
Cotton, C., 186 Discursive dilemma, 81, 95–6
Council Disobedience, civil, 50–52
city, 17, 82, 90 Dispersion,
of the European Union, 272, 274 of information, 154
United Nations Security, 272, 277 of payoffs, 77
Counteroffer, 106, 118–9 of wealth, 214
Cournot, A.A., 53 Distribution of income, 115
Cournot oligopoly, 53–61, 239 Diversity, 227–8
compare Bertrand oligopoly, competition, preferences for, 227
competitive market, monopoly Divorce, 110, 143–145, 250
Coverage, health, 74 Doctrinal paradox see discursive dilemma
Credible threat, 209 Dominated strategy, 21, 25
Credibility, 5, 69, 70–73, 128, 197, 210–1 elimination of, 24, 28, 35
Credit card, 247 compare weakly dominated strategy
Credit market, 152, 245 Doomsday machine, 180–182
Criterion, 46, 242, 248, 260 Double round vote, 95, 100–103
election, 151 Drakert, William, 22
ex ante, 151, 155 Dropout, 264
justice, 155 Drug, 141
maximin, 38, 156 DUP see Democratic Unionist Party
prudent, 23 Dynamic game, 62–3, 66–68, 104, 122; see also
Rawl’s, 155 extensive form game
Cultural evolution, 133 compare static game, normal form game
Custody, of children, 247, 250 Dynamics, 26, 41, 54, 211, 223–236
adaptive, 54, 235–236
D behavioral, 41
Dation in payment, 254 best reply, 61, 228–231
Dawkins, Richard, 134 evolutionary, 223–226
284
INDEX
imitation, 224, 226–227 European parliament, 260

no weakly dominated strategies, 232–234 European Union, 27, 74, 247, 268, 272–274
social, 26, 253 Council of the, 272, 274
Eurovision song contest, 84
E Evidence, 112, 120, 163–4, 205, 240, 254, 264,
Economics 269
of information, 152, 159 empirical, 120, 163, 254, 264
behavioral, 15, 240, 254 experimental, 112, 240
Economic theory, 192 historical, 269
Education, 9, 156, 159–165, 214, 248–9 Evolution, 120, 132–134, 223–226
as a signal, 159, 161–164 cultural, 133
mandatory, 249 Evolutionarily stable strategy, 224–226, 237
Efficiency, 115–6, 170 Evolutionary game, 224, 232
energy, 244 Ex ante criterion, 151, 155
Elasticity, 247 Experience, 131, 169, 248
Effort, 48, 138–9, 146, 152–3, 160, 166–174 Experimental evidence, 112, 240; see also empirical
Elections, 4, 84, 87, 95–6, 102, 215, 217, 265–268 evidence; experiments; h istorical evidence
criterion, 151; see also vote; voting Experiments, 16, 106, 112, 120, 130–1, 134, 226,
Electoral platform, 42–45, 58, 87 257, 264
Emissions, 9, 243 on irrationality, 245–6, 255, 257
Emission permits, 9 laboratory, 11, 15, 109, 245–6
Emotion, 133 natural, 164, 217
Empathic preferences, 19, 140–1; see also altruism; on negotiation, 112
altruistic preferences; charity; solidarity on power indices, 264
compare selfish preferences, selfishness on prisoners’ dilemma, 11, 130–1
Empirical data, 20, 132 though, 156, 255
Empirical question, 50 on ultimatum game, 15–6, 106, 109; see also
Empirical research, 90 evidence
Empirical evidence, 120, 163, 254, 264; see also Explanatory power, 226
experimental evidence; historical evidence Expert, 152–3, 158
Empirical validation, 159 Expert opinion, 248
Employee, 166–169 Extensive form, 3–6, 6–7
Employer, 159, 167–169 game, 142–3, 158, 190, 203; see also dynamic
Empty city, 186 game
Empty fort strategy, 184, 186 compare normal form game, static game
Empty threat, 62, 67 Externality, 105, 115–118
Encrypting system, 183–4 of consumption, 59
Entry game, 70–1 of production, 114
Environment, 9, 60, 135, 223, 242, 256
Environmentally friendly, 135 F
Equality, 86–7, 214–5 Fanaticism, 113, 110–1
preferences for, 215 Fanatic’s preferences, 111
Equilibrium Fairness, 131, 151
change, 49 Fair price, 154
multiple, 1, 2, 4, 92, 127, 223 Faithfulness, 143–145, 157
pooling, 163–4 Federal country, 50, 102
separating, 163–4 Federal government, 51
Espionage, 177, 184 Federal state, 102
Ethics, 11, 187, 214, 241 Federer, 13
EU see European Union Fernández, J.R., 274
285
INDEX
Ferreira, José Luis, 208 Genetics, 133

