.

.
 ISBN: 978-1708284817

Giacomo Bonanno is Professor of Economics at the

University of California, Davis
http://faculty.econ.ucdavis.edu/faculty/bonanno/

Copyright © 2019 Giacomo Bonanno

All rights reserved. ISBN-13: 978-1708284817 ISBN-10: 1708284817
You are free to redistribute this book in pdf format. If you make use of any part of this
book you must give appropriate credit to the author. You may not remix, transform, or
build upon the material without permission from the author. You may not use the material
for commercial purposes.
 Preface
In the last three years I wrote three open access textbooks: one on
Game Theory (http://faculty.econ.ucdavis.edu/faculty/bonanno/GT_Book.html ), one on
Decision Making (http://faculty.econ.ucdavis.edu/faculty/bonanno/DM_Book.html ) and
the third on The Economics of Uncertainty and Insurance (http://faculty.econ.ucdavis.
edu/faculty/bonanno/EUI_Book.html ). This book is an extension of the last one: it in-
corporates it and augments it with the addition of several new chapters on risk sharing,
asymmetric information, adverse selection, signaling and moral hazard. It provides a
comprehensive introduction to the analysis of economic decisions under uncertainty and to
the role of asymmetric information in contractual relationships. I have been teaching an
upper-division undergraduate class on this topic at the University of California, Davis for
25 years and was not able to find a suitable textbook. Hopefully this book will fill this gap.
I tried to write the book in such a way that it would be accessible to anybody with min-
imum knowledge of calculus: the ability to calculate the (partial) derivative of a function
of one or two variables. The book is appropriate for an upper-division undergraduate class,
although some parts of it might be useful also to graduate students.
I have followed the same format as the other three books, by concluding each chapter
with a collection of exercises that are grouped according to that chapter’s sections. Com-
plete and detailed answers for each exercise are given in the last section of each chapter.
The book contains more than 150 fully solved exercises. It is also richly illustrated with
150 Figures.
I expect that there will be some typos and (hopefully, minor) mistakes. If you come
across any typos or mistakes, I would be grateful if you could inform me: I can be reached
at [email protected] . I will maintain an updated version of the book on my web
page at

http://www.econ.ucdavis.edu/faculty/bonanno/
I intend to add, some time in the future, a further collection of exercises with detailed
solutions. Details will appear on my web page.
I am very grateful to Elise Tidrick for teaching me how to use spacing and formatting
in a (perhaps unconventional) way that makes it easier for the reader to learn the material.
I would like to thank Mathias Legrand for making the latex template used for this book
available for free (the template was downloaded from http://www.latextemplates.com/
template/the-legrand-orange-book ).

v.2 6-21
 Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

I Insurance
2 Insurance: basic notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1 Uncertainty and lotteries 15
2.2 Money lotteries and attitudes to risk 16
2.3 Certainty equivalent and the risk premium 19
2.4 Insurance: basic concepts 21
2.5 Isoprofit lines 25
2.6 Profitable insurance requires risk aversion 30
2.6.1 Insuring a risk-neutral individual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6.2 Insuring a risk-averse individual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6.3 The profit-maximizing contract for a monopolist . . . . . . . . . . . . . . . . . . . . 32
2.6.4 Perfectly competitive industry with free entry . . . . . . . . . . . . . . . . . . . . . . 34
2.7 Exercises 36
2.7.1 Exercises for Section 2.2: Money lotteries and attitudes to risk . . . . . . . . . 36
2.7.2 Exercises for Section 2.3: Certainty equivalent and risk premium . . . . . . 38
2.7.3 Exercises for Section 2.4: Insurance: basic concepts . . . . . . . . . . . . . . . . 39
2.7.4 Exercises for Section 2.5: Isoprofit lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.7.5 Exercises for Section 2.6: Profitable insurance requires risk aversion . . . . 42
2.8 Solutions to Exercises 42
3 Expected Utility Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1 Expected utility: theorems 55
3.2 Expected utility: the axioms 63
3.3 Exercises 71
3.3.1 Exercises for Section 3.1: Expected utility: theorems . . . . . . . . . . . . . . . . . 71
3.3.2 Exercises for Section 3.2: Expected utility: the axioms . . . . . . . . . . . . . . . . 73
3.4 Solutions to Exercises 74

4 Money lotteries revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4.1 von Neumann Morgenstern preferences over money lotteries 79
4.1.1 The vNM utility-of-money function of a risk-neutral agent . . . . . . . . . . . . 79
4.1.2 Concavity and risk aversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1.3 Convexity and risk loving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.1.4 Mixtures of risk attitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.1.5 Attitude to risk and the second derivative of the utility function . . . . . . . 85
4.2 Measures of risk aversion 86
4.3 Some noteworthy utility functions 92
4.4 Higher risk 93
4.4.1 First-order stochastic dominance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.4.2 Mean preserving spread and second-order stochastic dominance . . . 95
4.5 Exercises 99
4.5.1 Exercises for Section 4.1: vNM preferences over money lotteries . . . . . . 99
4.5.2 Exercises for Section 4.2: Measures of risk aversion . . . . . . . . . . . . . . . . . 100
4.5.3 Exercises for Section 4.3: Some noteworthy utility functions . . . . . . . . . . 102
4.5.4 Exercises for Section 4.4: Higher risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.6 Solutions to Exercises 104

5 Insurance: Part 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

5.1 Binary lotteries and indifference curves 113
5.1.1 Case 1: risk neutrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.1.2 Case 2: risk aversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.1.3 Case 3: risk love . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.1.4 The slope of an indifference curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.2 Back to insurance 121
5.2.1 The profit-maximizing contract for a monopolist . . . . . . . . . . . . . . . . . . . 124
5.2.2 Perfectly competitive industry with free entry . . . . . . . . . . . . . . . . . . . . . 125
5.3 Choosing from a menu of contracts 127
5.3.1 Choosing from a finite menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.3.2 Choosing from a continuum of options . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.4 Mutual insurance 138
5.5 Exercises 140
5.5.1 Exercises for Section 5.1: Binary lotteries and indifference curves . . . . . 140
5.5.2 Exercises for Section 5.2: Back to insurance . . . . . . . . . . . . . . . . . . . . . . . 141
5.5.3 Exercises for Section 5.3: Choosing from a menu of contracts . . . . . . . 143
5.5.4 Exercises for Section 5.4: Mutual insurance . . . . . . . . . . . . . . . . . . . . . . . . 146
5.6 Solutions to Exercises 147

II Risk Sharing
6 Risk Sharing and Pareto Efficiency . . . . . . . . . . . . . . . . . . . . . . 165
6.1 Sharing an uncertain surplus 165
6.2 The Edgeworth box 167
6.3 Points of tangency 175
6.3.1 Risk averse Principal and risk neutral Agent . . . . . . . . . . . . . . . . . . . . . . . 175
6.3.2 Risk neutral Principal and risk averse Agent . . . . . . . . . . . . . . . . . . . . . . . 176
6.3.3 A general principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.3.4 Both parties risk averse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.3.5 Both parties risk neutral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.3.6 Pareto efficiency for contracts in the interior of the Edgeworth box . . . 182
6.4 Pareto efficient contracts on the sides of the Edgeworth box 183
6.4.1 Risk averse Principal and risk neutral Agent . . . . . . . . . . . . . . . . . . . . . . . 183
6.4.2 Risk neutral Principal and risk averse Agent . . . . . . . . . . . . . . . . . . . . . . . 184
6.4.3 Both parties risk averse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
6.5 The Edgeworth box when the parties have positive initial wealth 187
6.6 More than two outcomes 193
6.6.1 Risk-neutral Principal and risk-averse Agent . . . . . . . . . . . . . . . . . . . . . . . 194
6.6.2 Risk-averse Principal and risk-neutral Agent . . . . . . . . . . . . . . . . . . . . . . . 197
6.6.3 Both parties risk neutral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
6.6.4 Both parties risk averse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
6.7 Exercises 199
6.7.1 Exercises for Section 6.1: Sharing an uncertain surplus . . . . . . . . . . . . . . 199
6.7.2 Exercises for Section 6.2: The Edgeworth box . . . . . . . . . . . . . . . . . . . . . . 199
6.7.3 Exercises for Section 6.3: Points of tangency . . . . . . . . . . . . . . . . . . . . . . 201
6.7.4 Exercises for Section 6.4: Pareto efficient contracts on the sides of the
Edgeworth box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
6.7.5 Exercises for Section 6.5: The Edgeworth box when the parties have positive
initial wealth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.7.6 Exercises for Section 6.6: More than two outcomes . . . . . . . . . . . . . . . . 207
6.8 Solutions to Exercises 209

III Asymmetric Information: Adverse Selection

7 Adverse Selection or Hidden Type . . . . . . . . . . . . . . . . . . . . . 227
7.1 Adverse selection or hidden type 227
7.2 Conditional probability and belief updating 229
7.2.1 Conditional probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
7.2.2 Belief updating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
7.3 The market for used cars 234
7.3.1 Possible remedies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
7.3.2 Further remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
7.4 Exercises 243
7.4.1 Exercises for Section 7.2.2: Conditional probability and belief updating 243
7.4.2 Exercises for Section 7.3: The market for used cars . . . . . . . . . . . . . . . . . 245
7.5 Solutions to Exercises 247

8 Adverse Selection in Insurance . . . . . . . . . . . . . . . . . . . . . . . . . 253

8.1 Adverse selection in insurance markets 253
8.2 Two types of customers 254
8.2.1 The contracts offered by a monopolist who can tell individuals apart 256
8.3 The monopolist under asymmetric information 257
8.3.1 The monopolist’s profit under Option 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
8.3.2 The monopolist’s profit under Option 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
8.3.3 The monopolist’s profit under Option 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
8.3.4 Option 2 revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
8.4 A perfectly competitive insurance industry 276
8.5 Exercises 283
8.5.1 Exercises for Section 8.2: Two types of customers . . . . . . . . . . . . . . . . . . . 283
8.5.2 Exercises for Section 8.3: The monopolist under asymmetric information 284
8.5.3 Exercises for Section 8.4: A perfectly competitive insurance industry . . 286
8.6 Solutions to Exercises 287

IV Asymmetric Information: Signaling

9 Signaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
9.1 Earnings and education 297
9.2 Signaling in the job market 299
9.2.1 Signaling equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
9.2.2 Pareto inefficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
9.2.3 Alternative interpretation of a signaling equilibrium . . . . . . . . . . . . . . . . 302
9.3 Indices versus signals 305
9.4 More than two types 309
9.5 A more general analysis 311
9.6 Signaling in other markets 322
9.6.1 Market for used cars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
9.6.2 Advertising as a signal of quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
9.6.3 Other markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.7 Exercises 325
9.7.1 Exercises for Section 9.2: Signaling in the job market . . . . . . . . . . . . . . . . 325
9.7.2 Exercises for Section 9.3: Indices versus signals . . . . . . . . . . . . . . . . . . . . . 328
9.7.3 Exercises for Section 9.4: More than two types . . . . . . . . . . . . . . . . . . . . . 329
9.7.4 Exercises for Section 9.5: A more general analysis . . . . . . . . . . . . . . . . . . 330
9.7.5 Exercises for Section 9.6: Signaling in other markets . . . . . . . . . . . . . . . . 330
9.8 Solutions to Exercises 331

V Moral Hazard
10 Moral Hazard in Insurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
10.1 Moral hazard or hidden action 343
10.2 Moral hazard and insurance 344
10.2.1 Two levels of unobserved effort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
10.2.2 The reservation utility locus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
10.2.3 The profit-maximizing contract for a monopolist . . . . . . . . . . . . . . . . . . . 354
10.3 Exercises 359
10.3.1 Exercises for Section 10.2.1: Two levels of unobserved effort . . . . . . . . . 359
10.3.2 Exercises for Section 10.2.2: The reservation utility locus . . . . . . . . . . . . . 361
10.3.3 Exercises for Section 10.2.3: The profit-maximizing contract . . . . . . . . . . 362
10.4 Solutions to Exercises 363

11 Moral Hazard in Principal-Agent . . . . . . . . . . . . . . . . . . . . . . . 369

11.1 Moral hazard in Principal-Agent relationships 369
11.2 Risk sharing under moral hazard 370
11.3 The case with two outcomes and two levels of effort 374
11.4 The case with more than two outcomes 388
11.5 Exercises 393
11.5.1 Exercises for Section 11.2: Risk sharing under moral hazard . . . . . . . . . . 393
11.5.2 Exercises for Section 11.3: Two outcomes and two levels of effort . . . . . 394
11.5.3 Exercises for Section 11.4: The case with more than two outcomes . . . 398
11.6 Solutions to Exercises 400

12 Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
 1. Introduction

This book offers an introduction to the economic analysis of uncertainty and information.
Life is made up of a never-ending sequence of decisions. Many decisions – such as what to watch on
television or what to eat for breakfast – do not have major consequences. Other decisions – such as whether
or not to invest all of one’s savings in the purchase of a house, or whether to purchase earthquake insurance –
can have a significant impact on one’s life. We will concern ourselves with decisions that potentially have a
considerable impact on the wealth of the individual in question.
Most of the time the outcome of a decision is influenced by external factors that are outside the decision
maker’s control, such as the side effects of a new drug, or the future price of real estate, or the occurrence
of a natural phenomenon (such as a flood, or a fire, or an earthquake). While one is typically aware of the
existence of such external factors, as the saying goes “It is difficult to make predictions, especially about
the future”.1 Most decisions are shrouded in uncertainty and this book is about how uncertainty affects the
actions and decisions of economic agents.
We begin by examining, in Chapter 2, what explains the existence and profitability of insurance markets.
For this we simply appeal to the definition of risk aversion, without the need for the full power of expected
utility theory.
Chapter 3 develops the Theory of Expected Utility, which is central to the rest of the book.
In Chapter 4 we use the theory of expected utility to re-examine the notion of attitude to risk (risk
aversion, risk neutrality and risk love), discuss how to measure the degree of risk aversion of an individual
and develop a test for determining when, of two alternative risky prospects, one can unambiguously be
labeled as being more risky than the other.
With the help of expected utility theory, in Chapter 5 we study the demand side of insurance markets.
We then put together the analysis of the supply side of insurance, developed in Chapter 2, with the analysis
of the demand side, to determine the equilibrium of an insurance industry under two opposite scenarios: the
case where the industry is a monopoly and the case where there is perfect competition with free entry.
In Chapter 6 we address the issue of efficient risk sharing. We consider the case of an individual,
referred to as “the Principal”, who is contemplating hiring another individual, referred to as “the Agent”, to
perform a task, whose outcome is uncertain (because it is affected by external factors). We consider all the
possible forms of payment to the Agent (e.g. a fixed wage or a payment contingent on the outcome) and ask
1 This saying is often attributed to the physicist Niels Bohr, but apparently it is an old Danish proverb.
12 Chapter 1. Introduction

what contracts are Pareto efficient, in the sense that there is no other contract that they both prefer. We study
how Pareto efficiency relates to the optimal way of allocating risk between the two parties to the contract.
In Chapters 7-9 we turn to the issue of asymmetric information. It is often the case that one of the
two parties to a contract has more information than the other party about aspects of the transaction that are
relevant to both parties. For example, the seller of a used car has knowledge about the quality of the car
that the potential buyer cannot easily acquire before the purchase, or a job applicant knows more about
herself than the potential employer can find out from an interview. Asymmetric information can manifest
itself in different forms. One type of asymmetric information gives rise to the phenomenon of “adverse
selection”, which is studied in Chapters 7 and 8. Chapter 7 deals with the general phenomenon of adverse
selection, with particular focus on the market for used durable goods, while Chapter 8 is devoted to the
analysis of adverse selection in insurance markets. This is the situation where there are different types of
individuals, with different propensities to incur losses, and – while each individual knows his or her own
type – the insurance company does not. We study how the asymmetry of information affects the decisions
of the suppliers of insurance and re-examine the conditions for an equilibrium in the two types of industry
structure examined in Chapter 5, namely monopoly and perfect competition.
In Chapter 9 we study another phenomenon that arises in the context of asymmetric information, namely
the phenomenon of “signaling”. Signaling refers to the attempt by the informed party to credibly convey
information to the uninformed party. When the latter is uncertain about the characteristics, or “type”, of
individual he/she is about to sign a contract with, he/she might offer contractual terms that are unappealing to
some individuals, thereby creating an incentive for the “better” types to engage in costly activities that allow
them to “separate themselves” from the worse types and to credibly convey information about themselves.
While the asymmetric information studied in Chapters 7-9 is also referred to as “hidden type”, the
informational asymmetry studied in Chapters 10 and 11 is called “hidden action” or “moral hazard”. It refers
to situations where what cannot be observed by one of the two parties to a contract is not the type of the other
party, but his/her behavior. When such behavior has an effect on the outcome, it becomes important for the
uninformed party to design the contract in such a way that it creates an incentive for the other party to act in a
“desirable” way. For example, in the case of insurance, the probability that the insured individual will face a
loss – and thus apply for a reimbursement from the insurance company – may be affected by the behavior of
the individual, in particular by the effort and care exerted in loss prevention. In such a situation the insurance
company might want to offer only insurance contracts that will create an incentive for the insured to exert
appropriate effort towards reducing the probability of loss. Chapter 10 deals with the phenomenon of moral
hazard in insurance, while Chapter 11 revisits the Principal-Agent relationships studied in Chapter 6 and
analyses the effect of moral hazard in that context.

Whenever possible, throughout the book we have tried to illustrate the relevant concepts graphically in
two-dimensional diagrams. The book is richly illustrated with approximately 150 figures and tables.

At the end of each section of each chapter the reader is invited to test his/her understanding of the
concepts introduced in that section by attempting several exercises. In order not to break the flow of the
exposition, the exercises are collected in a section at the end of the chapter. Complete and detailed answers
for each exercise are given in the last section of each chapter. In total, the book contains more than 150 fully
solved exercises. Attempting to solve the exercises is an integral part of learning the material covered in this
book.

The book was written in a way that should be accessible to anyone with minimum knowledge of calculus,
in particular the ability to calculate the (partial) derivative of a function of one or two variables.

This book does not necessarily follow conventional formatting standards. Rather, the intention was to
break each argument into clearly outlined steps, highlighted by appropriate spacing.
I
Insurance

2 Insurance: basic notions . . . . . . . . 15

2.1 Uncertainty and lotteries
2.2 Money lotteries and attitudes to risk
2.3 Certainty equivalent and the risk premium
2.4 Insurance: basic concepts
2.5 Isoprofit lines
2.6 Profitable insurance requires risk aversion
2.7 Exercises
2.8 Solutions to Exercises

3 Expected Utility Theory . . . . . . . . . . 55

3.1 Expected utility: theorems
3.2 Expected utility: the axioms
3.3 Exercises
3.4 Solutions to Exercises

4 Money lotteries revisited . . . . . . . . 79

4.1 von Neumann Morgenstern preferences over
money lotteries
4.2 Measures of risk aversion
4.3 Some noteworthy utility functions
4.4 Higher risk
4.5 Exercises
4.6 Solutions to Exercises

5 Insurance: Part 2 . . . . . . . . . . . . . . . . 113

5.1 Binary lotteries and indifference curves
5.2 Back to insurance
5.3 Choosing from a menu of contracts
5.4 Mutual insurance
5.5 Exercises
5.6 Solutions to Exercises
 2. Insurance: basic notions

2.1 Uncertainty and lotteries

Most of the important decisions that we make in life are made difficult by the presence of uncertainty:
the final outcome is influenced by external factors that we cannot control and we cannot predict with
certainty. Because of such external factors, any given decision will typically be associated with different
outcomes, depending on what “state of the world” will actually occur. If the decision-maker is able to assign
probabilities to these external factors – and thus to the associated outcomes – then one can represent the
uncertainty that the decision maker faces as a list of possible outcomes, each with a corresponding probability.
We call such lists lotteries.
For example, suppose that Ann and Bob are planning their wedding reception. They have a large number
of guests and face the choice between two venues: a spacious outdoor area where the guests will be able
to roam around or a small indoor area where the guests will feel rather crammed. Ann and Bob want their
reception to be a success and their guests to feel comfortable. It seems that the large outdoor area is a better
choice; however, there is also an external factor that needs to be taken into account, namely the weather. If it
does not rain, then the outdoor area will yield the best outcome (success: denote this outcome by o1 ) but if
it does rain then the outdoor area will give rise to the worst outcome (failure: denote this outcome by o3 ).
On the other hand, if Ann and Bob choose the indoor venue, then the corresponding outcome will be a less
successful reception but not a failure (call this outcome o2 ). Let us denote the possible outcomes as follows:
o1 : successful reception
o2 : mediocre reception
o3 : failed reception.
Clearly they prefer o1 to o2 and o2 to o3 . At the time of deciding which venue to pay for, Ann and Bob do not
know what the weather will be like on their wedding day. The most they can do is consult a weather forecast
service and obtain probabilistic estimates. Suppose that the forecast service predicts a 30% chance of rain on
the day in question. Then we can represent the decision to book the outdoor venue as the following lottery
 
outcome: o1 o3
probability: 0.7 0.3
On the other hand, the decision to book the indoor venue corresponds to the following degenerate lottery:
 
outcome: o2
probability: 1
 16 Chapter 2. Insurance: basic notions

Throughout this book we will represent the uncertainty facing a decision-maker in terms of lotteries.1
This assumes that the decision-maker is always able to assign probabilities to the possible outcomes. We
interpret these probabilities either as “objective” probabilities, obtained from relevant past data, or as
“subjective” estimates by the individual. For example, an individual who is considering whether or not to
insure her bicycle against theft, knows that there are two relevant basic outcomes: either the bicycle will be
stolen or it will not be stolen. Furthermore, she can look up data on past bicycle thefts in her area and use the
proportion of bicycles that were stolen as an objective estimate of the probability that her bicycle will be
stolen; alternatively, she can use a more subjective estimate: for example she might use a lower probability
of theft than suggested by the data, because she knows herself to be very conscientious and – unlike other
people – to always lock her bicycle when left unattended.
In this chapter we will focus on lotteries where the outcomes are sums of money. More general lotteries
will be considered in Chapter 3.

2.2 Money lotteries and attitudes to risk

Definition 2.2.1 A money lottery is a probability distribution over alist of outcomes, where
 each
$x1 $x2 ... $xn
outcome consists of a sum of money. Thus, it is an object of the form with
p1 p2 ... pn
0 ≤ pi ≤ 1 for all i = 1, 2, ..., n, and p1 + p2 + ... + pn = 1.

We assume that the individual in question is able to rank any two money lotteries. For
example, 
if asked
$400
to choose between getting $400 for sure, which can be viewed as the degenerate lottery , and
1
 
$900 $0
the lottery2 1 1 , she will be able to tell us if she prefers one lottery to the other or is indifferent
2 2
between the two. In general, there is no “right answer” to this question, as there is no right answer to the
question “do you prefer coffee or tea?”: it is a matter of individual taste.
 
$x1 $x2 ... $xn
Definition 2.2.2 Given a money lottery L = , its expected value is the
p1 p2 ... pn
number E[L] = x1 p1 + x2 p2 + ... + xn pn .

For example, the expected value of the money lottery

 
$600 $180 $120 $30
1 1 5 1
12 3 12 6
1 1 5 1
is 12 600 + 3 180 + 12 120 + 6 30 = 165.

Definition 2.2.3 Let L be a non-degenerate money lottery (that is, a money lottery where at least
two different outcomes are assigned positive probability)a and consider the choice between L and the
degenerate lottery  
$E[L]
1
(that is, the choice between facing the lottery L or getting the expected value of L with certainty).
Then

1 For
some analysis of decision-making in situations where the individual is not able to assign probabil-
ities to the outcomes see my book Decision Making (http://faculty.econ.ucdavis.edu/faculty/
bonanno/DM_Book.html ).
2 We can think of this lottery as tossing a fair coin and then giving the individual $900 if it comes up

Heads and nothing if it comes up Tails.

2.2 Money lotteries and attitudes to risk 17

• An individual who prefers $E[L] for certain to L is said to be risk averse (relative to L).
• An individual who is indifferent between $E[L] for certain and L is said to be risk neutral (relative
to L).
• An individual who prefers L to $E[L] for certain is said to be risk loving or risk seeking (relative
to L).
 
a A money lottery $x1 $x2 ... $xn
is non-degenerate if, for all i = 1, 2, ..., n, pi < 1.
p1 p2 ... pn

Note that, if an individual

(1) is risk neutral relative to every money lottery,
(2) has transitive preferences3 over money lotteries and
(3) prefers more money to less,
then we can tell how that individual ranks any two money lotteries.

3 That is, if she considers lottery A to be at least as good as lottery B and she considers lottery B to be at
least as good as lottery C then she considers A to be at least as good as C.
18 Chapter 2. Insurance: basic notions

For example, how! would a risk-neutral

! individual rank the two lotteries
$30 $45 $90 $5 $100
L1 = 1 5 1
and L2 = 3 2
? We shall use the symbol  to denote strict
3 9 9 5 5
preference and the symbol ∼ to denote indifference.4 Since E[L1 ] = 45 and the individual is risk neutral,
L1 ∼ $45; since E[L2 ] = 43 and the individual is risk neutral, $43 ∼ L2 ; since the individual prefers more
money to less, $45  $43:

L1 ∼ $45  $43 ∼ L2 .
Thus, by transitivity, L1  L2 (see Exercises 2.10-2.13).

On the other hand, knowing that an individual is risk averse relative to every money lottery, has transitive
preferences over money lotteries and prefers more money to less, is not sufficient to predict how she will
choose between two arbitrary money lotteries. For!example, as we will see in Chapter 3, it is possible that one
!
$28 $10 $50
risk-averse individual will prefer L3 = (whose expected value is 28) to L4 = 1 1
1 2 2
(whose expected value is 30), while another risk-averse individual will prefer L4 to L3 .

Similarly, knowing that an individual is risk loving relative to every money lottery, has transitive
preferences over money lotteries and prefers more money to less, is not sufficient to predict how she will
choose between two arbitrary money lotteries.

R Note that “rationality” does not, and should not, dictate whether an individual should
be risk neutral, risk averse or risk loving: an individual’s attitude to risk is merely
a reflection of that individual’s preferences. It is a generally accepted principle
that de gustibus non est disputandum (in matters of taste, there can be no disputes).
According to this principle, there is no such thing as an irrational preference and thus
there is no such thing as an irrational attitude to risk.
From an empirical point of view, however, most people reveal through their choices
(e.g. the decision to buy insurance) that they are risk averse, at least when the stakes
are sufficiently high. It is also possible (as we will see in Chapter 4) for an individual
to have different attitudes to risk, depending on how high the stakes are (e.g. an
individual might display risk aversion, by purchasing home insurance, as well as risk
love, by purchasing a lottery ticket).

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 2.7.1 at the end of this chapter.

4 Thus
A  B means that the individual prefers A to B and A ∼ B means that the individual is indifferent
between A and B.
 2.3 Certainty equivalent and the risk premium 19

2.3 Certainty equivalent and the risk premium

Given a set of money lotteries L , we will assume that the individual under consideration has well-defined
preferences over the elements of L . As before, we shall use the symbol  to denote strict preference
(L1  L2 means that the individual prefers lottery L1 to lottery L2 ) and the symbol ∼ to denote indifference
(L1 ∼ L2 means that the individual is indifferent between L1 and L2 , that is, she considers L1 to be just as
good as L2 ). Finally, we use the symbol % to signify “at least as good as”: L1 % L2 means that the individual
considers L1 to be at least as good as L2 , that is, either she prefers L1 to L2 or she is indifferent between L1
and L2 . The following table summarizes the notation:

notation: interpretation:
L1  L2 the individual prefers L1 to L2
L1 ∼ L2 the individual is indifferent between L1 and L2
L1 % L2 the individual considers L1 to be at least as good as L2 ,
that is, either L1  L2 or L1 ∼ L2 .

We shall assume that the individual is able to rank any two lotteries (her preferences are complete) and her
ranking is transitive:

• (completeness) for every L1 and L2 , either L1 % L2 or L2 % L1 or both,

• (transitivity) if L1 % L2 and L2 % L3 then L1 % L3 .5

We shall also assume throughout that the individual prefers more money to less, that is,
! !
$x $y
 if and only if x > y. (2.1)
1 1

Suppose that, for every money lottery L there is a sum of money, denoted by CL , that the individual
considers to be just as good as the lottery L; then we call CL the certainty equivalent of lottery L for that
individual.

Definition 2.3.1 The certainty equivalent of a money lottery L is that sum of money CL such that
   
$x1 ... $xn $CL
L= ∼
p1 ... pn 1

Typically, the certainty equivalent of a given money lottery will be different for different individuals. However,
all risk-neutral individuals will share the same certainty equivalent; in fact, it follows from the definition of
risk neutrality (Definition 2.2.3) that

• for a risk-neutral individual, the certainty equivalent of a money lottery L coincides with the expected
value of L:
CL = E[L].
On the other hand, for a risk-averse individual (who, furthermore, prefers more money to less and whose
preferences are complete and transitive) the certainty equivalent of a money lottery will be less than the
expected value:

• for a risk-averse individual, for every money lottery L,

CL < E[L].
5 InExercises 2.10-2.13 the reader is asked to prove that transitivity of the “at least as good” relation
implies transitivity of strict preference and of indifference.
20 Chapter 2. Insurance: basic notions
 
$E[L]
In fact, by definition of risk aversion,  L and, by definition of certainty equivalent, L ∼
1
     
$CL $E[L] $CL
. Thus, by transitivity,  ; hence, by (2.1),
1 1 1
E[L] > CL . Similarly,

• for a risk-loving individual, for every money lottery L,

CL > E[L].

From the notion of certainty equivalent we derive another notion which can be used to compare the
degree of risk aversion across individuals.

Definition 2.3.2 The risk premium of a money lottery L, denoted by RL , is the amount by which the
expected value of L can be reduced to induce indifference between the lottery itself and the reduced
amount for certain:    
$x1 ... $xn $(E[L] − RL )
L= ∼
p1 ... pn 1

It follows from Definitions 2.3.1 and 2.3.2 that CL = E[L] − RL or, equivalently,

RL = E[L] −CL .

Thus, for a risk-neutral individual the risk premium is zero, while for a risk-averse individual the risk
premium is positive (and for a risk-loving individual the risk premium is negative). Furthermore, we can
label a risk-averse individual as more risk-averse (relative to lottery L) than another risk-averse individual
if the risk premium for the former is larger than the risk premium for the latter. In fact, the risk premium
can be interpreted as the price (relative to the expected value) that the individual is willing to pay to avoid
facing lottery L. For example, consider three individuals: Ann, Bob and Carla. They all have the same initial
wealth $6, 000 and they are facing the lottery L where with probability !50% their wealth is wiped out and
$0 $12, 000
with probability 50% their wealth is doubled: L = 1 1
. Suppose that Ann’s risk premium
2 2
for this lottery is RAnn Bob
L = 900, Bob’s is RL = 500 and Carla’s is RL
Carla = 0. Then Ann and Bob are risk

averse and Ann is more risk averse than Bob, while Carla is risk neutral. Ann would be willing to pay up to
$900 (thus reducing her wealth from $6, 000 to $5, 100) in order to avoid the lottery L, while Bob would
only be willing to pay up to $500 (thus reducing his wealth from $6, 000 to $5, 500) in order to avoid the
lottery L; on the other hand, Carla would not be willing to pay any amount of money to avoid L, since she is
indifferent between keeping her initial wealth of $6, 000 and playing lottery L.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 2.7.2 at the end of this chapter.
 2.4 Insurance: basic concepts 21

2.4 Insurance: basic concepts

Insurance markets are a good example of situations where uncertainty can be represented by means of money
lotteries.

Consider an individual who has an initial wealth of $W0 > 0 and faces the possibility of a loss in the
amount of $` (0 < ` ≤ W0 ) with probability p (0 < p < 1). For example, it could be an individual who owns
a plot of land worth $80, 000 and a house built on it worth $220, 000 (so that W0 = 80, 000 + 220, 000 =
300, 000). She is worried about the possibility of a fire destroying the house (thus ` = 220, 000) and,
according to publicly available data, the probability of this happening in her area is 2% (thus p = 0.02). An
insurance company offers her a contract and she has to decide whether or not to purchase that contract. An
insurance contract is typically expressed in terms of two numbers: the premium, which we will denote by h,
and the deductible, which we will denote by d. The premium can be thought of as the price of the contract:
it is paid no matter whether the loss is incurred or not. The deductible is the portion of the loss that is not
covered. If d = 0 we say that the contract offers full insurance, while if d > 0 then we say that the contract
offers partial insurance:

d=0 full insurance

d>0 partial insurance.

In the above example, if the deductible is $40, 000 then, if the loss occurs, the insurance company makes a
payment to the insured in the amount of $(` − d) = $(220, 000 − 40, 000) = $180, 000 (and, of course, if the
loss in not incurred then the insurance company does not make any payments to the insured).
22 Chapter 2. Insurance: basic notions

The following table summarizes the notation used in this book in the context of insurance:

W0 initial wealth
` potential loss
p probability of loss
h premium
d deductible
`−d insured amount
(h, d) insurance contract.

It will be useful to represent the initial situation and possible insurance contracts graphically. We shall
do so by using wealth diagrams where, on the horizontal axis, we represent the individual’s wealth if a loss
occurs, denoted by W1 , and, on the vertical axis, the individual’s wealth if there is no loss, denoted by W2 ;
we shall also refer to the former as wealth in the bad state and to the latter as wealth in the good state. The
no-insurance situation can be represented in the wealth diagram as the point NI = (W0 − `,W0 ), as shown in
Figure 2.1.

wealth in good state set of

W2 insurance
contracts

no insurance 45o line

NI
W0

0 W1
W0 − ` wealth in bad state

Figure 2.1: The no-insurance point (NI) and the set of possible insurance contracts (the
shaded triangle).

The purpose of an insurance contract is to protect the individual in case she experiences
a loss: thus an insurance contract can be thought of as a point in the diagram where the
horizontal coordinate is larger than W0 − ` (which is the individual’s wealth in the bad
state if she does not insure), while the vertical coordinate is smaller than W0 because of the
premium. The set of possible insurance contracts (encoded in terms of the corresponding
wealth levels for the individual, in the bad state and in the good state), is shown in Figures
2.1 and 2.2 as a shaded triangle. The “45o line”– which is the line out of the origin with an
2.4 Insurance: basic concepts 23

angle of 45o – is the set of points (W1 ,W2 ) such that W1 = W2 . As we will see below, the
points on the 45o line represent full-insurance contracts.6
How do we translate an insurance contract (h, d), expressed in terms of premium h and
deductible d, into a point in the (W1 ,W2 ) diagram? If the individual purchases contract
(h, d) then she pays the premium h in any case (that is, whether or not she incurs a loss) and
thus her wealth in the good state is equal to W2 = W0 − h; the premium reduces her wealth
also in the bad state, but in this state there is a further reduction due to the deductible, so
that W1 = W0 − h − d = W2 − d. Conversely, given a contract expressed as a point (W1 ,W2 )
we can recover the premium and deductible as follows: h = W0 −W2 and d = W2 −W1 . It
is clear from this that d = 0 if and only if W1 = W2 , that is, if and only if the point lies on
the 45o line.

wealth in good state

W2

NI 45o line
5,000
A
4,500
B
4,100

0 W1
3,000 3,900 4,100 wealth in bad state

Figure 2.2: The no-insurance point and two insurance contracts.

In Figure 2.2 three points are shown: the no-insurance point NI = (3, 000, 5, 000) and
two possible insurance contracts: A = (W1A = 3, 900, W2A = 4, 500) and
B B
B = (W1 = 4, 100, W2 = 4, 100). From NI we deduce that
W0 = 5, 000 and ` = 5, 000 − 3, 000 = 2, 000.
Let hA denote the premium of contract A and dA the deductible; then
hA = W0 −W2A = 5, 000 − 4, 500 = 500 and dA = W2A −W1A = 4, 500 − 3, 900 = 600.
6 Expressed in terms of premium and deductible, the set of insurance contracts is {(h, d) : h > 0, d ≥
0, h + d < `} ∪ {(0, `)}. We have added the trivial contract with h = 0 and d = ` for convenience: it is
equivalent to no insurance. Expressed in terms of wealth levels, the set of insurance contracts is {(W1 ,W2 ) :
W0 − ` < W1 ≤ W2 < W0 } ∪ {(W0 − `,W0 )}.
24 Chapter 2. Insurance: basic notions

Similarly, let hB denote the premium of contract B and dB the deductible; then

hB = W0 −W2B = 5, 000 − 4, 100 = 900 and dB = W2B −W1B = 4, 100 − 4, 100 = 0.

Hence A is a partial-insurance contract, while B is a full-insurance contract.

Figure 2.3 shows how to view the premium and deductible corresponding to a contract
A = (W1A ,W2A ).

W2

NI 45o line
W0

premium hA
A
W0 − hA

deductible dA

W0 − hA − dA

0 W0 − ` W0 − hA − dA
W1

Figure 2.3: The graphical representation of the premium hA and the deductible dA corre-
sponding to a contract A = (W1A ,W2A ).

As shown in Figure 2.1, there are many potential insurance contracts (the points in the
shaded triangle). Will an insurance company be willing to offer any of them? Would an
individual be willing to accept any of them? The first question has to do with the incentives
of the supplier of contracts (the insurer), while the second question has to do with the
incentives of the potential customer (the insured).
We will address the first question in the next sections and postpone a full analysis of
the second question to Chapter 4.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 2.7.3 at the end of this chapter.
 2.5 Isoprofit lines 25

2.5 Isoprofit lines

Throughout this book we shall assume that insurance companies are risk neutral and that
their objective is to maximize expected profits.7 Selling a contract (h, d) to a customer
corresponds to the following money lottery (in terms of profits) for the insurer:
 
$h $[h − (` − d)]
. (N)
1− p p
Given an insurance contract (h, d), we denote by π(h, d) the expected value of the corre-
sponding profit lottery (N), that is, the expected profit from the contract:
π(h, d) = (1 − p)h + p[h − (` − d)] = h − p` + pd. (NN)
By the assumption of risk neutrality, the insurance company will be indifferent between
any two contracts that yield the same expected profit. For example, if ` = 4, 000 and
5
p = 100 , the two contracts A = (hA = 800, dA = 1, 000) and B = (hB = 825, dB = 500)
yield the same expected profit:
5 5
π(A) = 800 − 100 3, 000 = 650 and π(B) = 825 − 100 3, 500 = 650.

Definition 2.5.1 A line in the (W1 ,W2 ) plane joining all the contracts that give rise to
the same expected profit is called an isoprofit line.

p
We want to show that an isoprofit line is a downward-sloping straight line with slope − 1−p .
Let A = W1A ,W2A and B = W1B ,W2B be two contracts that yield the same expected profit,

that is,

W0 −W2A −p` + p (W2A −W1A ) = W0 −W2B −p` + p (W2B −W1B ) .

| {z } | {z } | {z } | {z }
=hA =dA =hB =dB

Deleting W0 − p` from both sides of the equation and rearranging the terms we get

−(1 − p)W2A − pW1A = −(1 − p)W2B − pW1B

or, equivalently,
rise W A −W2B p
= 2A = −
run W1 −W1B 1− p
which gives the slope of the line segment joining points A and B. Note that the slope is a
constant, that is, it does not vary with the points A and B that are chosen.
7 The assumption of risk neutrality is not needed if the insurance company sells the same contract to
a large number of individuals. Let n be a large number of customers insured by the insurance company
with contract (h, d). Let n0 be the number of customers who do not suffer a loss and n1 be the number of
customers who suffer a loss (thus n0 + n1 = n). Then the insurer’s total profits will be (n0 + n1 )h − n1 (` − d),
so that profit per customer, or profit per contract, is nh−n1n(`−d) = h − nn1 (` − d). By the Law of Large
Numbers in probability theory, nn1 will be approximately equal to p (the probability of loss), so that the profit
per customer will be approximately equal to π(h, d) = h − p` + pd as defined above.
26 Chapter 2. Insurance: basic notions

Figure 2.4 shows an isoprofit line and two contracts, A and B, on this line.

(probability 1 − p)

W2

isoprofit line 45o line

A
W2A
rise B
W2B
run
slope of
isoprofit line:
W2A −W2B p
W1A −W1B
= − 1−p
W1
0
W1A W1B (probability p)

Figure 2.4: The slope of an isoprofit line.

Thus

• isoprofit lines are straight lines,

• isoprofit lines are downward-sloping or decreasing, since the slope is negative:

p
− 1−p < 0 because 0 < p < 1.

Consider two insurance contracts A = W1A ,W2A and B = W1B ,W2B . From the point

of view of the potential buyer, these two contracts correspond to the wealth lotteries
   
W0 − hA − dA W0 − hA W0 − hB − dB W0 − hB
A= and B =
p 1− p p 1− p

where hA = W0 −W2A is the premium of contract A and dA = W2A −W1A is the deductible of
contract A and, similarly, hB = W0 −W2B and dB = W2B −W1B . Let

π(A) = hA − p` + pdA and π(B) = hB − p` + pdB

be the expected profits from contracts A and B, respectively. We want to show that

π(A) = π(B) if and only if E[A] = E[B], (2.2)

2.5 Isoprofit lines 27

that is, A and B lie on the same isoprofit line if and only if the two wealth lotteries A and B
have the expected value. In fact,

E[A] = p (W0 − hA − dA ) + (1 − p)(W0 − hA ) = W0 − hA − pdA = W0 − (hA + pdA )

E[B] = p (W0 − hB − dB ) + (1 − p)(W0 − hB ) = W0 − hB − pdB = W0 − (hB + pdB )

Thus E[A] = E[B] if and only if hA + pdA = hB + pdB if and only if (subtracting p` from
both sides)
hA + pdA − p` = hB + pdB − p`
| {z } | {z }
=π(A) =π(B)

For each point in the (W1 ,W2 ) plane there is an isoprofit line that goes through that
p
point. Hence the plane is filled with parallel isoprofit lines (each with slope − 1−p ).
Let A = W1A ,W2A and B = W1B ,W2B be two insurance contracts and suppose that B lies

below the isoprofit line that goes through A, as shown in Figure 2.5, so that π(B) 6= π(A).
Does B represent a contract that yields higher or lower profits than A? In other words, if
the isoprofit line through one contract is below the isoprofit line through another contract,
which of the two lines corresponds to a higher level of profit?

W2
isoprofit line

A 45o line

C
B D
isoprofit
line

0 W1
W0 − hD
W0 − hC

Figure 2.5: A lower isoprofit line corresponds to a higher level of profit.

As shown in Figure 2.5, let C be the full-insurance contract that lies on the isoprofit
line through A and D the full-insurance contract that lies on the isoprofit line through
B. Then π(A) = π(C) and π(B) = π(D). Thus, if we show that π(D) > π(C) then it
follows that π(B) > π(A). The proof that π(D) > π(C) is straightforward: contract D
28 Chapter 2. Insurance: basic notions

guarantees a wealth of W0 − hD to the consumer (where hD is the premium of contract D)

and contract C guarantees a wealth of W0 − hC to the consumer (where hC is the premium
of contract C); since W0 − hD < W0 − hC , it follows that hD > hC which implies that
π(D) = hD − p` > π(C) = hC − p`.8
Thus we have shown that moving from an isoprofit line to a lower one corresponds to
moving from lower profits to higher profits. This is illustrated in Figure 2.6.

W2

direction of
increasing profits

π3 > π2 > π1

π1
π2 isoprofit
π3 lines
W1
0

Figure 2.6: The direction of increasing profits.

Among the isoprofit lines there is one which is of particular interest, namely the zero-
profit line, that is, the line that joins all the contracts that yield zero profits. Like all the
p
other isoprofit lines, this is a straight line with slope − 1−p . Furthermore, it goes through
the no-insurance point NI; in fact, NI can be thought of as a trivial contract with zero
premium and full deductible: such a contract obviously involves zero profits because the
insurance company receives no payment (hNI = 0) and makes no payment (dNI = ` so that
` − dNI = 0).

R The zero-profit line is also called the fair odds line.

8 In
Exercise 2.21 the reader is asked to give an alternative proof of the fact that, if contract B lies below
the isoprofit line through contract A, then π(B) > π(A).
2.5 Isoprofit lines 29

The zero-profit line is shown in Figure 2.7. Points above the line represent contracts that
yield negative profits and points below the line represent contracts that yield positive profits.

W2
zero-profit line
NI C 45o line
W0 B

A
π(B) = 0
π(A) > 0
π(C) < 0

0 W1
W0 − `

Figure 2.7: The zero-profit line.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 2.7.4 at the end of this chapter.
 30 Chapter 2. Insurance: basic notions

2.6 Profitable insurance requires risk aversion

2.6.1 Insuring a risk-neutral individual
Recall that the no-insurance option corresponds to the wealth lottery
 
W0 W0 − `
NI = whose expected value is E[NI] = W0 − p` (2.3)
1− p p
where, as usual, W0 is the initial wealth, ` the potential loss and p the probability of loss.
Now suppose that the individual is risk neutral and is offered an insurance contract (h, d)
that yields positive profits, that is,

h − p` + pd > 0 or, equivalently h + pd > p`. (2.4)

For the potential customer, such a contract corresponds to the wealth lottery
 
W0 − h W0 − h − d
A=
1− p p
whose expected value is

E[A] = W0 − h − pd = W0 − (h + pd) (2.5)

Using the fact that, by (2.4), h + pd > p`, we get that

E[A] < W0 − p` = E[NI]. (2.6)

By risk neutrality, the individual is indifferent between NI and E[NI] for sure and is also
indifferent between A and E[A] for sure. Assuming that the individual prefers more money
to less, by (2.6) she prefers E[NI] for sure to E[A] for sure: denoting, as before, indifference
by ∼ and strict preference by , we can write this as

NI ∼ E[NI]  E[A] ∼ A.

Assuming that the individual’s preferences are transitive, it follows that

NI  A,

that is, the individual strictly prefers not insuring to purchasing contract A. Hence it is not
possible for an insurance company to make positive profits by selling insurance contracts
to risk-neutral individuals: the individuals will simply not buy the offered insurance
contracts.

Although it is intuitively clear that also a risk-loving individual would reject any
insurance contract that would yield non-negative profits to the insurer, the proof requires
more tools than we have developed so far.9

Thus we are left with the case of a risk-averse individual, to which we now turn.

9 In
Exercise 2.22 (at the end of this chapter) the reader is asked to show that a risk-loving individual
would reject a full-insurance contract that yields zero profits to the insurance company.
 2.6 Profitable insurance requires risk aversion 31

2.6.2 Insuring a risk-averse individual

In this section we show that, if an individual is risk averse, it is possible for an insurance
company to make positive profits by offering a contract that will be accepted by the
individual.
The argument assumes that the individual’s preferences are continuous, in the sense
that if she prefers contract B to contract A then contracts that are sufficiently close to B
are still better than A. This is shown in Figure 2.8. Suppose that contract B is preferred to
contract A; then continuity of preferences says that, if we take some other contract C in a
“sufficiently small disk” around B, then it will be true also for C that it is better than A.10

W2

B if B  A
C then
CA

W1

Figure 2.8: Continuity of preferences.

Now consider a risk-averse individual. By definition of risk aversion, she will strictly prefer
the full-insurance contract B with premium h = p` to not insuring, since such contract
leaves her with a wealth of W0 − p` for sure and W0 − p` is the expected value of the
no-insurance lottery NI:
B  NI. (2.7)
Contract B yields zero profits for the insurer:11
π(B) = 0. (2.8)
By continuity of preferences and (2.7), any contract C sufficiently close to B will also be
such that
C  NI. (2.9)
10 This is similar to the property of real numbers that, if b > a then any number c in a sufficiently small
interval around b will also be greater than a.
11 In fact, contract B lies at the intersection of the zero-profit line and the 45o line.
 32 Chapter 2. Insurance: basic notions

If we choose such a contract C which is below the zero-profit line, then – as we saw in
Section 2.5 – π(C) > π(B) and thus, by (2.8), π(C) > 0 and, by (2.9), the individual will
purchase contract C, since it makes her better off relative to not insuring. Hence it is
possible to sell a profitable contract to a risk-averse individual.
The above argument is illustrated graphically in Figure 2.9.

W2

zero-profit line
45o line

NI

C B
Risk aversion ⇒ B  NI
Continuity ⇒ C  NI
C below 0-π line ⇒ π(C) > 0

W1
0

Figure 2.9: Contract C yields positive profits and is better than no insurance.

2.6.3 The profit-maximizing contract for a monopolist

Suppose that the insurance industry is a monopoly, that is, there is only one firm in the
industry. Would a profit-maximizing monopolist want to offer a full insurance contract
or a partial insurance contract to a risk-averse individual? Extending the argument of the
previous section, we can show that for the monopolist the profit-maximizing choice is to
offer full insurance.
Consider any partial insurance contract B = (hB , dB ) (thus dB > 0) that the potential
customer is willing to purchase (thus B  NI); note that the monopolist’s profit from this
contract is
π(B) = hB − p` + pdB .
We want to show that there is a full-insurance contract C = (hC , 0) which the potential
customer is willing to purchase (C  NI) and is such that π(C) > π(B), so that it cannot
be profit-maximizing to offer contract B.
Let A be the following full-insurance contract: A = (hB + pdB , 0). The monopolist’s profit
from this contract would be

π(A) = hB + pdB − p` = π(B),

2.6 Profitable insurance requires risk aversion 33

that is, A and B lie on the same isoprofit line and hence the monopolist is indifferent
between these two contracts. The customer, however, would strictly prefer contract A to
contract B:A  B. In fact, purchasing  contract B = (hB , dB ) can be viewed as playing
W0 − hB W0 − hB − dB
the lottery whose expected value is W0 − hB − pdB and the
1− p p
full-insurance contract A guarantees this amount with certainty; thus, by the assumed risk-
aversion of the customer, A  B. By continuity of preferences, any contract C sufficiently
close to A would still be such that C  B and thus, by transitivity (since, by hypothesis,
B  NI) C  NI. Choosing such a contract below the isoprofit line going through contracts
A and B ensures that π(C) > π(B). Hence if the monopolist were to switch from contract
B to contract C the customer would still purchase insurance and the monopolist’s profits
would increase. The argument is illustrated in Figure 2.10.

W2
45o line
isoprofit line

B
A
C
Hypothesis: BNI
Risk aversion ⇒ AB
Continuity ⇒ CB
Transitivity ⇒ CNI
C below isoprofit line ⇒ π(C)>π(B)

W1

Figure 2.10: Contract C yields higher profits than contract B and is still better than NI.

Thus a monopolist would offer a full-insurance contract to the potential customer. What
is the maximum premium that the monopolist would be able to charge for full-insurance
without turning the customer away? We can answer this question by appealing to the
notion of risk premium (Definition 2.3.2): the monopolist can set the premium up to the
amount
hmax = p` + RNI
 
W0 W0 − `
where RNI is the customer’s risk premium for the no-insurance lottery NI = 1− p p
;
that is, the maximum premium the customer would be willing to pay for full insur-
ance is equal to the expected loss, p`, augmented by the risk premium, RNI . In fact,
E[NI] = W0 − p` and thus, by definition of risk premium,
 
W0 − p` − RNI
NI ∼ .
1
 34 Chapter 2. Insurance: basic notions

In other words, if the customer purchases insurance at premium hmax = p` + RNI then she
is guaranteed the certainty equivalent of the no-insurance lottery. If offered full insurance
at this premium, the potential customer would be indifferent between insuring and not
insuring; thus the monopolist might want to offer full insurance at a slightly lower premium
in order to provide the customer with an incentive to purchase insurance.

2.6.4 Perfectly competitive industry with free entry

In the previous section we considered the extreme case of complete absence of competition,
that is, the case where the insurance industry is a monopoly. In this section we consider the
opposite extreme, namely an insurance industry where competition is so intense that profits
are driven down to zero. The story that is often told for such a mythical industry is that
there is free entry into the industry and thus, if firms in the industry are making positive
profits, then some new entrepreneur will enter seeking to share in these profits; entry of new
firms intensifies competition and drives profits down. We shall assume that all the potential
customers in the industry are identical, in the sense that they have the same preferences,
the same initial wealth and face the same potential loss with the same probability (the case
where potential customers are not identical will be analysed in Chapter 8). Furthermore,
we assume that if a new contract is introduced that the insured customers prefer to their
current contract, then they will switch to the new contract.

Define a free-entry competitive equilibrium as a situation where

1. each firm in the industry makes zero profits, and

2. there is no unexploited profit opportunity in the industry, that is, there is no currently
not offered contract that would yield positive profits to a (existing or new) firm that
offered that contract.

By adopting a simple extension of the argument used in the previous section, we now show
that at a competitive free-entry equilibrium all the active firms, that is, all the firms that
are actually selling insurance,12 offer the same contract, namely the “fair” full-insurance
contract with premium h = p`.

12 There could be inactive firms whose contracts nobody purchases: these firms are also trivially making
zero profits.
2.6 Profitable insurance requires risk aversion 35

The first step in the argument is that – by the zero-profit condition – any actually
purchased contract must lie on the zero-profit line. Suppose that there is a contract, call
it A, that is currently being purchased by some customers (thus A % NI) and is different
from the “fair” full-insurance contract with premium h = p`; call the latter contract B: see
Figure 2.11. By definition of risk aversion, it must be that B  A and, by continuity of
preferences, any contract sufficiently close to B must also be better than A (thus, since
A % NI, by transitivity of preferences such a contract B is better than no insurance). Pick
a contract C sufficiently close to B (so that C  A) and below the zero-profit line (see
Figure 2.11). Then π(C) > 0 and thus a firm that offered this contract would attract all the
customers who are currently purchasing contract A and would make positive profit, so that
the initial situation cannot be a free-entry competitive equilibrium.

W2

zero-profit line
45o line

NI
A

C B Hypothesis: A%NI
Risk aversion ⇒ BA
Continuity ⇒ CA
C below 0-π line⇒ π(C)>0

W1
0

Figure 2.11: Contract C yields positive profits and is better than A (and NI).

We have seen that, no matter whether the insurance industry is a monopoly or a

perfectly competitive industry, the outcome is qualitatively the same, namely that potential
customers are offered full insurance (and only full insurance). There is an important
difference, however: in a perfectly competitive industry the premium of the full-insurance
contract is the “fair” premium h = p`, while the premium that the monopolist charges for
full insurance is higher, namely h = p` + RNI (recall that RNI is the risk premium of the
no-insurance lottery).

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 2.7.5 at the end of this chapter.
 36 Chapter 2. Insurance: basic notions

2.7 Exercises
The solutions to the following exercises are given in Section 2.8 at the end of this chapter.

2.7.1 Exercises for Section 2.2: Money lotteries and attitudes to risk

Exercise 2.1 Consider the following money lottery:

 
$10 $15 $18 $20 $25 $30 $36
3 1 3 2 3
12 12 0 12 12 0 12

(a) What is its expected value?

(b) If a risk-neutral individual is given a choice between the above lottery and $23
for sure, what will she choose?


 
o1 o2 o3
Exercise 2.2 Consider the lottery 3 5 2 where
10 10 10

• o1 is the outcome where you get $100 and an A in the class on the Economics of
Uncertainty and Information,
• o2 is the outcome where you get a free trip to Disneyland (which would normally
cost $500) and a C in the class and
• o3 is the outcome where you get a $150 gift certificate at Amazon.com and a B in
the class.
If you are risk neutral, what sum of money would you consider to be just as good as the
lottery? 

Exercise 2.3 Given the choice between getting $18 for sure or playing the lottery
 
$10 $20 $30
3 5 2
10 10 10

James – who likes money (that is, prefers more money to less) – chooses to get $18 for
sure. Is he risk neutral? 

Exercise 2.4 Find the expected value of the following lottery

 
24 12 48 6
1 2 1 2 .
6 6 6 6

2.7 Exercises 37
 
o1 o2 o3
Exercise 2.5 Consider the lottery 1 1 1 where
4 2 4
• o1 = you get an invitation to have dinner at the White House,
• o2 = you get (for free) a puppy of your choice,
• o3 = you get $600.
What is the expected value of this lottery? 

Exercise 2.6 Consider the following money lottery

 
$10 $15 $18 $20 $25 $30 $36
L= 3 1 3 2 3
12 12 0 12 12 0 12

(a) What is the expected value of the lottery?

(b) Ann prefers more money to less and has transitive preferences. She says that,
between getting $20 for certain and playing the above lottery, she would prefer
$20 for certain. What is her attitude to risk?
(c) Bob prefers more money to less and has transitive preferences. He says that,
given the same choice as Ann, he would prefer playing the lottery. What is his
attitude to risk?


Exercise 2.7 Sam has a debilitating illness and has been offered two mutually exclusive
courses of action:
(1) take some well-known drugs which have been tested for a long time, or
(2) take a new experimental drug.
If he chooses (1) then for certain his pain will be reduced to a bearable level. If he
chooses (2) then he has a 50% chance of being completely cured and a 50% chance of
no benefits from the drug and possibly some harmful side effects. He chose (1). What
is his attitude to risk? 

Exercise 2.8 Shirley owns a house worth $200, 000. The value of the building is
$75, 000 and the value of the land is $125, 000. In the area where she lives there is
a 10% probability that a fire will completely destroy the building (on the other hand,
the land would not be affected by a fire). An insurance company offers a policy that
covers the full replacement cost of the building in the event of fire (that is, there is no
deductible). The premium for this policy is $7, 500 per year. What attitude to risk must
Shirley have in order to purchase the insurance policy? [Hint: think in terms of wealth
levels.] 
 38 Chapter 2. Insurance: basic notions

Exercise 2.9 Bill’s entire wealth consists of the money in his bank account: $12, 000.
Bill’s friend Bob claims to have discovered a great investment opportunity, which
would require an investment of $10, 000. Bob does not have any money and asks Bill
to provide the $10, 000. According to Bob, the investment could yield a return of
$150, 000, in which case Bob will return the initial $10, 000 to Bill and then give him
50% of the remaining $140, 000. According to Bob the probability that the investment
will be successful is 12% and the probability that the initial investment of $10, 000
will be completely lost is 88%. Bill decides to go ahead with the investment and gives
$10, 000 to Bob. What is Bill’s attitude to risk? 

2.7.2 Exercises for Section 2.3: Certainty equivalent and risk premium

Exercise 2.10 In Section 2.3 we defined three relations over money lotteries: the strict
preference relation (denoted by ), the indifference relation (denoted by ∼) and the ‘at
least as good’ relation (denoted by %). As a matter of fact, one can simply postulate
just one relation, the ‘at least as good’ relation %, and derive the other two from it as
follows:
• L1  L2 if and only if L1 % L2 and it is not the case that L2 % L1 ,
• L1 ∼ L2 if and only if L1 % L2 and also L2 % L1 .
Recall that a relation % over a set L of money lotteries is complete if, for every two
lotteries L1 , L2 ∈ L , either L1 % L2 or L2 % L1 (or both) and is transitive if, for every
three lotteries L1 , L2 , L3 ∈ L , if L1 % L2 and L2 % L3 then L1 % L3 .
Prove that if the ‘at least as good’ relation % is complete and transitive then the
derived ‘strict preference’ relation  is also transitive, that is, if L1  L2 and L2  L3
then L1  L3 . 

Exercise 2.11 As in Exercise 2.10 take the ‘at least as good’ relation % as primitive and
derive from it the indifference relation ∼. Prove that if the ‘at least as good’ relation %
is transitive then the derived indifference relation ∼ is also transitive, that is, if L1 ∼ L2
and L2 ∼ L3 then L1 ∼ L3 . 

Exercise 2.12 As in Exercise 2.10 take the ‘at least as good’ relation % as primitive
and derive from it the ‘strict preference’ relation  and the indifference relation ∼.
Prove that if the ‘at least as good’ relation % is transitive then if L1  L2 and L2 ∼ L3
then L1  L3 . 

Exercise 2.13 As in Exercise 2.10 take the ‘at least as good’ relation % as primitive
and derive from it the ‘strict preference’ relation  and the indifference relation ∼.
Prove that if the ‘at least as good’ relation % is transitive then if L1 ∼ L2 and L2  L3
then L1  L3 . 
 2.7 Exercises 39

2.7.3 Exercises for Section 2.4: Insurance: basic concepts

Exercise 2.14 Tom’s entire wealth consists of a boat which is worth $38, 000. He is
worried about the possibility of a hurricane damaging the boat. Typically, restoring a
damaged boat costs $25, 000. Unfortunately, because of global warming, the probability
of a hurricane hitting his area is not negligible: it is 12%. The diagrams requested below
should all be drawn, as usual, in the cartesian plane where on the horizontal axis you
measure wealth in the bad state (W1 ) and on the vertical axis wealth in the good state
(W2 ). Call such a diagram a “wealth diagram”.
(a) Represent the no-insurance option (NI) as a point in a wealth diagram.
(b) Suppose that an insurance company offers the following insurance contract, call
it B: the premium is $2, 000 and the deductible is $9, 000. Represent contract B
in the wealth diagram of Part (a).
(c) Suppose that another insurance company offers the following full-insurance
contract, call it C: the premium is $3, 000. Represent contract C in the wealth
diagram of Part (a).
(d) If Tom is risk neutral, how will he rank the three options: NI, B and C?
(e) If Tom is risk averse, has transitive preferences and prefers more money to less,
how will he rank the three options: NI, B and C?


Exercise 2.15 Refer to the diagram shown in Figure 2.12.

(a) Calculate the potential loss `.
(b) Calculate the premium hA and the deductible dA of contract A.
(c) Calculate the premium hB and the deductible dB of contract B.
(d) For each of the two contracts state whether it is a partial-insurance contract or a
full-insurance contract.


good state
W2

NI
600
A
500 B
450

W1
100 250 425 bad state

Figure 2.12: The diagram for Exercise 2.15

 40 Chapter 2. Insurance: basic notions

2.7.4 Exercises for Section 2.5: Isoprofit lines

Exercise 2.16 Consider again the information given in Exercise 2.14: Tom’s entire
wealth consists of a boat which is worth $38, 000; he is worried about the possibility
of a hurricane damaging the boat, in which case it would cost him $25, 000 to repair
the boat; the probability of a hurricane hitting his area is 12%. He is considering two
insurance contracts: contract B, with premium $2, 000 and deductible $9, 000, and
contract C, with premium $3, 000 and zero deductible.
(a) If Tom were to purchase contract B, what would the expected profit be for the
insurance company?
(b) If Tom were to purchase contract C, what would the expected profit be for the
insurance company?
(c) What is the slope of an isoprofit line?
(d) Find the equation of the isoprofit line that goes through contract B and draw it in
a wealth diagram.
(e) Find the equation of the isoprofit line that goes through contract C and draw it in
a wealth diagram.


Exercise 2.17 Consider again the information given in Exercise 2.15 (shown in Figure
2.12). Assume that the probability of loss is 20%.
(a) Calculate the expected profit from contract A.
(b) Calculate the expected profit from contract B.
(c) Draw the zero-profit line.
(d) Draw the isoprofit line that goes through contract A.
(e) Draw the isoprofit line that goes through contract B.


Exercise 2.18 The equation of the zero-profit line is W2 = 8, 100 − 19 W1 . The individ-
ual’s initial wealth is $7, 600.
(a) What is the probability of loss?
(b) Calculate the potential loss `.
(c) Find a full-insurance contract, call it A, that yields a profit of $40.
(d) Find a contract, call it B, that lies on the isoprofit line through A and has a
deductible of $1, 500.
(e) Find a contract, call it C, with deductible d = 2, 000 that yields a profit of $25.
(f) Write the equation of the isoprofit line that goes through contract A (in the wealth
diagram).
(g) Write the equation of the isoprofit line that goes through contract C (in the wealth
diagram).

2.7 Exercises 41

Exercise 2.19 Consider the wealth diagram shown in Figure 2.13. Let p = 0.2.
(a) Interpret each point (including NI) as an insurance contract and express it in
terms of premium and deductible.
(b) Calculate the expected profit from each contract.
(c) Find the equation of the isoprofit line that goes through each contract (including
the one that goes through point NI).


(good state)

W2

NI 45o line
1,600
1,556 A
1,480 B

W1
0 1,024 1,390 (bad state)

Figure 2.13: The wealth diagram for Exercise 2.19

Exercise 2.20 In a wealth diagram

(a) show the subset of the set of insurance contracts that contains the contracts that
yield non-negative profits to the insurer (recall that the set of insurance contracts
is the shaded triangle shown in Figure 2.1 on page 22),
(b) of all the contracts that yield non-negative profits to the insurer, find the one that
is most preferred by a risk-averse individual and explain why there is only one
such contract,
(c) of all the contracts that yield non-negative profits to the insurer, find the ones that
are most preferred by a risk-neural individual.

 42 Chapter 2. Insurance: basic notions

Exercise 2.21 In a wealth

diagram draw an isoprofit line and pick a partial-insurance

contract A = W1A ,W2A on this line. Next, pick an insurance contract B = W1B ,W2B
which is vertically below A, so that W1B = W1A and W2B < W2A .
Prove that π(B) > π(A). 

2.7.5 Exercises for Section 2.6: Profitable insurance requires risk aversion

Exercise 2.22 Prove that a risk-neutral individual is indifferent between no insurance

and a full-insurance contract that yields zero profits to the insurance company. 

Exercise 2.23 Prove that a risk-loving individual strictly prefers no insurance to a “fair”
full-insurance contract (that is, a full-insurance contract that yields zero profits to the
insurance company). 

Exercise 2.24 Ann has an initial wealth of $24, 000 and faces a potential loss of
$15, 000 with probability 20%. The risk premium of the no-insurance lottery for Ann is
$2, 000. If Ann is offered full insurance for a premium of $4, 920 will she take it? 

Exercise 2.25 Bob has an initial wealth of $18, 000 and faces a potential loss of
$10, 000 . The risk premium of the no-insurance lottery for Bob is $900. The maximum
premium that he is willing to pay for full insurance is $4, 000. What is the probability
of loss? 

2.8 Solutions to Exercises

Solution to Exercise 2.1.

(a) The expected value is

3 1 3 2 3 263
10 + 15 + 0 (18) + 20 + 25 + 0 (30) + 36 = = $21.92.
12 12 12 12 12 12
(b) A risk-neutral person is indifferent between the lottery and $21.92 for sure. Assum-
ing that she prefers more money to less, she will prefer $23 to $21.92. Thus, if her
preferences are transitive, she will prefer $23 to the lottery. 
3
Solution to Exercise 2.2. One might be tempted to compute the “expected value” 10 100 +
5 2
10 500 + 10 150 = 310 and answer: $310. However, this answer would be wrong, because
the given lottery is not a money lottery: the outcomes are not just sums of money (they do
involve sums of money but also what grade you get in the class). The definition of risk
neutrality can only be applied to money lotteries. 
2.8 Solutions to Exercises 43
3 5 2
Solution to Exercise 2.3. The expected value of the lottery is 10 10 + 10 20 + 10 30 = 19. If
James were risk-neutral he would consider the lottery to be just as good as getting $19 for
sure and would therefore choose the lottery (since getting $19 is better than getting $18).
Hence, he is not risk neutral. 
 
24 12 48 6
Solution to Exercise 2.4 The expected value of the lottery 1 2 1 2 is
6 6 6 6
1 2 1 2
6 24 + 6 12 + 6 48 + 6 6 = 18. 

Solution to Exercise 2.5 This was a trick question! There is no expected value because
the outcomes are not numbers. 

Solution to Exercise 2.6

(a) As already computed in Exercise 2.1, the expected value of the lottery
 
$10 $15 $18 $20 $25 $30 $36
L= 3 1 3 2 3
12 12 0 12 12 0 12

3 1 3 2 3 263
is E[L] = 12 10 + 12 15 + 0 (18) + 12 20 + 12 25 + 0 (30) + 12 36 = 12 = $21.92.
(b) Since Ann prefers more money to less, she prefers $21.92 to $20 ($21.92  $20).
She said that she prefers $20 to lottery L ($20  L). Thus, since her preferences are
transitive, she prefers $21.92 to lottery L ($21.92  L). Hence, she is risk averse.
(c) The answer is: we cannot tell. First of all, since Bob prefers more money to less,
he prefers $21.92 to $20 ($21.92  $20). Bob could be risk neutral, because a risk
neutral person would be indifferent between L and $21.92 (L ∼ $21.92); since Bob
prefers $21.92 to $20 and has transitive preferences, if risk neutral he would prefer
L to $20. However, Bob could also be risk loving: a risk-loving person prefers L to
$21.92 (L  $21.92) and we know that he prefers $21.92 to $20; thus, by transitivity,
he would prefer L to $20. But Bob could also be risk averse: he could consistently
prefer $21.92 to L and L to $20 (for example, he could consider L to be just as good
as $20.50). 

Solution to Exercise 2.7 Just like Exercise 2.5, this was a trick question! Here the basic
outcomes are not sums of money but states of health. Since the described choice is
not one between money lotteries, the definitions of risk aversion/neutrality/love are not
applicable. 

Solution to Exercise 2.8 The decision not to buy insurance is the decision to face the fol-
lowing lottery: with probability 0.9 Shirley’s wealth will be $200, 000, with probability 0.1
it will be $125, 000. The expected value of this lottery is: 0.9(200, 000) + 0.1(125, 000) =
$192, 500. The insurance policy guarantees a wealth of $200, 000 − 7, 500 = 192, 500.
Hence Shirley will buy the insurance policy if she is risk-averse, will be indifferent be-
tween buying and not buying if she is risk-neutral and will prefer not to buy if she is
risk-loving. 
44 Chapter 2. Insurance: basic notions

Solution to Exercise 2.9 If Bill refuses to invest, his wealth is $12,000 for sure. If Bill
$2, 000 $82, 000
gives $10, 000 to Bob to invest then he faces the following lottery: L = .
0.88 0.12
88 12
The expected value of L is E[L] = 100 2, 000 + 100 82, 000 = $11, 600. If Bill were risk
averse he would prefer $11,600 for sure to the investment (lottery L) and obviously he
will prefer $12,000 to $11,600; thus he would prefer $12,000 for sure to the investment;
since he decides to go ahead with the investment he is not risk averse. If Bill were risk
neutral he would be indifferent between $11,600 for sure and the investment (lottery L)
and obviously he will prefer $12,000 to $11,600; thus he would prefer $12,000 for sure
to the investment; since he decides to go ahead with the investment he is not risk neutral
either. Hence Bill is risk-loving. 

Solution to Exercise 2.10 Let L1 , L2 , L3 ∈ L be such that L1  L2 and L2  L3 . We need

to show that L1  L3 . Since L1  L2 , L1 % L2 and since L2  L3 , L2 % L3 . Thus, by
transitivity of %, L1 % L3 . It remains to prove that it is not the case that L3 % L1 . Suppose
that L3 % L1 ; then, since L1 % L2 it would follow from transitivity of % that L3 % L2 ,
contradicting the hypothesis that L2  L3 . 

Solution to Exercise 2.11 Let L1 , L2 , L3 ∈ L be such that L1 ∼ L2 and L2 ∼ L3 . We

need to show that L1 ∼ L3 . Since L1 ∼ L2 , L1 % L2 and since L2 ∼ L3 , L2 % L3 ; thus, by
transitivity of %, L1 % L3 . Similarly, since L1 ∼ L2 , L2 % L1 and since L2 ∼ L3 , L3 % L2 ;
thus, by transitivity of %, L3 % L1 . It follows from L1 % L3 and L3 % L1 that L1 ∼ L3 . 

Solution to Exercise 2.12 Let L1 , L2 , L3 ∈ L be such that L1  L2 and L2 ∼ L3 . We

need to show that L1  L3 . Since L1  L2 , L1 % L2 and since L2 ∼ L3 , L2 % L3 ; thus, by
transitivity of %, L1 % L3 . It remains to show that it is not the case that L3 % L1 . Suppose
that L3 % L1 ; then, in conjunction with L1 % L3 , we get that L3 ∼ L1 ; since, by hypothesis,
L2 ∼ L3 , it would follow from transitivity of ∼ (proved in Exercise 2.11) that L1 ∼ L2
contradicting the hypothesis that L1  L2 . 

Solution to Exercise 2.13 Let L1 , L2 , L3 ∈ L be such that L1 ∼ L2 and L2  L3 . We

need to show that L1  L3 . Since L1 ∼ L2 , L1 % L2 and since L2  L3 , L2 % L3 ; thus, by
transitivity of %, L1 % L3 . It remains to show that it is not the case that L3 % L1 . Suppose
that L3 % L1 ; then, in conjunction with L1 % L3 , we get that L3 ∼ L1 ; since, by hypothesis,
L2 ∼ L1 , it would follow from transitivity of ∼ (proved in Exercise 2.11) that L2 ∼ L3
contradicting the hypothesis that L2  L3 . 
2.8 Solutions to Exercises 45

Solution to Exercise 2.14

(a) See Figure 2.14.
(b) W1B = 38, 000−2, 000−9, 000 = 27, 000 and W2B = 38, 000−2, 000 = 36, 000. Con-
tract B is shown in Figure 2.14.
(c) W1C = 38, 000 − 3, 000 = 35, 000 = W2C . Contract C is shown in Figure 2.12.
 
13, 000 38, 000 12
(d) NI represents the lottery 12 88 whose expected value is 100 13, 000+
100 100  
88 27, 000 36, 000
100 38, 000 = 35, 000. Contract B represents the lottery 12 88 whose
100 100
12 88
expected value is 100 27, 000 + 100 36, 000 = 34, 920. Contract C represents the
35, 000
lottery whose expected value is 35, 000. Thus if Tom is risk neutral
1
then he is indifferent between NI and C and prefers either of them to B.
(e) By risk aversion, Tom strictly prefers C to NI. He also strictly prefers C to B: since
he prefers more money to less, he prefers $35, 000 to $34, 920 and, by risk aversion,
he prefers $34, 920 for sure to B; hence, by transitivity, he prefers C to B. On the
other hand, we cannot tell how he ranks NI relative to B. 

wealth in
good state
W2

NI 45o line
38,000
B
36,000 C
35,000

0
13,000 27,000 35,000 W1
wealth in
bad state

Figure 2.14: The diagram for Exercise 2.14

46 Chapter 2. Insurance: basic notions

Solution to Exercise 2.15

(a) ` = 600 − 100 = 500.
(b) hA = 600 − 500 = 100, dA = 500 − 250 = 250.
(c) hB = 600 − 450 = 150, dB = 450 − 425 = 25.
(d) Both contracts are partial-insurance contracts (neither of them lies on the 45o line).


Solution to Exercise 2.16

12 12
(a) π(B) = hB − p` + p dB = 2, 000 − 100 25, 000 + 100 9, 000 = 80.
12
(b) π(C) = hC − p` = 3, 000 − 100 25, 000 = 0.
12
p 100 3
(c) The slope of each isoprofit line is − 1−p =− 88 = − 22 = −0.136.
100
3
(d) Since isoprofit lines are straight lines with slope − 22 , they are of the form
3
W2 = a − 22 W1 . To find the value of a replace W2 with 36, 000 and W1 with 27, 000
and solve for a to get a = 436,500
11 = 39, 681.82. The isoprofit line is shown in Figure
2.15.
(e) This is the zero-profit line and thus it goes through the no-insurance point NI. Again,
3
it is of the form W2 = a − 22 W1 . To find the value of a replace both W1 and W2 with
35, 000 and solve for a to get a = 437,500
11 = 39, 772.73. The isoprofit line is shown
in Figure 2.15. 

(good state)

W2

45o line
NI

C
B

0 W1 (bad state)

Figure 2.15: The diagram for Exercise 2.16

2.8 Solutions to Exercises 47

Solution to Exercise 2.17

(a) π(A) = hA − p(` − dA ) = 100 − 15 (500 − 250) = 50.

(b) π(B) = hB − p(` − dB ) = 150 − 15 (500 − 25) = 55.

(c) See Figure 2.16. All three profit lines have a slope of − 0.2 1
0.8 = − 4 . The equation of
the zero-profit line is W2 = 625 − 14 W1 .

(d) See Figure 2.16. The equation of the line is W2 = 562.5 − 14 W1 .

(e) See Figure 2.16. The equation of the line is W2 = 556.25 − 14 W1 . 

(good state)
W2

NI
600
A
500 π =0
450 π = 50
B π = 55

W1
0 100 250 425 (bad state)

Figure 2.16: The diagram for Exercise 2.17

48 Chapter 2. Insurance: basic notions

Solution to Exercise 2.18

p
(a) Let p be the probability of loss. We know that the slope of any isoprofit line is − 1−p .
p
Thus, since the slope of the zero-profit line is − 91 , it must be that 1−p = 91 . Solving
1
for p we get that p = 10 .

(b) The no-insurance point is on the zero-profit line. We know that the vertical coordinate
of the no-insurance point is W0 = 7, 600. Thus, to find the horizontal coordinate,
which is equal to W0 − `, we must solve the equation 7, 600 = 8, 100 − 19 W1 which
gives W1 = 4, 500. Thus ` = 7, 600 − 4, 500 = 3, 100.
1
(c) A full-insurance contract with premium h yields a profit of h − p` = h − 10 3, 100 =
h − 310. Thus we want h to solve the equation h − 310 = 40. Hence contract A has
a premium of 350 (and, of course, zero deductible).

(d) A contract with premium h and deductible 1, 500 yields a profit of h − p`+1, 500p =
h − 310 + 150 = h − 160. Hence, since B is on the same isoprofit line as A, it must
be that h − 160 = 40, that is, h = 200. Thus contract B has a premium of 200 and a
deductible of 1, 500.

(e) Similar reasoning as in Part (d): we need h − p` + 2, 000p to be equal to 25. Hence
we must solve h − 310 + 200 = 25, which gives h = 135. Thus contract C has a
premium of 135 and a deductible of 2, 000.

(f) The equation of an isoprofit line is of the form W2 = a − 19 W1 . To find the value of
a for the isoprofit line that goes through contract A we must solve 7, 600 − 350 =
72,500
a − 19 (7, 600 − 350) to get a = 9 = 8, 055.56. Thus the equation of the isoprofit
line that goes through contract A is W2 = 8, 055.56 − 91 W1 .

(g) Again, the equation of an isoprofit line is of the form W2 = a − 19 W1 . To find

the value of a for the isoprofit line that goes through contract C we must solve
72,650
7, 600 − 135 = a − 19 (7, 600 − 135 − 2, 000) to get a = 9 = 8, 072.22. Thus the
equation the isoprofit line that goes through contract C is W2 = 8, 072.22 − 91 W1 .
2.8 Solutions to Exercises 49

Solution to Exercise 2.19

(a) First of all, note that the initial wealth is W0 = 1, 600 and the potential loss is
` = 1, 600 − 1, 024 = 576. Let us write each point as a pair (h, d) where h = W0 −W2
is the premium and d = W2 −W1 is the deductible. Thus
NI = (0, 576),
A = (1600 − 1556, 1556 − 1390) = (44, 166),
B = (1600 − 1480, 1480 − 1390) = (120, 90),
C = (1600 − 1390, 0) = (210, 0).

(b) The expected profit from contract (h, d) is π = h − p(` − d). Thus π(NI) = 0,
π(A) = −38, π(B) = 22.8 and π(C) = 94.8.

p
(c) All the isoprofit lines are straight lines and all have the same slope given by − 1−p =
1
− 5
4 = − 41 . Thus starting at a point (W1 ,W2 ), if you reduce W1 to 0 then the vertical
5
coordinate changes to W2 + 14 W1 , yielding the vertical intercept. Applying this to
point NI we get that by reducing the horizontal coordinate by 1,024, the vertical
1,024
coordinate increases by 4 = 256 to 1, 600 + 256 = 1, 856. Hence the equation
of the isoprofit line that goes through point NI (which is the zero-profit line) is
W2 = 1, 856 − 14 W1 . Applying the same procedure we get that

Equation of isoprofit line through A: W2 = 1, 903.5 − 14 W1

Equation of isoprofit line through B: W2 = 1, 827.5 − 14 W1

Equation of isoprofit line through C: W2 = 1, 737.5 − 41 W1

50 Chapter 2. Insurance: basic notions

Solution to Exercise 2.20

(good state,
probability 1 − p)

W2
zero-profit line
NI 45o line
W0
C
A
B

set of contracts
with π ≥ 0

0 W1
W0 − ` (bad state,
probability p)

Figure 2.17: The diagram for Exercise 2.20

(a) The set of contracts that yield non-negative profits is shown as a shaded triangle in
Figure 2.17.
(b) Contract C shown in Figure 2.17 is the contract that is most preferred by a risk-averse
individual among the contracts in the shaded triangle. C is a full-insurance contract
with premium h = p` guaranteeing a wealth of W0 − p`. Any other contract on the
zero-profit line is worse than contract C because it gives rise to a non-degenerate
lottery with expected value W0 − p`. Any other contract on the thick part of the 45o
line is worse than contract C because it has a higher premium (while still being a
full-insurance contract). Finally, any contract, say A, inside the shaded area lies on
an isoprofit line that goes through a contract which is on the thick part of the 45o
line below point C, call this point B (see Figure 2.17); then B is better than A (since
it guarantees the expected value of A) but is worse than C, so that C is better than A.
(c) The contracts on the thick line from NI to C: the individual is indifferent among all
these contracts, because they correspond to lotteries that have the same expected
value as the NI lottery. On the other hand, contracts below this line have an expected
value which is less than the expected value of the NI lottery and thus are worse than
NI.
2.8 Solutions to Exercises 51

Solution to Exercise 2.21 Let contract B be vertically below contract A as shown in Figure
2.18

(probability 1 − p)
W2
isoprofit line
45o line

A
W0 − hA
∆ B
W0 − (hA + ∆)
| {z }
hB isoprofit line π(B) − π(A)
= ∆ − p∆
= (1 − p)∆ > 0

0 W1 (probability p)
W0 − hA − dA
= W0 − (hA + ∆)
| {z }
hB −dB
hence dB − dA = −∆

Figure 2.18: The diagram for Exercise 2.21

Then W1B = W1A and W2B < W2A . From the latter we deduce that the premium of contract
B is higher than the premium of contract A. Let ∆ = W2A − W2B > 0 be the amount by
which B’s premium exceeds A’s premium. From the fact that W1B = W1A we calculate the
difference between the deductible of contract B and the deductible of contract A as follows:
=W2A −∆ 1 =W A
z}|{ z}|{
W B − W1B = W2A −W1A − ∆,
| 2 {z } | {z }
dB dA

so that
dB − dA = −∆.
Thus, when contract B is vertically below contract A, then hB is greater than hA and, letting
∆ be the amount by which hB exceeds hA (∆ = hB − hA ), dB − dA = −∆, that is, the
deductible of contract B is less than the deductible of contract A by an amount equal to
∆. From this we deduce that contract B yields higher profits that contract A, because it
yields an extra $∆ for sure (the premium is received with probability 1) and involves an
extra payment of $∆ to the insured only with probability p (the payment is made with
probability p < 1), that is,13
π(B) − π(A) = ∆ − p∆ = (1 − p)∆ > 0.
13 This conclusion can be verified directly, as follows: π(B) = W0 − W2B − p` + p(W2B − W1B ) and
52 Chapter 2. Insurance: basic notions

Solution
 to Exercise 2.22
 The no-insurance option corresponds to the wealth lottery
W0 W0 − `
NI = whose expected value is
1− p p

E[NI] = W0 − p`. (2.10)

Suppose that the individual is risk neutral and is offered a full-insurance contract (h, 0)
which yields zero profits, that is,

h − p` = 0 or, equivalently, h = p`. (2.11)

 
W0 − h
For the potential customer, such a contract corresponds to the wealth lottery C =
1
whose expected value is

E[C] = W0 − h =
|{z} E[NI]. (2.12)
using (2.10) and (2.11)

By risk neutrality, the individual is indifferent between NI and E[NI] for sure (NI ∼ E[NI])
and is also indifferent between C and E[NI] for sure (since – by (2.12) – C and E[NI] for
sure are the same contract):

NI ∼ E[NI] ∼ C.

Assuming that the individual’s preferences are transitive, it follows that

NI ∼ C.

π(A) = W0 −W2A − p` + p(W2A −W1A ) so that (recall that W1B = W1A and thus pW1B = pW1A )

π(B) − π(A) = (W2A −W2B ) − p(W2A −W2B ) = (1 − p) (W2A −W2B ) .

| {z }
∆
2.8 Solutions to Exercises 53

Solution to Exercise 2.23 The proof is a simple modification of theproof for Exercise

W0 W0 − `
2.22: the no-insurance option corresponds to the wealth lottery NI =
1− p p
whose expected value is

E[NI] = W0 − p`. (2.13)

Suppose that the individual is risk loving and is offered a full-insurance contract (h, 0)
that yields zero profits, that is,

h = p`. (2.14)
 
W0 − h
For the potential customer, such a contract corresponds to the wealth lottery C =
  1
E[C]
which is the same as = E[C] = W0 − h |{z}= E[NI]. Since the individual is risk
1
by (2.14)
loving, he prefers NI to E[NI] for sure (NI  E[NI]) and is indifferent between C and
E[NI] for sure, since it is the same lottery (C ∼ E[NI]); thus NI  E[NI] ∼ C. Assuming
that the individual’s preferences are transitive, it follows that NI  C.

Solution to Exercise 2.24 The maximum premium that Ann is willing to pay for full
insurance is hmax = p` + RNI = 15 15, 000 + 2, 000 = 5, 000. Since she is offered full
insurance at a lower premium, namely 4, 920, she will take it (she is better off with full
insurance at this premium than with no insurance).

Solution to Exercise 2.25 The maximum premium that Bob is willing to pay for full insur-
ance is hmax = p` + RNI = p10, 000 + 900. We are told that this is equal to 4, 000. Thus, to
find p we solve the equation p10, 000 + 900 = 4, 000 whose solution is
3,100
p= 10,000 = 0.31.
 3. Expected Utility Theory

3.1 Expected utility: theorems

As noted in the previous chapter, with the exception of risk-neutral individuals, even if
we restrict attention to money lotteries we are not able to say much – in general – about
how an individual would choose among lotteries. What we need is a theory of “rational”
preferences over lotteries that
(1) is general enough to cover lotteries whose outcomes are not necessarily sums of money
and
(2) is capable of accounting for different attitudes to risk in the case of money lotteries.
One such theory is the theory of expected utility, to which we now turn.
The theory of expected utility was developed by the founders of Game Theory, namely
John von Neumann and Oskar Morgenstern, in their 1944 book Theory of Games and
Economic Behavior. In a rather unconventional way, we shall first (in this section) state
the main result of the theory (which we split into two theorems) and then (in the following
section) explain the assumptions (or axioms) behind that result. The reader who is not
interested in understanding the conceptual foundations of expected utility theory, but wants
to understand what the theory says and how it can be used, can study this section and skip
the next.
Let O be a set of basic outcomes. Note that a basic outcome need not be a sum of
money: it could be the state of an individual’s health, or whether the individual under
consideration receives an award, or whether it will rain on the day of a planned outdoor
party, etc. Let L (O) be the set of simple lotteries (or probability distributions) over O.
56 Chapter 3. Expected Utility Theory

We will assume throughout this chapter that O is a finite set: O!= {o1 , o2 , ..., om } (m ≥ 1).
o1 o2 ... om
Thus, an element of L (O) is of the form with 0 ≤ pi ≤ 1, for all
p1 p2 ... pm
i = 1, 2, ..., m, and p1 + p2 + ... + pm = 1. We will use the symbol L (with or without
subscript) to denote an element of L (O), that is, a simple lottery. Lotteries are used to
represent situations of uncertainty.! For example, if m = 4 and the individual faces the
o1 o2 o3 o4
lottery L = 2 then she knows that, eventually, the outcome will be
5 0 51 25
one and only one of o1 , o2 , o3 , o4 , but does not know which one; furthermore, she is able
to quantify her uncertainty by assigning probabilities to these outcomes. We interpret
these probabilities either as objectively obtained from relevant (past) data or as subjective
estimates by the individual, as explained in Chapter 2 (Section 2.1).
The assignment of zero probability to a particular basic outcome is taken to be an
expression of belief, not impossibility: the individual is confident that the outcome will not
arise, but she cannot rule out that outcome on logical grounds or by appealing to the laws
of nature.
Among the elements of L (O) there are the degenerate lotteries that assign probability!1
o1 o2 o3 o4
to one basic outcome: for example, if m = 4 one degenerate lottery is .
0 0 1 0
To simplify the notation we will often denote
! degenerate lotteries as basic outcomes, that
o1 o2 o3 o4
is, instead of writing we will simply write o3 . Thus, in general,
0 0 1 0
!
o1 ... oi−1 oi oi+1 ... om
the degenerate lottery will be denoted by oi . As
0 0 0 1 0 0 0
another simplification, we will often omit those outcomes
!that are assigned zero probability.
o1 o2 o3 o4
For example, if m = 4, the lottery 1 will be written more simply as
3 0 23 0
!
o1 o3
1 2 .
3 3
Throughout this chapter we shall call the individual under consideration the Decision-
Maker, or DM for short. The theory of expected utility assumes that the DM has a complete
and transitive ranking % of the elements of L (O) (indeed, this is one of the axioms listed
in the next section). As in Chapter 2, the interpretation of L % L0 is that the DM considers
L to be at least as good as L0 . By completeness, given any two lotteries L and L0 , either
L  L0 (the DM prefers L to L0 ) or L0  L (the DM prefers L0 to L) or L ∼ L0 (the DM is
indifferent between L and L0 ). Furthermore, by transitivity, for any three lotteries L1 , L2
and L3 , if L1 % L2 and L2 % L3 , then L1 % L3 . Besides completeness and transitivity, a
number of other “rationality” constraints are postulated on the ranking % of the elements
of L (O); these constraints are the so-called Expected Utility Axioms and are discussed in
3.1 Expected utility: theorems 57

the next section.

