American Economic Association
Decentralization, Hierarchies, and Incentives: A Mechanism Design Perspective
Author(s): Dilip Mookherjee
Reviewed work(s):
Source: Journal of Economic Literature, Vol. 44, No. 2 (Jun., 2006), pp. 367-390
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/30032252 .
Accessed: 08/02/2012 05:42
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 [email protected].
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal
of Economic Literature.
http://www.jstor.org
Journal of Economic Literature
Vol. XLIV (June 2006), pp. 367-390
Decentralization,
Hierarchies,
and
Incentives:
Design
A
Mechanism
Perspective
DILIPMOOKHERJEE*
Separationof ownershipfrom management,multidivisionalfirm organizations,delegation
of productiondecisions to worker teams, delegationof pricing and advertisingdecisions to
retailfranchisers, relianceon intermediariesin trade orfinance, and distributionof regulatory authority across different agencies represent examplesof organizationsthat delegate
and distribute decision-makingauthority instead of centralizingit. This paper reviews literature on costs and benefits of delegated decision making in hierarchicalorganizationsor
contracting networkswith regard to problems of incentives and coordination.It starts by
describing incentive and coordinationcosts of delegation in simple canonical examples of
hierarchieswhere both informationand incentives of differentdecisionmakersdiffer One
class of models pertain to contexts where the classical Revelation Principle applies, i.e.,
where costs of contractual complexity, information processing, or communication are
absent, agents do not collude, and the mechanismdesigner can commit to the mechanism.
Delegationmay conceivablyentail a loss of controland coordinationarisingfrom the divergence of informationand incentives. Sufficientand necessary conditionsfor this loss to be
mitigated entirely include risk neutrality,top-down contracting,and monitoring of transfers or production assignmentsbetween subordinates.The next class of models introduces
communicationcosts that restrict the performanceof centralizedarrangementsrelative to
delegationowing to a resultingloss offlexibility, which has to be traded off againstpossible
control losses of delegation. Finally, consequencesof collusion among agents is discussed,
which typically enlarge the range of circumstancesunder which delegationcan attain optimal second-best outcomes. The paper concludes with a discussion of the relevance of this
theoretical literature to recently emerging empirical studies of industrial organizations
where delegated decision making plays an important role: adoption of innovative human
resourcemanagementpractices, new informationtechnologiesand retailfranchising.
1. Introduction
the roots of this topic go back to discussions
of the appropriaterole of the state in a marAclassic topic in economic theory is the
ket economy by the Mercantilistsand Adam
organizationof economic activity.While
Smith's Wealth of Nations, it became the
*
issue in the 1930s debates concerncentral
Mookherjee: Boston University.An earlier version of
this paper was presented at the Summer Meetings of the
the
ing
possibilityof efficient resource alloEconometric Society at Northwestern University, June
cation in a socialisteconomy.In this debate,
2003, in the session on the Theory of the Firm, under the
it was recognized that informationrelevant
title "Delegation and Contracting Hierarchies: An
Overview." I am very grateful to Victor Aguirregabiria,
to efficient resource allocationis dispersed
Roger Gordon, John McMillan, Marc Rysman, Masatoshi
throughoutagents in an economy.A decenTsumagari, and anonymous referees for their comments
tralized market economy distributes deciand suggestions, as well as to Sandeep Baliga and Rakesh
Vohrafor their support and encouragement.
sion-makingauthorityto individualagents,
367
368
Journal of Economic Literature, Vol. XLIV (June 2006)
who are motivatedby their self-interestand
are coordinatedby marketprices. This was
contrastedwith a socialist economy where
decision makingauthorityis vested in a central planner, on the basis of information
communicatedby individualagents.
Leonid Hurwicz (1972a, 1972b, 1973,
1986) has providedexcellent surveysof this
early literatureand its subsequentoffshoots
culminatingin the contemporarytheory of
mechanism design. The comparativeperformanceof these specificdecentralizedand
centralizedresourceallocationsystemswere
sought to be comparedon the basis of the
extent of communicationand information
processingthey entail, and the incentivesof
individualagentsthey give riseto. Arguments
pertainingto the superiorityof the decentralized market-basedsystem relied on the fact
that their communicationsrequirementsare
minimal (representedby the dimensionality
of the message space for communication),
the factthatit distributestasksof information
processingacrossagentsin the economy,and
is consistentwith self-interestedbehaviorof
agents.On the other hand, a centralizedsystem maybe betterequippedto combatweaknesses of the decentralizedsystemin dealing
with problemsof externalities,public goods,
increasingreturns,and distributionalequity.
In environmentswhere these latterproblems
are significant, hybrid systems combining
elements of centralizationand decentralization have been studied,e.g., in the literature
on economic planning(Edmond Malinvaud
1967, GeoffreyM. Heal 1971).
Versionsof the grandquestionof comparative economic systemscontinue to be relevant to a variety of modern areas in
economics.An exampleis the internalorganization of large firms, to which A. Michael
Spence (1975) and Roy Radner(1992) provide excellent introductions.A large corporationis in manyways a miniatureeconomy,
in the organizationof which the question of
centralizationversus decentralizationis fundamental.Should decisionsconcerningproduction, procurement,pricing, distribution,
budgeting, and compensationsbe centralized in a corporateheadquarteroffice akin
to a socialisteconomy?Or shouldthe corporation be organized into multiple divisions
that operate as relativelyindependent cost,
profit, or investment centers, whose managers are delegated substantial authority
over production,pricing,procurement,and
resource allocation decisions pertaining to
their respective divisions? Should human
resource managementsystems be based on
delegationof problem-solvingon production
shop floorsto teams of productionworkers?
Comparative systems of organization of
business enterprises-e.g., the evolution of
the multidivisionalstructurein U.S. corporationsin the early twentieth century,comparisons between American and Japanese
business organizations, the adoption of
interaction-orientedratherthan control-oriented human resource management systems-have been described in these terms
(e.g., Alfred D. Chandler 1962; Masahiko
Aoki 1990; Casey Ichniowskiand Kathryn
Shaw 1999;Jon Gant,Ichniowski,and Shaw
2002; Ichniowski, Shaw, and Giovanna
Prennushi1997).
Similar issues arise in the literature on
regulation of public utilities (David P.
Baron 1985; Baron and David Besanko
1992; Richard J. Gilbert and Michael H.
Riordan1995). For instance, should regulation of all utilities be centered in a single
federal regulatoryauthorityor should it be
distributedbetween several state regulatory bodies possessing overlappingjurisdictions with a federal regulator?Alternatively,
should procurement of raw materialsby a
regulated utility company from an
upstreamfirm be delegated by regulatorsto
the utilityor shouldthe transactionwith the
upstream firm also be subject to regulation? Such organizationalquestions inherently raise similar issues of the optimal
degree of decentralization
activity.
of economic
Modernindustrialorganizationtheoryalso
raisesquestionsconcerningthe organization
Mookherjee:Decentralization, Hierarchies, and Incentives
of contracting relationships and networks
among differentmarketagents. In procurement contracting,shouldthe purchasercontract with a single "prime" contractor,
delegatingthe responsibilityof subcontracting with diverse upstream suppliers? Or
should she contract directly with downstreamand upstreamsuppliers,and personally coordinate their actions? Should a
manufacturerdelegate decisions over distribution, advertising,and pricing to wholesalers and retailers? Should investors
contractdirectlywith borrowersor delegate
authority to financial intermediaries to
negotiate loans and supervise the latter?
Changesin the natureand extent of reliance
on intermediariesappears to be occurring
within firms as well as procurement contractingin recent years. Corporate"reengineering" involved elimination of middle
layers of management in order to create
more direct links between customers, top
managers,and productioncenters (Michael
M. Hammerand JamesChampy1993, Erik
Brynjolfsson and Lorin M. Hitt 2000).
"Supplychain management"involves more
extensive outsourcingand reduced reliance
on intermediaries.
Comparisons between centralized and
decentralizedsystemsof economicorganization therefore continue to be relevant to
manyareasof contemporaryconcern.In this
article,I providean overviewof recent theoreticalresearchon this questionbased on an
applicationof mechanism design theory. I
shall focus on comparisonsbetween a centralized resource allocation "revelation"
mechanism where a central "principal"
makes decisions concerningproductionand
distribution based on communication
received from "agents,"with mechanisms
where decision-makingauthorityis delegated to the latter. Most of the literaturehas
focused attentionwhere the patternof delegation is hierarchical,i.e., where agents are
organizedas nodes in an invertedtree, with
the "manager"of any divisionalbranchdelegated authorityover subordinatebranches.
369
The purposeof this theoryis to throwlight
on the costs and benefits of decentralization
from an incentive standpointand compare
performanceof differentpatternsof decentralization.For the most part,the literatureI
describe has abstractedfrom costs of communicationor informationprocessing.This
has largely been driven by the need for
tractability:existing theories of mechanism
design typicallyfocus on each of these phenomenain isolation.For example,team theory (Radner and Jacob Marschak1972) or
the message space literature(Hurwicz1960,
1972b; Kenneth Mount and Stanley Reiter
1974; Thomas Marschak and Stefan
Reichelstein1995, 1998)concernthemselves
only with costs of communication.Theories
of costly information processing (Radner
1993; Patrick Bolton and Mathias
Dewatripoint1994; Mount and Reiter 1995;
Timothy van Zandt 1996, 1997) have
abstractedentirelyfromincentiveconsiderations. Nevertheless,I shallarguein section 6
below that, in order to have useful applications, the theoryneeds to addressaspectsof
informationtechnologyandincentivessimultaneouslywithin a common framework.To
this end, I shall include discussionof some
models which incorporate communication
costs as well as incentiveproblems.
A central result in the theory is the socalled "RevelationPrinciple,"which relies
for its validityon the absenceof communication or informationprocessingcosts. Under
some additionalassumptions,this Principle
establishes that centralized control cannot
be dominated by any delegation arrangement. Specifically,it demonstratesthat the
outcome of any decentralized organization
can be mimickedby a centralizedorganization in which the responsibilityof each agent
is merely to communicatetheir information
to a central authorityand awaitinstructions
on whatto do. This argumentcorrespondsto
the claimsof OskarLange and Abba Lerner
in the 1930s socialismdebate that incentive
properties of the decentralizedmechanism
can be replicated by a suitably designed
370
Journal of Economic Literature, Vol. XLIV (June 2006)
* Abilityof principalto commitupfrontto
centralizedmechanism.1Conversely,one can
view Hayek'scounterarguments
in the sociala mechanismandnot renegotiateit later.
ism debate that excludingconsiderationsof
In light of this Principle, there are two
communicationcost or limited information possible ways of constructing a theory of
processingcapacityof the centralplanneris
delegation and contractual hierarchies.
like throwingout the babywiththe bathwater.
