Implicit Contracts Under Asymmetric Information Author(s): Sanford J. Grossman and Oliver D.

Hart Source: The Quarterly Journal of Economics, Vol. 98, Supplement (1983), pp. 123-156 Published by: Oxford University Press Stable URL: . Accessed: 15/06/2011 15:56
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact

Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to The Quarterly Journal of Economics.


A model is analyzed where the firm knows more about its own profit opportunities than do its workers. It is shown that because of this asymmetric information, shifts in the productivity of labor can lead to large variations in employment even when labor supply is very inelastic. This is because the employment level chosen by the firm reveals information about its own profitability. This information is useful in making insurance contracts incentive-compatible. Comparative statics results are derived that show the extent to which firm or worker risk aversion, and the probability distribution of labor productivity affect the optimal employment contract and the equilibrium degree of employment variability.


In a Walrasian economy where labor supply is very inelastic, variations in labor demand will lead to variations in the real wage with little or no variation in employment. However, cyclical fluctuations in output seem to be associated with large amounts of employment variability and only small amounts of wage variability (see Hall [1980]). The implicit contract model of employment attempts to explain the low variability of wages in the context of variable labor demand in terms of optimal risk-sharing arrangements between firms and workers (for a survey of this literature see Azariadis [1979]). Unfortunately, when the employer and employees are assumed to share common information, the theory can explain only half of the above phenomenon; the ability to share risk will lead less risk-averse firms to offer more risk-averse workers low wage variability contracts; but if workers' labor supply is inelastic, the optimal contract will also involve low variability in employment (see Akerlof and Miyazaki [1980] and Bryant [1978]). Once we drop the assumption that workers can observe the increase in profit associated with hiring additional labor , however, then this last conclusion is no longer true; the optimal contract may involve high variability of employment, even though the workers' labor supply is inelastic. In this paper we explore in detail the amount of unemployment that is generated due to workers having less information than firms. We analyze the effect on unemployment of risk aversion and the probability distribution of s. It is not difficult to see why under common (symmetric) information, the implicit contract model cannot explain employment fluctuations in excess of those predicted by the Walrasian model. The
(c)1983 by the President and Fellows of Harvard College. Published by John Wiley & Sons, Inc. The Quarterly Journal of Economics, Vol. 98, Supplement, 1983 CCC 0033-5533/83/030123-34$04.40



firm and workers agree on a wage-employment contract which is ex ante Pareto optimal. That is, if s is the random variable that determines the workers' marginal product, and it can be observed by all parties, then the firm and workers agree on a wage rule w(s) and employment rule l (s) with the property that there are no other rules that can make them all better off. The ex ante optimal rule w (s),l (s) must be ex post optimal in each state s. For if in any state s both the firm and workers could be made better off by deviating from w(s),l(s), then w(s),l(s) would be the wrong rule to choose ex ante. Consequently, the employment rule will call for employment of a worker if and only if his marginal product is above his reservation wage. Since the optimal ex ante employment rule leads to all gains from ex post trade being exploited, the actual employment outcomes with implicit contracts are identical to those that would arise if there was a spot market for labor for each realization of s.1 Consequently, if the supply of labor is very inelastic, then variations in the marginal revenue product will not lead to variations in employment. Recently, models have been developed in which firms and workers have asymmetric information about the state-in particular, where firms have better information than workers.2 Such models, which incorporate certain types of moral hazard or adverse selection, can be shown to provide a considerably richer theory of employment than the symmetric information models. In particular, Grossman and Hart [1981] have shown that the presence of asymmetric information will, under certain assumptions, cause unemployment to be greater than in the symmetric information case or than in the Walrasian model, and unemployment can occur when the marginal revenue product of labor exceeds the reservation wage.3 Our earlier result is based upon the idea that insurance creates moral hazard when the underlying source of risk is not observed by both parties to the insurance contract. In particular, if the net payments made by one party (the firm) must fluctuate in order to achieve insurance and that party is more informed than the other party (the workers), then moral hazard will disrupt the insurance. More importantly, both parties, being cognizant of moral hazard, may decide
1. The conclusion that the employment outcomes are the same as in a spot market depends on the assumption that the worker's reservation wage is independent of his income. We assume this in what follows. 2. See, e.g., Azariadis [1983], Calvo and Phelps [1977], Chari [1983], Green and Kahn [1983], Grossman and Hart [1981], and Hall and Lilien [1979] for some models of asymmetric information and unemployment. 3. The possible importance of asymmetric information has been recognized by a number of authors. See, e.g., Gordon [1976].

there is no clear reason for the workers to prefer one type of contract to the other. this requires stronger assumptions than are necessary for the firm to establish a reputation about layoffs. In particular. Thus. in the long run workers will be able to verify the truthfulness of the firm's announcements. actual wages will always be low. This means that its wage bill must vary with &.In particular. even though the contract may specify that wages should be high in good states and low in bad states. For then the workers must rely on the firm to tell them which state has occurred. In general. and so the "variable wage-fixed employment" contract becomes a "fixed (low) wage-fixed employment" contract. Matters change considerably. It might be thought that. For then workers need only check that the firm is telling the truth on average. the firm may wish to misreport the state. given the "variable wage-fixed employment" contract. however. it is always in the firm's interest to claim that a bad state has occurred. Thus. This is because its wage-employment implicit contract is defined in terms of quantities that are immediately observable.. generalize to "many-shot" situations in which the firm's shocks are serially correlated or the time horizon is relatively short or there is discounting. however. since the implicit contract model is based upon the idea that firms establish reputations. In one (the "variable wage-fixed employment" contract). . where average truthtelling cannot be verified over a long period. workers are guaranteed a particular wage rate but may be laid off (or made redundant) in bad states. assume that the firm is risk-averse with respect to its profit stream (In Section VI we provide a justification for this. even if s is correlated over time or the discount rate is positive (so Radner's result would not hold). its wage bill must be low when s is low. workers are guaranteed employment. the state in the industry as a whole. This problem will be less serious if workers can observe variables that are correlated with the state in this firm. Our results should. Consider two different contracts that the firm might offer its workers. however. In this paper we shall study a "one-shot" situation. since this allows the firm to reduce wages without sacrificing employment. To understand the basic issue. Each of these contracts permits the wage bill to vary with s and shifts some of the risk from the firm to the workers. if only the firm can observe the state. firms can develop a reputation for their wage-employment policy. The moral hazard problem will also be reduced if the firm and workers have a very long relationship with each other and (1) the shocks (states of the world) that the firm is subject to are independent over time and (2) there is no discounting. e.g. See Radner [1981].IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 125 to change the employment rule to be ex post inefficient as -a method of generating an ex ante optimal contract. However. the "fixed wage-variable employment" contract is 4.) This implies that it is never optimal for the firm to bear all the risk of the variations in profit caused by s. Workers can observe immediately whether the firm deviates from an announced wage-employment schedule w(l) to be defined below. but agree to take a wage rate cut in bad states of the world.4 In contrast. In the other ("fixed wagevariable employment"). In a world where both the firm and the workers observe the state ex post.

Here we shall be interested in how the amount of employment explained by the implicit contract-asymmetric information theory-over and above that explained by Walrasian theory-depends on such exogenous variables 5. We may conclude that the "fixed wage-variable employment" contract may be implementable in a way that the "variable wage-fixed employment" contract is not. . there will be a set of variable wage-variable employment contracts that are implementable (w(s). The firm and the workers agree ex ante on a Pareto optimal member of this class..e.l) . gain) of labor at a time when this labor is relatively productive (respectively. the implementable policies can be thought of as w(l). a wage schedule d(l) always exists that can make the firm and the workers better off and has the property that &'(l) > R.l(s). but that wage rates should remain constant. This twisting of the wage schedule means that in bad states of nature. or else the firm would not be choosing I to maximize profit. Therefore.. since at w'(l) = R the net marginal benefit (or loss) involved in a small deviation in employment from the optimum is zero. w'(l) = R). there will be states of nature where the workers' marginal product is larger than their reservation wage and hence there is underemployment. good) when in fact it is good (respectively. where w is the total wage bill and 1is labor supply.126 QUARTERLY JOURNAL OF ECONOMICS less obviously open to abuse.l(s)). This is clearly suboptimal when the firm is risk-averse. given a wage schedule that has ex post efficient employment (i. where the wage bill is taken to be a predetermined function of the labor utilization. the firm has a lower wage bill than under the contract w (1). Suppose instead that the firm and the workers agree on a new contract &(1) with wi(l) > R and d(0) < w(0). This must mean that w'(l) = R for all 1.5 Though Grossman and Hart [1981] show that there will be ex post inefficient unemployment. where f is the revenue function. For a small twist in the schedule. Hence. it has effective control over employment and so for a policy to be implementable l (s) must maximize the firm's ex post profit f (s. The main result of Grossman and Hart [1981] is that a Pareto optimal contract will generally involve some unemployment in states of nature where a worker's marginal product is above his reservation wage.w (l) for each s. w can vary with s only if it is accompanied by 1 varying with s.This means that the net income of the workers (net of disutility of effort) is w(l) . the efficiency losses due to &'(l) > R are of second order. This possibility does not seem very attractive. all the firm achieves is a loss (respectively. unproductive). With i'(l) > R. the firm has a less risky profit stream than under w (1).RI = w(0) which is a constant independent of the state s. This implies that the workers' wage bill is given by w (0) + RI. since only the firm observes s. Suppose that contraryto what we claim. under &(I). by claiming that the state is bad (respectively. Furthermore. For the reason given above. and the firm bears all the risk. workers are always employed to the point where their marginal product equals the reservation wage. i. for this contract specifies that layoffs should occur in bad states. Thus. Our argument assumes that the reservation wage is a constant R and is motivated as follows. when &is low.e. Thus. they do not analyze the determinants of its magnitude. As a consequence. bad). In general. the workers bear no risk.

