The determinants of interest margins and their effect on bank diversification: Evidence from Asian banks

(1)

ContentslistsavailableatScienceDirect

Journal

of

Financial

Stability

journal homepage:www.elsevier.com/locate/jfstabil

The

determinants

of

interest

margins

and

their

effect

on

bank

diversiﬁcation:

Evidence

from

Asian

banks

Jane-Raung

Lin

a

_,

_Huimin

_Chung

a

_,

_Ming-Hsiang

_Hsieh

a,∗

_,

_Soushan

_Wu

b

a_Graduate_Institute_of_Finance,_National_Chiao_Tung_University,₁₀₀₁_Ta-Hsueh_Road,_Hsinchu_30050,_Taiwan

b_National_Taiwan_Normal_University_and_at_the_Securities_and_Futures_Institute,_Taiwan

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received3February2011

Receivedinrevisedform19July2011

Accepted1August2011

Available online 9 August 2011

JELclassiﬁcation: G21 L25 Keywords: Bankdiversiﬁcation Interestmargins

Endogenousswitchingmodel

a

b

s

t

r

a

c

t

Anendogenousswitchingregressionmodelisemployedforthisstudy,categorizingthebanksintoregimes ofhighandlowdegreesofdiversification,withourresultsindicatingthatnetinterestmarginscanbe lesssensitivetofluctuationsinbankriskfactorsforfunctionallydiversifiedbanksascomparedtomore specializedbanks.Inturn,thisimpliesthatbydiversifyingtheirincomesources,thesebankscanreduce theshockstonetinterestmarginsarisingfromidiosyncraticrisk.Ourresultsshowthatpriorfindingscan holdwhenthebanksarelocatedinaregimewithalowdegreeofdiversification.

1. Introduction

Theissueofnetinterestmarginsontheoverallbusiness activi-tiesofbankshasalreadybeenaddressedintheconsiderablebody ofbankingresearchliterature.Thenetinterestmarginsthatare setbybanks,essentiallytocoverthecostofintermediation,reﬂect boththevolumeandmixofassetsandliabilities;Angbazo(1997) speciﬁcallysuggeststhatthenetinterestmarginsofbanksshould actuallyrepresentasummarymeasureoftheirnetinterestrateof return.1_In_addition,_with_an_increase_in_exposure_to_risk,_adequate

netinterestmarginsshouldproducesufﬁcientearningstoenhance thecapitalbase.Sincenetinterestmarginsareclearlyofsigniﬁcant importancetothebankingindustry,theissuespertainingtohow theyareoptimallydeterminedandhowtheyadjusttochangesin thebankingenvironmentmeritcloserscrutiny.

Inourattempttomodelbanknetinterestmarginswithinthe currentstudy,weconsiderthe‘dealer’model—anapproachthat viewsbanksasrisk-aversedealersintheloananddepositmarkets

∗ Correspondingauthor.Tel.:+88635712121x57075;fax:+88635733260.

E-mailaddresses:jrlin@faculty.nctu.edu.tw(J.-R.Lin),

felixhsieh@mail.tbb.com.tw(M.-H.Hsieh).

1_Net_interest_margins_(commonly_referred_to_as_bank_interest_margins)_are

usu-allydeﬁnedasthedifferencebetweeninterestrevenueandinterestexpenses,

expressedasapercentageofaverageearningassets(Angbazo,1997).

undertheconditionwhereloanrequestsanddepositfundsareboth foundtooccurnon-synchronouslyatrandomtimearrivals.This approachwasinitiallydevelopedbyHoandSaunders(1981),and hasbeenextendedwithinanumberofsubsequentstudies,2_as_well

asappliedunderseveraldifferentsettings.3

Important regulatory steps have been taken to expand the functional scope of banking institutions4_, _both _in _the _US _(the

Gramm–Leach–BlileyAct,1999)andintheEuropeanUnion(the SecondBankingDirective,1989).Furthermore,asaresultofthe 1997Asianfinancialcrisis,thestructuresofthevariousfinancial systemsinmanyAsiancountrieschangedfromcontrolledsystems tomoreliberalizedforms(Gochoco-Bautistaetal.,2000).However, eventhoughdiversificationactivityhasbecomeanimportanttrend inbankingmanagement,theexistingliteratureonthe determina-tionofbanknetinterestmarginsdoesnotadequatelyaccountfor theeffectsofsuchdiversification.Therefore,theprimaryaimof thepresentstudyistoexplorehowdiversificationofbusinessby

2_Examples_include_McShane_and_Sharpe₍₁₉₈₅₎_,_Allen₍₁₉₈₈₎_,_Angbazo₍₁₉₉₇₎_,

SaundersandSchumacher(2000)andDrakos(2003).

3_See_Maudos_and_Fernandez_de_Guevara₍₂₀₀₄₎_,_Carbo_and_Rodriguez₍₂₀₀₇₎_and

MaudosandSolís(2009).

4_Financial_{stabilization}_and_deregulation_have_had_important_implications_on_the

incomestatementsofbanks:therehasbeenashiftfromnetinterestincometo

non-interestincomenotdependentontraditionalﬁnancialintermediation(Albertazzi

andGambacorta,2009).

(2)

banksaffectsthedeterminantsofnetinterestmarginsforasample ofcommercialbanksthatoperatewithinAsia.

The banking systems throughout Asia was previously char-acterized by the existence of large, dominant corporate and family-ownedcorporationswiththeirownfinancialsubsidiaries; thisservedtocreateinternalmarketswithinthesefirms,thereby enablingthemtocircumventmanylegalrestrictions,mostnotably onoffshorefinancing.Thegovernmentsofthevariouscountries also played important roles, both in terms of bank ownership andthroughsubsidizingbanks,particularlyforlendingdirected towardsnationalgrowthstrategies(WilliamsandNguyen,2005). Withinthecontextoffinancialderegulation,manyofthe bank-ingsystemsinAsiancountrieshaverecognizedtheneedformajor changesmostoftenassociatedwithincreasedcompetition, reor-ganizationandconcentration.Thevariousbankshaveresponded tothisnewenvironmentbyadoptingaproactivestrategywithin which the range of products offered to their clients has been extendedconsiderably.Apotentialbenefitforthefinancial con-glomeratesistheircross-sellingability,wherebymultiplefinancial productsaresoldtosimilarcustomers(Baeleetal.,2007).

In practice, bank revenues from lending activities tend to be cyclical—theyare largely dependentonboth theneedsand strengthsofloancustomersandthestageoftheeconomiccycle. Given that fee-based services and financial advice constitutea more stable revenue stream, banks may place greater empha-sisonthesetypesofrevenuelinesinanattempttosmoothout theirfinancialperformance;assuch,theyarealsolikelyto pur-suefunctionaldiversificationthroughactivitiessuchascommercial banking,investmentbanking,insuranceand otherfinancial ser-vices potentially capable of generatingrevenue in a variety of differentways,includinginterest,transactionfeesand commis-sions.Further,weconsiderthediversificationactivitiesofbanks that occureither as a result of shifts betweeninterest income andnon-interestincomeactivities,ordiversificationacrossthese twotypesofincomegeneratingactivities.Whilebanksareableto diversifyintonon-interestincomeservicesandproductsthatare directlylinkedtoanexistinginterestincomegeneratingactivity, theycanalsodiversifywithineitherinterestincomeactivitiesor non-interestincomeactivities(Merciecaetal.,2007).We there-fore expect to findthat suchfunctional diversification enables thebankstorealizecomparativeadvantagesbyincreasingtheir incomesources,andthatthesebanksmayreducetheshockstonet interestmarginsarisingfromidiosyncraticrisk.

Usingatwo-regimeendogenousswitchingregression model, this study extends Angbazo’s (1997) model by including bank diversiﬁcationtoexplorehowdiversiﬁcationofbusinessbybanks affectsthedeterminantsofnetinterestmargins.5_We_assume_that_a

bankmayoperateinoneoftworegimes,witheitheraloworahigh

5_Hu_and_{Schiantarelli}₍₁₉₉₈₎_indicate_that_ﬁrms_are_partitioned_into_groups_on

thebasisofasingleindicatorthatmayormaynotbeagoodproxyfortheimperfect

substitutabilityoffunctionaldiversiﬁcation.Theuseofasingleindicatoristhought

topreventresearchersfromcontrollingforthemultiplicityoffactorsthathave

someinﬂuenceonthefunctionaldiversiﬁcationofabank.Inaddition,theissue

ofwhetherabankbelongstothespecializedorfunctionallydiversiﬁedgroupis

determinedexogenously,andisﬁxedovertheentiresampleperiod,despitethe

factthatabankmaymovefromoneregimetoanother.Thisproblemcanbe

over-comewhenbank-speciﬁccontrolvariablesinteractwithpossibletime-varyingbank

characteristicsthatswitchthebankbetweenlowandhighdegreesofdiversiﬁcation.