Definition 3.1.1 A ranking % of the elements of L (O) that satisfies the Expected Utility
Axioms (listed in the next section) is called a von Neumann-Morgenstern ranking.

The two theorems in this section are the key results in the theory of expected utility.
Theorem 3.1.1 [von Neumann-Morgenstern, 1944].
Let O = {o1 , o2 , ..., om } be a set of basic outcomes and L (O) the set of simple lotteries
over O. If % is a von Neumann-Morgenstern ranking of the elements of L (O) then
there exists a function U : O → R, called a von Neumann-Morgenstern utility function,
that assigns a number (called utility) to every basic  outcome and is such
 that, for any
!
o1 o2 ... om o1 o2 ... om
two lotteries L = and L0 =  ,
p1 p2 ... pm q1 q2 ... qm

L  L0 if and only if E[U(L)] > E[U(L0 )], and

L ∼ L0 if and only if E[U(L)] = E[U(L0 )]

where
! !
U(o1 ) U(o2 ) ... U(om ) U(o1 ) U(o2 ) ... U(om )
U(L) = , U(L0 ) = ,
p1 p2 ... pm q1 q2 ... qm

E[U(L)] is the expected value of the lottery U(L) and E[U(L0 )] is the expected value of
the lottery U(L0 ), that is,

E[U(L)] = p1U(o1 ) + p2U(o2 ) + ... + pmU(om ), and

E[U(L0 )] = q1U(o1 ) + q2U(o2 ) + ... + qmU(om ).

E[U(L)] is called the expected utility of lottery L (and E[U(L0 )] the expected utility of
lottery L0 ).
We say that any function U : O → R that satisfies the property that, for any two lotteries
L and L0 , L % L0 if and only if E[U(L)] ≥ E[U(L0 )] represents the preferences (or
ranking) %.

Before we comment on Theorem 3.1.1 we give an example of how one can use it.
Theorem 3.1.1 sometimes allows us to predict an individual’s choice between two lotteries
C and D if we know how that individual ranks two different lotteries A and B.
58 Chapter 3. Expected Utility Theory

For example, suppose we observe that Susan is faced with the choice between lotteries A
and B below and she says that she prefers A to B:
! !
o1 o2 o3 o1 o2 o3
A= B=
0 0.25 0.75 0.2 0 0.8
With this information we can predict which of the following two lotteries C and D she will
choose, if she has von Neumann-Morgenstern preferences:
! !
o1 o2 o3 o1 o2 o3
C= D= = o2 .
0.8 0 0.2 0 1 0
Let U be a von Neumann-Morgenstern utility function whose existence is guaranteed
by Theorem 3.1.1. Let U(o1 ) = a, U(o2 ) = b and U(o3 ) = c (where a, b and c are
numbers). Then, since Susan prefers A to B, the expected utility of A must be greater than
the expected utility of B: 0.25b + 0.75c > 0.2a + 0.8c. This inequality is equivalent to
0.25b > 0.2a + 0.05c or, dividing both sides by 0.25, b > 0.8a + 0.2c. It follows from this
and Theorem 3.1.1 that Susan prefers D to C, because the expected utility of D is b and the
expected utility of C is 0.8a + 0.2c. Note that, in this example, we merely used the fact
that a von Neumann-Morgenstern utility function exists, even though we do not know what
the values of this function are.
Theorem 3.1.1 is an example of a “representation theorem” and is a generalization
of a similar result for the case of the ranking of a finite set of basic outcomes O. It is
not difficult to prove that if % is a complete and transitive ranking of O then there exists
a function U : O → R, called a utility function, such that, for any two basic outcomes
o, o0 ∈ O, U(o) ≥ U(o0 ) if and only if o % o0 . Now, it is quite possible that an individual
has a complete and transitive ranking of O, is fully aware of her ranking and yet she is not
able to answer the question “what is your utility function?”, perhaps because she has never
heard about utility functions. A utility function is a tool that we can use to represent her
ranking, nothing more than that. The same applies to von Neumann-Morgenstern rankings:
Theorem 3.1.1 tells us that if an individual has a von Neumann-Morgenstern ranking of
the set of lotteries L (O) then there exists a von Neumann-Morgenstern utility function
that we can use to represent her preferences, but it would not make sense for us to ask the
individual “what is your von Neumann-Morgenstern utility function?” (indeed this was a
question that could not even be conceived before von Neumann and Morgenstern stated
and proved Theorem 3.1.1 in 1944!).
Theorem 3.1.1 tells us that a von Neumann-Morgenstern utility function exists; the
next theorem can be used to actually construct such a function, by asking the individual to
answer a few questions, formulated in a way that is fully comprehensible to her (that is,
without using the word ‘utility’). The theorem says that, although there are many utility
functions that represent a given von Neumann-Morgenstern ranking, once you know one
3.1 Expected utility: theorems 59

function you “know them all”, in the sense that there is a simple operation that transforms
one function into the other.

Theorem 3.1.2 [von Neumann-Morgenstern, 1944].

Let % be a von Neumann-Morgenstern ranking of the set of basic lotteries L (O), where
O = {o1 , o2 , ..., om }. Then the following are true.
(A) If U : O → R is a von Neumann-Morgenstern utility function that represents %,
then, for any two real numbers a and b, with a > 0, the function V : O → R defined
by V (oi ) = aU(oi )+b (for every i = 1, . . . , m) is also a von Neumann-Morgenstern
utility function that represents %.
(B) If U : O → R and V : O → R are two von Neumann-Morgenstern utility functions
that represent %, then there exist two real numbers a and b, with a > 0, such that
V (oi ) = aU(oi ) + b (for every i = 1, . . . , m).

Proof. The proof of Part A of Theorem 3.1.2 is very simple. Let a and b be two numbers,
with a > 0. The hypothesis is that U : O → R is a von Neumann-Morgenstern ! utility
o1 ... om
function that represents %, that is, that, for any two lotteries L = and
p1 ... pm
!
o 1 ... om
L0 = ,
q1 ... qm

L % L0 if and only if p1U(o1 ) + ... + pmU(om ) ≥ q1U(o1 ) + ... + qmU(om ) (3.1)

Multiplying both sides of inequality (3.1) by a > 0 and adding (p1 + · · · + pm ) b to the
left-hand side and (q1 + · · · + qm ) b to the right-hand side1 we obtain

p1 [aU(o1 ) + b]+ ... + pm [aU(om ) + b] ≥ q1 [aU(o1 ) + b]+ ... + qm [aU(om ) + b] (3.2)

Defining V (oi ) = aU(oi ) + b, it follows from (3.1) and (3.2) that

L % L0 if and only if p1V (o1 ) + ... + pmV (om ) ≥ q1V (o1 ) + ... + qmV (om ),

that is, the function V is a von Neumann-Morgenstern utility function that represents the
ranking %. The proof of Part B will be given later, after introducing more notation and
some observations. 

1 Note that (p1 + · · · + pm ) = (q1 + · · · + qm ) = 1.

60 Chapter 3. Expected Utility Theory

Suppose that the DM has a von Neumann-Morgenstern ranking of the set of lotteries
L (O). Since among the lotteries there are the degenerate ones that assign probability 1 to
a single basic outcome, it follows that the DM has a complete and transitive ranking of
the basic outcomes. We shall write obest for a best basic outcome, that is, a basic outcome
which is at least as good as any other basic outcome (obest % o, for every o ∈ O) and oworst
for a worst basic outcome, that is, a basic outcome such that every other outcome is at least
as good as it (o % oworst , for every o ∈ O). Note that there may be several best outcomes
(then the DM would be indifferent among them) and several worst outcomes; then obest
will denote an arbitrary best outcome and oworst an arbitrary worst outcome. We shall
assume throughout that the DM is not indifferent among all the outcomes, that is, we shall
assume that obest  oworst .

We now show that, in virtue of Theorem 3.1.2, among the von Neumann-Morgenstern
utility functions that represent a given von Neumann-Morgenstern ranking % of L (O),
there is one that assigns the value 1 to the best basic outcome(s) and the value 0 to the worst
basic outcome(s). To see this, consider an arbitrary von Neumann-Morgenstern utility
function F : O → R that represents % and define G : O → R as follows: for every o ∈ O,
G(o) = F(o) − F(oworst ). Then, by Theorem 3.1.2 (with a = 1 and b = −F(oworst )), G
is also a utility function that represents % and, by construction, G(oworst ) = F(oworst ) −
F(oworst ) = 0; note also that, since obest  oworst , it follows that G(obest ) > 0. Finally,
G(o)
define U : O → R as follows: for every o ∈ O, U(o) = G(o ) . Then, by Theorem
best
3.1.2 (with a = G(o1 ) and b = 0), U is a utility function that represents % and, by
best
construction, U(oworst ) = 0 and U(obest ) = 1. For example, if there are six basic outcomes
and the ranking of the basic outcomes is o3 ∼ o6  o1  o4  o2 ∼ o5 , then one can
take as obest either o3 or o6 and as oworst either o2 or o5 ; furthermore, if F is given by
o1 o2 o3 o4 o5 o6 o1 o2 o3 o4 o5 o6
then G is the function and U is
2 −2 8 0 −2 8 4 0 10 2 0 10
o1 o2 o3 o4 o5 o6
the function .
0.4 0 1 0.2 0 1

Definition 3.1.2 Let U : O → R be a utility function that represents a given von

Neumann-Morgenstern ranking % of the set of lotteries L (O). We say that U is
normalized if U(oworst ) = 0 and U(obest ) = 1.

The transformations described above show how to normalize any given utility function.
Armed with the notion of a normalized utility function we can now complete the proof of
Theorem 3.1.2.
3.1 Expected utility: theorems 61

Proof of Part B of Theorem 3.1.2. Let F : O → R and G : O → R be two von Neumann-

Morgenstern utility functions that represent a given von Neumann-Morgenstern ranking of
L (O). Let U : O → R be the normalization of F and V : O → R be the normalization of G.
First we show that it must be that U = V , that is, U(o) = V (o) for every o ∈ O. Suppose,
by contradiction, that there is an ô ∈ O such that U(ô) 6= V (ô). Without loss of generality !
obest oworst
we can assume that U(ô) > V (ô). Construct the following lottery: L =
p̂ 1 − p̂
with p̂ = U(ô) (recall that U is normalized and thus takes on values in the interval from 0
to 1). Then E[U(L)] = E[V (L)] = U(ô). Hence, according to U it must be that ô ∼ L (this
follows from Theorem 3.1.1), while according to V it must be (again, by Theorem 3.1.1)
that L  ô (since E[V (L)] = U(ô) > V (ô)). Then U and V cannot be two representations
worst )
of the same ranking. Now let a1 = F(o )−F(o 1
worst )
and b1 = − F(o F(o
)−F(oworst ) . Note that
best best
a1 > 0. Then it is easy to verify that, for every o ∈ O, U(o) = a1 F(o) + b1 . Similarly
worst )
1
let a2 = G(o )−G(o worst )
and b2 = − G(o G(o
)−G(oworst ) ; again, a2 > 0 and, for every o ∈ O,
best best

V (o) = a2 G(o) + b2 . We can invert the latter transformation and obtain that, for every
o ∈ O, G(o) = Va(o)
2
− ba22 . Thus, we can transform F into U, which – as proved above – is
the same as V , and then transform V into G thus obtaining the following transformation of
F into G:
a1 b1 − b2
G(o) = aF(o) + b where a = > 0 and b = .
a2 a2


R Theorem 3.1.2 is often stated as follows: a utility function that represents a von
Neumann-Morgenstern ranking % of L (O) is unique up to a positive affine transfor-
mation. An affine transformation is a function f : R → R of the form f (x) = ax + b
with a, b ∈ R. The affine transformation is positive if a > 0.
Because of Theorem 3.1.2, a von Neumann-Morgenstern utility function is usually
referred to as a cardinal utility function.

Theorem 3.1.1 guarantees the existence of a utility function that represents a given
von Neumann-Morgenstern ranking % of L (O) and Theorem 3.1.2 characterizes the set
of such functions. Can one actually construct a utility function that represents a given
ranking? The answer is affirmative: if there are m basic outcomes one can construct an
individual’s von Neumann-Morgenstern utility function by asking her at most (m − 1)
questions. The first question is “what is your ranking of the basic outcomes?”. Then
we can construct the normalized utility function by first assigning the value 1 to the
best outcome(s) and the value 0 to the worst outcome(s). This leaves us with at most
(m − 2) values to determine. For this we appeal to one of the axioms discussed in the next
section, namely the Continuity Axiom, which says that, for every basic outcome oi there
62 Chapter 3. Expected Utility Theory

is a probability pi ∈ [0, 1] such that the DM is indifferent between oi for certain and the
lottery that gives a best outcome !with probability pi and a worst outcome with probability
obest oworst
(1 − pi ): oi ∼ . Thus, for each basic outcome oi for which a utility has
pi 1 − pi
not been determined yet,! we should ask the individual to tell us the value of pi such that
obest oworst
oi ∼ ; then we can set Ui (oi ) = pi , because the expected utility of the
pi 1 − pi
!
obest oworst
lottery is piUi (obest ) + (1 − pi )Ui (oworst ) = pi (1) + (1 − pi ) 0 = pi .
pi 1 − pi

 Example 3.1 Suppose that there are five basic outcomes, that is, O = {o1 , o2 , o3 , o4 , o5 }
and the DM, who has von Neumann-Morgenstern preferences, tells us that her ranking
of the basic outcomes is as follows: o2  o1 ∼ o5  o3 ∼ o4 . Then we can begin by
assigning utility 1 to the best outcome!o2 and utility 0 to the worst outcomes o3 and
outcome: o1 o2 o3 o4 o5
o4 : . There is only one value left to be determined,
utility: ? 1 0 0 ?
namely the utility of o1 (which is also the utility of o5 , since o1 ∼ o5 ). To find this
value, we ask the DM
! to tell us what value of p makes her indifferent between the lottery
o2 o3
L= and outcome o1 with certainty. Suppose that her answer is: 0.4. Then
p 1− p
her normalized von Neumann-Morgenstern utility function is
!
outcome: o1 o2 o3 o4 o5
.
utility: 0.4 1 0 0 0.4

Knowing this, we can predict her choice among any set of lotteries over these five basic
outcomes. 

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 3.3.2 at the end of this chapter.
 3.2 Expected utility: the axioms 63

3.2 Expected utility: the axioms

We can now turn to the list of rationality axioms proposed by von Neumann and Morgen-
stern. This section makes heavy use of mathematical notation and, as mentioned in the
previous section, if the reader is not interested in understanding in what sense the theory
of expected utility captures the notion of rationality, he/she can skip it without affecting
his/her ability to understand the rest of this book.
Let O = {o1 , o2 , . . . , om } be the set of basic outcomes and L (O) the set of simple
lotteries, that is, the set of probability distributions over O. Let % be a binary relation on
L (O). We say that % is a von Neumann-Morgenstern ranking of L (O) if it satisfies the
following four axioms or properties.

Axiom 1 [Completeness and transitivity]. % is complete (for every two lotteries L and
L0 either L % L0 or L0 % L or both) and transitive (for any three lotteries L1 , L2 and L3 , if
L1 % L2 and L2 % L3 then L1 % L3 ).

As noted in the previous section, Axiom 1 implies that there is a complete and transitive
ranking of the basic outcomes. Recall that obest denotes a best basic outcome and oworst
denotes a worst basic outcome and that we are assuming that obest  oworst , that is, that the
DM is not indifferent among all the basic outcomes.

! !
obest oworst obest oworst
Axiom 2 [Monotonicity]. % if and only if p ≥ q.
p 1− p q 1−q

! For every basic outcome oi there is a pi ∈ [0, 1] such that

Axiom 3 [Continuity].
obest oworst
oi ∼ .
pi 1 − pi

Before we introduce the last axiom we need to define a compound lottery.

!
x1 x2 ... xr
Definition 3.2.1 A compound lottery is a lottery of the form
p1 p2 ... pr
where each xi is either an element of O or an element of L (O).

!
o1 o2 o3 o4
For example, let m = 4. Then L = 2 is a simple lottery (an element of
5 0 15 25
64 Chapter 3. Expected Utility Theory

L (O)), while
 ! ! 
o1 o2 o3 o4 o1 o2 o3 o4
1 1 1 1 o1 1
0 15 53
 
C= 3 6 3 6 5
 

 1 1 1 
2 4 4

is a compound lottery.2 The compound lottery C can be viewed graphically as a tree, as

shown in Figure 3.1.

1 1
2 1 4
4

1 1 1 1 1 1 3
3 6 3 6 5 5 5

o1 o2 o3 o4 o1 o1 o3 o4

Figure 3.1: A compound lottery

Next we define the simple lottery L(C) corresponding to a compound lottery C. Before
introducing the formal definition, we shall explain in an example how to construct such
a simple lottery. Continuing with the example of the compound lottery C given above
and illustrated in Figure 3.1, first we replace a sequence of edges with a single edge and
associate with it the product of the probabilities along the sequence of edges, as shown in
Figure 3.2.

1 1 1 1 1 1 1 3
6 12 6 12 4 20 20 20

o1 o2 o3 o4 o1 o1 o3 o4

Figure 3.2: Simplification of Figure 3.1 obtained by merging paths into simple edges and
associating with the simple edges the products of the probabilities along the path.

   
o1 o2 o3 o4 o1 o2 o3 o4
2 With r = 3, x1 = 1 1 1 1 , x2 = o1 , x3 = 1 1 3 , p1 = 21 , p2 = 1
4 and
3 6 3 6 5 0 5 5
p3 = 41 .
3.2 Expected utility: the axioms 65

Then we add up the probabilities of each outcome,as shown in Figure3.3. Thus, the
o1 o2 o3 o4
simple lottery L(C) that corresponds to C is L(C) = 28 5 13 14 , namely the
60 60 60 60
lottery shown in Figure 3.3.

28 5 13 14
60 60 60 60

o1 o2 o3 o4

Figure 3.3: Simplification of Figure 3.2 obtained by adding, for each outcome, the proba-
bilities of that outcome.
 
x1 x2 ... xr
Definition 3.2.2 Given a compound lottery C = the correspond-
  p1 p2 ... pr
o1 o2 ... om
ing simple lottery L(C) = is defined as follows. First of all, for
q1 q2 ... qm
i = 1, . . . , m and j = 1, . . . , r, define


 1 if x j = oi
0 if x j = ok with k 6= i

oi (x j ) = 
o1 ... oi−1 oi oi+1 ... om
 si if x j =


s1 ... si−1 si si+1 ... sm
r
Then qi = ∑ p j oi (x j ).
j=1

Continuing the above example where

     
o1 o2 o3 o4 o1 o2 o3 o4
1 1 1 1 o1 1

3 6 3 6 5 0 15 35 
C= 
 1 1 1 
2 4 4
we have that
   
o1 o2 o3 o4 o1 o2 o3 o4
r = 3, x1 = 1 1 1 1 , x2 = o1 and x3 = 1 ,
3 6 3 6 5 0 15 35

so that
o1 (x1 ) = 13 , o1 (x2 ) = 1, and o1 (x3 ) = 1
5
66 Chapter 3. Expected Utility Theory

and thus q1 = 12 31 + 14 (1) + 14 1

28 1 1

1 1 1 5
5 = 60 . Similarly, q2 = 2 6 + 4 (0) + 4 (0) = 12 = 60 ,

q3 = 12 31 + 14 (0) + 14 51 = 13 and q4 = 12 16 + 14 (0) + 41 35 = 14

60 60 .

Axiom 4 [Independence or substitutability]. Consider an arbitrary basic

! outcome oi and
o1 ... oi−1 oi oi+1 ... om
an arbitrary simple lottery L = . If L̂ is a simple
p1 ... pi−1 pi pi+1 ... pm
lottery such that oi ∼ L̂, then L ∼ M where M is the simple lottery
! corresponding to the
o1 ... oi−1 L̂ oi+1 ... om
compound lottery C = obtained by replacing oi
p1 ... pi−1 pi pi+1 ... pm
with L̂ in L.

We can now prove the first theorem of the previous section.

Proof of Theorem 3.1.1. To simplify the notation, throughout this proof we will assume
that we have renumbered the basic outcomes in such a way that obest = o1 and oworst = o!
m.
o1 om
First of all, for every basic outcome oi , let ui ∈ [0, 1] be such that oi ∼ .
ui 1 − ui
The existence of such a value ui is guaranteed by the Continuity Axiom (Axiom 3); clearly
u1 = 1 and um = 0. Now consider an arbitrary lottery
!
o1 ... om
L1 = .
p1 ... pm

First we show that

 
o1 om
L1 ∼  m m  (3.3)
∑ pi ui 1 − ∑ pi ui
i=1 i=1

This is done through a repeated application of the Independence Axiom (Axiom 4), as
follows. Consider the compound lottery
 ! 
o1 om
 o1 o3 ... om 
C2 =  u2 1 − u2 
p1 p2 p3 ... pm
!
o1 om
obtained by replacing o2 in lottery L1 with the lottery that the DM
u2 1 − u2
3.2 Expected utility: the axioms 67

considers to be just as good as o2 . The simple lottery corresponding to C2 is

!
o1 o3 ... om−1 om
L2 = .
p1 + p2 u2 p3 ... pm−1 pm + p2 (1 − u2 )

Note that o2 is assigned probability 0 in L2 and thus we have omitted it. By Axiom 4,
L1 ∼ L2 . Now apply the same argument to L2 : let
 ! 
o1 om
o1 ... om−1 om
C3 =  u3 1 − u3
 

p1 + p2 u2 p3 ... pm−1 pm + p2 (1 − u2 )

whose corresponding simple lottery is

!
o1 ... om
L3 = .
p1 + p2 u2 + p3 u3 ... pm + p2 (1 − u2 ) + p3 (1 − u3 )

Note, again, that o3 is assigned probability zero in L3 . By Axiom 4, L2 ∼ L3 ; thus, by

transitivity (since L1 ∼ L2 and L2 ∼ L3 ) we have that L1 ∼ L3 . Repeating this argument
we get that L1 ∼ Lm−1 , where
!
o1 om
Lm−1 = .
p1 + p2 u2 + ... + pm−1 um−1 pm + p2 (1 − u2 ) + ... + pm−1 (1 − um−1 )

Since u1 = 1 (so that p1 u1 = p1 ) and um = 0 (so that pm um = 0),

m
p1 + p2 u2 + ... + pm−1 um−1 = ∑ pi ui
i=1

and
m m−1 m m−1
p2 (1 − u2 ) + ... + pm−1 (1 − um−1 ) + pm = ∑ pi − ∑ pi ui = p1 + ∑ pi − ∑ pi ui − p1
i=2 i=2 i=2 i=2

m m−1 m
= (since u1 =1 and um =0) ∑ pi − ∑ piui − p1u1 − pmum = 
m
since ∑ pi =1
 1 − ∑ pi ui .
i=1 i=2 i=1
i=1
68 Chapter 3. Expected Utility Theory
 

 o1 om 

Thus, Lm−1 = 

, which proves (3.3). Now define the following

 m m 
∑ pi ui 1 − ∑ pi ui
i=1 i=1
utility function U : {o1 , ..., om } → [0, 1]: U(oi ) = ui , where, as before, for every basic
 
 o1 om 
 
outcome oi , ui ∈ [0, 1] is such that oi ∼ 

. Consider two arbitrary lotteries

 
ui 1 − ui
   
 o1 ... om   o1 ... om 
 and L0 =  . We want to show that L % L0 if and only
   
L=
   
   
p1 ... pm q1 ... qm
m m
if E [U(L)] ≥ E [U(L0 )], that is, if and only if ∑ pi ui ≥ ∑ qi ui . By (3.3), L ∼ M, where
  i=1 i=1  
 o1 om   o1 om 
 and also L0 ∼ M 0 , where M 0 = 
   
M=
  
.

 m m   m m 
∑ pi ui 1 − ∑ pi ui ∑ qi ui 1 − ∑ qi ui
i=1 i=1 i=1 i=1
Thus, by transitivity of %, L % L0 if and only if M % M0; by the Monotonicity Axiom

m m
(Axiom 2), M % M 0 if and only if ∑ pi ui ≥ ∑ qi ui . 
i=1 i=1

The following example, known as the Allais paradox, suggests that one should view
expected utility theory as a “prescriptive” or “normative” theory (that is, as a theory
about how rational people should choose) rather than as a descriptive theory (that is,
as a theory about the actual behavior of individuals). In 1953 the French economist
Maurice Allais published a paper regarding a survey he had conducted in 1952 concerning
a hypothetical decision problem. Subjects “with good training in and knowledge of the
theory of probability, so that they could be considered to behave rationally” were asked to
rank the following pairs of lotteries:
! !
$5 Million $0 $1 Million $0
A= 89 11 versus B = 90 10
100 100 100 100

and
! !
$5 Million $1 Million $0 $1 Million
C= 89 10 1 versus D = .
100 100 100 1
3.2 Expected utility: the axioms 69

Most subjects reported the following ranking: A  B and D  C. Such ranking violates
the axioms of expected utility. To see this, let O = {o1 , o2 , o3 } with o1 = $5 Million,
o2 = $1 Million and o3 = $0. Let us assume that the individual in question prefers more
money to less, so that o1  o2  o3 and has a von Neumann-Morgenstern ranking ! of
$5 Million $0
the lotteries over L (O) . Let u2 ∈ (0, 1) be such that D ∼ (the
u2 1 − u2
existence of such u2 is guaranteed by the Continuity Axiom). Then, since D  C, by
transitivity
!
$5 Million $0
 C. (3.4)
u2 1 − u2

Let C0 be the simple lottery corresponding to the compound lottery

 ! 
$5 Million $0
 $5 Million $0 
 u2 1 − u2 .
89 10 1
100 100 100
!
$5 Million $0
Then C0 = 89 10 89 10

.
100 + 100 u2 1 − 100 + 100 u2
By the Independence Axiom, C ∼ C0 and thus, by (3.4) and transitivity,
! !
$5 Million $0 $5 Million $0
 89 10 89 10

.
u2 1 − u2 100 + 100 u2 1 − 100 + 100 u 2

89 10
Hence, by the Monotonicity Axiom, u2 > 100 + 100 u2 , that is,

89
u2 > 90 . (3.5)

Let B0 be the simple lottery corresponding to the following compound lottery, constructed
!
$5 Million $0
from B by replacing the basic outcome ‘$1 Million’ with :
u2 1 − u2
 ! 
$5 Million $0
$0 
u2 1 − u2 .


90 10
100 100

Then
!
$5 Million $0
B0 = 90 90 .
100 u2 1 − 100 u2

By the Independence Axiom, B ∼ B0 ; thus, since A  B, by transitivity, A  B0 and therefore,

70 Chapter 3. Expected Utility Theory
89 90 89
by the Monotonicity Axiom, 100 > 100 u2 , that is, u2 < 90 , contradicting (3.5).
Thus, if one finds the expected utility axioms compelling as axioms of rationality, then one
cannot consistently express a preference for A over B and also a preference for D over C.

Another well-known paradox is the Ellsberg paradox. Suppose that you are told that an
urn contains 30 red balls and 60 more balls that are either blue or yellow. You don’t know
how many blue or how many yellow balls there are, but the number of blue balls plus the
number of yellow ball equals 60 (they could be all blue or all yellow or any combination
of the two). The balls are well mixed so that each individual ball is as likely to be drawn as
any other. You are given a choice between bets A and B, where
A = you get $100 if you pick a red ball and nothing otherwise,
B = you get $100 if you pick a blue ball and nothing otherwise.
Many subjects in experiments state a strict preference for A over B: A  B. Consider now
the following bets:
C = you get $100 if you pick a red or yellow ball and nothing otherwise,
D = you get $100 if you pick a blue or yellow ball and nothing otherwise.
Do the axioms of expected utility constrain your ranking of C and D? Many subjects in
experiments state the following ranking: A  B and D % C. All such people violate the
axioms of expected utility. The fraction of red balls in the urn is 30 1
90 = 3 . Let p2 be the
fraction of blue balls and p3 the fraction of yellow balls (either of these can be zero: all
we know is that p2 + p3 = 60 2
90 = 3 ). Then A, B,C and D can be viewed as the following
lotteries:
! !
$100 $0 $100 $0
A= 1 , B=
3 p 2 + p3 p2 13 + p3
! !
$100 $0 $100 $0
C= 1 , D= 2 1
3 + p3 p2 p2 + p3 = 3 3

Let U be the normalized von Neumann-Morgenstern utility function that represents the
individual’s ranking; then U($100) = 1 and U(0) = 0. Thus,
E [U(A)] = 13 , E [U(B)] = p2 , E [U(C)] = 31 + p3 , and E [U(D)] = p2 + p3 = 23 .
Hence, A  B if and only if 13 > p2 , which implies that p3 > 31 , so that E [U(C)] = 13 + p3 >
E [U(D)] = 32 and thus C  D (similarly, B  A if and only if 13 < p2 , which implies that
E [U(C)] < E [U(D)] and thus D  C).

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 3.3.2 at the end of this chapter.
 3.3 Exercises 71

3.3 Exercises
The solutions to the following exercises are given in Section 3.4 at the end of this chapter.

3.3.1 Exercises for Section 3.1: Expected utility: theorems

Exercise 3.1 Ben is offered a choice between the following two money lotteries:
! !
$4, 000 $0 $3, 000
A= and B = .
0.8 0.2 1

He says he strictly prefers B to A. Which of the following two lotteries, C and D, will
Ben choose if he satisfies the axioms of expected utility and prefers more money to
less?
! !
$4, 000 $0 $3, 000 $0
C= , D= .
0.2 0.8 0.25 0.75


Exercise 3.2 There are three basic outcomes, o1 , o2 and o3 . Ann satisfies the axioms of
expected utility theory and her preferences over lotteries involving these three outcomes
can be represented by the following von Neumann-Morgenstern utility function:

V (o2 ) = a > V (o1 ) = b > V (o3 ) = c.

Normalize the utility function. 

Exercise 3.3 Consider the following lotteries:

! !
$3000 $500 $3000 $500
L1 = 5 1 , L2 = 2 1 ,
6 6 3 3
! !
$3000 $2000 $1000 $500 $2000 $1000
L3 = 1 1 1 1 , L4 = 1 1 .
4 4 4 4 2 2

Jennifer says that she is indifferent between lottery L1 and getting $2,000 for certain.
She is also indifferent between lottery L2 and getting $1,000 for certain. Finally, she
says that between L3 and L4 she would chose L3 . Is she rational according to the theory
of expected utility? [Assume that she prefers more money to less.] 
72 Chapter 3. Expected Utility Theory

Exercise 3.4 Consider the following basic outcomes:

• o1 = a Summer internship at the White House,
• o2 = a free one-week vacation in Europe,
• o3 = $800,
• o4 = a free ticket to a concert.
Rachel says that her ranking of these outcomes
! is o1  o2! o3  o4 . She also says
o2 o1 o4
that (1) she is indifferent between and 4 1 and (2) she is indifferent
1 5 5
! !
o3 o1 o4
between and 1 1 . If she satisfies the axioms of expected utility theory,
1 2 2 ! !
o1 o2 o3 o4 o1 o2 o3
which of the two lotteries L1 = 1 2 3 2 and L2 = 1 3 1 will
8 8 8 8 5 5 5
she choose? 

Exercise 3.5 Consider the following lotteries:

! !
$30 $28 $24 $18 $8 $30 $28 $8
L1 = 2 1 1 2 4 and L2 = 1 4 5 .
10 10 10 10 10 10 10 10

(a) Which lottery would a risk neutral person choose?

(b) Paul’s von Neumann-Morgenstern utility-of-money function is U($m) = ln(m),
where ln denotes the natural logarithm. Which lottery would Paul choose?


Exercise 3.6 There are five basic outcomes. Jane has a von Neumann-Morgenstern
ranking of the set of lotteries over basic outcomes
 that can be represented 
by either of
o1 o2 o3 o4 o5
the following utility functions U and V :  U : 44 170 −10 26 98 .
 

V : 32 95 5 23 59
(a) Show how to normalize each of U and V and verify that you get the same
normalized utility function.
(b) Show how to transform U into V with a positive affine transformation of the form
x 7→ ax + b with a, b ∈ R and a > 0.

 3.3 Exercises 73
! !
$28 $10 $50
Exercise 3.7 Consider the following lotteries: L3 = , L4 = 1 1 .
1 2 2
(a) Ann has the following von Neumann-Morgenstern utility function: UAnn ($m) =
√
m. How does she rank the two lotteries?
(b) Bob has the following von Neumann-Morgenstern utility function: UBob ($m) =
m4
2m − 100 3 . How does he rank the two lotteries?

(c) Verify that both Ann and Bob are risk averse, by determining what they would
choose between lottery L4 and its expected value for certain.


3.3.2 Exercises for Section 3.2: Expected utility: the axioms

Exercise 3.8 Let O = {o1 , o2 , o3 , o4 }. Find the simple lottery corresponding to the
following compound lottery
 ! ! ! 
o1 o2 o3 o4 o1 o3 o4 o2 o3
 2 1 3 1 o2 1 1 3 1 2 
5 10 10 5 5 5 5 3 3
 
 
 1 1 1 1 
8 4 8 2


Exercise 3.9 Let O = {o1 , o2 , o3 , o4 }. Suppose that the DM has a von Neumann-
Morgenstern ranking of L (O) and states the following indifference:
! !
o2 o4 o3 o4
o1 ∼ 1 3 and o2 ∼ 3 2 .
4 4 5 5

Find a lottery that the DM considers just as good as

!
o1 o2 o3 o4
L= 1 2 1 1 .
3 9 9 3

Do not add any information to what is given above (in particular, do not make any
assumptions about which outcome is best and which is worst). 
 74 Chapter 3. Expected Utility Theory

Exercise 3.10 — More difficult. Would you be willing to pay more in order to reduce
the probability of dying within the next hour from one sixth to zero or from four sixths
to three sixths? Unfortunately, this is not a hypothetical question: you accidentally
entered the office of a mad scientist and have been overpowered and tied to a chair. The
mad scientist has put six glasses in front of you, numbered 1 to 6, and tells you that
one of them contains a deadly poison and the other five contain a harmless liquid. He
says that he is going to roll a die and make you drink from the glass whose number
matches the number that shows up from the rolling of the die. You beg to be exempted
and he asks you “what is the largest amount of money that you would be willing to
pay to replace the glass containing the poison with one containing a harmless liquid?”.
Interpret this question as “what sum of money x makes you indifferent between
(1) leaving the poison in whichever glass contains it and rolling the die, and
(2) reducing your wealth by $x and rolling the die after the poison has been replaced by
a harmless liquid”. Your answer is: $X.
Then he asks you “suppose that instead of one glass with poison there had been four
glasses with poison (and two with a harmless liquid); what is the largest amount of
money that you would be willing to pay to replace one glass with poison with a glass
containing a harmless liquid (and thus roll the die with 3 glasses with poison and 3 with
a harmless liquid)?”. Your answer is: $Y .
Show that if X > Y then you do not satisfy the axioms of Expected Utility Theory.
[Hint: think about what the basic outcomes are; assume that you do not care about
how much money is left in your estate if you die and that, when alive, you prefer more
money to less.] 

3.4 Solutions to Exercises

Solution to Exercise 3.1 Since Ben prefers B to A, he must prefer D to C.
Proof. Let U be a von Neumann-Morgenstern utility function that represents Ben’s
preferences. Let U($4, 000) = a,U($3, 000) = b and U($0) = c. Since Ben prefers more
money to less, a > b > c. Then E[U(A)] = 0.8U($4, 000) + 0.2U($0) = 0.8a + 0.2c and
E[U(B)] = U($3, 000) = b. Since Ben prefers B to A, it must be that b > 0.8a + 0.2c. Let
us now compare C and D: E[U(C)] = 0.2 a + 0.8c and E[U(D)] = 0.25b + 0.75c. Since
b > 0.8a + 0.2c, 0.25b > 0.25(0.8a + 0.2c) = 0.2a + 0.05c and thus, adding 0.75c to both
sides, we get that 0.25b + 0.75c > 0.2a + 0.8c, that is, E[U(D)] > E[U(C)], so that D  C.
Note that the proof would have been somewhat easier if we had taken the normalized utility
function, so that a = 1 and c = 0. 
3.4 Solutions to Exercises 75
1
Solution to Exercise 3.2 Define the function U as follows: U(x) = a−c c
V (x)− a−c = V (x)−c
a−c
1
(note that, by hypothesis, a > c and thus a−c > 0). Then U represents the same preferences
as V . Then U(o2 ) = V (oa−c
2 )−c
= a−c V (o1 )−c
a−c = 1, U(o1 ) = a−c =
b−c
a−c , and U(o3 ) = V (oa−c
3 )−c
=
c−c b−c
a−c = 0. Note that, since a > b > c, 0 < a−c < 1. 

Solution to Exercise 3.3 We can take the set of basic outcomes to be {$3000, $2000, $1000,
$500}. Suppose that there is a von Neumann-Morgenstern utility function U that represents
Jennifer’s preferences. We can normalize it so that U($3000) = 1 and U($500) = 0. Since
Jennifer is indifferent between L1 and $2000, U($2000) = 56 (since the expected utility
of L1 is 56 (1) + 61 (0) = 56 ). Since she is indifferent between L2 and $1000, U($1000) = 23
(since the expected utility of L2 is 23 (1) + 31 (0) = 23 ). Thus, E[U(L3 )] = 14 (1) + 14 65 +

1 2

1 5 1 5

1 2
3 3 5
4 3 + 4 (0) = 8 and E[U(L 4 )] = 2 6 + 2 3 = 4 . Since 4 > 8 , Jennifer should prefer

L4 to L3 . Hence, she is not rational according to the theory of expected utility. 

Solution to Exercise 3.4 Normalize her utility function

! so that U(o ! 1 ) = 1 and U(o4 ) = 0.
o2 o1 o4
Then, since Rachel is indifferent between and 4 1 , we have that U(o2 ) =
1 5 5
! !
4 o3 o1 o4
5 . Similarly, since she is indifferent between and 1 1 , U(o3 ) = 12 . Then
1 2 2
!
o1 o2 o3 o4
the expected utility of L1 = 1 2 3 2 is 81 (1) + 28 ( 45 ) + 38 ( 12 ) + 28 (0) = 41
80 =
8 8 8 8 !
o1 o2 o3
0.5125, while the expected utility of L2 = 1 3 1 is 15 (1) + 35 ( 45 ) + 15 ( 12 ) =.
5 5 5
39
50 = 0.78. Hence, she prefers L2 to L1 . 

Solution to Exercise 3.5

2 1 1 2 4
(a) The expected value of L1 is 10 (30) + 10 (28) + 10 (24) + 10 (18) + 10 (8) = 18 and
1 4 5
the expected value of L2 is 10 (30) + 10 (28) + 10 8 = 18.2. Hence, a risk-neutral
person would prefer L2 to L1 .

1 1 1 1 2
(b) The expected utility of L1 is 5 ln(30) + 10 ln(28) + 10 ln(24) + 5 ln(18) + 5 ln(8) =
1 2 1
2.741 while the expected utility of L2 is 10 ln(30) + 5 ln(28) + 2 ln(8) = 2.713. Thus,

Paul would choose L1 (since he prefers L1 to L2 ). 

76 Chapter 3. Expected Utility Theory

Solution to Exercise 3.6

(a) To normalize U first add 10 to each value and then divide by 180. Denote the
normalization of U by Ū. Then

o1 o2 o3 o4 o5
54 180 0 36 108
Ū : 180 = 0.3 180 = 1 180 = 0 180 = 0.2 180 = 0.6

To normalize V first subtract 5 from each value and then divide by 90. Denote the
normalization of V by V̄ . Then

o1 o2 o3 o4 o5
27 90 0 18 54
V̄ : 90 = 0.3 90 =1 90 =0 90 = 0.2 90 = 0.6

(b) The transformation is of the form V (o) = aU(o) + b. To find the(values of a and
44a + b = 32
b plug in two sets of values and solve the system of equations .
170a + b = 95
The solution is a = 21 , b = 10. Thus, V (o) = 12 U(o) + 10. 

Solution to Exercise 3.7

√
(a) Ann prefers L3 to L4 (L3 Ann L4 ). In fact, E [UAnn (L3 )] = 28 = 5.2915 while
√ √
E [UAnn (L4 )] = 12 10 + 12 50 = 5.1167.

28 4
(b) Bob prefers L4 to L3 (L4 Bob L3 ). In fact, E [UBob (L3 )] = 2(28) − 1003 = 55.3853
h 4
i h 4
i
10
while E [UBob (L4 )] = 21 2(10) − 100 1 50
3 + 2 2(50) − 1003 = 56.87.

(c) The expected value of lottery L4 is 12 10 + 12 50 = 30; thus, a risk-averse person

would strictly prefer $30 with certainty to lottery L4 . We saw in Part (a) that for
√
Ann the expected utility of lottery L4 is 5.1167; the utility of $30 is 30 = 5.4772.
Thus, Ann would indeed choose $30 for certain over the lottery L4 . We saw in
Part (b) that for Bob the expected utility of lottery L4 is 56.87; the utility of $30
304
is 2(30) − 100 3 = 59.19 . Thus, Bob would indeed choose $30 for certain over the

lottery L4 . 

!
o1 o2 o3 o4
Solution to Exercise 3.8 The simple lottery is 18 103 95 24 . For example, the
240 240 240 240

1 1
+ 14 (1) + 18 (0) + 12 1 103



probability of o2 is computed as follows: 8 10 3 = 240 . 
3.4 Solutions to Exercises 77

Solution to Exercise 3.9 Using the stated indifference, use lottery L to construct the
compound lottery
 ! ! 
o2 o4 o3 o4
 1 3 3 2 o3 o4 
 4 4 5 5
,
 
1 2 1 1
3 9 9 3
!
o1 o2 o3 o4
whose corresponding simple lottery is L0 = 1 11 121 . Then, by the Indepen-
0 12 45 180

dence Axiom, L ∼ L0 . 

Solution to Exercise 3.10 Let W be your initial wealth. The basic outcomes are:
1. you do not pay any money, do not die and live to enjoy your wealth W (denote this
outcome by A0 ),
2. you pay $Y , do not die and live to enjoy your remaining wealth W − Y (call this
outcome AY ),
3. you pay $X, do not die and live to enjoy your remaining wealth W − X (call this
outcome AX ),
4. you die (call this outcome D); this could happen because (a) you do not pay any
money, roll the die and drink the poison or (b) you pay $Y , roll the die and drink the
poison; we assume that you are indifferent between these two outcomes.

Since, by hypothesis, X > Y , your ranking of these outcomes must be A0  AY  AX 

D. If you satisfy the von Neumann-Morgenstern axioms, then your preferences can be
represented by a von Neumann-Morgenstern utility function U defined on the set of basic
outcomes. We can normalize your utility function by setting U(A0 ) = 1 and U(D) = 0.
Furthermore, it must be that

U(AY ) > U(AX ). (3.6)

The maximum amount $P that you are willing to pay is that amount that makes you
indifferent between (1) rolling the die with the initial number of poisoned glasses and (2)
giving up $P and rolling the die with one less poisoned glass.

Thus – based on your answers – you are indifferent between the two lotteries
! !
D A0 AX
1 5 and
6 6 1
78 Chapter 3. Expected Utility Theory

and you are indifferent between the two lotteries:

! !
D A0 D AY
4 2 and 3 3 .
6 6 6 6

Thus,

1
U(D) + 56 U(A0 ) = U(AX ) and 4
U(D) + 26 U(A0 ) = 36 U(D) + 36 U(AY ) .
|6 {z } |6 {z } | {z }
1 5 5 4 2 2 3 3
= 6 0+ 6 1= 6 = 6 0+ 6 1= 6 = 6 0+ 6 U(AY )

5 2
Hence, U(AX ) = 6 and U(AY ) = 3 = 46 , so that U(AX ) > U(AY ), contradicting (3.6). 
 4. Money lotteries revisited

4.1 von Neumann Morgenstern preferences over money lotteries

In this section we revisit the notions of risk aversion/neutrality/love in the context of von
Neumann-Morgenstern preferences. From now on we will use the abbreviation “vNM” for
“von Neumann Morgenstern” and, unless explicitly stated otherwise, we will assume that
the individual in question has vNM preferences.

4.1.1 The vNM utility-of-money function of a risk-neutral agent

 
$m1 $m2 ... $mn
Recall from Chapter 2 (Section 2.2) that, given a money lottery L = ,
p1 p2 ... pn
an individual is said to be risk neutral
 relative to L if she is indifferent between L and the
$E[L]
expected value of L for sure: L ∼ . If the individual has vNM preferences over
1
money lotteries then, by Theorem 3.1.1 (Chapter 3), there exists a utility function U (that
assigns a real number to each sums of money) such that
 
$E[L]
L∼ if and only if E[U(L)] = U (E[L]) , that is, if and only if,
1
| {z }
L is as good as E[L]
  (4.1)

p U(m1 ) + ... + pnU(mn ) = U

|p1 m1 + {z
... + pn mn  .

|1 {z } }
E[U(L)] E[L]

It is clear that one utility function that satisfies (4.1) is the identity function U(m) = m. In
fact, when U is the identity function

p1U(m1 ) + ... + pnU(mn ) = p1 m1 + ... + pn mn = U (p1 m1 + ... + pn mn ) . (4.2)

 80 Chapter 4. Money lotteries revisited

R The utility-of-money function U(m) = m represents the vNM preferences of an

individual who is risk neutral relative to all money lotteries, or risk neutral for short.
Hence, by Theorem 3.1.1 (Chapter 3), any function of the form U(m) = am + b
with a > 0 is an alternative vNM utility function representing the preferences of a
risk-neutral person.

In Chapter 3 we assumed that the set of possible monetary outcomes (over which lotteries
were defined) was finite. If we allow for any non-negative amount of money then the vNM
utility-of-money function of a risk-neutral individual is represented by the 45o line, as
shown in Figure 4.1.

utility
U

45o line

0 m
money

Figure 4.1: The identity function U(m) = m represents the vNM preferences of a risk-
neutral individual.

4.1.2 Concavity and risk aversion

 
$m1 $m2 ... $mn
In Chapter 2 we said that, given a non-degenerate money lottery L = ,
p1 p2 ... pn
an individual is defined to be risk averse relative to L if she prefers the expected value of
L for sure to the lottery: E[L]  L.1 If the individual has vNM preferences over money
lotteries then, by Theorem 3.1.1 (Chapter 3), there exists a utility-of-money function U
 
1 It $E[L]
would be more precise to write  L instead of E[L]  L, but from now on we shall use
1
the latter, simpler, notation.
4.1 von Neumann Morgenstern preferences over money lotteries 81

such that
E[L]  L if and only if U (E[L]) > E[U(L)], that is, if and only if,
 
(4.3)
U  p1 m1 + ... + pn mn 

| {z }

> p1U(m1 ) + ... + pnU(mn ) .
| {z }
E[L] E[U(L)]

The inequality in (4.3) is shown graphically in Figure 4.2 when n = 2, that is, with reference
 
$m1 $m2
to the lottery L = with m1 < m2 and 0 < p < 1.
p 1− p

utility
U

U(m2 )

U (E[L]) = U (pm1 + (1 − p)m2 )

E[U(L)] = pU(m1 ) + (1 − p)U(m2 )

U(m1 )

 
$m1 $m2
L=
p 1− p 0 m1 m2 m
money
pm1 + (1 − p)m2
| {z }
E[L]

Figure 4.2: A graphical representation of the inequality U (E[L]) > E[U(L)], that is
U (pm1 + (1 − p)m2 ) > pU(m1 ) + (1 − p)U(m2 ).

In Figure 4.2, the point pm1 + (1 − p)m2 is a point on the horizontal axis between m1
and m2 (the closer to m1 , the closer p is to 1). Since the individual is assumed to prefer
more money to less, U(m1 ) < U(m2 ). The point pU(m1 ) + (1 − p)U(m2 ) is a point on the
vertical axis between U(m1 ) and U(m2 ). To find that point, draw a straight-line segment
from the point (m1 ,U(m1 )) to the point (m2 ,U(m2 )) (the dashed line in Figure 4.2) and
go vertically up from the point pm1 + (1 − p)m2 on the horizontal axis to the dashed line
and from there horizontally to the vertical axis. By (4.3) this point must be below the point
U (pm1 + (1 − p)m2 ).
82 Chapter 4. Money lotteries revisited

It follows from the above discussion that, if we draw a continuous vNM utility-of-
money function for a risk-averse individual, the corresponding curve must lie above the
straight-line segment joining any two points on the graph, as shown in Figure 4.3. This
property is know as strict concavity.

utility
U

U(m2 )

U (pm1 + (1 − p)m2 )
pU(m1 ) + (1 − p)U(m2 )

U(m1 )

0 m1 m2 m
money
pm1 + (1 − p)m2

Figure 4.3: The utility function of a risk-averse individual is strictly concave.

Definition 4.1.1 A function f : R+ → R (where R+ denotes the set of non-negative

real numbers) is stricly concave if, for every x, y ∈ R and for every p ∈ (0, 1),

f (px + (1 − p)y) > p f (x) + (1 − p) f (y).

R The vNM utility-of-money function of a risk-averse individual (that is, of an individual

who is risk averse relative to every non-degenerate money lottery) is strictly concave.
 4.1 von Neumann Morgenstern preferences over money lotteries 83

4.1.3 Convexity and risk loving

 
$m1 $m2 ... $mn
Given a non-degenerate money lottery L = , a risk-loving indi-
p1 p2 ... pn
vidual prefers the lottery L to its expected value for sure: L  E[L]. If the individual has
vNM preferences over money lotteries then, by Theorem 3.1.1 (Chapter 3), there exists a
utility-of-money function U such that
L  E[L] if and only if E[U(L)] > U (E[L]) , that is, if and only if,
 
(4.4)
p U(m1 ) + ... + pnU(mn ) > U ... + pn mn  .
|p1 m1 + {z

|1 {z } }
E[U(L)] E[L]

An argument similar to the one used in the previous section leads to the conclusion that,
if we draw a continuous vNM utility-of-money function for a risk-loving individual, the
graph must lie below the straight-line segment joining any two points on the graph, as
shown in Figure 4.4. This property is know as strict convexity.

utility
U

U(m2 )

E[U(L)] = pU(m1 ) + (1 − p)U(m2 )

U (E[L]) = U (pm1 + (1 − p)m2 )

U(m1 )

0 m1 m2 m
money
pm1 + (1 − p)m2

Figure 4.4: The utility function of a risk-averse individual is strictly concave.

Definition 4.1.2 A function f : R+ → R is strictly convex if, for every x, y ∈ R and for
every p ∈ (0, 1),
f (px + (1 − p)y) < p f (x) + (1 − p) f (y).

R The vNM utility-of-money function of a risk-loving individual (that is, of an individual

who is risk loving relative to every non-degenerate money lottery) is strictly convex.
 84 Chapter 4. Money lotteries revisited

4.1.4 Mixtures of risk attitudes

While we will tend to concentrate on risk neutrality and risk aversion – and we will typically
assume that an individual is risk-neutral or risk-averse relative to every non-degenerate
money lottery – it is possible for people to display different attitudes to risk for different
money lotteries. Consider, for example, an individual whose vNM utility-of-money
function is as shown in Figure 4.5. This individual displays risk love for money lotteries
that involve small sums of money, risk neutrality for lotteries involving intermediate sums
of money and risk aversion for lotteries involving “big stakes”: the function is strictly
convex for values of m between 0 and m1 , a straight line for values of m between m1 and
m2 and strictly concave for values of m larger than m2 . Such an individual might be willing
to buy a Powerball lottery ticket for $1 (thus displaying risk love) while at the same time
purchasing fire insurance for her house worth $400,000 (thus displaying risk aversion).

utility
U

0 m1 m2 m
money

Figure 4.5: The utility function of an individual with different attitudes to risk for different
lotteries.
 4.1 von Neumann Morgenstern preferences over money lotteries 85

4.1.5 Attitude to risk and the second derivative of the utility function
If the vNM utility-of-money function of an individual is a smooth function (or, at least,
twice differentiable) then we can relate the attitude to risk (that is, the shape of the graph
of the function) to the second-derivative of the utility function, using the following result
from calculus.
Let f : R+ → R be a twice differentiable function, then

d2 f
• f is strictly concave if and only if (x) < 0, for every x ∈ R+ ,
dx2

d2 f
• f is strictly convex if and only if (x) > 0, for every x ∈ R+ ,
dx2

d2 f
• the graph of f is a straight line if and only if (x) = 0, for every x ∈ R+ .
dx2

√
For example, an individual whose vNM utility-of-money function is U(m) = m is risk
1 d 2U √1
averse, since dU
dm = 2 m and thus dm2 = − 4 m3 which is negative for every m > 0. Figure
√
4.6 shows the graph of this function.

utility
√
U(m) = m

0 m
money

√
Figure 4.6: The graph of the utility function U(m) = m.
 86 Chapter 4. Money lotteries revisited
2
On the other hand, an individual whose vNM utility-of-money function is U(m) = m8
d 2U 1
is risk loving, since dU m
dm = 4 and thus dm2 = 4 > 0 . Figure 4.7 shows the graph of this
function.

utility
2
U(m) = m8

0 m
money

m2
Figure 4.7: The graph of the utility function U(m) = 8 .

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 4.5.1 at the end of this chapter.

4.2 Measures of risk aversion

Let us now focus on the case of risk aversion. Is it possible to measure the degree of risk
aversion of an individual?
One possible measure of risk aversion is provided by the notion of risk premium defined
in Chapter 2 (Section2.3), which we can now re-write using the notion of expected utility.
 an individual, whoseinitial wealth is $W ≥ 0, and a non-degenerate money lot-
Consider
$x1 $x2 ... $xn
tery M = (if xi < 0 then we assume that |xi | ≤ W , that is, if xi rep-
p1 p2 ... pn
resents a loss then we assume that the loss is not larger than the initial wealth). As usual, let
E[M] = p1 x1 + · · · + pn xn be the expected value of M. In terms of wealth levels, lottery M
$(W + x1 ) $(W + x2 ) ... $(W + xn )
corresponds to the non-degenerate lottery L =
p1 p2 ... pn
4.2 Measures of risk aversion 87

(given our assumption about the size of potential losses, W + xi ≥ 0, for every i = 1, . . . , n),
whose expected value is
E[L] = W + E[M].
If the individual has vNM preferences represented by the utility-of-money function U(m),
the expected utility of L is

E[U(L)] = p1U(W + x1 ) + · · · + pnU(W + xn ).

The risk premium associated with lottery L and utility function U is that amount of
money RLU such that the agent is indifferent between lottery L and the sum of money
$(E[L] − RLU ) for sure:
U (E[L] − RLU ) = E[U(L)].
Thus RLU is the maximum amount that an individual with vNM utility function U is willing
to forego to exchange the risky prospect L for a non-risky (i.e. certain) one with the same
expected value.
For example, consider the money lottery
 
$11 $20
M= 3 2
5 5

whose expected value is E[M] = 35 11 + 25 20 = $14.6, and an individual whose initial wealth
is W = $5√and whose vNM preferences can be represented by the utility-of-money function
U(m) = m. Then, in terms of wealth levels, the individual is facing the lottery
 
$16 $25
L= 3 2
5 5

whose expected√value is√E[L] = $(14.6 + 5) = $19.6; the expected utility of lottery L

is E[U(L)] = 35 16 + 25 25 = 4.4. Thus, √ for this lottery and this individual, the risk
premium is the solution to the equation 19.6 − RLU = 4.4, which is RLU = $0.24.
On the other hand, if the individual’s initial wealth is $110 then lottery M corresponds
to the wealth lottery  
0 $121 $130
L = 3 2
5 5

whose expected value√ is E[L√0 ] = $(14.6 + 110) = $124.6; the expected utility of lottery
L0 is E[U(L0 )] = 53 121 + 25 130 = 11.1607. Hence, √ for this lottery and this individual,
the risk premium is the solution to the equation 124.6 − RL0U = 11.1607, which is
RL0U = $0.0388.
Thus – as measured by the risk premium– the degree of risk aversion (incorporated
in a given vNM utility function U) towards a given money lottery M, varies with the
individual’s initial wealth. In the above example, when the individual’s initial wealth is
only $5, she is prepared to avoid the risky prospect M by reducing the expected value of
the corresponding wealth lottery by an amount of up to 24 cents, but if her initial wealth is
$110 then she is only prepared to reduce the expected value of the corresponding wealth
lottery by up to 4 cents.
88 Chapter 4. Money lotteries revisited

R From now on we will use the expression wealth lottery to refer to a lottery whose
outcomes are levels of wealth for the individual (such as lotteries L and L0 above,
constructed from the money lottery M by adding the individual’s initial wealth to
every outcome in M).
The reader should try to prove the following: suppose that, given a wealth lottery L
and a vNM utility function U, the risk premium is $r; then the risk premium remains
$r if the utility-of-money function U is replaced by a positive affine transformation
V of U, that is, for every m, V (m) = aU(m) + b, with a > 0. In other words, if, for
every m, V (m) = aU(m) + b, with a > 0, then RLV = RLU .

Instead of comparing, for a fixed utility function, the risk premium of a given money
lottery across different levels of initial wealth, we can also compare the risk premium,
for a fixed wealth lottery, for different utility functions (representing different prefer-
ences, hence different individuals).
 Figure 4.8 shows the risk premium for the wealth
$m1 $m2
lottery L = and two different utility functions, U and V . Let RLU be the
p 1− p
risk premium associated with U and RLV be the risk premium associated with V ; then
RLV = [pm1 + (1 − p)m2 ] − m̂V > RLU = [pm1 + (1 − p)m2 ] − m̂U , where m̂V is the cer-
tainty equivalent of lottery L for V and m̂U is the certainty equivalent of lottery L for U
(see Chapter 2, Section 2.3 for the notion of certainty equivalent).

utility

U(m2 ) = V (m2 )
V U
E[U(L)] = E[V (L)]
= U(m̂U ) = V (m̂V )
risk premium for V

U(m1 ) = V (m1 ) risk premium for U

wealth
m1 m̂V m̂U m2

pm1 + (1 − p)m2
| {z }
E[L]

Figure 4.8: The graphs of two utility functions U and V and the risk premia corresponding
to the wealth lottery that gives $m1 with probability p and $m2 with probability (1 − p).
4.2 Measures of risk aversion 89

Let us now focus on the issue of comparing different utility functions in terms of the
the extent to which they express risk aversion. Using the notion of risk premium, one
possibility is given in the following definition.

Definition 4.2.1 Let U and V be two concave vNM utility-of-money functions. We

say that V incorporates more risk aversion than U if, for every non-degenerate wealth
lottery L, RLV ≥ RLU (with strict inequality for at least one lottery).

Since we identified risk aversion with concavity of the vNM utility function, it seems
that the more concave the utility function, the more risk averse the individual is. This
intuition is confirmed in Figure 4.8 (on page 88): V is more concave than U and indeed it
incorporates more risk aversion, as measured by the size of the risk premium.
Since we identified concavity with negativity of the second derivative of the utility function,
one might be tempted to conclude that an individual with vNM utility-of-money function V
is more risk-averse than an individual with vNM utility-of-money function U if, in absolute
d 2V d 2U
value, the second derivative of V is larger than the second derivative of U: dm 2 > dm2 .

Unfortunately, this is not correct, because it violates the requirement that, if V is a positive
affine transformation of U, then V and U represent the same preferences and thus the same
degree of risk aversion (as was pointed out in the remark on page 88). For example, let
d 2V d 2U d 2U
V (m) = 2U(m), for every m ≥ 0. Then, for every m, dm 2 (m) = 2 dm2 (m) > dm2 (m)

and yet V and U represent the same preferences.

An expression involving the second derivative of the utility function is the following,
which is known as the Arrow-Pratt measure of absolute risk aversion, denoted by AU (m):2

U 00 (m)
AU (m) = − .
U 0 (m)

The minus sign makes this expression positive (since U 0 (m) > 0, because the individual is
assumed to prefer more money to less, and U 00 (m) < 0, since the individual is assumed to
be risk averse).
Note that the Arrow-Pratt measure of absolute risk aversion is a local measure, since it
varies with the amount of money considered; that is, typically, if m1 6= m2 then AU (m1 ) 6=
AU (m2 ).
Let us verify that the Arrow-Pratt measure of risk aversion is invariant to affine
transformations. Let a and b be real numbers, with a > 0, and let V (m) = aU(m) + b, for
(m) 00
(m) U (m) 00 00
every m ≥ 0. Then V 0 (m) = aU 0 (m) and V 00 (m) = aU 00 (m), so that VV 0 (m) = aU
aU 0 (m) = U 0 (m)
and thus AV (m) = AU (m).
Using the Arrow-Pratt measure of risk aversion we can introduce a second definition
of “more risk averse”.

2 We denote the first derivative of U interchangeably by either U 0 (m) or dU

dm (m) and the second derivative
d 2U
interchangeably by either U 00 (m) or dm2
(m)
90 Chapter 4. Money lotteries revisited
Definition 4.2.2 Let U and V be two concave vNM utility-of-money functions. We
say that V incorporates more risk aversion than U if, for every level of wealth m > 0,
AV (m) ≥ AU (m) (with strict inequality for at least one m).

√
For example, according to Definition 4.2.2, which of m and ln(m) incorporates
greater risk aversion? Let us compute the Arrow-Pratt measure of risk aversion for these
two functions.
d √ d2 √
Since dm m = − 2√1 m and dm 2 m = − √1 3 , A√ (m) = 2m 1
.
4 m
1 d2
On the other hand, since d
dm ln(m) = m and dm2
ln(m) = − m12 , Aln (m) = m1 .
1 1
Thus, since, for every m > 0, m > 2m we have that, for every m > 0, Aln (m) > A√ (m)
and thus, according to Definition√ 4.2.2, the utility function ln(m) incorporates more risk
aversion than the utility function m.

Note that, in general, there may be utility functions U and V that cannot be ranked
according to Definition 4.2.2. For example, it may be that case AU (m) > AV (m) for values
of m in some interval and AU (m) < AV (m) for values of m in some other interval: see
Exercise 4.11.

Yet a third definition of “more risk averse” relies on the intuition that “more concave”
means “more risk averse”:

Definition 4.2.3 Let U and V be two concave vNM utility-of-money functions. We say
that V incorporates more risk aversion than U if there exists a strictly increasing and
concave function f : R → R such that, for every m ≥ 0, V (m) = f (U(m)). In this case
we say that V is a concave transformation of U.

√
increasing, concave function, V (m) = ln ( m)
For example, since ln(x) is a strictly √
is a concave transformation of U(m) = m and thus, according to Definition 4.2.3, V
incorporates more risk aversion than U.

Of course, having three different definitions of greater risk aversion is rather confusing:
which of the three is the “correct” definition? Furthermore, while the condition in Definition
4.2.2 is somewhat easier to verify, the condition in Definition 4.2.1 is not very practical,
since it would require considering all possible wealth lotteries, and the condition in
Definition 4.2.3 is also hard to verify: how can one tell if one function is a concave
transformation of another? Luckily, it turns out that the three definitions are in fact
equivalent. The following theorem was proved by John Pratt in 1964.3

3 John W. Pratt, Risk aversion in the small and in the large, Econometrica, Vol. 32, No. 1/2, 1964, pp.
122-136.
4.2 Measures of risk aversion 91

Theorem 4.2.1 Let U(m) and V (m) be two functions. Then the following conditions
are equivalent:
1. RLV ≥ RLU , for every non-degenerate wealth lottery L.
2. AV (m) ≥ AU (m), for every m.
3. There exists a strictly increasing and concave function f : R → R such that
V (m) = f (U(m)), for every m ≥ 0.

The Arrow-Pratt measure of absolute risk aversion is not invariant to a change in

the units √
of measurement. For example, if the agent’s vNM utility-of-money function is
U(m) = m, where m is wealth measured in dollars, then her Arrow-Pratt measure of
1
absolute risk aversion is, as we saw above, A√ (m) = 2m ; for example,when the agent’s
1 1
wealth is $10, her Arrow-Pratt measure of absolute risk aversion is A√ (10) = 2(10) = 20 =
0.05. Suppose now that we want to change our units of measurement from dollars to cents.
√
The utility function then would be written as V (y) = y where y is wealth measured in cents
1
(thus y = 100m) and her Arrow-Pratt measure of absolute risk aversion is A√ (y) = 2y ; so
that when y = 1, 000 cents, that is, $10, her Arrow-Pratt measure of absolute risk aversion
1 1
is A√ (1, 000) = 2(1,000) = 2,000 = 0.0005: a different number, despite the fact that we are
looking at the same preferences and the same wealth.
A related measure of risk aversion, which is immune from this problem (that is, is
invariant to changes in units of measurement) is the Arrow-Pratt measure of relative risk
aversion, denoted by rU (m):
U 00 (m)
rU (m) = −m 0 .
U (m)
Thus rU (m) = mAU (m).
While the Arrow-Pratt measure of absolute risk aversion measures the rate at which
marginal utility (that is, the first derivative of the utility function) decreases when wealth is
increased by one monetary unit (e.g. $1), the Arrow-Pratt measure of relative risk aversion
measures the rate at which marginal utility decreases when wealth is increased by 1%.4

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 4.5.2 at the end of this chapter.

4 In other words, rU (m) is the absolute value of the elasticity of marginal utility, U 0 (m), with respect to m.
 92 Chapter 4. Money lotteries revisited

4.3 Some noteworthy utility functions

In the previous section we considered some specific utility-of-money functions.
√
For the square root function m we found that the Arrow-Pratt measure of absolute risk
1
aversion is A√ (m) = 2m , which is decreasing in m. Thus an individual with this vNM
utility function displays less and less risk aversion as her wealth increases.5
The natural logarithm function ln(m) is similar: the Arrow-Pratt measure of absolute risk
aversion is also decreasing in m: Aln (m) = m1 .6
Consider now the following utility-of-money function, whose graph is shown in Figure
4.9:
1
U(m) = 1 − e−m = 1 − m .
e

utility
U(m) = 1 − e−m

0 m
money

Figure 4.9: The graphs of the utility function U(m) = 1 − e−m .

−m ) = e−m d2
Since d
dm (1 − e and dm2
(1 − e−m ) = −e−m it follows that the Arrow-Pratt
measure of absolute risk aversion for this function is a constant:

−e−m
A(m) = − = 1.
e−m
5 On the other hand, the Arrow-Pratt measure of relative risk aversion is constant: r√ (m) = mA√ (m) =
1
m 2m = 12 .
6 And the Arrow-Pratt measure of relative risk aversion is constant: rln (m) = mAln (m) = m m1 = 1.
 4.4 Higher risk 93

Indeed, this is a special case of the class of CARA (Constant Absolute Risk Aversion)
utility functions, which is the class of functions of the form

1
U(m) = 1 − e−λ m = 1 −
eλ m
   
d2
where λ is a positive constant. In fact, d
dm 1 − e−λ m = λ e−λ m and dm2
1 − e−λ m =
−λ 2 e−λ m so that the Arrow-Pratt measure of absolute risk aversion is equal to λ .
We saw above that the utility function ln(m) is characterized by constant relative risk
aversion. That is not the only function with this property. The class of CRRA (Constant
Relative Risk Aversion) utility functions contains, besides the natural logarithm function,7
the functions of the form

m(1−λ )
U(m) = with λ > 0, λ 6= 1.
1−λ
λ
For these functions the Arrow-Pratt measure of absolute risk aversion is AU (m) = m so
that the Arrow-Pratt measure of relative risk aversion is rU (m) = mAU (m) = λ .

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 4.5.3 at the end of this chapter.

4.4 Higher risk

In the previous section we answered the following questions:
- For a given money lottery M involving changes in wealth, how does a risk-averse
individual view the corresponding wealth lottery at different levels of initial wealth?
Typically (but not necessarily), individuals display less risk aversion as their initial
wealth increases.
- How can we determine if one individual, whose utility-of-money function is U(m),
is more or less risk averse than another individual, whose utility-of-money function
is V (m)? We considered three alternative definitions of “more risk averse” and saw
that they are equivalent.
In this section we ask a different question, namely: when can we say that a money
lottery, L, is “more risky” than another money lottery, M?
In order to address this issue we fix a set of non-negative monetary prizes, $m1 , . . . , $mn
with the convention that they are ordered from smallest to largest, that is,

0 ≤ m1 < m2 < · · · < mn .

A lottery over {m1 , m2 , . . . , mn } coincides with a probability distribution over this set.
7 And, more generally, the logarithmic functions loga (m) with a > 1.
 94 Chapter 4. Money lotteries revisited

Let L be one such lottery, whose probabilities are {p1 , p2 , . . . , pn } (pi is the probability of
prize $mi , for every i = 1, 2, . . . , n)8 and let M be another lottery, whose probabilities are
{q1 , q2 , . . . , qn }:
   
$m1 $m2 . . . $mn $m1 $m2 . . . $mn
L= and M= .
p1 p2 . . . pn q1 q2 . . . qn

R Note that we allow for the possibility that some of the pi ’s and qi ’s are zero.

We shall denote by P : {m1 , m2 , . . . , mn } → [0, 1] the cumulative distribution corre-

sponding to the distribution {p1 , p2 , . . . , pn } and by Q : {m1 , m2 , . . . , mn } → [0, 1] the
cumulative distribution corresponding to the distribution {q1 , q2 , . . . , qn }, that is, for every
i = 1, 2, . . . , n (denoting P(mi ) by Pi and Q(mi ) by Qi )

Pi = p1 + · · · + pi and Qi = q1 + · · · + qi

(clearly, P1 = p1 , Q1 = q1 and Pn = Qn = 1). For example, if

 
$12 $26 $40 $58 $80 $96
L= 1 7 4 8
20 20 0 20 0 20

then the corresponding cumulative distribution is as follows:

$12 $26 $40 $58 $80 $96

1 8 8 12 12 20
cumulative P : 20 20 20 20 20 20

4.4.1 First-order stochastic dominance

The following definition captures one (rather obvious) way in which a lottery M can be
viewed as unambiguously better than another lottery L.

Definition 4.4.1 Given two lotteries

   
$m1 $m2 . . . $mn $m1 $m2 . . . $mn
L= and M=
p1 p2 . . . pn q1 q2 . . . qn

we say that L first-order stochastically dominates M (and write L >FSD M) if

Pi ≤ Qi for ever i = 1, 2, . . . , n, with at least one strict inequality.

8 Thus p is a function p : {m1 , . . . , mn } → [0, 1] and, for every i = 1, . . . , m, we denote p(mi ) by pi .
 4.4 Higher risk 95

For example,
   
$26 $40 $58 $80 $96 $26 $40 $58 $80 $96
L= 7 3 1 9 >FSD M = 7 4 9
20 0 20 20 20 20 0 20 0 20

as can be seen from the cumulative distributions:

$26 $40 $58 $80 $96

7 7 10 11
cumulative for L, P: 20 20 20 20 1
7 7 11 11
cumulative for M, Q: 20 20 20 20 1

It should be clear that if lottery L first-order stochastically dominates lottery M, then

L assigns higher probabilities to higher prizes relative to M. It follows that the expected
value of L is greater than the expected value of M: E[L] > E[M]; thus a risk-neutral person
prefers L to M. However, the same is true for any attitude to risk.9

Theorem 4.4.1 Let L and M be two money lotteries (over the same set of prizes). Then

L >FSD M

if and only if

E[U(L)] > E[U(M)], for every strictly increasing utility function U.

4.4.2 Mean preserving spread and second-order stochastic dominance

The notion of first-order stochastic dominance does not really capture the fact that a money
lottery is less risky than another one; it captures a different notion, namely that of a money
lottery being unambiguously better than another one. On the other hand, the notion of
second-order stochastic dominance does capture the property of being unambiguously less
risky. Second-order stochastic dominance is based on the notion of a “mean preserving
spread”.
Intuitively, a mean preserving spread of a probability distribution is an operation that
takes probability from a point and moves it to each side of that point in such a way that the
expected value remains the same.

9 The theorem can be proved using Abel’s Lemma, which says that if a1 , . . . , an and b1 , . . . , bn are real
n n−1
numbers, then, letting Ai = a1 + · · · + ai and Bi = b1 + · · · + bi , ∑ ai bi = ∑ Ai (bi − bi+1 ) + An bn (see
i=1 i=1
https://planetmath.org/abelslemma ). To prove Theorem 4.4.1 using Abel’s Lemma, let ai = qi − pi
and bi = U(mi ).
96 Chapter 4. Money lotteries revisited

Let L be the following money lottery, whose expected value is E[L] = 5:

 
$2 $3 $4 $5 $9
L= 1 .
3 0 13 0 13

Now let us construct a new lottery M by taking the probability assigned to the prize $4,
namely 13 , and spreading it equally between the prizes $3 and $5, as shown inn Figure 4.10.
It is easy to check that the expected value of M is the same as the expected value of L,
namely 5.

$2 $3 $4 $5 $9
L: 13 0 31 0 13

M: 1 1 1 1
3 6 0 6 3

Figure 4.10: A mean-preserving spread.

Intuitively, a risk-averse person should dislike the change from L to M because it involves
more risk: the prize of $4 has been replaced with a non-degenerate “sub-lottery” with
expected value of $4. By definition of risk aversion, the sub-lottery is worse than its
expected value.
To perform a ‘worsening” of lottery L it is not even necessary to spread out the entire
probability of prize $4; for example, we could merely take away half that probability,
namely 16 , and spread it equally between the prizes $3 and $5 thus obtaining the alternative
lottery  
0 $2 $3 $4 $5 $9
M = 1 1 1 1 1 .
3 12 6 12 3
It is easy to check that also the expected value of M0
is 5. In Exercise 4.18√the reader
is asked to verify that an individual with utility-of-money function U(m) = m strictly
prefers L to M 0 and M 0 to M.
In fact, according to the next definition, M is a mean-preserving spread of M 0 , which,
in turn, is a mean-preserving spread of L.
Before giving the formal definition of a mean-preserving spread, let us gain some
intuition, as follows. Start with a lottery
 
$m1 $m2 . . . $mn
L=
p1 p2 . . . pn

and fix three monetary prizes mi , m j and mk with mi < m j < mk and assume that m j has
positive probability in L, that is, p j > 0. Since m j is strictly between mi and mk , there is a
δ ∈ (0, 1) such that m j = (1 − δ )mi + δ mk , in fact
m j − mi
δ= . ()
mk − mi
4.4 Higher risk 97

Now focus on the part of lottery L that involves the three prizes mi , m j and mk :
 
. . . $mi . . . $m j . . . $mk . . .
. . . $pi . . . $p j . . . $pk . . .
Let α ∈ (0, 1) and let us reduce the probability of m j from p j to p j − α p j and spread the
probability α p j between mi and mk in the proportions (1 − δ ) and δ , respectively, where
δ is given by (). This is shown in Figure 4.11.

... $mi ... $m j ... $mk ...

... pi ... pj ... pk ...

. . . pi +(1 − δ )α p j . . . p j −α p j . . . pk +δ α p j . . .

Figure 4.11: A mean-preserving spread.

Let M be the new lottery so constructed. Thus L and M differ only in the probabilities
assigned to prizes mi , m j and mk . The contribution of these prizes to the calculation of the
expected value of the initial lottery L is
pi mi + p j m j + pk mk
while the contribution of these prizes to the calculation of the expected value of the new
lottery M is
     
pi + (1 − δ )α p j mi + p j − α p j m j + pk + δ α p j mk
 
= pi mi + p j m j + pk mk + α p j (1 − δ )mi + δ mk − m j
| {z }
=0 because (1−δ )mi +δ mk =m j

= pi mi + p j m j + pk mk .
Hence E[L] = E[M].
We are now ready to give the definition of a mean-preserving spread.10
10 The notion of a mean preserving spread of a probability distribution was introduced by Michael

Rothschild and Joseph Stiglitz in “Increasing risk I: A definition”, Journal of Economic Theory, 1970, Vol.
2, pp. 225-243. Their definition involved re-assigning probabilities across four different points, but that
definition is equivalent to one that is based on reassigning probabilities across three different points, as
shown by Eric Rasmusen and Emmanuel Petrakis, “Defining the mean-preserving spread: 3-pt versus 4-pt”,
in: Decision making under risk and uncertainty: new models and empirical findings, edited by John Geweke,
Amsterdam: Kluwer, 1992. Rasmusen and Petrakis’ definition is different from ours, but equivalent to it.
Their definition is as follows. Take three points a1 , a2 , a3 with a1 < a2 < a3 ; a mean-preserving spread
is a triple of probabilities γ1 , γ2 , γ3 such that (1) γ1 + γ3 = γ2 ≤ p2 (where p2 is the initial probability of
a2 ) and (2) γ1 a1 − γ2 a2 + γ3 a3 = 0. The initial probability of ai is pi and the modified probabilities are
p1 + γ1 , p2 − γ2 , p3 + γ3 . To convert it into our definition let α = γ1p+γ
2
3
and δ = aa23 −a1
−a1 . Invert the operations
to go from our definition to theirs.
98 Chapter 4. Money lotteries revisited
Definition 4.4.2 Let
! !
$m1 $m2 . . . $mn $m1 $m2 . . . $mn
L= and M= .
p1 p2 ... pn q1 q2 ... qn

be two money lotteries. We say that M is obtained from L by a mean-preserving spread,

and write
L →MPS M

if there are three prizes mi , m j , mk with mi < m j < mk such that:

(1) for very t ∈ {1, . . . , n} \ {i, j, k}, qt = pt and

(2) for some α ∈ (0, 1]

m j − mi
qi = pi + (1 − δ ) α p j , q j = p j − α p j, qk = pk + δ α p j with δ = .
mk − mi

Definition 4.4.3 Let

! !
$m1 $m2 . . . $mn $m1 $m2 . . . $mn
L= and M= .
p1 p2 ... pn q1 q2 ... qn

be two money lotteries. We say that L second-order stochastically dominates M and

write
L >SSD M

if M can be obtained from L by a finite sequence of mean-preserving spreads, that is, if

there is a sequence of money lotteries hL1 , L2 , . . . , Lm i (with m ≥ 2) such that:
(1) L1 = L,
(2) Lm = M and
(3) for every i = 1, . . . , m − 1, Li →MPS Li+1 .
 4.5 Exercises 99

As remarked above, a risk-averse person ought to be made worse off by a mean

preserving spread. This intuition is confirmed by the following theorem.11

Theorem 4.4.2 Let L and M be two money lotteries (over the same set of prizes). Then

L >SSD M

if and only if

E[U(L)] > E[U(M)], for every strictly increasing and concave utility function U.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 4.5.4 at the end of this chapter.

4.5 Exercises
The solutions to the following exercises are given in Section 4.6 at the end of this chapter.

4.5.1 Exercises for Section 4.1: vNM preferences over money lotteries

Exercise 4.1 Jennifer’s von Neumann-Morgenstern utility-of-money function is U(m) =

√
20 m − 4. Consider the following lottery, where the outcomes are possible levels of
wealth for Jennifer:  
$8 $18 $24 $28 $30
L= 2 1 1 1 1
5 5 10 10 5

(a) What is the expected value of L?

(b) What is the expected utility of L?
d
(c) Calculate dm U(m).
d 2
(d) Calculate dm 2 U(m).

(e) Is Jennifer risk-averse, risk-neutral or risk-loving?



11 This result is best known to economists from the 1970 paper by Rothschild and Stiglitz mentioned in
Footnote 10. However, in a later article (Michael Rothschild and Joseph Stiglitz, Addendum to ‘Increasing
risk I: A definition’, Journal of Economic Theory, 1972, Vol. 5, p. 306) the authors themselves acknowledged
that their main result could have been derived from earlier contributions by mathematicians.
 100 Chapter 4. Money lotteries revisited

Exercise 4.2 Consider again the wealth lottery of Exercise 4.1, but a different agent:
√
Jim, whose vNM utility-of-money function is U(m) = m. Answer the same questions
as in Exercise 4.1 but referring to Jim. 

Exercise 4.3 What attitude to risk is incorporated in the following vNM utility-of-
money functions? Base your answer on the sign of the second derivative of the utility
function.
(a) ln(m + 1)
(b) 8 + m1.65
(c) 2 + 7m.


Exercise 4.4 Let m denote the amount of money (measured in millions of dollars) and
suppose that it varies in the interval [0, 1]. John’s utility-of-money function is given by:

U(m) = −m2 + 2m − 4.

(a) What is John’s attitude to risk?

Jenny, on the other hand, has the following utility function:

V (m) = −3 m2 − 2m .

(b) What is Jenny’s attitude to risk?

(c) Do John and Jenny have the same preferences?
(d) Give an example of two utility functions that incorporate the same attitude to risk
but do not represent the same preferences for lotteries.


4.5.2 Exercises for Section 4.2: Measures of risk aversion

Exercise 4.5 As in Exercise 4.1, consider Jennifer, whose vNM utility-of-money
√
function is U(m) = 20 m − 4, and the lottery
 
$8 $18 $24 $28 $30
L= 2 1 1 1 1
5 5 10 10 5

(a) Calculate the risk premium for lottery L for Jennifer.

(b) Calculate Jennifer’s Arrow-Pratt measure of absolute risk aversion for m = 900
and for m = 1, 600.

4.5 Exercises 101

Exercise 4.6 As in Exercise 4.2, consider Jim, whose vNM utility-of-money function
√
is V (m) = m, and the lottery
 
$8 $18 $24 $28 $30
L= 2 1 1 1 1
5 5 10 10 5

(a) Calculate the risk premium for lottery L for Jim.

(b) Calculate Jim’s Arrow-Pratt measure of absolute risk aversion for m = 900 and
for m = 1, 600.


Exercise 4.7 As in Exercise 4.4, let m denote the amount of money, measured in
millions of dollars, and suppose that it varies in the interval [0, 1]. John’s utility-of-
2

by: U(m) = −m + 2m − 4 while Jenny’s utility function is:
money function is given
2
V (m) = −3 m − 2m . Calculate the Arrow-Pratt measures of absolute and relative
risk aversion for John and Jenny and compare them. 

 
$24 $12 $48
Exercise 4.8 Amy faces the wealth lottery 2 3 1 and tells you that she
6 6 6
considers it equivalent to getting $18 for sure.
(a) Calculate the risk premium for lottery L for Amy.
(b) What is Amy’s attitude to risk?
√
(c) Could Amy’s vNM utility-of-money function be U(m) = m ?


Exercise 4.9 Bill is risk neutral.

(a) How does he rank the following lotteries?
   
$24 $12 $48 $6 $180 $0 $90
L1 = 1 2 1 2 L2 = 1 17 2
6 6 6 6 20 20 20

(b) What is the risk premium associated with lottery L1 for Bill?
(c) What is the risk premium associated with lottery L2 for Bill?

 102 Chapter 4. Money lotteries revisited

Exercise 4.10 Consider the following money lottery, where the outcomes are changes
in wealth:  
−$50 $120
M= 1 3 .
4 4
Berta’s vNM utility-of-money function is U(m) = ln(m).
(a) Suppose that Berta’s inital wealth is $80. Write the wealth lottery corresponding
to lottery M above and calculate the risk premium for this lottery for Berta.
(b) Suppose that Berta’s inital wealth is $200. Write the wealth lottery corresponding
to lottery M above and calculate the risk premium for this lottery for Berta.


 
$120 $180 $260
Exercise 4.11 Consider the wealth lottery L = 2 2 1 and the fol-
5 5 5
lowing vNM utility-of-money functions defined for m ∈ [0, 300]:
√ m 2 m
U(m) = m and V (m) = − − 36 + + 1, 400.
10 20
(a) Write an equation whose solution gives RLU (the risk premium for lottery L and
utility function U) and verify that the solution is RLU = 3.6949.
(b) Write an equation whose solution gives RLV (the risk premium for lottery L and
utility function V ) and verify that the solution is RLV = 6.848.
(c) Using the Arrow-Pratt measure of absolute risk aversion, which of U and V
incorporates greater risk aversion?


4.5.3 Exercises for Section 4.3: Some noteworthy utility functions

Exercise 4.12 Plot the following utility functions in the same diagram:

U(m) = 1 − e−m and V (m) = 1 − e−3m .

Exercise 4.13 Consider the utility-of-money function U(m) = ma , where a is a constant

such that 0 < a < 1. For this function is the Arrow-Pratt measure of absolute risk
aversion decreasing, constant or increasing? 

2
Exercise 4.14 Consider the quadratic utility-of-money function U(m) = cm − m2 ,
where c is a positive constant and m ∈ [0, c). For this function is the Arrow-Pratt
measure of absolute risk aversion decreasing, constant or increasing? 
 4.5 Exercises 103

4.5.4 Exercises for Section 4.4: Higher risk

Exercise 4.15 Consider the following lotteries:

   
$26 $40 $58 $80 $96 $26 $40 $58 $80 $96
L= 6 4 2 1 7 and M = 5 4 2 2 7
20 20 20 20 20 20 20 20 20 20

Does one dominate the other in terms of first-order stochastic dominance? 

Exercise 4.16 Consider the following lotteries:

   
$26 $40 $58 $80 $96 $26 $40 $58 $80 $96
L= 6 4 2 8 and M = 5 4 2 2 7
20 20 20 0 20 20 20 20 20 20

Does one dominate the other in terms of first-order stochastic dominance? 

Exercise 4.17 Consider the lotteries of Exercise 4.16. Since it is not the case that M
dominates L in terms of first-order stochastic dominance, by Theorem 4.4.1 there must
be an increasing utility-of-money function U such that E[U(L)] > E[U(M)]. Construct
such a function. Note that you don’t need to define a function over the entire set of non-
negative real numbers: it is enough to define a function over the set {26, 40, 58, 80, 96}.


Exercise 4.18 Consider the following lotteries, which were discussed at the beginning
of Section 4.4.2:  
$2 $3 $4 $5 $9
L= 1
3 0 31 0 13
   
0 $2 $3 $4 $5 $9 $2 $3 $4 $5 $9
M = 1 1 1 1 1 and M = 1 1 .
3 12 6 12 3 3 6 0 16 13
√
Show that an individual with utility-of-money function U(m) = m strictly prefers L
to M 0 and M 0 to M. 
 104 Chapter 4. Money lotteries revisited

Exercise 4.19 Consider the following lotteries:

   
$4 $16 $25 $36 $49 $4 $16 $25 $36 $49
L= 3 9 18 8 2 and M= 23 9 3 8 17 .
40 40 40 40 40 200 40 8 40 200

(a) Calculate E[L].

(b) Calculate E[M].
(c) Calculate the expected
√ utility of L for an individual whose utility-of-money
function is U(m) = m.
(d) Calculate the expected
√ utility of M for an individual whose utility-of-money
function is U(m) = m.
(e) Show that M is a mean-preserving spread of L according to Definition 4.4.2.


