Public Disclosure Authorized

Public Disclosure Authorized

Public Disclosure Authorized
Public Disclosure Authorized

POLIcy

RESEARCH

WORKING

PAPER

1345

TheDecentralizationQusinabt
of Public Services

decenralizationin
govemnment are questions

about the allocationof

Lessonasfrom the Theory of the Firm

corntrolrights. How much to

decentralize
dependson
which level or govertment

Jacques Cremer
Antonio FEstache
Paul Seabrigbt

Backgro:d paper for WorldDevelopmnt ReIpat1994

The World Bank

Officeof the VicePre-ident
Development

August1994

EcDonmics

wfillhave the most incentive

to bring about desired

OUECOmes.

POLICY RESEARCH WORKING PAPER 1345

Summary findings
The literature on the theory of the firm is rich in
theoretical and practical insights.The key messagesin
this overview are the following:
The modem theory of the firm provides many
insights into poiitical organization, for political
jurisdictionscan be viewed as pseudo-firmsthat provide
services and that group together various kinds of
decisionmakingactivities,
- Questionsabout decentralizationin government are
questions about the allocation of control rights. If
contractual relations were complete, it would not matter
whether power were decentralized, as contracts would

specifyeverything to be done at each level of

government. There would be no need for discretion.
* How much to decentralizedepends on which ievel
o&government will have the most incentiveto bring
about desired outcomes.Centralizedgovemments may
be better at coordinating things but tend to be less
accountable than decentralizedgovemments (with
important exceptions).
* The organizationaldesign of government affects not
only incentivesto make decisionsbut also incentivesto
gather the information on which those decisionsare
based.

This paper - a productof the Officeof the VicePresident,DevelopmentEconomics- is one in a seriesof backgroundpapers
preparedfor WorldDeuelopment Report 1994 oninfrastructure.
Copiesofthepaperareavailable
freefromtheWerldBank,1818
H StreetNW,Washington,DC 20433. PleasecontactAntonioEstache,roomQ7-123,extension81442 (51 pages).August1994.

The PolicyResarch WorkingPaperSeriesdisseminatesthe findhs of work i progressto enoiaage the xchangeof ides abous
deelopmcnt issus An obi clw of the sries is to get thefcdings out qukly, een ifbe presenation areksMuy
poIshed The
paprs carry the namesof the aths and shod be sd and cited accordngly.The nungs, itierpresatimn&
and condiorms are the
authors'oam and shouldnoabe artribted to the WorldBank its ExecutiueBoardof Direaors,or any of as membercoun&res.

Producedby the PolicyRcsearchDisseminationCenter

Lessons from the Theory of the Firm

Jacques Cremer

(Institut d'Economie Industrielle, Universitd des Sciences Sociales de Toulouse)

Antonio Estache
(ITheWorld Bank and ECARE)
and
Paul Seabright
(Cambridge University)

of a Public EconomicSeminarat the WorldBank

We would like to thankparTficipants
and RichardBird for usefulcommentsand suggestions.

itp:

The large expansion of government powers in some of the Western countres in the
post World War II period was given theoretical backdngby the theory of market failure, and
the achievements of the studies of externalities, public goods, and optimal taxation. Many of
these studies have found their way into the analysis of economic development issues. During
the 1980s, these theories, or at least their applications in the public debate, have been justly
criticized for their lack of attention to government failures. Moreover, in recent years
government intervention in the economy has come to be regarded with more suspicion than
in the past and an increasing number of policymakers in countries as different as Argentina,
Mexico, China or Hungary have adopted a more pragmatic approach to development favoring
reliance on the market to allocate goods and service.
The theoretical intuition behind the superiority of market allocation is wel known.
Given the efficiency of market exchanges, agents should be left to make as many decisions
for themselves as possible. When these decisions are transferred to the public sector, there
are asymmetries of information and the associated opportunities for rent seeking actvities,
and the loss of efficiency may be large enough for even the poorest members, who are
supposed to benefit from these policies, to lose out
These insights have been transfered to the orgnization of the public sector itself. if
political economy teaches us that as many decisions as possible should be decentalized
through the market, doesn't it follow that within the public sector deisions should also be
decentralized as much as is feasible? What one cannot leave to private initiative should be

left to the lowest possible level of government.

Things are not that simplebecausethese policyinsightsare based on a very

unbalancedtheoreticalunderpinning.Economistshave a good theoryof the functioningof
markets,their benefitsand their failures. Our understandingof govemmentsand of public
administrationis much moreimperfect,in great part becausethe topic is more difficult.
When one comparesmarketsto public provision,one comparessomethingunderstood
relativelywell to somethingone does not understandvery well at all. The situationis even
more complexwhen one wants to comparetwo types of publicprovision, for instancea very
centralizedgovernmentwith a federationthat leaves muchpower to local units. One
comparestwo situationsof which neitheris understoodvery well.
We believe that improvementsin the analyticaltreatmentof the iternal functioning
and of the optimalstructureof governmentshouldbe accessiblethroughthe tansfer of the
intuitionsdevelopedin the theory of the firm in the last twentyyears. Somework bas been
done on this topic, but much more can be done. In particular,the very importantadvancesof
contracttheory in the last ten years have not been adequatelyintegratedinto the discussion.
The aim of this paper is to open channelsof thoughtsalong these lines, with a special
emphasison the provisionof infrastructure.This is not to implythat results from the theory
of the firm can be directly imported. The public sector has very specific fears,

some of

which will be discussedbelow. What can be importedare intuitionsand tchniques for the
study of organizationalproblems.Furthermore,normativeissues are of much greater
importancein the choice of optimalpoliciesdta appearsfrom the summarythat follows, and
we neglect them for the majorpart of the sequel.

The paper is organizedas follows. In section2, we describea simple model that lies,
at least implicidy, at the basis of much of recent policydiscussion.In section3, we enich
this modelthrough the use of contracttheory, as developedin the 1970's and the begnning
of the 1980's. This enablesus to discussthe problemsof transfersamongjurisdictionsas
well as aspects of financing.In section4, we discussthe shortcomingsof these approaches
and show that recent developmentsin contracttheoryand the theory of

ions

incompletecontracts and renegotiation- providea naturallanguageto speak of some of

these issues. The sequelof the paper is used to developapplicationsof this methodology.In
section5 we try to provide some preliminaryreflectionson the conceptof political
jurisdictions,parallelingthe famous "nature of the finn" debate in organzation theory.
Finally, section6 providesa pot pourri of applications.

2. The receivedwisdomon decentralization

This sectionsummarizesthe receivedwisdomon decen

on of responsiilities

for regulationand provisionon goodsand servicesin the standardpublic economics

literatu. It outlines the key assumptionsunderlyingthe mainpolicy guidelinesof the
standardpublic financeliterature on fiscal federalism,focusingin particularon those that can
much moreeasily be addressedby relying on a contractualapproachto the issues raised by
the structureof government.

0<.f--;2.1 A Quick Sure

The standardpublic financeliteratureprovidesboth ex-anteand ex-postargumentsfor

decentmalization.
The classicex-ante case for decentralizationis providedby Oatesin 1972.
He points out that public goods can have spatialcharacteristicsreflectingthe specific
boundaries-global,regionalor local-of the benefitsthey provideto their consumers.Oates'
decentralizationtheoremn
suggeststhat, for a given informationstructureabout demands,each
tyeofgood shouldbe providedby a level of government- say central, state and municipal-enjoyinga comparativeadvantagein accountingfor the diversityof preferencesin its choice
of servicedelivery. Club theory providesrelated argumentsto make the case for an optimum
size of local authorities.
Based on these demandelementsand on the recognitiontha spilloversand other
forms of externalitiesin productionand consumptioncan matter, a 'standard lst o
desirable expenditureassignments-interms of provisionand/or responsibility-providesthec
basis of the policy guidelinesfor countriesconsideringa decnrliainof

government

structures.For instance,airways and railways shouldbe a centralconcern in view of the

nationalscope of their benefitsand costs, whileprovisionof servicesin highwayscould be
national,state or local-reflectingthe balancebetweenspilloversand local preferences.
Where does the comparativeadvantageof the variousgovernmentlevels come from?
it reflects an assumedgovernmentfailure. In one version, the failure stemsfrom an
informnation
asymmetrybetweencentral and subnationalgovernments.Cenftralgovernments
are assumedto have some institutionaldisadvantagein pickingup local informationon

preferences.More decentralized governments have better knowledge of local preferences,

either in the sense of having access to information denied to central government, or in the
sense of observing preferences with less noise. In a sMMod
version of the government failure
argument, the central government is assumed to be institutionally compelled to adopt a
uniformity constraint in service delivery. Even if it were able to identify the diversity of
demands across regions and municipalities, it could not diversify its supply accordingly.
These failures reveal the second best nature of the decentalization argument.
The classic ex-post case for decentralization was made by Tiebout in 1956. liebout's
point was that population mobility reinforces ex post the ex-ante case based on superior
information or flexibility of local government. This mobility should eventnlly ensure a
perfect match between public sector supply and consumer preferences. Consumers vote with
their feet and move to the municipality or region which provides them with what they want
both in terns of services and in terms of financing mix - taxes, user fees, etc. In sum, it
explains how jurisdictions of optimal size come to be established or how individuals take
membership in these "local clubs".

2.2

Shortcomings of the Tiebout approach

Since most analysts of fiscal federalism have used generalized versions of the liebout
model in which the incentives of the different districts are fundamentally linked to the abit
to attct

new inhabitants and/or new capital, it is worth focusing on some of its limiation

as a model of political decentralization in practice:

1)

Ihe role of the nationalgovemmentis not endogenous,or rather it can only be

consideredas trying to circumventthe negativeaspectsof the competitionbetween

jurisdictions.

2)

The governments'objectivesin the Tieboutmodelare assumedto be profit

maximization(see Bewley, 1981), withoutany discussionof how such objectives

mightaise from a realisticmodel of the politicalprocess. The latter issue is generAly
only addressedexplicitlyin public choice models.
3)

Government failures (including corruption) can arise at the subnational level

just as much as they arise at the central level.

