Principal and Agent

JOSEPH E. STIGLITZ

The principal-agent literature is concerned with how one individual, the principal
(sayan employer), can design a compensation system (a contract) which motivates
another individual, his agent (say the employee), to act in the principal's interests.
The term principal-agent problem is due to Ross (1973). Other early contributions
to this literature include Mirrlees (1974, 1976) and Stiglitz (1974, 1975).
A principal-agent problem arises when there is imperfect information, either
concerning what action the agent has undertaken or what he should undertake.
In many situations, the actions of an individual are not easily observable. It
would be very difficult for a landlord to monitor perfectly the weeding activity
of his tenant. A bank cannot monitor perfectly the actions of those to whom it
lends money. The employer cannot travel on the road with his salesman, to
monitor precisely the effort he puts into his salesmanship. In each of these
situations, the agent's (tenant's, borrower's, employee's) action affects the
principal (landlord, lender, employer). Clearly, if an individual's actions are
unobservable, then compensation cannot be based on those actions. In some cases,
even if an individual's actions are not directly observable, it may be possible to
infer his actions. Thus, if output were a function just of effort [Q = F(e)] then
even if effort were unobservable, if output were observable, and the relationship
between output and effort were known, then effort could be inferred with perfect
accuracy.
The principal-agent literature focuses on situations where an individual's
actions can neither be observed nor be perfectly inferred on the basis of observable
variables; thus, for instance, it is usually assumed that output is a function of
effort and an unobservable random variable, 0: Q = F(e, 0).
Moreover, in many circumstances, the principal wishes the agent to take actions
based on information which is available to the agent, not the principal. Indeed,
this is the very reason that individuals delegate responsibility. Because of the
asymmetry of information, the principal does not know whether the agent

241
Allocation, Information and Markets

undertook the action the principal would himself have undertaken, in the given
circumstances. Hence, even if the principal can observe the action, he may not
know whether that action was appropriate.
Since, in general, the pay-offs to the agent will differ from those to the principal,
the agent will not in general take the action which the principal would like him
to take, or that they would contract for in the presence of perfect information.
For instance, the employee may not adjust his effort as the situation requires, or
he may engage in too much or too little risk taking.
The principal-agent problem is, then, the central problem of economic
incentives.
In spite of the importance attached to economic incentives, until recently
economic theory had little to say on the matter. In the standard theory, individuals
were paid for performing a particular task. If they performed the task, they
received their compensation; if they failed to perform the task, they did not.
Individuals thus always had an incentive to perform the contracted-for service.
Only if the employer were so foolish as to pay the worker whether he performed
the task or not would an incentive problem arise.
The standard theory was based on the assumption that what action the
'principal' wished his agent to perform was perfectly known, and that the action
could be perfectly and costlessly monitored. Neither assumption is plausible and,
indeed, relatively few workers are paid solely on the basis of their observed inputs.

ORIGINS OF PRINCIPAL-AGENT PROBLEMS. Principal-agent problems arise whenever

one individual's actions have an effect on another individual. The question arises,
then, why cannot economic relationships be designed to avoid this kind of
dependency? Under what circumstances do these interdependencies arise? For
instance, if a landlord were to sell or rent his land to his tenant, then the workers'
effort would have no effect on him. If an employer were to sell or rent his
capital to his worker, then the workers' effort would again have no effect on
him. Traditional neoclassical analysis emphasized the symmetry in economic
relationships: one could describe the employer-employee relationship as the
employee hiring capital just as well as one could describe it as the employer
hiring labour. (This Wicksellian description of economic relationships always
seemed peculiar to me; it seemed to suggest the absence within neoclassical
analysis of certain important aspects of economic relationships; it is those aspects
which are the subject of scrutiny here.)
There are three important reasons for the existence of principal-agent
problems. Two have to do with the essential intertemporal nature of certain
relationships: insurance and credit. When two individuals enter into an insurance
contract, one individual (a) promises to pay the other (b) a certain amount if
event A occurs, while the other (b) promises to pay (a) a certain amount if event
B occurs. If there are actions which one of the individuals can undertake between
the date of the contract and the event which will affect the outcome, then there
is a principal-agent relationship between the two. This particular form of the
principal-agent problem is referred to within the insurance literature as the moral

