IMF-信貸來源的風(fēng)險與經(jīng)濟(jì)活動的下行風(fēng)險（英）

TheRiskinessofCreditOriginsandDownsideRiskstoEconomic

Activity

PreparedbyClaudioRaddatz,DulaniSeneviratne,Jér?me

Vandenbussche,PeichuXie,andYizhiXu

WP/24/72

IMFWorkingPapersdescriberesearchin

progressbytheauthor(s)andarepublished

toelicitcommentsandtoencouragedebate.

TheviewsexpressedinIMFWorkingPapersare

thoseoftheauthor(s)anddonotnecessarily

representtheviewsoftheIMF,itsExecutive

Board,orIMFmanagement

2024

MAR

NAr

?2024InternationalMonetaryFund

WP/24/72

IMFWorkingPaper

MonetaryandCapitalMarketsDepartment

TheRiskinessofCreditOriginsandDownsideRiskstoEconomicActivity

PreparedbyClaudioRaddatz,DulaniSeneviratne,Jér?meVandenbussche,PeichuXie,andYizhiXu

*

AuthorizedfordistributionbyDavidHofman

March2024

IMFWorkingPapersdescriberesearchinprogressbytheauthor(s)andarepublishedtoelicit

commentsandtoencouragedebate.TheviewsexpressedinIMFWorkingPapersarethoseofthe

author(s)anddonotnecessarilyrepresenttheviewsoftheIMF,itsExecutiveBoard,orIMFmanagement.

ABSTRACT:Weconstructacountry-levelindicatorcapturingtheextenttowhichaggregatebankcreditgrowthoriginatesfrombankswitharelativelyriskierprofile,whichwelabeltheRiskinessofCreditOrigins(RCO).

Usingbank-leveldatafrom42countriesovermorethantwodecades,wedocumentthatRCOvariationsovertimeareafeatureofthecreditcycle.RCOalsorobustlypredictsdownsideriskstoGDPgrowthevenafter

controllingforaggregatebankcreditgrowthandfinancialconditions,amongotherdeterminants.RCO’s

explanatorypowercomesfromitsrelationshipwithassetquality,investorandbankingsectorsentiment,aswellasfuturebankingsectorresilience.Ourfindingsunderscoretheimportanceofbankheterogeneityfortheoriesofthecreditcycleandfinancialstabilitypolicy.

RECOMMENDEDCITATION:ClaudioRaddatz,DulaniSeneviratne,JeromeVandenbussche,PeichuXie,andYizhiXu,?TheRiskinessofCreditOriginsandDownsideRiskstoEconomicActivity”,IMFWorkingPaperNo.24/72(2024)

JELClassificationNumbers:E44,E47,G01,G21,G28

Keywords:

Privatesectordebt;creditgrowth;creditorigin;creditcycle;banksoundness;creditrisk;financialvulnerability;investorsentiment;financialstability

Authors’emailaddresses:

clraddatz@fen.uchile.cl;

dseneviratne@IMF.org;

jvandenbussche@IMF.org;

xiepeichu@;

yxu@IMF.org

*ClaudioRaddatz(UniversidaddeChile);DulaniSeneviratne,Jér?meVandenbussche,andYizhiXu(InternationalMonetaryFund);andPeichuXie(CitadelLLC).

WORKINGPAPERS

TheRiskinessofCreditOrigins

andDownsideRiskstoEconomic

Activity

PreparedbyClaudioRaddatz,DulaniSeneviratne,Jér?me

Vandenbussche,PeichuXie,andYizhiXu

1

1TheauthorswouldliketothankTobiasAdrianaswellasparticipantsataseminarattheIMFandthe2023RIDGEconferenceonfinancialstabilityforcomments,andKen(Zhi)GanandDiegoVillalobosforexcellentresearchassistance.

