Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
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 support@jstor.org.
RAND Corporation is collaborating with JSTOR to digitize, preserve and extend access to The Bell Journal of
Economics.
http://www.jstor.org
1. Introduction
* The economictheoryof agencyprovidesa frameworkfor studyingsituationsin which
a personor a groupof people delegatesthe selectionof an action to anotherindividual.
Much of the workin agencytheoryhas been done in a single-periodcontext (Ross, 1973,
1974; Harrisand Raviv, 1978, 1979; Mirrlees,1974, 1976; Holmstrom, 1979; Shavell,
1979;Grossmanand Hart, 1983). Little is known about the extent to which the insights
and conclusionswhichhavebeenderivedin single-periodsettingsgeneralizeto multiperiod
relationships.
The purposeof this articleis to take a preliminarystep in extendingagency theory
to multiperiodrelationships.Our analysis differs in several respects from multiperiod
agency problems studied elsewhere.Radner (1980) and Rubinstein and Yaari (1980)
examine infinite horizon models in which the same single-periodsituation is repeated
over time. In addition,they do not allow utilitiesto be discountedover time. In contrast,
we analyzeagencyrelationshipswhich last (a maximum of) T periods,where T is finite.
Our productionfunctionsare requiredto be separableover time; however,we do allow
the productionfunctionsto change in a deterministicfashion over time. We also allow
the principaland the agent to discount their utilities.
Radner(1981) considersa model in which the same one-periodsituationis repeated
a finite numberof times. He analyzes strategieswhich form an "approximate"equilib-
rium, called an epsilon equilibrium.He shows that if the principaland the agent do not
discount their utilities and if T is large enough, these strategiesallow them to come
arbitrarilyclose to the infinitehorizonresults.We discussthe infinitehorizonresults,and
comparethem with our own in Section 4.
Our approachis to derive optimal equilibriumstrategiesin a game in which a type
of binding precommitmentis allowed. In particular,we examine the role of long-term
contractsin controllingmoralhazardproblems.By offeringthe agenta long-termcontract,
the employerhas a mechanismfor precommittingto a strategythat is feasible,legal, and
bindingwithin our institutionalframework.We initiallyassume that the agent also pre-
commits to the long-termcontract by guaranteeingto remain with the firm for all T
periods.We derive the optimal long-termcontract, and we find that the agent's com-
* Northwestern University.
I would like to thank Amin Amershi, Rick Antle, Joel Demski, Mark Wolfson, an anonymous referee,
and the Editorial Board for their helpful comments. The financial support of the Accounting Research Center
of the J.L. Kellogg Graduate School of Management is gratefully acknowledged.
441
2. The model
* To keep the analysissimple, we shall restrictourselvesto a two-periodmodel. At the
end of Section 3, we shall indicatehow to extend the analysisto more than two periods.
In each period the agent selects an action, a, E A,. This action, in conjunction with a
randomstate of nature,producesa cash flow, x,. We assume that X,, the set of possible
cash flows in periodt, is a finite set. The cash flow in each periodcan be representedas
a randomvariablewhose distributiondepends on the actions taken by the agent. Let p,
denote the probabilityfunction for the period t cash flow. We assume that the state
variablesare independentlydistributedover time, and that effort in one period has no
effecton the cash flow in any other period. We can write
remainwith the firm for both periods.Figure 1 illustratesthe sequenceof choices in the
model.
The principaland the agent are assumedto possess utility functions which are ad-
2
FIGURE1
TIME LINEFOR CHOICES
PERIOD1 PERIOD2
Note that since S2 may depend on x,, the agent'schoice of a2 may also depend on
the first-periodoutcome. That is, the function s2(x2)may be parameterizedby the first-
periodoutcome. If this is the case, the agent is facinga differentsecond-periodcontract
for differentx,'s. A differentsecond-periodaction may thereforebe optimal for different
first-periodoutcomes. Also note that since a, affectsthe distributionof x,, it may affect
the distribution(as assessedat the beginningof period one) of which a2 will be selected.
Since the argmaxconditions as listed in constraints(2) and (3) above are difficult
to work with, we make the simplifyingassumptionthat the agent's choice of effort in
eachperiodcan be representedby his first-orderconditionon effort.'Assumingan interior
solution for a, and a2(x,) for all xl, the agent will choose his effortto satisfy:
'Grossman and Hart (1983) discuss problems associated with modeling the agent's actions by using his first-
order conditions. It is possible to derive restrictions on the distribution functions to ensure the concavity of the
agent's expected utility with respect to his effort. For example, Lambert ( 198 1) shows the if the distribution satisfies
the monotone likelihood ratio property and the cumulative distribution function is a concave function of the
agent's effort, the argmax constraints in equations (2) and (3) can be replaced by the corresponding first-order
conditions. The results of the article also go through using Grossman and Hart's approach to modeling the agent's
action choice.
z U2[s2(x1,x2]p'2(x2la2(x1))
- [a2(XI)] = 0 for each xi (4)
E {UI[sI(xi)] + EH2[s2(x1, x2), a2(x1)]}p' (x,1la,)- V(a,) = 0. (5)
Equation(4) is analogousto the agent'sfirst-orderconditionon effortin a one-period
model. The agent selects his effort so that its effect on his expected marginalutility of
income is equal to his marginaldisutilityof effort.Equation(5) is similarto (4) except
that the agent must also considerthe effect that a, has (throughits effect on x,) on his
second-periodeffortand second-periodincome.