The first approachstayswithinthe frameHow then does this theory succeed in
work of the RevelationPrincipleand studies
conditionsunderwhich delegationis an optithrowingany light on the centralizationversus decentralizationquestion? Its principal mal mechanism, i.e., where a hierarchical
virtue is to provide an analysisof the potenmechanismwithdelegationreplicatesthe pertial costs of decentralizationfrom an incenformanceof the best centralizedmechanism.
tive standpoint, while abstractingfrom its
Thisapproachhasbeen pursuedby Baronand
Besanko(1992), Nahum D. Melumad,Dilip
principalvirtues.These costs arise from the
fact that agents that are delegated decision- Mookherjee,and Reichelstein(1992, 1995),
and Sergei Severinov(1999), among others.
making authority act in their own selfrather
than
of
the
as
The
mainproblemwith this approachis that
interest,
organization
a whole or of the central designer. the conditionsmaybe restrictiveand, even if
Accordingly,it gives rise to a problem of
they are not, it cannotexplainwhy delegation
"loss of control" or "abuse of power." In
can be superiorto centralization.
contexts where these incentive problems
The second approach departs from the
can be overcome by judicious design of the
comfortable premises of the Revelation
delegation arrangement(i.e., incentives of
Principle. Different theories can be classito
whom
has
been
agents
power
delegated), fied by the precise avenueof departurefrom
the presumption is that decentralization the set of assumptions underlying the
will be the superior organizationalmode
RevelationPrinciple:
to
its
with
to
com(a) Informationprocessing costs (Radner
economy
respect
owing
munication cost and distributionof infor1993; Mount and Reiter 1995; Bolton
mationprocessingresponsibility.Whereasif
and Dewatripoint 1994; van Zandt
the incentive costs are substantial, the
1996, 1997)
choice between centralizationand decen(b) Costly communication,contract comtralizationwill involve a trade-off between
plexity (Melumad, Mookherjee, and
costs
and
Reichelstein 1992, 1997; Jean-Jacques
(incentive)
(communication,
informationprocessing)benefits.
Laffont and David Martimort 1998;
To furtherexplainthe differentbranchesof
Ilya Segal 2001)
the theoryand the organizationof this paper,
(c) Collusionamongagents(SandeepBaliga
it helps to review the differentassumptions
and TomasSjostrom1998; Laffontand
Martimort 1998; Antoine Faureunderlyingthe RevelationPrinciple2:
* Absence of: (1) costs of communication
Grimaud,Laffontand Martimort2003;
between agents and principal,(2) inforGorkemCelikforthcoming;
Mookherjee
mationprocessingcostsfor the principal,
and MasatoshiTsumagari
2004)
and (3) contractcomplexitycosts.
(d) IncompleteCommitmentand Renego* Absence of collusionamongagents.
tiation (Paul Beaudry and Michel
Poitevin 1995; Poitevin 1995; Wouter
Dessein 2002; Faure-Grimaud and
1 For the most general statement of this Principle, see
For
Martimort2001; Baliga and Sjostrom
B.
a
wider
see
discussion,
Roger
Myerson (1982).
Oliver Williamson (1985).
2001), IncompleteContracts(Philippe
2 In what follows we refer to the "principal"as the de
Aghion and Jean Tirole 1997; Andreas
facto designer and residual claimant, and "agents"as all
Roider2003).
others with a production or informationgeneration role.
Decentralization,
Hierarchies,and Incentives
Mookherjee:
In this paper,I will present an overviewof
the firstapproach,and avenues(b) and (c) of
the second approach.The emphasiswill be
on papersthat deal with the costs and benefits of delegationrelativeto centralization,in
which incentiveconsiderationsplayan active
role. I thus exclude discussionof avenue (a)
since it has so far abstractedfrom incentive
considerations.I do not discuss the important avenue (d) based on incomplete commitment mainlybecause an excellent survey
of this is alreadyavailablein Poitevin(2000).
I also exclude the literature based on
incomplete contractswhich involves ad hoc
restrictionson the set of allowablemechanisms.3These restrictionshave been sought
to be rationalizedby some authorsowing to
the possibility of contract renegotiation.
Segal (1999), for instance, explores conditions under which "simple"contractswhich
do not incorporateanycontingenciesachieve
the same performanceas optimalcontingent
contracts in a bilateral trading problem,
when contractscan be renegotiated.The literatureI discussin the first part of this survey (as well as the second part dealingwith
collusion) poses the analogous question of
when "simple" but commonly observed
schemes such as delegated decision-making
are optimal.
Clearly, there are some similarities
between the literature surveyed here and
the literature exploringthe foundationsof
incomplete contracts.Both search for additional constraintson contracts (apart from
standardnoncooperativeincentive compatibility constraints)that rationalize"simple"
real-worldmechanisms.Both are concerned
about the allocation of decision rights,
though they use somewhat different terminologies: decentralizationratherthan property rights.Nevertheless,the two literatures
differ in some importantdetails. The literature surveyedhere concerns a static context
3 See Tirole (1999) and the symposium on incomplete
contracts in the Review of Economic Studies 1999 for a
critical review of this literature.
371
with a single principaland manyproduction
agents privately informed about their
respective costs. Mechanismsmay be constrainedwith regardto extent of communication among agents, or by the presence of
collusivebehavior.The literatureon foundations of incomplete contracts deals instead
with bilateraltradesbetween two agentsthat
make ex ante relation-specificinvestments
under conditions of symmetricinformation
that cannot be verified by third parties.
Contractsmaybe subjectto the possibilityof
ex post renegotiation.
Returning to the theme of this essay,
some additional caveats are in order. The
literatureI shall discuss does not seem relevant to the question of boundaries of the
firm, where the incomplete contractframework seems more fruitful.5 It therefore
applies to delegation arrangements that
may or maynot be intrafirmand has little to
say on whether and why firms should outsource some production rather than inhouse. Agents in this frameworkcannot be
distinguishedby whether they happen to be
within or outside a firm.
Moreover, my intention is to provide a
selective guide to the literature that I am
personally familiar with, rather than an
exhaustivesurvey.So importantpapers may
havebeen missedor glossedover.I alsowant
to keep the overviewbrief and readable,so
will eschew formalismof any kind and try to
communicatethe essence of key resultsand
ideas. Those interested in more detailed
statementsof model assumptionsand results
should consultthe originalsources.
Section 2 sets out the canonicalsetting of
production with multiple agents privately
informedabouttheirown costs, employedin
most of the literature. Section 3 describes
models belonging to the "first"approach
4 Of course, these comparisonspertain to polar types of
models studied in the two literatures. Baliga and Sjostrom
(2001) is an example of a principal-multiagent model
where decentralization turns out to be optimal owing to
the possibility of contract renegotiation.
5 For an overview, see Oliver Hart (1995).
372
Journal of Economic Literature, Vol. XLIV (June2006)
described above, which explore conditions
for optimalityof delegationwithinthe traditional confines of the RevelationPrinciple.
Section 4 then discusses avenue (b) within
the second approach,based on costs of communication or contract complexity, while
section 5 is devoted to implicationsof allowing agents to collude. Section 6 assesses the
potential relevance of this literature to
empirical industrial organizationresearch.
Finally,section 7 concludeswith a summary
of the principalinsightsand shortcomingsof
this literature, and suggests directions for
future research.
2. The CanonicalSettingof Joint
Productionwith AdverseSelection
Most of the literaturefocuses on models
where agents have private information
about their costs, ratherthan moralhazard,
and we will do the same. The typicalmodel
is usually of the following kind. There is a
single principal (P), one or two producing
suppliers(A1,A2),plus one manager/middleman/monitorM who plays no role in production. The gross benefit of P depends on
joint output whose production depends on
contributionsor supplies of the two agents,
described by the production function q =
f(ql,q2), where Ai producesqi at privatecost
Ci(qi,O,)satisfying a set of single-crossing
conditions.6 The parameter O6affects the
productivityor cost of the agent and Ai is
privately informed about realization of O.
The cost shocks 01,02are independentlydistributed,and satisfya monotonehazardrate
condition that allowsone to ignore nonlocal
incentive constraintsin the classicalsetting.
The manageror monitorM observessignals
71,172which are informative about 0,02
respectively.P, A1,A2are risk neutral,while
M is risk neutral or risk averse. Outside
6 The key condition is that marginalproductioncosts are
increasingin 6,, but these are usuallysupplemented by conditions that total costs and the rate of increase of marginal
costs are also increasing. For the most general statement,
see Melumad, Mookherjee, and Reichelstein (1995).
options for all agents are normalizedto 0.
Agents observetheir costs priorto contracting and so earn informationrents. The main
trade-off in designing contracts or delegation is between productive efficiency and
payinginformationrents to agents.
Many papers employ considerably
restrictedversionsof this model, e.g., where
there is one productive agent rather than
two, or particularproduction functionse.g., perfect complementaritybetween ql,q2,
indivisibilityof inputs or outputs,linearcost
functions,or two point distributionsfor cost
shocksand signals.
The role of M in the organizationis informational "expertise"relative to the principal, acquiredeither throughprior expertise
or from ability to monitor cost realizations
of the agents. The process by which M
acquires this information is treated as
exogenous;the theory exploresthe implications of such informationalspecialization.
This contrastswith an earlier literature on
contracting hierarchies (e.g., based on
Guillermo A. Calvo and StanislawWellisz
1978) focused on moral hazardin supervision, but not on delegation questions.7
Some of the interesting questions
addressed by the theory concern the relative desirabilityof delegating to M rather
than the productive agents themselves.
This concerns the widespread phenomenon of "management" as a specialized
activity, distinct from "production" (as
argued persuasively by Radner 1992).
Accordingly,the organizationalalternatives
usually compared include the following
(see figure 1):
7 In that literature, supervisors are not delegated any
authorityto contract with those they supervise. The focus
was on whether moral hazard in supervision creates organizational diseconomies of scale. Indeed, Calvo and
Wellisz and others found that explaining scale diseconomies through this channel was typicallynot as straightforward has had been originally imagined, e.g., by
Williamson (1967). The literature on delegation described
here in contrast does provide conditions for emergence of
"controlloss" in hierarchies under suitable conditions, so
can be thought of as an alternative formalization of this
phenomenon.