Exercise 4.20 Show that L >SSD M, where

   
$6 $23 $44 $51 $70 $6 $23 $44 $51 $70
L= 1 1 1 1 1 and M= 77 11
3 12 6 12 3 192 94 0 0 4349
9024

by constructing a two-step mean-preserving spread from L to M.

[Hint: in the first step take the probability assigned to $44 and re-allocate it to $6 and
$70.] 

4.6 Solutions to Exercises

Solution to Exercise 4.1

2 1 1 2 4
(a) The expected value of L is 10 × 30 + 10 × 28 + 10 × 24 + 10 × 18 + 10 × 8 = 18.
(b) The expected utility of L is
2
√ 1
√ 1
√ 2
√ 4
√
10 (20 30−4)+ 10 (20 28−4)+ 10 (20 24−4)+ 10 (20 18−4)+ 10 (20 8−4)

= 77.88.
√
d
(c) dm (20 m − 4) = 20 2√1 m = 10
√
m
.

d2 √ 3
(20 m − 4) = 10 − 12 m− 2 = − √5 3 < 0, for every m > 0.


(d) dm2 m
(e) Jennifer is risk-averse since the second derivative of her utility function is negative
for every m > 0. 
4.6 Solutions to Exercises 105

Solution to Exercise 4.2

(a) The expected value is, of course, the same, namely 18.
(b) The expected utility of L is
2
√ 1
√ 1
√ 2
√ 4
√
10 30 + 10 28 + 10 24 + 10 18 + 10 8 = 4.094.

Note that this is equal to 77.88 1

20 + 5 (recall that 77.88 was the expected utility for Jen-
nifer). Indeed, Jim’s utility function, call it V , can be obtained from Jennifer’s
utility function, call it U(m), by applying the following affine transformation
1
V (m) = 20 U(m) + 15 ; hence Jennifer and Jim have the same preferences.
√
d m 1
dm = 2 m .
(c) √
√
d2 m
(d) dm2
= − √1 3 .
4 m
(e) Jim is risk-averse (he has the same preferences as Jennifer). 

Solution to Exercise 4.3

d2 1
(a) dm2
ln(m + 1) = − (m+1)2 < 0, for every m ≥ 0. Thus risk aversion.

d2
(b) dm2
(8 + m1.65 ) = 1.0725
m0.35
> 0, for every m > 0. Thus risk love.
d2
(c) dm2
(2 + 7m) = 0. Thus risk neutrality. 

Solution to Exercise 4.4

(a) U 00 (m) = −2 < 0. Thus John is risk averse.
(b) V 00 (m) = −6 < 0. Thus Jenny is risk averse.
(c) Since V (m) = 3U(m) + 12, that is, V is an affine transformation of U, John and
Jenny have the same preferences. √
(d) There are, of course, many examples. One example is U(m) = m and
V (m) = ln(m + 1). 

Solution to Exercise 4.5

(a) Recall from Exercise 4.1 that the expected√value of L is 18. The risk premium is
the value of R that solves the equation 20 18 − R − 4 = 77.88. The solution is
R = $1.24.
(b) The Arrow-Pratt measure of absolute risk aversion is
5
U 00 (m) −√ 3 1
A(m) = − 0 = − 10m =
U (m) √ 2m
m

1 1
Thus A(900) = 1,800 and A(1, 600) = 3,200 . 
106 Chapter 4. Money lotteries revisited

Solution to Exercise 4.6

(a) Again, the expected

√ value of L is 18. The risk premium is the value of R that solves
the equation 18 − R = 4.094. The solution is R = $1.24: the same as for Jennifer
(as it should be, since they have the same preferences).
(b) The Arrow-Pratt measure of absolute risk aversion is

1
V 00 (m) − √ 3 1
A(m) = − 0 = − 41 m =
V (m) √ 2m
2 m

the same as for Jennifer (as it should be, since they have the same preferences). Thus
1 1
A(900) = 1,800 and A(1, 600) = 3,200 . 

Solution to Exercise 4.7

We already know from Exercise 4.4 that John and Jennifer have the same preferences. This
is confirmed by the fact that the Arrow-Pratt measures are the same for both individuals:

1 m
AU (m) = AV (m) = and rU (m) = rV (m) = .
1−m 1−m

Solution to Exercise 4.8  

$24 $12 $48
(a) The expected value of lottery 2 3 1 is 26 24 + 36 12 + 16 48 = 22. Thus
6 6 6
the risk premium is $(22 − 18) = $4.
(b) Amy is risk-averse since she considers the lottery to be equivalent to a sum of money
which is less than the expected value of the lottery (hence she prefers the expected
value of the lottery for sure to the lottery).
(c) If U(m) is Amy’s vNM utility-of-money function, then it must be that
U(18) = E[U(L)], where E[U(L)] = 26 U(24) + 36 U(12) + 16 U(48). Since
√ √ √ √
18 = 4.2426, while 26 24 + 63 12 + 61 48 = 4.5197, it cannot be that
√
U(m) = m. 

Solution to Exercise 4.9

(a) The expected value of both lotteries is 18, hence Bill is indifferent between the two.
(b) Zero.
(c) Zero. 
4.6 Solutions to Exercises 107

Solution to Exercise 4.10

(a) When
 Berta’s initial
 wealth is $80, the corresponding wealth lottery is
$30 $200
L= 1 3 , whose expected value is $157.5. The risk premium is given
4 4
by the solution to

ln(157.5 − R) = 41 ln(30) + 34 ln(200)

which is $33.0304.
(b) When
 Berta’s initial
 wealth is $200, the corresponding wealth lottery is
$150 $320
L= 1 3 , whose expected value is $277.5. The risk premium is given
4 4
by the solution to

ln(277.5 − R) = 41 ln(150) + 34 ln(320)

which is $12.7199. 

 
$120 $180 $260
Solution to Exercise 4.11 We are considering the wealth lottery L = 2 2 1 .
5 5 5
The expected value of L is 172.
√ √ √ √
(a) RLU is the solution to 172 − R = 25 120 + 25 180 + 15 260. The solution is
RLU = 3.6949.

(b) RLV is the solution to

172−R

2
− 10 − 36 + 172−R
20 + 1, 400 =
h
2 120 i
2 120
5 − 10 − 36 + 20 + 1, 400
h i
2
+ 52 − 180 180


10 − 36 + 20 + 1, 400
h 2
i
1 260 260


+ 5 − 10 − 36 + 20 + 1, 400 .

The solution is RLV = 6.848.

1 1
(c) AU (m) = 2m and AV (m) = 362.5−m . The two are equal when m = 120.833,
AU (m) > AV (m) for m ∈ (0, 120.833) and AU (m) < AV (m) for m ∈ (120.833, 300].
Thus U incorporates greater risk aversion than V for values of m in the interval
(0, 120.833) and less risk aversion than V in the interval (120.833, 300]. 
108 Chapter 4. Money lotteries revisited

Solution to Exercise 4.12 See Figure 4.12. 

utility

1 − e−3m

1 − e−m

money
0

Figure 4.12: The graphs of 1 − e−m and 1 − e−3m .

Solution to Exercise 4.13

d a a d2 −a(1−a) 1−a
dm m = m1−a
and dm2
ma = m2−a
. Thus A(m) = m which is decreasing in m. 

Solution to Exercise 4.14

   
d m2 d2 m2 1
dm c m − 2 = c − m and dm2
c m − 2 = −1. Thus A(m) = c−m which is increasing
d 1 1


in m. In fact, dm c−m = (c−m)2 > 0. 

Solution to Exercise 4.15

The lotteries are:
   
$26 $40 $58 $80 $96 $26 $40 $58 $80 $96
M= 5 4 2 2 7 and L = 6 4 2 1 7
20 20 20 20 20 20 20 20 20 20

By constructing the corresponding cumulative distribution functions one can see that
M >FSD L (Qi ≤ Pi for every i = 1, . . . , 5 and Q3 < P3 ):

m1 = $26 m2 = $40 m3 = $58 m4 = $80 m5 = $96

6 10 12 13
cumulative for L, P: 20 20 20 20 1
5 9 11 13
cumulative for M, Q: 20 20 20 20 1


4.6 Solutions to Exercises 109

Solution to Exercise 4.16

The lotteries are:
   
$26 $40 $58 $80 $96 $26 $40 $58 $80 $96
L= 6 4 2 8 and M= 5 4 2 2 7
20 20 20 0 20 20 20 20 20 20

By constructing the corresponding cumulative distribution functions one can see that,
according to the criterion of first-order dominance, it is neither the case that L dominates
6 5
M (since, for example, P1 = 20 > Q1 = 20 ) nor the case that M dominates L (since
13 12
Q4 = 20 > P4 = 20 ):

$26 $40 $58 $80 $96

6 10 12 12
cumulative for L, P: 20 20 20 20 1
5 9 11 13
cumulative for M, Q: 20 20 20 20 1


Solution to Exercise 4.17
The lotteries are:
   
$26 $40 $58 $80 $96 $26 $40 $58 $80 $96
L= 6 4 2 8 and M= 5 4 2 2 7
20 20 20 0 20 20 20 20 20 20

1
Since L assigns an additional probability of 20 to $96 (relative to M), it is sufficient to have
a “big jump” in utility going from $80 to $96. For example, consider the following utility
function:
$26 $40 $58 $80 $96
Utility U : 1 2 3 4 10
6 4 2 8 5 4 2 2
Then E[U(L)] = 20 1 + 20 2 + 20 3 + 20 10 = 5 and E[U(M)] = 20 1 + 20 2 + 20 3 + + 20 4+
7
20 10 = 4.85. Thus an individual with this (strictly increasing) utility function prefers
lottery L to lottery M. One can also easily construct a strictly increasing utility function
according to which lottery M is preferred to lottery L (big jump at $80, small jump
at $96). 

Solution to Exercise 4.18

The lotteries are:  
$2 $3 $4 $5 $9
L= 1
3 0 13 0 13
   
0 $2 $3 $4 $5 $9 $2 $3 $4 $5 $9
M = 1 1 1 1 1 and M = 1 1 .
3 12 6 12 3 3 6 0 16 31
√
1
√ √
E[U(L)] = 3 2 + 31 4 + 31 9 = 2.1381
√ √ √ √ √
> E[U(M 0 )] = 13 2 + 12
1
3 + 16 4 + 121
5 + 13 9 = 2.1354
√ √ √ √
> E[U(M)] = 31 2 + 16 3 + 16 1
5 + 13 9 = 2.1328.


110 Chapter 4. Money lotteries revisited

Solution to Exercise 4.19

The lotteries are:
   
$4 $16 $25 $36 $49 $4 $16 $25 $36 $49
L= 3 9 18 8 2 and M= 23 9 3 8 17 .
40 40 40 40 40 200 40 8 40 200

3 9 18 8 2
(a) E[L] = 40 4 + 40 16 + 40 25 + 40 36 + 40 49 = 24.8.
23 9 3 8 17
(b) E[M] = 200 4 + 40 16 + 8 25 + 40 36 + 200 49 = 24.8.
√ 3 9
√ √ √ √
(c) E[U(L)] = 4 + 40
40 16 + 18 8 2
40 25 + 40 36 + 40 49 = 4.85
23
√ 9
√ √ √ √
(d) E[U(M)] = 200 4 + 40 16 + 38 25 + 40
8 17
36 + 200 49 = 4.8.

(e) We have that m1 = 4, m2 = 16, m3 = 25, m4 = 36, m5 = 49, p2 = q2 and p4 = q4 .

Thus the change involves prizes m1 , m3 and m5 , that is, i = 1, j = 3,
k = 5. To find α solve 18 18 3
40 − α 40 = 8 which gives α = 61 . Then verify that
  
3 −m1 m3 −m1
p1 + 1 − mm5 −m1 α p3 = q1 and p5 + m5 −m1 α p3 = q5 ;
3
+ 1 − 25−4

1 18
23 2 25−4 1 18


17
indeed 40 49−4 6 40 = 200 and 40 + 49−4 6 40 = 200 

Solution to Exercise 4.20

The lotteries are:
   
$6 $23 $44 $51 $70 $6 $23 $44 $51 $70
L=  and M= .
1 1 1 1 1 77 11 4349
3 12 6 12 3 192 94 0 0 9024

Let us perform a first mean-preserving spread (MPS) on L by reducing the probability of

m3 = 44 to 0 (hence α = 1) and spreading it out to m1 = 6 and m5 = 70 (thus δ = 44−6
70−6 =
38 1 38 1 77


64 ); then the probability of m1 becomes 3 + 1 − 64 6 = 192 and the probability of m5
becomes 31 + 38 0

1 83
64 6 = 192 . Call the resulting lottery M . Then

 
$6 $23 $44 $51 $70
M0 =  .
77 1 1 83
192 12 0 12 192

Now perform a second MPS on M 0 by reducing the probability of m4 = 51 to 0 (hence

α = 1) and spreading it out to m2 = 23 and m5 = 70 (thus δ = 51−23 28
70−23 = 47 ); then the
1 28 1 11


probability of m2 becomes 12 + 1 − 47 12 = 94 and the probability of m5 becomes
4.6 Solutions to Exercises 111
83 28 1 4349


192 + 47 12 = 9024 thus yielding

 
$6 $23 $44 $51 $70
M= .
77 11 4349
192 94 0 0 9024

It can be verified that E[L] = E[M 0 ] = E[M] = 38.8333. 

 5. Insurance: Part 2

5.1 Binary lotteries and indifference curves

In this chapter we complete the analysis of insurance that we started in Chapter 2 by
considering the point view of the potential customer. Before we do so, we need to develop
the analysis of binary money lotteries, which are lotteries that involve only two prizes.
Fix a value of p (with 0 < p < 1) and consider all the lotteries of the form
 
$x $y
with x ≥ 0 and y ≥ 0.
p 1− p
Thus we think of x and y as variables, while p is a constant.
We can identify a binary lottery with a point (x, y) in the positive quadrant of the
cartesian plane. If x = y then the lottery (x, x) lies on the 45o -line out of the origin and
represents the situation where the individual gets x with probability p and x with probability
(1 − p), that is, she gets x for sure; if x > y the point lies below the 45o -line and if x < y
the point lies above the 45o -line.
Consider an individual whose utility-of-money function is U(m). We assume that
U 0 (m) > 0 (for every m ≥ 0), that is, that the individual prefers more money to less. Given
a lottery (x, y), the individual’s expected utility is given by: pU(x) + (1 − p)U(y). Given
two lotteries A = (x1 , y1 ) and B = (x2 , y2 ), the individual will prefer A to B if and only if
E[U(A)] = pU(x1 ) + (1 − p)U(y1 ) > E[U(B)] = pU(x2 ) + (1 − p)U(y2 ),
she will prefer B to A if the above inequality is reversed and will be indifferent between
A and B if E[U(A)] = E[U(B)]. For example, if p = 14 and the individual is risk neutral
(so that we can take the identity function U(m) = m as her vNM utility function) then the
individual will be indifferent among the following lotteries, since their expected value is
the same (namely 85): (130, 70), (100, 80), (85, 85) and (16, 108).
 114 Chapter 5. Insurance: Part 2
Definition 5.1.1 An indifference curve is a set of points (lotteries) in the (x, y) plane
among which the individual is indifferent. For every point (x, y) there is an indifference
curve that goes through that point. Since U 0 (m) > 0, for every m, each indifference
curve will be downward-sloping.a
a In order for expected utility to remain constant, if one coordinate is increased then the other coordinate

must be decreased.

We want to relate the shape of the indifference curves of an individual to her attitude
towards risk.

5.1.1 Case 1: risk neutrality

As remarked above, for a risk-neutral person we can take the identity function U(m) = m
as her vNM utility-of-money function, so that expected utility and expected value coincide.
Fix an arbitrary lottery A = (xA , yA ) and let us try to find another lottery B = (xB , yB ) that
lies on the same indifference curve. Then it must be that pxA +(1− p)yA = pxB +(1− p)yB
which can be written as
rise
z }| {
yA − yB p
=− .
xA − xB 1− p
| {z }
run
p
Thus indifference curves are straight lines with slope − 1−p , as shown in Figure 5.1.

y
(probability 1 − p)

yA A

yA −yB p

 xA −xB = − 1−p
rise


yB B
| {z }
run

x
0 xA xB (probability p)

p
slope: − 1−p

Figure 5.1: An indifference curve for a risk-neutral individual.

 5.1 Binary lotteries and indifference curves 115

5.1.2 Case 2: risk aversion

Now consider the case of a risk-averse individual. Recall from Chapter 4 that the utility-
of-money function U(m) of a risk-averse individual is strictly concave, that is, for every
x > 0 and y > 0 and for every t ∈ (0, 1),

U (tx + (1 − t)y) > tU(x) + (1 − t)U(y). (5.1)

We now show that, if we take two lotteries A and B that yield the same expected utility
(so that they lie on the same indifference curve), then all the lotteries on the line segment
joining A and B (apart from A and B themselves) correspond to higher levels of expected
utility than A and B . Hence, since the utility function is assumed to be strictly increasing,
it follows that the indifference curve to which A and B belong, must lie below the line
segment that joins A and B, that is, the indifference curve must be convex towards the
origin.

As before, fix an arbitrary p ∈ (0, 1) and consider all the lotteries of the form
 
$x $y
p 1− p

which can be identified with points in the positive quadrant of the cartesian plane (x, y).

Let A = (xA , yA ) and B = (xB , yB ) lie on the same indifference curve, that is,

pU(xA ) + (1 − p)U(yA ) = pU(xB ) + (1 − p)U(yB ) = û.

| {z } | {z }
=E[U(A)] =E[U(B)]

Fix an arbitrary t ∈ (0, 1) and consider the point C = tA + (1 − t)B on the line segment
joining A and B, which represents the lottery
 
txA + (1 − t)xB tyA + (1 − t)yB
C= .
p 1− p

Then

E[U(C)] = p U (txA + (1 − t)xB ) + (1 − p) U (tyA + (1 − t)yB ) . (5.2)

By (5.1),

U (txA + (1 − t)xB ) > tU(xA ) + (1 − t)U(xB ) (5.3)

U (tyA + (1 − t)yB ) > tU(yA ) + (1 − t)U(yB ). (5.4)
116 Chapter 5. Insurance: Part 2

Thus, from (5.2)-(5.4) we get that

E[U(C)] > p [tU(xA ) + (1 − t)U(xB )] + (1 − p) [tU(yA ) + (1 − t)U(yB )]

= t [pU(xA ) + (1 − p)U(yA )] + (1 − t) [pU(xB ) + (1 − p)U(yB )]

= tE[U(A)] + (1 − t)E[U(B)]

= t û + (1 − t)û = û.

All of this is illustrated in Figure 5.2.

y
(probability 1 − p)

yA A

tyA + (1 − t)yB C = tA + (1 − t)B

E[U(C)] > û

yB B
EU = û
x
0 xA xB (probability p)

txA + (1 − t)xB

Figure 5.2: Indifference curves for a risk-averse individual are convex.

 5.1 Binary lotteries and indifference curves 117

5.1.3 Case 3: risk love

Now consider the case of a risk-loving individual. Recall from Chapter 4 that the utility-of-
money function U(m) of a risk-loving individual is strictly convex, that is, for every x > 0
and y > 0 and for every t ∈ (0, 1),

U (tx + (1 − t)y) < tU(x) + (1 − t)U(y). (5.5)

With an argument similar to the one used in the previous section, one can show that, if
we take two lotteries A and B that yield the same expected utility – so that they lie on the
same indifference curve – all the lotteries on the line segment joining A and B (apart from
A and B themselves) correspond to lower levels of expected utility than A and B . Hence,
since the utility function is assumed to be strictly increasing, it follows that the indifference
curve to which A and B belong, must lie above the line segment that joins A and B, that is,
the indifference curve must be concave towards the origin, as shown in Figure 5.3.

y
(probability 1 − p)

A
yA

tyA + (1 − t)yB
tA + (1 − t)B = C
yB E[U(C)] < û
B

EU = û
x
0 xA xB (probability p)

txA + (1 − t)xB

Figure 5.3: Indifference curves for a risk-loving individual are concave.

 118 Chapter 5. Insurance: Part 2

5.1.4 The slope of an indifference curve

We saw above that the indifference curves of a risk-neutral individual are straight lines and
p
thus have a constant slope, which is equal to − 1−p . On the other hand, the indifference
curves of a risk-averse individual are convex towards the origin and thus do not have a
constant slope: indeed the slope decreases as we move along the curve in the direction of
an increase in the horizontal coordinate (and a decrease in the vertical coordinate). For a
risk-loving individual the opposite is true: the slope of an indifference curve increases as
we move along the curve in the direction of an increase in the horizontal coordinate.
How can we compute the slope of an indifference curve at a point? Let A = (xA , yA )
and consider a point B on the same indifference curve as A, so that E[U(A)] = E[U(B)].
Let us choose this point B to be “very close” to A, so that B = (xA + δ , yA + ε) with δ and
ε close to 0 (one must be positive and the other negative). By hypothesis,

pU(xA ) + (1 − p)U(yA ) = pU(xA + δ ) + (1 − p)U(yA + ε) (5.6)

| {z } | {z }
E[U(A)] E[U(B)]

Since B is close to A (that is, δ and ε are small), we can approximate the values of
U(xA + δ ) and U(yA + ε) using the derivative of U (that is, using a first-order Taylor
expansion):

U(xA + δ ) = U(xA ) +U 0 (xA ) δ

(5.7)
U(yA + ε) = U(yA ) +U 0 (yA ) ε.
Replacing (5.7) into (5.6) we get

pU(xA ) + (1 − p)U(yA ) = p U(xA ) +U 0 (xA )δ ) + (1 − p) U(yA ) +U 0 (yA )ε

   
(5.8)
= pU(xA ) + (1 − p)U(yA ) + pU 0 (xA )δ + (1 − p)U 0 (yA )ε
from which we get that

pU 0 (xA )δ + (1 − p)U 0 (yA )ε = 0,

that is,
rise
p U 0 (xA )
z}|{
ε
=− .
δ
|{z} 1 − p U 0 (yA )
run

Thus the slope of an indifference curve at a point A = (xA , yA ) is given by1

p U 0 (xA )
− (5.9)
1 − p U 0 (yA )
1 Alternatively, one can derive the slope of an indifference curve at a point by using the implicit func-
tion theorem, which says the following. Let F : R2 → R be a continuously differentiable function and
(x0 , y0 ) ∈ R2 a point such that F(x0 , y0 ) = c; if ∂∂Fy (x0 , y0 ) 6= 0 then there is an interval (x0 − ε, x0 + ε) and a
differentiable function f : (x0 − ε, x0 + ε) → R such that (1) F(x0 , f (x0 )) = y0 , (2) F(x, f (x)) = c for every
∂F
(x0 ,y0 )
x ∈ (x0 − ε, x0 + ε) and (3) f 0 (x0 ) = − ∂∂Fx . To apply the implicit function theorem in this context, let
∂ y (x0 ,y0 )
F(x, y) = pU(x) + (1 − p)U(y) and let A = (xA , yA ) be a point where pU(xA ) + (1 − p)U(yA ) = û.
5.1 Binary lotteries and indifference curves 119
(xA ) 0
In the case of risk neutrality U 0 is constant and thus U 0 (xA ) = U 0 (yA ) so that U
U 0 (yA ) = 1;
p
hence the slope becomes − 1−p at every point, consistently with what we saw above.

Now let us see what (5.9) implies for a concave utility-of-money function, that is, for
the case of risk aversion. When the utility function is concave, the second derivative is
negative (U 00 (m) < 0), which means that the first derivative is decreasing, that is,
U 0 (m1 )
 
0 0
if m1 < m2 then U (m1 ) > U (m2 ) or 0 >1 ,
U (m2 )
as shown in Figure 5.4.

Utility Marginal utility

U(m) U 0 (m)

Utility Marginal utility

U 0 (m1 )

U 0 (m2 )
money money
0 m m1 m2 m

Figure 5.4: When the utility function is concave, marginal utility is decreasing.

U 0 (x)
R At a point A = (x, y) above the 45o line (where x < y) we have that U 0 (y) > 1 so that

p U 0 (x) p p U 0 (x) p
− <− or > .
1 − p U 0 (y) 1− p 1 − p U 0 (y) 1 − p
p
Hence the indifference curve is steeper than the straight line with slope − 1−p .

Conversely, at a point A = (x, y) below the 45o line (where x > y) we have that
U 0 (x)
U 0 (y) < 1 so that

p U 0 (x) p p U 0 (x) p
− 0
>− or 0
< .
1 − p U (y) 1− p 1 − p U (y) 1 − p
p
Hence the indifference curve is less steep than the straight line with slope − 1−p .

U 0 (x)
Finally at a point on the 45o line (where x = y) we have that U 0 (y) = 1 so that

p U 0 (x) p
− =− ;
1 − p U 0 (y) 1− p
p
hence the straight line with slope − 1−p is tangent to the indifference curve.
120 Chapter 5. Insurance: Part 2
2 √
 Example 5.1 This example is illustrated in Figure 5.5. Let p = and U(m) = m and
5
√
 
$x $y 0 (m) = √ 1 U 0 (x) y
consider all the lotteries of the form 2 3 . Since, U ,
2 m U 0 (y)
= √
x
(for
5 5
x > 0 and y > 0).
Consider three points: (25,100), (64,64) and (121,36). The expected utility of these three
lotteries is the same, namely 8; hence these three points belong to the same indifference
curve.2
• Point (64,64) is on the 45o line and the slope of the indifference curve at that point is
√ !
p 64 p 2
− √ =− =− .
1− p 64 1− p 3

• Point (25,100) is above the 45o line and the slope of the indifference curve at that
point is
√ !  
p 100 2 10 4
− √ =− =− .
1− p 25 3 5 3
• Point (121,36) is below the 45o line and the slope of the indifference curve at that
point is
√ !  
p 36 2 6 4
− √ =− =− .
1− p 121 3 11 11


y
(probability 35 )

45o line
EU = 8

100

64
36 4
slope: − 11
x
0 (probability 25 )
25 64 121
slope:
slope: p
= − 1−p
− 43
= − 32

Figure 5.5: The graph for Example 5.1.

2 The 2√ 3√
equation of the indifference curve is obtained by solving for y the equation 5 x + 5 y = 8. The
√ 2
solution is y = 49 (20 − x) .
 5.2 Back to insurance 121

We will omit the case of risk love (convex utility function, concave indifference curves).
The reader should convince himself/herself that in this case an indifference curve is less
p
steep than the line of slope − 1−p at a point above the 45o line and steeper at a point below
o
the 45 line.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 5.5.1 at the end of this chapter.

5.2 Back to insurance

We can now return to the topic of insurance, which we partially analyzed in Chapter 2. We
begin by recalling the general set-up.
Consider an individual whose current wealth is $W0 . She faces the possibility of a loss
in the amount of $` (0 < ` ≤ W0 ) with probability p (0 < p < 1). An insurance contract
can be expressed as a pair of wealth levels (W1 ,W2 ), where W1 is wealth in the bad state (if
the loss occurs) and W2 is wealth in the good state (if the loss does not occur); the amount
(W0 −W2 ) is the contract’s premium and the amount W2 −W1 is the deductible. If W1 = W2
the contract offers full insurance, while if W1 < W2 the contract offers partial insurance.
We saw in Chapter 2 that, through any point in the (W1 ,W2 ) plane, we can draw an
isoprofit line which contains all the contracts that yield the same profit to the insurer. Recall
that
p
isoprofit lines are straight lines with slope − .
1− p
The isoprofit line that goes through the no-insurance point NI = (W0 − `,W0 ) is the zero-
profit line. Points below the zero-profit line represent profitable contracts, while points
above the zero-profit line correspond to contracts that would involve a loss for the insurer.
Thus no insurer would be willing to offer a contract that lies above the zero-profit line.
We can now ask the question: what contracts would be acceptable to the individual
under consideration?
If the individual purchases insurance contract (W1 ,W2 ) then she faces the following
money lottery:  
$W1 $W2
.
p 1− p
We focus on a risk-averse individual who has von Neumann-Morgenstern preferences,
so that her preferences over possible insurance contracts can be represented by means
of a vNM utility-of-money function U(m) (which is increasing and strictly concave). A
contract (W1 ,W2 ) will be acceptable to the individual if it yields at least as high an expected
utility as the no-insurance option, that is, if

pU(W1 ) + (1 − p)U(W2 ) ≥ pU(W0 − `) + (1 − p)U(W0 ).

Using the tools developed in this chapter, we can draw the individual’s indifference curve
122 Chapter 5. Insurance: Part 2

that goes through the no-insurance point NI: it will be a decreasing and convex curve; we
shall call it the reservation indifference curve. Points below the reservation indifference
curve represent contracts that would yield lower expected utility than the no-insurance
option; thus the individual would reject any such contracts, if offered to her. Only contracts
represented by points on or above the reservation indifference curve will be acceptable to
the individual.
Thus the set of mutually beneficial insurance contracts is given by the area bounded
below by the reservation indifference curve, bounded above by the zero-profit line and
bounded on the right by the 450 line; it is shown as a shaded area in Figure 5.6.

Wealth in
good state
(probability 1 − p)
W2

NI 45o line
W0

C zero-profit
line

reservation
indifference
curve

W1 Wealth in
W0 − ` bad state
0 (probability p)

Figure 5.6: The shaded area is the set of mutually beneficial insurance contracts.

Contract C in Figure 5.6 is above the reservation indifference curve and thus yields
higher expected utility than the no-insurance option (that is, C is strictly preferred to NI by
the potential customer) and is below the zero-profit line and thus yields positive profit to
the insurer.
In Figure 5.6 the reservation indifference curve that goes through the no-insurance (NI)
point is steeper at that point than the zero-profit line. This follows from the analysis in
Section 5.1.4. Indeed this is true of any point that lies above the 45o line. Let A = (W1A ,W2A )
be a contract that lies above the 45o line. Then, by (5.9) the slope, at point A, of the
indifference curve that goes through A is equal to
 0 A 
p U (W1 )
− (5.10)
1 − p U 0 (W2A )
5.2 Back to insurance 123

Recall that the slope of the isoprofit line that goes through any point in the wealth diagram
p
is − 1−p . By the remark on page 119,

• At any point above the 45o line the indifference curve is steeper than the isoprofit
line that goes through that point.

• At any point on the 45o line the indifference curve is tangent to (has the same slope
as) the isoprofit line that goes through that point.

This is shown in Figure 5.7. Any contract that lies in the area above the indifference curve
that goes through contract A and below the isoprofit line through A, such as point B
in Figure 5.7, represents a contract that is better than A for the potential customer
(B yields higher expected utility than A) and is better than A for the insurance company (B
yields higher profits than A).

Wealth in
good state
(probability 1 − p)
W2
NI
W0 45o line

A
W2A

B
zero-profit
line
iso-profit
line

W1 Wealth in
bad state
0 W1A
(probability p)
W0 − `

Figure 5.7: The relative slope of an indifference curve and an isoprofit line.
 124 Chapter 5. Insurance: Part 2

5.2.1 The profit-maximizing contract for a monopolist

Without making use of expected utility theory, we showed in Chapter 2 (Section 2.6.3)
that a profit-maximizing monopolist would offer a full-insurance contract to a potential
customer, at a premium that makes her indifferent between insuring and not insuring. We
can now confirm this result with the tools developed in this chapter. Consider an arbitrary
partial-insurance contract that is acceptable to the potential customer (that is, that lies
on or above the reservation indifference curve), such as point A in Figure 5.7. Such a
contract is not profit maximizing, because the monopolist could replace it with a contract
above the indifference curve through A and below the iso-profit line through A (such as
contract B in Figure 5.7) and (1) the potential customer would be even happier with the
new contract and (2) the monopolist would increase its profits. Since this argument applies
to any partial-insurance contract (that is, to any point above the 45o line), we deduce
that a profit-maximizing monopolist would offer a full-insurance contract.3 Of all the
full-insurance contracts that are acceptable to the potential customer (that is, that are not
below the reservation indifference curve) the one that yields the highest profit to the insurer
is at the intersection of the reservation indifference curve and the 45o line: contract C in
Figure 5.8. The corresponding premium, denoted by hmax , is such that:

U(W0 − hmax ) = pU(W0 − `) + (1 − p)U(W0 ). (5.11)

Wealth in
good state
(probability 1 − p)
W2

reservation
NI indifference 45o line
W0 curve

hmax

W0 − hmax C

W1 Wealth in
W0 − ` W0 − hmax bad state
0 (probability p)

Figure 5.8: Contract C is the full-insurance contract that would be offered by a monopolist.
3 Recall that, at any point on the 45o line, the indifference curve is tangent to the isoprofit line.
 5.2 Back to insurance 125

Recall the definition of risk premium, RL , of a money lottery L: it is the amount by

which the expected value of lottery L can be reduced to leave the individual indifferent
between the amount $(E[L] − RL ) for sure and the lottery itself. Using this definition and
(5.11) it is clear that, since E[NI] = W0 − p`,

hmax = p` + RNI

 
W0 − ` W0
where NI = is the no-insurance lottery. That is, hmax is equal to the
p 1− p
expected loss plus the risk premium of the no-insurance lottery.
1 √
For example, if W0 = 1, 600, ` = 700, p = 10 and U(m) = m then hmax is given by
the solution to the equation
p 1p 9p
1, 600 − h = 1, 600 − 700 + 1, 600
10 10

which is hmax = 79. Hence the risk premium of the NI lottery is

1
RNI = hmax − p` = 79 − 700 = $9.
10

5.2.2 Perfectly competitive industry with free entry

Without making use of expected utility theory, we showed in Chapter 2 (Section 2.6.4)
that, at an equilibrium in a perfectly competitive industry with free entry, all the insurance
firms offer the same contract, namely the full insurance contract with “fair” premium equal
to the expected loss p`. We can now confirm this result with the tools developed in this
chapter.

Recall that a free-entry competitive equilibrium is a situation where

1. each firm in the industry makes zero profits, and

2. there is no unexploited profit opportunity in the industry, that is, there is no currently-
not-offered contract that would attract some custmers and yield positive profit to a
firm that offered that contract.
126 Chapter 5. Insurance: Part 2

By the zero-profit condition (Point 1), any equilibrium contract must be on the zero-
profit line. By the no-profitable-opportunity condition (Point 2), it cannot be a partial-
insurance contract, such as contract A in Figure 5.9, because a new entrant (or an existing
firm) could offer a contract in the region above the indifference curve through point A
and below the iso-profit line through point A, such as contract B in Figure 5.9; such a
contract would induce all those customers who were purchasing contract A to switch to B
and would yield positive profits to the insurance firm offering it. The only contract that is
immune to this is the contract at the intersection of the zero-profit line and the 45o line
(contract D in Figure 5.9).

Wealth in
good state
(probability 1 − p)
W2
NI
W0 45o line
A

B D

zero-profit
line

W1 Wealth in
bad state
0 W0 − ` (probability p)

Figure 5.9: Contract D is the full-insurance contract that would be offered at a free-entry
competitive equilibrium.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 5.5.2 at the end of this chapter.
 5.3 Choosing from a menu of contracts 127

5.3 Choosing from a menu of contracts

It is often the case that insurance companies offer, not just a single contract, but a menu
of contracts and potential customers are allowed to choose which contract to purchase
from this menu. Typically, customers are given a choice between a higher premium with
higher coverage (= lower deductible) or a lower premium with lower coverage (= higher
deductible). The offered menu consists of either a list of contracts or a formula that relates
premium to deductible. In this section we discuss how the potential customer chooses a
contract from a given menu.

5.3.1 Choosing from a finite menu

In the case where the menu consists of a finite list of contracts, the potential customer
will first determine which of the offered contracts is best for her (that is, yields the
highest expected utility) then choose the best contract, provided that it is better than the
no-insurance alternative.
For example, consider an individual whose initial wealth is $1,000. He faces a potential
loss of $400,
√ with probability 20% and has the following vNM utility-of-money function
U(m) = m. Suppose that the insurance company offers the following options:

premium deductible
Contract 1: $82 0
Contract 2: $62 $100
Contract 3: $40 $200

The expected utility of each contract is as follows:

√
Contract 1: 1, 000 − 82 = 30.2985
√ √
Contract 2: 0.2 1, 000 − 162 + 0.8 1, 000 − 62 = 30.2911
√ √
Contract 3: 0.2 1, 000 − 240 + 0.8 1, 000 − 40 = 30.3007

This if he decides to insure, he will choose Contract 3. To see if he does decide to insure
we need to compare the expected utility of the √ best contract, namely√ Contract 3, with
the expected utility of no insurance, which is 0.2 1, 000 − 400 + 0.8 1, 000 = 30.1972.
Since Contract 3 (the best of the three offered contracts) is better than no insurance, he
will purchase Contract 3.
 128 Chapter 5. Insurance: Part 2

5.3.2 Choosing from a continuum of options

Suppose now that the insurance company offers a continuum of options in the form of
a formula relating premium and deductible. For example, consider an individual who is
facing a potential loss of $4,100 and is told by the insurance company that she can choose
any deductible d ∈ [0, 4100]; the corresponding premium h is then calculated according to
the following formula:
1
h = 820 − d. (5.12)
5
Thus the following are some of the many possible contracts that the individual can choose
from:

deductible premium
0 $820
$100 $800
$140 $792
$260 $768
... ...

The set of possible choices is infinite, since any d ∈ [0, 4100] can be chosen by the
individual. Thus we can think of (5.12) as a line, similar to the budget line faced by a
consumer. We shall call it the insurance budget line.
It is useful to translate the line of equation (5.12) – which is expressed in terms of premium
and deductible – into a line in the wealth diagram (W1 ,W2 ). This is easily done by recalling
that h = W0 −W2 and d = W2 −W1 :

 
1 5W0 W1
W0 −W2 = 820 − (W2 −W1 ) , that is, W2 = − 1025 − . (5.13)
5 4 4

For example, if W0 = 6, 000 then (5.13) becomes W0 −W2 = 820 − 15 (W2 −W1 ), that is,

W1
W2 = 6, 475 − . (5.14)
4
Note that the line (in the wealth space) corresponding to equation (5.14) goes through the
no-insurance point NI = (1900, 6000); in fact, replacing W1 with the value 1900 in (5.14)
we get W2 = 6000.
5.3 Choosing from a menu of contracts 129

The insurance budget line of equation (5.14) is shown in Figure 5.10.

Wealth in
good state
(probability 1 − p)
W2

45o line
insurance budget line
W2 = 6475 − 41 W1

NI
6,000
5,180

W1 Wealth in
bad state
0 1,900 5,180
(probability p)

Figure 5.10: The insurance budget line W2 = 6, 475 − W41 .

We will consider insurance budget lines defined by equations of the form

h = a−bd with a > 0, b > 0, d ∈ [0, `] and a − b ` ≥ 0, (5.15)

which – translated into the wealth space (by replacing h with (W0 − W2 ) and d with
(W2 −W1 )) – becomes

W0 − a b
W2 = − W1 . (5.16)
1−b 1−b

If (5.15) is such that a − b ` = 0 then the insurance budget line in the wealth space (defined
by (5.16)) goes through the no-insurance point NI = (W0 − `,W0 ),4 while if a − b ` > 0
then the insurance budget line in the wealth space (defined by (5.16)) goes through a point
vertically below NI.5

4 For example, the zero-profit line falls into this category.

5 For example, an isoprofit line corresponding to a positive level of profit will fall into this category.
130 Chapter 5. Insurance: Part 2

What contract, if any, would the individual choose from the insurance budget line?

We start with the case where the insurance budget line goes through the NI point. There
are three cases to consider.

Case 1: the reservation indifference curve is, at NI, as steep as, or less steep than, the
insurance budget line, as shown in Figure 5.11. It follows that the entire insurance budget
line lies below the reservation indifference curve and thus the individual will choose not
to insure.

Wealth in
good state
(probability 1 − p)
W2

NI 45o line
W0

reservation
insurance indifference curve
budget line

W1 Wealth in
W0 − ` bad state
0 (probability p)

Figure 5.11: Case 1: the insurance budget line lies below the reservation indifference
curve.

In the example above, where W0 = 6, 000, ` = 4, 100 and the insurance budget line is
given by the equation W2 = 6, 475 − W41 , we will be in Case 1 if and only if

 0 
p U (1, 900) 1
0
≤ .
1 − p U (6, 000) 4

19
For instance, if U(m) = ln(m), then Case 1 occurs if and only if p ≤ 259 = 0.0734.
5.3 Choosing from a menu of contracts 131

For Cases 2 and 3 below we assume that the reservation indifference

curve is steeper at NI than the insurance budget line.

Case 2: the indifference curve that goes through the point at the intersection of the 45o
line and the insurance budget line is steeper than, or as steep as, the insurance budget line
at that point, as shown in Figure 5.12.6

Wealth in
good state
(probability 1 − p)
W2

45o line
NI
W0

W1 Wealth in
bad state
0 W0 − `
(probability p)

Figure 5.12: Case 2: the individual chooses full insurance.

In this case, insurance is better than no insurance and the best contract (that is, the
contract that yields the highest expected utility) is the full insurance contract (the point at
the intersection of the 45o line and the insurance budget).
In the example above, where the insurance budget line is given by the equation
W2 = 6, 475 − W41 , we will be in Case 2 if and only if (recall that the slope of any in-
p
difference curve at any point on the 45o line is − 1−p )
p 1 1
≥ i.e. p≥ .
1− p 4 5

6 The indifference curve will be, at that point, as steep as the insurance budget line if and only if the
p
insurance budget line is the zero-profit line, since that slope will be − 1−p . It is steeper if and only if
the insurance budget line is less steep than the zero-profit line, which implies that all the contracts on the
insurance budget line – with the exception of the NI point – yield negative profits; thus it is unlikely that an
insurance company would offer such menu (unless it is subsidized by the government).
132 Chapter 5. Insurance: Part 2

Case 3: the indifference curve that goes through the point at the intersection of the 45o
line and the insurance budget line is less steep than the insurance budget line at that point,
as shown in Figure 5.13.

Wealth in
good state
(probability 1 − p)
W2

45o line
NI
W0

W1 Wealth in
bad state
0 W0 − `
(probability p)

Figure 5.13: Case 3: the individual chooses partial insurance.

In this case, there are partial-insurance contracts on the insurance budget line that are
better than full insurance (and than no insurance). Thus the individual will choose a partial
insurance contract.
Which of the many partial-insurance contracts will she choose? It cannot be a contract
where the indifference curve through it is either steeper, or less steep, than the insurance
budget line: in the former case there would be contracts on the budget line to the right of
that contract that would be better and in the latter case there would be contracts on the
budget line to the left of that contract that would be better. Hence the best contract is the
one at which the slope of the indifference curve at that point is equal to the slope of the
insurance budget line, that is, it is a contract at which the indifference
curve through it is
tangent to the insurance budget line: it is that contract C = W1C ,W2C such that, letting
W2 = α − βW1 be the equation of the budget line (with α > 0 and β > 0),

 0 C 
p U (W1 )
W2C = α − βW1C and = β. (5.17)
1 − p U 0 (W2C )
5.3 Choosing from a menu of contracts 133

The optimal contract (which satisfies (5.17)) is shown in Figure 5.14.

Wealth in
good state
(probability 1 − p)
W2

NI 45o line
W0
C

W1 Wealth in
W0 − ` bad state
0 (probability p)

Figure 5.14: The best contract in Case 3.

In the example above, where W0 = 6, 000, ` = 4, 100 and the insurance budget line is
given by the equation W2 = 6, 475 − W41 , we will be in Case 3 if and only if7
 0 
p U (1, 900) 1 p 1
0
> and < .
1 − p U (6, 000) 4 1− p 4
19
For instance, if U(m) = ln(m), then Case 3 occurs if and only if p > 259 and p < 15 , that
 19 1 
is, if and only if p ∈ 259 , 5 . Fix a value of p in this range and continue to assume that
U(m) = ln(m). Then the optimal contract C = (W1C ,W2C ) is given by the solution to
 
W1 p W2 1
W2 = 6, 475 − and = .
4 1 − p W1 4

For example, if p = 17 then the optimal contract is C = (3700, 5550), that is, the individual
will choose a deductible of 5, 550 − 3, 700 = $1, 850 with a corresponding premium of
6, 000 − 5, 550 = $450.

7 The first inequality says that the reservation indifference curve is steeper at NI than the insurance budget
line. The second inequality says that the indifference curve that goes through the point on the 45o line that
lies on the insurance budget line is less steep at that point than the budget line.
134 Chapter 5. Insurance: Part 2

So far we have assumed that the insurance budget line goes through the no-insurance
point NI. If it does not then, given the assumptions stated in (5.15) on page 129, it will go
through a point vertically below the no-insurance point. There are several possibilities.

A first possibility is that the insurance budget line lies entirely below the reservation
indifference curve (this situation is similar to Case 1 considered above). In such a case
the individual will choose not to insure, since any of the offered contracts yields a lower
expected utility than no insurance.

A second possibility is that there are points on the insurance budget line that are
above the reservation indifference curve, as well as points that are below the reservation
indifference curve. This case can be further subdivided into two sub-cases.
The first sub-case is that there is only one point of intersection between the insurance
budget line and the reservation indifference curve, as shown in Figure 5.15. In this case
the entire segment of the budget line to the right of the intersection point lies above the
reservation indifference curve.

Wealth in
good state
(probability 1 − p)
W2

NI 45o line
W0

W1 Wealth in
W0 − ` bad state
0 (probability p)

Figure 5.15: The case where the insurance budget line does not go through NI.
5.3 Choosing from a menu of contracts 135

In this case,
• if the indifference curve that goes through the point at the intersection of the 45o
line and the insurance budget line is - at that point - steeper than, or as steep as,
the insurance budget line (as shown in the left panel of Figure 5.16), then the best
contract for the individual is the full-insurance contract;
• if the indifference curve that goes through the point at the intersection of the 45o
line and the insurance budget line is - at that point - less steep than the insurance
budget line (as shown in the right panel of Figure 5.16), then the best contract for the
individual is a partial-insurance contract, namely a point at which there is a tangency
between the budget line and the indifference curve that goes through that point (point
C in the right panel of Figure 5.16).

W2 W2
NI 45o NI 45o
W0 line W0 line

W1 W1
0 W0 − ` 0 W0 − `

Figure 5.16: In the case on the left the individual chooses full insurance, in the case on the
right partial insurance.

For example, let W0 = 3, 600, ` = 1, 100 and p = 15 . Suppose that the insurance
company is willing to offer any contact that yields a profit of $10. Then premium and
deductible are related by the equation h − p(` − d) = 10, that is, h = 230 − d5 . Let us
translate it into a line in the wealth space: 3, 600 −W2 = 230 − 15 (W2 −W1 ), that is,
W1
W2 = 4, 212.5 − .
4
This is the equation of the isoprofit line corresponding to a profit-level of 10. Call B the
point of intersection of the budget line and the 45o line. Then the slope of the isoprofit
line is equal to the slope, at point B, of the indifference curve that goes through point B.
Hence we are in the subcase shown in the left panel of Figure 5.16 and the individual will
choose the full-insurance contract, with a premium of $230 (assuming, of course, that
the utility-of-money function is such that the reservation indifference curve crosses the
insurance budget line).
On the other hand, if the insurance budget line is steeper than an isoprofit line, then
we are in the case shown in the right panel of Figure 5.16 and the individual will choose a
partial-insurance contract.
136 Chapter 5. Insurance: Part 2

The second sub-case is that there are two points of intersection between the insurance
budget line and the reservation indifference curve, as shown in Figure 5.17.

Wealth in
good state
(probability 1 − p)
W2

NI 45o line
W0

W1 Wealth in
W0 − ` bad state
0 (probability p)

Figure 5.17: In this case the individual chooses partial insurance.

In this case the individual will choose a partial-insurance contract, namely the contract
on the insurance budget line at which there is a tangency between the budget line and the
indifference curve that goes through that point (shown as point C in Figure 5.17).

We will not consider cases where the insurance budget line does not cover the entire
range [0, `] of possible deductibles and is thus a smaller segment than considered so far; it
should be clear, however, that the method would be the same, namely based on comparing
the relative slopes of indifference curves and the budget line. An example of this is given
in Exercise 5.16.
5.3 Choosing from a menu of contracts 137

We can summarize the discussion of this section as follows. Let

W2 = α − βW1
be the equation of the insurance budget line (with α > 0 and β > 0). To determine whether
the individual will buy insurance and, if so, what contract she will choose, we can proceed
as follows.8
Let B be the full-insurance contract on the insurance budget line (thus W1B = W2B = 1+β
α
)
p
and recall that the slope, at point B, of the indifference curve that goes through B is 1−p
(where, as usual, p denotes the probability of loss).

1. If E[U(B)] ≥ E[U(NI)], then the individual will buy insurance9 and

p
(a) if 1−p ≥ β she will choose the full-insurance contract B;
p
(b) if 1−p < β she will choose partial insurance, namely that contract C =
W1 ,W2C
C

such that
 0 C 
C C p U (W1 )
W2 = α − βW1 and = β.
1 − p U 0 (W2C )

2. If E[U(B)] < E[U(NI)] then

(a) if the insurance budget line does not intersect the reservation indifference
curve, that is, if the there is no solution to the equation
pU(W0 − `) + (1 − p)U(W0 ) = pU(W1 ) + (1 − p)U(α − βW1 )
in the range (W0 − `,W0 ), then the individual will not buy insurance;
(b) if the insurance budget line intersects the reservation indifference curve at two
points, that is, if the there are two solutions to the equation
pU(W0 − `) + (1 − p)U(W0 ) = pU(W1 ) + (1 − p)U(α − βW1 )
in the range (W0 − `,W0 ), then the
individual will choose partial insurance,
namely the contract C = W1C ,W2C such that
 0 C 
C C p U (W1 )
W2 = α − βW1 and = β.
1 − p U 0 (W2C )

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 5.5.3 at the end of this chapter.

8 The reader should convince herself/himself that, indeed, the following steps cover all the cases considered

in this section.
9 Here and elsewhere we are implicitly assuming that if the individual is indifferent between insuring and

not insuring then she will choose to insure.

 138 Chapter 5. Insurance: Part 2

5.4 Mutual insurance

After an exceptional spate of wildfires in the western United States in 2017 and 2018,
several insurance companies notified residents in brush-heavy areas that their homeowner-
insurance policies would not be renewed because the location of their homes posed an
unacceptable fire risk. For example, on August 6, 2018, CBS News reported that10

facing mounting losses, some property insurers are pulling back from selling
policies in California and other western states where wildfire risk is elevated.
In California alone, damages had mounted to at least $12 billion by early
2018.

Clearly, the unavailability of insurance makes a risk-averse individual worse off. Is there
anything that he/she can do mitigate the welfare loss due to the inability to but insurance?
Imagine that Ann and Bob are friends who live in high-risk areas that insurers have
decided to no longer cover. To make things simple, imagine that Ann and Bob have the
same initial wealth $W0 and their homes are of equal value, so that they both face the same
potential loss of $` if a fire occurs; furthermore, assume that they face the same probability
p of a fire occurring. Without insurance they both face the same money lottery, namely
 
W0 W0 − `
NI =
1− p p

One possible course of action for Ann and Bob is to resort to mutual-insurance, that is, to
insure each other by signing the following contract:

I agree to the following:

1. in the event that we both suffer a loss (due to a fire) or in the event
that none of us suffers a loss, then no transfer of money will take place
between us,
2. in the event that only one of us suffers a loss of $` (due to a fire), the
other one (who did not suffer a loss) will give $ 2` to the person who
suffered the loss, that is, will cover 50% of his/her loss.

Would such a contract make them better off relative to no insurance? Let us assume
they leave in locations that are far apart, so that the event of a fire in Ann’s area can
plausibly be treated as independent of the event of a fire in Bob’s area. When two events
are independent, the probability of them jointly occurring is equal to the product of the
individual probabilities. Thus the probabilities can be computed as follows:

event: no fire fire only fire only fire at

at Ann’s at Bob’s both locations
probability: (1 − p)2 p(1 − p) p(1 − p) p2

Thus, each individual will suffers a loss of $ 2` (either in the form of a payment to a less
lucky friend or in the form of a loss of ` followed by a reimbursement, in the amount of 2` ,
10 https://www.cbsnews.com/news/california-wildfires-property-insurers-cancel-policies-because-of-risk/
5.4 Mutual insurance 139

from the luckier friend) with probability 2p(1 − p), so that the contract will give rise to
the following money lottery for each of them (‘MI’ stands for ‘Mutual Insurance’):

W0 − 2`
 
W0 W0 − `
MI =
(1 − p)2 2p(1 − p) p2
Assuming that Ann has vNM preferences over money lotteries and prefers more money to
less (a similar argument applies to Bob), we can represent her preferences by means of a
normalized vNM utility function as follows, with 0 < a < 1:

money: W0 W0 − 2` W0 − `
utility: 1 a 0
Then the expected utilities of the two lotteries are:

E[U(NI)] = 1 − p, and

E[U(MI)] = (1 − p)2 + 2p(1 − p)a = (1 − p)[1 + p(2a − 1)].

Consider first the case where Ann is risk neutral. Then it must be that a = 12 . In fact,
W0 − 2`
   
W0 W0 − `
the lottery and the lottery 1 1 have the same expected value,
1 2 2
namely W0 − 2` and thus Ann must be indifferent between them. The expected utility of the
former is a and the expected utility of the latter is 12 ; thus a = 12 . When a = 12 , (2a − 1) = 0
and thus E[U(NI)] = E[U(MI)] so that Ann does not gain from signing the contract: with
the contract she is as well off as without the contract.

Next consider first the case where Ann is risk averse. Then it must be that a > 21 . In
W0 − 2`
 
fact, the lottery , whose expected utility is a, gives for sure the expected value
1
 
W0 W0 − `
of the lottery 1 1 , whose expected utility is 12 . By definition of risk aversion,
2 2
Ann prefers the former lottery to the latter. Hence a > 21 . When a > 12 , (2a − 1) > 0 and
thus (1 − p)[1 + p(2a − 1)] > 1 − p so that E[U(MI)] > E[U(NI)], that is, Ann is better
off with the contract than without the contract.
Thus we have shown that two risk-averse individuals can make themselves better off by
signing a mutual-insurance agreement, according to which they share the losses equally .
The astute reader will have realized that there was no need for a detailed proof of
the fact that a risk-averse individual will prefer lottery MI to lottery NI, since it follows
from the analysis of Chapter 4 (Section 4.4.2): lottery NI can be obtained from lottery
MI by means of a mean-preserving spread. First of all, the reader should verify that
E[NI] = E[MI] = W0 − p`. Secondly, by taking the probability of outcome W0 − 2` ,

namely 2p(1 − p), and spreading it equally to each of the other outcomes we obtain lottery
NI; in fact, (1 − p)2 + 21 2p(1 − p) = 1 − p and p2 + 12 2p(1 − p) = p.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 5.5.4 at the end of this chapter.
 140 Chapter 5. Insurance: Part 2

5.5 Exercises

The solutions to the following exercises are given in Section 5.6 at the end of this chapter.

5.5.1 Exercises for Section 5.1: Binary lotteries and indifference curves

 
$x $y
Exercise 5.1 Consider all the lotteries of the form 1 2 with x ≥ 0 and y ≥ 0.
3 3
Assume that the individual in question is risk neutral.
(a) Write the equation of the indifference curve that goes through point
(x = 6, y = 10).
(b) Write the equation of the indifference curve that goes through point (10,8).
(c) Write the equation of the indifference curve that goes through point (4,9).


 
$x $y
Exercise 5.2 Consider all the lotteries of the form 1 2 with x ≥ 0 and y ≥ 0.
3 3
Consider an individual with von Neumann-Morgenstern utility-of-money function
U(m) = ln(m).  
$10 $40
(a) Calculate the expected utility of lottery A = 1 2 .
3 3
 
$10 $10
(b) Calculate the expected utility of lottery B = 1 2 .
3 3

(c) Calculate the slope of the indifference curve at point A = (10, 40).
(d) Calculate the slope of the indifference curve at point B = (10, 10).
(e) In the (x, y)-plane draw the indifference curve that goes to point A = (10, 40) and
the indifference curve that goes to point B = (10, 10).


Exercise 5.3 Repeat Parts (a)-(e) of the previous question for the case of an individual
who is risk neutral. 
 5.5 Exercises 141
 
$x $y
Exercise 5.4 Consider all the lotteries of the form 2 1 with x ≥ 0 and y ≥ 0.
3 3
Consider an individual with von Neumann-Morgenstern utility-of-money function
U(m) = ln(m).
(a) Write an equation whose solutions give the setof lotteries that the individual
$4 $4
considers just as good as lottery A = 2 1 .
3 3

(b) Solve the equation of Part (a) and obtain a function y = f (x) whose graph is the
indifference curve that goes through point (4, 4).
(c) Write an equation whose solutions give the setof lotteries that the individual
$9 $4
considers just as good as lottery B = 2 1 .
3 3

(d) Solve the equation of Part (c) and obtain a function y = g(x) whose graph is the
indifference curve that goes through point (9, 4).
(e) Calculate the slope of the indifference curve at point A = (4, 4).
(f) Calculate the slope of the indifference curve at point B = (9, 4).


 
$x $y
Exercise 5.5 Consider all the lotteries of the form 1 4 with x ≥ 0 and y ≥ 0.
5 5
Let A = (100, 25), B = (4, 49) and C = (40, 40).
(a) Draw the indifference curves that go through points A, B and C for an√individual
with von Neumann-Morgenstern utility-of-money function U(m) = m.
(b) Draw the indifference curves that go through points A, B and C for a risk-neutral
individual.


5.5.2 Exercises for Section 5.2: Back to insurance

1
Exercise 5.6 Adam’s current wealth is $80, 000. With probability 20 he faces a loss of
$30, 000. His vNM utility-of-money function is U(m) = ln(m).
(a) Calculate the slope of Adam’s reservation indifference curve at the no-insurance
point NI.
(b) Calculate the slope of the iso-profit curve that goes through point NI.
(c) Calculate the maximum premium that Adam is willing to pay for full insurance.
(d) Calculate the increase in Adam’s utility relative to no insurance if he obtains full
insurance at the “fair” premium (that is, at a premium that yields zero profits to
the insurer).
(e) Consider contract A = (80, 000 − h, 80, 000 − h). Calculate the slope at point A
of Adam’s indifference curve that goes through point A.

142 Chapter 5. Insurance: Part 2

1
Exercise 5.7 Frank has a wealth of $W0 . With probability p = 10 he faces a loss of
$`. The maximum he is willing to pay for full insurance is $800. The risk premium
associated with the lottery corresponding to no insurance is $500.
(a) What is the value of `?
(b) What is the maximum profit that a monopolist can make by selling insurance to
Frank?


Exercise 5.8 Bob owns a house. The value of the land is $75, 000 while the value of
the building is $110, 000. The rest of his wealth consists of the balance of his bank
account, which is $10, 000. Thus his current wealth is $195, 000. Bob lives in an area
where there is a 5% probability that a fire will completely destroy his house during any
year (while the land will not be affected by a fire). Bob’s utility function is given by:

U(m) = 800 − (20 − m)2

where m ∈ [0, 20] denotes money measured in $10, 000 (thus, for example, m = 11
means $110, 000).
(a) What is Bob’s expected loss if he does not insure?
(b) What is Bob’s expected wealth if he does not insure?
(c) What is Bob’s expected utility if he does not insure?
(d) What is Bob’s expected utility if he purchases an insurance contract with premium
$1, 200 and deductible $20, 000?
(e) What is the slope of Bob’s reservation indifference curve at the no-insurance
point?
(f) Let A be the point in the wealth diagram that corresponds to the insurance contract
with premium $1, 200 and deductible $20, 000. What is the slope, at point A, of
Bob’s indifference curve that goes through point A?
(g) Does the indifference curve that goes through point A of Part (f) lie above or
below the reservation indifference curve?
(h) What is the maximum premium that Bob would be willing to pay for full insur-
ance?


Exercise 5.9 Beth’s vNM utility-of-money function is U(m) = α − β e−m , where α

d x
and β are positive constants. [Recall that e ≈ 2.71828 and dx e = ex .]
(a) What is Beth’s attitude to risk?
(b) What is Beth’s Arrow-Pratt measure of absolute risk aversion?
(c) Show that if Beth’s initial wealth is W0 and she is faced with a potential loss
` with probability p, the maximum premium that she is willing to pay for full
insurance is the same whatever her initial wealth, that is, it is independent of W0 .

 5.5 Exercises 143

5.5.3 Exercises for Section 5.3: Choosing from a menu of contracts

Exercise 5.10 Barbara has a wealth of $80,000 and faces a potential loss of $20,000
√
with probability 10%. Her utility-of-money function is U(m) = m . An insurance
company offers her the following menu of contracts:
premium deductible
Contract 1: $2,340 $500
Contract 2: $2,280 $1,000
Contract 3: $2,220 $1,500
Contract 4: $2,160 $2,000

(a) What is Barbara’s expected utility if she does not insure?

(b) For each contract calculate the corresponding expected utility and determine
which contract, if any, Barbara will choose.


Exercise 5.11 You have the following vNM utility-of-money function: U(m) = ln(m).
Your initial wealth is $10,000 and you face a potential loss of $4,000 with probability
1
6 . An insurance company offers you the following menu of choices: if you choose
deductible d (with 0 ≤ d ≤ 4, 000) then your premium is h = 800 − 0.2 d.
(a) Translate the equation h = 800 − 0.2d into an equation in terms of wealth levels.
(b) Compare the slope of the reservation indifference curve at the no-insurance point
NI to the slope of the insurance budget line. Are there contracts that are better for
you than no insurance?
(c) Which contract will you choose from the menu?
(d) Compare expected utility if you do not insure with expected utility if you purchase
the best contract from the menu.
(e) What is the insurance company’s expected profit from the contract of Part (c)?
(f) Prove the result of Part (c) directly by expressing expected utility as a function of
the deductible d and by maximizing that expression.


√
Exercise 5.12 You have the following vNM utility-of-money function: U(m) = m.
1
Your initial wealth is $576 and you face a potential loss of $176 with probability 16 . An
insurance company offers you the following menu of choices: if you choose deductible
d (with 0 ≤ d ≤ 176) then your premium is h = 19 (176 − d).
(a) Translate the equation h = 91 (176 − d) into an equation in terms of wealth levels.
(b) Compare the slope of the reservation indifference curve at the no-insurance point
NI to the slope of the insurance budget line. Are there contracts that are better for
you than no insurance?

144 Chapter 5. Insurance: Part 2

Exercise 5.13 [Note: in this exercise the data is the same as in Exercise 5.12, but we
1
have changed the probability of loss from to 17 .]
16
√
Your vNM utility-of-money function is U(m) = m; your initial wealth is $576 and
you face a potential loss of $176 with probability 17 . An insurance company offers you
the following menu of choices: if you choose deductible d (with 0 ≤ d ≤ 176) then
your premium is h = 91 (176 − d).
(a) Translate the equation h = 91 (176 − d) into an equation in terms of wealth levels.
(b) Compare the slope of the reservation indifference curve at the no-insurance point
NI to the slope of the insurance budget line. Are there contracts that are better for
you than no insurance?
(c) Which contract will you choose from the menu?
(d) Compare expected utility if you do not insure with expected utility if you purchase
the best contract from the menu.
(e) What is the insurance company’s expected profit from the contract of Part (c)?
(f) Confirm the result of Part (c) by expressing expected utility as a function of the
deductible d and by finding the maximum of that function in the interval [0, 400].


Exercise 5.14 David’s vNM utility-of-money function is U(m) = 1 − (m + 1)−1 ,

where m is money measured in thousands of dollars (thus, for example, m = 6 means
$6, 000).
(a) What is David’s attitude to risk?
(b) Calculate the Arrow-Pratt measure of absolute risk aversion for David.
David’s initial wealth is $8,000 and he faces a potential loss of $3,000 with probability
1
10 . An insurance company offers him insurance at the following terms:

choose the amount (of your potential loss) that you would like to be covered
(thus a number in the range from 0 to 3,000); for every dollar of coverage
you will pay $γ as premium (with 0 < γ < 1).

(c) Write an equation that expresses the premium in terms of the deductible. [Recall
that the deductible is that part of the loss that is not covered.].
(d) Translate the equation of Part (c) into an equation in terms of David’s wealth
levels.
(e) Does the insurance budget line of Part (d) go through the no-insurance point NI?
(f) For what values of γ will David choose not to insure?
(g) For what values of γ will Davis purchase full insurance?
(h) Assuming that the value of γ is in the range where David chooses partial insurance,
write a system of two equations whose solution gives the contract that David will
choose.

5.5 Exercises 145

Exercise 5.15 Anna’s vNM utility-of-money function is U(m) = ln(m). Her initial
wealth is $3,600 and she faces a potential loss of $2,700 with 25% probability. An
insurance company is offering Anna any contract such that premium h and deductible d
3
satisfy the following equation: h = 810 − 10 d.
3
(a) Translate the equation h = 810 − 10 d into an equation in terms of wealth levels.
(b) Does the equation found in Part (a) correspond to an isoprofit line?
(c) Does the insurance budget line of Part (a) go through the no-insurance point?
(d) Are there any contracts on the insurance budget line that Anna prefers to no
insurance?
(e) What is the best contract on the insurance budget line for Anna?
(f) Calculate Anna’s expected utility at the following points:
(1) NI (no insurance),
(2) the full-insurance contracts that belongs to the budget line, and
(3) the contract found in Part (e).
(g) Prove the result of Part (e) directly by expressing expected utility as a function of
the deductible d and by maximizing that expression.


Exercise 5.16 Kate has an initial wealth of W0 = $1, 600 and faces a potential loss
of ` = $576 with probability
√ 20%. Her von Neumann-Morgenstern utility-of-money
function is U(m) = m. An insurance company is offering the following menu of
contracts:
3
h = 152 − d.
10
However, the deductible is restricted to the interval [0, 360] (hence the largest deductible
is 360 rather than 576).
Let A = W1A ,W2A be the point in wealth space corresponding to the contract

(h = 44, d = 360).
3
(a) Translate the insurance budget line h = 152 − 10 d (d ∈ [0, 360]) into a budget
line in wealth space (give the range of values).
(b) Does contract A lie on, below or above the reservation indifference curve?
(c) Calculate the slope, at point A, of the indifference curve that goes through A,
compare it to the slope of the insurance budget line and deduce which contract
from the offered menu will be chosen by Kate.

 146 Chapter 5. Insurance: Part 2

5.5.4 Exercises for Section 5.4: Mutual insurance

Exercise 5.17 Carla and Don have the same initial wealth, namely $32,400, and face
the same potential loss, namely $18,000, with the same probability, namely 15 .
Carla’s vNM utility-of-money function is
√
UC (m) = m

while Don’s is
1
UD (m) = 1 − m .
10,000 +1
They are unable to obtain insurance on the market, so they have decided to write a
mutual insurance contract, according to which any losses are shared equally between
them.
Assuming that the event that Carla suffers a loss is independent of the event that Don
suffers a loss, show that signing the mutual insurance contract has made each of them
better off relative to no insurance. 

Exercise 5.18 Ann, Carla and Dana have the same initial wealth, namely $40,000,
and face the same potential loss, namely $30,000, with the same probability, namely
1
5 . They have the √same vNM preferences, represented by the vNM utility-of-money
function U(m) = m. They are unable to obtain insurance on the market and have
decided to sign a mutual insurance contract, according to which any losses suffered by
any of them will be shared equally by all three of them. Assume that the event that any
of them suffers a loss is independent of the event(s) that the other(s) also suffer a loss.
(a) Calculate the probabilities of the following events:
1. All three of them suffer a loss.
2. Exactly two of them suffer a loss.
3. Exactly one of them suffers a loss.
4. None of them suffers a loss.
(b) Show that each of them is better off with the mutual insurance contract relative to
no insurance.

 5.6 Solutions to Exercises 147

Exercise 5.19 In this exercise we consider the benefit of mutual insurance when
independence fails to hold: one person suffering a loss makes it more likely that the
other person would also suffer a loss.
We consider two individuals, Albert an Ben, who have the same initial wealth,
namely $40,000, and face the same potential loss of $30,000 due to wildfire, with the
same probability, namely 15 . They have the√ same vNM preferences, represented by the
vNM utility-of-money function U(m) = m.
Since they live not far apart from each other if a fire occurs at one of the two
properties then the probability that there will be a fire at the other property is greater
than 15 . If the events were independent then the probabilities would be as follows:

fire at both fire at one only no fires

1 8 16
25 25 25

However, due to correlation, the probabilities are as follows:

fire at both fire at one only no fires

3 7 20
30 30 30

Suppose that Albert and Ben are unable to obtain insurance on the market. Are they
better off with no insurance or with a mutual insurance agreement (according to which
any losses suffered by either of them will be shared equally by both)? 

5.6 Solutions to Exercises

Solution to Exercise 5.1

1
The slope of every indifference curve is − 3
2 = − 21 . Thus the equation of any indifference
3
curve is of the form y = a − 12 x.
(a) To find the value of a for the indifference curve that goes through point (6,10) solve
the equation 10 = a − 12 6 to get a = 13. Thus the equation of the indifference curve
that goes through point (6,10) is y = 13 − 21 x.
(b) To find the value of a for the indifference curve that goes through point (10,8) solve
the equation 8 = a − 12 10 to get a = 13. Thus the equation of the indifference curve
that goes through point (10,8) is y = 13 − 12 x. Hence the two lotteries (6,10) and
(10,8) lie on the same indifference curve; indeed, they have the same expected value,
namely 26 3.
(c) To find the value of a for the indifference curve that goes through point (4,9) solve
the equation 9 = a − 12 4 to get a = 11. Thus the equation of the indifference curve
that goes through point (4,9) is y = 11 − 21 x. 
148 Chapter 5. Insurance: Part 2

Solution to Exercise 5.2  

$10 $40
(a) The expected utility of lottery A = 1 2 is 13 ln(10) + 23 ln(40) = 3.227.
3 3
 
$10 $10
(b) The expected utility of lottery B = 1 2 is 13 ln(10) + 23 ln(10) = 3.303.
3 3

(c) The indifference curve that goes through A = (10, 40) is the set of lotteries that
yield an expected utility of 3.227. It is a convex curve since the utility function U
incorporates risk aversion. The slope of the indifference curve at point A is equal to
1
!
 0  1
p U (10) 3 10
− =− 2 = −2.
1 − p U 0 (40) 3
1
40

(d) Similarly, the indifference curve that goes through B = (10, 10) is the set of lotteries
that yield an expected utility of 2.303. It is a convex curve. The slope of the
indifference curve at point B is equal to
1 1
p U 0 (10) 1
− 0
= − 32 10
1
=− .
1 − p U (10) 3 10
2

(e) The two indifference curves are shown in Figure 5.18. 

y
(probability 32 )

A
40

B
10
x
0 (probability 13 )
10
slope: − 12 slope: −2

Figure 5.18: The graph for Part (e) of Exercise 5.2.

5.6 Solutions to Exercises 149

Solution to Exercise 5.3

Since the individual is risk neutral, we can take his/her utility-of-money function to be
the identity function U(m) = m. Thus the expected utility of a lottery coincides with the
expected value.  
$10 $40
(a) The expected utility (= value) of lottery A = 1 2 is 13 10 + 23 40 = 30.
3 3
 
$10 $10
(b) The expected utility of lottery B = 1 2 is 13 10 + 23 10 = 10.
3 3

(c) The indifference curve that goes through A = (10, 40) is the set of lotteries that yield
an expected utility (= value) of 30. It is a straight line because of risk neutrality. The
slope of the indifference curve at point A is equal to
 0  1  
p U (10) 3 1 1
− =− =− .
1 − p U 0 (40) 2
3
1 2

(d) Similarly, the indifference curve that goes through B = (10, 10) is the set of lotteries
that yield an expected utility (= value) of 10. It is a straight line because of risk
neutrality. The slope of the indifference curve at point B is equal to
 0  1
p U (10) 3 1
− =− 1=− .
1 − p U 0 (10) 2
3
2

(e) The two indifference curves are shown in Figure 5.19. 

y
(probability 23 )

A
40

B
10 slope: − 12
x
0 (probability 13 )
10

slope: − 12

Figure 5.19: The graph for Part (e) of Exercise 5.3.

150 Chapter 5. Insurance: Part 2

Solution to Exercise 5.4

2 1
(a) 3 ln(x) + 3 ln(y) = ln(4).

(b) First of all, rewrite the above equation as 2 ln(x) + ln(y) = 3 ln(4), from which we get
 2
that e(2 ln(x)+ln(y)) = e3 ln(4) . Now, e(2 ln(x)+ln(y)) = e2 ln(x) eln(y) = eln(x) y = x2 y.
Similarly, e3 ln(4) = 43 = 64. Thus the equation becomes x2 y = 64 from which we
64
get y = 2 .
x

2 1 2 1
(c) 3 ln(x) + 3 ln(y) = 3 ln(9) + 3 ln(4).

(d) Repeating the steps of Part (b): from 2 ln(x) + ln(y) = 2 ln(9) + ln(4) we get
324
e(2 ln(x)+ln(y)) = e(2 ln(9)+ln(4)) , which becomes x2 y = 92 4 which yields y = 2 .
x

(e) The slope of the indifference curve at point A = (4, 4) is equal to11

 0  2
p U (4) 3
− =− 1 = −2.
1 − p U 0 (4) 1
3

(f) The slope of the indifference curve at point B = (9, 4) is equal to12

2 1
 0  !
p U (9) 3 9 8
− =− =− .
1 − p U 0 (4) 1
3
1
4
9

11 Alternatively, using the function f (x) = 64

x2
of Part (b), f 0 (x) = − 128
x3
so that f 0 (4) = − 128
43
= −2.

12 Alternatively, using the function g(x) = 324

x2
of Part (d), g0 (x) = − 648
x3
so that g0 (9) = − 648
93
= − 89 .
5.6 Solutions to Exercises 151

Solution to Exercise 5.5

√ √ √ √
(a) E[U(A)] = 15 100 + 45 25 = 2 + 4 = 6 and E[U(B)] = 15 4 + 54 49 = 6. Thus A
√
and B lie on the same indifference curve. On the other hand, E[U(C)] = 40 =
6.3246. Thus C lies on a higher indifference curve. See Figure 5.20.13

(b) The expected value of A, B and C is the same, namely 40. Thus the three points lie
on the same indifference curve, which is a straight line with slope − 14 . 

y
(probability 45 )

B
49
C
35

A EU = 6.3246
25 EU = 6 x
0 (probability 15 )
4 60 100

Figure 5.20: The graph for Exercise 5.5.

13 Note that the scale of the axes has been distorted to make the qualitative properties of the graph easier to
see.
152 Chapter 5. Insurance: Part 2

Solution to Exercise 5.6

d 1
(a) NI = (50000, 80000) and dm ln(m) = m . Thus the slope of the reservation indiffer-
ence curve at NI is

1
!
1
20 50,000 8
− 19 1
=− = −0.0842.
20 80,000
95

1
20 1
(b) The slope of every isoprofit line is − 19 = − 19 = 0.0526.
20
(c) The maximum premium that Adam is willing to pay for full insurance is given by
the solution to the equation

1 19
ln(80, 000 − h) = ln(50, 000) + ln(80, 000)
20 20

which is hmax = $1, 858.10

1
(d) The fair premium is equal to 20 (30, 000) = 1, 500. If Adam does not insure, his
1 19
expected utility is 20 ln(50, 000) + 20 ln(80, 000) = 11.2663. If Adam obtains full
insurance at premium $1, 500, his utility is ln(80, 000 − 1, 500) = 11.2709. Thus the
increase in utility is 11.2709 − 11.2663 = 0.0046.
(e) The slope of any indifference curve at any point on the 45o line is equal to the slope
1
1
of any isoprofit line, namely − 20
19 = − 19 = 0.0526. 
20

Solution to Exercise 5.7

 
W0 W0 − `
(a) If Frank does not buy insurance he faces the lottery 9 1 whose expected
10 10
1
value is W0 − 10 `. If hmax is the maximum premium that he is willing to pay for
9
full insurance [that is, hmax is the solution to the equation U(W0 − h) = 10 U(W0 ) +
1
10 U(W0 − `)] and RNI is the risk premium associated with the no-insurance lottery
1 9 1
[that is, RNI is the solution to the equation U(W0 − 10 ` − R) = 10 U(W0 ) + 10 U(W0 −
1
`)], then hmax = 10 ` + RNI . Thus we have the following equation

1
800 = ` + 500,
10

whose solution is ` = 3, 000.

(b) The monopolist would sell Frank the full-insurance contract with premium $800 and
1
thus make an expected profit equal to 800 − 10 3, 000 = $500. 
5.6 Solutions to Exercises 153

Solution to Exercise 5.8

5
(a) Bob’s expected loss is 100 110, 000 = $5, 500.

(b) Bob’s expected wealth if he does not insure is 195, 000 − 5, 500 = $189, 500.

(c) Bob’s expected utility if he does not insure is

5   95 
800 − (20 − 8.5)2 + 800 − (20 − 19.5)2 = 793.15.

100 100

(d) The insurance contract with premium $1, 200 and deductible $20, 000 corresponds to
the following point in the wealth diagram: (173,800, 193,800). The corresponding
expected utility is

5   95 
800 − (20 − 17.38)2 + 800 − (20 − 19.38)2 = 799.29.

100 100

(e) U 0 (m) = 40 − 2m. Thus the slope of Bob’s reservation indifference curve at the
no-insurance point is

5  
100 40 − 2(8.5)
− 95 = −1.2105.
100
40 − 2(19.5)

(f) From Part (d) we have that A = (173, 800, 193, 800). The slope, at point A, of Bob’s
indifference curve that goes through point A is

 
5 40 − 2(17.38)
− = −0.2224.
95 40 − 2(19.38)

(g) Since E[U(A)] = 799.29 > 793.15 = E[U(NI)], the indifference curve that goes
through point A lies above reservation indifference curve.

(h) The maximum premium that Bob would be willing to pay for full insurance is given
by the solution to the equation U(W0 − h) = E[U(NI)], that is,
800 − [20 − (19.5 − h)]2 = 793.15, which is 2.1173, that is, $21, 173 (slightly less
than four times the expected loss). 
154 Chapter 5. Insurance: Part 2

Solution to Exercise 5.9

(a) U 0 (m) = β e−m > 0 and U 00 (m) = −β e−m < 0, thus Beth is risk averse.
00
(m)
(b) A(m) = − UU 0 (m) = 1, a constant.

(c) The maximum premium hmax is determined by the solution to

U(W0 − h) = pU(W0 − `) + (1 − p)U(W0 ), that is,

 
α − β e(−W0 +h) = p α − β e(−W0 +`) + (1 − p) α − β e−W0


h i
= pα + (1 − p)α − β pe(−W0 +`) + (1 − p)e−W0
h i
−W0 ` −W0
= α − β pe e + (1 − p)e
h i
= α − β e−W0 pe` + 1 − p .

Subtracting α from both sides and multiplying by − β1 we get

h i
e−W0 eh = e−W0 pe` + 1 − p ,

and multiplying both sides by eW0 , we are left with

eh = pe` + 1 − p.

Thus hmax = ln(pe` + 1 − p), independent of W0 . 

Solution to Exercise 5.10

√ √
(a) Barbara’s expected utility if she does not insure is: 0.9 80, 000 + 0.1 60, 000 =
279.053.
(b) Barbara’s
√ expected utility√from a contract with premium h and deductible d is
0.1 80, 000 − h − d + 0.9 80, 000 − h. Thus,

premium deductible expected utility

√ √
Contract 1: 2, 340 500 0.1 77, 160 + 0.9 77, 660 = 278.5856
√ √
Contract 2: 2, 280 1, 000 0.1 76, 720 + 0.9 77, 720 = 278.6031
√ √
Contract 3: 2, 220 1, 500 0.1 76, 280 + 0.9 77, 780 = 278.6204
√ √
Contract 4: 2, 160 2, 000 0.1 75, 840 + 0.9 77, 840 = 278.6375.

None of the contracts gives her higher expected utility than no insurance. Hence she
will not buy insurance. 
5.6 Solutions to Exercises 155

Solution to Exercise 5.11

(a) Letting h = 10, 000 −W2 and d = W2 −W1 we get 10, 000 −W2 = 800 − 0.2(W2 −
W1 ), that is
1
W2 = 11, 500 − W1 .
4
(b) From Part (a) we have that the slope of the insurance budget line is − 14 . The slope
of the reservation indifference curve at NI = (6000, 10000) is
1 1
!
6 6,000 1
− 5 1
= − .
6 10,000
3

Thus the reservation indifference curve is steeper, at NI, than the insurance budget
line and, therefore, there are contracts that are better than no insurance.
(c) Since the slope of any indifference curve at any point on the 45o line is equal to
1
6
− 5 = − 51 , the indifference curve that goes through the point at the intersection of
6
the insurance budget line and the 45o line is less steep than the insurance budget line
and thus we are in Case 3 of Section 5.3.2 and the optimal contract is given by the
solution to the following two equations:
1 1
!
1 6 W1 1
W2 = 11, 500 − W1 and − 5 1
=−
4 6 W
4
2

which is W1 = 7, 666.67 and W2 = 9, 583.33. Thus the chosen deductible is

d = 9, 583.33 − 7, 666.67 = $1, 916.66
and the corresponding premium is
h = 10, 000 − 9, 583.33 = $416.67.
(d) Expected utility from no insurance is 16 ln(6, 000) + 56 ln(10, 000) = 9.1252, while
expected utility from the contract of Part (c) is 16 ln(7, 666.67) + 65 ln(9, 583.33) =
9.1306.
(e) Expected profits from the contract of Part (c) is 416.67 − 16 (4, 000 − 1, 916.66) =
$69.45.
(f) Expected utility from contract (h, d) is: 16 ln(10, 000 − h − d) + 56 ln(10, 000 − h).
Replacing h with 800 − 0.2d we get the following function:
f (d) = 16 ln(10, 000 − 800 + 0.2d − d) + 65 ln(10, 000 − 800 + 0.2d)
= 16 ln(9, 200 − 0.8d) + 56 ln(9, 200 + 0.2d).
To maximize this function we must solve the equation f 0 (d) = 0, that is
   
1 1 5 1
(−0.8) + 0.2 = 0
6 9, 200 − 0.8d 6 9, 200 + 0.2d

The solution is d = 5,750 3 = 1, 916.66 with corresponding premium

h = 800 − 0.2(1, 916.67) = 416.67, confirming the conclusion of Part (c). 
156 Chapter 5. Insurance: Part 2

Solution to Exercise 5.12

(a) Letting h = 576 −W2 and d = W2 −W1 we get 576 −W2 = 19 (176 −W2 +W1 ), that
is
1
W2 = 626 − W1 .
8
(b) From Part (a) we have that the slope of the insurance budget line is − 18 . The slope
of the reservation indifference curve at NI = (400, 576) is

√1
!
1
2 400 2
− 16 = − .
15 √1 25
16 2 576

Thus the reservation indifference curve is less steep, at NI, than the insurance budget
line and, therefore, the insurance budget line lies below the reservation indifference
curve, that is, there are no contracts that are better than no insurance. Thus we are in
Case 1 of Section 5.3.2 and your best decision is not to insure. 

Solution to Exercise 5.13

(a) The answer is, of course, the same as in Exercise 5.12: W2 = 626 − 81 W1 .
(b) The slope of the insurance budget line is − 18 . The slope of the reservation indiffer-
ence curve at NI = (400, 576) is

1 √1
!
7 2 400 1
− 6 1
=− .
7
√ 5
2 576

Thus the reservation indifference curve is steeper, at NI, than the insurance budget
line and, therefore, there are contracts that are better than no insurance.
(c) Since the slope of any indifference curve at any point on the 45o line is equal to
1
− 76 = − 16 , the indifference curve that goes through the point at the intersection of
7
the insurance budget line and the 45o line is steeper than the insurance budget line
and thus we are in Case 2 of Section 5.3.2 and the optimal contract is given by the
full-insurance contract (the point of intersection between the insurance budget line
and the 45o line), that is, by the solution to the following two equations:

1
W2 = 626 − W1 and W2 = W1
8

which is W1 = W2 = 5,008 9 = 556.44. Thus the chosen deductible is zero and the
premium is h = 576 − 556.44 = $19.56.√ √
(d) Expected utility from no insurance is 17 400 + 67 576 = 23.4286, while expected
√
utility from the contract of Part (c) is 556.44 = 23.589.
(e) Expected profits from the contract of Part (c) is 19.56 − 71 400 = −$37.58, thus a
loss.
5.6 Solutions to Exercises 157
√ √
(f) Expected utility from contract (h, d) is: 17 576 − h − d + 76 576 − h. Replacing h
with 91 (176 − d) we get the following function:
q q
f (d) = 71 576 − 176
9 + 1
9 d − d + 6 176 1
7 576 − 9 + 9 d
q q
= 7 556.44 − 9 d + 7 556.44 + 19 d.
1 8 6

We know from Part (c) that the maximum of the function f (d) is achieved at a corner
(that is, not in the interior of the interval [0, 400]) and thus it cannot be found by
solving the equation f 0 (d) = 0 (indeed, the solution to this equation is −398.36
which is outside the interval [0, 400]). Another way to see that the solution is at a
corner, is to calculate the value f 0 (d) at d = 0: f 0 (0) = −0.0007 < 0, indicating that
increasing the deductible from 0 reduces expected utility. 

Solution to Exercise 5.14

(a) The utility function is U(m) = 1− m+11
. Thus U 0 (m) = (m+1)
1 00 2
2 and U (m) = − (m+1)3 .

Since U 00 (m) < 0 (given that m ≥ 0), David is risk averse.

(m) 00
(b) The Arrow-Pratt measure of risk aversion is AU (m) = − UU 0 (m) = 2
m+1 .

(c) Since coverage = loss − deductible, the equation is h = γ(` − d) = γ(3000 − d).
(d) Replacing h with W0 −W2 = 8, 000 −W2 and d = (W2 −W1 ) in the equation of Part
(c) we get
8, 000 − 3, 000γ γ
W2 = − W1 .
1−γ 1−γ
(e) Yes: replacing W1 with 5, 000 (= W0 − ` = 8, 000 − 3, 000, the horizontal coordinate
of NI) in the above equation we get W2 = 8, 000 (the vertical coordinate of NI).
(f) David will choose not to insure when the insurance budget line is steeper than the
reservation indifference curve at the NI point, that is, when (recall that money is
measure in thousands of dollars)
 0 
1 (8 + 1)2
 
γ p U (5) 9 1
> = = = .
1−γ 1 − p U 0 (8) 9 (5 + 1)2 36 4
1
Solving γ
1−γ > 4 we get γ > 15 .
(g) David will choose full insurance when the slope of the indifference curve at the
p
offered full-insurance contract (which is − 1−p = − 91 ) is, in absolute value, greater
than or equal to the slope of the budget line (which is − 1−γγ
). Solving 19 ≥ 1−γ
γ
we
1
get γ ≤ 10 .
(h) David will choose partial insurance when
1. he prefers insurance to no insurance (that is, as seen in Part (f), when γ < 51 )
and
2. the slope of the indifference curve at the offered full-insurance contract is, in
absolute value, less than the slope of the budget line (that is, as seen in Part (g),
1
when γ > 10 ).
158 Chapter 5. Insurance: Part 2

Thus David will choose partial insurance when

1 1
<γ < .
10 5

1
Assume that
10 < γ < 51 . Then David will choose that contract
C C
C = W1 ,W2 at which there is a tangency between the budget line and the in-
difference curve through C, that is, C must satisfy the following equations (recall
that wealth levels are expressed in dollars while the argument of the utility function
is expressed in thousands of dollars):

 2 
W2C
8, 000 − 3, 000γ γ 1 +1 
1,000 γ
W2 = − W1 and 2
= .
1−γ 1−γ 9 1−γ
 C  
W1
1,000 + 1

Solution to Exercise 5.15

(a) Replacing h with W0 − W2 = 3, 600 − W2 and d with (W2 − W1 ) in the equation
3
h = 810 − 10 d we get
27, 900 3
W2 = − W1 .
7 7
1
p 4
(b) No, because the slope of an isoprofit line is − 1−p =− 3 = − 31 6= − 37 .
4
(c) Yes: replacing W1 in the equation of Part (a) with 900 (= W0 − ` = 3, 600 − 2, 700,
the horizontal coordinate of NI) we get W2 = 3, 600 (= W0 , the vertical coordinate
of NI).
(d) Yes, because the absolute value of the slope of the reservation indifference curve at
NI, namely 13 3,600 4 0 1
900 = 3 (recall that U(m) = ln(m) and U (m) = m ) is greater than
the absolute value of the slope of the insurance budget line, namely 37 .
(e) Since the slope of the indifference curve at the full-insurance contract is − 13 , which
is less – in absolute value – than the slope of the budget line (in absolute value),
the best contract is a partial-insurance contract. It is found by solving the following
equations:
 
27, 900 3 1 W2 3
W2 = − W1 and = .
7 7 3 W1 7

20,925
The solution is: W1 = 2, 325 and W2 = 7 = 2, 989.29. Thus the premium is h =
3, 600 − 2, 989.29 = 610.71 and the deductible is d = 2, 989.29 − 2, 325 = 664.29.
5.6 Solutions to Exercises 159

(f) Let F be the full-insurance contract (obtained by solving the equation

W1 = 27,900 3
7 − 7 W1 ) and B the contract of Part (e). Then

NI = (900, 3600), F = (2790, 2790) and B = (2325, 2989.29)

Thus
1
• E[U(NI)] = 4 ln(900) + 34 ln(3, 600) = 7.8421.
• E[U(F)] = ln(2, 790) = 7.9338.
1
• E[U(B)] = 4 ln(2, 325) + 34 ln(2989.29) = 7.94.
(g) Expected utility from contract (h, d) is: 14 ln(3, 600 − h − d) + 34 ln(3, 600 − h).
3
Replacing h with 810 − 10 d we get the following function:

f (d) = 41 ln(3, 600 − 810 + 0.3d − d) + 34 ln(3, 600 − 810 + 0.3d)

= 14 ln(2, 790 − 0.7d) + 34 ln(2, 790 + 0.3d).

