banks as multi output oligopolies an empirical evaluation of the retail and corporate banking mar

21
Banks as Multioutput Oligopolies: An Empirical Evaluation of the Retail and Corporate Banking Markets Author(s): Sigbjørn Atle Berg and Moshe Kim Source: Journal of Money, Credit and Banking, Vol. 30, No. 2 (May, 1998), pp. 135-153 Published by: Blackwell Publishing Stable URL: http://www.jstor.org/stable/2601206 Accessed: 05/04/2010 07:59 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=black . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].  Blackwell Publishing is collaborating with JSTOR to digitize, preserve and extend access to  Journal of Money, Credit and Banking. http://www.jstor.org

Upload: siddu4040

Post on 07-Apr-2018

221 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 1/20

Banks as Multioutput Oligopolies: An Empirical Evaluation of the Retail and CorporateBanking MarketsAuthor(s): Sigbjørn Atle Berg and Moshe KimSource: Journal of Money, Credit and Banking, Vol. 30, No. 2 (May, 1998), pp. 135-153Published by: Blackwell PublishingStable URL: http://www.jstor.org/stable/2601206

Accessed: 05/04/2010 07:59

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at

http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless

you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

http://www.jstor.org/action/showPublisher?publisherCode=black .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed

page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of 

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

 Blackwell Publishing is collaborating with JSTOR to digitize, preserve and extend access to Journal of Money,

Credit and Banking.

http://www.jstor.org

Page 2: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 2/20

SIGBJ0RN ATLE BERG

MOSHE KIM

Banks as MultioutputOligopolies:

An EmpiricalEvaluationof the Retail

and CorporateBankingMarkets

The distinctionbetween retailand corporatebankingmarkets s of much importance nreal life banking organizations.The two marketsdiffer with respect to concentration,the importanceof infolmational asymmetries, and the extent of customer mobility.

Withina standard onjecturalvariationmodel estimatedon cost-efficientbanksas wellas on the full sample of banks, we empirically characterize he strategic behavior ineach of these markets,and also focus on cross-market nteractions o see whether nitial

moves in one marketaffect the equilibrium n the other market.We compareour find-ings to the predictions hatwould follow from merely consideringconcentration atios,such as the Heffindahl ndex.

ONEOFTHEMAJORNTERESTSxpressedby researcherswho

are engaged in empiricalbankingstudies is the assessmentof the competitiveviabili-

ty of the industry.For over a decade of research n the area,researchershave tried to

address his issue by applyingvariousmethods o differingdatasets for the purposeofestimating the degree of scale economies. Realizing the multiplicity of products n

banking,recentresearchhas also focussed on the propermeasurement f productmix

economies or what has been termed economies of scope. 1 This line of researchhas

been centeredon the propertiesof the cost function,and most studies have not incor-

porated he market nfluences on the behavior of banks, and the potential impact of

these forces on competitive viability and behaviorof the industry.Very recent litera-

ture has been testing for competition and marketpower in the banking sector. A1-

though marketpower may be inferred rom various concentration atios such as the

Herfindahl ndex, these measures may not accuratelydepict the true marketpower.The purposeof the presentpaper s to investigate these market nfluences, focussing

on the degree of marketpower and on the interactionsbetween differentmarketseg-

ments in an oligopolistic marketsetting.

Multioutputbanking firms sell their outputs n a numberof differentmarketseg-

ments, which are characterized y customersof differing nformational atheringand

The authors hankAllen Berger, David Humphrey,Diana Hancock, SherrilShaffer, and participants ta numberof seminars or most insightfulcomments on previousversions of this paper.

1. See, for example,the papersby Murray ndWhite ( 1983), Clark 1984), Gilligan, Smirlock,andMar-shall ( 1984), Kim ( 1986), Berger,Hanweck,and Humphrey 1987), Hancock( 1991), and many others.

SIGBJ0RN ATLE BERG is assistant irector f the CentralBankof Norway ndprofessor feconomics,Universityf Oslo.MOSHE KIM is professor f economics,Universityf IIaifa.

Journal f Money,Credit, ndBanking,VO1. 0 NO. (May 1998)Copyright 1998bY he Ohio State University Press

Page 3: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 3/20

136 : MONEY,CREDIT,AND BANKING

processingabilitiesandmobilityacross product haracteristics.Thesedifferencesareimportant easonswhybanksmaydevelopdifferentstrategicbehavioracross marketsegments,perhapscharacterized y some kindof retaliation Porter1980;Salop and

Scheffman1983) or forbearance due to fear of retaliation).2Recent studies of suchbehaviorin bankingmarketsinclude Spiller and Favaro(1984), Hancock (1986),GelfandandSpiller(1987), Shaffer 1993), ShafferandDiSalvo (1994), andBerg andKim (1994). Only the paperby Gelfandand Spillerincludes a cross-marketanalysisof oligopolisticcompetition.None of thepapers,however,takeintoaccountthepos-sibility thatX-inefficiency maymaketheiroptimizationmodelsirrelevant,as will bethe case forinefficientbanksthatarenoton theirfrontiercost function.

In this studywe focus on cost-efficientbanks n order o avoid thatpitfall. We fur-therattempt o

uncovertheoligopolisticbehaviorof multioutputbanksthat sell theiroutputs n two distinct marketscommon to all banking ndustries,namely,retail andcorporatebankingmarkets.Thatdistinction s new tothis line of study,3and its ratio-nale is based on two mainobservations.Firstandgenerally,the distinctionbetweenthese activities(outputs) s intrinsic o the operationof banks and is of much impor-tancein practice.Infact, banksthatare moreretailoriented endto rely moreheavilyon produceddeposits, and thus use more extended branchingnetworks, whereasbankswhich aremore oriented owardcorporate ustomers endto rely moreheavilyon the purchaseof funds- ather hanon produceddeposits.4Second, these activities

reflectdifferent ypesof customerswho differsubstantiallywithrespectto their abil-ity to gatherandprocessthe relevant nformation n financialmarketsandthereforetheirmobility across banksdiffers.5In addition,customerscannotsubstituteretailloans forcorporate oansor vice versa,whereasbanks can anddo divertfundsfromonemarket o theother.6As a resultof thesedifferencesamongcustomers, hedegreeof competition will differ, and the elasticity of demandas perceived by individualbankswill be different n thesemarkets.

Sincecustomerscannotsubstitute,demand nthe retailandcorporatebankingmar-kets will be independent.Even so, any shock to eitheroligopolistic marketsegment

hasboth a directandan indirecteffect on theprofitsof a bank,whencosts of produc-tionacrossmarkets or a given bankareinterrelated.7After the shockthebanks' ini-tial allocationof outputsbetween the two markets s no longer profitmaximizing,sincethemarginalgainfrom selling a unitin one of the marketshaschangedand thebank will reoptimize. As has been shown by Bulow, Geanakoplos,and Klemperer

2. FeinbergandSherman 1982).

3. Hunter,Timme, andYang(1990) are theonly ones we areawareof to have used a similardistinctionbetweendifferent oan market egments,albeit for a differentpurpose,by specifying retailandwholesaleoutputs n their studyof cost subadditivity n largeU.S. banks.

4. Also, largerbanks end to employa higherproportion f purchasedunds.

5. Thisobservation s also madeby Klemperer1987) who examinesmarketsnwhichconsumers'costsof switchingbetweenbrandsof aproductmakeit easier for a firm o sell to customerswho purchasedromit in a previousperiod(market).

6. Orengage incross-subsidizationpractices.

7. We treat he notion of interrelatedosts asjointness inproduction.Thisphenomenonmay ormay notresultin scope economies.In ourpresentcontextjointness is sufficientfor the strategic nteractionamongmultioutput ligopolies.