Final agreement, 113 Genetics and Insurance Committee, 151
Financial aid, 18 Gilboa, Itzhak, 232
Financial market, 152, 245 Gini index, 214
Fines, 218–9; see also penalties George, Peter, 182
Finitely repeated games, 121–3 Gibbard, Allan, 87
compare with indefinitely repeated games, Gibbard-Satterthwaite theorem, 87, 97
infinitely repeated games Give up control, 72
First welfare theorem, 12 God, 200, 204–206, 221
Fiscal battle, 41–2; see also tax war Goddess, 69
Fiscal benefits, 50–1 Golden Rule, 11
Food, 223 Good,
healthy, 254 public, 9, 130, 137, 213–4, 235
Football, 21 game, 137
Forgiving, 135 voluntary contribution to the provision
Fraternity, 86 of, 235
Free rider, 174, 214–5 second-hand see used good
Free trade, 192–194, 199 used good, 139, 152, 158
agreement, 193–4 The great book of game theory, 26
Freixas. J., 274 Great Recession, 194
Friendly, 79, 135, 242 Greece and the Troika 177, 194–199
environmentally, 135 Greek parliament, 197
Frustration, 200–204 Groves, T., 92
Function Groves-Clarke mechanism, 91–2, 94
payoff, 21, 29, 66 Grubb, M., 245, 246
utility, 29, 33–4, 155–6, 240 Guang, Li, 186
Future self, 69 Guarantees, 2, 23–4, 165, 173, 217
Gulf, second war, 187
G
Game H
bargaining, 104, 106, 115 Harinck, Fieke, 112
dynamic, 62–3, 66–68, 104, 122 Harsanyi, John, 156
cooperative, 104, 182, 260–262, 270–272 Hawk-dove game, 223–226
extensive form, 142–3, 158, 190, 203 Haywood, O.G., 178
evolutionary, 224, 232 Health, 75, 148, 166, 173, 250, 254
majority, 260–264 coverage, 74
normal form, 17, 20–1 index, 148
repeated, 120–136, 192–3, 209–12 indicator, 170
zero-sum, 12, 14, 22–28, 108, 134, 178, 182, insurance 75, 148–158, 170, 173
186, 223; see also under individual game system, 74–5, 138, 147, 159
names Healthcare, 73–75, 248
Gas station, 45 Healthcare game, 151
Gates, Scott, 153 Healthy food, 254
Gazelle, 166 Heaven, 205, 221
Genes, 133 Hell, 205, 221
Genetic Heterogeneity, 131
code, 133 High-risk, 149–50
material, 166 Hijacking, 113–4
predisposition, 154 Historical evidence, 269; see also empirical evi-
test, 151 dence; experimental evidence
286
INDEX
History, 122–125, 130, 134, 182, 187, 192, 197, Inspectors, 188–9
223, 247, 268 Insurance, 73–4, 139, 147–154, 157, 170–173
of behavior, 25 car, 245
Ho, Teck-Hua, 130–1, 137 company, 139, 148–9, 154, 166, 170–1
Hobbes, 11–2 health, 75, 148–158, 170, 173
Hofstadter, Douglas, 10, 56 market, 148–150, 246
Hostage, 113 medical, 139, 152
Human capital, 159, 164 policy, 73–4, 148–154
Hungary, 84 Intelligence, 134
Hussein regime, 187 Intelligence services, 187, 189, 191
International
I agreement, 41, 66
Identity, 86–7, 117 community, 198
Ideology, 42–45, 214–5 Iriberri, Nagore, 200
Imamura, Admiral, 177–8 Irrationality, 6, 130, 180, 223, 244–5, 254–5
Impatience, 107–109, 113, 116 compare limited rationality, rationality
Imperfection, 81, 86–7, 154 Iterative process, 54, 235
Income distribution, 115
Incoherence, 87 J
Inconsistency, time 175–6, 242–3 Jiménez, N., 274
Incumbent, 117, 225–6 Jelmov, A., 187
Indefinitely repeated games, 123–4, 127, 209 Job market, 159, 164–166
compare finitely repeated games, infinitely Justice, 86, 155
repeated games criterion, 155
Indemnity, 148–9, 170, 171, 172, 173
Independence, 50, 51, 52, 61, 84, 96, 97, 98, 99, K
102, 103, 268
Kahneman, 254
Independence of irrelevant alternatives, 85,
Kang, S.C., 90
86, 87
Kant, Immanuel, 11
Indices
Kelling, George, 50
Gini, 214
Kenney, General, 177–8
of health, 148
Keeping up with the Joneses, 58, 60
of power, 260–277; see also Banzhaf power
KGB, 180
index; Shapley-Shubik power index
King Solomon, 166, 176
Indoctrination, 253–4, 258–9
Kino, 154
Inefficiency, 117–8
Kóczy, László Á., 274
Inequality, 214–5
Kropp, D., 164
Infinitely repeated games, 124–30, 136–7, 199,
209–10, 221–2; see also finitely repeated Kubrick, Stanley, 182
games; indefinitely repeated games
Information, 177, 180, 183–4, 190, 240, 244, 247, L
252–3, 334 Labor market, 159, 164–166
complete, 62 Labor party, 267–8
dispersion, 154 Lang, K., 164
economics, 152, 159 Language, 228, 232
private, 62, 91, 138, 154 majority, 200, 220
revelation of, 92, 138, 154, 177, 183 minority, 201–3, 220
set, 138, 140–144, 160, 168, 176, 190 game, 200–203
Inspection, 39, 53, 57, 87, 187–8, 189–191 Lehendakari, 268
game, 188–9 Leisure, 58–9, 117
cost, 189–91 Lemons, market for, 152, 159
287
INDEX
Lewis, David, 232 Maximum price, 257

