OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

When we talk about the market for, say, wheat or toothpaste, what are we
talking about? To a farmer, there is really no such thing as “wheat.” There
are literally hundreds of different species that we call “wheat,” and until
recently a farmer’s crop might be a mixture of quite a few of them. This
made buying and selling “wheat” difficult because while the farmer knows
what he is selling, the buyer does not.
But we now have markets for a limited number of species and grades
of wheat. These markets came about not because nature conveniently
produced a standardized product called “hard red winter wheat #2” but
because an economic organization—the Chicago Board of Trade—in the
mid-nineteenth century adopted classifications and policies to create
homogeneous categories of grain so as to facilitate transactions.2
Wheat once referred to a diverse collection of products, with size, genetic
strain, and quality differing from one sack of wheat to another. For two
farmers—Jones and Svenson—the supply, demand, and price for farmer
Jones’s wheat differed from the corresponding supply, demand, and price
for farmer Svenson’s wheat. The markets for the two farmers differed even
though they both sold wheat.
But, thanks to the Chicago Board of Trade, different grades of white
winter wheat, red winter wheat, spring wheat, and many other standardized
categories came to be of such uniform quality that the ownership of grain
no longer referred to any specific sack or particular lot of wheat, but to
a contract entitling the owner to the delivery of a specified amount of
some particular grade of wheat. As a result the biodiversity of wheat on the
American Great Plains fell, but a limited number of new markets, resembling
what one sees in an economics textbook, came into being.
Each type of grain has become a homogeneous good, for which traders
can write enforceable and complete contracts—simply for an amount with-
out worrying about the quality. The categories of grain are now like elec-
tricity: you purchase any amount of it by the kilowatt hour without caring
or knowing which power plant generated it. You do not worry about the
quality of the electricity you buy. What you buy is what you get. And if you
paid for electricity that you did not receive, you can get your money back.
Grain is not unusual in this respect. Goods like Sugar Number 11, Corn
Number 2 Yellow, or Light LA Sweet (that’s crude oil) are not gifts of
nature. Rather, they are created by a deliberate process of standardization
to eliminate difficult-to-monitor differences in quality.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

But the case of wheat is exceptional: it was possible by standardization to

make reliable information about the quality of what was being purchased
available to buyers, facilitating the exchange process.
But in most economic interactions there are some important bits of
information that are known to some of the actors but not to the others. The
language training instructor knows how successful he is in teaching fluency
in some new language; but those signing up and paying for his classes have
no idea of what quality of instruction to expect. The worker knows how
hard she worked yesterday, her employer may not. The borrower knows
how the loan will actually be used—prudently or recklessly, for example—the
banker may not. These are examples (as you already know) of asymmetric
information.
Other bits of information may be known to a buyer or seller, but not
admissible as evidence in a court of law (that is, non-verifiable). In most
legal systems, for example, the employer’s account that he found the
worker asleep at her desk, unless substantiated by witnesses, would not
be considered to be verifiable and could not be used as evidence against
the employee, for example to recover the wages paid to the employee for
her nap time.

Unlike electricity and red winter wheat #2 many of the commodities

transacted in a modern economy are not homogeneous goods or services.
The limited nature of information is the reason why, in many markets,
contracts are incomplete: traders cannot easily verify quality, or they can
observe the quality but such quality cannot be verified in a court of law or
by some other third party to enforce the contract. If they are disappointed
with the quality, there is often no way to get their money back. Think of
hiring someone to care for an elderly relative or to take care of your children
while you are at work. How can you know the quality of the care they are
giving? And even if you later found out that they had been careless, could
you go to court to get back the wages you paid them? Almost certainly not.
Contract enforcement by courts or other third parties is called exoge-
nous enforcement because the enforcers are not parties to the exchange
(they are outside the exchange). But for many exchanges this is not the case:
the contract is incomplete.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

Complete and incomplete contracts share two features:

• Mutual gain: a transaction is based on the mutual expectation of gain by

all the participants.
• Conflict of interest: there is a conflict of interest over how these gains will
be divided among the parties to the exchange.

The key difference is that when contracts are incomplete the division of
the gains is not enforced entirely by an external body—the courts—based
on terms specified in the contract. The terms of the contract matter and
courts may be involved, but the outcome of the exchange is also the
result of strategic interactions among the participants involving rewards,
punishments, and the exercise of power.
These include threats, promises, and the creation of incentives through
offering repeated interactions. The preferences of the parties to an
exchange also matter, for example a commitment to telling the truth
about the condition of the used car you are selling, or a worker’s intrinsic
motivation to do high-quality work.
This is what Durkheim meant—quoted in the introduction to this part of
the book—when he said that “not everything in the contract is contractual”
and that market exchanges are socially regulated. We call this the endoge-
nous enforcement of the terms of an exchange.
Students are experts on incomplete contracts. Suppose as a condition of
your employment at a consulting firm following graduation you contracted
to “learn microeconomics.” How would it be determined that you had
fulfilled your obligation? The contract might have been written “pass this
particular course in microeconomics” but the employer would hardly be
satisfied that this would guarantee that you were able to do the kinds of
work they need. Or suppose you did poorly in the exam. Had you failed
to learn microeconomics? Or did the exam not test your knowledge of
microeconomics? Or did you have a particularly bad allergy attack the day
of the exam? How could your employer ever enforce this contract?
Here are some other examples of exchanges under incomplete
contracts.

• Owners of firms want managers to maximize the value of the owners’

assets, but managers have their own objectives (first-class air travel,
lavish offices, “on-the-job leisure”) and managerial contracts fall far short
of having an enforceable requirement to maximize the owner’s wealth.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• In many countries, families devote a sizable fraction of their budgets to

purchasing educational services for their children, the quality of which is
rarely specified in a contract (and would be unenforceable if it were).
• Parents, in turn, hope and expect that their children will care for them,
if needed, in their old age; but there is no way to write this into an
enforceable contract.

Three of the most important examples of incomplete contracts are the

subject of this chapter and the next two chapters. Here we consider the
case where information on the quality of a good is known to the seller but
not to the buyer—whether it is a used car or a piece of clothing provided to
Benetton by a subcontractor. In Chapters 11 and 12 we study incomplete
contracts in the labor market where information on the effort you put
into your job is not readily available to your employer, and the credit
market in which your promise to repay the money you borrowed may be
unenforceable if you are broke. As these examples suggest, contractual
incompleteness is the rule rather than the exception in economic transac-
tions. Here are five reasons why.

Asymmetric or non-verifiable information: Third-party enforcement of

contracts requires information that is available to both parties and can
be verified by third parties such as courts of law. Information is verifiable
if it can be used in court to enforce a contract. Non-verifiable infor-
mation such as hearsay, or even direct but uncorroborated eyewitness
observation, generally cannot be used to enforce contracts. Information
is asymmetric if something is known by one party but not by another.
Time: A contract is generally executed over a period of time as when a
contract specifies that Party A does X now and Party B does Y later. But
what if what B does later depends on other things that cannot now be
determined? A complete contract must specify what the parties must do
in every possible future situation or contingency or “state of the world.”
In general, people cannot completely specify these future states, and in
any case, it is not ordinarily cost-effective to specify what to do in each
contingency.
Measurability: Many of the services or goods involved in the exchange
process are inherently difficult to measure or to describe precisely
enough to be written into a contract. The restaurant owner would like
his serving staff to interact in a pleasant manner with customers, but
how can this be observed by the owner, and even if it were to be
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

observed, how could it be measured or considered to be verified for use

in a legal proceeding?
Authority: For some transactions there is no institution—no court or
other relevant third party—capable of enforcing a contract. Many inter-
national transactions are of this type. For example, if a country defaults
on its debt to international creditors, no third party enforces the claims
on the debt. This has happened in a variety of countries internationally,
such as Lebanon’s default on $1.2 billion of Eurobonds in 2020, or
Argentina’s debt restructuring that took place repeatedly during the
period 2005–2016.
Motivation: Even where the nature of the goods or services to be
exchanged would permit a more complete contract, traders may favor a
less complete contract for motivational reasons. Intrusive surveillance
of workers by employers to establish verifiable information on work
activities, for example, may backfire if the employer’s distrust angers
workers, leading to less satisfactory work performance.

As the final reason suggests, how incomplete a contract will be is in

some measure a matter of choice. For example, where the parties to an
exchange are trusting and trustworthy people committed to reciprocity in
their dealings, perhaps having reciprocal or altruistic preferences like those
we saw in Chapter 2, they may deliberately leave some important aspects of
the exchange unspecified, even when the relevant enforceable contractual
clauses could be written.
The first and fourth reason for incomplete contracts above—lack of
verifiability and authority—make it clear that whether a particular good or
service is subject to complete contracting will differ from one legal system
to another. The completeness and enforceability of contracts depends on
legal institutions in other ways as well. The ability of a lender to enforce
a debt contract against a borrower may be greatly influenced by whether
legal institutions include bankruptcy or other forms of limited liability that
protect some of the borrower’s assets from being taken by the lender, or,
at the other end of the spectrum, imprisonment of delinquent debtors on
behalf of creditors.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

A principal–agent relationship (also called an agency problem) arises when

two conditions hold:

• Conflict of interest: the actions or attributes of the agent affect the payoffs
of the principal in such a way that there is a conflict of interest between
the principal and the agent. The employer, for example would like the
employee to work harder; the employee would like to go home a little
less exhausted at the end of the day.
• Incomplete contract: the agent’s actions or attributes are not known to the
principal (or, if known, they are not verifiable) and so cannot be subject
to enforceable contract. How hard the worker works cannot be specified
in an enforceable contract.