Page 4: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 4/20

SIGBJ0RN ATLE BERG AND MOSHEKIM : 137

(1985) this reoptimizationor marginalchange in strategycan have first-order ffects

on the bank's profits n oligopolistic markets.8

The cross-market ffect modeled by Bulow, Geanakoplos,and Klempererdepend-

ed primarilyon interrelated osts. But they also showed thatcross-market ffects ob-served n practicemay dependon the natureof oligopolistic competitionoccurring n

the two segments, and on short-run nput-output onstraints, uch as the banks' abili-

ty to fund desired increases in total overall lending to both markets.9Cost comple-

mentarity,which should perhaps be expected between retail and corporatemarket

outputs,would ceteris paribus nduce parallelmovements of bank outputs n the two

marketsegments, as long as short-run onstraintsare not too severe. However, in a

near competitivemarketsegment we would not expect significanteffects originating

from anothermarket,whereas in a highly noncompetitivesegment such effects may

be significant.

The classification of banking output into retail and corporatemarket activities

seems a natural tartingpoint for providingus with a more coherentpictureas to the

oligopolistic natureof interactionamong banks. By taking account of the process by

which output s supplied, he varyingprice elasticitiesof demand n each of these mar-

kets, and the possible cross-market trategiceffects, we hope to gain further nsights

into the competitivebehaviorof banking irms.

The paper s organized n the following manner.Section 1 presentsthe model. In

section 2 we discuss the dataused. Section 3 presentsestimationand results,and sec-tion 4 concludesthe paper.

1. THE MODEL

Similarto Gelfandand Spiller (1987), we consider a set of multioutputbanks, i =

1, ..., 1, each supplying their output in two credit markets (retail and corporate)at

pricespk (k = 1,2). The ith bankchooses outputquantities ik' k = 1,2, to maximize

the profitszi,

zi = Plqil + P2qi2- Ci(qil,qi2) .t. (1) pk = pk(qk) k= 1 2

(2) {Rjk;7& k = 1, 2} (1)

wherepk(.) iS the inverse demand unction or productk,qk iS the total supplyof prod-

uct k, Rjks the expected reaction of bankj in marketk (to be explicitly specified

below), qik is the quantityof productkproducedby bank and Ci(.) is the multioutput

variablecost functionof the ith bank.In the case of imperfectlycompetitivemarkets,

each bankwill recognize thata change in one of its outputswill have an effect on both

8. The reason s thatsmall changes in bank 's equilibrium trategy n market1 will cause small changesin its competitorj' marginalprofitschedule and thus induce small changes in j' s market2 strategy.Thesesmall changes inj' s strategyhave first-order ffects on i' s profits.

9. The limitationon credit ines from money market undingmay be a functionof the economy's open-ness, the tightnessof regulatory egimes and the state of healthof the bankingsystem.

Page 5: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 5/20

138 : MONEY, CREDIT,AND BANKING

productprices throughall other banks'outputreactions.In general therewould also

have been aneffect through he cross-priceelasticities.However, in our specific case

there is no substitutability n demandbetween the two markets,since these markets

are definedby two separategroupsof customersand no arbitrage pportunities xist.The first-order onditionsfor each bank are

zk = pk + (qk _ )(1 + 4 qj ) + (qim-)( 4 Ij*i-)- Cik= 0 (2)

where m, k = 1, 2, but m 7&k; i = 1, ..., 1;andzik, Cikarepartialderivativesof zi, C

with respect to qik,respectively. The termsaqJ.klaQikre bank i's expectationsof reac-

tions from bankj in marketk to its own output decision in that market.The term

aqJmlaQikS the expectedreactionfrom bank in the other market,m 7&k,to the same

initiation.

The mainobjective of thispaper s to analyze the natureof theseconjecturesand to

determinewhether heychange through imebetween and within the marketsof retail

andcorporatebanking.Generally, t wouldbe very useful to utilize a dynamicmodel

of banks'interaction n order o capture he evolution of strategies.Unfortunately,a

data set containingonly threetime seriesobservations years) is far too shortto look

seriously at evolutionary trategies.Thus,the method used in our paperrelies on thecomparativestatics of profit-maximizing irms. This method is by now standard

(Bresnahan1989) and is especially useful due to its relativetractability s far as em-

piricalresearch s concerned.We must keep in mind though,thatstaticmodels assess

only outcomes and do not (pretend o) locate the source of collusion (if present).In-

deed, due to its empiricaltractability, t has been appliedearlier by researchers n

banking ike Berg and Kim (1994) and Shaffer and DiSalvo (1994). It is well known

that theoreticalmodels of games can yield any behavioraloutcome for any structure.

This is truefor both the static as well as dynamicspecifications.Moreover,when us-

ing dynamicmodeling, oneneeds to introduce nformational ssumptions: pen-loop,feedback, or closed-loop informationalstructures.In an open-loop informational

structure, layerscannotobserve the state aftert = O.Thus,an open-loopcalculation,

since it involves choice only at time zero, is essentiallystatic. With a feedback nfor-

mationalstructure layersknow only the current tate, andhence agents have (again)

staticormemoryless nformation.So what one is left with is the closed-loopinforma-

tional structure ndequilibria.Unfortunately, losed-loop equilibriamay notbe sub-

gameperfect, and hence one cannotrefinethe equilibrium oncept to ruleout certain

implausibleNash equilibria. n this case, dueto the multiplicityof equilibria, hereare

in generalno testablerestrictions see Slade 1995). We havediscussedthis somewhat

lengthily in orderto highlight the problematicnatureof the empiricalapplicationof

truedynamicconduct models. However, we shouldemphasize that, in what follows,

we allow conjectures o be time variantwhich is consistentwith the outcome of dy-

namicmodels (Bresnahan1989). Thoughnot a truedynamicmodel in the theoretical

Page 6: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 6/20

SIGBJ0RN ATLEBERGAND MOSHEKIM : 139

sense,onecan (as we do) still test for the(non)constancyof conduct hrough he sam-

ple period.10

To obtaina parsimonious epresentation f conjecturalvariations,we shallassume

that eithertheabsolute aggregateresponse orthe relativeaggregateresponseexpect-ed fromotherbanksare constantacrossbanks n each year:

8(j wi qJ ) I aqi eJmk (3)

or

Alog(>i qJ ) / aloGqi eJmk

Uponsubstitution f equations 3) or (4) intothe first-order onditionsspecifiedinequation(2) andrearrangement,we arriveat the first-order onditionsfor bank i in

marketk,

pk[l - -k k (1 + (0)kk)] p m m 8jmk Ci

or

1 qk qk 1 qn qm

pk{l k[ k + (1 -k)0kk]} -p - k (1 - -)0mk= Ck (6)

where k,m = 1, 2 representmarket ndexes, andEk = _(8qk/8pk)(pk/qk) iS the price

demandelasticityin marketk. Equation 5) or (6) aboveeach representsa systemof

two equations one for eachmarket) o be estimated ointly withthe cost function.No-

tice that his proceduren factamounts oestimatinga profit unction,sinceit involves

boththe revenueandthe cost sideof bankproduction.

2. DATA

Theabove model is appliedto a panelof data fromthe Norwegianbankingsector

for the threeyears 1990-1992, andit contains453 observationsn total. 2 This panel

covers the climax of an extensive bankingcrisis with huge loan losses, andalso the

startof a recoveryperiod.Most bankswere cuttingcosts during hese threeyears,and

systematically econsideringheirmarket trategies.Due to the ongoingcrisisand se-

10. Thedynamicmodelswhichappear n theliterature iffer n the expectedtime-seriesbehaviorof theconductparameters, ince the exact equationsdetermining he passagefrom one regimeto the other varybetween theories.See Porter 1985) andRotembergandSaloner(1986).