Getting some idea of the importance of these factors would seem to be an essential step toward assessing the empirical significance of the implicit contract-asymmetric information theory of employment. the workers' reservation wage. In Section II the model is described. THE MODEL The model is based on that of Grossman and Hart [1981]. concluding remarks appear in Section VI. where W (respectively. The paper is organized as follows. etc. his utility is U(W . Here R represents the worker's . are observed by the workers. Let G be the distribution function of s. (respectively. If the worker is employed. and perhaps other variables too. W') is the worker's wealth when employed. Finally. In Section III we analyze how unemployment is affected by changes in the risk aversion of the firm and worker. but not how many other workers are employed by his firm. the productivity of labor. All potential workers are assumed identical. We shall therefore simplify matters considerably by confining our attention to the case where a representative firm makes a contract with. his utility is U(W'). II. Consider a (representative) firm that can employ at most one worker. a single worker. and if he is unemployed. G (s) = 1 for some s < s-. unemployed). Each worker has a von Neumann-Morgenstern utility function U. the workers' risk aversion.R). but it should be noted that our results generalize to the case of a firm which makes a contract with many workers under the assumption that each worker can observe only whether or not he is employed. If employed. Analyzing the effect of risk aversion and productivity on unemployment in a general model in which each firm makes a contract with many workers is difficult. This is obviously a very special case. We assume that G (s) = 0. In Sections IV and V we consider how unemployment is affected by changes in the probability distribution of the marginal product of labor. which is positive for s < s < g.We assume that G has a continuous density function g(s). It is to be hoped also that our results will be a useful starting point for the analysis of the many-worker case when total employment within the firm.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 127 as the firm's risk aversion. and employs. The comparative statics results that we obtain may also provide a way of distinguishing at the empirical level between the implicit contract-asymmetric information theory and other theories of unemployment (see also Section VI). the output or revenue (we shall not distinguish between the two) of the worker is given by the random variable s.

.) Thus. 7. It is also assumed that if the worker signs a contract with this firm. If production does occur at date 1. i. 8. but that cannot readily be transferred to other firms.128 JOURNAL OF ECONOMICS QUARTERLY disutility of effort.7 We suppose that at date 0. see Chari [1983] and Green and Kahn [1983].. No actual production or employment occurs at this date. What we are trying to model is the idea that the worker is generally less productive in other firms at date 1 than in the firm with which he has signed the contract-that is.8 Thus. the firm learns the realization of s.e. when the contract is signed. See Section VI and footnote 4 for a motivation for the asymmetric information assumption. we have in mind that the period between date 0 and date 1 is a training period during which the worker acquires skills that are useful for this firm. that at date 1. i.. (We also assume that U. the marginal product of labor exceeds the reservation wage with positive probability. which could be relaxed.6 We shall assume that s > R. production and employment can occur. For models in which the reservation wage is allowed to depend on wealth. and R are known to both parties at date 0. There is no difficulty in generalizing the model to the case where at date 1 the worker can earn a certain amount R' > 0 elsewhere if he is laid off by this firm. the firm's output is zero. We shall assume that the worker'sreservation wage is independent of his wealth W. At date 1.e.. as described above. We shall suppose that the attitudes to risk of the firm's owners can be represented by a von Neumann-Morgenstern utility function of profit V(wr). If production does not occur. before any employment decisions have to be made. V. in contrast. At date 0. but that the realization of s is known to neither. there is asymmetric information at date 1. the firm and worker have symmetric information. we suppose that there is a minimum level of expected utility U that the worker must be offered by the firm if it wishes the workerto sign a contract at date 0. We shall call R the worker's reservation wage. particularly if workers have better information about the realization of R' than does the firm. This is an extreme assumption. however. The worker. Matters become complicated (and more interesting) if R' is a random variable. the firm decides whether to sign a contract with a worker. Finally. One can 6. the worker is unemployed. The contract will specify under what conditions they will occur. We shall suppose.e. it is the amount that he must be paid to compensate him for working. The model is a two-period one. i. or to put it another way. is assumed not to observe s. G is known to both parties. his productivity elsewhere is zero. the firm's output is s. however. if we assume that a contract has been signed. he cannot work anywhere else at date 1. The only effect this has is to increase the effective opportunity cost of labor or reservation wage from R to (R + R'). the worker is employed. at date 0.

.. Note that we allow w(s) > 0 when L(s) = 0. a contract under symmetric information is a pair of functions (w(s). and that U is the market-clearing "wage" (in expected utility terms) in the market for contracts. Allowing layoff-redundancy pay changes the form of the optimal employment rule both when s is common knowledge and when it is observed only by the firm..s < s < s). i. 10. It is important to emphasize the different assumptions that we make about the mobility of labor at dates 0 and 1. i. see Azariadis [1979]. U(x) = -O. i. U' > 0.L(s). oo) of the real line and lim. it is common knowledge. mobility is assumed to be much reduced (in the extreme case considered here it is taken to be zero).9 The Optimal Contract We wish to study the form of the optimal contract between the firm and worker at date 0. a contract that maximizes the firm's expected utility subject to the constraint that the worker's expected utility is at least U.10 When s is common knowledge. Thus. we permit layoff (or redundancy) pay. U is defined and twice differentiable on an interval Y = (a. labor is highly mobile. In (Al) and (A2) we allow for the possibility that a or a' =O. In addition. It is helpful to begin by considering the case where.oo) of the real line and limx-a' V(x) = -O. A2. i. we make the following assumptions about U and V. U" < o on Y. Note first that any contract which is ex ante optimal at date 0 before s is known must be ex post optimal at date 1 for each realization of s (for if both the firm and the worker can be made better off in some 9. This is in contrast to much of the implicit contract literature. . Al. by the time firm-specific skills have been acquired at date 1.e. V is defined and twice differentiable on an interval 5 = (a/. V' > 0. where L(s) = 0 or 1 for each s. in contrast to our above asymmetric information assumption.e. The market for contracts at date 0 is taken to be a standard Walrasian one. In contrast. In order to proceed. In addition.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 129 imagine that there are many identical firms and workers trying to find partners at date 0.e. V" < Oon 5'. indicating for each state (1) whether the worker is employed (L (s) = 1) or unemployed (L(s) = 0).e. Thus. See Grossman and Hart [1981].. there seems no reason to rule out layoff-redundancy pay in the present model. it is easy to characterize an optimal contract. Unless other considerations are introduced.. Then a contract can make both wages and employment at date 1 conditional on the realization of ?. At date 0 it is assumed that workers have not yet acquired firm-specific skills and therefore could work in a large number of different firms. (2) his wage w(s). a. s is observed by both parties at date 1. our model does not explain (involuntary or non-Walrasian) unemployment at date 0.

Matters can be simplified considerably by appealing to results from the incentive-compatibility literature. whether L(s) = 0 or 1. s is observed only by the firm. e. There it is shown that in studying the optimal contract attention can be confined to the case where truthtelling is an optimal strategy for all agents (see.. We suppose. The reason w(s) is constant over unemployment states is that social output is constant (equal to zero) over such states.R + E. then both parties can be made better off by letting production occur and increasing the wage from w(s) to w(s) + R + E. The contract then specifies a wage w(s) and an employment level L(s^)(equal to zero or one).w(s)) (21) *~ ~~~V U (W. both parties can be made better off by stopping production and reducing the wage from w (s) to w (s) . and U'(w(s) . which satisfy L(s) = 0. a constant. this involves choosing the wage function w (s) such that w (s) = we. If layoff-redundancy pay is ruled out. it can be shown that without layoff-redundancy pay the ex ante optimal employment rule is employ if and only if s > k. Now a necessary condition for ex post Pareto optimality is that production should occur if and only if s > R.11 The remaining part of the optimal contract when s is common knowledge at date 1 concerns the wage function w(s). We turn now to the more interesting case where. This will be chosen to ensure optimal risk sharing between the firm and the worker.g. at date 1. 11. We suppose that.) V(-Wu) over states s. at date 1. We analyze the optimal contract in terms of the following revelation game.It follows that any ex ante optimal contract must specify the employment rule: employ if and only if s > R. then their expected utilities at date 0 can be increased). Under (A1)-(A2). conditional on the announcement s = s'. This argument depends on the possibility of layoff-redundancy pay. if production occurs when s < R.1) is the condition that the marginal rate of substitution between income in employment and unemployment states should be the same for the firm and worker. For if production does not occur in some state s > R. Finally. it is no longer the case that ex ante optimality implies ex post efficiency. where L(s) = 1. In fact. i.R) V'(s .130 QUARTERLY JOURNAL OF ECONOMICS state s and no worse off in any other state. See Azariadis [1979]. where k is a number less than R.. Equation (2. There is an obvious and unimportant ambiguity when s = R. conversely. we will be chosen so that the worker's expected utility equals U.e. . where E > 0 is small. the firm is asked to announce the realization of g it has observed. however. over the states s. that the worker does observe whether he is employed at date 1.