4)

The actual mobilityof factorsis generallymuch more limitedthan assumedby

Tiebout and the effectof positivedecisionsby the governmentof one districtwill

influencethe competitivepositionof the districtonly in the long run. From an
empiricalpoint of view, as pointed out by Walsh(1992), the main relevanceof the
modelis in a metropolitansetting-i.e. withinsuburbslocated near other communides
and hencewith exit optionsthan in communitieswith no or fewer exit options.
5)

Mobilitycan give rise to externalitieswhichare not accountedfor in the

design of the financingmodeof thejurisdiction.

6)

In practice,politiciansoften try to favor specificinterest groups in their

districts (for instance, US congressmenwill try to introducein the law tax breaks
designatedto favor specificfirms whose headquartersare in their districts). Tiebout
theory would predict more stress on favoringthe transferof fators of productionto
the district.

7)
*

Respnsibilities are actuallyoften shared by various levelsof govemnment.

For

~~~instance,
governmentsmustdecide on aggregateroad expendituresinia jurisdiction,

and on the allocfon of this expenditure among different types of -ads. The firt

decisioncan be left to the federalgovernmentand the secondto the local government.

Similarly,the federalgovernmentcan set pollutionstandards,and the local
governmentcan be in charge of monitoringtheir implementation(see Estache and
Zheng (1992)and (1993)and Jack (1992)).The difficultyin the study of this
rLesponisibility
sharing mechanismstems from the modellingof the incentivesof the
different levels of governments.

In additionto appreciatingthe particularshortcomingsof the Tieboutframework,it is

'_ Hr '~~~~~~~~~

importat to appreciatewhat the underlyingapproachcan and cannotbe expectedto achieve.

Tiebout's basic insight was that populationmobilitycan enable governmentsto overcomethe
well-knownproblemof inducingcitizensto reveal theirpreferencesfor public goods. This
problemarises if governmentshave no way to exclude from consumptionthose citizenswho
state that they would prefer not to pay for the provisionof a certainpublic good. If goods
are "local*, however, they can be providedonly for the citizensof a given localjurisdiction
(togetherwith a local tax), therebyexclucingfrom consumptionall iion-residentsand
allowingcitizensin their choice of residenceto select theirpreferred combinationof public
goods. In the limit, when mobilityis costlessand various other technicalconditionsare
satisfied,the public goods problemdisappearsand the allocationof resourcesis efficient.
The TLiebout
approachhas nothingto say, however,about how the appropriatelocal

bundlesof publicgoods mightcome to be established.This might, of course, happen as the

result of spiritedcompetiton betweenindependentlocal govemments(though to do so they
wouldhave to be profit maximizing,and it is far from clear how sucha go;l might emerge
from any plausiblepoliticalprocess). But there is no reasonin principlewhy a central
governmentcould not also put the Tiebout insight to use, and differentiateits supplyof
publicgoods by localityin order to inducethe revelationof preferences.In short, the
Tiebout modeldescribesthe virtues of local differentiadonof publicgoods supply, not the
virtuesof decentralizationof power. In order to understandthe pros and cons of
decentrdlizationas such, it thereforeneeds sapplementingby an accountof the respective
abilitiesof centra and local governmentsto undertakethe kind of local differentiationwhose
virtues the Tieboutmodel has described.
Such an account wouldreturn us essentiallyto the ex an approachto
decentrlization. Underlying many scholarly as well as popular accounts of

ntralization

is

a view of the essentialinformationaladvantagesof local government,and an implicit model

with the followingelements':

1) A oDuntryis dividedinto n districts,and there are no other politicaldivisions,

either above the district or below it.
2) Each district has a local governmentwhich has knowledgeof the relevant
characteristicsof that district.
3) Each local governmenttakes decisionsin the interest of its own disktict,without
A model with the some of the features
Groot

(1988).
9

discussed

below in presented

in Do

takng into account the desiresof other districtsexcept if this is imposedby the
federalgovernment.
4) There is a centra governmentthat takesdecisionsin the interest of the countryas
a whole (whateverexactlythis may mean).
5) The central governmentknows only the average of the characteristicsof the
districts. Consequentlyit can do no better than implementa uniformpackageof
public goods for all districte

For each issue, the power to take decisionsis given either to the central governmenit
or to the local governments.
The types of issuethat we have in mind are the levels of
expenditureon elementaryeducation,pollutionstandards,the mileageof new roads to be
built.
Such a modelyields results that are quite close to the 'common' wisdomon
decentralization: there is a fundamentaltradeoffbetweendecisionstha are more appropriate
to local conditionsand the internalizationof externaities. There are of course other benefits
from coordinateddecisionsthan simply the internalizationof externalities:those due to the
exploitationof increasingreturns, for instance.
The drawbackof the model, however,is that it leaves the informationaladvantageof
local governmentessentiallyunexplained.To explorethis issue further we turn to a

2 This inference
is not correct
in a dynamic model in which the central.
government,
while initially
uninformed,
experimentu
with differentiated
public
good supply
and thereby
learns
about the dif ferent
characteriotics
of the
districts.
This would be similar
to a tatonnement
proceus
in Walrasian
modeim.
We are not aware of such a dynamic model explicitly
developed
in the public

finance literature.

10

discussionof the implicationsof recent developments

in contracttreory.

3.

A simplecontractmodelof decentralizadon

In this section,we revisitthe precedingmodelwith the helpof asimple ontactual

approach. This allowsus to enrichthe descriptionof the communicadonsbetweenfederl
and local governments.We begin by a brief expositionof revelationcontracts,and apply the
insightsto the problemof decentralization.

3.1

Asymmetry of information and contracts

It is hard to overestimatethe revolutionthat tookplace at the beginingof the 1970sin

economictheory. For a longtime, economistshad talkedinformallyaboutthe strategicuse
of information(see for instancethe famousplanningdebateof the 1930'sbetwe Hayek and
Lange(1938)),but it was not until over thirty yearslater that the strategicuse of inhrmation
wa-sfor the first time formallyintegratedinto our mrodels.This revolutionchangedour
understandingof markets(followingthe path breakingcontributionsof Akerlof(1970)and
Spence(1973))and of incentivesin organizations(followingthe no less pathbreaking
contribudonsof Groves(1973)and Weitzman(1974)).
The basicframeworkis very simple.A fedeal governmentwantsa municipalityto
builda sewagesystemin a poor neighborhood.It has no administativepowerto forceit to
do so, and mustoffer a cost sharingformula.Typically,this formulawil be part of a

~~~~~~~11

F.-

contract that linls the amount paid by the federal go.ernment to the charcteistics of the

sewagesystemthat is built. If the costs of buildingthe sewagesystemare better knownto

the local governmentthan to the federalgovemment,the former will be able to extract an
informationalrent. This informationalrent is not only a transfer from one governmentlevel
to another, it has efficiencycosts: it order to reduce it the federalgovenimentwill build a
cost sharingformula that will inducethe local governmentto choosea system that is not
optimallydimensioned.
It is possibleto show that the federalgovernmentof our examplenever looses
anything by offering a contract of the following form: "Tell me everything you know that is

relevantto the estimateof cost. As a functionof what you tell me I will, accordingto a
predeterminedformula, transferfunds to you and order you to build a systemof
predeterminedcharacteistics".
Furthermore,it is possibleto choosea contractsuch that the local governmentwil
truthfullyreveal its information.The equivalenceof any cost sharingformula to one of these
"revelationmechanisms"in which the informedparty is given incentivesto announceits
informationtruthfully, is called the revelationprinciple.
The intuitionbehindthis principleis that, given any initialcontact, the federal
governmentcan calculateunder what circumstancesthe local governmentwould have
incentivesto misreportits information.It can then modifythe contractso as to increasethe
tansfers to which the local governmentwouldbe eligibleunderthese cicumstnces, so as to
leave it no worse off if it tells the truth. The increasedtranfers which it must therefore
implementare a necessarycost of the initialasymmetryof information,and cannot be

12

reducedby anystrategicmanipulation.
The validityof theprincipledepends,however,on
theassumptionthatthefederalgovernment
knowsenoughaboutthepayoffsof the local
governmentto calculateits incentivesfor lying;even moreimportantly,
it dependson the
assumptionthatit has no difficultycommitting
itself to a truth-inducing
contract.We shall
discussthe limitationsof theseassumptions
below, but shallconcentrate
for the momenton
thetheory'sstrengths.
Manyvariationson this themeare possible,of whichone is speciallyrelevantto our
purposes.The revelationprinciplecan easilybe adaptedto a situationwheretheconstruction
of the sewagesystemextendsovera longperiodof time andwheretheinformation
becomes
availableto the localgovernmentprogressively.
Then,any contractbetweenthe twoparties
is equivalentto a contractin whichthelocal government
informsthefederalgovernmentas
soonas information
becomesavailable.
The fact chatanycontractis equivalentto a revelatonmechanismn
is an extremely
usefultechnicaltool. It also bringsout in verystarkformthe limnitations
of thetheory.The
Constitution
of the UnitedStatescanbe consideredthe foundingcontractof tha country.
Usingcontract:
theoryto analyzethepresent,one wouldneedto assume(implicitly)thatthe
FoundingFathershadforeseenpresentconditionsas one of the contingencies
in the conftact
theydrewup.