11. Equations(5) and (6) containthe conventional optimalitycondition thatfirmmarginalrevenueisequalto marginal ost. However,since strategic nteractions re allowedfor in themodel, through he con-ductparameters , marketmarginalrevenuewill deviatefrom marginal ost to a degree determinedby theconjecturalvariation conduct)parameters.

Page 7: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 7/20

140 : MONEY,CREDIT,AND BANKING

rious bankfailures, the system as a whole was very reluctant o extendnew credit; n

fact,throughout ur sampledyears,thetotal (sample) stockof bankcreditdeclinedbyroughly 5 percent n realterms.Themarket nteractionsmay thuswell have changed

even over this short time interval.Despite the crisis in the banking markets,themacroeconomywas not disturbed o any importantdegree. GDP growthrates variedfrom 1.8 percent n 1990 to 2.9 percent n 1992.

Bank outputsarespecified as loansextended to retail andcorporatecustomers,re-

spectively.l3Three variable actors of productionarespecified; abor,materials,andborrowedmoney. We furtherspecify two quasi-fixedfactors, namely,producedde-posits andmachine capital.Produceddeposits arecustomerdeposits obtainedfrom

othersthanfinancial nstitutions,and is exclusive of moneymarket unding.

Thebank-specificpricesof laborarecomputedas totallaborexpendituresperman-hour.Borrowedmoney is definedas the amountborrowed romother financial nsti-tutionsand from the national and internationalmoney markets.Its unit price is themoney marketrate,assumedequal to all banks. 4 The materials nputprice is mea-sured by a national price index and thus variesthroughtime, but is constantacrossbanks. Theexpenditureon materials ncludes thecosts (exclusive of depreciational-lowances) of operatingmachinesand transport quipment,external EDP, postage,

telephoneandtelex, andothernonlaborandnoncapitaloperating osts. Thecost func-

tion is taken to represent he costs of these threevariable nputs,namely,labor, bor-

rowed money, andmaterials,and to exclude othercosts of bankproduction.Produceddeposits aretreatedas a quasi-fixed actor andthereforeappearas quan-

tity (measuredasaverageof start-,andend-of-the-yearbalances) n thecost function.This treatment s based on the view (Flannery 1982) thatproduceddepositsinvolvetransaction pecific (set-up) costs to bothcustomersand banks,giving each a strongincentiveto continuetherelationship n order o amortize hesecosts. Thus, the level

of produceddeposits is changeable only at increasingcosts rendering ts quantity

quasi-fixed.

The numberof brancheswill be included n thecost function to representnetwork

density.Brancheswill be assumedfixed, for the samereasons as given forproduced

12. Therewere 157, 149, and 147 banks n the 1990, 1991, and 1992sets, respectively.This is thecom-plete set of Norwegianbanks,exclusive of seven oreight subsidiariesof foreignbanks n eachyear andonecommercialbankthatwas established n 1992.

13. We haveabstracted rom the issue of uncertainty nd riskconsiderations,as is done in similarprevi-ous papers.The sourcesof riskare importantdeterminants or its implicationsforconduct.Thus,authorstrying to specify variousmeasuresof risks (stemming from differentsources), like Shafferand DiSalvo(1994), documentambiguityof results.Consequently,clear orderingof banksaccordingto risk levels isvery problematic.Ourexisting distinctionof corporateversusretailloans alreadycapturesan importantcomponentof the riskdifferentialnherent n the data.Sincebusiness cycle effects cannotbe well forecast-ed, our oandistinctionwould be the mainpiece of information hatcouldbe knownwith some certaintybe-

forehand.We would like to stress the fact that loanprices did not varymuch for the differentrisksduringtheperiod of ouranalysis andthus, this is not a very important ource of a problemand hence its implica-tions for ourestimatesof conductwould notbe significant.Moreimportantly, hecost of riskmay be deter-mined by managementbehavior [Berger and DeYoung (1995) show that lower managementqualityGranger-causes roblem oans] and thus,one wouldnot want to "correct"or risk in the regression,as it isendogenous (not reallyexogenousas through hebusinesscycle). See also Hughesand Mester(1993).

14. Pricevariationsdue to riskdifferencesamongbankshave not been importantduring he yearscov-eredln oursample.

Page 8: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 8/20

SIGBJ0RN ATLE BERG AND MOSHEKIM : 141

deposits.ls The bulk of small and medium-sizedNorwegian banks have in fact kept

theirnumberof branchesconstanteven during he somewhat urbulent eriodconsid-

ered here.

The retail output s defined as loans granted o privatenoncommercial nstitutions,

personalenterprises,and ndividuals.The corporate utput s definedas loans granted

to other sectors, includinggovernment, inancial nstitutions, imited companies,and

foreigncompanies.Investments n bonds andothersecuritiesare not included.The re-

tail and corporateoutputsare measuredby the averageof start-and end-of-year oan

balances. Theircorrespondingunit prices are assumedconstantacross banks, since a

numberof nationwide banks compete directly on the most important ocal markets,

and since the loan ratescharged end to be uniformwithinthe entirenetworksof these

banks.Data on prices of retailand corporate ending are not directlyavailable.But market

marginal evenuesare arrivedat by the weighted least squarecross-sectionregression

of total loan revenues of individualbanks with respect to the loan quantitybalances.

The interceptsare statistically nsignificant,and the R-squaresare 0.998 or higher.

The coefficient estimates are thus marginal and average) revenues, and are used as

prices n the appropriate quations.The estimatedmarginal evenues(standard rrors

in parentheses) or retail and corporateoutputsrespectively are:0.1436 (0.0017) and

0.1194 (0.0015) for 1990; 0.1562 (0.0013) and 0.0834 (0.0010) for 1991; and 0.1384

(0.0008) and 0.0869 (0.0006) for 1992.16The estimationprocedurealso requiresdata on the marketdemandelasticities of

the two outputs.These are assumedconstantover time, and are derived from aggre-

gate time series data 1988-91, by regressing otal retailand corporate ending quanti-

ties on the average nterestrates chargedby banks according o a quarterly urvey,l7

and on the inflationrate. There are sixteen quarterly bservations,and a logarithmic

functional form is imposed. The (negative) demand elasticities are estimated as

0.9017 (0.2728) in the retail marketand 0.8619 (0.3256) in the corporatemarket.l8

Notice thatthese are marketdemandelasticities, and thatthe demandelasticities per-

ceived by individualbanksmay differmore between the two markets,due to differentdegrees of market mperfections.

15. We have checked that treatingdeposits and branchesas variable nputs,by deleting them from thecost function,does not change our results n any importantway.

16. We triedadjusting he lending volumes for nonperformingoans, but thatdid not make a significantdifferenceto the estimates.

17. Since quarterlydata are required, he retail and corporateoutputprices estimatedabove can not beused.

18. Our conclusions with respect to banks' strategicbehavior will be robust to significantdeviations

from these pointestimates.We tested this assertionby reestimatingour model with demandelasticities (£1,£2) in the interval 0.1, 3.0). We could not estimateboth elasticities as coefficients in the model, but we didmanage to estimate one of the demand elasticities, given the other. The estimated elasticities were thenslightly above unity. In none of these cases were our qualitativeconclusions in section 3 changed in anyway. We shouldstressthatdemandelasticities shouldnot have changedmuch (if at all) during he periodofour analysis, since the macroeconomywas quite stable during his period;cfr. section 2. This assertionwasconfirmedwhen we included a time-varyingparameter or the elasticity whose estimate was -0.04 withstandard rrorof 0. 11.