Lex talionis, 208 May’s theorem, 87, 96
Leynard, John, 130–1 McKelvey, R., 256
Leyton-Brown, K., 257 Mechanism see vote
Liang, Zhuge, 186 Median voter, 41–58, 88–90
Liberal Democrats, 267–8 theorem, 87–89, 100–1
Liberty, 86–7 Medical insurance, 139, 152
Limited rationality, 136, 240–257 Message, 5, 78, 166, 190–1, 232; see also declaration
compare irrationality, rationality Minimum participation, 83–4, 99
Liu, C., 186 Minor, 247–253
Location game, 45 Minority, 90, 96, 153, 226–7, 238
Logrolling, 91 language, 200–203, 220
López, J.J., 274 language game, 200–203
Low-risk, 149–50, 154, 166 Mistakes, 246, 257
Mixed strategy, mixed, 4, 14, 27, 31, 135, 189
M Moctezuma, 79
MAD see mutually assured destruction Moehler, Michael, 156
Majority, 46, 51–2, 82–3, 87–91, 94–96, 100, Monitoring, 49, 132
102, 226–7, 260–277 Monolingual, 200–203, 220
absolute, 261, 265, 267, 269, 275–6 Monopoly, 55, 57–8, 78, 154
age of, 247–250, 253 power, 154
game, 260–264 price, 56–7
language, 200, 220 compare competitive, market price
qualified, 261, 272, 274 compare competition, oligopolistic model
vote, 96, 100 Monotonicity, 83–4, 86–7, 100
simple, 272 Montenegro, 84
Mandatory Moral hazard, 159, 166–175
education, 164, 249 Morality, 113, 214, 255
insurance, 151 Moral preferences, 112, 156
pension, 241 Morgenstern, Oskar, 24, 26, 182
policy, 151 Mortgage, 254
universal healthcare, 75 Multiplicity of equilibria, 1, 2, 4, 92, 127, 223
Manipulation Music, 240–1
of the agenda, 82, 99 Mutant, 225, 226, 237
of behavior, 255 Mutation, 224–5,
Market Mutually assured destruction, 180–182
competitive, 16, 19, 55, 57, 70, 104, 147, 154, 246
financial, 152, 245 N
insurance, 148–150, 246 Nadal, 13
for lemons, 152, 159 Nannicini, T., 217
labor, 159, 164–166 Nash, John 11, 26
second hand goods see used goods market Nash equilibrium
sharing agreement, 66 definition, 22, 26–28
for used goods, 139, 152, 158 finding, 29, 30, 34
Market price, 53–4, 104; see also competitive price in mixed strategies, 31
compare monopoly price non-existence, 36
Matching pennies game, 12–14, 23, 27, 177, 239 theorems, 28
Matsui, Akihiko, 232 Negotiation, 7, 96, 104–119, 132, 153, 177, 194–197
Maximin criterion, 38, 156 cost, 115, 117
Maximin strategy, 21–24, 28, 223 games, 104–117
288
INDEX
peace, 153 Parliament, 45, 82, 84, 94, 96, 100, 151, 197,
power, 107–113, 153, 194 260–277
Nervous system, 133–4 Basque, 268–270
Neutrality, 87 British, 151, 267, 277
New Democracy, 196 European, 260
Newcomer, 5, 196, 226 Greek, 197
Nixon, 180 Spanish, 262, 264–268, 276
Nobel Prize, 11, 86, 115, 132, 152, 156, 159, Participation, minimum, 83–4, 99
182, 254; see also under individual Partido Socialista de Euskadi, 268–70
names Parties, political, 39, 42–46, 50, 58, 82, 90, 94,
Non-cooperative game theory, 182, 260 152–3, 196, 215–217, 261, 265, 267–8
compare cooperative game theory nationalist, 267–8; see also under individual
Normal form game, 17, 20–1; see also static names
game Pascal, Blaise, 204
compare dynamic game, extensive form game Pascal’s wager, 204–206, 221
Normandy, 22 PASOK, 196
Nuclear war, 178, 182 Patience, 33, 105, 113, 128
Nuclear weapons, 179, 182, 198 Payment, 75–6, 92, 113–4, 244–247, 254, 261
Nudges, 254–5 dation in, 254
of the debt game, 75–77
O fixed 166–170
Oak Park, 228 by performance, 166–170
Odysseus and the Sirens, 69–70, 79 compare copayment
Offers, 9, 15, 43–4, 56, 74, 105–119, 138–175, Payoff function, 21, 29, 66
216, 245–6, 254 Peacocks, 166
Take-it-or-leave-it offer, 115–6, 119 Pearls, 154
Old-me, 240–242 Pearl dealer, 154
compare young-me Pearl of the World, 154
Oligopolistic model, 16, 53, 55, 58, 123; see also Penalties, 9, 38, 50, 52, 65, 79–80, 116, 132, 148;
Bertrand oligopoly; Cournot oligopoly see also fines
compare competition, monopoly Penalty kick, 19
OPEC, 132 Pension plan, 241; see also retirement
Open borders policy, 193 Pension system, 195
Opera, 9–10 Performance, payment by, 166–170
Operation Opera, 187 Philosophy, 81, 156, 200, 214
Ostrom, Elinor, 132
Pigouvian taxes, 115, 244
Overdraft, 147, 245
Pirates, 105, 113–4
Overconfidence, 245–247
Pivotal voter, 97–8, 103, 262–264, 270–272
Overfishing, 116, 132
Plaid Cymru, 206–7
Overrepresentation, 267
Players set, 21
Plurality vote, 94, 215–217, 268
P
PNV, 268–70
Pairwise confrontation, 135, 137
Pairwise vote, 89, 101 Poker, 62, 185–6
Palfrey, T., 256 Policy, 24, 182, 192–3, 217, 219, 247, 254–5
Paradox, 6, 41, 166, 208 insurance, 73–4, 148–154
Bertrand, 57 MAD, 182
Condorcet’s, 83 mandatory, 151
doctrinal, 81, 95–6 open border, 193
surprise exam, 206–7, 221 protectionist, 192
Parking, 218–220, 222 public, 247
289
INDEX
Political party see parties Principal-agent problem, 138