242
 Principal and Agent

hazard problem (see Arrow, 1965), and, by extension, the term has been applied
to the principal-agent problem more generally.
Similarly, in credit relationships, one individual gives another some resource
(money), in return for a promise to repay that money at some later date. So long
as there is some probability of default, which can be affected by the actions of
the borrower, there is a moral hazard or principal-agent problem (provided that
that action cannot be perfectly monitored by the lender).
Many economic relationships have an important element of insurance within
them. The landlord-tenant sharecropping relationship can be viewed as if the
tenant pays a fixed rent, and then receives an insurance policy from the landlord,
in which the landlord agrees to pay the tenant a certain amount if output is low
(equal to the difference between his share and the fixed rent); and the tenant
agrees to pay a premium equal again to the difference between the share and the
fixed rent, when output is high.
Indeed, the credit 'problem' can be viewed as a special form of an insurance
relationship: the lender provides an insurance policy, such that if the borrower's
resources are less than the amount owed, the lender agrees to pay the borrower
the difference (which the borrower then immediately repays to the lender). The
premium is the difference between the rate of interest on a perfectly safe loan
and the rate of interest charged on this risky loan.
Insurance (spreading and transferring risk) provides one of the explanations
of sharecropping; were workers to rent the land, they would have to absorb all
the risk associated with output variations. With sharecropping, the risk is shared
between the landlord and the tenant. Since the wealth of tenants is usually much
less than that of landlords, there is some presumption that the landlords are
better able to absorb this risk.
But even if the tenants were risk neutral, there might be a principal-agent
problem. We suggested above that if the landlord were to rent his land to the
tenant, there would be no principal-agent problem. But this is not quite correct.
If the tenant did not have sufficient resources to pay the rent before production,
then the landlord would have to lend the tenant the money. (If he receives the
rent at the end of the period, then it is as if he is lending the individual the
money.) And then, if there are actions which the individual can undertake which
affect the likelihood of not being able to repay the debt (pay the rent), then there
is a moral hazard problem.
There is a second reason why renting land might not solve the moral hazard
problem. There may be actions which the tenant can take which affect the quality
of the land. To the extent that those actions are monitorable, the rental agreement
may specify the actions to be undertaken (e.g. concerning what crops are to be
grown, or grazing patterns). But these actions are not perfectly monitorable, and
thus, even with rental agreements there are important principal-agent problems.
(The same issues arise, of course, with the rental of any durable good.)
Again, one should ask, cannot these principal-agent problems be alleviated,
e.g. by selling the asset. But this entails precisely the two problems we identified
before as giving rise to principal-agent relationships: The agent (tenant,