13

In light of contract theory, the basic modeltha we have sketchedabove can be

substanti-ally
modified.There is no reason to assumethat the federa governmentwill accept
the fact that the infornnationof the local governmentis niotaccessible.It can propose
contracts (implicitor explicit)that ask the local governmentsto reveal their informationand
commitit to some actionsas a functionof this information.All the appartats of contract
theory can be used to study the end resulL
A revelationmechanismneed not only be interpretedas a contract signedbetweenthe
federalgovernmentand the local governments.It can also be thoughtof as a systemof
taxation,where the tax is leviedaccordingto the observableactionsand characteristicsof the
distri'ct.The same fundamentaltrade-offwill be involved:efficiencyrequires large
informationalrents to the more fortunatedistricts.This makesa very sftrog cas for
perormncebased
grants, thoughit also underlinesthat such grants may conflictwith
considerationsof equity. The informationalrents will not necessarilybe allocatedaccording
to any nornativelyjustifiableformula.
In order to limit informationalrents, the federalgovernmentwill find it desimable
to
accept inefficiencies.The servicethat the districtprovidewill not be priced in such a way
that marginalcost is equal to its socialmarginal-value.It will only be for thesedistrictswho
are the most efficientat providingthe servicethat the equalitywill hold.
An example may help. In a countryin whichthe federalgovernmentsucnrcsto
the local governmentsmaintenanice
of the roads, the latter would try to exploittheir spro

14

knowledgeof local technicalconditionsand of the taste of the local population.They would

have a tendencyto overstatecosts and benefitsof additionalmaintenance.The theory tells us
tha they will be able to reap some advantagefrom this superiorknowledge.Furthermore,in
order to limit this advantagethe federalgovernmentwill accepttha some inefficienciescreep
into the system, and the directionof theseinefficienciescan be predicted:there will be too
litte maintenance.

The rigour and clarity of the theoryof revelationmechanismshas an associatedcost.

By appearingto suggestthat a central governmentcan, if it wishes, overcomethe
informational advantages of local govemnment,it undermines the entire basis of
nt Tade-off
between information-gatheTingand the interalization of extemnalitiesthat constituted the
conventional wisdom as we characterized it in section 2. The conclusion might appear to be
dat centralization is always to be preferred to decentralization, but in fact te subveion of
the traditional approach goes further than that. For if a central govemrmentcan sign contra

with local governments,what is to prevent local governmentsfrom signingcontractswith

each other without the intermediation of a central authority at all?

3.3

-Why do we need a central

government
in order to internalize externalities?

irt management of river basins represents a typical example of decisions that one
should expect to leave to levels above the local level (assuming, of course,
ttat,

as is usually

the case, the river flows through many localities). Indeed there are strong xteract

with

the upstream localities able literally to dump their tash on the downstream localities. If this

15

is the case, why can't the externality be resolved through direct negotiations between the
localities themselves?
More generally, the conventional wisdom has generally assumed that we need an
authority that can order subordinate units to limit negative externalities, or encourage them to
produce more positive extemalities. There is no explanation of why we cannot apply a
version of the reasoning in the Coase theorem, which states that in the absence of negotiation
costs, bargaining will lead to optimal outcomes. One may acknowledge that the Coase
theorem is more seductive at the theoretical level than as guide to reality, and one cannot
expect a totally satisfactory solution to this problem. However, without at least some thought
given to the discrepancy between theoretical and practical intuition, it is difficult to provide
guidelines on the circumstances in which it would be better to let districts negotiate between
each other (see Gatsios & Seabright, 1989, and Jack, 1992, for discussions of this point).
So far, then, the application of contract theory to the analysis of decenalization has
yielded a somewhat nihilistic conclusion. Not only has it shown that the supposed advantages
of decentralization are not really advantages; it has also shown that the supposed
disadvantages of decentralization are not really disadvantages. In consequence, without
further refinement it implicitly makes it impossible to discuss the allocation of rerponsibilities
among the different levels of government at all. For example, the literature contains
arguments for the decentralizatori of secondary schools to local governments as well as
arguments why it should be run at a regional or national level. How would we use standard
contract theory to study these two situations? In both cases, contracts, implicit or explicit,
would be signed, and some parties would benefit from informational rents due to their

16

superior "time and place" knowledge.There wouldreally be no reason to have very

differentoutcomesin one case and the other. Even if the objectivesof the regionaland the
local governmentswere different,they wouldpresumablybe drawnto bargaintogetherin
such a way that the terms of the contractwould representboth of their objectivefunctions.
The main differencewould be in the identificationof the party responsiblefor paying the
informationalrents to the other.
The implicationof this reasoningis that studyingthe imperfectionsof contratual
bargainingis not just the result of a pedanticdesire for a baroquemodellingstrategy.On the
contrary, it is absolutelycentral to understandingthe strengthsand weaknessesof
decentralizationin government.If contratual bargainingworked-perfectly,the extent of
decentralizationin governmentwould be an irrelevance.In section4, therefore,we discuss
three importantaspectsof the imperfectionsof contractualbargainingthat are of particular
importance:the endogeneityof informationacquisition,the incompletenessof contract, and
the possibility.of renegotiation.

4.

Contracttheory with imperfectionsin bargaining

4.1

Th'eendogeneityof informationacquisition

Both the conventionalwisom on decentralizationand first-generationcontracttheory

treated the allocation of information as exogeneous to the model. In reality the information
possessed by agents depends on the cost of acquiring it and the cost of processing it, and

17

thereforeon the incentivesthat they have to incur these costs. A good allocationof
responsibilitiesshouldgive the proper incentivesto agents to incur these costs
For instance,the controlof politiciansby the electoratedependscruciallyon the
informationthat the electors have on the performanceof the differentgovernmententities to
which they belong. As we discussbelow, the allocationof responsibilitiesamonglevelsof
governmentaffects this acquisitionof information,by increasingthe value of the information
to voters in their efforts to controlpoliticians.
The neglect of this point can also lead to underestimadonof the cost of cidzen
participationin the running of governmentand can lead to recommendationsfor the creation
of overly complicatedadministrativestructures,with eachcitizen belongingto a multiplicity
of districts,one by each type of public good.
Closer attentionto the endogeneityof informationacquisitionis the first step in
understandingwhetherand to what extent localjurisdictionsenjoy any kind of informatonal
advantageover central ones. Let us begin with an example.Shoulda public tstation
systemin a provincialcity be run by a local agencyor by a branch of a nationalagency?The
receivedwisdomwould answer that it should. Externalitiesbetweencities are negligible,and
hencethe informationaladvantagesof proximitydominate.Upon reflectionthough,it is
difficultto understandwhat theseinformationaladvantagesare. A nationaltransportation
agencycould and presumablywould employrepresentativeslivingin the city, who would
have accessto the samesourcesof informationas the representativesof a local agency.
There seem to be two commonreasonswhy local authoritiesare assumedto be better
informedabout local conditionsthan nationalones. First, there is direct observation.Because

18

7-te-,

local politiciansand civilservantslive in the district, they have directaccessto information

(qualityof service,traffic problems)whichis not availableto individualswho live far away.
Furthermore,the informationthat they do not gather directlycan easilybe providedby
acquaintances,familymembers,or simplecitizenswho have easy accessto them. Second,
there are formal mechanisms.Governmentsoftenput in place formalproceduresthrough
whichthey collectinformationfrom their constituents:consultativeassemblies,public
hearings,advisoryboards.
If local governmentsindeedhave betterinformationthan cental governments,it must
be eitherbecausesome of thesetechniquesare availableto themand not to the central
goverment,or becausethey have betterincentivesto use them. However,it is clear that there
is no reasonwhy centralgovernmentcannotuse any of the techniquesavailableto local
government.Centralgovernmentsdo nqamerepresentativesto localareas who collect
inforrnationon theirbehalf(for instance,the Frenchprefets);and they do use mechanismsby
whichcitizensof a local area can expresstheirviewsand provideinformation.
If anything,the balanceof advantageas far as the availabilityof infonnation
gatheringtechniques! concernedprobablylies with centrl government.Thereare
economiesof scopein the collectionof information:a centralagencywhichruns many
trnsportation systemscan transfertechniquesacquired-in one of them to the others at low
cost; it can also afford to hire techniciansin morespecializedareas an can a local agency.
If centralgovernmentsdo not in fact use the techniquesavailableto them as
assiduouslyas local governments(a matteron which empiricalevidenceis unclearand which
will differ accordingto the typeof informationin question),this mustbe becausethey have

19

less incentive to use them.3 In the city transportation case, it will be because local
politicians, who live among and are up for re-election by local citizens, have more incentive
to listen to what those local citizens say. National politicians could find out if they wished to
the views and opinions of local citizens; but they will have less incentive to take theem
seriously. Understanding why this is so depends critically on appreciating the significance of
the incompleteness of political contracts.

4.2

The incompleteness of contracts

4.2.1 The notion of incompleteness

Whether in economic or in political life, a contract allocating responsibility to various

parties for the accomplishment of certain tasks can never cover all the eventualities that may
occur. This may be because the parties simply do not foresee all relevant eventualities; it Wi
certainly be because the cost of writing contracts to cover them all would be prohibitive; and
more subtly, some eventualities, though foreseeable and describable, could not be verified by
any agency capable of enforcing the contract. For example, suppose the national government
tried to sign an undertaking with the citizens of the provincial city to indemnify them
financially if the performance of their transportation system were inadequate. Even in
circumstances where .he citizens and the government both knew perfecty well that the
3 Note that the higher incentive
of local politicians to use local
information does not imply that corruption may be a lesser problem in
decentralized
governments.
In fact, corruption
may actually
become a more serious
problem
if
increased
access
to information
is not matched
by increased
accountability of local authorities.
20

performanoeof the systemwas inadequate,sucha clause wouldbe unenforceable.