Page 9: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 9/20

142 : MONEY, CREDIT,AND BANKING

3. ESTIMATIONAND RESULTS

3.1 EmpiricalSpecificationsTo assess the marginalcosts of different outputs, two alternativeproceduresare

used in previous studies of oligopolistic markets. Spiller and Favaro (1984) and

Gelfand and Spiller (1987) in studies of banking industriesused the money marketrateas theirestimateof marginalcosts. RobertsandSamuelson(1988) in a nonbank-

ing studyderived he marginalcosts from anestimatedmultioutput ost function.We

adopt the latter more general approach,but unlike RobertsandSamuelson's study,

ourconjecturalvariationparameterswill be estimatedsimultaneouslywith the cost

functionparameters, husallowing for the conduct parameters o be consistent with

theshort-run onstraintson total lending imposed by thetechnology (that s, the vari-able cost function).Generatingmarginalcosts from an auxiliary cost function and

then using them as data n the estimationof the behavioralequations withconjectur-

al variationparameters), s is done by RobertsandSamuelson,disregards onstraints

posed by technology andmay introduce naccuracies esulting n variousbiases in the

econometric esting.

As discussed earlier we estimate a multioutputvariablecost function with pro-

duced deposits and machinecapitalspecifiedas quasi-fixed actors,and withbranch-

es as an indicatorof network density (confer Kim and Ben-Zion 1989). Thus, the

variablecost function s

C = g(w, q, BR, Z) (7)

where C is variablecost, w is a vectorof variable nputprices, q is thetwo-outputvec-

tor, BR is the numberof branches,l9and Z is a vectorrepresenting hequasi-fixed n-

puts. The empiricalcost function s taken o be a second-order ranslogapproximation

to a twice differentiablearbitrary ost function, where all right-hand-sidevariables

aremeasuredas deviationsfrom the samplemean:

logC = cxO Ei OLilOgQi + Ej >logwj + wBlogBR+ Ek aklOgZk

+2(i Ej Aijl°gqil°gqj + Ej Ek BjklogwjlogWk)

+ EimEjn ijlogqilogw;+ Eim vilogqilogBR (8)

+ Eim Ik iklogqilogZkEjnHjlogwjlogBRSk KklogBRlogZk

+ Ej Ek MjklogwjlOgZk + 2 (NBB(lOgBR) + Ek Eh Pkhl°gZkl°gZh)-

19. Produceddeposits and the numberof branchesare naturallyhighly correlated 0.81 between thelogs), but not to a degree that nterfereswith theidentificationof marginalcosts.

20. We haveintroduced imedummies nto the interceptandfirst-orderoefficients.Bank-specificdum-mies could not be introduceddue to the strongdominanceof the cross-section observations n the panel.However, estimatingeach yearseparatelyhas producedno appreciable hange in results, thuspointing tothe robustnatureof our estimationresults.

Page 10: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 10/20

SIGBJ0RN ATLE BERG AND MOSHEKIM : 143

This cost functionwill be assumedequal for all years, except for the first-order ara-

metersand the interceptwhere differentvalues fordifferentyears will be allowed for.

This limits the numberof parameterso be estimated,and thus allows the pooling of

observations romseveral years.20

Standard ractice s followed by estimating he cost function ointlywith cost share

equations:

Sj= 0J}jE-k BjklogWk + Ei rijl°gqi + HjlogBR + EkMjkl°g k ( )

whereSj sthe cost shareof inputj.

Restrictions nsuringhomogeneityof degree one in factorprices andsymmetryare

imposed according o

Aij Aji,tjoi Bjk Bkj,Vk,j; Phk= Pkh'Vksh;

Ejpflj= ;

EnkBjk= °, j = 1, . . . ,n; Ejnrij = °, i = 1, . . . ,m;

Ej j ; EnMjk=°s = 1,...,n;

The expressionsfor the marginalcost of retail andcorporate oans for each obser-

vation are

MCi =-dl =-(°ti + EjmAijlogqjEjnijlogwy

+ AilogBR + Ejn ijl gZi) (10)

These expressions depend on thebank's level of retailand corporate oans as well as

on its input prices, networkcharacteristics s expressedby the numberof branches,and quasi-fixed nputs. Thus, the marginalcosts of the two outputswill vary across

banksand overtime.

The marginal ost expressions( 10) are nserted ntothe first-order onditions(5) or

(6), which togetherwith the cost function (8) and two of the threeshareequations(9)

form the systemof equations o beestimated.This implicitprofit unction(confer sec-

tion 1 is estimatedby the methodof maximum ikelihood,using the Davidon-Fletch-

er-Powell algorithm.

The first-orderonditionswerederived n the two alternative orms(5) and (6), de-

pending on whether we choose to model absolute or relative expected responses.However, we always find, by Akaike's InformationCriterion,and by the Bartlett-

Nauda-Pillai racecriterion,2l hat the logarithmic orm (6) fits the datamuch better

21. Akaike's InformationCriterion AIC)is definedas AIC =-210g L(f ) + 2n, whereL(f ) is the like-lihood function andn is the numberof observations.The Bartlett-Nauda-Pillairacecriterion confer An-derson 1984) is definedas R2 = 1-IQRI IQAI I, whereQRandQAare the covariancematricesof residualswithin the constantrelativeand constantabsoluteresponsemodels,respectively.

Page 11: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 11/20

144 : MONEY,CREDIT,AND BANKING

thanthe absoluteresponsealternative. n whatfollows we shall thereforeonly report

resultsfromthe relativeresponsespecification 6).

The estimationof the aboveconditionsfor profitmaximization mplicitlyassumes

thateach bank s operatingon its cost function(8). However,from the recentresearch

literatureon bank efficiency we know thatX-inefficiency is an importanteature of

bankingmarkets BergerandHumphrey1991). That makesa case for restricting he

analysis to those banks where X-inefficiencyis not a seriousproblem.Only then can

we be confidentthatbanksare on the estimatedcost function,so that our modelcap-

turesreal behavior.22

3.2 The CostFunction

The model was thereforeestimatedboth on the entire sampleof 453 observations,and on a subsampleof 120 cost-efficientobservations.Thissubsamplewasconstruct-

ed by selectingthe four banks with the lowest ratio of variablecosts to total loans in

each of ten size classes in eachof the threeyears.This corresponds losely to the most

efficient quartileof banks and thus providesa "thick frontiercost function," as in

BergerandHumphrey 1991).23We shall be reporting esultsfrom both samples,and

we shall see that they were very similar. But since X-inefficiency might potentially

make the marginalcost conceptindeterminaten the full sample,we have chosen to

focus mainlyon the resultsfrom the thick frontierexercise.