Both conditions are necessary. If there were not a conflict of interest,

then the agent would simply do what the principal desired (both would
desire the same thing) without an enforceable contract. It would be as if
the principal himself carried out the necessary action.
If a complete contract covered all of the agent’s actions that mattered to
the principal, then a conflict of interest would not make the relationship
special: “purchasing” the agent’s action would be no different from the
principal buying some quantity of wheat or electricity. The models of
exchange with complete contracts—used earlier in the book—would be
perfectly adequate.
We can classify principal–agent problems into two categories:

• Hidden actions: These are things that an agent does that the principal
has some interest in, but does not know (or lacks verifiable information
about), such as the effort of an employee or the business practices of
a borrower. An example is that a person whose home is fully insured
against fire may take less care to avoid fires. Insurance companies call
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

this behavior moral hazard, and that term is sometimes used to apply to
any hidden action problem.
• Hidden attributes: There are characteristics of an agent—what the agent
is—that the principal has some interest in, but does not know (or lacks
verifiable information about) such as which drivers are reckless, or which
patients are seriously ill.

To understand how mutually beneficial exchanges take place without

complete contracts and how the benefits are divided among the parties, we
need to use the concepts we have already developed—best response, first
mover, fallback option, and non-clearing markets—to construct a new set of
analytical tools called principal–agent models. Especially important among
the methods you have already learned is the theory of repeated games
introduced in Chapter 5. Most jobs are not one-shot interactions: they go
on year after year. So the game between the employer and the worker is
repeated, and what each party does in one period depends on what they
expect to happen as a result later periods. Remember that in a repeated
game what is called the stage game—like a simple Prisoners’ Dilemma—is
played more than once with the same players.
Models based on repeated games of this type are used to study transac-
tions between employers and workers, lenders and borrowers, and a wide
set of other exchanges as shown in Table 10.1. They range from the landlords
and sharecroppers that we mentioned in Chapter 2 to a fundamental
problem of democracy: citizens trying to control their governments.
In the right-hand column we list some of the strategies followed by prin-
cipals to get agents to act in their interest. In the first row—about employers
as principals and workers as agents—the term contingent renewal means
that because employment is repeated game, the principal has the option
to not renew the relationship (fire the worker) and whether this happens is
contingent on (depends on) whatever information the employer gets on the
worker’s job performance.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

A “lemon” (in American slang) is a used car you discover is defective after
you buy it. A “peach” is a used car you discover works better and costs less to
maintain than you expected after you buy it. The problem the existence of
lemons poses to economic markets can be illustrated in a model of a used-
car market where the principals are the prospective buyers, the agents are
the sellers, and the hidden attribute (whether the car is a lemon) is known
only by the seller.
Consider the following example (summarized in Table 10.2):

• Every day, ten owners of ten used cars consider selling.

• The cars differ in quality, which we measure by the monetary value of the
car to its owner. Quality ranges from zero to 9,000 in equal steps: there
is one worthless car, one worth 1,000, another worth 2,000, and so on.
The average value of the cars is therefore 4,500.
• There are many prospective buyers, and each would happily buy a car for
a price equal to its true value, but not more, which is their willingness to
pay for a car.
• Sellers do not expect to receive the full value of their vehicle, but they
are willing to sell if they can get even just a little more than half the true
value, which is their willingness to sell for the car.
• The potential economic mutual gain—the sum of buyers’ and sellers’
surpluses—will be the difference between the willingness to pay and to
sell, or half the price of the car.

Imagine that prospective buyers could ascertain the quality of each car,
and approach each seller to bargain over the price. Then if sellers also knew
each buyer’s willingness to pay, by the end of the day all of the cars (except
for the entirely worthless one) will have been sold at a price somewhere
between their true value and half the true value. All mutually beneficial
trades would have taken place.
But if potential buyers cannot ascertain the quality of any particular car
that is for sale, the market will not work. Suppose that those who bought
cars yesterday find out the true value of their purchase and post it on social
media. Then today, the potential buyers will know the true value of the cars
sold the previous day. They still do not know the true value of any of the
cars for sale today. But they might reasonably adopt the rule that the most
they are willing to pay for a car today will be the average value of the cars
sold yesterday.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

Now suppose that ten cars had been offered on the market on the first
day. We use a proof by contradiction to show that, one by one, the highest-
quality cars will drop out of the market, until there is no market in used
cars. Consider the market on the second day:

• On the first day all the cars (as we assumed at the start) were put on the
market and sold at their true value which is the highest price at which
they possibly could have been sold.
• The average value of these cars was $4,500, so the most a buyer is willing
to pay today for any car will be $4,500.
• At the beginning of the second day, each prospective seller expects a
price of $4,500 at the most. Most of the sellers are happy: $4,500 is more
than half the true value of their car.
• But one owner isn’t pleased. The owner of the best-quality car ($9,000)
would not sell unless the price exceeds half the value of his car: more than
$4,500.
• On the second day the owner of the best car will not offer it for sale. No
one with a car worth $9,000 will be willing to participate in the market
on the second day.
• The rest of the cars will sell on the second day: their value averages
$4,000.
• On the third day, buyers will know the average value of the cars sold on
the second day, and will be willing to pay at most $4,000 for a car.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• The owner of the second day’s highest-quality car (the one worth $8,000)
will know this, and will not offer her car for sale on the third day.
• As a result, the average quality of cars sold on the third day will be $3,500.
The owner of the third-best car will not put his car up for sale on the
fourth day.
• And so it goes on, until, after ten days, only the owner of a lemon worth
$1,000 and a totally worthless car will remain in the market.
• If cars of these two values sell on the tenth day, then, on the eleventh
day, uninformed buyers will be willing to pay at most $500 for a car of
any quality.
• Knowing this, the owner of the car worth $1,000 will decide she would
rather keep her car than try to sell on the eleventh day.
• The only car on the market on the eleventh day will be worth nothing: the
cars that remain on the market are lemons, because only the owner of a
worthless car would be prepared to offer that car for sale.

Economists call this process adverse selection, because the prevailing price
selects which cars will be left in the market. The market of uninformed
buyers selects against quality, and is adverse to the interests of potential
buyers and sellers holding high-quality goods.
The lemons problem emphasizes the dependence of real-world markets
on the nature of the information available to the exchanging parties and
to the courts. Unless information is available to sustain the differentiation
of products, contracts will converge to the lowest-common-denominator
level for products. Lemons problems are particularly acute in markets for
insurance and credit, where information is at a premium, but can occur in
a wide variety of other situations.
Whether the information available is sufficient to allow markets to work
will depend on the economic and legal institutions governing the exchange
process. In the case of a commodity like #2 red winter wheat, the informa-
tion required to sustain the market is provided as a kind of public good by the
agency that sponsors the market and certifies the quality of the product.
There are many factors that could help to ensure that the lemons prob-
lem does not persist. Quality assurance bodies, like those for the wheat
described above, are one such solution. Another is the introduction of
legislation by third parties (states) to create compensation or liability rules
such that those who purchase defective products in one-shot exchanges
can receive full compensation or exchange their goods. This idea is demon-
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

strated in the ‘lemon laws’ in most developed countries nationally, or across

a variety of US states.
Another method by which parties can exchange to reduce the severity
of the lemons problem is to repeat the interaction. When interactions are
repeated, players can build up and sustain a reputation. Reputations are
costly to build and maintain and if a player acts contrary to their reputation,
then they may lose the investment they have built up in their reputation.
Examples abound. In online markets, second-hand goods are sold by sellers
with a star rating saying whether or not it is worthwhile purchasing from
them and the rating information is symmetric. Car dealerships build up
reputations for honest dealing and good-quality vehicles through word of
mouth and customer satisfaction reports.
Finally, as we have seen in Chapter 2, people often wish to uphold social
norms and to punish those who do not. Among these norms are honestly
reporting the nature or quality of a good one is selling, providing no
guarantees, but sometimes providing the social regulation of the contract
that Durkheim mentioned as essential to buying and selling.
When we move from the problem of hidden attributes to hidden actions
we encounter a new set of reasons why sustaining mutually beneficial trade
may be difficult, even when mutual gains are technically possible.

The lemons problem illustrates adverse selection due to hidden attributes.

The problem is far more general than the used-car market. To see why,
think about health insurance. Imagine hypothetically that you will be born
into a population, but do not know whether you would be born with some
serious health problem, or might contract such a problem later in life, or
perhaps be entirely healthy until you die of old age.
Now imagine that before you were born, or even knew who your parents
would be you were asked this question: Would you buy health insur-
ance if the premium—that is, the cost of the insurance to you—which is
the same for everyone, is just sufficient to pay for the medical services
required across the entire population if everyone agreed to purchase
it? This kind of hypothetical decision-making process is called making a
decision behind a veil of ignorance because it is a thought experiment in
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

which you are invited to think about how you would act or what policies
you would favor if you did not know the state of your actual health.
Most people would be willing to purchase health insurance behind the
veil of ignorance, because they would rather pay a premium representing
the average costs of healthcare to the whole population rather than be
individually responsible for paying for the treatment of a serious illness,
which, even if it is very unlikely to strike them, would impose high costs
that most families could not pay. The benefit of protecting oneself and
one’s family from a financial catastrophe (or the possibility that you can’t
afford healthcare when you need it) is worth the insurance premium
on average.
But the thought experiment is unrealistic: we cannot sign people up for
health insurance before they know how healthy they will be. The reason is
that this would require signing them up for insurance before they are born.
Though most people would buy fairly priced health insurance if they did not
know about their future health status, the situation changes dramatically if
they can choose whether to buy health insurance knowing something about
their current and future health status.
Let’s look at the situation from the standpoint of the insurance
provider (the principal) paired with a prospective buyer of insurance
(the agent):

• People are more likely to purchase insurance if they know that they are
ill or likely to become ill. The average health of people buying insurance
will be lower than the average health of the population.
• This information is asymmetric: The person buying the insurance knows
more than the insurance company about how healthy they are.
• Insurance companies selling insurance to people who are sicker than
average will be profitable only if they charge higher premiums than if
everyone bought the same insurance.
• This will lead some—those who are reasonably certain that they are
healthy and will remain so—to not purchase insurance.
• To remain in business, the insurance companies will have to charge
even higher premiums. Eventually the vast majority of the people buying
insurance will be those who know they already have or are likely to have
a serious health problem.

