Edinburgh Research Explorerv3.globalcube.net/clients/eacb/content/medias/...Probability of Default...
Transcript of Edinburgh Research Explorerv3.globalcube.net/clients/eacb/content/medias/...Probability of Default...
Edinburgh Research Explorer
Probability of default and efficiency in cooperative banking
Citation for published version:Fiordelisi, F & Mare, D 2013, 'Probability of default and efficiency in cooperative banking' Journal ofInternational Financial Markets, Institutions, and Money, vol 26, pp. 30-45., 10.1016/j.intfin.2013.03.003
Digital Object Identifier (DOI):10.1016/j.intfin.2013.03.003
Link:Link to publication record in Edinburgh Research Explorer
Document Version:Author final version (often known as postprint)
Published In:Journal of International Financial Markets, Institutions, and Money
Publisher Rights Statement:This is an Author's Accepted Manuscript of an article whose final and definitive form, the Version of Record, hasbeen published in the Journal of International Financial Markets, Institutions, and Money, 26, 2013. © 2013Elsevier B.V. The final publication is available online at: http://dx.doi.org/10.1016/j.intfin.2013.03.003
Fiordelisi, F., & Mare, D. (2013). Probability of default and efficiency in cooperative banking. Journal ofInternational Financial Markets, Institutions, and Money, 26, 30-45. 10.1016/j.intfin.2013.03.003
General rightsCopyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s)and / or other copyright owners and it is a condition of accessing these publications that users recognise andabide by the legal requirements associated with these rights.
Take down policyThe University of Edinburgh has made every reasonable effort to ensure that Edinburgh Research Explorercontent complies with UK legislation. If you believe that the public display of this file breaches copyright pleasecontact [email protected] providing details, and we will remove access to the work immediately andinvestigate your claim.
Download date: 26. Aug. 2014
ProbabilityofDefaultandEfficiencyinCooperativeBanking
FrancoFiordelisia,b,DavideSalvatoreMarec*
aFacultyofEconomics,UniversityofRomeIII,Italy
bBangorBusinessSchool,BangorUniversity,U.K.
cBusinessSchool,CreditResearchCentre,TheUniversityofEdinburgh,U.K.
Abstract
Cooperative banks are small credit institutions, and they aremore likely than commercialbanks to default in periods of financial stability. Focusing on Italy (one of the largestcooperativebankingmarkets),weanalysethecontributionofefficiencytotheestimationofthe probability of default of cooperative banks. We estimate several measures of bankefficiency, and we run a discrete‐time survival model to determine whether differentmanagerial abilities play different roles in predicting bank failures. We show that higherefficiency levels (both in cost minimization and revenue and profit maximization) have apositiveandstatisticallysignificantlinkwiththeprobabilityofsurvivalofcooperativebanks.Wealsofindthatcapitaladequacyreducestheprobabilityofdefault,supportingtheviewthathighercapitalbuffersprovideadditionallossabsorbencyandreducemoralhazardproblems.
JEL‐Classification:C23,G21,G28
Keywords:Bankfailure,Smallbanks,Efficiencymeasures,Hazardmodel.
*Correspondingauthor:CreditResearchCentre,UniversityofEdinburghBusinessSchool,29BuccleuchPlace,EH89JU,Edinburgh,UK.Phone:+44(0)1316515077,Fax:+44(0)1316513197,E‐mail:[email protected]
1
1 Introduction
CooperativebanksplayakeyroleintheEuropeanbankingindustry.In2010,cooperative
bankswere a driving force for socially committed business at the local level through their
3,900memberbanks,65,000branches,morethan770,000employees,50millionmembers,
and 180 million clients (European Association of Co‐operative Banks, 2011). Overall,
cooperative banks account for approximately one fifth of the European banking system
(market shares of deposits and credits are 21% and 19%, respectively). Various studies
(Groeneveld and de Vries, 2009; Cihák and Hesse, 2007; Groeneveld, 2012) suggest that
cooperativebanksare,onaverage,morestable thancommercialbanksbecausetheyhavea
great deal of soft information (which is hard to collect) on the creditworthiness of
members/customers, and therefore they are much less likely to make lending mistakes.