However,ifasinglecharacteristicisusedintheseinteractions,itmaybeinadequate

tocapturetheseverityofbankdiversiﬁcationinformation;ifmorethanone

char-acteristicisused,thenumberofparameterstobeestimatedincreasesrapidlyand

mayleadtoimpreciseinferences.Thus,thisstudyattemptstodealwithbothstatic

anddynamicmisclassiﬁcationproblemsbyemployinganendogenousswitching

regressionmodel(withunknownsampleseparation)toinvestigatetheimpactsof

functionaldiversiﬁcationonthenetinterestmarginsofbanks.

degreeofdiversification.Theprobabilityofoperatingwithineach regimeisdeterminedbytheHuandSchiantarelli(1998) switch-ingmodel associatedwiththefollowing features. First, wecan directlytesttheeffectsofdifferentfactorsonthelikelihoodofa bankbeingfacedwithahighdegreeofdiversificationbyestimating theswitchingfunction.Second,astheswitchingfunctionisdefined asafunctionofthefinancialvariablesofthebank,aswellasother characteristicsthatproxyfortheseverityofbankdiversification information,thedeterminationofwhetherthebankislocatedina regimewithahighdegreeofdiversificationisundertaken endoge-nouslyineachperiod;themodelcantherebyeffectivelycapturethe dynamiceffectsofbank-specificvariablesonbankdiversification. Theresultsofthepresentstudydemonstratethatthesignson thecoefficientofriskfactorsaspredictedbyAngbazo(1997)can holdonlywhenthebanksarelocatedinaregimewithalowdegree ofdiversification,althoughtheseconclusionscannotbeconfirmed whenthe banksarelocatedin a regime witha highdegreeof diversification.Inaddition,priorstudiesprovidesomewhat incon-clusiveresultsregardingtheeffectsofbankrevenuediversification onrisk;forexample,Baeleetal.(2007)findthatmostbankscan reducetheiridiosyncraticriskbyengaginginrevenue diversifica-tion,whileLepetitetal.(2007)demonstratethathigherreliance onnon-interestgeneratingactivitiesisalsoassociatedwithhigher risk.6_In_this_study,_we_find_that_for_functionally_diversified_banks,

themarginscanbelesssensitivetofluctuationsinbankriskfactors thanthoseofspecializedbanks.Thisimpliesthatbydiversifying theirincomesourcesandplacingmoreemphasisontheserevenue linestosmooththeirfinancialperformance,thesebankscanreduce theshockstonetinterestmarginsarisingfromidiosyncraticrisk. SincethesepriorstudiesfocusedonEuropeanbanks,thepresent studyfillsanimportantgapwithintheliteratureontheeffectsof revenuediversificationonAsianbanksbyprovidingsomeindirect supportforBaeleetal.(2007).

Theremainderofthispaperisorganizedasfollows.Areview ofthepriorliteratureispresentedinSection2,followedin Sec-tion3byadescriptionofthedatasetandanexplanationofthe methodologyadoptedforthisstudy.Ourempiricalresultsare pre-sentedinSection4.Finally,theconclusionsdrawnfromthisstudy arepresentedinSection5.

2. Literaturereview

Thefirststudyundertakingananalysisofthedeterminantsof interestmarginswasprovidedbyHoandSaunders(1981),who approachedtheissuefromtheperspectiveofbankingfirms act-ingasmereintermediariesbetweenlendersandborrowers;they reportedthatinterestmarginsarecomprisedoftwobasic compo-nents:thedegreeofcompetitionwithintheassociatedmarketsand theinterestraterisktowhichthebankswereexposed.Allen(1988) extendedthesingleproductmodelofHoandSaunders(1981)to includeheterogeneousloansanddeposits,andpositedthatproduct diversificationmayresultinareductioninpureinterestspreads.

McShaneand Sharpe(1985) re-conceptualizedthesourceof

interestraterisk,regardingitastheuncertaintyexistingwithin moneymarkets,asopposed totheinterestratesoncreditsand

6 _Using_stock_market_data_to_analyze_the_long-term_performance_and_riskiness_of

banksfordifferentdegreesoffunctionaldiversiﬁcation,Baeleetal.(2007)showed

thatforsomebanks,diversiﬁcationcanactuallyreduceidiosyncraticrisk,thereby

makingthemsafer.Ontheotherhand,fromtheirinvestigationintotherelationship

betweenbankriskandproductdiversiﬁcationwithinthechangingstructureofthe

Europeanbankingindustry,Lepetitetal.(2007)concludedthatbanksexpanding

intonon-interestincomeactivitieswillinvariablyﬁndthemselvesexposedtohigher

riskandagreaterriskofinsolvencythanthosebanksthataremainlyengagedin

(3)

deposits.Basedonthedevelopmentofanempiricalmodelthat incorporatedcreditriskintotheexistingmodel,Angbazo(1997) notedthatthenetinterestmarginsofcommercialbanksreﬂect both thedefault and interest-raterisk premiums,andalsothat banksofdifferentsizesaresensitivetodifferenttypesofrisk.

HasanandSarkar’s(2002)separateexaminationsoftheeffects

of interest rate changes on existing loans (loans-in-place) and potentialloans(loans-in-process)ledtotheﬁndingthat‘low-slack’ banksareindeedexposedtosigniﬁcantlygreaterinterestraterisk than‘high-slack’banks.MaudosandFernandezdeGuevara(2004) subsequentlywentontoincludeaverageoperatingcostsasa deter-minantoftheintermediationmargin,usingtheLernerindexof marketpowerasadirectmeasureofthedegreeofcompetition.

CarboandRodriguez(2007)furtherextendedthemodelby

incor-poratingtheimportanceof non-traditionalactivities,proposing a multi-output model withtheoverall aimof determining the natureoftherelationshipbetweenbankmarginsand specializa-tion.Chen(2007)demonstratedtheeffectofbankingderegulation oncreditrisk,andrevealedthatcompetitionintensiﬁedfollowing thecompletionoftheSecondBankingDirective,whileloan qual-ityimprovedinmostmarkets.Theevidenceshowedthattheloan qualityimprovementisassociatedwithlowerinterestmargin.

MaudosandSolís(2009)recentlymodelednetinterestmargins

withthesimultaneous inclusion of operating costs, diversiﬁca-tionandspecializationasthedeterminantsofthemargins;their resultsindicatedthathighmarginsarelargelyexplainedby aver-ageoperatingcostsandmarketpower.Furthermore,basedontheir explorationofthesources ofrisk asimportantdeterminantsof thecorporatestructureof differentbankswhenexpandinginto newmarkets,Dell’AricciaandMarquez(2010)foundthatcorporate structurehasdirecteffectsonrisktakingandafﬁliatesize.

Otherstudieshavefoundthatdiversiﬁcationhasasigniﬁcantly positiveimpactonthevolatilityofearnings.DeYoungandRoland

(2001),for example, concluded that fee-based activities,which

representagrowingshareofbankingactivities,raisetheoverall levelofvolatilityinbankrevenue.Asimilarresultwasobtainedby

Stiroh(2004),whodemonstratedagrowingcorrelationbetween

netinterestincomeandnon-interestincome.Whenemployinga portfolioframeworktoassesstheimpactofincreasednon-interest income on equity market measures of return and risk within USfinancialholdingcompanies,Stiroh(2006)couldfindnolink betweennon-interestincomeexposureandaveragereturnsacross banks,althoughasignificantlypositivelinkbetweennon-interest incomeandthevolatilityofmarketreturnswasdiscernible.

Lepetit et al. (2008) subsequently investigated the ways in

which the expansionby banksinto fee-based services affected theirinterestmarginsandloanpricing,andfoundthat:(i)greater relianceonfee-basedactivitieswasassociatedwithlower lend-ingrates;and (ii)borrowerdefaultrisk wasunderpricedin the lendingrateschargedbythosebankswithgreaterproportionsof feeincome.Hence,theirﬁndingssuggestthatbanksmaytendto useloansasalossleader,whichinturnraisestheissueofhow cross-sellingstrategiesshouldbeaddressedbyregulatorsinorder tocontrolforbankrisk.

Berger et al. (2010) concluded that all dimensions (loans,

deposits,assets, and geography) of diversificationwere associ-atedwith highercosts and reduced profits. Theseresults were robustregardless ofalternative measuresof diversificationand performance. Moreover,theyobserved that bankswith foreign ownershipandthosewithconglomerateaffiliationwereassociated withfewerdiseconomiesof diversification,suggestingthat for-eignownershipandconglomerateaffiliationmayplayimportant mitigatingroles.

Despitethefactthatextensiveresearchonnetinterest mar-ginswithinUScommercialbanks,andtoalesserextentEuropean

financial institutions, has already been undertaken, relatively littleresearchhasbeencarriedouttodeterminebanknet inter-est margins within financial institutions in Asia. By analyzing a sample ofbanksin Asian countries, thepresent studymakes a two-fold contribution that complements the extant litera-ture.First, wehandle bothstatic anddynamic misclassification problems byemployingtheHuand Schiantarelli(1998)model, with unknown sample separation, to investigate the ways in which functional diversification affects bank net interest mar-gins. Second, we show that by diversifying intonew activities and placinggreateremphasisontheserevenuelinestosmooth theirfinancialperformance,bankscanreducetheiridiosyncratic risk.

3. Dataandmethodology

This study employed data on commercial banks in nine Asian countries(China, India, Indonesia,Japan, thePhilippines, Singapore,SouthKorea,TaiwanandThailand)coveringtheyears 1997–2005. Asoursample coversthe nine-year period follow-ingregionalderegulation,itshouldenableustodetectlong-term effectsofdiversiﬁcationonbothbankperformanceandrisk.The timeframeofourstudysamplealsoensuresthatmultiplebusiness cycleswererepresented.