IMFWORKINGPAPERSTheRiskinessofCreditOriginsandDownsideRiskstoEconomicActivity

INTERNATIONALMONETARYFUND

2

Contents

Acronyms/Glossary 3

I.Introduction 4

II.TheoreticalUnderpinningsandFurtherLinkstotheLiterature 6

III.RiskinessOfCreditOriginsMeasurementandSamplesConstruction 8

IV.RiskinessOfCreditOriginsandDownsideRiskstoGrowth 13

V.WhyDoesRCOPredictDownsiderisks?ExploringtheChannels 15

VI.Conclusion 20

References 22

INTERNATIONALMONETARYFUND

3

Acronyms/Glossary

BLSBankLendingStandards

CACurrentAccount

EDFExpectedDefaultFrequency

FCIFinancialConditionsIndex

GDPGrossDomesticProduct

GFCGlobalFinancialCrisis

GUOGlobalUltimateOwner

IFSInternationalFinancialStatistics

LLPLoanLossProvisions

OLSOrdinaryLeastSquares

NPLNonperformingLoans

RCARiskinessofCreditAllocation

RCORiskinessofCreditOrigins

IMFWORKINGPAPERSTheRiskinessofCreditOriginsandDownsideRiskstoEconomicActivity

INTERNATIONALMONETARYFUND

4

I.Introduction

Abundantempiricalevidencesupportstheviewthatperiodsoflargeaggregatecreditexpansionstendtobe

followedbyadversemacroeconomicoutcomesandtheoccurrenceoffinancialcrises(Jordaetal.2011,

SchularickandTaylor2012,Mianetal.2018,amongothers),especiallywhenthecreditexpansiontakesplaceinanenvironmentofeasyfinancialconditionsandbuoyantcreditsentiment(KrishnamurthyandMuir2017,

López-Salidoetal.2017,Kirti2021,Adrianetal.2022,Greenwoodetal.2022).However,existingcross-

countryempiricalstudiesfocusonaggregatemeasuresofthevolumeandpriceofcreditandleaveasidetherolethatthecompositionofcreditoriginationandlenderheterogeneitymayplayinaggregaterisk-takingandfinancialstability.

Anecdotalevidencesuggeststhatfasterbank-levelcreditgrowthduringaboomisassociatedwithworse

performanceduringtheensuingbustandthatthestrengthoffinancialinstitutionsdrivingtheexpansionmattersforfutureaggregateoutcomes.DuringtheGlobalFinancialCrisis(GFC),severaliconicfailureswerefinancialintermediariesthathadfollowedaveryaggressiveexpansionstrategy.IntheUnitedStates,Countrywide

FinancialandWashingtonMutualbecamethefirstandthirdlargestmortgageoriginatorsoverashortperiodbeforethecrisis,lostbillionsonsubprimeexposures,andhadtoberesolvedin2008(UnitedStatesSenate,2010).Spanishsavingsbanks,whichwereattheepicenteroftheSpanishbankingcrisisadecadeago,hadexperiencedacontinuousriseintheirloanmarketshareintherun-uptothecrisis(Santos,2018).Anglo-IrishBank,theonlyIrishbanknationalizedduringtheIrishbankingcrisisof2008-2010,hadthefastestpre-crisiscreditgrowthamongmajorIrishbanks(ReglingandWatson,2010).Goingfurtherbackintime,duringthe

creditboominFinlandandSwedenintheearly1990s,themostaggressivelendersweretheweakestincapitalizationandunderlyingprofitability(EnglundandVihriala,2010).

Theoreticalmodelsoffinancialamplificationandfinancialcriseshavelongrecognizedtheimportanceof

accountingforheterogeneityacrosseconomicagents(BernankeandGertler1989;KiyotakiandMoore1997;BrunnermeierandSanikov2014)

2

.Itisonlyrecentlythatsomemacrofinancialmodelshavefocusedon

heterogeneityacrossfinancialintermediariesandshownhowthisheterogeneitymattersforthedynamicsofaggregaterisk-takingandfinancialstability(Geanakoplos2010,KorinekandNowak2017,CoimbraandRey2018and2023).