Assumingwe can representthe agent's choice of effort by the correspondingfirst-
order conditions,we now replaceconstraint(2) in the principal'sproblemby equation
(4) and constraint(3) by (5). We furtherassume that we can representthe principal's
problemby using a Lagrangianformulation.2If we let X be the Lagrangemultiplieron
the expectedutility constraint,,u, be the Lagrangemultiplieron the agent's first-period
effort constraint, and A^2(xl)be the multipliers associated with the second-period effort
constraints,the Lagrangianis
L[sl(xl,) SOXI, X2), a,, a0(X), X, tl, A'2(X,)]
= z {G,[xi - s,(x,)] + z GX-s2(x1, x2)]p2(x2la2(xI))}p1(x,la1)
+ X{ { U1[s1(x1)] + z U2[s2(x1, X2)]p2(X21a2(X1)) - V2[a2(x1)]}p1(x1la,) - V1(a1) - 0}
+ I{ Ul [Is (X1)] + z U2[s2(xI, X2)]p2(X21a2(X1)) - V2[a2(x1)]}p',(xIla,) - V,(a,)}
+ z 82(X,){X U2[s2(XI, x2)]p2(x2a2(x1l)) -V'[a2(X )] }
2 See Grossman and Hart (1983) for a discussion of modeling issues and an examination of the existence
of a solution in a single-period agency model.
p'1(xllal)
I A
______
AX
Note that if we define X2(xI)= X + , I and82(X,) = we can wnte
p1(xlIa)nl L(l) -p,(xIa,)'
the second-periodsharingrule as:
s2(x1,X2)]= X2(X()+ Ap'2(X 2I 2(x1))
- (7a)
X2)]
U'2[S2(XI, P(X21a0(x))
subjectto
E U,[s,(x,)]p,(x,1a,)- V,(a,)> II,
a, E argmaxz Ut[s,(x,)]p,(x,la,)- V,(a,).
Proposition3: ,u > 0.
This propositionimpliesthatthe incentiveproblemassociatedwith the agent'schoice
of first-periodeffortis not eliminated.Therefore,the principal'sabilityto use both first-
and second-periodrewardsand penaltiesdoes not lead to the "full ex post settling up"
phenomenon discussedby Fama (1980). Using equation (7), ,u > 0 implies that the
second-periodsharingrule willdependon the first-periodoutcome.In addition,ifp,(x,Ia,)
satisfiesthe monotone likelihoodratio property,it is easy to show that
s, is an increasingfunctionof xl,
EH2[s2(x1, x2), a2(x1)]is an increasingfunction of xl, and
EG2[s2(xI, x2), a2(x1)]is a decreasingfunction of x,.
Note that the contractensuresthat the second-periodincentivesreinforcethe first-period
incentivesin motivatingthe agent'sfirst-periodeffort.In otherwords,both his first-period
compensationand his expectedsecond-periodutilityare increasingfunctionsof the first-
period cash flow.
It is relativelyeasyto extendthis modelto morethantwo periods.Usinga formulation
analogousto problem(1), let Xbe the Lagrangemultiplieron the constraintthat the agent
receiveat least 0 over the T-periodhorizon. Let u,(X,_,)be the Lagrangemultiplieron
the agent's choice of effort in period t when the prior sequence of cash flows is
x,-I = (xi, ..., x,_,). It is easy to show that the optimal sharing rule in period t will
satisfy(assumingan interiorsolution):
G'[x,- S,(x-1, xi)] ')= a,(x-i) + M(X( )
where
A,x,_I-) = A + M(1)p'j(xjIaj(xj-1))
Aj(xj-l) IjXa(',
pi(x11a1(x1,))'
and
H pj(xjlaj(xj-,))
j=,
We can also show that u,(X,_,) > 0 for all X,-l. Just as in the two-period analysis, the
principaluses both current-periodand future-periodincentives to motivate the agent's
effortin each period.
4. Interpretation of results
N In a single-period model, the agency problem exists because the uncertainty prevents
the principalfrom using the cash flow to determineunambiguouslythe agent's action.
