0034-6527/95/00240541$02.00
Reviewof EconomicStudies(1995) 62, 541-555
? 1995The Reviewof EconomicStudiesLimited
Credit
and
Efficiency
in
Centralized and
DecentralizedEconomies
M. DEWATRIPONT
DULBEA and ECARE (Universite Libre de Bruxelles), CEPR and CORE
and
E. MASKIN
Harvard University
First version received August 1990; final ver'sion accepted Apr-il1995 (Eds.)
We study a credit model where, because of adverseselection, unprofitableprojectsmay
neverthelessbe financed.Indeedthey may continueto be financedeven when shown to be lowoffersa way
qualityif sunkcosts havealreadybeenincurred.We show that creditdecentralization
for creditorsto commitnot to refinancesuch projects,therebydiscouragingentrepreneursfrom
undertakingthem initially.Thus, decentralizationprovidesfinancialdiscipline.Nevertheless,we
arguethat it puts too high a premiumon short-termreturns.
The model seemspertinentto two issues: "soft budgetconstraint"problemsin centralized
financingpractices.
economies,and differencesbetween"Anglo-Saxon"and "German-Japanese"
1. INTRODUCTION
We investigatehow the degree to which credit marketsare centralizedaffectsefficiency
whenthereis asymmetricinformation.Specifically,we arguethat decentralizationof credit
may promoteefficientprojectselectionwhencreditorsare not fullyinformedex ante about
projectquality.
Our startingpoint is the idea that, althoughan entrepreneur(projectmanager)may
have a relativelygood idea of her project'squality from the outset, creditorsacquirethis
informationonly later on, by which time the criteriafor profitabilitymay have changed.
Thus, a poor project (one whose completion time is too long to be profitableex ante)
may neverthelessbe financed,since a creditorcannot distinguishit at the time from a
good (quick) project.Moreover,the projectmay not be terminatedeven afterthe creditor
has discoveredits quality,if significantsunkcosts have alreadybeen incurred.If the threat
of terminationdeterredentrepreneursfrom undertakingpoor projectsin the first place,
creditorswould wish to commit ex ante not to refinancethem. But, sunk costs may well
renderthis threatincredible:ex post, both creditorand entrepreneurcould be better off
carryingon with the project,i.e. refinancingit.
We conceiveof a decentralized
How can decentralizationhelp in such circumstances?
credit marketas one in which ownershipof capital is diffuse,so that the capital needed
to refinancea poor projectmay be availablebut not in the hands of the initial creditor.
This creditor,we assume,can monitorthe projectand therebyenhanceits value.However,
monitoringis not observableto subsequentcreditors.Consequently,the initialcreditor's
541
542
REVIEW OF ECONOMIC STUDIES
incentiveto monitor is blunted (relativeto a centralizedmarketwhere he owned all the
capital) because he cannot fully appropriatethe marginalreturnfrom doing so. With
incentivesreduced,he will monitor less than undercentralization,which in turn reduces
the value of the projectand thereforethe profitabilityof refinancing.That is, refinancing
is less likelythanin a centralizedmarket;the threatto terminatea projectis morecredible.'
Entrepreneursare therebyinducednot to undertakepoor projectsin the first place, and
this enhancesefficiency.2
Decentralizationtends to deterprojectsthat dragon too long, but for similarreasons
may also discourageprofitableprojectsthat are slow to pay off. That is, the samefeatures
that strengthencommitmentsto terminatepoor projectsfosteran over-emphasison shorttermprofit opportunities.
model we have sketched is
To see this, suppose that the slow-and-quick-project
enrichedso that not only poor projectsbut also highly profitableprojectsrequirelongtermfinancing.Poor (i.e. inept) entrepreneursare stuckwith poor projects,but good (i.e.
capable) entrepreneurshave a choice about whethertheir projectis to be long-termand
highly profitableor short-termand only moderatelyprofitable.Finally, suppose that the
That is, owners
degreeof decentralizationin the creditmarketis determinedendogenously.
of capital can come togetherand choosewhetherto form a few big "banks"or a lot of
small banks.
In such a model multiple equilibria (and, hence coordinationproblems) may well arise.
If, in equilibrium,banks are small, even good entrepreneurswill have trouble getting
continuedfinancingfor long-termprojectsfor the reasonsmentionedabove. Thus they
will choose the short-termoption. But given that they do so, it will pay banksto be small
(a single big bank would be overrun with unprofitablelong-termprojects from poor
entrepreneurs).Thus, an equilibriumwith only short-termprojectsand smallbanksexists.
But another equilibriumis also possible, one in which all banks are big. With a
profusion of big banks, good entrepreneurscan get long-termfinancingand so choose
highly lucrativeprojects.The profits from these projectsoutweighthe losses that banks
incur from poor projects (which because of adverseselection are also financed). Such
a "long-term"equilibriumcan, in fact, be shown to Pareto-dominatethe "short-term"
equilibrium.
We believethat our frameworkmay be relevantfor two widely-discussedissues: the
"soft budget constraint"problem of centrally-plannedeconomies and the contrast in
financingpracticesand investmenthorizons between economies of the "Anglo-Saxon"
and "Japanese-German"modes.
Kornai(1979, 1980)hasemphasizedthat the absenceof bankruptcythreatsin socialist
economiesresultedin the proliferationof inefficiententerprises.Firms realizedthat their
losses would be coveredby the state, and so operatedquiteindependentlyof profitconsiderations. The pervasivenessof these soft budget constraintsunder socialism is widely
acknowledged,and attemptsto hardenthemarecentralfeaturesof severalrecentproposals
for reformin easternEurope.
But althoughthe consequencesof soft budgetconstraintshavebeenintensivelyinvestigated, the same is not true of their causes. Most explanationshave focused on political
1. Lack of commitment in centralized settings has been the focus of the ratchet effect literature. (See, for
example, Freixas et al. (1985), Laffont and Tirole (1988), and Schaffer (1989)). What remains unsettled in this
particular literature, however, is why lack of commitment slhould pertain particularly to centralization. Our
paper attempts an answer to that question.
2. As in Stiglitz and Weiss (1981), creditors face an adverse selection problem. In the Stiglitz-Weiss model,
credit rationing is a way to deal with this problem and improve the mix of projects being financed. In our setting,
by contrast, it is the threat of termination that serves as the device for screening out poor projects.