243
Allocation, Information and Markets

employee) may not have sufficient capital (and thus must borrow to make the
purchase, creating a credit principal-agent problem); and if there is any risk
associated with the future value of the land, it imposes a risk on the agent. Any
attempt to alleviate those risks (through insurance) again gives rise to a moral
hazard problem.
The third major source of principal-agent relationships is rather different. It
arises from the attempt of the principal to extract as much rent (surplus) from
the agent as possible. The employer does not know how difficult the task is that
he would like the worker to perform. He could pay the worker the full value of
his output, but that would leave him no profits. He might pay much less, but
that might result in the worker refusing to work, ifthe task is in fact quite difficult;
and thus he would lose profits that he might otherwise obtain. This rent extraction
problem has been particularly well studied in the context of public utilities: the
government does not know the minimum amount of compensation required to
keep the utility producing. The rent extraction problem may be alleviated within
competitive environments by holding auctions: the individual for whom the asset
(franchise) is most valuable will bid the most. But there may not be enough
bidders to extract all the rents through an auction mechanism; and at least in
the case of utilities, the government may care not only about the rents received,
but also about the actions undertaken by the franchisee. (In some cases, the rent
extraction problem and the insurance problem are closely related: the average
value of rents received may be increased if rents can be varied with the weather,
the state of nature; again, we can think of decomposing the rent payment into
a fixed rent and an insurance payment.)
This list of reasons for the origins of principal-agent relations is not meant
to be exhaustive: yet many of the other reasons cited may be reduced to one of
these explanations. For instance, consider the problem of a production line on
which there are many workers; the output of the production line depends on all
of their efforts. In the absence of risk aversion and credit problems, the incentive
problem could be solved by giving each worker the total value of net output.
He would purchase the right to the job by paying a fixed fee. With such a
compensation scheme, the worker would have full incentives for maximizing the
firm's output. But such a compensation scheme imposes on the worker an
intolerable level of risk; and the fixed fee he would be required to pay necessitates
his borrowing large amounts of money.

THE BASIC PRINCIPAL-AGENT PROBLEM. In the standard principal-agent problem,

one looks for that contract (compensation scheme) which maximizes the expected
utility ofthe principal, given that (a) the agent will undertake the action(s) which
maximizes his expected utility, given the compensation scheme; and (b) given
that he must be willing to accept the contract.
The second set of constraints (which are nothing more than the standard
reservation utility constraints) are sometimes referred to as the individual
rationality constraints.
There are two standard mathematical formulations. One is a direct generalization

244
 Principal and Agent

of the insurance-moral hazard problem. There are a set of observable events,

such as whether an accident occurs. The probability that an event i occurs is a
function of the actions undertaken (effort at accident avoidance):
Pi = Pi(e),
where e may be a vector. The wealth of the individual in state i, in the absence
of insurance, is Wi' and with insurance it is Yi' Thus
hi=Yi-Wi
is the net payment from (to) the insurance company (the principal) in state i.
The expected utility of the insured (the agent) is then just
U= L Ui(Yi,e)Pi(e)
i

while that of the principal is

V= L V;(h;)Pi(e).
i

{h;} is chosen to maximize V subject to U ~ G.

Notice that the employer-employee relationship may be cast in this form: the
observable events are the levels of output. Assume for simplicity, that we measure
outputs in round numbers (say, bushels of wheat). Then state i refers to the
number of bushels produced. Pi then is the probability that i bushels will be
produced. Assume that the individual's wealth, apart from this contractual
arrangement with his employer, is zero. Then Yi is the individual's pay if output
is i. If the employer is risk neutral,
V;(h;) = qi - hi = qi - Yi'
where q is the price of output (of a bushel of wheat), assumed to be independent of i.
Although the employer-employee relationship can be cast in this form, it is
more naturally represented by a formulation in which the probabilities of the
states (weather) are fixed, where the states are unobservable, but where what is
affected by the employee is the output in each state.
We can represent this formally in the following way. Let S be a set of state
variables (like weather) observable to the agent. Let Q be a set of output variables
(assumed observable to the principal and agent). And let A be a set of inputs
(actions) by the agent assumed observable only by the agent.
Then a compensation scheme is a payment from the principal to the agent
which is a function of all variables that are observable to both the agent and
principal,
Y = (p(Q).
The agent chooses his actions to maximize his expected utility which depends
both on his income and his actions, given
max EU( Y, A, S)

245
Allocation, Information and Markets

where outputs (actions), A, are related to the inputs by a production function

Q = Q(A,S).
We denote the solution to this by
A =H(S).
Finally, we can calculate the expected utility of the principal; his utility depends
on the agent's actions, the payments he makes to the agent and his state (the
actions may affect the principal either directly, or via their effect on outputs, or
via their effects on payments),
EV = EV( </>( Q), Q, A, S).