However,intimesoffinancialstability,regulatorsaremorepronetoletadistressedbankgo
intodefaultifitisasmallcooperativebank.ThisoutcomeisconsistentwiththeToo‐Big‐To‐
Fail policy (i.e., regulators avoid letting the largest and most powerful banks go out of
business in order to prevent panic in financial markets) and the Too‐Important‐To‐Fail
argument (i.e., regulators avoid letting the most well‐known and systematically important
banks go out of business in order to prevent the risk that many banks fail together). For
instance,thedefaultrateofItaliancooperativebankswasalmostfourtimeshigherthanthat
ofcommercialbanksintheperiodbeforethefinancialcrisis(1997–2006).Specifically,there
were 44 default cases among cooperative banks (default rate 1.04%) and 8 among
commercialbanks(defaultrate0.28%).
Ourpaperanalysesthedeterminantsoftheprobabilityofsurvivalofcooperativebanks.
What drives thedefault of banks? Is efficiency adeterminant in thedefault of banks?Does
2
managerialskillplayaroleinthefinancialdistressofsmallcreditinstitutions?Thepurpose
ofthispaperistoempiricallyaddressthesequestionsinregardtocooperativebanks.Because
there is evidence thathigher efficiency reducesbank risk‐taking (e.g.,Berger andDeYoung,
1997;Fiordelisietal.,2011;CihákandSchaeck,2013),wepositthatalowerexposuretorisky
assets increases the survival time of a bank. Consequently,we argue that higher efficiency
favoursbanksoundness.Surprisingly,thereislimitedavailableempiricalevidencesupporting
this expectation.As such,weposit thatbank survival is related to themanagerial ability to
save costs (cost efficiency), maximize revenues (revenue efficiency), and maximize profits
(operatingandinterestefficiency).
Wehavethreemainresults.First,weshowthatmoreefficientbanks(efficienteither in
costsavingorinrevenuemaximization)haveahigherprobabilityofsurvival.Second,wefind
thatwhenabank’smanagerial ability tominimize costs andmaximize revenuesare jointly
considered(e.g.,efficiencyingeneratinginterestincome),moreskilfulmanagementincreases
thebank’ssurvivaltime.Third,wefindevidencetosupporttheviewthattraditionalfinancial
ratiosareconsistentpredictorsofbankdistress.Inthisregard,weshowthatcapitalisakey
determinantofbanksoundness.
Weanalysealargesamplewithmorethan4,200observations‐alltheItaliancooperative
banksbetween1997and2009.Weestimatetheprobabilityofdefaultbyrunningadiscrete‐
timesurvivalmodelthatrelatesachangeinthehazardratetoanabsolutechangeinagiven
covariate,allelsebeingequal.WefocusonItalybecausethiscase isparticularly interesting
forvariousreasons.First,cooperativebanksplayacrucialroleintheItalianbankingmarket‐
3
Italian cooperative banks1have approximately 36,000 employees, 6.7 million clients, 1.1
millionmembers2,and7.3%ofthemarketshareofdeposits3.Second,theItaliancooperative
banking sector is the fourth largest inEuropeafterGermany,France, andAustria (in 2010,
6.7%4ofthetotalassetsundermanagementintheEU27cooperativebankingsector).Finally,
Italypresents auseful laboratory setting to analyse the impact of theeconomic, social, and
demographic conditions of local areas on bank efficiency. The Italian regions display very
differentconditionsthatmustbeconsideredtoaccuratelyestimatetheprobabilityoffailure.
Theremainderofthepaperisorganizedasfollows:Section2brieflyreviewstherelevant
literature. In Section 3, we formulate our research hypotheses and describe the data
employedintheempiricalanalysis.Section4summarizesthemethodology.Section5reports
the results of the analysis. Section 6 addresses the predictive accuracy of the model, and
Section7concludesandoffersfinalremarks.
2 Literaturereview
Our paper joins two separate streams of the economic literature. The first stream
concernsefficiencyestimationwiththeaimofcomparingcooperativeandcommercialbanks.
One group of studies compares cooperative and commercial banks by estimating, first, a
specific efficiency frontier for each typeof bank and, second, a common frontier that pools
togethercooperativeandcommercialbanks.Thesepapersprovidemixedevidenceaboutthe
1Notethat ItalianBanchePopolariarenotcovered in thepresentanalysisbecause, in termsofgovernance, theymorecloselyresemblejoint‐stockcompanies(Fonteyne,2007).2Sourceofdata:ItalianFederationofCooperativeBanks(Federcasse),estimateddataat31/12/2011.3Sourceofdata:EuropeanAssociationofCo‐operativeBanks(2011).4Sourceofdata:owncalculationusingdatafromtheEuropeanAssociationofCooperativeBanks(2011).Cooperativenetworks,suchastheDutchRabobankandtheFrenchCréditAgricole,arenotconsideredinthecalculation.