DEWATRIPONT & MASKIN
CREDIT AND EFFICIENCY
543
constraints,such as the need to avoid unemploymentor sociallycostly relocation.While
not denyingthe importanceof such constraints,we wish to suggestthat economicfactors
model outlinedabove (and
may also be relevant.Specifically,the slow-and-quick-project
presentedin detail in Sections 2 and 3) offers an explanationof soft budget constraints
in which "softness"arises from the profitabilityof refinancingpoor projects.Indeed, in
our framework,softness is the "normal"state of affairs;the pertinentquestion is how,
in some circumstances(e.g. a decentralizedcredit market), budget constraintscan be
hardened.3
We can also apply the frameworkto explaindifferencesbetweenAnglo-Saxon(U.S.
and U.K.) and German-Japanesecorporatefinance.Severaleconomistshave noted that
large German or Japanesefirms have been more likely to obtain financingfrom banks
than their Americanor Britishcounterparts(which have reliedmore on equity or bonds
for externalfinance). Moreover,these banking relationshipshave typicallyhad a longtermstructurein whichbanksassumedan activemonitoringrole. (See Aoki (1990), Baliga
and Polak (1994), Corbett (1987), Edwardsand Fischer (1994),4Mayer and Alexander
(1989) and Hoshi, Kashyap and Scharfstein(1988, 1989).) Most importantfrom our
financialcontrastseemsto be markedby
standpoint,the Anglo-Saxon/German-Japanese
differencesin projectlength.Specifically,Germanand Japanesecorporationshave seemed
(see for exampleCorbett(1987) and TheEconomist(1990)).
less proneto "short-termism"
Althoughhighly stylized,the enrichedmodel sketchedabove, is consistentwith these
differences.The "long-term"equilibriumaccordswith German-Japaneseexperience,and
the "short-term"equilibriumwith that of the U.S. and U.K.
We proceedas follows. In Section2, we presenta very simple(in some respects,oversimplified)model and show how credit decentralizationcan improveefficiency.We then
discuss severalalternativespecificationsthat lead to the same conclusions.In particular,
we arguethat the contrastbetweencentralizationand decentralizationis only heightened
if we suppose, following one tradition, that the central financingauthoritymaximizes
social surplusratherthan profit.
Section 2 distinguishesdecentralizationfrom centralizationrathercrudelyby identifying the formerwith two creditorsand the latterwith one. In Sections 3 and 4 we turn
to a richermodel in which the market structureis determinedendogenously.Section 3
establishesthat the main qualitativeconclusionsof Section 2 carryover to a framework
in which marketstructureis determinedendogenously.Finally,Section4 introducesprofitable long-runprojectsas an additionaloption for good entrepreneursand shows that
therecan be two (Pareto-ranked)equilibriamarkedby differentaverageprojectlengths.
2. DECENTRALIZATION AS A COMMITMENT DEVICE
a. The Model
Thereare threeperiods,one entrepreneur,and eitherone or two creditors(banks). Contractingbetweenthe entrepreneurand a bank occursin period0, and projectsare carried
out in periods I and 2. If a project remains incomplete at the end of period 1, the
entrepreneurand bankcan renegotiatethe termsof the contractto theirmutualadvantage.
3. Qian and Xu (1991) and Qian (1994) have used this approachto show how soft budgetconstraints
both interferewith innovationand can contributeto the endemicshortagesthat plaguesocialism.
4. Edwardsand Fischer(1994) note, however,that the relianceon externalfinanceamongGermanbanks
has not been so greatas commonlysupposed.
544
REVIEWOF ECONOMICSTUDIES
The entrepreneur'sproject can be either good (g) or poor (p). A good project is
completed after one period; a poor project requirestwo periods for completion. (We
identifythe qualityof a projectwith that of its entrepreneur;thus, we shall referto good
and poor entrepreneurs).The projectgeneratesan observable(and verifiable)monetary
returnonly at its completion.Whethergood or poor, it requiresone unit of capital per
period (all returns,capitalinputs, and payoffsare denominatedin money).
The entrepreneurhas no capital herself and so has to obtain financingfrom the
bank(s). Banks have capital but cannot initiallydistinguishbetweengood and poor projects. Let a be the prior probabilitythat the projectis good. All partiesare risk neutral,
i.e. they maximizeexpectedprofit.
For the time being we will assign no bargainingpower to the entrepreneur(we will
relaxthis assumptionin Section3). Thus, in negotiatingfinancialterms,a bankcan make
a take-it-or-leave-itoffer to the entrepreneurand therebyextract the entire observable
return.The entrepreneuris limitedto unobservableprivatebenefitssuch as the perquisites
she can command,the enhancementof her human capital and reputation,or what she
can divertfrom the projectinto her own pocket.
Let Eg be a good entrepreneur'sprivate benefit.E, is a poor entrepreneur'sbenefit
when her project is terminatedafter the first period, whereasEp is her benefit from a
completedproject.We assumethat Ep? E,. This inequalitymakessenseif we imaginethat
the entrepreneurcan extract more from a projectthe longer it continues. It would also
follow from a more elaboratemodel in which her reputationis enhancedif the projectis
completed.In any case, it musthold in any modelin whichpoor projectsareeverrefinanced
(providedthat the entrepreneuralwayshas the option of quittingafter the firstperiod).'
We allow for the possibilitythat any of Eg, E,, and Epmay be negative,6which could
occur, for example,if privatebenefitsincludethe cost of effortthat the entrepreneurmust
incur to set the projectup.
Considercentralizationfirst. In this case, thereis a single bank B endowedwith two
units of capital. In period0, the entrepreneurE (whose type is privateinformation)turns
up and requestsfinancing(i.e. a loan of one unit of capital). B makes a take-it-or-leaveit contractofferin which the repaymenttermsdependon the observablereturnand when
it is realized(becauseE has no endowment,the repaymentcannot exceed the observable
return7).Assume that a good projectgeneratesobservablereturnRg> 1, which, given its
bargainingpower, B can fully extract(providedthat Eg>O; if Eg<O, B can extractonly
Rg+ Eg becauseE will requirean inducement-Eg to undertakethe project).
If the projectis poor, B obtainsnothingunlesshe agreesto refinancingat the beginning
of period 2, i.e. agreesto loan anotherunit of capital8(since the observablereturnis zero
at the end of the first period). Moreover,we assume that regardlessof the first period
agreement,B cannot commit himselfnot to refinance(or, rather,that any such commitment can be renegotiated).If refinanced,the poor project'sobservablereturnat the end
of the second period is a randomvariableRp,whose realizationis either0 or Rp, where
0 < Rp,,.(We could allow Rgto be a randomvariableas well, but this would not matterin
view of the parties'risk neutrality.)One can interpretR,,as the liquidationor resalevalue
of the completedproject.We suppose that, in addition to its role as lender,B serves to
5. And it is precisely the problem created by refinancing poor projects that is of interest to us.
6. As we shall see, in fact, the major case of interest for our purposes is where E, < 0 and Ep>0.
7. This is not necessarily true if thle private return is known to be positive and bounded away from zero.
But as long as B is uncertain about the value of this private return, he will not be able to extract it fully.