under asymmetric information. We rule out this indeterminacy by adopting the convention that s ' k ' s. . since by doing this. a contract in effect specifies two wages: wage we when the worker is employed at date 1. similarly. we can deduce the wage when employed. the extra amount the firm has to pay to employ him at date 1. Thus. Thus. if s d E. This argument assumes that E and U are nonempty. > s) will do. a contract can be characterized by two numbers: wu.12 For whenever s > k. Let E = {s IL (s) = 1} be the set of employment states and U = Is IL(s) O}be the set of unemployment states designated by the contract. we may assume without loss of generality that the contract is such that the firm always wishes to report the true state. Then the condition that truthtelling is an optimal strategy tells us that E tsIL(s) = 1}= sls > k} and U IsIL(s) = O}= Isls < k}. and if s d U. Hammond. the wage when the worker is unemployed at date 1. Note also that. we = wu + k. In other words. if w(s1) > w(s2) where sl. see Remark 1 below and Section V). unemployment always). However. and wage wu when he is not employed at date 1. s < k. where s1. For. and if E is empty. 12. and Myerson [1978]). Neither of these possibilities is consistent with truthtelling. the firm will pretend that the state which has occurred lies in U. In both cases E = IsIs > k . From this. we can set we = wu + s. and U = IsIs < k}. and the employment rule: employ if and only if s > k. the firm will pretend that s = S2 when s = si. there is an indeterminacy in k: any k ' s (respectively. it can reduce wages without sacrificing employment.s and k = s without changing anything. the firm will pretend that the state which has occurred lies in E. s > k. and k. k = s. The state s = k can lie either in E or in U. and Maskin [1979]. Then the condition that truthtelling is optimal implies that w (s) equals a constant we on E and that w(s) equals a (possibly different) constant Wuon U.s2 e E.wu). Let k = (We . we can always set wu = we. Instead we can imagine that given the wages wu and (wu + k). Note that in the case where there is full employment always (respectively. given the form of the optimal contract. the firm gains more from employing the worker than it has to pay for this employment.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 131 Dasgupta. since it occurs with probability zero (however. the firm simply chooses whether or not to employ the worker at date 1. 13. if U is empty.s2 E U. We do not bother about this state. if w(s1) > w(s2). there is no need for the firm actually to report s at date 1.13 We see then that in the case where only the firm observes the state at date 1. and vice versa when s < k.

w (s). However.132 QUARTERLY JOURNAL OF ECONOMICS DEFINITION. we have confined our attention to deterministic contracts. since the insurance is conditional on the announcement s and. (wu + k .. In the above.3) gives the worker's expected utility. 15. the firm will always wish to go to a (risk-neutral) insurance company and arrange that. the worker receives a net income of (wu + k . Note that there is no moral hazard in such an arrangement. there are k . Under (A1)-(A2). L (s) is the outcome of the lottery. Expression (2. An optimal contract under asymmetric information is a pair (wuk). Maskin [1981] has shown that it may be desirable to make the contract stochastic.wu sat14. once this announcement is made. Such schemes make the firm's profit sL (s) . s < k s. i. when the set of feasible employment and output levels is convex.R)EY. Alternatively.k) An optimal contract is one that maximizes the firm's expected utility subject to the worker's expected utility being at least U.R) when employed and wu when unemployed.e.G(k)) + U(wu)G(k) 2 U WUEf.w>) for some function W. Thus.iV (s) a stochastic function of s. the firm can always increase profit by reducing wu.2) is the firm's expected utility. to specify a random wage iv (s) and employment level L (s) conditional on the announcement s. (A1)-(A2) also ensure that as long as the constraint set is nonempty.3) will hold with equality. In particular. one can imagine that the firm earns profit 7ro elsewhere which is sufficient to cover its wage bill. Recently.4) V(s . We assume that V is defined for some levels of negative income (if W.k)g(s) ds + V(-wu)g(s) ds U(wu + k .V(. that is. which maximizes Ss k (2. i. having signed such a contract. where 7ro2 Wu. we have assumed that the contract specifies a certain wage w (s) and employment level L(s) for each s announced by the firm. (2. . where w(s).R)(1 . the insurance company will guarantee the firm a certain income of E[sL(s) . since when s > k the worker is employed and paid (k + wu). A contract when only the firm observes the state at date satisfying s < k < . if the worker cannot monitor (and hence prevent) such insurance. ? 0). if not.w) = W(7ro. note that it may be difficult to implement a random contract. since.Wu. -wuEY.. 15 The left-hand side of (2.C(s)] in exchange for the firm giving the insurance company sL (s) .. the firm's actual profit will be a deterministic function of s. it can be shown that stochastic contracts have no advantage over deterministic contracts. while when s < k the worker is unemployed and paid wu. In the situation given in the text. if at date 1 the firm announces that the state is s. DEFINITION. where the realization of &(s) and L(s) is determined by an objective lottery.3) and (2.14 1 is a pair (w.e. Under such conditions. that is. so there is some benefit from randomization that will disappear when L is a continuous choice variable and the production function is concave. L is discrete. the realization of L (s) and a(s) are determined by an objective lottery over which the firm has no control.2) subject to (2.

This is because the maximization problem in (2. Let (WUk) be an optimal contract. k = max(s. PROPOSITION 1.. i.3)-(2. If the firm is risk-neutral. there may be more than one optimal contract. unemployment will not occur in states close to Th Proof of Proposition 1.) The proposition below.4) is nonconvex. the optimal contract is unique.e. (In contrast. The nonemptiness of the constraint set is guaranteed by (A3). Since the firm's expected utility 16.16 Proposition 1(2) says that the probability of unemployment will still be G (R) under asymmetric information if the firm is risk-neutral. which was also proved in Grossman and Hart [1981]. i. then k > R.e. there may or may not be unemployment under asymmetric information. and where there is simply a spot market for labor at date 1.1) is w(s) = wu + R for s satisfying L(s) = 1. V"(wr)< 0 for all 7 E Y'. V. an optimal contract exists. this wage path can be realized under asymmetric information by setting k = R. and this yields efficient employment.. and (b) R > s.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 133 isfying (2. Proposition 1 says that the optimal k may exceed s or it may equal s. in an optimal contract. provides some information about the form of an optimal contract. the optimal contract under symmetric information can be implemented even when there is asymmetric information. the probability of unemployment is G(R)). Then (1) s-> k > R. i. unless both the firm and the worker are risk-neutral.e. In general. ex post efficiency is always achieved and so unemployment occurs if and only if s < R. (2) if the firm is risk-neutral. When we speak of the Walrasian model. depending on the exact form of U.e.. Assume (A1)-(A3). i.G. Note finally that Proposition 1(1) also tells us that. (3) if R > s and the firm is risk- averse. V" (or) = 0 for all wrEY'. We will consider this case further in Section V.R).. If R < s. For when V' = constant..4).2)-(2. . Proposition 1(1) tells us that the probability of unemployment at date 1 will never be lower under asymmetric information than under symmetric information or than in the usual Walrasian model of the labor market (in the latter two cases. However. there is full employment under symmetric information. U(w-) = U. when s is common knowledge. if the probability of unemployment is positive under symmetric information. we are referring to the standard model in which there are no training costs or locking-in effects. Proposition 1(3) says that the probability of unemployment will be greater than G (R) under asymmetric information if (a) the firm is risk-averse.e. the solution of (2. that is. There exists i satisfying wd0. -wd0'. i.

G(k)).G(k)). it matters whether s = k is an employment state or an unemployment state. the left-hand side (LHS) of (2.G(k)). Note also that the first-order conditions are not satisfied at k = s-. wages cannot be conditioned on s directly.G(k)). it is clearly optimal to make s = k an employment state if k > R (since then we = Wu+ k R > wu and so the worker prefers to be employed) and an unemployment state if k < R (since then we = wu + k .R))g(k) U'(wu + k . we may have k = R). Consider the first-order conditions for the problem (2. with equality in (2. These yield V'(s -kWd (2.R)(1 . the right-hand side of (2. If the worker cannot observe s. since k > s. For when k = R. although part (3) may not (i. since U'(wu + k .D.134 QUARTERLY JOURNAL OF ECONOMICS can clearly be no higher under asymmetric information than under symmetric information (any contract that is feasible under asymmetric information is feasible under symmetric information). Proposition 1 supports the intuition given in the introduction. if k = s-.k) dG + V'(-wu) dG U'(wu + k .E.G(k)) + (U(wu) . we must replace the equality by >.Wu. Hence (2.4).3).G(k)) + U'(wu)G(k) if s < k < s-. Note that when G is discrete. such as em- .by <. (If k = s. However. since V'(s . Q. this proves part (2).On the other hand.k) < V'(-wu) when s > k.5) -fk)dG V'(s .R) > U'(wu) and U(wu) > U(wu + k .2)-(2. Hence part (1) is true.R)(1 .R). Remark 1. but only on observed variables determined by s.5). Since the firm is indifferent between employment and unemployment when s = k. Note that part (1) clearly holds if R < s. This proves that k > R.5) > (1 .5) cannot hold when s < k < R < s-. the above argument shows that k = R also does not satisfy (2. Proposition 1 holds for the case where G has a density function g.R < wu and so the worker prefers to be unemployed).) Since V'(s . This establishes part (3).5) = (1 . We prove part (1) for the case R > s. the first-order conditions are also not satisfied if s = k < R. it can be shown that parts (1) and (2) still hold.e. If G is discrete. while the left-hand side < (1 . If V" < 0.U(wu + k .k) < V'(-wu) for s > k.5) < (1 .Wu.Wu . the right-hand side (RHS) of (2. if k < R..