Thanks to Coase's (1937)insight that a good theoryof the firm had to be built on a
precisedescriptionof the costs of transactions,theoristsof the firm have long focusedtheir
attentionon the difficultiesof writingcontracts.If one can write explicitcontractsfor the
deliveryof some goods and services,this trnsaction can be conductedbetweenfirms without
any loss of efficie-nicy.
When explicitcontractshave limitations,there mightbe place for
internalorganization.
The first formal modelof incompletecontractsis the model of the employment
relationshipdue to Simon (1951)4.Simonargued that an employercannotpredict accurately
the tasks it would like a future employeeto do; it is even impossibleto write down explicitly
a list of which tasks have to be accomplishedunderwhichcircumstances.Under these
conditionsan employmentcontractspecifiesa salary and a set of tasks that the employeris
allowed to ask the employeeto do. Williamson(1975),who providedan essentiallink
betweenthis literatureand the modemformal contractingliterature, conductsa carefuland
exhaustivestudy of the difficultiesof contracting.
FollowingGrossmanand Hart (1986),this has been the subjectof muchrecent work
which shows that even relativelysimpletransactionsnecessitatevery complexcontracts.The
sale of an airplanegives rise to a contractof severalhundredpages. Redistributionof income
among politicaldistrictsnecessitatesfornulae whoseconsequencesare not alwaysclearly
anticipatedby their designers.In practice, the complexityis reducedby writing contrac that
do not discriminateas finelyas ideallydesirablebetweendifferentcircumstances.For
4 Alchian

and Demsetz

(1973)

present

Simon.
21

counterarguments

to the

thesis

of

instance,the proportionof the-costof a projectto be paid by differentlevels of government

will be dependenton the valuesof a subsetof all the variablesthat economictheorywould
deem relevant. In practice it is importantto discriminatebetweentwo tpsof

-variables:

I) those variablesthat dependon the actionsof the parties. For instance,

redistributionof funds betweendifferentareas shouldin principledepend on all the
effort they make to provideservicesto the poorestsegmentsof the population.
2) those variables that are exogeneousto the actionsof the parties. For instance,
redistributionshoulddependon the relative prices of a numberof internationally
traded goods which influencethe economicwell being of the poor or the cost of
providingthem with services.

:;

.W::--~~~~~~~~

In both cases we observe real-lifearrangementsthat take into accountonly sonic of

the many variablesthat are potentiallyrelevant.

4.2.2 Implicationsf6irthe-theoryof organizations

This renewal of contracttheory has had powerfulimplicationsfor the theoryof

organizations.If it is impossibleto specify fully in a contractwhat actionsshould be taken
by which party, the best the contractcan do may be to decide whichparty has the discretion
to take the relevantdecision. In other words, contractsbecomeless a matter of specifying
particularactionsand morea matterof allocatingrights of control. By contrast,complete

22

K. contractswouldremovethe need for discretionand therebymakeredundantthe notionof

control.
The notionof controlpavesthe way for decisionswith a "0-lu &ruyes-.nou chramcter
*

in contracttheory.This shedslighton issuesof integration:organizationsare or are not

verticallyintegratedwith eachother. This type of sharp distinctionis not easilyavailablein

standardcontractingmodels.
The idea that partiesto a contractwill needto agree on allocationsof controlrights
also helpsto explan the powerof incumbencyin organizations.Witha completelongterm
contractthe only benefitsof incumbencycome from the accumulationof specificcapitalor
otherphysicalchangesin the system.All the rentsassociatedwiththesebenefitscan in
generalbe extractedby the partywho has the mostbargainingpowerat the outset.With
incompletecontracts,the situationchanges.One mustoftenleave somepowerax-postto the
agent, who may share in the rents associatedwith the contracteven if there is no asymmetry
of information.

4.2.3 Implicationsfor the theoryof decentralization

This insightcan be appliedto the problemof the differencebetweengivingthe right

to controlan agencyto a local governmentor to a ce'ntralgoverment.The contractbetween
the agencyand its controllingbody specifiescertainactionsthat it musttake, certainrights
that it has, but also certainactionsthat the controllinggovernmentbodyis entitledto take. In
order to understandthe consequencesof the contract,it is thereforecrucial to understandthe

23

=-'

-:

objectives of the controlling body, and hence to know its identity. So in the case of the city

transportationauthority,answeringto a localcontrollingbody will meanhavingits policy

determinedby strictlylocal considerations,whichis desirableso longas spillovereffectson
and from other regionsare small. Answeringto a nationalcontrollingbody wouldmean
possiblybenefittingfrom the nationalcoordinationof policybut also being run by those with
many other prioritiesthan the welfareof citizensof the city concerned.Centalizationmay
be said thereforeto improvecoordinationbut reduceaccountability.
Similarly,the contractbetweenthe electorateand a politicianelectedto officeis very
incomplete.The actionstakenby the politicianwill dependon a numberof factors,including
the other tasks for whichhe is responsible.This is rathertrite, but it is very difficultto
integratethose considerationsin a completecontractmodel,where the electoratecan give
sophisticatedinstructionsto the politicianand adjustpreciselythe rewardto the actiontaken.
Likewise,contractsbetweenjurisdictionswill be very incomplete,and there will
consequentybe limitsto the extent to whichexternalitiescan be internalizedpurelyby
bargainingbetweenjurisdictions.
A stylizedmodelof politicalaccountabilitythat explicitlyincorporatesthe
incompletenessof contractsis presentedin Seabright(1994).We describethe basicmodelin
the Appendixand derive its fundamentalrelationships;for applicationsand versionsthat
relax some of the moresimplisticassumptions,the reader is referred to the originalpaper.
Essentiallyit is a modelin whichvoters seek to providepoliticianswith incentivesto act in
their interests.However,the 'performance"of the politiciansdependson factorsthat are not
verifiableand cannotbe madethe subjectof an explicitcontract.Thereforethe only way to

24

provide politicians with the right incentivesis to give voters the power to eject them if they
are dissatisfied.
Centralizationin this model involvestwo features. Its advantageis that by allowing
the central govemment to control more levers of policy it internalisesany externalities
between the regions. Its disadvantageis that any onejregion loses its ability to eject the
goverment purely according to its own preferences; it faces the risk that in some
circumstancesa governmentwill be re-elected whom it would have wished to eject, because
of the preferences of other regions.

4.3

Renegotiationof contracts

A federal gcvernment promises support for the constructionby a local governmentof

an irrigation system if a certain number of conditionsare met: a design made by a reputable
engineering firm, a cost benefit analysis which proves the desirabilityof going ahead, and a
public hearing of the users. The local governmentcomplieswith the first two conditions,but
not with the third, which it feels politicallydangerous. If it has been smart enough to keep
the design of the system within acceptablebounds it can, explicitlyor implicitly,propose to
the federal governmentthe followingdeal: "Why don't you accept to fund the project, even
though some conditionshave not been met? You will be better off than if you did not fund it,
and so would wet. Under the general principle ftat bygones are bygones, the federal

25

governmentshould accept this deal5. The original contractis not "renegotiationproof". The
awarenessof this fact will weakenthe incentivesof the local governmentto respect the terms
of the original agreement.
The general idea is straightforward:if during the executionof a contractit becomes
clear that it can be modifiedto increase the utility of all the pardes involved, the original
contractwill be modifiedat this point. Foreseeingthis modification,the parties may modify
their behaviorin previous periods.
This has importantpolicy consequences,and makesit possibleto understandpolicy
choices that seem mysterious.For instance,penalitiesfor delay in large projects are often
renegotiatedaway when the contractoris late. The expectationof this renegotation
encouragesdelay. However, this is often unavoidable.Once the project is late, the client will
often prefer not to impose the penalty so as to keep carrots with which to encouragenot too
untimelycompletion(see Cramer and Seabright(1994, forthcoming)).

4.4

Unresolvedissues

In the excitementof new discovery, it is easy to underestimatethe number and

importanceof topics whichare importanton applied groundsand on which the literature has
difficultyprogressing.We mentionfour.

We are ignoring
some important real world considerations,
desirability for the federal government to appear "tough"
subsequentnegotiations.

26

such ao the
in possible

Dynamicsand tmnsitions: We do not understand well the dynamics of organizations.

There is informal evidence that transitionsbetween one line of activity and another, between
one industryand another, between one organizationalform and another are difficult and
require time. Analyticaltreatment of these phenomenais very difficult.

The constitutionalframework: The contracts, implicit or explicit, that govern the life
of an organizationhave different statuses. Someare more fundamentalthan others, rather as,
in the legal realm, clauses of constitutions have a higher status and are more difficult to
change than ordinary laws. This hierarchy of rules, important in private firms and even more
so in governments, is poorly understood.

Multidimensionaluncertainty: Most of current theory has been developed in settings

where uncertaintybears on the value of one real variable. The extension to multidimensional
uncertainty creates very difficult technicalproblems, and has been successfullycompleted
only in very special cases. It is difficult to give precise advice on methods to circumvent the
problem, but one should be aware of this fact when using the insights of the theory.

Networks of contracts: Recently a number of authors have stressed the fact that finns
should be thought of as networks or nexuses of contracts or of treaties (see C-rmer and
Riordan, 1987; Aoki, Gustafson& Williamson, 1990). However, very little formal work has
been done on the interrelationshipsbetween differentcontracts signed by the same agent. For
instance, a provincial government will be party to contracts signed with the federal

27

government,withthe districtsat a lower level, with the unionsrepresentingits employees,

implicitlywith the voters,etc. We do not understandvery well how thesecontract interact
with each other.

5.

What is p jurisdictionand where shouldits boundarieslie?,

Traditionalmicroeconomictheory takesthe definitionof the firm and its objectives

as given. The theory of the firm began makingsubstantialprogress by askingthe seemingly
very abstractquestion:"what is a firm?" Even thoughthe answerswe have are still
imperfect,the work done to find them has yieldedlessonsthat have very concretepolicy
implications.It is thereforea reasonablestrategyto try to ask the question:'whiatis a
politicaljurisdiction?"

5.1 Coordination

One importantaspectof the definitionof the fim is that it groups togethernumerous

types of activities.Similarly,if we are to developa good understandingof political
jurisdictions,we need to take seriouslythe notionthat they integratethe power to make
decisionsin differentdomains.Indeed,jurisdictionscan be thoughtof as ogmztosta
grouptogetherthe power.totake politicaldecisions,and 'wecan ask what determinesthe
groupingof activitiesby analogywith the groupingof activitieswithina finn.
Simon(1973)has arguedthat what determinesthe groupingof activitiesis the

28

necessty of coordinationof activitiesin the short run. Crdmer(1981)has analyzedformally

this coordination.The followingexample,drawn from this paper, provides fth necessary
intuition.
The allocationof responsibilitiesfor transportationwithina countryis usuallydoneby
mode of transportation.This seemsto neglectthe fundamentalnatureof the good. From the
pointof viewof consumers,a plane trip betweenParis and Lyonshas more similaritywith a
train trip betweenthe sametwo cities than with a planetrip betweenParis and Toulouse.
Yet, it is the two train trips that are the responsibilityof the same'agency,not the two trips
betweenParis and ]Lyons.Why is that? In the day to day managementof the trnsortation
system,the decisionsto be takenare decisionsthatinvolvethe planesand the crewstha wfll
flow betweenall towns.Very fe"v'short run decisionsinvolveplanesand trains at the same
time. It is tru-ethat for planningpurposesone shouldweighcarefulflythe optionsof plane and
train transportationone againstthe other. Thesedecisionsare madein circumstanceswhere
time is not a crucialfaictor,and hencethe extra difficultyof coordinationbetweenagenciesis
not very important.