The principal's problem is to choose </> to maximize his expected utility,

maxEV,

recognizing the dependence of the agent's action on </> and recognizing that he
must pay the agent enough to induce him to accept the job,
EU~ 0. (RU)
POOLING VERSUS SEPARATING EQUILIBRIUM. Much of the literature has focused
on situations where the principal wishes to induce the agent to take different
actions in different states. That is, in the simplest case where only output is
observable by the principal, if A*(S) is the action desired in state S, then the
compensation scheme must be such that
EU[</>(Q(A*, S», A*, S] ~ EU[(Q(A, S», A, S] for all feasible A.
These constraints are referred to as the self-selection or incentive compatibility
constraints.
When the individual takes actions in two different states, so that the observable
variables are the same, i.e. so that the principal cannot distinguish which of the
two states has occurred, we say that there is a pooling equilibrium. When the
individual takes actions so that the principal can identify which state has occurred,
we say that there is a separating equilibrium. (This terminology was introduced
within the context of the adverse selection literature by Rothschild and Stiglitz
(1976).) A basic result of the principal-agent literature establishes conditions
under which the optimal contract involves complete or partial separation.
ADVERSE SELECTION. The variable S can be thought of as a characteristic of an
individual, rather than as the state of nature. Then the self-election constraint
says that individuals of type S prefer action A(S) to any other feasible action.
If the self-selection constraints are satisfied, we can identify who is of what type.
The action may consist of nothing more than making a choice. In the adverse
selection interpretations of the model, the constraint (RU) needs to be replaced
by the set of constraints,
U(</>(Q(A,S»,A,S)~ O(S), for all s

246
 Principal and Agent

that is, there is a reservation utility level for each individual (an individual
rationality constraint for each type). (Note that a similar set of constraints is
relevant if the contractual arrangement between the principal and agent is not
binding, i.e. the individual can quit after he sees what the state of nature is.)
Some examples follow.
(i) The partially discriminating monopoly (see, e.g. Salop, 1977; Stiglitz, 1977).
The firm knows that different individuals have different indifference curves
between the good he sells and other goods, and different reservation utility levels,
but he does not know who is of which type. Q may be the quantity of some
commodity chosen by an individual in which case ¢(Q) can be interpreted as
the payment to the monopolist. (If one individual unambiguously has stronger
preferences for the good, in the sense that at any quantity and payment, the extra
amount he is willing to pay for a marginal unit is greater, then some separation
is always desirable; this property is called the single crossing property.)
(ii) Optimal tax structures (Mirrlees, 1971). The government wishes to impose
differential taxation on different individuals; it may want to impose a higher tax
on the more able, but cannot tell who is the more able. Neither the individual's
productivity nor the number of hours a week he works is observable, but his
income is observable. The income tax schedule specifies a level of consumption
corresponding to each level of income. The individual chooses (by the amount
of work he undertakes) a point on that schedule. A schedule which results in the
more able earning (choosing) higher incomes is one which separates. This will
be desirable if the indifference curves between consumption and income are flatter
for the more able - they require less of an increase in consumption to compensate
for an increase in income. This will be true, for instance, if the underlying
indifference curves between hours worked and consumption are the same for all
individuals.
(iii) Pareto efficient tax structures (Stiglitz, 1982a). In the previous problem,
the government maximized the sum of utilities, subject to the self-selection
constraints, the revenue constraints and the individual rationality constraints
(which simply required that the individual desire to work). The revenue constraint
was equivalent, in this problem, to the profits (revenues) of the landlord; that is,
while in the landlord problem we maximize the revenue, subject to the expected
utility of the individual satisfying a certain constraint, here the dual ofthis problem
is analysed. The 'sum of utilities' is equivalent to 'expected utility' - where the
probability of each state S is identical. We can directly generalize this by imposing
constraints on the level of utility attained by all individuals other than the first;
we then maximize the first individual's utility subject to these constraints (and
subject to the self-selection constraints, and the revenue constraints). This is the
problem of Pareto efficient taxation. It is equivalent to the problem of maximizing
a weighted sum of individuals' utilities.
(iv) Implicit contracts with asymmetric information. (For surveys, see Hart, 1983;
Stiglitz, 1986; Azariadis and Stiglitz, 1983.) With perfect information, the
employer would provide insurance to the employee, to stabilize the employee's
income. If, for instance, the workers' utility function was separable between hours