. that the level of employment under asymmetric . This is because this is the only way that the (risk-averse) firm can reduce its wage bill in times when s is low (the firm would like to reduce wages without sacrificing employment.R > wj) and hence the worker regrets the fact that he is unemployed.e. In the taxation problem. In contrast. For this reason. in order to get the right degree of risk sharing. in terms of this paper.> We = w. Note also that our result that k > R is exactly analogous to the result that the optimal marginal tax rate is positive-both results imply that the utility of the uninformed agent (the utility of the worker or the revenue received by the government) is higher the better is the state or ability of the informed agent (the firm or the consumer-worker). The firm would like to make wages depend on this information directly.. It is voluntary in the sense that the worker is a willing party to the contract at date 0. Finally. the consumer-worker has information about his ability that the government does not. but because of moral hazard this is not possible). As a result. Note that in the implicit contract model one considers a contract between a worker and one firm which may end up in one of a large number of states. in particular in states satisfying k > s > R. i. The government would like taxes to depend on ability directly. i. There is a close formal relationship between the model presented here and optimal income tax theory (see Mirrlees [1971]). In particular. it may be necessary for employment to vary more than is desirable from an efficiency standpoint. i. In our model the firm has information that the worker does not. the worker (the uninformed agent) in our model corresponds to the government (the uninformed agent) in the tax model and the firm to the consumer-worker. It is involuntary in the sense that the worker is better off in employment states than in unemployment states (for k > R . employment levels. Thus.. whereas in the tax model one thinks of the government taxing a large number of different people in different states. our result that unemployment will not occur in states close to s. it may be desirable at date 0 to agree to let unemployment occur in states where the marginal product of labor exceeds the reservation wage. i.. the government is risk-neutral. the government in the tax model needs to balance demands and supplies only in average or expected terms. Remark 2. One may ask whether the unemployment that occurs at date 1 is voluntary or involuntary.e.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 135 ployment.e. but instead must base taxes on the observed consequences of ability. labor supply or income. but instead must base wages on the consequences of this information. in the contract model one is interested in the case where the worker may be risk-averse.e. + k .

3) is increasing in w1. we can draw indifference curves for the firm in Figure 1.G (k)) + U(w.R)(1 . w. In this section we consider how this excess unemployment depends on the degree of risk aversion of the worker and firm.and therefore wu(k) < 6-.1) exceeds U when wu = w-.) Thus.where U(MO) U (the existence of UT is guaranteed by (A3)). In order to answer this and other comparative statics questions. The model presented here is also an example of the principal-agent problem. (Note that EV is decreasing in wu and in k. (k) + k . is analogous to the result that the marginal tax rate should be zero for the most able (see Sadka [1977]).3) when there is equality. We know from the proof of Proposition 1 that. 17. but may be increasing or decreasing in k (an increase in k raises the wage when employed but reduces the probability of employment).1) U(w. .thatis. Also. How THE RISK AVERSION OF THE FIRM AND WORKER AFFECTS THE LEVEL OF EMPLOYMENT In the last section we saw that an optimal contract under asymmetric information is generally characterized by unemployment in excess of the spot market level or of the unemployment under symmetric information. it is useful to investigate the constraint (2. It is therefore natural to ask whether the probability of unemployment increases as the firm becomes more risk-averse. is as in Figure I. (s) = U-. is Weitzman [19801. given by (2. and that obtains comparative statics results for some special cases. Hence w. For the remainder of the paper we shall assume that the firm does find it in its interest to make a contract with a worker at date 0. the left-hand side of (3. = Clearly w. if R > s. (3. (R) = w.k. is also a function of wu.(k). Finally.2). Proposition 1 tells us that the probability of unemployment will exceed the Walrasian level only if the firm is risk-averse.R).136 QUARTERLYJOURNAL OF ECONOMICS information is equal to the Walrasian level in the best state of nature. Proposition 1 is generalized to the case of a firm that employs many workers in Grossman and Hart [19811. for s > k > R. A paper that adopts a somewhat similar approach to that taken here. The left-hand side of (2. We therefore use (2.3) to write wu as a function of k: Wu=wu(k). for max (s. (k))G (k) = U. The firm's expected utility EV. is differentiable for s > k > max(s.17 III.R) < k < s.

if s < R. The optimal contract is at k = k in Figure I. the firm's indifference curves are less steep than the worker's at k = R. since (1) these are necessary but not generally sufficient and (2) there may be more than one optimal contract.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION Wu 137 Wu w k k EV= constant EV = constant FIGURE I dwl dk E V=constant - -LHS of (2.j) is an optimal contract before the change and (&'. every optimal contract will satisfy k > R. See Appendix. In other words.. Let the firm's utility function V be replaced by H(V). PROPOSITION 2. strictly concave Assume that (A1)-(A3) function defined on the range of V(wx). if (&. h' > i. hold before and after the change. An increase in k increases the wage that the worker is paid in employment states. increasing. The intuition behind Proposition 2 is the following. Proof. Furthermore. it is not sufficient to differentiate the first-order conditions (2. This will change the slope of the firm's indifference curves.P') is an optimal contract after the change.5).5) dw| dk ElIT=constant at k = R. Suppose now that the firm becomes more risk-averse. Note that to prove this proposition.5) > -RHS of (2. and this enables the firm to reduce wu(k). P' > I. which restates Proposition 1. It is thus clear from the diagram that if R > s. We now show that this will have the effect of increasing the optimal k and thus increasing the probability of unemployment G(k). where H is a differentiable. the firm's profits become less risky in a well-defined sense-they rise in bad unem- . As a result. Then.

See Appendix. Hence if (wu. minimizeswu(k). this will also generally lead to a change in U.1). let k* be the smallestof these). Proposition 2 says that k increases as the firm becomes more risk-averse. PROPOSITION Assume (A1)-(A3). We shall write U = U(0) as in (A3) and assume that the worker can continue to get the certain income w if he goes elsewhere at date 0. Proposition 3 is easy to understand.k < k*. We now consider the influence of the worker's risk aversion. there exists a such that. then U is replaced by Cf(-i). . It should be noted that when the worker's risk aversion changes.k < Proof. so that if the utility function U is replaced by U. however. Propositions 2 and 3 tell us how the probability of unemployment depends on the firm's risk aversion. Then (1) if (wuk) is an optimalcontract. it would be paying a higher wage both when the workeris unemployedand employed.This opportunity reduceriskis moreattractive.where wu(k) satisfies (3.e. Let k* be the smallest k that 3. which makes the worker's net income riskier.k . it is helpful to look at Figure I. to "maximin. This is done by setting k = k*.Is there an upper limit to the optimal k as the firm becomes more and more risk-averse?To answer this question.wu(k). The next we proposition shows that as the firm becomes more and more riskaverse. The opposite is true of the worker. and real= izingfeweremploymentstates.fall).138 QUARTERLY JOURNAL OF ECONOMICS ployment states and fall in good employmentstates (expectedprofit may. (2) given E > 0. Wu wu(k*). k will convergeto k*. We have argued that an increase in k makes the firm's profit stream less risky (in some sense). This suggests that an increase in the worker's risk aversion may reduce the optimal k. Let k* be the value of k that minimizes (k) Wu (if there is morethan one k that minimizeswu(k). the worker's opportunity cost of labor at date 0. Hence the firm wants to maximize -wu(k). It is clearthat no firm. if (. i. to the more risk-averseis the firm. its would ever select k > k*..V"(r)/V'(w)) > a for all ir." Profit in unemployment states -w. A very risk-averse firm wishes approximately to maximize income in the worst state.than if it selectedk = k*.. musthavek < k*.whatever attitudesto risk. For if it did so.k)is an optimalcontract. any optimal contract has the property that Ik . (k) is less than profit in employment states s . The next proposition establishes this for the case where the worker'sand firm's utility functions exhibit constant absolute risk aversion. an increase in k causes wages when employed to rise but when unemployed to fall.

The the probabilityof unemploymentwill be higherthe morerisk-averse firm is. Proposition 4 does not appear to hold for generalutility functions. We may summarizethe results of this section as follows. It follows fromthese results that.G(k)). as the next propositionshows. See Appendix.R .G(k)). there are no wealth effects.G(k)). In the exponential case the probability of unemployment will also be higher the less risk-aversethe workeris. then E'< A. Proof.R) underthese conditions. and so k* is the smallest value of k that maximizes (k .max(s. there exists a such that if (-U" (w)IU'(w)) > a for all w. The intuition behind Proposition5 is that if the workeris very risk-averse. See Appendix./)is an optimal conis tract when a = ao and (dW. ranging from G(R). Furthermore. Then.we can askwhathappensto the optimal k as the worker'srisk aversionbecomes very large.The next proposition tells us that k will be very close to max(s. given e > -e -bi and the worker's utility function by U(w) = -e -aw.where k* is defined in Proposition3.R)(1 . PROPOSITION 6. in the exponential case. Callthis E.the probabilityof unemploymentis at its greatest when the firm is infinitely risk-averseand the worker is risk-neutral. Let T be the smallest value of k that maximizes (k . Let the firm's utility function be given by V(r) = ab > O. Assume (A1)-(A3).R)(1 .. to R if R > s. reachingits minimumlevel G(R) when the workeris infinitely risk-averse(the last statement is also true in the general non-exponentialcase). Assume (A1)-(A3).R)f < e. Proof. As in the case of the firm.R)(1 . PROPOSITION 5. the Walrasianlevel. to G(k*).IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 139 PROPOSITION 4. wu(k) = U-(k . if (Wd. It is worthnoting that the proofof Proposition4 establishesthat the optimal k (or k's if there is more than one) is independent of U in the exponential is optimal for the firm to offer the workersomething This it does by settingk close close to completenet incomeinsurance.Also let U = -e-aw. Then.wu(k). fi < k. any optimal contract has the propertythat Ik .if s < R. Then (1) if (wak) is an . if we regardas variableboth the firm'sand the worker'sutility functions. since then wu(k) + k . when the firm is risk-neutral(see Proposition1). this last result is true in the generalnon-exponential case. i.k') an optimal contract when a = al > a. when the firm is infinitely risk-averse. Note that when the workeris risk-neutral.e. where 0.In fact.