Pollution, 9, 105, 143, 244 Prisoners’ dilemma, 7- 12, 18, 33, 38–9, 140–1,
Pope Benedict XV, 121 174, 179, 181–2, 192–3
Positive response, 87 book by Poundstone, William, 180
Pooling equilibrium, 163–4 with a previous contract, 64–66
Poor, 12, 214–5; see also rich repeated, 120, 123–126, 129–137
Poundstone, William, 182 Private information, 62, 91, 138, 154
Powell, G.B., 90 Prizes, 123, 126
Power Process
explanatory, 226 cognitive, 257; see also cognitive bias; cognitive
monopoly, 154 science; cognitive system
negotiation, 107–113, 153, 194 iterative, 54, 235
separation of, 72–3, 104, 114 Productivity, 114, 146–7, 157, 163–4
Power indices, 260–277 Profile of strategies, 20, 26, 29–30
Shapley-Shubik, 262–3, 267 Property rights, 115, 117
Banzhaf, 263–4 Protectionism, 192–3
Predators, 135, 166 Prudent criterion, 23
Predatory strategy, 135 Prudent strategy, 22–3
Preemptive war, 187, 191 PSE see Partido Socialista de Euskadi
Preferences, 3, 37, 49 Public allocation, 104
aggregated, 81, 83, 86–7 Public goods, 9, 130, 137, 213–4, 235
altruistic, 18, 120, 123; see also altruism voluntary contribution to the provision of, 235
changing with time, 240, 252 Public goods game, 137
for discrimination, 228 Public policy, 247
for diversity, 227 Punishment, 7, 9, 123, 126, 128–9, 132–135, 193,
empathic, 19, 140–1 209–10, 213
for equality, 215
Q
fanatic’s, 111
Quality, 139, 147, 151–153, 165, 218
moral, 112, 156
Qualified majority, 261, 272, 274
on relative consumption, 58–9
Quantal best response, 256
selfish, 106, 131, 141
Quantal level-k model, 256–7
single peaked, 88–9, 101
Quota, 261–264, 267–272, 274–5, 277
social see aggregated
time, 126, 242–244 R
changing with, 240, 252 Random circumstances, 138
transitive, 83 Random distribution, 227
Presidents, 28, 100, 182, 262, 265, 268 Random strategy see mixed strategy
Price Random walk, 231
competitive, 154 Randomization, 14
discount, 244 Randomize, 14, 23, 31–2, 199
fair, 154 Rationality, 3, 10, 26, 130–1, 153, 204, 223, 231,
market, 53–4, 104 235, 247, 254–6
maximum, 257 compare limited rationality, irrationality
monopoly, 56–7 Rationalizable strategy, 25–27, 223
compare competitive price, market price Rawls, John, 138, 151, 155, 156
regulation, 247 Rawl’s
war, 77–8, 166 original position, 151, 155
Price, George, 226 veil of ignorance, 138, 155
Prime minister, 45, 94, 101–2, 196–7, 262 criterion, 155
Principal, 138–9, 147, 152–154, 166, 170 Rawls-Harsanyi dispute, 156
290
INDEX
Rebel, 5, 153, 182 high risk, 149, 150