373
Mookherjee:Decentralization, Hierarchies, and Incentives
p
p
A,
A1
A2
A2
CENTRALIZATION
(C)
DELEGATION
TO SUPPLIER
(DS)
P
P
M
A1
A2
M
A1
A2
DELEGATION TO
MANAGER
MIDDLEMAN
(DM)
CENTRALIZATION
WITH SUPERVISION
(CM)
Alternatives
Figure1. Organizational
Centralization without Supervision (C):
P retains all control, contracts and communicateswith A1,A2
Centralization with Supervision (CM): P
retainsall control, contractsand communicateswith A1,A2and M
Delegation to Supplier (DS): P contracts
onlywithA1,delegatesauthorityover contractingwith A2
Delegation to Manager/Middleman
(DM): P contractsonly with M, delegates
authorityto M overcontractingwithA1,A2.
Manypaperspose the questionof delegation versus centralizationas DS versus C;
others pose it as DM versus CM; yet others
as DM versus C. Accordinglyto understand
the relation between different papers, we
shall make this explicit in discussing their
resultsusingthis terminology.It is alsouseful
to clarifythatwe use the term "delegation"
to
connote a setting of hierarchicalcontracting
with multipleverticallayers,where contracting and communicationare restricted only
between adjacentlayers.For instance,in DS,
P does not contractor communicatewith the
subcontractorA2at all.
With only two productiveagents, most of
the literaturecomparesthe centralizedtwo
layerhierarchywith a three layerone where
the agent at the intermediatelayer is delegated authorityover contractingwith those
at the third layer.The main focus is thus on
problems of vertical control. Problems of
horizontal coordination across different
branchesof a hierarchyinherent in delegation have therefore received little attention
(with few exceptions, described in section
3.4). Issues concerningdesign of more complex hierarchies(e.g., span of controlversus
number of vertical layers, trading off horizontal coordination problems and vertical
control loss), or comparisonwith nonhierarchicalformsof delegation,must awaitfuture
research.
Additional details concerning delegation
concernexactlywhatis observedby the principal and the sequence of contractingin the
hierarchy.Insofaras they help determinethe
374
Journal of Economic Literature, Vol. XLIV (June 2006)
natureof controlloss from delegation,these
will be described in subsequent sections.
Collusionbetween agentsis modeled as hidden side-contractsbetween the agents, and
further details of observabilityand timing
are involvedhere, whichwill be explainedin
section 5.
3. First Approach:Conditionsfor
Optimalityof Delegationwithin
the TraditionalFramework
3.1 Conditionsfor DS = C
Considerfirstthe optimalityof delegating
to a "prime"supplier the authorityto subcontract with the other supplier,i.e., comparisonof DS with C. When the Revelation
Principle applies, we know that C weakly
dominates DS, so the question is when DS
achieves the outcome under the optimal
centralized mechanism. In this case, delegation entails no control loss at all. Insofar
as contractingwith one agent rather than
two is simpler or easier for the principal,
this can provide a rationalizationof delegation as a way of implementing the optimal
outcome.
The potential control problem with delegationinvolvesa formof "moralhazard":the
"prime"supplier'sincentive to allocateproduction and paymentsbetween himself and
the subcontractormay differ from what the
principal desires. In particular,the prime
contractorhas monopsonypower over the
subcontractors,and the principalin turn has
monopsonypower over the prime contractor. Both sets of relationshipsare subject to
adverseselection, so monopsonywill generate distortionsin productionand payments
that raise contractingcosts for the principal.
The verticalcontrolprobleminherentin delegation thereforeis essentiallythat of "double marginalization of rents" (DMR),
stressed in the context of sequences of
monopoly relationships in the industrial
organizationliterature.
The monopsonypower of the prime contractor (A1, say) over the subcontractor (A2)
resultsin two problems.First,there tends to
be "toolittle outsourcing"from subcontractors (or too low a supplyprice offeredby A,).
This resultsfromthe attemptby A, to garner
maximalinformationalrents for himself at
the expense of A2 and P. Roughlyspeaking,
the volume of informationrents are proportional to the level of productionallocatedto
an agent, so A1 allocates himself a larger
share of production than either A2 or P
would desire. Second, A1 is privately
informedaboutsubcontractingcost (whathe
has to pay A2)vis-a-visthe Principal,owing
to the lackof directcommunicationbetween
P and A2.This adds a dimensionof adverse
selectionin A1'srelationshipwith P, over and
above privacyof informationabout his own
cost of production. There is a consequent
"cascading"of informationrents: first such
rents are paid to A2 by A1, and then this is
subjectto an additionalmarkupowingto priinformationvis-t-vis P regarding
vacy of
AI's costs. This is as far as the princisubcontract
pal'swelfare is concerned; from the standpoint of socialefficiencyonly the production
misallocationmatters.
An importantresultin the literatureis that
these controlproblemscan be eliminatedin
DS under the followingconditions:
(3.1) Observability of subcontract costs
or allocation: either q, the amount
produced by the prime supplier,or x,
the subcontractingcost incurred.
(3.2) Top-Down Contracting: P contracts
with A1 before A1 communicates or
contractswith A2.
(3.3) Risk-neutrality, absence of limited
liability constraints: for A1
The mainidea underlyingthis resultis that
condition(3.1) enables the principalto subsidizeoutsourcingor taxin-houseproduction
by A1 to correctthe productivemisallocation
between the two agents. Moreover(3.2) and
(3.3) allow P to "tax"awayupfrontinformation rents derivingfromprivacyof the prime
contractor'sknowledgeof subcontractcost.
Each of these three conditionsare necessary for delegation to be optimal, barring
Mookherjee:Decentralization, Hierarchies, and Incentives
exceptionalcases. The productionmisallocation can arise only in the presence of some
substitutabilitybetween the inputs supplied
by the two agents. Otherwiseif there is perfect complementarity,there is no scope for a
monopsony distortion between A1 and A2,
and condition (3.1) is unnecessary (Baron
and Besanko 1992; Gilbert and Riordan
1995). Otherwise,conditions(3.1) and (3.2)
are necessary(given risk neutrality(3.3)) if
the production function is smooth
(Melumad, Mookherjee, and Reichelstein
(MMR, hereafter) 1995). The necessity of
risk-neutrality(3.3) is also suggested by
MMR 1995, and reinforcedby the literature
to be describednext.
3.2 Comparisonof DM with C
Another way of posing the question of
optimalityof delegation is to compare DM
(ratherthan DS) with C. R. PrestonMcAfee
and John McMillan(1995) considerthe case
with only one productiveagent, and a single
monitor M who has no better information
than P about the agent'scost. They focus on
the costs of intermediationarisingfrom delegation of procurementto M who is subject
to limited liability.The main idea here is
that M can earn information rents with
respect to privacyof knowledge of subcontract costs, since there is no direct communication between the supplier and the
principal.These rents cannot be taxed away
upfront owing to limited liability constraints.This is very similarto the implications of violation of (3.2)-wherein A1 can
contract or communicate with A2 before
responding to the contract offered by P
(MMR 1995).
Faure-Grimaud and Martimort (2001)
replacelimited liabilityby riskaversionof M
and obtainan agencycost of delegation(with
a single productiveagent) that is qualitatively similar (though different in detail).
Inabilityof M to costlesslybear riskprevents
congruence of interests of M with P at the
time M designs subcontractsfor productive
agents. This is somewhat akin to a classical
375
moral hazard problem that trades off risk
sharingand incentiveswith a single agent.
3.3 Effectsof SupplierConsolidation:
ComparingDM with C with Two
ProductiveSuppliers
Now considerthe relatedquestionof how
DM relates to C in the case of two productive agents A1,A2.Assume M is perfectly
informed ex ante about realizationof the
agents' costs: so there is no upfront uncertainty faced by M. Delegation to M implies
that M ratherthan A1,A2earn informational
rents, owing to informational asymmetry
between M and P concerning subcontract
costs.The principalnow contractswith a single supplierof both inputswho incurssupply
cost equal to the sum of the two agents'cost.
The comparisonof C with DM is thus really
a questionof effects of consolidatingthe two
suppliersinto a single supplier.
The main result here is that such consolidation benefits P if the inputs are perfectly
complementaryin production under some
added distributionalconditions (Baron and
Besanko 1992, Gilbert and Riordan 1995),
but hurts P if they are substitutes(Severinov
1999; Mookherjee and Tsumagari2004).8
The intuitive idea underlyingthis result is
the following. In C there are externalities
between the two separate suppliers,which
are internalizedwith consolidation.If they
supply substitutes,then competitionis suppressed and cost reports increase (on average) with consolidation.Consolidationalso
convertsa pair of one-dimensionalincentive
problemsinto a single two-dimensionalone
(both in terms of dimensionsof information
and numberof goods delivered).This tends
to strengthenincentiveconstraints(owingto
the abilityof the single consolidatedagent to
coordinate reports). On both counts the
8 The definition of substitutes and complements is in
terms of the way that the demand for input delivered by
one suppliervarieswith the cost reported by the other supplier. Mookherjee and Tsumagarishow how this relates to
notions of substitutes and complements in terms of the
elasticity of substitution of the production function.
376
Journal of Economic Literature, Vol. XLIV (June 2006)
principalis worse off when the two inputs
are substitutes.If they supplycomplements
instead,the internalizationof biddingexternalitiescauses cost reportsto fall on average
with consolidation.But opposingthis is the
problem that the adverseselection problem
is a higher dimensionalone with consolidation. Under particulardistributionalconditions the higher dimensionalityposes no
additional problems, and then P attains
superiorexpected profitfrom DM.9
That delegation scores above centralization here may occasionsome surprise,since
the Revelation Principle still applies.
However the set of agents is not the same
between DM and C: in the former there is
an additional agent M who is better
informed than P. The superiorityof DM
thus flows from access to this additional
source of informationunavailablein C by
assumption. Indeed, the relevant centralized benchmarkfor DM is CM ratherthan
C, where M belongs to the organizationand
P retainsall controland contractswith M, A1
and A2. In such a setting M is relegated to
the status of a supervisor or consultant,
rather than being a manager with control
rights over the suppliers. The Revelation
Principle asserts that CM always weakly
dominates DM. Moreover,in CM the principal can costlessly acquire all of M's information, given the absence of any collusion
between M and the productive agents
supervised. Whereas in DM the principal
has to encounterthe problem of controlling
M. Hence in generalP would be strictlybetter off retaining control (CM) rather than
delegating (DM).10This implies that explanations of the widespread phenomenon of
delegation to intermediariesuninvolved in
actual production must perforce depart
9 The distributionalcondition is however quite restrictive: i.i.d. exponential cost shocks with a lower bound of 0.
Extensions to a wider class of distributions remains an
open question.
10 This is easiest to see when M has perfect information
about costs-then CM attains the first-best, which DM
cannot.
from the traditional setting of the
RevelationPrinciple.We shall returnto this
point below.