Theannualbalancesheetandincomestatementsofthe com-mercialbanksobtainedfromBankscopeFitchIBCAwereusedto constructthevariablesforempiricalanalysisinthepresentstudy. In order to enhance cross-country comparability, we excluded bankswithmissing dataonbasicaccountingvariables, suchas assets, loans,deposits,equity, interestincome andnon-interest income.Wealsoexcludedalloutliersbyeliminatingextremebank observationsforeachconsideredvariable.Inaddition,mostofthe estimates werebased ona balanced sampleof those banksfor whichthedatawerecontinuouslyavailablethroughouttheentire sampleperiod.Themainreasonforusingabalancedsamplewas toallowus tocheckwhetherourresultswererobustin terms ofmodelingthefirm-specificeffectsasafunctionofpre-sample averagevaluesofthefirm-specificexplanatoryvariablesinthenet interestmarginandswitchingfunction—themeaningfulnessofthis methodincreasesifwetakeaveragesoveracommonperiod.Our finalpaneldatasetwascomprisedof262banks,providingatotal of2358bank-yearobservations.Table1reportsthemedian val-uesofthebankvariables bycountry,while Table2depictsthe Pearsoncorrelationcoefficients7_of_all_of_the_variables_used_in_this

study.

WeextendedthemodelofAngbazo(1997)utilizinga switch-ingmodelofnetinterestmarginsinanattempttodeterminethe importanceofbankdiversiﬁcation;themodelwasbasedontheHu

andSchiantarelli(1998)endogenousswitchingregressionmodel.

Dependingontheswitchingfunction,thenetinterestmargin equa-tioncanbeineitheroftworegimes,bothofwhichareunobserved bytheresearcher,andcharacterizedbydifferentvaluesofthe coef-ﬁcientsofthebank-speciﬁccontrolvariables.

Theestimation oftheswitchingfunctionallowsustoassess thestatisticalandeconomicsignificanceofthecharacteristicsof thedifferentbanksindeterminingtheprobabilityofbeinginone of tworegimes: a‘high degreeofdiversification’(hd) ora ‘low degreeofdiversification’(ld).Thebasicspecificationofthe switch-ingmodelofnetinterestmarginsisdefinedasfollows,withthe

7_Regressions_were_checked_for_{multicollinearity}_using_the_variance_inﬂation

fac-tor(VIF).ThemaximumVIFofanyofourexplanatoryvariableswas3.63,indicating

(4)

Table1

Summarydescriptivestatistics,bycountry.

Variables Countries Mean Median S.D.

China India Indonesia Japan S.Korea Philippines Singapore Taiwan Thailand

Nim 2.63 2.56 6.36 1.91 2.67 4.32 2.27 2.41 2.90 2.64 2.26 3.66 Mgmt 93.83 86.04 89.30 95.37 89.09 83.20 89.32 90.74 87.86 91.59 93.54 7.18 Lev 9.35 5.81 12.99 4.46 4.66 15.13 11.80 7.13 7.16 6.47 5.12 5.77 Opp 5.04 6.05 3.04 1.49 6.36 5.06 3.94 3.05 2.97 3.27 2.27 3.08 Imp 1.31 0.87 3.80 1.88 2.80 3.06 0.97 2.06 3.69 1.96 1.79 4.16 Liq 28.73 37.96 26.80 15.26 10.84 33.18 37.27 13.46 13.05 21.46 16.09 15.77 Int −61.10 −50.95 −49.78 −77.16 −70.26 −51.34 −48.84 −75.45 −70.65 −67.34 −74.36 17.21 Cdt 0.15 1.55 4.10 0.98 2.09 2.45 0.90 1.34 2.38 1.47 0.84 15.07 Ni 10.33 38.49 23.31 16.86 34.08 39.68 24.98 17.93 32.04 23.11 17.78 21.48 Lta 52.75 44.55 47.06 68.20 58.84 48.36 54.76 68.33 68.85 59.98 63.29 14.96 Rd 21.01 65.53 39.72 29.59 53.93 63.15 49.96 36.07 55.90 40.37 34.72 24.62 Ad 80.65 83.94 64.21 62.50 79.86 75.05 87.71 63.13 55.87 69.05 69.63 18.32

Themedianvaluesofthebank-speciﬁccontrolvariablesandthefunctionaldiversiﬁcationmeasures;thecontrolvariablesincludenetinterestmargin(Nim),management

efﬁciency(Mgmt),capitalbase(Lev),opportunitycostofreserves(Opp),implicitinterestpayments(Imp),liquidityrisk(Liq),interestraterisk(Int),andcreditrisk(Cdt);the

functionaldiversiﬁcationmeasuresincludetheratioofnon-interestincometototaloperatingincome(Ni),theloans-to-assetsratio(Lta),revenuediversity(Rd)andasset

diversity(Ad).Thediversitymeasuresaredeﬁnedinthisstudyasfollows:Diversity=1−|2x−1|,wherexiseithertheloans-to-assetsratioortheratioofnon-interestincome

tototaloperatingincome.Thediversityvariables,whichtakevaluesofbetween1and0,increasewiththedegreeofdiversiﬁcation(Baeleetal.,2007;LaevenandLevine,

2007).Allﬁguresrefertopercentages.

netinterestmarginequationforbanki,operatinginalowdegree ofdiversiﬁcationregime,attimet,being:

Nimi,t=Xitˇld+ε1,it (1)

if Zit+uit<0 (2)

Conversely,thenetinterestmarginequationforbanki, operat-inginahighdegreeofdiversiﬁcationregimeattimet,isdeﬁned as:

Nimi,t=Xitˇhd+ε2,it (3)

if Zit+uit≥0 (4)

whereXit=(1,Mgmtit,Levit,Oppit,Impit,Liqit,Intit,Cdtit,Intit×Cdtit,

CD,YD);Zit=(1,Niit,Ltait,Rdit,Adit,CD,YD);andNimi,tistheratioof

netinterestincome(beforeprovisionsforloanlosses)toaverage earningsassets.

Inthenetinterestmarginfunction,Xit=(1,Mgmtit,Levit,Oppit,

Impit,Liqit,Intit,Cdtit,Intit×Cdtit,CD,YD),whereMgmtitis

man-agementefﬁciency; Levit is thecapital base;Oppit referstothe

opportunitycostof reserves;Impit denotestheimplicitinterest

payments;Liqitreferstotheliquidityrisk;Intitistheinterestrate

risk;andCdtitreferstothecreditrisk.Theempiricalmodel

vari-ables,theirproxiesandthepredictedcoefﬁcientsigns,whichare allbasedonthesameassumptionsmadeinAngbazo(1997),are summarizedinTable3.TheZitvectorineachofthespeciﬁcationsof

theswitchingfunctionrepresentsasetofdiversiﬁcationvariables, comprisedoftheratioofnon-interestincometototaloperating income (Niit), theloans-to-assets ratio(Ltait), revenue diversity

(Rdit)andassetdiversity(Adit).Inaddition,countrydummies(CD),

andyeardummies(YD)areincludedtocaptureunobservedtime andcountryheterogeneity.

Wefollowed Baeleetal.(2007)toadopta pragmatic defini-tionofthedegreeoffunctionaldiversificationforourempirical analysis,relyingononeasset-basedmeasureandonebroad mea-sureofrelativediversification,bothofwhicharepubliclyavailable andwidelyusedbyanalystsandinvestorstoassessthelong-term potential and risk of banks.8 _Any _bank _with_a _lower

loans-to-8_The_asset-based_measure_is_the_{loans-to-assets}_ratio,_which_captures_the

propor-tionofloansrelativetototalassets.Therevenuemeasureistheratioofnon-interest

incometototaloperatingincome,wherethehighertheratio,themoreabank

focusesonnon-traditionalbankactivities(seeBaeleetal.,2007).

assetsratioora higherproportionofnon-interest revenue was regardedasbeingmoreorientedtowardsnon-traditional bank-ing activities.An alternative approach is to follow Baele et al.

(2007)andLaevenandLevine(2007)toconstructmeasuresofasset

andrevenuediversity;assetdiversityisbasedonthestock vari-ables,whilerevenuediversityisbasedontheflowvariables,with thesediversitymeasuresdefinedasfollows:Diversity=1−|2x–1|, where xis either the loans-to-assets ratioor theratio of non-interestincometototaloperatingincome.Thediversityvariables, whichtakevaluesbetween1and0,increasewiththedegreeof diversification.9

Existingtheoreticalargumentssuggestthatabankismorelikely tooperateinaregimewithahighdegreeofdiversificationwhen assetdiversity, revenue diversity and theratio of non-interest incometototaloperatingincomearehigh,andwhenthe loans-to-assetsratio is low. In ourformulation, this implies thatthe coefficientsontheratioofnon-interestincometototaloperating income,revenuediversityandassetdiversityshouldbepositive, whilethecoefficientontheloans-to-assetsratioshouldbe nega-tive.Furthermore,inthemoregeneralspecification,weallowedthe switchingfunctiontobedependentonthecost-to-incomeratio,as wellasthenaturallogarithmofbanksize.10_We_expected_to_find_a

greaterlikelihoodofabankoperatinginthehighdegreeof diver-siﬁcationregimewhenthecost-to-incomeratioislowandwhen thenaturallogarithmofbanksizeishigh.