4
costandprofitefficiencyofsuchbanks.Battagliaetal.(2010)showthatsampleheterogeneity
also occurs when estimates are obtained from a single efficient frontier estimated for
cooperativebanksonly.Cooperativebankshaveastrong linkwith thegeographicalarea in
which they operate; therefore, the levels of efficiency are influenced by the social,
demographic,andeconomicconditionsofthatspecificarea.
Various papers also compare commercial and cooperative banks by focusing on issues
other than efficiency. Cihák and Hesse (2007) use individual bank data to test whether
cooperativebanksreducethestabilityofotherbanksandrespondslowlytofinancialdistress.
Contrarytothefindingsofpreviousstudies(Brunneretal.,2004;Goodhart,2004;Fonteyne,
2007),CihákandHessefindcooperativebankstobemorestablethancommercialbanks:the
lowervolatilityof thecooperativebanks’ returnsmore thanoffsets their lowerprofitability
andcapitalization.Groeneveld(2012)comparescommercialandcooperativebanks,focusing
onthemeanvaluesofsomeindicators(returnonequity,Tier1capital,andZ‐score).Overall,
the author concludes that in Europe cooperative banks are less profitable andmore stable
thancommercialbanks.
The second stream of literature addresses the estimation of bank failures and is
characterizedbytwoapproaches:micro‐andmacro‐approaches.The“micro”strandfocuses
on individual banks’ balance sheet data, possibly integrated with financial market data, to
predict bank failures. This approach stems from the seminal papers of Altman (1968) and
Beaver(1966),whouseaccountingdatatodiscriminatebetweensoundandtroubled firms.
Sincethen,manystudieshaveassessed theabilityof financial ratios topredict the financial
health of bank operations (see, amongmany others, Meyer and Pifer, 1970; Sinkey, 1975;
SantomeroandVisno,1977;West,1985;Estrellaetal.,2000).Variouspapershavealsotested
5
thesuperiorityofonespecificassessmenttechniqueoveranother(Martin,1977;Espahbodi,
1991;Shumway,2001;Glennonetal.,2002;Boyaciogluetal.,2009).Recently,Demyanykand
Hasan (2010) reviewed this extensive literature and showed that the combination of
operational research techniques with statistical methods substantially improves the
predictionofbankfailures.Otherrecentworkshavefocusedonthesubprimemortgagecrisis
andthesubsequentbankingfailures(DavisandKarim,2008a;Jinetal.,2011;ColeandWhite,
2012).
The “macro” approach investigates banking crises by focusing on macroeconomic
determinants(Demirguc‐KuntandDetragiache,1998;González‐Hermosillo,1999;Davisand
Karim,2008b).Thesestudiestypicallyanalysealargesampleofcountriestodeterminewhich
macroeconomic factors signalabankingcrisis inadvance.For instance,Demirguc‐Kuntand
Detragiache (1998) argue that GDP growth, excessively high real interest rates, and high
inflationsignificantlyincreasethelikelihoodofsystemicbankingcrises.Otherrecentstudies
(DeYoung,2003;Arena,2008;MännasooandMayes,2009;Mare,2012)haveusedboththe
micro andmacroperspectives, highlighting that combiningdifferent sourcesof information
increasestheaccuracyofpredictionsofbankfinancialdistress.
Variouspapershaveestimatedtheprobabilityofbankfailureorsurvival.Laneetal.(1986)
pioneered the field by using duration analysis. The authors were the first to use the Cox
proportional hazards model to predict US commercial bank failures. Estrella et al. (2000)
estimate the likelihood of failure using cross‐sectional logit regressions and then analyse
time‐dependency in the conditional probability of failure through a proportional hazards
model. Wheelock and Wilson (2000) use a competing‐risks hazard model to identify
characteristicsleadingtoeitherfailureoracquisition.TheauthorsdemonstratethatUSbanks
6
thathavelowcapitalization,highleverage,lowliquidity,poor‐qualityloanportfoliosandlow
earningsaremorepronetofailure.Moreover,theauthorssuggestthatinefficiency,measured
throughmanagementquality,increasesthelikelihoodofbankfailure.