This is a case of adverse selection again. The reasoning above shows why
the hidden attributes of initial health status can result in an unraveling of
the healthcare market. Why? Insurance companies make profits by insuring
people who are healthy. Healthy people who want to buy insurance in case
they fall ill in the future are priced out of the market, and will not buy
insurance. In the extreme case, the health insurance premium will be so
high that only people who know they are likely to become seriously ill will
buy insurance (people with initial conditions of poor health).
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

In this case we have what is called a missing market. It is a market that

could exist, but it would only exist if health information were symmetrical
and verifiable (ignoring for the moment the problem of whether everyone
would want to share their health data). Under those imaginary conditions,
the market could provide benefits to both insurance company owners and
people who wanted to insure themselves. Not having such a market is
Pareto inefficient.
To address the problem of adverse selection due to asymmetric informa-
tion and the resulting missing markets for health insurance, many coun-
tries have adopted policies of compulsory enrollment in private insurance
programs or universal tax-financed coverage, such as the National Health
Service in the UK, or similar services in Canada, France, and elsewhere.
When we move from the problem of hidden attributes to hidden actions
we encounter a new set of reasons why sustaining mutually beneficial trade
may be difficult even when mutual gains are technically feasible.

Insurers, whether private or governmental, face problems other than hid-

den attributes. There is also the problem of hidden actions: buying the
insurance policy may make the buyer more likely to take exactly the risks
that have been insured against. A person who has purchased full coverage
for his car against damage or theft may as a result take less care in
driving or in securing his vehicle than someone who had not purchased the
insurance.
Insurers typically place limits on the insurance. For example, insurance
coverage may not apply or be more expensive if someone other than the
insured person is driving, or if it is parked on a daily basis in a theft-prone
location. These provisions can be written into an insurance contract and are
enforceable. But, the insurer cannot enforce a contract about how fast you
drive or whether you drive after having had a drink. These are the actions
that are hidden from the insurer because of the asymmetric information:
you know these facts, but the insurance company does not.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

Here is a model of incomplete contracting in the case of hidden actions,

inspired by Benetton, the global casual wear marketer. To understand the
model, let’s first explore some of the history of the Benetton company.

The Benetton family—three brothers and a sister born during the Great
Depression and World War II—began with a small company selling sweaters
to shops near the town of Treviso in northern Italy half a century ago.
Benetton grew to become one of the world’s largest designers, producers,
and sellers of casual wear and other garments. A key to its early success was
a highly decentralized system of production: the labor-intensive aspects of
production—primarily sewing—were carried out by hundreds of small sub-
contractors, working to designs and schedules and with materials supplied
by Benetton.8 A few processes were done by Benetton itself—notably dying,
performed at the last minute so as to keep the product in tune with fashion
trends. In 2020 Benetton’s subcontractors outnumbered their employees
by a factor of 17 to 1.9
Importantly, quality control and marketing were centralized, performed
by Benetton’s own staff. Subcontractors who reliably produced goods of
the specified quality benefited from permanent orders, quick payment by
Benetton, and other benefits of their long-standing relationship with the
company.
As a result of this decentralized subcontracting structure, Benetton saved
on the costs of establishing its own production facility. The most valuable
asset of the company was not garment-making factories but instead the
trademark, the Benetton name itself (called an intangible asset), which,
once the garments were acquired from the subcontractors, was of course
attached to the product before sale. Many contemporary companies have
similar structures, from manufacturing to design to developing code and
online applications.

To model these relationships we introduce a buyer (the principal, Patrisia)

who purchases a good, say, a shirt, from a supplier (the agent, Armin, a
subcontractor) for a price, p and then puts her trademark on the shirt and
sells it to a consumer (who is not a player in this game). Their interaction is
depicted in Figure 10.2.
The good—the shirt—has an aspect, its quality, q that is important to the
principal, and costly or difficult for the agent to provide. Patrisia (who might
be the Benetton company in this model) would like to pay a low price for a
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

q p
tq

high-quality shirt, and Armin (one of the subcontracting shirt producers)

would like to receive a high price for a shirt of lesser quality. So their
interests conflict.
The principal–agent problem arises from this conflict of interest along
with the fact that the quality of the shirt cannot be readily determined
by Patrisia. She has the problem that shirts are sometimes rejected by
customers, and Patrisia then has to refund the price when they return
the good.
The challenge that the principal, Patrisia, faces is that she knows that
Armin would prefer to produce lower-quality goods, but it is not cost-
effective to determine the actual quality of each unit that she purchases.
So she comes up with the following plan:

• Repeated relationship: Offer Armin an ongoing (or repeated) subcontract-

ing relationship with her, so he can count on her buying his products year
after year.
• Termination probability: But let him know that if he provides a shirt that
is rejected by a consumer, she will terminate the relationship.
• Enforcement rent: pay him enough for the shirts he provides so that he
does not want the relationship to be terminated, giving him a motive to
provide quality shirts.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

In other words she structures her interaction with Armin as a repeated

game that she can terminate if she is not satisfied with the quality he
provides.
Patrisia interacts with a large number of subcontractors like Armin. We
simplify the problem by assuming that each period she purchases a given
number of shirts (it could be one or one thousand) from Armin and each
of the other subcontractors. So she is really choosing the price and the
number of subcontractors from whom to buy.
Here is the structure of the game. The principal is first mover and seeks
to maximize her profit by deciding on:

• price: the price to offer the agent (and the other subcontractors); and
• quantity: how many shirts to purchase.

The agent seeks to maximize the expected value of his utility over the
duration of his doing business with the principal; he has just one decision:

• Quality: the quality of the good to supply.

The principal sets the price knowing the agent’s best-response function,
that is, the quality he will supply for every price she could offer. The agent
chooses the quality to provide knowing the price the principal has offered
and the probability that the transaction will be terminated for every level
of quality he could provide.
To determine what the first mover (the principal) will do we need first
to derive the best-response function of the second mover (the agent). The
level of quality the agent supplies will depend on how valuable it is to the
agent to continue the relationship with the principal. This is the rent that
the agent will receive, and the reason why he will supply higher quality than
he would provide otherwise.

Because this is a repeated game, how important it is to the agent that the
relationship with the principal continue will depend on the following:

• the utility that the agent experiences in each period;

 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• the number of periods that the principal and agent will interact; and
• the utility the agent will experience if terminated by the principal, that is
u
the agent’s fallback option.
q uq

The seller, Armin, prefers to receive a higher price p and to provide a lower
level of quality q (which can take any value from 0 to 1, representing a range
from a quality of 0 percent to 100 percent). Armin prefers to provide lower u
quality because it is costly—in terms of effort or care—for him to produce 1 q

high-quality goods.
q uq
For concreteness we will express the general form of his utility function
(below on the left) by a specific form (on the right below):

u u
Agent’s utility upq p (10.1) uq
1 q q

u
uq
Equation 10.1 says the following: 1 q 2

• Because the marginal utility of the price, up 1 0, Armin considers p to

be a “good” something he would like more of because getting more money
is valuable to Armin.
• Because the marginal utility of providing quality uq 0, Armin considers
q to be a “bad,” so he would rather not exert the effort required to provide
q 0.
u
The second term on the right hand side ( ), is Armin’s disutility of
1 q
providing higher quality shown in Figure 10.3. We can see three things:
q 1
• if he provides no quality (he just hands over a product with the lowest u
q 1
possible quality, i.e. q 0) his disutility is u; 1 q

• the higher the quality he provides, the more disutility he experiences: the
curve is upward sloping; and
• the marginal disutility of quality is increasing: the curve is steeper at point
j than at point k.

From Equation 10.1 you can also see that he will never choose to produce a
perfect good (100% quality or q 1) because at q 1 his disutility is infinite
or undefined.
There are two confusions to avoid here, one about slopes and the other
about signs. First be clear about the difference between:

• the height of the curve which is the disutility of providing the level of
quality indicated on the x-axis quality; and
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

40
Maximum
u level of
Disutility =
(1 q) quality, q
Disutility of providing quality

Slope = uq
u
=
(1 q)2
k

u=5

0
0 1
Quality, q

• the slope of the curve, which is the marginal disutility of quality, namely
the effect on his disutility of providing a bit more quality when he is
already providing the level indicated on the x-axis.

The second is the distinction between:

• the marginal utility of providing quality, uq which is negative (providing

quality reduces Armin’s utility); and
• the marginal disutility of quality which is just the same thing with a
negative sign, uq , and so is positive.

We can construct indifference curves for Armin based on his utility

function: this is another example in which the choices include both a good
(receiving the price) and a bad (providing quality). As a result, for any
quality q on the vertical axis in Figure 10.4 Armin would prefer to obtain
a higher price, p, which would place him on a higher indifference curve;
and for any price, p, on the horizontal axis Armin would prefer to provide a
lesser quality. The negative of the slope of the indifference curve is Armin’s
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

marginal rate of substitution between more pay (the price he receives) and
more quality. We derive this in M-Note 10.1.
We already know from Figure 10.3 and Equation 10.1 that increasing q
reduces Armin’s utility (increases his disutility) and does so more and more
as q approaches 1. This is the reason why in Figure 10.4 Armin’s indifference
curves become almost flat when Armin provides high levels of quality:
Armin experiences an extremely high marginal disutility of providing any
additional quality. At already high levels of quality providing additional
quality (moving up in the figure) can be compensated only by a very large
increase in price (moving to the right), as shown by the flattening of his q
indifference curves as q gets larger.
p

p q
p q u u 5

2 1 0

Maximum level of quality, q

1

Slope = mrs =
(1 q)2
u
a b
Quality, q

u0 u1 u2
0
0 5 10 20 40
Price, p
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

u upq
q p

dp dq

du dq uq dp up 0
u u
uq up
q p

dp uq

dq uq
dq
dp

dq up
dp uq

p
mrs p q
up
uq
1
uq

mrs

uq
q uq
u
upq p
1 q
u
uq 1 1
1 q 2

u
1 q 2

up 1
mrs p q
uq uq
1
u
1 q 2

2
1 q
u
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

p mrs p q

u
upq
q 04 q 08
mrs p q u 5
mrs

To make her plan work, Patrisia has to do two things:

• get some information on the quality Armin provides as the basis for
terminating the contract if necessary; and
• offer him a high enough price so that he receives a rent, and therefore
he will prefer continuing doing business with Patrisia rather than being
terminated.