The estimatesof the year-specific irst-orderermsof thevariablecost function(8),

along with the conjecturalvariationparameters rom equations(6), are presented n

Table 1. The large numberof second-ordercost functionparametersare assumed

equal across years and will not be presented.We have checked for concavitywith re-

spect to variable nputprices,which shouldbe met by any well-behavedvariablecost

function.Concavitywas foundat nearlyall observationpoints.24

The measureof short-run ay scale economies is [I-S(dlog Cllog Zk)][z(dlog

Cldlog qi) + dlog Cldlog BR]-1. Note thatwe included the term (dlog CldlogBR) to

insure thatwe do indeed measurereturns o scale (the variation n unit costs with re-spect to proportional hange in both networksize, as measuredby the number of

branches,andthe provisionof bankingservices)rather hanreturns o networkdensi-

ty (the variation n unit costs caused by changingbankingservice levels withina net-

work of a given size) (confer Kim and Ben-Zion 1989). Most cost-efficientbanks

were foundto experienceslightly increasingreturns o scale. At the sample mean the

estimatedcost scale elasticitieswere 0.95, 0.95, and 0.88 for the three years consid-

22. In the cost-efficient subsamplebanks are assumedto conjectureresponses from all otherbanks, n-cluding those not included n the subsample.The behavioralmodel is thus exactly the same as for the full

sampleof banks.23. As is shownin a recentsurvey (Bergerand Humphrey1997) of five different rontier echniquesap-

plied to banking nstitutions n a numberof differentcountries(based upon 119 studies), the averageeffi-ciency values obtained from the different techniquesare quite similar. As well, the rank correlationsoffirm-levelefficiencyvalues, while not perfectlypositivelycorrelatedacross the five different rontier ech-niques, were oftenabove 0.50 andsometimes as high as 0.75. Thus, we canexpect similarresultswith oth-er frontier fficiencytechniques.

24. In the cost efficient subsamplethere were two violations among the 120 observations.In the fullsamplethere were twenty-oneviolationsamong the 453 observations.

Page 12: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 12/20

TABLE 1

PARAMETERSTIMATESROMHESIMULTANEOUSSTIMATIONFTHECOST UNCTIONANDCONJECTURALARIATIONQUATIONS

Thick frontier ost-efficient sample Full sample

Parameter 1990 1991 1992 1990 1991 1992

aO (intercept) 12.283 12.295 12.170 12.130 12.230 12.243(0.055) (0.061) (0.057) (0 030) (0 033) (0.034)

al (retail) 0.388 0.436 0.222 0.415 0.417 0409(0.039) (0.040) (0.036) (0.019) (0.019) (0.018)

a2 (corporate) 0.502 0.559 0.457 0.380 0.390 0.431(0.040) (0.044) (0.038) (0.020) (0.020) (0.023)

,81 labor) 0.286 0.328 0.364 0.262 0.265 0.290(0.021) (0.021) (0.026) (0.009) (0 009) (0.010)

,82(materials) 0.224 0.257 0.280 0.186 0.197 0.234(0.019) (0.016) (0.021) (0.010) (0.009) (0.011)

YB (branches) 0.112 0.120 0.157 0.191 0.193 0.185(0.033) (0.036) (0.039) (0.018) (0.018) (0.019)

y1 (deposits) 0.012 -0.061 0.240 -0.049 -0.072 -0.081(0.077) (0.086) (0.078) (0.033) (0.033) (0.034)

'2 (machines) 0.035 0.003 0.020 0.096 0.106 0.100(0.046) (0.045) (0.048) (0.026) (0.028) (0.030)

01 (retail, 0.722 0.746 0.763 0.611 0.641 0.632retail) (0.013) (0.012) (0.011) (0.011) (0.009) (0.011)021 (corporate, -0. 140 -0. 300 -0.084 -0.008 -0.008 -0.005retail) (0.026) (0.048) (0.024) (0.002) (0.002) (0.002)012 (retail, -0.0 19 -0.020 -0.000 -0.025 -0.054 -0.094corporate) (0.004) (0.003) (0.003) (0.014) (0.024) (0.023)

022 (Corporates 0.257 -0.110 -0 016 0.105 -0.264 -0.343corporate) (0.065) (0.107) (0.077) (0.049) (0.069) (0.073)Sample size 40 40 40 157 149 147

NOTES: or the sake of brevity he second-order ermsof thecost functionare not shown here. Full details can be made availableuponrequest.o indicates the expected response n market to an initialchange in market . Subscript I represents he retail and subscript2 the corporatebankingmarkets espectively.

Standard rrors n parentheses.

ered. Constantreturns o scale in all three years are statisticallyrejectedat the 1 per-

cent test level. The same was true for the results obtained on the full sample of

banks.25

As has been discussed above, strategicbehaviorof multioutput anksmayemanate

fromjointness in production, or example, cost complementarities.A proper est for

the existence of jointness in production s a test of aiaj + Aij< O.26This test wasper-

25. The cost scale elasticities for the largest bank in the sample are 0.97, 0.97, and 0.93, for the threeyearsrespectively, whereas or the smallestbank n the samplethey are0.92,0.93, and0.86. The scaleelas-ticities for the averagebank are 0.95, 0.95, and 0.88. This indicates hatthe averagecost curve is quadraticas expected.

26. This is a test of 82VC/8ql aq2 < 0, withsubscripts ndexingthe retailandcorporateoutputs.Notethat

this test is a test of a local property,performedat the point of approximation the sample mean). It is aprop-er test forjointness, but may not be a proper est for economies of scope because a rejectionof therestrict-ed form might mply82VC/8q aq2 > ° somewhere n the data.However, in the presentcontext, ointness issufficient or a change in equifibrium trategyregarding he level of one output o affect the optimized ev-el of the other output.For tests using approximationanalysis see Denny and Fuss (1977). White (1980)casts doubts on the general validity of the translogto provide reliable informationabout local properties(elasticities)of unknown unctions.However, as predictors, he approximations ave desirableproperties.White notes thatthe parameters stimatedconverge to those of the least squaresapproximation nd nottothe "true" arameters xcept when no misspecificationoccurs. Ourresultspoint to quite a good approxima-tion in termsof its ability to satisfy theregularity onditions which are derivativeproperties.Thereareonly

SIGBJ0RN ATLE BERG AND MOSHE KIM : 145

Page 13: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 13/20

146 : MONEY,CREDIT,AND BANKING

formedusing the log likelihood ratiostatistic.27The computedx2 values were 4.98

and 1235.0, respectively,for the cost-efficient sample andthe full sample.Both val-

ues exceed the criticalx2value for the 5 percentsignificance evel. Thus,nonjointness

inproduction f retailandcorporateoutputswas rejected.Since these two outputsare

not substitutable n demand, mplying thatcross-priceelasticities are vanishing,pro-

duction ointness may be one of the majordrivingforcesbehindthe strategic nterac-. . .

tlon ln tSllSsector.

The estimatedmarginalcosts of retailand corporate ending were on average for

cost-efficientbanks0.03 and0.11, respectively.In the full sampleboth averageswere

aboutone percentagepointhigher.Themarginal osts of corporate endingwere near-

ly always much higher thanof retaillending. The main reason for this is that cross-

section variations n retail endingcorresponds losely to variations n thevolume of(quasi-fixed)produceddeposits. We notice thatthe marginalcost estimateswere un-

correlatedwith bank size.

3.3 X-Inefficiency

Since we make a point of focussing on the subsampleof cost-efficientbanks, we

havecheckedfor theimportance f X-inefficiency n thepresentsample.Thebanks n

the full sample wereevaluatedrelativeto our thick frontier mplicit profit unction.28

We do thatby computing he ratiobetween actualprofits(loan revenuesminus vari-

able operating osts) andthebest practiceprofits hatwould have accrued f the banks

were operatingon the thickcost functionfrontierand obtainedaverageloan prices.

The averageefficiency score acrossthe full sampleof bankswas 0.89.

Forcomparison,we also estimateda thick frontierstandard no conduct parame-

ters) profit unctionfI(p, w,Z, BR) andcomputed he ratiobetween actualprofitsand

the best practiceprofitspredicted. n thatcase the averageefficiency scorewas only

0.74. The computedefficiency scores are also lower in the standardprofitfunction

framework or most individualobservations.The reason is, of course, that the stan-

dardmodeloverstates he benefitsfromincreasingoutput,by not taking nto accountthe reactionof otherbanksto such outputchange. This is a very importantpoint con-

cerning the use of eithernonparametric r frontier echniques.The abstraction rom

noncompetitivebehaviormay contribute o misleadingresultsconcerning he stateof

efficiency of firms n an industry.The introduction f properbehavioralequations s

of utmost mportance or a properanalysis.