Consider first what happens if there is a uniform shift to the right in the distribution of s. k < k. and R increases by the same amount. there exist a. Under symmetric information this will lead to a lower probability of employment. See Appendix.k < E. 7. the probability of unemployment = prob(g < k) will not change. PROPOSITION Let the firm's utility function be given by V(ir) = -ebr and the worker's by U(w) = -e-aw. is G(k) - G(R). Let the random variable &be replaced by (&+ a) and R by (R + a). If we substitute R' = R + a. &' = s + a. we have 8. It tells us that. i. Then. which can be explained by the implicit contract-asymmetric information theory. the probability of unemployment can lie anywhere between G(R) and G(k).2)-(2. where ab > 0. any optimal contract has the property that Ik ..140 QUARTERLY JOURNAL OF ECONOMICS optimal contract. but s stays the same. k' = k + a in (2. w = wu. Proposition 7 says that the same is true under asymmetric information.2)-(2. Proof. Under asymmetric information. (2) given E > 0. where a > 0. However. IV. Proof.4). How CHANGES IN THE DISTRIBUTION FUNCTION G AFFECT THE PROBABILITY OF UNEMPLOYMENT In the last section we considered comparative statics with respect to a change in the firm's and worker's attitudes to risk. Proposition 6 is useful if we know the distribution function G but do not know the form of the utility functions U and V. Let R . We now analyze the effect of changes in the distribution function G. it appears difficult to say what will happen in general. given G.k) is an optimal contract initially. where a > 0.4 > 0 such <? that if (-V"(7r)/V'(7r)) > a for all ir and (-U"(x)/U'(x)) for all x. PROPOSITION Assume (A1)-(A3). In the case of symmetric information the probability of unemployment = prob(9 < R) is unchanged by this. if (w>. depending on how risk-averse the firm and worker are.D. In particular. (w>.E. problem (2. We consider now what happens if there is an increase in R.4) remains the same.e. &is replaced by (9 + a) and R by (R + a). if U and V are exponential.k + a) will be an optimal contract after the change. Q. the maximum increase in the probability of unemployment relative to the Walrasian case. In particular.

2) subject to (2.. Furthermore.In also multiplied by A.e.e. Q. R. (dP.i.D. Then. we suppose that s and R are multipliedby X > 0. 9. Then we know by Proposition 7 that k increasesby a. if (Wu. Proof. where X > 0. and that w-.b > O.the latterif it is the logarithm .This means that all productionactivities (including the opportunitycost of laborat date 0) are scaled up or down by the same amount. the probabilityof unemploymentfalls. I' <I + a. respectively.the probability of unemploymentstays the same. either W(XAro Air) = = + 0(X)W(iro+ -r). long as 7ro also subject to the scaling effect. Under symmetricinformation. we must have iP'> i. If we combine Propositions 7 and 8. AXR.the probabilityof employment is unchangedby this. Proof. where a > 0./i') are optimal contracts. before and after the change. Furthermore. if ( W>. PROPOSITION Supposethat U(w) exhibits constant relative risk + aversionand that V(ir)can be writtenas W(7ro ir) for all ir for > some 7ro 0. we must have I' < (I + a). (AWuA) is an optimal contractafter the change. Then. where a > 0. See Appendix.I) is an optimalcontractinitially.u).Let s be reand the worker's by U(w) placed by the random variable (&+ a).e.if s < R. Assumethat (A1)-(A3) hold before and after the change. Consider the problem. Proposition 9 says that the same is true under asymmetric information in the case where U exhibits constant relative risk aversionand V(r) . respectively. k falls. or W(XAro XA7r) 0(X) + W(iro+ 7r)(the former holds if W(ro + ir) is a powerfunction. where W also exhibits constant relative risk averand sion. where W exis hibits constant relativerisk aversion. givenby (A3). before and after the change.E. Airo. by Proposition 8. Proof. Now reduceR by a. the probability of unemployment rises.maximize (2. I' > I. ifs < R./).Then if (wi.3).W(ro + ir).Let the firm's utility function be given by V(r) == -e-aw. i. Suppose that s and R both increaseby a. Then.. w. (wtI') are optimal contracts. Let s. + Under constant relative risk aversion. We next considerthe case of a multiplicativeshift. Propositions7 and 8 are concernedwith additive shifts in s and R. AR. b COROLLARY. where a.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 141 be replaced by (R + a).. and irobe replaced by As.-we get immediately the following corollary.

Therefore.R)(1 . k' = Xk. wi = Xwn. The same is true for U. but this will not matter for what follows.142 QUARTERLY JOURNAL OF ECONOMICS function).1) h is the smallest value of k that maximizes (k . In this case.. the only interesting case is where R < SL < SH. the firm cannot pay the worker a positive wage when it does not employ him. It follows that if we substitute R' = XR. An illuminating special case is where s can take on only two values. parts (1) and (2) of Proposition 1 continue to hold (see Remark 1 of Section II). it appears difficult to establish further results about the behavior of k as a function of the distribution function G. k 2 SL. The best way to think of wo is as the output (revenue) of the firm coming from sources other than the worker's production. where (5.E. however. Then under symmetric information we get em- .4) remains the same. In particular. In general. we see that there is no difference between symmetric and asymmetric information. The reason for introducing ro is that constant relative risk aversion utility functions are defined only for nonnegative wealth.D. It follows that when s can take on only two values. the optimal k will be close to k. then Proposition 1(1) tells us that employment will occur if and only if s = sH. V. if iro = 0. 7rL > 0.G(k)). THE CASE WHERE THE FIRM IS EXTREMELY RISK-AVERSE AND THE WORKER IS APPROXIMATELY RISK-NEUTRAL In the last section we obtained some results about how the probability of unemployment is affected by changes in the distribution function G. and xH > 0.2)-(2. It is then quite natural to scale iro by X too. SL with probability WLand SH with probability 7rH. i. In this section we obtain some further results for the case where the firm is extremely risk-averse and the worker is approximately risk-neutral. &' = As. as we noted in Proposition 6 of Section III. where rL + 7rH = 1. problem (2.e. In the next section. Q. This case violates our assumption that G has a density function. If SL ? R < SH. we shall show that some results can be obtained for the case where the firm is extremely risk-averse and the worker is approximately risk-neutral. Since this employment rule is also optimal under symmetric information. V' = XaJ.

i.G(k)) is replaced by prob[9 > k]). G(s) = 7rLfor SL < s < SH. is given as follows for SH > k ' SL = max(SL.R. the firm's profits in the worst state are SL . Except for the discontinuity. respectively.SL) or Wu(SH) < Wu(SL). s = k is always considered an employment state when k > R.a7 + 7rH(SH . (limk+SLWU (k) > wI (SL)).R)). (k) at the point k = SL.-J.aT. It is possible to have either wu(sH) > Wu(SL) = (W + R .e.3) holds.2) U(wu(k) +k-R) U(wu(k)+k-R)7rH+U(w.. We therefore have as a special case of Proposition 6: PROPOSITION10. the two cases yield profit for the firm in the worst state of (SL .)Furthermore. then every optimal contract (-U"(w)/U'(w)) .1).3) holds.(k))7rL =U ifsH>k>SL- As noted in Remark 1 of Section II. The optimal contract involves either employment in both states or unemployment when s = SL.V satisfy (A1)-(A3) and (-V"(7r)/V'(7r)) > a for all ir. in the discrete case. Consider an infinitely risk-averse firm and a risk-neutral worker.R): = UU(i) if k =SL (5. In the first case k = SL and the worker's wage when employed is w + R (see (5. V.3) SL -R < rH(SH R). to set k = SH. Then a necessary and sufficient condition for there to be some UV satisfying (A1)-(A3) for which there is an optimal contract (wuk) with k > SLis that (5. whereas under asymmetric information we may get unemployment when S = SL (note that we shall always have employment when S = sH-see Proposition 1). Suppose that &takes on just two values-SL . wU(k). (k > SLmeans that unemployment occurs in the bad state s = SL. In the second case the worker's wage when unemployed = Wu= H(k -R). as defined in (3. As in Proposition 6 this is also a necessary and sufficient condition for unemployment to occur in the bad state for some utility function U. that is. <? for all w. it is obviously optimal to maximize k. If R < SL < SH. G(s) = 0 for 0 < s < SL.2))./ > 0 such that if U.R) and (. where R < SL < SH. w>(k) is decreasing in k. If the firm is infinitely risk-averse. Note that there is a discontinuity in w.SH. It is easy to show that Proposition 6 continues to hold in the discrete case (where (1 . if (5. G(s) = 1 for s > SH. Thus. from which it follows that the optimal contract will involve unemployment when the firm is infinitely risk-averse and the worker is risk-neutral if (5. there exist a.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 143 ployment in both states.Thus.

3) holds. approximate versions of Propositions 8 and 9 can be established for the many-state case when the firm is extremely risk-averse and the worker is approximately risk-neutral. then (5. since we switch from the regime in which s = SL is an unemployment state to the regime in which there is employment in both states.144 QUARTERLY JOURNAL OF ECONOMICS satisfies k > SL. (k .1).Then we know from Proposition 10 that. In contrast.G(k))+ has two maxi18. In fact.G(k))+ denotes the probability that s > k. while a proportionate change in s and R will lead to the same proportionate change in k. V.R)). given an dition (5.3) will hold for all irL < (SH . after a certain point. the relationship is not monotonic. (B) the lower is the net social return in the bad state relative to that in the good state. In particular.SL)/(SH SL)/(SH . the probability of the bad state. Note that in the limit when 7lL = ((SH ((SH . The probability of unemployment reaches its supremum as rL . Proposition 10 also enables us to determine for what two-point distribution the probability of unemployment will be highest. SL and R are all increased by the same proportion. as shown in Figure II. Similarly. the expected efficiency loss that results from making s = SL an unemployment state is small. This is because by making s = SL an unemployment state the firm is able to reduce wages in this state. then it will continue to hold if R increases. Thus.G(k))+ (see Proposition 6) is maximized at both k = SL and k = SH-18 That (k . For if (5. . and this is very important to a highly risk-averse firm even if the probability of s = SL is low.R)(1 .3) holds and the firm is sufficiently risk-averse and the worker is approximately risk-neutral.3) is unaffected.R)(1 . If 7rL or (SL . an increase in rTL reduces the probability of unemployment. Condition (5. if (5. unemployment will occur for some U. then an increase in R can be shown to reduce k. Suppose that we keep (SH . if k satisfies (5.3) tells us that unemployment is more likely to occur in the bad state (A) the lower is the probability of the bad state. In particular. Clearly. if the firm is extremely riskaverse.R) is small.R) constant and vary rjL. ConR). if SH. (A) and (B) are fairly intuitive. That is. (1 . unemployment will occur when s = SL. the gain from improved risk sharing may be large. (SL . It is clear from Proposition 10 that in the two-state case Propositions 8 and 9 generalize to the case where the firm is extremely risk-averse and the worker is approximately risk-neutral.SL)/(SHextremely risk-averse firm and an approximately risk-neutral worker. if (5. The probability of unemployment will then be rjL.R). we have a relationship between the probability of unemployment and 7rL.R)).3) holds initially.