247
Allocation, Information and Markets

worked, I, and income y,

u= u(y) - v(l),
then with complete information, and risk neutral firms, y will be the same in all
states, but 1will be higher in states where labour productivity is higher. Thus, if
the employer knew the state, but the worker did not, the employer would have
an incentive always to say that it was a good state (since what he paid the worker
was the same, but workers are required to work more in good states). The optimal
contract will induce the employer to announce that it is bad when it is in fact
bad, i.e. it will separate (at least partially).

QUALITATIVE RESULTS. It is clear that many economic relationships fall within

the scope of the 'principal-agent' model. Many of the basic qualitative results
emerge from a detailed analysis of the insurance model:
(a) There is a risk-incentive trade-off; since the risks undertaken will be a
function of the quantity of insurance purchased, if the latter is observable, the
premium will depend on it, and in equilibrium, there will be quantity rationing,
i.e., the individual would like to purchase morl! insurance, at the going benefit
premium ratio (Pauly, 1968). The amount of insurance will be greater, the more
risk averse the individual.
(b) Indifference curves (between benefits and premia) are not in general
quasi-concave, and feasibility sets (the set of insurance premia satisfying the
Hon-negative profit constraint) are not convex; this has important consequences
for the existence of competitive equilibria. The amount of insurance purchased
may not be a continuous function of the price of insurance; and the level of effort
may not be a continuous function of the amount of insurance purchased.
(c) Competitive equilibrium, when it exists, will not in general be Pareto efficient
(Arnott and Stiglitz, 1986; Greenwald and Stiglitz, 1986); the profits of one
insurance firm are affected both by the terms at which other firms offer insurance
contracts (whether for similar accidents or not), and by the prices at which goods
(whether complements or substitutes for accident avoidance or accident inducing
activities) are sold; there exist a set of Pareto improving subsidies and taxes. In
some instances, firms may attempt to internalize some of these 'externalities'.
This leads to interlinkage of markets, both across time (the same insurance firm
insures the individual over time), and at the same time (the same insurance firm
insures the individual for many different risks) (Braverman and Stiglitz, 1982).
The frequently observed interlinkage between credit and land markets in less
developed countries has been interpreted in this light.

VARIANTS OF THE GENERAL MODEL. Further results have been obtained for various
variants of the general model. We discuss a few of the more important versions
below:
(i) Adverse selection model. The major qualitative results of this model (other
than the specification of the conditions under which pooling or separation occurs,
discussed above) entail an analysis of the distortions (relative to perfect