8. Here we are assuming for convenience that E cannot contribute any of what she may have saved from
the first loan to reduce the size of the second.
DEWATRIPONT & MASKIN
CREDIT AND EFFICIENCY
545
TABLE 1
PaYoffs under centralization
Good project
(assumingEg>0)
Entrepreneur
Bank
Poor project
withoutrefinancing
E,
-1
Eg
Rg-I
Poor project
with refinancing
Ep
Hlp- 2
monitor the project.9This is modeledby assumingthat, throughhis efforts,B can influence
the distributionof Rp.'0Assumethat B learnsE's type at the beginningof period 1. If E
is poor, B can expendmonitoringeffort ae [0, 1] to raisethe expectationof Rp.Specifically,
let a be the probabilityof Rp. As a rises, so does the cost of B's efforts.Let I(a) denote
this cost, with V'> 0, V"> 0, V/(0)= V/'(0)= 0, and M'(1) = oo. These assumptionsensure
an optimal effort level a*e (0, I) such that Rp= yl'(a*) and, given its bargainingpower,
an expectedreturnfor B (gross of its capital investment)of n*_ a*Rp-yd(a*).
To summarize,the payoffs (net of the cost of capital) of the entrepreneurand bank
undercentralizationare displayedin Table 1.
Under decentralization,the model is much the same as above, but now assumethat
there are two banks, B, and B2, each with only one unit of capital. The entrepreneur
presents herself to B, in at the beginningof period I (we will postpone the issue of
competitionbetweenbanks until Section 3). If she turns out to be good, the analysisis
as above. The same is true if she is poor but not refinanced.If, however, she is to be
refinanced,she must turn to B2, since by then B1 has no capitalleft." Supposethat any
monitoringthat B, has done in period I is unobservableto B2.
For the sake of comparability,we assign B2 no bargainingpower so that, as in the
case of centralization,B1 can make take-it-or-leave-itoffers. The problem for B, is to
convinceB2 to loan a secondunit of capitalin exchangeof a shareof Rp. The higherB2's
expectationof Bl's monitoringeffort in period 1, the smallerthis sharecan be. We claim
that equilibriummonitoringeffort is less than a* (the effort level under centralization),
despite the fact that endowing B1 with all the bargainingpower maximizeshis incentive
to monitor.To see this, let a be B2'sassessmentof the expectedlevel of Bl's monitoring
activity.Then, to induce B2 to participate,the repaymenthe receivesmust be Il/a if Rip=
Rp. This means that B, chooses a to maximize
a(Rp- I /la) -(a),
I/a= yl'(a).'2 Now, in equilibrium, a must be correct, so that if a** is
the equilibriumeffort level, a** satisfiesRp= yl'(a**)+ I/a**.I3 Clearly,a** is less than
a* (becauseB, concedespart of the marginalreturnfrom monitoringto B2). Therefore,
np * _ a**Rp- yr(a** ) is less than nH*.
i.e. to satisfy Rp-
9. In the 1990versionof this paper,we assumedthat, insteadof monitoring,B acquiresinformationabout
the projectthat it can use to affectthe distributionof Rp,.
10. We could also assumethat monitoringaffectsthe realizationof Rg. Becausesuch monitoringwould
play no role in our analysis,however,we do not considerit.
11. Actually,all that is neededfor our purposesis that B1 shouldnot be willingor able to undertakeall
the refinancingitself. Indeed,even if B, had morethan 1 unit of capitalleft, B2would still have to be brought
in if B1 weresufficientlyrisk averse.
12. This first-orderconditionis valid providedthat R,,> l/a. Otherwise,the maximizingchoice of a is
a=0.
13. If thereis no solutionto this equation,then a**= 0 (see footnote 12). If thereare severalsolutions,
due simplyto coordination
choose the one that maximizesa(Rp- I/a) - v(a), in orderto ruleout inefficiencies
failure.
REVIEWOF ECONOMICSTUDIES
546
TABLE 2
Payoffs under decentralization
Good project
(if Eg> 0)
E
B1
Eg
Rg -1
B2
0
Poor project
with no refinancing
Poor project
with refinancing
E,
Ep
**-2
0
0
Recappingwe exhibitthe (net) equilibriumpayoffsunderdecentralizationin Table2.
We are interestedin comparingthe (perfectBayesian)equilibriaundercentralization
and decentralization,and, especially,in investigatinghow these two alternativesfare in
deterringpoor entrepreneurs.For these purposes,it makes sense to suppose that poor
projects generate negative "social surplus" (nf + Ep < 2),14 that good projects have positive
surplus (Rg + Eg> 1), and that poor entrepreneurs are deterred only by termination'5
(E, < 0 < Ep). We shall (briefly) consider the other cases after Proposition I and in Section
3. (Not surprisingly,centralizationand decentralizationperformvery similarlyin most of
those other cases.)
Proposition 1. Assume that Ep> 0 > E,. Undereither centralizationor decentralization,
there exists a uniqueequilibrium.For parameter values such that somefinancing is undertaken
in equilibrium,a necessary and sufficient conditionfor project selection to differ in the two
> 1> p**. If this condition holds, only a good project is financed under
equilibria is npH
decentralization (the socially efficient outcome); both good and poor p-ojects are financed
(and the latter refinanced) under centralization.'6
< 1, then it is inefficientto refinancea poor project under
Sketch of Proof. If npH
centralization(and a fortior-i underdecentralization).Thus, a poor entrepreneurwill not
seek financing(since E, <0), and so only a good projectis financedunderboth centralization and decentralization.If Ip* > 1, then once even a poor projectis started,partieswill
end up refinancingit underdecentralization(and afortiori undercentralization).Because
Ep> 0, we conclude that a poor entrepreneurwill gain by getting funded and so, under
14. Even if n, + El,> 2, a poor project may not necessarily be desirable. In view of the unobservability of
the entrepreneur's private return, she cannot be made to compensate the centralized creditor for its negative
profit nZ- 2. Thus the project's desirability will depend on the creditor's and entrepreneur's relative weights in
the social welfare function. However, if npI+ Ep< 2, then a slow project is unambiguously inefficient.
15. Poor entrepreneurs might be threatened by legal sanctions (e.g. the threat of being thrown in jail),
which could have a deterrent effect. However, if these entrepreneurs are needed for the completion of the project
in the second period, such threats may not be very credible.