She uses consumer complaints about quality as the basis for her termina-
tion decision and as a result she will terminate the relationship with Armin
with probability t 1 q. This means that:

• If Armin provides no quality at all (q 0), then there will surely be a

complaint, so he will surely be terminated (t 1).
• If Armin were to provide q 1, then no consumer would ever complain, so
he would be certain that Patrisia would continue buying from him (t 0).
• If Armin provides q 0 5, whether he gets terminated at the end of each
period is a coin toss: 50 percent chance of keeping the contract and
50 percent chance of losing the contract.

If he provides quality of q, then at the end of the first period he will lose his
contract with probability 1 q. If he is lucky and does not lose the contract,
then the first period is just repeated—she offers the same price and so he
offers the same quality. As a result, the next period will be his last with
the same probability 1 q and so on until his luck runs out and he is
terminated.
How long will his contract with Patrisia last? Armin does not know for
sure: he has to think about probabilities. To do this, imagine a game of
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

flipping a coin, in which you flip it once and if it came up heads you flip
it again, and continue doing this until it comes up tails, at which point the
game ends. If you did this many times, the game would sometimes end right
after the first flip (because it came up tails); sometimes it would continue
for many flips.
But on average how many flips would you expect to make, including the
first one? The answer (which we show in the Mathematics Appendix) is two
periods, which is just one divided by one-half, the probability that the game
will end after each flip. So the expected number of periods that Armin’s
contract will last, T q will be one divided by the probability of termination
at the end of each period, or t q 1 t q , where t q 1 q.
The value he gets from doing business with Patrisia is the utility he gets
in a single period times the number of periods he expects to do business
with her.
vpq
• The utility per period is u p q .
• The expected lifetime (number of periods) of the contract is T q , which
is equal to 1 divided by the probability that he gets terminated in a given
1
period, or T q 1tq .
1 q
• So:

Expected value Utility per period Expected lifetime of contract

vpq upq Tq
1 1
upq upq (10.4)
1 q 1 q

This is how much he values being Patrisia’s sub–contractor. To under-

stand how motivated he will be by the threat of her ending the contract we
need to know about his other options. How much he would like to keep his
job is the difference between how much it is worth to him, v, and what he
would be able to get if his relationship with Patrisia were to be terminated.
This next-best opportunity is his fallback option, which is finding another
buyer and attempting to sell him the (presumably low-quality) good he has
made. Later we will discuss his fallback option but for now we assume that
his utility is zero if Patrisia terminates their relationship.
Because his fallback option is zero, it follows that v itself is Armin’s
enforcement rent, the loss of which he would like to prevent by supplying
higher quality than he would otherwise do. If v 0, then the following
will hold:
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• It is a rent because it is how much his current situation is preferable to vp vq

his next-best alternative.
v
• It is an enforcement rent because the rent motivates him to provide more
p q
quality than he otherwise would (he doesn’t want to lose the contract).

In Figure 10.5 we show Armin’s iso-value curves. Each curve is made

up of all of the combinations of p and q that give the same values of v
in Equation 10.3 with v4 v3 v2 v1 v0 0. The iso-value curves differ
from the single-period utility indifference curves in Figure 10.4: they are not vp
mrs p q
vq
uniformly upward-sloping. The upward-sloping portions of the iso-value
curves are easy to understand. In Figure 10.5 Armin is indifferent between
point f—providing high quality and getting paid a high price—and point e—
providing lesser quality at a lesser price.
But why is Armin indifferent between point e and point g, where he is
providing less quality and getting a higher price, which would seem to be a
better deal than point e?

p q v u 5

q vp
mrs p q
p vq

v0 0
x p 5
q 0 x p 5

Iso–v curves

f
Quality, q

1
n Better for agent
2
e

Slope of iso–v
= –mrs
g vp
v0 = 0 v1 v2 v3 v4 =–
vq
0
0 u 10 = 2 u 20 = 4u 40
Price, p
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

The answer is that when Armin takes account of the likelihood of being
terminated, providing more quality is not necessarily a “bad.” The reason is
that providing more quality will increase his chance of keeping his contract
with Patrisia which (if v 0) he values. At some low levels of quality,
providing a little more quality will increase v because it will prolong the
duration for which he receives the per-period utility. The iso-value function
has a different shape from the single-period utility indifference curves in
Figure 10.4 because the game is repeated, and Armin has an interest in
continuing his relationship with Patrisia.

2u
q 05 v 0

upq
up0

For any particular price that Patrisia offers, there are three things that
Armin could do:

• refuse the contract and have utility u 0;

• accept the contract but deliver q 0 in which case his contract would
be terminated with certainty, so he would receive u p 0 for a single
period, and the value of his transacting with Patrisia would be v p 0
u p 0 ; or
• accept the contract, deliver some level of quality q 0, and receive u p q
1
for an expected number of periods, , and receive v p q .
t q

Armin will choose quality (q) to maximize his value (v), taking account of
the fact the higher q will reduce the probability of termination t (remember
at the end of any period, t 1 q). Suppose hypothetically in Figure 10.6
that the price offered by Patrisia is p 20. Then we can think of Armin’s
optimizing problem in the following way, as we did in the other best-
response functions we have derived in previous chapters. Starting at some
low level of quality (point g) he would see that as he provides more
quality, he reaches higher iso-value functions (not shown) until he reaches
point n, the tangency of the iso-value curve labeled v4 with the vertical
line indicating the hypothetical price. If he proceeds upward—offering more
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

quality—then he will be crossing ever lower iso-value curves, such as the one
at point f.
The quality the agent offers and price indicated by point n is one point on
Armin’s best-response function. Two others, constructed in the same way
but at lower prices are points e and d.
Here is how we can derive an equation giving the agent’s best response
to the principal’s price, that is, the best-response function shown as the
q
purple line in the Figure 10.6. He will reject the zero quality option and want
to supply more quality as long as the disutility of doing that—the marginal p
cost of quality—is less than the marginal benefit that he derives from
providing additional quality. That is he will want to compare two negative
quantities:

• Marginal cost: uq , namely the reduction in his utility associated with

providing more quality (the marginal utility of effort).

2u 10 q 0

u 2u

Iso v curves

f
Best response
function (ICC)
Quality, q

1
2 n
Participation e
constraint
(PC)
d

Slope of iso v
= mrs
g vp
v0 = 0 v1 v2 v3 v4 =
vq
0
0 u 10 = 2u 20 = 4u 40
Price, p
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• Marginal benefit: tq v, namely the reduction in likelihood of being ter-

minated made possible by providing more quality times the effect of
working harder on the probability of keeping the job.

So, as shown in M-Note 10.3 the level of quality that will maximize his
value is that at which the marginal costs and marginal benefits are equal
(remember both uq and tq are negative):

mc mb
uq q tq q v p q (10.5)

Marginal utility Reduction in termination probability

enforcement rent

The values of q satisfying this equation for each value of p is the best-
q
response function.
To see what Equation 10.5 means, return to Armin’s iso-value curves as
p
shown in Figure 10.6. We show in M-Note 10.4 that if we use the specific
utility function in Equation 10.1, then Armin’s best-response function (for
values of p 2u) is given by:
2u
Armin’s best response: qp 1 (10.6)
p
This is the best-response function we derived in the figure, using u 5.
You can see from Equation 10.6 why at prices lower than p 10 2u Armin
will not provide any quality at all. The agent would accept the contract and
supply zero quality if the price were between 2u and u. This is because
over that price range, the participation constraint is satisfied (so the agent
“participates” in the contract) but prices in that range do not provide
sufficient incentives for the agent to raise quality above zero. At a price
below u the agent will not accept the contract.

vpq
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

1
vpq upq
tq
q

uq t u tq
vq
t2
uq t u tq
0 vq
t2
0 uq t u tq
uq t u tq
u
t uq tq
t
u pq
vpq
t q

uq tq v

u t v p q
uq p q tq q v p q

uq
v p q
qp

dq vp
mrs p q
dp vq

q
vq 0 vq 0
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

uq tq v
t 1 q tq 1

u u 1
1 p
1 q 2 1 q 1 q

1 q
u u
p
1 q 1 q
2u
p
1 q
1 p
2u
1 q 2u
2u
1 1 q
p
pq
2u
q q 1
p

2u
qp 1
p
p

The principal makes two decisions: how many units to purchase (which is
equivalent to how many subcontractor–agents to engage) and how much
to pay them for each unit. We have set up the problem so that we can focus
on the second—the principal–agent problem. We have already analyzed
models to address the first question of how many shirts she should sell to
consumers, in Chapters 8 and 9.
To study the price-setting process in the principal–agent relationship,
we proceed in three steps:

• Patrisia, the principal, knowing Armin’s best response q p , determines

the price pN that will minimize the cost of acquiring quality p q (we use
the N superscript because this will be the Nash equilibrium price).
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• Armin, the agent, best responds to the price offer by choosing the Nash
equilibrium quality, qN , the quality level that maximizes his value given
the price (using his best-response function).
• When pN is offered and qN is the response, then the expected number of
1 1
periods that Armin’s relationship with Patrisia will last is TN N N
.
t q 1 q

To show how the principal will set the price p in order to minimize the cost
of acquiring quality, we show a variety of different potential ratios of price to pq
quality to price (p q) in Figure 10.7 as isocost rays. Along a given isocost ray, 00
Patrisia has the same cost of quality, as can be seen by comparing points b
and d. This is why they are called isocost (“equal cost”) rays. Patrisia prefers
isocost c1 to c2 to c3 because c1 c2 c3 . Comparing points a on c3 , with b

q p c3 c2 c1

q 05
c1 q 0 25 q 0 125
c2 c3 c2

qp 00
0 25
q 0 25 p 10 qp 0 025
10
05
q 05 p 20 qp 0 025
20

1 c1 c2

q
Slope =
p
Better for
principal
Quality, q

1 c3
2 c d

1
4 b

0
0 10 20 40
Price, p
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

on c2 , and c on c1 , we can see that at the same price, the quality at point c is
higher. So, steeper rays are better for the principal (corresponding to lower
cost): the slope of any ray is q p.
Think of these rays as similar to the indifference curves representing the
objectives of a decision maker or the iso-profit curves representing
the objectives of the owners of a firm. Here the isocost rays represent
the cost-minimizing objectives of the principal in her relationship with the
subcontractor. The slope of the isocost ray (that is q p) is the negative of
the marginal rate of substitution.