. This shows that.. letk(7r) maxiri. with probabilities 1. .(k . Note that by regarding s1.R) lsi>k - 0}. under the conditions of Proposition 11. Proof. < * < sn.. then... the probability of unemployment tends to one in the former case. the limiting distribution in which G (k).7r-ir0) (sn -R)nr.INFORMATION145 IMPLICIT CONTRACTS UNDER ASYMMETRIC Probability of unemployment SH-SL SH-R A450 SH-SL 1 7L ~LSHR FIGURE II miners is no coincidence. . ..G (k)-...7r?) = (S3-R)(1 . . while it equals zero in the latter case.r > G(k))+ .R)(1 - = {7rERn 1l ri = 1. if the distribution function G can be chosen arbitrarily....R)(1 . . .R) and in the limit 7r?. < sn. . the maximum probability of unemployment when both ir and the risk aversion of the worker and firm are variable is (Sn . .G(k))+ is constant for k equal to the mass points of s. Let G (x).R). where each si > R.EA (sn - si)/(sn R).. For as s1 R. For each vector of probabilities 7r = imize (h G(k(r))-= (Tro. Suppose that s can take on n fixed values s1 < S2 <. respectively. The next proposition shows the following: if s can take on n values s. Thus. has a supremum equal to (Sn .Si)/(Sn . .irn ) EAn . See Appendix. .si)/(sn .4) (s. there is no limit to the amount of unemployment which can be explained by the implicit contract-asymmetric information model relative to the implicit contract-symmetric information model.. with each si > as R. PROPOSITION 11. If we return to the case where s takes on two values. .R) = (s2 .. Urn.R) as close to one as we like (simply let si R).1). if we regard the wri's variables. rn as variable."achieves" the supremum has the property that (k . . Then sup.R)(1 . we can make (Sn . where k is given by ( the probability that s < x. we can also . In other words. the probability of unemployment. sn as well as 7r1.sV)/(sn .. ir satisfy (5.

R) > SL . R = 0.R.146 QUARTERLY JOURNAL OF ECONOMICS use Proposition 10 to obtain some negative results.R and so if p is not observed. let R = 0. SH = 10. and so this corresponds to a mean preserving spread in the sense of Rothschild-Stiglitz. SL = 6.R when *L is close to one. This suggests that an increase in the riskiness (in the sense of Rothschild-Stiglitz [1970]) of "first-best" profit or net social return max(9 .R. prob[s = SL IP = Pi] = (WTiLWrL)/Xi.Then lrH(SH . it is not true that an improvement in the worker's information about s necessarily leads to an increase in the probability of employment. for each p. In particular. 7TOL= 1/4. We have seen that increases in k allow the firm to transfer some of the risk it is bearing to the worker. Now let SH = (7 . and Xo = X = 1/2. one uses the posterior probabilities prob[s. Then SH > SH. To see this.irL. 71L = 3/4. A contract now is a vector (wu(po).k(pi)). the desire to shed risk will be greater for the firm the riskier is profit when k = R. Using Proposition 10.k(po). since wages can be made to depend on p. and so if the firm is sufficiently risk-averse and the worker is approximately risk-neutral. and 7rL = 1/2.R) > SL . In particular. However. where now. where7rL is close S = SL. Assume now that at date 1 the worker or receives a signal p about s. SL = 4. where 7rOH+7rlH = rOL+lrlL = 1.Then lrOH(SH .R) < SL . and so now there will be full employment for all utility functions U. assume that p = PO P1. For similar reasons.R.7TH = 1/2. V.33/4 rL)/( -lrL). 7T1H = 1/4. the probability of unemployment is zero if the firm is sufficiently risk-averse and the worker is approximately risk-neutral. where Xi = prob[p = Pi= 7riH-H + lriLlrL. and that the conditionalprobabilitythat p = pi given s = sH is lriH and that p = pi given s = SL is 7riL. Ceteris paribus. to compute expected utility. suppose that SH = 10. It follows that if the firm is sufficiently risk-averse and the worker is approximately risk-neutral. . to one. An optimal contract is defined in the obvious way. which is common knowledge.R = wu = the wage in unemployment states). SL <SL.wu(p1). AH(sH . Then 7rH(SH . there will now be unemployment when p = po in the state s = SL. there is unemployment when SL = 33/4. 7rHSH + 7rLSL =7= 7rHSH + 7LsL. the firm is bearing all the risk and the worker none (the wage in employment states = wu + k . respectively. Suppose again that s = SH or SL with probability irH.R) < SL . and 7lH = 7L = 1/2.SH= P = Pi] = (wriHw7H)/Xi. Now suppose that p is observed and 70OH= 3/4. This is not true in general.O) may lead to an increase in k and possibly to an increase in unemployment. we may easily construct examples in which the probability of unemployment rises as a result of the signal p being observed. however. for when k = R.

R)(1 . the wage bill cannot be made an arbitrary function of the true ?. Equation (5.5) will have a unique solution if the hazard rate is increasing. This is because a drop in employment makes the firm bear a cost of decreasing the wage bill that is high in states where &is high and low in states where ? is low. Unfortunately.5) is the reciprocal of the hazard rate of the distribution G. In a world where &is common knowledge. since workers cannot determine ?.5) k-R = (1 - G(k))/G'(k). VI. this need have no implication regarding employment (especially if labor supply is inelastic). When &is not observed by workers. Then k maximizes (k .G). Thus. (5. which yields the firstorder conditions. among other things. optimal risk sharing requires a fall in the firm's wage bill when &is low. For then we move from the case of asymmetric information to the case of symmetric information and the probability of unemployment must fall (or stay the same). the above cannot happen if the signal p gives perfect information about s. Clearly how a change in the distribution function G affects the probability of unemployment will depend. If the firm's owners are riskaverse. This is true for a number of well-known distributions such as the normal (see Barlow and Proschan [1975]).G(k)). Further insight into why increases in risk or improvements in information do not have unambiguous effects on unemployment can be obtained by returning to the general case where there are many states. employment is a good instrument for inducing incentive-compatible risk sharing. However a contract which specifies that the wage bill is an appropriately increasing function of total employment can induce the firm to cut wages only when &is low. on how it affects the hazard rate G'/(1 .IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 147 Of course. We have been concerned with analyzing what determines how much unemployment the asymmetry of information between firms and workers can cause in situations where labor supply is inelastic (over the region where wages exceed the reservation wage). CONCLUSIONS This paper has been concerned with analyzing the implications of the assumption that a worker's marginal revenue product &is observed by the firm but not its workers. The . The right-hand side of (5. increases in riskiness and improvements in informativeness do not generally affect the hazard rate in a simple way.

As we showed in Section III. then our implicit contracts model implies that there will be less (or the same) employment in each state than there would be in a spot market model (with employment being equal in both models in the best state of nature). Note that all our results generalize to the many-worker case if each worker can observe only whether or not he is employed. Since the benefits in terms of risk-sharing are unaffected by the magnitude of (s . Note that once s is unobservable. In Grossman and Hart [1981]. the insurance company unable to observe s would find it in its interest to condition income transfers on the level of employment. The cost of unemployment is that it is ex post inefficient. employment is a useful screening device for any party that the firm wishes to share risk with.148 JOURNAL OF ECONOMICS QUARTERLY benefit of unemployment is that the firm is able to reduce the wage bill in bad states and hence to bear less risk (the worker correspondingly bears more risk). reduces the probability of unemployment." we mean that even though labor supply is inelastic under symmetric information (over the region where wages are larger than the reservation wage R). this suggests that the optimal amount of unemployment will be lower the smaller is the probability mass of states close to R.. Even if the firm uses an insurance company to achieve risk sharing. gross profit s is not common knowledge. It is intuitively clear that the cost of introducing ex post inefficient unemployment in state s is smaller the closer the state is to R.19 Our analysis makes two important assumptions: (1) that the owners or managers of the firm are unwilling to bear all of the profit risk associated with variations in s. By "unemployment will be higher.e. variations in productivity s (in the region where s > R) will lead to variations in employment under asymmetric information.where ax> 0. the wage function can no longer be represented by just two numbers-a wage when employed and a wage when unemployed-as in the one-worker model. we showed that if L can take on any finite number of values. then the owners of the firm could condition income transfers directly on the realization of g. The difficulty in generalizing the comparative statics results of this paper to the case of many workers stems from the fact that. 19. As a result. In this paper we have concentrated on the case where L = 0 or L = 1.R). If s could be observed by the market. which showed that a shift in each s from s to s + ax. in particular. wages can be conditioned on this variable. . the more risk-averse is the firm and the less risk-averse is the worker. if workers can observe the employment level in the firm. benefits from risk sharing will be higher. however. and thus unemployment will be higher. and (2) that the only variable which is common knowledge is employment. i. they are risk-averse. This was made precise in the Corollary to Proposition 8 in Section IV. but not the employment level in the firm as a whole.