To maximize this function we must solve the equation f 0 (d) = 0, that is

0.225 0.175
− =0
2, 790 + 0.3d 2, 790 − 0.7d

The solution is d = 664.29 with corresponding premium h = 610.71, confirming the

conclusion of Part (e). 

Solution to Exercise 5.16

Contract A is the following point in the (W1 ,W2 ) space: W2A = 1, 600 − 44 = 1, 556,
W1A = 1, 556 − 360 = 1, 196, that is, A = (1196, 1556).
3
(a) Replacing h with (1, 600 −W2 ) and d with (W2 −W1 ) in the equation h = 152 − 10 d
we get
14, 480 3
W2 = − W1 .
7 7
√ √ √ √
(b) E[U(NI)] = 15 1, 024 + 45 1600 = 38.4 and E[U(A)] = 15 1, 196 + 45 1, 556 =
38.4736. Thus point A lies above the reservation indifference curve.
(c) The slope of the indifference curve that goes through point A is, at point A, equal to
 0  √ 
p U (1, 196) 1 1, 556
− = √ = −0.2852.
1 − p U 0 (1, 556) 4 1, 196

Since 0.2852 < 37 = 0.4286, the indifference curve is less steep, at point A, than the
insurance budget line; thus the insurance budget line lies below the indifference
curve that goes through point A and, therefore, the best contract on the budget line is
contract A. 
160 Chapter 5. Insurance: Part 2

Solution to Exercise 5.17

For both Carla and Don, no insurance corresponds to the lottery
 
$14, 400 $32, 400
NI = 1 4
5 5

whose expected utility is:

√ √
for Carla: 51 14, 400 + 45 32, 400 = 168

for Don: 15 1 − 1.44+1

1

4 1


+ 5 1 − 3.24+1 = 0.7294.

Mutual insurance corresponds to the lottery

 
$14, 400 $23, 400 $32, 400
MI = 1 8 16
25 25 25

whose expected utility is:

1√ 8√ √
for Carla: 25 14, 400 + 25 23, 400 + 16
25 32, 400 = 168.9506
1 1

8 1

16 1


for Don: 25 1 − 1.44+1 + 25 1 − 2.34+1 + 25 1 − 3.24+1 = 0.7369.

Thus they are both better off with mutual insurance than with no insurance. 

Solution to Exercise 5.18

(a) The probabilities are as follows:

3 losses 2 losses 1 loss no losses

1 3 1 1 2 4 12 1 4 2 48 4 3 64





5 = 125 3 5 5= 125 3 25 5 = 125 5 = 125

(b) Expected utility from no insurance is

1p 4p
E[U(NI)] = 10, 000 + 40, 000 = 180
5 5
while expected utility from mutual insurance is

1 p 12
q
E[U(MI)] = 10, 000 + 40, 000 − 13 60, 000
125 125
48 64 p
q
+ 40, 000 − 13 30, 000 + 40, 000
125 125
= 183.29.

Thus mutual insurance is better than no insurance for each of them.


5.6 Solutions to Exercises 161

Solution to Exercise 5.19

Expected utility from no insurance is

1p 4p
E[U(NI)] = 10, 000 + 40, 000 = 180.
5 5

Expected utility form a mutual insurance (MI) agreement is

3p 7 20 p
q
E[U(MI)] = 10, 000 + 40, 000 − 12 30, 000 + 40, 000 = 180.23.
30 30 30

Thus mutual insurance would still be preferred by both to no insurance. 

II
Risk Sharing

6 Risk Sharing and Pareto Efficiency

165
6.1 Sharing an uncertain surplus
6.2 The Edgeworth box
6.3 Points of tangency
6.4 Pareto efficient contracts on the sides of the
Edgeworth box
6.5 The Edgeworth box when the parties have pos-
itive initial wealth
6.6 More than two outcomes
6.7 Exercises
6.8 Solutions to Exercises
 6. Risk Sharing and Pareto Efficiency

6.1 Sharing an uncertain surplus

In this chapter we will consider contractual relationships between two parties, one called
the Principal and the other the Agent. The Principal hires the Agent to perform a task,
whose outcome is a surplus (a sum of money) that is to be divided between the two parties,
according to the terms of the contract. If there were no uncertainty concerning the surplus,
then the terms of the contract would merely reflect the relative bargaining power of the two
parties. Instead, we are interested in the case where the size of the surplus is partly affected
by random factors and thus is ex ante uncertain. In such a case there are two issues that
arise in determining the type of contract that will be agreed upon: the relative bargaining
power of the parties and the allocation of risk between the two parties. This chapter will
focus on the latter issue.
The following are examples of the contractual relationships considered in this chapter.
• A client (the Principal) hires a lawyer (the Agent) to sue a third party for damages.
The surplus is the amount that will be awarded by the jury or the judge. Typically,
this amount cannot be determined in advance with certainty: it may depend on the
type of evidence that the defendant will produce during the trial, on the composition
and leaning of the jury, etc. There are several types of contracts that the client and
the lawyer could sign: they could agree on a fixed payment to the lawyer (that is,
independent of the awarded damages), or they could agree that the lawyer will retain
a specified percentage of the awarded damages, or they could agree that the client
will be guaranteed a certain amount and the lawyer will keep the residual amount.
166 Chapter 6. Risk Sharing and Pareto Efficiency

• The owner of a store (the Principal) hires a sales assistant (the Agent) to run the
store. The surplus is the profit generated during a specified period of time. The size
of the profit is typically affected by random factors which are not under the control
of either party, such as the number of customers that will show up during that period
of time. Again, many contractual possibilities arise: the employee (the Agent) could
be hired at a fixed salary, or could be offered a percentage of the profits generated by
the store, or could be guaranteed a base salary as well as a percentage of the profits,
etc.
• A landowner (the Principal) gives his land to a farmer (the Agent) to cultivate. The
surplus is the sale value of the crop. The size and the value of the crop depend on
random factors, such as the weather and the market price. A possible contractual
arrangement is to have the Agent give a part of the crop to the Principal as rent
(this is called sharecropping). Another possible contractual arrangement is for the
Principal to pay a fixed salary to the Agent and keep the crop for himself. There are,
of course, several alternative contracts that could be agreed upon.
• The shareholders of a firm (the Principal) hire a manager (the Agent) to run the firm.
The surplus is the profit generated by the firm, which can vary depending on such
factors as the cost of inputs, the intensity of competition, the state of the economy,
etc. Possible compensation schemes are: a fixed salary to the manager, stock options,
bonuses, etc.
We assume that the Principal and the Agent are able to agree on the possible levels
of surplus, denoted by X1 , X2 , . . . , Xn (n ≥ 2) and are able to assign probabilities to them,
denoted by p1 , p2 , . . . , pn . Thus if they enter into a contractual relationship they will,
together, generate the money lottery
 
$X1 $X2 ... $Xn
p1 p2 ... pn

with Xi > 0 and 0 < pi < 1, for all i = 1, 2, . . . , n, and p1 + p2 + · · · + pn = 1.

We will describe a possible contract between the two parties as a list of n sums of
money w1 , w2 , . . . , wn with the interpretation of wi (i = 1, . . . , n) as the amount of money
that will be paid to the Agent if the surplus turns out to be Xi . Unless we explicitly state
otherwise, we shall assume that, for all i = 1, ..., n,

0 ≤ wi ≤ Xi .

A contract C = (w1 , w2 , . . . , wn ) is viewed differently by the two parties:

• for the Agent the contract corresponds to the money lottery
 
$w1 ... $wn
p1 ... pn

• while for the Principal the contract corresponds to the money lottery
 
$(X1 − w1 ) ... $(Xn − wn )
.
p1 ... pn
 6.2 The Edgeworth box 167

As in previous chapters, we use the following notation: if C and D are two contracts
and j is one of the parties to the contract (thus j ∈ {P, A}, where P denotes the Principal
and A the Agent),
C j D means that party j strictly prefers C to D.
C ∼j D means that party j is indifferent between C and D.
C %j D means that party j considers C to be at least as good as D
(that is, either C  j D or C ∼ j D).

Definition 6.1.1 Let C and D be two possible contracts. We say that C Pareto dominates
D (or that D is Pareto dominated by C) if one party prefers C to D and the other considers
C to be at least as good as D, that is, if
- either C P D and C %A D,
- or C %P D and C A D.
We say that a contract is Pareto efficient if there is no other contract that Pareto dominates
it.a
a The notions of Pareto dominance and Pareto efficiency are named after the Italian economist Vilfredo
Pareto (1848-1923) who introduced them.

Thus a contract C is Pareto efficient if and only if, for every other contract D, if one of
party prefers D to C then the other Party prefers C to D. Clearly, it would be irrational for
the parties to sign a Pareto dominated contract, since there is another contract that one of
the parties prefers and the other also prefers or considers to be just as good. Our objective
is to characterize the set of Pareto efficient contracts.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 6.7.1 at the end of this chapter.

6.2 The Edgeworth box

In order to be able to illustrate with figures, we shall - until Section 6.6 - focus on the case
where n = 2, that is where there are only two possible levels of surplus. Thus we can think
in terms of two possible states: the “good" state where the surplus is $X g and the “bad"
state where the surplus is $X b , with

X g > X b > 0.

A contract is specified as a pair (wg , wb ) where wg is what the agent gets in the good state
(so that the Principal gets the residual amount X g − wg ) and wb is what the Agent gets in the
bad state (so that the Principal gets the difference: X b − wb ). The set of possible contracts
can be represented graphically by means of an Edgeworth box, which is a rectangle whose
long side is equal to X g and whose short side is equal to X b , as shown in Figure 6.1.1
1 The Edgeworth box is named after the English economist Francis Ysidro Edgeworth (1845-1926) who
introduced it as a way of representing possible distributions of resources between two individuals.
168 Chapter 6. Risk Sharing and Pareto Efficiency

origin for
the Agent
wg 0A

The
length
of
this
side
is X b − wb wb
Xb C

0P X g − wg
origin for
the Principal

The length of this side is X g

Figure 6.1: The Edgeworth box representing the set of possible divisions of the uncertain
surplus between the Principal and the Agent.

The lower left-hand corner of the box is taken to be the origin for the Principal and the
upper right-hand corner to be the origin for the Agent. Any point in the box represents
a possible contract. Consider, for example, point C in Figure 6.1. Going vertically up
from C to the top horizontal side of the box one obtains the amount wg (measured from
the origin 0A for the Agent) that the Agent receives if the surplus turns out to be X g , while
going vertically down from point C to the bottom horizontal side of the box one obtains
the amount (X g − wg ) (measured from the origin 0P for the Principal) that goes to the
Principal if the surplus turns out to be X g . Similarly, going horizontally from point C to
the right vertical side of the box one can read the amount wb that goes to the Agent if the
surplus turns out to be X b , while going horizontally to the left vertical side one can read
the amount (X b − wb ) that goes to the Principal when the surplus is X b .
6.2 The Edgeworth box 169

What about points outside the Edgeworth box? Do they represent feasible con-
tracts? For example, suppose that X g = 1, 000 and X b = 600 and consider the contract
(wg = 300, wb = −200) which is shown as point D in Figure 6.2.

D
800 −200
origin for
the Agent
0A
300

The
length
of
this
side
is
600

0P 700
origin for
the Principal

The length of this side is 1, 000

Figure 6.2: A contract outside the Edgeworth box.

How is such a contract to be interpreted? If the outcome is $1,000 then the Agent will get
$300 and the rest, namely $700, will go to the Principal; on the other hand, if the outcome
is $600 then the Agent will give $200 to the Principal so that the Principal will get the
entire surplus of $600 plus the additional sum of $200 contributed by the Agent, for a total
of $800. Is such a contractual arrangement feasible? Only if the Agent has an initial wealth
which is greater than or equal to $200 so that she can draw from that wealth to make a
payment to the Principal in case the outcome turns out to be $600.
For the time being we shall assume that neither the Principal nor the Agent have
independent funds from which they can draw to make payments to the other party:

until Section 6.5 we will assume that both the Principal

and the Agent have zero initial wealth so that
contracts outside the Edgeworth box are not feasible.
170 Chapter 6. Risk Sharing and Pareto Efficiency

Two possible types of contracts in the Edgeworth box are worth highlighting.

• The first type are contracts that guarantee the Agent a fixed income, no matter what
the surplus turns out to be: these are contracts of the form (wg , wb ) with wg = wb .
The set of such contracts is represented by the 45o line out of the origin for the
Agent. Contract D in Figure 6.3 is an example of a fixed-income, or income-certainty,
contract for the Agent.

• The second type are contracts that guarantee the Principal a fixed income, no matter
what the surplus turns out to be: these are contracts of the form (wg , wb ) with
X g − wg = X b − wb . The set of such contracts is represented by the 45o line out of
the origin for the Principal. Contract E in Figure 6.3 is an example of a fixed-income,
or income-certainty, contract for the Principal.

origin for
the Agent
wgD 0A
45o
income certainty
for the Principal
The
length
of D wbD
this
side X b − wbE E
is
Xb
income certainty
45o for the Agent

0P X g − wgE
origin for
the Principal

The length of this side is X g

Figure 6.3: Contract D = (wgD , wbD ) is such that wgD = wbD and thus guarantees income
certainty to the Agent. Contract E = (wgE , wbE ) is such that X g − wgE = X b − wbE and thus
guarantees income certainty to the Principal.

We shall assume that both the Principal and the Agent have preferences over money
lotteries that satisfy the axioms of Expected Utility Theory (see Chapter 5) and can thus
be represented by von Neumann-Morgenstern utility-of-money functions: UP (m) for the
Principal and UA (m) for the Agent.
6.2 The Edgeworth box 171
√
For example, suppose that the Principal is risk averse with utility function UP (m) = m,
while the Agent is risk neutral, so that we can take her utility function to be the identity
function: UA (m) = m. Suppose that in the good state the surplus is X g = $1, 000 and in the
bad state the surplus is X b = $600; furthermore the probability of the good state is pg = 13
(so that probability of the bad state is pb = 32 ). Would it be rational for the these two
individuals to sign the contract C = (wg = 400, wb = 400) that guarantees income certainty
to the Agent? √ If they did sign such
√ a contract, the Principal’s expected utility would be
E[UP (C)] = 13 1, 000 − 400 + 23 600 − 400 = 17.5931 and the Agent’s utility would be
UA (400) = 400. Consider the alternative contract D = (wg = 676, wb = 276). With this
contract, the Principal gets the same amount of money no matter what the state (namely,
1, 000 − 676
√ = 324 in the good state and 600 − 276 = 324 in the bad state) and thus his
utility is 324 = 18, while the Agent has an expected utility of 13 676 + 23 276 = 409.33.
Hence both parties strictly prefer contract D to contract C and thus it would be irrational
for them to agree on contract C. In other words, contract C is not a rational choice because
there is another contract, namely D, that Pareto dominates it.
In order to identify the set of Pareto efficient contracts, we need to add indifference
curves to the Edgeworth box.
Recall from Chapter 5 that the indifference curves of a risk-averse individual are convex
towards the origin. Figure 6.4 illustrates the case of a risk-averse Principal.

origin for
the Agent
0A

direction of
increasing utility
for the Principal

0P
origin for
the Principal

Figure 6.4: Indifference curves of a risk-averse Principal. Higher indifference curves (that
is, indifference curves farther away from the origin for the Principal) correspond to higher
levels of utility for the Principal.
172 Chapter 6. Risk Sharing and Pareto Efficiency

In Chapter 5 we also saw that the indifference curves of a risk-neutral individual are
straight lines. Figure 6.5 illustrates the case of a risk-neutral Agent.

origin for
the Agent
0A

direction of
increasing utility
for the Agent

0P
origin for
the Principal

Figure 6.5: Indifference curves of a risk-neutral Agent. Lower indifference curves (that is,
indifference curves farther away from the origin for the Agent) correspond to higher levels
of utility for the Agent.
6.2 The Edgeworth box 173

0A
Principal’s indifference curve

Agent’s indifference curve

A
E

C
D B

0P

Figure 6.6: Contract C is not Pareto efficient.

We now show that any contract at which the indifference curve of the Principal and the
indifference curve of the Agent cross cannot be Pareto efficient. Consider again the case
where the Principal is risk averse and the Agent is risk neutral. Contract C in Figure 6.6 is
a contract at which the indifference curves of the two parties cross.
Let us compare alternative contracts to contract C. The terms "above" and "below" are
to be understood as taking the point of view of the origin for the Principal (the bottom,
left-hand side of the box).
• Any contract, such as contract A, that lies above the Principal’s indifference curve
through C and above the Agent’s indifference curve through C, is preferred to C by
the Principal, but considered to be worse than C by the Agent: A P C and C A A.
• Any contract, such as contract B, that lies below the Principal’s indifference curve
through C and above the Agent’s indifference curve through C, is considered to be
worse than C by both Principal and Agent: C P B and C A B.
• Any contract, such as contract D, that lies below both indifference curves, is consid-
ered to be worse than C by the Principal but better than C by the Agent: C P D and
D A C.
• Finally, any contract, such as contract E, that lies between the two indifference
curves, is preferred to contract C by both parties: E P C and E A C.
Thus, whenever the indifference curves of the two parties that go through a contract C
cross at point C, the area between the two curves (the shaded area in Figure 6.6) represents
contracts that are preferred by both the Principal and the Agent to contract C. Hence C is
Pareto dominated by any of those contracts, implying that C is not Pareto efficient.
174 Chapter 6. Risk Sharing and Pareto Efficiency

origin for
the Agent
0A
Principal’s indifference curve

Agent’s indifference curve

0P
origin for
the Principal

Figure 6.7: Contract C is Pareto efficient.

On the other hand, a contract at which the two indifference curves are tangent must
be Pareto efficient. Contract C in Figure 6.7 is an example of such a contract: any other
contract must be either above the Principal’s indifference curve, in which case it is worse
than C for the Agent, or below the Agent’s indifference curve, in which case it is worse
than C for the Principal, or between the two indifference curves, in which case it is worse
than C for both parties.

Thus we conclude that

any contract at which the indifference curves

are tangent is Pareto efficient.

In the next section we identify the set of points where the indifference curves of
Principal and Agent are tangent.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 6.7.2 at the end of this chapter.
 6.3 Points of tangency 175

6.3 Points of tangency

We will consider four separate cases.

6.3.1 Risk averse Principal and risk neutral Agent

Let us begin with the case where the Principal is risk averse, so that his indifference curves
are convex towards the origin for the Principal, and the Agent’s indifference curves are
straight lines.
Recall from Chapter 5 that the indifference curves of a risk-neutral individual are
straight lines with slope (since we are measuring the outcome of the good state on the
horizontal axis):

pg
− (6.1)
1 − pg

where pg is the probability of the good state. On the other hand, the slope of the Principal’s
indifference curve at a contract where the Agent gets wg in the good state (so that the
Principal gets X g − wg ) and wb in the bad state (so that the Principal gets X b − wb ) is (recall
that UP (·) is the utility-of-money function of the Principal and UP0 (·) is its first derivative):

pg UP0 (X g − wg )
− . (6.2)
1 − pg UP0 (X b − wb )

If the two indifference curves are tangent at point C = (wg , wb ) then they have the same
slope there, that is, it must be that

pg pg UP0 (X g − wg )
− = − (6.3)
1 − pg 1 − pg UP0 (X b − wb )

which requires that

UP0 (X g − wg )
= 1, that is, that UP0 (X g − wg ) = UP0 (X b − wb ).
UP0 (X b − wb )

Since the Principal is risk averse, the second derivative of his utility function is negative,
which means that the first derivative is strictly decreasing; hence
 176 Chapter 6. Risk Sharing and Pareto Efficiency

UP0 (X g − wg ) = UP0 (X b − wb ) if and only if X g − wg = X b − wb ,

that is, if and only if point C = (wg , wb ) lies on the 45o line out of the origin for the
Principal, as shown in Figure 6.8.

origin for
the Agent
0A
Principal’s
indifference curve

45o Agent’s
indifference curve
0P
origin for
the Principal

Figure 6.8: When the Principal is risk averse and the Agent is risk neutral, the 45o line out
of the origin for the Principal is the set of points where there is a tangency between the
Principal’s indifference curve and the Agent’s indifference curve.

6.3.2 Risk neutral Principal and risk averse Agent

Now consider the symmetric case where the Principal is risk neutral and the Agent is risk
pg
averse. In this case the Principal’s indifference curves are straight lines with slope − 1−p g,

while the Agent’s indifference curves are convex towards the origin for the Agent. The
slope of the Agent’s indifference curve at contract C = (wg , wb ) is
pg UA0 (wg )
−
1 − pg UA0 (wb )
where UA (·) is the utility-of-money function of the Agent and UA0 (·) is its first derivative.
As before, if the two indifference curves are tangent at point C = (wg , wb ) then they have
the same slope there, that is, it must be that
pg pg UA0 (wg )
− = − (6.4)
1 − pg 1 − pg UA0 (wb )
6.3 Points of tangency 177

which requires that

UA0 (wg )
0 b
= 1, that is, that UA0 (wg ) = UA0 (wb ).
UA (w )

Since the Agent is risk averse, the second derivative of her utility function is negative,
which means that the first derivative is strictly decreasing; hence

UA0 (wg ) = UA0 (wb ) if and only if wg = wb ,

that is, if and only if point C = (wg , wb ) lies on the 45o line out of the origin for the Agent,
as shown in Figure 6.9.

origin for
the Agent
0A
Principal’s 45o
indifference
curve

Agent’s
indifference curve
0P
origin for
the Principal

Figure 6.9: When the Agent is risk averse and the Principal is risk neutral, the 45o line
out of the origin for the Agent is the set of points where there is a tangency between the
Principal’s indifference curve and the Agent’s indifference curve.
 178 Chapter 6. Risk Sharing and Pareto Efficiency

6.3.3 A general principle

From the analysis of Sections 6.3.1 and 6.3.2 we can draw a general principle concerning
the optimal (i.e. Pareto efficient) allocation of risk between two individuals, one of whom
is risk averse and the other is risk neutral:

When one party is risk neutral and the other is risk averse,
a contract C = (wg , wb ) such that 0 < wg < X g and 0 < wb < X b
is Pareto efficient if and only if all the risk is borne by the risk-neutral party,
so that the risk-averse party is guaranteed a fixed income.

The reason why the above principle requires the restriction that 0 < wg < X g and
0 < wb < X b will be explained in Section 6.4.

6.3.4 Both parties risk averse

When both the Principal and the Agent are risk averse, a tangency between the two
indifference curves can only occur at a point strictly between the two 45o lines.

Recall from Chapter 5 that the absolute value of the slope of the indifference curve of a
risk-averse individual is

pg
grater than 1−pg at a point above the 45o line

pg
equal to 1−pg at a point on the 45o line (6.5)

pg
less than 1−pg at a point below the 45o line.

Consider a point which is above (such as point A in Figure 6.10) or on (such as point B
in Figure 6.10) the 45o line out of the origin for the Principal. Then, by (6.5), the slope of
the Principal’s indifference curve at that point is greater than (point A), or equal to (point
pg
B), 1−pg ; on the other hand, from the point of view of the Agent, that point is below her
45o line and thus the slope of her indifference curve at that point is, in absolute value, less
pg
than 1−pg . Hence the Principal’s indifference curve is steeper at that point than the Agent’s
indifference curve, so that the two curves are not tangent and thus the contract is not Pareto
efficient (any contract between the two indifference curves Pareto dominates it).
6.3 Points of tangency 179

0A
Pic
45o
Aic

A D
Pic
Pic

Aic Aic
B C
Pic
45o
Aic

0P

Figure 6.10: When both the Principal and the Agent are risk averse a tangency between
the Principal’s indifference curve and the Agent’s indifference curve cannot occur at points
that are not between the two 45o lines. Aic stands for "indifference curve of the Agent" and
Pic stands for "indifference curve of the Principal".

Similarly, consider a point which is below (such as point C in Figure 6.10) or on (such
as point D in Figure 6.10) the 45o line out of the origin for the Agent. Then, by (6.5), the
slope of the Agent’s indifference curve at that point is greater than (point C), or equal to
pg
(point D), 1−pg ; on the other hand, from the point of view of the Principal, that point is
below his 45o line and thus the slope of his indifference curve at that point is, in absolute
pg
value, less than 1−pg . Hence the Principal’s indifference curve is less steep at that point
than the Agent’s indifference curve, so that the two curves are not tangent and thus the
contract is not Pareto efficient (any contract between the two indifference curves Pareto
dominates it).
180 Chapter 6. Risk Sharing and Pareto Efficiency

A Pareto efficient contract is shown as point E in Figure 6.11. It lies between the two
45o lines and thus involves risk-sharing: neither party is guaranteed a fixed income.

origin for
the Agent
0A
45o

Pic
Aic

45o

0P
origin for
the Principal

Figure 6.11: Contract E is Pareto efficient and involves risk-sharing. Aic stands for
"indifference curve of the Agent" and Pic stands for "indifference curve of the Principal".

If E = (wg , wb ) is a contract at which there is a tangency between the indifference

curve of the Principal and the indifference curve of the Agent, then it must be that

pg UP0 (X g − wg ) pg UA0 (wg )

− = −
1 − pg UP0 (X b − wb ) 1 − pg UP0 (wb )

that is,

UP0 (X g − wg ) UA0 (wg )

= . (6.6)
UP0 (X b − wb ) UA0 (wb )

Equation (6.6) is necessary and sufficient for a contract E = (wg , wb ) with 0 < wg < X g and
0 < wb < X b to be Pareto efficient and provides us with a quick test for Pareto efficiency.
 6.3 Points of tangency 181

For example, suppose that

- the surplus in the good state is X g = $1, 000 and the surplus in the bad state be
X b = $321,
√
- the Principal’s von Neumann-Morgenstern utility-of-money function is UP (m) = m
and
- the Agent’s von Neumann-Morgenstern utility-of-money function is UA (m) = ln(m).
Is Contract C = (600, 200) Pareto efficient? We need to check if (6.6) is satisfied at point
C. Since
√1
UP0 (X g − wg )) 2 400 11
0 b b
= 1 =
UP (X − w ) √ 20
2 121
and
1
UA0 (wg ) 600 1
= =
UA0 (wb ) 1
200
3
11
and 20 6= 13 , contract C is not Pareto efficient.

6.3.5 Both parties risk neutral

The only case that remains to be considered is the case where both Principal and Agent are
risk neutral. We want to show that in this case every contract is Pareto efficient.
There are several ways to prove this. First of all, when an individual is risk neutral
he/she prefers one money lottery to another if and only if the expected value of the first
is greater than the expected value of the second. Pick an arbitrary contract C = (wCg , wCb ).
To show that it is Pareto efficient it is sufficient to show that, for every other contract
D = (wgD , wbD ), if the Principal prefers D to C then the Agent prefers C to D. Being risk
neutral, the Principal prefers D to C if and only if

pg (X g − wgD ) + (1 − pg )(X b − wbD ) > pg (X g − wCg ) + (1 − pg )(X b − wCb )

that is, if and only if

pg X g − pg wgD + (1 − pg )X b − (1 − pg )wbD
(6.7)
> pg X g − pg wCg + (1 − pg )X b − (1 − pg )wCb ).

Subtracting pg X g + (1 − pg )X b from both sides of (6.7) we get

−pg wgD − (1 − pg )wbD > −pg wCg − (1 − pg )wCb

that is,

pg wCg + (1 − pg )wCb > pg wgD + (1 − pg )wbD . (6.8)

The left-hand side of (6.8) is the expected value of the money lottery for the Agent
corresponding to contract C and the right-hand side of (6.8) is the expected value of the
money lottery for the Agent corresponding to contract D. Thus we conclude that contract
D is preferred to contract C by the Principal if and only if the Agent has the opposite
ranking, that is, she prefers C to D.
 182 Chapter 6. Risk Sharing and Pareto Efficiency

An alternative proof that any contract is Pareto efficient makes use of the tangency
argument: gwhen both parties are risk neutral, their indifference curves are straight line with
p
slope − 1−p g and thus the indifference curve of the Principal that goes through a point C
coincides with the indifference curve of the Agent that goes to point C and thus the two
are tangent.

6.3.6 Pareto efficiency for contracts in the interior of the Edgeworth box
A contract C = (wg , wb ) is in the interior of the Edgeworth box if and only if
0 < wg < X g and 0 < wb < X b .
Let us focus on contracts that are in the interior of the Edgeworth box. We showed in
Section 6.3.4 that, for such contracts, equality (6.6) (page 180) is necessary and sufficient
for Pareto efficiency when both parties are risk averse. We now show that, for contracts
in the interior of the Edgeworth box, equality (6.6) is necessary and sufficient for Pareto
efficiency no matter what the risk attitude of the parties, that is, in all cases.
First of all, note that when an individual is risk neutral, we can take the identity function
as his von Neumann-Morgenstern utility-of-money function, that is, we can take U(m) = m.
0 (m )
Hence U 0 (m) = 1 and, for any two sums of money m1 and m2 , U 1
U 0 (m ) = 1. Thus
2

- If the Principal is risk averse and the Agent is risk neutral (Section 6.3.1), equation
(6.6) reduces to
UP0 (X g − wg )
=1
UP0 (X b − wb )
which requires (as shown in Section 6.3.1) that X g − wg = X b − wb , that is, the
Principal (the risk-averse party) must be guaranteed a fixed income.
- If the Principal is risk neutral and the Agent is risk averse(Section 6.3.2), (6.6)
reduces to
U 0 (wg )
1 = A0 b
UA (w )
which requires (as shown in Section 6.3.2) that wg = wb , that is, the Agent (the
risk-averse party) must be guaranteed a fixed income.
- If both Principal and Agent are risk neutral, then equation (6.6) reduces to 1 = 1,
which is true at every point.
Hence we conclude this section with a general principle:

A contract C = (wg , wb ) such that 0 < wg < X g and 0 < wb < X b

UP0 (X g −wg ) U 0 (wg )
is Pareto efficient if and only if UP0 (X b −wb )
= UA0 (wb ) .
A

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 6.7.3 at the end of this chapter.
 6.4 Pareto efficient contracts on the sides of the Edgeworth box 183

6.4 Pareto efficient contracts on the sides of the Edgeworth box

In the previous section we focused on Pareto efficient contracts in the interior of the
Edgeworth box: these are the contracts where there is a tangency between the indifference
curve of the Principal and the indifference curve of the Agent. In this section we show that
there are also contracts on the sides of the Edgeworth box that are Pareto efficient, even
though the indifference curves of the two parties are not tangent there.

0A
C
Pic
Aic

45o

0P

Figure 6.12: Contract C satisfies Pareto efficiency subject to the constraint that only
contracts in the Edgeworth box are feasible. Aic stands for "indifference curve of the
Agent" and Pic stands for "indifference curve of the Principal".

6.4.1 Risk averse Principal and risk neutral Agent

Let us revisit the case of Section 6.3.1, where the Principal is risk averse (so that his
indifference curves are convex towards the origin for the Principal) and the Agent is risk
neutral (so that her indifference curves are straight lines). Consider a point, such as point C
in Figure 6.12, on the upper side of the Edgeworth box to the right of the point at which the
45o line out of the origin for the Principal intersects that side. From (6.5) (page 178) we
know that at such a point the indifference curve of the Agent is steeper than the indifference
curve of the Principal, as shown in Figure 6.12. Thus there would potentially be contracts
that are Pareto superior to contract C (all those in the shaded area in Figure 6.12, that is,
all those that lie between the two indifference curves). However, such contracts are not
feasible, because they lie outside the Edgeworth box. Thus there are no contracts in the
Edgeworth box that are Pareto superior to C; in other words, contract C satisfies Pareto
 184 Chapter 6. Risk Sharing and Pareto Efficiency

efficiency subject to the constraint that only contracts in the Edgeworth box are feasible.2
Putting together the results of this Section and of Section 6.3.1 we conclude that when the
Principal is risk averse and the Agent is risk neutral, the set of Pareto efficient contracts is
given by the union of the 45o line out of the origin for the Principal and the portion of the
top side of the Edgeworth box from the point of intersection with the 45o line out of the
origin for the Principal to the origin for the Agent, as shown in Figure 6.13.

origin for
the Agent
0A

45o

0P
origin for
the Principal

Figure 6.13: The set of Pareto efficient contracts when the Principal is risk averse and the
Agent is risk neutral is given by the thick line (the union of the 45o line out of the origin
for the Principal and the portion of the top side of the Edgeworth box from the origin for
the Agent to the point of intersection with the 45o line out of the origin for the Principal).

6.4.2 Risk neutral Principal and risk averse Agent

Now let us revisit the case of Section 6.3.2, where the Principal is risk neutral (so that his
indifference curves are straight lines) while the Agent is risk averse (so that her indifference
2A contract (wg , wb ) in the shaded area in Figure 6.12 is such that wb < 0 (while 0 ≤ wg < X g ) and thus
it would imply a payment from the Agent to the Principal in the bad state (so that the Principal would receive
an amount greater than X b ). Such a payment is feasible only if the Agent has independent funds from which
to draw if the state is bad. Since we assumed that neither the Principal nor the Agent have independent funds
at their disposal, contracts outside the Edgeworth box are not feasible. The case where the parties do have
independent funds will be discussed in Section 6.5.
6.4 Pareto efficient contracts on the sides of the Edgeworth box 185

curves are convex towards the origin for the Agent). The analysis in this case is symmetric
to that of the previous section.
Consider a point, such as point D in Figure 6.14, on the bottom side of the Edgeworth
box to the left of the point at which the 45o line out of the origin for the Agent intersects
that side. From (6.5) (page 178) we know that at such a point the indifference curve of
the Principal is steeper than the indifference curve for the Agent, as shown in Figure 6.14.
Thus there would potentially be contracts that are Pareto superior to contract D (all those
in the shaded area in Figure 6.14, that is, all those that lie between the two indifference
curves). However, such contracts are not feasible, because they lie outside the Edgeworth
box. Thus there are no contracts in the Edgeworth box that are Pareto superior to D;
in other words, contract D satisfies Pareto efficiency subject to the constraint that only
contracts in the Edgeworth box are feasible.3

origin for
the Agent
0A

45o

0P D
origin for
the Principal
Aic
Pic

Figure 6.14: Contract D satisfies Pareto efficiency subject to the constraint that only
contracts in the Edgeworth box are feasible. Aic stands for "indifference curve of the
Agent" and Pic stands for "indifference curve of the Principal".
3A contract (wg , wb ) in the shaded area in Figure 6.14 is such that wb > X b (while 0 < wg ≤ X g ) and thus
it would imply a payment from the Principal to the Agent in the bad state. Such a payment is feasible only if
the Principal has independent funds from which to draw if the state is bad. Since we assumed that neither the
Principal nor the Agent have independent funds at their disposal, contracts outside the Edgeworth box are
not feasible. The case where the parties do have independent funds will be discussed in Section 6.5.
 186 Chapter 6. Risk Sharing and Pareto Efficiency

Putting together the results of this Section and of Section 6.3.2 we conclude that when
the Principal is risk neutral and the Agent is risk averse, the set of Pareto efficient contracts
is given by the union of the 45o line out of the origin for the Agent and the portion of
the bottom side of the Edgeworth box from the origin for the Principal to the point of
intersection with the 45o line out of the origin for the Agent, as shown in Figure 6.15.

origin for
the Agent
0A

0P
origin for
the Principal

Figure 6.15: The set of Pareto efficient contracts when the Principal is risk neutral and the
Agent is risk averse is given by the thick line (the union of the 45o line out of the origin for
the Agent and the portion of the bottom side of the Edgeworth box from the origin for the
Principal to the point of intersection with the 45o line out of the origin for the Agent).

6.4.3 Both parties risk averse

The same logic applies to the case where both the Principal and the Agent are risk averse:
points on the top side of the Edgeworth box where the Principal’s indifference curve is less
steep than the Agent’s indifference curve and points on the bottom side of the Edgeworth
box where the Principal’s indifference curve is steeper than the Agent’s indifference curve
correspond to Pareto efficient contracts (subject to the constraint that only contracts in the
Edgeworth box are feasible).
 6.5 The Edgeworth box when the parties have positive initial wealth 187

The thick curve in Figure 6.16 shows the set of Pareto efficient contracts for a situation
where both the Principal and the Agent are risk averse.

Aic
0A

Pic

Pic
Aic

0P
Aic
Pic

Figure 6.16: The set of Pareto efficient contracts when both the Principal and the Agent
are risk averse is shown as a thick curve. Aic stands for "indifference curve of the Agent"
and Pic stands for "indifference curve of the Principal".

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 6.7.4 at the end of this chapter.

6.5 The Edgeworth box when the parties have positive initial wealth
So far we have assumed that only contracts inside the Edgeworth box are feasible or,
equivalently, that both the Principal and the Agent have zero initial wealth. If, on the other
hand, one or both parties have a positive initial wealth then, as shown in Figure 6.2 on page
169, it is feasible for the parties to agree on a contract that lies outside the Edgeworth box.
As before, let the surplus in the good state be X g and the surplus in the bad state X b .
Let W P ≥ 0 be the initial wealth of the Principal and W A ≥ 0 the initial wealth of the
Agent. Construct an extended Edgeworth box as follows: the length of the horizontal side
is X g +W P +W A and the length of the vertical side is X b +W P +W A .
188 Chapter 6. Risk Sharing and Pareto Efficiency

R Whereas before we represented a contract as a pair (wg , wb ), where wg was the portion,
given to the Agent, of the surplus X g generated in the good state (thus 0 ≤ wg ≤ X g )
and wb was the portion, given to the Agent, of the surplus X b generated in the bad
state (thus 0 ≤ wb ≤ X b ), we now need to represent a contract differently in order to
allow for the possibility of payments from one party to the other. Thus in this more
general setting we represent a contract as a pair (zg , zb ) where zg is the wealth of the
Agent in the good state (thus 0 ≤ zg ≤ X g + W P + W A ) and zb is the wealth of the
Agent in the bad state (thus 0 ≤ zb ≤ +X b + +W P +W A ).4

Figure 6.17 shows the extended Edgeworth box. The dashed box inside the extended box
is the Edgeworth box that would apply if the parties had zero initial wealth.

WA 0A

WA
The D
The length length
of this of this
side is side is
b
X +W P +W A Xb C
WP
The length of this side is X g

0P WP

The length of this side is X g +W P +W A

Figure 6.17: The extended Edgeworth box. The dashed box inside the extended box is the
Edgeworth box that would apply if the parties had zero wealth.

Point C in Figure 6.17 represents a contract where the Agent gets all the surplus in each
state and the Principal retains his initial wealth of W P . Note that a wealth of W P for sure is
an outside option for the Principal: he can guarantee himself such a wealth by refusing
to enter into an agreement with the Agent. Thus the indifference curve of the Principal
that goes through point C represents a reservation level of utility for the Principal: if the
Principal is rational, he will not agree to any contract that gives him an expected utility
less than UP (W P ).
4 For example, if X g = $1, 500, X b = $900 and the parties agree that the Agent will get one third of the
surplus in each state (thus $500 in the good state and $300 in the bad state) then we represent this contract
not as the pair (500, 300) but as the pair (zg = 500 +W A , zb = 300 +W A ), where W A is the Agent’s initial
wealth. Hence, from the Principal’s point of view, this contract will yield him a wealth of 1, 000 +W P in the
good state and 600 +W P in the bad state, where W P is the Principal’s initial wealth.
6.5 The Edgeworth box when the parties have positive initial wealth 189

Similarly, point D in Figure 6.17 represents a contract where the Principal gets all the
surplus in each state and the Agent retains her initial wealth of W A . A wealth of W A for
sure constitutes an outside option for the Agent: she can guarantee herself such a wealth
by refusing to enter into an agreement with the Principal. Thus the indifference curve of
the Agent that goes through point D represents a reservation level of utility for the Agent:
if the Agent is rational, she will not agree to any contract that gives her an expected utility
less than UA (W A ).
We call a contract that gives each party at least his/her reservation utility individually
rational. Thus the set of individually rational contracts is the set of points in the extended
Edgeworth box that lie between the two reservation indifference curves: the indifference
curve of the Principal that goes through point (W P ,W P ) (measured from the origin for the
Principal) and the indifference curve of the Agent that goes through the point (W A ,W A )
(if measured from the origin for the Agent, or (X g +W P , X b +W P ) if measured from the
origin for the Principal).
Figure 6.18 shows the reservation indifference curve of a of a risk-averse Principal and
the reservation indifference curve of a risk-neutral Agent. The area between the two curves
consist of the set of individually rational contracts.

WA 0A
Agent’s reservation
indifference curve

WA
D
The length
of this Xb
side is
b
X +W P +W A C Xg
WP
Principal’s reservation
indifference curve

0P WP

The length of this side is X g +W P +W A

Figure 6.18: The reservation indifference curves of a risk-averse Principal and a risk-neutral
Agent.

To find the Pareto efficient contracts we proceed as in the previous sections. The only
difference is that rationality of the parties now requires not only Pareto efficiency but also
individual rationality. Thus we want to identify the set of contracts that are both Pareto
efficient and individually rational; let us call it the set of rational contracts. It is given by
the intersection of the set of Pareto efficient contracts and the set of individually rational
contracts.
190 Chapter 6. Risk Sharing and Pareto Efficiency

1. For the case where the Principal is risk averse and the Agent is risk neutral, the
set of rational contracts consists of the portion of the 45o line out of the origin for
the Principal from the point (W P ,W P ) to the top side of the extended Edgeworth
box, followed by the portion of the top side of the extended Edgeworth box from
the intersection with the 45o line to the point of intersection with the reservation
indifference curve of the Agent. The set of rational (i.e. Pareto efficient and
individually rational) contracts is shown as a thick line in Figure 6.19.

WA 0A
set of
rational contracts
WA

Xb

Xg
WP
C Agent’s reservation
Principal’s reservation indifference curve
indifference curve

0P WP

Figure 6.19: The thick line shows the set of rational (i.e. Pareto efficient and individually
rational) contracts when the Principal is risk averse and the Agent is risk neutral.
6.5 The Edgeworth box when the parties have positive initial wealth 191

2. For the case where the Principal is risk neutral and the Agent is risk averse,
the set of rational contracts consists of the portion of the 45o line out of
the origin for the Agent from the point (W A ,W A ) to the bottom side of the
extended Edgeworth box, followed by the portion of the bottom side of the
extended Edgeworth box from the intersection with the 45o line to the point of
intersection with the reservation indifference curve of the Principal. The set of
rational (i.e. Pareto efficient and individually rational) contracts is shown as a
thick line in Figure 6.20.

WA 0A
Principal’s reservation
indifference curve D
WA
set of
Xb rational contracts

Xg
WP
Agent’s reservation
indifference curve

0P WP

Figure 6.20: The thick line shows the set of rational (i.e. Pareto efficient and individually
rational) contracts when the Principal is risk neutral and the Agent is risk averse.
192 Chapter 6. Risk Sharing and Pareto Efficiency

3. For the case where both the Principal and the Agent are risk neutral, the set of
rational contracts coincides with the set of individually rational contracts (since
every contract is Pareto efficient). It is shown as the shaded area between the two
reservation indifference curves in Figure 6.21.

WA 0A

D
WA

Xb

Xg
WP Agent’s reservation
C indifference curve
Principal’s reservation
indifference curve

0P WP

Figure 6.21: The shaded area between the two reservation indifference curves is the set of
rational contracts when both parties are risk neutral.
 6.6 More than two outcomes 193

4. For the case where both the Principal and the Agent are risk averse, the set of
rational contracts is a curve between the two 45o lines together with some segments
of the top and bottom sides of the extended Edgeworth box. It is shown as a thick
curve in Figure 6.22.

WA
0A

WA

Principal’s Agent’s
reservation reservation
indifference indifference
curve curve

WP

0P WP

Figure 6.22: The thick curve shows the set of rational (i.e. Pareto efficient and individually
rational) contracts when both the Principal and the Agent are risk averse.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 6.7.5 at the end of this chapter.

6.6 More than two outcomes

In the previous sections the analysis was restricted to the case of two outcomes (two states)
in order to be able to illustrate with diagrams (Edgeworth boxes). In this section we briefly
discuss the general case where there are n outcomes, with n ≥ 2. For simplicity we assume
that the initial wealth of both parties is zero.5
5 The case of positive initial wealth is dealt with in a manner similar to the analysis of Section 6.5. Let
W P be the initial wealth of the Principal and W A be the initial wealth of the Agent. If Xi is the monetary
outcome in state i, let X i = Xi +W P +W A . Then proceed as explained in this section but replacing Xi with
X i . The only difference is that each party now has a reservation utility level given by the outside option
of keeping his/her wealth and not reaching an agreement and, as a consequence, neither party will accept
a contract that gives him/her a level of utility less than the reservation level; that is, to the requirement of
Pareto efficiency we add the requirement of individual rationality.
 194 Chapter 6. Risk Sharing and Pareto Efficiency

Let there be n states (n ≥ 2) and let $Xi be the outcome in state i ∈ {1, . . . , n}, which
occurs with probability pi (0 < pi < 1, for every i ∈ {1, . . . , n}, and p1 + p2 + · · · + pn = 1).
We order the outcomes in such a way that

X1 < X2 < · · · < Xn . (6.9)

A contract is an n-tuple (w1 , w2 , . . . , wn ) where wi is what the Agent gets in state i (with
0 ≤ wi ≤ Xi ).

6.6.1 Risk-neutral Principal and risk-averse Agent

The general principle that if one party is risk averse and the other is risk neutral then, for
Pareto efficiency, the risk-neutral party should bear all the risk applies in this case too.
Consider first the case where the Principal is risk neutral and the Agent is risk averse.
It is easy to see that a contract of the form C̃ = (w̃, w̃, . . . , w̃) (where the risk-averse party,
namely the Agent, is guaranteed a fixed income of w̃) is Pareto efficient. Take any other
contract D = (w1 , w2 , . . . , wn ) 6= C̃ that the Principal considers to be at least as good as C̃;
then the expected value of D is less than or equal to the expected value of C̃ (which is w̃),
that is, p1 w1 + p2 w2 + · · · + pn wn ≤ w̃.6 On the other hand, the Agent prefers contract C̃ to
D: if E[D] = w̃ then C̃ A D by definition of risk aversion and if E[D] < w̃ then the Agent
prefers $w̃ to $E[D], because she prefers more money to less, and she considers $E[D]
for sure to be at least as good as D (better, if D is non-degenerate), by definition of risk
aversion; thus, by transitivity, C̃ A D.
However, as in the case where n = 2, a guaranteed fixed income for the risk-averse
party is only sufficient for Pareto efficiency: it is not necessary. There are contracts on the
edges of the "Edgeworth hypercube" that are Pareto efficient. The following proposition
helps identifying those contracts.
Proposition 6.6.1 Let the Principal be risk neutral and the Agent risk averse.
Let i, j ∈ {1, 2, . . . , n} with i < j (so that, by (6.9), Xi < X j ) and consider a contract
C = (w1 , w2 , . . . , wn ) with wi 6= w j . If C is Pareto efficient then
(A) w j > wi , and
(B) wi = Xi .
6 Since the Principal is risk neutral we can take the identity function as his utility-of-money function so
that the expected utility of contract C̃ is
n n
∑ pi (Xi − w̃) = ∑ pi Xi − w̃
i=1 i=1

and the expected utility of contract D is

n n n
∑ pi (Xi − wi ) = ∑ pi Xi − ∑ pi wi .
i=1 i=1 i=1

The Principal considers D to be at least as good as C̃ if and only if

n n n n
∑ pi Xi − ∑ pi wi ≥ ∑ pi Xi − w̃, that is, if and only if ∑ pi wi ≤ w̃.
i=1 i=1 i=1 i=1
6.6 More than two outcomes 195

Proof. (A) Suppose that w j < wi (recall that Xi < X j ) and let
pi pj
ŵ = wi + w j. (6.10)
pi + p j pi + p j

Then w j < ŵ < wi .7 Modify contract C by replacing both wi and w j with ŵ and leaving
the other coordinates unchanged. Call D the resulting contract. Then the expected value of
D is equal to the expected value of C.8 Hence the Principal is indifferent between contract
C and contract D.9 On the other hand, the Agent prefers contract D to contract C, since C
is a mean-preserving spread of D (see Chapter 4, Section 4.4.2). Thus D Pareto dominates
C and therefore C is not Pareto efficient.
(B) Because of Part (A) we can assume that wi < w j (by hypothesis, wi 6= w j ). Suppose
that wi < Xi ; we want to show that C is not Pareto efficient. We need to consider two cases.
Case B.1: ŵ ≤ Xi where ŵ is defined in (6.10) (and thus wi < ŵ < w j ). Consider the
contract obtained from C by replacing both wi and w j with ŵ. Then, repeating the argument
developed in Part (A), the Principal is indifferent between C and D, because E[C] = E[D],
while the Agent prefers D to C because C is a mean-preserving spread of D. Hence D
Pareto dominates C and therefore C is not Pareto efficient.
Case B.2: ŵ > Xi (where ŵ is defined in (6.10); note that this implies that w j > Xi ).
Construct a new contract, call it D, obtained from C by replacing wi with Xi and w j with w̃˜
where
pi
w̃˜ = w j − (Xi − wi ) . (6.11)
pj
pi
Note that, since - by hypothesis - wi < Xi , pj (Xi − wi ) > 0 and thus w̃˜ < w j . Furthermore,
w̃˜ > Xi since
pi
wj − (Xi − wi ) > Xi if and only if p j w j − pi Xi + pi wi > p j Xi
pj
if and only if pi wi + p j w j > (pi + p j )Xi
pi pj
if and only if wi + w j > Xi
pi + p j pi + p j
if and only if (by (6.10)) ŵ > Xi

which is our hypothesis.

7 Since pi p j
pi and p j are strictly between 0 and 1, so are pi +p j
and pi +p j
and their sum is equal to 1; hence ŵ
is a convex combination of wi and w j , that is, a point strictly between wi and w j .
n
8 Let a = ∑ pk wk − pi wi − p j w j . Then E[C] = a + pi wi + p j w j and E[D] = a + (pi + p j )ŵ. Since
k=1
(pi + p j )ŵ = pi wi + p j w j , E[C] = E[D].
9 For the Principal (who, by hypothesis, is risk neutral) the expected utility of contract C is

n n n n
∑ pk (Xk − wk ) = ∑ pk Xk − ∑ pk wk = ∑ pk Xk − E[C]
k=1 k=1 k=1 k=1

n
and, similarly, the expected utility of contract D is ∑ pk Xk − E[D].
k=1
196 Chapter 6. Risk Sharing and Pareto Efficiency

First we show that E[C] = E[D], so that the Principal is indifferent between C and D.
Clearly, E[C] = E[D] if and only if pi Xi + p j w̃ = pi wi + p j w j , which is true since
 
pi
pi Xi + p j w̃˜ = pi Xi + p j w j − (Xi − wi ) = pi Xi + p j w j − pi Xi + pi wi = pi wi + p j w j .
pj

Next we show that the Agent prefers D to C, so that D Pareto dominates C and therefore C
is not Pareto efficient. The Agent prefers D to C because D dominates C in the sense of
second-order stochastic dominance (see Chapter 4, Definition 4.4.3); indeed the lottery
corresponding to contract C can be obtained from the lottery corresponding to contract D
by a sequence of two mean-preserving spreads. To see
 this, consider the following lotteries
wk
where the dots stand for entries of the form for k ∈ {1, 2, . . . , n} \ {i, j}.
pk

. . . wi Xi . . . w̃˜ w j . . .
 
LC = (6.12)
. . . pi 0 . . . 0 p j . . .

w̃˜ w j . . .
 
. . . wi Xi . . .
LD = (6.13)
. . . 0 pi . . . pj 0 ...

w̃˜
 
. . . wi Xi . . . wj ...
LE = (6.14)
. . . α pi 0 . . . p j (1 − α)pi . . .

Xi . . . w̃˜
 
... wi wj ...
LF = (6.15)
. . . α pi + β p j 0 . . . 0 (1 − α)pi + (1 − β )p j . . .

(6.16)

where

w j − Xi w j − w̃˜
α= and β= (6.17)
w j − wi w j − wi

LC is the lottery (for the Agent) corresponding to contract C and LD is the lottery (for the
Agent) corresponding to contract D. We want to show that LE is a mean-preserving spread
of LD , LF is a mean-preserving spread of LE and LF = LC .
• LE is a mean-preserving spread of LD because (1) wi < Xi (by hypothesis),
(2) w j > w̃˜ (as shown above) and (3) α pi wi + (1 − α)pi w j = pi Xi (so that the
expected value of E equals the expected value of D).10

10 In fact,
 
w j − Xi Xi − wi wi w j − wi Xi + w j Xi − wi w j
α pi wi + (1 − α)pi w j = pi wi + pi w j = pi =
w j − wi w j − wi w j − wi
w j − wi
= pi Xi = pi Xi .
w j − wi
 6.6 More than two outcomes 197

• LF is a mean-preserving spread of LE because (1) wi < w̃˜ (since, by hypothe-

sis, wi < Xi and, as shown above, Xi < w̃˜ ), (2) w j > w̃˜ (as shown above) and
(3) β p j wi + (1 − β )p j w j = p j w̃˜ (so that the expected value of F equals the expected
value of E).11
• Finally, LF = LC since α pi + β p j = pi and (1 − α)pi + (1 − β )p j = p j .12


6.6.2 Risk-averse Principal and risk-neutral Agent

The case where the Principal is risk-averse and the Agent is risk neutral is symmetric to
the case considered in the previous section. Thus we will merely state, without proof, the
counterpart to Proposition 6.6.1.
Proposition 6.6.2 Let the Principal be risk averse and the Agent risk neutral.
Let i, j ∈ {1, 2, . . . , n} with i < j (so that, by (6.9), Xi < X j ) and consider a contract
C = (w1 , w2 , . . . , wn ) with wi 6= w j . If C is Pareto efficient then
(A) X j − w j > Xi − wi , and
(B) wi = 0.

6.6.3 Both parties risk neutral

As in the case of two outcomes, if both parties are risk neutral then every contract is Pareto
efficient. This is because if one party prefers a contract, say D, to another contract, say C,
then the expected value of the lottery - for that party - corresponding to contract D is higher
than the expected value of the lottery - for that party - corresponding to contract C, which
implies that the expected value of the lottery - for the other party - corresponding to contract
D is lower than the expected value of the lottery - for the other party - corresponding to
contract C; hence the other party prefers C to D.

11 In fact,

w j − w̃˜ w̃˜ − wi w j − wi
β p j wi + (1 − β )p j w j = p j wi + p jw j = p j w̃˜ = p j w̃˜.
w j − wi w j − wi w j − wi

12 In fact,
h   i
w j − Xi w j − w̃˜ (w j − Xi ) pi + w j − w j − ppij (Xi − wi ) pj
α pi + β p j = pi + pj =
w j − wi w j − wi w j − wi

(w j − Xi ) pi + (Xi − wi ) pi w j − wi
= = pi = pi
w j − wi w j − wi
and
w j − w̃˜ w̃˜ − wi
   
w j − Xi Xi − wi
(1 − α) pi + (1 − β ) p j = 1 − pi + 1 − pj = pi + pj
w j − wi w j − wi w j − wi w j − wi

Xi − wi w j p j − (Xi − wi ) pi − wi p j w j − wi
= pi + = p j = p j.
w j − wi w j − wi w j − wi
 198 Chapter 6. Risk Sharing and Pareto Efficiency

6.6.4 Both parties risk averse

In the case of only two outcomes we saw that a contract in the interior of the Edgeworth
box is Pareto efficient only if the indifference curves of the two parties through that point
are tangent. This requirement, expressed as equation (6.6) (page 180) extends to the case
of more than two outcomes as follows.
As before, let there be n states (n ≥ 2) and let $Xi be the outcome in state i ∈ {1, . . . , n},
which occurs with probability pi (0 < pi < 1, for every i ∈ {1, . . . , n}, and p1 + p2 + · · · +
pn = 1). We order the outcomes in such a way that

X1 < X2 < · · · < Xn .

Let C = (w1 , w2 , . . . , wn ) be an interior contract, that is, a contract such that, for every
i = 1, 2, . . . , n, 0 < wi < Xi . If C is Pareto efficient then13

UP0 (X1 − w1 ) UA0 (w1 )

= ,
UP0 (X2 − w2 ) UA0 (w2 )

UP0 (X2 − w2 ) UA0 (w2 )

=
UP0 (X3 − w3 ) UA0 (w3 )

...

UP0 (Xn−1 − wn−1 ) UA0 (wn−1 )

= .
UP0 (Xn − wn ) UA0 (wn )

As in the case of two outcomes, there are also Pareto efficient contracts that are not interior
contracts, that is, contracts where, in some states, one of the parties gets the entire surplus.
We will omit a discussion of necessary conditions for Pareto efficiency for such contracts.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 6.7.6 at the end of this chapter.

13 Note that if the n − 1 equations below are satisfied, then, for every i, j = 1, . . . , n with i 6= j,
UP0 (Xi −wi ) U 0 (w ) U 0 (X −w ) U 0 (w ) U 0 (X −w )
UP0 (X j −w j )
= U A0 (w i ) . For example, one can get UP0 (X1 −w1 ) = UA0 (w1 ) by multiplying the equation UP0 (X1 −w1 ) =
A j P 3 3 A 3 P 2 2
UA0 (w1 ) UP0 (X2 −w2 ) UA0 (w2 )
UA0 (w2 )
by the equation UP0 (X3 −w3 )
= UA0 (w3 )
.
 6.7 Exercises 199

6.7 Exercises
The solutions to the following exercises are given in Section 6.8 at the end of this chapter.

6.7.1 Exercises for Section 6.1: Sharing an uncertain surplus

Exercise 6.1 A Principal wants to hire an Agent to run his firm. The Principal’s utility-
of-money function is UP (m) = ln(m), while the Agent’s utility-of-money function is
UA (m) = 2m + 2. There are three possible profit levels: X1 = $2, 400, X2 = $1, 600 and
X3 = $900. The corresponding probabilities are p1 = 15 , p2 = 25 and p3 = 25 . Let w1
be the payment to the Agent if the outcome is X1 and similarly for w2 and w3 . Show
that contract A = (w1 = 700, w2 = 700, w3 = 700) is Pareto dominated by contract
B = (w1 = 1, 200, w2 = 1, 000, w3 = 200). 

Exercise 6.2 John’s von Neumann-Morgenstern utility-of-money function is U(m) =

√
m. He owns a firm and is thinking of hiring Joanne to run the firm for him. Joanne’s
von Neumann-Morgenstern utility-of-money function is V (m) = 2m. In the past there
were good times when the firm’s yearly profits were $4,000 and bad times when the
firm’s yearly profits were $1,600. About 23 of the time it was a good year and 13 of the
time it was a bad year. John offered Joanne a contract, call it A, that pays her a fixed
salary of $1,000 per year.
(a) What is John’s attitude to risk?
(b) What is Joanne’s attitude to risk?
(c) Show that the alternative contract, call it B, that pays Joanne $1,500 if the profit
turns out to be $4,000 and pays her nothing if the profit turns out to be $1,600, is
as good as contract A for Joanne but gives John a higher expected utility. That is,
contract B Pareto dominates contract A.


6.7.2 Exercises for Section 6.2: The Edgeworth box

Exercise 6.3 Let the surplus in the good state be X g = $200 and the surplus in the bad
state be X b = $120. Consider the following contracts, where the first coordinate is the
amount of money that the Agent receives in the good state and the second coordinate
is the amount of money that the Agent receives in the bad state: C = (60, 80) and
D = (100, 60). The probability of the good state is 25% (so that the probability of the
bad state is 75%).
(a) Represent the two contracts as points in an Edgeworth box.
(b) If the Principal is risk neutral, how does he rank the two contracts?
(c) If the Agent is risk neutral, how does she rank the two contracts?

200 Chapter 6. Risk Sharing and Pareto Efficiency

Exercise 6.4 Consider the contracts shown in Figure 6.23.

(a) What is the Principal’s attitude to risk?
(b) What is the Agent’s attitude to risk?
(c) How does the Principal rank the five contracts A, B,C, D and E?
(d) How does the Agent rank the five contracts A, B,C, D and E?
(e) Among the five contracts A, B,C, D and E find all the pairs of contracts such that
one Pareto dominates the other.


0A
Pic Pic
Aic

A
E

D B

0P

Figure 6.23: The Edgeworth box for Exercise 6.4. Pic stands for "indifference curve of the
Principal" and Aic for "indifference curve of the Agent".

Exercise 6.5 Let the surplus in the good state be X g = $200 and the surplus in the bad
state be X b = $120. Consider the following contracts, where the first coordinate is the
amount of money that the Agent receives in the good state and the second coordinate
is the amount of money that the Agent receives in the bad state: C = (60, 80) and
D = (100, 60). The probability of the good state is 25% (so that the probability of the
bad state is 75%).
√ The Principal’s von Neumann-Morgenstern utility-of-money function
is UP (m) = m and the Agent’s von Neumann-Morgenstern utility-of-money function
is UA (m) = ln(m). Does one of the two contracts Pareto dominate the other? 
 6.7 Exercises 201

6.7.3 Exercises for Section 6.3: Points of tangency

0A
45o

A D

B C

45o

0P

Figure 6.24: The Edgeworth box for Exercise 6.6.

Exercise 6.6 Consider the Edgeworth box of Figure 6.24.

(a) Suppose that the Principal is risk neutral and the Agent is risk averse. Of the four
contracts A, B,C and D, which (if any) is Pareto efficient?
(b) Suppose that the Principal is risk averse and the Agent is risk neutral. Of the four
contracts A, B,C and D, which (if any) is Pareto efficient?
(c) Suppose that both the Principal and the Agent are risk averse. Of the four contracts
A, B,C and D, which (if any) is Pareto efficient?
(d) Suppose that both the Principal and the Agent are risk neutral. Of the four
contracts A, B,C and D, which (if any) is Pareto efficient?


Exercise 6.7 Let the surplus in the good state be X g = $800 and the surplus in the
bad state be X b = $400. The Principal’s von Neumann-Morgenstern utility-of-money
m 2


function is UP (m) = 82 − 10 − 100 while the Agent’s von Neumann-Morgenstern
utility-of-money function is UA (m) = 2m + 4. Consider the following contract (where,
as usual, the first coordinate is the amount of money that the Agent receives in the good
state and the second coordinate is the amount of money that the Agent receives in the
bad state): C = (400, 200).
(a) Is contract C Pareto efficient? Explain.
(b) Represent contract C as a point in an Edgeworth box and sketch the indifference
curves of Principal and Agent that go through that point.
(c) In the Edgeworth box show a contract D such that D P C and D ∼A C (that is,
the Principal prefers D to C and the Agent finds D to be just as good as C).

202 Chapter 6. Risk Sharing and Pareto Efficiency

Exercise 6.8 Let the surplus in the good state be X g = $1, 206 and the surplus in the
bad state be X b = $676. The probability of the good state is 25%. The Principal’s von
Neumann-Morgenstern utility-of-money function is UP (m) = m √ while the Agent’s von
Neumann-Morgenstern utility-of-money function is UA (m) = m.
(a) Find a Pareto efficient contract in the interior of the Edgeworth box at which the
Principal’s expected utility is 232.5.
(b) Calculate the Agent’s expected utility at the contract of Part (a).
(c) Find a Pareto efficient contract on one of the sides of the Edgeworth box at which
the Agent’s expected utility is 27.
(d) Calculate the Principal’s expected utility at the contract of Part (c).


Exercise 6.9 The owner of a firm (the Principal) hires a manager (the Agent) to
run the firm. The Principal’s utility-of-money function
h is U(m)
i = ln(m), while the
x
−( 100 )
Agent’s utility-of-money function is V (m) = 100 1 − e (where e is the number
2.71828. . . ). The profit of the firm is affected by random events and can turn out to
be X g = $1, 000 (this is expected to happen with probability 14 ) or X b = $600 (with
probability 34 ). The Principal offers the following contract to the Agent: if the profit
turns out to be $1,000, I will pay you wg = $400, otherwise I will pay you wb = $200.
(a) What is the attitude to risk of the Principal?
(b) What is the attitude to risk of the Agent?
(c) Is the proposed contract Pareto efficient? Explain.
(d) If you were to propose a contract that both Principal and Agent prefer to the one
considered above, which of the following suggestions would you make? Explain
your answer by making use of an Edgeworth box.
1. Increase both wg and wb .
2. Decrease both wg and wb .
3. Increase wg and decrease wb .
4. Decrease wg and increase wb .

6.7 Exercises 203

1,000 0A
Pic

E
D
650 550
C
Aic

45o

0P 1,000

Figure 6.25: The Edgeworth box for Exercise 6.10.

Exercise 6.10 Consider the Edgeworth box of Figure 6.25. Clearly, the Agent is risk
neutral while the Principal is risk averse.
√ The Principal’s von Neumann-Morgenstern
utility-of-money function is UP (m) = m. The probability of the good state is 25 .
(a) What is the value of X g , the surplus in the good state?
(b) What is the value of X b , the surplus in the bad state?
(c) Calculate the expected utility of both parties at contract C.
(d) Find the coordinates of contract D and calculate the expected utility of both
parties at that contract.
(e) Find the coordinates of contract E and calculate the expected utility of both parties
at that contract.
(f) Find three contracts that Pareto dominate contract C.


Exercise 6.11 Let the surplus in the good state be X g = $1, 000 and the surplus in the
bad state be X b = $400.
√ The Principal’s von Neumann-Morgenstern utility-of-money
function is UP (m) = m while the Agent’s von Neumann-Morgenstern utility-of-
money function is UA (m) = ln(m). Is contract C = (600, 300) (where, as usual, the first
coordinate is what the Agent gets in the good state and the second coordinate is what
the Agent gets in the bad state) Pareto efficient? 
 204 Chapter 6. Risk Sharing and Pareto Efficiency

6.7.4 Exercises for Section 6.4: Pareto efficient contracts on the sides of the
Edgeworth box

Exercise 6.12 Let the surplus in the good state be X g = $1, 000 and the surplus in the
bad state be X b = $600. The probability of the good state is 23 . The Agent is risk neutral,
while the Principal is risk averse with von Neumann-Morgenstern utility-of-money
function UP (m) = ln(m). Consider contract C = (250, 150) (where, as usual, the first
coordinate is what the Agent gets in the good state and the second coordinate is what
the Agent gets in the bad state).
(a) Is contract C Pareto efficient?
(b) Represent contract C in an Edgeworth box and sketch the indifference curves of
Principal and Agent through point C.
(c) Find a Pareto efficient contract, call it D, such that the Agent is indifferent between
C and D while the Principal prefers D to C.
(d) Represent contract D in the Edgeworth box of Part (b).


Exercise 6.13 Let the surplus in the good state be X g = $2, 000 and the surplus in the
bad state be X b = $1, 200. The probability of the good state is 79 . The Principal is risk
neutral, while the Agent
√ is risk averse with von Neumann-Morgenstern utility-of-money
function UA (m) = m. Consider contract C = (1559, 1004) (where, as usual, the first
coordinate is what the Agent gets in the good state and the second coordinate is what
the Agent gets in the bad state).
(a) Is contract C Pareto efficient?
(b) Represent contract C in an Edgeworth box and sketch the indifference curves of
Principal and Agent through point C.
(c) Find a Pareto efficient contract, call it D, such that the Agent is indifferent between
C and D while the Principal prefers D to C.
(d) Represent contract D in the Edgeworth box of Part (b).


Exercise 6.14 As in Exercise 6.13, let the surplus in the good state be X g = $2, 000
and the surplus in the bad state be X b = $1, 200. The probability of the good state is 79 .
However, consider now the case where both the Principal and the Agent are risk√averse
√ von Neumann-Morgenstern utility-of-money function: UP (m) = m and
with the same
UA (m) = m. Consider again contracts C = (1559, 1004) and D = (1493.04, 1200) of
Exercise 6.13.
(a) Is contract C Pareto efficient?
(b) How does the Principal rank the two contracts C and D?
(c) How does the Agent rank the two contracts C and D?
(d) Is contract D Pareto efficient?

6.7 Exercises 205

Exercise 6.15 Let the surplus in the good state be X g = $3, 000 and the surplus in the
bad state be X b = $1, 800. Both the Principal and the Agent are risk averse with the
same von Neumann-Morgenstern utility-of-money function:
   
m m
UP (m) = ln 1 + and UA (m) = ln 1 + .
1, 000 1, 000

(a) Find the range of values of wg for which contracts of the form (wg , 0) with
0 ≤ wg ≤ X g are Pareto efficient. [Hint: refer to Figure 6.15 on page 186.]
(b) Find the range of values of wg for which contracts of the form (wg , 1800) with
0 ≤ wg ≤ X g are Pareto efficient.


Exercise 6.16 Mr. P wants to hire Ms. A to run his firm. If Ms. A works for Mr.
P, one of two outcomes will occur: the profit of the firm will be $520 (this occurs
with probability 45
98 ) or it will be $200 (with probability
√
53
98 ). Mr. P’s von Neumann-
Morgenstern utility-of-money function is U(m) = m, while Ms. A is risk neutral.
Consider the following contract, call it C: Ms. A will get $144 if the profit of the firm
turns out to be $520, and she will get $90 if the profit of the firm turns out to be $200.
(a) What is Mr. P’s expected utility from this contract?
(b) What is Ms. A’s expected utility from this contract?
(c) Find a Pareto efficient contract in the Edgeworth box, call it D, that Mr. P
considers to be just as good as C and Ms. A prefers to C.
(d) Find a a Pareto efficient contract in the Edgeworth box, call it E, that Mr. P
prefers to C and Ms. A considers to be just as good as C.
(e) How does Mr. P rank the three contracts, C, D and E?
(f) How does Ms. A rank the three contracts, C, D and E?

 206 Chapter 6. Risk Sharing and Pareto Efficiency

6.7.5 Exercises for Section 6.5: The Edgeworth box when the parties have positive
initial wealth

initial wealth is W P = $1, 000, is risk averse with

Exercise 6.17 The Principal, whose √
utility-of-money function UP (m) = m + 1 − 1. The Agent, whose initial wealth is
W A = $800, is risk neutral. If they agree on a contract then the surplus in the good
state will be X g = $2, 000 and the surplus in the bad state will be X b = $1, 000. The
probability of the good state is pg = 15%.
Consider the following possible contracts:
1. Contract C: in each state the surplus is split equally between the Principal and the
Agent.
2. Contract D: in the good state the entire surplus goes to the Principal and, further-
more, the Agent gives $200 to the Principal, while in the bad state the Agent gets
the entire surplus (and there are no further payments by either party).
(a) What are the coordinates of point C in the extended Edgeworth box?
(b) Is contract C individually rational?
(c) Is contract C Pareto efficient?
(d) What are the coordinates of point D in the extended Edgeworth box?
(e) Is contract D individually rational?
(f) Is contract D Pareto efficient?
(g) Find a contract that is individually rational and leaves the Principal with zero
wealth in the good state.
(h) Is the contract of Part (g) Pareto efficient?


initial wealth is W P = $200, is risk averse with

Exercise 6.18 The Principal, whose √
utility-of-money function UP (m) = m + 1. The Agent, whose initial wealth
is
m
W A = $100, is risk averse with utility-of-money function UA (m) = ln 100 + 1 . If
they agree on a contract then the surplus in the good state will be X g = $700 and the
surplus in the bad state will be X b = $300. The two states are equally likely.
Consider the following contracts (where the first coordinate is the Agent’s wealth
in the good state and the second coordinate is the Agent’s wealth in the bad state):
C = (350, 0), D = (500, 300), E = (700, 600).
(a) Represent the three contracts as points in an extended Edgeworth box.
(b) Is contract C individually rational?
(c) Is contract C Pareto efficient?
(d) Is contract D individually rational?
(e) Is contract D Pareto efficient?
(f) Is contract E individually rational?
(g) Is contract E Pareto efficient?
(h) In the extended Edgeworth box, sketch the indifference curves of Principal and
Agent through the three contracts C, D and E.

 6.7 Exercises 207

6.7.6 Exercises for Section 6.6: More than two outcomes

Exercise 6.19 Suppose that there are three possible outcomes as shown in the following
table:
outcome: X1 = $500 X2 = $600 X3 = $800
1 3 12
probability: 16 16 16
The Principal
√ is risk averse with von Neumann-Morgenstern utility-of-money function
UP (m) = m, while the Agent is risk neutral. Consider the contract, call it C, where
the Agent is given $400 no matter what the outcome.
(a) Calculate the expected utility from contract C for the two parties.
(b) Find a contract, call it D, that (1) Pareto dominates C and (2) is Pareto efficient.


Exercise 6.20 Suppose that there are five possible outcomes:

outcome : X1 = $141 X2 = $164 X3 = $461 X4 = $645 X5 = $749

1 1 1 1 1
probability : 5 5 5 5 5

The Principal
√ is risk averse with von Neumann-Morgenstern utility-of-money function
UP (m) = m, while the Agent is risk neutral. Suppose that the Principal offers the
following contract, call it C, to the Agent: the Agent will be paid $20 no matter what
the outcome.
(a) What is the Principal’s expected utility from contract C? What is the Agent’s
expected utility?
(b) Consider an alternative contract, call it D: the Agent will be paid nothing is the
outcome is $141 or $164 or $461, while she will be paid $60 if the outcome is
either $645 or $749. Is contract D Pareto superior to contact C?


For Exercises 6.21 and 6.22 make use of Propositions 6.6.1 and 6.6.2 and the logic used in
the proof of Proposition 6.22.

Exercise 6.21 Suppose that there are four possible outcomes:

outcome : X1 = $141 X2 = $164 X3 = $461 X4 = $645

1 3 3 3
probability : 10 10 10 10

The Principal is risk neutral while the Agent
is risk averse with von Neumann-Morgenstern
m
utility-of-money function UP (m) = ln 100 . Consider the following contract (where the
first coordinate is what the Agent gets if the the outcome is $141, the second coordinate
is what she gets if the outcome is $164, etc.): C = (141, 164, 400, 500).
Is contract C Pareto efficient? If your answer is ‘Yes’, then explain why, if your answer
is ‘No’ then find a contract that Pareto dominates C. 
208 Chapter 6. Risk Sharing and Pareto Efficiency

Exercise 6.22 Suppose that there are three possible outcomes:

outcome : X1 = $164 X2 = $461 X3 = $645

3 3 4
probability : 10 10 10

The Agent is risk neutral while the Principal

√ is risk averse with von Neumann-Morgenstern
utility-of-money function UA (m) = m. Consider the following contracts (where the
first coordinate is what the Agent gets if the the outcome is $164, the second coordinate
is what she gets if the outcome is $461, etc.): C = (144, 100, 500), D = (0, 80, 200).
(a) Is contract C Pareto efficient? If your answer is ‘Yes’, then explain why, if your
answer is ‘No’ then find a contract that Pareto dominates C.
(b) Is contract D Pareto efficient? If your answer is ‘Yes’, then explain why, if your
answer is ‘No’ then find a contract that Pareto dominates D.


Exercise 6.23 Suppose that there are three possible outcomes:

outcome : X1 = $264 X2 = $561 X3 = $745

5 3 2
probability : 10 10 10

The Principal
√ is risk averse with von Neumann-Morgenstern utility-of-money function
UP (m) = m and the Agent is risk averse

with von Neumann-Morgenstern utility-of-
m
money function UA (m) = ln 100 . Consider the following contracts:
C = (w1 = 100, w2 = 100, w3 = 100), D = (w1 = 164, w2 = 336, w3 = 450).
(a) Is contract C Pareto efficient?
(b) Is contract D Pareto efficient?


Exercise 6.24 Suppose that there are three possible outcomes:

outcome : X1 = $400 X2 = $900 X3 = $1, 600

probability : p q 1− p−q

with 0 < p + q < 1.

The Principal and the Agent are risk neutral. Consider the following contract:
C = (w1 = 400, w2 = 0, w3 = 800). For what values of p and q is contract C Pareto
efficient? 
 6.8 Solutions to Exercises 209

6.8 Solutions to Exercises

Solution to Exercise 6.1.  

$700
Contract A gives rise to the following money lottery for the Agent: and therefore
1
an expected utility of UA (700) = 2(700) + 2 = 1402. The corresponding money lottery for
the Principal is
   
$(2, 400 − 700) $(1, 600 − 700) $(900 − 700) $1, 700 $900 $200
1 2 2 = 1 2 2
5 5 5 5 5 5

with corresponding expected utility of 51 ln(1, 700) + 25 ln(900) + 52 ln(200) = 6.328.

Contract B gives rise to the following lottery for the Agent:
 
$1, 200 $1, 000 $200
1 2 2
5 5 5

with an expected utility of 15 (2(1, 200) + 2) + 52 (2(1, 000) + 2) + 25 (2(200) + 2) = 1, 442.

The corresponding lottery for the Principal is
   
$(2, 400 − 1, 200) $(1, 600 − 1, 000) $(900 − 200) $1, 200 $600 $700
1 2 2 = 1 2 2
5 5 5 5 5 5

with corresponding expected utility of 51 ln(1, 200) + 25 ln(600) + 52 ln(700) = 6.597.

Thus both Principal and Agent prefer contract B to contract A, that is, contract A is
Pareto dominated by contract B. 

Solution to Exercise 6.2.

(a) Since U 00 (m) = − √1 < 0, John is risk averse.
4 m3
(b) Since V 00 (m) = 0, Joanne is risk neutral.
(c) Contract A gives rise to the following lottery for John:
 
$(4, 000 − 1, 000) = $3, 000 $(1, 600 − 1, 000) = $600
2 1
3 3
√ √
with corresponding expected utility of 23 3, 000 + 13 600 = 44.68. Joanne’s utility
from contract A is V (1,
 000) = 2(1, 000) = 2, 000. Contract B gives
 rise to the follow-
$(4, 000 − 1, 500) = $2, 500 $1, 600
ing lottery for John: 2 1 with corresponding
√ √ 3 3
expected utility of 23 2, 500 + 13 1, 600 = 46.67. Thus John prefers contract B to
contract A. For Joanne contract B gives rise to the following lottery
 
1, 500 0
2 1
3 3

with corresponding expected utility of 23 V (2, 500)+ 13 V (0) = 23 2(1, 500)+ 31 2(0) =
2, 000. Thus Joanne is indifferent between the two contracts. 
210 Chapter 6. Risk Sharing and Pareto Efficiency

Solution to Exercise 6.3.

(a) The Edgeworth box is shown in Figure 6.26.

origin for
the Agent
100 60 0A

60 60
D
40 80
C

0P 100 140
origin for
the Principal

Figure 6.26: The Edgeworth box for Exercise 6.3.

(b) From
 the point 
of view of the Principal, contract C corresponds to the
 money lottery

$140 $40 $100 $60
1 3 and contract D corresponds to the money lottery 1 3
4 4 4 4
The expected value of C is 14 140 + 43 40 = 65 and the expected value of D is
1 3
4 100 + 4 60 = 70. Thus a risk-neutral Principal prefers contract D to contract
C.

(c) From
 the point
 of view of the Agent, contract C corresponds to the
 money lottery

$60 $80 $100 $60
1 3 and contract D corresponds to the money lottery 1 3
4 4 4 4
The expected value of C is 14 60 + 34 80 = 75 and the expected value of D is
1 3
4 100 + 4 60 = 70. Thus a risk-neutral Agent prefers contract C to contract D. 
6.8 Solutions to Exercises 211

Solution to Exercise 6.4.

(a) The Principal is risk averse.

(b) The Agent is risk neutral.
(c) The Principal’s ranking is as follows:

A P E P C P B P D.

(d) The Agent’s ranking is as follows:

D A E A C ∼A B A A.

(e) The following pairs:

• C Pareto dominates B (since C P B and C ∼A B).
• E Pareto dominates B (since E P B and E A B).
• E Pareto dominates C (since E P C and E A C). 

Solution to Exercise 6.5.

For the Principal the expected utility of contract C is
√ √
E[UP (C)] = 41 140 + 34 40 = 7.7 and the expected utility of contract D is
√ √
E[UP (D)] = 14 100 + 34 60 = 8.31. Thus the Principal prefers D to C.

For the Agent the expected utility of contract C is E[UA (C)] = 41 ln (60) + 43 ln (80) = 4.31
and the expected utility of contract D is E[UA (D)] = 41 ln (100) + 34 ln (60) = 4.22. Thus
the Agent prefers C to D.

Hence it is neither the case that C Pareto dominates D nor the case that D Pareto
dominates C. 

Solution to Exercise 6.6.

(a) Only contract D.

(b) Only contract B.
(c) None of them.
(d) All of them. 
212 Chapter 6. Risk Sharing and Pareto Efficiency

Solution to Exercise 6.7.

(a) The Principal is risk averse (UP00 (m) = − 5,000

1
< 0) and the Agent is risk neutral
00
(UA (m) = 0). Thus, since contract C is in the interior of the Edgeworth box (that
is, 0 < wCg < X g and 0 < wCb < X b ), Pareto efficiency requires that the contract lie
on the 45o line out of the Principal’s origin, which is not the case with contract C
since X g − wCg = 800 − 400 = 400 6= 200 = 400 − 200 = X b − wCb . Hence contract
C is not Pareto efficient.

(b) See Figure 6.27.

(c) One such contract is contract D shown in Figure 6.27. There are many more: any
contract on the portion of the straight-line indifference curve of the Agent which lies
above the indifference curve of the Principal. Of all those contracts, contract D is
the only one that is Pareto efficient. 

400 0A

200
C Pic
45o
Aic

0P
Figure 6.27: The Edgeworth box for Exercise 6.7. Pic stands for "indifference curve of the
Principal" and Aic for "indifference curve of the Agent".
6.8 Solutions to Exercises 213

Solution to Exercise 6.8.

(a) Since the Principal is risk neutral and the Agent is risk averse, a Pareto efficient
contract in the interior of the Edgeworth box must be on the 45o line out of the origin
for the Agent, that is, it must be a contract that guarantees a fixed amount w to the
Agent. Thus w is the solution to

1 3
(1, 206 − w) + (676 − w) = 232.5
4 4

which is w = 576.
√
(b) The Agent’s expected utility is 576 = 24.
√ √
(c) It must be a contract of the form C = (wg , 676) with 14 wg + 34 676 = 27. The
solution is wg = 900. Thus the contract is C = (900, 676).
(d) At contract C = (900, 676) the Principal’s expected utility is 14 (1, 206 − 900) +
3
4 (0) = 76.5. 

Solution to Exercise 6.9.

(a) The Principal is risk-averse, since U 00 (m) = − m12 < 0.

1 − 100 m
(b) The Agent is also risk-averse, since V 00 (m) = − 100 e < 0 (recall that d x
dx e = ex ).
(c) Since the proposed contract is in the interior of the Edgeworth box, Pareto efficiency
0 (X g −wg ) 0 g) x
requires that U
U 0 (X b −wb )
= VV 0 (w
(wb )
. Since U 0 (m) = m1 and V 0 (m) = e− 100 ,

1
U 0 (X g − wg ) U 0 (600) 600 400 2
= = = = = 0.67
U 0 (X b − wb ) U 0 (400) 1
400
600 3

and

V 0 (wg ) V 0 (400) e−4

0 b
= 0 = −2 = e−2 = 0.135.
V (w ) V (200) e

Thus the contract is not Pareto efficient.

(d) From the above calculations we see that the Principal’s indifference curve through
the proposed contract is steeper than the Agent’s indifference curve. Thus any
contract in the shaded region in Figure 6.28 is Pareto superior to the given contract.
Hence wg should be decreased and wb should be increased. 
214 Chapter 6. Risk Sharing and Pareto Efficiency

400 0A
Pic
45o

200

Aic

45o

0P

Figure 6.28: The Edgeworth box for Exercise 6.9.

Solution to Exercise 6.10.

(a) X g = 1, 000 + 1, 000 = 2, 000.
(b) X b = 650 + 550 = 1, 200.
(c) Since the Agent is risk neutral, we can take the identity function as her utility
function: UA (m) = m. Thus E[UA (C)] = 25 (1, 000) + 35 (550) = 730. The expected
√ √
utility for the Principal is E[UP (C)] = 25 1, 000 + 35 650 = 27.9462.
(d) Contract D = (wgD , wbD ) is such that 2, 000 −q
wgD = 1, 200 − wbD (since it is on the 45o
line out of the origin for the Principal) and 2, 000 − wgD = 27.9462 (since it is on
the Principal’s indifference curve through point C). From the latter equation we get
wgD = 1, 219.01 and thus, from the first equation we get wbD = 419.01. At contract
D the Principal’s expected utility is the same as at C, namely 27.9462, while the
Agent’s expected utility is E[UA (D)] = 25 (1, 219.01) + 35 (419.01) = 739.01. Thus
contract D Pareto dominates contract C.
(e) Contract E = (wgE , wbE ) is such that 2, 000 − wgE = 1, 200 − wbE (since it is on the
45o line out of the origin for the Principal) and 25 (wgE ) + 35 (wbE ) = 730 (since it is
on the Agent’s indifference curve through point C). Solving this system of two
equations we get wgE = 1, 210 and wbE = 410. At contract E the Agent’s expected
utility
√ is the same as at C, namely 730, while the Principal’s expected utility is
2, 000 − 1, 210 = 28.1069. Thus contract E Pareto dominates contract C.
(f) In Parts (d) and (e) we already found two contracts that Pareto dominate contract C,
namely contracts D = (1219.01, 419.01) and E = (1210, 410). To find a third one
we can take the mid point between D and E on the 45o line out  of the origin for the
1,219.01+1,210 419.01+410
Principal, call it point F; then F = 2 , 2 = (1214.51, 414.51).
Of course, any other point between the two indifference curves would have been a
possible choice. 
6.8 Solutions to Exercises 215

Solution to Exercise 6.11.