Rebel community, 153 insurance 139
Rebellion, 253 low risk, 149, 150, 154, 166
Rebel without a cause, 5 risk aversion, 38, 77, 105, 109, 113, 148, 156, 169
Reciprocity, 16, 133–4 risk neutrality, 77, 109, 118, 148
Red Alert, 182 type see risk group
Referendum, 50–1, 83–4, 96–99, 197 Rock-paper-scissors, 19, 40
Refugees, 84 Rule see vote
Regret, 2, 4, 13, 240, 242–3, 248, 253, 258 Runoff elections, 215, 217
Regulation of price, 247
Rehabilitation, 208, 213, 242 S
Relative consumption, preferences on, 58–9 Salary, 59, 119, 139, 146–7, 157, 160–170
Religion, 214 reservation salary, 146–7, 157, 161, 163, 169, 175
Rent seeking, 117 Satterthwaite, Mark, 87
Repeated games, 120–136, 192–3, 209–212; See Saving, 52, 163, 240
also finitely repeated games; indefinitely energy, 242
repeated games; infinitely repeated games face-saving, 6
Repeated prisoners’ dilemma, 120, 123–126, 129–137 lives, 255
Reply, best see best response Saving-spending game, 240
Reputation, 5–6, 120, 129, 195–6, 209–211 Schelling’s segregation model, 226–7, 237
Research, 10, 81, 90, 117, 138, 143, 164, 248 Schelling’s tipping model see Schelling’s
Basic, 9 segregation model
Agenda, 120 Schelling, Thomas, 182, 226, 228
Line, 2, 41, 81, 223, 254 Science
Researcher, 10, 38, 44, 95, 112, 164 cognitive, 10; see also cognitive bias; cognitive
social, 112 process; cognitive system
cognitive science, 10 Scottish National Party, 267–8
Reservation salary, 146–7, 157, 161–163, 169, 175 Second Gulf war, 187
Resource, 9, 39, 79, 116–7, 132–3, 173, 178, 194, Second-hand goods market see used good market
213–4, 223–226, 255, 264–5 Security, 86
common, 132 Segerson, K., 244
natural, 9 Segmentation, 214
Response Segregation, 226–228, 237–8; see also Schelling’s
best, 11, 30, 33–37, 41–44, 53, 235 segregation model
quantal, 256, 257 Self-inflicted injuries, 232
positive, 87 Self-referential proposition, 206
Responsibility, 86 Selfishness, 1, 1–2, 16, 131; see also selfish preferences
Responsiveness, 135 compare, altruism, altruistic preferences, charity,
empathic preferences, solidarity
Restitution, 208
Selfish preferences, 106, 131, 141; see also selfishness
Retirement, 240–1; see also pension plan
compare, altruism, altruistic preferences, charity,
Results, payment by, 166–170
empathic preferences, solidarity
Revelation of information, 92, 138, 154, 177, 183
Send the dossier game, 65–6
Revenge, 208
Sentence, 8, 114, 208, 212
Rich, 214–5, 222; see also poor
Separating equilibrium, 163–4
Rights
Separation of powers, 72–3, 104, 114
minor’s 249
Set
property 115, 117
information, 138, 140–144, 160, 168, 176, 190
Risk, 4, 37, 71, 77, 105, 109, 114, 155, 185, 213
of players, 21
attitude, 37, 156
of strategies, 21, 25, 27, 66; see also strategy
group, 139, 149–154, 166, 170 combination; strategy profile
291
INDEX
Shapley, L.S., 277 rationalizable, 25–27, 223