Inthispaper,weprovidenovelempiricalevidencethattheextenttowhichthegrowthinaggregatebank

lendingactivityconcentratesinriskierbanksvariesoverthecreditcycleand,moreimportantly,thatithelps

predictdownsideriskstoeconomicgrowth.

3

Furthermore,weprovidecountry-levelandbank-levelanalysestoexplorethemechanismsunderlyingourkeyresult.

Specifically,usingalargesampleof3071banksacross42countriesoverthe1990–2019period,weconstructanaggregatemeasureoftheextenttowhichcreditisoriginatedbyrelativelyriskierbanks(asmeasuredbythewithin-country,relativez-score),takinginspirationfromtheapproachofGreenwoodandHanson(2013)for

2Thesemodelsgenerallyimposeconditionsthatleadtotheseparationofheterogeneousagentsinborrowers,lenders,or

intermediariesinequilibrium.Mosttraditionalmodelseitherassumethateachsectorisrepresentedbyasingleagentorthatthereisperfectrisksharingwithinasector,sothatheterogeneitywithinasector—thatis.acrossborrowersorfinancialintermediaries—doesnotmatter.

3Inthepaper,weusetheexpressions“riskierbank”and“weakerbank”interchangeably.

INTERNATIONALMONETARYFUND

5

capturingthecompositionofaggregatedebtissuanceacrossheterogeneousborrowers.Wepresentevidencethatourmeasure,whichwelabeltheRiskinessofCreditOrigins(RCO),riseswhenaggregatecreditgrowthincreasesandwhenfinancialconditionsbecomelooser.Inaddition,weprovidecomplementarybank-level

evidencedocumentingtheunderlyingmechanismatthemicrolevel.Thesepatternsinthecross-sectionof

bankrisk-takingoverthecreditcyclecapturedorproxiedbyRCOarenotonlyofintrinsicinterestasa

characterizationofthecycle,buttheyalsohelpshedfurtherlightonwhylargecreditexpansionspresentariskforfinancialstability.

WeshowthatanincreaseinRCOpredictsdownsideriskstoGDPgrowth,evenaftercontrollingforkey

determinantspreviouslyhighlightedintheliterature,includingaggregatecreditgrowthandfinancialconditions.Themagnitudeoftheeffectswedocumentissizable.Aone-standard-deviationincreaseinRCOshiftsthelefttailoftheaveragecumulativetwo-year-aheadGDPgrowthdistributionbyabout30basispointinourbaselinespecification.Ourfindingsarerobusttoabatteryofrobustnessteststhatincludeusingadditionalcontrols

(includinganaggregatemeasureofbankingsectorriskiness),analternativemeasureofbank-levelriskiness,arestrictedsampleofbanksintheanalysis,oranalternativequantileregressionestimationmethod.

Finally,weexplorethreepossible—andsomewhatrelated—channelsunderlyingourkeyfinding.Wefirst

examineacreditqualitychannel.Atthemicrolevel,weinvestigatewhetherriskierbankslendmoretoriskier

borrowers,leadingtoaweakerfutureloanportfolioperformance,andhowthisrelationshipdependsonbank-levelrelativecreditgrowth.Wedocumentthatbanksthatexpandcreditrelativelyfasterexperienceagreaterincreaseinloanlossprovisionsandnonperformingloanratioslaterandthatthisincreaseisevenstronger

whenthebankisex-anteriskier(thatis,whenithasalowerrelativez-score).Atthemacrolevel,wealso

analyzewhetherRCO’sexplanatorypowerfordownsideriskstogrowthisaffectedbytheinclusionofa

variablecapturingariskierallocationofcredit(BrandaoMarquesetal.2022)inthespecification.Wefindthatitdoesathorizonsuptotwoyears.