Shumway (2001) develops a discrete‐time hazard model to determine probability
estimates for corporationsateachpoint in time.Theauthorargues thathazardmodels are
theoreticallypreferabletosingle‐periodclassificationmodels(staticmodels)becausehazard
modelsconsiderthatfirmschangeoverandthroughtime.Hence,theresultantprobabilitiesof
defaultconsistentlyapproximatethetrueprobabilitiesoffailure.Arena(2008)performsboth
cross‐sectional logit estimation and survival time analysis to prove that bank‐level
fundamentals, the banking system, and macroeconomic variables significantly affect the
likelihood of bank failures in the case of banking crises in East Asia and Latin America. In
addition,theauthorsuggeststhatsystemicmacroeconomicandliquidityshocksdestabilized
notonlythebanksthatwerealreadyweakbeforethecrises,butalsothosebanksthatwere
relatively stronger ex‐ante. This result suggests that negative effects triggered by systemic
crisescanalsoaffectsoundbanks.MännasooandMayes(2009)useadiscretecomplementary
log‐log model to link banks’ hazard rates to macroeconomic, structural, and bank‐specific
factors.Thestudysuggeststhatchanges inbankearnings,efficiency(measuredbythecost‐
incomeratio),andtherelativesizeofthecreditportfolioarenotearlywarningindicators.
Although various studies (e.g., Berger andDeYoung, 1997; Fiordelisi et al., 2011; Cihák
andSchaeck,2013)haveanalysed therelationshipbetweenbankefficiencyandrisk‐taking,
nostudieshavedirectlyrelateddifferentmanagerialskillstotheoccurrenceofbankfailure.
This is surprising because efficiency is one of the key factors behind bank performance
(Fiordelisi,2007;FiordelisiandMolyneux,2010)anditguaranteesbanksurvivalovertime.
7
The recent crises of credit institutions have shown the importance of assessing how well
managementcontributestobanksurvival(intermsofminimizingcosts,maximizingrevenues,
or maximizing various measures of profits). The accurate prediction of bank survival is
fundamentalforpractitioners,investors,academics,andregulators.Whilstthisistrueforall
banks, it is critical for cooperative banks because their failure has been historically more
likelythanthatofcommercialbanksandmightgeneratehighersocialcostsatthelocallevel.
3 Researchhypothesesanddata
Inthissection,weformulateourresearchhypotheses,andthenweoutlinethedatausedfor
theestimation.Thekeyquestionaddressedinthepaperistoverifywhetherthemanagerial
ability of a bank to reduce costs and/or increase revenues plays a role in avoiding bank
default. Our approach entails an empirical analysis of the link between various efficiency
measures and the probability of default of cooperative banks. Moreover, we estimate four
different efficiencymeasures using stochastic frontier analysis and test our results using a
balance‐sheet measure of operating efficiency (i.e., cost‐income ratio, as in Mannasoo and
Mayes,2009).Wespecifythreehypothesesthatfocusonvariousefficiencyconcepts.
First,wepositthat ifabankismorecostefficientthanitscompetitors, it is lesslikelyto
default.Theunderlyingassumptionisthatifbankmanagershavesuperiorskillsinmanaging
costsandinducinghighercostefficiency,thiswillhelpthebanksurviveduringindividualor
sector distress. We call this assumption “cost‐management excellence”. A competing
hypothesis is“cost‐skimping”(originallypositedbyBergerandDeYoung,1997,andrecently
testedbyFiordelisietal.,2011):ifabankismorecostefficientthanitscompetitors,thereisa
8
higher likelihoodofdefault.Theunderlying idea is thatcost‐efficientbanksprobablydevote
fewerresourcestocreditscreeningandmonitoring,whichimpliesatrade‐offbetweenshort‐
termcostefficiencyandfuturerisk‐taking.
HypothesisI(H1):ifabankismorecostefficientthanitscompetitors,ithasalowerprobability
ofdefault(“cost‐managementexcellence”hypothesis).
AlternativeHypothesis I (H1A): if a bank ismore cost efficient than its competitors, it has an
increasedprobabilityofdefault(“cost‐skimping”hypothesis).
Second,we argue that if a bank ismore revenue efficient than its competitors, it has a
lowerlikelihoodofdefault.Theunderlyingassumptionisthatifbankmanagershavesuperior
skills inmanagingrevenueandstimulatinghigherrevenueefficiency,thiswillhelpthebank
survive in cases of individual or sector distress. We name this assumption “revenue‐
managementexcellence”.Acompetinghypothesis is “short‐termrevenue”: if abank ismore
revenueefficientthanitscompetitors,ithasahigherlikelihoodofdefault.Theunderlyingidea
isthatrevenue‐efficientbanksprobablyhavealowerqualityloanportfolio,andcustomersare
therefore willing to pay higher interest rates. This implies a trade‐off between short‐term
revenueefficiencyandfuturebank‐soundness.