For the managementof infrastructure,this mighthave someimportantconnces.

For instance,one could thinkthat mainirigation channelscould be, let us say, a provincial
responsibility,whereassmallerbranchescouldbe a Municipalresponsibility.This can only
be true in circumstanceswherethere is no necessityto coordinaterapidlythe managementof
the two parts of the system.

29

5.2 Contlbyl voters

The thieoryof thefirm has spent muchenergy tryingto understanddie mechanisms

throughwhich managersof corporationsare controlled.Stcckholdershave bounded
rationality,and both the acquisitionof informationand the exerciceof the right to vote on
corporatemattershave relativelylarge costs. Furthermore,the benefitsof any improvement
in corporateperformancegeneratedby suchactivitiesare sharedwith all the other
stockholders.It wouldseem thereforethat corporationsare undercotrolled. The mechanisms
used to remedythis situationare of two kinds. First, the controlof the day to day
managementof the firm canibe delegatedto specialists(i.e. the boardof directors).Second,
underperformingfirmnswill be targets for takeoversif outsidersbelievethat they can run the
firm in ways that increaseits profits.There is generalconsensusthat the questionof control
affectsconsiderablythe structureand the behaviorof ognztnS.
Differentformsof groupingtogetheractivitiesaffectconsiderablythe natureof the
controlthat is exercisedon politicalmanagers.First, relyingon cooperationbetweenlayers
of governmentoftenblurs responsibilityfor the executionof particularprojects,and hence
mighthave negativeeffectson efficiency.This is a commonproblemin water projectsin
some parts of Latin Americafor instancewher-ecoordinationefforts becomea highlytime
consumingpart of projectmanagement.Second,even where responsibilitiesare separateit
makessenseto dividethem accordingto principlesthat are easy to remember.For example,
allocatingresponsibilityfor differenttypesof road to differentlevelsof governmnent
sometimesresults in difficultiesin identifyingwho is in chargeof what.

30

Because voters invest in informationgatheringabout the performanceof local

governments,there is someorganizationalcapitalinvestedin the qualityof participationin

publiclife at differentlevels. For instance,in a systemwhere municipalgovernmentsare
powerful,voterswill pay attentionto the actionsof the mayor, newspaperswill develop
sourcesin townhalls,and so on. Undertheseconditionsthere will be strongbenefitsto
allocatenew responsibilitiesto townhalls,ratherthan to weaker,and thereforeless
controlled,politicalentities.One can expecta certain stabilityof the allocationof powerin
the system.
If theseintuitionsresist further analysis,and if they are as importantin practical
terms as we suspectthat they are, the consequencesfor the organizationof the publicsector
are strildng.Becauseone cannotdeterminethe optimallevel of governmentfor specifictasks
in isolation,it does not makesenseto studythe decentralizationof decisionsabout
infrastructurein isolationfrom the allocationof other publicresponsibilities.For instance,
landfillmanagagementwill be assignedto the municipalities,becausethey are responsiblefor
garbagecollection,and they are responsiblefor garbagecollecetion
becausethey have better
informationaboutthe use of land, havinggatheredthis informationin order to collect
propertytaxes.
Note that there is a cost to pushingtoo muchresponsibilityonto one level of
government.If the centralgovernment,for instance,is responsiblefor too manyactivities
each of themwill have relativelylittle weightin the decisionsof the voters, and henceit will
be easierto run someof them for privateratherthan publicbenefits.

31

6.

SomeApplicaions

6.1

The optimalsize of a jurisdiction

promotedin the
Decentralization
belowthe nationallevelis relativelysystematically
literature.Very often, one sees lists statingthat this decisionis to be takenat the federal
level,this other at the statelevel,and yet anotherat the local level.Why wouldthe
efficiencycriterionbe a certainnumberof levelsbelowthe nationallevel?Somecountries
in other countries.It would
havea smallernationalpopulationthan statesor municipalities
seemthat population,or area, or someother measureof the servicesto be renderedwouldbe
a morelogicalcriterion,and if this is the case,it doesnot makemuchsenseto push
decentralization
on smallercountries.
On the otherhand, the notionof controlmighthelp explainwhy the optimalnumber
of levelsmightvary in sucha way that the smallerdistrictis largerin large countriesthan in
smallercountries.The boundedrationalityof voterspreventsthemfrom exercisingadequate
controlover morethan a certainnumberof govemmentalidentities,a numberwhichis
independentof the size of the country.

6.2

Doesdecentralization
promotedemocracy?

It is oftenarguedthat decentralization
willpromotedemocracy:local governmentsare
moresensitiveto localneeds, hencean individualhas morepossibilitiesto influence.

32

-,

'Is

decisions that affect his life and his welfare under greater decentralization. This statement
can be mnademore precise in two ways. First, under decentalization, there is a greater
responsivenessof decisions with respect to the changes of preference of a single agent or
group of similar agents. Second, under decentralization,there is a higher probability that a
given agent or group of agents is pivotal, and hence determines the choice of a solution
between two altematives.
However, these generalizationswill hold, if they do, only when all other factors are
equal. In particular, it is important to notice that decentralizationmight lower the political
power of the poorest citizens (see Seabright, 1994, section 3). Under decentralization,
government will indeed be more sensitive to local needs, but it will still be local needs as
perceived by the local political system. Hence, the groups that find it easier to organize
political activity at the local level will be favored by decentralization.Orgaization of
political forces is easier for groups where the benefits of influencingpublic policy is shared
among a small number of agents. Groups that have more difficulty organizing will in general
prefer larger political districts so that organizing is easier. Some political backing comes
from the fact that political forces that have favored income equality have generally favored
centralized governments.

6.3

Why should decentralizationlead to more experimentation?

The argument is often made that decentralizationpromotes experimentation:free from

the strictures of central administration, local governmentscan try out new ideas and methods.

33

Decentralizationof decisions in water supply matters for instance has led Fortaleza in the
North East of Brazil to adopt cheaper technologicalsolutions such as community water
pumps not yet tried in the South of the country where traditional large utilities have typically
been the norm. The diversity of experiences iniurban tranport in France or in India for
instance also illustrates what municipalitiescan do when given the freedom 50choose.
However, one must be careful with this argument. If the regions of the country are
very different, it does indeed seem likely that local control will lead to more diversification.
On the other hand, experimenting with a new technique creates positive externalities for
other districts: they do not suffer the consequences of potental failures but benefit from the
information generated.
It might seem that one could have the best of both worlds: use a decentralized system
to benefit from local adaptation, and have the centml government fund experimentation.But
note that it is very difficult to write a contract that will indeed incite a district to experiment
with a different institutional arrangements. It is difficult to define precisely what is the
innovation; typically, institutional innovationsare refined during implementation. It is also
difficult to check whether it has indeed been implemented.

6.4

Decentralizing education

Seventy percent of French people believe that regional governments should be given
more power in the Financingand the management of schools; 60% believe that they should
be given more power in the recruiting of teachers. On the other hand, only 30% call for a

34

L%r

greater role of regionsin the developmentof curriculaand 12%are willingto give the
responsibilityfor the grantingof diplomat. It seemsfair to summarizethese resultsby
saying that Frenchpeoplebelieve that programsand standardsshouldbe decidedat the
mntional
level, but that schoolsshouldbe managedlocally.We shouldlike to show that the
desirabilityof suchan allocationcan be fruitfullyanalyzedusingthe conceptualframework
sketchedin this paper.
To considerthe allocationof powers over the managementof schools,we must first
weighexternalitiesbetweenlocalitiesand knowledgeof local conditions.The extenalities
stem from migration:studentstrainedin a districtmaygo and spendtheir adultlives in
anotherdistrict, and thereforethe socialbenefitsof trainingare only partiallyenjoyedby the
districtof schooling.The nationalsetting of the curriculum,and the verificationthrough
nationalexamsthat the curriculumis actuallylearnedby the students,preventfree riding by
districts.On the other hand, decentralizingthe managementof the schoolsallowsfor better
adaptationto local tastesby a bettersettingof the datesof vacations,hiring of teacherswho
are sensitiveto local culture,and so on.
This analysis,whichdrawsentirelyon the conventionalwisdomon decentalizationas
we describedit in section2, has strongpolicyimplications,some of which we do not believe
wouldbe generallyaccepted.For instance,belowa certaindegreeof inter districtmnigration,
responsibilityfor curriculumand standardsbecomethe responsibilityof regions. Theanalysis
is also basedon assumptionswhich seemdifficultto defend.For instance,if tastesare
6 More precisely, 12% agree with the statement that "...leu r6gionu
devraient
pouvoir remplacer
les diplomes nationaux
par des diplames r&gionauxw.
These statistics are taken from Le Monde de l'Education,octobre 1993 (from poll
conducted by SCP Communication of 858 persons, representative of the French
population.
35

, ._

regionallydifferentiatedfor vacationsand personalityof teachers,why are they not

regionallydifferentiatedfor curriculumand the organizationof examinations?
Of course, all the elementssketchedabove are importantin reality. However,we
believethat an essentialelementof realityis missedif we do not thinkin more contractual
terms. To use a contractualapproach,we must first identifythe fundamentalcontacting
difficultiesbetweenthe publicand the schoolsystem. As a first approximation

1) consumersof schoolservices(parentsand/or pupils)havegood informationabout

the pleasurethat they derivefrom the schoolsystem(facilities,friendlinessof
teachers...).
2) on the other hand, they have poor informationabout the difficultyand the quality
of the curriculum.