Holmstrom(1979) suggeststhat "when the same situation repeatsitself over time, the
effectsof uncertaintytend to be reducedand dysfunctionalbehavioris more accurately
revealed,thus alleviatingthe problem of moral hazard."Under these conditions, if the
agentselectsthe first-bestlevel of effortin each period,the cash flow will be independent
and identicallydistributedover time. As the number of periodsincreases,the variance
of the agent'saverageoutput, if he selectsthe first-bestlevel of effortin each period,gets
smaller.Note that for this diversificationeffectto occur, it is necessarythat the principal
evaluatethe agent'seffortover the entirehistoryof his employment,ratherthan evaluate
each period'sperformanceseparately.
Radner(1980) and Rubinsteinand Yaari(1980) considerthe extremecase in which
therearean infinitenumberof observations.They show that the principalcan eventually
T T7
Let Hf =, Hi. We thereforehave Hf = H, + H,+,. In a manneranalogousto the two-
3 They also demonstratethat the principalcan providethe agentwith a marginof safetyto give him the
benefitof the doubt in case he has been selectingthe correctactionsbut has been experiencingbad luck.
4 To see this, supposethe agentis requiredto selectthe same amountof effortin each period.In this case,
it is optimalto evaluatethe agenton the basisof his averageperformanceif the likelihoodfunctionis a linear
functionof x,.
S If the productionfunctionvariesstochastically over time, theremay be some circumstancesunderwhich
it is optimal to abandonthe firm or hire a different"type"of agent, one with a differentset of skills, etc.
Similarly,if the agent'stastes or skills vary stochasticallyover time, situationsmay arise under which it is
optimalfor the agent to leave the firm.These considerationswill tend to counteractthe value of a long-term
contractwhich binds the agentto the firm.
period analysis, we can show that F/(H/)> F1(H1)+ F;+i(H/+i) MovingF1(H1)to the left-
hand side and subtracting I F1(H1)from both sides yield
) =1+ 1
7 7
5. No commitment
* To this point we have assumed that both the principaland the agent precommitto
a long-termcontract.In many cases it is not feasibleor legal for the agent to commit to
remainingwith the firm. If the agent findsa more attractiveemploymentopportunityin
a later period, he is free to leave the firm to take it. In this section we incorporatethis
into the analysisby relaxingthe assumptionthat the agent is committed to remaining
with the firm for both periods.
The principalis assumedto be committedto any two-periodcontractthat he offers;
however,the agent has the option to leave the firm after the firstperiod. To model this
situationwe must make an assumptionregardingthe utility that the agent can achieve
in the second period.For convenience,we assumethat 02, the amount of utility that the
agent can obtain, is independentof xl.6 The agent'sthreat of leaving places additional
constraintson a feasible two-period contract. In addition to satisfying H(s, a) 2 0
= 01 + 02, a feasibletwo-periodcontractmust also satisfyE[H2[s2(x1,x2), a2(x1)11x11
2 02
for all xl.
We can replicatethe analysis in Section 3 to show that the optimal contract will
satisfy equations (6) and (7a), where now X2(xI) = max X + 41 P'l(l I) X2), and
-A2 is the slope of the 12 frontierat the point wherethe agent'sexpectedsecond-period
utilityis 02. Recallthat the agent'sexpectedsecond-periodutilityis an increasingfunction
of X2(xI). Essentially,the external labor market provides the agent with insuranceby
limitingthe amount that the principalcan penalizethe agent for a bad first-periodper-
formance.No matterhow bad his first-periodperformancehas been, the agent is guar-
anteedan expectedutility of at least 02 in the second period. The principalmust match
this to keep the agent from leaving.7
Becausethe penaltiesthat can be imposed upon the agent have been limited, it will
generallybe the case that the principalcannot motivate the agent'sfirst-periodeffortso
efficientlyas he could when the agent precommittedto a two-periodcontract.This will
lead to a welfareloss. However, note that it is still optimal to rewardthe agent in the
second period for a good first-periodperformance.Therefore,the agent'ssecond-period
compensationwill still depend on both his first-periodand second-periodperformance.
To lose the dependenceof s2 on xl, it is necessarythat neitherthe principalnor the
agent precommitto a long-termcontract,and that the externallabor marketnot adjust
the utilityvalue of the agent'soutsideemploymentopportunitiesas a functionof his first-
period performance.Under these circumstances,the multiperiodprincipal-agentrela-
tionship decomposesinto a sequenceof one-periodgames. In each period the principal
and the agent act exactly as they would if that were the only period in the game. The
agentwill receivean expectedutilityof 02 in the second periodregardlessof what happens
in the firstperiod.