Asecondplausiblechannelissentiment.InthespiritofLópez-Salidoetal.(2017)whoproxycreditsentimentbyfinancialvariablesthatpredictfuturechangesincreditspreads,weexaminewhetherRCOpredictsfuturechangesinaggregatebanklendingstandardsandfinancialconditions.Wefindthatitdoesathorizonsuptotwoyearsforbanklendingstandardsandfinancialconditions.Bothfindingsstronglysupportasentiment

channel.

4

Finally,RCOcouldcaptureadimensionofaggregatebankingsectorvulnerabilityrelatedtothedistributionofbank-levelvulnerabilities.Byconstruction,RCOmeasurestheextenttowhichbanksthatarerelativelyriskier

contributetotheexpansionofbankingsectorcredit.Whiletherelativenatureoftheinputstothemeasuredoesnotimplyamechanicalrelationship,wespeculatethatperiodswhenRCOiselevated,especiallyiftheypersist,couldresultinalargerfractionofaneconomy’sloanportfoliobeingconcentratedinriskierbanks.Totheextentthatriskierbanksaremorelikelytoreducetheirlendinginthefutureinresponsetoanadverseshock,andthatborrowersfacefrictionswhentryingtoshiftlenders,thiscouldresultinanaggregatecontractioninlendingandactivity.Insupportoftheexistenceofthisthirdchannel,wefindthatbankriskinessisadeterminantoffuturebank-levellendingactivityfollowinglargenegativeshocks.WealsofindthatRCOpredictsleftwardshiftsofthe

4Notethatalthoughthecreditqualityandbankingsectorsentimentchannelsbearsomeresemblance,theyareconceptually

distinct.Inthecreditqualitychannel,poorfutureaggregateperformanceisduetoadeteriorationoflendingqualitybyriskierbanks.Inthebankingsectorsentimentchannel,poorfutureaggregateperformanceiscouldbedueadeteriorationinlendingqualityacrosstheboard.

INTERNATIONALMONETARYFUND

6

extremelefttailofbankingsectorstockreturns,whichisalsoconsistentwiththepresenceofaresiliencechannel.

5

Therestofthepaperisstructuredasfollows.SectionIIdiscussesthetheoreticalunderpinningsofthe

relationshipbetweenbankriskiness,risk-taking,andcreditcycle,andreviewstherelevanttheoreticaland

empiricalliteratures.SectionIIIintroducesourmeasureoftheRCO.SectionIVanalyzesitsco-movementwithaggregatechangesinbankcreditandprovidesrelatedbank-levelevidence.SectionVdocumentsRCO’s

predictivepowerforfuturedownsideriskstogrowthwhileSectionVIpresentsouranalysisofthethreepossiblechannelsunderlyingthisrelationship.SectionVIIconcludes.Appendicesprovideadditionalinformationondatasources,variablesconstruction,sampleconstruction,andadditionalrobustnessanalyses.

II.TheoreticalUnderpinningsandFurtherLinkstotheLiterature

Therelationshipbetweenbankriskiness—theprobabilitythatabankwilldefaultonitsobligations—andrisk-takingistheoreticallyambiguous.Ontheonehand,classicrisk-shiftingincentivesduetolimitedliability

(JensenandMeckling1976)naturallygenerateapositiveassociationbetweenthetwo.

6

Inaddition,lowbankcapitalizationreducestheincentivestomonitorloanqualitybecauseofmarketimperfections(HolmstromandTirole1997;Allenetal.2011).

7

Evenifbankcreditorsareawareoftheseincentivesandaskforcompensationthroughahighercostofbankdebtorattempttoexertdisciplineonmanagersthroughgreaterrelianceon

runnabledemanddeposits(CalomirisandKahn1991;DiamondandRajan2000,2001),theexistenceof

depositinsuranceorimplicitgovernmentguaranteescouldlimitmarketdisciplineorefficiency(Gortonand

Huang,2004;FarhiandTirole,2012).Ontheotherhand,thethreatofrunsmaybeastrongincentiveforbankstoavoidrisk-shiftingbehavior(JacklinandBattacharya,1988;DiamondandRajan,2000;Iyeretal.,2016).Theabilityofbondholderstoimposecovenants(Ashcraft,2008)orregulatoryconstraintsmayalsolimittheabilityofbankstotakerisks(DewatripontandTirole2012).