The isocost rays tell the principal which regions of the graph she prefers,
roughly, which way is up. But she is constrained by having to provide the
agent with incentives to implement the level of quality she desires. So the
q she wants to motivate Armin to provide must lie on his best-response
function. This equation—the incentive compatibility constraint—tells her
what is feasible. In Figure 10.8 she would prefer to be, for example, at point
c rather than at points d or n, because c is on a lower (steeper) isocost ray.
But point c is infeasible because it is above Armin’s best-response function.
The resulting cost-minimizing price will be somewhere on the incentive
compatibility constraint, but where?
It is simpler to find the price that maximizes q p than that which mini-
mizes costs (p q). The two are equivalent. So, putting her objectives together
with what is feasible, Patrisia wants to find the p that will:

q 2u
maximize , where q has to be such that q 1
p p

The price that accomplishes this is given by the following condition for cost
minimization (as shown in M-Note 10.5):

q
Condition for cost minimization: qp (10.14)
p
slope of isocost slope of agent’s BRF
Meaning mrs mrt

This condition is shown in Figure 10.8: the price is chosen so that the
slope of the agent’s best-response function qp is equal to the slope of the
isocost ray with the highest ratio of quality to cost, q p.
Equation 10.14, which requires equating the slopes of the best-response
function and an isocost line, is another example of the mrs mrt rule.

• The slope of an isocost ray is the (negative of the) marginal rate of

substitution between quality and price in the eyes of the principal, namely
the increase in quality that would just compensate for a unit increase in
the price.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

pN qN
qp
2u p2 q p 8u q 1 2u p

1 c2

q 1
Slope = =
p 8u

Best-response
function (ICC)
Nash
Quality, q

1 c equilibrium n
2 2u
Slope = qp =
p2

0
0 2u 4u 40
Price, p

• The slope of the best-response function is the (negative of the) marginal

rate of transformation of paying a higher price into receiving higher
quality.

The allocation pN qN is a Nash equilibrium because it is a mutual best

response:

• Patrisia’s choice: given Armin’s strategy choice—his best-response

function—offering the price (pN ) is the best Patrisia can do (we know
this because it is the result of the constrained maximum problem she
just solved); and
• Armin’s choice: given Patrisia’s price offer, providing (qN ) is the best Armin
can do (we know this because qN is a point on his best-response function).

p
qp p
pq
qp
p
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

d qp qp p q1
dp p p2
p
qp p q
0
p2
p2 qp p q 0
q
qp
p

2u
• qp
p2
q
•
p

2u q
mrt qp mrs
p2 p

q qp 1 p
q q p p

p p2
q 2u
p p2
p2 pq 2u
2u
q p
q
q
2u
q 1
p

2u
p
q
2u
p 2u
1
p
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

2u
p 1 2u
p
2u 2u
p 1
p p
4u
1
p
p pN 4u
pN

2u
q 1
p
2u
pN 4u q 1
4u
qN 05
N
q
1 1
T qN 2
1 qN 1 05

u 5 u 25

qN pN

The Nash equilibrium allocation pN qN is not Pareto efficient. We can see

this in Figure 10.9. Remember that the principal prefers allocations that are
higher and to the left (more quality for a lower price) and the agent prefers
the allocations to the right. So any point that is both above the isocost line
through the Nash equilibrium (better for the principal) and to the right of
the iso-value curve v4 (better for the agent), is a Pareto improvement over n.
In Figure 10.9 the shaded lens is the set of allocations that have these two
properties and therefore are Pareto superior to the Nash equilibrium. You
can see, for example, that point f is better for both Armin and Patrisia than
is the Nash equilibrium, n. So f is Pareto superior to n. It follows that n is
not Pareto efficient.
We know that points like f in the Pareto-improving lens must exist
because the rule that the principal implemented to minimize her cost
of quality–Equation 10.14—selected a price that equated the slope of her
steepest possible isocost ray with the slope of the agent’s best-response
function. In other words she followed the mrs mrt rule.
But a Pareto-efficient outcome requires a different rule: equating the
principal’s marginal rate of substitution (mrsP ) with the agent’s marginal
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

v5

q 1 2u p c2 q p 4u

1 c2

Better for
principal

Pareto improving lens f

Best-response
function (ICC)
Quality, q

1 Nash equilibrium
2 n

Better for agent

v4 v5
0
0 2u 4u 40
Price, p

rate of substitution (mrsA ). This would have required that the principal
vp
choose a price that would equate the slope of an isocost ray with the slope
dq dp
vq of the agent’s iso-value curve.
vq 0 You are already familiar with the reason why the Nash equilibrium is inef-
ficient. Like the fishermen overexploiting their resource and diminishing
each other’s catch, the subcontracting agent, in deciding on his action—the
level of quality to provide—is not taking account of the effect of his choice
on another person, in this case the buyer-principal. The reason this occurs
is that the contract is incomplete: it does not cover the quality he provides,
which is then another example of an external effect.
We will see in section 10.11 that if the principal could just purchase quality
(offering a particular p for some amount of q as if she were buying electricity
or red winter wheat #2), then the result will be Pareto efficient. If the con-
tract is complete, Patrisia minimizing the cost of quality by implementing
the mrs mrt rule also unwittingly implements the mrsP mrsA rule.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

Rent at Iso-v curves

point n

Best-response
function (ICC)
Quality, q

1 c
2 n
Participation
constraint
(PC)

v0 = 0 v4
0
0 u p0 = 2u p = 4u 40
Price, p

A key element in Patrisia’s strategy, remember, and the reason why it works
is that she pays him enough so that he wants to avoid being terminated.
His enforcement rent is shown in Figure 10.10 as the distance between n
and c. To see this, imagine Armin were at the Nash equilibrium, n, and we
asked: hypothetically holding constant the level of quality he provides, how
much less could he be paid and still be no worse off than at his fallback
option, namely zero? To answer the question, we imagine moving to the
left (lowering the price) from point n. We eventually hit the participation
constraint v0 0. At that point, c he receives no rent. So the difference in
price between n and c namely, 2u v4 v0 is his per-period rent.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

The rent that the agent receives each period is the reason why terminating
the exchange is an effective threat. If Patrisia cuts him off, Armin will not
immediately find another principal to transact with on terms as good as his
current transaction.
The key feature of these equilibrium contracts is that principals trans-
act with agents who receive economic enforcement rents and prefer the
current transaction to their next-best alternative. Because some agents
receive enforcement rents in ongoing contracts, there must exist some
other identical agents who are quantity-constrained, namely, the suppliers
who fail to make a sale. If this were not the case, then immediately upon
termination the agent could find another principal so the termination
would not impose a cost on the agent. It is the fact that the market does
not clear that makes the threat of termination effective.
In the Benetton model there are more agents looking for transactions
than there are transactions being offered by principals. This is why, if
Armin were terminated from his current transaction, it would take him time
and a costly search process to find another. (For simplicity, we made the
assumption that his next-best alternative is to get nothing, but as will see
in section 10.13, this is not an essential feature of the model.)
Patrisia, the employer and other principals in similar principal–agent
relationships, are on the short side of the market. Armin, the employee and
other agents are on the long side; the long side includes agents who would
like a transaction but cannot secure one. Armin fears being one of these:
unable to sell his product.
We use the term short-side power to describe the power that Patrisia is
able to exercise over Armin because she is on the short side of a market
that does not clear. In Chapter 9 we studied out-of-equilibrium markets in
which supply does not equal demand. In the Benetton model of principals
and agents, when contracts are incomplete the market does not clear even
when it is in equilibrium (this will be the case also in labor and credit
markets).
Essential to Patrisia’s ability to get Armin to do what she wanted is her
ability to threaten to impose a major economic cost on him. This cost—
called a sanction, or penalty—is part of our definition of power.
The definition can be applied to the Benetton principal–agent interac-
tion. In equilibrium the following conditions hold:

A B A
B A
B B A
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• Sanctions affect behavior. Armin provides a shirt of higher quality than he

would have in the absence of the Patrisia’s threatened termination.
• To the advantage of the person exercising power. Patrisia benefits from
this.
• The relationship is asymmetrical. Armin could not get Patrisia to act in a
way beneficial to him by threatening her with termination.

To see why Armin did not have power similar to Patrisia, imagine that he
threatened to end their relationship unless Patrisia raised the price above
pN . Patrisia would refuse to raise the price, knowing the following:

• because she is on the short side of the market, she could easily find
another supplier (remember some of them cannot find a buyer); and
• it would not be in Armin’s interest to carry out the threat because he
would then have to find another buyer, and as a long-sider, this might
not be possible.