248
 Principal and Agent

information) engendered by the self-selection constraints; in the optimal income

tax, the reduction in work (income) of the less able (associated with a positive
marginal tax rate); in the asymmetric information implicit contract model, in the
existence of overemployment in good states (with the separable utility function
and risk neutral firms), or underemployment in bad states (with very risk averse
firms). To discriminate among individuals, firms may engage in socially wasteful
activities, such as random pricing or long queues. Generally, one group in the
population (the most risk averse in the insurance model, the highest ability in
the optimal income tax model) chooses a contract which does not distort its
behaviour.
(ii) Incentive model with actions taken before state is known. When the random
elements have bounded support, then a first best can be achieved simply by
imposing a large enough penalty for performances below a given threshold. The
individual will exert enough effort to avoid this. (See Mirrlees, 1974; Stiglitz, 1975.)
(iii) Theory of contests. If the output of others performing similar tasks in
similar situations is observable, then one will employ compensation schemes
based on relative performance; these will do better than individualistic
compensation schemes. If there are enough individuals, simple schemes, based
only on individuals' rankings, can approximate the first-best outcomes.
(iv) Models in which the utility constraint is not binding. In some cases, when
the principal maximizes his expected utility, subject to the workers' reservation
utility constraint, the latter constraint will not be binding. Such models give rise
to unemployment. A particularly important variant of these models is described
next.
(v) Models in which quality is affected by price. If the probability of default
increases with the rate of interest charged (either because individuals undertake
more risks when the interest rate is higher, or because those who are less risky
stop applying for loans at high interest rates), then banks may not raise interest
rates, even in the presence of an excess demand for loans. Similarly, if the
productivity of a worker increases with the wage paid (either because individuals
exert greater effort at higher wages or because those who are recruited at higher
wages are more productive), then firms may not lower wages, even in the presence
of an excess supply of labour.
(vi) Terminations. In multiperiod models, it has been shown that the optimal
contract may entail the termination of a relationship when performance is
unsatisfactory; this is shown to be preferable to the imposition of other penalties.
(See Stiglitz and Weiss, 1983.)
(vii) Infinite period models. Long-term relationships may ameliorate some of
the incentive problems (see Radner, 1981). Over an infinite lifetime, the principal
(insurer) can make good inferences concerning the actions ofthe agent (insured);
the relative frequency of accidents will converge to the accident probability
corresponding to the individual's effort level. Not surprisingly, then, with low
enough discount rates, incentive schemes can be designed which approximate
the first best outcomes. The interpretation of this result is, however, subject to
some controversy. Since with low discount rates, the change in lifetime income

249
Allocation, Information and Markets

which would be associated with the individual bearing the full risk ofthe outcome
for any period is negligible, it is as if the individual is risk neutral; and with risk
neutrality we know that first best optimum can be obtained (if bankruptcy is
ignored).

THE SET OF ADMISSIBLE CONTRACTS. One of the important and general results to
emerge from the principal-agent literature is that the nature of the equilibrium
contract depends on the set of admissible contracts. Contracts can depend only
on the available information; typically, it is desirable to use all of the available
information, though in practice, many variables which ought to be relevant (have
information value) are not included within the compensation scheme.
Similarly, if one could costlessly implement a non-linear incentive scheme, such
schemes "would, in general, be preferable to linear schemes. Though in practice,
again, most observed schemes seem relatively simple (linear, piece-wise linear,
etc.), much of the literature has been concerned with characterizing in admittedly
simple situations the optimal non-linear scheme.
In a variety of situations, if one could make pay a stochastic function, it would
be desirable to do so, even with risk averse individuals. (Arnott-Stiglitz (1987)
and Holmstrom (1979) show that with separable utility functions, this will not
happen.) The intuition behind this, in the case when actions have to be taken
prior to the agent obtaining information about the state, is that the possibility
that he receives a low compensation so induces him to work hard that the
employer (landlord) can reduce the dependence (on average) of pay on output,
and thus reduce the variability of income.
Though optimal schemes may thus appear to be fairly complex, in practice
most schemes employed are relatively simple. There is an ongoing controversy
between those who seek to consider increasingly complex schedules, dismissing
work which has analysed simple linear schedules as ad hoc, and those who seek
to explain the kinds of compensation schedules actually employed; these dismiss
the complex solutions as being irrelevant. They would argue that efforts should
be devoted to understanding why actually employed schemes take on the form
they do.
One possible explanation of the use of simple schedules is that they may be
more robust. That is, as technology changes or the probability distribution of
states changes (the exogenous parameters in the principal-agent problem) the
optimal compensation scheme changes. But in practice, revisions to compensation
schemes are costly, and one must find a scheme that works under a variety of
situations. Simple, linear schemes may possess this property of robustness.
Another important characteristic of the set of admissible schemes relates to
commitments. Can, for instance, the worker commit himself not to leave, or can
the employer commit himself not to terminate the relationship?
A closely related issue is the set of punishments (rewards) which are admissible.
It makes a great deal of difference ifthere are limits on the negative compensations
that can be provided in the presence of bad outcomes.
We have noted the role of observability in the design of contracts. In some

250
 Principal and Agent

cases an important distinction may arise between observability and verifiability.