FollowingHu and Schiantarelli(1998), we assumed that in thebanknetinterestmarginandswitchingfunctions,thevector of the error terms (ε1,it, ε2,it, uit) is jointly normally

indepen-9 _Asset_and_revenue_diversity_are_similar_in_spirit_to_the_{Hirschmann–Herﬁndahl}

indexofconcentrationofassetactivitiesorrevenuestreams.Theuseofthelatter

measureofactivityconcentrationisfoundinseveralstudies,suchasStirohand

Rumble(2006),Baeleetal.(2007)andLaevenandLevine(2007).

10 _The_{cost-to-income}_ratio_(ratio_of_operating_expenses_as_a_proportion_of_the

sumofnetinterestandnon-interestrevenue)measurestheoperationalefﬁciency

ofeachbank,withefﬁcientbanksbeingexpectedtohaveahigherfranchisevalue

(Baeleetal.,2007).Banksizeishighlycorrelatedwiththemeasuresoffunctional

diversiﬁcation;indeed,Merciecaetal.(2007)showedthatsmallbanksdonotgain

bydiversifyingoutsidetheirtraditionallinesofbusiness,therebysuggestingthat

itmaybedifﬁcultforsuchinstitutionstoachieveastrongfootholdinnon-interest

activities.Moreover,Lepetitetal.(2008)foundthatsmallbanks(withtotalassets

oflessthanD 1billion)signiﬁcantlyincreasetheirriskexposurewhenengagingin

(5)

Table 2 Correlation matrix. Nim Mgmt Lev Opp Imp Liq Int Cdt Ni Lta Rd Mgmt 0.3321 (<0.0001) Lev 0.1743 (<0.0001) − 0.1718 (<0.0001) Opp 0.0325 (0.1150) − 0.3592 (<0.0001) 0.1113 (<0.0001) Imp 0.7616 (<0.0001) 0.3914 (<0.0001) − 0.1467 (<0.0001) − 0.0722 (0.0004) Liq 0.1783 (<0.0001) − 0.1708 (<0.0001) 0.3456 (<0.0001) 0.3692 (<0.0001) − 0.0013 (0.9486) Int 0.2153 (<0.0001) − 0.2634 (<0.0001) 0.4797 (<0.0001) 0.5305 (<0.0001) − 0.0213 (0.3004) 0.8344 (<0.0001) Cdt 0.0467 (0.0233) − 0.0440 (0.0328) − 0.0023 (0.9097) 0.0090 (0.6615) 0.1125 (<0.0001) − 0.0254 (0.2181) 0.0243 (0.2387) Ni 0.4694 (<0.0001) − 0.5587 (<0.0001) 0.1098 (<0.0001) 0.3072 (<0.0001) − 0.4823 (<0.0001) 0.1714 (<0.0001) 0.3223 (<0.0001) 0.0412 (0.0452) Lta − 0.0267 (0.1958) 0.4917 (<0.0001) − 0.2830 (<0.0001) − 0.4308 (<0.0001) 0.1817 (<0.0001) − 0.6651 (<0.0001) − 0.6770 (<0.0001) − 0.0184 (0.3710) − 0.4255 (<0.0001) Rd 0.1151 (<0.0001) − 0.2156 (<0.0001) 0.0593 (0.0040) 0.3553 (<0.0001) 0.0471 (0.0221) 0.3023 (<0.0001) 0.3900 (<0.0001) 0.0361 (0.0799) 0.5340 (<0.0001) − 0.3630 (<0.0001) Ad 0.1409 (<0.0001) − 0.0556 (0.0070) − 0.0102 (0.6195) 0.2911 (<0.0001) 0.0858 (<0.0001) 0.2705 (<0.0001) 0.3012 (<0.0001) 0.0039 (0.8511) 0.1493 (<0.0001) − 0.3780 (<0.0001) 0.3980 (<0.0001) The Pearson correlation coefﬁcients between all of the variables used in this study, with p -values shown in parentheses.

dently distributed with mean zero and covariance matrix ˙,

where (ε1,it,ε2,it,uit)∼N(0,˙), ˙=

2 1 12 1u 21 22 2u u1 u2 u2

. Thenon-zerocovariancebetweenε1,it,ε2,itanduitallowsthe

shockstonetinterestmarginstobecorrelatedwiththeshocksto theﬁnancialcharacteristicsandothercharacteristicsofthebanks; thus,themodelusedinthepresentstudyisanendogenous switch-ingregressionmodel(Maddala,1986).

Althoughwecannotdirectlyobservetheregimewithinwhich thebankislocated,wecanspecifyandcalculatetheprobabilityof theoccurrenceofeachregimeasfollows:

Prob(Nimi,t=Nimldi,t) =Prob(Zit+uit<0)

=Prob(uit<−Zit)

=˚(−Zit)

Prob(Nimi,t=Nimhdi,t) =Prob(Zit+uit≥0)

=Prob(uit≥−Zit)

=1−˚(−Zit)

The likelihood density function for each observation fit is a

weightedconditionaldensityfunctionofε1,itandε2,it,withweights

ofProb(uit<–Zit)andProb(uit≥–Zit),

f(Nimi,t)=(ε1,it

Zit+uit<0 )˚(−Zit)

+(ε2,it

Zit+uit≥0)[1−˚(−Zit)] =(ε1,it,1)˚

−Zit

−(1u/₁2)ε1,it 1−(2 1u/21)

+(ε2,it,2)

1−˚

−Zit

−(1u/22)ε2,it 1−(2 2u/22)

(5)

where(·)isthenormaldensityfunctionand˚(·)isthe cumu-lativedistributionfunction;thus,(εJit/·)andJ=1,2denotethe

conditionaldensity,and(εJit,J)denotesthemarginaldensity.

Thesecondequalitysignusesthefactthatthejointdensityis equaltotheproductoftheconditionaldensitymultipliedbythe marginaldensityandthepropertiesofthebivariatenormal.Asin theprobitmodel,we canonlyestimate/u inEqs.(1)–(4),as

opposedtotheseparateestimationsof andu,withu being

normalizedasequalto1.Furthermore,12isinestimable,sinceit

doesnotappearinEq.(5).ForapanelofNﬁrms,withTi

observa-tionsforﬁrmi,thelog-likelihoodfunctionforalltheobservations isgivenbyF=

Ni=1

Ti

t=1log(fit).Theˇhd,ˇldand parameters

canbeestimatedbymaximizingthelog-likelihoodfunction. 4. Empiricalresults

4.1. Cross-sectionalmodelregressionresults

The empiricaldeterminants ofbankinterest marginscanbe foundinModel(9)ofTable4.Managementefﬁciency(Mgmt)and Capitalbase(Lev,aproxyforsolvencyrisk)arepositivelyrelatedto bankmargins.Thepositiverelationshipbetweenthecapitalbase and netinterest marginsis consistentwiththeincreased aver-agecostofcapital,sinceequitycapitalisincreasedbysubstituting equityfordebt.11_The_coefﬁcient_on_the_opportunity_cost_of_reserves

11_This_in_turn_leads_to_a_requirement_for_higher_net_interest_margins_(as_noted_by

(6)

Table3

Empiricalmodelvariablesanddescriptions.

Variables Description Predictedsign Rationale

PanelA:bank-speciﬁccontrolvariables

Mgmt Managementefﬁciency:

Earningassets/totalassets

+ Thisratioisincludedtoestimatethecomponentoftheinterestmargins

attributabletomanagementefﬁciency.Sincemanagementdecisionsaffectthe

compositionofassetsthatearn(high)interest,thesechangeswillbereﬂectedin

highernetinterestmargins

Lev Capitalbase:

Bookvalueequity/totalassets

+ Sinceequityisamorecostlyfundingsource,anincreaseinequitycapitalmay

increasetheaveragecostofcapital;bankswillthereforerequireahighernet

interestmargininordertocompensateforthehighercostofcapital

Opp Opportunitycostofreserves:

Non-interestbearingreserves/total

assets

+ Theopportunitycostofreservesistheaveragereturnonearningassetsforegone

byholdingdepositsincash,whichincreasesthecostoffundsbeyondtheobserved

rate.Bankswillraisetheirnetinterestmarginsinordertocompensateforthis

Imp Implicitinterestpayments:

(non-interestexpenditure–

non-interestrevenue)/earning

assets

+ Implicitinterestpaymentsreﬂectextrapaymentstodepositorsthroughservice

chargeremissionarisingfromcompetitioninthemarketfordeposits.Theseextra

interestpaymentsshouldbereﬂectedinhigherinterestmargins

Liq Liquidityrisk:

Liquidassets/totalliabilities

− Withanincreaseintheproportionoffundsinvestedincashorcashequivalents,

thereisadeclineinliquidityrisk,leadingtoalowerliquiditypremiuminthenet

interestmargin

Int Interestraterisk:

Netshorttermassets/bookvalue

equity

− Thematurity-mismatchhypothesissuggeststhatinterestrateriskexposurehasa

negativecorrelationwiththeaveragematurityofassets(FlanneryandJames,

1984);thus,thehigherthelevelofshort-termassets,thelowerthesensitivityto

near-terminterestratechanges,whichshouldresultinalowerinterestraterisk

premium

Cdt Creditrisk:

Loanlossprovisions/totalloans

+ Bankswithmoreriskyloanswillrequireahighernetinterestmarginto

compensateforthegreaterriskofdefault

PanelB:functionaldiversiﬁcationmeasures

Ni Ratioofnon-interestincometo

totaloperatingincome:

Non-interestincome/total

operatingincome

+ Thismeasureeffectivelycapturesallofthesourcesofthenon-interestincome

generatedbydiversiﬁedbanks.Thehighertheratio,thegreaterthefocuson

non-traditionalbankingactivities

Lta Loans-to-assetsratio:

Totalloans/totalassets

− Theloans-to-assetsratiocapturestheproportionofloansrelativetototalassets,

withaveryhighvalueindicatingthatthebankspecializesinloanmaking

Rd Revenuediversity:

Diversity=1−|2x−1|,wherexis

theratioofnon-interestincometo

totaloperatingincome

+ Therevenuediversityvariables,whichtakevaluesbetween0and1,increasewith

thedegreeofdiversiﬁcation

Ad Assetdiversity:

Diversity=1−|2x−1|,wherexis

theloans-to-assetsratio

+ Theassetdiversityvariables,whichtakevaluesbetween0and1,increasewiththe

degreeofdiversiﬁcation

Theempiricalmodelvariablesanddescriptionsforbankinterestmargins,alongwiththeirpredictedcoefﬁcientsignsandtheeconomicrationalefollowingAngbazo(1997),

Baeleetal.(2007)andLaevenandLevine(2007).