Hypothesis II (H2): if a bank is more revenue efficient than its competitors, it has a lower
probabilityofdefault(“revenue‐managementexcellence”hypothesis).
AlternativeHypothesisII(H2A):ifabankismorerevenueefficientthanitscompetitors,ithasan
increasedprobabilityofdefault(“short‐termrevenue”hypothesis).
9
Third,weassumethattheprobabilityofdefaultisnotonlyrelatedtothebank’sabilityto
manageeither itscostsorrevenues,butalso to itsability toconcurrentlymanagecostsand
revenues to achieve higher profits. Specifically, a cost‐efficient bank could be disastrous at
managing revenue or the reverse. We posit that if a bank is more profit efficient than its
competitors, it has a lower likelihood of default. The underlying assumption is that higher
profitefficiencyimpliesthatbankmanagershavesuperiorskills inmanagingbothcostsand
revenues,whichwillhelpthebanksurviveintimesofindividualorsectordistress.Wecallthis
assumption“managementexcellence”.SimilartohypothesisH2A,analternativeandcompeting
hypothesisisthatof“short‐termprofits”:ifabankismoreprofitefficientthanitscompetitors,
itwillhaveahigher likelihoodofdefault (due toa lowerquality loanportfolio), implying a
trade‐offbetweenshort‐termprofitsandthefuturesoundnessofthebank.Wetestthesetwo
competingassumptionsusingtwodifferentprofitmeasures:operatingprofits(whichincludes
alloperatingcostsandrevenues)andtheinterestmargin(whichincludesonlyinterestcosts
andrevenues).
Hypothesis III (H3): if a bank is more profit efficient than its competitors, it has a lower
probabilityofdefault(“management‐excellence”hypothesis).
AlternativeHypothesisIII(H3A): ifabankismoreprofitefficientthanitscompetitors,ithasan
increasedprobabilityofdefault(“short‐termprofits”hypothesis).
Our sample comprises more than 4,200 observations and includes the financial
statements of almost all the Italian cooperative banks. The data cover the period between
10
1997and2009,althoughwedonotincludetheyears2007and2008becausetheycontainno
defaultevents.EachbankcontributesTtrowsofdata, correspondingtothenumberof time
periodstinwhichitwasatriskoffailure.
Thedataset for theexplanatoryvariables iscomprehensive, combiningbank‐leveldata,
geographical information, and efficiency measures. Market information is not considered
becausecooperativebanksarenotpubliclytradedandhaveverylittlemarketactivity.
We collect data from various sources. Data on distressed banks are retrieved from the
ItalianCentralBank(BankofItaly);accountingdataareobtainedfromtheItalianAssociation
of CooperativeBanks (Federcasse); andwe garner local geographical information from the
Italian National Institute of Statistics (ISTAT). The bank‐level data, a potential leading
indicator of failure, are drawn from the banks’ financial statements. Data are publicly
available formostkey items ‐ liquidity,balance sheets,profitsand losses, off‐balance sheet
items, large depositors, and large exposures. Themajor constraints are information on the
sectorpatternoflending(includingexposuretothepropertysector)andtheinterestrateson
liabilitiesandassets.
4.Methodology
FollowingMännasoo andMayes (2009), we estimate a discrete‐time survival model to
determinetheprobabilityoffailureateachpointintime.Werunatwo‐stageanalysis.First,a
complementary log‐log model (cloglog) is estimated using various efficiency measures,
macroeconomic information, and bank risk‐taking variables. Second, we test out‐of‐sample
theaccuracyofthepredictionofthemodelandtherobustnessoftheresults.
11
4.1Thehazardmodel
We estimate the survival model in discrete time because our data set only provides
observationsdiscretely.Wefocusonasingle‐statemodel,andweassumethatwehavesingle‐
spelldataforeachbank.Ourmodel implicitlyassumesthatallrelevantdifferencesbetween
banks can be summarized by the observed explanatory variables. We also assume that
bankruptcyonlyoccursatdiscretepointsintime(t=1,2,3,...,n).Moreover,eachbankeither
fails during the sample period or survives. If banks merge or are liquidated or if the
identification variable (Abi) is not available for the whole observation window, they are
omittedfromthesample.Thus,weconsiderexitsfromasinglestate(soundness)toasingle
destination(failure).