The separationof powersdefinedaboveprovidesa way to circumventthe second

difficulty,while maintainingthe advantagesof controlby consumerson the aspectthey
controlbest. On the aspectsthat the consumersunderstandbest, powershouldbe local
because"voice" can be exercicedmoreeasilyat that level. On the otherhand, it is very
costlyfor the citizensto judge the appropriatenessof locallyset standards.There exist strong
incentivesfor the local governmentto hide low standardsin order to improvethe pecption
of the qualityof the educationthat is provided.
By detachingthe right of controlover standardsfrom the right of controlover
managementof the school, one limitsthe incentivesto tinker withstandards.Comison

36

of

the results in nationalexamsbetweendifferentdistrictsprovidesthe basis for a fastjudgment

by citizensof the qualityof their schooldistricts.

A completeanalysiswouldincludemanyother factors:
I) a discussionof the burdensthat changesof standardscan imposeof the districts
and theirnon internalizationby the nationalauthority;
2) a discussionof the tensiondue to the fact that the consequencesof neglectof the
educationalsector has consequencesthat may not be felt untilafter the nextelection;
3) a discussionof the reallocationof resourcesdependingon the wealthof the
districts.

Nevertheless,our analysisalreadyhas strongpolicy implications.In particularthe

desirabilityof the proposedorganizationdoesdependon a certainnationalhomogeneityof
tastes for curriculum,but not at all on the intensityof migrationbetweendistricts.

6.5

Earmarking

In manyEast Asianand LatinAmericancountries,earmarkingis or has beena

commonof way for subnationalgovernmentsto financesome of their majorinfrastructure
expendituresand hencedeservesa carefulassessmentin a discussionof decentralition.
McCleary(1991)providesa usefulsurveyof the argumentsfor and againstcarmarkingof
revenues.Definingearmarldngas "thepracice of assigningrevenue- genrally through

37

statuteor constitutionalclause

--

from specifictaxesor groupsof taxes to specific

governmentactivitiesor areas of activity",he developsa numberof criteriathat allow

identificationof the cases when it is justified.Thesecriteriareflectthe lessonsof experience
but some lessonsfrom standardwelfareeconomics. Theycan be revisitedin the languageof
incompletecontracts.

6.5.1 Earmarlingarisesbecauseof incomplete

contracts

McClearyrecognizesthat earmarkingis a secondbest solution:in the absenceof

imperfectionsin the governmentbudgetarysystemit shouldno.be used. For instance,in
some countriestaxeson fuelsand vehiclesare earmarkedfor a road constructionfund. This
practicehas been defendedas a meanto provide"morestablefunding(that) wouldencourage
competition". Translatedinto our framework,this argumentgoes as follows:

1) Entrepreneurswill enter the road constructionindustryonly if they can be

relativelyconfidentthat the governmenthas a long run commitmentto the buildingof
roads so that theyhave a reasonableprobabilityof recoveringtheirinvestments.
2) The completecontractapproachto this problemwouldhave the governmentmake
a contractwith entrepreneursdefiningconditionsunderwhichroads would oe built,
the prices whichwouldbe paid, and the penaltiesthat the governmentis willig to
incurr if it does not pursue this program.Suchcontractsare not feasible.
3) Therefore,a secondbest approachis for the governmentto commititself thrugh a

38

second best, albeit imperfectcontract:the resourcescomingfrom a specificsource

should go to a specificuse.

This way of statingthe problemallows a few refinementsof MJcCleary'sanalysis.For

instance, McCleauyargues that "when moneyis tight, earmarkedfunds may be temporarily
frozen (as happenedin Ghana)or divertedto other uses (as in Mali), or the governmentand
publicentrerises may stop payingtheir fuel bills and hencefuel taxes (as they did in
Zaire)". This shows that a commitmentto earmadcingis not fully credible;this mightimply
that earmarkingcannotachievewhat it sets out to achieveand that it wouldbe betternot to
attemptit. However,it is also recognizedthat "the morereliablefundinghas made it easier
to use privatecontractorsthroughcompetitivebidding".Ex-ante,the earmarkingwas
recognizedby privateentrepreneursas a way to increasegovernmentcommitment.
We believethat an approachthat takes into account fr-omthe outset the fact that all
contractsare incompleteshouldbe able to weighbetter the costs and benefitsof earmiarking.
It does not work perfectly,becausecommitmentto earmarkingis sometimesreversibleand
because it potentiallydistortsgovernmentallocationof funds. But if we can understand
exactly how and in what circumstancesit increasesthe cost to governmentof reversingits
expenditurecommitments,this will help to quantifythe benefitsit can provide.
Thseincompletecontractsapproachcan also guard againstcertain misconceptions.For
instance,McClearyargues that earmarkingshouldonly take place when there is "anx
appropriateinvestmentprogr-amand a clear set of rules to regulateinvestmentdec-isions,the
mix of spendingon capital, maintenanceand rehabilitation,and administrativeoverheads"

39

(ibid. p.102).
Noticethat if there did exist a clear set of rules, and a credibleinvestmentprogram,
there wouldbe no need for earmarking.It wouldbe possibleto write a rather explicitset of
rules that the governmentshouldfollow.This approachseemsto be in conhadictionwith the
rmcognition
that it is the difficultyof writingcontractsthat makesearmarkingnecessary.
To see this, let us use an example.Considera country,suchas Columbia,where the
for
federalgovernmenthas responsibilityfor road buildingbut stateshave responsibilit.y
maintenance.Presumably,the federalgovernmenthas responsibilityfor the constructionof
roads becausethere are externalitiesand the stategovemmentswouldnot of their own
volitionbuildenough.If this is the case, left to their own they wouldalso underinvestin
maintenance.The federalgovernmentshouldthereforefind some techniqueto encouragethe
local governmentto conductmore maintenance(we assumefor the momentthat state
governmentshave a comparativeadvantagein maintenatice,perhapsbecauselocal monitoring
of road qualityis easier).
If there were clear cut indicatorsof the qualityof mamtenanceundertakenby state
governments,the federalgovernmentcouldeasilysign a contractwiththem linkng a
paymentto the quantityand qualityof the maintenancethat had been done, as well as to
factorsover whichthey did not havecontrol. It is becausesuchindicatorsare not available
that other techniquesmustbe found. Earmarkingmay be one of themsinceit can be
reasonableto order the local governmentto earmarkall the receiptsfrom local gasolinetaxes
and from tolls to maintenance.Its has the advantageof providingan automaticlink between
the usageof the transportationnetworkand maintenance.It can also have a high opprftnity

40

cost in terms of other decentralizedexpenditureneeds(education,health,...) when local

resourcesconstraintsare important.

7. =Qnc1uding[remarks

This paper has done no morethan scratchthe surfaceof a literaturerich in both

theoreticaland practicalinsights.Its key messagesare the following:

1)

The moderntheoryof the firm providesmanyinsightsinto political

organization,for politicaljurisdictionscan be thoughto as pseudo-firmsthat provide

servicesand whichgrouptogetherdecision-maldng
activitiesof variousldnds.

2)

in governmentare questons about the

Questionsaboutdecentralization

allocationof rightsof control.If contractualrelationswere completeit would not

matterwhethernower were decentralizedor not, sincecontractswouldspecify
everythingto be done at eachlevel of governmentand there wouldbe no need for
discretion.

3)

dependsupon whichlevelof
The appropriatedegreeof decentralization

govermentwill have the mostincentiveto act to bring aboutthe desiredoutcomes.In

particular,centralizedgovernmentscan reap benefitsof coordination,but tend to be
less accountablethan decentralizedones, thoughthere are iinportantexceptionsto this

41

generalrule.

4)

Theorganizaidonal
designof government
affec not onlyincentivesto

takedecisionsbut also incentivesto gatherthe informationon whichtho decisions

willbe based.

References
Akerlof,,George A. 1970. "The Market for 'Lemons:' QualityUncertaintyand the Marke-t
Mechanism."7TheQuarterlyJournal of Ecowomics,84:3, 488-500.
Aold, M., B. Gustafsoon,and O.E. Williamson.1990. The Kim as a Nexus of Treaties.
London: EuropeanStage.
Coase, Ronald. 1937. 'The Natureof the Firm.' Economica,4, 386-405.
Cr6mer,Jacques. 1981. "A Partial Theory of the OptimalOrganizationof a Bureaucracy."
The Bell Journal of Economics,11:2, 683-93.
Cr6mer, Jacques. 1993. "Arm's LengthRelationship.' mimeo.
Cramer, Jacques. 1993. "A Theory of VerticalIntegrationBasedon MonitoringCosts.'
mimeo, forthcoming.

Cr6mer, Jacquesand MichaelH. Riordan. 1987. "On GoverningMultilatera Transactions

with BilateralContracts."Rand Journal of Economics,18:3, 436-51.
De Giroot,Hans. 1988. "DecentralizingDecisionsin Bureaucraciesas a Principal-agent
Problem."Journal of PublicEconomics,37, 323-37.
Estache,Antonioand KangbinZeng. 1992. "ManagingPollutionControlin Brazil: The
PotentialUse of Taxes and Fines by Federal and State Governments."World.PAnk
Working Paper 929, Washington, D.C.: World Bank.

Grossman,SandfordJ. and OliverD. Hart. 1986. "The Costs and Benefitsof Ownership:A
Theory of Verticaland Lateral Integration.' Journal of PollikialEconoMy,94:4, 691719.
Groves,Thbeodore.1973. "Incentivesin Teams." Econometrics,3 1:6, 617-.63.
Jack, William. 1992. "PowerSharingand PollutionControl:CoordinatingPoliciesamnong
Levels of Governments."World Bank WorkingPaper 887, Washington,D.C.: World
Bank.
Lange, 0. 1938. 'On the EconomicTheory of Socialism,"in On the EconomicTheoryof
Socialism.eds. 0. Lange and F.M. Taylor, Minneapolis:Universityof Minnesota
Press, 57-143.