Since contract C = (600, 300) is such that 0 < wg < X g and 0 < wb < X b (that is, it is in
the interior of the Edgeworth box) a necessary and sufficient condition for it to be Pareto
efficient is
UP0 (X g − wg ) UA0 (wg ) UP0 (400) UA0 (600)
= that is, = .
UP0 (X b − wb ) UA0 (wb ) UP0 (100) UA0 (300)
√
UP0 (400) 100 1 UA0 (600) 300 1
Indeed, = √ = and = = .
UP0 (100) 400 2 UA0 (300) 600 2
Thus the contract is Pareto efficient. 
Solution to Exercise 6.12.
(a) Contract C is not Pareto efficient, because it is in the interior of the Edgeworth box
and yet does not lie on the 45o line out of the Principal’s origin.
(b) See Figure 6.29.
(c) For the Agent we can take as utility-of-money function the identity function:
UA (m) = m. Thus the Agent’s expected utility from contract C is its expected value,
namely 23 250 + 13 150 = 6503 = 216.667. Inside the Edgeworth box there is no point
on the 45o line out of the Principal’s origin that has the same expected value as point
C. In fact it would have to be a point (wg , wb ) such that (1) 1, 000 − wg = 600 − wb
and (2) 23 wg + 13 wb = 650
3 and the only point that satisfies these two inequalities is
g b
the point (w = 350, w = −50) which lies outside the Edgeworth box. Thus the
point we are seeking must be on the top side of the Edgeworth box, that is it must be
a point D = (wg , 0) such that 23 wg = 650
3 ; hence D = (325, 0).
(d) See Figure 6.29. 

Aic

325 250 0A
D
Pic

150
C
Pic

45o

0P

Figure 6.29: The Edgeworth box for Exercise 6.12.

216 Chapter 6. Risk Sharing and Pareto Efficiency

Solution to Exercise 6.13.

(a) Contract C is not Pareto efficient, because it is in the interior of the Edgeworth box
and yet does not lie on the 45o line out of the Agent’s origin.

(b) See Figure 6.30.

√ √
(c) For the Agent the expected utility from contract C is 79 1, 559 + 29 1, 004 =
37.7512. There is no point inside the Edgeworth and on the 45o line out of the
Agent’s origin that has the same expected utility√ as point √
C. In fact it would have to be
g b g b 7 2
a point (w , w ) such that (1) w = w and (2) 9 w + 9 wb = 37.7512 and the only
g

point that satisfies these two inequalities is the point (wg = 1425.16, wb = 1425.16)
which lies outside the Edgeworth box (since wb = 1425.16 > X b = 1, 200). Thus
the point we are seeking must be on the bottom√side of√the Edgeworth box, that
is it must be a point D = (wg , 1200) such that 79 wg + 29 1200 = 37.7512; hence
D = (1493.04, 1200).

(d) See Figure 6.30. 

1,559 0A
45o

Aic
C
196 1,004
D
0P 441

Pic

Figure 6.30: The Edgeworth box for Exercise 6.13.

6.8 Solutions to Exercises 217

Solution to Exercise 6.14.

(a) Contract C is not Pareto efficient, because it is in the interior of the Edgeworth box
and the indifference curves of Principal and Agent are not tangent at that point:

√
UP0 (441) 196 2
0 =√ = = 0.667
UP (196) 441 3

while
√
UA0 (1, 559) 1, 004
0 =√ = 0.8025.
UA (1, 004) 1, 559
√ √
(b) For the Principal the expected utility of contract C is 79 441 + 29 196 = 175 =
7√ 2
√9
19.4444 and the expected utility of contract D is 9 2, 000 − 1, 493.04 + 9 0 =
17.5123. Thus the Principal prefers contract C to contract D.

(c) As calculated in Exercise 6.13, the Agent is indifferent between contract C and
contract D.

(d) It follows from Parts (b) and (c) that contract C Pareto dominates contract D; hence
contract D is not Pareto efficient. 

Solution to Exercise 6.15.

(a) A contract of the form (wg , 0) lies on the top side of the Edgeworth box and thus
is Pareto efficient if and only if the Principal’s indifference curve at that point is as
steep as, or less steep than, the Agent’s indifference curve (see Figure 6.15 on page
186), that is, if and only if

UP0 (3, 000 − wg ) UA0 (wg ) 2, 800 1, 000

≤ 0 that is ≤ g .
UP0 (1, 800) UA (0) 4, 000 − wg w + 1, 000

6,000
The solution is wg ≤ 19 = 315.7895.
(b) A contract of the form (wg , 1800) lies on the bottom side of the Edgeworth box and
thus is Pareto efficient if and only if the Principal’s indifference curve at that point is
as steep as, or steeper than, the Agent’s indifference curve (see Figure 6.15 on page
186), that is, if and only if

UP0 (3, 000 − wg ) UA0 (wg ) 1, 000 2, 800

≥ that is ≥ g .
UP0 (0) UA0 (1, 800) 4, 000 − wg w + 1, 000

51,000
The solution is wg ≥ 19 = 2, 684.2105. 
218 Chapter 6. Risk Sharing and Pareto Efficiency

Solution to Exercise 6.16.

45
√ √
(a) Mr. P’s expected utility from contract C is: 98 520 − 144+ 53
98 200 − 90 = 14.576.
(b) For Ms. A we can take as utility function the identity function, so that the expected
utility of a contract coincides with its expected value. The expected value of contract
C is: 45 53
98 144 + 98 90 = 114.796.
(c) Since P is risk-averse and A is risk neutral, the Pareto efficient contracts in the interior
of the Edgeworth box must lie on the 45o out of the origin for the Principal. Let us
see if there is such a contract which, furthermore, is as good as contract C for the
Principal.
√ Then it must be a contract (wg , wb ) such that (1) 520 − wg = 200 − wb and
(2) 520 − wg = 14.576. From (2) we get that wg = 307.54 and from this and (1)
we get that wb = −12.46. This contract is not feasible, because it does not lie inside
the Edgeworth box. Thus we need to find a contract on the q top side of the Edgeworth
√
box, that is, a contract of the form D = (wgD , 0) such that 45 g 45
98 520 − wD + 98 200 =
14.576. The solution is wgD = 292.38. Hence D = (292.38, 0). Ms. A indeed prefers
contract D to contract C since the expected value of D is 45
98 292.38 = 134.256.
(d) A similar reasoning would lead us to look for a point on the 45o out of the origin for
the Principal which lies on the straight-line indifference curve of Ms. A that goes
through point C, that is a point (wg , wb ) such that (1) 520 − wg = 200 − wb and (2)
45 g 53 b
98 w + 98 w = 114.796. However, such a point lies outside the Edgeworth box (it
is the point (287.86, −32.14)). Hence we need to find a point on the top side of the
Edgeworth box that Ms. A finds just as good as C. Call this point E = (wgE , 0). Then
g g
it must be that 45
98 wE = 114.796. The solution is wE = 250, that is, E = (250, 0).
√ √
(e) Mr. P’s ranking is E P C ∼P D since E[UP (E)] = 45 98 520 − 250 + 53
98 200 =
15.1935 > 14.576 = E[UP (C)] = E[UP (D)].
(f) From the previous calculations it is clear that D A C ∼A E (E[D] = 134.256 >
E[C] = E[E] = 114.796). 

Solution to Exercise 6.17.

(a) C = (1800, 1300) (the first coordinate is the Agent’s wealth in the good state and the
second coordinate is her wealth in the bad state).
(b) Yes, contract C is individually rational.

85The√ expected utility
of contract C for
15 √
the Principal is 100 2, 000 + 1 − 1 + 100 √1, 500 + 1 − 1 = 38.641 which is
greater than the utility from his initial wealth 1, 000 + 1 − 1 = 30.639. For the
Agent we can take as utility function the identity function so that the expected utility
15 85
of contract C is 100 1800 + 100 1300 = 1, 375 which is greater than the utility from
her initial wealth, namely 800.
(c) Contract C is not Pareto efficient because it is in the interior of the extended Edge-
worth box and not on the 45o line out of the origin for the Principal (who is the only
risk-averse party).
(d) D = (600, 1800) (again, the first coordinate is the Agent’s wealth in the good state
and the second coordinate her wealth in the bad state).
6.8 Solutions to Exercises 219
15 √


(e) The expected utility of contract D for the Principal is 100 3, 200 + 1 − 1 +
85 √


100 1,
√ 000 + 1 − 1 = 34.379 which is greater than the utility from his initial
wealth 1, 000 + 1 − 1 = 30.639. For the Agent the expected utility of contract D is
15 85
100 600 + 100 1800 = 1, 620 which is greater than the utility from her initial wealth,
namely 800. Thus contract D is individually rational.
(f) Like contract C, contract D is not Pareto efficient because it is in the interior of the
extended Edgeworth box and not on the 45o line out of the origin for the Principal
(the only risk-averse party).
(g) Let us see if there is a contract of the form (3800, x) (again, the first coordinate is
the Agent’s wealth in the good state and the second coordinate her wealth in the bad
state) that the Principal considers to be at least
√ as good
 as keeping his initial wealth
15 85 √


with no agreement, that is, such that 100 1 − 1 + 100 2, 800 − x + 1 − 1 ≥
√
1, 000 + 1 − 1 = 34.379; then it must be that x ≤ 1, 428.64.
Let us try contract E = (3800, 1428). The Principal then prefers contract E to
keeping his initial wealth with no agreement. For the Agent the expected utility of
15 85
contract E is 100 3800 + 100 1428 = 1, 783.8, greater than her reservation utility of
800. Thus E is individually rational.
(Any other contract of the form (3800, x) with 270.59 ≤ x ≤ 1, 428.64 is also indi-
vidually rational.)
(h) Since contract E lies on the left side of the extended Edgeworth box it is not Pareto
efficient. In fact, at that point the slope of the Principal’s indifference curve is, in
15
3
absolute value, greater than 100
85 = 17 (because point E lies above the 45o line for the
100
Principal), which is the absolute value of the slope of the straight-line indifference
curve of the Agent. Hence there are contracts inside the extended Edgeworth box
that lie between the two indifference curves through point E and such contracts
Pareto dominate contract E. 

Solution to Exercise 6.18.

(a) See Figure 6.31
(b) It is clear that the Principal prefers contract C to keeping his wealth of $200 with no
agreement. Thus we only need to check if the Agent views contract C to be at least
as good as keeping her wealth of $100 with no agreement. UA (100) = ln(2) = 0.693;
the Agent’s expected utility from contract C is 12 ln(4.5) + 12 ln(1) = 0.752. Thus the
Agent prefers contract C to keeping her wealth. Hence contract C is individually
rational.
(c) Contract C is Pareto efficient if and only if the indifference curve of the Principal at
point C is less steep than, or as steep as, the indifference curve of the Agent at that
point (see Figure 6.16 on page 187), that is, if and only if
√1 1
UP0 (650) UA0 (350) 2 651 450
≤ , that is, ≤
UP0 (600) UA0 (0) √1 1
100
2 601

√1 1
2 651 450
which is not the case since = 0.961 and = 0.222. Thus contract C is not
√1 1
2 601 100
220 Chapter 6. Risk Sharing and Pareto Efficiency

700 500 350 100

0A
C
100

300
D
200

600
0P 200 E

Figure 6.31: The Edgeworth box for Part (a) of Exercise 6.18.

Pareto efficient.
√ √
(d) For the Principal the expected utility of contract D is 12 501 + 12 301 = 19.866
√
which is greater than 201 = 14.177 (the utility of his initial wealth), hence the
Principal prefers contract D to keeping his initial wealth. For the Agent the expected
utility of contract D is 12 ln(6) + 12 ln(4) = 1.589 which is greater than ln(2) = 0.693
(the utility of her initial wealth), hence the Agent prefers contract D to keeping her
initial wealth. Hence contract D is individually rational.
(e) Contract D is Pareto efficient if and only if the indifference curve of the Principal at
point D has the same slope as the indifference curve of the Agent at that point, that
is, if and only if

√1 1
UP0 (500) UA0 (500) 2 501 600
= , that is, =
UP0 (300) UA0 (300) √1 1
400
2 301

√1 1
2 501 600
which is not the case since = 0.775 while = 0.667. Thus contract D is
√1 1
2 301 400
not Pareto efficient.
√ √
(f) For the Principal the expected utility of contract E is 12 301 + 12 1 = 9.175 which
√
is less than 201 = 14.177 (the utility of his initial wealth), hence the Principal
prefers keeping his initial wealth to contract E, so that contract E is not individually
rational.
(g) Contract E is Pareto efficient if and only if the indifference curve of the Principal
is steeper than, or as steep as, the indifference curve of the Agent at that point (see
6.8 Solutions to Exercises 221

Figure 6.16 on page 187), that is, if and only if

√1 1
UP0 (300) UA0 (700) 2 301 800
≥ 0 , that is, ≥
UP0 (0) UA (600) 1
√ 1
700
2 1

√1 1
2 301 800
which is not the case since 1 = 0.058 and 1 = 0.875. Thus contract E is not
√
2 1 700
Pareto efficient.
(h) See Figure 6.32. 

700 500 350 100

0A
C
100

300
D
200

600
0P 200 E Pic

Aic

Figure 6.32: The Edgeworth box for Part (h) of Exercise 6.18.

Solution to Exercise 6.19.

(a) Expected for the Principal is
1√ 3√ 12 √
500 − 400 + 600 − 400 + 800 − 400 = 18.277.
16 16 16
For the Agent we can take the identity function as her utility function and thus her
utility from contract C is 400.
(b) Since the Principal is risk averse and the Agent is risk neutral, Pareto efficiency
requires that the Agent bear all the risk (i.e. the Principal must be guaranteed
a fixed income). There are many contracts that are Pareto efficient and Pareto
dominate contract C. One example is that fixed-income (for the Principal) con-
tract, call it D, that the Principal considers to be just as good as contract C; to
222 Chapter 6. Risk Sharing and Pareto Efficiency
√
find such a contract solve the equation x = 18.277 to obtain x = 334.05. Then
D = (500−334.05, 600−334.05, 800−334.05) = (165.95, 265.95, 465.95) and the
Agent prefers D to C since for her the expected utility from contract D is
1 3 12
165.95 + 265.95 + 465.95 = 409.7
16 16 16


Solution to Exercise 6.20.

(a) The Principal’s expected utility from contract C is
1√ 1√ 1√ 1√ 1√
141 − 20 + 164 − 20 + 461 − 20 + 645 − 20 + 749 − 20
5 5 5 5 5
1
= (11 + 12 + 21 + 25 + 27) = 19.2
5
For the Agent we can take the identity function as her utility function and thus her
utility from contract C is 20.
(b) Under contract D the Agent’s expected utility is 35 0 + 25 60 = 24 and thus the Agent
prefers contract D to contract C. Under contract D the Principal’s expected utility is:
1 √ √ √ √ √ 
141 + 164 + 461 + 645 − 60 + 749 − 60 = 19.3174.
5
Thus also the Principal prefers contract D to contract C. Hence contract D Pareto
dominates contract C. 

Solution to Exercise 6.21.

It follows from Proposition 6.6.1 that contract C is not Pareto efficient, because
w3 = 400 < w4 = 500 and yet w3 6= X3 = 461. To find a contract that Pareto dominates
contract C, let us use the method employed in the proof of Proposition 6.6.1: replace w3
and w4 with ŵ = p3p+p
3
4
w3 + p3p+p
4
4
w4 = 12 400 + 12 500 = 450; that is, consider the contract
B = (141, 164, 450, 450). For the Agent

1 3 6
E[UA (B)] = ln(1.41) + ln(1.64) + ln(4.5) = 1.0852
10 10 10
1 3 3 3
E[UA (C)] = ln(1.41) + ln(1.64) + ln(4) + ln(5) = 1.0815.
10 10 10 10
Thus the Agent prefers B to C. For the Principal we can take the identity function as his
utility-of-money function so that
1 3 3 3
E[UP (B)] = 0 + 0 + (461 − 450) + (645 − 450) = 61.8
10 10 10 10
1 3 3 3
E[UP (C)] = 0 + 0 + (461 − 400) + (645 − 500) = 61.8
10 10 10 10
Thus the Principal is indifferent between B and C and therefore contract B Pareto dominates
contract C. 
6.8 Solutions to Exercises 223

Solution to Exercise 6.22.

(a) Contract C is not Pareto efficient because, contrary to Proposition 6.6.2, X2 − w2 =
361 > X3 −w3 = 145. From the point of view of the Principal, contract C corresponds
to the money lottery
 
$20 $361 $145
3 3 4 .
10 10 10
Let us create a new lottery for the Principal, call it E, where we replace outcomes
$361 and $145 with an amount x such that the Principal is indifferent between the
new lottery and the lottery corresponding to contract C. Since for the Principal

3√ 3√ 4√
E[UP (C)] = 20 + 361 + 145 = 11.858,
10 10 10
we want x to be such that
3√ 7√
E[UP (E)] = 20 + x = 11.858.
10 10
Thus x = 225.70. It remains to show that the Agent prefers contract E to con-
tract C. E is the contract (w1 = 144, w2 = 461 − 225.70, w3 = 645 − 225.70) =
(144, 235.3, 419.3). Since the Agent is risk neutral, she ranks contracts according to
their expected value:

3 3 4
E[C] = 144 + 100 + 500 = 273.2 and
10 10 10

3 3 4
E[E] = 144 + 235.3 + 419.3 = 281.51.
10 10 10
Thus the Agent prefers E to C and hence E Pareto dominates C.
(b) Contract D is not Pareto efficient because, contrary to Proposition 6.6.2, although
X1 − w1 = 164 < X2 − w2 = 461 − 80 = 381 < X3 − w3 = 645 − 200 = 445, it is not
the case that w2 = 0. Let us find an alternative contract, call it F, that the Principal
considers just as good as contract D and gives the Principal the same income is states
2 and 3. Since
3√ 3√ 4√
E[UP (D)] = 164 + 381 + 445 = 18.1357,
10 10 10
we need to solve the equation

3√ 7√
164 + x = 18.1357.
10 10
The solution is x = 416.966. Thus F = (0, 461−416.966, 645−416.966) = (0, 44.034, 228.034).
Since the Agent is risk neutral, she ranks contracts according to their expected value:
3 4 3 4
E[D] = 10 80 + 10 200 = 104 and E[F] = 10 44.034 + 10 228.034 = 104.424. Thus
the Agent prefers contract F to contract D, so that D is Pareto dominated by F.

224 Chapter 6. Risk Sharing and Pareto Efficiency

Solution to Exercise 6.23.

(a) Contract C = (100, 100, 100) is not Pareto efficient because
√
UP0 (X1 − w1 ) UP0 (164) 461 UA0 (w1 ) UA0 (100)
= = √ = 1.6766 6
= = = 1.
UP0 (X2 − w2 ) UP0 (461) 164 UA0 (w2 ) UA0 (100)

(b) Contract D = (164, 336, 450) is not Pareto efficient because

√
UP0 (X1 − w1 ) UP0 (100) 225 UA0 (w1 ) UA0 (164)
= = √ = 1.5 6
= = = 2.0488.
UP0 (X2 − w2 ) UP0 (225) 100 UA0 (w2 ) UA0 (336)

. 

Solution to Exercise 6.24.

When both parties are risk neutral, every contract is Pareto efficient, no matter what the
probabilities are. Thus the answer is: for all (meaningful) values of p and q. 
III
Asymmetric Information:
Adverse Selection

7 Adverse Selection or Hidden Type

227
7.1 Adverse selection or hidden type
7.2 Conditional probability and belief updating
7.3 The market for used cars
7.4 Exercises
7.5 Solutions to Exercises

8 Adverse Selection in Insurance 253

8.1 Adverse selection in insurance markets
8.2 Two types of customers
8.3 The monopolist under asymmetric information
8.4 A perfectly competitive insurance industry
8.5 Exercises
8.6 Solutions to Exercises
 7. Adverse Selection or Hidden Type

7.1 Adverse selection or hidden type

The expression ‘adverse selection’, or ‘hidden type’, refers to situations in which one
party to a contract (e.g. a buyer) possesses information relevant to the contract that is
not available to the opposing party (e.g. a seller). Thus it is a situation of asymmetric
information.
For example, in the context of health insurance, the insurance company (the seller
of insurance) is typically aware of the fact that there are individuals who – because of
their family history – are at a higher risk of developing a condition that requires extensive
medical services, while other individuals represent a lower risk. If the insurance company
offers a contract that would, on average, be profitable if everybody (high-risk and low-risk
individuals) were to purchase that contract, it might discover that its costs are much higher
than expected, because only (or mostly) the high-risk individuals ended up purchasing
the contract.1 In a situation of adverse selection the uninformed party realizes that the
terms of a proposed contract determine the composition of the pool of individuals who
will find that contract acceptable and a change in the contract that would be beneficial to
the uninformed party, if everybody found the contract acceptable, might have undesirable
consequences because of an adverse change in the proportion of “bad types” within the
pool of applicants. Adverse selection in insurance markets will be discussed in the next
chapter. In this chapter we will discuss how adverse selection arises in other markets and
what its consequences are.

1 Thiswas the rationale behind the original provision of the Affordable Care Act that established a
mandate for individuals to purchase health insurance.
228 Chapter 7. Adverse Selection or Hidden Type

The phenomenon of adverse selection was brought to the attention of economists by

George Akerlof in 1970.2 Akerlof analyzed the market for second-hand cars, which will
be discussed in Section 7.3. Here we begin with a simple illustration of the phenomenon
of adverse selection in a different context.
Suppose that you have just opened an all-you-can-eat buffet restaurant in a small facility
that can accommodate up to a maximum of 50 customers. Since this is your first business
venture, you are uncertain as to how much to charge and decide to start with a low price of
$6. On the first day you notice that, at this price, there is excess demand: your restaurant is
full (all 50 seats are taken) and you had to turn away some potential customers. At the end
of the first day you calculate that, on average, each customer consumed an amount of food
that cost you $4.50, so that you made a profit of $1.50 per customer, for a total profit of
$(1.50 × 50) = $75. Having recently graduate with a degree in economics you remember
that when there is excess demand the price should be increased. The potential downside
of increasing the price is that the higher price might drive away too many customers and
you might end up not being able to fill your restaurant; so you decide to first try a modest
price increase: from $6 to $7. To your delight, on the second day - with the higher price
of $7 - there is again excess demand: all the 50 seats are taken and, again, some people
had to be turned away. At the end of the second day, you calculate your total profit for the
day, expecting it to be $(7 × 50 − 4.5 × 50) = $125 but to your dismay you realize that it
is only $50, less than on the first day when your price was lower! What happened?
The reduction in profit is not due to a drop in demand, because you still served the same
number of customers, namely 50. In fact, your revenue did increase: from $(6×50) = $300
to $(7 × 50) = $350; hence the adverse effect of the price increase occurred on the cost
side of the equation: your total costs increased from $225 to $300. How could this happen?
A simple explanation is as follows. There are two types of potential customers: the light
eaters, denoted by L, and the heavy eaters, denoted by H. Each L-type consumes a small
amount of food that costs $3, while each H-type consumes a large amount of food that
costs $6. Because of their moderate consumption, the L-types are not willing to pay more
than $6 for a buffet, while the H-types are willing to pay up to $7 (or even more). There
are as many L-types as H-types in the population and their arrival at your restaurant is
random, so that the probability of each served customer being of one type or the other is
the same, namely 50%, provided that both types are in fact willing to enter your restaurant:
this is the case if the price is $6 but not if the price is $7. In other words, if you charge
only $6, admitting a customer corresponds to playing a lottery where, with probability
1 1
2 the cost of serving that customer will be $3 and with probability 2 the cost of serving
that customer will be $6, so that the expected cost of serving any one customer is $4.50
and thus the expected profit from any one customer is $(6 − 4.50) = $1.50. On the other
hand, if you charge a price of $7, then only the H-types will enter your restaurant and
thus admitting a customer corresponds to playing a lottery where, with probability 1, the
cost of serving that customer will be $6, with a corresponding profit of $(7 − 6) = $1 per
customer. In other words, increasing the price from $6 to $7 changed the composition of

2 In
“The market for ‘lemons’: qualitative uncertainty and the price mechanism”, Quarterly Journal of
Economics, 1970, Vol. 84, pp. 488-500. The Nobel Memorial Prize in Economics was awarded in 2001
to George Akerlof, Michael Spence and Joseph Stiglitz "for their analyses of markets with asymmetric
information".
 7.2 Conditional probability and belief updating 229

your 50 customers
   
Type L Type H Type H
from to .
50% 50% 100%

This simple example illustrates the logic of the adverse selection phenomenon.
In order to be able to develop a more general analysis, we need to take a side tour into
the topic of conditional probability and belief updating. To illustrate the reason for this,
consider te following, slightly more complex, version of the above example, where instead
of two types there are three: L, M and H. The types are characterized as follows (where
‘reservation price’ is the maximum price that the individual is willing to pay for a buffet):

type: L M H
2 3 1
fraction of total population: 6 6 6
reservation price: $6 $6.50 $7
cost of food consumed: $3 $6 $9

If you charge $6 for the buffet then all types will be willing to enter your restaurant.
Thus (assuming that the probability of any given customer being of a particular type is
equal to the fraction of that type in the population) from the point of view of cost admitting
a customer correspond to playing the following lottery:
 
cost of serving the customer: $3 $6 $9
 
2 3 1
probability: 6 6 6

so that the expected cost of serving each customer is 26 3 + 63 6 + 16 9 = $5.50. On the other
hand, if you charge $6.50 then the L-types will not enter your restaurant and thus, from
the from point of view of cost, admitting a customer correspond to facing the following
lottery:  
cost of serving the customer: $6 $9
 
probability: p 1− p
What is the value of p? This question is answered in the following section.

7.2 Conditional probability and belief updating

So far we have been focussing on money lotteries representing uncertainty concerning
possible levels of wealth or possible levels of profit or cost, etc. More generally, we can use
lotteries to represent uncertainty about anything. Let as call a possible item of uncertainty
a “state of the world”. Then an individual’s uncertainty can be represented by a set, listing
all the states of the world that the individual considers possible.
Consider, for example, the state of uncertainty of a doctor who, after listening to
her patient’s symptoms, reaches the conclusion that there are only five possible causes:
(1) a bacterial infection (call this state a), (2) a viral infection (state b), (3) an allergic
reaction to a drug (state c), (4) an allergic reaction to food (state d) and (5) environmental
 230 Chapter 7. Adverse Selection or Hidden Type

bacterial viral drug food environmental

infection infection allergy allergy factors
a b c d e

Figure 7.1: The doctor’s initial state of uncertainty

factors (state e). Then we can represent the doctor’s state of uncertainty by the set
{a, b, c, d, e}, as shown in Figure 7.1.
Of all the states that are possible, the individual might consider some to be more likely
than others and might even dismiss some states as “extremely unlikely” or “implausible”.
To represent the individual’s probabilistic beliefs we use a probability distribution over the
set of states.
Continuing the doctor’s example, suppose that - based upon her past experience with
similar cases - she considers a bacterial infection (state a) to be twice as likely as a viral
infection (state b), she considers a food allergy (state d) to be three times as likely as a drug
allergy (state c), which - in turn - she considers to be as likely as environmental factors
(state e); that is, she has the following beliefs:
 
state: a b c d e
 .
4 2 1 3 1
probability: 11 11 11 11 11
Suppose that the doctor can acquire further information by ordering a blood test. How
should she revise her beliefs when she learns the result of the blood test? In order to
answer this question we need to introduce the notion of conditional probability and be
more precise about what we mean by information.

7.2.1 Conditional probability

Let us frame the issue of uncertainty in general terms. Let U (‘U’ stands for ‘universal
set’) be a finite set of possibilities. An element s ∈ U is called a state and any subset E of
U (E ⊆ U) is called an event. A probability distribution P on U is a function P → [0, 1]
that assigns to every state s ∈ U a number P(s), called the probability of state s, such that
0 ≤ P(s) ≤ 1. It is required that the probabilities add up to 1: ∑ P(s) = 1. Given an
s∈U
event E ⊆ U, we denote by P(E) the probability of event E and define it as the sum of the
probabilities of the states in E: P(E) = ∑ P(s). Denote the empty set (which is a subset
s∈E
of U) by ∅; then we set, by definition, P(∅) = 0.
Let s ∈ U be a state and E ⊆ U an event. While P(s) is the unconditional probability of
state s, we can also define the probability of s conditional on event E, denoted by P(s|E).
It is meant to capture the new probability that one should assign to state s if informed that
the "true state" is an element of E and it is defined as follows:

 0 if s ∈
/E
if P(E) > 0 then P(s|E) = . (7.1)
 P(s) if s ∈ E
P(E)
 7.2 Conditional probability and belief updating 231

Note that the probability conditional on an event E is defined only if P(E) > 0, that is, if
E is an event that has positive unconditional probability.
Conditional probability can also be extended to events, as follows. Let D and E be
events with P(E) > 0. Then the probability of D conditional on E, denoted by P(D|E) is
defined as3
P(D|E) = ∑ P(s|E). (7.2)
s∈D

7.2.2 Belief updating

Sometimes uncertainty can be reduced by acquiring information. Indeed, information can
be thought of as “reduction of uncertainty”. Continuing the example of the doctor, whose
initial beliefs are represented by the probability distribution
 
state: a b c d e
 ,
4 2 1 3 1
probability: 11 11 11 11 11

suppose that the doctor can order a blood test: a positive blood test will reveal that there is
an infection and rule out states c, d and e, while a negative blood test will reveal that there is
no infection, thus ruling out states a and b. We can represent the information obtained from
a blood test as a set of states (an event): the set {a, b} represents the information conveyed
by a positive blood test, while the set {c, d, e} represents the information conveyed by a
negative blood test.
Suppose that the doctor orders the blood test and it comes back positive. How should
she revise her beliefs in light of this piece of information?
The issue of how to “rationally” modify one’s initial beliefs - expressed as a probability
distribution P on a finite set U - after receiving an item of information (represented by a
subset F of U) has been studied extensively by philosophers and logicians. Two different
situations may arise:
• In one case, the item of information F was not ruled out by the initial beliefs, in
the sense that event F was assigned positive probability (P(F) > 0). Information
might still be somewhat surprising, in case P(F) is small (close to zero), but it is not
completely unexpected. We call this case belief updating.
• The other case is where the item of information was initially dismissed, in the sense
that it was assigned zero probability (P(F) = 0). In this case the information received
is completely surprising or completely unexpected. We call this case belief revision.
In this book we will only address the issue of belief updating.4
It is generally agreed that the rational way to update one’s beliefs is by conditioning
the initial beliefs on the information received, that is, by using the conditional probability
formula (7.1).
3 Typically, conditional probability is defined directly for events as follows: if D, E ⊆ U and P(E) > 0
then P(D|E) = P(D∩E)P(E) where D ∩ E is the intersection of D and E, that is, the set of states that belong to
both D and E. The reader should convince himself/herself that the conjunction of (7.1) and 7.2 is equivalent
to this definition (interpreting P(s|E) as P({s}|E) and recalling that P(∅) = 0).
4 The issue of belief revision is discussed in my book Decision Making (see
http://faculty.econ.ucdavis.edu/faculty/bonanno/).
232 Chapter 7. Adverse Selection or Hidden Type
Definition 7.2.1 We use the expression belief updating or Bayesian updating to refer
to the modification of initial beliefs (expressed by an initial probability distribution
P) obtained by applying the conditional probability rule; this assumes that the belief
change is prompted by the arrival of new information, represented by an event F such
that P(F) > 0.

Thus, when receiving a piece of information F ⊆ U such that P(F) > 0, one would
change one’s initial probability distribution P into a new probability distribution Pnew by
• reducing the probability of every state in U \ F (the complement of F) to zero (this
captures the notion that the information represented by F is trusted to be correct),
and
• setting Pnew (s) = P(s|F) for every state s ∈ F.

Thus, for every state s ∈ U,


 0 if s ∈
/F
Pnew (s) = P(s|F) = (7.3)
 P(s) if s ∈ F
P(F)

(recall the assumption that P(F) > 0). Thus, for every event E ⊆ U, Pnew (E) = ∑ Pnew (s) =
s∈E
∑ P(s|F) = P(E|F).
s∈E

For instance, in the doctor’s example, belief updating requires the following. Recall
that the doctor’s initial beliefs are:
 
state: a b c d e
 .
4 2 1 3 1
probability: 11 11 11 11 11

6
Let + be the event that the blood test is positive (that is, + = {a, b} and thus P(+) = 11 ).
5
Let – be the event that the blood test is negative (that is, – = {c, d, e} and thus P(–) = 11 ).
Then

state a b c d e
4 2 1 3 1
initial beliefs 11 11 11 11 11
4 2
beliefs updated on information + 11
6 = 2
3
11
6 = 1
3 0 0 0
11 11
1 3 1
beliefs updated on information – 0 0 11
5 = 1
5
11
5 = 3
5
11
5 = 1
5
11 11 11
7.2 Conditional probability and belief updating 233

As a further example, suppose that there are only three students in a class: Ann, Bob
and Carla. The professor tells them that in the last exam one of them got 95 points (out of
100), another 78 and the third 54.
We can think of a state as a triple (a, b, c), where a is Ann’s score, b is Bob’s score and c is
Carla’s score. Then, based on the information given by the professor, the set of possible
states is:

U = {(95, 78, 54), (95, 54, 78), (78, 95, 54), (78, 54, 95), (54, 95, 78), (54, 78, 95)} .

Suppose that in all the previous exams Ann and Bob always obtained a higher score than
Carla and often Ann outperformed Bob. Then Ann might consider states (95,78,54) and
(78,95,54) much more likely than (78,54,95) and (54,78,95).
For example, suppose that Ann’s beliefs are as follows:
 
(95, 78, 54) (95, 54, 78) (78, 95, 54) (54, 95, 78) (78, 54, 95) (54, 78, 95)
 .
16 8 4 2 1 1
32 32 32 32 32 32

Suppose that, before distributing the exams, the professor says “I was surprised to see that,
this time, Ann did not get the highest score”. Based on this information, how should Ann
revise her beliefs? The information is that Ann did not receive the highest score, which is
represented by the event

F = {(78, 95, 54), (54, 95, 78), (78, 54, 95), (54, 78, 95)}.

Conditioning on this event yields the following updated beliefs:

 
(95,78,54) (95,54,78) (78,95,54) (54,95,78) (78,54,95) (54,78,95)
 . (7.4)
4 2 1 1
0 0 8 8 8 8

These updated beliefs can be represented more succinctly as follows, by not listing the
states that are ruled out by information F (that is, the states in the complement of F, which
are zero-probability states in the updated beliefs):
 
(78, 95, 54) (54, 95, 78) (78, 54, 95) (54, 78, 95)
 . (7.5)
4 2 1 1
8 8 8 8

The belief updating rule can also be applied sequentially if one first receives information
F (with P(F) > 0) and later receives a further piece of information E (with E ⊆ F and
P(E|F) > 0 or, equivalently, P(E ∩ F) > 0). For instance, in the above example, suppose
that the professor first informs the students that Ann did not get the highest score and later
tells them that Carla received a higher score than Bob. Call F the first piece of information
and E the second piece of information. Then, as we saw above, F is the set of states where
it is in fact true that Ann did not get the highest score:

F = {(78, 95, 54), (54, 95, 78), (78, 54, 95), (54, 78, 95)}.
 234 Chapter 7. Adverse Selection or Hidden Type

On the other hand, E is the set of states where it is in fact true that Carla received a higher
score than Bob:
E = {(95, 54, 78), (78, 54, 95), (54, 78, 95)}.
Ann’s updated beliefs after learning information F are given above in (7.4). Updating
those beliefs by conditioning on E yields
 
(95, 78, 54) (95, 54, 78) (78, 95, 54) (54, 95, 78) (78, 54, 95) (54, 78, 95)
 
1 1
0 0 0 0 2 2

Clearly, this is the same as conditioning the initial beliefs on

E ∩ F = {(78, 54, 95), (54, 78, 95)}.

Expressing all of this more succinctly, we have that the updated beliefs after learning F
are as given in (7.5) above, namely
 
(78, 95, 54) (54, 95, 78) (78, 54, 95) (54, 78, 95)
 
4 2 1 1
8 8 8 8

and the final beliefs are obtained by conditioning these beliefs on E or, equivalently, by
conditioning the initial beliefs on E ∩ F:
 
(78, 54, 95) (54, 78, 95)
 .
1 1
2 2

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 7.4.1 at the end of this chapter.

7.3 The market for used cars

Akerlof’s seminal paper studied the role of adverse selection in the market for used cars
and showed that it can lead to considerable market inefficiencies. We will also take the
market for used cars as our starting point. The model described in this section is a much
simplified version of Akerlof’s model: its purpose is to illustrate the phenomenon as simply
as possible and it is not intended to be realistic.
Suppose that there are two groups of individuals in the population: Group S (‘S’ stands
for potential ‘seller’) and Group B (‘B’ stands for potential ‘buyer’). The number of
individuals in Group B is at least as large as the number of individuals in Group S. Each
Group-S individual owns a car, while no Group-B individual owns a car.
7.3 The market for used cars 235

Cars are of different qualities. Let there be n possible quality levels: q1 , q2 , . . . , qn with
qi denoting a lower quality level than qi+1 , for every i = 1, 2, . . . , n − 1. We express this as
follows:

q1 ≺ q2 ≺ · · · ≺ qn . (7.6)

We assume that each current owner of a car of quality qi (a Group-S individual) considers
owning that car to be just as good as having $Sqi ; thus we can think of Sqi as her reservation
price for the car, in the sense that she would be happy to sell her car at any price P > Sqi .5
We call Sqi the value of a car of quality qi to its initial owner. Given the assumed ranking
of possible qualities (see (7.6)), it is natural to postulate that higher-quality cars have a
higher value to their owners than lower-quality cars, that is,

Sq1 < Sq2 < · · · < Sqn . (7.7)

On the other side of the market, we assume that each individual within Group B considers
becoming the owner of a car of quality qi to be just as good as having $Bqi ; thus we can
think of Bqi as his reservation price for such a car, in the sense that he would be happy to
buy a car of quality qi at any price P < Bqi .6 We call Bqi the value of a car of quality qi
to a Group-B individual. Given the assumed ranking of possible qualities (see (7.6)), it is
natural to postulate that higher-quality cars have a higher value to potential buyers than
lower-quality cars, that is,

Bq1 < Bq2 < · · · < Bqn . (7.8)

Now we introduce two crucial assumptions:

• For every quality level qi (i = 1, 2, . . . , n),

Bqi > Sqi . (7.9)

This means that, for every car, there are potential gains from trade, in the sense
that for a car of quality qi there is a price P such that (1) an individual in Group B
would be willing to pay $P for that car and (2) the owner of such a car (a Group-S
individual) would be happy to sell at price P and (3) the trade would make each
party better off (indeed this would be true for every P such that Sqi < P < Bqi ).
5 For example, the utility-of-money function of the owner of a car of quality qi could be

 m + Sq if she has a car of quality qi
i
UqSi (m) = . Let m be her initial endowment of money; then, since
 m if she has no car
she owns a car of quality qi her current utility is m + Sqi , while if she sells her car for $P then her utility is
m + P. Thus she is better off selling her car if and only if P > Sqi (and she is indifferent if P = Sqi ).
6 For example, his utility-of-money function could be

 m + Bq if he has a car of quality qi
i
UqBi (m) = . Let m be his initial endowment of money; then, since
 m if he has no car
he currently does not own a car, his current utility is m, while if he buys a car of quality qi at price $P then
his utility is m − P + Bqi . Thus he is better off buying such a car if and only if P < Bqi (and he is indifferent
if P = Bqi ).
236 Chapter 7. Adverse Selection or Hidden Type

• Information about quality is asymmetric: each car-owner knows the quality of her
own car (because she has had extensive experience with it and knows how well she
took care of it), while a potential buyer is not able to determine the quality of a car
by mere inspection.7 Of course, a potential buyer can ask the owner what the quality
of her car is, but the owner would have an incentive to claim that it is higher than it
actually is, hoping to sell the car for a higher price.

By (7.9) the initial situation (where cars are currently in the hands of Group-S individuals)
is Pareto inefficient: it would be a Pareto improvement to have each car sold to a Group-
B individual (at an appropriate price). We now show that the postulated asymmetric
information prevents the potential gains from trade from being realized, in the sense that
some cars will remain in the "wrong hands", that is, in the hands of Group-S individuals.

The crucial observation is that, although Group-B individuals have different reservation
prices for different quality levels, there cannot be different prices on the market for cars of
different qualities, because buyers are unable to determine the quality of the cars that are
offered for sale and thus would not know if they should pay a high price or a low price;
furthermore, they realize that the owners of low-quality cars would have an incentive to
claim that their cars are of high quality in order to obtain a higher price and thus any claims
made by the sellers would have to be discarded as not credible. Hence there can only be
one price for all second-hand cars.

Finally, we make two additional assumptions:

• The potential buyers (Group-B individuals) are risk neutral.8

• The potential buyers know, for every quality qi , the fraction of cars of that quality
within the total population of second-hand cars.9 Denote that fraction by pi (thus
0 < pi < 1). Hence, if a buyer were to be able to pick a second-hand car at random,
the probability that he would end up with a car of quality qi is pi .

Let us begin with an example. Let there be six quality levels. For each quality level,
the fraction of cars of that quality and the value of a car of that quality to its current owner
and to a potential buyer is shown in the following table:

quality: q1 q2 q3 q4 q5 q6
1 3 4 1 2 1
proportion: 12 12 12 12 12 12
value to current owner Sqi : $900 $1, 800 $2, 700 $3, 600 $4, 500 $5, 400
value to potential buyer Bqi : $1, 000 $2, 000 $3, 000 $4, 000 $5, 000 $6, 000

7 Some features of a car, such as mileage, exterior damage, etc., can be ascertained by a potential buyer
upon inspection. Other features cannot: for example, whether it was driven with care, whether the car was
affected by a flood or involved in an accident, etc. These are hidden features and it is natural to assume that
they are known to the current owner, but cannot be ascertained by a potential buyer.
8 We make this assumption in order to show that the inefficiencies that arise have nothing to do with the

buyers being risk averse. An example where the buyers are risk averse is given in Exercise 7.10.
9 Without this information, the potential buyers would find themselves in an even worse position.
7.3 The market for used cars 237

A naïve potential buyer would reason as follows (recall the assumption of risk
neutrality):
“For me, picking a car at random corresponds to facing the money lottery
 
$1, 000 $2, 000 $3, 000 $4, 000 $5, 000 $6, 000
  (7.10)
1 3 4 1 2 1
12 12 12 12 12 12

whose expected value is

1 3 4 1 2 1
12 1000 + 12 2000 + 12 3000 + 12 4000 + 12 5000 + 12 6000 = 3, 250.

Thus, as long as the price P of a second-hand car is less than $3,250, I expect
to make a positive net gain by buying a car.”
Suppose that, in fact, the price of a second-hand car is $3,000. Should the potential
buyer decide to buy a car that is offered for sale at that price, expecting a gain of
$(3, 250 − 3, 000) = $250? The answer is negative, because when the price is $3,000
not every car will be offered for sale: the owners of cars of qualities q4 , q5 and q6 will not
be willing to sell.10 Thus only cars of qualities q1 , q2 and q3 will be offered for sale when
the price is $3,000, so that buying a car at that price means facing the following lottery,
where the probabilities are obtained from (7.10) by conditioning on the event {q1 , q2 , q3 }:
 
$1, 000 $2, 000 $3, 000
  (7.11)
1 3 4
8 8 8

whose expected value is

1 3 4
8 1000 + 8 2000 + 8 3000 = 2, 375.

Thus the buyer would be paying $3,000 expecting a benefit of only $2,375! That is, buying
a car for $3,000 means facing an expected loss of $625.
Might paying a price somewhat below $2,375 be a good idea? For example, should the
buyer be willing to pay $2,000 for a second-hand car? The answer is again negative: when
the price is $2,000 the owners of cars of quality q3 will drop out of the market (because
they value their cars at $2,700: more than the price) and thus buying a car for $2,000
means facing the lottery11
 
$1, 000 $2, 000
  (7.12)
1 3
4 4

whose expected value is

1 3
4 1000 + 4 2000 = 1, 750.
10 Since, for i = 3, 4, 5, Sqi > 3, 000. That is, the owners have a reservation price that exceeds the market
price and thus selling at that price would make them worse off.
11 The probabilities are obtained from (7.10) by conditioning on the event {q , q } or, equivalently, from
1 2
(7.12) by conditioning on the event {q1 , q2 }.
238 Chapter 7. Adverse Selection or Hidden Type

Thus the buyer would be paying $2,000 expecting a benefit of only $1,750! That is, buying
a car for $2,000 means facing an expected loss of $250.
It is only when the price P is in the range between $900 and $1,000, that buying a
car at that price is rational for the buyer: when 900 < P < 1, 000 then only cars of the
lowest quality, namely q1 , are offered for sale and the buyer knows this (such cars are
referred to as “lemons”; hence the expression ‘the market for lemons’ used by Akerlof in
his seminal paper). Thus in this example we have an extreme inefficiency: of all the used
cars in existence, only the lowest-quality cars are being traded in the market, despite the
fact that for every car that is not traded there is a potential buyer who would be willing to
pay a price that exceeds the reservation price of the owner of that car.

We can now define what an equilibrium price in the market for used cars is in the
presence of adverse selection.

Definition 7.3.1 A price P for used cars is an equilibrium price if there is an

m ∈ {1, 2, . . . , n} such that
1. Sqm < P < Sqm+1 .

2. E[L] > P, where L is the money lottery

 
 $Bq1 . . . $Bqm 
L=  mp1 . . . mpm
.
 (7.13)
∑ pi ∑ pi
i=1 i=1

The first condition in Definition 7.3.1 says that at price P all the qualities q1 , . . . , qm are
offered for sale and none of the qualities above qm are offered for sale.12 Thus, conditioning
on the fact that the qualities offered for sale are precisely {q1 , . . . , qm } the buyer realizes
that buying a used car at price P means facing the following money lottery in terms of
potential net gain:

 
 $(Bq1 − P) . . . $(Bqm − P) 
 p1 pm
 (7.14)
 m ... m

∑ pi ∑ pi
i=1 i=1

whose expected value is E[L] − P where L is given by (7.13). Thus, condition 2 of

Definition 7.3.1 says that every potential buyer expects to gain from purchasing a car that
is offered for sale.

12 Note that, by (7.7), (1) Sqm < P implies that, for every i = 1, . . . , m − 1, Sqi < P and
(2) Sqm+1 > P implies that, for every i = m + 2, . . . , n, Sqi > P.
7.3 The market for used cars 239

The example analyzed above illustrates the logic of the adverse selection phenomenon,
which is as follows. In order for all the used cars to be put up for sale, the price P must
exceed the value of the top-quality car for the owner of such a car, that is, it must be that
P > Sqn . When all the used cars are offered for sale, a potential buyer faces the following
lottery in terms of potential net gain:
 
$(Bq1 − P) . . . $(Bqn − P)
  (7.15)
p1 ... pn


whose expected value is p1 Bq1 + · · · + pn Bqn − P. Note that


p1 Bq1 + · · · + pn Bqn < Bqn

(since it is a convex combination of all the values Bqi , of which Bqn is the largest).
Our hypothesis is
that Bqn > Sqn and thus it is, in principle, possible that
p1 Bq1 + · · · + pm Bqn > Sqn , in which case one can find a price P such that


Sqn < P < p1 Bq1 + · · · + pn Bqn .

If that is the case, then at such a price all the cars would be traded and thus Pareto efficiency
would be achieved. In the above example, on the other hand, we had that


p1 Bq1 + · · · + pn Bqn < Sqn (7.16)

and thus at any price P that would induce all the car-owners to offer their cars for sale
(that is, at any P such that P > Sqm ) no potential buyer would be willing to buy, so that the
highest-quality cars would not be traded, leading to a Pareto inefficient situation.
Then the reasoning can repeated for the case of all qualities but the top one being
offered for sale. This would require a price P such that Sqn−1 < P < Sqn . Then the lottery
faced by a potential buyer contemplating buying a used car offered for sale would be
 
$(Bq1 − P) . . . $(Bqn−1 − P)
  (7.17)
p1 ... pn−1


whose expected value is p1 Bq1 +
· · · + p n−1 Bq n−1 − P. Again, the important observation
is that p1 Bq1 + · · · + pn−1 Bqn−1 < Bqn−1 . Given our
assumption that Bqn−1 > Sqn−1 it is,
in principle, possible that p1 Bq1 + · · · + pn−1 Bqn−1 > S
qn−1 , in which case one can find
a price P such that Sqn−1 < P < p1 Bq1 + · · · + pn−1 Bqn−1 , in which case all the qualities
except for the top one would be traded.
However, it is also possible - as in our example
above - that p1 Bq1 + · · · + pn−1 Bqn−1 < Sqn−1 , in which case there cannot be a price at
which all the qualities up to qn−1 are traded. This argument can be repeated for every
quality level. In the extreme case - as in our example above - the only possible equilibrium
is one where only the lowest-quality cars are traded. Note that, given our assumptions
(namely (7.9), in particular Sq1 < Bq1 ), an equilibrium whereonly the lowest-quality cars
are traded always exists: any price P such that Sq1 < P < min Bq1 , Sq2 would give rise to
such an equilibrium.
240 Chapter 7. Adverse Selection or Hidden Type

Depending on the difference between the seller’s value and the buyer’s value of each
quality many possibilities arise:
- If the difference is small - as in our example above - then it may be that the only
possible equilibrium is one where only the lowest quality cars are traded. Thus this
is a situation where the bad-quality cars drive the good-quality cars out of the market
and the lack of trading for good-quality cars gives rise to a Pareto inefficiency.
- If the difference is somewhat large then there may be several possible equilibria
that can be “Pareto ranked” in terms of efficiency (this is illustrated in the example
below).
- If the difference is substantially large then among the possible equilibria there is also
the most efficient one where all the cars are traded.
The following is an example of the second possibility listed above. Let there be three qual-
ities: L (for ‘low’), M (for ‘medium’) and H (for ‘high’). The corresponding proportions
and valuations are as shown in the following table:

quality: L M H
1 3 1
fraction: 5 5 5
value to current owner: $10, 000 $14, 000 $16, 000
value to potential buyer: $12, 000 $15, 000 $18, 000

Let us see what values of P (the market price of a used car) give rise to an equilibrium.13
1. If P > 16, 000 then all the cars will be offered for sale. A potential buyer then faces
the following net-gain lottery if he buys a car:
 
$(12, 000 − P) $(15, 000 − P) $(18, 000 − P)
 
1 3 1
5 5 5

whose expected value is 15, 000 − P < 0. Thus there is no equilibrium at any such
price.
2. If 14, 000 < P ≤ 16, 000 then only cars of qualities L and M are offered for sale. A
potential buyer then faces the following net-gain lottery if he buys a car:
 
$(12, 000 − P) $(15, 000 − P)
 
1 3
4 4

whose expected value is 14, 250 − P. This is positive if and only if P < 14, 250. Thus
- If 14, 250 ≤ P ≤ 16, 000 there is no equilibrium at such price.
- If 14, 000 < P < 14, 250 then, for every such price there is an equilibrium where
all the L-quality and all the M-quality cars are traded, while the H-quality cars
are not.
13 We assume that, if indifferent between offering her car for sale and not offering it, the owner chooses
not to offer it and, if indifferent between buying and not buying, the potential buyer chooses not to buy.
 7.3 The market for used cars 241

3. If 10, 000 < P ≤ 14, 000 then only the L-cars are offered for sale, so that a potential
buyer realizes that he would buy a car worth $12,000 to him. Thus
- If 12, 000 ≤ P ≤ 14, 000 there is no equilibrium at such price.
- If 10, 000 < P < 12, 000 then, for every such price there is an equilibrium
where all and only the L-quality cars are traded.
4. If P ≤ 10, 000 then no cars are offered for sale and thus there is no trading.
Clearly, the equilibria under Point 2 are Pareto superior to the equilibria under Point 3,
since more cars are traded. None of them, however, is Pareto efficient, since the H-quality
cars are not traded and remain “in the wrong hands”.

7.3.1 Possible remedies

The root cause of the market inefficiency highlighted in the previous section is the asymmet-
ric information between the seller (who is informed) and the buyer (who lacks information).
Are there any solutions to this problem?
The owner of a good-quality car finds herself unable to sell her car because of the
presence of low-quality cars in the market. Claiming (truthfully) that her car deserves a
higher price because it is is of high quality is pointless, because every seller would want to
imitate that claim (if seen to be successful). So any claims made by the seller have to be
credible, not just “cheap talk”. Is there a way to make the claim credible? The answer is
affirmative and involves the notion of signalling, which will be covered in Chapter 9. Thus
we postpone this discussion to Chapter 9, where we will consider warranties as signals of
high quality.
While a seller’s claims should be discarded as not credible, perhaps a knowledgeable
third party, such as a mechanic, could be consulted to ascertain the quality of a car offered
for sale. This solution, however, is problematic for a number of reasons. First of all, there
is a moral hazard problem (the notion of moral hazard will be discussed in Chapter 10):
would the mechanic have an incentive to do due diligence, that is, to thoroughly inspect the
car? If the mechanic is paid a fixed fee then he might decide to save time and perform only
a cursory inspection. Secondly, who would pay the mechanic’s fee? If the fee is paid by the
potential buyer, then sellers will not face any costs in claiming that their cars are of high
quality and will thus have no disincentive to lie. On the other hand, if the fee is paid by the
seller, then it could play the role of a signal (see Chapter 9) and be effective in separating
high-quality cars from lower-quality ones. However, the buyer might be suspicious of a
high-quality certification provided by a mechanic chosen by the seller, since there might
be collusion between the two. On the other hand, the seller might be reluctant to let the
buyer choose the mechanic, because the latter might be in cahoots with the former (the
mechanic might certify the car to be of low quality, thus giving an excuse to the buyer not
to buy, and then share the fee with the buyer).

7.3.2 Further remarks

One might wonder whether the phenomenon pointed out in this section, namely that
“low-quality cars drive high-quality cars out of the market” applies also to new cars. The
answer is negative: a dealership does not acquire, by use, knowledge of the quality of any
given car (and thus cannot keep the better cars supplied by the manufacturers for itself and
242 Chapter 7. Adverse Selection or Hidden Type

offer for sale the lower-quality ones); in other words, in the market for new cars there is
no asymmetric information between buyer and seller. When buying a new car a buyer is
just as likely to obtain an above-average-quality car as a below-average-quality car; hence
new cars sell for a market price that reflects the average quality of all new cars: there is no
adverse selection in the market for new cars.

One could object to the analysis of this section by pointing out that there are observable
characteristics of used cars, such as mileage and age, that are easily verified by the buyer
and can be the basis for price differences: typically, a 2014 Toyota Corolla will sell for a
lower price than a 2018 Toyota Corolla, or a 2014 Toyota Corolla with 30,000 miles will
sell for a higher price than a 2014 Toyota Corolla with 85,000 miles. The answer to this
objection is that, indeed, observable differences can be associated with different prices, but
within each category of car, e.g. a 2014 Toyota Corolla with approximately 80,000 miles,
there typically are better cars and worse cars and information about the car’s quality is
asymmetric: the seller has that information, the buyer does not. Thus the “lemons problem”
arises within each well-defined category of used car.

The adverse selection problem arises in many different markets. In this chapter we
considered the market for used car, or - more generally - the market for durable goods.
In the next chapter we will analyze adverse selection in the market for insurance. Stglitz
and Weiss14 suggested that adverse selection could be the cause of credit rationing in the
market for unsecured loans, that is, loans for which the borrower cannot offer a collateral.
Consider, for example, a bank that offers loans not knowing the credit worthiness of the
borrowers: some borrowers are “good quality” in that they are low risk, either because
they only borrow if they expect to be able to repay the loan and/or they plan to invest in a
low-return but reasonably safe project, while other borrowers are “low quality” in that they
are high risk, either because they borrow without worrying about defaulting and/or they
plan to invest in high-return but high-risk projects. The low-risk types are more sensitive
to increases in the rate of interest and drop out of the market when the rate of interest is
high, while the high-risk types are less sensitive to increases in the rate of interest and
continue to borrow even if the rate is high. In this case lenders might prefer to keep the rate
of interest low, so as to have a better mix of “good” and “bad” types, even if at that low
rate there is excess demand for loans. That is, if the bank faces excess demand for loans, it
might prefer not to raise its interest rate to clear the market, because by doing so it would
drive away some low-risk borrowers and increase the proportion of high-risk borrowers in
the pool of applicants, thus facing a higher probability of defaulting borrowers.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 7.4.2 at the end of this chapter.

14 Stiglitz,
Joseph and Andrew Weiss, “Credit rationing in markets with imperfect information”, The
American Economic Review, 1981, Vol. 71, pp. 393-410.
 7.4 Exercises 243

7.4 Exercises
The solutions to the following exercises are given in Section 7.5 at the end of this chapter.

7.4.1 Exercises for Section 7.2.2: Conditional probability and belief updating

Exercise 7.1 Consider the following probability distribution:

 
z1 z2 z3 z4 z5 z6 z7
 .
3 1 3 2 3
12 12 0 12 12 0 12

What is the probability of the event {z2 , z3 , z5 , z6 , z7 }? 

Exercise 7.2 Let the universal set be U = {z1 , z2 , z3 , z4 , z5 , z6 , z7 , z8 }.

Let A = {z2 , z4 , z5 , z7 } , B = {z3 , z6 , z8 } , C = {z2 , z6 } , D = {z3 , z4 } and E = {z7 , z8 }.
21 5 3
You are given the following data: P(A ∪ B) = 24 , P(A ∩ C) = 24 , P(B ∩ C) = 24 ,
2 3 7 2
P(A ∩ D) = 24 , P(B ∩ D) = 24 , P(B) = 24 and P(E) = 24 .
(a) Find the probability P(zi ) for each i = 1, ..., 8.
(b) Calculate P ((A ∪ B) ∩ (C ∪ D)).


Exercise 7.3 Let U = {a, b, c, d, e, f , g, h, i} and consider the following probability

 
 a b c d e f g h i 
distribution:  .
 
 
11 7 9 16 5 4 8
60 0 60 60 60 60 60 60 0

(a) Let E = {a, f , g, h, i}. What is the probability of E?

(b) List all the events that have probability 1.


Exercise 7.4 There is an urn with 40 balls: 4 red, 16 white, 10 blue and 10 black. You
close your eyes and pick a ball at random. Let E be the event “the selected ball is either
red or white”.
(a) What is the probability of E?
(b) Now somebody tells you: “the ball in your hand is not black”. How likely is it
now that you picked either a red or a white ball?

244 Chapter 7. Adverse Selection or Hidden Type

Exercise 7.5 Suppose there are 3 individuals. It is known that one of them has a virus.
A blood test can be performed to test for the virus. If an individual does have the virus,
then the result of the test will be positive. However, the test will be positive also for an
individual who does not have the virus but has a particular defective gene.
It is known that exactly one of the three individuals has this defective gene: it could be
the same person who has the virus or somebody who does not have the virus. A positive
test result will come up if and only if either the patient has the virus or the defective
gene (or both).
Suppose that Individual 1 takes the blood test and the result is positive. Assuming that
all the states are equally likely, what is the probability that he has the virus? [Hint: think
of the universal set (or sample space) U as a list of states and each state tells you which
individual has the virus and which individual has the defective gene.] 

Exercise 7.6 In a remote rural clinic with limited resources, a patient arrives complain-
ing of low-abdomen pain. Based on all the information available, the doctor thinks that
there are only four possible causes: a bacterial infection (b), a viral infection (v), cancer
(c), internal bleeding (i). Of the four, only the bacterial infection and internal bleeding
are treatable at the clinic. In the past the doctor has seen 600 similar cases and they
eventually turned out to be as follows:

b : bacterial infection v : viral infection c : cancer i : internal bleeding

140 110 90 260

The doctor’s probabilistic estimates are based on those past cases.

(a) What is the probability that the patient has a treatable disease?
There are two possible ways of gathering more information: a blood test and an
ultrasound. A positive blood test will reveal that there is an infection, however it could
be either bacterial or viral; a negative blood test rules out an infection and thus leaves
cancer and internal bleeding as the only possibilities. The ultrasound, on the other hand,
will reveal if there is internal bleeding.
(b) Suppose that the patient gets an ultrasound and it turns out that there is no internal
bleeding. What is the probability that he does not have a treatable disease? What
is the probability that he has cancer?
(c) If instead of getting the ultrasound he had taken the blood test and it had been
positive, what would the probability that he had a treatable disease have been?
(d) Now let us go back to the hypothesis that the patient only gets the ultrasound and
it turns out that there is no internal bleeding. He then asks the doctor: “if I were
to take the blood test too (that is, in addition to the ultrasound), how likely is it
that it would be positive?”. What should the doctor’s answer be?
(e) Finally, suppose that the patient gets both the ultrasound and the blood test and
the ultrasound reveals that there is no internal bleeding, while the blood test is
positive. How likely is it that he has a treatable disease?

 7.4 Exercises 245

7.4.2 Exercises for Section 7.3: The market for used cars

Exercise 7.7 There are 200 cars in total. Cars are of different quality. Quality is
measured in terms of the expected number of years of residual life of the car. Make use
of the following information:

Quality: 4 8 10 16
1 3 2 2
Fraction of total number: 8 8 8 8
Value of car to current owner: $2, 000 $3, 000 $4, 000 $5, 000

Fill in the following table:

If the price is: Number of cars Average quality of cars

offered for sale offered for sale
$2, 500
$3, 100
$4, 600
$5, 225
$6, 100


Exercise 7.8 There are two groups of individuals. Group 1 individuals own cars, while
Group 2 do not. There are four possible quality levels of cars as shown in the following
table, together with the total number of cars of each quality.

Quality: A B C D
Number of cars: 1, 000 2, 000 1, 000 4, 000

The value that a Group-1 individual attaches to a car of a given quality is lower than the
value of the same quality car to a Group-2 individual, as shown in the following table:

Quality: A B C D
Value to Group 1 individuals: $5, 400 $4, 500 $3, 600 $2, 700
Value to Group 2 individuals: $6, 000 $5, 000 $4, 000 $3, 000

The quality of a car is known to the owner but not to the prospective buyer. All
individuals are risk-neutral.
(a) Write the lottery for a Group-2 individual that corresponds to picking a car at
random from the entire pool of cars and calculate the expected value.
(b) Suppose that the price of a second-hand car is $3,800. Should a Group-2 individ-
ual be willing to buy a car at that price? Explain your answer.
(c) Let P be the price at which second-hand cars are traded. What are the possible
values of P such that, at that price, 4,000 cars are traded?

246 Chapter 7. Adverse Selection or Hidden Type

Exercise 7.9 Making use of the information in Exercise 7.8, for every positive price P
determine if there is trading of second-hand cars at that price. Assume that, if indifferent
between selling and not selling, a Group-1 individual would decide to sell and, if
indifferent between buying and not buying, a Group-2 individual would decide to buy. 

Exercise 7.10 Let the quality of a second-hand motorcycle be denoted by θ ∈ {1, 2, 3},
where θ is the number of comprehensive tune-ups that the motorcycle received in the
past. The owner knows the value of θ but a potential buyer does not. The value of
a motorcycle of quality θ to the seller is $800θ . Each potential buyer has an initial
wealth
√ of $9,025 and the utility of purchasing a motorcycle √ of quality θ at price P
is 9, 025 − P + 1, 000θ , while the utility of not buying is 9, 025 = 95. Let the
proportion of motorcycles of each quality be as follows (where q is a number strictly
between 0 and 31 ):
 
θ: 1 2 3
 .
2 1
proportion: q 3 −q 3

Suppose that the price of a second-hand motorcycle is P = $1, 744. [In the following
assume that, if indifferent between selling and not selling, the owner of a motorcycle
would sell and, if indifferent between buying and not buying, a potential buyer would
buy.]
(a) Are there values of q such that ALL motorcycles are traded?
(b) Are there values of q such that all motorcycles of quality θ = 1 and θ = 2 are
traded?
(c) Are there values of q such that only motorcycles of quality θ =1 are traded?


Exercise 7.11 Suppose that you have just opened an all-you-can-eat buffet restaurant.
The capacity of your restaurant is 50, that is, you can accommodate at most 50 customers.
Suppose it costs $2 to provide one serving of food (e.g. a standard-size plate). Assume
that you are risk neutral. Each potential customer can be described by a pair (r, c)
where r is the reservation price (the customer will come to your restaurant if and only
if the admission price p is less than or equal to r) and c is the number of servings of
food he/she consumes. There is a total of 240 potential customers. Their types and
corresponding numbers are given in the following table:

Customer type: ($6, 1) ($6, 1.5) ($6.50, 1.5) ($6.50, 2.2) ($7, 2.2) ($7, 3.5)
Number: 20 20 40 60 40 60

What price should you charge if you want to maximize your profits? 
 7.5 Solutions to Exercises 247

7.5 Solutions to Exercises

Solution to Exercise 7.1.
1 2 3 1
P ({z2 , z3 , z5 , z6 , z7 }) = ∑ P(zi ) = +0+ +0+ = .
i∈{2,3,5,6,7}
12 12 12 2


Solution to Exercise 7.2.

(a) Since {z1 } is the complement of A ∪ B, P(z1 ) = 1 − 21
24 =
3
24 .
5
Since {z2 } = A ∩C, P(z2 ) = 24 .
3 3 2
Similarly, P(z6 ) = P(B∩C) = 24 , P(z3 ) = P(B∩D) = 24 and P(z4 ) = P(A∩D) = 24 .
7 3 3 1
Thus, P(z8 ) = P(B) − P(z3 ) − P(z6 ) = 24 − 24 − 24 = 24 .
2 1 1
Hence, P(z7 ) = P(E) − P(z8 ) = 24 − 24 = 24 .
6
Finally, P(z5 ) = 1 − ∑ P(zi ) = 24 . Thus, the probability distribution is:
i6=5

 
 z1 z2 z3 z4 z5 z6 z7 z8 
 
 
 
3 5 3 2 6 3 1 1
24 24 24 24 24 24 24 24

(b) A ∪ B = {z2 , z3 , z4 , z5 , z6 , z7 , z8 } , C ∪ D = {z2 , z3 , z4 , z6 }.

Hence, (A ∪ B) ∩ (C ∪ D) = C ∪ D = {z2 , z3 , z4 , z6 }
5 3 2 3 13
so P ((A ∪ B) ∩ (C ∪ D)) = P(z2 ) + P(z3 ) + P(z4 ) + P(z6 ) = 24 + 24 + 24 + 24 = 24 .


Solution to Exercise 7.3.

The probability distribution is:

 
 a b c d e f g h i 
.
 

 
11 7 9 16 5 4 8
60 0 60 60 60 60 60 60 0

(a) Let E = {a, f , g, h, i}.

11 5 4 8 28 7
Then P(E) = P(a) + P( f ) + P(g) + P(h) + P(i) = 60 + 60 + 60 + 60 +0 = 60 = 15 .
(b) The probability-1 events are: {a, c, d, e, f , g, h} = U \ {b, i}, {a, b, c, d, e, f , g, h} =
U \ {i}, {a, c, d, e, f , g, h, i} = U \ {b} and {a, b, c, d, e, f , g, h, i} = U. 
248 Chapter 7. Adverse Selection or Hidden Type

Solution to Exercise 7.4.

(a) P(E) = 4+16 1
40 = 2 .
30
(b) Let F be the event “the selected ball is not black”. Then, initially, P(F) = 40 = 34 .
1
P(E∩F) P(E)
Furthermore, E ∩ F = E. Thus, P(E|F) = P(F) = P(F) = 2
3 = 23 . 
4

Solution to Exercise 7.5.

First we list the possible states. A state is a complete description of the external facts that
are relevant: it tells you who has the virus and who has the gene. Let us represent a state
as a pair (x, y) interpreted as follows: individual x has the virus and individual y has the
defective gene.
Then U = {a = (1, 1), b = (1, 2), c = (1, 3), d = (2, 1), e = (2, 2),
f = (2, 3), g = (3, 1), h = (3, 2), i = (3, 3)}.
Let V1 be the event “Individual 1 has the virus”. Then V1 = {a, b, c}.
Let G1 be the event “Individual 1 has the defective gene”. Then G1 = {a, d, g}.
Since every state is assumed to have probability 19 , P(V1 ) = P(G1 ) = 19 + 19 + 19 = 13 . Let
1+ be the event that a blood test administered to Individual 1 comes up positive. Then
1+ = {a, b, c, d, g} and P(1+ ) = 59 .
We can now compute the requested conditional probability (note that V1 ∩ 1+ = V1 ):

1
P(V1 ∩ 1+ ) P(V1 ) 3 3
P(V1 |1+ ) = = = 5
= 5 = 60%.
P(1+ ) P(1+ ) 9


Solution to Exercise 7.6. The probabilities are as follows:
b v c i
140 14 110 11 90 9 260 26
600 = 60 600 = 60 600 = 60 600 = 60

(a) The event that the patient has a treatable disease is {b, i}.
14 26
P({b, i}) = P(b) + P(i) = 60 + 60 = 23 .
(b) A negative result of the ultrasound is represented by the event {b, v, c}.
A non-treatable disease is the event {v, c}. Thus,
11 9
P ({v, c} ∩ {b, v, c}) P ({v, c}) 60 + 60
P ({v, c}|{b, v, c}) = = = 14 11 9
= 10
17 = 58.82%.
P ({b, v, c}) P ({b, v, c}) 60 + 60 + 60
9
P (c) 60 9
P (c|{b, v, c}) = = 14 11 9
= 34 = 26.47%.
P ({b, v, c}) 60 + 60 + 60
(c) A positive blood test is represented by the event {b, v}. A treatable disease is the
event {b, i}. Thus,
14
P ({b, i} ∩ {b, v}) P (b) 60 14
P ({b, i}|{b, v}) = = = 14 11
= 25 = 56%.
P ({b, v}) P ({b, v}) 60 + 60
7.5 Solutions to Exercises 249

(d) Here we want

14
P ({b, v}) 60 + 11
60 25
P ({b, v}|{b, v, c}) = = 14
= 34 = 73.53%.
P ({b, v, c}) 60 + 11 9
60 + 60

(e) We are conditioning on {b, v} ∩ {b, v, c} = {b, v}; thus, we want P ({b, i}|{b, v})
which was calculated in Part (c) as 14
25 = 56%. 

Solution to Exercise 7.7.

If the price is: Number of cars Average quality of cars

offered for sale offered for sale
1
$2, 500 8 200 = 25 4
4 1 3
$3, 100 8 200 = 100 44 + 48 = 7
6 1 3 2
$4, 600 8 200 = 150 6 4 + 6 8 + 6 10 = 8
1 3 2 2
$5, 225 200 8 4 + 8 8 + 8 10 + 8 16 = 10
1 3 2 2
$6, 100 200 8 4 + 8 8 + 8 10 + 8 16 = 10

Solution to Exercise 7.8.

 
$6, 000 $5, 000 $4, 000 $3, 000
(a) The lottery is 1 2 1 4 . The expected value is
8 8 8 8
4, 000.
(b) No, because if the price is $3,800, only owners of cars of quality C and D will offer
 
$4, 000 $3, 000
their cars for sale and thus buying a car means facing the lottery 1 4 ,
5 5
whose expected value is $3,200, less than the price of $3,800.
(c) Any price P such that $2, 700 < P < $3, 000 would lead to all and only cars of
quality D being traded and thus 4,000 cars. 

Solution to Exercise 7.9.

• If P ≥ 5, 400 then every car will be offered for sale, but then - as calculated in Part
(a) of Exercise 7.8 - a potential buyer would be facing a lottery whose expected value
is less than the price (namely $4,000) and thus would not be willing to buy. Thus if
P ≥ 5, 400 there is no trading.
• If 4, 500 ≤ P < 5, 400 then only cars of qualities B, 
C and D will be offered for sale.

$5, 000 $4, 000 $3, 000
Thus a potential buyer would be facing the lottery 2 1 4 ,
7 7 7
whose expected value is $3,714.29, less than the price, and thus would not be willing
to buy. Hence if 4, 500 ≤ P < 5, 400 there is no trading.
250 Chapter 7. Adverse Selection or Hidden Type

• If 3, 600 ≤ P < 4, 500 then only cars of qualities C and

 D will be offered for sale.
$4, 000 $3, 000
Thus a potential buyer would be facing the lottery 1 4 , whose
5 5
expected value is $3,200, less than the price. Thus if 3, 600 ≤ P < 4, 500 there is no
trading.

• If 2, 700 ≤ P < 3, 600 then only cars of qualities D will be offered for sale and thus
a potential buyer realizes that she would be buying a car worth $3,000 to her. Hence
– If 3, 000 < P < 3, 600 then potential buyers are not willing to buy and thus
there is no trading.
– If 2, 700 ≤ P ≤ 3, 000 then all (and only) the cars of quality D are traded.

• If P < 2, 700 no cars will be offered for sale and thus there is no trading. 

Solution to Exercise 7.10.

(a) Since the owner of a motorcycle of quality θ = 3 values the motorcycle at $2,400,
when P = $1, 744 she will not be willing to sell. Hence the answer is No.
(b) The owner of a motorcycle of quality θ = 1 values it at $800 and the owner of a
motorcycle of quality θ = 2 values it at $1, 600; hence both qualities will be offered
for sale when the price is $1,744 (but not those of quality θ = 3). Thus, in terms of
quality, the buyer faces the following lottery:15
 
θ =1 θ =2
3
2q 1 − 32 q

which corresponds to the money lottery

 
$(9, 025 − 1, 744 + 1, 000) $(9, 025 − 1, 744 + 2, 000)
3
2q 1 − 32 q

whose expected utility is:

3
p
p
2q 8, 281 + 1 − 23 q 9, 281 = 136.5q + 96.34 − 144.51q = 96.34 − 8.01q.
√
The buyer will be willing to buy if 96.34 − 8.01q ≥ 9025 = 95, that is, if
q ≤ 0.1965. So the answer is: Yes all the values of q ≤ 0.1673.

(c) No. If q ≤ 0.1673 then both qualities θ = 1 and θ = 2 are traded and if q > 0.1673
then both qualities θ = 1 and θ = 2 are offered for sale, but buyers are not willing
to buy.

15 By q 3
the conditional probability rule, P(θ = 1|{θ = 1, θ = 2}) = 2 = 2q and
q+ 3 −q
2
P(θ = 2|{θ = 1, θ = 2}) = 3 −q = 1 − 23 q.
2
q+ 3 −q
7.5 Solutions to Exercises 251

Solution to Exercise 7.11. First of all, let us convert numbers into proportions:

Customer type: ($6, 1) ($6, 1.5) ($6.50, 1.5) ($6.50, 2.2) ($7, 2.2) ($7, 3.5)
1 1 2 3 2 3
Proportion: 12 12 12 12 12 12

Secondly, note that it cannot be profit maximizing to charge a price different from either
$6 or $6.50 or $7.16
• If you charge $6 then all the 240 people are interested in eating at your restaurant.
You will only be able to accommodate 50 of them. Assuming that the order of arrival
is random, the probability that a served customer is of any given type coincides
with the fraction of that type in the population. Thus letting in any one customer
corresponds to playing the following “consumption”’ lottery:

consumption: 1 1.5 2.2 3.5

1 1 2 3 3 2 5 3
probability: 12 12 + 12 = 12 12 + 12 = 12 12

1 3 5 3
whose expected value is 12 1 + 12 (1.5) + 12 (2.2) + 12 (3.5) = 2.25. Hence the ex-
pected cost of a single customer is $(2 × 2.25) = $4.50 so that the expected profit
from a single customer is $(6 − 4.50) = $1.50. Hence your total expected profit is
$(1.50 × 50) = $75 .

• If you charge $6.50 then only 10 12 of the potential customers are interested in eating
at your restaurant, for a total of 10
12 240 = 200: still more than you can accommodate.
Assuming, again, that the order of arrival is random, the probability that a served
customer is of any given type coincides with the fraction of that type in the population
of interested customers. Applying the conditional probability rule, their types and
(updated) proportions are:

Customer type: (6.50, 1.5) (6.50, 2.2) (7, 2.2) (7, 3.5)
2 3 2 3
Fraction: 10 10 10 10

Thus, letting in any one customer corresponds to playing the following consumption
lottery:
consumption: 1.5 2.2 3.5
2 3 2 5 3
probability 10 10 + 10 = 10 10

whose expected value is 2.45. Hence the expected cost of a single customer is
$(2 × 2.45) = $4.90 so that the expected profit from a single customer is $(6.50 −
4.90) = $1.60. Hence your total expected profit is $(1.60 × 50) = $80 .

16 For example, if you charge $5.50 then your revenue will be $(5.50 × 50) = $275 and if you increase
your price to $6 then it will still be the case that at least 50 customers want to enter your restaurant and your
revenue will increase to $(6 × 50) = $300.
252 Chapter 7. Adverse Selection or Hidden Type
5
• If you charge $7 then only 12 of the potential customers are interested in eating at
5
your restaurant, for a total of 12 240 = 100: still more than you can accommodate.
Applying the conditional probability rule, their types and (updated) proportions are:

Customer type: (7, 2.2) (7, 3.5)

2 3
Fraction: 5 5

Thus, letting in any one customer corresponds to playing the following consumption
lottery:
consumption: 2.2 3.5
2 3
probability: 5 5
whose expected value is 2.98. Hence the expected cost of a single customer is $(2 ×
2.98) = $5.96 so that the expected profit from a single customer is
$(7 − 5.96) = $1.04. Hence your total expected profit is $(1.04 × 50) = $52 .

Thus the profit-maximizing price is $6.50. 

 8. Adverse Selection in Insurance

8.1 Adverse selection in insurance markets

As we noted in the previous chapter, the phenomenon of adverse selection arises naturally
in the context of insurance: the insurance company (the seller of insurance) is typically
aware of the fact that there are individuals who – because of their family history – are at a
higher risk of developing a condition that requires extensive medical services, while other
individuals represent a lower risk. If the insurance company offers a contract that would,
on average, cover its expected costs if everybody (high-risk and low-risk individuals) were
to purchase that contract, it might discover that its costs are much higher than expected,
because only the high-risk individuals ended up purchasing the contract.

In Chapter 5 we considered the case of only one type of individual, with a constant
probability of loss. In Chapter 10 we will consider the case where the probability of loss
is affected by the individual’s behavior: a situation referred to as "moral hazard". In this
chapter we will continue to assume that the probability of loss is constant, that is, not
affected by the individual’s behavior, but we will consider the case where there are different
types of individuals, each type with a different probability of loss. In the case of two types,
one type has a higher probability of loss and will be called the high-risk type and the other
will be called the low-risk type. There is uncertainty on the part of the insurance company
due to do the insurance company’s inability to tell high-risk from low-risk individuals
apart (while being fully aware that it faces two different types of potential customers).
 254 Chapter 8. Adverse Selection in Insurance

8.2 Two types of customers

Suppose that there are two types of individuals. They are all identical in terms of initial
wealth, denoted by W0 , and in terms of the potential loss that they face, denoted by ` (with
0 < ` ≤ W0 ). They also have the same vNM utility-of-money function U(m). What they
differ in is the probability of loss: it is pH for type-H (= high-risk) individuals and pL for
type-L (= low-risk) individuals, with

0 < pL < pH < 1.

It follows that type-H individuals have steeper indifference curves than type-L individuals,
as shown in Figure 8.1. In fact, fix an arbitrary point C = (W1C ,W2C ) in the wealth plane.
The slope of the indifference curve going through this point is

pH U 0 (W1C )
− for type-H individuals
1 − pH U 0 (W2C )

and
pL U 0 (W1C )
− for type-L individuals.
1 − pL U 0 (W2C )

Since pL < pH ,
pL pH
< .
1 − pL 1 − pH

Wealth in
good state
W2

C
W2C

indifference curve
for L-type
indifference curve
for H-type

Wealth in
W1 bad state
0 W1C

Figure 8.1: The indifference curve through point C for type-H is steeper than the indiffer-
ence curve through point C for type-L.
8.2 Two types of customers 255

Since the H-type indifference curve is steeper than the L-type indifference curve at any
point, this must be true also at the no-insurance point, that is, the reservation indifference
curve for the H-type is steeper than the reservation indifference curve for the L-type, as
shown in Figure 8.2.1

Wealth in
good state
W2

NI 45o line
W0

W0 − hLmax FL
reservation indifference
W0 − hH
max FH curve for L-type
reservation indifference
curve for H-type

Wealth in
W1 bad state
0 W0 − `

Figure 8.2: The reservation indifference curve of an H-type is steeper than the reservation
indifference curve of an L-type.

It follows that the maximum premium that the L-type individuals are willing to pay for
full insurance, denoted by hLmax , is smaller than the maximum premium that the H-type
individuals are willing to pay for full insurance, denoted by hH L H
max : hmax < hmax , as shown
in Figure 8.2. Letting FL be the full-insurance contract that the L-type individuals consider
to be just as good as no insurance and FH the full-insurance contract that the H-type individ-
uals consider to be just as good as no insurance, we have that FL = W0 − hLmax ,W0 − hLmax

and FH = W0 − hH H L H


max ,W0 − hmax , with W0 − hmax > W0 − hmax .

1 Recallthat the reservation indifference curve is defined to be the indifference curve that goes through
the no-insurance point NI.
 256 Chapter 8. Adverse Selection in Insurance

Let NH > 0 be the number of H-type individuals in the population and NL > 0 be the
number of L-type individuals and define

NH NL
qH = so that 1 − qH = . (8.1)
NH + NL NH + NL

Then 0 < qH < 1 and 0 < 1 − qH < 1.

8.2.1 The contracts offered by a monopolist who can tell individuals apart
As a benchmark, we shall first consider the case of a monopolist who is able to tell whether
an individual who applies for insurance is an H-type or and L-type. For example, a health
insurance company might be legally allowed to require applicants to submit to a DNA test
that reveals whether a defective gene is present, in which case the individual is more likely
to develop a particular disease requiring extensive medical care.
In the perfect-information case, from the point of view of the monopolist there are
effectively two separate insurance markets: one for the H-types and one for the L-types.
Then we can apply the analysis of Chapter 5 (Section 5.2.1) and conclude that the monop-
olist would offer the full-insurance contract FL to type-L individuals and the full-insurance
contract FH to type-H individuals so that its expected profits would be2
hH L



total profits: max − pH ` NH + hmax − pL ` NL (8.2)

hH L



profit per customer: max − pH ` qH + hmax − pL ` (1 − qH ) . (8.3)
Which type of individual is “better” for the insurance company, that is, which type
yields higher profits? The H-type is better in that she is willing to pay a higher premium
for full insurance, but on the other hand she will submit a claim with higher probability,
that is, the H-type yields higher revenue but also higher cost. Thus, in principle, either type
could be more profitable. The answer depends on the specific values of the parameters.

√ is W0 = 3, 600, potential loss is ` = 2, 000 and

For example, suppose that initial wealth
the utility-of-money function is U(m) = m.
1 4
• Let pL = 10 and pH = 10 . To find hLmax solve the equation
1p 9p √
1, 600 + 3, 600 = W .
10 10
The solution is W = 3, 364 so that hLmax = 3, 600 − 3, 364 = $236 and thus the
expected profit from an L-type is
1
236 − (2, 000) = $36.
10
To find hH
max solve the equation

4p 6p √
1, 600 + 3, 600 = W .
10 10
2 To obtain profit per customer from total profits, divide by (NH + NL ) and use (8.1).
 8.3 The monopolist under asymmetric information 257

The solution is W = 2, 704 so that hH

max = 3, 600 − 2, 704 = $896 and thus the
expected profit from an H-type is
4
896 − (2, 000) = $96.
10
Thus in this case insuring an H-type is more profitable than insuring an L-type.
3 8
• Let pL = 10 and pH = 10 . To find hLmax solve the equation
3p 7p √
1, 600 + 3, 600 = W .
10 10
The solution is W = 2, 916 so that hLmax = 3, 600 − 2, 916 = $684 and thus the
expected profit from an L-type is
3
684 − (2, 000) = $84.
10
To find hH
max solve the equation

8p 2p √
1, 600 + 3, 600 = W .
10 10
The solution is W = 1, 936 so that hH
max = 3, 600 − 1, 936 = $1, 664 and thus the
expected profit from an H-type is
8
1, 664 − (2, 000) = $64.
10
Thus in this case insuring an L-type is more profitable than insuring an H-type.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 8.5.1 at the end of this chapter.

8.3 The monopolist under asymmetric information

We now turn to the case of asymmetric information, where each individual knows her own
probability of loss, but the monopolist only knows that there are NH high-risk individuals
with probability of loss pH and NL low-risk individuals with probability of loss pL .
We will consider three options for the monopolist:
Option 1. Cater only to the high-risk individuals by offering one insurance contract,
designed in such a way that only the H-type will purchase it.
Option 2. Cater to both types of individuals, by offering one insurance contract that is
attractive to both the L-type and the H-type.
Option 3. Offer a menu of two contracts: one – call it CH – targeted to the H-type and
the other – call it CL – targeted to the L-type.
 258 Chapter 8. Adverse Selection in Insurance

The shaded area in left pane of Figure 8.3 shows the set of insurance contracts that
are attractive to the H-type, in that they yield at least the reservation utility to this type
of individuals, while the shaded area in right pane shows the set of insurance contracts
that are attractive to the L-type, in that they yield at least the reservation utility to this type
of individuals. It is clear from Figure 8.3 that if an insurance contract is attractive to an
L-type then it is also attractive to an H-type.

W2 W2

NI 45o line NI 45o line

L-type L-type

H-type H-type

W1 W1
0 0

Figure 8.3: The shaded area in the left pane shows the set of insurance contracts that are
acceptable to the H-type; the shaded area in the right pane shows the set of insurance
contracts that are acceptable to the L-type.

8.3.1 The monopolist’s profit under Option 1

If the monopolist chooses Option 1 then it will offer a contract that lies in the shaded area
shown in Figure 8.4: the area between the two reservation indifference curves.

W2

NI 45o line

L-type
H-type FH

W1
0

Figure 8.4: Contracts in the shaded area are acceptable to the H-type but not to the L-type.
 8.3 The monopolist under asymmetric information 259

Under Option 1, the monopolist caters only to one type of individuals, namely the
H-type, and thus we can use the analysis of Chapter 5 (Section 5.2.1) and conclude that,
in order to maximize its profits, it will offer the full-insurance contract that leaves the
H-type indifferent between insuring and not insuring (shown as point FH in Figure 8.4).
Let hH
max be the premium of this contract. Then the monopolist’s total profits under Option
1, denoted by π1 , will be

π1 = NH hH


max − pH ` . (8.4)

For example, if the individuals’ initial wealth is W0 = 3, 600, the potential loss is
√ 8
` = 2, 000, the utility-of-money function is U(m) = m and pH = 10 , then, as calculated
at the end of the previous section, hH
max = $1, 664 and the expected profit from a single
8
contract would be 1, 664 − 10 (2, 000) = $64, so that total profits would be $64NH .

8.3.2 The monopolist’s profit under Option 2

If the monopolist chooses Option 2 then it will offer a contract that lies in the shaded area
shown in the right pane of Figure 8.3: the area on and above the reservation indifference
curve of the L-type. Which of these contracts maximizes the monopolist’s profits?
One might be tempted to infer from the analysis of Chapter 5 that the monopolist would
offer the full-insurance contract at the intersection of the reservation indifference curve
of the L-type and the 45o line. However this conclusion is not correct: the monopolist –
under Option 2 – would prefer to offer (to everybody) a partial-insurance contract.
To see this, recall the reasoning developed in Chapter 5: the crucial step in that reasoning
was to note that, at any point not on the 45o line, the reservation indifference curve of the
pL
L-type is steeper than the isoprofit line with slope − 1−p L
, so that there are points to the
right of the point under consideration which represent contracts that are acceptable to the
L-type and yield higher profit to the insurer if sold only to the L-types. In other words,
pL
the line with slope − 1−p L
is a relevant isoprofit line only under the assumption that the
insurer is dealing only with L-type individuals. However, as remarked above, any contract
that is acceptable to an L-type is also acceptable to an H-type and thus offering such a
contract implies that the expected profit from this contract is not [h − pL (` − d)] (where h
is the premium and d the deductible), because the probability of receiving a claim from a
customer should reflect the fact that some customers are L-types and others are H-types.
This is the essence of the notion of adverse selection: the contract that is offered
determines the composition of the pool of applicants: if the insurer offers the full insurance
contract with premium hH max determined in Section 8.3.1, then the pool of applicants will
consist entirely of H-types, while if the insurer offers a contract that is acceptable to the
L-types then the pool of applicants will consist of all the individuals, L-types and H-types.
What is the probability of receiving a claim from a customer if the insurance contract is
purchased by both types? Recall that qH is the fraction of individuals in the population who
are H-types (and (1 − qH ) is the fraction of L-types; see (8.1) on page 256). Thus we can
take qH as the probability that any particular customer taken from the set of customers who
submit a claim is an H-type (and (1 − qH ) as the probability that she is an L-type). Thus
the expected profit from a contract with premium h and deductible d which is purchased
260 Chapter 8. Adverse Selection in Insurance

by both types is:

h − [qH pH (` − d) + (1 − qH ) pL (` − d)] = h − [qH pH + (1 − qH ) pL ] (` − d) . (8.5)

We call the number [qH pH + (1 − qH ) pL ] the average probability of loss and denote it
by p̄:

average probability of loss: p̄ = qH pH + (1 − qH ) pL . (8.6)

Note that, since pL < pH and 0 < qH < 1,

pL < p̄ < pH . (8.7)

Thus when both types are insured with the same contract, the relevant isoprofit line is a
straight line with slope − 1−p̄ p̄ ; we call isoprofit lines with this slope average isoprofit lines.
It follows from (8.7) that
pL p̄ pH
< < . (8.8)
1 − pL 1 − p̄ 1 − pH

Of course, it is still true that at a point on the 45o line the slope of the L-type indifference
pL
curve is − 1−p L
; however the straight line with this slope is no longer a relevant isoprofit
line: the relevant isoprofit line has a slope of − 1−p̄ p̄ and is thus steeper than the L-type
indifference curve at that point. Figure 8.5 shows this for the reservation indifference curve
of the L-type.