3.4 BehavioralEquations

The two first-order onditions in the preferred ogarithmic orm (6) contain four

-conjecturalvariationparameters, escribing he expectedrelative responsesto initial

two violationsout of the 120 observations n the cost-efficient sample.This may indicate that the leastsquaresestimatesare not too farfrom the gradient.

27. The likelihood ratio test statistic is computedas-210g(L,/L ) which is distributedas a %2 variatewith degrees of freedomequal to the numberof independent estrlctions, n this case one. L andL are thelikelihoodfunctionvalues of therestrictedandunrestrictedmodels,respectively.

28. Notice thatthe profit unctionsconsideredarerestrictedby thequasi-fixed actors.

Page 14: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 14/20

SIGBJ0RN ATLEBERGAND MOSHE KIM : 147

changesin output.Positiveown-market eactions 011> ° (322 > °) represent xpect-ed retaliation,andimply that themarket s oligopolistic andcharacterized y signifi-

cant strategic nteraction.Negativeown-market eactions(01 < 0, 0 < 0) represent

expectedaccommodation orforbearance), nd thusdepict a marketapproachinghecompetitivestructure.29

Thecross-market eactions,given by (312and 321 shouldbe interpretedn conjunc-

tion withtheown-market onjectures.Equalsigns of own-andcross-marketeactionsmay be explained by the existence of jointness in production,as establishedabove.Forinstance,with a retaliatory ituation n the firstmarket, ointness may lead us toexpect a positivecross-market eaction.Oppositesignsmay also be given reasonableexplanations.Forinstance, a positive own-market eactionmay be consistentwith anegativecross-market eaction,since funds may be withdrawn rom one marketto

make possibleanexpansionof lendingin theothermarket.

Theestimatedvalues of theconjecturalparameters rereported n Table 1. Thees-

timates orown-retail-marketnteraction01 arestronglypositivein all threeyearsforboth samples,implying thatbanksexpect an increase n theirretail ending to be metwithretaliation romotherbanks n theretailmarket.Theestimatesfor owncorporatemarketnteraction322iS in bothsamplespositive for 1990, butnegativefor 1991and1992. This implies thatbanks expect a weak response, or an accommodatingre-sponse, in the corporatemarket o an initial increasein their corporate ending.The

behavioralpatternsare thus very different in the two markets.The retail market scharacterizedby significantstrategicinteractions,whereas the corporatebankingmarketappears o be muchcloser to a competitivetype of marketstructure,with ac-

commodatingexpectedresponses n 1991 and 1992. We have tested whether heoli-gopolisticnatureof banks'behavior s significantlydifferent romaCournotconduct,which is representedby all the conjecturalvariationparameters 0ij)equalling zero.To carry out this taskwe have performedsix sequential(likelihood-ratio) ests, onefor eachmarket n each year.Results shown in Table2 largelypoint to the rejectionofCournotconduct.

Notice thatthe estimateddifferencebetween the two marketsoriginatesfromthemuchlowermarginal evenuesobtained romcorporate endingthan rom retail end-ing (confer section 2 above).Theestimatedmarginalcosts arealso on averagehigherfor corporate ending thanfor retail lending(confersection 3.2). The mark-upovercosts is thusmuchmore modest,and in many cases negative, in thecorporatemarket.Thiscorrespondswell to theconventionalviews on thesetwo market egments.

However, it does not correspondwell to inferences thatwould have been made

aboutmarketpower f one followed theconventionalparadigmandlooked atconcen-

trationratios such as the Herfindahl ndex. Specifically, we have computed theHerfindahlndex for the retail andcorporatemarkets n 199>92, as summarizednTable 3. Theentries n thattabledepict a nonconcentrated etailmarketand a highly

concentrated orporatebankingmarket.30Onthatbasis one would haveexpectedthe

29. Note thatCournot onductwouldimply0 . = O, i,j. Hadthe absoluteresponseform been selectedby the data(equation3), thenokk =-1 wouldrepJresentperfectlycompetitivemarket.

30. Note that the numbers n Table 3 may somewhatexaggerate the trueconcentration atios in both

Page 15: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 15/20

TABLE 2

TESTING FOR COURNOT BEHAVIOR

Thick frontier ost-efficient sample Full sample

Actual Critical Accept Actual Critical AcceptHypothesis x2 X 0005 /Reject X2 X 0005 /Reject

01l(1990) = 02l(1990) = 0 324.42 10.6 Reject 795.06 10.6 Reject

01l(l99l) = 02l(1991) = 0 347.67 10.6 Reject 988.0 10.6 Reject

01l(1992) = 02l(1992) = 0 362.64 10.6 Reject 768.2 10.6 Reject

022(199°) = ol2(1990) = 0 36.66 10.6 Reject 12.18 10.6 Reject

022(1991) = ol2(1991) = 0 37.88 10.6 Reject 25.56 10.6 Reject

022(1992) = Ol2(1992) = 0 0.02 10.6 Accept 26.84 10.6 Reject

*The total significance evel of 0.05 was allocatedacross the six stages with approximately .005 Ievel for each stage, according o the Benfer-oni principle.

retailmarket o be the morecompetitive.We findthatthe opposite is the case and can

readily explain that result by the substantial nformationaldifferences between the

two markets.Thus, our analysis showsthatconcentration ndices may be unreliableas

a tool for measuringmarketpower.Other nfluences may in practicebe far more im-

portantdeterminants.3l

The oligopolistic structure dentified n the retailbankingmarketcorresponds o a

situationwhere, for reasonsof forbearance fear of retaliation),banksare not very ac-

tively competingfor customers.This is consistent with our notion of retailcustomers

as being less informedand less likely to switch from one bankto another hanare cor-porate customers. It is also consistentwith informationalasymmetryon the supply

side, with each bank possessing some exclusive informationabout its currentretail

customers,and thus an absoluteadvantage n the evaluationof theircreditworthiness

(confer,for example, Stiglitz and Weiss 1988; Sharpe 1990; Vale 1993).

Furthermore,he more competitive structureof the corporatemarket s consistent

with the observation hatcorporate ustomers n generalhave a stronger ncentive and

abilityto gather nformation nd seek price offers from severalbanks.Since most cor-

poratecustomersmust file publiclyavailable financialstatements, he informational

asymmetryon the supply side might also be less important n thatmarket.

The parameters321 and (312 representexpected cross-marketresponses, and are

both found to be negative for all threeyears in both samples (see Table 1). Banks do

not expect their competitors to retaliate in another market than where the initial

change in lending took place. Theestimatednegative response n the corporatebank-

ing market, ollowing an increase nretail ending (321) may mainly reflectthe rivals'

expected diversionof funds from thecorporate o the retailmarket, o generate he ex-

pected retaliation n the retail market.Given the presence of cost complementarity

(section 3.2), this aggregate cross-marketeffect may seem unreasonable at first

markets, ince nonbanksdo also supply retailand corporate oans. This bias may be more important n themarket or corporate oans. On the other hand, retail banking and to some extent corporatebankingtakesplace in local markets,andhence ourconcentrationmeasuresmay tendto underestimaterueconcentration.These combinedfacts are, however, unlikely to affect the qualitativeconclusion drawn.

31. Nathanand Neave (1989), Shaffer(1993), and Shafferand DiSalvo (1994) also reporthigh degreesof competition n structurally oncentrated ankingmarkets.