The random variable T denotes the time to exit from the sample (failure) and t the
realizationthereof.Thediscrete‐timedurationmodelimpliesthatweobservetheprobability
ofsurvivalofcooperativebanksatdistinctpointsintime.Becausethesampledatarefertoan
observationwindowof tenyears, thesurvival timedata set is right‐censored,meaning that
weobservethestartdateofthespell(year1996)butnotthetotallengthoftransitionoutof
the current state (fromsoundness to failure). It is alsoassumed that theprocess that gives
risetocensoringisindependentfromthesurvival‐timeprocess.Theprobabilityofexitwithin
thejthintervalisexpressedasfollows:
(1)
wherea1,a2,…..,akaretheintervalboundarydates(years);F(aj)isthecumulativedistribution
functionofTatdurationtime j(failurefunction);andS(aj) isthesurvivalfunctionattime j.
Thediscretehazardrateistheconditionalprobabilityofexitintheinterval(aj‐1,aj]definedas:
1 1 1Pr( ) ( ) ( ) ( ) ( )j j j j j ja T a F a F a S a S a
12
(2)
Thediscrete‐timesurvivorfunctionistheproductofprobabilitiesofnotexperiencingthe
eventineachoftheintervalsuptoandincludingthecurrentone.Writtenintermsofinterval
hazardrates,itcorrespondsto:
(3)
Weallowthehazardratetovarybetweenbanksdependingontheircharacteristics,and
we summarize this information in a vector of variables. Time‐varying covariates offer an
opportunitytodynamicallyexaminetherelationshipbetweenthedistressprobabilityandthe
changing conditionsunderwhich thedistress takesplace.Thehazard rate and the selected
characteristicsarelinkedthroughanindexfunction.FollowingMännasooandMayes(2009),
weemployacomplementarylog‐logmodel(cloglog):
( , ) 1 exp[ exp( ' ' ' )]j j j j jh j X EF CAL ENV (4)
where contains time‐varying covariates; , , and are the vectors of coefficients;
denotes efficiency measures; jCAL captures bank‐level fundamentals;
represents environmental variables; and is the log of the difference between the
integrated baseline hazard evaluated at the end of the interval and at the beginning of the
interval(aj‐1;aj].Eachregressioncoefficientsummarizestheeffectonthehazardofabsolute
changesinthecorrespondingcovariates.Thecoefficientsdonotvarywithsurvivaltime.
1 11
( )Pr( | ) 1
( )j
j j jj
S aa T a T a
S a
1 2 11
( ) (1 ) (1 ) ..... (1 ) (1 ) (1 )j
j j kk
S j h h h h h
jX ' ' '
jEF jENV
j
13
4.2Variables
4.2.1Eventoffailure
Following previous studies (Arena, 2008; Männasoo and Mayes, 2009, among others),
bank default is defined as the occurrence of public intervention to solve a critical distress
situation.Wemodelbankfailuresusingacategoricalvariablethatequals1ifbankifailedat
timetandequals0otherwise.FollowingItalianlaw,wedefineabankasbeingindefaultifit
underwent either of the two following events between January 1, 1997 and December 31,
2006: a) it entered extraordinary administration (e.g., conservatorship), or b) it entered
liquidation.Eachofthesegovernmentinterventionsobjectivelyshowsthatabankisunableto
continueitsoperations.
WecollectdataondistressedbanksfromtheBankofItaly.Overall,therewere44casesof
governmentintervention(eitherextraordinaryadministrationsorliquidations)intheperiod
analysed,asshown inTable1.Accountingdata forall cooperativebanksareobtained from
theItalianFederationofCooperativeBanks(Federcasse).
<InserthereTable1>
4.2.2Explanatoryvariables
Thesetofpotentialexplanatoryvariablesischoseninordertoexplaintheprobabilityof
failure as a consequence of the better management of bank operations quantified by the
efficiencymeasures.