Shapley value, 262 set, 21, 25, 27, 66
Shapley-Shubik power index, 262–3, 267 tit for tat, 135–137
compare Banzhaf power index trigger, 125–130, 192–3, 209–211
Sharing agreement, 66 weakly dominated, 25, 28
Shubik, M., 262 Strom, Kaare, 153
Sincere vote, 81–2 Subgame, 68, 122–125, 142–3
compare strategic vote Subgame perfect Nash equilibrium, 62, 66,
Signaling, 159–166, 175, 211 122–125, 142–3
Single peaked preferences, 88–9, 101 Sunstein, Cass, 254
Single round, 215, 217 Surgery, 248, 250
Single transferable vote, 95 Surprise, 2, 83, 121, 179, 182, 187, 196,
Sinn Féin, 267 206–208, 221
Smith, Adam, 12 Surprise exam paradox, 206–7, 221
Smith, John Maynard, 134, 226 Symmetry, 10–1, 55, 224, 276–7
Smoking, 49, 117, 166 Syriza, 196–7
SNP see Scottish National Party System
Sober, E., 206 health, 74–5, 138, 147, 159
Social choice, 86 cognitive, 134; see also cognitive bias; cognitive
Social contract, 215 process; cognitive science
Social preferences see aggregated preferences T
Social pressure, 9 Tabellini, G., 217
Solidarity, 86, 213–215 Take-it-or-leave-it offer, 115–6, 119
Solomon, King, 166, 176 Tax war, 24–5; see also fiscal battle
Spanish parliament, 262, 264–268, 276 Taxes, 17, 24–5, 39, 41–2, 50, 90, 115, 121–123,
Spence, M., 159 217, 243
Standardization, 214, 245 Pigouvian taxes, 115, 244
Standards, choice of, 2, 4, 38–9, 228 Teams, 16, 139, 159, 174–176
Stango, V., 247 Teenagers, 248
Static game, 41; see also normal form game Telephone companies, 247
compare dynamic game, extensive form Telephone services, 247
game Teleology, 133
Steinbeck, John, 154 Tennis, 13–4
Stiglitz, J., 159 Terms of agreement, 10
Dr. Strangelove, 182 Thaler, Richard, 254
Stratagems see strategy Theorems
Strategic vote, 81–2, 87–8, 97, 100; see also sincere Arrow’s impossibility, 81, 86–7
vote Coase, 105, 114–118
Strategy, 21 first welfare, 12
combination, 20, 26, 29–30 Gibbard-Satterthwaite, 87, 97
desperate, 184 May’s, 87, 96
dominated strategy, 21, 25 median voter, 87–89, 100–1
elimination of, 24, 28, 35 Nash equilibrium, 28
empty fort, 184–186 Theory
maximin, 21–24, 28, 223 behavioral, 26, 54, 223, 240, 256
mixed, 4, 14, 27, 31, 135, 189 broken windows theory, 41, 48–50
predatory, 135 of cooperative games, 104, 182, 260
profile, 20, 26, 29–30 non-cooperative game theory, 182, 260
prudent, 22–3 Thieves, 208–213, 221–2
random see mixed strategy Thirty-Six Strategies, 184
292
INDEX
Time von Neumann, John, 24, 26, 182