148 : MONEY, CREDIT,AND BANKING

Page 16: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 16/20

TABLE 3

THE HERFINDAHL INDEX OF MARKET CONCENTRATION

Year Retailmarket Corporatemarket

1990 886.4 2168.81991 870.6 2135.31992 842.2 2 160.2

NOTES: he Herfindahlndex (H) is the sum of squaredmarket hares n percentages.The US Department f Justiceguidelines considerH <1000 as an unconcentratedmarket,and H > 1800 as a highly concentrated ne (Salop, 1987).

SIGBJ0RN ATLE BERG AND MOSHEKIM : 149

glance, butmay be explainedby the factthatduring hesample years two of the three

large banks32dominating his aggregateexperiencedsevere problems,which limited

their access to the money marketwhereany additional unding must be obtained.33

Thus, mostbankshad reason to expectthattheirrivals'retaliatory apacity wouldbe

limited.

The estimatednegativeresponse in the retail market, ollowing anincrease n cor-

porate ending (312) iS numericallyverysmall. However, it combineswith an expect-

ed negativeor close to zeroown-market esponse n thecorporatemarket,andmay be

due to perceivedcost complementarity etween the two outputs.We note that 10l21

1(321 thatis, the expectedresponse (to a change in the other market) s more accom-

modating n the corporatemarket han t is in the retailmarket.Thisis consistent withthe corporatemarketbeing the more competitive.

The precedinganalysiscan be enhancedby payingattention o the change in thepa-

rameter stimatesthrough ime. Thus, we performed estsof conjecturalvariationpa-

rameterequality through he three years of our sample, using likelihoodratio tests.

These tests, along with therestrictedmodel resulting rom them, arepresented n Ta-

bles 4 and5. We could not in any of thetwo samplesrejectequalityof the own-retail-

marketresponses 01 through he threesampledyears. The cross-market ffects (312

were equalin 1990 and 1991, while the othercross-market ffects (321were found to

differ amongeach of the three sampleyears in the cost-efficientsample. The own-corporate-marketffect (322was foundto be equal for 1991 and 1992,but not equalto

the 1990value. We note thatthe strategicbehavior dentified n the corporatebanking

market became systematically more forbearing, indicating an increasing degree

of competitionin that market.Similartrends are not apparent n the retail banking

market.

The aboveresults are in accordancewith the prevalentviews on how the Norwe-

gian bankingmarketshavedeveloped duringandafter he generalcrisisthatpeaked n

1991. Thedegree of competitionhas been perceivedto be increasing, n particular n

the corporatemarketwhere a numberof new entrantshave tried o gain foothold since

the mid-1980s. Corporateending, beingbased primarilyon price, becausesmall dif-

ferences hereadd up to a large value dueto the larger-sized oans, is an easier market

32. The threelargest banksin the sample hadin 1992 56 percentof total retail andcorporate ending.

33. The bankcrisis particularlyhit the largest banks and impaired heir funding possibilities.This wasnot trueto the same extent of themajorityof smallbanks.

Page 17: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 17/20

TABLE 4

TESTING TIME EQUALITY OF CONJECTURAL VARIATION PARAMETERS

A. Thickfrontier cost-efficient sample

Hypothesis Computed%2 Critical %005 Accept Reject

01 (1990) = 011(1991) 3.60 7.88 Accept

01 (1990) = 011(1991) = 011(1992) 5.02 7.88 Accept

02l(1990) = 021(1991) 48.24 7.88 Reject

021 l99l ) = 021 1992) 37.22 7.88 Reject

022(199°) = 022(1991) 58.06 7.88 Reject

022(l99l) = 022(1992) 0 04 7.88 Accept

0l2(1990) = 012(1991) 0.22 7.88 Accept

012(199°) = ol2(l99l) = 012(1992) 30.02 7.88 Reject

B. Full sample

Hypothesis Computed %2 Critical %2)eX)5 Accept/Reject

01l(1990) = 01l(l99l) 6.00 7.88 Accept

01l(1990) = 01l(1991) = 011(1992) 0.20 7.88 Accept

02l(199°) = 021(1991) 1.32 7.88 Accept

021(199°)= 02l(l99l) =021(1992) 9.44 7.88 Reject

022(199°)= 022(1991) 60.8 7.88 Reject

022(l99l) = 022(1992) 0.68 7.88 Accept

0l2(1990) = 012(1991) 0.52 7.88 Accept

012(199°)=0l2(l99l) = 012(1992) 2.38 7.88 Accept

NOTE: hetotal significance evel of 0.05 was allocatedacross the eight stages withapproximately .005 level for each stage,according o theBenferoniprinciple.

to enter han s the retailbankingmarketwherenew entrantshave been far less impor-

tant.

4. CONCLUSIONS

We have characterizedoligopolistic behaviorin the retail and corporatebanking

markets,within a conjecturalvariationmodel estimated on cost-efficientbanks. Wehave found evidence of strongdifferences between the two markets,with expecta-

tions of retaliatorybehavior n the retail marketandaccommodatingbehavior n the

corporatemarket.These differencesare contrary o what an inspectionof the Herfin-

dahl indices of supply-side concentrationwould lead us to expect. Thus, similar to

what was documented in earlierresearch(Shaffer 1993 and Shaffer and DiSalvo

1994) concentrationmust notprecludesubstantially ompetitiveconduct.

Banks are found to have substantialmarketpower in theretail market,most likely

becauseretailcustomersoften do not have theresources o search or the bestoffer inthe market,andbecause of importantnformational symmetrieson the supplyside of

the market.Both factors contribute o impede themobility of customers n the retail

market.

The customers n the corporatemarkethave a much stronger ncentive and ability

to search or the best offer among anumberof banks.The informational dvantagesof

eachbanktoward ts current ustomersmay alsobe less important n this market,and

150 : MONEY, CREDIT,AND BANKING

Page 18: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 18/20

TABLE 5

EXPECTEDCONSTRAINED) ELATIVERESPONSES

Thick frontier ost-efficientsample Fullsample

Parameter 1990 1991 1992 1990 1991 1992

011 (retai1, retai1) 0.743 0.743 0.743 0.627 0.627 0.627(0.016) (0.016) (0.016) (0.010) (0.010) (0.010)

021 (COrPOrateSretai1) -0.138 -0.298 -0.086 -0.026 -0.026 -0.089(0.032) (0.057) (0 034) (0.014) (0.014) (0.024)

012 (retai1, C°rPOrate) -0.020 -0.020 -0.00 1 -0.007 -0.007 -0.007(0.003) (0.003) (0.004) (0.002) (0.002) (0.002)

022(COrPOrate,COrPOrate) 0.341 -0.122 -0.122 0.162 -0.304 -0.304

(0.074) (0.105) (0.105) (0.043) (0.070) (0.070)

SamP1e SiZe 40 40 40 157 149 147

NOTES: E.ndicates he expectedresponse n market to an initialchangein marketj. SubscriptI representshe retailand subscript2 the cor-poratebankingmarkets espectively.

Standardrrors n parentheses.

SIGBJ0RN ATLE BERGAND MOSHEKIM : 151

the banksareseen topossess less marketpower.Themobilityof customerswithinthe

corporatemarket s potentiallymuch more mportanthan n the retailmarket,and en-

tryis thus ess costly in thismarket egment.

Cross-market ffectsarefoundboth fromthe retail o the corporatemarketand vice

versa. In particular,while thedirectresponseto an initialincrease n retail endingis

expectedto be retaliatory,heindirectresponse n thecorporatemarkets expectedto

benonretaliatory.Andwhile a weak or accommodativedirectresponse s expectedto

an initialincrease n corporateending, the indirectresponse n the retailmarket s al-

ways expected to be nonretaliatory.The negative cross-marketeffects may be ex-

plainedby thenearcompetitivenatureof thecorporatemarket,andby the restrictions

on moneymarket undingwhichmayhave appliedto a few largeproblembanks.