(Opp)isfoundtobepositiveandsignificantatthe5%level,which leadstoarequirementforhighernetinterestmarginsto compen-satethebanksfor theinterestforegoneonearningsassets.The coefficientontheimplicitinterestpayments(Imp)revealsthatthe increasingrelianceonimplicit interestleadstoacorresponding increase inthenetinterest marginsofthebanks.Firmswitha greaterproportionoffundsinliquidassets(Liq)havelowermargins toreflecttheirreducedliquidityriskpremiums.Further,the coeffi-cientonthecreditrisk(Cdt)ispositive,indicatingthatbankswith moreriskyloanswilltendtoselecthighernetinterestmargins.

However,whentheinterestraterisk(Int)isalsoincludedwith thebank-specificcontrolvariables,thecoefficientisfoundtobe bothpositiveandsignificant.Thisfindingdiffersmarkedlyfromthe negativesignreportedinAngbazo(1997),whereitwassuggested thatanyincreaseinthenetshort-termassets—whichimplieslower interestrateriskexposure—willleadtoarequirementforlower interestrateriskpremiums.Giventhatweuseasimilarmeasure ofinterestrateriskinthepresentstudy,themostprobable expla-nationforthedifferenceisthenetshort-termassets.Thismaybe duetothefactthatAsia’sfinancialinstitutionsoftenexcessively relyonshort-termfunding,particularlywheninterestcostsand marginsarelow(ArnerandPark,2010);therefore,theaveragenet short-termassetsmaybelessthanzero,whichmayexplainthe findingthatbankswithlowerinterest-rateriskexposurehavea requirementforlowerinterestrateriskpremiums.Theinteraction

termisfoundtobepositiveandsigniﬁcantinthefullsample regres-sion;withtheexceptionof(Int),alloftheseﬁndingsarebroadly consistentwiththoseofAngbazo(1997).

4.2. Endogenousswitchingmodelregressionresults

TheresultsforthebasicmodelarereportedinPanelAofTable5, wherewefindthatthesensitivityofnetinterestmarginstobank risk factors varies withthe degreeof bank diversification. The coefficientsoncapitalbase,opportunitycostofreserves,implicit interestpayments,liquidityrisk,interestraterisk(Int),creditrisk (Cdt)andtheinteractionbetweenthetwo(Int×Cdt)areallfound tobesignificantandhigherinthelowdiversificationregime,as comparedtothehighdiversificationregime.Thisindicatesthat banksinthehighdiversificationregimeseemtobelesssensitive tothesebank-specificvariables;italsoimpliesthatbydiversifying intonewactivitiesandplacegreateremphasisontheserevenue linesas ameansof smoothing outtheirfinancial performance, bankscanreducetheshockofidiosyncraticriskonnetinterest margins,whichisconsistentwithourhypothesis.

Sincebanksareallowedtodiversifyfunctionally,manyAsian bankshaveintegratedmutualfunddistributionorinsurance activ-itiesintotheirretailnetworks.Theseactionsmayhaveincreased theacceptanceofone-stopshoppingbycustomers,andmayalso havehelpedbankstoextractreputationalrentsfromsuch

(7)

activi-Table 4 Cross-sectional regression estimates. Variables Model (1) Model (2) Model (3) Model (4) Model (5) Model (6) Model (7) Model (8) Model (9) Constant − 0.1288 (− 14.14) *** 0.0192 (17.23) *** 0.0251 (22.84) *** 0.0132 (24.56) *** 0.0175 (13.96) *** 0.0572 (19.23) *** 0.0262 (34.62) *** 0.0263 (34.79) *** − 0.0375 (− 4.49) *** Mgmt it 0.1694 (17.09) *** 0.0642 (9.20) *** Lev it 0.1106 (8.59) *** 0.1533 (17.44) *** Opp it 0.0386 (1.58) 0.0420 (2.40) ** Imp it 0.6699 (57.03) *** 0.6626 (57.70) *** Liq it 0.0414 (8.79) *** − 0.0007 (− 0.14) Int it 0.0458 (10.70) *** 0.0281 (4.90) *** Cdt it 0.0114 (2.27) ** 0.0225 (1.51) (Int *Cdt )it − 0.0065 (− 0.99) 0.0418 (2.16) ** R 2 0.11 0.03 0.11 0.58 0.03 0.05 0.01 0.01 0.69 F -stat. 291.97 73.82 2.49 3252.97 77.32 114.48 0.02 0.32 648.01 The ordinary least squares estimates of the net interest margin regressions on management efficiency (Mgmt ), capital base (Lev ), opportunity cost of reserves (Opp ), implicit interest payments (Imp ), liquidity risk (Liq ), interest rate risk (Int ), credit risk (Cdt ) and the interaction between the latter two (Int × Cdt ) based on the following model: Nim i,t = Xit ˇ + εit Xit = (1 , Mgmt it , Le vit , Opp it , Imp it , Liq it , Int it , Cdt it , Int it × Cdt it ). The dependent variable is bank net interest margins (Nim ), which is the ratio of net interest income (before provisions for loan losses) to average earning assets; figures in parentheses are t-ratios, and the total number of observations is 2358. ** Statistical significance at the 5% level. *** Statistical significance at the 1% level.

ties,therebyenablingthemtorealizecomparativeadvantagesby diversifyingtheirincomesourcesandwideningtheirrangeof prod-ucts.Banksalsogleaninformationfromtheirlendingrelationships thatmayfacilitatetheefficientprovisionofotherfinancialservices. Moreover,information acquiredviaother financialservicescan alsoimproveloanoriginationandcreditriskmanagement.Assuch, ifintegrationweretoleadtooperationalsynergies,thenrelativeto specializedbanks,theoperatingcostsoffinancialconglomerates wouldbelower(asaresultofeconomiesofscope).

Withinaregimecharacterizedbyalowdegreeof diversifica-tion,thecoefficientsoncapitalbase,opportunitycostofreserves, implicitinterestpaymentsandcreditriskareallsignificantly pos-itive, while the coefficient on interest rate risk is significantly negative.Theseresultsarebroadlyconsistentwiththosereported

in Angbazo (1997); nevertheless, theycannot beconﬁrmed for

ahighdegreeofdiversificationregime,essentiallybecausesuch bank-specificcontrolvariablesasmanagementefficiency, opportu-nitycostofreserves,liquidityriskandcreditdonotyieldsignificant coefficientsforthenetinterestmargin.Wethereforesuggestthat thesignsonthecoefficientofriskfactorsaspredictedbyAngbazo

(1997)canholdonlywhenthebanksarelocatedinalowdegree

ofdiversiﬁcationregime.

Inthepresentstudy,thecoefficientsontheratioofnon-interest income to total operating income, revenue diversity and asset diversityareallfoundtobesignificantandpositive,indicatingthat withanincreaseintheseratios,abankismorelikelytobefaced withahighdegreeofdiversification.Thenegativecoefficientonthe loans-to-assetsratioindicatesthatthehigherthisratio,thegreater thelikelihoodofthebankbeingfacedwithalowdegreeof diver-sification.ThesefindingsareconsistentwiththoseofBaeleetal.

(2007)andLaevenandLevine(2007).12

Wenowattempttoexplainwhetherthedataarebetter char-acterizedbyamodelthatallowsfortworegimesasopposedtoa singleregime.FollowingHuandSchiantarelli(1998)andGoldfeld

andQuandt(1976),weﬁndthatthelikelihoodratiotestresultis

7.70,withsigniﬁcanceatthe5%level,leadingtoadecisive rejec-tionofthesingleregimehypothesis.13_Hence,_in_accordance_with

thetheoreticalmodelpresentedbyAngbazo(1997),weemploythe endogenousswitchingregressionmodel,dividingthebanksample intoregimesofhighandlowdegreesofdiversiﬁcation.