16. As modelled, negotiation between the entrepreneur and the bank can occur only after period I has
elapsed, i.e., after one unit of capital has already been sunk. Let us consider what would happen if regeneration
were also permitted befoorethe capital is sunk (but after the initial financing contract has been signed). In that
case, the bank could propose returning the first period's capital unused in exchange for a fee of E, + ?. A poor
entrepreneur would accept this deal, whereas a good entrepreneur would not (provided that Ep+ E < Eg). Moreover, given our assumption that nH+ Ep-2<0, the bank would be better off. To rule out such a peculiar
outcome, we can suppose that, in addition to good and poor projects, there is a third type that is so dreadful
that refinancing is never desirable but for which the entrepreneur's payoff is positive if financed for even one
period. Let us suppose that, with high probability, the quality of such a project is detected by the bank before
the capital is sunk. Nevertheless if the probability is less one, dreadful entrepreneurs will still seek financing.
Therefore, the bank will thwart its detection mechanism and seriously interfere with efficiency if it offers the
above deal.
DEWATRIPONT& MASKIN
CREDIT AND EFFICIENCY
547
both decentralizationand centralization,both types of projectswill be financed.'7Finally,
if lnp*< I < Hp*,refinancingis efficientundercentralizationbut not underdecentralization.
Hence, a poor projectwill be fundedin the formercase but not the latter. 11
Hence,eithercentralizationand decentralizationlead to the sameprojectselectionin
equilibrium,'8or else decentralizationis strictly better, i.e. it selects efficientlywhereas
centralizationis subjectto a soft budgetconstraint.
We have been assumingthat E,<O < Ep. If E,>O, then terminationdoes not detera
poor entrepreneurfrom seekingfinancing,and both poor and good projectsare financed
under either centralizationor decentralization(although that is not to say that the two
systems are equally efficient; see footnote 17). If Ep<O, then only good projects are
financedundereithersystem.
b. Alternative Specifications
We have modeled the initial projectselectionas a problemof adverseselectionand refinancingas one of moral hazard,but these imperfectionscan readilybe switchedaround.
Specifically,suppose that insteadof projectlength being given exogenously,E can affect
it through (unobservable)effort. Under centralization,B could rewardthe entrepreneur
for early completion, but such a rewardmight make financingunattractivefrom B's
perspective.The advantageof decentralizationwould be to induce E to complete early
without having to rewardher; the threat of terminationwould be inducementenough.
Such an alternativemodel should yield qualitativelyvery similarresults. Undoubtedly,
both specificationsare relevantin reality.
By the same token, B2'sinformationaldisadvantagehas been formallyexpressedas
a problemof moral hazardbut could alternativelybe derivedfrom adverseselectionand
collusionbetweenE and Bl. Let us, for example,drop Bl's effortfrom the model (so that
Rp'sdistributionbecomesexogenous)but also abandonthe assumptionthat Rp'srealization is verifiable.InterpretBl's informationaladvantageas the abilityto prove to a court
that Rp=Rp, if that equalityholds. As in models of hierarchies(Tirole (1986), Kofman
and Lawarree(1993)), collusion betweentwo partieswho share some informationmay
preventa third party without access to that informationfrom sharingthe benefits.Here
it would be in B,'s and E's joint interestto agree to conceal the evidencethat Rp=Rp
(puttingaside the unresolvedtheoreticalissue of how such an agreementwould be enforced) in orderto preventB2from extractingsome of the return.Hencedecentralization,by
givingrise to collusion,reducesthe incentiveto refinancepoor projects,as in subsectiona.
In our model, it is the non-transferabilityof informationthat makes multi-creditor
financialarrangementsproblematic.But there is a related(yet informal)'9idea from the
financeliteraturethatwouldserveour purposesjust as well:the principlethatrenegotiation
17. This reliesoniour assumptionthat some financingis undertakenin equilibrium.If this assumptionis
violated,then it is possiblethat no projectsare financedundereithersystem,or even that both are financed
undercentralizationand neitherunderdecentralization.The latter possibilityis an artefact,hiowever,of the
crude way we have modelleddecentralization.If the marketstructureis determinedendogenously(as in the
disappears.
model of Section3), this particulardiscrepancybetweencentralizationand decentralization
18. But not necessarilythe same degreeof efficiency.If 1,,**> 1, both centralizationand decentralization
select the same projects,but the formeris more efficient,since nH> fl**. However,this discrepancyderives
(see footnote 16). In the more satisfactorymodelof Section
model of decentralization
from our over-simplified
areequallyefficientin the case wheretheymakethe sameprojectselection.
3, centralizationand decentralization
19. See Boltonand Scharfstein(1994)and Hartand Moore(1995) for two recentcontributionsthat build
upon this insight.
548
REVIEW OF ECONOMIC STUDIES
becomesmore difficultto coordinatethe more partiesare involved.From this standpoint,
havingtwo creditorsreducesthe chancesof refinancingbecausegettingthem to agreeto
it is harder.
We haveendowedthe creditorsin both the centralizedand decentralizedmodelswith
the same objective:expectedprofitmaximization.But ever since Langeand Lernerit has
been commonpracticeto have the centrein plannedeconomymodelsmaximizeexpected
socialsurplus.To do so herewould,in fact, only aggravatethe inefficiencyof centralization.
To see this, recallthat centralization'sshortcomingis that it promotes"too much" refinancing.Now if, at the beginningof period2, the creditortakes into accounttotal social
surplusratherthanjust its own profit (see footnote 14, however,for why social surplus
is not unambiguouslythe best measureof efficiencyin this model), the criterionfor refinancing would become Ep+ n* > 1, i.e. it would be more relaxed than before and so
refinancingwould occur even more readily.
3. EQUILIBRIUMIN A DECENTRALIZEDCREDIT MARKET
in nature,involvThe contractingmodelof the previoussectionis rather"microeconomic"
andat mosttwo creditors.For thecase of centralization,assuming
ing a singleentrepreneur
only a single creditorseems quite reasonable;in many centralizedeconomies, the state
has beenthe only significantlender.However,to equatedecentralizationwith the existence
of two banksis fairlyheroic(or foolhardy).Moreover,our model leavesout two ingredients that are important features of decentralizedcredit markets, namely, competition
among creditorsand the endogenousdeterminationof the marketstructure.
Thus in this section, we enrich the previousmodel of decentralizationby assuming
that there is an indefinitelylarge populationof (identical)investors,each endowedwith
a smallamountof capital.Thus, as in the introduction,a decentralizedmarketis one with
diffuseownership,in the sense that thereare many smallinvestors.Investors,however,are
allowedto joinforces at the beginningof period I, to form banks. Each bank has capital
equal to the sum of its investors'endowmentsand should be viewedas a cooperative,i.e.
as managedjointly with all membershaving access to the informationacquiredwhen
monitoringa project. (Actually,given our risk-neutralityassumption,we could alternatively assumethat joining forces entails settingup a lottery that gives each participanta
chanceto receiveall the capital).But the transferof informationacrossbanksis assumed
to be impossible.