43

- 0McCleaTy, William. 1991. "The Earmarkingof GovernmentRevenue:A Reviewof Some

World Bank Experience."WorldBank ResearchObserver,6:1, 81-104.
Oates, WallaceF. 1969. "The Effectsof PropertyTaxes and Local PublicSpendingon
Property Values:An EmpiricalStudyof Tax Capitalizationand TieboutHypothesis.
Journal of Political,Economy,77:6 957-71.
Seabright,Paul. 1994. "Accountabilityand Decentralizationin Government:An Incomplet
ContractsModel." CEPR DiscussionPaper 889.
Simon, Herbert A. 1973. "ApplyingInformationTechnologyto OrganizationDesign."
Public Administration33:3, 268-78.
Spence, Michael. 1973. "Job Market Signalling."The Quarerly Jounal of Economics,87:3,
561-72.
Weitzman,Martin L. 1974. "Prices VersusQuantities."The Rewewof EconomncStudies,
41:4?, 477-91.

44

Centralisation versus Depentralisation: a simple model

The model of this appendix is drawn from Seabright (1994). The population in a
certain area is divided between two regions 1 and 2. The people will elect a government; this
may be either a govemment for each region, or a single central government. After the
election the govemments have to implement a policy vector a: the regional governments
would choose simultaneously the value of scalars xl and x2 respectively, while a central
govemment would choose both scalars. In this sense we can think of xl as being the policy
instrument of region 1, in that it is an instrument available to whichever authority governs
region 1. Without loss of generality we assume that the welfare of the population in region I
is strictly increasing in xl, and the welfare of the population in region 2 is strictly increasing
in x2. The welfare of the population in region 1 may also depend positively or negatively on
x2, and that in region 2 on the value of x2. To help intuition, where these externalities are
positive we shall characterise the policies as "efficiency-enhancing", since policies that help
one region also help the other; where they are negative we shall call them 'redistributive',
since policies that help one region are at the expense of the other.
The values of xl and x2 chosen by the governments are not directly observable by the
population but remain the private information of the government. The welfare of the
population in each region, while observable by both the population and the government, is
not verifiable; consequently the constitution cannot specify rewards or penalties for the
government conditional on the welfare attained. By contrast, the extemalities between regions
are verifiable, and the constitution can make the assignment of powers conditional upon tieir
magnitude. Note that the constitution may not make explicit reference to the magnitude of
externalities, but may assign powers according to implicit judgments by the framers of the
constitution about their magnitude. Alternatively, where externalities vary from case to case
in ways that are verifiable the constitution may specify an allocation of powers explicitly
contingent on their magnitude (see note 3 below).
The unobservability of govremmentaction matters in this model because higher values
of xl and x2 are costly for the governments (they involve 'effort"). Indeed in this simple
model we can interpret xl and x2 directly as the levels of effort undertaken by the
governments concerned. Since these levels are unobservable, governments must be induced
to undertake such effort by the threat that they will not be re-elected if the populations are
not satisfied with their levels of welfare. Re-election therefore has a value for thiem(the
spoils of office), which we shall represent by W for the regional governments and aW for
the central government. The value of not being re-elected is normalised to zero. The
population's welfare is also, however, affected by additive but unobserved region-specific
shocks yl and y2; if their welfare net of these shocks falls short of some reservation level C
(which can be interpreted as the welfare they might expect from a rival political party) they
will wish to throw out the govemment. Regional governments will fail to be re-elected if the
welfare level of the population in the region falls below C; a central government, however,
will fail to be re-elected only if both regions' welfare falls below C.
45

_V

Centralizationin this model, therefore,involvestwo features.Its advantageis that by

allowingthe cental governmentto controlbothxl and x2 it internalisesany externalities
betweenthe regions.Its disadvantageis that any one regionloses its abilityto qejctthe
govemnentpurelyaccordingto its own preferences;it facesthe risk that in some
circumstancesa governmentwill be re-electedwhomit wouldhave wishedto eject, because
of the preferencesof the other region(we couldalternativelyhaveassumedthat it faceda
risk that a governmentwouldbe ejectedwhomit wouldhave wishedto re-elect, but nothing
significantwouldbe changedby assumingthis).
The order of eventsis as follows:
1)
A form of govenment (centralisedor decentralised)is chosen.
2)
The govemment(s)chooseK.
3)
The region-specificshocksyl and y2 are realised(but not observedby the
populations).
4)
The populations'welfareis realisedand theydecidewhetherto re-electtheir
government(s).
The welfareof regionI is givenby Ui(xl,x2) + yi, whereUi is strictlyincreasing
and concavein xi, and concavein its other argument.For simplicity,we assumeyi to be
uniformlydistributedwith support[O,Yi],and we assumethat this support(andthe value of
C) are such as to generateinteriorsolutions.
The welfareof regionalgovernmentPi is givenby Vi(xi) + W if relected and
Vi(xi) if not re-elected,whereVi is strictlydecreasingand concave.The welfareof natonal
govemmentE is givenby V(xi,x2)+ aW if re-electedand V(xl,x2) if not We assumethat
dV(dxi)/dxi= dVi(xi)/dxi;this is to ensure that neitherform of governmenthas an intinsic
advantageover the otherarising from a differingmarginaldistufilityof effort
We can now comparethe resultsunder two differentformsof government
RegionalGovernment
The governmentin eachregioni will choosexi to maximise
(2.1) B [ Vi(xi) + W]
= Vi(xi) + W.prIui(xl,x2)+yi >

C]

= Vi(xi) + W.pr(yi > = C-Ui(xl,x2)]

= Vi(xi) + W(Yi+Ui(xl,x2)-C)IYifor interiorsolutions,given the uniform
distributionof yi.

46

The first order conditionsfor this problemare

.(2.2)

dVi(xi)/dxi = (W/Yi)(dUi/dxi)

What this meansis that the disutilityof effort is equatedto the value of stayingin
office multipliedby the marginalincreasein probabilityof re-election.Note that Vi and Yi
need not be the same for each region;consequentlythe valuesof xl and x2 chosenby the
regionalgovernmentsmay not be the same.
-Cntml

Goverment

The central governmentE will choosexl and x2 to maximise

(2.3) E [V(xI,x2) + aW].
= V(xl,x2) + aW.pr[Ul(xl,x2)+yl

> = C or U2(xl,x2)+y2 >=

= V(xl,x2) + aW.(I-pr[Ul(xl,x2)+yl

C]

< C and U2(xl,x2)+y2 < C])

To simplifynotation, denoteby ql the event that Ul(xI,x2)+yl < C and by q2 the event
that U2(xl,x2)+y2 c C. Then we can rewriteE's objectivefunctionas
(2.4) V(xl,x2) + aW.(l-pr(qi).pr(qj Jqi))
*

for each i

- = V(xl,x2) + caW.(l-pr(qi).pr(qjI qi))

=V(xl,x2)

+ aW.(l-((C-Ui(xl,x2))/Yi).pr(qjI qi))

The first order conditionsfor this problemare

(2.5)

dVi(xi)/dxi= (crW/Yi)(dUi(xi)/dxi).pr(qj
I qi)
+ aWpr(qi)(dUj/dxi)(dpr(qjf qi)/dUj)

for each i

And comparingthis to equation(2.2) we can see what differencecen

ion makes
to the solution.The disutilityof effort in implementingeach policy instrumentis under
regionalgovernmentset equal to the marginalincreasein probabilityof re-electionmultiplied
by the value of being re-elected.The probabilityof re-electionis just the probabilityth fthe
:welfareof the populationof the regionconcernedfalls below its thresholdlevel. Under
cental governmentthis secondterm is multipliedby the probabilit hat the welfareof the
populationof the other regionfalls below its thresholdlevel, to reflect the fact that it takes
both regions to be dissatisfiedfor the governmentto be ejected. Other thingsequal, this will
' lower the value of the right hand side of equation(2.5) and result in a lower effort by the
-governmentand-a lower utility for the population;this is a dimctconsequenceof the reduced
47

acoutablit
cosequent upon centralisation.But now there are two othereffectsto take
intoaccunt.Oneis
due to the possiblygreaterincentivefor effort due to greaterspoilsof
offce
t atina leel(represented by the factora). The secondis representedby the second
term on the right hand side of (2.5): the externalitydue to the impactof xi on regionj's
Z. welfarehas now been intemnalised.
How mnuch
differenceit makesto the government'seffort
wili in turn dependon the differenceit makesto the probabilityof re-election,whichis
preciselyequalto the marginalincreasein the probabilitythat the populationof regionj are
satisfied,conditionalon the populationof regioni's being dissatisfied.And whetherit
increasesor reducesthe government'seffort will depend,of course, on whetherthe
externalitywas positiveor negativein the first place.
s

If the lattertwo effectsoutweighthe first, a centralgovernmentwouldmake more

effort to implementthe policyinstrumentof regioni than wouldregionalgovernments.
Whetherthe populationof regio'ni is betteroff as a consequencewil depend,of course, on
what happensto the effort the governmentmakesto implementthe instrumentof regionj, as
well as on whetherthis has a positiveor negativeexternalimpacton regioni.
It is worthnotingthat, if each region'swelfarewere verifiableas well as being
observable,the constitutioncouldspecifyrewardsand penaltiesconditionaluponwelfare,
and the adverseimpactof centralisationon accountabilitywoulddisappear.IdWeed,it is
straightforwardto see that, sicne there is no risk-efforttrade-offin the model,sucha
constitutioncouldalwaysimplementthe first-bestunder eitherform of government.
We can summariseequations(2.2) and (2.5)in words:
Underregional govemmuent:
Marginal
disutiLity
of effort