The question is associated with how a contract is to be enforced. If the contract
is to be enforced through the courts, it must be the case that any violation can
be verified by an outside third party. Both the principal and the agent might
know that the contract has been violated, i.e. they both may observe that S (and
not S') has occurred, and therefore that the payment should be that corresponding
to S (and not S'). But unless it can be proved, the principal might attempt to
cheat the agent. Knowing this, the agent would refuse to sign a contract based
on unverifiable variables.
On the other hand, if the contract is enforced by a reputation mechanism,
good behaviour may be enforced so long as the state is observable by both parties.
CONCLUDING REMARK. We have focused here on a discussion of general principles.
It should be emphasized, however, that the principal-agent model has provided
important insights into the nature of a variety of economic relationships, in
labour, land, credit and product markets. These detailed applications of the
general theory represent an important area of on-going research.

BIBLIOGRAPHY
Allen, F. 1981. The prevention of default. Journal of Finance 36, May 271-6.
Arnott, R. and Stiglitz, J.E. 1982. The welfare economics of moral hazard. Discussion
Paper No. 465, Queen's University, Kingston, Ontario, March.
Arnott, R. and Stiglitz, J.E. 1985. Labor turnover, wage structures, and moral hazard: the
inefficiency of competitive markets. Journal of Labor Economics 3(4), 434-62.
Arnott, R. and Stiglitz, lE. 1986. Moral hazard and optimal commodity taxation. Journal
of Public Economics 29(1), February, 1-24.
Arnott, R. and Stiglitz, J.E. 1987. Randomization with asymmetric information: a simplified
exposition. Bell Journal of Economics 17.
Arrow, K.J. 1965. Aspects of the Theory of Risk-Bearing. Helsinki: Yrjo Jahnsson
Foundation.
Azariadis, e. and Stiglitz, J.E. 1983. Implicit contracts and fixed price equilibria. Quarterly
Journal of Economics 98(3), Supplement, 1-22.
Braverman, A. and Stiglitz, J.E. 1982. Sharecropping and the interlinking of agrarian
markets. American Economic Review 72(4), September, 695-715.
Eaton, land Gersovitz, M. 1981. Debt with potential repudiation; theoretical and empirical
analysis. Review of Economic Studies 48,289-309.
Fellingham, J.e., Kwon, Y.K. and Newman, D.P. 1982. Ex ante randomization in agency
models. Rand Journal of Economic Studies 15,290-301.
Gjesdal, F. 1982. Information and incentives: the agency information problem. Review of
Economic Studies 49, 373-90.
Green, J. and Stokey, N. 1983. A comparison of tournaments and contests. Journal of
Political Economy 91, 349-64.
Greenwald, B. and Stiglitz, J.E. 1986. Externalities in economies with imperfect information
and incomplete markets. Quarterly Journal of Economics 101(4), May, 229-64.
Grossman, S. and Hart, Q. 1983. An analysis of the principal-agent problem. Econometrica
51(1), January, 7-45.
Hart, O. 1983. Optimal labor contracts under asymmetric information: an introduction.
Review of Economic Studies 50, 3-35.