We assumethat thereis a populationn of entrepreneurs,each drawnindependently
from a distributionin which there is a probabilitya of being good. Although n should
be thought of as large, the indefinitesupplyof capital (which we may suppose takes the
form of a liquidasset with interestratenormalizedto zero) ensuresthat everyprojectcan
in principlebe financed.Operationally,this will have the effectof drivingcreditors'profits
to zero throughcompetition(i.e. the entrepreneurwill now retainsome of the observable
returnherself).
As for centralization,we modify the model of subsection2a only by adopting the
and by supposingthat the singlecreditorhas enough
above assumptionof n entrepreneurs
The
earlieranalysis of equilibriumin the centralized
them
all.
capital to accommodate
over
carries
case clearly
completely.
As modelled,centralizationdiffersfrom decentralizationin two respects:ownership
of capital and transferabilityof information.It is this combinationof attributesthat
generatesour results.Of course, we are idealizingthe quality of the flow of information
withina centralizedhierarchy,and our perspectiveis quite "un-Hayekian"in that respect.
DEWATRIPONT& MASKIN
CREDIT AND EFFICIENCY
549
Still this flow, howeverimperfect,is likelyto be betterthan the transferabilityof information betweenseparate(and competing)hierarchies(e.g. rival banks).
The timingof our modifieddecentralizationmodel is as follows. At the beginningof
period 1, investorscan join forces to form banks (in equilibrium,not all investorsneed
do so). An investorcan contributehis capitalto a bank of any "size"he chooses (because
all creditorsare identicaland thereare indefinitelymany of them, he will be able to find
sufficientlymany other like-mindedinvestorsin equilibriumto actuallyform the bank).
At the same time, each bank/creditoroffersa set of contracts(a contractis the same as
in Section2). Entrepreneursthen choose amongcontracts.If more than one entrepreneur
chooses the same contract, then there has to be rationing(see below). If after period 1
someprojectsarenot yet complete,existingor newcreditorscan offerrefinancingcontracts.
The affectedentrepreneursthen choose among thesecontracts(again,possiblywith some
rationing).
Becauseeveryoneis risk neutral,there is no advantageto diversificationper se, and
so in equilibriumthe largestcreditorthat need form is one with two units of capital.We
shall refer to creditorswith one and two units of capital20as small and large creditors,
respectively.
Notice that what we are referringto as a bank's "size" is more accuratelythought
of as the bank's liquidity-how much of its assets are availableto be loaned out-which
may bear little relation to its literal size, i.e. total assets. Thus, the terms "small"and
"large"creditormightmoreproperlybe relabelled"illiquid"and "liquid"creditor.(From
this perspective,soft budgetconstraintsarise in a centralizedeconomybecausethe centre
is too liquid, e.g. it can print money to refinanceprojects.)
A smallcreditormust investall its capitalin a singleprojectif it is to do any financing
in period 1. In this case the refinancingproblemis the same as in subsection2a.
A big creditorhas two choices: it can fund a singleprojectand keep its second unit
of capitalliquid,or it can financetwo projects,thus sinkingall its capital.Such a creditor
is, respectively,denoted diversifiedor undiversified(the usage here is not quite standard
because,as noted, ordinarydiversificationplays no role). A poor entrepreneurfinanced
by a diversifiedcreditorknows that its chance of being refinancedis the same as in the
centralizedmodelof subsection2a. Whenthe creditoris undiversified,however,refinancing
possibilitiesdependon its mix of projects.Indeed,a poor entrepreneurfinancedby such
a creditoris in the same situation, if the creditor'sother projectis also slow, as though
financedby a small creditor.In this sense, lack of diversificationis a substitutefor being
small. However,it is not a perfectsubstitutebecausethe poor entrepreneurcan obtain refinancingif the otherprojectis good. (The creditorcan use the returnon the good project
eitherto refinancethe poor one directlyor-if this returnis realizedtoo late-as collateral
againsta loan from anotherbank.)
As we have mentioned,entrepreneurshave to be rationedif more than one chooses
the same financingcontract. By a rationingschemlewe mean a rule that, for any set of
contractsthatcould be offered,specifies,for eachcontractin the set andeachentrepreneur,
the probabilitythat the contract is assignedthis entrepreneur.For our purposes,many
differentschemes would do. For concreteness,we concentrateon the following simple
scheme:
20. Notice that it is of no valueand possiblyactuallyharmfulto have strictlybetweenone and two units
of capital.To preventrefinancingfromoccurring,it is betterto have one unit. And if refinancingdoes occur,it
is betterto have two.
550
REVIEW OF ECONOMIC STUDIES
of a giventypeare firstallocateduniformly
TheRationingScheme:21 All entrepreneurs
over the set of their favouritecontracts(this reflectsthe attemptby an entrepreneurto
choose the best contractfor herself); if thereare fewerentrepreneursof a given type than
favouritecontracts,the entrepreneursare allocatedat randomto these contracts.If only
one entrepreneuris allocated to a given contract, she is assignedto that contract with
probabilityone. If more than one is allocated,each has an equalchanceof beingassigned.
At the end of this round, the procedureis repeatedwith all entrepreneursand contracts
not yet assigned.The process continuesiterativelyuntil either the supply of unassigned
entrepreneursor that of desirablecontracts(those that are preferredto no contractat all)
is exhausted.
Insteadof modellingentrepreneurs'behaviourexplicitly,we shall subsumeit within
the rationingscheme(whichis appliedboth after the period 1 and period2 contractsare
offered).We can thus defineequilibriumin termsof creditors'behaviouralone.22
Equilibrium. An equilibriumis a configurationof creditors,each creditor'sset of
period I contracts(possiblyempty), and each creditor'srefinancingstrategy(the period
2 contractsit offers as a function of what happenedin the first period) such that, given
the rationingscheme,
(i) eachcreditorearnsnon-negativeexpectedprofiton eachof its contracts(whether
first or second period) given other creditors'contracts and their refinancing
strategies;
(ii) thereis no otherset of contractsthat a creditorcould offerand no otherrefinancing strategythat, given others'behaviour,would earn higherexpectedprofit;
(iii) thereis no groupof inactiveinvestors(i.e. investorswho do not alreadyform a
bank) who could come togetherto becomea creditorwith a set of contractsand
a refinancingstrategythat, given the behaviourof the alreadyexistingcreditors,
makes strictlypositiveexpectedprofit.
We will focus on pure-strategyequilibria(where,moreover,all creditorsof a given
size offer the same contracts).