The experiencesfrom the Europeanbankingmarketsof the past decade indicate

thatnew entrantswill mostoften tryto establishsomeinitialpresence n the corporatebankingmarket, and may frequentlychoose to ignore the retail marketaltogether.

Thisis whatwe wouldexpectfromthe estimatedown-market ffectsabove,wherewe

find that retaliation s expectedin the retail market,but not in the corporatemarket.

Theprioritygiven of incumbentbanksto defendingmarket hares n retailbanking s

alsoemphasizedby the strongernegativecross-market ffect in the corporatemarket.

This may mean thatincumbentbanksperceivetheir competitiveviability as being

mainlydependenton theirabilityto maintain heirretailcustomerrelationships.

LITERATURECITED

Anderson,T.W. An Introductiono MultivariateStatisticalAnalysis.New York:JohnWiley &Sons, 1984.

Berg, S. A., and M. Kim. "OligopolisticInterdependence nd the Structureof Production nBanking:An EmpiricalEvaluation.?' Journal of Money, Credit,and Banking26 (1994),309-22.

Page 19: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 19/20

152 : MONEY, CREDIT,AND BANKING

Berger, A. N., and R. DeYoung. Problem Loans and Cost Efficiency in CommercialBanks.Workingpaper,Boardof Governorsof the FederalReserve System, 1995.

Berger,A. N., G. A. Hanweck,andD. B. Humphrey."CompetitiveViability n Banking:Scale,

Scope, and ProductMix Economies."Journal of MonetaryEconomics 20 (1987), 501-20.Berger,A. N., and D. B. Humphrey."TheDominanceof Inefficienciesover Scale andProduct

Mix Economies in Banking."Journalof MonetaryEconomics 28 (1991), 11748.

. "Efficiency of FinancialInstitutions: nternational urvey and Directions for FutureResearch."EuropeanJournal of OperationalResearch98(1997), 175-212.

Bresnahan,T. "EmpiricalStudies of Industrieswith MarketPower."In Handbookof Industri-al Organization,vol. 2, edited by K. Schmalensee and R. D. Willig, pp. 1011-57. Amster-dam:North-Holland,1989.

Bulow, J. I., J. D. Geanakoplos,and P. D. Klemperer."MultimarketOligopoly: StrategicSub-stitutesand Complements." ournal of Political Economy93 (1985),488-511.

Clark, J. A. "Estimationof Economies of Scale in Banking Using a GeneralizedFunctionalForm."Journal of Money, Credit,and Banking 16 (1984),53-68.

Denny, M., and M. Fuss. "TheUse of ApproximationAnalysis to Test for Separability nd Ex-istence of ConsistentAggregates."AmericanEconomicReview 67 (1977),404-18.

Feinberg,R. M., and R. Sherman.An Experimental nvestigationof MutualForbearance.Dis-cussion paperno.126, Universityof Virginia,Charlottesville,1982.

Flannery,M. J. "RetailBank Deposits as Quasi-FixedFactorsof Production."AmericanEco-nomic Review72 (1982),527-36.

Gelfand,M. D., and P. T. Spiller. "EntryBarriersand MultiproductOligopolies."Internation-

al Journal of IndustrialOrganization (1987), 101 13.Gilligan, T. W., M. L. Smirlock,and W. Marshall."Scale and Scope Economies in the Multi-

ProductBankingFirm."Journalof MonetaryEconomics 13 (1984), 393-405.

Hancock,D. "A Model of the FinancialFirm with ImperfectAsset and Deposit Elasticities."Journalof Bankingand Finance 10 (1986),37-54.

. A Theoryof Production or the Financial Firm. Boston: KluwerAcademicPublishers,1991.

Hughes,J. P., andL. Mester."A QualityandRisk-AdjustedCost Function or Banks:Evidenceon the "Too-Big-To-Fail"Doctrine."Journal of ProductivityAnalysis 4 (1993), 293-315.

Hunter,W. C., G. S. Timme, andW. K. Yang. "AnExamination f Cost Subadditivity ndMul-

tiproduct Production n Large U.S. Banks." Journal of Money, Credit, and Banking 22(1990),504-25.

Kim, M. "BankingTechnology and the Existence of a ConsistentOutputAggregate."Journalof MonetaryEconomics 18 (1986),181-95.

Kim, M., and U. Ben-Zion. "On he Structure f Technology n a BranchBankingFirm."Jour-nal of Business and EconomicStatistics7 (1989), 489-96.

Klemperer,Paul D. "TheCompetitivenessof Marketswith Switching Costs."RandJournal ofEconomics 18 (1987),138-51.

Murray, . D., and R. W. White. "Economiesof Scale and Economiesof Scope in MultiproductFinancialInstitutions:A Study of BritishColumbiaCreditUnions."Journal of Finance 38(1983), 887-901.

Nathan,A., and E. H. Neave. "Competition nd Contestability n Canada'sFinancialSystem:EmpiricalResults."CanadianJournalof Economics 22 (1989),576-94.

Porter,M. E. Competitive trategy:TechniquesforAnalyzing ndustriesand Competitors.NewYork:Free Press, 1980.

Porter,R. H. "On he Incidenceand Durationof Price Wars."Journal of IndustrialEconomics33 (1985),415-26.

Page 20: Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

8/6/2019 Banks as Multi Output Oligopolies an Empirical Evaluation of the Retail and Corporate Banking Mar

http://slidepdf.com/reader/full/banks-as-multi-output-oligopolies-an-empirical-evaluation-of-the-retail-and 20/20

SIGBJ0RN ATLE BERGAND MOSHEKIM : 153

Roberts,M. J., andL. Samuelson."AnEmpiricalAnalysisof Dynamic,NonpriceCompetitionin an OligopolisticIndustry."RandJournalof Economics19 (1988), 200-20.

Rotemberg,J., and G. Saloner."A Supergame-TheoreticModel of Business Cycle andPrice

WarsduringBooms."AmericanEconomicReview76 (1986), 390-407.Salop,S. C. "Symposiumon Mergersand Antitrust." ournalof EconomicPerspectives1 (Fall

1987),3-12.

Salop, S. C., and D. T. Scheffman."RaisingRivals' Costs."AmericanEconomicReview73(1983), 267-71.

Shaffer,S. "A Testof Competition n CanadianBanking." ournalofMoney,Credit,andBank-ing 25 (1993), 49-61.

Shaffer,S., andJ. DiSalvo. "Conductn a BankingDuopoly."Journalof Bankingand Finance18(1994), 1063-82.

Sharpe,S. "Asymmetric nformation,Bank Lending,andImplicitContracts:A StylizedMod-el of CustomerRelationships." ournalof Finance45 (1990), 1069-87.

Slade, M. "EmpiricalGames: The Oligopoly Case." Canadian Journal of Economics 28(1995), 368401.

Spiller,P. T., andE. Favaro."TheEffectsof EntryRegulationon OligopolisticInteraction:TheUruguayanBankingSector."RandJournalof Economics15 (1984), 244-54.

Stiglitz,J., andA. Weiss. Banks as Social Accountantsand ScreeningDevicesfor theAlloca-tionof Credit.Workingpaperno. 2710, NationalBureauof EconomicResearch,1988.

Vale, B. "TheDual Role of DemandDepositsunderAsymmetricInformation." candinavianJournalof Economics95 (1993),77-95.

White,H. "UsingLeastSquaresto ApproximateUnknownRegressionFunctions."Interna-tionalEconomicReview21 (1980), 149-70.