3.4 MoreComplexHierarchies:
OrganizationalDiseconomiesof Scale;
HorizontalCoordination
Now suppose there are more than two
productive agents. Then there is a choice
between different hierarchicalpatterns of
delegation. Abstractingfrom the possibility
of employingsupervisorsor intermediaries,
the n producingagents could be organized
in a linear vertical chain, with agent Ai
authorized to contract with A+1..Or they
could be organized into two horizontal
departments,each of which contains a subset of producers,and involvesa verticallinear chain within the department. Or we
could have a hierarchywhere each agent is
authorizedto contractwith m subordinates
(with k vertical layers). See figure 2 for
examples.And so on: there are a large number of possible hierarchicalstructures,with
varyingspans of control (horizontalbranches) and vertical layers. The questions that
can be posed in this frameworkinvolveboth
questionsof the design with a given number
of agents(e.g., spanof controlversusvertical
layers, how to group agents into departments, how to organizecommunicationand
contractingsystems),as well as how the performanceof the organizationchangesas the
numberof agentsincreases.The latterquestion is of interest to understandinglimits to
the size of firmsor networksthat arisesolely
from incentiveand coordinationproblems.
Delegation is now potentially prone to
problems both of vertical control loss (cascading across verticallayers) and coordination across horizontalbranches. The latter
problem is not addressed by any of the
models described so far, though it has
received some attention in the team-theoretic literature which has abstracted from
incentive considerations (e.g., Marschak
and Reichelstein 1995, 1998). Jacques
Cremer and Riordan(1987) representsone
Mookherjee:Decentralization, Hierarchies, and Incentives
p
377
P
Al
A1
A3
A2
A4
A2
A3
TWO LAYER
TWO
DEPARTMENT
HIERARCHY
A4
LINEAR CHAIN
FOUR LAYER
HIERARCHY
withMoreThanTwoAgents
Figure2. Hierachies
of the first attempts to deal with complex
hierarchies (and also nonhierarchicalcontract networks), wherein conditions for
implementabilityof first-best efficient allocations with ex ante contractingwith riskneutral agents were established.11 In
particularthey describe a sophisticatedsystem of contractingand communicationthat
enables such a contracting network to
resolve both incentive compatibility and
coordination to achieve ex post efficient
outcomes.
Extensionsof the theory to implementation of optimal mechanismsfor a principal
(rather than ex post efficient outcomes) in
the presence of adverseselection are considered in Mookherjeeand Reichelstein(1997,
2001). They confine attentionto contracting
hierarchies that are "consistent"with the
technologyin the sense that one dimensional reportingwould help the same hierarchy
11Ex ante contractingrefers to the situationwhere contracts are negotiated before agents receive any private
information, so adverse selection problems do not arise.
achievefirst-bestoutcomesin the absenceof
any incentiveproblems.In other words,the
productiontechnologyhas a constantreturns
recursive structure, wherein production is
hierarchicallydecomposedinto departments
and subdepartments. In the absence of
incentive problems,coordinationacrossdifferent departments can be achieved by a
hierarchicalcommunicationsystem, where
each manager receives (one-dimensional)
cost reports from subordinates,aggregates
them into a departmentalcost, and reports
this to his superiorin turn. Cost reportsflow
up the hierarchyin this way, up to the very
top. There aggregate output decisions are
made,which are subsequentlydisaggregated
into departmentaltargets,with these flowing
subsequentlydown the hierarchy.
This mechanismcan be extendedto incorporate incentive problems as well, under
exactlythe same conditions (3.1)-(3.3) that
enables DS to achieve optimaloutcomes in
the two agent case. Specifically,Mookherjee
and Reichelstein show that if assumptions
(3.1)-(3.3) hold, any hierarchy "consistent"
378
Journal of Economic Literature, Vol. XLIV (June 2006)
with the technology (in the sense defined
above) can achieve the same expectedprofit
and production allocation as the optimal
centralized mechanism. This implies that
conditions (3.1)-(3.3) as well as "consistency" are sufficient (as well as necessary)for
hierarchicaldelegation to implement optimal centralized outcomes. Once vertical
controlproblemscan be overcomewith one
level of delegation, they can be overcome
with multipleverticallayers.Moreoverhorizontal coordinationacross departmentscan
also be costlesslyassuredby the mechanism.
The mechanisminvolvesthree stages:first
contractsflow down the hierarchy,then cost
reportsflow up and get aggregatedalongthe
way, determiningproductiontargets which
flow down at the third stage. Specifically:
* at each layeran agent managesa "profit
center"
* managerialincentive schemes are linear
in a measureof profit of the center
* departmental costs equal aggregate
payments authorized to subordinate
departments
* departmentaloutputis valuedat willingness to pay (ala Groves-Vickrey)of the
managerat next higherlevel
* managers self-select profit targets for
themselvesat the first(contracting)stage,
then reportcost (bid on projects)to their
bosses at the secondstageafterreceiving
cost reportsfromtheir subordinates
Some of these featuresresembleperformance budgetingsystemsdescribedin management accountingtextbooks(e.g., CharlesT.
Horngren and George Foster 1991). The
mechanismdistributesinformationprocessing
tasks throughoutthe hierarchy,as managers
aggregate cost reports for their respective
departmentsandallocateproductionbetween
subordinatesand themselves.With incentive
problemsovercome,design of the hierarchy
can be based on considerationsof efficient
distributionof informationprocessingi la theories of Radner(1993) or van Zandt
(1996,
1997). In particular,these considerations
explainwhy the two extreme organizational
forms (a "flat"two layer centralizedmechanism and a "steep"linear chain) would be
dominatedby intermediateformswitha number of verticallayersandhorizontalbranches.
Note finallythat the result providesa set of
benchmarkconditionsunderwhich there are
no organizational
diseconomiesof scale.
3.5 Shortcomingsof the First ("Traditional")
Approach
As we have seen above,the conditionsfor
optimalityof delegation are restrictive.For
instance,absence of risk aversionor limited
liability constraints limits applicability to
managerialhierarchies.Implementabilityof
second-best allocationsvia delegation may
be more relevant in procurement settings
where the prime contractoris a large firm
with deep pockets. On the flip side, the theory explainsinformationrents of managers
that cascade across vertical layers. If intermediate managersare more risk-aversethan
the owners, this makes it difficultto ensure
coincidence of their objectives. These
results suggest that costs will grow with the
scale of the firm'soperations,causingorganizational diseconomies of scale, though
explicitmodels of this phenomenonare still
awaited.
A more serious problem with this
approachis that it cannot ever explainwhy
delegation may dominate centralization.
This is simplya consequenceof stayingwithin the confines of the RevelationPrinciple.
Moreover, the constructed delegation
mechanisms (e.g., under assumptions
(3.1)-(3.2) of observability and top-down
contracting) are vulnerable to collusion
between agents. This is because the principal has to subsidize outsourcing from
subcontractors, creating incentives for
artificiallyexaggeratingoutsourced cost by
the prime contractor.12The prime contractor also would have an incentive to
contract and communicate with the subcontractorbefore respondingto the principal's offer, which may be difficult for the
principal to prevent.
Mookherjee:Decentralization, Hierarchies, and Incentives
379
Yet anotherproblemis that this approach
cannot explain delegation to middlemen or
managersthatplayno directproductiverole.
As explained above, it is generally strictly
better for P to retain control and treat M as
an informationprovider.This stems partly
from the assumptionof noncollusivebehavior between supervisor and productive
agents.
This motivatesinterestin literaturewhich
confrontsthese problems,by departingfrom
the confines of the RevelationPrinciple.
authority. While this is undoubtedly an
appealing idea, it requires modeling limits
on informationprocessing of the principal
and agents. Modelingincentive compatibility constraintsin such a context is difficult,
since it requiresmodelinghow agentsdivide
their limited informationprocessingcapacity between attemptingto game the principal
for their own self-interest,and carryingout
the computationsthat they are appointedto
perform.13
In order to avoid this foundationalproblem, an alternativeapproachis to introduce
4. Benefitsof Delegation:Costly
restrictions on communication between
Communicationand Contract
agents, ratherthan constraintheir information processing capacities. Mechanismsare
Complexity
constructed under the assumption of
Reconsider the choice of DS versus C,
"unboundedrationality,"
but subjectto these
where there are two productiveagents. The
communication
restrictions.
MMR (1992)
questionis whetherthe principalshoulddelcentralized
and
decentralcompareoptimal
egate contractingwith A2to A1.A commonly ized
mechanisms
with
an
restricalleged advantageof decentralizationis that tion on the size of the exogenous that
message space
it utilizes the benefits of local information.
This presumes that centralizationis unable can be used by any agent to communicate
to take advantageof such local information. with others.In particular,if the upperbound
In the framework where the Revelation on the size of the message space is not large
enough to permit agents to communicate
Principleapplies,however,communicationis
costless and centralizeddecision makingcan everythingthat they know,centralizeddecision-makingcannot access all the informaaccess the same informationas any decention
that delegationmechanismsutilize. For
tralized mode. Formalizationof the "local
the mechanismDS allowsproducinstance,
information"advantageof decentralization
tion
allocations
between the two agentsto be
necessitates incorporationof constraintson
decided
on
by A1 the basisof his information
what agents can report to the principal,or
about
his
personalcost 01,which he can only
alternativelyon the extent of information
communicate
to the principal.The
partially
that can be processedby the principal.
restrictions
on
the
size
of
the messagespace,
Models of costly informationprocessing
are
not
however,
explicitly modeled, and
in the mould of Radner (1993), Mount and
would
likely requiresome underlyingmodel
Reiter (1995), or van Zandt(1996, 1997) are
of
"expertise"of suppliers not shared by P
still in their infancy and await extensions
which
restrictthe vocabularyof communicathat incorporate incentive considerations.
tion
between
them, or the need to make
They model computationstaking place in
decisions in real time where messages take
real time, and model advantagesof decentime
to be communicated.
tralizationin distributinginformationproThe
result in MMR (1992) is
cessing tasks throughout the organization, that:DSprincipal
dominates
C if(a) there is a
strictly
rather than concentratingit in one central
12 The prime contractorcould claim to have outsourced
more than the true amount, in return for a side payment
from the subcontractor.
13 Indeed, this decision itself is a higher order problem,
quickly giving rise to an infinite regress of the form familiar in any model with "decision costs."