First,weestimatecostefficiencybyusingBatteseandCoelli’s(1995)stochasticfrontiermodel,
asdetailed inAppendix1.Wealso compute the cost‐to‐income ratio as adirect test tomeasure
14
efficiency in generating operating income.We then include other variables that are likely to
contribute to the survival of a cooperative bank.We take into consideration the economic
conditionsofthelocalareaenvironment(measuredbytheannualgrowthrateinItalianGDP
percapitaandtheemploymentgrowthintheregionalworkforce)becausethesearelikelyto
affectthesurvivalofcooperativebanks(Mare,2012).Asecondgroupofvariablesfocuseson
bank‐specificfactors.Weselecttheindicatorsfollowingthewell‐knownCAMELframework5:
specifically, we measure Capital Adequacy as the ratio between the capital in excess of
regulatory requirements and the minimum capital requirement; we estimate bank Asset
Quality using the ratio of annual loan loss provisions to total loans and we quantify the
LiquidityRiskastheratioofbankdepositstocustomerdeposits.ThetworemainingCAMEL
categories(i.e.,ManagementQualityandEarnings)areexplicitlyincludedintheestimationof
the efficiency measures. We also control for a bank’s asset size and for a bank’s credit
orientation.Table2describesthevariablesemployedintheanalysisandreportstheexpected
signsoftheestimatedparameters.
<InserthereTable2>
The descriptive statistics for the explanatory variables are reported in Table 3. We
separateallbanks into twocategories: failedandnon‐failedbanks.T‐testsarecomputedto
detectstatisticallysignificantdifferencesinunivariatecomparisons.
5CAMEL is the acronym referring to the following five factors introducedbyUS regulators inNovember1979: “C” stands forCapital Adequacy, “A” for AssetQuality, “M” forManagementQuality, “E” for Earnings, and “L” for Liquidity. In 1996, CAMELevolvedintoCAMELS,where“S”istheSensitivitytoMarketRisk.Cooperativebanksengageinverylittlemarketactivity;thus,theCAMELframeworkismoreappropriatetoanalyzetheirbanksoundness.
15
<InserthereTable3>
5.Results
WereporttheresultsfromthehazardfunctioninTable4.FollowingMännasooandMayes
(2009), the explanatoryvariables are laggedbyoneyear. Thediscrete‐time survivalmodel
relates a change in the hazard rate due to an absolute change in a given regressor, ceteris
paribus.Thecharacteristicsoftheeconomicenvironment(atboththelocalandnationallevel)
andbank‐specific indicators are introduced to the analysis to control for factors that could
influencethelinkbetweenbankefficiencyandtheprobabilityofdefault.Thecontrolvariables
are standardized tomake it easier to compare the individual contributions of the different
factorstothesurvivalofcooperativebanks.
<InserthereTable4>
To test our research hypotheses, we run our cloglog model (Equation 4) using five
differentspecificationstolinkefficiencyandbankfailure(respectively,oneforeachefficiency
measureandoneusingthecost‐to‐incomeratio).
Our results support the cost‐management hypothesis (H1): we find that higher cost
efficiencyengendersahigherprobabilityofsurvival.Theestimatedcost‐efficiencycoefficient
(inEstimation1ofTable4)isnegativeandstatisticallysignificantatthe1%confidencelevel.
Asaresult,wecanrejectthecost‐skimpinghypothesis(H1A).
16
In regard toabank’s ability tomaximize its revenues, our results support the revenue‐
managementhypothesis (H2), suggesting thatbanks that are able to extractmore revenues
fromservicesprovidedtocustomersandachievehigherreturnsfromtheirinvestmentshave
ahigherprobabilityofsurvival.Theestimatedrevenue‐efficiencycoefficient(inEstimation2
ofTable4)isnegativeandstatisticallysignificantatthe10%confidencelevel.Consequently,
wecanrejecttheshort‐termrevenuehypothesis(H2A).
Turning to profits, our findings support the management‐excellence hypothesis (H3)
becausemoreprofit–efficientbanksarefoundtohaveahigherprobabilityofsurvival.Weuse
two measures of profitability: operating income and interest margin. In both cases, the
coefficient(inEstimations3and4ofTable4)isnegativeandstatisticallysignificantatthe1%
confidencelevel.Consequently,wecanalsorejecttheshort‐termprofitshypothesis(H3A).In
Estimation5inTable4,thecost‐to‐incomeratioispositivelyrelatedtoabank’sriskoffailure,
as banks that are less able to contain their operating costs are more likely to be poorly
managedandshowahigherhazardrate.
Wefindthat theCAMELindicatorsarestatisticallysignificant(atthe10%levelor less)
and strongly related to the hazard rate. Specifically, higher capital levels reduce the
probability of default. This supports the view that higher capitalization provides additional
lossabsorbencyandreducesabank’smoralhazard(Laneetal.,1986;Fiordelisietal.,2011;
HaqandHeaney,2012,amongmanyothers).Assuch,strengtheningthecapitalrequirement
in the cooperative banking sector, as proposed in the Basel III agreement6, may help to
preventbankdistress.