6. Concluding remarks
* In a dynamic model with incentiveproblemswe have demonstratedthat the agent's
second-periodcompensationwill dependon his first-periodperformance.This allowsthe
principalto diversifyaway some of the uncertaintysurroundingthe agent's actions. In
addition,this allows the principalto smooth the agent'sincome over time by spreading
the riskof the first-periodoutcome over both periods.If a capitalmarketfor transferring
wealthovertime exists,the principaland the agentmay be ableto deferthis latterfunction
of the long-termcontractto the capital market.It seems difficultto fully incorporatea
capitalmarketinto the analysis.The primaryproblemis that the agent'seffortstrategy
will not be separablefromhis savingsdecisions.Forexample,the agent'schoice of second-
period effortwill be a decreasingfunction of the amount he has saved, ceteris paribus.
In this case, the cross partialderivativesbetween effort and saving become important,
and this considerablycomplicatesthe signingof the Lagrangemultiplierson the agent's
effort.
There are many interestingaspectsof multiperiodagency relationshipswhich have
not been addressedhere.We consideredsituationsin which the productionfunctionsare
6
The analysisalso holds if 02 is a randomvariablewhose distributionis independentof xi.
7Alternatively, supposethe agent leaves the firm and receivesan expectedutility of
02 elsewhere.The
principalmust then hire an equivalentagentand providethe new agentwith an expectedutilityof at least82.
Appendix
* Proofsof Propositions2 and 3 follow.
Proofof Proposition2: Since the left-handside of (6) is greaterthan zero, so is the right-
hand side. This implies that X2(xI) is greaterthan zero. Let S2f(X2) be the sharingrule
which satisfiesG'[x2 - s2f(x2)] = X2(X,)U'2[S2f(X2)].
By using the same proof as in Holmstrom(1979), ,A2(x0)< 0 implies
- s2(x, x2)]p'2(x2la2(x1))
Z G2[X2 > 0,
2 Z G2[x2- s2f(x2Dp'2(x2Ia2(x1))
Similarly,for all x, E X-, we have EGAs2(x1, x2), a2(x1)] < r2x. Multiplyingboth
sides by p'1(xla,) < 0 gives
EGA[S(XI, x2), a2(X,)]P'(x,|aj) 2 1'2XP'1(xj |aj)
{G,G[xl
z - s,(xj)] + EGAs2(x1, x2)a2(x1)] > 0.
}?P'(xjla1)
which contradictsAl < 0 (by using equation(8)). Q.E.D.
References
FAMA,E. "Agency Problems and the Theory of the Firm." Journal of Political Economy (April 1980), pp.
288-307.
GROSSMAN,S. AND HART, 0. "An Analysis of the Principal-Agent Model." Econometrica (January 1983),
pp. 7-45.
HARRIS,M. AND HOLMSTROM, B. "A Theory of Wage Dynamics." Review of Economic Studies (July 1982),
pp. 315-333.
ANDRAVIV,A. "Some Results on Incentive Contracts with Applications to Education, Health Insurance,
and Law Enforcement." American Economic Review (March 1978), pp. 20-30.
AND . "Optimal Incentive Contracts with Imperfect Information." Journal of Economic Theory
(December 1979), pp. 231-255.
HOLMSTROM, B. "Moral Hazard and Observability." Bell Journal of Economics (Spring 1979), pp. 74-91.
LAMBERT, R. "Managerial Incentives in Multiperiod Agency Relationships." Unpublished Ph.D. Dissertation,
Stanford University, 1981.
MILGROM,P. "Good News and Bad News: Representation Theorems and Applications." Bell Journal of Eco-
nomics (Autumn 1981), pp. 380-391.
MIRRLEES, J. "Notes on Welfare Economics, Information, and Uncertainty" in Balch, McFadden, and Wu,
eds., Essays on Economic Behavior under Uncertainty, Amsterdam: North-Holland, 1974, pp. 243-261.
. "The Optimal Structure of Incentives and Authority within an Organization." Bell Journal of Economics
(Spring 1976), pp. 105-131.
RADNER, R. "Does Decentralization Promote Wasteful Conflict?" Bell Laboratories Economic Discussion Paper,
1980.
. "Monitoring Cooperative Agreements in a Repeated Principal-Agent Relationship." Econometrica
(September 1981), pp. 1127-1148.
Ross, S. "The Economic Theory of Agency: The Principal's Problem." American Economic Review (March
1973), pp. 134-139.
. "On the Economic Theory of Agency and the Principle of Similarity." Proceedings of the NBER-NSF
Conference on Decision-Making and Uncertainty, 1974.
RUBINSTEIN, A. AND YAARI, M. "Repeated Insurance Contracts and Moral Hazard." Working Paper, The
Hebrew University, 1980.
SHAVELL, S. "Risk Sharing and Incentives in the Principal and Agent Relationship." Bell Journal of Economics
(Spring 1979), pp. 55-73.