and this can be achieved in an incentive compatible manner only if it is accompanied by a decrease in employment. An essential feature of our theory is that there is an asymmetric response to productivity shocks.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 149 The assumption that owners are risk-averse and profits are not observed does not seem unreasonable for the case of small ownermanaged firms. involve the manager holding a significant fraction of the firm's shares. Under this interpretation. when s is high. which means that the manager will be risk-averse with respect to the firm's profit stream. however. will not be a perfect signal about the firm's performance. Thus. and where owners can diversify away risk through the stock market. applies to this case too. Then in order to induce the manager to take the right actions. When s is low. the owners will prevent the manager from diversifying (assuming they can monitor this). There is no moral hazard . it is an endogenous variable like employment. a moral hazard arises with respect to the bad states of nature because it is there that the firm asks workers to take a wage cut (the firm would always claim the state is bad if there was no accompanying employment reduction). Note that under these conditions. since profit depends on unobserved actions of the manager. i. Consider a corporation that is run by a manager who takes actions which the owners cannot observe.e. it will still be the case that the firm will have to create too much unemployment (from an efficiency standpoint) in bad states in order to show workers that times really are bad. Thus.. However. under certain assumptions our underemployment result can be shown to generalize. the manager will have no incentive to manage the corporation well and the incentive scheme will be self-defeating. profit)-such an incentive scheme may. even if observed. It is less reasonable as a description of public corporations that publish profit statements and pay dividends. the moral hazard problems discussed in this paper remain. As a result. with some modifications.e. for example.g. the owners will give the manager an incentive scheme that makes the manager's utility depend positively on the firm's performance (e. ex post profit. the firm can always increase wages to get more employment and better risk sharing.. In particular. i. the owners and the manager are in a principal-agent relationship. while it can be shown that in an optimal contract wages will now be conditioned on profits as well as on employment. Obviously if the manager can then diversify away risk by selling off these shares.. A formal model along these lines is developed in Hart [1983]. To put it differently. optimal risk sharing implies that there should be a cut in the wage bill. We wish to argue. V refers to the risk aversion of the manager. that our model.

Lucas [1972] proposed a model where short-run labor supply depends on the expected interest rate (which in his model is the real return to holding money-the only asset in the model). there are versions of Lucas' model where the driving shocks are real (not nominal) and the inefficiency in output also depends on expectational errors. Though Lucas [1972] and Barro [1976] consider models where "inefficient" unemployment occurs because of monetary nonneutrality. The fact that in our model bad states create non-Walrasian unemployment but good states do not create non-Walrasian employment suggests a method for distinguishing our theory from some other asymmetric information theories of employment. In linearized versions of Lucas' [1972] model such as Barro [1976].E [x IInfo]. and this leads to labor supply fluctuations. while bad states lead to non. Good states lead to Walrasian levels of unemployment.20 In all of these models this is a symmetric variable. there is no reason (or emphasis in the logic of the theory) for L (-) to respond in an asymmetric way: agents make "up side" and "down side" forecast errors. As a consequence of this. In all of these Lucas-like models. Further. the above asymmetry implies that relative demand shocks decrease employment by more than they would under complete information. Workers in a particular submarket have incomplete information about the real interest rate. In our model this is not the case. and Maskin [1982] develop a general equilibrium model of implicit contracts in order to explore the above asymmetric effect. For example. Grossman and Weiss [1982] considered a model where there are real productivity shocks and employment is a linear function of the difference between the observed productivity r and the anticipated real opportunity cost of investment: L(r . Grossman. In particular. Hart.P). It is at the very essence of the model that labor supply response is asymmetric. In a model of relative demand shocks across industries. we were able to show that the level of employment under asymmetric information equals the Walrasian level in the best state of nature.Walrasian levels of unemployment.150 JOURNAL OF ECONOMICS QUARTERLY with respect to the best state of nature because the firm would never claim that times are very good and give workers an increase in pay if this were not truly the case. Workers in all submarkets make correlated errors in forecasting interest rates because they all condition their forecasts on the observed price level in their submarket which is in turn affected by the economy-wide money supply.E[r Iinfo]). . the driving variable that converts long-run inelastic labor supply into a short-run elastic labor supply is a variable like x . labor supply is a linear function of the difference between the expected economy-wide price level and the actual price level: L (E [PI Info] . A model where the relative demand shocks are caused by wealth re20.

Let F2 satisfy the single crossing property with respect to F1.. and hence there are more employment states under k'. X2 is riskier than X1 in the sense of Diamond-Stiglitz [1974]. Q. respectively. kI'< possible to have U < w'U.) on the whole real line.k. But this means that the firm's net profit max(s . That is. by continuity there exists X such that Egx-x(U(XD)) = Eg-x(U(X2)). Consider a person more risk-averse than person U in the sense that the new person's utility function is g(U(x)). the individual with utility function gl(U) must prefer X1 to X2. that under (wuik. that is.E. We do not require that EX1 = EX2.IMPLICIT CONTRACTS UNDER ASYMMETRIC INFORMATION 151 distributions associated with unanticipated movements in the price level leads to implications that are very different from those of Lucas-like models of the same phenomenon. and that a risk-averse person with U'(x) > 0 weakly prefers X1 to X2. wd <iWU <iWU+ k' <Wu+ k. Proof of Proposition 2: Step 2.) the firm pays less both when the worker is employed and when he is unemployed. define F1(. Therefore. Suppose that Eg(U(X1)) < Eg( U(X2)). in view of the single crossing property.wu) is lower for small s and higher for W- .. Note first that it is im+ + kI. defined bygx (U) = Xg(U) + (1 . Let Xi(s).which means since then wU./') to (Wuk). where g(-) is a strictly increasing.D. Hence wu > dU.i.wu . We say that F2 satisfies the single crossing property with respect to F1 (written "F2 scp F1'")if there exists an x such that < 0 for x < x0 F1(x)-F2(x) F1(x) .e. 0 < X < 1. The distribution function of s induces distribution functions on X1 and X2. and assume that F2(x) # Fi(x) for some x. X2(s) be real-valued functions of s. strictly concave function. (W/k) is never chosen. However. Consider the class of concave transforms gx (-). A similar argumentshows that Wu+ k' < wu + k. We shall sometimes say that X1 is less risky than X2 in the above case.X)U for all U. Contradiction. Suppose that k' < k.F2(x) > 0 for x > x0.e. For convenience. LEMMA 1. APPENDIX Proof of Proposition 2: Step 1. the individual with utility function g-X(U) is indifferent between X1 and X2. Then Eg(U(X1)) > Eg(U(92))Proof. Our proof will involve the use of the following lemma. EU(X1) > EU(X2). i. say F1(-) and F2(-). But this means that whatever V is the firm will prefer (Wu. Suppose that Xl(s) and X2(s) are fuietions of s. . It follows that. Eg(U(Xl)) > Eg(U(X2)). Since Eg1(U(X)) < Eg1(U(X2)) and Ego(L(X1)) > Ego(U(X2)). since an individual with utility function g1(U) is more risk-averse than an individual with utility function g-x (U).

Let I.EV(I1)je+a2 =-E =-Ee-a(I2-2) + [e-(I2'2) b I2 - e-(I1'2)] e-a(I1-I2) dF + 1 e-a(I1-I2) dF.E. a firm with utility function -e-at prefers the profit stream with the higher value in the worst state. and so k = k* does not satisfy the first-order conditions for problem (2. - . w- k'.2)-(2. we may conclude that wu(k*) < w (k (a)) < wu (k (a)) + k (a) < wu (k*) + k*. Let Ii infix Iprob(Ii < x) LEMMA2. {EV(12) . EV(11) < EV(I2). Then. again by the proof of Proposition 2. if a is sufficiently large. -wu) is negative for s < some s* and positive for s > s*. U A Proof. Q. Q. -K u) satisfies the single crossing property with respect to the distribution of max(s By Lemma 1.D. Q. for a large. Proof of Proposition 3: Step 2. Suppose that I < 12. By definition.5) > 0. This proves that k' > k. since otherwise k (a) would not be preferred to k*. -w4) - max(s - .Ik) and a less risk-averse firm (weakly) prefers (i.E. i. kI') to (i . the firm's profit stream when k = k (a) will satisfy the single crossing property relative to the profit stream when k = k*. As a goes to infinity. . Now dwu/dk = -RHS of (2. To establish the last part of the proposition.D. the left-hand side of the expression is positive./k) to (KIk). a firm with utility function V prefers k = k (a) to k*.5) = 0 at k = k*.. Therefore. We show that if a person has constant absolute risk aversion of a sufficiently large size.I). To prove (2). for a sufficiently large. It follows that the distribution of max(s . But LHS of (2. suppose that it is false. Hence k < k*. Then we can find a sequence of optimal k (a) converging to k < k* as a a) . Using the argument of the proof of Proposition 2. one shows that k' = k violates the first-order conditions (2. then he will always choose the gamble with the higher minimum value. We shall need the following lemma.4). But this contradicts Lemma 2. Note that as a a ). Suppose that V(x) = -e > 0 for i 1.e.152 QUARTERLY JOURNAL OF ECONOMICS large s under (dIP) u(s) max(s - than under (&. Hence. and I2 be two functions of s that are bounded above by a number b. Proof of Proposition 3: Step 1.W'k'.E.2. Wu (a)) (k wu (k) > wu(k*).D. the first two terms are bounded. which says that. It follows from Lemma 1 that a firm with utility function -e-a will prefer k = k (a) to k = k*. this contradicts the fact that a more risk-averse firm (weakly) prefers (A. We have shown in Section II that k < k*. while the third term goes to positive infinity.5).- k.