consistency, 242 Vote, 4, 45–6, 51, 81–103
inconsistency, 175–6, 242–3 majority, 96, 100
preferences, 126, 242–4 pairwise, 89, 101
changing with, 240, 252 plurality, 94, 215–217, 268
Tit for tat strategy, 135–137 sincere, 81–2
Tooth Fairy, 205–6 strategic, 81–2, 87–8, 97, 100
Tories see British Conservative Party Voter
Tosca, 9–10 median, 41–58, 88–90
Tragedy of the commons, 116, 132 pivotal, 97–8, 103, 262–264, 270–272
Trade Voting see vote
free, 192–194, 199
war, 194 W
Transparency, 147 War, 11, 40, 61, 79, 120–1, 178–9, 186
Transitive preferences, 83 civil, 153
Transitivity, 83, 86–7 cold, 177, 180–182
Trigger strategy, 125–130, 192–3, 209–211 nuclear, 178, 182
Troika, 177, 194–197, 199 preemptive, 187, 191
Trojan war, 69 price, 77–8, 166
Trolley problem, 255 second Gulf, 187
Truce, Christmas, 120, 125 tax, 24–5; see also fiscal battle
Truth, 91, 93–4 trade, 194
Truthful, 91, 93, 176 Trojan, 69
Truth-revealing mechanism 92 U-boat, 183
Tsipras, 197 World War I, 120
Tsvetanov, T., 244 World War II, 177
TV, 3, 45 Weakly dominated strategy, 25, 28
Tversky, 254 compare dominated strategy
Welfare, 12, 139, 200, 246
U First welfare theorem, 12
U-boat war, 183 Wealth, 12, 75–6, 115, 170, 214–5, 222, 254
Ultimatum game, 14–16, 18–9, 105–6, 130 distribution, 115, 214–5
Uncertainty, 109, 221 indicator, 170
Unfaithful, 143–145, 157 Weapons of mass destruction, 187, 192
Union, 104, 118–9 Wilson, James, 50
Unionist, 267 Winner, Condorcet, 89, 95
United Nations Security Council, 272, 277 World War I, 120
Uriarte, José Ramón, 200 World War II, 177
Used goods market, 139, 152, 158 Wright, J.R., 257
Utility function, 29, 33–4, 155–6, 240 Wright, Robert, 134
Expected utility, 37–8, 145, 151, 162, 169, 191,
250–252 Y
Young-me, 241–2
V compare old-me
Valuation, 48, 91–94, 218
Values, 86 Z
Verbal agreement, 65, 174 Zamora, Jesús, 208
Voltaire, 205–6 Zinman, J., 247
Voluntary contribution, 9, 80, 130, 213, Zero-sum game, 12, 14, 22–28, 108, 134, 178,
215, 235 182, 186, 223
293