380
Journalof EconomicLiterature,Vol.XLIV(June2006)
finite messagespacefor each agent, whereas
costs lie on a continuum;(b) assumptions
(3.1)-(3.3) of cost observability,top-down
contractingand risk-neutralityhold;and (c)
the cost function of agent Ai is multiplicatively separablebetween quantityproduced
qi and the cost shock Oi.Note that this result
applies irrespectiveof the size or nature of
the message spaces, as long as they apply
uniformly to both organizationalregimes.
Hence the comparisondoes not requireany
explicitmodel of communicationcosts. Note
also this result pertains to DS and C with
particular reporting structures (e.g., in C
where agents either communicate simultaneously or sequentially in two stages). So
this does not establish the global optimality of delegation among the class of all possible mechanisms subject to the same
message space restrictions. Part of the
problem is that there is no general characterization so far available(analogousto the
Revelation Principle) of the set of all possible feasible mechanismsin the presence of
communicationconstraints.14
Nevertheless,the result capturesthe idea
that delegated decision makingmay be better able than most commonversionsof centralization to utilize "local" information.
Productionassignmentsare selected by A1 in
DS on the basisof his informationaboutown
cost 01,which is finer than can be communicould
cated to P. In general,this "flexibility"
be abusedby A1 to pursuehis own interestat
the expense of the principal:the restriction
on communicationalso limitsthe abilityof P
to calibratethe outsourcing.subsidy in DS
preciselyenough to amelioratethe problem
of double marginalization of rents.
Nevertheless MMR show that the added
control loss is always outweighed by the
advantage of better informed production
decisions,providedthe principalcan monitor
14 In
particular, the result described above does not
exclude the possibilitythat a centralized mechanism with a
more general multistage communication mechanism may
be able to match the performance of DS. This is an open
question for future research.
productionassignmentsorpaymentsbetween
the agentsex post.
An alternativerestrictionwhich yields the
same result concernsthe numberof contingencies in contracts.A contingencycan be
viewed as a statementof productiontargets
and paymentsmade to an agent, conditional
on reportscommunicatedby the latter.One
measureof the complexityof a contractis the
numberof contingencies,since these haveto
be written(oftenwith the aid of lawyers)and
read by thirdpartyenforcers(e.g., courts)at
some cost. Accordinglycontractswith more
contingenciesinvolvehighercosts of writing
and verification.
Of course in some situationsthis is not a
good measureof complexity,e.g., when it is
possibleto expressthe contractin the formof
a simple mathematicalformula(the simplicity of which maybe contrastedwith the number of elements in the graphof the function,
the complexity measure being considered
here). A deeper analysisof complexityshould
be based on measuresof complexityor computabilityof the mathematicalformulathat
representsthe contract.
The performanceof DS and C can be
comparedwith identical restrictionson the
number of contingencies in contracts in
either regime. Restrictingcontractsto finite
complexityimposeslimitsnot just on the size
of messagespacesof agents,but also of decisions (concerningproductionand payments)
thatcan be takenon the basisof suchreports.
This furtherlimitsthe flexibilityadvantageof
DS over C, comparedwith the case where
messagespace size is restricted.Despite this,
it can be shown that the superiorityof DS
over C still prevails under the same conditions as where message space sizes are
restricted (MMR 1997). Productionallocations can no longerbe variedcontinuouslyby
A1;the contractcan onlyspecifya finite set of
alternativeproductiontarget configurations.
Yetthere is scope in DS forA1 to choose flexibly among these different configurations
based on his "local"informationwhich cannot be accessedby a centralizedmechanism.
Mookherjee:Decentralization, Hierarchies, and Incentives
On the other hand, if one or more of
assumptions (3.1)-(3.3) do not hold then
examples can be constructed whereby the
rankinggets reversed,i.e., the exacerbation
of the control loss outweighs the flexibility
advantage,renderingC superior.Accordingly
this theory succeeds in providingconditions
when either of the systems (strictly)dominates the other. Howeverthere still remains
the need to provide a deeper foundationof
the notions of contractual complexity or
communication costs employed in this
branchof the literature.
5. CollusionAmongAgents
We now describea more recent and active
strandof the literatureevaluatingdelegation
when agents collude, while all other traditionalassumptionsunderlyingthe Revelation
Principleare retained.Collusionaffects the
performanceof both delegationand centralization. For instance we have alreadyseen
thatthe mechanismsconstructedfor DS that
replicate outcomes of the best centralized
mechanismunder conditions(3.1)-(3.3) are
vulnerable to collusion among the agents.
Optimal centralized mechanisms are also
frequently vulnerable to collusion.15The
questionis how the two regimescomparein
the presence of collusion.
One possible intuition for the virtues of
delegation is that it already incorporates
side-contractingamong the agents, unlike
centralization.However we shall see below
that this intuitionis difficultto makeprecise:
when one models hidden side contractsin a
particularway, a version of the Revelation
Principle reappears,whereby centralization
can replicatethe outcomesof anydelegation
arrangement. Nevertheless, one can then
explorethe idea that delegationand centralizationare equivalentunder a broaderset of
circumstanceswhen agentscollude. In other
15 Consider for instance a second price auction in which
bidders can coordinate their bids and enter into hidden
side-contracts that reallocate the good among themselves.
There often exist such side contracts that are interim
Pareto superior from the standpoint of the bidders.
381
words, delegation may be an optimal
responseto the presence of collusionamong
agents-if side contractingcannot be prevented, the principalmayas well authorizeit
explicitly.
An additional reason to study collusion
was mentionedpreviously:manyinstancesof
delegation take the form of DM, where
authorityis delegated to expert intermediaries or managerswho play no role in actual
production.In the absenceof collusionP can
costlesslyobtainM'sinformationanduse this
to design contractsfor supplierspersonally.
However, the agents then have a stake in
bribing M to withhold his information.
Collusion would therefore prevent costless
acquisitionof M'sinformationundercentralization. Delegation to M would not permit
costless extractionof M'sinformationeither,
since it would form a source of information
rents for M in contractingwith the principal.
M will earn rents under either regime in the
presence of collusion, so the superiorityof
centralizationis no longer obvious.
5.1 ModelingCollusion
Collusion is typically modeled as a side
contract between agents which is unobservedby P, and subjectto asymmetricinformation within the coalition. Nevertheless
manyimportantdetails need to be specified
concerningthe nature of the side-contracting game. How are these contractsnegotiated, how does their timing relate to the
proposaland acceptance stages of the contract offered by the principal?What is the
rangeof decisionsthe agents can coordinate
on, and what does the principal observe?
How is the side contractenforced?
The complexityof analysisof coalitional
behaviorunderprivateinformationhas motivated most authors to adopt simplifying
assumptionsthat permit a tractableanalysis.
Mostof the literaturefollowsthe trendset by
Tirole (1986) by ignoring all enforceability
constraints,by assuminglong-termcontracts
or self-enforcingrelationshipsamongagents.
Only restrictions imposed by asymmetric
382
Journal of Economic Literature, Vol. XLIV (June 2006)
informationwithinthe coalition(besideslimited liability,if applicable)are imposed. At
the same time the contractualrelationship
between agents and principalis short term
and restrictedto that defined by the formal
contract.
It is also commonto assumethat side contractallowsagentsto coordinatereportsto P,
and to reallocate production assignments
andpaymentsbetween themselves.This corresponds to auction contexts where coordinated bid-riggingand hidden "resale"of the
good cannot be prevented by the seller.
Since collusion occurs with asymmetric
information,the Coase Theorem does not
apply.Hence actualoutcomes (e.g., production assignments)will depend on the allocation of bargaining power within the
coalition. In turn this depends on who has
the power to propose the side contract.
Laffontand Martimort(1998) assumethat
the organizationalvariantaffects allocation
of bargainingpower:in DS A1 makesa takeit-or-leave-itside contractoffer to A2,while
in C a neutral third party designs the side
contractfor A1and A2.This may seem natural in some ways: if the principaldelegates
authority for contractingwith one of the
agentsto the other,it augmentsthe bargaining power of the latter. Yet in the Laffont
and Martimort (1998) formulation this is
built in as an exogenous shift in bargaining
power. An alternativewould be to explain
this shift endogenously. For instance, the
reason that A2 has more bargainingpower
vis-.-vis A1 undercentralizationis thatA2has
the option of turning down the contract
offered by A1 and then playingthe contract
offered by the principal noncooperatively.
Such an outside option is not availableto A2
in DS-thus conferringmonopsonypower
to A1in the delegationarrangement.
The more recent literature models the
endogenous effect of organizationalstructure on bargainingpower within the coalition. Specifically,it is commonly assumed
that one of the agents (denoted A*,say) has
the power to make a take-it-or-leave-it
side-contractto the others. If the principal
delegates subcontractingto A*, then the
other agents do not have the option of
rejecting the subcontractand dealing with
the principalon their own. Under centralizationin contrast,they have the opportunity of rejecting the subcontractand playing
the principal'sgame noncooperatively.This
raisestheir outside optionvis-h-visA*,effectively gainingbargainingpower. The extent
to which they do so depends endogenously
on the contract offered by the principal.
The latter simultaneouslydefines the stakes
of collusion for the agents, as well as their
relativebargainingpower.
A consequence of this formulationis that
the differencebetween delegationand centralizationboils down to this. If P delegates
to A*,then all other agents have no alternative to the subcontractoffered by A*.Under
centralization,P offers a contractto all the
agents,who now havethe optionof rejecting
the side contractoffered by A*and responding to P's offer noncooperatively.Then centralizationreduces to delegationif P offers a
null contractto all agents but A*.Therefore
a version of the RevelationPrinciple reappears:if the structureof the side-contracting
game is the same in differentorganizational
variants(in the sense of who has the right to
designthe side contract),then centralization
is always weakly preferred to delegation.
Such a formulationcannot thereforepermit
delegation to dominate centralization.We
are back to a question similarto that in the
traditionalfirst approachdescribedabove:is
delegationcostly relativeto centralization?
5.2 Costsof Delegationwith Collusion
In order to emphasizethat centralization
regime now incorporatescollusion,we shall
refer to the two variants of centralization
(withandwithoutM) by CMC and C respectively. As before, we can phrase the key
question either as comparisonof C and DS
(shouldP delegate to one of two suppliers?),
or of CMC and DM (shouldP delegate to a
monitor/managerM?).