Regardlessofthesignoftherelationshipbetweenbankriskinessandrisk-takinginordinarybankcreditmarketconditions,riskierbanks’incentivesforrisk-takingarelikelyrelativelygreaterduringbuoyantaggregatecreditexpansionsforvariousreasons.First,theoreticalmodelswithrationalagentsindicatethatlendingstandards

areprocyclicalbecauseofendogenousvariationintheprofitabilityofscreeningortheinformationonthequalitycompositionofborrowers(Ruckes2004;Dell’AricciaandMarquez2006),orbecauseoflossininstitutional

memory(BergerandUdell2004).Sincescreeningbenefitsarearguablylowerforweakerbanksbecauseofthe

5WealsoexplorewhetherRCO’spredictivepowerfordownsideriskstogrowthisaffectedbytheinclusionoftheskewnessofthedistributionofbankleverage(CoimbraandRey,2018)inthespecification,andfindnoconsistentevidencethatitdoes.Asaby-product,wealsofindthattheleverageskewnessmeasureisnotstatisticallysignificantinourregressionresults.

6Likeothertypesoffirms,banksprotectedbylimitedliabilityhavesuchincentivesbecauseoftheoptionvalueofequity:abanktakingariskwillreapthebenefitswhenthegamblepaysoffandwillleaveitscreditorsholdingthebucketwhenitdoesnot.Theseincentivesarestrongerwhenbanksolvencyislower.

7Conversely,underlimitedliability,bankswithhigherriskappetitechoosetobemoreleveragedandriskier(CoimbraandRey,2023).

IMFWORKINGPAPERSTheRiskinessofCreditOriginsandDownsideRiskstoEconomicActivity

INTERNATIONALMONETARYFUND

7

debtoverhangproblem(Myers1977),therelaxationofstandardsingoodtimesislikelystrongeramongthem.InCoimbraandRey(2023),loweraggregatefundingcostsencouragebankswithahigherriskappetiteto

expandtheircreditprovisionandleveragerelativelymore.Second,withboundedlyrationalagents,thepriceofriskistoolowduringtheexpansionaryphaseofthecreditcyclebecauseofdiagnosticexpectations(Bordaloetal.2018)orneglectofcrashrisk(BaronandXiong2017).Theresultingeasieraccesstodebtfinancingwouldfacilitaterisk-takingbybankswithrelativelyhigherincentivestoengageinthisbehavior.

Altogether,thesetheoreticalconsiderationssuggestthatthecreditcycleshouldbeanimportantdriverofcross-sectionaldifferencesinbankrisk-takingthroughloanportfoliogrowth,whichiswhatourRCOmeasure

captures.Yetthishypothesishassofarremaineduntested.CoimbraandRey(2018)constructthewithin-

countryskewnessoftheleveragedistributionacrossbanks.Theirindicatorisanaggregatemeasureofbankingsectorriskinessbasedonasingledimension(bankleverage),whileourscapturestwodimensionsby

combiningthebank-levelriskinessdimensionwithinformationontheflowofcredittocreateanindicatoroftheRCOatanygivenpointintime.

Ontheempiricalside,ourcyclicalityanalysisrelatestopriorbank-levelevidencesuggestinganassociation

betweenbankriskinessandbankrisk-taking.IganandTamirisa(2008)andIganandPinheiro(2011)findthatweakerbanksgrowtheirloanportfoliosmoreslowlythanstrongerbanksinnormaltimesbutgrowthematthesamepaceasotherbanksduringcreditbooms.Ourloangrowthregressionresultsechotheirs,butour

empiricalspecificationismoreparsimonious,andourkeymacrodriverisaggregatecreditgrowthratherthanadummycapturingepisodesofcreditbooms.Ourcyclicalityanalysisalsorelatestotheliteratureontherisk-

takingchannelofmonetarypolicy,inwhichvariouspapershaveusedgranularsupervisorydatatoshowthat

loosermonetarypolicyinducesbankstotakemoreriskandthatthiseffectdependsonbanksolvency(Jimenezetal.2014,Dell’Aricciaetal.2017).Wecomplementthisliteraturebyfocusingonabroadersampleof

countriesandonthecreditcycleratherthanonchangesinmonetarypolicy.