If Armin were to threaten to sanction Patrisia should she not raise the
price (for example, to supply lower quality shirts), his threat would not be
credible. Namely, it would not be in the interest of the actor to carry out, if
the threat did not have its desired effect.
The definition of power emphasizes its use to advantage one person over
the other. But the exercise of power is also essential to making mutually
beneficial exchanges possible when contracts are incomplete.
Imagine, for example, that Patrisia were prevented from threatening to
replace Armin with another supplier. Then he would provide only low-
quality shirts, which she could not sell. As a result she would not purchase
shirts from Armin. They would then both be worse off.
The case which we have analyzed—the exercise of power by the buyer
(principal) over the producer (agent)—is just one example of power
relationships that are sustainable as the Nash equilibrium of a system
of voluntary competitive exchanges among private individuals. Other
examples include the power that employers wield over employees,
or lenders over borrowers, and the other hidden action examples in
Table 10.1.
The principal–agent model of hidden actions shows the following:

• that the exercise of power is essential to the principal’s strategy;

• that a non-clearing market is essential to the principal’s ability to exercise
power; and
• that the result of the principal’s actions—along with the other actors in
the market all acting independently—is a Nash equilibrium in which the
market does not clear, thereby providing conditions for them to exercise
power in their interactions with agents.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

The last point is important and a bit counterintuitive. In a competitive

environment, no principal, acting singly, can create markets that do not
clear. But we have shown that, without intending to do so and without
coordinating in any way, principals’ choices–setting prices so that agents
receive enforcement rents in order to minimize their costs—taken together
create markets that do not clear.
In introductory economics, the exercise of power is commonly associ-
ated with limited competition in a market, for example what is sometimes
called the “market power” of a price-setting monopolist. Markets that do
not clear are typically attributed to government policies (such as rent
control) or “market imperfections” such as “sticky prices” that do not
adjust. The above three bullets show that where contracts are incom-
plete, the exercise of power and markets that do not clear occur in the
absence of government policies and markets with unlimited competition—
approximating the perfectly competitive model.
None of this would be possible if markets cleared. To see this, we imagine
a case in which quality was subject to a complete contract in the Benetton
model.

Suppose that Patrisia has discovered some magical device that can deter-
mine exactly the quality offered by Armin, and that this information is
verifiable. So a complete contract is now possible. She can just name a price
and the exact amount of quality that she would like in return. If he does not
deliver the goods of the specified quality, he does not get paid.

Armin will provide that amount, as long as she pays a price that makes Armin
even just a bit better off than he would be without the contract. Of course
she would not require him to deliver anything close to q 1 because that
would be so costly to him to achieve that she would have to pay a very high
price in order for him to accept the contract.
The big difference that the complete contract makes is that Patrisia is
no longer constrained by Armin’s best-response function—the incentive
compatibility constraint—but instead by Armin’s participation constraint.
This requires that Armin not be worse off agreeing to Patrisia’s proposal
than he would be were he to walk away, that is, to receive instead his
fallback, which as before we assume to be zero.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

She no longer has to provide him with a rent—a utility greater than zero,
his fallback option—along with a threat to terminate the contract if the
goods he has supplied are returned by disgruntled consumers. She can
simply refuse to pay for the goods when he delivers them if they are of
less quality than she has specified.
Her interaction with the agent is no longer repeated, it is a one-shot
game. And it is no longer a principal–agent interaction: the conflict of
interest over quality remains, but the second characteristic of a principal–
agent relationship is now missing: quality is now something that can be
enforced by contract.

Here is the new complete contracting game:

• Patrisia is first mover and she makes a take-it-or-leave-it offer to pay

Armin pC to purchase his product of quality qC .
p q
• Armin either accepts the offer and the exchange is carried out or rejects.
• In either case, this ends the game. p

We use the C superscript to indicate the hypothetical complete contracting

Nash equilibrium.
Because, due to the complete contract, she can just purchase quality, we
can now interpret Armin’s participation constraint as the minimum price
at which he is willing to sell his products with the quality indicated by up uq u0
the height of the curve. So in Figure 10.11 he would be willing to sell his
products with quality of q 0 5 if the price were 2u. And the same goes for
all of the price–quality combinations that make up the willingness to sell
(participation constraint) curve.
Just as in the case where contracts are incomplete, her maximum will be
where the ray from the origin with slope q p is tangent to the constraint.
But this is now the participation constraint (Armin’s iso-value curve v0 0)
not the incentive compatibility constraint (Armin’s best-response function).
In Figure 10.11 (b) you can see that the outcome of the game is point c, with
qC 0 5 and pC 2u. This is the principal’s offer that equates the slope of
her isocost line with his iso-value curve, implementing the mrs mrt rule.
We can confirm that this is a Nash equilibrium: q p

• Given Armin’s utility function and fallback option (described by his par-
ticipation constraint) Patrisia is doing the best she can by offering the
contract: qC 0 5 and pC 2u.
• Given her offer and his fallback position, Armin is doing the best he can
by accepting the contract.

Under the complete contract, Patrisia has paid half as much but received
the same quality as when the contract was incomplete; so her miracle
machine for detecting quality has cut her costs in half. Were Patrisia to offer
the complete contracting price pC 2u under an incomplete contract, you
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

pC qC C
u u0 0
pN qN
q pN
N
qN
N
p
2u p2 pN qN
c1 q p 4u c2
q p 8u q 1 up q 1 2u p

1 c1 c2 1 c1 c2
Quality, q

Quality, q

1 c Participation constraint c Participation

qC = (willingness to sell) qN= qC = 1 n constraint
2 2
(willingness to
sell)
Best-response
function
(ICC)
v0=u0
u0=0 =0
0 0
0 u pC = 2u 40 0 u pC =2u pN=4u 40
Price, p Price, p
(a) The complete contracting game (b) Both complete and incomplete contracting

can see from Figure 10.11 that Armin would provide zero quality (his best-
response function q p q 2u 0).
Three characteristics of the Nash equilibrium with complete contract
are:

• Pareto efficiency: Patrisia offered Armin a contract that implemented

the mrs mrt rule, she also (without intending to do so) implemented
the mrsP mrsA rule guaranteeing the Pareto efficiency of the Nash
equilibrium. (The “Reminder” on the previous page explains why the two
rules coincide.) It is not possible—by choosing some allocation other than
qC 0 5 and pC 2u—for Patrisia to be better off without Armin being
worse off. We know from Chapters 4, 5, and 9 that when one actor
optimizes subject to a participation constraint (rather than an incentive
compatibility constraint) of another actor, the outcome is Pareto effi-
cient. This follows from the definition of a Pareto-efficient outcome: if
Patrisia has indeed minimized her costs subject to the requirement that
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

Armin have utility of at least zero, then it must be that any other allocation
would make at least one of them worse off.
• No rent for the agent: The supplier’s utility under the contract is identical
to his next-best alternative. Remember we used the distance between
point n and point c to calculate the rent that Armin received under
the incomplete contract, so you can see that completing the contract
transferred the rent from the agent to the principal.
• The market clears: The fact that the agent receives no rent means—by the
definition of a rent—that he is no better off with the transaction than he
would be at his fallback option, that is, if it were terminated. But this in
turn means that his fallback option—what he gets if terminated—must be
to immediately secure the same deal—providing qC for ihe price pC from
some other principal. And for this to be true, it must be that there are no
other agents just like Armin who are unable to sell their goods, because
if there were, then he would be among them looking for a buyer, that
is, worse off than he was with Patrisia. This is why the absence of rents
indicates that markets clear.

u
upq p 0
1 q
u
p
1 q

qp
p

q q 1 q q
p u u
1 q

q
q
d
p 1
1 q 1 q 1
dq u
1 2q
u
qc
pc
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

1 2q
0
u
1 2q
1
qc
2
pc 2u

u 0

In order to implement point c Patrisia did not need to specify the quality
that she required Armin to deliver. She did not have to have take-it-or-
leave-it power. To see this we modify the complete contracting game.

• Patrisia as before is first mover and can commit to some price at which
she will pay for Armin’s product depending on its quality: pC pq She is
simply buying whatever quality he provides at the price given.
• Armin responds to her price either by rejecting the contract (if the price
is too low) or by delivering a good of a quality level of his choosing.
• Patrisia measures the quality of the good and pays pC pq
• This ends the game.

Because the contract is complete, Patrisia knows that she will get exactly
what she pays for, nothing more and (more important to her) nothing
less. So she realizes that she did not need to have take-it-or-leave-it
power—dictating the quality as well as the price—in order to transfer the
entire rent from the transaction to herself. He could have offered a different
contract simply specifying a price: she would purchase any q that Armin
provided at a price p 4qu. This means that Armin could pick not only point
c, but also any other point on the isocost line c1 through point c. This line is
now Armin’s constraint, and he would like to be on the highest indifference
curve as possible.
Which point would he pick? Recalling that points to the right and below
are better for Armin (higher price and lower quality), point c is the point he
would pick (all of the other points in c1 are worse than c, in fact they give
him negative utility).
So he would deliver his goods to Patrisia, she would measure their quality,
finding q to be one-half, and pay him the price 2u that is p 4qu, as
promised.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

Compared to a complete contract, the incomplete contracting contin-

gent renewal Nash equilibrium has six important characteristics, shown
in Table 10.3. These characteristics are general features of contingent
renewal incomplete contracts, and do not depend on the utility function, qN 05 qC
the fallback option or any of the other special assumptions we have made
in the Benetton model:

Equilibrium rents: Under the incomplete contract, the agent receives a

rent above his next-best alternative. In Table 10.3 you see that he has
a utility of 2u for each period and that because he provides qN 0 5 the
transaction is expected to last for the inverse of this, that is, two periods,
giving him a rent of 4u.
The principal is a price maker: The reason why the principal does not
treat the price as given is that she can benefit from changing the
price, given the contractual incompleteness concerning the quality
of the good. Price-making in the incomplete contracting context does
not derive from any noncompetitive aspect of the assumed market
structure, such as monopoly power. The Benetton model can include
many principals and many agents and there are no barriers to entry,
but contractual incompleteness means that principals can benefit by
setting prices.
Competitive equilibrium without market clearing: Because the agent
receives a rent, we know that unlike the complete contracting case,
his next-best alternative cannot be to just walk across the street and
contract with some other principal on identical terms. Instead, he will
have to search for a new partner, during which time he will be without a
transaction. But for this to be the case there must be other agents also
searching for buyers: if he were the only buyer without a contract, then
given the ordinary turnover from jobs (quits, deaths and other reasons
for leaving a job) he would find a similar job without delay. So the fact
that agents on the supply side of the market who are transacting with
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

principals are getting a rent is diagnostic about the state of the market:
there must be excess supply. It is not the case that there is excess supply
because agents are receiving rents. We will turn to the question—why
does the market not clear when contracts are incomplete—in Chapter 11.
Durable Transactions: The principal (buyer) and agent (supplier) will
interact over many periods, even though there are many identical
buyers and suppliers. Competitive equilibrium with contingent renewal
incomplete contracting will be characterized by a series of durable
bilateral trading islands rather than a sea of anonymous traders engaged
in one-shot interactions in spot markets.
Endogenous claim enforcement through the exercise of short-side power:
The buyer (principal) in the contingent renewal contract minimizes
costs by threatening to terminate the ongoing relationship with the
seller (agent). Because of this threatened sanction, the agent acts in the
principal’s interests by providing a product of higher quality.
Pareto-inefficient equilibrium: Because the buyer maximized taking
the supplier’s best-response function as the (incentive compatibility)
constraint, rather than the supplier’s participation constraint, and
because when contracts are incomplete the two constraints differ, the
noncontractual equilibrium will not be Pareto efficient.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

In the model discussed so far, the agent could choose any level of quality.
By restricting the agent to two feasible quality levels and reducing the
game to a one-shot game, we get a simpler model that focuses on the
most important aspects of the problem and that we will find useful in other
contexts, such as the employment contract in Chapter 11.

In this stripped-down version of the scenario, Armin the agent and supplier
of the variable-quality shirt may offer either low or high quality, at a
disutility cost u and u respectively where u u. In order to protect the
reputation of the brand, if the quality is low, then Patrisia will not put her
brand’s label on the product (e.g. Benetton) and she will not be able to sell
it. Patrisia (the buyer) never mistakenly thinks that a high-quality shirt is
low-quality. But if Armin has provided her with a low-quality shirt she will
detect this with probability t and Armin knows this.
Here is the game.

• Principal offers a price: Patrisia is first mover and offers a price p for
Armin’s product and announces that she will not pay if she detects that
the good is of low quality.
• Agent decides on the quality: Given the price offered, Armin decides to
produce high or low quality and delivers the good to Patrisia.
• Agent may be terminated: If he has delivered a low-quality good, then
with probability t Patrisia detects this and refuses to pay, in which case
Armin must sell the good to another buyer at a lower price.
• Principal pays for the good: If the good is of high quality, or if it is of low
quality but not detected by Patrisia, she pays for the good.

To ensure that Armin will provide high quality, Patrisia must offer a price
high enough so that his expected income from offering high quality, p u,
is not less than his expected income when he offers low quality. Patrisia has
to make it a best response for Armin to provide high quality.
The game with just two levels of quality is similar to the games in
Chapter 1 where the players had just two choices, for example fishing ten
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

or 12 hours. So here the incentive compatibility constraint (ICC) that limits

Patrisia’s optimization problem is not a range of prices p that she must offer
if she wants quality q. It is a single price that will make it in Armin’s interest
to choose high over low quality.
The interaction along with the agent’s payoffs are shown in the game tree
in Figure 10.12. The right-hand side branch gives the result for when the
agent provides high quality: he bears the cost of producing high quality u
and is paid the price p so his utility is p u.
The left-hand branch shows how the game proceeds if he produces low
quality. There are two outcomes that might occur if he chooses not to
produce high quality:

• Agent’s contract is terminated: With probability t Patrisia detects the low

quality, refuses to pay Armin, and he gets his fallback price pz selling the
good to some alternative buyer and so has utility pz u.

Agent

Low quality High quality

(disutility = u) (disutility = u)

Chance

Terminated Not terminated

(t) (1 t)

pz u p u p u
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

• Agent’s contract is not terminated: With probability (1 t) Patrisia does not

detect the low quality, so she pays him p and his utility is p u.

His expected income is the income he receives in these two events,

multiplied by the probability of each of them occurring. This is the right-
hand side of Equation 10.21.
Not terminated Terminated

ICC: p u 1 t p u t pz u (10.21)

To find the lowest price Patrisia can offer that will induce Armin to
provide high quality, we rearrange Equation 10.21 to isolate p (as shown in
M-Note 10.8) and find the Nash equilibrium price, pN :

u u
Nash equilibrium price: pN pz (10.22)
t
Patrisia will set the price pN and Armin will provide high quality.
This is a Nash equilibrium because at the price pN Armin would not
do better by providing low quality, and given the incentive compatibility
constraint based on what she knows about Armin’s behavior, Patrisia cannot
do better than to offer pN . If she offered a higher price than pN , she would
be throwing away money. If she paid less than pN , then he would produce
low quality.
From Equation 10.22 for the equilibrium price, the lowest price that
Patrisia can offer compatible with Armin supplying high quality will be
higher:

• the greater is the difference in cost to the agent (u u) to provide high

rather than low quality;
• the greater is the agent’s fallback price, pz ; and
• lower: the more difficult it is for the principal to detect low quality, that
is, the lower is t.

p u 1 t p u t pz u

p u 1 t p u t pz u
z
p u tp p p u
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

u p pN
u u t pz p
u u
pz p
t
u u
pN pz
t

pN u
pz u
pN u pz u
u u
pN pz u pz u
t
u u
u u
t
u u tu u
t t
1 t
u u
t

1 t
u u
t

These results replicate some of the economics you have already learned
about exchange with incomplete contracts. But in one respect this model
goes further: it provides us with a measure of how incomplete a contract is,
and to see why this matters. It gives us a measure of how asymmetric the
relevant information is.
Armin knows if he has produced a low-quality good; but Patrisia will
discover this only with probability t. So if we denote the quality of Armin’s
knowledge as 1 (100 percent), then the difference between this and the
quality of Patrisia’s knowledge (t) or 1 t is an indicator of the extent of
information asymmetry. The accuracy of her information compared to his—
t itself—is a measure of how complete the contract is.
We now can see the effect of contractual incompleteness on the dis-
tribution of income between the principal and the agent. To do this, let’s
assume that Patrisia can sell a high-quality shirt for some given price pB (for
Benetton). Then her maximum willingness to pay Armin for a good shirt is
pB . If she were to sell at that price, her profits would be zero. We do not need
to consider the case in which she ends up with a low-quality good and has
to (attempt to) sell that. If she pays Armin pN he will not deliver low-quality
goods.
Armin’s minimum willingness to sell is the price at which he could have
sold a low-quality good, plus compensation for the extra cost of producing
high quality, or u u pz . The difference between her willingness to pay
B
p and his willingness to sell u u pz is the total rent to be gained from
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

pB u u pz pB

u u z
pN(t) = +p
t

pB = 1 P's Willingness
to pay

Principal's profit

Total
Price, p

economic
surplus

Agent's rent

u + (pz u) A's Willingness

to sell

0 t 1
Degree of contractual completeness, t

exchange of one unit. This is the quantity that will be divided up between
them in the form of her profits and his rent. How it is divided up will depend
on how complete the contract is.
Figure 10.13 plots Equation 10.22, showing that the Nash equilibrium
price differs for different degrees of contractual completeness. It illustrates
how the distribution of the gains from exchange depend on the extent
of information asymmetry and contractual incompleteness. The vertical
distance between the agent’s willingness to sell and principal’s willingness
to pay is the “pie” that is to be divided up into the two “slices”: the principal’s
profit and the agent’s rent.
Equation 10.22 shows that if the contract were complete (t 1 so that
low quality could be detected with certainty), Patrisia need do nothing
more than to meet Armin’s participation constraint paying his minimum
willingness to sell price:

Complete contract price pC u u pz (10.28)

OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

Even at this low price he would provide high quality because in this case
his best response to that price is to produce high quality. To see this, think
about Armin’s options. He could produce high quality and get pN u for
sure, or produce low quality and get pz u for sure.
If the contract is incomplete, meaning t 1, however, Figure 10.13 shows
that Patricia will offer Armin a price greater than his willingness to sell.
You can also see that the size of the resulting rent is greater, the more
incomplete the contract is (going from right to left in the figure).
As expected, the more complete the contract, the larger is the share
of the principal’s profits. Do not conclude, however, that from the agent’s
standpoint the more asymmetric the information the better. There is some
level of information asymmetry (t in the figure) beyond which the best the
principal can do would be to pay the agent more than the price at which she
can sell the trademarked good to a consumer. In this case there is no way
for the transaction to satisfy the principal’s participation constraint. So she
will stop purchasing shirts from suppliers, trademarking them, and selling
them to consumers. With no exchange taking place there are no gains from
trade to be shared.
We can use this simple model of the quality problem to better understand
a fast-growing kind of work in many countries: the gig economy.

pN t
pZ
pZ
t pB
t t

A ‘gig’ for a musician or comedian is a single appearance for which they will
be paid not by the hour, but an agreed sum for the performance. The gig
economy is not about jokes and tunes, however, it refers to the combined
activities of Uber or Lyft drivers, TaskRabbits, UpWorkers, Mechanical
Turkers, and others who transport people and goods, home-assemble
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

online purchased furniture, and perform other well-defined tasks for which
they are paid a fixed rate.
In many legal jurisdictions, gig workers are considered private contrac-
tors and not employees. They provide their own cars or tools and gain
access to their gigs by means of a two-sided platform that connects those
who will pay for the gig, and those who perform it. The company—Uber,
for example—sets the prices and determines the number of people who are
allowed to use their app.
The gig economy is a small portion (in the US not more than 2 percent of
employment) even of those high-income economies where, for example,
ride services like Uber and Lyft have made significant inroads against
conventional taxi firms. The gig economy is growing because modern
information technology makes it easier both:

• to match buyers and sellers—drivers and those needing a ride, or those

needing a task done in their home—and others with the time and skill to
do the job; and
• to define tasks sufficiently precisely—the exact time taken for a delivery
for example—that gig workers can be paid by the task and not by the hour.