Wealsocarryoutaregressionestimationofthebasicmodelon anunbalancedsampleofbankswithatleastsixsuccessiveyears ofobservationsovertheperiodfrom1997to2005,theresultsof whicharereportedinPanelBofTable5.Asdepictedinthetable, thecoefficientsonmanagementefficiency,capitalbase,implicit interestpayments,liquidityriskandinterestrateriskareallfound tobesignificantand higherinthelowdegreeofdiversification regimethaninthehighdegreeofdiversificationregime.In addi-tion,thesignsofthecoefficientsonthediversificationvariables intheswitchingfunctionarefoundtobethesameasinthe bal-ancedsample,despitetheircoefficientshavingbeendetermined lessprecisely.

12_Baele_et_al.₍₂₀₀₇₎_and_Laeven_and_Levine₍₂₀₀₇₎_both_concluded_that_a_bank

thatismoreorientedtowardsnon-traditionalbankingactivitieshasalower

loans-to-assetsratioorahighernon-interestrevenueshare.Furthermore,lowervaluesof

thesediversityindicesimplygreaterspecialization,whereashighervaluessignify

thatthebankengagesinamixtureoflendingandnon-lendingactivities.

13_Hu_and_{Schiantarelli}₍₁₉₉₈₎_indicated_that_the_testing_is_complicated_by_the_fact

thattheparametersoftheswitchingfunctionarenotidentiﬁedunderthe

restric-tionthatthecoefﬁcientsinthetwodegreesofdiversiﬁcationequationsareequal.

GoldfeldandQuandt(1976)showedthatusingaChi-squareddistributionforthe

likelihoodratiotestwithdegreesoffreedomequaltothenumberofconstraints

plusthenumberofunidentiﬁedparametersyieldsatestthatfavorsnon-rejection

(8)

Table5

Estimationresultsofthebasicversionofthebalancedandunbalancedpanelswitchingregressionmodel,1997–2005.

Variables Netinterestmarginfunction Switchingfunction

Lowdegreeofdiversiﬁcation Highdegreeofdiversiﬁcation Coeff. t-statistic

Coeff. t-statistic Coeff. t-statistic

PanelA:balancedpanel:Zit=(1,Niit,Ltait,Rdit,Adit,CD,YD)

Constant −0.0514 −3.77*** _0.0293 _2.76*** _−2.0937 _−6.19*** Mgmtit 0.0077 0.66 −0.0032 −0.34 – – Levit 0.2168 18.40*** 0.0790 7.05*** – – Oppit 0.1188 3.56*** 0.0208 1.26 – – Impit 0.8233 51.99*** 0.2000 9.11*** – – Liqit 0.0302 3.42*** −0.0041 −0.73 – – Intit −0.0573 −5.85*** 0.0003 0.06 – – Cdtit 0.4456 10.34*** 0.0171 1.39 – – (Int× Cdt)it 1.2158 15.17*** 0.0196 1.24 – – Niit – – – – 1.4354 7.36*** Ltait – – – – −3.1398 −8.51*** Rdit – – – – 6.2193 20.95*** Adit – – – – 1.4469 5.87*** Loglikelihood:5825.957

Totalno.ofobservations:2358

PanelB:unbalancedpanel:Zit=(1,Niit,Ltait,Rdit,Adit,CD,YD)

Constant −0.2577 −18.93*** _−0.0063 _−0.51 _−12.4693 _−14.20*** Mgmtit 0.2417 21.83*** 0.0336 3.04*** – – Levit 0.3123 26.73*** 0.1227 9.90*** – – Oppit −0.0552 −1.76* 0.0294 1.92* – – Impit 0.2718 20.18*** −0.0414 −1.03 – – Liqit 0.0595 6.56*** 0.0061 0.86 – – Intit −0.0632 −6.54*** 0.0063 1.06 – – Cdtit 0.0068 1.11 0.0928 6.59*** – – (Int× Cdt)it 0.0382 4.52*** 0.0237 0.81 – – Niit – – – – 10.9601 16.66*** Ltait – – – – −1.5009 −3.15*** Rdit – – – – 5.4965 16.52*** Adit – – – – 8.1062 12.60*** Loglikelihood:6113.2852

PanelC:balancedpanel:Zit=(1,Niit−1,Ltait−1,Rdit−1,Adit−1,CD,YD)

Constant −0.0851 −5.78*** _−0.1007 _−5.72*** _−5.5897 _−14.12*** Mgmtit 0.0888 8.49*** _0.1429 _10.10*** _– _– Levit 0.1802 15.00*** 0.1606 7.25*** – – Oppit 0.0478 1.75* 0.0235 0.76 – – Impit 0.6653 43.19*** 0.5556 26.42*** – – Liqit 0.0138 1.44 0.0075 0.66 – – Intit −0.0260 −2.33** 0.0274 2.33** – – Cdtit 0.0308 1.79* −0.1050 −1.21 – – (Int×Cdt)it 0.0803 3.26*** −0.1316 −1.23 – – Niit−1 – – – – 3.2194 16.06*** Ltait−1 – – – – −0.8650 −1.88** Rdit−1 – – – – 4.5946 16.60*** Adit−1 – – – – 1.4345 4.64*** Loglikelihood:4849.42

Theresultsofthebasicversionoftheswitchingregressionmodelonmanagementefﬁciency(Mgmt),capitalbase(Lev),opportunitycostofreserves(Opp),implicitinterest

payments(Implicit),liquidityrisk(Liq),interestraterisk(Int),creditrisk(Cdt)andtheinteractionbetweenthelattertwo(Int×Cdt)inthenetinterestmarginequation,as

wellasthediversiﬁcationvariablesintheswitchingfunction,includingtheratioofnon-interestincometototaloperatingincome(Ni),theloans-to-assetsratio(Lta),revenue

diversity(Rd)andassetdiversity(Ad).Althoughnotreportedhere,countrydummy(CD)andyeardummy(YD)variablesareincludedinallspeciﬁcations.Thedependent

variableisthebanknetinterestmargin(Nim),whichistheratioofnetinterestincome(beforeprovisionsforloanlosses)toaverageearningassets.Themodelisdeﬁnedas

follows:

Nimi,t=Xitˇld+ε1,it if Zit+uit<0, Nimi,t=Xitˇhd+ε2,it if Zit+uit≥0

Xit=(1,Mgmtit,Levit,Oppit,Impit,Liqit,Intit,Cdtit,Intit×Cdtit,CD,YD) .

*_Statistical_{signiﬁcance}_at_the_10%_level.

**_Statistical_{signiﬁcance}_at_the_5%_level.

***_Statistical_{signiﬁcance}_at_the_1%_level.

Theresultsreportedaboveprovidegeneralsupportforthe argu-mentthatfunctionaldiversificationisimportanttobankinterest margins;however,themodelestimatedthusfarincludes contem-poraneousdiversificationvariablesastheregressor,whichcangive risetothepotentialproblemofendogeneity.Inordertoaccount forthispossibility,were-estimatethemodelwiththe contempora-neousdiversificationvariablesreplacedbytheirlaggedvalues.The

estimatesforthebalancedsamplearereportedinPanelCofTable5, wherewefindthattheresultsconfirmthoseobtainedpreviously. Anadditional potential problemwith the estimation model underdiscussionhereisthatthereisinsufficientconsiderationof thefirm-specificeffectswithintheestimations.Weassumethat wecanmodelthefirm-specifictime-invarianteffectsinthenet interestmarginandswitchingfunctionsasalinearfunctionofthe

(9)

Table6

Estimationresultsofthebasicversionoftheswitchingregressionmodelwithﬁrm-speciﬁceffects,2001–2005.

PanelA:withﬁrm-speciﬁceffects:ld

i = ¯Xiıld, hdi = ¯Xiıhd, i= ¯Ziıs Constant 0.0954 6.32*** _−0.0159 _−0.78 _−5.7805 _−9.25*** Mgmtit −0.0368 −2.81*** −0.0632 −4.44*** – – Levit 0.1188 7.20*** 0.0443 2.53** – – Oppit 0.1014 3.35*** −0.0190 −0.72 – – Impit 0.9454 62.59*** 0.2977 4.29*** – – Liqit 0.0389 3.06*** 0.0002 0.01 – – Intit −0.0390 −2.51** 0.0053 0.30 – – Cdtit 0.0081 0.50 −0.1225 −1.16 – – (Int×Cdt)it 0.0159 0.76 −0.2131 −1.00 – – Niit – – – – 0.9784 1.91* Ltait – – – – −1.3695 −1.77* Rdit – – – – 6.1913 13.28*** Adit – – – – 3.0258 5.92*** Loglikelihood:3747.284

PanelB:withoutﬁrm-speciﬁceffects(ld

i, hd i andiexcluded) Constant 0.0722 4.97*** _0.0386 _2.43** _−5.4671 _−9.44*** Mgmtit −0.0701 −7.82*** −0.0199 −1.59 – – Levit 0.1344 9.26*** 0.0773 5.21*** – – Oppit 0.0934 3.21*** 0.0232 1.13 – – Impit 0.9267 64.75*** 0.4478 6.82*** – – Liqit −0.0033 −0.31 −0.0102 −0.98 – – Intit 0.0096 0.76 −0.0062 −0.53 – – Cdtit −0.0148 −0.99 −0.0478 −0.44 – – (Int× Cdt)it −0.0139 −0.73 0.0356 0.17 – – Niit – – – – 1.8163 5.40*** Ltait – – – – −2.2635 −3.74*** Rdit – – – – 5.7779 14.73*** Adit – – – – 2.7544 6.26*** Loglikelihood:3704.5416