As in Section 2, we are interestedin comparingequilibriaundercentralizationand
decentralization.Once again, the interestingcase (i.e. the case wherethereis a significant
difference)is Ep>0 > E,, and so we shall stick to this assumption.We shall also continue
to assume that Eg+Rg >I (good projectsare efficient),and, that Ep+n*< 2, i.e. poor
projectsare inefficient(but see the discussionof equilibriumefficiencyafter Proposition
3, wherethis is relaxed).
When np>l> np*, we have seen that the centralizedoutcome entails inefficient
projectselection:both good and poor projectsarefinanced.Accordingto the simplemodel
of Section2, decentralizationhardensthe budgetconstraintand inducesan efficientoutcome in which only good projectsare funded.We now observethat the same conclusion
i.e. the possibilitythat an
21. We ignore the issue of strategicbehaviouron the part of entrepreneurs,
will choose a less favouredcontractbecauseshe has a betterchanceof beingassignedit. However,
entrepreneur
are, etc., such behaviourwould not be optimalin
with enough uncertaintyabout wlhothe otherentrepreneurs
any case.
22. Actually,it is investors,ratlherthan creditors,who are the basicdecision-makingunits.We findit too
cumbersome,however,to defineequilibriumin termsof investorbehaviour.Whicheverway one does it, the
"natural"notion of equilibriumis not entirelyclear.This is becauseif an investorcontemplatesjoininga bank
of a givensize lhemustcomparethe correspondingpayoffwith what he wouldget if he joined some otherbank.
But what is he to supposehappensto the firstbank if he does not join it? (The answermay well be relevantto
his payoff.) That it finds a replacementfor him?That it does not form at all? Implicitly,our definitionof
equilibriumadoptsthe formerhypothesis.
DEWATRIPONT& MASKIN
CREDIT AND EFFICIENCY
551
obtainsfor our more elaboratemodel (Proposition2 shows that an equilibriumwith this
hardeningfeature exists, and Proposition 3 demonstratesthat it is essentiallyunique).
Basically,this is because creditorswould like to avoid financingpoor projects.Hence,
whether or not there is competitionamong them, they will extract all the observable
surplus from such projects.And so, even in this more elaboratemodel, the condition
HI>> fI * continues to imply that refinancingwill occur with big creditorsbut not
small.
*H . There exists an equilibriun in which each of
Proposition 2. Suppose p*> 1> np
n + 1 or more smnall(one-unit) creditors offers a first-period contract that just breaks even
on good entrepreneurs.No big creditors (two or more units) offerfirst-period contracts.
Proof Eachof thesesmallcreditorsearnszero profitbecauseit breakseven on good
entrepreneursand does not attractpoor entrepreneurs(since they cannot be refinanced).
Moreover,none of these creditorscould make positive profit by deviatingbecause any
contract that earned positive profit on good entrepreneurswould not (in view of the
rationingscheme) be allocatedany of them since there are enough other small creditors
(that is, at leastn) offeringmorefavourabletermsto accommodateall good entrepreneurs.
Finally, no new creditorcan enter and make positive profit: it cannot make money on
good entrepreneursfor the reasonjust given, and if it attractedpoor entrepreneurs(which
> 1> npV*),it would lose
would requirethat it consist of two or more units since
money on them since n < 2.
Proposition 3.
Proposition 2.23
If np > 1> [p *, then the only equilibriumnis that described in
Proof. We first show that there cannot be an equilibriumin which a big creditor
offers any first-periodcontracts. If there were such contractsin equilibrium,then there
would be one to which a poor entrepreneuris assignedwith positiveprobability.(A poor
entrepreneurcan earn a positive returnonly from big creditors,contractsbecause,since
np > I >Hn**,only these are refinanced.Indeed, if such a contract is refinanced,the
entrepreneur'sreturnis certainlypositive. This will be the case when the big creditoris
diversified,but also when undiversifiedprovidedthat the other projectfinancedis good.
Thus, it cannot be the case that every big creditorcontractis assignedonly good entrepreneurs.)Of the contractsthat are assignedpoor entrepreneurswith positiveprobability,
let cobe the one that gives poor entrepreneursthe best terms.Contractcoearnsa negative
returnon poor entrepreneurs(sinceflp*< 2), and so, in orderto earna non-negativereturn
over all, it must earn a strictly positive returnon good entrepreneursand be assigned
them with positive probability.Suppose that a group of investorswho are inactive in
equilibriumcome togetheras a small creditorand offer a contractcoowith slightlymore
favourabletermsfor good entrepreneursthan co (i.e. the contractcooslightly"undercuts"
c?). This contractcoomust be assignedgood entrepreneurs.But becauseit is not refinanced
(since nH*< 1) it will not be assignedpoor entrepreneurs.Therefore,it makes positive
profit overall, a contradiction.We conclude that big creditorscannot offer first-period
contractsin equilibrium.
23. Actually,Proposition2 describesa inultiplicityof equilibriain which the numberof activecreditors
can vary as long as it exceedsit + I. However,this sort of non-uniquenessis clearlynot essential.
552
REVIEWOF ECONOMICSTUDIES
We next observethat the only contractthat is acceptedwith positive probabilityin
equilibriumis the break-evencontract for good entrepreneurs.A contract that offered
more favourableterms to good entrepreneurswould lose money, and a less favourable
contractwould makepositiveprofitand so induceentryand slightundercuttingas above.
Finally, theremust be at least n + I small creditorsofferingthe break-evencontract
in equilibrium.Otherwise,a smallcreditorcould enterand offera contractthat, if assigned
to a good entrepreneur,wouldmakea profit(and also would be preferredby the entrepreneurto no contractat all). Becausetherearefewerthann othersmallcreditors,therewould
couldfindfinancingelsewhereand
be a positiveprobabilitythat not all good entrepreneurs
thereforewould be assignedthis contract. 11
The proof of Proposition3 is somewhatinvolved, but the idea that underliesit is
very simple: If nHl>> nl*, then small creditors have the advantage over their big
counterpartsof not attractingpoor entrepreneurs.Thus they are more efficientand so, in
equilibriumwith free entry, drive the big creditorsout of the market.
We have been consideringthe case in which Ep+ np <2. If instead this inequality
goes the other way (but all other inequalitiesremainthe same,in particularnp < 2), then,
accordingto the criterionof social surplus,slow projectsare efficient(see footnote 14,
however,for why social surplusmay not be the rightcriterion).Nonetheless,Propositions
2 and 3 continueto hold. That is, only good projectsare financedunderdecentralization.
This follows because creditors ignore the entrepreneurs'private benefits in deciding
whetheror not to fund a project,and suggeststhat there may be an excessivetendency
in decentralizedcreditmarketsto focus on short-term(i.e. one-period)projects(because
banks are too illiquidto make efficientloans). For a less ambiguousillustrationof this
tendency(one that does not rely on this questionablemeasureof efficiency),see the next
section.