ThemainanalysisinourpaperrelatingRCOtodownsideriskstoGDPgrowthisdirectlyconnectedtothe

bankingcrisisliterature(Gourinchasetal.2001,Obstfeld2012,SchularickandTaylor2012,Dell’Aricciaetal.

2016,Jordàetal.2021,amongothers)andthegrowth-at-riskliterature(Giglioetal.2016,Adrianetal.2019,Adrianetal.2022)whichhaveinvestigatedtheroleplayedbyaggregatecreditgrowth,financialconditions,andstandardaggregatebankingsoundnessindicatorsindrivingadversemacrofinancialoutcomes.Weaddto

theseliteraturesbydemonstratingtheimportantroleoftheoriginsofbankcredit.

Ourmicroanalysisoftheassetqualitychannelbuildsonseveralempiricalpapersthathaveexaminedthe

bank-levelrelationshipbetweensizeofloangrowthandsubduedfutureperformance.Thesepapershave

shownthatbankswhoseloanportfoliogrowsfastest(relativetodomesticpeers)sufferfromarelativelyweakerperformancewithinafewyears,regardlessofwhetherperformanceismeasuredbythenon-performingloan

ratio(JimenezandSaurina2006;ChavanandGambacorta2019),loanlossprovisions(Foosetal.2010),

stockreturns,orreturnonassets(Fahlenbrachetal.2018).Wecomplementthesestudies,allfocusedon

singlecountries,byexaminingthisrelationshipinabroadsampleofcountriesand,mostimportantly,by

showingthatbank-levelriskinessamplifiestheeffectofrelativesizeofloanportfoliogrowthinaffectingfutureperformance.Inaddition,inasmallersampleofbanks,wedocumentthatexantecreditquality(measuredbytheshareofleveragedloansissuanceintotalloanissuance)isgreaterinbanksthatareriskierandgrowtheirloanbookrelativelyfaster.Ourdiscussionoftheassetqualitychannelatthecountrylevelrelatestothemacro

IMFWORKINGPAPERSTheRiskinessofCreditOriginsandDownsideRiskstoEconomicActivity

INTERNATIONALMONETARYFUND

8

literatureonlendingstandardsandGDPgrowth(GreenwoodandHanson2013,Kirti2021,BrandaoMarquesetal.2022).

Ourdiscussionoftheresiliencechannelisindirectlyrelatedtothemicroliteratureonrelationshipbanking,

whichhasextensivelydocumentedthecostsofswitchingbanksforborrowers(James1987;andPetersenandRajan1994;ElyasaniandGoldberg2004;Hubbardetal.2002;Schwert2018).Atthemacrolevel,Coimbra

andRey(2018)showthataggregatecreditgrowthismoreresponsivetofundingcostswhentheskewnessofthebankleveragedistributionincreases.Bycontrastwithourwork,CoimbraandRey(2023)donotrelatetheirindicatortofinancialstabilityoutcomevariables,aswedo.

Whilethetypicalinterpretationintheliteratureofthepositiverelationshipbetweencreditgrowthandfuture

recessionsorcriseshasbeenthatfastercreditgrowthimplieshigherfinancialvulnerabilitiesbecauseofhigherleverageintheeconomy,theevidenceweprovidealsosuggeststhatcompositioneffectserodebankingsectorresilienceduringtheupwardphaseofthecreditcycleastherelativelymorefragilebanks’contributionsto

creditgrowthandrisk-takingincrease.SuchcompositioneffectisafeatureofKorinekandNowak(2017)’s

model,inwhich,becauseofimperfectrisk-sharing,asequenceofpositiveaggregateshocksallowsthemarketshareofintermediarieswithahigherriskappetitetogroworganicallyand,therefore,increasesthevulnerabilityoftheeconomytobadshocks.