Theresultsofthebasicversionoftheswitchingregressionmodelwithﬁrm-speciﬁceffects.FollowingHuandSchiantarelli(1998),weestimatethemodeloverthe2001–2005

periodwhileusingthe1997–2000periodtocalculatetheaveragesforthefirm-specificvariables, ¯Xiand ¯Zi,whicharethemeansofthefirm-specificcomponents,XitandZit,

andıld_,_ıhd_and_ıs_,_which_are_the_(column)_vectors_of_the_parameters._The_model_is_deﬁned_as_follows:

Nimi,t=Xitˇld+ld_i +ε1,it if Zit+uit+i<0, Nimi,t=Xitˇhd+hd_i +ε2,it if Zit+uit+i≥0

Xit=(1,Mgmtit,Levit,Oppit,Impit,Liqit,Intit,Cdtit,Intit×Cdtit,CD,YD)

Zit=(1,Niit,Ltait,Rdit,Adit,CD,YD)

.

averagevaluesofthefirm-specificvariablesincludedwithineach ofthem.Assuch,wemustminimizethepotentialfor endogene-ityarisingfromthecorrelationbetweentheerrortermsinthenet interestmarginequationsandtheswitchingfunction,aswellasin theproxiesforthefirm-specificeffects.WethereforefollowHuand

Schiantarelli(1998)14_to_estimate_the_model_over_the_2001–2005

period,whileusingthe1997–2000periodtocomputetheaverages fortheﬁrm-speciﬁcvariables.

The estimation results of themodel with contemporaneous controlvariablesandtime-invariantfirm-specificeffectsoverthe 2001–2005periodarepresentedinPanelAofTable6,withthe resultsshowingthatinthelowdegreeofdiversificationregime, thecoefficientsoncapitalbase,opportunitycostofreservesand implicitinterestpaymentsareallpositiveandsignificant; how-ever,interestrateriskisfoundtohaveasignificantandinverse relationshipwithnetinterestmargins.Theseresultsarebroadly

14 _Hu_and_{Schiantarelli}₍₁₉₉₈₎_indicated_that_while_this_approach_is_{unsatisfactory,}

becauseitimposesrestrictionsonthenatureoftheﬁrm-speciﬁceffectsandrequires

estimationsthemodelonareducedsampleperiod,itdoesallowsuserstocheckon

therobustnessoftheresultsreachedthusfar.

consistentwiththeﬁndingsreportedbyAngbazo(1997)fortheUS bankingmarket.

The same estimation period is used in Panel B of Table 6, althoughthefirm-specificeffectsaresetasequaltozero. Exam-iningthecontrolvariables,wefindthatthecoefficientsoncapital base,opportunitycostofreservesandimplicitinterestpayments allremainsignificantandhigherinthelowdegreeofdiversification regime,ascomparedtothehighdegreeofdiversificationregime. However,thecoefficientsonliquidityrisk,interestrateriskand creditriskarenotsignificant.ConsistentwiththefindingsofBaele

etal.(2007),ourresultsalsoshowthatthecoefﬁcientsontheratio

ofnon-interestincometototaloperatingincome,loans-to-assets ratio,revenuediversityandassetdiversityarestatisticallyand eco-nomically signiﬁcant,which revealsthat a bank witha greater likelihoodofbeinginahighdegreeofdiversiﬁcationwillhavea lowerloans-to-assetsratioorahigherproportionofnon-interest revenue.

Further,wecarryoutanumberofrobustnesschecks,allofwhich arespecificationrelated.AswecanseefromPanelAofTable7, whenthecost-to-incomeratioisaddedtothediversification vari-ables,itscoefficienthasthepredictednegativesign.Animportant resultisthefindingthatbankswithgreatercostinefficiency,as

(10)

Table7

Estimationresultsofageneralversionoftheswitchingregressionmodel,1997–2005.

PanelA:Zit=(1,Lnsizeit,Cirit,Levt−1,Cdtt−1,ROEt−1,CD,YD)

Constant −0.0640 −4.40*** _0.0357 _3.16*** _−2.9398 _−5.12*** Mgmtit 0.0556 4.83*** _−0.0197 _−1.98** _– _– Levit 0.2377 16.78*** _0.0534 _5.69*** _– _– Oppit 0.1138 2.80*** _0.0246 _1.76* _– _– Impit 0.7438 44.24*** _0.1680 _6.79*** _– _– Liqit 0.0052 0.55 0.0083 1.30 – – Intit −0.0110 −1.05 −0.0101 −1.48 – – Cdtit 0.1916 4.80*** _0.0134 _1.31 _– _– (Int× Cdt)it 0.5898 8.74*** _0.0163 _1.24 _– _– Lnsizeit – – – – 0.2866 8.99*** Cirit – – – – −0.7225 −3.15*** Levit−1 – – – – 2.7297 3.66*** Cdtit−1 – – – – 0.6405 1.84* ROEit−1 – – – – 0.2196 2.50** Loglikelihood:4862.969

PanelB:Zit=(1,Niit,Ltait,Rdit,Adit,Lnsizeit,Cirit,Levt−1,Cdtt−1,ROEt−1,CD,YD)

Constant −0.0969 −8.46*** _0.1602 _7.79*** _−5.0561 _−7.2*** Mgmtit 0.0987 10.56*** −0.1452 −9.69*** – – Levit 0.1809 17.38*** 0.0306 1.43 – – Oppit 0.1703 3.82*** −0.0291 −0.89 – – Impit 0.7134 50.83*** 0.1724 4.13*** – – Liqit 0.0041 0.59 0.0061 0.42 – – Intit −0.0208 −2.81*** −0.0061 −0.37 – – Cdtit 0.3066 10.97*** −0.0110 −0.56 – – (Int×Cdt)it 0.8787 16.64** −0.0161 −0.64 – – Niit – – – – 43.1017 17.07*** Ltait – – – – −1.2787 −2.96*** Rdit – – – – 0.4640 1.95* Adit – – – – 1.1756 3.76*** Lnsizeit – – – – 0.1307 3.38*** Cirit – – – – −0.3562 −1.95* Levit−1 – – – – 2.0224 2.22** Cdtit−1 – – – – 2.8910 3.80*** ROEit−1 – – – – 0.2193 2.24** Loglikelihood:5118.609

Theresultsofamoregeneralversionoftheswitchingregressionmodelwithinwhichthefollowingvariablesareincludedinthenetinterestmarginequation:management

efﬁciency(Mgmt),capitalbase(Lev),opportunitycostofreserves(Opp),implicitinterestpayments(Imp),liquidityrisk(Liq),interestraterisk(Int),creditrisk(Cdt),andthe

interactionbetweenthelattertwo(Int×Cdt).Thefollowingdiversiﬁcationvariablesareincludedintheswitchingfunction:non-interestincometototaloperatingincome

ratio(Ni),loans-to-assetsratio(Lta),revenuediversity(Rd),assetdiversity(Ad),thenaturallogarithmofbanksize(Lnsize),cost-to-incomeratio(Cir),capitalbase(Lev),credit

risk(Cdt)andreturnonequity(ROE).Themodelisdeﬁnedasfollows:

Nimi,t=Xitˇld+ε1,it if Zit+uit<0, Nimi,t=Xitˇhd+e2,it if Zit+uit≥0

Xit=(1,Mgmtit,Levit,Oppit,Impit,Liqit,Intit,Cdtit,Intit×Cdtit,CD,YD) .

measuredbythecost-to-incomeratio,arelesslikelytobeinthe highdegreeofdiversiﬁcationregime.Thisimpliesthatbankswith superiormanagementskills orbettertechnologiesarealso per-ceivedashavingappropriatefunctionaldiversiﬁcation.

Thecoefficientonthenaturallogarithmofbanksizeisfound tobepositiveand significant,which indicatesthatasthis ratio increases,thelikelihoodofthebankbeingfacedwithahighdegree ofdiversificationalsoincreases.15_When_a_bank_applies_to_engage_in

somenewformofbusiness,therelevantauthoritieswillconsider thebank’sﬁnancialsituationasacriticalreferenceforapproval;

15_This_suggests_that_it_may_be_difﬁcult_for_small_banks_to_achieve_a_strong_foothold

innon-interestactivities,whichmaybeduetothefactthattraditionalinterest

incomeactivitiesarethoselinesofbusinesswheresmallbankshavethemost

exper-tise,whereassmallbanksmayhavelessexperienceinnon-interestactivities;see

alsoLaevenandLevine(2007).

hence,thecapitalbase(Lev),creditrisk(Cdt)andreturnonequity (ROE)areallincludedasadditionalregressorswithintheswitching function.Thecoefficientsonthecapitalbase,creditriskandreturn onequityarefoundtobesignificantandpositive,indicatingthat withanincreaseinanyoftheseratios,thebankismorelikelyto befacedwithahighdegreeofdiversification.

PanelBofTable7providestheresultsforthecasewherethe ratioofnon-interestincometototaloperatingincome, loans-to-assetsratio,revenuediversity,assetdiversity,naturallogarithmof banksize,cost-to-incomeratio,capitalbase,creditriskandreturn onequity are alltreated as diversiﬁcationvariables withinthe switchingfunction.