In the case n> I >Inp*, the marketoutcomereproducesthe featuresof the decentralized model of subsection2a. When n* > i, mattersare more complicatedbecauseboth
types of entrepreneurswill be financed regardlessof the size of creditors.A potential
problemof non-existenceof equilibriummay arise if a creditoris able to affect its mix
of entrepreneurs(the relativeprobabilitiesof good and poor entrepreneurschoosing its
contracts)sharplyby slightlychangingthe terms it offers. Such a problemis similarto
those arisingin insurancemodels 'ala Rothschild-Stiglitz(1976) and Wilson (1977). To
avoid all this, we introducethe followingmild assumption:
AssumiptionA. Slow entrepreneurshave an (arbitrarily)small probabilityof completingtheir projectsin one period.
This assumptionlimits the effectthat improvingthe termsofferedto good entrepreneurshas on a creditor'smix of entrepreneurs;any improvementwill be attractiveto poor
as well as to good entrepreneurs.AssumptionA enablesus to derivethe followingresult:
A, the-e is a uniqueequilibrium(where
Proposition 4. Let n *> 1, UnderAssunmption
uniqueness is qualified the same way as in Proposition 3) in which (i) only big creditors are
active in the maarket,and (ii) at least n + I of them offer contracts that break even on average
across good andpoo- projects and that extract the entire observable returnf-om poor projects.
Proof. We will show that the behaviour described constitutes an equilibrium.
Uniquenesscan be establishedas in the proof of Proposition3. Clearly,it is not optimal
DEWATRIPONT& MASKIN
CREDIT AND EFFICIENCY
553
to leave any observablereturnto poor entrepreneurs:a creditorwould only improveits
mix of entrepreneursby loweringthe returnofferedto poor ones. UnderAssumptionA,
however,a creditorcannotimproveits mix by offeringbettertermsto good entrepreneurs,
since such improvementwould attractall the poor entrepreneursas well. Therefore,if at
least n other big creditorsofferbreak-evencontractsas describedin the proposition,a big
creditorcan do no better than to follow suit.
As for smallcreditors,theycannotavoid attractingpoor as wellas good entrepreneurs
since Hp**>1. However, they are less efficientin monitoringpoor projectsthan are big
creditors.Thus if the lattercreditorsbreakeven, the formerlose money. 11
Thus, in this model, a decentralizedmarketleads to efficientcreditorliquidity.When
neitherlarge nor small creditorscan commit not to refinancepoor entrepreneurs,large
creditorsare more efficientbecausethey have the incentiveto providebettermonitoring.
Therefore,they drive small creditorsout of the market.
4. DECENTRALIZATIONAND SHORT-TERMISM
We now introduce a third project: a two-period but very profitable undertaking
denoted by the subscript v. This project requiresone unit of capital per period and
generatesa returnRv> 2 aftertwo periods.For simplicity,we supposethat R"is deterministic and does not requiremonitoring.A good entrepreneurcan choose betweena good
or veryprofitableproject(poor entrepreneursare stuckwith poor projects),but her choice
is unobservable.24Moreover,poor and very profitableprojectsare indistinguishableto
creditors at the end of period 1.25 A good entrepreneur'sprivate benefit from a very
profitableprojectis E, (if the projectis terminatedafter one period) or EV(if the project
is completed).We adopt the naturalassumptionthat E, > Ep.
The timingis much the same as that of Section 3. But we now must insertthe choice
betweengood and very profitablecontracts,which we assumeis made at the same time
as creditorsoffer contracts. A creditor'smonitoringintensitydepends on its beliefs in
period I about projectquality.Specifically,a largecreditorwill expendefforta*(a') such
that (1 - a')Rp-/= '(a*(a')) if it believes that a' is the probability,the project is very
profitableand I - a' is the probabilityit is poor. Note that if a' = 0, the model reducesto
that of Section 3. Hence a*(0) =a*, and the financialreturn (gross of capital) is npH.
Similarly,for a smallcreditor,we definea**(a'), and obtaina**(O)=a**, whichgenerates
gross financialreturnn*. Refinancingdecisionsclearly also depend on a'. Pessimistic
beliefs (i.e. low values of a') lead to short-termismn-i.e.the choice of good over very
profitableprojects-because good entrepreneursforecastthat long-termprojectswill not
be refinanced:
Proposition 5. If [Ip* < I there exists an equilibriumin which only small creditors are
active and only good projects are chosen.26
24. We thank Ian Jewitt for suggestingthat we replaceour earlieradverseselectiontreatmentof very
profitablelong-runprojectswith the currentmoralhazardformulation.
25. To simplifyanalysis,however,we supposethattheseprojectsa-e distinguishable
at the end of period2.
26. As the model stands,this resultdependsto some extent on the timing.If good entrepreneurs
choose
projectsbefore creditorsmove, nothingis changedsincethe entrepreneurs'
decisionsare unobservableanyway.
are
But if the creditorsmove first,then a groupof investorsmightforma bankso big that good entrepreneurs
encouragedto choose very profitableprojects.Still, the bank may have to be very big indeed-big enoughto
a good entrepreneur
accommodatea largefractionof all entreprenetirs-otherwise,
may face too high a risk of
not being assignedto one of this bank's contractsif she chooses the very good project.Thus, if there are
reasonablelimitson creditorsize/liquidity,our resultsshouldnot be very sensitiveto timingafterall.
554
REVIEWOF ECONOMICSTUDIES
Proof. Supposethat n + I or more smallcreditorsare active (and no other creditors
are) and offer the contractthat breakseven on good projects.Suppose,furthermore,that
creditorsbelieve that, if a projecthas to be refinanced,then with high probabilityit is
poor. Under these circumstances,all good entrepreneurswill choose good projects,since
Hlp* < 1 and the creditors'pessimisticbeliefs togetherimply that two-periodprojectswill
not be refinanced.Hence the creditors'beliefsare justified.Now, a small creditorclearly
cannot do betterthan breakeven. Supposethen that a big creditorenters.It will attract
all the poor entrepreneursand only its share of good projects.But since the formerare
unprofitable(Hp*< 2), the creditorwill lose money on average. 11
The equilibriumof Proposition5 can be highly inefficient.As in Section2, let a be
the fraction of entrepreneurswho are good. Notice that the Proposition5 equilibrium
exists no matterhow close a is to 1. Howeverif Rvis big, then, for a near 1, it is clearly
betterfrom a social standpointto put up with poor projectsfor the sake of the very good
ones. Indeed, for big enough Rv, there exists another and more efficientequilibrium,
provided that a is sufficientlynear 1. If E, -=Eg, the precise condition we requirefor
existenceof this other equilibriumis
aRv + (1 - a)a*(a)RRp- VI(a*(a)) -2> a(Rg-1).