Mookherjee:Decentralization, Hierarchies, and Incentives
5.2.1 C versus DS
This question is posed by Baliga and
Sjostrom(BS, hereafter) (1998) in a model
with moralhazardand limited liability.Two
suppliersA1,A2jointlyproducean indivisible
output for P. The probabilityof "success"is
increasingin the effort of each agent, which
takes one of two possiblevalues ("shirk"and
"work").Productionoccurssequentially,and
the effort of one agent (A1)is observed by
the other (A2)but not vice versa.The principal cannot observe efforts of either agent,
and is constrainedto pay them nonnegative
wages (with zero as their respective outside
options). The combinationof moral hazard
with limited liabilitygives rise to "efficiency
wage" informationrents, and renders the
Coase Theoreminapplicable.The stakesfor
collusionarisehere wheneverP tries to elicit A2'sinformationabout Al's effort:A1can
bribeA2to withholdadverseinformation.BS
do not model the actual process by which
the side contractis negotiated,using instead
a notion of an equilibriumside-contractas a
Pareto-undominatedside contract,given the
principal'smechanism.
The main resultsof their paper are as follows: (i) In C it is alwaysoptimalto use simple contractswith no communication,where
P paysA, a wage wi in event of success and 0
otherwise;(ii) For a large range of parameter values, delegation is optimal (i.e., the
best simple contracthas either w, or w2= 0);
and (iii) There exist other cases for which
delegation is not optimal (e.g., if neither
agentA, is "essential"enoughthat delegating
to the other agentAjwould induce the latter
to pay formera positiverent in event of success). The model thus capturesthe idea that
in a large range of cases decentralizationis
an optimal response to collusion-the principal can implementthe optimalcentralized
outcomeundercollusionby contractingwith
only one of the two agents and leaving that
agentto subcontractwith the other.The payment to the delegated agent is such that it
motivatesthat agent to pay the other agent
383
exactlythe same efficiencywage as the principal desires,and leave the same amountfor
the delegated agent as well. However (iii)
shows that such an implementationis not
globallypossible.
A similarcomparisonis carriedout in an
adverse selection framework of the kind
describedin previoussectionsby Mookherjee
and Tsumagari(MT,hereafter)(2004). They
find in contrastto the BS paper that delegation to a supplieris always strictlydominated by centralization.The essential reasonis
that the shift of bargainingpower to A2
under centralizationallows the productive
distortion inherent in DS to be reduced.
Recall that this distortiontook the form of
"insufficientoutsourcing"fromA2.The need
for A1 to offer higher rents to A2 in the centralized setting (where P offers a contract
more favorableto A2 than emerges in the
solution to DS) forces A, to subcontract
more to A2,thus reducingthe extent of the
productivemisallocation.In a sense, this is
qualitativelysimilarto the cases studied by
BS in which decentralizationwas not optimal: there centralizationoffersA2 a positive
efficiency wage which raises his effort,
reducingthe distortionarisingin delegation
when the delegatedagentA1prefersto keep
all the rents for himself and so does not pay
any efficiencywage to A2.The differencein
resultsbetween the two paperscould therefore reflect the fact that outputs and inputs
are assumedto be divisiblein the MT paper,
enlarging the scope of distortions arising
from Al's monopsony power in delegation.
Alternativelythere are important distinctions between the moralhazardand adverse
selection settingsconsideredthat drive their
respectiveresults.
5.2.2 DM versus CMC
Faure-Grimaud,Laffont, and Martimort
(FLM, hereafter) (2003) and Celik (2002)
both consider the case of one productive
agentA and one supervisorM. The formeris
privatelyinformed about his own cost. The
two papers adopt different assumptions
384
Journal of Economic Literature, Vol. XLIV (June 2006)
about the natureof informationavailableto
M aboutthis cost. FLM assumethat the cost
of the suppliertakestwo possiblevalues,and
M observesan informativesignalwhich also
takes two possible values. In contrast,Celik
assumesan arbitraryfinite numberof possible costs, with the informationof M represented by a connected partition over the
state space. Specifically,M can narrowdown
the agent'scost to a (connected) subset of
possible cost levels, but cannot distinguish
between different costs within this subset.
Apartfrom the informationstructure,FLM
allowsfor risk-aversionof M, whereas Celik
assumesriskneutrality,but this differenceis
less fundamental.
The results of these papers are strikingly
different:FLMfind that delegationis always
equivalent to centralization,whereas Celik
finds that delegation is inferior in general
(e.g., wheneverthe optimalcontractfor the
agentin the absenceof M is strictlymonotone,
and M is not perfectly informedabout the
agent'scost). The intuition in Celik'spaper
seems similarto that in the MT paper:delegationto M is subjectto a monopsonydistortion, causing M to procure an insufficient
amountfrom A that is not in the principal's
interest. In particular,the optimalallocation
involvesincentiveconstraintsthatbind solely
in the "downward"direction, where more
productivetypes of the agent are indifferent
between underreportingand reportingtruthfully. Centralizationallows the principalto
raiseA'soutsideoptionin bargainingoverthe
subcontractwith M, reducingthe extent of
this monopsonydistortion.In particular,since
the agent'sreservationutilitywith respect to
the side contractis strictlyincreasingin his
productivity(owing to the agent'soption of
playingP's mechanismnoncooperatively),it
incentivesfor
createsa set of "countervailing"
the agentto over-ratherthanunderrepresent
its productivity.This relaxationof the downward incentive constraintspermits a reduction in the extent of underproduction.The
patternof incentiveconstraintsand associated productivedistortionsin the FLM context
incentiveconare different,where "upward"
straintsbind with delegation.In such a context, creating countervailingincentives by
raisingthe agent'soutsideoptionturnout not
to be valuablefor the principal.
Finally, MT (2004) compare CMC, DM,
and C in the case of two productiveagents
A1, A, and a perfectly informed M. Their
motivationis to explore the idea that collusion mayrationalizedelegationto intermediaries uninvolved in production. This
necessitatesthe presence of more than one
productiveagent.16
With M perfectlyinformed,there is effectively no asymmetricinformationwithin the
coalition,so inducingchanges in bargaining
powerwithinthe coalitionis not valuablefor
the principaland DM = CMC. Hence delegation is costless relativeto the centralized
alternativewhere P retainscontroland treats
M as an informationprovider.
What is the value of hiringM when there
are two productiveagents?With M perfectly informed, DM reduces effectively to the
case where the principaldeals with a single
consolidatedagent that deliversboth inputs
at a cost equalto the sum of productioncosts
of A1 and A,. Whereasif M is not hired, we
have already explained that centralization
(C) is the better alternativethan delegating
to one of the suppliers(DS). So the value of
hiring(anddelegatingto M) is effectivelythe
same question as the effect of consolidating
diverse suppliers,except that now we must
consider the presence of collusion among
the suppliersunder centralization.MT show
that exactlythe same resultsconcerningconsolidation effects (discussed in section 3.3
above) hold in the presence of collusion as
well. Specifically,DM < C if the two agents
produce substitutes,while DM > C if they
produce complements and some additional
distributionalconditionsare satisfied.In the
latter case of supplier complementarity,
16With a single agent, the coalition of M and the agent
behaves exactly the same way as the agent does, as a consequence of the Coase Theorem which applies when M is
perfectly informed about the agent's cost.
Mookherjee:Decentralization, Hierarchies, and Incentives
therefore,DM is optimalamongall the organizational modes considered: P is strictly
better off hiringM, and then can delegate to
him at no cost.17 But in the substitutescase,
it is strictly better for the principalto not
hire M and contract personallywith both
suppliers.
6. Applicationsto IndustrialOrganization
The literaturesurveyedin this articlehas
so far been primarilytheoretical, without
anysignificantinteractionwith more applied
workon the internalorganizationof firms.In
this section, I shall argue that the issues
addressedby these theories are potentially
relevantto appliedworkon industrialorganization, and then discuss possible reasons
why this potentialhas not yet been realized.
Partof the problemunderlyingthe lackof
interactionwith appliedworkis that most of
the evidence concerning internal organization of firms comprises case studies and
business reports, instead of large scale
empiricaldatasets. Nevertheless, some systematic empirical evidence has begun to
become availablein select contexts, which
are discussedbelow.
6.1 HumanResourceManagement
Practices
Ichniowskiand Shaw(1999, 2003) discuss
how U.S. businessesin the last twentyyears
have dramaticallyincreased their use of
"innovative"human resource management
(HRM) practices.A detailed analysisof the
nature of these practicesand their productivity impact in the context of a sample of
steel finishing productionlines is provided
in Ichniowski,Shaw, and Prennushi (1997)
and Gant, Ichniowski, and Shaw (2002).
Gantet al. (2002) providea detaileddescription of different componentsof these practices, such as problem-solving teams,
17 However this result is driven by the assumption that
M is perfectly informed about agents' costs. If M is imperfectly informed one expects delegation to M to be costly
relative to CMC, but these costs ought to be small if M is
"sufficiently"well-informed.
385
incentive pay, flexible job design, information sharing among workers directly,training, careful hiring and selection, and
employment security. They summarize by
statingthat
. . the underlyingtheme of the value of innovativeHRM practicesis thatplantswith innovative practicesexpecttheiremployeesto do more
problemsolving,both on and off the production
line (Gant,Ichniowski,and Shaw2002, p. 296).
Workersmake decisionsboth individually
and collectively as they work on the line,
thus necessitatinggreaterexchangeof information within work teams, flexibilityof job
workers.
assignmentsand hiringof "smarter"
In contrast, traditional "control oriented"
HRM practices concentrate decision-making responsibilityamonga few hierarchically
superioragents,such as productionforemen
and supervisors. Ichniowski, Shaw, and
Prennushi (1997) estimate the productivity
impact of adoptionof these practicesto be
statistically and quantitatively significant.
They also find significantcomplementarities
between differentcomponentsof the HRM
practices:each componentby itself does not
improveproductivity.
The contrastbetween the innovativeand
traditional HRM practices corresponds
closely to the distinction between mechanisms DS or DM and C describedin previous sections. In the delegationmechanisms,
productiondecisions are made by a team of
agentsA1,A2andM on the basisof intrateam
communicationand group-based incentive
pay for the team leader, who subsequently
allocates production assignments and payments among team members. If the key to
the new HRM practices lie in delegationof
productiondecisions to worker teams, their
value cannot be explained by any conventional incentive-based theory of adverse
selection or moral hazard that conformsto
the Revelation Principle. The outcome of
any such decentralized mechanism can be
replicated by a centralized mechanism
which resemblestraditionalcontrol-oriented
386
Journal of Economic Literature, Vol. XLIV (June2006)
HRM practicesin which each workercommunicates and contracts separately with
central managementthat makes all production decisions. Hence traditionalincentive
theory cannot explainthe superiorperformance of innovative HRM practices. An
appropriatetheory would have to confront
the issues raised in the literaturedescribed
in this essay.
Naturallyenough, it is evidentthat a theory explaining the superiorityof innovative
HRM practices would need to incorporate
costs of communication,informationprocessing, or contract renegotiation. In an
empirical analysisof workplacereorganization in a large cross-section of 300 large
U.S. firms during 1995-96, Timothy F.