Wealth in
good state
W2

line with slope

NI 45o line
− 1−p̄ p̄
L-type
line with
pL
slope − 1−p L

H-type

Wealth in
W1 bad state
0

Figure 8.5: The reservation indifference curve of the L-type is less steep, at a point on the
45o line, than the average isoprofit line, whose slope is − 1−p̄ p̄ .
8.3 The monopolist under asymmetric information 261

Given the relative slope of the L-type reservation indifference curve and the average
isoprofit line at a full-insurance contract, such a contract cannot be profit-maximizing
under Option 2: there will be contracts to the left of it (thus partial-insurance contracts) that
are acceptable to the L-type (and thus to both types) and are below that average isoprofit
line (and thus yield higher expected profits). Of course, this argument applies to any
contract where the average isoprofit line is steeper than the L-type reservation indifference
curve. Hence, the profit-maximizing choice for the monopolist under Option 2 is that
contract on the L-type reservation indifference curve where there is a tangency between
the indifference curve and the average isoprofit line, as shown in Figure 8.6.

Wealth in
good state
W2

NI 45o line

C
L-type
p̄
line with slope − 1− p̄

H-type

Wealth in
W1 bad state
0

Figure 8.6: C is the profit-maximizing contract for the monopolist under Option 2.

The existence of such a contract is guaranteed if and only if the slope of the L-type
reservation indifference curve at the no-insurance point is, in absolute value, greater than
p̄ 3
1− p̄ . On the other hand, if the slope of the L-type reservation indifference curve at the
no-insurance point is, in absolute value, less than 1−p̄ p̄ , then Option 2 cannot yield positive
profits. The reason for this is that the average isoprofit line that goes through the no-
insurance point is the zero-profit line and thus every contract that is acceptable to the
L-types will be above the zero profit line, which means that it would yield negative profits.
3 If p̄
this condition is satisfied, then at NI the slope of the indifference curve is larger than 1− p̄ and at a
p̄
point on the 45o line it is smaller than 1− p̄ ; thus, by the Intermediate Value Theorem, there must be a point
along the curve where it is equal.
262 Chapter 8. Adverse Selection in Insurance

If Option 2 is profitable, the profit-maximizing contract under this option, let us denote
it by C = (W1C ,W2C ), is given by the solution to the following equations:

pLU(W1C ) + (1 − pL )U(W2C ) = pLU(W0 − `) + (1 − pL )U(W0 ) (8.9)

U 0 (W1C )
 
pL p̄
= . (8.10)
1 − pL U 0 (W2C ) 1 − p̄

Equation (8.9) states that an L-type individual is indifferent between contract C and no
insurance, that is, contract C lies on the reservation indifference curve for the L-type;
equation (8.10) states that, at point C, the reservation indifference curve of the L-type is
tangent to (has the same slope as) the average isoprofit line.

As an example, let us revisit the case considered at the end of Section 8.2.1 where
√ is W0 = 33, 600, potential
the individuals’ initial wealth
8
loss is ` = 2, 000, the utility-of-
money function is U(m) = m, pL = 10 and pH = 10 ; furthermore, let NH = 2, 400 and
8
NL = 3, 900, so that qH = 21 . Thus the average probability of loss is
   
8 8 13 3 103
p̄ = + = .
21 10 21 10 210

To see if Option 2 is profitable, we check if the L-type reservation indifference curve is

steeper, at the no-insurance point NI = (1600, 3600), than the average isoprofit line; that
is, we check if
 0 
pL U (1, 600) p̄
0
> . (8.11)
1 − pL U (3, 600) 1 − p̄

Since !
3 1
U 0 (1, 600)
 
pL 10 80 9
= = = 0.6429
1 − pL U 0 (3, 600) 7
10
1
120
14
and
103
p̄ 210
= 107
= 0.9626,
1 − p̄ 210
inequality (8.11) is not satisfied and thus Option 2 is not profitable.

Let us now change the value of NL from 3,900 to 44,000: NL = 44, 000 (while everything
2,400 3
else remains as above). Then qH = 2,400+44,000 = 58 ; thus
   
3 8 55 3 189 p̄ 189
p̄ = + = so that = = 0.483
58 10 58 10 580 1 − p̄ 391

and thus inequality (8.11) is satisfied and Option 2 is profitable. The profit-maximizing
contract under Option 2, denoted by C = (W1C ,W2C ), is given by the solution to the following
equations (which correspond to equations (8.9) and (8.10)):
 8.3 The monopolist under asymmetric information 263
3 7 3p 7p
q q
W1C + W2C = 1, 600 + 3, 600
10 10 10 10
 1 
3 √ C 189
10  2 W1  580
7 1
= 391
.
10
√ C 580
2 W2

The solution is W1C = 2, 456.53 and W2C = 3, 124.97, that is, C = (2456.53, 3124.97); the
corresponding premium is 3, 600 − 3, 124.97 = $475.03 and the deductible is 3, 124.97 −
2, 456.53 = $668.44 so that the expected profit from a single contract is
189
475.03 − p̄(2, 000 − 668.44) = 475.03 − (1, 331.56) = $41.13
580
and total expected profits are

41.13 (NL + NH ) = 41.13(2, 400 + 44, 000) = $1, 908, 432.

8.3.3 The monopolist’s profit under Option 3

If the monopolist chooses Option 3 then it will offer two contracts: one contract– call it
CH – targeted to the H-type and the other contract – call it CL – targeted to the L-type. Let
us express these contracts in terms of premium and deductible and denote them by

CH = (hH , dH ) and CL = (hL , dL ) .

We shall use the following abbreviations:

EH [U(NI)] H-type’s expected utility from no insurance
EL [U(NI)] L-type’s expected utility from no insurance
EH [U(CH )] H-type’s expected utility from contract CH
EL [U(CH )] L-type’s expected utility from contract CH
EH [U(CL )] H-type’s expected utility from contract CL
EL [U(CL )] L-type’s expected utility from contract CL .
Thus
EH [U(NI)] = pH U (W0 − `) + (1 − pH )U (W0 )
EL [U(NI)] = pLU (W0 − `) + (1 − pL )U (W0 )
EH [U(CH )] = pH U (W0 − hH − dH ) + (1 − pH )U (W0 − hH )
EL [U(CH )] = pLU (W0 − hH − dH ) + (1 − pL )U (W0 − hH )
EH [U(CL )] = pH U (W0 − hL − dL ) + (1 − pH )U (W0 − hL )
EL [U(CL )] = pLU (W0 − hL − dL ) + (1 − pL )U (W0 − hL ) .
In order for contract CH to be chosen by H-type individuals two conditions must be
satisfied:

EH [U(CH )] ≥ EH [U(NI)] (IRH )

EH [U(CH )] ≥ EH [U(CL )] (ICH )
264 Chapter 8. Adverse Selection in Insurance

The first condition, (IRH ), is called the Individual Rationality constraint for type H and
says that the H-types must consider the contract targeted to them to be at least as good as
no insurance. The second condition, (ICH ), is called the Incentive Compatibility constraint
for type H and says that the H-types must consider the contract targeted to them to be at
least as good as the other contract that is offered (namely CL ).
Similarly, in order for contract CL to be chosen by L-type individuals two conditions must
be satisfied:

EL [U(CL )] ≥ EL [U(NI)] (IRL )

EL [U(CL )] ≥ EL [U(CH )] (ICL )

The first condition, (IRL ), is the Individual Rationality constraint for type L: it says that the
L-types must consider the contract targeted to them to be at least as good as no insurance.
The second condition, (ICL ), is he Incentive Compatibility constraint for type L: it says
that the L-types must consider the contract targeted to them to be at least as good as the
other contract that is offered (namely CH ).

Wealth in
good state
W2
contract CL
NI 45o line

L-type
H-type H-type
contract CH

Wealth in
W1 bad state
0

Figure 8.7: Two contracts that satisfy the four constraints.

Figure 8.7 shows a pair of contracts that satisfy all four constraints as strict inequalities:
(1) contract CH is strictly above the H-type reservation indifference curve and thus (IRH )
is satisfied as a strict inequality, (2) contract CL is to the left of the H-type indifference
curve that goes through contract CH and thus (ICH ) is satisfied as a strict inequality,
(3) contract CL is strictly above the L-type reservation indifference curve and thus (IRL ) is
8.3 The monopolist under asymmetric information 265

satisfied as a strict inequality and (4) contract CH is worse than contract CL for the L-type
(indeed, it is even worse than no insurance).
On the other hand, Figure 8.8 shows a pair of contracts where the (ICH ) and (IRL )
constraints are satisfied as equalities while the other two constraints are satisfied as strict
inequalities.

Wealth in
good state
W2

NI 45o line

CL

CH L-type
H-type H-type

Wealth in
W1 bad state
0

Figure 8.8: Another pair of contracts that satisfy the four constraints.

From now on, we will assume that, if indifferent between contract CH and contract
CL the H-types will choose contract CH . Furthermore, we will continue to assume that, if
indifferent between insuring and not insuring, each individual will choose to insure.
If the monopolist offers a menu of two contracts, CH and CL , that satisfy the four
constraints, then the H-types will purchase contract CH = (hH , dH ) and the L-types will
purchase contract CL = (hL , dL ) and thus the monopolist’s expected total profits will be
π3 = NH [hH − pH (` − dH )] + NL [hL − pL (` − dL )] .
Thus the monopolist, under Option 3, faces the following maximization problem:
Max π3 = NH [hH − pH (` − dH )] + NL [hL − pL (` − dL )]
hH ,dH ,hL ,dL
subject to
(IRH ) EH [U(CH )] ≥ EH [U(NI)]
(ICH ) EH [U(CH )] ≥ EH [U(CL )]
(IRL ) EL [U(CL )] ≥ EL [U(NI)]
(ICL ) EL [U(CL )] ≥ EL [U(CH )]
266 Chapter 8. Adverse Selection in Insurance

Let us study this maximization problem.

We showed at the beginning of this section (see Figure 8.3) that, if an insurance contract
is acceptable to the L-type (in that it lies on or above the L-type’s reservation indifference
curve), then it is acceptable to the H-type too (that is, it lies on or above the H-type’s
reservation indifference curve); thus

EL [U(CL )] ≥ EL [U(NI)] implies that EH [U(CL )] ≥ EH [U(NI)]. (8.12)

It follows that the (IRH ) constraint can be derived from the (IRL ) and (ICH ) constraints:

- by (IRL ), EL [U(CL )] ≥ EL [U(NI)], which, by (8.12) implies

EH [U(CL )] ≥ EH [U(NI)]; (8.13)

- by (ICH ), EH [U(CH )] ≥ EH [U(CL )] and this, together with (8.13) yields the (IRH )
constraint: EH [U(CH )] ≥ EH [U(NI)].
Thus:
 First observation: the (IRH ) constraint is redundant.

Hence the monopolist’s maximization problem can be simplified to:

Max π3 = NH [hH − pH (` − dH )] + NL [hL − pL (` − dL )]

hH ,dH ,hL ,dL

subject to:
EH [U(CH )]
z }| {
(ICH ) pH U (W0 − hH − dH ) + (1 − pH )U (W0 − hH )

≥ pH U (W0 − hL − dL ) + (1 − pH )U (W0 − hL )
| {z }
EH [U(CL )]

EL [U(CL )]
z }| { (8.14)
(IRL ) pLU (W0 − hL − dL ) + (1 − pL )U (W0 − hL )

≥ pLU (W0 − `) + (1 − pL )U (W0 )

| {z }
EL [U(NI)]

EL [U(CL )]
z }| {
(ICL ) pLU (W0 − hL − dL ) + (1 − pL )U (W0 − hL )

≥ pLU (W0 − hH − dH ) + (1 − pL )U (W0 − hH )

| {z }
EL [U(CH )]

 Second observation: at a solution of the above maximization problem, the (ICH )

constraint must be satisfied as an equality, that is, contracts CH and CL must be on
the same indifference curve for the H-type (as illustrated in Figure 8.8 on page 265).
8.3 The monopolist under asymmetric information 267

To see this, start with two contracts (CH ) and (CL ) that satisfy the above three constraints
and suppose that EH [U(CH )] > EH [U(CL )]. Modify contract CH by increasing the premium
hH up to the point where (ICH ) is satisfied as an equality, that is, up to the point where
EH [U(CH )] = EH [U(CL )].4 Then profits will increase, since π3 is increasing in hH , and
thus the initial pair {CH ,CL } could not have been a solution to the maximization problem.5
So far we have concluded that the solution to the initial constrained optimization
problem requires that the two contracts CH and CL must be on the same indifference curve
for the H-type.

 Third observation: at a solution of the maximization problem (8.14), contract

CL – which, by the second observation, must be on the same H-type indifference
curve as contract CH – must be above contract CH .

To see this, suppose that contract CH were above contract CL , as shown in Figure 8.9.

Wealth in
good state
W2

45o line

CH

L-type
CL
H-type

Wealth in
W1 bad state
0

Figure 8.9: Contract CH – which, by the second observation, must be on the same H-type
indifference curve as contract CL – cannot be above contract CL .

Then we can draw the indifference curve of the L-type that goes through contract CH : it
will be less steep than the indifference curve of the H-type and thus contract CL will be
below this L-type indifference curve, implying the the L-type would strictly prefer contract
CH to contact CL , contradicting the incentive compatibility constraint for the L-type, (ICL ).

4 The right-hand side of (ICH ) is independent of hH while the left-hand side is decreasing in hH .
5 Note that an increase in hH does not affect the (IRL ) constraint (both sides of it are independent of
hH ), while it reinforces the (ICL ) constraint, since the left-hand side of (ICL ) is independent of hH , while
the right-hand side is decreasing in hH ; thus if the (ICL ) constraint was satisfied to start with, then it will
continue to be satisfied after the increase in hH .
268 Chapter 8. Adverse Selection in Insurance

Thus, by the second and third observation, the two contracts CH and CL must be on the
same H-type indifference curve, with CL above CH , as shown in Figure 8.10.6

Wealth in
good state
W2

45o line

CL

L-type
CH H-type

isoprofit line with

pH
slope − 1−p
H
Wealth in
W1 bad state
0

Figure 8.10: Contract CH cannot be above the 45o line.

 Fourth observation: at a solution of the maximization problem (8.14), contract CH

must be a full-insurance contract.

To see this, suppose that CH is not a full-insurance contract, that is, suppose that it lies
above the 45o line. Then, as we know from Chapter 5, the H-type indifference curve is
pH
steeper at point CH than the isoprofit line with slope − 1−p H
, as shown in Figure 8.10. This
is indeed a relevant isoprofit line, because – by the (ICL ) constraint – contract CH will be
purchased only by the H-types. Hence there are points below contract CH , on the H-type
indifference curve, that will yield higher profits to the monopolist, since any such contract
would still be bought only by the H-types.7
6 To reach this conclusion one also needs to rule out the possibility that C = C . This is a consequence
H L
of the fourth observation below: starting from CH = CL above the 45o line, the monopolist could increase its
profits by separating CH from CL and moving it, along the H-type indifference curve, towards the 45o line;
on the other hand, if – to start with – CH = CL is already on the 45o line, then, by the (ICL ) constraint it must
be on or above the reservation indifference curve for the L-type and we know from the analysis of Option 2
that this is not a profit-maximizing configuration.
7 Moving contract C towards the 45o line, along the H-type indifference curve, will not alter the (IR )
H L
constraint (which is independent of hH and dH ) and will make contract CH even less attractive than contract
CL for the L-type, that is, the (ICL ) constraint will still hold.
8.3 The monopolist under asymmetric information 269

 Fifth observation: at a solution of the maximization problem (8.14), the (IRL )

constraint must be satisfied as an equality, that is, contract CL must be on the
reservation indifference curve of the L-type.

To see this, consider the situation depicted in Figure 8.11 where contract CL is above
the reservation indifference curve of the L-type (and – in accordance with the previous
observations – CL and CH lie on the same indifference curve of the H-type; furthermore,
CH is a full-insurance contract, that is, it lies on the 45o line). Draw the isoprofit line with
pL
slope − 1−p L
that goes through contract CL . We know from Chapter 5 that the H-type
pH
indifference curve at point CL is steeper than the line with slope − 1−p H
which, in turn, is
pL
steeper than the line with slope − 1−pL . Thus there are points on the H-type indifference
curve (to which both CL and CH belong) that are below this isoprofit line. Modify contract
CL by moving it along the H-type indifference curve up to the point where it intersects the
reservation indifference curve of the L-type, that is, until the (IRL ) constraint is satisfied
as an equality. Then the (ICH ) constraint is not affected (it is still satisfied as an equality)
and the (ICL ) constraint is also not affected, since contract CH is still worse, for the L-type,
than the modified contract CL . Thus the new CL contract is still purchased only by the
L-types and hence yields higher profits to the monopolist than the original CL contract
pL
(since the new contract is below the isoprofit line with slope − 1−p L
that goes through the
original contract).

Wealth in
good state
W2
isoprofit line with
pL
slope − 1−p
NI L 45o line
CL

L-type
H-type
CH

Wealth in
W1 bad state
0

Figure 8.11: Contract CL cannot be above the L-type reservation indifference curve.
270 Chapter 8. Adverse Selection in Insurance

We conclude from the above five observations that

The pair of contracts (CH ,CL )

is a solution of the maximization problem (8.14) only if
• CH is on the 45o line and
• CL lies at the intersection of
(1) the indifference curve of the H-type that goes through CH and
(2) the indifference curve of the L-type that goes through NI.

Fix an arbitrary premium hH for the full-insurance contract CH (thus dH = 0), such
that CH lies on the segment of the 45o line between FH and FL , where FH is the point of
intersection of the reservation indifference curve of the H-type and the 45o line and FL is
the point of intersection of the reservation indifference curve of the L-type and the 45o line:
see Figure 8.12. Then the contract CL at the intersection of the reservation indifference
curve of the L-type and the indifference curve of the H-type that goes through contract CH
is uniquely determined; hence we can think of this contract CL as a function of hH :

CL (hH ) = (hL (hH ), dL (hH )) . (8.15)

Wealth in
good state
W2

NI 45o line

CL
L-type
H-type FL
H-type CH
FH

Wealth in
W1 bad state
0

Figure 8.12: Contract CL is uniquely determined by the choice of contract CH on the

segment of the 45o line between points FH and FL .
8.3 The monopolist under asymmetric information 271

Then the constrained maximization problem (8.14) can be reduced to the following
unconstrained maximization problem:
Max π3 (hH ) = NH (hH − pH `) + NL [hL (hH ) − pL (` − dL (hH ))] (8.16)
hH ∈[hLmax ,hH
max ]

where hLmax is the maximum premium that the L-type is willing to pay for full insurance
(the premium of contract FL in Figure 8.12) and hH max is the maximum premium that the
H-type is willing to pay for full insurance (the premium of contract FH in Figure 8.12).

R Note that one possible choice under Option 3 is to set hH = hHmax , that is, to choose
CH = FH (see Figure 8.12), in which case the corresponding contract for the L-type
is the trivial contract with hL = 0 and dL = `, that is, CL = NI. This amounts to
insuring only the H-types, with the full-insurance contract that was obtained as
the profit-maximizing contract under Option 1. Thus Option 1 is a special case of
Option 3.

Before we discuss the solution to (8.16) in general, let us consider a numerical example.

√ W0 = 1, 600, poten-
Let the following be common to all individuals: initial wealth
tial loss ` = 700 and vNM utility-of-money function U(m) = m. The H-types face a
1
probability of loss pH = 15 while the L-types face a probability of loss pL = 10 .

• To find hLmax solve the equation

1√ 9p p
900 + 1, 600 = 1, 600 − h.
10 10
The solution is hLmax = 79.

• To find hH
max solve the equation
1√ 4p p
900 + 1, 600 = 1, 600 − h.
5 5
The solution is hH
max = 156.

Given hH ∈ [79, 156], contract CL is given by the solution to the following pair of equations:

1
√ 9√ 1√ 9√
10 900 + 10 1, 600 = 10 1, 600 − hL − dL + 10 1, 600 − hL
(8.17)
1√ 4√ √
5 1, 600 − hL − dL + 5 1, 600 − hL = 1, 600 − hH .
The first equation states that CL lies on the reservation indifference curve of the L-type and
the second equation states that CL lies on the indifference curve of the H-type that goes
through the full-insurance contract CH with premium hH . The solution to (8.17) is
p
hL (hH ) = hH + 156 1, 600 − hH − 6, 084 (8.18)
p
dL (hH ) = 80 hH + 5, 460 1, 600 − hH − 219, 260. (8.19)
272 Chapter 8. Adverse Selection in Insurance

Let there be a total of N individuals and let qH be the fraction of type-H individuals (thus
NH = qH N), so that (1 − qH ) is the fraction of type-L individuals (thus NL = (1 − qH )N).
Then the monopolist’s objective is to choose that value of hH ∈ [79, 156] that maximizes
the function

π3 (hH ) = N [qH (hH − pH 700) + (1 − qH ) [hL (hH ) − pL (700 − dL (hH ))] (8.20)

where hL (hH ) and dL (hH ) are given by (8.18) and (8.19), respectively. The solution of this
maximization problem depends on the value of qH , that is, on how many H-types there are
relative to the L-types.

1
For example, if qH = 20 then the graph of the profit function (8.20) is shown in Figure
8.13. The profit-maximizing value of hH is an interior point of the interval [79, 156], namely
96.64. Replacing this value in (8.18) and (8.19) we obtain hL = 61.26 and dL = 172.80.
Thus the monopolist offers a full-insurance contract CH = (hH = 96.64, dH = 0) and
a partial-insurance contract CL = (hL = 61.26, dL = 172.80) and the H-types purchase
contract CH while the L-types purchase contract CL . We call such a situation a separating
equilibrium, since the monopolist – through the menu of contracts it offers – is able to
induce a separation of the types: all the individuals of one type purchase one contract and
all the individuals of the other type purchase the other contract.

π3 (hH )

5.94N

5.5N

0.8N
hH
79 96.64 156

1
Figure 8.13: The graph of the profit function (8.20) when qH = 20 .
8.3 The monopolist under asymmetric information 273

Now consider the case where qH = 15 . For this case the graph of the profit function
(8.20) is shown in Figure 8.14. It is clear from the graph that in this case the solution of the
profit-maximizing problem is a corner solution at hH = hH
max = 156 and only the H-types
insure (the contract targeted to the L-types is the trivial contract with hL = 0 and dL = 700:
see the remark on page 271).

π3 (hH )

3.2N

hH
79 156

−5N

Figure 8.14: The graph of the profit function (8.20) when qH = 15 .

 274 Chapter 8. Adverse Selection in Insurance

The above example illustrates a general feature of the profit-maximization under

Option 3:

~ When qH is “sufficiently close to 1” (that is, when the number of H-types is large
relative to the entire population of potential customers), the monopolist will choose
to insure only the H-types by offering the full-insurance contract with premium hHmax
(that is, the solution to the profit maximization problem (8.16) given on page 271
is the corner solution, as illustrated in Figure 8.14). As a consequence, the (IRH )
constraint is satisfied as an equality.

~ Otherwise the monopolist will offer two contracts:

(1) a full-insurance contract CH with premium hH < hH max , which will be purchased
by the H-types, who therefore enjoy a surplus (that is, they are strictly better
off than with no insurance or, in other words, the (IRH ) constraint is satisfied
as a strict inequality), and
(2) a partial-insurance contract which will be purchased by the L-types, located
at the intersection of the reservation indifference curve of the L-types and the
indifference curve of the H-types that goes through contract CH (thus the (IRL )
and (ICH ) constraints are satisfied as equalities while the (IRH ) and (ICL )
constraints are satisfied as strict inequalities).
The expression “qH sufficiently close to 1” is rather vague: its exact meaning depends
on the specific values of the parameters. For instance, it can be shown that, in the example
9
considered above, “qH sufficiently close to 1” means qH ≥ 47 = 0.1915. 8

8.3.4 Option 2 revisited

We saw above that Option 1 can be viewed as a subcase of Option 3. Now we compare
Option 2 and Option 3.
Recall that the profit-maximizing contract under Option 2 (if it exists) is that contract
on the reservation indifference curve of the L-type where the slope of the indifference
curve is equal to the slope of the average profit line, which is − 1−p̄ p̄ , where p̄ is the average
probability of loss, that is,
p̄ = qH pH + (1 − pH ) pL .
Denote the maximum profits that the monopolist can make under Option 2 by π2∗ . If
B = (hB , dB ) is the profit-maximizing contract under Option 2 then
π2∗ = (NH + NL ) [hB − p̄ (` − dB )] ,
where NH is the number of individuals of type H and NL is the number of individuals of
type L.
8 Thus, 9
in that example, if qH ≥ 47 then the monopolist will choose to insure only the H-types, while if
9
qH < then the monopolist will implement a separating two-contract solution. To find this critical value of
47
qH , first calculate the derivative of the function π3 (hH ) given in (8.20) and evaluate it at the corner point
hH = 156; this will give an expression in terms of qH and then set this expression equal to zero and solve for
9 9
qH . In this case the solution is qH = 47 ; hence if qH ≥ 47 then the function π3 (hH ) is increasing or constant
9
at hH = 156 and thus the function is maximized at hH = 156, while if qH < 47 then the function π3 (hH ) is
decreasing at hH = 156 and thus the function is maximized at a point to the left of hH = 156.
8.3 The monopolist under asymmetric information 275

We want to show that π2∗ < π3∗ , where π3∗ is the maximum profit that the monopolist
can make under Option 3.
Figure 8.15 shows the contract, denoted by B, that maximizes profits under Option 2.
If the monopolist offers only this contract, then both types of individuals will purchase it.
Suppose that the monopolist switches from a one-contract menu containing only contract
B to a two-contract menu {B, F} obtained by adding to contract B also the full-insurance
contract F given by the intersection of the H-type indifference curve that goes through B
and the 45o line, as shown in Figure 8.15. Then the L-types will continue to buy contract B
(since it is strictly better for them that the newly added contract F), while the H-types will
switch to contract F.9 Thus profits from the L-types will not change, but profits from the
H-types will increase.10 The pair of contracts {B, F} satisfies all the constraints considered
under Option 3 (namely, IRH , ICH , IRL , ICL ) and yields profits that are larger than π2∗ ;
thus π3∗ (which may be even larger) is greater than π2∗ .

Wealth in
good state
W2

NI 45o line

L-type
H-type
p̄
line with slope − 1− p̄
F

Wealth in
W1 bad state
0

Figure 8.15: Option 2 is inferior to Option 3.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 8.5.2 at the end of this chapter.

9 As usual, we assume that, if indifferent, the H-types will choose the contract targeted to them. Otherwise
the newly offered contract would have to be slightly above contract F on the 45o line.
10 Because of the usual argument that at a point above the 45o line, such as point B in Figure 8.15, the
pH
H-type indifference curve is steeper than the isoprofit line with slope − 1−p H
, which is the relevant isoprofit
line if we consider contracts on the portion of the H-type indifference curve between point B and point F,
since those contracts would be purchased only by the H-types.
 276 Chapter 8. Adverse Selection in Insurance

8.4 A perfectly competitive insurance industry

In Chapters 2 and 5 (Sections 2.6.4 and 5.2.2) we studied the equilibrium in a perfectly
competitive insurance industry with free entry when there is only one type of potential
customer. Now we extend the analysis to the case of asymmetric information with two
types of individuals. We continue to assume that all the individuals are identical in terms
of initial wealth W0 and potential loss ` (with 0 < ` ≤ W0 ); furthermore, they also have the
same vNM utility-of-money function U(m). What they differ in is the probability of loss:
it is pH for type-H (= high-risk) individuals and pL for type-L (= low-risk) individuals,
with 0 < pL < pH < 1.

Recall that a free-entry competitive equilibrium is defined as a situation where

1. each firm in the industry makes zero profits, and

2. there is no unexploited profit opportunity in the industry, that is, there is no currently
not offered contract that would yield positive profits to a (existing or new) firm that
offered that contract.

In the one-type case we saw that at the free-entry competitive equilibrium every firm
offers the full-insurance contract that lies at the intersection of the zero-profit line and the
45o line. Recall that the no-insurance point NI can be thought of as a trivial contract with
zero premium and deductible equal to the full loss `: such a “contract” obviously yields
zero profits; thus the zero-profit line goes through point NI. In the two-type context the
situation is complicated by the fact that there are three zero-profit lines:

pL
• a “low-risk” line (through NI) with slope − 1−p L
, which is a relevant isoprofit line
if and only if the contracts on this line are sold only to the low-risk individuals,

• an “average-risk” line (through NI) with slope − 1−p̄ p̄ , where

p̄ = qH pH + (1 − qH )pL ,

which is a relevant isoprofit line if and only if the contracts on this line are sold to
both types of individuals,

pH
• a “high- risk” line (through NI) with slope − 1−p H
, which is a relevant isoprofit line
if and only if the contracts on this line are sold only to the hight-risk individuals.

Since 0 < pL < pH < 1 and 0 < qH < 1, pL < p̄ < pH and thus

pL p̄ pH
< < . (8.21)
1 − pL 1 − p̄ 1 − pH

Hence the low-risk zero-profit line is less steep that the average risk zero-profit line, which,
in turn, is less steep than the high-risk zero-profit line.
8.4 A perfectly competitive insurance industry 277

Figure 8.16 shows the three zero-profit lines.

W2

NI 45o line

pL
line with slope − 1−p L

line with slope − 1−p̄ p̄

pH
line with slope − 1−p H

W1
0

Figure 8.16: Three zero-profit lines in the two-type case.

We also note that, at any insurance contract, the slope of the H-type indifference curve
pH
is – in absolute value – greater than or equal to 1−p H
and thus, by (8.21), it is greater than
the absolute value of the slope of the average isoprofit line through that point, as shown in
Figure 8.17.

W2

H-type 45o line

p̄
line with slope − 1− p̄

W1
0

Figure 8.17: At any (full- or partial-) insurance contract the H-type indifference curve is
steeper than the average isoprofit line.
278 Chapter 8. Adverse Selection in Insurance

In what follows we shall assume that the reservation indifference curve of the L-type is
steeper, at the no-insurance point, than the average zero-profit line, that is,

U 0 (W0 − `)
 
pL p̄
> . (8.22)
1 − pL U 0 (W0 ) 1 − p̄

Inequality (8.22) ensures that it is possible to make positive profits by insuring both types
of individuals with the same contract.11

In principle, a free-entry competitive equilibrium could be one of two types:

• a pooling equilibrium where all the firms in the industry offer the same contract,
which is bought by both types of individuals, or

• a separating equilibrium where two contracts are offered in the industry: one
contract, denoted by CH , which is purchased only by the H-types and the other
contract, denoted by CL , which is purchased only by the L-types.

Let us begin by considering the possibility of a pooling equilibrium. We want to show

that such an equilibrium is not possible. By Condition 1 of the definition of free-entry
competitive equilibrium, the contract in question, call it B = (hB , dB ) (with 0 ≤ dB < `),
must be on the average zero profit line, that is,

hB − p̄ (` − dB ) = 0. (8.23)

Consider the function πL (h, d) = h − pL (` − d) that gives, for every insurance contract
(h, d) the expected profit from that contract if it is bought only by the low-risk individuals,
that is, only by the L-types. Since pL < p̄, pL (` − dB ) < p̄(` − dB ) and thus it follows
from (8.23) that

πL (hB , dB ) = hB − pL (` − dB ) > 0. (8.24)

Since the function πL (h, d) is a continuous function, it follows from (8.24) that, for every
insurance contract A = (hA , dA ),

if A is sufficiently close to B then πL (hA , dA ) = hA − pL (` − dA ) > 0. (8.25)

Thus, if we can find a contract, close to B, that would be considered better than B by the
L-types, but worse than B by the H-types, then any firm that introduced contract A would
attract only the L-types and thus, by (8.25), it would make positive profits, contradicting
the second requirement of the definition of a free-entry competitive equilibrium. Does
such a contract exist?
11 By (8.22) there are contracts that are above the reservation indifference curve of the L-type (and thus
attractive to both types) and below the average zero-profit line (and thus yielding positive profits if purchased
by both types).
8.4 A perfectly competitive insurance industry 279

The answer is affirmative, as shown in Figure 8.18: draw the indifference curves of the
two types that go through contract B. The L-type indifference curve is less steep than the
H-type indifference curve and thus there are contracts, such as contract A in Figure 8.18,
which are below the H-type indifference curve through B and above the L-type indifference
curve through B. Thus if such a contract A were to be introduced, the L-types would
switch from the original contract B to the new contract A while the H-types would stay
with contract B. Hence, by (8.25), the firm that introduced contract A would make positive
profits, contradicting the second requirement of the definition of free-entry competitive
equilibrium.

Wealth in
good state
W2

45o line
A
B
L-type
H-type

Wealth in
W1 bad state
0

Figure 8.18: If, initially, both types purchase contract B and then contract A is added as an
option, then the H-types will not switch to A while the L-types will.

Thus we conclude that, if there is a free-entry competitive equilibrium, then it must

be a two-contract separating equilibrium. We now turn to the question of whether a
two-contract {CH ,CL } equilibrium exists with all H-types purchasing contract CH and all
L-type purchasing contract CL .

• By the zero-profit condition (the first requirement of a free-entry competitive equi-

librium), contract CH must be on the high-risk zero-profit line (the line through NI
pH
with slope − 1−p H
) and contract CL must be on the low-risk zero-profit line (the line
pL
through NI with slope − 1−p L
).
280 Chapter 8. Adverse Selection in Insurance

• Furthermore, contract CH must be the full-insurance contract on the high-risk zero-

profit line because, if it were above the 45o line, then – by the usual argument
based on the observation that at such a point the H-type indifference curve is steeper
than the high-risk zero-profit line – there would be contracts below that zero profit
line and above that indifference curve that would yield positive profits to a firm
that introduced such a contract (which would induce the H-types to switch to it),
contradicting the second requirement of a free-entry competitive equilibrium.
It remains to determine where on the low-risk zero-profit line contract CL should be. Draw
the indifference curve of the H-type that goes through contract CH and call the point at
the intersection of this indifference curve and the low-risk zero-profit line contract C, as
shown in Figure 8.19.

Wealth in
good state
W2

H-type 45o line

NI A
CB
low-risk zero-profit line
high-risk
zero-profit CH
line

Wealth in
W1 bad state
0

Figure 8.19: Where on the low-risk zero-profit line could contract CL be?

• Suppose that contract CL is on the low-risk zero-profit line to the right of point C,
such as point B in Figure 8.19. Then such a contract B would be preferred to CH by
both types,12 giving rise to a situation where all types purchase the same contract,
which – as we saw above – cannot be a free-entry competitive equilibrium.

12 B is preferred to CH by type H because B is to the right of the H-type indifference curve through CH and
B is preferred to CH by type L because CH is below the L-type indifference curve through B: this indifference
curve is not shown in Figure 8.19, but it is less steep than the indifference curve of the H-type through B
(also not shown); hence, since CH is below the latter, it is also below the former.
8.4 A perfectly competitive insurance industry 281

• Suppose that contract CL is on the low-risk zero-profit line to the left of point C, such
as point A in Figure 8.19. Then – by the usual argument based on the observation that
at such a point the L-type indifference curve is steeper than the low-risk zero-profit
line – there would be contracts below that zero profit line and above the indifference
curve of the L-type that goes through A that would attract the L-types, and only the
L-types, and yield positive profits, contradicting the first requirement of a free-entry
competitive equilibrium.

Thus we conclude that contract CL must be at the intersection of the low-risk zero-profit
line and the indifference curve of the H-type that goes through contract CH , that is, it must
coincide with point C in Figure 8.19.
We have determined that necessary conditions for a pair of contracts {CH ,CL } to be a
free-entry competitive equilibrium are:

(1) CH is at the intersection of the high-risk zero-profit line and the 45o line, and
(2) CL is at the intersection of the low-risk zero-profit line and the H-type indifference
curve through CH .
While necessary, the above two conditions are not sufficient for a free-entry competitive
equilibrium. To see this, consider the situation depicted in Figure 8.20.

Wealth in
good state
W2

H-type 45o line

NI CL
low-risk zero-profit line
D average zero-profit line
high-risk L-type
zero-profit
line
CH

Wealth in
W1 bad state
0

Figure 8.20: A case where the pair of contracts {CH ,CL } is not a free-entry competitive
equilibrium because of the existence of contracts like D.
282 Chapter 8. Adverse Selection in Insurance

Suppose that the contracts currently offered in the industry are CH and CL and one of the
existing firms, or a new firm, introduces contract D. The H-types will switch to contract D,
since it is better than the contract that they are currently purchasing, namely CH (D is to
the right of the H-type indifference curve through CH ) and the L-types will also switch to
D, since it is better than the contract that they are currently purchasing, namely CL (D is to
the right of the L-type indifference curve through CL ). Then, for the firm that introduced
contract D, the relevant isoprofit lines are the average isoprofit lines; since contract D is
below the average zero-profit line, it yields positive profits to the firm, contradicting the
second requirement of a free-entry competitive equilibrium.
Thus, in order for the two-contract configuration {CH ,CL } described above to be a
free-entry competitive equilibrium, it is also necessary that there be no contracts such as
contract D described above, that is, it must not be the case that the average zero-profit line
crosses the L-type indifference curve that goes through contract CL . In other words, the
average zero-profit line must be entirely below the L-type indifference curve through CL ,
as shown in Figure 8.21.
Since the higher the value of qH (that is, the larger the number of H-types in the
population relative to the number of L-types), the closer the average zero-profit line will be
to the high-risk zero-profit line, this additional requirement for the existence of a free-entry
competitive equilibrium can be understood in terms of qH being “sufficiently large”.

W2

H-type 45o line

NI CL
low-risk zero-profit line
high-risk L-type
zero-profit average zero-profit line
line
CH

0 W1

Figure 8.21: The pair of contracts {CH ,CL } is a free-entry competitive equilibrium.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 8.5.3 at the end of this chapter.
 8.5 Exercises 283

8.5 Exercises

The solutions to the following exercises are given in Section 8.6 at the end of this chapter.

8.5.1 Exercises for Section 8.2: Two types of customers

Exercise 8.1 Sara and Mary have the same initial wealth of $400 and face the same
potential loss of $280. They also have the same vNM utility-of-money function
U(m) = ln(m). They differ, however, in the probability of loss, which is pS = 12
for Sara and pM = 15 for Mary.
(a) Calculate the slope of Sara’s reservation indifference curve at the no-insurance
point.
(b) Calculate the slope of Mary’s reservation indifference curve at the no-insurance
point.
(c) Find the maximum premium that Sara would be willing to pay for full insurance
and calculate the insurance company’s expected profit from selling that full-
insurance contract to Sara.
(d) Find the maximum premium that Mary would be willing to pay for full insurance
and calculate the insurance company’s expected profit from selling that full-
insurance contract to Mary.


Exercise 8.2 Diana and Fran have the same initial wealth of $4,096 and face the same
potential√loss of $2,800. They also have the same vNM utility-of-money function
U(m) = m. They differ, however, in the probability of loss, which is pD = 14 for Diana
1
and pF = 16 for Fran.
Consider a monopolist seller of insurance who knows all of the above information about
Diana and Fran.
(a) What insurance contract would the monopolist offer to Diana?
(b) What insurance contract would the monopolist offer to Fran?
(c) Assume that, when indifferent between not insuring and insuring, both Diana
and Fran will choose to insure. Calculate the monopolist’s expected profit from
selling insurance to Diana and Fran.

 284 Chapter 8. Adverse Selection in Insurance

8.5.2 Exercises for Section 8.3: The monopolist under asymmetric information

Exercise 8.3 There are 16,000 individuals, all identical in terms of initial wealth
W0 = $4, 000, potential
 loss ` = $2, 500 and vNM utility-of-money function
m
U(m) = 10 ln 1,000 . Of these individuals, 12,000 have a high probability of loss
pH = 51 and 4,000 have a low probability of loss pL = 15 1
. A monopolist seller of in-
surance is considering using Option 1 (Section 8.3.1). What will its expected total
profits be? [As usual, assume that – if indifferent between insuring and not insuring –
individuals will choose to insure.] 

Exercise 8.4 Consider again the case described in Exercise 8.3: there are 16,000 individ-
uals, all identical in terms of initial wealth W0 = $4,  000, potential loss
m
` = $2, 500 and vNM utility-of-money function U(m) = 10 ln 1,000 ; of these individ-
1
uals, 12,000 have a high probability of loss pH = 5 and 4,000 have a low probability of
1
loss pL = 15 .
(a) Show that, for a monopolist seller of insurance, Option 2 (Section 8.3.2) is not
profitable.
(b) Calculate what the monopolist’s total expected profits would be if it offered a
full-insurance contract that makes the L-type indifferent between insuring and
not insuring. [As usual, assume that – if indifferent between insuring and not
insuring – individuals will choose to insure.]


Exercise 8.5 Consider again the case described in Exercise 8.3, but reduce the number
of high-risk individuals from 12,000 to 6,000; thus there are 10,000 individuals, all
identical in terms of initial wealth W0 =  potential loss ` = $2, 500 and vNM
 $4, 000,
m
utility-of-money function U(m) = 10 ln 1,000 ; of these individuals, 6,000 have a high
1 1
probability of loss pH = 5 and 4,000 have a low probability of loss pL = 15 .
(a) Show that, for a monopolist seller of insurance, Option 2 (Section 8.3.2) is
profitable, by calculating the slopes of the relevant curves at the no-insurance
point.
(b) Write two equations, whose solution gives the contract that maximizes the mo-
nopolist’s profits under Option 2. If you are able to, compute the solution and
determine the monopolist’s expected total profits if it offers that contract. [As
usual, assume that – if indifferent between insuring and not insuring – individuals
will choose to insure.]

8.5 Exercises 285

Exercise 8.6 There are 6,000 individuals, all with the same initial wealth W0 = 16, 000,
potential loss ` = 7, 000 and with the same vNM utility-of-money
facing the same √
function U(m) = m. Of these 6,000 individuals, 1,000 are high-risk with a probability
2
of loss pH = 10 while the remaining 5,000 are low-risk with a probability of loss
1
pL = 10 .

(a) Calculate the monopolist’s total expected profits if it decides to pursue Option 1.
(b) Calculate the average probability of loss p̄.
(c) Write two equations whose solution gives the profit-maximizing contract under
Option 2.
(d) Suppose that, under Option 3, the monopolist decides to offer a full-insurance
contract with premium $1,400. Write a pair of equations whose solution gives
the other contract that the monopolist will offer.
(e) The solution of the pair of equations in Part (d) is the following contract, expressed
in term of wealth levels: C = (10169.41, 15832.48). Calculate the monopolist’s
profits if it offers the two contracts of Part (d).


Exercise 8.7 There are 8,000 individuals, all with the same initial wealth W0 = 10, 000,
facing the same potential loss ` = 6, 000 and with the same vNM utility-of-money
function U(m) = ln(m). Of these 8,000 individuals, 1,500 are high-risk with a proba-
bility of loss pH = 41 while the remaining 6,500 are low-risk with a probability of loss
1
pL = 16 .

(a) Calculate the monopolist’s total expected profits if it decides to pursue Option 1.
(b) Calculate the average probability of loss p̄.
(c) Write two equations whose solution gives the profit-maximizing contract under
Option 2.
(d) Suppose that, under Option 3, the monopolist decides to offer a full-insurance
contract with premium $2,000. Write a pair of equations whose solution gives
the other contract that the monopolist will offer.
(e) The solution of the pair of equations in Part (d) is the following contract, expressed
in term of wealth levels: C = (4120.36, 9980.26). Calculate the monopolist’s
profits if it offers the two contracts of Part (d).

 286 Chapter 8. Adverse Selection in Insurance

8.5.3 Exercises for Section 8.4: A perfectly competitive insurance industry

Exercise 8.8 Consider again the information given in Exercise 8.6: there are 6,000
individuals, all with the same initial wealth W0 = 16, 000, facing the same potential
√
loss ` = 7, 000 and with the same vNM utility-of-money function U(m) = m; of
2
these 6,000 individuals, 1,000 are high-risk with probability of loss pH = 10 while the
1
remaining 5,000 are low-risk with probability of loss pL = 10 .
(a) Find the pair of contracts that is the only candidate for a free-entry perfectly
competitive equilibrium.
(b) Calculate the average probability of loss p̄.
(c) Write a pair of equations such that if it does not have a solution then the pair of
contracts of Part (a) is a free-entry perfectly competitive equilibrium.
(d) Suppose that the pair of equations in Part (c) does have a solution. Write a third
equation such that (1) if it is satisfied (at the solution of the other two equations),
then the pair of contracts of Part (a) is a free-entry perfectly competitive equi-
librium and (2) if it is not satisfied (at the solution of the other two equations),
then the pair of contracts of Part (a) is not a free-entry perfectly competitive
equilibrium.


Exercise 8.9 Consider again the information given in Exercise 8.7: there are 8,000
individuals, all with the same initial wealth W0 = 10, 000, facing the same potential
loss ` = 6, 000 and with the same vNM utility-of-money function U(m) = ln(m); of
these 8,000 individuals, 1,500 are high-risk with probability of loss pH = 14 while the
1
remaining 6,500 are low-risk with probability of loss pL = 16 .
(a) Find the pair of contracts that is the only candidate for a free-entry perfectly
competitive equilibrium.
(b) Calculate the average probability of loss p̄.
(c) Write a pair of equations such that: (1) if it has a solution then the pair of contracts
of Part (a) is not a free-entry perfectly competitive equilibrium and (2) if it does
not have a solution then the pair of contracts of Part (a) is a free-entry perfectly
competitive equilibrium.

 8.6 Solutions to Exercises 287

8.6 Solutions to Exercises

Solution to Exercise 8.1

(a) The slope of Sara’s reservation indifference curve at NI = (120, 400) is

!
1 1
U 0 (120)
 
pS 2 120 10
− =− =− = −3.33.
1 − pS U 0 (400) 1− 1
2
1
400
3

(b) The slope of Mary’s reservation indifference curve at NI = (120, 400) is

!
1 1
U 0 (120)
 
pM 5 120 5
− =− = − = −0.83.
1 − pM U 0 (400) 1 − 15 1
400
6

(c) To find the full-insurance contract that makes Sara indifferent between insuring and
not insuring we need to solve the equation ln(W ) = 12 ln(120) + 21 ln(400). The
solution is W = 219.09, so that the maximum premium that Sara is willing to pay
for full insurance is 400 − 219.09 = $180.91. If Sara purchases this full-insurance
contract then the insurance company’s expected profit is 180.91 − 12 (280) = $40.91.
(d) To find the full-insurance contract that makes Mary indifferent between insuring
and not insuring we need to solve the equation ln(W ) = 15 ln(120) + 45 ln(400). The
solution is W = 314.40, so that the maximum premium that Mary is willing to pay
for full insurance is 400 − 314.40 = $85.60. If Mary purchases this full-insurance
contract then the insurance company’s expected profit is 85.60 − 15 (280) = $29.60.


Solution to Exercise 8.2

(a) The monopolist will offer Diana the full-insurance contract that makes her indif-
ferent
√ between
√ insuring
√ and not insuring, which is determined by the solution to
W = 14 1, 296 + 34 4, 096. The solution is W = 3, 249, so that the premium of
the full-insurance contract is 4, 096 − 3, 249 = $847. If Diana purchases this full-
insurance contract then the insurance company’s expected profit from this contract
is 847 − 14 (2, 800) = $147.

(b) The monopolist will offer Fran the full-insurance contract that makes her indif-
ferent
√ between insuring√and not insuring, which is determined by the solution to
1√
W = 16 1, 296 + 15
16 4, 096. The solution is W = 3, 875.06, so that the pre-
mium of the full-insurance contract is 4, 096 − 3, 875.06 = $220.94. If Fran pur-
chases this full-insurance contract then the insurance company’s expected profit is
1
220.94 − 16 (2, 800) = $45.94.

(c) The monopolist’s expected profit from insuring Diana and Fran is 147 + 45.94 =
$192.94. 
288 Chapter 8. Adverse Selection in Insurance

Solution to Exercise 8.3

The monopolist would offer the full-insurance contract that makes the H-types indif-
ferent between insuring and not insuring. To find that contract, solve the equation
U(W ) = pH U(W0 − `) + (1 − pH )U(W0 ), that is,
 
W 1 4
10 ln = 10 ln(1.5) + 10 ln(4).
1, 000 5 5
The solution is W = 3, 287.50. Thus the offered full-insurance contract has a premium of
4, 000 − 3, 287.50 = $712.50. This contract will be purchased only by the H-types (the
L-types are better off without insurance). Thus the monopolist’s expected total profits are
 
1
12, 000 712.50 − (2, 500) = $2, 550, 000.
5


Solution to Exercise 8.4

(a) We need to show that the L-type reservation indifference curve is less steep at the
no-insurance point, than the average isoprofit line, that is,
 0 
pL U (W0 − `) p̄
0
≤ (8.26)
1 − pL U (W0 ) 1 − p̄
12,000
First we compute the average probability of loss p̄. Since qH = 16,000 = 43 ,
   
3 1 1 1 1
p̄ = + = .
4 5 4 15 6
Thus
p̄ 1
= = 0.2.
1 − p̄ 5
On the other hand,
10
U 0 (W0 − `)
 
pL 1 1,500 4
= = = 0.1905.
1 − pL U 0 (W0 ) 14 10
4,000
21

Thus inequality (8.26) is indeed satisfied.

(b) To find that contract, solve the equation U(W ) = pLU(W0 − `) + (1 − pL )U(W0 ),
that is,  
W 1 14
10 ln = 10 ln(1.5) + 10 ln(4).
1, 000 15 15
The solution is W = 3, 746.81. Thus the offered full-insurance contract has a pre-
mium of 4, 000 − 3, 746.81 = $253.19. This contract would be purchased by both
types. Thus the monopolist’s expected total profits would be N (253.19 − p̄`):
 
1
16, 000 253.19 − (2, 500) = $ − 2, 615, 626.67,
6
that is, a loss. 
8.6 Solutions to Exercises 289

Solution to Exercise 8.5

(a) We need to show that the L-type reservation indifference curve is steeper, at the
no-insurance point, than the average isoprofit line, that is,
 0 
pL U (W0 − `) p̄
> (8.27)
1 − pL U 0 (W0 ) 1 − p̄

6,000
First we compute the average probability of loss p̄. Since qH = 10,000 = 53 ,
   
3 1 2 1 11
p̄ = + = .
5 5 5 15 75

Thus
p̄ 11
= = 0.1719.
1 − p̄ 64
On the other hand,
10
U 0 (W0 − `)
 
pL 1 1,500 4
= = = 0.1905.
1 − pL U 0 (W0 ) 14 10
4,000
21

Thus inequality (8.27) is indeed satisfied.

(b) The equations are as follows (see (8.9) and (8.10) on page 262):

pLU(W1 ) + (1 − pL )U(W2 ) = pLU(W0 − `) + (1 − pL )U(W0 )

U 0 (W1 )
 
pL p̄
= .
1 − pL U 0 (W2 ) 1 − p̄

that is,
   
1 W1 14 W1 1 14
10 ln + 10 ln = 10 ln(1.5) + 10 ln(4)
15 1, 000 15 1, 000 15 15
10
!
1 11
15 W1 75
14 10
= 64
.
15 W2 75

The solution is W1 = 1, 650.99 and W2 = 3, 972.69. Thus the premium of the offered
contract is 4, 000 − 3, 972.69 = $27.31 and the deductible is 3, 972.69 − 1, 650.99 =
$2, 321.70. Both types purchase this contract. Thus the monopolist’s expected total
profits are:
 
11
10, 000 27.31 − (2, 500 − 2, 321.70) = $11, 593.33.
75


290 Chapter 8. Adverse Selection in Insurance

Solution to Exercise 8.6

(a) First calculate the maximum premium that the H-types are willing to pay for full
insurance by solving
2p 8p p
9, 000 + 16, 000 = 16, 000 − h.
10 10
The solution is hHmax = $1, 560. Thus under Option 1 the monopolist would only
offer full insurance at a premium of $1,560, attracting only the H-types; its total
expected profits would be
 
2
1, 000 1, 560 − 7, 000 = $160, 000.
10
(b) The average probability of loss is
   
1 2 5 1 7
p̄ = qH pH + (1 − qH ) pL = + = .
6 10 6 10 60
(c) The equations whose solution gives the profit-maximizing contract under Option 2
are:

pLU(W0 − `) + (1 − pL )U(W0 ) = pLU(W0 − h − d) + (1 − pL )U(W0 − h)

 0 
pL U (W0 − h − d) p̄
= , that is,
1 − pL U 0 (W0 − h) 1 − p̄

1p 9p 1p 9p
9, 000 + 16, 000 = 16, 000 − h − d + 16, 000 − h
10 10 10 10
1 √1 !
7
10 2 16,000−h−d 60
9
= .
√ 1 53
10 2 16,000−h 60

(d) The assumption is that CH = (hH = 1400, dH = 0). To find contract CL = (hL , dL )
solve the following equations:
2p 8p p
16, 000 − h − d + 16, 000 − h = 16, 000 − 1, 400
10 10
1 p 9p 1p 9p
16, 000 − h − d + 16, 000 − h = 9, 000 + 16, 000
10 10 10 10
(e) The two contracts are: CH = (hH = 1400, dH = 0), which will be bought by the
H-types, and CL = (hL = 167.52, dL = 5663.07), which will be bought by the L-
types.13 Thus the monopolist’s expected total profits will be
   
2 1
1, 000 1, 400 − 7, 000 +5, 000 167.52 − (7, 000 − 5, 663.07) = $169, 135. 
10 10
13 The
contract was given in terms of wealth levels as (10169.41, 15832.48), from which we obtain the
premium as 16, 000 − 15, 832.48 = 167.52 and the deductible as 15, 832.48 − 10, 169.41 = 5, 663.07.
8.6 Solutions to Exercises 291

Solution to Exercise 8.7

(a) First calculate the maximum premium that the H-types are willing to pay for full
insurance by solving
1 3
ln(4, 000) + ln(10, 000) = ln(10, 000 − h).
4 4
The solution is hHmax = $2, 047.29. Thus under Option 1 the monopolist would only
offer full insurance at a premium of $2,047.29, attracting only the H-types; its total
expected profits would be
 
1
1, 500 2, 047.29 − (6, 000) = $820, 935.
4

(b) The average probability of loss is

   
3 1 13 1 25
p̄ = qH pH + (1 − qH ) pL = + = .
16 4 16 16 256
(c) The equations whose solution gives the profit-maximizing contract under Option 2
are:
pLU(W0 − `) + (1 − pL )U(W0 ) = pLU(W0 − h − d) + (1 − pL )U(W0 − h)
 0 
pL U (W0 − h − d) p̄
0
=
1 − pL U (W0 − h) 1 − p̄
that is,
1 15 1 15
ln(4, 000) + ln(10, 000) = ln(10, 000 − h − d) + ln(10, 000 − h)
16 16 16 16
1
!
1 25
16 10,000−h−d 256
15 1
= 231
.
16 10,000−h 256

(d) The assumption is that CH = (hH = 2000, dH = 0). To find contract CL = (hL , dL )
solve the following equations:
1 3
ln(10, 000 − h − d) + ln(10, 000 − h) = ln(10, 000 − 2, 000)
4 4
1 15 1 15
ln(10, 000 − h − d) + ln(10, 000 − h) = ln(4, 000) + ln(10, 000)
16 16 16 16
(e) The two contracts are: CH = (hH = 2000, dH = 0), which will be bought by the H-
types, and CL = (hL = 19.74, dL = 5859.9), which will be bought by the L-types.14
Thus the monopolist’s expected total profits will be
   
1 1
1, 500 2, 000 − 6, 000 +6, 500 19.74 − (6, 000 − 5, 859.9) = $821, 394.38. 
4 16
14 The
contract was given in terms of wealth levels as (4120.36, 9980.26), from which we obtain the
premium as 10, 000 − 9, 980.26 = 19.74 and the deductible as 9, 980.26 − 4, 120.36 = 5, 859.9.
292 Chapter 8. Adverse Selection in Insurance

Solution to Exercise 8.8

(a) Contract CH is the full-insurance contract that yields zero profits if bought only by
the H-types. Thus its premium is given by the solution to hH − pH ` = 0, that is,
2
hH = 10 (7, 000) = $1, 400. Contract CL is given by the intersection of the H-type
indifference curve through CH and the low-risk zero-profit line; thus it is given by
the solution to

p 2p 8p
16, 000 − 1, 400 = 16, 000 − h − d + 16, 000 − h
10 10
1
h− (7, 000 − d) = 0
10

which is CL = (hL = 109.32, dL = 5906.81).

(b) The average probability of loss is

   
1 2 5 1 7
p̄ = qH pH + (1 − qH ) pL = + = .
6 10 6 10 60

(c) We need to express the fact that there is a contract at which the average zero-profit
line intersects the L-type indifference curve through contract
CL = (hL = 109.32, dL = 5906.81):

7
h− (7, 000 − d) = 0
60
1p 9p
16, 000 − 109.32 − 5, 906.81 + 16, 000 − 109.32
10 10
1p 9p
= 16, 000 − h − d + 16, 000 − h.
10 10

(d) Let E = (hE , dE ) be a contract that is a solution to the two equations of Part (c),
that is, E lies both on the average zero-profit line and on the L-type indifference
curve through contract CL = (hL = 109.32, dL = 5906.81). The existence of such
a contract E is compatible with the fact that the pair of contracts of Part (a) is a
free-entry competitive equilibrium if and only if there is a tangency at contract E
between the indifference curve and the average zero-profit line, that is, if the two
slopes are the same: a condition that is expressed by the following equation (the
left-hand side is 1−p̄ p̄ and the right-hand side is the absolute value of the slope of the
indifference curve at contract E):
 √ 
7 1 16, 000 − hE
= √ .
53 9 16, 000 − hE − dE


8.6 Solutions to Exercises 293

Solution to Exercise 8.9

(a) Contract CH is the full-insurance contract that yields zero profits if bought only by
the H-types. Thus its premium is given by the solution to hH − pH ` = 0, that is,
hH = 14 (6, 000) = $1, 500. Contract CL is given by the intersection of the H-type
indifference curve through CH and the low-risk zero-profit line; thus it is given by
the solution to
1 3
ln(10, 000 − 1, 500) = ln(10, 000 − h − d) + ln(10, 000 − h)
4 4
1
h− (6, 000 − d) = 0
16
which is CL = (hL = 90.98, dL = 4544.31).
(b) The average probability of loss is
   
3 1 13 1 25
p̄ = qH pH + (1 − qH ) pL = + = .
16 4 16 16 256

(c) We need to express the fact that there is a contract at which the average zero-profit
line intersects the L-type indifference curve through contract
CL = (hL = 90.98, dL = 4544.31):

25
h− (6, 000 − d) = 0
256
1 15
ln(10, 000 − 90.98 − 4, 544.31) + ln(10, 000 − 90.98)
16 16
1 15
= ln(10, 000 − h − d) + ln(10, 000 − h).
16 16

IV
Asymmetric Information:
Signaling

9 Signaling . . . . . . . . . . . . . . . . . . . . . . . . . 297
9.1 Earnings and education
9.2 Signaling in the job market
9.3 Indices versus signals
9.4 More than two types
9.5 A more general analysis
9.6 Signaling in other markets
9.7 Exercises
9.8 Solutions to Exercises
 9. Signaling

9.1 Earnings and education

It is well-known that education can be viewed as an investment, since - typically - higher
levels of education are associated with higher earnings. Table 9.1 shows the relationship
between level of education and median annual income for the year 2018.1

Table 9.1: Median annual earnings by level of education, 2018.

Educational attainment Median

annual earnings
Less than a high school diploma $28,756
High school diploma $37,960
Some college, no degree $41,704
Associate’s degree $44,824
Bachelor’s degree $62,296
Master’s degree $74,568
Doctoral degree $94,900
Professional degree $97,968

Thus, for example, in 2018 a worker with a Bachelor’s degree had an annual income
1 Thedata in Table 9.1 refers to persons of age 25 and over, who are full-time wage and salary workers.
Source: Current Population Survey, U.S. Department of Labor, U.S. Bureau of Labor Statistics, https:
//www.bls.gov/emp/tables/unemployment-earnings-education.htm . We have converted the data
from weekly to annual figures.
298 Chapter 9. Signaling

that was 64% higher than the income of a worker with only a High-school diploma, and
a Master’s degree translated into a further 20% increase in income. Given the positive
correlation between level of education and earnings, it is natural to ask the following
questions:
1. Why do some people choose lower levels of education than others?
2. Why do employers reward higher levels of education with higher salaries?
3. Does the availability of educational institutions make society better off?
The answer to the first question is that, while education offers clear benefits in terms of
earnings, it also involves direct and opportunity costs. The opportunity cost of pursuing
a college degree, for example, is the income that one foregoes during the four years of
college. The direct cost of a college degree is the amount of money that one spends during
those four years in tuition, books, etc., as well as the subsequent interest cost on loans
taken during the college years. For some individuals those costs are lower than for others:
for example, children of more affluent parents do not need to borrow in order to pay tuition
costs or are able to obtain zero-interest loans from family members. Furthermore, there are
also differential psychological costs associated with college life: some find studying more
difficult and more burdensome than others. Thus, for some, a cost-benefit calculation leads
to the decision to acquire a lower level of education, while others pursue higher levels of
education.
The answer to the second question seems obvious too: employers pay more educated
workers more, because those workers have become more productive and more valuable
to the employer as a consequence of the additional education. In other words, employers
recognize that there is an objective, causal relationship between education and productivity:
the more education you acquire, the more productive you become. But is there truly
such a causal relationship between education and productivity? Is it at all possible that
employers are objectively wrong in postulating such a relationship between education and
productivity? What if a person’s productivity was determined at birth by his/her genetic
makeup and education had no effect at all on productivity? Could this be possible?
Before the 1974 seminal contribution of Michael Spence,2 an economist would have
denied such a possibility by arguing as follows. It is certainly possible that employers
might start with some incorrect hypothesis about the relationship between education and
productivity, but then their experience would reveal that, sometimes, highly educated
employees turn out to be less productive than, or as productive as, less educated employees.
The employers would then adjust their beliefs, so that, with sufficient time and experience,
they would discover that their initial hypothesis was wrong and would therefore no longer
automatically offer higher salaries to more educated job applicants. Spence’s contribution
was to show the possibility self-confirming beliefs: the employers’ (objectively wrong)
beliefs induce them to make education-dependent wage offers, which - in turn - induce
prospective employees to make choices of education levels which then confirm the em-
ployers’ beliefs, thus generating an equilibrium where nobody has any reason to change
his/her actions and beliefs. Such a situation is called a signaling equilibrium and is the
object of the next section.

2 MichaelSpence, Market signaling: Informational transfer in hiring and related screening processes,
Harvard University Press, 1974.
 9.2 Signaling in the job market 299

9.2 Signaling in the job market

9.2.1 Signaling equilibrium
Let us begin with a simple example. In Table 9.2 we measure the possible levels of
education of a job applicant in terms of the number of years spent in school, denoted by y,
and associated educational certificate:

Table 9.2: Possible levels of education and corresponding certificates.

Years Corresponding
of schooling education certificate
y=6 Elementary school
y = 12 High school diploma
y = 16 Bachelor’s degree
y = 18 Master’s degree
y = 21 Ph.D. degree

Suppose that employers believe that education increases productivity, so that the more
educated a candidate is, the more valuable he/she is to the employer. These beliefs are
manifested through a wage schedule that associates higher salaries with higher levels of
education, as shown in the following table:

Table 9.3: Offered wage as a function of education.

Education Wage offered

y=6 $6,000
y = 12 $20,000
y = 16 $25,000
y = 18 $30,000
y = 21 $32,000

Thus, for example, a job candidate who only has a High-school diploma will be offered
a salary of $20,000, while a job candidate with a Bachelor’s degree will be offered a higher
salary, namely $25,000.
Now the reader is asked to imagine a world where education has no effect on a person’s
productivity. In particular, in this hypothetical world there are only two types of individuals:
Type H whose productivity is worth $30,000 to potential employers (and is unaffected by
education), and type L whose productivity is worth $20,000 to potential employers (and is
also unaffected by education). The proportion of individuals of type H in the population
is pH , with 0 < pH < 1, so that the proportion of type L is (1 − pH ). We assume that
each individual knows her own type, while a potential employer is unable to determine
the type of a job applicant. Thus we are in a situation of asymmetric information. One
300 Chapter 9. Signaling

last assumption is that the two types have different monetary costs of acquiring education,
which are as follows:
(
0 if y ≤ 6
For type H : CH (y) =
1, 000(y − 6) if y > 6

(
0 if y ≤ 6
For type L : CL (y) =
2, 000(y − 6) if y > 6

Thus every year of schooling beyond 6th grade costs $1,000 to an H-type and $2,000 to an
L-type.
Each type will choose that level of education that yields the largest net income. The
cost-benefit analysis for a type-H person is shown in Table 9.4.

Table 9.4: The cost-benefit analysis of a type-H person.

y w CH w −CH
years in school gross wage cost net wage
y=6 $6,000 $0 $6,000
y = 12 $20,000 $6,000 $14,000
y = 16 $25,000 $10,000 $15,000
y = 18 $30,000 $12,000 $18, 000
y = 21 $32,000 $15,000 $17,000

Thus an H-type will maximize her net income by pursuing a Master’s degree (corre-
sponding to 18 years of schooling).
On the other hand, a type-L person faces the cost-benefit calculations shown in Table
9.5.

Table 9.5: The cost-benefit analysis of a type-L person.

y w CL w −CL
years in school gross wage cost net wage
y=6 $6,000 $0 $6,000
y = 12 $20,000 $12,000 $8, 000
y = 16 $25,000 $20,000 $5,000
y = 18 $30,000 $24,000 $6,000
y = 21 $32,000 $30,000 $2,000
 9.2 Signaling in the job market 301

Hence a type-L person will maximize his net income by obtaining a High-school
diploma (corresponding to 12 years of schooling).
In this hypothetical world type-L individuals, whose productivity is $20,000, will
present themselves at a job interview with a High-school diploma and, according to the
employers’ wage schedule (Table 9.3), they will be offered a wage of $20,000 which
matches their true productivity. On the other hand, type-H individuals, whose productivity
is $30,000, will present themselves at a job interview with a Master’s degree and, according
to the employers’ wage schedule, they will be offered a wage of $30,000 which matches
their true productivity. After hiring candidates according to the wage schedule shown in
Table 9.3 - which reflects their initial beliefs that more education causes higher productivity
- employers will observe that, indeed, those with a Master’s degree turn out to be more
productive than those who only have a High-school diploma, thus confirming their initial
(wrong) beliefs; furthermore, every employee is paid an amount that matches her/his true
productivity. Hence employers have no reason to change their beliefs and their hiring
practices. Since the wage schedule remains the same, both types have no reason to change
their education choices. Thus we have a situation of equilibrium. Following Spence, we
call such a situation a signaling equilibrium.

9.2.2 Pareto inefficiency

In the previous section we asked the question: “Does the availability of educational
institutions make society better off?”. Let us continue to analyse the hypothetical world
described in the previous section and ask whether it is possible that, were the government
to ban all institutions of higher education, everybody would be at least as well off as before
the government intervention and some individuals would be better off. That is, we ask
whether closing down educational institutions beyond High school would lead to a Pareto
improvement.
Suppose that, in the new world, every individual were required to obtain a High-
school diploma, but no further education were available. Furthermore, suppose that
employers were instructed to pay each job applicant a wage equal to the average wage
that they were paying in the pre-intervention world.3 Let N be the number of workers
hired by an employer. Before the government intervention, since the fraction pH of the
population is of type H, the employer would have employed pH N people of type H at
a salary of $30,000 (those who had a Master’s degree) and (1 − pH )N people of type
L at a salary of $20,000 (those who had a High-school diploma), for a total wage bill
of pH N30, 000 + (1 − pH )N20, 000 = [pH 30, 000 + (1 − pH )20, 000]N and thus a wage
of [pH 30, 000 + (1 − pH )20, 000] per worker. If the employer were required to pay this
amount to each job applicant in the post-intervention world, then the employer would be
as well off as before.
In the post-intervention world each type-L worker would be better off, since [pH 30, 000+
(1 − pH )20, 000] > 20, 000, that is, for these individuals, the new-world wage is higher
than the old-world wage (and the education level, and associated cost, are the same). What
about the type-H workers? A type-H individual will be better off if her net income is
higher; in the old world her net income was $18,000 (equal to a wage of $30,000 minus
the cost of acquiring a Master’s degree, which is $12,000), while in the new world her net
3 This could be the initial wage, which can then be adjusted, up or down, later when the employer
discovers the true productivity of each employee.
 302 Chapter 9. Signaling

income is [pH 30, 000 + (1 − pH )20, 000] − 6, 000 (where $6,000 is her cost of obtaining a
High-school diploma). Thus, a type-H individual is made better off by the government
intervention if and only if
[pH 30, 000 + (1 − pH )20, 000] − 6, 000 > 18, 000,
that is, if and only if pH > 52 . That is, if type-H individuals constitute at least 40% of the
population, then eliminating all the institutions of higher education would lead to a Pareto
improvement!4

9.2.3 Alternative interpretation of a signaling equilibrium

In Section 9.2.1 we interpreted a signaling equilibrium as a situation where the employers’
objectively wrong beliefs become self confirming: employers manifest their beliefs in
education-dependent wage offers, which induce different types of prospective employees to
make different choices of education, which in turn leads to employers getting information
that confirms their initial beliefs, so that the situation is self-sustaining and there is no
reason for anybody to change his/her beliefs and choices.
An alternative interpretation of a signaling equilibrium is not based on the assumption
that employers hold wrong beliefs. Employers may be fully aware that education has no
effect on productivity, but they are also aware of the fact that in the population there are
different types of individuals with different productivity. As in Section 9.2.1 suppose that
there are only two types of individuals: type H with high productivity and type L with low
productivity; furthermore, employers cannot tell individuals apart, while each prospective
employee knows her own type. Then high-productivity individuals face the problem of
how to credibly convey to employers that they are of type H, in order to be paid a higher
wage. Simply claiming to be a high-productivity type would not work, because every
individual would make such a claim and thus it would not be credible. What is needed is a
costly signal that is too costly for a type-L individual to use but not too costly for a type-H
individual, so that only the latter avail themselves of it. Such a signal would allow the
better types to separate themselves from the worse types. A signaling equilibrium is then a
separating equilibrium in which the H types send the costly signal, while the L types do
not and thus employers can distinguish between job applicants on the basis of the presence
or absence of such a signal. In the job market education can be such a signal.
We shall now give a more precise definition of a signaling equilibrium in the more
general case where education can potentially affect productivity. For the moment we
continue to assume that there are only two types in the population: type H and type L,
with proportions pH and (1 − pH ), respectively. Suppose that productivity is a (possibly
constant) function of the amount of education y, as follows:
For type H : a + sH y (9.1)
For type L : b + sL y (9.2)
with 0 ≤ b ≤ a, 0 ≤ sL ≤ sH and either b < a or sL < sH (or both). Thus, if b < a and
sL = sH = 0, then we are in the world considered in Section 9.2.1 where education has
4 Ifalso High schools were to be eliminated, then every worker would be even better off, since workers
would now choose y = 6 and a type-L individual would save $12,000 in education costs, while a type-H
individual would save an additional $6,000 relative to obtaining a High-school diploma . Indeed, in this case
for a Pareto improvement it is sufficient to have pH 30, 000 + (1 − pH )20, 000 > 18, 000, that is, pH > 15 .
9.2 Signaling in the job market 303

no effect on productivity, while if sH > sL then more education translates into higher
productivity (for both types if sL > 0 or only for the H type if sL = 0).5
As in Section 9.2.1, the additional crucial assumption is that education is more costly
for individuals of type L than for individuals of type H. Let the monetary cost of education
be:
For type H : cH y (9.3)

For type L : cL y (9.4)

with 0 < cH < cL , cL > sL and cH > sH ; the last two inequalities mean that, for each type,
the cost of an extra unit of education exceeds the benefit in terms of the extra productivity
gained.
We assume that employers know the relationship between education and productivity
given by (9.1) and (9.2) and thus are willing to offer higher wages to more educated job
applicants. However, if a job applicant has level of education ŷ, then the employer would
not know whether to offer a wage equal to a + sH ŷ or a wage equal to b + sL ŷ, because the
employer does not know if the applicant is of type H or of type L. Suppose, therefore, that
employers choose an arbitrary threshold level of education y∗ and announce that anybody
who applies with education level y ≥ y∗ will be presumed to be of type H and offered a
wage equal to a + sH y and anybody with education level y < y∗ will be presumed to be of
type L and offered a wage equal to b + sL y.
Definition 9.2.1 A separating signaling equilibrium is a triple (y∗ , yH , yL ) where y∗ is
the threshold level set by employers, yH is the education level chosen by every individual
of type H and yL is the education level chosen by every individual of type L, such that:
1. yH ≥ y∗ > yL , so that a type H is paid a wage equal to a + sH yH and a type L is
paid a wage equal to b + sL yL ,
2. the net income of a type H is maximized when she chooses education level yH ,
3. the net income of a type L is maximized when he chooses education level yL .

The term ‘separating’ means that the two types make difference choices of education.

R Note that, since cL > sL , at a signaling equilibrium it must be that yL = 0, because

if yL > 0 then the net income of a type L is b + sL yL − cL yL = b + (sL − cL )yL < b,
violating Point 3 of Definition 9.2.1: a type L would get a higher net income, namely
b, if he were to choose y = 0.
Similarly, since cH > sH , at a signaling equilibrium it must be that yH = y∗ , because
if yH > y∗ then the net income of a type H is a + sH yH − cH yH = a + (sH − cH )yH <
a + (sH − cH )y∗ , violating Point 2 of Definition 9.2.1: a type H would get a higher
net income, namely a + (sH − cH )y∗ , is she were to switch from yH to y∗ .
The same argument shows that every individual - whatever her type - will only
consider choosing either y = 0 or y = y∗ : any choice strictly between 0 and y∗ is
worse than 0 and any choice greater than y∗ is worse than y∗ .

5 The initial example considered by Spence is one where a = 2, b = 1 and sL = sH = 0.

304 Chapter 9. Signaling

Thus (y∗ , yH , yL ) is a separating signaling equilibrium (given the employers’ wage

schedule, according to which a candidate with y ≥ y∗ is classified as type H and everybody
else is classified as type L) if and only if:6

1.
a + sH y∗ − cH y∗ ≥ b (9.5)

(this inequality says that a type H prefers to choose y = y∗ and be classified correctly
as type H, rather than choosing y = 0 and be classified incorrectly as type L, where
‘prefers’ means ‘obtains a higher net income’),

2.
b ≥ a + sH y∗ − cL y∗ (9.6)

(this inequality says that a type L prefers to choose y = 0 and be classified correctly
as type L, rather than choosing y = y∗ and be classified incorrectly as a type H, in
which case he would be paid a wage equal to a + sH y∗ ).

It should be noted that there may be a multiplicity of separating signaling equilibria (indeed,
even a continuum). For example, let us consider the case where a = 2, b = 1, sH = sL = 0,
cH = 12 and cH = 1 (thus, in this case, productivity is not affected by education). Then
inequalities (9.5) and (9.6) become:

2 − 12 y∗ ≥ 1

1 ≥ 2 − y∗

which are equivalent to 1 ≤ y∗ ≤ 2. Thus, for every y∗ ∈ [1, 2], the triple (y∗ , yH = y∗ , yL =
0) is a separating signaling equilibrium. These equilibria are Pareto ranked: the lower the
value of y∗ the better off individuals of type H are (since their net income is 2 − 12 y∗ ); on
the other hand, the L types are indifferent among all the equilibria, since their net income is
always 1 and the employers are also indifferent because their total wage bill is the same at
each of those equilibria, namely 2pH N + (1 − pH )N (where, as before, pH is the proportion
of type H in the population and N is the number of people employed), which, in turn is
equal to total productivity.
To reiterate the point made in Section 9.2.2, we show that also in this example it
is possible for a signaling equilibrium to be Pareto inefficient. Consider the alternative
situation where
(1) the possibility of signaling through education is removed, so that every individual is
forced to choose y = 0 and
(2) everybody is paid a wage equal to the average productivity of the population, which is
6 Whether we use weak or strict inequalities depends on what we assume in case an individual is indifferent.

Since we use weak inequalities, our implicit assumption is that, when indifferent between y = 0 and y = y∗ ,
a type H would choose y = y∗ while a type L would choose y = 0.
 9.3 Indices versus signals 305

2pH + (1 − pH ) = 1 + pH .
Then employers are as well off in the new situation and type L individuals are better off
(their wage is now 1 + pH whereas in the previous situation it was 1). Type H individuals
are better off in the new situation if and only if 1 + pH > 2 − 12 y∗ , which is possible: for
example if y∗ = 1.5 and pH > 14 .

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 9.7.1 at the end of this chapter.

9.3 Indices versus signals

A signal is an observable characteristic that can be modified by the agents concerned and
is available to every type of individual. Examples of signals are: education, hair style, the
way one dresses, whether one wears make-up or not, etc. If employers indicate that they
take a particular signal as a sign of higher productivity - by offering a higher salary to
those who display that signal - then in principle every individual can choose that signal and
qualify for a higher wage. For example, if employers believed that people with short hair
were more productive, then everybody could choose to cut their hair short and obtain a
higher wage. The reason why some people do choose a particular signal (e.g. a sufficiently
high level of education) and others do not is that the former face a lower cost of acquiring
the signal.
There are also other characteristics that cannot be modified (or are prohibitively
expensive to modify) by the individual, such as race, gender, height, skin color, etc. Any
such characteristic is called an index (the plural is ‘indices’). If employers believed that
taller people were more productive, and thus offered a higher salary to those whose height
was above a certain threshold, then there would be nothing that short people could do to
qualify for the higher salary.
At the beginning of the previous section we offered an interpretation of a signaling
equilibrium in the job market in terms of self-confirming - although objectively wrong
- beliefs (in that context the wrong beliefs were about the causal relationship between
education and productivity). It seems that this phenomenon cannot happen when the wrong
beliefs involve an index rather than a signal. To continue the example of height, if there
is no objective correlation between height and productivity and yet employers believed
that taller people were more productive, then initially they would offer taller people higher
salaries, but over time they would observe that shorter employees, on average, were as
productive as taller employees and the evidence would force them to change their beliefs.
Spence, however, showed that this intuition is wrong: the possibility of objectively
wrong beliefs concerning an index can be self-confirming if the index is combined with a
signal.
We will show this by continuing the example analyzed in Section 9.2.1. We consider
a world where education has no effect on a person’s productivity. In particular, in this
hypothetical world there are only two types of individuals: type H whose productivity is
worth $30,000 to potential employers (and is unaffected by education), and type L whose
306 Chapter 9. Signaling

productivity is worth $20,000 to potential employers (and is also unaffected by education).

The proportion of individuals of type H in the population is pH , with 0 < pH < 1 and
the proportion of type L is thus (1 − pH ). We also assume that, within each group, men
and women are equinumerous, that is, 50% of type H are men and 50% are women and,
similarly, 50% of type L are men and 50% are women. We also assume that employers
hold beliefs that are doubly wrong: they believe that
1. education increases productivity, and
2. women are less able to learn and thus require more education than their male
counterpart in order to achieve a higher level of productivity.
Employers manifest their beliefs by offering different wage schedules to men and women:
there is no difference up to High-school diploma, but then certificates of higher education
are remunerated differently across men and women. The wage schedules are as shown in
Table 9.6 (note that, for men, the wage schedule coincides with Table 9.3 in Section 9.2.1).

Table 9.6: Offered wage as a function of education and gender.

Education Wage offered Wage offered

to men to women
Elementary school (y = 6) $6,000 $6,000
High-school diploma (y = 12) $20,000 $20,000
College degree (y = 16) $25,000 $23,000
Master’s degree (y = 18) $30,000 $26,000
Ph.D. degree (y = 21) $32,000 $30,000

Thus employers’ beliefs embody a wrong view of the effect of education on productivity
(recall that productivity is assumed to be constant and thus independent of the amount of
education) as well as a prejudice against women (women are thought to be innately less
able to convert education into productivity).
We continue to assume, as in Section 9.2.1, that the two types of individuals have
different monetary costs of acquiring education:

 0 if y ≤ 6
For type H (man or woman): CH (y) =
 1, 000(y − 6) if y > 6


 0 if y ≤ 6
For type L (man or woman): CL (y) =
 2, 000(y − 6) if y > 6

Thus every year of schooling beyond 6th grade costs $1,000 to an H-type, whether it is a
man or a woman, and $2,000 to an L-type, whether it is a man or a woman.
9.3 Indices versus signals 307

The cost-benefit analysis for a man of type L is the same as in Table 9.5 in Section
9.2.1, with the conclusion that a man of type L would choose y = 12.
A woman of type-L faces the cost-benefit calculations shown in Table 9.7.

Table 9.7: The cost-benefit analysis of a type-L woman.

y w CL w −CL
years in school gross wage cost net wage
y=6 $6,000 $0 $6,000
y = 12 $20,000 $12,000 $8, 000
y = 16 $23,000 $20,000 $3,000
y = 18 $26,000 $24,000 $2,000
y = 21 $30,000 $30,000 $0

Thus also a woman of type L would choose y = 12.

The cost-benefit analysis for a man of type H is the same as in Table 9.4 in Section
9.2.1, with the conclusion that a man of type H would choose y = 18.
Finally, for a woman of type H the cost-benefit calculations are shown in Table 9.8.

Table 9.8: The cost-benefit analysis of a type-H woman.

y w CL w −CL
years in school gross wage cost net wage
y=6 $6,000 $0 $6,000
y = 12 $20,000 $6,000 $14,000
y = 16 $23,000 $10,000 $13,000
y = 18 $26,000 $12,000 $14,000
y = 21 $30,000 $15,000 $15, 000

Thus a type-H woman would choose y = 21.

Hence we have a separating equilibrium where all type-L individuals (men and women)
choose to stop at a High-school diploma and are hired at a salary of $20,000, which
corresponds to their true productivity; type-H men choose to obtain a Master’s de-
gree and are hired at a salary of $30,000, which corresponds to their true productivity;
type-H women choose to obtain a Ph.D. degree and are hired at a salary of $30,000,
which corresponds to their true productivity. At this equilibrium employers never observe
308 Chapter 9. Signaling

men and women with the same graduate degree: if they did, they would realize that their
beliefs concerning women’s abilities were wrong and would revise those beliefs, thereby
abandoning their prejudices. What they do observe is that those whose productivity is
$30,000 are either men with 18 years of schooling or women with 21 years of schooling
and would thus find confirmation of their beliefs that “women need to spend more time in
school in order to achieve the higher level of productivity”. In other words, the employers’
doubly wrong beliefs are self-confirming at a separating equilibrium.
To sum up, the question that the presence of indices gives rise to is the following: can the
informational structure of the market bring about persistent and consistent discrimination
between objectively identical individuals? We considered the case where the unchangeable
attribute (index) is gender. We supposed that within each gender there were both low-
productivity and high-productivity individuals. Signaling costs (in our example, costs of
acquiring education) were different for low-productivity and high-productivity people, but
individuals with the same level of productivity had the same signaling costs, no matter
whether they were men or women. We also assumed that people (both men and women)
with the same productivity had the same preferences and the same objective: to maximize
their income net of education costs. Therefore the index (gender) should be absolutely
irrelevant: in principle, people with the same opportunities and the same preferences ought
to make similar decisions and end up in similar situations. The informational structure
of the market, however, can destroy this principle. If employers believe that gender
(besides education) is correlated with productivity, they might offer a wage schedule that
is differentiated on the basis of education and gender, thus presenting otherwise identical
individuals with different opportunity sets. As a consequence, the employer’s beliefs may
force high-productivity women to invest in education more than their male counterpart
(that is, more than high-productivity men). The reason why this situation can persist is
that employers will interpret the incoming data separately for the two groups of men and
women. If different levels of education were associated with the same observed level of
productivity within the same group, employers would be forced to revise their beliefs.
For example, if within the group of men different levels of education were accompanied
by the same observed productivity, then employers would conclude that, at least above
a certain level, education no longer increases productivity (at least for men). But since
men and women are judged separately and independently (data on men is not used to
classify women and vice versa), employers can consistently think that women need to
acquire more education than their male counterpart in order to compensate for a “genetic”
handicap. The employers’ beliefs induce women to invest more in education than men,
thereby confirming the prejudice that women require more education in order to achieve
the same level of productivity as men (despite the fact that those beliefs have no objective
grounds: they constitute, indeed, a prejudice). As a consequence, women end up being
over-qualified for their jobs as compared to men.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 9.7.2 at the end of this chapter.
 9.4 More than two types 309

9.4 More than two types

In the previous sections we restricted attention to the case where there are only two types
of individuals. However, the reasoning developed in those sections applies also to the case
of more than two types. Let us consider a simple example with three types.
Suppose that education has no effect on productivity and that the are three types of
individuals: Type 1 with productivity 4, Type 2 with productivity 5 and Type 3 with
productivity 6. As in the previous sections, the crucial assumption is that the cost of
education is inversely related to productivity: the higher the (innate) productivity of an
individual, the lower the cost of education. Let us suppose that the cost of education is as
follows:

Type Productivity Cost of education

1 4 C1 (y) = 3y
2 5 C2 (y) = 2y
3 6 C3 (y) = y

Suppose that employers set two threshold levels of education, y∗1 and y∗2 , and announce that
wage offers will depend on the level of education y of the applicant, as follows:

Level of eduction y Wage offered

y < y∗1 4
y∗1 ≤ y < y∗2 5
y ≥ y∗2 6

That is, applicants with y < y∗1 are presumed to be of type 1, applicants with y∗1 ≤ y < y∗2
are presumed to be of type 2 and applicants with y ≥ y∗2 are presumed to be of type 3.
For what values of y∗1 and y∗2 do we have a separating signaling equilibrium?
Since - except for the threshold values - increasing y increases the cost of education
but leaves the wage constant, each individual will only consider choosing either y = 0 or
y = y∗1 or y = y∗2 . At a separating equilibrium, type-1 individuals must choose y = 0 in
order to be paid a wage equal to their true productivity, type-2 individuals must choose
y = y∗1 and type-3 individuals must choose y = y∗2 ; furthermore, it must be in their interest
to do so. Thus we need two inequalities to be satisfied for each type.
Let us begin with Type 1. The following table shows the net wage for a type-1 individual
for each of the three choices of education.

Type 1
choice wage cost net wage
y=0 4 0 4
y = y∗1 5 3y∗1 5 − 3y∗1
y = y∗2 6 3y∗2 6 − 3y∗2
310 Chapter 9. Signaling

• In order for y = 0 to be better than7 y = y∗1 we need

4 ≥ 5 − 3y∗1 , that is, y∗1 ≥ 31 . (9.7)

• In order for y = 0 to be better than y = y∗2 we need

4 ≥ 6 − 3y∗2 , that is, y∗2 ≥ 32 . (9.8)

The following table shows the net wage for Type 2 for each of the three choices of
education.