Thecoefficients ontheratioof non-interest income tototal operatingincome,revenuediversity,assetdiversity,natural log-arithmofbanksize,capitalbase,creditriskandreturnonequity areallsignificantlypositive,whilethenegativecoefficientsonthe loans-to-assets ratioand cost-to-income ratio indicate thatthe

(11)

highertheratio,thegreaterthelikelihoodofthebankbeingin alowdegreeofdiversificationregime.Thecoefficientsonallofthe diversificationvariablesmaintainthesamesignunderthemore restrictedspecification.

5. Conclusions

Inthepresentstudy,weextendtheAngbazo(1997)modelto includebankdiversificationaspartofourinvestigationintohow diversificationofbusinessbybanksaffectsthedeterminantsofnet interestmarginsfora sampleofcommercialbanksoperatingin Asia;weuseatwo-regimeendogenousswitchingmodelto catego-rizebanksintoregimesofhighandlowdegreesofdiversification. Wefindthatforfunctionallydiversifiedbanks,netinterest mar-ginscanbelesssensitivetofluctuationsinbankriskfactorsthan thenetinterest marginsof specializedbanks.Thisis consistent withourhypothesis,andimpliesthatbydiversifyingtheirincome sourcesandplacingemphasisontheserevenuelinestosmooth theirfinancialperformance,thebankscanreducetheimpactof idiosyncraticriskontheirnetinterestmargins.Wealsoshowthat thesignsonthecoefficientofriskfactorsaspredictedbyAngbazo

(1997)canholdonlywhenthebanksarelocatedinalowdegreeof

diversificationregime;however,theconclusionsofAngbazo(1997) cannotbeconfirmedwhenthebanksarelocatedinahighdegree ofdiversificationregime.

Inaddition,wefindthatthecoefficientsontheratioof non-interestincometototaloperatingincome,revenuediversityand assetdiversityaresignificantlypositive,indicating thatwithan increaseintheseratios,thebankwillbemorelikelytofaceahigh degreeofdiversification.Thenegativecoefficientonthe loans-to-assets ratioindicatesthat thehigher this ratio,thegreater the likelihoodthatthebankwillfacealowdegreeofdiversification. Assuch,weconfirmthefindingsofbothBaeleetal.(2007)and

LaevenandLevine(2007):banksthatareorientedtowards

non-traditionalbankingactivitieshavealowerloans-to-assetsratioor ahigherproportionofnon-interestrevenue.

However,withinthecontextofthecurrentfinancialcrisison aglobalscale,toomuchrelianceonnon-interesttypesofrevenue mayraisethelevelofriskforthesebanks.Itwouldthereforeappear thatthesecurityofsuchbankswillbelargelydependentuponthe waysinwhichtheyinteractwitheconomy-wideshocks,aswellas thetypesofdiversifyingactivitiesthattheychoosetoundertake, althoughtheseissuesarebeyondthescopeofthepresentstudy. Furtherresearchinthisareashouldaimtoinvestigatewhetherthe observedshifttowardsnon-interestincomeactivitieswillcontinue tobenefitthesefinancialconglomeratesoncethecurrentglobal financialcrisishassubsided.

Acknowledgement

HuiminChungwouldliketogratefullyacknowledgethe ﬁnan-cialsupportprovidedbytheMoEATUPlanatNationalChiaoTung University.TheauthorsthankIftekharHasan(theeditor)andtwo anonymousrefereesfortheirhelpfulcommentsandsuggestions. References

Albertazzi,U.,Gambacorta,L.,2009.Bankproﬁtabilityandthebusinesscycle.Journal

ofFinancialStability5,393–409.

Allen,L.,1988.Thedeterminantsofbankinterestmargins:anote.Journalof

Finan-cialandQuantitativeAnalysis23,231–235.

Angbazo,L.,1997.Commercialbanknetinterestmargins,defaultrisk,interest

rateriskandoff-balancesheetbanking.JournalofBankingandFinance21,

55–87.

Arner,D.W.,Park,C.Y.,2010.GlobalFinancialRegulatoryReforms:Implicationsfor

DevelopingAsia,AsianDevelopmentBankWorkingPaperSeriesonRegional

EconomicIntegrationNo.57.

Baele,L.,DeJonghe,O.,VanderVennet,R.,2007.Doesthestockmarketvaluebank

diversiﬁcation?JournalofBankingandFinance31,1999–2023.

Berger,A.N.,Hasan,I.,Zhou,M.,2010.Theeffectsoffocusversusdiversiﬁcation

onbankperformance:evidencefromChinesebanks.JournalofBankingand

Finance34,1417–1435.

Berger,A.N.,1995.Therelationshipbetweencapitalandearningsinbanking.Journal

ofMoneyCreditandBanking27,432–456.

Carbo,S.,Rodriguez,F.,2007.ThedeterminantsofbankmarginsinEuropean

bank-ing.JournalofBankingandFinance31,2043–2063.

Chen,X.,2007.Bankingderegulationandcreditrisk:evidencefromtheEU.Journal

ofFinancialStability2,356–390.

Drakos,K.,2003.Assessingthesuccessofreformintransitionbanking10yearslater:

aninterestmarginsanalysis.JournalofPolicyModeling25(3),309–317.

DellAriccia,G.,Marquez,R.,2010.Riskandthecorporatestructureofbanks.Journal

ofFinance65,1075–1096.

DeYoung,R.,Roland,K.P.,2001.Productmixandearningsvolatilityat

commer-cialbanks:evidencefromadegreeoftotalleveragemodel.JournalofFinancial

Intermediation10,54–84.

Flannery, M.C., James, C.M., 1984. The effects of interest rate changes on

thecommonstockreturnsofﬁnancialinstitutions. JournalofFinance39,

1141–1153.

Gochoco-Bautista,S.,Oh,S.-N.,Rhee,S.G.,2000.IntheeyeoftheAsianﬁnancial

maelstrom:bankingsectorreformsintheAsia-Paciﬁcregion.In:Risingtothe

ChallengeinAsia:AStudyofFinancialMarkets:Volume1–AnOverview.Asian

DevelopmentBank,Manila,pp.50–106.

Goldfeld,S.M.,Quandt,R.E.,1976.Techniquesforestimatingswitching

regres-sions.In:Goldfeld,S.M.,Quandt,R.E.(Eds.),StudiesinNon-LinearEstimation.

Ballinger,Cambridge,MA,pp.3–36.

Hasan,I.,Sarkar,S.,2002.Banks’optiontolend,interestratesensitivity,andcredit

availability.ReviewofDerivativesResearch5(3),213–250.

Ho,T.S.Y.,Saunders,A.,1981.Thedeterminantsofbankinterestmargins:theoryand

empiricalevidence.JournalofFinancialandQuantitativeAnalysis16,581–600.

Hu,X.,Schiantarelli,F.,1998.Investmentandcapitalmarketimperfections:a

switchingregressionapproachusingUSﬁrmpaneldata.ReviewofEconomics

andStatistics80(3),466–479.

Laeven,L.,Levine,R.,2007.Isthereadiversiﬁcationdiscountinﬁnancial

conglom-erates?JournalofFinancialEconomics85,331–367.

Lepetit,L.,Nys,E.,Rous,P.,Tarazi,A.,2007. Bankincomestructureandrisk:

anempiricalanalysisofEuropeanbanks.JournalofBankingandFinance32,

1452–1467.

Lepetit,L.,Nys,E.,Rous,P.,Tarazi,A.,2008.TheexpansionofservicesinEuropean

banking:implicationsforloanpricingandinterestmargins.JournalofBanking

andFinance32,2325–2335.

Maddala, G.S., 1986. Disequilibrium,Self-Selection andSwitching Models.In:

Griliches,Z.,Intrilgator,M.D.(Eds.),HandbookofEconometrics,vol.3.

Ams-terdam,pp.1633–1688.

Maudos,J.,FernandezdeGuevara,J.,2004.Factorsexplainingtheinterestmarginin

thebankingsectorsoftheEuropeanunion.JournalofBankingandFinance28,

2259–2281.

Maudos,J.,Solís,L.,2009.Thedeterminantsofnetinterestincomeinthe

Mexi-canbankingsystem:anintegratedmodel.JournalofBankingandFinance33,

1920–1931.

McShane,R.W.,Sharpe,I.G.,1985.Atime-seriescross-sectionanalysisofthe

deter-minantsofAustraliantradingbankloan-depositinterestmargins:1962–1981.

JournalofBankingandFinance9,115–136.

Mercieca,S.,Schaeck,K.,Wolfe,S.,2007.SmallEuropeanbanks:beneﬁtsfrom

diver-siﬁcation?JournalofBankingandFinance31,1975–1998.

Saunders,A.,Schumacher,L.,2000.TheDeterminantsofBankInterestRate

Mar-gins:AnInternationalStudy.JournalofInternationalMoneyandFinance19(6),

813–832.

Stiroh,K.J.,2004.Diversiﬁcationinbanking:isnon-interestincometheanswer?

JournalofMoneyCreditandBanking36,853–882.

Stiroh,K.J.,2006.Aportfolioviewofbankingwithinterestandnon-interest

activi-ties.JournalofMoneyCreditandBanking38,1351–1361.

Stiroh,K.J.,Rumble,A.,2006.TheDarkSideofDiversiﬁcation:TheCaseofUS

Finan-cialHoldingCompanies.JournalofBankingandFinance.30,2131–2161.

Williams,J.,Nguyen,N.,2005.Financialliberalisation,crisisandrestructuring:a

comparativestudyofbankperformanceandbankgovernanceinSouthEast