Bresnahan, Brynjolfsson,and Hitt (2002)
find evidence of significantcomplementarities between adoption of new information
technology (IT), innovativeHRM practices,
and hiring of skilled workers. Thomas N.
Hubbard (2000) studies the effect of onboard computersin trucking,and its effects
on the value of delegation. Massimo G.
Colombo and Marco Delmastro (2004)
examine the effect of communicationtechnology on whether authorityis delegated to
plant managers.Theoreticalexplanationsof
these empiricalfindingswould need to consider the implications of lowered costs of
communicationand informationprocessing
on the tradeoff between centralized and
decentralized decision-making. Clearly,
models that simultaneously incorporate
informationcosts and incentives are necessary to address this question. The relative
paucity of such models is one possible reason for the limited interchangebetween the
theoreticaland empiricalliteratures.
Nevertheless, the potential relevance of
the theoreticalliteratureis indicatedby the
model with communicationcost described
in section 4. Since centralizeddecision making necessitates greater exchange of information between production agents and
decision makers,it might appearthat lower
communication cost should enhance the
relative profitabilityof centralizeddecision
making. Recall, however, that delegated
decision makinggives rise to a control loss,
which has to be traded off against the
enhancedflexibilityof productiondecisions.
The ability of the principal to moderate
such control losses depends upon whether
the principalcan monitordecisionsmade by
agents in a decentralizedsetting. Advanced
IT enables such monitoring to be carried
out by central managementat much lower
cost, as described by Brynjolfssonand Hitt
(2000).18Hence the net effect of improved
IT on the value of delegation is ambiguous
in general, and consistent with observed
complementaritiesbetween IT and innovative HRM practices. Needless to say, models addressingsuch questions explicitlyare
still awaited.
6.2 RetailFranchising
Benito Arrunada,Luis Garicano,and Luis
Vazquez (2001) describe retail franchise
agreements between automobile manufacturers and dealers in Spain. Retailers are
delegated decisions concerning product
pricing, purchases from manufacturers,
advertisingand after-salesservice.This gives
rise to a classicproblem of "doublemarginalizationof rents"acrossretailand manufacturing stages: decisions made by retailers
affectprofitsearnedby manufacturers.They
also give rise to horizontal externalities
acrosssales effort of differentdealers:some
customers gather information from one
dealer and purchasefrom other dealers. To
control these externalities, manufacturers
retaincertainmonitoringrights:
18 For example, they write that "Computer-basedsupply chain integration has been especially sophisticated in
the consumer packaged goods industries . . . Because
many consumer products have long shelf lives, retailers
tended to buy massive amounts during promotional periods, which increased volatilityin manufacturingschedules
and distorted manufacturers' view of their market. In
response, manufacturerssped up their packagingchanges
to discourage stock-piling of products and developed
internal audit departments to monitor retailers' purchasing behavior for contractual violations"(Brynjolfssonand
Hitt 2000, p. 28).
Mookherjee:Decentralization,Hierarchies, and Incentives
...
they are authorized to directly inspect the
inputs of the dealership(machinery,personnel
etc.) and to measure dealer performance in
termsof achievementof salestargets.Most contracts also allow manufacturersto monitorcustomer satisfaction,as measuredby polls. Finally,
most manufacturersmay monitordealersfinancial performance by regularly auditing their
financial statements. Concerning termination
rights,all contractsexplicitlyassignmanufacturers the rightto terminatethe contractswhen the
dealer infringes certain duties (Arrunada,
Garicano,and Vazquez2001, p. 264).
Traditionalincentivetheorycannotexplain
why decisionsover productpricing,purchases from manufacturers,advertisingor aftersales serviceshouldbe delegatedto retailers
by manufacturerswho are clearlyin a position to design the franchisingrelationship.
When the RevelationPrincipleapplies,delegated arrangementscannot outperform a
centralizedarrangementwhere all decision
rights are retained by manufacturers.An
obvious explanation of such a delegation
arrangementis that it is either impossible,
too expensiveor time consumingfor retailers
to communicateall relevantlocalinformation
to manufacturers.
At the same time, in order
to limitcontrolloss, manufacturersretainthe
rightto monitordealersin a varietyof ways.
Arrunada,Garicano,and Vazquez (2001)
explainthe prevalenceof delegation by the
presence of "two-sidedmoral hazard,"i.e.,
the possibilityof opportunisticex post renegotiation of contracts by manufacturers.
Consistent with this hypothesis, they find
that manufacturerstend to delegate more to
retailerswhen manufacturersare more reputable and thus less subjectto renegotiation
constraints.Nevertheless,the decisions that
are subject to more or less delegation concern monitoring and enforcement of contractualterms.Decisionsconcerningpricing,
advertisingor after-salesservice tend to be
uniformlydelegated,even amongstthe most
reputed manufacturers.The theoreticalliteraturedescribedin this essay will be needed to explainthe widespreadprevalenceof
such kinds of delegationeven for reputable
387
manufacturersfor whom the moral hazard
problemis of the "one-sided"variety.
7. ConcludingComments
In summary,the most importantlacunaof
existingtheoreticalincentive-basedliterature
is thatit focuseson costs ratherthanthe benefits of delegation.The latterare difficultto
incorporateinto traditionalcontracttheory.
Perhapsthe most importantbenefit of delegationis the distributionof informationprocessingtasks,but no progresshas occurredin
theories that marryinformationprocessing
costs with incentives. Some progress has
been possiblewith communicationcosts and
simple measuresof contractcomplexity,but
these need better foundations.
What have we learned from the existing
literature?It identifies a number of potential costs of delegation: moral hazard for
intermediariesowing to noncoincidence of
their own objectives with the principal's,
and their monopsony power over subordinates. These can result in production distortions (insufficient sourcing from
subordinates), cascading of information
rents acrossverticallayers,and problemsof
coordinatingdifferent horizontalbranches.
If agentsdo not collude,these agencycosts
of delegationcan be avoidedif (and only if)
the principalcan monitorsubcontractcosts
or quantities,if contractsflow downthe hierarchy,and agentsare risk-neutral.If any one
of these conditionsdo not hold then agency
costs cannotbe avoided.The only significant
problem pertains to vertical control loss; if
they can be avoided (i.e., under the above
mentioned conditions) then incentive considerations do not complicate horizontal
coordinationacross branches of the hierarchy: "group"-basedincentive contracts can
be designed to costlessly internalize these
horizontalexternalities.On the other hand,
managerialrisk aversionor limited capacity
for principalsto monitorlocal conditionsor
agent decisionscan cause significantcontrol
losses from delegation, that grow with the
388
Journal of Economic Literature, Vol. XLIV (June2006)
size and complexityof the organization.This
disprovidesan explanationof organizational
economies of scale, i.e., why larger firms
tend to be more "bureaucratic"
and less able
to controlcosts.
If agents collude, centralizationis also
subject to unobserved side contracting
among agents, limiting the ability of the
principal to moderate "control loss."
However, centralizationpotentially allows
greater control over side contracting outcomes by the principal offering outside
options to subordinatesthat limit monopsonypower of intermediaries.Dependingon
the precise distortions engendered, this
added dimensionof control may or may not
be valuable. Overall,the presence of collusion amongagents enlargesthe rangeof circumstances where delegation implements
optimalallocations.
There are numerous open questions and
fruitful avenues for future research. I conclude by listing some of these.
First, a better understandingof effects of
collusion is still needed. The few papers on
this topic emerge with different results the
intuitive basis for which is not very clear.
One hopes a more unified perspectivewill
emerge in due course. There is a need to
explore implications of different formulations of side-contracting,e.g., more general
assignmentof bargainingpowerwithincoalitions, or alternative timing assumptions.
Baron and Besanko (1999) provide an
intriguingmodel in which agentsthemselves
decide ex ante whetherto consolidatethemselves into a single entity, a decision which
the principalobservesandtakesinto account
before offeringa contract.In the models we
described,the principalcan anticipatea particularpattern of side contracting,but cannot observe whether or not the agents
actually do side-contract.In contexts with
more agents and verticallayers,the possibility of collusion-within-collusion further
complicatesthe analysis.
Second,moreeffortneeds to be devotedto
explainingthe potentialbenefitsof delegation.
Models integratinginformationprocessing
or communicationcosts with incentive considerations are needed to provide a fullblown theory of the trade-off between
centralization and decentralization. This
would render the theory useful in applied
work assessing the effectiveness of innovative human resource management practices and their complementaritywith new
informationtechnology.
A third possible avenue would consider
applications and extensions to contexts
involvingmore productiveagentsand a richer specificationof the productiontechnology. Questionsconcerningthe optimalshape
of hierarchiescan then be addressed, e.g.,
trade-offsbetween spanof controland number of vertical layers, how to group agents
within departments, organizationaldiseconomies of scale, and the advantagesof nonhierarchicalorganizations.One hopes that
both theoryand empiricaldatasetsregarding
these organizationalattributescan be developed interactively,permittingbetter understanding of their productivityimplications,
and how they respondto changesin market
competitionor informationtechnology.
REFERENCES
Aghion,Philippe,and Jean Tirole. 1997. "Formaland
RealAuthorityin Organizations."
Journalof Political
Economy,105(1):1-29.
Aoki,Masahiko.1990. "Towardan EconomicModelof
the JapaneseFirm."Journalof EconomicLiterature,
28(1): 1-27.
Arrow,Kenneth. 1974. The Limits of Organization.
New York: W Norton.
W.
Arrunada,Benito, Luis Garicano,and Luis Vazquez.
2001. "Contractual
Allocationof Decision Rightsand
Incentives:The Case of AutomobileDistribution."
Journalof Law,Economics,and Organization,17(1):
257-84.
Baliga, Sandeep, and Tomas Sjostrom. 1998.
"Decentralization and Collusion." Journal of
EconomicTheory,83(2): 196-232.
Baliga,Sandeep,and TomasSjostrom.2001. "Optimal
Design of Peer Review and Self-Assessment
Schemes." RAND Journal of Economics, 32(1):
27-51.
Baron,DavidP. 1985. "NoncooperativeRegulationof a
Nonlocalized Externality." RAND Journal of
Economics, 16(4): 553-68.
Baron,DavidP.,andDavidBesanko.1992."Information,
Control, and OrganizationalStructure."Journal of
Hierarchies,and Incentives
Decentralization,
Mookherjee:
Economicsand ManagementStrategy,1(2):237-75.
Baron, David P., and David Besanko. 1999.
"Informational Alliances." Review of Economic
Studies,66(4): 743-68.