The second bullet means that app-based ride–hail, delivery, and other
parts of the gig economy provide an illuminating contrast with the model of
hidden actions with variable quality studied in this unit. The key difference
is that in some cases the tasks performed are sufficiently well defined and
easily measured so that a virtually complete contract is possible: if the
person is not delivered from the hotel to the airport, the Lyft driver does not
get paid; if the Ikea shelves purchased online are not assembled properly,
the TaskRabbit does not make a penny.
This is equivalent to the simple model of the quality problem in which the
principal is the person who has engaged the gig worker (the agent) to do a
job. We can illustrate what this means with the simple ‘two levels of quality’
model just introduced. Think about some task, for example, assembling a
bicycle that a person has just purchased in a kit. A Rabbit tasker might take
on this job for an agreed-upon fee. If the task is not performed—the bicycle
does not work properly when assembled, for example—this information will
be available to the purchaser and the tasker will not be paid.
In terms of the model, the probability that low quality (a nonfunctioning
bike) will be detected, t, is much closer to 1 than in conventional jobs. Figure
10.13 shows that if t 1, then the agent—that is the gig worker—is paid her
minimum willingness to sell and the principal—so that the owner of the
platform—TaskRabbit or Uber—receives the entire economic rent.
A result is that gig performers in this economy face extraordinary eco-
nomic insecurity: they are not guaranteed a fixed schedule of hours and
pay, nor do they receive health insurance benefits, maternity leave, holiday
pay, or pension contributions through their employer. Working full-time,
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

the hourly earnings of the vast majority of app-based drivers for ride–hail
services in New York City would place them below the official poverty level
for New York.
The reason, also clear from the model and Figure 10.13, is that gig
companies do not need to pay substantially more than the gig worker’s
next-best alternative. Beyond compensating the tasker for her time and
trouble, they do not need to motivate the tasker to do the job as specified:
if it is not done, the tasker will not be paid. A result is that the gig economy
can often produce services at a lower cost and price than are available
from conventional firms that, as we will see in the next chapter, pay their
employees substantially more than their next-best alternative.
The structure of the gig economy not only reduces the pay that workers
receive, it also depresses their fallback option. An important feature of the
gig economy is that the only way that drivers or taskers can get gigs is
through the platforms owned by a few firms such as Uber, Lyft, TaskRabbit,
Mechanical Turk, and others. This is equivalent to Patrisia being the only
buyer or just one of few buyers to whom Armin could sell his goods. This
means that those performing the gigs have no real bargaining power. If a
Rabbit tasker objects to the terms, there will always be another tasker to
take her place, but few if any other ways that the disgruntled tasker could
find a gig.
The computer platform that allows those who need a gig performed to
connect to those performing the gigs makes possible substantial mutual
benefits in putting together gig workers who have free time and the skills,
a vehicle, or other equipment required with those willing to pay for a
completed gig. But the distribution of these gains among those who hire
the gig worker, the platform owners, and gig workers is highly unequal.
The gig work illustrates an important truth about the modern economy:
the nature of our social interactions at work—whether they be personal or
anonymous, long-lasting or ephemeral, for example—is strongly affected
by whether the contract governing our transactions is complete or not.
You have already seen a hint of this: the complete contracting game was
a one shot, while incomplete contracts were modeled by a repeated game.
Experiments confirm that the nature of the contracts influences the kinds
of social interactions and social norms that are part of the exchange
process.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

When contracts are complete, you get what you pay for. So, for a given
price, there is little economic reason to be concerned about one’s exchange
partner’s psychological makeup or moral commitments. If you do not get
what you paid for, you get your money back at no cost to yourself. This
is what a complete contract means: its terms are enforced—if necessary—
by the courts, not by the parties to the exchange. And this is also why we
care about who we interact with a lot more in cases where contracts are
incomplete.
To see this, put all of the people with whom you have any economic
interactions into two groups: those whose names you know and those
whose names you do not know. You probably do know the names of your
employer, your doctor, the person you consult with for legal advice, and
perhaps your car mechanic. These are all exchanges in which the contract
is substantially incomplete. Do you also know the name of the gas station
attendant or the checker at the supermarket?
When contracts are incomplete, parties to an exchange—whether buyers
or sellers—will favor social interactions where exchange is personal and
durable. Exchange is personal, rather than anonymous, when the parties to
exchange have personal knowledge of each other, such as personal histories
and knowledge of whether the other party is trustworthy or untrustworthy.
Exchange is durable, rather than one shot, when it results in long-term
repeated interactions between the parties to exchange, such as when you
regularly go to a hair stylist you like, or you trust a car mechanic or a
babysitter you’ve known for a long time.
The relationship between contractual incompleteness and market struc-
ture can be seen in the contrasting structures of the rice and raw rubber
trade in Thailand. Buying and selling in the wholesale rice market—where
the quality of the product is easily determined by the buyer—buyers and
sellers hardly know one another. This contrasts with the personalized
exchange based on trust in the raw rubber market. In the raw rubber market
quality is impossible to determine at the moment of purchase. As a result,
buyers purchased rubber repeatedly from the same sellers rather than
shopping around, a strategy that gave them the kind of short-side power
that Patrisia has over Armin as a way of controlling the quality.
Similarly, in villages like Palanpur (in India), wheat and rice as well as
seeds and fertilizer are standardized, easily measured commodities and are
subject to relatively complete contracting. These inputs are bought and
sold in region-wide markets in which transactions are governed by little
more than the going price and the budget constraints of the participants.
The markets are impersonal and anonymous.
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

By contrast, exchanges concerning labor, credit, the use of land, and the
services of farm assets such as bullocks take place almost entirely within
the village, and often within the same caste. Moneylenders in villages rarely
extend loans to people they don’t know or who don’t live in the same
village. The village markets in goods or services with incomplete contracts
are personalized.

A number of experiments also show differences in behavior depending on

the possibility of complete contracting.
Economists have investigated a variety of experimental markets to
understand the decisions people make when buying and selling goods.12
Experimental researchers can change the “institutions” under which
participants interact. Remember, institutions are rules of the game, so
in an experimental game, changing institutions just means changing the
rules. For example, one experiment could have the institutional structure
of complete contracts, and another incomplete contracts.
Economists Martin Brown, Armin Falk, and Ernst Fehr designed a mar-
ket experiment to explore the effects of contractual incompleteness on
patterns of trading. The good exchanged varied in quality, with higher
quality more costly to provide. In the complete contracting condition, the
experimenter enforced the level of quality promised by the supplier, while
in the incomplete contracting condition the supplier could provide any level
of quality (irrespective of any promise or agreement with the buyer).13
Buyers and sellers knew the identification numbers of those they were
interacting with, so they could use information they had acquired in previ-
ous rounds as a guide to whom they would like to have as trading partners,
and the prices and quality to offer. Buyers had the opportunity to make a
private offer (rather than broadcasting a public offer) to the same seller in
the next period, therefore attempting to initiate an ongoing relationship
with the seller.
Very different patterns of trading emerged under the complete and
incomplete contracting conditions. In the complete contract condition,
90 percent of the trading relationships lasted less than three periods (and
most of them were one shot). By contrast, under the incomplete contracting
condition only 40 percent of the relationships were fewer than three
periods, and most traders formed trusting relationships with their partners.
Buyers in the incomplete contracting condition offered prices consid-
erably higher than the cost of providing quality (just as in the principal-
agent shirt quality model). When buyers were disappointed by the qual-
ity supplied, they terminated the relationship, withdrawing the implied
enforcement rent from the supplier. Other differences are summarized in
Table 10.4. The behavioral differences in complete and incomplete con-
tracting treatments were particularly pronounced in later rounds of the
game, suggesting that the subjects updated their behaviors according to
experience.
 OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

These experimental results suggest that there may be a two-way rela-

tionship between trust, reciprocity, and other social preferences, on the
one hand, and the degree of contractual completeness on the other.

• Long-standing relationships: Where contracts are incomplete—as in the

above experiment—economic interactions may endure over long periods
during which people develop trusting and reciprocal relationships; while
this is unlikely to be the case where contracts are complete.
• Social preferences and contracts are substitutes: Where people are trusting
and reciprocal, making the contract “as complete as possible” may not be
worth the legal costs and possible offense to one’s trading partners. But if
people are entirely self-interested, trying to complete the contract may
be the only way to do business.

The most important organizations governing exchanges in modern

economies are firms, whose managers in order to produce and market
goods and services combine other people’s labor and (what Adam Smith
called) “other people’s money,” neither of which are subject to complete
contracting. Labor and credit markets are typical of the many important
exchanges in which what is transacted are not well-defined and easily
measured objects, like the nuts and apples in Ronald Coase’s example in
the head quote for this chapter. In these markets the transaction involves
something quite different and much more difficult to enforce (the promise
to repay the loan, for example) or to measure (e.g. the promise to work
hard on the job). Coase put it this way: “what are traded on the market are
not, as is often supposed by economists, physical entities, but the rights
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

to perform certain actions . . . the objects of exchange are complex bundles

of obligations and claims concerning who should do what under what
conditions.”15
In the next two chapters we use the principal–agent model you have just
learned to study how the owners and managers of firms—as both employers
of workers and borrowers from banks and other lenders—structure the
rights to perform actions concerning other people’s labor and other peo-
ple’s money, respectively.

mrt mrs
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

mrs mrt
P
mrs mrsA
OUP UNCORRECTED PROOF – REVISES, 9/9/2021, SPi

q
p
pB
u
u
u
u
vpq

t
tq
T
1
t q