(*)
Condition (*) impliesthat if all good entrepreneurschoose very profitableprojects,big
creditorscan offer them better termsthan on good projects,while still breakingeven. In
such a case, creditors'optimisticexpectationsare self-fulfilling.
Proposition 6. Suppose that Hp**< 1, E, = Eg, and (*) is satisfied. Then there exists
an equilibriumin which only big creditorsform and all good entrepreneursselect very profitable projects.
Proof. Supposethat thereare at leastn + 1 big creditorsand each offersthe contract
whichgives the entrepreneurnothing(excepther privatereturn)if the projectturnsout
to be poor, T, if the projectis very profitable,and Tgif the projectis good where
(1)
a(Rv-Tv) + (I-a)a*(a)Rp-yV(a*(a))-2 = O
2,
and
Rg- I-Tg=O.
(2)
From (1), c just breakseven if creditors'beliefs that all good entrepreneurschoose very
profitableprojectsare correct.Now, good entrepreneurswill choose these projectsprovided that
Ev+ Tv> Eg+ Tg.
(3)
From (1) and (2) and becauseE, = Eg, (3) can be rewrittenas
aR,,+(l
-
a)n*- 2 >a(Rg -1),
whichis just (*). Hence,good entrepreneurswill selectveryprofitableprojectsas claimed.
The argumentsthat no big creditorcan do betterby deviatingand that any creditorcan
profit from enteringare the same as in the proof of Proposition4. 11
Propositions5 and 6 implythat the sameeconomymay end up in two quite different
equilibria.In the equilibriumof Proposition5, creditorsare small and projectsare shortterm. In that of Proposition6, creditorsare big and projectsare long-term.Note that,
DEWATRIPONT & MASKIN
CREDIT AND EFFICIENCY
555
even ignoring entrepreneurs' private benefits, (*) implies that the latter equilibrium is more
efficient. Including the private benefits only aggravates the discrepancy (it would entail
adding (I - a)Ep to the left-hand side of (*)). Indeed, the equilibrium of Proposition 6
Pareto-dominates that of Proposition 5.
To conclude, let us note that Propositions 5 and 6 have some connection with those
of von Thadden (1995). Von Thadden argues that a commitment not to refinance projects
may be an optimal screening device for creditors facing an adverse selection problem, even
though it can induce short-termism on the part of good entrepreneurs. Although the set
of technological opportunities available to entrepreneurs and the initial asymmetry of
information in his paper are similar to those in our model, von Thadden takes a different
perspective, since he does not explicitly address ex post incentives to refinance or the role
of creditor liquidity. Rather, he concentrates on a one-creditor problem. In his model,
bank finance can reduce short-termism thanks to economies of scale (a' la Diamond
(1984)), which make direct inspection of project types profitable.
Acknowtledgements. We thank Patrick Bolton, Jeremy Edwards, Ian Jewitt, Charles Kahn, Janos Kornai,
Colin Mayer, John McMillan, John Moore, Yingyi Qian, Gerard Roland, David Scharfstein, Chenggang Xu,
and two referees for useful comments. This research was supported by the NSF, and by the Belgian Government
under PAI grant No. 26.
REFERENCES
AOKI, M. (1990), "Towardan EconomicModel of the JapaneseFirm",Journal of Economic Literature, 28,
1-27.
BALIGA,S. and POLAK,B. (1994), "CreditMarketsand Efficiency"(mimeo).
CORBETT,J. (1987), "InternationalPerspectiveson Financing:Evidencefrom Japan", Oxford Reviewl of
Economic Policy, 3, 30-55.
DIAMOND, D. (1984), "FinancialIntermediationand DelegatedMonitoring",Reviewsof EconomiiicStudies,
51, 393-414.
The Econiomnist
(1990), "Puntersor Proprietors?:A Surveyof Capitalism",May 5-11.
EDWARDS,J. and FISCHERK. (1994) Banks, Finianceand Investment in Germiany(Cambridge: Cambridge
UniversityPress).
FREIXAS,X., GUESNERIE,R. and TIROLE,J. (1985), "PlanningunderIncompleteInformationand the
RatchetEffect",Reviev of Economic Studies, 52, 173-192.
HOSHI,T., KASHYAP,A. and SCHARFSTEIN,D. (1988),"CorporateStructure,Liquidity,and Investment:
Evidencefrom JapaneseIndustrialGroups"(mimeo).
HOSHI, T., KASHYAP,A. and SCHARFSTEIN,D. (1989), "Bank Monitoringand Investment:Evidence
from the ChangingStructureof JapaneseCorporateBankingRelationships"(mimeo).
KOFMAN, F. and LAWAREE,J. (1989), "Collusionin HierarchicalAgency"(mimeo).
versusDemand-Constrained
Systems",Econiometrica,47, 801-819.
KORNAI,J. (1979), "Resource-Constrained
KORNAI,J. (1980) The Economiiicsof Slhortage(New York: North-Holland).
LAFFONT,J. J. andTIROLE,J. (1988),"TheDynamicsof IncentiveContracts",Econometrica,56, 1153-1175.
MAYER,C. and ALEXANDER,1. (1990), "Banksand SecuritiesMarkets:CorporateFinancingin Germany
and the UK" (mimeo).
QIAN, Y. (1994),"A Theoryof Shortagein SocialistEconomiesBasedon the Soft BudgetConstraint",American
Econiomiiic
Review, 84, 145-156.
QIAN, Y. and XU, C. (1991), "Innovationand FinancialConstraintsin Centralizedand DecentralizedEconomies" (mimeo,LondonSchool of Economics).
ROTHSCHILD,M. and STIGLITZ,J. (1976), "Equilibriumin CompetitiveInsuranceMarkets:An Essayon
the Economicsof ImperfectInformation",Quarterl.yJournal of Economics, 90, 629-649.
Problemin the Center-Enterprise
Relationship",Journal
SCHAFFER,M. (1989), "TheCredible-Commitment
of ComlparativeEconomics, 13, 359-382.
STIGLITZ,J. and WEISS,A. (1981), "CreditRationingin Marketswith ImperfectInformation",American
Econom1icReviewv,71, 393-410.
On the Role of Collusionin Organizations",Journal of
TIROLE,J. (1986), "Hierarchiesand Bureaucracies:
Law1,Economics anid Organizationis,2, 181-214.
VON THADDEN, E. L. (1995), "BankFinanceand Long-TermInvestment",Review,of Economic Studies, 62,
557-575.
WILSON, C. (1977), "A Model of InsuranceMarketswith IncompleteInformation",Journal of Economic
Theory, 16, 167-207.