Type 2
choice wage cost net wage
y=0 4 0 4
y = y∗1 5 2y∗1 5 − 2y∗1
y = y∗2 6 2y∗2 6 − 2y∗2

• In order for y = y∗1 to be better than y = 0 we need

5 − 2y∗1 ≥ 4, that is, y∗1 ≤ 12 . (9.9)

• In order for y = y∗1 to be better than y = y∗2 we need

5 − 2y∗1 ≥ 6 − 2y∗2 , that is, y∗2 ≥ 21 + y∗1 . (9.10)

The following table shows the net wage for Type 3 for each of the three choices of
education.

Type 3
choice wage cost net wage
y=0 4 0 4
y = y∗1 5 y∗1 5 − y∗1
y = y∗2 6 y∗2 6 − y∗2

• In order for y = y∗2 to be better than y = 0 we need

6 − y∗2 ≥ 4 that is, y∗2 ≤ 2. (9.11)

• In order for y = y∗2 to be better than y = y∗1 we need

6 − y∗2 ≥ 5 − y∗1 that is, y∗2 ≤ 1 + y∗1 . (9.12)

7 Asin the previous sections, we assume that, if indifferent between two levels of education, a Type 1
would choose the lower level.
 9.5 A more general analysis 311

From (9.7) and (9.9) we get that

1
3 ≤ y∗1 ≤ 12 . (9.13)
It follows from (9.13) that
(1) 1 + y∗1 < 2 so that if (9.12) is satisfied then (9.11) is also satisfied, and
(2) 12 + y∗1 > 32 so that if (9.10) is satisfied then (9.8) is also satisfied.
Thus from (9.10) and (9.12) we get that
1
2 + y∗1 ≤ y∗2 ≤ 1 + y∗1 . (9.14)
Whenever inequalities (9.13) and (9.14) are satisfied we have a separating signaling
equilibrium. For example, both inequalities are satisfied if y∗1 = 12
5
and y∗2 = 1.
In general, if there are n types then, for each type, (n − 1) inequalities need to be
satisfied, for a total of n(n − 1) inequalities. In Exercise 9.9 the reader is asked to find the
relevant inequalities for the case where n = 4.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 9.7.3 at the end of this chapter.

9.5 A more general analysis

In the previous four sections we assumed that individuals were facing a monetary cost of
education and that their objective was to choose that level of education that maximized
their net income. A crucial assumption was that lower-productivity people faced higher
monetary costs of acquiring education. Although we indicated some reasons why this
might be the case, it would be preferable to carry out the analysis assuming that, while the
monetary cost of education is essentially the same for everybody, it is the psychological
cost of education that differs across different types of individuals. In other words, different
types have different preferences over pairs of income and education levels.
We assume that every individual has a utility function U(m, y) over income, denoted
by m, and education, denoted by y. Individuals prefer more money to less, that is, for
every level of education y and any two levels of income m1 and m2 , with m1 < m2 ,
U(m2 , y) > U(m1 , y). In other words, utility is an increasing function of income; this can
be expressed by saying that the partial derivative of U with respect to m is positive:
∂U
for all (m, y), (m, y) > 0. (9.15)
∂m
The second assumption is that acquiring education involves effort and thus a psychological
cost, so that - at any level of income m - the individual prefers less education to more edu-
cation, that is, if y1 < y2 then U(m, y1 ) > U(m, y2 ). In other words, utility is a decreasing
function of education; this can be expressed by saying that the partial derivative of U with
respect to y is negative:
∂U
for all (m, y), (m, y) < 0. (9.16)
∂y
312 Chapter 9. Signaling

It follows from (9.15) and (9.16) that indifference curves in the (y, m) plane must be
increasing: starting from a point (ŷ, m̂) if another point (y, m) with y > ŷ is to give the same
utility as (ŷ, m̂) (that is, is to be on the indifference curve that goes through (ŷ, m̂)) it must
be that m > m̂. We shall consider the case where indifference curves are convex, as shown
in Figure9.1.8

income
m
direction of
increasing
utility indifference
curve

y
education
0

Figure 9.1: An indifference curve in the (y, m) plane.

Indifference curves can be used to obtain the monetary equivalent of the psychological
cost (or “disutility”) associated with an increase in the amount of education. Starting from
an education-income pair (ŷ, m̂) (which lies on the indifference curve corresponding to
some level of utility ū), if education is increased by one unit and income is kept constant,
then utility goes down, that is, the individual is worse off. One can thus ask: How much
extra income should the individual be given to compensate for the reduction in utility
due to the increase in y? In other words, what is the amount by which income has to be
increased in order to keep the individual on the same indifference curve?

8 For √
example, if U(y, m) = m − y then the equation of the indifference curve corresponding to utility
2
level ū is m = (ū + y) , which is increasing and convex in y.
9.5 A more general analysis 313

This is shown in Figure 9.2: starting from the education-income pair (ŷ, m̂), if education
is increased by one unit to ŷ +1 then the individual requires an additional amount of income
equal to (m1 − m̂) in order to remain on the same indifference curve; if education is then
increased by one more unit, from ŷ + 1 to ŷ + 2, then the individual requires an additional
income compensation equal to (m2 − m1 ). Note that, since the indifference curve is
convex, (m2 − m1 ) > (m1 − m̂), that is, every additional unit of education requires a larger
compensation in terms of income.

income
m

indifference
curve

m2
m1
m̂
y
0 ŷ ŷ + 1 ŷ + 2 education

Figure 9.2: The monetary equivalent of the psychological cost of education.

Recall that the crucial assumption in the analysis of the previous sections was that
higher-productivity people had a lower monetary cost of education. We can now express
this difference in education costs in terms of preferences, as follows. Let us focus on
the simple case where there are only two types of individuals: Type H and Type L; the
latter are the less productive ones. We want to capture, using indifference curves, the fact
that education is more “psychologically costly” for type-L individuals than for type-H
individuals: for any given increase in education, the income compensation required by an
L-type is larger than the income compensation required by an H-type. This will be the case
if and only if the indifference of an L-type that goes through a given education-income
pair (ŷ, m̂) is steeper than the indifference curve of an H-type through (ŷ, m̂).
314 Chapter 9. Signaling

This is shown in Figure 9.3: at income level m̂, increasing education from ŷ to ŷ + 1
makes every type worse off; in order to restore a type-L individual to the original utility
level (that is, in order to keep him on the L-type indifference curve through (ŷ, m̂)) an
increase in income equal to (mL − m̂) is required, while for a type-H individual the required
increase in income is (mH − m̂). As shown in Figure 9.3, (mL − m̂) > (mH − m̂).

income
m

Type L
Type H

mL
mH
m̂
y
0 ŷ ŷ + 1 education

Figure 9.3: The monetary compensation for an extra unit of education required by an
L-type is larger than that required by an H-type.

Let us assume that education does increase productivity, but at different rates for the
two types. Specifically, we assume that productivity, denoted by Π, is as follows, with
0 < kL < kH :

Type L : ΠL (y) = kL y
Type H : ΠH (y) = kH y.
9.5 A more general analysis 315

The productivity functions for the two types are shown in Figure 9.4.

productivity
Π

Type H

Type L

y
0 education

Figure 9.4: Productivity as a function of education.

What education levels would the two types choose in a perfect-information world in
which employers were able to determine the type of each job applicant? The answer is
illustrated in Figure 9.5.

productivity
Π Type H

Type H

Type L

Type L

y
0 yL yH education

Figure 9.5: The education choices of the two types under perfect information.
316 Chapter 9. Signaling

In the perfect-information case (where employers can determine the type of a job
applicant and offer a wage equal to his/her productivity), the optimal choice of education
for a type is given by the horizontal coordinate of the point at which there is a tangency
between the productivity line of that type and an indifference curve of that type. Indeed, if
an individual chooses a level of education corresponding to a point at which the indifference
curve crosses the productivity line, such as point A in Figure 9.6 for type L, then there are
points to the left of that point (corresponding to lower levels of education) that lie on a
higher indifference curve (and thus are better for the individual).

productivity
Π

A Type L

y
0 yL ŷ education

Figure 9.6: An education level, such as ŷ, at which the indifference curve crosses the
wage/productivity line is not optimal (this is shown for a type-L individual).

We denote the perfect-information choice of education for type L by yL and the perfect-
information choice of education for type H by yH .

Let us now return to the case where employers are not able to determine the type of a
job applicant. We assume, as in the previous sections, that employers set a threshold level
of education, denoted by y∗ , and announce that anybody with a level of education less than
y∗ will be classified as a type-L individual and will be offered a wage equal to kL y (where y
is the level of education he/she has) and anybody with a level of education greater than or
equal y∗ will be classified as a type-H individual and will be offered a wage equal to kH y .
9.5 A more general analysis 317

This threshold, education-dependent, wage schedule is shown in Figure 9.7 as the

union of the two thick lines.

wage
offered

(Type H)

(Type L)
kH y∗

y
0 y∗ education

Figure 9.7: The wage offered by employers as a function of education, with a threshold
value y∗ .

We want to determine under what circumstances a separating signaling equilibrium

exists. Let us consider first the case where the threshold value y∗ is greater than the level
of education that would be chosen by an H-type under perfect information, that is, the case
where
y∗ > yH .

Then the indifference curve of an H-type that goes through the point (y∗ , kH y∗ ) crosses the
H-type productivity line at that point, so that points to the right of it are worse choices for
the H-type. Hence a type-H individual will only consider either education level y∗ (that
will have her classified correctly as type H) or a lower education level (that would have her
classified incorrectly as type L). Two cases are possible:
1. the indifference curve of the H-type that goes through the point (y∗ , kH y∗ ) lies
entirely above the segment of the wage schedule to the left of that point,
2. the indifference curve of the H type that goes through the point (y∗ , kH y∗ ) crosses
the segment of the wage schedule to the left of that point.
In Case 1, an H-type would choose level of education y∗ , while in Case 2 an H-type would
choose a level of education less than y∗ and thus be classified incorrectly as type-L, with
the consequence that there would be no separating signaling equilibrium.
318 Chapter 9. Signaling

Case 1 is shown in Figure 9.8 while Case 2 is shown in Figure 9.9.

productivity
Π
(Type H)

kH y∗

y
0 yH y∗ education

Figure 9.8: Case 1: the optimal choice of education for an H-type is y∗ .

productivity
Π
(Type H)

kH y∗

y
0 yH y∗ education

Figure 9.9: Case 2: the optimal choice of education for an H-type is less than y∗ .
9.5 A more general analysis 319

Thus, in order for a separating signaling equilibrium to exist, it is necessary that we are
in Case 1 above where the indifference curve of the H-type that goes through the point
(y∗ , kH y∗ ) lies entirely above the segment of the wage schedule to the left of that point.
This guarantees that the H-type individuals will choose level of education y∗ and will thus
be correctly classified as type-H and remunerated according to their true productivity.
The second requirement for the existence of a separating signaling equilibrium is that
the L-types choose a level of education less than y∗ (so that they will be correctly classified
as type-L and remunerated according to their true productivity). The relevant indifference
curve for the L-type is the one that goes through the point corresponding to the L-type
choice of education under perfect information, namely point (yL , kL yL ).9
One possibility, illustrated in Figure 9.10, is that this indifference curve of the L-type
cuts the second segment of the wage schedule. In this case the L-type would choose a
level of education greater than, or equal to, y∗ and thus there is no separating signaling
equilibrium.

productivity
Π

(Type L)

kL yL
y
0 yL y∗ education

Figure 9.10: The case where the optimal choice of education for an L-type is greater than,
or equal to, y∗ .

The other possibility is that the indifference curve of the L-type that goes through point
(yL , kL yL ) lies entirely above the second segment of the wage schedule, that is, the segment
corresponding to levels of education greater than or equal to y∗ . In this case the optimal
choice of education for the L-type is yL , with the consequence that type-L individuals are
correctly classified as type L and are paid according to their true productivity.

9 Note that, since we are looking at the case where yH < y∗ , and yL < yH , the point (yL , kL yL ) lies on the
first segment of the wage schedule, that is, the segment corresponding to choices of education less than y∗ .
320 Chapter 9. Signaling

The latter case is illustrated in Figure 9.11.

productivity
Π

(Type L)

y
0 yL y∗ education

Figure 9.11: The case where the optimal choice of education for an L-type is yL .

Putting together the cases illustrated in Figures 9.8 and 9.11 we obtain the conditions
that yield the existence of a separating signaling equilibrium:

1. the indifference curve of the H-type that goes through the point (y∗ , kH y∗ ) lies
entirely above the segment of the wage schedule to the left of that point (that is, the
segment from (0, 0) to (y∗ , kL y∗ ),

2. the indifference curve of the L-type that goes through point (yL , kL yL ) lies entirely to
the left of the second segment of the wage schedule (that is, the segment that starts
at the point (y∗ , kH y∗ )
9.5 A more general analysis 321

Figure 9.12 illustrates the case where a separating signaling equilibrium exists: all type-
H individuals choose level of education y∗ and are paid kH y∗ , while all type-L individuals
choose level of education yL and are paid kL yL .

productivity
Π

(Type H
(Type L)

y
0 yL y∗ education

Figure 9.12: The case where a separating signaling equilibrium exists.

Note that, since y∗ > yH , at the separating signaling equilibrium illustrated in Figure
9.12 there is over-investment in education: the H-types invest in education more than
they would in a perfect-information world. Thus the asymmetry of information between
employers and job applicants involves a cost for society that would not be incurred if there
were perfect information.

We considered the case where the threshold value y∗ is greater than the level of
education that would be chosen by an H-type under perfect information, that is, the case
where y∗ > yH . The remaining two cases, namely y∗ < yL and yL < y∗ < yH are examined
in Exercises 9.10 and 9.11.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 9.7.4 at the end of this chapter.
 322 Chapter 9. Signaling

9.6 Signaling in other markets

We conclude with a brief discussion of other markets in which signaling can take place.

9.6.1 Market for used cars

We saw in Chapter 7 that, due to adverse selection, it is possible that in the market for
second-hand cars some qualities (in the extreme case, all but the lowest quality) cannot
be traded, giving rise to an inefficiency if potential buyers value those cars more than the
owners. Thus the owner of a good-quality car gets “stuck” with her car, unable to convince
a potential buyer that her car is of good quality and thus justifies a higher price than the
low market price. Merely claiming that her car is of good quality is “cheap talk” which
can be imitated by the owners of cars of inferior quality. Credible communication with
potential buyers would require a costly signal which other sellers would not find profitable
to imitate. What could constitute such a signal? The answer is: offering a warranty to
cover future repairs.
We shall not develop a formal analysis of this, since - as explained below - this is more
of a theoretical possibility than a practical one. However, a formal example is developed in
detail in Exercise 9.12, which the reader is urged to attempt.
Presumably, in the case of cars ‘high quality’ means ‘unlikely to break down and
require an expensive repair’ and ‘low quality’ means ‘likely to require an expensive repair’.
Indeed, the buyer of a used car will typically worry that she will end up with a “lemon” into
which she will have to sink a lot of money for maintenance. The owner of a good-quality
car knows that her car is unlikely to require major repairs and would in principle be willing
to offer a warranty against such future repairs, that is, she will be willing to contractually
commit herself to paying for the cost of any such repairs (within a specified period of
time). This commitment will end up involving a low expected cost for her, because of the
low probability of a major breakdown (since the car is, after all, a good-quality car). On
the other hand, the owner of a low-quality car would not be willing to undertake such a
commitment, because she recognizes that the probability of her having to pay for a major
repair in the future is, in fact, quite high. Thus offering to sell with warranty can indeed
work as a credible signal of high quality and induce buyers to correspondingly pay a higher
price for cars sold with warranty.
However, from a practical point of view there are several factors that work against this
form of signaling. First of all, while a warranty offered by an established dealership is
perhaps sufficiently credible,10 a warranty offered by a private individual is less credible:
the seller might “disappear” (e.g. move to a different state) and become untraceable when
the warranty becomes relevant (due to a needed repair) or the seller might declare personal
bankruptcy or become insolvent, etc. There is also an issue of moral hazard involving the
buyer, which is a source of concern for the seller: if the car is insured against future repairs,
the new owner might abuse the car and cause a breakdown that could be avoided with due
care. The topic of moral hazard will be discussed in the next chapter.

10 However, even dealerships can go bankrupt, leaving the buyer with a worthless warranty.
 9.6 Signaling in other markets 323

9.6.2 Advertising as a signal of quality

Advertising is pervasive in our daily lives: we are exposed to it as we drive on the freeway,
as we watch television, while we browse the internet, etc. Why do firms advertise?
What is the purpose of advertising? What information, if any, do consumers obtain from
advertisements?
It is clear that there cannot be a single answer to the above questions, because advertis-
ing falls into several different categories. In terms of content, for example, advertising can
be informative or uninformative. The former makes consumers aware of something they
did not know, such as the existence of a product, or its properties, or its price, or where
it can be obtained. Most advertising, however, does not seem to fall into such category:
for example, an ad showing a pop-star or an athlete drinking Coca Cola does not seem
to convey to consumers any information that they did not already have (are there many
people who do not know what Coca Cola is?); an ad claiming that a particular model of
car is “the best” seems to be just “cheap talk”, etc.
In this section we focus on “cheap talk” advertising, that is, claims made about a
product that are easily imitated by other producers. Does this type of advertising make
sense? Should consumers pay attention to it?
Nelson11 provided an explanation for this type of advertising in terms of what we now
call signaling. He observed that the characteristics that differentiate products in a given
market can be classified into search characteristics and experience characteristics.
Search characteristics are those that can be ascertained before purchasing the good
(e.g. the style of a dress). Since consumers can verify by themselves what the properties
of the good are - before buying it - in this case there is no scope for false or misleading
advertising: a false or exaggerated claim by the seller will not induce any purchases and
will only damage the seller’s reputation.
Experience characteristics, on the other hand, can only be learned through use and
therefore after the purchase (e.g. the quality of a stereo system or the taste of a brand of
canned tuna fish). In this case there is scope for false or misleading advertising: some
consumers might be induced to buy the product by claims that will turn out to be false.
However, Nelson argues that high-quality brands will obtain more repeat purchases, ceteris
paribus, than low-quality brands, because consumers will be satisfied with the product
and will buy it again in the future and/or recommend it to their friends. A dollar spent on
advertising for a high-quality good generates future revenue that justifies the advertising
expense. Producers of low-quality goods, on the other hand, might be able to fool a
consumer into buying the first time, but will experience no repeat purchases. In other
words, advertising is an expensive signal, which is more costly for the producer of a low-
quality product than for the producer of a high-quality product and can thus be used by the
latter to separate herself from the competitors who sell inferior goods. Hence advertising
is informative, no matter what its content: the information conveyed by advertising is that
the advertiser can afford the cost of advertising because its product is successful due to its
high quality: advertising is a signal that the advertised product is of high quality.

11 Phillip Nelson, Advertising as information, Journal of Political Economy, 1974, Vol. 82, No. 4, pp.
729-754.
 324 Chapter 9. Signaling

9.6.3 Other markets

There are more markets where the phenomenon of signaling takes place. For exam-
ple, Spence12 examines signaling in the market for loans and household credit and
Bhattacharya13 shows that firms may use their dividend policy as a signal of their fi-
nancial strength. We shall not pursue this topic any further.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 9.7.5 at the end of this chapter.

12 Michael Spence, Market signaling: Informational transfer in hiring and related screening processes,
Harvard University Press, 1974.
13 Sudipto Bhattacharya, Imperfect information, dividend policy and the “Bird in the Hand” fallacy, Bell

Journal of Economics, 1979, Vol. 110, pp. 257-70.

 9.7 Exercises 325

9.7 Exercises
The solutions to the following exercises are given in Section 9.8 at the end of this chapter.

9.7.1 Exercises for Section 9.2: Signaling in the job market

Table 9.9: Wage offered as a function of the number of years of schooling.

Education Wage offered

y=6 $10,000
y = 12 $24,000
y = 16 $29,000
y = 18 $34,000
y = 21 $36,000

Exercise 9.1 Suppose that employers offer the wage schedule shown in Table 9.9.
There are two types of individuals in the population: type H with productivity $34,000
(independent of education) and type L with productivity $10,000 (also independent of
education). Education is costly and the monetary cost of acquiring education for the
two types is as follows:

 0 if y ≤ 6
For type H : CH (y) =
 1, 600(y − 6) if y > 6


 0 if y ≤ 6
For type L : CL (y) =
 2, 500(y − 6) if y > 6

(a) Is there a separating signaling equilibrium?

(b) Let pH be the proportion of H-types in the population. For what values of pH
would the following situation be Pareto superior to the signaling equilibrium of
Part (a)? Institutions of education beyond sixth grade (that is, beyond y = 6) are
abolished and everybody is hired at a wage equal to the average productivity of
the population.

326 Chapter 9. Signaling

Exercise 9.2 Suppose that there are two groups of individuals: Group L with produc-
tivity 2 (independent of education) and Group H with productivity 3 (independent of
education). Group H constitutes 13 of the population. Workers of both types are able to
buy education, at a cost. The amount of education y is a continuous variable and is fully
verifiable (e.g. through a certificate).
Type-L individuals face a higher cost of acquiring education than type-H individuals:
CL (y) = y and CH (y) = 2y .
Employers believe that anybody with a level of education less than y∗ has a produc-
tivity of 2 (and thus is offered a wage of 2) while anybody with a level of education
greater than or equal to y∗ has a productivity of 3 (and thus is offered a wage of 3).
(a) What values of y∗ give rise to a separating signaling equilibrium?
(b) If the government intervened and forced everybody to choose y = 0 and employers
to pay the same wage to everybody, equal to the average productivity in the
population, would there be a Pareto improvement?


Exercise 9.3 Consider the following situation. Education does increase productivity.
There are two groups in the population.
People in Group L have a productivity of 1 + 4y (where y is the amount of education)

and their cost of acquiring y units of education is $y.

People in Group H have a productivity of 2 + 4y and the cost of acquiring y units of

education is $ 2y
Find all the signaling equilibria, when the employers’ beliefs are as follows: if a job
applicant has y < y∗ , then he/she must be from Group L (and thus will be offered a wage
of 1 + 4y ), while if he/she has y ≥ y∗ , then he/she must be from Group H (and thus will
be offered a wage of 2 + 4y ). 

Exercise 9.4 Consider the following situation, where education increases productivity.
There are two groups in the population.
People in Group L have a productivity of (4 + 2y) (where y is the amount of education)
and their cost of acquiring y units of education is $5y.
People in Group H have a productivity of (6 + 3y) and the cost of acquiring y units of
education is $4y
Find all the signaling equilibria, when the employers’ beliefs are as follows: if a job
applicant has y < y∗ , then he/she must be from Group L (and thus will be offered a wage
of 4 + 2y), while if he/she has y ≥ y∗ , then he/she must be from Group H (and thus will
be offered a wage of 6 + 3y). 
9.7 Exercises 327

Table 9.10: Wage offered as a function of the number of years of schooling.

Education Wage offered

y=6 $10,000
y = 12 $15,000
y = 16 $20,000
y = 18 $25,000
y = 21 $30,000

Exercise 9.5 Suppose that employers offer the wage schedule shown in Table 9.10.
There are two types of individuals in the population: type H with productivity πH (inde-
pendent of education) and type L with productivity πL (also independent of education).
Education is costly and the monetary cost of acquiring education for the two types is as
follows:

 0 if y ≤ 6
For type H : CH (y) =
 900(y − 6) if y > 6


 0 if y ≤ 6
For type L : CL (y) =
 1, 400(y − 6) if y > 6

(a) For what values of πH and πL is there a separating signaling equilibrium?

(b) Assume the values of πH and πL found in Part (a). Let pH be the proportion of
H-types in the population. For what values of pH would the following situation be
Pareto superior to the signaling equilibrium of Part (a)? Institutions of education
beyond sixth grade (that is, beyond y = 6) are abolished and everybody is hired at
a wage equal to the average productivity of the population.


Exercise 9.6 There are two types in the population: type H and type L. Productivity
and cost are functions of the amount of education y, as follows:

For type H : productivity: a + 4y cost: 5y

For type L : productivity: b + 3y cost: 6y

Employers believe that applicants with y ≥ y∗ > 0 are of type H and are thus offered a
wage equal to a + 4y and everybody else is offered a wage equal to b + 3y.
For what values of a and b is there a separating signaling equilibrium? 
 328 Chapter 9. Signaling

9.7.2 Exercises for Section 9.3: Indices versus signals

Exercise 9.7 Suppose that there are two groups of individuals:

Group L with productivity 1 (independent of education), and
Group H with productivity 2 (independent of education).
Group H constitutes 50% of the population. Workers of both types are able to buy
education, at a cost. The amount of education y is a continuous variable and is fully
verifiable (e.g. through a certificate).
Type-L individuals face a higher cost of acquiring education than type-H individuals:
y
CL (y) = y and CH (y) = .
2
Within each group there are as many men as women. Thus the composition of the
populations is as follows:

Type/gender percentage of population

men of type H 25%
women of type H 25%
men of type L 25%
women of type L 25%

Note that productivity is innate and is not influenced by either education or gender.
Suppose that employers mistakenly believe, not only that education affects pro-
ductivity, but also that women are somewhat “genetically handicapped” and therefore
need to acquire more education in order to become more productive. Accordingly, the
employers offer the following wage schedules, with y∗w > y∗m .

If applicant is MALE If applicant is FEMALE

education wage education wage
y < y∗m 1 y < y∗w 1
y ≥ y∗m 2 y ≥ y∗w 2

Are there values of y∗m and y∗w that give rise to a separating signaling equilibrium
where the employers beliefs are confirmed? 
 9.7 Exercises 329

9.7.3 Exercises for Section 9.4: More than two types

Exercise 9.8 Let y denote the amount of education. There are three types of potential
workers:
those (Group I) with productivity 36 (a constant, thus independent of education),
those (Group II) with productivity (60 + 3y), and
those (Group III) with productivity (72 + 2y).
Each worker knows whether she belongs to Group I or Group II or Group III, while the
potential employer does not. The cost of acquiring y units of education is
24y for Group I,
12y for Group II, and
6y for Group III.
Let 0 < a < b. The potential employer believes that those applicants with education
less than a belong to Group I, those with education at least a but less than b belong
to Group II and those with education at least b belong to Group III and offers each
applicant a wage equal to the applicant’s estimated productivity, given the applicant’s
level of education (the level of education can be verified by the employer during the job
interview).
(a) Write a list of inequalities (involving the parameters a and b) that are necessary
and sufficient for the existence of a signaling equilibrium.
(b) Explain why a = 2 and b = 2.5 is not a signaling equilibrium.
(c) Is a = 2.5 and b = 4 a signaling equilibrium?


Exercise 9.9 Suppose that there are four types in the population with the following
productivity and cost of education (y is the amount of education), with a1 < a2 < a3 < a4
and b1 > b2 > b3 > b4 > 2.

Type 1 Type 2 Type 3 Type 4

productivity a1 a2 a3 + y a4 + 2y
cost of education b1 y b2 y b3 y b4 y

Employers set three threshold levels of education, y∗1 , y∗2 and y∗3 and offer the following
wages, depending on the level of education y of the job applicant:

Education y wage
y < y∗1 a1
y∗1 ≤ y < y∗2 a2
y∗2 ≤ y < y∗3 a3 + y
y ≥ y∗3 a4 + 2y

Write the inequalities that need to be satisfied in order to have a separating signaling
equilibrium (where any two types make different choices of education). 
 330 Chapter 9. Signaling

9.7.4 Exercises for Section 9.5: A more general analysis

Exercise 9.10 Continuing the analysis of Section 9.5, consider the case where the
threshold value y∗ is less than the level of education that would be chosen by an L-type
under perfect information, that is, the case where y∗ < yL .
Draw a diagram showing the wage schedule and the relevant indifference curves of the
two types and determine whether there can be a separating signaling equilibrium in this
case. 

Exercise 9.11 Continuing the analysis of Section 9.5, consider the case where the
threshold value y∗ is greater than the level of education that would be chosen by an
L-type under perfect information, and less than the level of education that would be
chosen by an H-type under perfect information that is, the case where yL < y∗ < yH .
Draw a diagram showing the wage schedule and the relevant indifference curves of the
two types and determine whether there can be a separating signaling equilibrium in this
case. [Hint: you need to distinguish two cases.] 

9.7.5 Exercises for Section 9.6: Signaling in other markets

Exercise 9.12 Consider the market for second-hand cars. There are four possible
qualities with the following proportions and values to sellers:

quality of car: A B C D
1 2 2 3
proportion: 8 8 8 8
value to seller: $15,000 $10,000 $8,000 $6,000.

All sellers and buyers are risk neutral. While each seller knows the quality of his car,
each buyer is unable to determine the quality of any car offered for sale.
A car of quality i ∈ {A, B,C, D} has probability qi of requiring a major repair
1
within the next 5 years with qA = 10 , qB = 12 , qC = 54 , qD = 100
99
. A major repair costs
$15,000. The buyer’s valuation bi of a car of quality i reflects this, in the sense that, for
all i ∈ {A, B,C, D}, bi = $(24, 000 − 15, 000qi ).
(a) Compute the value of each quality to the buyers.
(b) Suppose first that there is only one price for used cars, p = $9, 000. What cars
are traded?
(c) Suppose now that each seller can offer his car for sale either with or without
warranty. If a car is sold with warranty, the seller credibly undertakes to pay for
the repair himself, should a repair become necessary within the next 5 years. In
this new situation, there can be two prices in the market: a price pN for cars sold
without warranty and a price pW for cars sold with warranty. A car of quality A
sold with warranty is worth $24,000 to the buyer. Suppose that pW = $17, 000
and pN = $11, 000. What cars are sold with warranty and what cars are sold
without warranty?

 9.8 Solutions to Exercises 331

9.8 Solutions to Exercises

Solution to Exercise 9.1.
(a) The cost-benefit analysis of a type-H person is shown in Table 9.11. Thus a type-
H individual will choose y = 18 and be paid $34,000 which is equal to her true
productivity.

Table 9.11: The cost-benefit analysis of a type-H person.

y w CH w −CH
years in school gross wage cost net wage
y=6 $10,000 $0 $10,000
y = 12 $24,000 $9,600 $14,400
y = 16 $29,000 $16,000 $13,000
y = 18 $34,000 $19,200 $14, 800
y = 21 $36,000 $24,000 $12,000

The cost-benefit analysis of a type-L person is shown in Table 9.12. Thus a type-
L individual will choose y = 6 and be paid $10,000 which is equal to his true
productivity.
Hence we do have a separating signaling equilibrium.

Table 9.12: The cost-benefit analysis of a type-L person.

y w CL w −CL
years in school gross wage cost net wage
y=6 $10,000 $0 $10, 000
y = 12 $24,000 $15,000 $9, 000
y = 16 $29,000 $25,000 $4,000
y = 18 $34,000 $30,000 $4,000
y = 21 $36,000 $37,500 $-1,500

(b) In the new world everybody chooses y = 6 and faces zero costs of education. The
average productivity is 34, 000pH + 10, 000(1 − pH ) = 10, 000 + 24, 000pH . Type-
L individuals will be better off because 10, 000 + 24, 000pH > 10, 000. Type-H
individuals will now have a net income of 10, 000 + 24, 000pH , which is greater than
or equal to their net income at the signaling equilibrium of Part (a), namely $14,800,
if and only if pH ≥ 15 . Thus we have a Pareto improvement if and only if pH ≥ 15 . 
332 Chapter 9. Signaling

Solution to Exercise 9.2.

(a) Nobody will want to acquire a level of education y > y∗ or a level 0 < y < y∗ . Thus
every individual will limit herself to choosing between y = 0 and y = y∗ .
For a type-L, y = 0 yields a net income of 2 and y = y∗ yields a net income of 3 − y∗ .
Thus he will choose y = 0 if and only if 2 ≥ 3 − y∗ , that is, if and only if y∗ ≥ 1.
∗
For a type-H, y = 0 yields a net income of 2 and y = y∗ yields a net income of 3 − y2 .
∗
Thus she will choose y = y∗ if and only if 3 − y2 ≥ 2, that is, if and only if y∗ ≤ 2.
Hence every y∗ ∈ [1, 2], yields a separating signaling equilibrium.

(b) The average productivity is 13 3 + 32 2 = 73 . In the new situation employers are as

well off as before and type-L individuals are better off. Type-H individuals are not
∗
worse off if and only if 73 ≥ 3 − y2 , which is the case if and only if y∗ ≥ 43 . Thus if
y∗ ∈ 43 , 2 then the government intervention leads to a Pareto improvement, while
 

if y∗ ∈ 1, 43 , then there is no Pareto improvement because type-H individuals are



made worse off by the government intervention. 

Solution to Exercise 9.3.

First of all, note that the only sensible choices are y = 0 and y = y∗ . In fact, if you increase
y by 1 unit, you get an extra $0.25 but you pay more than this (you pay an extra $1 if you
belong to Group L and an extra $ 0.50 if you belong to Group H).
Decision for a Group-L person:
education wage cost net income
y=0 1 + 40 = 1 0 1
y∗
y = y∗ 2+ 4 y∗ 2 − 34 y∗

Thus a Group-L person will choose y = 0 and be paid a wage equal to his true productivity

if and only if 1 ≥ 2 − 34 y∗ , that is, if and only if y∗ ≥ 4

3 .

Decision for a Group-H person:

education wage cost net income
y=0 1 + 04 = 1 0 1
y∗ y∗
y = y∗ 2+ 4 2 2 − 14 y∗

Thus a Group-H person will choose y = y∗ and be paid a wage equal to her true productivity
if and only if 2 − 14 y∗ ≥ 1, that is, if and only if y∗ ≤ 4 .

Thus every y∗ ∈ 43 , 4 gives rise to a separating signaling equilibrium.

 

9.8 Solutions to Exercises 333

Solution to Exercise 9.4.

As usual, the only sensible choices are y = 0 and y = y∗ .
Decision for a Group-L person:
education wage cost net income
y=0 4 0 4
y = y∗ 6 + 3y∗ 5y∗ 6 − 2y∗
Thus a Group-L person will choose y = 0 and be paid a wage equal to his true productivity
if and only if 4 ≥ 6 − 2y∗ , that is, if and only if y∗ ≥ 1 .
Decision for a Group-H person:
education wage cost net income
y=0 4 0 4
y = y∗ 6 + 3y∗ 4y∗ 6 − y∗
Thus a Group-H person will choose y = y∗ and be paid a wage equal to her true productivity
if and only if 6 − y∗ ≥ 4, that is, if and only if y∗ ≤ 2 .
Thus every y∗ ∈ [1, 2] gives rise to a separating signaling equilibrium. 
Solution to Exercise 9.5.
(a) The cost-benefit analysis of a type-H person is shown in the following table. Thus a
type-H individual will choose y = 21 and be paid $30,000.

y w CH w −CH
y=6 $10,000 $0 $10,000
y = 12 $15,000 $5,400 $9,600
y = 16 $20,000 $9,000 $11,000
y = 18 $25,000 $10,800 $14,200
y = 21 $30,000 $13,500 $16, 500

The cost-benefit analysis of a type-L person is shown in the following table. Thus a
type-L individual will choose y = 6 and be paid $10,000.
Hence in order to have a signaling equilibrium it must be that πH = 30, 000 and
πL = 10, 000.

y w CL w −CL
y=6 $10,000 $0 $10, 000
y = 12 $15,000 $8,400 $6, 600
y = 16 $20,000 $14,000 $6,000
y = 18 $25,000 $16,800 $8,200
y = 21 $30,000 $21,000 $9,000

(b) In the new world everybody chooses y = 6 (the only available choice). The average
productivity is 30, 000pH + 10, 000(1 − pH ) = 10, 000 + 20, 000pH . Thus type-L
individuals will be better off, since 10, 000 + 20, 000pH > 10, 000. Type-H individu-
als will now have a net income of 10, 000 + 20, 000pH ; this is greater than, or equal
to, their net income at the signaling equilibrium of Part (a), namely $16,500, if and
13 13
only if pH ≥ 40 . Thus we have a Pareto improvement if and only if pH ≥ 40 . 
334 Chapter 9. Signaling

Solution to Exercise 9.6.

As usual, we only need to compare the choices y = 0 and y = y∗ . An H type will choose
y = y∗ if and only if

a + 4y∗ − 5y∗ ≥ b that is, a − b ≥ y∗ . (9.17)

An L type will choose y = 0 if and only if

b ≥ a + 4y∗ − 6y∗ that is, a − b ≤ 2y∗ . (9.18)

Thus any two values of a and b such that y∗ ≤ a − b ≤ 2y∗ will give rise to a separating
signaling equilibrium. 

Solution to Exercise 9.7.

Let us construct an equilibrium where (1) all type-L individuals (male and female) choose
y = 1, (2) type-H men choose y = y∗m and (3) type-H women y = y∗w .

1. For (1) we need 1 ≥ 2 − y∗m (for men), that is y∗m ≥ 1 and 1 ≥ 2 − y∗w (for women),
that is y∗w ≥ 1. Since y∗w > y∗m , both conditions are satisfied if

1 ≤ y∗m . (9.19)

y∗m
2. For (2) we need 2 − 2 ≥ 1, that is,

y∗m ≤ 2. (9.20)

y∗w
3. For (3) we need 2 − 2 ≥ 1, that is,

y∗w ≤ 2. (9.21)

Since y∗w > y∗m , (9.21) implies (9.20). Thus any pair of values (y∗m , y∗w ) such that
1 ≤ y∗m < y∗w ≤ 2 gives rise to a separating equilibrium where employers’ beliefs are
confirmed in every respect. In particular, they obtain no evidence against their belief
that, in order to become more productive, women need to spend more time in school: a
man acquires a productivity of 2 by spending y∗m years in school, while women acquire a
productivity of 2 by spending more time in school, namely y∗w years rather than y∗m . 
9.8 Solutions to Exercises 335

Solution to Exercise 9.8.

(a) First of all, since for every group the cost of one extra unit of education exceeds the
benefit (in terms of increased salary), everybody will only consider only y = 0, y = a
and y = b. The inequalities are as follows.
For Group I:
(1) 36 ≥ 60 + 3a − 24a, that is, a ≥ 87 .
18
(2) 36 ≥ 72 + 2b − 24b, that is, b ≥ 11 .
For Group II:
(3) 60 + 3a − 12a ≥ 36, that is, a ≤ 83 .
(4) 60 + 3a − 12a ≥ 72 + 2b − 12b, that is, b ≥ 56 + 10
9
a.
For Group III:
(5) 72 + 2b − 6b ≥ 36, that is, b ≤ 9.
(6) 72 + 2b − 6b ≥ 60 + 3a − 6a, that is, b ≤ 3 + 34 a.
(b) When a = 2 and b = 2.5, inequality (4) is violated. Thus Group II individuals
would be better off pretending to be Group III by choosing y = b = 2.5 (instead of
y = a = 2).
(c) Yes, when a = 2.5 and b = 4, all the above inequalities are satisfied. 

Solution to Exercise 9.9.

There are 12 inequalities, which are as follows.
For Group 1:
(1) a1 ≥ a2 − b1 y∗1 .
(2) a1 ≥ a3 + y∗2 − b1 y∗2 .
(3) a1 ≥ a4 + 2y∗3 − b1 y∗3 .
For Group 2:
(4) a2 − b2 y∗1 ≥ a1 .
(5) a2 − b2 y∗1 ≥ a3 + y∗2 − b2 y∗2 .
(6) a2 − b2 y∗1 ≥ a4 + 2y∗3 − b2 y∗3 .
For Group 3:
(7) a3 + y∗2 − b3 y∗2 ≥ a1 .
(8) a3 + y∗2 − b3 y∗2 ≥ a2 − b3 y∗1 .
(9) a3 + y∗2 − b3 y∗2 ≥ a4 + 2y∗3 − b3 y∗3 .
For Group 4:
(10) a4 + 2y∗3 − b4 y∗3 ≥ a1 .
(11) a4 + 2y∗3 − b4 y∗3 ≥ a2 − b4 y∗1 .
(12) a4 + 2y∗3 − b4 y∗3 ≥ a3 + y∗2 − b4 y∗2 . 
336 Chapter 9. Signaling

Solution to Exercise 9.10.

There is no signaling equilibrium in this case, which is illustrated in Figure 9.13.14 The
H-type will choose education level yH , because the point (yH , kH yH ) is the best along the
productivity line Π = kH y and hence also better than any point on the lower productivity
line Π = kL y and thus it is the utility-maximizing point on the wage schedule. Type
L individuals will choose an education level y ≥ y∗ for the following reason: the point
(yL , kL yL ) is the best point for the L-type along the productivity line Π = kL y and the point
(yL , kH yL ) is even better, because it involves a larger income; it follows that (yL , kH yL )
is better than any point on the productivity line Π = kL y, in particular, the segment of
it which belongs to the wage schedule. Since (yL , kH yL ) is available (it belongs to the
wage schedule) the L-types will not make a choice of education that leads to them being
remunerated according to their true productivity. Hence, even if their choice of education
is different from the choice made by the H-types, so that we would have a separating
outcome, it would not be an equilibrium, because employers would discover that they are
overpaying some of the employees and would want to change the wage schedule.

wage
offered (Type H)
(Type L)

y
0 y∗ yL yH education

Figure 9.13: Every point on the first segment of the wage schedule is worse for the L-types
than some point on the second segment of the wage schedule.

14 We have drawn the case where the point (yH , kH yH ) lies above the L-type indifference curve that goes
through the point (yL , kL yL ), but this is not a relevant fact: even if this were not the case, the argument would
remain valid.
9.8 Solutions to Exercises 337

Solution to Exercise 9.11.

Since yH > y∗ , the H-types will choose education level yH and be remunerated according
to their true productivity. The issue is what choice the L-types will make. We need to
distinguish two cases.
Case 1: The point (y∗ , kH y∗ ) lies above the L-type indifference curve that goes through
the point (yL , kL yL ). This case is illustrated in Figure 9.14.

wage
offered (Type H)

(Type L)

y
0 yL y∗ yH education

Figure 9.14: The case where the point (y∗ , kH y∗ ) lies above the L-type indifference curve
that goes through the point (yL , kL yL ).

In this case the point (y∗ , kH y∗ ) - which is on the wage schedule - is better than the point
(yL , kL yL ), which is the best point on the first segment of the wage schedule (indeed the
best point on the entire line corresponding to the productivity of the L-type: Π = kL y).
Thus the L-type will not choose a level of education less than y∗ and thus will not be
remunerated according to their true productivity. As remarked in the answer to Exercise
9.10, even if the choice of education of the L-types is different from the choice made by
the H-types (namely yH ) - so that we would have a separating outcome - it would not be
an equilibrium, because employers would discover that they are overpaying some of the
employees and would want to change the wage schedule.
338 Chapter 9. Signaling

Case 2: The point (y∗ , kH y∗ ) lies to the right of the L-type indifference curve that goes
through the point (yL , kL yL ). This case is illustrated in Figure 9.15.

wage
offered (Type H)

(Type L)

y
0 yL y∗ yH education

Figure 9.15: The case where the point (y∗ , kH y∗ ) lies to the right of the L-type indifference
curve that goes through the point (yL , kL yL ).

In this case every point on the second segment of the wage schedule (the segment that
starts at (y∗ , kH y∗ )) is worse, for the L-type, than the point (yL , kL yL ), which - in turn - is
the best point on the first segment of the wage schedule. Thus the L-types will choose
level of education yL and the H-types will choose level of education yH and both types
will be remunerated according to their true productivity. Hence in this case we do have
a separating signaling equilibrium. Note that it coincides (in terms of education choices
and remuneration) with what would happen in the case of perfect information. Thus in
this case the asymmetry of information does not have negative consequences; in particular
there is no over-investment in education by the H-types (unlike the case where y∗ > yH
considered in Section 9.5).
9.8 Solutions to Exercises 339

Solution to Exercise 9.12.

(a) bA = 22, 500, bB = 16, 500, bC = 12, 000, bD = 9, 150.
(b) Since p = 9, 000, only qualities C and D will be offered for sale. Thus buying a car
means facing the lottery  
$12, 000 $9, 150
2 3
5 5
whose expected value is $10,290 which is greater than the price of $9,000; hence
buyers are willing to buy.15 Hence only cars of quality C and D are traded.
(c) The owner of a car of quality A is not willing to sell at price pN = 11, 000. If he sells
at price pW = 17, 000 then he faces the lottery
 
$17, 000 $(17, 000 − 15, 000)
9 1
10 10

whose expected value is 15,500, greater than the value of the car. Thus he is willing
to sell with warranty at price pW = 17, 000.
If the owner of a car of quality B sells at price pW = 17, 000 then he faces the lottery
 
$17, 000 $(17, 000 − 15, 000)
1 1
2 2

whose expected value is 9,500 which is less than the value of the car to him, namely
10,000. Thus he is not willing to sell with warranty.
If the owner of a car of quality C sells at price pW = 17, 000 then he faces the lottery
 
$17, 000 $(17, 000 − 15, 000)
1 4
5 5

whose expected value is 5,000 which is less than the value of the car to him, namely
8,000. Thus he is not willing to sell with warranty.
If the owner of a car of quality D sells at price pW = 17, 000 then he faces the lottery
 
$17, 000 $(17, 000 − 15, 000)
1 99
100 100

whose expected value is 2,150 which is less than the value of the car to him, namely
6,000. Thus he is not willing to sell with warranty.
Thus the only cars offered for sale with warranty are the cars of quality A. Since
the buyer values a car of quality A sold with warranty at $24,000, he is willing to
pay $17,000, realizing a gain equal to $7,000. Thus all cars of quality A will be sold
with warranty at price pW = 17, 000.
15 Alternatively one could say that the lottery - after paying the price - is
 
$3, 000 $150
2 3
5 5

whose expected value is $1,290, hence a positive net benefit from buying.
340 Chapter 9. Signaling

Now let us look at the market for cars sold without warranty. Since pN = 11, 000
the owners of cars of qualities B, C and D are willing to sell (they value their cars
less than 11,000). Thus buying a car without warranty means facing the lottery
 
$16, 500 $12, 000 $9, 150
2 2 3
7 7 7

which has an expected value of 12,064.29, greater than the price of 11,000. Thus
consumers are willing to buy.
In conclusion, cars of quality A will be sold with warranty at a price of $17,000 and
all the other cars will be sold without warranty at a price of $11,000.
Thus, offering to sell with warranty is a signal used by the owners of the best cars
(those of quality A) to credibly convey information to potential buyers about the
quality of their cars.
V
Moral Hazard

10 Moral Hazard in Insurance . . . . . 343

10.1 Moral hazard or hidden action
10.2 Moral hazard and insurance
10.3 Exercises
10.4 Solutions to Exercises

11 Moral Hazard in Principal-Agent 369

11.1 Moral hazard in Principal-Agent relationships
11.2 Risk sharing under moral hazard
11.3 The case with two outcomes and two levels of
effort
11.4 The case with more than two outcomes
11.5 Exercises
11.6 Solutions to Exercises

12 Glossary . . . . . . . . . . . . . . . . . . . . . . . . . 409

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
 10. Moral Hazard in Insurance

10.1 Moral hazard or hidden action

In Chapters 7-9 we analysed the phenomenon of asymmetric information, namely situations
where one party to a potential transaction has some relevant characteristics that are known
to her but cannot be ascertained (at the time of signing the contract) by the other party. For
this reason asymmetric information is also referred to as “hidden type”. The expression
“moral hazard”, on the other hand, refers to situations where the behavior of one of the
parties to a contract cannot be observed by the other. For this reason it is also called
“hidden action”.
A parent who hires a babysitter to look after his child while he is at work cares very
much about the behavior of the sitter: will she be keeping a constant eye on the child or
will she be updating her social media profile? Will she comfort the child if he/she cries or
let him/her cry for a long time?
The owner of a store who hires a shopkeeper to take care of business will want the
employee to be helpful to customers and do what he can to induce them to make a purchase;
but if the owner is not present, she cannot tell whether the employee is working hard or is
shirking.
An individual who hires a lawyer to file a lawsuit on her behalf will want the lawyer to
spend time and effort studying the case and preparing well-informed briefs. Unfortunately,
it is typically the case that the client cannot monitor the lawyer’s effort.
What all these cases have in common is the fact that the outcome of the contractual
relationship depends on the level of care and effort that one of the two parties puts into the
endeavor and monitoring this party’s behavior is not possible or too expensive for the other
party. Thus, if possible, one has to design the contract in such a way that the un-monitored
party has an incentive to behave in a way that is aligned with the interest of the other party.
In this chapter we will examine the role of moral hazard in insurance markets and in
the next chapter we will revisit the Principal-Agent environment of Chapter 6 and add the
 344 Chapter 10. Moral Hazard in Insurance

possibility of moral hazard.

10.2 Moral hazard and insurance

In the chapters that dealt with insurance (Chapters 2, 5 and 8) we assumed that the
probability of loss was constant and the individual’s decision was merely whether or not to
insure and which contract to choose from a given menu. When we say that “the probability
of loss is constant” we mean that it cannot be affected by the individual’s behavior.
In some cases this is a reasonable assumption: for example, there is nothing that a
shopkeeper can do to make it less likely that there will be a riot, or an earthquake, or a
meteorite strike.
In other cases, however, there is a causal link between the behavior of the insured
person and the probability that she will suffer a loss: for example, the probability that her
bicycle will be stolen is higher if she leaves it unattended and unlocked, and lower if she
locks it to a permanent fixture with a sturdy cable and padlock.
In cases where the probability of loss can be affected by the individual’s actions we
say that the insurance company faces a situation of moral hazard or hidden action. Below
are a few more examples of possible actions that the individual can take to reduce the
probability of loss:
◦ Install a home security alarm, to make a robbery less likely.
◦ Always lock the door(s) to one’s house, to make a robbery less likely.
◦ Clear the brush around the house, to make it less likely that a brush fire will reach
the building.
◦ Drive carefully and below the speed limit, to make it less likely that one will be
involved in a car accident.
◦ Exercise regularly and eat healthy food to reduce the probability of vascular or
cardiac disease.
As the above examples illustrate, what an individual can do to reduce the probability
of loss can be either
• taking an action involving extra effort, or
• incurring an extra expense.
We shall refer to both of the above as making extra effort. The crucial element of both is
that they involve “disutility”, that is, they make the individual worse off, ceteris paribus.
Because of this, the insurance company realizes that the individual will prefer to avoid the
extra effort if she is protected from the consequences of not exerting it. For example, if
the individual has full insurance, then she has no incentive to incur extra expenses or exert
extra effort in order to reduce the probability of suffering a loss: if the loss occurs, she will
be fully reimbursed by the insurance company. On the other hand, the insurance company
cares very much about the behavior of its customers, because the more careless they are,
the more likely it is that the insurance company will have to cover their losses. In order to
incentivize the customer to exert extra effort, the insurance company might want to make
it more costly for the insured to suffer a loss, by requiring a substantial deductible.
 10.2 Moral hazard and insurance 345

If the customer’s behavior can be observed by the insurance company and verified by
a court of law, then the insurance company can specify it in the contract and make any
payments conditional on the customer’s actions. For example, the insurance company could
require the customer to install a security system in her house and make any reimbursements
due to theft conditional on proof that the security system was in fact installed. However,
in most cases it is impossible, or prohibitively expensive, for the insurer to monitor the
behavior of its customers. For example, no insurance company will find it worthwhile to
have an insurance agent follow the customer around to make sure that she always locks her
bicycle, when left unattended!
We will assume that the customer’s behavior cannot be observed by the insurance
company; however, the insurance company can try to figure out what the potential customer
would do under different insurance contract.

10.2.1 Two levels of unobserved effort

We will limit ourselves to the binary case, where the individual has only two choices in
terms of effort: either exert effort, denoted by E, or exert no effort, denoted by N. The
individual’s choice affects the probability of loss: it will be lower in she exerts effort, that
is, letting pE be the probability of loss in the case of effort and pN the probability of loss
in the case of no effort,
0 < pE < pN < 1. (10.1)
As in previous chapters, we denote the individual’s initial wealth by W0 and the potential
loss by ` (with 0 < ` ≤ W0 ). We continue to assume that the individual has vNM prefer-
ences; however, the outcomes of the lotteries now include not only wealth levels but also
the “inconvenience” or “cost” of exerting effort. Thus we can think of the individual as
having two utility-of-money functions: one if she exerts effort, denoted by UE (m), and the
other if she exerts no effort, denoted by UN (m).1
The fact that effort is costly (either in a psychological or in a monetary sense) is
captured by the following assumption:
UN (m) > UE (m), for every m ≥ 0.
To simplify the analysis we will consider the following special case: let U(m) be a strictly
increasing and concave function then
UN (m) = U(m) (10.2)

UE (m) = U(m) − c, with c > 0. (10.3)

Note that, since pE < pN ,2

pE pN
< . (10.4)
1 − pE 1 − pN
1 Equivalently,
the vNM utility function has two arguments: money and level of effort.
2
This can be seen as follows:
pE pN
(1) 1−p E
< 1−p E
because the denominator is the same and pE < pN , and
pN pN
(2) 1−pE < 1−pN because the numerator is the same and (1 − pE ) > (1 − pN ) (since pE < pN ).
346 Chapter 10. Moral Hazard in Insurance

It follows from (10.2) and (10.3) that, for any point (W1 ,W2 ) in the wealth space,
- with No-effort, the slope of the indifference curve at point (W1 ,W2 ) is
 0 
pN U (W1 )
− (10.5)
1 − pN U 0 (W2 )

- with Effort, the slope of the indifference curve at point (W1 ,W2 ) is
 0 
pE U (W1 )
− . (10.6)
1 − pE U 0 (W2 )

Thus, for any point A = (W1A ,W2A ) in the wealth space, we deduce from (10.4), (10.5) and
(10.6) that

the indifference curve (through point A) corresponding to Effort

is, at point A, less steep than,
the indifference curve (through point A) corresponding to No-effort.

Wealth in
good state
W2

NI 45o line
W0

indifference curves
with Effort

indifference curves
with No-effort

Wealth in
W1 bad state
0 W0 − `

Figure 10.1: Indifference curves corresponding to Effort are less steep than those corre-
sponding to No-effort.

Figure 10.1 shows two sets of indifference curves: one corresponding to Effort (the
less steep ones) and one corresponding to No-effort (the steeper ones).
 10.2 Moral hazard and insurance 347

When offered an insurance contract, the individual will have four possible choices:
1. remain uninsured and choose Effort,
2. remain uninsured and choose No-effort,
3. purchase the insurance contract and choose Effort,
4. purchase the insurance contract and choose No-effort,
and she will choose the option that yields the highest expected utility.

In general, the individual’s decision problem can be framed as follows:

◦ first determine the best effort level if uninsured,
◦ then, for each offered insurance contract, determine the best effort level if that
contract is purchased,
◦ compare the expected utility from the best choice of effort under each option and
choose that option that yields the largest expected utility.
For example, consider an individual who has an initial wealth of $50,000 and faces a
potential loss of $30,000 with the following probability:
1
(
pE = 10 if she chooses Effort
probability of loss =
1
pN = 5 if she chooses No-effort.

Her vNM preferences are represented the following vNM utility-of-money function
(for m > 0):3
10 − 10,000
(
m if she chooses Effort
U(m) =
10.01 − 10,000
m if she chooses No-effort.
Suppose that she is offered the following menu of insurance contracts:

premium deductible
Contract 1: $1,000 $5,000
Contract 2: $500 $8,000
Contract 3: $100 $12,000

Will she insure and, if so, which contract will she choose?

Step 1. Determine the best effort level in case of no insurance NI = (20000, 50000):
1
10 − 21 + 10

9
10 − 15 = 9.77 .


Effort: E[UE (NI)] = 10
No-effort: E[UN (NI)] = 51 10.01 − 12 + 45 10.01 − 15 = 9.75.

Thus, if uninsured, the individual will choose Effort.

3 This is an instance of (10.2) and (10.3) with c = 0.01.

 348 Chapter 10. Moral Hazard in Insurance

Step 2. Determine the best effort level for Contract 1, namely C1 = (44000, 49000):
1 1
  9 1

Effort: E[UE (C1 )] = 10 10 − 4.4 + 10 10 − 4.9 = 9.7936.
1
No-effort: E[UN (C1 )] = 15 10.01 − 4.4 + 45 10.01 − 4.9
1
   
= 9.8013 .
Thus, under Contract 1, the individual will choose No-effort.

Step 3. Determine the best effort level for Contract 2, namely C2 = (41500, 49500):
1 1
  9 1

Effort: E[UE (C2 )] = 10 10 − 4.15 + 10 10 − 4.95 = 9.7941.
No-effort: E[UN (C2 )] = 15 10.01 − 4.15
1
+ 45 10.01 − 4.95
1
   
= 9.8002 .
Thus, under Contract 2, the individual will choose No-effort.

Step 4. Determine the best effort level for Contract 3, namely C3 = (37900, 49900):
1 1
  9 1

Effort: E[UE (C3 )] = 10 10 − 3.79 + 10 10 − 4.99 = 9.7933.
No-effort: E[UN (C3 )] = 15 10.01 − 3.79
1
+ 45 10.01 − 4.99
1
   
= 9.7969 .
Thus, under Contract 3, the individual will choose No-effort.

Of all these options, the one that gives the highest expected utility is Contract 2 with No-
effort. Thus the individual would purchase Contract 2 and choose No-effort. The insurer’s
expected profit from Contract 2 is thus h − pN (` − d) = 500 − 15 (30, 000 − 8, 000) =
$ − 3, 900: a loss! Hence the insurance company would not want to offer Contract 2 (or
any of the other two contracts, since they all involve a loss). Before we address the issue
of what contract(s) would be offered by an insurer, we need to re-examine the notion of
“reservation indifference curve” in the context of moral hazard.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 10.3.1 at the end of this chapter.

10.2.2 The reservation utility locus

In previous chapters we defined the reservation indifference curve as the indifference curve
that goes through the no-insurance point NI. We can no longer do so in the present context,
because, for every point in the wealth space, there are now two indifference curves, one
corresponding to Effort and the other to No-effort. Since a reservation indifference curve
is supposed to contain all the contracts that yield the same expected utility as no insurance,
we first need to determine the “reservation utility” of the individual, that is, the maximum
utility that she can obtain if she does not insure. Letting E[UE (NI)] be the expected utility
under no insurance if the individual chooses Effort and E[UN (NI)] be the expected utility
under no insurance if the individual chooses No-effort, the reservation utility is:

EUNI = max {E[UE (NI)], E[UN (NI)]} . (10.7)

The interesting case is where under no insurance the individual will choose Effort; it is
interesting because insurance might provide an incentive for the individual to switch to
10.2 Moral hazard and insurance 349

No-effort. Thus in this section and the next we will assume that E[UE (NI)] > E[UN (NI)]
so that, by (10.7),

EUNI = E[UE (NI)]. (10.8)

The individual will reject any contract which, with the best choice of effort, will yield a
utility which is less than E[UE (NI)], that is, E[UE (NI)] provides the reservation level of
utility. Does this mean that the indifference curve corresponding to Effort that goes through
NI can be taken to be the reservation indifference curve, that is, the set of contracts that
give an expected utility equal to E[UE (NI)]? The answer is negative, because the fact that
the individual prefers to choose Effort if uninsured does not imply that she will continue
to choose Effort when insured. For example, if fully insured then she will be better off by
choosing No-effort.
Figure 10.2 shows the two indifference curves that go through the NI point: the less
steep one corresponds to Effort and a utility level of ū = E[UE (NI)] and the steeper one to
No-effort and a utility level of û = E[UN (NI)].

Wealth in
good state
W2

NI 45o line
W0

utility
ū F assume: ū > û
W2F = W1F
utility indifference curve
û with Effort; utility: ū
indifference curve
with No-effort; utility: û

Wealth in
W1 bad state
0 W0 − ` W1F

Figure 10.2: The two indifference curves through NI.

In accordance with (10.8) we assume

F F

that ū > û, that is, that when uninsured, the
o
individual
chooses Effort. Let F = W1 ,W1 be the point at the intersection of the 45 line and the
indifference curve corresponding to Effort that goes through NI; then E[UE (F)] = UE (W1F ).
By definition of indifference curve, UE (W1F ) = E[UE (NI)] = ū; however, if offered the
full-insurance contract F the individual can achieve a higher level of utility by purchasing
350 Chapter 10. Moral Hazard in Insurance

the contract and switching to No-effort. In fact, by (10.2) and (10.3),

UN (W1F ) = UE (W1F ) + c = ū + c > ū.

Let us use the expression reservation utility locus4 to denote the set of points (contracts) in
the wealth plane that give an expected utility equal to ū, when the individual makes the
best choice of effort; then, while NI belongs to it, point F does not. Thus the reservation
utility locus does not coincide with the indifference curve corresponding to Effort that goes
through NI.
How do we determine the reservation utility locus?
Continue to denote by ū the individual’s expected utility when uninsured and choosing
Effort. We need to consider the expected utility from choosing No-effort. By hypothesis,

E[UN (NI)] < ū

and, as shown above, UN (F) > ū.

Thus, by continuity, there must be a point, call it A = W1A ,W2A , on the indifference curve

corresponding to Effort that goes through NI, such that (see Figure 10.3)

E[UN (A)] = ū.

Wealth in
good state
W2

NI 45o line
W0
utility: ū
A assume: ū > û
F
utility: indifference curve
û with Effort; utility: ū
indifference curves
with No-effort

Wealth in
W1 bad state
0 W0 − `

Figure 10.3: Point A is such that E[UN (A)] = E[UE (A)] = ū.

Thus, since the reservation utility level (that is, the maximum utility that the individual
can achieve if not insured) is ū (= E[UE (NI)]), the reservation utility locus is a kinked
4 Instead of ‘reservation indifference curve’, since it does not coincide with an indifference curve.
10.2 Moral hazard and insurance 351

curve consisting of the initial segment from NI to point A of the indifference curve
through NI corresponding to Effort and the segment from point A to the 45o line of
the indifference curve through A corresponding to No-effort, where point A is such that
E[UN (A)] = E[UE (A)] = E[UE (NI)].
The reservation utility locus is shown as a thick continuous kinked curve in Figure
10.4.

Wealth in
good state
W2

NI 45o line
W0

indifference curve
with Effort; utility: ū
indifference curve
with No-effort; utility: ū

Wealth in
W1 bad state
0 W0 − `

Figure 10.4: The reservation utility locus is the union of the thick continuous curves.

Let us illustrate all of the above in an example.

Consider an individual whose vNM utility-of-money function is

( √
UN (m) = m if she chooses No-effort
√ with c > 0.
UE (m) = m − c if she chooses Effort

The individual’s initial wealth is W0 and she faces a potential loss of `. The probability
of her incurring a loss is pE if she chooses Effort and pN if she chooses No-effort, with
0 < pE < pN < 1.
352 Chapter 10. Moral Hazard in Insurance

 Let us first determine for what values of c she will choose Effort when not insured.

Her expected utility if she has no insurance and chooses No-effort is

p p
E[UN (NI)] = pN W0 − ` + (1 − pN ) W0

and her expected utility if she has no insurance and chooses Effort is
p  p 
E[UE (NI)] = pE W0 − ` − c + (1 − pE ) W0 − c
p p
= pE W0 − ` + (1 − pE ) W0 − c.

Then it must be that

 p p   p p 
c < pE W0 − ` + (1 − pE ) W0 − pN W0 − ` + (1 − pN ) W0 ,

that is,
p p 
c < (pN − pE ) W0 − W0 − ` (10.9)

For the rest of this section let us fix the following values of the parameters, which
satisfy (10.9):5

1 1 15
W0 = 2, 500 ` = 1, 600 pE = pN = c=
20 10 16

 What is the individual’s reservation utility level?

Since (10.9) is satisfied, the individual – if uninsured – will choose Effort and thus her
reservation utility is:

1√ 19 p 15 769
E[UE (NI)] = 900 + 2, 500 − = = 48.0625.
20 20 16 16

 Let us find the contract, call it A, that would make the individual indifferent between
1. not insuring and choosing Effort,
2. purchasing contract A and choosing Effort,
3. purchasing contract A and choosing No-effort.
5 In fact, with these values, the right-hand side of (10.9) is equal to 1.
10.2 Moral hazard and insurance 353

Let A = W1A ,W2A . Then it must be that E[UE (NI)] = E[UE (A)] and E[UE (A)] = E[UN (A)],

that is,

1 19 15
q q
48.0625 = W1A + W2A − and (10.10)
20 20 16
1 19 15 1 9
q q q q
W1A + W2A − = W1A + W2A . (10.11)
20 20 16 10 10

The solution is W1A = 972.66 and W2A = 2, 493.75; thus A is the contract with premium
2, 500 − 2, 493.75 = $6.25 and deductible 2, 493.75 − 972.66 = $1, 521.09.

 Suppose that the individual is offered contract A and she breaks her indifference by
purchasing the contract. What is the insurer’s expected profit from this contract?
With contract A the individual is indifferent between Effort and No-effort. Thus expected
profit will be
1
◦ h − pE (` − d) = 6.25 − 20 (1, 600 − 1, 521.09) = $2.31 if the individual chooses
Effort.
1
◦ h − pN (` − d) = 6.25 − 10 (1, 600 − 1, 521.09) = $ − 1.641 if the individual chooses
No-effort.

 Let us find the full-insurance contract, call it F, that makes the individual indifferent
between purchasing the contract and not insuring.
We saw above that, without insurance, the individual can achieve a level of utility of
48.0625 (by opting for Effort). On the other hand, with any full-insurance contract, the
individual will maximize√ her utility by choosing No-effort. Thus we are looking for a level
of wealth W such that W = 48.0625. The solution is W = 2, 310. Thus F = (2310, 2310),
that is, a contract with premium 2, 500 − 2, 310 = $190 and zero deductible.

 Suppose that the individual is offered contract F and she breaks her indifference by
purchasing the contract. What is the insurer’s expected profit from this contract?

1
Expected profit will be h − pN ` = 190 − 10 1600 = $30.

 What is the reservation utility locus for this individual?

It is the union of the following two curves: (1) the portion of the indifference curve
through NI corresponding to Effort, from NI to point A , followed by (2) the portion of the
indifference curve through A corresponding to No-effort from point A to point F.

In the above example, an insurer would be better off offering contract F than contract
A. We now turn to the issue of what contract(s) would be offered by a monopolist.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 10.3.2 at the end of this chapter.
 354 Chapter 10. Moral Hazard in Insurance

10.2.3 The profit-maximizing contract for a monopolist

In Chapter 5 (Section 5.2.1) we showed that – in the case where the probability of loss p
is fixed and thus there is no issue of moral hazard – a monopolist would offer only one
contract, namely the full-insurance contract with the maximum premium that the individual
is willing to pay for full insurance, namely hmax = p` + RNI , where p` is expected loss
and RNI is the risk premium associated with the no-insurance lottery. Such contract is
determine by the intersection of the reservation indifference curve and the 45o line.
In the case of moral hazard, the situation is more subtle. First of all, there are now
two sets of isoprofit lines: one set corresponds to the case where the individual chooses
pE
Effort (so that the probability of loss is pE and the slope of each isoprofit line is − 1−p E
)
and the other set corresponds to the case where the individual chooses No-effort (so that
pN
the probability of loss is pN and the slope of each isoprofit line is − 1−p N
). Since pE < pN ,
pE pN
1−pE < 1−p N
and thus the isoprofit lines in the former set are less steep than the ones in
the latter set. In order to use the correct isoprofit lines, the monopolist must first figure out
what choice of effort the individual will make.
As in the case considered in Chapter 5, the monopolist will want to offer a contract
that extracts the maximum surplus from the individual, that is, the contract that leaves the
individual just indifferent between insuring and not insuring.
As in the previous section, we will continue to assume that, if uninsured, the individual
will choose Effort, that is,
E[UE (NI)] > E[UN (NI)].
Then the monopolist’s problem is to find that contract on the reservation utility locus that
maximizes its profits. Figure 10.5 (which reproduces Figure 10.4) shows the reservation
utility locus (it is the union of the two thick, continuous curves).

Wealth in
good state
W2

NI 45o line
W0

indifference curve
with Effort
indifference curve
with No-effort

Wealth in
W1 bad state
0 W0 − `

Figure 10.5: The reservation utility locus is the union of the two thick, continuous curves.
10.2 Moral hazard and insurance 355

Let us begin by considering the first segment of the reservation utility locus, namely
the portion of the indifference curve through NI corresponding to Effort, from point NI
to point A (where, as before, A is the contract that yields the reservation utility no matter
whether the individual chooses Effort or No-effort). It is shown in Figure 10.6.

Wealth in
good state
W2
isoprofit line
pE
NI slope: − 1−p 45o line
W0 E

A
indifference curve
with Effort

Wealth in
W1 bad state
0 W0 − `

Figure 10.6: Point A is such that E[UN (A)] = E[UE (A)].

If the insurance company wants to offer a contract that

1. leaves the customer with no surplus (that is, her expected utility with the offered
contract is her reservation utility, call it ū) and
2. induces the customer to choose Effort,
then the insurance company has to offer a contract on this portion of the reservation
utility locus. Note that if the individual purchases a contract that lies on this portion of
the reservation utility locus then she will indeed choose Effort. In fact, since contract
A is such that the individual is indifferent between choosing Effort and No-effort, that
is, E[UE (A)] = E[UN (A)] = ū, any point B to the left of point A on this portion of the
reservation utility locus is such that E[UN (B)] < ū (since it lies on a lower N-indifference
curve than the N-indifference curve that goes through A). On the other hand, if offered
contract A the individual is indifferent between choosing Effort and No-effort; we will
assume that in this case she will choose Effort.6
Since, for any of the contracts being considered, the individual will choose Effort,
the relevant probability of loss is pE and thus the relevant isoprofit lines are those with
pE
slope − 1−p E
. Figure 10.6 shows one such isoprofit line, namely the one that goes through
point B. Recall from Chapter 5 (Section 5.1.4) that at any point above the 45o line the
Effort-indifference curve through that point is steeper than the pE -isoprofit-line. Thus
6 Without this assumption, instead of contract A the insurance company would offer a contract slightly to
the left of point A (on the portion of the curve under consideration) in order to provide the customer with an
incentive to choose Effort (and thus reduce the probability of loss). To simplify the exposition we assume
that, with contract A, the individual would choose Effort.
356 Chapter 10. Moral Hazard in Insurance

contract B cannot be profit-maximizing for the insurer, since there are points on this portion
of the reservation utility locus that lie below the isoprofit line through B and thus yield
higher profits than B.
Thus we conclude that, of all the points on the portion of the reservation utility locus
considered so far, point A represents the profit-maximizing contract.

Let us now turn to the other portion of the reservation utility locus, namely the segment
of the indifference curve through A corresponding to No-effort from point A to the 45o
line, shown in Figure 10.7, where the point of intersection between the indifference curve
and the 45o line is denoted by F.

Wealth in
good state
W2

NI 45o line
W0

C isoprofit line
pN
F slope: − 1−p
N

indifference curve
with No-effort

Wealth in
W1 bad state
0 W0 − `

Figure 10.7: Point A is such that E[UN (A)] = E[UE (A)].

Note that if the individual purchases a contract that lies on this portion of the reservation
utility locus then she will choose No-effort. In fact, since contract A lies on the indifference
curve corresponding to Effort and utility level ū = E[UE (NI)], any point to the right of A
on this portion of the reservation utility locus will be lie on an Effort-indifference curve
corresponding to a utility level less than ū (while it lies on the No-effort-indifference curve
corresponding to ū).7
Since, for any of the contracts being considered, the individual will choose No-effort,
the relevant probability of loss is pN and thus the relevant isoprofit lines are those with
pN
slope − 1−p N
. Figure 10.7 shows one such isoprofit line, namely the one that goes through
point C. Recall from Chapter 5 (Section 5.1.4) that at any point above the 45o line the
No-effort-indifference curve through that point is steeper than the pN -isoprofit-line. Thus
contract C cannot be profit-maximizing for the insurer, since there are points on this portion
of the reservation utility locus that lie below the isoprofit line through C and thus yield
7 As for point A we continue to assume that the individual would choose Effort.
10.2 Moral hazard and insurance 357

higher profits than C.

Thus we conclude that, of all the points on the portion of the reservation utility locus under
consideration, point F represents the profit-maximizing contract.

Let A = W1A ,W2A and F = W F ,W F so that the corresponding premia and de-

ductibles are
premium deductible
Contract A : hA = W0 −W2A dA = W2A −W1A
Contract F : hF = W0 −W F dF = 0

Furthermore, let πA be the expected profit from contract A and πF be the expected profit
from contract F, that is,

πA = hA − pE (` − dA )
πF = hF − pN `.

Then we can conclude from the above analysis that the monopolist will offer

• contract A if πA > πF ,

• contract F if πF > πA

and be indifferent between the two contracts if πA = πF .

In the numerical example considered at the end of the previous section, the monopolist
would opt for the full-insurance contract F, since πF = 30 and πA = 2.31 (as shown on
page 353).
Now we give an example where the monopolist prefers to offer the partial-insurance
contract A. Let us keep the same data as in the example considered at the end of the
1 1
previous section, but change the value of pE from 20 to 40 . Thus we have that the vNM
utility-of-money function is
( √
UN (m) = m if she chooses No-effort
√ 15
UE (m) = m − 16 if she chooses Effort

1 1
W0 = 2, 500 ` = 1, 600 pE = pN =
40 10

Note that, if uninsured, the individual would choose Effort, since

1√ 39 p 15
E[UE (NI)] = 900 + 2, 500 − = 48.5625
40 40 16
and
1√ 9p
E[UN (NI)] = 900 + 2, 500 = 48.
10 10
358 Chapter 10. Moral Hazard in Insurance

Contract A is given by the solution to

 q q
1
48.5625 =

 40 W1 + 40 W2A − 15
A 39
16

 q q
1 9
W1 + 10 W2A
A

48.5625 =

10

The solution is W1A = 1, 392.22 and W2A = 2, 481.29, that is, A = (1392.22, 2481.29). Thus
hA = 2, 500 − 2, 481.29 = 18.71 and dA = 2, 481.29 − 1, 392.22 = 1, 089.07.
√
Contract F is given by the solution to W F = 48.5625, which is W F = 2, 358.32 so that
F = (2358.32, 2358.32) and hF = 2, 500 − 2, 358.32 = 141.68. Hence8

1
πA = 18.71 − (1, 600 − 1, 089.07) = $5.94
40
1
πF = 141.68 − (1, 600) = $ − 18.32
10

so that the monopolist would offer contract A, thereby inducing the individual to reduce
the probability of loss by choosing Effort.

What would the monopolist’s profit be if it offered the full-insurance contract G which
is given by the intersection of the 45o line and the indifference curve corresponding to
Effort that goes through the NI point? In this example contract G is obtained by solv-
√ 15
ing W F − 16 = 48.5625, that is, G = (2450.25, 2450.25) with corresponding premium
hG = 2, 500 − 2450.25 = 49.75. Thus one might be tempted to answer that the monopo-
1
list’s profit would be 49.75 − 40 (1, 600) = $9.75, but this answer is wrong! The individual
would be very happy to purchase contract G because her best choice would then be
√
No-effort, with an expected utility of 2, 450.25 = 49.5 (instead of 48.5625) and thus, as
1
a matter of fact, the monopolist’s profits would turn out to be πG = 49.75 − 10 (1, 600) =
$ − 110.25!

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 10.3.3 at the end of this chapter.

8 Recall the assumption that, with contract A, the individual would choose Effort: see Footnote 6 on page
355.
 10.3 Exercises 359

10.3 Exercises

The solutions to the following exercises are given in Section 10.4 at the end of this chapter.

10.3.1 Exercises for Section 10.2.1: Two levels of unobserved effort

Exercise 10.1 Emily has an initial wealth of $80,000 and faces a potential loss of
$36,000. The probability of loss depends on the amount of effort she puts into trying to
avoid the loss. If she puts a high level of effort, then the probability is 5%, while if she
exerts low effort the probability is 15%. Her vNM utility-of-money function is
√
 m if low effort
U(m) = √
 m − 1 if high effort.

(a) If Emily remains uninsured, what level of effort will she choose?
(b) If Emily is offered a full insurance contract with premium $2,250 and she accepts
it, what level of effort will she choose?
(c) If Emily is offered a full insurance contract with premium $2,250 will she accept
it?
(d) What is the insurance company’s expected profit from a full insurance contract
with premium $2,250?


Exercise 10.2 Susan has an initial wealth of $10,000 and faces a potential loss of
$1,900. The probability of loss depends on the amount of effort she puts into trying to
1
avoid the loss. If she puts a high level of effort, then the probability is 10 , while if she
4
exerts low effort the probability is 10 . Her vNM utility-of-money function is
√
 m if low effort
U(m) = √
 m − 2 if high effort

(a) If Susan remains uninsured, what level of effort will she choose?
(b) If Susan is offered a partial insurance contract with premium $800 and deductible
$200 and she accepts it, what level of effort will she choose?
(c) If Susan is offered the contract of Part (b) will she accept it?

360 Chapter 10. Moral Hazard in Insurance

Exercise 10.3 Bob owns a house near Lake Tahoe. The house is worth $949,000.
He also has $1,000 in his bank account, so that his entire wealth is $950,000. The
1
probability that there will be a forest fire next year is 10 . If a forest fire occurs then the
house will incur damages equal to $400,000. However, by spending $x on protective
1
measures Bob can reduce the probability that the fire will reach the house from 10 to
1 x
10 − 15,000 . Thus the more he spends, the lower the probability. The most he can spend
is $1,000. Bob’s vNM utility of money function is U(m) = 10 ln(m).
(a) If Bob is not insured, which of the following four options will he choose (assuming
that these are the only options)?
(1) x = 0,
(2) x = $400,
(3) x = $750,
(4) x = $1, 000.

(b) If Bob is offered a full-insurance contract with premium h, what value of x will
he choose?

(c) Suppose that Bob is offered a full insurance contract with premium h = $40, 000.
Will he purchase it?


Exercise 10.4 In this exercise we consider the case where the individual’s effort has an
effect not on the probability of loss but on the size of the loss.
Mike’s initial wealth is $6,400 and he faces a potential loss with probability 14 . The
size of the loss depends on his choice of effort: it he chooses Effort then the loss is
`E = $471, while if he chooses No-effort then the loss is `N = $1, 216. Mike’s vNM
utility-of-money function is
 √
UN (m) = m if he chooses No-effort
UE (m) = √m − 1 if he chooses Effort

(a) If Mike is uninsured, will he choose Effort or No-effort?

(b) If Mike purchases a full-insurance contract with premium h, will he choose Effort
or No-effort?
(c) Suppose that Mike is offered an insurance contract with premium h = $80 and
deductible d = $471. Will he purchase it?

 10.3 Exercises 361

10.3.2 Exercises for Section 10.2.2: The reservation utility locus

Exercise 10.5 Consider again the information given in Exercise 10.2: Susan has an
initial wealth of $10,000 and faces a potential loss of $1,900; the probability of loss
depends on the amount of effort she puts into trying to avoid the loss; if she puts a high
1
level of effort, then the probability is 10 , while if she exerts low effort the probability is
4
10 ; her vNM utility-of-money function is
(√
m if low effort
U(m) = √
m− 2 if high effort

(a) What is Susan’s reservation level of utility?

(b) Write two equations whose solution gives that point, call it A, on the indifference
curve through NI that corresponds to high effort such that Susan’s expected utility
at A if she chooses low effort is equal to her reservation level of utility.
(c) Find a full-insurance contract, call it F, that yields Susan the same expected utility,
when she chooses low effort, as contract A of Part (b). Calculate the premium of
contract F
(d) Describe in words how you would draw the reservation utility locus for Susan.


Exercise 10.6 In this exercise we consider a slightly different type of vNM utility-of-
money function, where instead of UE (m) = UN (m) − c we have that UE (m) = αUN (m)
with 0 < α < 1.
Tom has an initial wealth of $4,900 and faces a potential loss of $1,300; the probability
of loss depends on the amount of effort he puts into trying to avoid the loss; if he puts a
high level of effort, then the probability is 18 , while if he exerts low effort the probability
is 38 ; his vNM utility-of-money function is
(√
m if low effort
U(m) = √ with 0 < α < 1.
α m if high effort

(a) For what values of α would Tom choose high effort if uninsured?
(b) Let α = 0.9. What is Tom’s reservation level of utility?
(c) Let α = 0.98. What is Tom’s reservation level of utility?
(d) Assume that α = 0.98. Write two equations whose solution gives that insurance
contract, call it A, such that if Tom purchases this contract he gets his reservation
level of utility, no matter whether he chooses low effort or high effort.
(e) Continue to assume that α = 0.98. Find a full-insurance contract, call if F, such
that Tom is indifferent between not insuring and purchasing contract F.
(f) Continue to assume that α = 0.98. Describe in words how you would draw the
reservation utility locus for Tom.

 362 Chapter 10. Moral Hazard in Insurance

10.3.3 Exercises for Section 10.2.3: The profit-maximizing contract

Exercise 10.7 Carol has an initial wealth of $10,000 and faces a potential loss of
$4,000; the probability of loss depends on the amount of effort she puts into trying to
1
avoid the loss; if she puts a high level of effort, then the probability is 10 , while if she
1
exerts low effort the probability is 2 ; her vNM utility-of-money function is

m

10 ln( 1,000 ) if low effort
U(m) = with c > 0.
m
10 ln( 1,000 )− c if high effort


(a) For what values of c would Carol choose high effort if uninsured?
(b) Assume that c = 1.8. What insurance contract would a monopolist offer to Carol?
(c) Assume that c = 1.5. What insurance contract would a monopolist offer to Carol?

 10.4 Solutions to Exercises 363

10.4 Solutions to Exercises

Solution to Exercise 10.1

(a) If Emily chooses low effort, her expected utility is

p p
0.15 80, 000 − 36, 000 + 0.85 80, 000 = 271.881

and if she chooses high effort, her expected utility is

p  p 
0.05 80, 000 − 36, 000 − 1 + 0.95 80, 000 − 1 = 278.189.

Thus she will choose high effort.

√
(b) Emily will choose low effort, because her √ utility will be 80, 000 − 2, 250 while
with high effort it would be less, namely 80, 000 − 2, 250 − 1.
√
(c) She will accept the contract, because her utility if she accepts it is 80, 000 − 2, 250 =
278.837, while her best alternative would be to remain uninsured and choose high
effort with an expected utility of 278.189.
(d) Since Emily will indeed buy insurance (and exert low effort), expected profits will
be 2, 250 − 0.15(36, 000) = $ − 3, 150, that is, a loss. Thus it would not be a good
idea for the insurance company to offer this contract. 

Solution to Exercise 10.2

(a) If Susan is uninsured and chooses high effort, her expected utility is
p  p 
0.1 10, 000 − 1, 900 − 2 + 0.9 10, 000 − 2 = 97

while if she chooses low effort, her expected utility is

p p
0.4 10, 000 − 1, 900 + 0.6 10, 000 = 96.

Thus she will choose high effort.

(b) The contract under consideration (premium of $800, deductible of $200) corresponds
to the following point in the wealth space: C = (9000, 9200). If Susan purchases
this contract and chooses high effort, then her expected utility is
p  p 
0.1 9, 000 − 2 + 0.9 9, 200 − 2 = 93.81

while if she chooses low effort, her expected utility is

p p
0.4 9, 000 + 0.6 9, 200 = 95.5.

Thus, under this contract, she would choose low effort.

(c) She will not accept the contract, because her highest utility if she accepts it is 95.5,
while her best alternative would be to remain uninsured and choose high effort with
an expected utility of 97. 
364 Chapter 10. Moral Hazard in Insurance

Solution to Exercise 10.3

(a) If Bob chooses to spend $x on preventive measures his expected utility – when not
insured – is
 
1 x
NI(x) = − 10 ln (950, 000 − 400, 000 − x)
10 15, 000
  
1 x
+ 1− − 10 ln (950, 000 − x).
10 15, 000
Since NI(0) = 137.096, NI(400) = 137.237, NI(750) = 137.361 and NI(1, 000) =
137.449, of the four options he will choose x = $1, 000.
(b) If he is fully insured at premium h then his utility, if he does not spend any money
on preventive measures, is 10 ln(950, 000 − h), while if he spends $x (with x > 0)
then his utility is less, namely 10 ln(950, 000 − h − x). Thus he will choose x = 0.
(c) As determined in Part (b), if he buys the full insurance contract with premium
$40, 000 then he will choose x = 0, so that his utility will be 10 ln(950, 000 −
40, 000) = 10 ln(910, 000) = 137.21. This is less than his expected utility if he
remains uninsured and spends $1, 000 on preventive measures, which is 137.449 as
calculated in Part (a). Thus he will not accept the contract. 

Solution to Exercise 10.4

(a) If Mike is uninsured and chooses No-effort then his expected utility is
1p 3p
E[UN (NI)] = 6, 400 − 1, 216 + 6, 400 = 78
4 4
while if he chooses Effort his expected utility is
1p 3p
E[UE (NI)] = 6, 400 − 471 + 6, 400 − 1 = 78.25.
4 4
Thus he will choose Effort, if not insured.
(b) With a full-insurance contract with premium h his utility if he chooses Effort is
√ √
6, 400 − h−1 while his utility if he chooses No-effort is higher, namely 6, 400 − h.
Thus he would choose No-effort (he does not care about the size of the loss, since he
gets fully reimbursed by the insurance company if the loss occurs).
(c) When the deductible is $471 (= `E , loss with Effort) it would not make sense for
Mike to purchase the contract and choose Effort, because he would be better off by
not insuring and choosing Effort (his wealth would be larger by an amount equal
to the premium). Thus he would compare expected utility from no insurance with
Effort (which was calculated to be 78.25 in Part (a)) with expected utility with the
contract and No-effort which is (recall that the premium is $80)
1p 3p
6, 400 − 80 − 471 + 6, 400 − 80 = 78.74.
4 4
Thus Mike would purchase the contract and choose No-effort. 
10.4 Solutions to Exercises 365

Solution to Exercise 10.5

(a) We saw in Part (a) of Exercise 10.2 that if Susan is uninsured and chooses low effort,
her expected utility is 96 and if she chooses high effort her expected utility is 97.
Thus her reservation level of utility is 97: she would not accept a contract that did
not give her an expected utility of at least 97.
(b) Let A = W1A ,W2A . The first equation says that A should lie on the indifference

curve through NI corresponding to high effort. Since, as computed in Part (a) of

Exercise 10.2, E[UE (NI)] = 97, the first equation is
q  q 
0.1 A
W1 − 2 + 0.9 A
W2 − 2 = 97.

The second equation says that contract A gives the same expected utility whether
Susan chooses high effort or low effort:
q q q  q 
A
0.4 W1 + 0.6 W2 = 0.1A A
W1 − 2 + 0.9 A
W2 − 2 .

The solution of these two equations is: W1A = 8, 649 and W2A = 9, 933.44. Thus A is
the contract with premium 10, 000 − 9, 933.44 = $66.56 and deductible
9, 933.44 − 8, 649 = $1, 284.44.
(c) We are looking for a level of wealth W such that
√
W = 97.

The solution is W = 9, 409. Thus F = (9409, 9409), that is, a contract with premium
10, 000 − 9, 409 = $591 and zero deductible.
(d) The reservation utility locus for Susan is the union of the following two curves:
(1) the portion of the indifference curve through NI corresponding to high effort
from NI to point A, followed by (2) the portion of the indifference curve through A
corresponding to low effort from point A to point F. 

Solution to Exercise 10.6

(a) When uninsured, Tom’s expected utility is:

3p 5√ 265
if low effort: 3, 600 + 4900 = = 66.25
8 8 4
1 p 7 √ 275
if high effort: α 3, 600 + α 4900 = α = 68.75α.
8 8 4
Thus, when uninsured, Tom will choose high effort if 68.75α > 66.25, that is, if
α > 53
55 = 0.9636.
(b) The assumption is that α = 0.9. Since in this case α < 0.9636, Tom – when
uninsured – will choose low effort. Thus his reservation level of utility is 66.25, as
computed in Part (a).
366 Chapter 10. Moral Hazard in Insurance

(c) The assumption is that α = 0.98. Since in this case α > 0.9636, Tom – when
uninsured – will choose high effort. Thus his reservation level of utility is 68.75(0.98)
= 67.375, as computed in Part (a).