e. Assume that k' > A. i.(k) < tu (k)+ k < Cvuk') + A'. It follows that A' ? k. Consider (2.2) yields an objective function for the firm of the form U(-b/a)4. When a is large..E.(k). Then the worker's net income is w (k) < w if unemployed and wu(k) + k . In the exponential case. If k = s > R. If k = R. That A'< k if s < R follows from the fact that A'= k is not consistent with the first-order conditions (2. this can be written as e-alwuq(k) = U. The proof breaks down because V changes when the worker becomes more risk-averse and the optimal k is not independent of V. the workeris risk-neutral.. it is enough to establish that k* < k when U is nonlinear. One proceeds as in the proof of Proposition 3.(k ') < w. But this means that Lemma 2). It follows that the optimal k is independent of U in the exponential case.s + R satisfies (2.e.. Note first that k is by construction the smallest value of k that minimizes wu(k).Use the equation.5).R with probability (1 . WU + k .1). to express wUas a function of k: wU = vu(k). Therefore. the firm pays in the limit hi when the worker is unemployed and more than (h + R) when the worker is employed. defined in (3.3) for all utility functions U. then wu = W. Suppose that k* > k. Substituting in (2. wu + k . the worker is very risk-averse.R). i. where 0 is a function of k. Then wU(k*) < w. i.IMPLICIT CONTRACTSUNDER ASYMMETRICINFORMATION153 Proof of Proposition 4.R if employed. This proves that k(a) . wu (k. (k). The above argument shows that in the exponential case the optimal k will be independent of V.3) for all utility functions U. Proof of Proposition 6. for the case where U is linear. The next point to notice is that we can equally well think of an optimal contract as maximizing the worker's expected utility subject to the firm's expected utility being equal to some V (V of course depends on U). increase a. Q.R when employed) is riskier in the sense of the single crossing property when k = k' than when k = k. Clearly the firm could do better by setting k = max(s. Q.e. wu(k) must be close to w-for the worker to (k) be indifferent between the lottery (wu(k) with probability G (k).R). In both cases the worker'snet income is ff with probability 1. by Lemma 1. In view of Proposition 3.E. to prove part (1). This contradicts the fact that the worker prefers k to k' when a = ao and A'to k when a = a. Take k > max(s.3).max(s. then wu = w-satisfies (2. Proof of Proposition 5. (A.D. where k* is defined in Proposition 3. Hence fix V and let the worker become more risk-averse.D. Hence k' > Aimplies that a. the worker's net income (wu when unemployed..R). where wu refers to the nonlinear utility .R) as a -a co. It is easy to see that WU is decreasing in k and (wU+ k) is increasing in k.1) f V(s-kw-u) dGJ+ V(-wU) dG = V.G(k))) and w.with certainty (see o a. In particular.a) -*wh as a if k is bounded away from max(s. Proposition 4 does not appear to hold for general utility functions.

R). (k*) + k* . .R with probability (1 .G (k*)) > (k . As in the proof of Proposition 4. To show that the probability of unemployment cannot exceed (Sn .G (k))). it follows that w.R). This shows that employment will (R+ a). it follows that OEU/IR is larger the larger is k (since (ith Ck) + k) is increasing in k). Note that by Proposition 1(1) we always have employment when s = Sn.R).154 QUARTERLY JOURNAL OF ECONOMICS function U. This contradicts the definition of k.Si)/(Sn . Suppose that k' < k. note that if the worker is risk-neutral. Therefore.7r?) = (Sn .2) with respect to R. Proposition 3(2J tells us that there exists a such that (-V"(w )/V'(ir)) > a ==k .G(k*))) and the lottery L2 = (wa (k) with probability G(k). Since the probability of employment is lower under L1.5).R.3) (R. From this. But this contradicts the fact that I' is optimal under This proves that I' ' I.. for the probability of unemployment to exceed (Sn .S)/(Sn . But this means that L1 is riskier in the sense of Diamond-Stiglitz [1974] than L2.R)(1 = G(k)) is greater at k = s1 than at k = Sn. To prove (2). and so a risk-neutral worker will prefer L1 to L2. The latter constraint may be used to solve for wu = Jv (k). Proof of Proposition 11. suppose that it does. i. Finally.ik) ' EU(R. But this means that a risk-averse worker with utility function U is indifferent between the lottery L1 = (w.R)(1 -G(k*)) > wu(k)G(k) + (wu(k) + k - R)(1 -G(k)) which implies that (k* .R). By choosing 7rclose to the 7r. I' = fIviolates the first-order conditions (2. w (k) + k . Hence we may write the worker's expected utility as (A. (k*) with probability G(k*). Q. we must have -rn < (s .E.k) = U('u(k))G(k) + U(wu(k) + k - R)(1 - G(k)). Since we know that EU(R. since (k . k < sn-1. Thus.G (k)).R)(1 .which solves (5.G(k))+ is maximized at k = Sn.k I < e.R). where Cvu is decreasing in k and (vu (k) + k) is increasing (k) in k. Differentiate (A. Thus. wu(k*)G(k*) + (wu(k*) + k*.Si)/(Sn .D. the probability of unemployment can be made arbitrarily close to (1 . Hence k < Sn.R > wu(k) + k .G(k)) cannot reach a maximum except at a mass point. This yields (A.2) EU(R.( . it follows that EU(R + a.k) __5_R_ = -U'(vu (k) + k - R)(1 -G (k)).k) > EU(R + ak') given that k > k'.k') (I is initially optimal). That the same is true if the worker is approximately riskneutral follows from a continuity argument. Proof of Proposition 8.4). But then (k . weLk*) + k* -R with probability I .R)(1 . we can ensure that (k .e.R)/(sn . we may think of the optimal contract as maximizing the worker's expected utility subject to the firm's expected utility being at least V. since wu(k*) < wu(k).R)(1 .

E." this Journal. Grossman. 185-216. Barro. 1159-60. P. R.R) < (s . Maskin." Journal of Monetary Economics. and Winston..." mimeo. R.. 1979). J. Inc." Review of Economic Studies." mimeo. XCVIII (1983). V. 321-38. V (1977). 1979. II (1976). Supplement. XCVIII (1983)." this Journal. and H." Working Paper #79-17. Supplement." Review of Economic Studies." American Economic Review. III (1976).. "An Annotation of 'Implicit Contracts and Underemployment. But this means that (k . O. R." Journal of Economic Theory. Supplement. and E. Green.and D. and so k < Sn-2." Journal of Monetary Economics.. --. the probability of unemployment is zero. Statistical Theory of Reliability and Life Testing (New York: Holt.R)/(s . G.. 301-07. "Employment with Asymmetric Information. "Wage Employment Contracts.. "Optimal Labour Contracts Under Asymmetric Information: An Introduction. "Recent Developments in the Theory of Inflation and Unemployment. Grossman. 1980). Diamond.. Gordon. "Implicit Contracts. "Randomization in the Principal-Agent Problem. 157-72. and 0. Hence k < Sn-1." this Journal. Supplement. .. UNIVERSITY OF CHICAGO LONDON SCHOOL OF ECONOMICS REFERENCES Akerlof.INFORMATION 155 IMPLICIT CONTRACTS UNDER ASYMMETRIC occur when s = sn-1 as well as when s = s. "Implicit Contracts and Related Topics: A Survey. Maskin. Journal of Political Economy. and L.R)/(sn. 1981. 696-727..D. 1982. and F. IV (1972). LXXI (1981). Hart. and Unemployment. E. Lucas. XC (1982). 107-22.. and E. 160-68. Stiglitz.R).. Thus. R. 1979). Maskin. "Employment Contingent Wage Contracts. But this means that employment occurs in all states. P.. Kahn. LXIX (Dec. Proschan.. Barlow. J. "Employment Fluctuations and Wage Rigidity. "Unemployment with Observable Aggregate Shocks. 337-60." Brookings Papers of Economic Activity (1980).e. Bryant. 173-87. "The Implicit Contract Theory of Unemployment Meets the Wage Bill Argument. we must have Wn-l + r. Lilien. VIII (1974).. Hall.13-34. which is a contradiction." Journal of Economic Theory. R." Review of Economic Studies... .' "Journal of Political Economy. Hammond. S. "Rational Expectations and the Role of Monetary Policy. S. Moral Hazard.. . "Involuntary Unemployment and Implicit Contracts.R). Calvo. . S.. "Increases in Risk and Risk Aversion. "Efficient Wage Bargains Under Uncertain Supply and Demand. < (si . LXXXVI (1978). Grossman. Continuing this argument yields k = sl. C.G(k)) is greater at k = s1 than at k = Sn1.. Hart.. i. and C." Journal of Monetary Economics.. Hart. XCVIII (1983). "Heterogeneous Information and the Theory of the Business Cycle. 868-79. Chari. 1983)." American Economic Review. 91-124. L (Jan. Q. and J. 1975). 185-219. 1-32.R)(1 . XLVI (April. "The Implementation of Social Choice Rules: Some General Results on Incentive Compatibility. Phelps. University of Pennsylvania. Miyazaki. Weiss. 0. Azariadis. "Expectations and the Neutrality of Money. Dasgupta. G. and E. XLVII (Jan. for the probability of unemployment to exceed (s. P. Rinehart.

XLVII (Jan.156 QUARTERLY JOURNAL OF ECONOMICS Mirrlees." Review of Economic Studies. Myerson. 225-43. "Incentive Compatibility and the Bargaining Problem.. M. "On Income Distribution. and J. "Increasing Risk. XXXVIII (April 1971).. E.. Radner. . 261-67. R. Incentive Effects. J. "Monitoring Cooperative Agreements in a Repeated Principal-Agent Relationship. 61-74." Review of Economic Studies. 1127-48. Rothschild. "Efficient Incentive Contracts. Sadka... and Optimal Income Taxation. II (1970). M. Stiglitz. 1978)." Econometrica." this Journal. "An Exploration in the Theory of Optimum Income Taxation. Weitzman.. R. 175-208. XCIV (June 1980). XLIII (1976)." Journal of Economic Theory. 719-730." Econometrica. II (1981).