Beaudry,Paul, and Michel Poitevin. 1995. "Contract
Renegotiation: A Simple Framework and
Implications for OrganizationTheory." Canadian
Journalof Economics,28(2): 302-35.
Bolton, Patrick,and MathiasDewatripont.1994. "The
Firm as a Communication Network." Quarterly
Journalof Economics,109(4):809-39.
Bresnahan,TimothyF., Erik Brynjolfsson,and Lorin
M. Hitt. 2002. "InformationTechnology,Workplace
Organization,and the Demand for Skilled Labor:
Firm-Level Evidence." Quarterly Journal of
Economics,117(1):339-76.
Brynjolfsson,Erik, and Lorin M. Hitt. 2000. "Beyond
InformationTechnology,Organizational
Computation:
andBusinessPerformance."
Transformation
Journalof
EconomicPerspectives,14(4):23-48.
Calvo, Guillermo A., and Stanislaw Wellisz. 1978.
"Supervision,Loss of Control, and the Optimum
Size of the Firm."Journal of Political Economy,
86(5): 943-52.
Celik,Gorkem.Forthcoming."MechanismDesignwith
Reviewof EconomicStudies.
CollusiveSupervision."
Chandler, Alfred D. 1962. Strategy and Structure.
Cambridge:MIT Press.
Colombo, MassimoG., and Marco Delmastro.2004.
"Delegationof Authorityin BusinessOrganizations:
An EmpiricalTest."Journalof IndustrialEconomics,
52(1): 53-80.
Cremer,Jacques,and MichaelH. Riordan.1987. "On
Governing MultilateralTransactionswith Bilateral
Contracts."RAND Journal of Economics, 18(3):
436-51.
andCommunication
Dessein, Wouter.2002. "Authority
in Organizations."
Reviewof EconomicStudies,69(4):
811-38.
Faure-Grimaud,Antoine,and David Martimort.1999.
"Political Stabilization by an Independent
LondonSchoolof Economics.Mimeo.
Bureaucracy."
Faure-Grimaud,Antoine,and David Martimort.2001.
"On Some Agency Costs of Intermediated
Contracting."EconomicsLetters,71(1): 75-82.
Faure-Grimaud,Antoine, Jean-JacquesLaffont, and
David Martimort.2000. "A Theory of Supervision
with Endogenous TransactionCosts." Annals of
Economicsand Finance,1(2):231-63.
Faure-Grimaud,Antoine, Jean-JacquesLaffont, and
David Martimort.2003. "Collusion,Delegation and
Supervision with Soft Information." Review of
EconomicStudies,70(2):253-79.
Gant,Jon, Casey Ichniowski,and KathyrnShaw.2002.
"SocialCapitaland OrganizationalChangein HighInvolvementand TraditionalWork Organizations."
Journal of Economics and ManagementStrategy,
11(2):289-328.
Gilbert, RichardJ., and Michael H. Riordan. 1995.
Products:A Comparative
"Regulating
Complementary
InstitutionalAnalysis."RANDJournalof Economics,
26(2):243-56.
Hammer, Michael M., and James Champy. 1993.
389
Reengineeringthe Corporation.New York:Harper
Collins.
Hart, Oliver. 1995. Firms, Contracts,and Financial
Structure.Oxfordand New York:OxfordUniversity
Press and ClarendonPress.
Heal, Geoffrey M. 1971. "Planning, Prices, and
IncreasingReturns."Review of Economic Studies,
38(115):281-94.
Horngren,CharlesT., and George Foster. 1991. Cost
Accounting:A Managerial Emphasis. Englewood
Cliffs, N.J.:Prentice-Hall.
Hubbard, Thomas N. 2000. "The Demand for
MonitoringTechnologies:The Case of Trucking."
QuarterlyJournalof Economics,115(2):533-60.
Hurwicz,Leonid. 1960. "Optimalityand Informational
Efficiency in Resource Allocation Processes," in
MathematicalMethods in the Social Sciences. K.
Arrow, S. Karlin, and P. Suppes, eds. Stanford:
StanfordUniversityPress,27-46.
Hurwicz,Leonid. 1972a. "On InformationallyDecentralizedSystems,"in Decision and Organization.R.
Radner and C. McGuire, eds. Amsterdam:NorthHolland,297-336.
Hurwicz, Leonid. 1972b. "On the Dimensional
Requirements of Informationally Decentralized
Pareto-SatisfactoryProcesses."Paper presented at
the Decentralization Conference, Northwestern
University.
Hurwicz, Leonid. 1973. "The Design of Mechanisms
for Resource Allocation." American Economic
Review,63(2): 1-30.
Hurwicz, Leonid. 1986. "Incentive Aspects of
Decentralization,"in Handbook of Mathematical
Economics,Vol.3. KennethJ. Arrowand MichaelD.
Intrilligator, eds. Amsterdam: North-Holland,
1442-81.
Ichniowski, Casey, and Kathryn Shaw. 1999. "The
Effects of Human Resource ManagementSystems
on Economic Performance: An International
Comparison of U.S. and Japanese Plants."
ManagementScience,45(5): 704-21.
Ichniowski,Casey,and KathrynShaw.2003. "Beyond
Incentive Pay: Insiders' Estimates of the Value of
Complementary Human Resource Management
Practices."Journalof EconomicPerspectives,17(1):
155-80.
Ichniowski, Casey, Kathryn Shaw, and Giovanna
Prennushi.1997. "The Effects of Human Resource
ManagementPracticeson Productivity:A Study of
Steel FinishingLines."AmericanEconomicReview,
87(3): 291-313.
Laffont, Jean-Jacques,and David Martimort.1998.
"Collusion and Delegation." RAND Journal of
Economics,29(2): 280-305.
Malinvaud,Edmond.1967. "DecentralizedProcedures
for Planning,"in ActivityAnalysisin the Theoryof
Growth and Planning. E. Malinvaud and M.
Bacharach,eds. London:MacMillan,170-208.
Marschak,Thomas, and Stefan Reichelstein. 1995.
"Communication Requirements for Individual
Agents in Networks and Hierarchies," in The
Economics of Informational Decentralization:
Complexity, Efficiency, and Stability. John O.
390
Journal of Economic Literature, Vol. XLIV (June 2006)
Ledyard,ed. Boston:KluwerAcademic,311-46.
Marschak,Thomas, and Stefan Reichelstein. 1998.
"Network Mechanisms, InformationalEfficiency,
and Hierarchies."Journal of Economic Theory,
79(1): 106-41.
McAfee, R. Preston, and John McMillan. 1995.
Diseconomies of Scale."Journalof
"Organizational
Economicsand ManagementStrategy,4(3):399-426.
Melumad,Nahum D., Dilip Mookherjee,and Stefan
Reichelstein. 1992. "A Theory of Responsibility
Centers."Journal of Accounting and Economics,
15(4):445-84.
Melumad,Nahum D., Dilip Mookherjee,and Stefan
Reichelstein. 1995. "HierarchicalDecentralization
of Incentive Contracts." RAND Journal of
Economics,26(4): 654-72.
Melumad, Nahum D., Dilip Mookherjee,and Stefan
Reichelstein. 1997. "Contract Complexity,
Incentives,and the Valueof Delegation."Journalof
Economicsand ManagementStrategy,6(2): 257-89.
Mookherjee, Dilip, and Stefan Reichelstein. 1997.
"Budgetingand HierarchicalControl."Journal of
AccountingResearch,35(2): 129-55.
Mookherjee, Dilip, and Stefan Reichelstein. 2001.
"Incentives and Coordination in Hierarchies."
Advancesin TheoreticalEconomics,1(1).
Mookherjee, Dilip, and MasatoshiTsumagari.2004.
"The Organizationof SupplierNetworks:Effects of
Delegation and Intermediation." Econometrica,
72(4): 1179-1219.
Mount, Kenneth, and Stanley Reiter. 1974. "The
InformationalSize of Message Spaces."Journal of
EconomicTheory,8(2): 161-92.
Mount, Kenneth,and StanleyReiter.1995. "ATheory
of Computingwith Human Agents."Northwestern
University Department of Economics Working
Paper.
Myerson, Roger B. 1982. "Optimal Coordination
Mechanisms in Generalized Principal-Agent
Problems."Journal of Mathematical Economics,
10(1):67-81.
Poitevin, Michel. 1995. "ContractRenegotiationand
Organizational
Design."CIRANOWorkingPaper95-3.
Poitevin,Michel.2000. "Canthe Theoryof Incentives
Explain Decentralization?"Canadian Journal of
Economics,33(4): 878-906.
Radner, Roy. 1992. "Hierarchy:The Economics of
Management."Journal of Economic Literature,
30(3): 1382-1415.
Radner,Roy.1993."TheOrganizationof Decentralized
Information Processing." Econometrica, 61(5):
1109-46.
Radner, Roy, and Jacob Marschak.1972. Economic
Theoryof Teams.New Haven:YaleUniversityPress.
Roider,Andreas.2003. "Delegationof Authorityas an
Optimal(In)completeContract."Universityof Bonn
DiscussionPaper.
Segal, Ilya. 1999. "Complexityand Renegotiation:A
Foundation for Incomplete Contracts."Review of
EconomicStudies,66(1):57-82.
Segal, Ilya. 2001. "CommunicationComplexityand
Coordinationby Authority."
Universityof California,
Berkeley.Mimeo.
Severinov, Sergei. 1999. "Optimal Organization:
Centralization, Decentralization or Delegation?"
Universityof Wisconsin,Madison.Mimeo.
Spence,A. Michael.1975. "TheEconomicsof Internal
Organization:An Introduction."Bell Journal of
Economics,6(1): 163-72.
Tirole,Jean.1986."Hierarchiesand Bureaucracies:On
the Role of Collusionin Organizations."
Journalof
Law, Economics,and Organization,2(2): 181-214.
Tirole,Jean. 1999. "IncompleteContracts:Where Do
We Stand?"Econometrica,67(4): 741-81.
van Zandt,Timothy.1996. "DecentralizedInformation
Processing in the Theory of Organizations,"in
ContemporaryEconomic Development Reviewed,
Vol. 4: The Enterpriseand its Environment.Murat
Sertel,ed. London:MacmillanPress, 125-60.
van Zandt, Timothy. 1997. "Real-TimeHierarchical
ResourceAllocation."PrincetonUniversity.Mimeo.
Williamson,Oliver. 1967. "HierarchicalControl and
Optimal Firm Size."Journal of Political Economy,
85(2): 123-38.
Williamson,Oliver.1985. TheEconomicInstitutionsof
Capitalism.New York:Free Press.