VaLueof
reeLection

MarginaL
increase
X

inprobabiLity

to regionaL
goverriment

thatregionis
seti1sfijed

Undercentral govermieent:
NarginaLdisutiLity
of effortn

Pwlu

value of
reeLection
to central
government

VaLueof
reeLection
to centraL
goverrsient

Spittover on
weLfareof
other region

MarginaLincrease
inpro)abiLity
X
that region is
satisfied

Effect of welfare
increase on probabiLity that other
region is satisfied

ProbabiLity
thatthis
region's
wetlare
determines
reelection
X

ProbabiLity
that other
region's
wetfare
determineis
reelection

What kinds of conclusion can be drawn from thiii analysis? We begin by considering
48

the simplest case where policies are efficiency-enhancing (so spillovers are positive and
therefore increases in effort are unambiguouslya good thing for both regions) and where the
value of reelection is the same at regional and national level (so a is equal to one). Eight
conclusions can be drawn, some obvious, some less so:
1)
First, the analysis confirms the basic intuition that the case for centrdlisation is
stengthned if there are significant spillovers between regions. That is to say, the likelihood
that centralisation will increase government effort and consequently population welfare is
increased if spillovers are large.
2)
It also gives a precise sense to the notion that the cost of centralisation is a loss of
local accountability. Here the loss of accountabilityis the fact that the welfare of a region
now has a probability less than one of being the decisive factor in whether or not the
government is reelected.
3)
A less obvious conclusion is that a positive correlation between region-specific shocks
strengthens the case for centralisation. To see this, note that the probability that any region's
welfare is decisive in whether or not the central government is re-elected is the probability
that the other region is dissatisfied (since it takes two dissatisfied regions to eject a central
government). But this is not an unconditional probability; rather it is the probability that the
other region is dissatisfied, conditional on this region's being dissatisfied (thatis, it is not
pr(qj) in equation (2.5) but pr(qj I qi)). And this conditional probability is higher the greater
is the correlation between the shocks to the two regions. At the extreme, if the shocks are
perfectly correlated, the risk that any region will find itself unable to eject a government with
which it is satisfied because the other region is of a different view becomes negligible, and
there is no loss of accountability in centralisation.
4)
Note that this does not at all the same thing as saying that regional similarity
strengthens the case for centralisation. If differences between regions are incorporated in
their different utility functions or the different distributions of their region-specific shocks
(variables that are known to populations and governments before any decision are taken),
then both regional and central governments are entirely capable of setting different values of
xl and x2 to reflect these differences. Centralisation makes neither easier nor more difficult
the regional differentiation of policy. It is the degree of correlation of shocks (which are nif
observed by the populations) that affects the degree of centralisation. To put it another way,
what weakens local accountability is not the risk that regions will require different policies; it
is the risk that regions will be differentially satisfied with whatever policies they have.
5)
Centralisation may benefit some regions and not others. In particular, the magnitude
of spillovers may vary between regions. In such circumstances the recipients of large
spillovers from other regions are more likely to benefit from centralisation than are the
recipients of small ones.
6)

The more entrenched' governments are, in the sense that the more unlikely voters
49

are to wish to eject them, the strongerthe case for regionalas opposedto central
government.This is because,if it is unlikelythat one region
will want to eject its
governmentit is relativelyeven moreunlikelythat both regionswill wish to, so the loss of
-accountability
from centralisationis relativelygreat.
7)
The interestof citizensin one or other formof governmentis not necessarilyshared
by theirpoliticalrepresentatives.Regionalpoliticianswill lose, and central politiciansgain,
from centralisation,whateverthe benefitsto theircitizens.Furthennore,if some politicians
are potentiallymobile,in the sensethat they have a significantprobabilityof formingthe
governmentundereither centralor regionalarrangements,their incentivesare perverse.They
will have an interestin arguingfor centralisationonly whenits impactin reducing
accountabilityoutweighsits beneficialimpacton intemnalising
spilovers, since underthese
circumstancesthe level of effort they have to undertakewill be reduced.
8)
A choicebetweencentralisedand decentralisedformnsof governmentneed not always
be made once and for all, but can sometimesbe undertakenon a case-be-casebasis if it is
possibleto estimatesome of the relevantvariables(suchas the sizeof the spilloversf.
We can now ask.what wouldbe the effect of r-elaxingthe assumptionsthat al and
that spillovers.are positive.It mightseemnaturalto think that the valueof re-electionwill
alwaysbe greater undercentralthan regionalgovernment,becausecentra governments
commandlarger resourcesgeneraly, but it is importantto distinguishbetweenthe value of
initia electionand the value of re-election.Formercentralpoliticianstypicallyhave a higher
publicprofileand betteralternativeemploymentopportunitiestha formerlocalpoliticians,
and the value of re-electionis measuredrelativeto these alternativeopportunities.
Nevertheless,to the extentthat a exceedsunity, the case for centamlisation
will be
strengthened.If a is greater than the reciprocalof the termpr(qj I qi)) in equation(2.5), it
mayeven happenthat centralisationleadsto highergovernmenteffort levelstan
decentralisationeven withoutthe internalisationof spillovers,becausethe greaterdesire of
politiciansfor re-electionoutweighstheir diminishedaccountability.
Takinginto accountthe possibilityof redistributivepolicies(thosewith negative
spillovers)somewhatcomplicatesthe conclusionsreachedabove.First of all, it is no longer
true that higherlevelsof effortare alwaysgood for the welfareof populations.Taking
externalitiesinto accountby centralisationmay actuallylowereffortlevels, and this may be
desirableif they were previouslyabove the efficientlevel (as they may have been through
failingto take the externalitiesinto account).If, however,effort was previouslybelowthe

7 A good example of this in practice is European community

merger control, where since 1990 the question whether the EC or
member states have jurisidction over mergers has been settled with
reference to the estimated magnitude of the spillovers between
member states generated by the merger in question.

50

07

-O'

efficientlevel (because, for example,W was too low to motivatepoliticiansadequately),

takingspilloversinto account mightparadoxicallymakeboth regions worse off. Region 1
could benefitfrom a reductionin the value of x2, but lose even more by a reductionin xl
consequenton the government'staing into account the spilHover
on region2.
Secondly,if a is significantlygreater than one, centralisationmay increasethe
incentivefor redistributivepoliciesthat harm some regiont. Suppose,for example, that high
valuesof xt imposelarge costs on region2, but that the increasedspoils from centralised
power make the government'senthusiasmfor pleasingregion 1 outweighits concern for
region2; it may then increasethe value of xl and make region 2 worse off. Such an
outcomemay seemextreme in this simple model,but it is evidentlymuchmore likely in
circumstanceswhere there are manyregions,some of whichenjoy positiveand some suffer
negativespilloversfrom the policiesof region i. The increasedincentivedue to internalising
the positive spilloverson the fortunateregions may outweighthe negativespilloverson the
others, especiallyif these latter are unlikelyto be the pivotalvoters in the centralisedsystem.

Note that this is not the same thing as saying that

centralisation makes available certain redistributive policies that
did not exist before, though the latter is certainly a very
important phenomenon.
8

51

Policy Research Working

Tltlb

Paper Serles
Contact
for paper

Author

Date

WPS1331 The Myth of Monopoly.A NowVlow

of Industrial Structuro In Russia

Annotlo N, Brown
Blirry W. Ickos
RandiRylorrnan

August 1994

M. Borg
30959

WPS1332 Povertyand HousoholdSizo

Polor Lanjouw
Mnilin Ravatilon

August 1994

P. Cook
33902

WPS1333 A Test of iho Inlomational

Convorgenco HypothesisUsing
Panol Date

Nornuin V. Loayza

August 1994

R. Mariln
39026

WPS1334 Taxation, Public Services, and the

InformalSoctorin a Modelof
EndogonousGrowth

Juan Braun
NormanV. Loayza

August 1994

R. Martin
39026

WPS1335 Labor Regulationsand the Informal

Economy

NormanV. Loayza

Augusl 1994

R. Martin
39026

WPS1336 ModernizingPaymentSystems In
EmergingEconomies

Robert Listlield
Augusl 1994
FemandoMontes-Negret

F. Monles-Negret
37832

WPS1337 The Countrywido Effects of Aid

HowardWhite
Joke Lutilk

August 1994

E. Khine
37471

WPS1338 CommodityExportsand the Adding- Maurice Schifl

Up Problem in DevelopingCountrles:
Trade, Investment,and Lending Policy

August 1994

A. Kim
35715

WPS1339 China's Emergence: PRospects,

Opportunities,and Challenges

August 1994

J. Queen
33740

WPS1340 Opportunity Cost and Prudentiality: HerbertL. Baer

An Analysis of FuturesClearinghouse Virginia G. France
Behavior
James T. Moser

August 1994

P. Sintim-Aboagye
38526

WPS1341 ExplainingPakistan's High Growth

Performnance
Over the PastTwo
Decades:Can It Be Sustained?

SadiqAhmed

August 1994

A. Bhalia
84440

WPS1342 Winners and Losers in Transiticn:

Retums to Education, Experience,
and Genderin Slovenia

Peter F. Orazem
MilanVodopivec

August 1994

Andres Boltho
Ur Dadush
DongHe
Shigeru Otsubo

WPS1343 Strategic Interdependencein the

Wafik Grais
East-WestGas Trade: A Hierarchical KangbinZheng
StackelbergGame Approach

August`1994

K Zheng
36974

Policy Research Working Paper Series

Tiltle

Author

Date

Contact
for paper

WPS1344 Which Foreign InvestorsWony About Eric Bond

Foreign ExchangeRisk In South
Antonio Estache
Asia and Why?

August 1994

A. Eslache
81442

WPS1345 The Decentralizationof Public

Services: Lessonsfrom the Theory
of the Firm

Jacques Cremer
Antonio EstRche
Paul Seabright

Augusi 1994

A. Estachs
81442

WPS1346 Linking Compelillonand Trade

Policies In Central and Eastem
EurcpeanCountries

Bemard M. Hoekman
Petros C. Mavroldis

August 1994

F. Hatab
35835

WPS1347 Anlitnust-BasedRemedlesand
Dumping In InternationalTrade

BemardM. Hoekman
PetrosC. Mavroidis

August 1994

F. Hatab
35835