Game Theory - An Applied Introduction

Uploaded by

Copyright:

Available Formats

You might also like

Game Theory - An Applied Introduction

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Game Theory - An Applied Introduction

Uploaded by

Copyright:

Available Formats

José Luis Ferreira

JOSÉ LUIS FERREIRA

All rights reserved. No reproduction, copy or transmission of this publication may be

No portion of this publication may be reproduced, copied or transmitted save with

First published 2020 by

Red Globe Press in the UK is an imprint of Springer Nature Limited, registered in E

List of Figures and Tables xi

1. The most famous games 1

2. Building the theory for simultaneous games 20

6. Negotiation games 104

7. Repeated games 120

7.5 Other strategies in the repeated game 128

8. Agency problems: Adverse selection 138

9. Agency problems: Signaling and moral hazard 159

10. Seven applications of game theory 177

11. Seven more applications 200

12. Dynamics 223

13. Limited rationality and behavioral economics 240

14. Power indices 260

4.7a Odysseus-sane vs. Odysseus-insane 69

10.2d Using the nuclear weapons first 179

12.4b Player 1 burns a unit of money 232

11.1 Parties’ preferences 216

My first game theory professor during my undergraduate years at the University of

A basic undergraduate course of 40 hours can cover Chapters 1 through 4 quite

A solutions manual is available to lecturers on the companion website at macmillanihe.

1.1 The coordination game

Figure 1.1 Coordination game

Figure 1.2 Choice of standards

1.3 The battle of the sexes

Figure 1.3 Battle of the sexes

-Darling, we’ ll meet at the stadium gates.

1.4 The chicken game

Figure 1.4 The chicken game

-Darling, you better swerve.

1.5 The prisoners’ dilemma

Figure 1.5a The prisoners’ dilemma

Figure 1.5b Prisoners’ dilemma with empathy

Figure 1.5c Prisoners’ dilemma with punishment

At this time, the reader may already guess Scarpia’s thoughts:

Now, if reasoning dictates an answer, then everyone should independently

Whatsoever therefore is consequent to a time of war, where every man is

2 Immanuel Kant (1785), Grounding for the Metaphysics of Morals.

Figure 1.6a Matching pennies

Figure 1.6b Nadal-Federer game

1.7 The ultimatum game

3 An example is Walker and Wooders (2001).

Figure 1.7 The ultimatum game

Exercises for Chapter 1

2.1 The normal form game

The definition of the normal form game

The payoff function for Sylvie is uS ( F , F ) = 2, uS ( F , M ) = 0, uS ( M , F ) = 0,

Formal definition of a maximin strategy

2. Next, take the maximum of those values:

Formal definition of a two-person, zero-sum game, and of its solution

The elimination of dominated strategies

Figure 2.2b Tax war after elimination of dominated strategies

Definition of a dominated strategy

List of Figures and Tables xi

1. The most famous games 1

2. Building the theory for simultaneous games 20

6. Negotiation games 104

7. Repeated games 120

7.5 Other strategies in the repeated game 128

8. Agency problems: Adverse selection 138

9. Agency problems: Signaling and moral hazard 159

10. Seven applications of game theory 177

11. Seven more applications 200

12. Dynamics 223

13. Limited rationality and behavioral economics 240

14. Power indices 260

4.7a Odysseus-sane vs. Odysseus-insane 69

10.2d Using the nuclear weapons first 179

12.4b Player 1 burns a unit of money 232

11.1 Parties’ preferences 216

2. Next, take the maximum of those values:

1. Detect and delete from the game all dominated strategies.