(d) Let the contract be A = W1A ,W2A . The equations are:

1 7 3 5
q q q q
(0.98) W1A + (0.98) W2A = W1A + W2A
8 8 8 8

1 7
q q
(0.98) W1A + (0.98) W2A = 67.375.
8 8
The solution is W1A = 4, 088 and W2A = 4, 821.57.

√ assumption is that α = 0.98. Contract F is given by the solution to:

(e) The
W = 67.375 which is W = 4, 539.39. Thus F = (4539.39, 4539.39).
(f) The reservation utility locus for Tom is the union of the following two curves: (1) the
portion, from point NI to point A, of the indifference curve through NI corresponding
to high effort followed by (2) the portion, from point A to point F, of the indifference
curve through A corresponding to low effort. 

Solution to Exercise 10.7

(a) If Carol is uninsured and exerts low effort, her expected utility is:
1 1
10 ln(6) + 10 ln(10) = 20.4717
2 2
while with high effort her expected utility is
1 9
(10 ln(6) − c) + (10 ln(10) − c) = 22.515 − c.
10 10
Thus she will choose high effort if 20.4717 < 22.515 − c, that is, if c < 2.0433.
(b) The assumption is that c = 1.8, so that – if uninsured – Carol would choose high
effort and her expected utility would be 22.515 − 1.8 = 20.715. The monopolist
would only consider two options:
• the partial-insurance contract A given by the intersection of the low-effort
indifference curve corresponding to a utility of 20.715 and the high-effort
indifference curve corresponding to a utility of 20.715, and
• the full-insurance contract F that makes Carol indifferent between (1) purchas-
ing F and exerting low effort and (2) not insuring and exerting high effort.
Contract A is given by the solution to the following two equations:
       
1 W1 1 W2 1 W1 9 W2
2 10 ln 1,000 + 2 10 ln 1,000 = 10 10 ln 1,000 + 10 10 ln 1,000 − 1.8

   
1 W1 W2
2 10 ln 1,000 + 21 10 ln 1,000 = 20.715
10.4 Solutions to Exercises 367

which is W1 = 6, 337.6 and W2 = 9, 939.33, so that A = (6337.6, 9939.33) with corre-

sponding premium hA = 10, 000 − 9, 939.33 = $60.67 and deductible
dA = 9, 939.33 − 6, 337.6 = 3, 601.73.
 
W
Contract F is given by the solution to 10 ln 1,000 = 20.715, which is
W = 7, 936.72; thus F = (7936.72, 7936.72) with corresponding premium
hF = 10, 000 − 7, 936.72 = $2, 063.28
Assuming that with contract A Carol would choose high effort, the expected profit
1 1
with contract A is hA − 10 (` − dA ) = 60.67 − 10 (4, 000 − 3, 601.73) = $20.84. On
the other hand, expected profit from Contract F is hF − 12 ` = 2, 063.28 − 12 (4, 000) =
$63.28. Thus the monopolist would offer the full-insurance contract F.
(c) The assumption is that c = 1.5, so that – if uninsured – Carol would choose high
effort and her expected utility would be 22.515 − 1.5 = 21.015. The monopolist
would only consider two options:
• the partial-insurance contract A given by the intersection of the low-effort
indifference curve corresponding to a utility of 21.015 and the high-effort
indifference curve corresponding to a utility of 21.015, and
• the full-insurance contract F that makes Carol indifferent between (1) purchas-
ing F and exerting low effort and (2) not insuring and exerting high effort.
Contract A is given by the solution to the following two equations:
       
1 W1 1 W2 1 W1 9 W2
2 10 ln 1,000 + 2 10 ln 1,000 = 10 10 ln 1,000 + 10 10 ln 1,000 − 1.5

   
1 W1 1 W2
2 10 ln 1,000 + 2 10 ln 1,000 = 21.015

which is W1 = 6, 780.16 and W2 = 9, 865.07, so that A = (6780.16, 9865.07) with

corresponding premium hA = 10, 000 − 9, 865.07 = $134.93 and deductible
dA = 9, 865.07 − 6, 780.16 = 3, 084.91.
 
W
Contract F is given by the solution to 10 ln 1,000 = 21.015, which is
W = 8, 178.43; thus F = (8178.43, 8178.43) with corresponding premium
hF = 10, 000 − 8, 178.43 = $1, 821.57
Assuming that with contract A Carol would choose high effort, the expected profit
1 1
with contract A is hA − 10 (` − dA ) = 134.93 − 10 (4, 000 − 3, 084.91) = $43.42.
On the other hand, expected profit from Contract F is hF − 12 ` = 1, 821.57 −
1
2 (4, 000) = $−178.43, a loss. Thus the monopolist would offer the partial-insurance
contract A. 
 11. Moral Hazard in Principal-Agent

11.1 Moral hazard in Principal-Agent relationships

In Chapter 6 we considered the case of a party, called the Principal, who intends to hire
another party, called the Agent, to perform some activity, whose outcome is uncertain,
because it is affected by external factors that cannot be controlled by either party. In that
chapter we assumed that the uncertainty was completely exogenous, that is, not affected in
any way by the parties involved, and we focused on the issue of efficient risk sharing.
In this chapter we want to extend the analysis by considering the possibility that the
actions of the Agent might have some effect on the probabilistic outcomes. That is, we
want to add moral hazard to that framework. Moral hazard is present whenever the Agent’s
behavior cannot be observed by the Principal.
The following table provides several examples of this situation.

Principal Agent Agent’s unobserved action

Owner of the firm Manager Amount of time/effort
spent running the firm
Client Lawyer Amount of time/care
devoted to the case
Client Doctor Amount of time/care devoted to
studying patient’s symptoms
Insurance Company Policyholder Care to avoid theft/loss
Land owner Farmer Farming effort
Landlord Tenant Upkeep of building
 370 Chapter 11. Moral Hazard in Principal-Agent

Although, typically, the Agent cannot fully control the outcome, she can have some
influence on it by exerting sufficient effort. For example, while a lawyer cannot guarantee
that her client will win the lawsuit, she can make a victory more likely if she devotes a lot
time and effort to preparing the case. The lawyer’s behavior cannot be observed by the
client, but the outcome of the lawsuit is observed and thus the client might want to make
his payment to the lawyer dependent on the outcome, in order to provide the lawyer with
the incentive to work hard. Examples of payments that are contingent on the observed
outcome are: contingency fees for lawyers (if the client loses he owes nothing to the lawyer,
if he wins he owes a percentage of what he has been awarded), stock options for managers
(part of their salaries is in the form of a percentage of the profits of the firm), deductibles
for policy-holders, etc.

11.2 Risk sharing under moral hazard

In this section we show that, in the presence of moral hazard, the principles of optimal risk
sharing discussed in Chapter 6 no longer hold.
We shall use the notation shown in the following table:

Table 11.1: Notation

Xi (i ∈ {1, 2, . . . , n}) Possible monetary outcomes

X1 < X2 < · · · < Xn

wi (i ∈ {1, 2, . . . , n}) Payment to the Agent if Xi occurs

UP (m) Principal’s vNM

UP0 (m) > 0, U 00 (m) ≤ 0 utility-of-money function

e Agent’s effort

UA (m, e) Agent’s vNM utility function

∂UA (m,e) ∂ 2UA (m,e) ∂UA (m,e)
∂m > 0, ∂ m2
≤ 0, ∂e <0

u Agent’s reservation utility

The Agent’s reservation level of utility u is that level of utility that the Agent can obtain
with an “outside option”, that is, by not signing a contract with the Principal. Thus any
contract that is signed must give the Agent an expected utility of at least u.
11.2 Risk sharing under moral hazard 371

We shall assume that the Agent’s effort influences the outcome, in the sense that greater
effort increases the probability of better outcomes. Consider first the case where there are
only two outcomes: X1 and X2 (with X1 < X2 ), and two possible levels of effort: low-effort
eL and high effort eH . Let pL1 be the probability of the lower outcome (outcome X1 ) when
the Agent exerts low effort (so that the probability of the higher outcome is (1 − pL1 )) and
pH1 be the probability of the lower outcome (outcome X1 ) when the Agent exerts high effort
(so that the probability of the higher outcome is (1 − pH 1 )). Then the assumption is that

pL1 > pH
1.

Let us begin with a simple example. The Principal is risk neutral:

UP (m) = m,

while the Agent is risk averse, for any level of effort, and dislikes effort:
√
UA (m, e) = m − e.

There are two possible levels of effort for the Agent: eL = 10 and eH = 12. The Agent is
currently employed elsewhere at a job that involves low effort eL and pays a salary of $169,
so that her reservation level of utility is
√
u = 169 − eL = 13 − 10 = 3. (11.1)

Let
X1 = $1, 500 and X2 = $9, 000
and suppose that, if a contract is signed and the Agent exerts low effort, then the probability
of the lower outcome is 80%, while if the Agent exerts high effort then the probability of
the lower outcome is 50%:
4 1
pL1 = and pH
1 = .
5 2
We know from Chapter 6 that when one party is risk neutral and the other is risk averse,
Pareto efficient risk sharing requires that all the risk be borne by the risk-neutral party;
in other words, the risk-averse party (in this case the Agent) ought to be given a fixed
income: w1 = w2 . However, this principle clashes with the need to provide the Agent
with an incentive
√ to exert high effort: if √
the Agent is given a sure payment of $w then her
utility is w − 10 is she choose eL and w − 12 is she choose eH and therefore she will
choose eL . The Principal realizes this and figures out that the best (from his point of view)
fixed-salary contract that he can offer the Agent is one that gives the Agent exactly her
reservation utility, that is, contract B = (w1 = 169, w2 = 169), in which case the Principal
will face the money lottery
 
 $(1, 500 − 169) $(9, 000 − 169) 
 
 
4 1
5 5
372 Chapter 11. Moral Hazard in Principal-Agent

whose expected utility for the Principal is

E[UP (B)] = 45 (1, 500 − 169) + 15 (9, 000 − 169) = 2, 831. (11.2)

Consider now the following alternative contract, which exposes the Agent to risk:
C = (w1 = 100, w2 = 484). Will the Agent accept this contract and, if she does, what
level of effort will she exert? If the Agent exerts low effort, then she faces the money
lottery  
 $100 $484 
 
 
4 1
5 5

whose expected utility for her is

√  √ 
4 1
E[UA (C, eL )] = 100 − 10 + 5
5 484 − 10 = 2.4

which is below her reservation utility of 3.

If the Agent exerts high effort, then she faces the money lottery
 
 $100 $484 
 
 
1 1
2 2

whose expected utility for her is

√  √ 
1
E[UA (C, eH )] = 2 100 − 12 + 12 484 − 12 = 4 (11.3)

which exceeds her reservation utility. Hence the Agent will accept contract C and exert
high effort.
Realizing this, the Principal knows that if he signs contract C then he will face the money
lottery  
 $(1, 500 − 484) $(9, 000 − 100) 
 
 
1 1
2 2

whose expected utility for him is

E[UP (C)] = 12 (9, 000 − 484) + 21 (1, 500 − 100) = 4, 958. (11.4)

By (11.3) the Agent prefers contract C to contract B and by (11.2) and (11.4) also the
Principal prefers C to B, so that B is not Pareto efficient, despite the fact that it allocates
risk optimally. Hence, when moral hazard is present, the principles of optimal risk sharing
established in Chapter 6 no longer hold.
Let us now consider an example with three possible outcomes and three possible levels
of effort. As before, the Principal is risk neutral and the Agent has the following utility
11.2 Risk sharing under moral hazard 373
√
function: UA (m, e) = m − e with e ∈ {10, 12, 24}. The Agent’s reservation utility is 3.
The outcomes and corresponding probabilities are as follows:

outcome : $196 $1, 000 $5, 000

1 1 1
probability if e = 10 : 2 4 4

1 1 1
probability if e = 12 : 4 4 2

1 1 5
probability if e = 24 : 8 4 8

Consider the following contracts:

C = (w1 = 196, w2 = 196, w3 = 196) , D = (w1 = 0, w2 = 200, w3 = 1, 680) .
Under Contract C the Agent’s utility is:
√
• if she chooses e = 10: 196 − 10 = 4,
√
• if she chooses e = 12: 196 − 12 = 2,
√
• if she chooses e = 14: 196 − 24 = −10.
Thus she would choose e = 10.
Under Contract D the Agent’s expected utility is:
• if she chooses e = 10:
√  √  p 
1 1 1
2 0 − 10 + 4 200 − 10 + 4 1, 680 − 10 = 3.78
• if she chooses e = 12:
√  √  p 
1 1 1
4 0 − 12 + 4 200 − 12 + 2 1, 680 − 12 = 4.03
• if she chooses e = 24:
√  √  p 
1 1 5
8 0 − 24 + 4 200 − 24 + 8 1, 680 − 24 = 5.15
Thus she would choose e = 24.

Hence the Principal’s expected utility is:

1 1 1


• under contract C: 2 196 + 4 1, 000 + 4 5, 000 − 196 = 1, 402

• under contract D: 18 (196 − 0) + 41 (1, 000 − 200) + 58 (5, 000 − 1, 680) = 2, 299.5.
Thus both Principal and Agent prefer contract D to contract C so that the latter is not
Pareto efficient, despite the fact that it allocates risk optimally, according to the principles
established in Chapter 6.
What are the Pareto efficient contracts in a situation where there is moral hazard? We
will approach this question in several steps.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 11.5.1 at the end of this chapter.
 374 Chapter 11. Moral Hazard in Principal-Agent

11.3 The case with two outcomes and two levels of effort
In this section we assume that there are only two outcomes: X1 and X2 (with X1 < X2 ), and
two possible levels of effort: low-effort eL and high effort eH . Let pL1 be the probability of
X1 when the Agent exerts low effort and pH 1 the probability of X1 when the Agent exerts
high effort. We assume that

pL1 > pH
1. (11.5)

We shall also assume that the Principal is risk neutral, so that his vNM utility-of-money
function can be taken to be the identity function:

UP (m) = m,

while the Agent is risk-averse and dislikes effort; in particular, we take her utility function
to be1
(
U(m) if e = eL
UA (m, e) = c > 0, U 0 (m) > 0, U 00 (m) < 0. (11.6)
U(m) − c if e = eH

The reader might have noticed the similarity between the situation under consideration
and the insurance context analysed in Chapter 10 (compare (10.2) and (10.3) to (11.6)).
Also in this case we can view a possible contract as a point in the positive quadrant of
the (w1 , w2 ) Cartesian plane and we can draw indifference curves for the Agent and the
Principal.
Let us begin with the Agent. As in the context studied in Chapter 10, there are two
indifference curves for the Agent that go through a point (ŵ1 , ŵ2 ):

 one - corresponding to high effort eH - that joins all the contracts (w1 , w2 ) that yield
the same utility to the Agent as contract (ŵ1 , ŵ2 ) if the Agent chooses eH , and

 one - corresponding to low effort eL - that joins all the contracts (w1 , w2 ) that yield
the same utility to the Agent as contract (ŵ1 , ŵ2 ) if the Agent chooses eL .

We shall call the indifference curve corresponding to eH the “H-indifference curve” and
the indifference curve corresponding to eL the “L-indifference curve”. As explained in
Chapter 10, at any point (w1 , w2 ),

 the slope of the H-indifference curve is

pH U 0 (w1 )
 
− 1 H (11.7)
1 − p1 U 0 (w2 )

 the slope of the L-indifference curve is

pL U 0 (w1 )
 
− 1 L (11.8)
1 − p1 U 0 (w2 )
1 One can start with UA (m, eL ) = V (m) − eL and UA (m, eH ) = V (m) − eH and then define U(m) = V (m) +
eL (recall from Chapter 3 that adding a constant is an “allowed transformation" of a vNM utility function) so
that UA (m, eL ) = U(m) and UA (m, eH ) = V (m) − (eH − eL ) and define c = eH − eL and thus obtain (11.6).
11.3 The case with two outcomes and two levels of effort 375

It follows from (11.5) that (see Footnote 2 on page 345)

pL1 pH
1
L
> (11.9)
1 − p1 1 − pH
1

that is,

the indifference curve through point (w1 , w2 ) corresponding to high effort eH

is, at point (w1 , w2 ), less steep than,
the indifference curve through point (w1 , w2 ) corresponding to low effort eL .

This fact is illustrated in Figure 11.1.

w2
(payment if X2 )

45o line

eH

eL
w1
(payment if X1 )
0

Figure 11.1: At point A, the indifference curve corresponding to high effort eH is less steep
than the indifference curve corresponding to low effort eL .

The 45o line represents the fixed-wage contracts, that is, contracts (w1 , w2 ) with
w1 = w2 .
Note that, in the case of insurance (Chapter 10), we restricted attention to contracts
above the 45o line, because - in that context - a contract below the 45o line could not be
interpreted as an insurance contract. On the other hand, in the Principal-Agent case, a
contract below the 45o line is a meaningful contract and should not be ruled out in principle.
However, by the following proposition, such contracts will not be observed in practice (if
the parties are rational) since they are Pareto inefficient.

Proposition 11.3.1 Any contract (w1 , w2 ) such that w1 > w2 is Pareto inefficient.
376 Chapter 11. Moral Hazard in Principal-Agent

Proof. The proof of Proposition 11.3.1 is done in two steps.

Step 1. We first show that, if offered a contract below the 45o line, the Agent will choose
low effort eL (assuming that she finds the contract acceptable, that is, that the contract gives
her at least her reservation utility). We can prove this either algebraically or graphically.
The algebraic proof is as follows. Let C = (w1 , w2 ) be such that
w1 > w2 . (11.10)
Then

E[UA (C, eL )] − E[UA (C, eH )]

= pL1 U(w1 ) + 1 − pL1 U(w2 ) − pH H



1 U(w1 ) + 1 − p1 U(w2 ) − c

= pL1 − pH L H



1 U(w1 ) − p1 − p1 U(w2 ) + c

= pL1 − pH


1 [U(w1 ) −U(w2 )] + c. (11.11)

By (11.5), pL1 − pH


1 > 0; by assumption, c > 0; by (11.10) and the fact that U(m) is
increasing in m, [U(w1 ) −U(w2 )] > 0. Thus (11.11) is positive, that is, E[UA (C, eL )] >
E[UA (C, eH )].
Graphically, consider contract C in Figure 11.2.

w2
(payment if X2 )

45o line

D
mL
B
mH C
wC2 eH
eL
w1
mH mL (payment if X1 )
0 wC1

Figure 11.2: E[UA (C, eH )] = E[UA (B, eH )] = U(mH ) − c; E[UA (C, eL )] = E[UA (B, eL )] =
U(mL ).

By definition of indifference curve,

11.3 The case with two outcomes and two levels of effort 377

◦ E[UA (C, eH )] = E[UA (B, eH )] = U(mH ) − c, and

◦ E[UA (C, eL )] = E[UA (D, eL )] = U(mL ).
Since mH < mL , U(mH ) < U(mL ).
Thus E[UA (C, eH ) < E[UA (C, eL ).
Step 2. Next we show that, for every contract C below the 45o line, there is a contract
D on the 45o line such that: (1) the Agent is indifferent between C and D and (2) the
Principal prefers D to C, so that C is not Pareto efficient. Figure 11.3 reproduces Figure
11.2 without the H-indifference curve, since in the previous step we showed that with
contract C the Agent will choose low effort eL . Let D be the contract at the intersection
of the L-indifference curve through C and the 45o line and let mL be the fixed wage of
contract D.

w2
(payment if X2 )

45o line

D
mL
C
wC2
eL
w1
0 mL wC1
(payment if X1 )

Figure 11.3: Contract D is Pareto superior to contract C.

 
wC1 wC2
For the Agent, contract C corresponds to the lottery C =   whose expected
pL1 1 − pL1
value is E[C] = pL1 wC1 + 1 − pL1 wC2 . By definition of risk-aversion, the Agent strictly

prefers E[C] for sure to lottery C, that is, UA (E[C], eL ) = U(E[C]) > E[UA (C, eL )] and,
since D is on the same indifference curve as C, E[UA (C, eL )] = E[UA (D, eL )] = U(mL ).
Thus U(E[C]) > U(mL ), which implies, since U(m) is increasing in m, that

E[C] = pL1 wC1 + 1 − pL1 wC2 > mL .

(11.12)
378 Chapter 11. Moral Hazard in Principal-Agent
!
X1 − wC1 X2 − wC2
For the Principal contract C corresponds to the lottery whose
pL1 1 − pL1
expected value (and expected utility for the Principal, since he is risk neutral) is
 
 
pL1 X1 − wC1 + 1 − pL1 X2 − wC2 = pL1 X1 + 1 − pL1 X2 − E[C]


(11.13)

On the other hand,! for the Principal contract D corresponds to the lottery
X1 − mL X2 − mL
whose expected value (and expected utility) is
pL1 1 − pL1

pL1 X1 + 1 − pL1 X2 − mL .


(11.14)

It follows from (11.12), (11.13) and (11.14) that the Principal strictly prefers contract D to
contract C. Thus contract C is Pareto inefficient. 

It follows from Proposition 11.3.1 that – if we are interested in Pareto efficiency – we

can restrict attention to contracts on or above the 45o line.
We now want to determine which contracts on or above the 45o line are Pareto efficient.
Consider contract C in Figure 11.4.

w2
(payment if X2 )

eH eL 45o line

wC2 C
mH E
mL D

w1
mL mH (payment if X1 )
0 wC1

Figure 11.4: What level of effort will the Agent choose under contract C?

What level of effort will the Agent choose under contract C? The expected utility from
contract C under low effort eL is the same as with contract D under low effort and the latter
is U(mL ). The expected utility from contract C under high effort eH is the same as with
contract E under high effort and the latter is U(mH ) − c. Thus
11.3 The case with two outcomes and two levels of effort 379

- the Agent will choose eL if U(mH ) −U(mL ) < c,

- the Agent will choose eH if U(mH ) −U(mL ) > c,
- the Agent will be indifferent between eL and eH if U(mH ) −U(mL ) = c.
Now fix a level of utility û for the Agent, with û ≥ u (recall that u is the Agent’s reservation
utility: she will not accept a contract that gives her a utility below u). We want to find the
set of contracts on or above the 45o line that give the Agent utility û when she chooses the
best level of effort for each of these contracts. Following the terminology of Chapter 10,
we will call it the û-utility-locus.
Let m̂ be the solution to the equation U(m) = û (that is, U(m̂) = û). Then one contract
on the û-utility-locus is the fixed-wage contract D̂ = (w1 = m̂, w2 = m̂) and we know that
under such contract the Agent will choose low effort eL . Draw the two indifference curves
through contract D̂: see Figure 11.5.

w2
(payment if X2 )

eH eL 45o line
eH
eH
wC2 C
E
m̂ D̂

w1
(payment if X1 )
0 wC1 m̂

Figure 11.5: The two indifference curves through point D̂ and two more H-indifference
curves.
The H-indifference curve that goes through point D̂ (the continuous less steep curve)
corresponds to a level of utility less than û (namely U(m̂) − c = û − c < û). Consider
now the dashed H-indifference curve: it is above the one that goes through D̂ and thus
corresponds to higher utility. As we consider higher and higher H-indifference curves the
level utility keeps increasing. At some stage we must reach an H-indifference curve that
corresponds to utility û. Suppose it is the dotted curve in Figure 11.5. Let C = (wC1 , wC2 ) be
the point of intersection between this H-indifference curve and the L-indifference curve
that goes through D̂. Then it must be that E[UA (C, eH )] = E[UA (C, eL )] = û, that is,
pH C H C L C L C



1 U(w1 ) + 1 − p1 U(w2 ) − c = p1 U(w1 ) + 1 − p1 U(w2 ) = û.
380 Chapter 11. Moral Hazard in Principal-Agent

Once we reach point C, in order to keep utility constant at the value û we need to stop
traveling along the L-indifference curve and switch to the dotted H-indifference curve.
Thus the û-utility-locus if given by the union of two segments of two indifference curves,
just like in the case of insurance.
Figure 11.6 shows the û-utility-locus, where û = UA (m̂, eL ) = U(m̂), as the union of
the two thick curves.

wage if X2
w2

45o line

m̂ indifference curve
D̂ with high effort eH
indifference curve
with low effort eL

w1 wage if X1
0 m̂

Figure 11.6: The û-utility-locus (where û = UA (m̂, eL ) = U(m̂)) is the union of the two
thick curves.

What contracts on the û-utility-locus are Pareto efficient? We will show that the only
candidates are contracts C and D̂ in Figure 11.6.
So far we have focused on the indifference curves of the Agent. Now we turn to the
Principal, whom we have assumed to be risk neutral. Recall from Chapter 5 (Section 5.1.1)
that the indifference curves of a risk-neutral individual are downward-sloping straight lines
p
with slope − 1−p , where p is the probability of the outcome measured on the horizontal
axis. In the present context, there are two possible values of p: pL , if the Agent chooses
low effort eL , and pH , if the Agent chooses high effort eH , with pL > pH . Thus through
any contract C on or above the 45o line we can draw two straight-line indifference curves
for the Principal:
pL
• a steeper one with slope − 1−p L
, connecting all the contracts that give the same utility
to the Principal, conditional on the Agent choosing eL , and
pH
• a less steep one with slope − 1−p H
, connecting all the contracts that give the same
utility to the Principal, conditional on the Agent choosing eH .
11.3 The case with two outcomes and two levels of effort 381

Which indifference curve for the Principal is relevant at contract C depends on the choice
of effort of the Agent under contract C. Figure 11.7 shows the two indifference curves of
the Principal through a point C.

wage if X2
w2

45o line

if eH

if eL
w1 wage if X1
0

Figure 11.7: The two possible indifference curves of the Principal through contract C.

With an argument similar to the one used in Chapter 2 (Section 2.5), one can show that

1. given two contracts A and B, both of which would induce the Agent to choose
high effort eH , if B lies below the Principal’s indifference curve through A then the
Principal gets higher utility from B than from A, that is, he prefers B to A,

2. given two contracts D and E, both of which would induce the Agent to choose
low effort eL , if E lies below the Principal’s indifference curve through D then the
Principal gets higher utility from E than from D, that is, he prefers E to D.

The reader is asked to prove Point 1 in Exercise 11.6 and Point 2 in Exercise 11.7. The
above two claims are illustrated in Figures 11.8 and 11.9.
382 Chapter 11. Moral Hazard in Principal-Agent

wage if X2
w2

45o line

B if eH

w1 wage if X1
0

Figure 11.8: Assuming that the Agent would choose eH under both contract A and contract
B, the Principal prefers contract B to contract A.

wage if X2
w2

45o line

if eL
w1 wage if X1
0

Figure 11.9: Assuming that the Agent would choose eL under both contract D and contract
E, the Principal prefers contract E to contract D.
11.3 The case with two outcomes and two levels of effort 383

Let us focus on an L-indifference curve for the Agent and consider two points: point
F = (wF , wF ) lying on the 45o line and point D = (wD D o
1 , w2 ) lying above the 45 line (so
that wD D
1 < w2 ), and assume that, under contract D, the Agent would choose low effort eL .
By (11.8) (page 374), the slope of the Agent’s indifference curve at point F is
pL1 U 0 (wF ) pL1
− = − ;
1 − pL1 U 0 (wF ) 1 − pL1
thus, at point F, the Principal’s L-indifference curve (which is the relevant one, since we
know that at a point on the 45o line the Agent chooses low effort eL ) is tangent to the
Agent’s L-indifference curve. On the other hand, the slope of the Agent’s indifference at
point D is
pL U 0 (wD )
− 1 L 0 1D .
1 − p1 U (w2 )
Since wD D 0 00
1 < w2 and U (m) is decreasing in m (because U (m) < 0, by risk aversion),
U 0 (wD
1)
U 0 (wD
> 1 and thus
2)
pL1 U 0 (wD 1) pL1
>
1 − pL1 U 0 (wD
2) 1 − pL1
so that the Principal’s indifference curve through point D is less steep than the Agent’s
indifference curve through D (recall the assumption that, under contract D, the Agent
would choose eL ).
Figure 11.10 shows the relative slopes of the indifference curves of Principal and Agent
at a point above the 45o line (point D) and at a point on the 45o line (point F).

wage if X2
w2

eL 45o line
D

w1 wage if X1
0

Figure 11.10: The relative slopes of the indifference curves of Agent and Principal for the
case where the Agent chooses low effort eL .
384 Chapter 11. Moral Hazard in Principal-Agent

It follows from the above that a contract on an L-indifference curve of the Agent which
is above the 45o line (such as point D in Figure 11.10) is not Pareto efficient (as long
it yields higher utility to the Agent under low effort than under high effort): there are
contracts in the region between the two curves - such as contract E in Figure 11.10 -
that are better for both the Agent and the Principal. On the other hand, a contract at the
intersection of an L-indifference of the Agent and the 45o line (such as point F in Figure
11.10) is a candidate for Pareto efficiency.
Next we use a similar argument for contracts that are on an H-indifference curve of the
Agent and above the 45o line. Fix such a contract, call it A, and assume that under contract
A the Agent is better off with high effort eH than with low effort eL . Figure 11.11 shows
the relative slopes of the indifference curves of Principal and Agent at a point above the
45o line (point A) for a contract that induces the Agent to choose high effort eH . Since the
H-indifference curve of the Principal through point A is less steep than the H-indifference
curve of the Agent at point A, there are contracts between the two (such as point B in
Figure 11.11) that are better for both. Thus points such as point A in Figure 11.11 are not
Pareto efficient (given the hypothesis that under contract A the Agent would choose high
effort eH ).

wage if X2
w2

eH 45o line
A
B

w1 wage if X1
0

Figure 11.11: The relative slopes of the indifference curves of Principal and Agent at a
point above the 45o line (point A) for a contract that induces the Agent to choose high
effort eH .
11.3 The case with two outcomes and two levels of effort 385

We can now put together Figures 11.6 (reproduced below as Figure 11.12), 11.10 and
11.11 to complete the argument that, for any level of utility û of the Agent, there are only
two contracts on the û-utility-locus that are candidates for Pareto efficiency.
The top-left shaded area in Figure 11.12 consists of contracts that would induce the
Agent to choose high effort eH (since they are above the H-indifference curve for level
of utility û and to the left of the L-indifference curve for level of utility û), while the
bottom-right shaded area consists of contracts that would induce the Agent to choose low
effort eL (these contracts are above the L-indifference curve for level of utility û and below
the H-indifference curve for level of utility û). Thus the argument illustrated in Figure
11.11 rules out all the contracts on the H-indifference curve (for level of utility û) to the left
of point C (the first segment of the û-utility-locus) and the argument illustrated in Figure
11.10 rules out all the contracts on the L-indifference curve (for level of utility û) to the
right of point C (the second segment of the û-utility-locus) except for point D̂ on the 45o
line.

wage if X2
w2

eH 45o line
eL

m̂
D̂

w1 wage if X1
0 m̂

Figure 11.12: Copy of the û-utility-locus of Figure 11.6. The top-left shaded area consists
of contracts that would induce the Agent to choose high effort eH , while the bottom-right
shaded area consists of contracts that would induce the Agent to choose low effort eL .

R Recall that, at any contract F on the 45o line, the (straight-line) L-indifference curve
of the Principal is tangent to the (convex) L-indifference curve of the Agent (see
Figure 11.10). Hence any contract E that induces the Agent to choose low effort eL
and gives the Agent higher utility than F must lie above the Agent’s L-indifference
curve through F and thus must be worse for the Principal than F.
386 Chapter 11. Moral Hazard in Principal-Agent

From now on we will assume that, given a contract under which the Agent is indifferent
between choosing low effort eL and choosing high effort eH , the Agent will choose high
effort eH .2
Lemma 11.1 Let D̂ = (m̂, m̂) be a contract on the 45o line and let û be the utility of the

C C

D̂ (the Agent will choose eL under this contract and thus û = U(m̂)).
Agent under contract
Let C = w1 , w2 be a contract that gives utility û to the Agent no matter whether she
chooses eL or eH . Then
1. if D̂ P C (that is, the Principal prefers contract D̂ to contract C), then D̂ P E
(the Principal prefers D̂ to E) for every contract E 6= D̂ on the û-utility-locus of the
Agent,
2. if C P D̂ (that is, the Principal prefers contract C to contract D̂), then C P E (the
Principal prefers C to E) for every contract E 6= C on the û-utility-locus of the Agent.
Proof. .
1. Assume that D̂ P C. Pick an arbitrary contract E above the 45o line and on the
û-utility-locus of the Agent. By the Remark on page 385, if E induces the Agent
to choose eL then D̂ P E. Suppose, therefore, that E induces the Agent to choose
high effort eH . Then either E = C, and thus D̂ P E by hypothesis, or E 6= C, in
which case E lies on the first segment of the û-utility-locus of the Agent, to the left
of point C. Then, by the argument illustrated in Figure 11.11, C P E and thus, by
transitivity (since D̂ P C), D̂ P E.
2. Assume that C P D̂. Pick an arbitrary contract E 6= C above the 45o line and on the
û-utility-locus of the Agent. If E induces the Agent to choose high effort eH then
E lies on the first segment of the û-utility-locus of the Agent, to the left of point C;
then, by the argument illustrated in Figure 11.11, C P E. Suppose, therefore, that E
induces the Agent to choose low effort eL . Then E lies on the second segment of the
û-utility-locus of the Agent, to the left of point D̂. Then, by the argument illustrated
in Figure 11.10, D̂ P E and thus, by transitivity (since C P D̂), C P E.

Proposition 11.3.2 Let D̂ = (m̂, m̂), û and C be as in Lemma 11.1. Then
1. if D̂ P C then D̂ is the only Pareto efficient contract on the û-utility-locus of the
Agent,
2. if C P D̂ then C is the only Pareto efficient contract on the û-utility-locus of the
Agent,
3. if C ∼P D̂ (that is, the Principal is indifferent between C and D̂) then C and D̂ are
the only Pareto efficient contracts on the û-utility-locus for the Agent.
Proof. We already proved that, among the contracts on the û-utility-locus of the Agent, C
and D̂ are the only candidates for Pareto efficiency. It is clear that if the Principal prefers
D̂ to C then D̂ Pareto dominates C and thus C is not Pareto efficient, so that D̂ is the only
candidate for Pareto efficiency; similarly, if the Principal prefers C to D̂ then C Pareto
dominates D̂ and thus D̂ is not Pareto efficient, so that C is the only candidate for Pareto
2 Wemade the same assumption in Chapter 10. See the remarks in that section about this assumption: the
same considerations apply here.
11.3 The case with two outcomes and two levels of effort 387

efficiency.
Next we prove that if D̂ P C (the Principal prefers D̂ to C) then D̂ is Pareto efficient.
Suppose, by contradiction, that D̂ is not Pareto efficient. Then there exists a contract F 6= D̂
that is better than D̂ for one of the parties and at least as good as D̂ for the other party. It
cannot be that F ∼A D̂ (the Agent is indifferent between F and D̂) because then it must be
that F lies on the û-utility-locus of the Agent and F P D̂ (the Principal prefers F to D̂),
which contradicts Point 1 of Lemma 11.1. Thus it must be that F A D̂ (the Agent prefers
F to D̂) and

F %P D̂ (the Principal considers F to be at least as good as D̂). (11.15)

By (11.15) and the Remark on page 385, it must be that F induces the Agent to choose
high effort eH . Let E be the point at the intersection of the H-indifference curve of the
Agent corresponding to utility û and the H-indifference curve of the Principal through
pH
point F (the straight line with slope − 1−p1 H ). Then E ∼P F (the Principal is indifferent
1
between E and F) and, by Point 1 of Lemma 11.1, D̂ P E so that, by transitivity, D̂ P F,
contradicting (11.15).
Next we prove that if C P D̂ then C is Pareto efficient. Suppose, by contradiction, that C
is not Pareto efficient. Then there exists a contract F 6= C that is better than C for one of the
parties and at least as good as C for the other party. It cannot be that F ∼A C because then
it must be that F lies on the û-utility-locus of the Agent and F P C, which contradicts
Point 2 of Lemma 11.1. Thus it must be that F A C and

F %P C. (11.16)

Suppose first that contract F induces the Agent to choose low effort eL . Let E be the
point at the intersection of the L-indifference curve of the Agent corresponding to utility û
and the L-indifference curve of the Principal through point F (the straight line with slope
pL
− 1−p1 L ). Then E ∼P F and, by Point 2 of Lemma 11.1, C P E so that, by transitivity,
1
C P F, contradicting (11.16). Suppose, therefore, that contract F induces the Agent to
choose high effort eH . Let G be the point at the intersection of the H-indifference curve of
the Agent corresponding to utility û and the H-indifference curve of the Principal through
pH
point F (the straight line with slope − 1−p1 H ). Then G ∼P F and, by Point 2 of Lemma
1
11.1, C P G so that, by transitivity, C P F, contradicting (11.16).
The proof of Point 3 is along the same lines and we will omit it. 

In Exercises 11.10-11.12 the reader is asked to characterize and draw (1) the set of
contracts that make the Principal indifferent between the Agent exerting low effort and
the Agent exerting high effort and (2) the set of contracts that make the Agent indifferent
between low effort and high effort.
 388 Chapter 11. Moral Hazard in Principal-Agent

We conclude this section with two observations.

R First of all, it follows from Proposition 11.3.2 that, if a contract is Pareto

efficient and is not a fixed-salary contract, then it is characterized by a larger
payment to the Agent if the outcome is the better outcome X2 and a smaller
payment if the outcome is the worse outcome X1 . Such contracts are called
incentive contracts because their role is to provide an incentive to the Agent to
exert high effort (thereby increasing the probability of the better outcome X2 ).

Secondly, if the Principal has all the bargaining power (because, for example,
there is “perfect competition” among possible Agents) and thus the Principal
is able to offer a contract that keeps the Agent at her reservation level of
utility, call it u, then the Principal will offer to the Agent that contract on
the u-utility-locus of the Agent that maximizes his own expected utility. By
Lemma 11.1, such a contract is either a fixed-wage contract or that incentive
contract that makes the Agent indifferent between choosing low effort and
choosing high effort (see Exercise 11.13).

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 11.5.2 at the end of this chapter.

11.4 The case with more than two outcomes

So far we have restricted attention to the case where there are only two outcomes and two
levels of effort. The logic of the Principal-Agent problem remains the same if there are three
or more outcomes and possibly more than two levels of effort. However, the conclusion
highlighted in the above Remark, namely that an incentive contract is characterized by
payments that are an increasing function of the outcome, does not generalize to the case of
more than two outcomes. For example, suppose that there are three possible outcomes X1 ,
X2 and X3 with X1 < X2 < X3 . As usual, a contract can be written as a triple (w1 , w2 , w3 ),
where wi is the payment to the Agent if the outcome is Xi (i ∈ {1, 2, 3}). Then it is not
necessarily the case that a Pareto efficient incentive contract C = (w1 , w2 , w3 ) is such that
w1 < w2 < w3 . We will give an example of this below.
In the case of two outcomes, X1 and X2 with X1 < X2 , we could express that fact that
high effort eH makes better outcomes more likely by postulating that pL1 > pH L
1 , where p1
H
is the probability of the worse outcome X1 when the Agent chooses low effort eL and p1 is
the probability of the worse outcome X1 when the Agent chooses high effort eH . When
there are more than two outcomes (and possibly more than two levels of effort), then the
fact that higher effort makes better outcomes more likely can be expressed using the notion
11.4 The case with more than two outcomes 389

of first-order stochastic dominance introduced in Chapter 4 (Definition 4.4.1). Thus the

general setting is as follows:

possible outcomes: X1 , X2 , . . . , Xn (n ≥ 2)
levels of effort: e1 , e2 , . . . , em (m ≥ 2)
probability distribution
over {X1 , . . . , Xn } given ei : Pi

X1 < X2 < · · · < Xn

assumptions: e1 < e2 < · · · < em

Pm >FOC Pm−1 >FOC · · · >FOC P2 >FOC P1


where Pi >FOC Pj means that the probability distribution Pi dominates the probability
distribution Pj in the sense of first-order stochastic dominance (Chapter 4, Definition 4.4.1).
We will not characterize the solution to the general problem, since it is rather complex
and requires more advanced mathematical tools than we have employed so far (in particular,
the notion of constrained optimization).
We conclude this section with an example that shows that, when there are more than
two outcomes, a Pareto efficient incentive contract does not necessarily involve payments
that are an increasing function of the outcomes. Suppose that the Agent can choose
between a low level of effort eL and a high level of effort eH and, as usual, the Principal
cannot monitor the Agent’s effort. The Agent’s reservation utility is 10 and thus she will
not accept any contract that (with the optimal choice of effort) will give her a utility of less
than 10. The Principal is risk neutral while the Agent’s utility function is given by:
√
 m if eL
UA (m, e) = √
 m − 1 if e
H

There are three possible outcomes with the probabilities listed in the following table:3

outcome: X1 = $100 X2 = $120 X3 = $200

1 1
probability if eL : 2 2 0
2 1 1
probability if eH : 5 10 2

Suppose that the Principal has all the bargaining power, so that he will only offer contracts
that give the Agent her reservation utility of 10. Among such contract that, furthermore,
induce the Agent to choose low effort eL the one that gives the Principal the largest expected
utility is the fixed-wage contract F = (w1 = 100, w2 = 100, w3 = 100) in which case the
Principal’s expected utility is 12 (100 −100)+ 21 (120 −100)+0(200 −100) = 10. Consider
3 Note that the probability distribution associated with eH dominates that associated with eL in the sense
of first-order stochastic dominance.
390 Chapter 11. Moral Hazard in Principal-Agent

now the following incentive contract C = (w1 = 81, w2 = 100, w3 = 163.84). Under this
contract the Agent’s expected utility is:

1
√ √
if she chooses eL : 2 81 + 12 100 = 9.5
2
√ 1
√ 1
√
if she chooses eH : 5 81 + 10 100 + 2 163.84 − 1 = 10.

Thus the Agent would accept contract C and exert high effort eH . With contract C the
1
Principal’s expected utility is 25 (100 − 81) + 10 (120 − 100) + 21 (200 − 163.84) = 27.68.
Thus the Agent is indifferent between F and C, while the Principal prefers C to F; hence C
Pareto dominates F.
Next we show that contract C is not the contract that maximizes the Principal’s ex-
pected utility among the contracts that induce the Agent to choose eH and give her an
expected utility of 10. To find such a contract, we need to solve the following constrained
maximization problem:
2 1 1
choose w1 , w2 , w3 to maximize 5 (100 − w1 ) + 10 (120 − w2 ) + 2 (200 − w3 )
= 152 − 52 w1 − 101
w2 − 12 w3
subject to

2√ 1√ 1√
5 w1 + 10 w2 + 2 w3 − 1 ≥ 10 (IR)

2√ 1√ 1√ 1√ 1√
5 w1 + 10 w2 + 2 w3 − 1 ≥ 2 w1 + 2 w2 . (IC)

(IR) is the individual rationality constraint for the Agent: it says that the proposed contract
must give the Agent an expected utility of at least 10 if she chooses high effort eH and
(IC) is the incentive compatibility constraint for the Agent: it says that, under the proposed
contract, the Agent cannot get higher expected utility by choosing low effort eL . We now
show that, at a solution to the above maximization problem, both constraints must be
satisfied as equalities. We will do so without explicitly employing the tools of constrained
maximization (the Kuhn-Tucker conditions).
First we argue that, at a solution of the maximization problem, the (IR) constraint must
be satisfied as an equality. Pick an arbitrary triple (w1 , w2 , w3 ) at which the (IR) constraint
is satisfied as a strict inequality and the (IC) constraint is also satisfied (possibly as an
equality). Now decrease the value of w2 by a small amount, so that the (IR) constraint is
still satisfied. Then, since the maximand, that is the function 152 − 52 w1 − 10 1
w2 − 12 w3 , is
decreasing in w2 , its value will increase. The only concern is that decreasing w2 might
lead to a violation of the (IC) constraint; however, if w2 is decreased, the left-hand-side
(LHS) of the (IC) constraint decreases less than the right-hand-side (RHS) of it and thus
the constraint will be preserved.4 Since we were able to increase the value of the maximand
without violating the constraints, the triple we started with could not have been a solution
to the maximization problem. Thus the (IR) constraint must be satisfied as an equality.
 
4 The 1 √1
derivative, with respect to w2 , of the LHS of (IC) is 10 2 w2 which is less than the derivative,
 
with respect to w2 , of the RHS of (IC) which is 12 2√1w2 .
11.4 The case with more than two outcomes 391

From the (IR) constraint satisfied as an equality, namely

2√ 1√ 1√
5 w1 + 10 w 2 + 2 w3 − 1 = 10
√
we can solve for w3 to obtain
√ √ √
w3 = 22 − 54 w1 − 15 w2 . (11.17)
√
Defining new variables yi = wi (i = 1, 2, 3) (so that (11.17) can be written as
y3 = 22 − 45 y1 − 15 y2 ), we can rewrite the maximization problem as follows:

2
choose y1 , y2 to maximize = 152 − 25 (y1 )2 − 10
1
(y2 )2 − 12 22 − 45 y1 − 15 y2
subject to

2 1 1
22 − 45 y1 − 15 y2 − 1 ≥ 1 1


5 y1 + 10 y2 + 2 2 y1 + 2 y2 (IC)

The next step in the argument is to show that if the (IC) constraint is satisfied as

2
a strict inequality then the function 152 − 25 (y1 )2 − 101
(y2 )2 − 12 22 − 45 y1 − 15 y2 is not
maximized. This part of the proof is developed in Exercise 11.15.
Thus the (IC) constraint must be satisfied as an equality, that is, it must be that
2 1 1 4 1
− 1 = 12 y1 + 12 y2


5 y1 + 10 y2 + 2 22 − 5 y1 − 5 y2

which simplifies to

y2 = 20 − y1 (11.18)

Using (11.18) we can reduce the maximization problem to the following unconstrained
one:

choose y1 to maximize
2
152 − 52 (y1 )2 − 10
1
(20 − y1 )2 − 12 22 − 54 y1 − 15 (20 − y1 )


74 17 2
= 5 y1 − 25 y1 − 50.

Let f (y1 ) = 74 17 2 0 74 34
5 y1 − 25 y1 − 50; then f (y1 ) = 5 − 25 y1 and setting this equal to zero and
solving we obtain y1 = 185 185 155
17 ; from this and (11.18) we get that y2 = 20 − 17 = 17 ; finally,
replacing these two values into y3 = 22 − 45 y1 − 15 y2 we get that y3 = 19517 . Thus, since
√
yi = wi we get the following solution to the original maximization problem:
185 2 155 2 195 2




w1 = 17 = 118.43, w2 = 17 = 83.13, w3 = 17 = 131.57.

Let D be the corresponding contract: D = (w1 = 118.43, w2 = 83.13, w3 = 131.57). The

Agent’s expected utility from this contract is
√ √ √
E[UA (D)] = 25 118.43 + 10 1
83.13 + 12 131.57 − 1 = 10

and the Principal’s expected utility is

E[UP (D)] = 52 (100 − 118.43) + 10

1
(120 − 82.13) + 12 (200 − 131.57) − 1 = 30.53
392 Chapter 11. Moral Hazard in Principal-Agent

Under contract D the Agent is paid the least if the intermediate outcome ($120) obtains.
If the outcome turns out to be $120 then she is paid approximately $83, while if the
outcome turns out to be lower ($100) then she is paid more: approximately $118! The
reason is that from a low outcome of $100 one cannot infer that the Agent did not work
hard: it is almost as likely that she exerted high effort as it is that she exerted low effort. On
the other hand, the intermediate outcome ($120) is much more likely if the Agent chooses
low effort than if she chooses high effort, hence it makes sense to associate a relatively low
payment to that outcome.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 11.5.3 at the end of this chapter.
 11.5 Exercises 393

11.5 Exercises
The solutions to the following exercises are given in Section 11.6 at the end of this chapter.

11.5.1 Exercises for Section 11.2: Risk sharing under moral hazard

Exercise 11.1 A risk-neutral Principal wants to hire an Agent to run her firm. The
Agent’s utility depends on two things: money and effort. Denote effort by e and assume
that it can take on only two values: eL (for low) and eH (for high). Let the Agent’s
utility function be given as follows (where m denotes money)
√
 90 + m − 9 if e = eL
UA (m, e) = √
 90 + m − 10 if e = e .
H

Assume that if they don’t sign a contract they both get a utility of zero. The Agent’s
effort cannot be observed by the Principal and thus cannot be made part of the contract.
There are only three possible outcomes (profit levels): $ − 100, $100, $500. If the Agent
works hard, better outcomes are more likely than if the Agent does not work hard:

outcome (profits): $ − 100 $100 $500

1 1 1
probability if e = eL : 2 4 4
1 1 1
probability if e = eH : 4 4 2

Show that, of the following contracts, C is Pareto superior to B.

CONTRACT B (fixed wage): the Agent will be paid w = 10, no matter what the profits
of the firm.
CONTRACT C (contingent wage): the Agent will get nothing (w = 0) if the profit of
the firm is either −100 or 100, while she will get w = 100 if the profit of the firm is 500.

 394 Chapter 11. Moral Hazard in Principal-Agent

Exercise 11.2 Mr. O owns a firm. He can either run the firm himself or hire Ms. M
to run it for him. If he runs the firm himself he gets a utility of 0. If he hires Ms. M,
he will not be able to check whether she works hard or not. Under the management of
Ms. M, the firm’s profit levels and corresponding probabilities would be as follows (e
denotes effort, eL is low effort and eH is high effort):

profit: $0 $100 $800

1 1 1
probability if e = eL : 2 4 4
1 1 1
probability if e = eH : 4 4 2

Mr. O is risk neutral. Ms. M is currently unemployed and her current utility is 0; her
utility function is as follows (m denotes money and e denotes effort):
(
m − 8 if e = eL
UA (m, e) =
m − 10 if e = eH

Consider the following contracts.

CONTRACT A: Mr. O hires Ms. M and pays her a fixed wage of $10 (i.e. Ms. O’s
wage will be $10 no matter what the profits of the firm).
CONTRACT B: Mr. O hires Ms. M on the following terms: if the profit of the firm is
less than $800, Ms. M will get nothing; if the profit is $800 Ms. M will get $24.
(a) Which of the two contracts would Ms. M find acceptable?
(b) How does Ms M rank the three options: (1) sign contract A, (2) sign contract B,
(3) remain unemployed.
(c) How does Mr O rank the two contracts?


11.5.2 Exercises for Section 11.3: Two outcomes and two levels of effort
Exercise 11.3 Suppose that there are only two outcomes, X1 = $2, 000 and
X2 = $5, 000, and two levels of effort: low effort eL and high effort eH . The Prin-
cipal is risk neutral, while the Agent’s utility function is given by
(√
m if e = eL
UA (m, e) = √
m − 4 if e = eH .

The probability of X1 is 14 if the Agent chooses eL and 10 1

if the Agent chooses eH .
Consider contract C = (w1 = 1, 600, w2 = 2, 500).
(a) Calculate the slope, at point C, of the Agent’s indifference curve corresponding
to low effort eL that goes through contract C.
(b) Calculate the slope, at point C, of the Agent’s indifference curve corresponding
to high effort eH that goes through contract C.
(c) Calculate the Agent’s expected utility from contract C.
(d) Calculate the Principal’s expected utility from contract C.

11.5 Exercises 395

w2
(payment if X2 )

45o line

600 C
E
D

w1
(payment if X1 )
0 210

Figure 11.13: The figure for Exercise 11.4.

Exercise 11.4 Suppose that there are only two outcomes, X1 and X2 (with X1 < X2 ),
and two levels of effort: low effort eL and high effort eH . The Principal is risk neutral,
while the Agent’s utility function is given by
(
ln(m) if e = eL
UA (m, e) =
ln(m) − 1 if e = eH .

The probability of X1 is 25 if the Agent chooses eL and 51 if the Agent chooses eH .

(a) Under contract C in Figure 11.13, would the Agent choose eL or eH ?
(b) Calculate the coordinates of contract D in Figure 11.13.
(c) Calculate the coordinates of contract E in Figure 11.13.


Exercise 11.5 Suppose that there are only two outcomes, X1 = $1, 000 and X2 =
$1, 500, and two levels of effort: low effort eL and high effort eH . The Principal is risk
neutral, while the Agent’s utility function is given by (with a > 0)
(√
m if e = eL
UA (m, e) = √
m − a if e = eH .

The probability of X1 is 12 if the Agent chooses eL and 15 if the Agent chooses eH .

Consider contract C = (w1 = 625, w2 = 900).
(a) For what values of a would the Agent choose eH under contract C?
(b) Let a = 1. Find a fixed-wage contract D such that the Agent is indifferent between
C and D.

396 Chapter 11. Moral Hazard in Principal-Agent

Exercise 11.6 Let A and B be two contracts above the 45o line such that the Agent
would choose high effort eH under either contract. Prove that, if B lies below the
Principal’s indifference curve through A, then the Principal gets higher utility from B
than from A, that is, he prefers B to A. 

Exercise 11.7 Let D and E be two contracts above the 45o line such that the Agent
would choose low effort eL under either contract. Prove that, if E lies below the
Principal’s indifference curve through D, then the Principal gets higher utility from E
than from D, that is, he prefers E to D. 

Exercise 11.8 Suppose that there are only two outcomes, X1 = $1, 000 and
X2 = $1, 500, and two levels of effort: low effort eL and high effort eH . The Prin-
cipal is risk neutral, while the Agent’s utility function is given by
(√
m if e = eL
UA (m, e) = √
m − 1 if e = eH .

The probability of X1 is 12 if the Agent chooses eL and 25 if the Agent chooses eH . Assume
that, if a contract makes the Agent indifferent between choosing eL and choosing eH ,
then the Agent chooses eH .
(a) Find the fixed-wage contract, call it D, that gives the Agent a utility equal to 24.
(b) Find the contract, call it C, that (1) makes the Agent is indifferent between
choosing eL and choosing eH and (2) gives the Agent an expected utility of 24.
(c) Is contract D Pareto efficient?
(d) Is contract C Pareto efficient?


Exercise 11.9 Repeat Exercise 11.8 with the same data with one exception: the proba-
1
bility of X1 is 3 if the Agent chooses eH . 

Exercise 11.10 .
(a) What properties must contract C = (w1 , w2 ) satisfy in order for the Principal to
be indifferent between the Agent exerting low effort eL and the Agent exerting
high effort eH ?
(b) For every w1 ≥ 0, let w2 = f (w1 ) be such that the contract (w1 , f (w1 )) makes the
the Principal indifferent between the Agent exerting low effort eL and the Agent
exerting high effort eH . Find the expression for the function f (w1 ).
(c) Draw the graph of the function w2 = f (w1 ) in the (w1 , w2 ) plane and identify two
regions: (1) the region where the Principal prefers eL , (2) the region where the
Principal prefers eH .

11.5 Exercises 397

Exercise 11.11 Suppose that there are only two outcomes: X1 and X2 , with X1 < X2 ,
and two levels of effort: low effort eL and high effort eH . The Agent’s utility function is
given by (where m denotes money, e denotes effort and c ≥ 1 is a constant)
(
ln(m) if e = eL
UA (m, e) =
ln(m) − c if e = eH .

The probability of X1 is pL1 if the Agent chooses eL and pH

1 if the Agent chooses eH ,
with pL1 > pH
1 .
Let the function w2 = g(w1 ) be such that, for every w1 ≥ 0, the contract (w1 , g(w1 ))
has the property that the Agent is indifferent between choosing eL and choosing eH .

(a) Find the expression for the function w2 = g(w1 ).

(b) Draw the graph of the function w2 = g(w1 ) and identify two regions:
(1) the region where the Agent prefers eL ,
(2) the region where the Agent prefers eH .


Exercise 11.12 Suppose that there are only two outcomes: X1 and X2 , with X1 < X2 ,
and two levels of effort: low effort eL and high effort eH . The Agent’s utility function is
given by (where m denotes money, e denotes effort, c is a positive constant, U 0 (m) > 0,
U 00 (m) < 0)
(
U(m) if e = eL
UA (m, e) =
U(m) − c if e = eH

The probability of X1 is pL1 if the Agent chooses eL and pH

1 if the Agent chooses eH ,
L H
with p1 > p1 .
Let the function w2 = g(w1 ) be such that, for every w1 ≥ 0, the contract (w1 , g(w1 ))
has the property that the Agent is indifferent between choosing eL and choosing eH .
Let A = (wA1 , wA2 ) be a point (above the 45o line) at which the Agent is indifferent
between eL and eH ; thus wA2 = g(wA1 ). Find the slope of the function g(·) at point A.
[Hint: use the method employed in Chapter 5 (Section 5.1.4) to determine the slope of
an indifference curve (or, if you are familiar with it, use the implicit function theorem).]

 398 Chapter 11. Moral Hazard in Principal-Agent

Exercise 11.13 Ann’s current fixed salary is $16,900. She is bored with her job and
she will accept any new employment that does not make her worse off relative to
her current situation. Peter is thinking of starting a business and hiring an agent to
run it for him. Peter will not be able to monitor the agent, but knows that the agent
can either be lazy or work hard. There are two possible levels of income from the
business: $22,000 and $36,000. If the agent is lazy the probability of the income being
$22,000 is 50% while if the agent works hard it is 40%. Peter likes money (prefers to
have more rather than less) and is risk neutral. Ann also likes money, but she is risk
averse; furthermore, she dislikes effort. Her vNM utility function is as follows (where
m denotes money and(√e effort: eL means that she is lazy and eH means that she works
m if e = eL
hard): UA (m, e) = √ . What contract will Peter offer to Ann? 
m − 3 if e = eH

11.5.3 Exercises for Section 11.4: The case with more than two outcomes

Exercise 11.14 A Principal is considering hiring an Agent to run his firm. The Principal
will not be able to monitor the Agent’s effort, which can be either low eL or high eH .
There are three possible outcomes with probabilities that are affected by the Agent’s
level of effort, as shown in the following table:

outcome: X1 = $1, 000 X2 = $2, 000 X3 = $3, 000

1 1 1
probability if eL : 2 4 4
1 1
probability if eH : 2 0 2

The Agent’s utility function is

(√
m if e = eL
UA (m, e) = √
m − 1 if e = eH

and her reservation utility is 20. Consider the following contract (where wi is the
payment to the Agent if the outcome is Xi ) C = (w1 = $484, w2 = $900, w3 = $961).
(a) If offered contract C, will the Agent accept it?
(b) What level of effort will the Agent choose if offered contract C?
(c) Show that contract C is not Pareto efficient, by constructing an alternative contract
D that Pareto dominates C. [Hint: outcome X2 is evidence that the Agent chose
low effort, thus the value of w2 can be used to induce the Agent to choose high
effort.]

11.5 Exercises 399

Exercise 11.15 Consider the maximization problem given on page 391, namely

choose y1 , y2 to maximize

2
f (y1 , y2 ) = 152 − 25 (y1 )2 − 10
1
(y2 )2 − 12 22 − 45 y1 − 15 y2
subject to
2 1 1
22 − 54 y1 − 15 y2 − 1 ≥ 1 1


5 y1 + 10 y2 + 2 2 y1 + 2 y2 (IC)

The objective of this exercise is to show that, if the (IC) constraint is satisfied as a strict
inequality, then the function f (y1 , y2 ) is not maximized. This can be done by showing
that at every point (y1 , y2 ) such that y1 ≥ 0, y2 ≥ 0 and y1 + y2 < 20 (the shaded area in
Figure 11.14) either ∂∂y f (y1 , y2 ) > 0 or ∂∂y f (y1 , y2 ) > 0 (or both), so that the value
1 2
of the function can be increased by increasing either y1 or y2 , until the (IC) constraint
becomes an equality.
∂
(a) Calculate the partial derivative ∂ y1 f (y1 , y2 ) and show in the (y1 , y2 ) plane the
∂
region where ∂ y1 f (y1 , y2 ) > 0.
∂
(b) Calculate the partial derivative ∂ y2 f (y1 , y2 ) and show in the (y1 , y2 ) plane the
∂
region where ∂ y2 f (y1 , y2 ) > 0.
(c) Verify that the shaded area shown in Figure 11.14 is entirely contained in the
union of the two regions identified in Parts (a) and (b).


y2

20

y1
0 20

Figure 11.14: The shaded area is the set of points (y1 , y2 ) such that y1 ≥ 0, y2 ≥ 0 and
y1 + y2 < 20.
 400 Chapter 11. Moral Hazard in Principal-Agent

11.6 Solutions to Exercises

Solution to Exercise 11.1

√
CONTRACT B. If the Agent works hard, she will get a utility of UA (10, e√H ) = 90 + 10 −
10 = 0; if she does not work hard, her utility will be: UA (10, eL ) = 90 + 10 − 9 = 1.
Hence she will choose e = eL , that is, she will not work hard. The Principal can figure this
out and will realize that his expected utility from this contract is:
1 1 1
2 (−100 − 10) + 4 (100 − 10) + 4 (500 − 10) = 90.
√
CONTRACT C. If the Agent works hard, her expected utility will be: 14 ( 90 − 10) +
1
√ 1
√
4 ( 90 − 10) + 2 (√ 90 + 100 − 10) √
= 1.635. If she does not work hard, her expected
√
utility will be: 2 ( 90 − 9) + 4 ( 90 − 9) + 41 ( 90 + 100 − 9) = 1.561. Hence she will
1 1

choose eH , that is, she will work hard. The Principal can figure this out and will realize
that his expected utility from this contract is: 14 (−100) + 41 (100) + 12 (500 − 100) = 200.
Thus contract C is Pareto superior to contract B, despite the fact that it does not allocate
risk optimally. 

Solution to Exercise 11.2

Let us first see what wold happen under each contract.
CONTRACT A. If Ms. M chooses e = eL her utility is 10 − 8 = 2, if she chooses e = eH
her utility is 10 − 10 = 0. Hence she will choose e = eL . Hence Mr. O’s expected utility is
1 1 1
2 (−10) + 4 (100 − 10) + 4 (800 − 10) = 210.
CONTRACT B. If Ms. M chooses e = eL her expected utility is 12 (−8) + 14 (−8) + 14 (24 −
8) = −2. If she chooses e = eH her expected utility is 14 (−10) + 14 (−10) + 12 (24 − 10) = 2.
Thus she will choose e = eH . Hence Mr. O’s expected utility is 14 (0) + 14 (100) + 21 (800 −
24) = 413.
(a) Ms. M would find both contracts acceptable, since they give her (with the appropriate
choice of effort) higher utility than her current level of utility.
(b) Ms. M is indifferent between the two contracts and prefers either of them to
remaining unemployed.
(c) Mr. O prefers contract B to contract A. 

Solution to Exercise 11.3

(a) The slope is
1 √1
4 2 1,600 5
−3 =− = −0.4167.
√1 12
4 2 2,500

(b) The slope is √

1 2, 500 5
− √ = − = −0.1389.
9 1, 600 36
(c) First we have to figure out whether the Agent will choose eL or eH under
contract C.
√ √
• E[UA (C, eL )] = 41 1, 600 + 34 2, 500 = 47.5.
1√ 9√
• E[UA (C, eH )] = 10 1, 600 + 10 2, 500 − 4 = 45.
11.6 Solutions to Exercises 401

Thus the Agent will choose eL and her expected utility is 47.5.
(d) E[UP (C)] = 14 (2, 000 − 1, 600) + 43 (5, 000 − 2, 500) = 1, 975. 

Solution to Exercise 11.4

First of all, note that the steeper indifference curve corresponds to eL and the less steep
one to eH .
(a) The Agent’s expected utility from contract C is:
(1) if she chooses eL : 25 ln(210) + 53 ln(600) = 5.977,
(2) if she chooses eH : 15 ln(210) + 45 ln(600) − 1 = 5.187.
Thus the Agent would choose eL .
(b) D = (mD , mD ) where mD is the solution to ln(m) = 25 ln(210) + 53 ln(600) = 5.977.
Thus mD = 394.26.
(c) E = (mE , mE ) where mE is the solution to ln(m) − 1 = 15 ln(210) + 45 ln(600) − 1 =
5.187. Thus mE = 486.38. 

Solution to Exercise 11.5

(a) The Agent’s expected utility from contract C is:
√ √
(1) if she chooses eL : 12 625 + 12 900 = 27.5,
√ √
(2) if she chooses eH : 51 625 + 45 900 − a = 29 − a.
Thus the Agent will choose eH if a < 1.5 (and will be indifferent if a = 1.5).
(b) When a = 1 under contract C the Agent√ chooses eH and gets a utility of 28. Thus
D = (m, m) where m is the solution to m = 28, that is, m = 784 (under a fixed-wage
contract the Agent chooses eL ). 

Solution to Exercise 11.6

This is a repetition of the argument used in Chapter 2 (Section 2.5). Let A0 be the
point at the intersection of the 45o
line and the Principal’s (straight-line) H-indifference
curve through point A = wA1 , wA2 and let B0 be the point at the intersection of the 45
o
line and the Principal’s (straight-line) H-indifference curve through point B = wB1 , wB2 .
Then A0 = (mA0 , mA0 ), where H A H A 0


H B H

B mA0 = p1 w1 + 1 − p1 w2 , and B = (mB0 , mB0 ), where
mB0 = p1 w1 + 1 − p1 w2 . Since point B lies below the Principal’s indifference curve
through A,

mB0 < mA0 . (11.19)

For the Principal, the expected utility of A is the same as the expected utility of A0
(since they belong to the same indifference curve) and the latter is equal to EH [X] − mA0 ,
where EH [X] = pH H


1 X1 + 1 − p1 X2 ; similarly, the expected utility of B is the same as
0
the expected utility of B (since they belong to the same indifference curve) and the latter
is equal to EH [X] − mB0 . Thus, by (11.19), the expected utility of B0 is greater than the
expected utility of A0 ; therefore, the expected utility of B is greater than the expected utility
of A. 
402 Chapter 11. Moral Hazard in Principal-Agent

Solution to Exercise 11.7

This is a repetition of the argument used in Exercise 11.6. Let D0 be the point at the
intersection of the 45 o
D D

line0 and the Principal’s (straight-line) L-indifference curve through
o line and the Principal’s
point D = w1 , w2 and E be the point at the intersection of the 45
(straight-line) L-indifference
curve through point E = wE1 , wE2 . Then D0 = (mD0 , m


L wD , and E 0 = (m 0 , m 0 ), where m 0 = pL wE + 1 − pL wE .

D0 ),
where mD0 = pL1 wD 1 + 1 − p 1 2 E E E 1 1 1 2
Since point E lies below the Principal’s indifference curve through D,

mE 0 < mD0 . (11.20)

For the Principal, the expected utility of D is the same as the expected utility of D0 (since
they belong to the same
indifference curve) and the latter is equal to EL [X] − mD0 , where
EL [X] = pL1 X1 + 1 − pL1 X2 ; similarly, the expected utility of E is the same as the expected
utility of E 0 (since they belong to the same indifference curve) and the latter is equal to
EL [X] − mE 0 . Thus, by (11.20), the expected utility of E 0 is greater than the expected utility
of D0 ; therefore, the expected utility of E is greater than the expected utility of D. 

Solution to Exercise 11.8

√ contract, the Agent will choose eL . Thus D = (m, m) where
(a) Since it is a fixed-wage
m is the solution to m = 24, that is, m = 576.
(b) Contract C = (w1 , w2 ) is given by the solution to the following equations:
2 √ 3 √ 1√ 1√
5 ( w1 − 1) + 5 ( w2 − 1) = 2 w1 + 2 w2

2 √ 3 √
5 ( w1 − 1) + 5 ( w2 − 1) = 24

The solution is w1 = 361 and w2 = 841. Thus C = (361, 841).

In order to answer Parts (c) and (d) let us see how the Principal ranks the two contracts C
and D. E[UP (D)] = 12 (1, 000 − 576) + 12 (1, 500 − 576) = 674, while

E[UP (C)] = 25 (1, 000 − 361) + 35 (1, 500 − 841) = 651.

Thus D P C. Hence, by Proposition 11.3.2, D is the only Pareto efficient contract on the
utility locus of the Agent corresponding to a utility of 24. Thus
(c) Yes, contract D is Pareto efficient.
(d) No, contract C is not Pareto efficient. 

Solution to Exercise 11.9

(a) The answer to this question remains the same: D = (576, 576).
(b) Contract C = (w1 , w2 ) is given by the solution to the following equations:
1 √ 2 √ 1√ 1√
3 ( w1 − 1) + 3 ( w2 − 1) = 2 w1 + 2 w2

1 √ 2 √
3 ( w1 − 1) + 3 ( w2 − 1) = 24

The solution is w1 = 441 and w2 = 729. Thus C = (441, 729).

11.6 Solutions to Exercises 403

In order to answer Parts (c) and (d) let us see how the Principal ranks the two contracts C
and D. As calculated in the previous exercise, E[UP (D)] = 674, while

E[UP (C)] = 13 (1, 000 − 441) + 32 (1, 500 − 729) = 700.33.

Thus C P D. Hence, by Proposition 11.3.2, C is the only Pareto efficient contract on the
utility locus of the Agent corresponding to a utility of 24. Thus
(c) No, contract D is not Pareto efficient.
(d) Yes, contract C is Pareto efficient. 

Solution to Exercise 11.10

(a) It must be that

pL1 (X1 − w1 ) + 1 − pL1 (X2 − w2 ) = pH H

1 (X1 − w1 ) + 1 − p1 (X2 − w2 )

that is (rearranging the terms),
pL1 − pH L − pH (X − w ) and,

1 (X 1 − w1 ) = p 1 1 2 2
dividing both sides by pL1 − pH 1 ,

X1 − w1 = X2 − w2 . (11.21)

(b) It follows from (11.21) that

w2 = X2 − X1 + w1

which is the equation of a straight line with slope 1.

(c) See Figure 11.15. 

w2
(payment if X2 )
eL ∼P eH

45o line
eL P eH

eH P eL

X2 − X1

w1
(payment if X1 )
0

Figure 11.15: The straight line is the set of contracts that make the Principal indifferent
between the Agent choosing low effort eL and the Agent choosing high effort eH . For
contracts above the line the Principal is better off if the Agent chooses low effort and for
contracts below the line the Principal is better off if the Agent chooses high effort.
404 Chapter 11. Moral Hazard in Principal-Agent

Solution to Exercise 11.11

(a) A contract (w1 , w2 ) makes the Agent indifferent between eL and eH if and only if

pL1 ln(w1 ) + 1 − pL1 ln(w2 ) = pH H

1 ln(w1 ) + 1 − p1 ln(w2 ) − c.

Rearranging the terms in the above equation we get

c
ln(w2 ) − ln(w1 ) = .
pL1 − pH
1

c
Note that, since c ≥ 1 and 0 < pL1 − pH 1 < 1, pL −pH > 1. From the above equation
1 1
we get that (e denotes the irrational number 2.71828...)

c c
ln(w2 )−ln(w1 ) pL H w2 pL −pH
e =e 1 −p1 that is =e 1 1 .
w1

c
L H
Thus g(w1 ) = e p1 −p1 w1 .

(b) See Figure 11.16. 

w2 eL ∼A eH
(payment if X2 )

45o line
eH A eL

eL A eH

c
pL H
1 −p1
e w1
(payment if X1 )
0

Figure 11.16: The straight line is the set of contracts that make the Agent indifferent
between choosing low effort eL and choosing high effort eH . For contracts above the line
the Agent is better off choosing high effort and for contracts below the line the Agent is
better off choosing low effort.
11.6 Solutions to Exercises 405

Solution to Exercise 11.12

Let A = wA1 , wA2 be a contract (above the 45o line) that makes the Agent indifferent

between choosing eL and choosing eH . Then

pL1 U(wA1 ) + 1 − pL1 U(wA2 ) = pH A H A

1 U(w1 ) + 1 − p1 U(w2 ) − c,

that is,

c
U(wA2 ) −U(wA1 ) = . (11.22)
pL1 − pH
1

Let B = wB1 , wB2 be a contract close to A that also makes the Agent indifferent between

choosing eL and choosing eH . Then

c
U(wB2 ) −U(wB1 ) = . (11.23)
pL1 − pH
1

We want to calculate the slope of the segment that joins A and B. Since B is close to A,
 
U(wB1 ) ' U(wA1 ) +U 0 (wA1 ) wB1 − wA1 , (11.24)
 
U(wB2 ) ' U(wA2 ) +U 0 (wA2 ) wB2 − wA2 . (11.25)

Replacing (11.24) and (11.25) in (11.23) we get

    c
U(wA2 ) −U(wA1 ) +U 0 (wA2 ) wB2 − wA2 −U 0 (wA1 ) wB1 − wA1 = L
| {z } p1 − pH
1
c
= by (11.22)
pL H
1 −p1

from which it follows that

   
0 0 A
U (wA2 ) B A B A
w2 − w2 −U (w1 ) w1 − w1 = 0,

that is,
rise
z }| {
wB2 − wA2 U 0 (wA1 )
= 0 A ,
wB1 − wA1 U (w2 )
| {z }
run

U 0 (wA )
that is, the slope of the curve w2 = g(w1 ) at point A is U 0 (w1A ) . Since point A is above the 45o
2
line, wA1 < wA2 and thus (since U 0 (m) is decreasing because U 00 (m) < 0), U 0 (wA1 ) > U 0 (wA2 )
U 0 (wA )
so that U 0 (w1A ) > 1. Hence the curve is increasing with a slope greater than 1 everywhere.
2
406 Chapter 11. Moral Hazard in Principal-Agent

Solution to Exercise 11.13

Since Ann is currently
√ on a fixed salary, her current choice of effort is eL and thus her
current utility is 16, 900 = 130; call this value her reservation utility. Peter will thus make
sure that he offers Ann a contract that gives her exactly her reservation utility, not more.
Thus Peter will offer that contract on the 130-utility-locus of Ann that maximizes Peter’s
own utility. By Lemma 11.1 we know that this contract is either the fixed-wage contract
D = (16900, 16900) or that contract, call it C = (w1 , w2 ), that makes Ann indifferent
between choosing eL and choosing eH (and gives her an expected utility equal to 130). To
find contract C we must solve the following system of two equations:
2 √ 3 √ 1√ 1√
5 ( w1 − 3) + 5 ( w2 − 3) = 2 w1 + 2 w2

2 √ 3 √
5 ( w1 − 3) + 5 ( w2 − 3) = 130.

The solution is w1 = 13, 225 and w2 = 21, 025. Thus C = (13225, 21025). It only remains
to check which of C and D gives higher utility to Peter.
Peter’s expected utility from D is: 12 22000 + 12 36000 − 16900 = 12, 100.
Peter’s expected utility from C is: 25 (22000 − 13225) + 35 (36000 − 21025) = 12, 495.
Thus Peter will offer Ann the contingent contract C = (13225, 21025). 

Solution to Exercise 11.14

The Agent’s expected utility under contract C is:
√ √ √
if low effort eL : 21 484 + 41 900 + 14 961 = 26.25
√ √
if high effort eH : 12 484 + 21 961 − 1 = 25.5.

(a) Since the Agent can achieve a utility of at least 25.5 and her reservation utility is 20,
she will accept contract C.
(b) Under contract C the Agent will choose low effort, since it gives her higher utility
than high effort.
(c) Since the middle outcome X2 can only occur if the Agent chooses low effort, the
Principal can provide an incentive to the Agent to choose high effort by setting
w2 = 0. Let us keep the value w1 = 484 and find a contract with w2 = 0 that
gives
√ the Agent a utility of 26.25 if she chooses high effort, by solving the equation
1 1√
2 484 + 2 w3 − 1 = 26.25. The solution is w3 = 1, 056.25. Now consider contract
D = (w1 = 484, w2 = 0, w3 = 1056.25). The Agent’s expected utility under contract
D is:
√ p
if low effort eL : 12 484 + 14 1, 056.25 = 19.125
√ p
if high effort eH : 12 484 + 12 1, 056.25 − 1 = 26.25.

Thus the Agent chooses high effort eH . The Principal’s expected utility is:
1 1 1
with C : 2 (1000 − 484) + 4 (2000 − 900) + 4 (3000 − 961) = 1, 042.75
1 1
with D : 2 (1000 − 484) + 2 (3000 − 1056.25) = 1, 229.875.

Thus contract D Pareto dominates contract C. 

11.6 Solutions to Exercises 407

Solution to Exercise 11.15

(a) The partial derivative of f with respect to y1 is

∂ 1
∂ y1
f (y1 , y2 ) = 25 (440 − 36y1 − 4y2 ) .

∂
Thus ∂ y1 f (y1 , y2 ) > 0 if and only if y2 < 110 − 9y1 . The set of points (y1 , y2 ) such
∂
that ∂ y1 f (y1 , y2 ) > 0 is shown as the shaded area in Figure 11.17.

y2

110

The coordinates of this

intersection point are
20 y1 = 11.25 and y2 = 8.75

y1
0 110
9
20

Figure 11.17: The shaded area is the set of points (y1 , y2 ) such that y2 < 110 − 9y1
(equivalently, such that ∂∂y f (y1 , y2 ) > 0). (The drawing is not accurate in terms of scale,
1
because we wanted the coordinates to be spaced sufficiently apart to be readable.)

(b) The partial derivative of f with respect to y2 is

∂ 1
∂ y2
f (y1 , y2 ) = 25 (110 − 4y1 − 6y2 ) .

55
Thus ∂
∂ y2 f (y1 , y2 ) > 0 if and only if y2 < 3 − 23 y1 . The set of points (y1 , y2 ) such

∂
that ∂ y2 f (y1 , y2 ) > 0 is shown as the shaded area in Figure 11.18.
408 Chapter 11. Moral Hazard in Principal-Agent
y2

20
55
3

15

y1
0 5 20 27.5

Figure 11.18: The shaded area is the set of points (y1 , y2 ) such that y2 < 55 2
3 − 3 y1 (equiva-
lently, such that ∂∂y f (y1 , y2 ) > 0). (The drawing is not accurate in terms of scale, because
2
we wanted the coordinates to be spaced sufficiently apart to be readable.)

(c) The shaded area in Figure 11.19 shows the union of the two shaded areas of Figures
11.17 and 11.18 which contains the set of points below the line of equation y2 =
20 − y1 . Algebraically, this can be seen as follows. The point of intersection
between the line of equation y2 = 110 − 9y1 and the line of equation y2 = 20 − y1 has
coordinates (y1 = 11.25, y2 = 8.75); from y1 = 0 to y1 = 11.25 the area under the
line of equation y2 = 20 − y1 is entirely contained within the area under the line of
equation y2 = 110 − 9y1 (that is, for y1 ≤ 11.25, if y2 < 20 − y1 then y2 < 110 − 9y1 ).
The point of intersection between the line of equation y2 = 55 2
3 − 3 y1 and the line
of equation y2 = 20 − y1 has coordinates (y1 = 5, y2 = 15); from y1 = 5 to y1 = 20
the area under the line of equation y2 = 20 − y1 is entirely contained within the area
under the line of equation y2 = 55 2
3 − 3 y1 (that is, for y1 ≥ 5, if y2 < 20 − y1 then
y2 < 55 2
3 − 3 y1 ). 

y2

110

20
55
3

y1
0 110
9
20 55
2

Figure 11.19: The shaded area shows the union of the two shaded areas of Figures 11.17
and 11.18 which contains the set of points below the line of equation y2 = 20 − y1 .
 12. Glossary

Adverse selection or hidden type

A situation of asymmetric information where the uninformed party realizes that the
terms of a proposed contract determine the composition of the pool of individuals
who will find that contract acceptable; contractual terms that would be beneficial
to the uninformed party, if everybody found the contract acceptable, might have
undesirable consequences because of an adverse effect on the proportion of “bad
types” within the pool of applicants.

Asymmetric information
A situation where one of the two parties to a potential transaction has valuable
information that is not available to the other party.

Deductible
The amount, specified in an insurance contract, for which the insured is liable on
each loss (so that, in case of loss, the insurance provider will reimburse only an
amount equal to the difference between the loss and the deductible).

Expected value of a money lottery

A number associated with money lotteries, obtained by multiplying each monetary
prize by its probability and adding up all the corresponding numbers.

Full insurance
An insurance contract with zero deductible.

Incentive contract
A contractual arrangement according to which payments to one of the parties
are based, at least in part, on the observed outcome (taken as an indicator of
performance).
410 Chapter 12. Glossary

Index
An observable characteristic (such as race or gender) that cannot be changed.

Indifference curve
A graphical representation of the set of alternatives (e.g. contracts) that an individ-
ual considers to be just as good as some specified alternative (thus the individual
is indifferent among all of them).

Lottery
A list of possible outcomes with corresponding probabilities.

Money lottery
A lottery where the possible outcomes are sums of money.

Monopoly
A industry consisting of only one firm.

Moral hazard or hidden action

A situation where the behavior of one of the parties to a contract cannot be
monitored by the other party. Such behavior is relevant in that it affects the
probabilities associated with the various outcomes. The un-monitored party
behaves differently when fully shielded from risk as compared to the case where
he/she is partly or fully exposed to the risk.

Pareto efficiency
A situation X is Pareto efficient if there is no other feasible situation Y such that
everybody considers Y to be at least as good as X and at least one person strictly
prefers Y to X.

Partial insurance
An insurance contract with positive deductible.

Premium
The amount paid by the holder of an insurance policy for coverage under the
contract.

Principal-Agent relationship
A contractual relationship where one party (the Principal) hires another party (the
Agent) to perform a task whose outcome is typically affected by external factors
that are not under the control of either party.

Risk aversion
One of the possible attitudes to risk, where the individual considers a money
lottery to be worse than the expected value of the lottery for sure.

Risk neutrality
One of the possible attitudes to risk, where the individual considers a money
lottery to be just as good as the expected value of the lottery for sure.
 411

Risk love
One of the possible attitudes to risk, where the individual considers a money
lottery to be better than the expected value of the lottery for sure.

Risk sharing
The distribution of risk between the parties to a contract (typically a Principal-
Agent relationship) by means of a payment scheme.

Separating equilibrium
A situation where individuals are presented with a menu of choices and different
types of individuals make different choices, thereby separating themselves from
the other types.

Signal
An observable characteristic (such as an educational certificate) which is available
to, and can be chosen by, everybody.

Signaling
The phenomenon by which some types of individuals expend resources on a signal
in order to credibly convey information about themselves, since other types find it
too costly to imitate the signaling activity.
 Index

A D
adverse selection . . . . . . . . . . . . . . 227, 322 dominance
affine transformation . . . . . . . . . . . . . . . . . 61 first-order stochastic . . . . . . . . . 94, 389
Allais paradox . . . . . . . . . . . . . . . . . . . . . . 68 second-order stochastic . . . . . . . . . . 95
Arrow-Pratt measure . . . . . . . . . . . . . . . . . 89
asymmetric information . . . . . . . . 227, 299 E
attitude to risk
risk aversion . . . . . . . . . . . . . . . . . . . . 17 Edgeworth box . . . . . . . . . . . . . . . . . . . . . 167
risk love . . . . . . . . . . . . . . . . . . . . . . . . 17 Ellsberg paradox . . . . . . . . . . . . . . . . . . . . 70
risk neutrality . . . . . . . . . . . . . . . . . . . 17 expected utility . . . . . . . . . . . . . . . . . . . . . . 57
expected value . . . . . . . . . . . . . . . . . . . . . . 16
B
F
basic outcome . . . . . . . . . . . . . . . . . . . . . . . 55
fair odds line . . . . . . . . . . . . . . . . . . . . . . . . 28
belief . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
revision . . . . . . . . . . . . . . . . . . . . . . . . 231
H
updating . . . . . . . . . . . . . . . . . . . . . . . 229
hidden
C action . . . . . . . . . . . . . . . . . . . . . 343, 344
features. . . . . . . . . . . . . . . . . . . . . . . .236
CARA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 type . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
certainty equivalent . . . . . . . . . . . . . . . . . . 19
concavity . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 I
conditional probability . . . . . . . . . . . . . . 229
convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 incentive compatibility constraint . . . . 264
credit rationing . . . . . . . . . . . . . . . . . . . . . 242 index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
CRRA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 indifference curve . . . . . . . . . . . . . . . . . . 114
414 INDEX

reservation . . . . . . . . . . . . . . . . . . . . . 122 Petrakis, Emmanuel . . . . . . . . . . . . . . . . . 97

slope . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Pratt, John . . . . . . . . . . . . . . . . . . . . . . . . . . 90
individual rationality . . . . . . . . . . . . . . . 189 preferences
individual rationality constraint . . . . . . 264 completeness . . . . . . . . . . . . . . . . 19, 63
information . . . . . . . . . . . . . . . . . . . . . . . . 231 continuity . . . . . . . . . . . . . . . . . . . . . . . 31
insurance indifference . . . . . . . . . . . . . . . . . . . . . 19
budget line . . . . . . . . . . . . . . . . . . . . . 128 strict preference . . . . . . . . . . . . . . . . . 19
contract . . . . . . . . . . . . . . . . . . . . . . . . . 21 transitivity . . . . . . . . . . . . . . . . . . . 19, 63
mutual . . . . . . . . . . . . . . . . . . . . . . . . 138 Principal-Agent . . . . . . . . . . . . . . . 165, 369
insurance contract
deductible . . . . . . . . . . . . . . . . . . . . . . . 21
R
mutually beneficial . . . . . . . . . . . . . 122
premium . . . . . . . . . . . . . . . . . . . . . . . . 21 Rasmusen, Eric . . . . . . . . . . . . . . . . . . . . . 97
isoprofit line . . . . . . . . . . . . . . . . . . . . . . . . 25 rationality
average . . . . . . . . . . . . . . . . . . . . . . . . 260 individual . . . . . . . . . . . . . . . . . . . . . . 189
reservation
L indifference curve . . . . . . . . . . 122, 189
level of utility . . . . . . . . . . . . . . . . . . 349
lottery
utility . . . . . . . . . . . . . . . . . . . . . . . . . 189
compound . . . . . . . . . . . . . . . . . . . . . . 63
money . . . . . . . . . . . . . . . . . . . . . . . . . . 16 utility locus . . . . . . . . . . . . . . . . . . . . 348
non-degenerate . . . . . . . . . . . . . . . . . . 17 revision . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
simple . . . . . . . . . . . . . . . . . . . . . . . . . . 55 risk
allocation . . . . . . . . . . . . . . . . . . . . . . 165
M sharing . . . . . . . . . . . . . . . . . . . . . . . . 165
risk premium . . . . . . . . . . . . . . . . . . . . . . . 20
mean preserving spread . . . . . . . . . . . . . . 95 Rothschild, Michael . . . . . . . . . . . . . . . . . 97
menu of contracts
finite . . . . . . . . . . . . . . . . . . . . . . . . . . 127 S
infinite . . . . . . . . . . . . . . . . . . . . . . . . 128
monopoly . . . . . . . . 32, 124, 256, 257, 354 self-confirming beliefs . . . . . . . . . . . . . . 298
moral hazard . . . . . . . . . 322, 343, 344, 369 separating equilibrium . . . . . . . . . 302, 303
multiplicity of equilibria . . . . . . . . . . . . 304 signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
mutual insurance . . . . . . . . . . . . . . . . . . . 138 signaling equilibrium . . . . . . 298, 301, 303
Stiglitz, Joseph . . . . . . . . . . . . . . . . . . . . . . 97
N stochastic dominance
normalized utility function . . . . . . . . . . . 60 first-order . . . . . . . . . . . . . . . . . . 94, 389
second-order . . . . . . . . . . . . . . . . . 95, 98
P
T
Pareto
dominance . . . . . . . . . . . . . . . . . . . . . 167 type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
efficiency . . . . . . . . . . . . . . . . . 167, 236
improvement . . . . . . . . . . . . . . 236, 301 U
inefficiency. . . . . . . . . . . . . . . .301, 304
ranking . . . . . . . . . . . . . . . . . . . . . . . . 304 updating . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
perfect competition . . . . . . . . . 34, 125, 276 utility-of-money function . . . . . . . . . . . . 79
INDEX 415

V
von Neumann-Morgenstern
ranking . . . . . . . . . . . . . . . . . . . . . . . . . 57
utility function . . . . . . . . . . . . . . . . . . 57

W
warranty. . . . . . . . . . . . . . . . . . . . . . . . . . .322
wealth diagram . . . . . . . . . . . . . . . . . . . . . . 22