251
Allocation, Information and Markets

Helpman, E. and Laffont, J.J. 1975. On moral hazard in general equilibrium. Journal of
Economic Theory 10,8-23.
Holmstrom, B. 1979. Moral hazard and observability. Bell Journal of Economics 10(1),
Spring, 74-91.
Holmstrom, B. 1982. Moral hazard in teams. Bell Journal ofEconomics 13, Autumn, 324-40.
Keeton, W.R. 1979. Equilibrium Credit Rationing. New York: Garland Publishing.
Lazear, E. and Rosen, S. 1981. Rank order tournaments as optimum labor contracts.
Journal of Political Economy 89,841-64.
Mirrlees, J. 1971. An exploration of the theory of optimum income taxation. Review of
Economic Studies 38(2), April, 175-208.
Mirrlees, J. 1974. Notes on welfare economics, information, and uncertainty. In Contributions
to Economic Analysis, ed. M.S. Balch, D.L. McFadden and S.Y. Wu, Amsterdam:
North-Holland.
Mirrlees, J. 1976. The optimal structure of incentives and authority within an organization.
Bell Journal of Economics 7(1), Spring, 105-31.
Nalebuff, B. and Stiglitz, J.E. 1983a. Information, competition and markets. American
Economic Review 72(2), May, 278-84.
Nalebuff, B. and Stiglitz, J.E. 1983b. Prizes and incentives: towards a general theory of
compensation and competition. Bell Journal of Economics 14, Spring, 21-43.
Pauly, M. V. 1968. The economics of moral hazard: comment. American Economic Review
58, June, 531-6.
Radner, R. 1981. Monitoring cooperative agreements in a repeated principal-agent
relationship. Econometrica 49, September, 1127-48.
Radner, R. and Stiglitz, J.E. 1983. A nonconcavity in the value of innovation. In Bayesian
Models in Economic Theory, ed. M. Boyer and R. Kihlstrom, Amsterdam: North-
Holland.
Ross, S. 1973. The economic theory of agency: the principal's problem. American Economic
Review 63(2), May, 134-9.
Rothschild, M. and Stiglitz, J.E. 1976. Equilibrium in competitive insurance markets: an
essay on the economics of imperfect information. Quarterly Journal of Economics 90( 4),
November, 629-49.
Salop, S. 1977. The noisy monopolist: imperfect information, price dispersion and price
discrimination. Review of Economic Studies 44, October, 393-406.
Salop, S. and Salop, J. 1976. Self-selection and turnover in the labor market. Quarterly
Journal of Economics 90, November, 619-28.
Sappington, D. and Stiglitz, J.E. 1987. Information and regulation. In Public Regulation:
New Perspectives on Institutions and Policies, ed. E. Bailey, Cambridge, Mass.: MIT
Press.
Spence, A.M. and Zeckhauser, R. 1971. Insurance, information, and individual action.
American Economic Review 61(2), May, 380-87.
Stiglitz, lE. 1974. Incentives and risk sharing in sharecropping. Review of Economic Studies
41, April, 219-55.
Stiglitz, lE. 1975. Incentives, risk, and information: notes toward a theory of hierarchy.
Bell Journal of Economics 6(2), Autumn, 552-79.
Stiglitz, J.E. 1977. Monopoly non-linear pricing and imperfect information: the insurance
market. Review of Economic Studies 44, October, 407-30.
Stiglitz, J.E. 1982a. Self-selection and Pareto efficient taxation. Journal of Public Economics
17,213-40.
Stiglitz, J.E. 1982b. Utilitarianism and horizontal equity: the case for random taxation.

252
 Principal and Agent

Journal of Public Economics 18, 1-33.

Stiglitz, J.E. 1987. On the causes and consequences of the dependence of quality on price.
Journal of Economic Literature 27, March, 1-48.
Stiglitz, J.E. 1986. Theories of wage rigidity. In Keynes' Economic Legacy, ed. J. Butkiewicz,
K. Koford and J. Miller, New York: Praeger Publishers, 153-221.
Stiglitz, J.E. and Weiss, A. 1981. Credit rationing in markets with imperfect information.
American Economic Review 71(3), June, 393-410.
Stiglitz, J.E. and Weiss, A. 1983. Incentive effects ofterminations: applications to the credit
and labor markets. American Economic Review 73, December, 912-27.
Weiss, L. 1976. On the desirability of cheating, incentives and randomness in the optimal
income tax. Journal of Political Economy 84, 1343-52.

253