III.RiskinessOfCreditOriginsMeasurementandSamplesConstruction

MeasuringtheRiskinessofCreditOrigins

WemeasuretheRCObasedontheapproachofGreenwoodandHanson(2013)fornonfinancialfirmsintheUnitedStates.Thisapproachconsistsoffoursteps,whichweapplytobanksforeachcountry-yearinour

sample.First,wesortthesebanksintodecilesaccordingtoanindicatoroftheirriskinessandassigneachbankitsdecilepositioninthedistribution(ahigherdecilecorrespondingtohigherriskiness).Second,wesortall

banksintotwogroupsaccordingtotheirannualloangrowthandclassifyallbankswithloangrowthequaltoorabove(below)themedianastop(bottom)lenders.Third,wecomputetheaveragelaggedriskinessdecile

amongtopandbottomlenders.

8

Finally,wetakethedifferencebetweenthesetwoaverages.Formally,themeasureisdefinedasfollows:

RCOc,t=ΣieToPc,tRisk(decile)i,c,t-1

一ΣieBottomc,tRisk(decile)i,c,t-1

(1)

whereRisk(decile)i,c,t-1isthedecileinthedistributionofbanki’sriskinessmeasureincountrycattimet-1,NandNmarethenumberofbanksinthetopandbottomhalfofthedistributionofloangrowthin

countrycattimet,respectively.BecausethepaperfocusesonthedynamicsofRCOwithincountriesandnot

8Weobtainverysimilarresultsifweusethecontemporaneousriskinessdecileinsteadofitslaggedvalueinthecountry-levelanalysispresentedlaterinthepaper.

IMFWORKINGPAPERSTheRiskinessofCreditOriginsandDownsideRiskstoEconomicActivity

INTERNATIONALMONETARYFUND

9

onitscross-countryvariation,wenormalizethisrawmeasurebysubtractingitscountry-specificmean.Thisadjustmentremovestheinfluenceofthecountry-specificsectoralcompositionofbanksandensuresgreatercross-countrycomparability.AnincreaseinRCOsignalsthatbanksexpandinglendingrelativelyfasterareriskier,indicatingariskieraggregateoriginofcredit.Byconstruction,theunitsofRCOcorrespondtodeciles,soavalueof1indicatesthattopissuershaveanaverageriskinesswhichisonedecileabovethatofbottomissuers.

Followingthebankingliterature,ourbaselinemeasureofbankriskinessistheoppositeofabank’sz-score,

definedasthesumofthereturnonaverageassetsandtheleverageratio,dividedbythehistorical(three-year)standarddeviationofreturnsonaverageassets.

9

Thez-scorecapturestheextenttowhichabank’scurrent

incomeandequitycapitalcanabsorbfluctuationsinincome,soahighervalueindicatesasaferbank.Foritsopposite,ahigher(lessnegative)valueindicatesariskierbank.

Asanalternativetothez-score,wealsoconstructameasureofbankriskinessbasedonbalancesheet

indicatorsofbankfundamentals.Followingtheliterature,weconsiderthefollowingsetofbankfundamentalsrelatedtotheCAMEL/CAELratingsapproachinitiallydevelopedbyU.S.banksupervisors(seePurnandaram,2007):(i)capitaladequacy,capturedbytheprincipalcomponentofabank’sratiooftotalequitytototalassetsanditsz-score(asdefinedabove);(ii)theratioofloanlossprovisionstototalassetscapturingassetquality;(iii)thereturnonaverageassetsasameasureofprofitability;(iv)thecost-to-incomeratioasaproxyfor

efficiency;and(v)liquidity,capturedb