6The Basel III framework comprehends a set of documents issued by The Basel Committee on Banking Supervision (firstpublicationinDecember2010).Theaimistoreducetheimpactofbankingcrisesinthefuture.TheBaselIIIagreementfollows
17
Wealsonote thatassetquality (measuredby the loan lossprovision ratio) ispositively
related(atthe1%level)totheprobabilityofdefault.Assuch,inthecaseofadecreaseinasset
quality(i.e.,theloanlossprovisionratioincreases),theprobabilityofdefaultforcooperative
banks increases. In linewithMännasooandMayes(2009),wefindthatthe liquidityratio is
positively and statistically significantly related to the probability of default. This result
suggests that when cooperative banks rely too heavily on the interbank market, they are
exposed to the sudden freezing of funds. This can lead banks to default, especially during
periods of systemic financial distress when funding is critical. Size is negatively related to
bankfailure,aslargercooperativebankshavelessconcentratedportfoliosandcandiversify
lendingtowardsdifferentindustrialsectors.
6.TestingtheModel’sPredictiveAbility
Wetest thepredictiveaccuracyofourmodelboth in‐sampleandbyusingholdoutdata.
Theanalysisofthegoodness‐of‐fitenablesustodeterminehowwelltheeconometricmodel
fits the observed phenomena7. We again run the five models in Table 4 using only the
variables thatwere statistically significant at the 10% level; therefore,we do not consider
employment growth in our computations. In Model 5 of Table 4, we include the cost‐to‐
the publication of the first accord on capital measurement and capital standards (Basel I, July 1988) and a second, morecomprehensiveframework(BaselII,June2004).7As a further test, we run a robustness check by comparing the impact of efficiency on the probability of bank default incooperativeandcommercialbanking.We retrieveaccountingdataon Italiancommercialbanks from theBankscopedatabase.Becausedataondefaultsforcommercialbanksareinsufficienttoestimateamodel(8defaultsovertheperiod1997–2006),wefocuson2009 (6defaults). All estimated coefficients for efficiencymeasures inboth commercial and cooperativebanks havesignsthatareconsistentwiththoseestimatedinColumn5ofTable4.However,theoverallstatisticalpoweroftheestimationispoor,mainlyduetothelownumberofdefaults.Resultsfromthistestareavailableuponrequest.
18
incomeratiobecauseitisstatisticallysignificantatthe10%levelafterremovingtheliquidity
variable,employmentgrowth,andGDPgrowth.
Table5reportsthein‐samplecheckoftheaccuracyofthepredictions.InPanelAofTable
5, we report six indicators: Sensitivity, Specificity, Overall Predictive Power, ROC Area,
Accuracy Ratio, and Brier Score. Efficiency measures, estimated through the stochastic
frontieranalysis,increasethepredictiveaccuracyofthemodelwhencomparedwithModel5,
inwhichthecost‐to‐incomeratioproxiesformanagerialabilitytoreducecosts(inputs)and
toproducerevenues(output).
Sensitivity quantifies the proportion of banks in default that are correctly identified as
such; Specificity measures the proportion of safe banks (e.g., sound) that are correctly
identified.ThesetwoindicatorsarecloselyrelatedtotheconceptsofTypeIandTypeIIerrors:
all estimated models have an in‐sample Sensitivity higher than 54% (70%, if we do not
consider Estimation5,whichhas the lowestperformance) andSpecificityhigher than75%
(80%, ifwe omit Estimation 5,which has the lowest performance), values that are largely
superior to those of a naïve model (i.e., 50%). The Overall Predictive Power is the ratio
betweenthesumofallsafeanddefaultedbanksaccuratelyclassifiedandthetotalnumberof
banks.Allestimationshavein‐sampleOverallPredictivePowerhigherthan74%(80%,ifwe
omit Specification 5,which has the lowest performance, and 83% in the best specification,
which includesoperating cost efficiency), a figure that is largely superior to that of anaïve
model(i.e.,50%).
We compute the ROCCurve, which measures the impact of changes in the probability
threshold,i.e.,thedecisionpointusedbythemodelforclassification.TheareaundertheROC
Curvemeasuresthediscriminatingabilityofabinaryclassificationmodel:thelargerthisarea,
19
thehigher the likelihoodthatanactualdefault casewillbeassignedahigherprobabilityof
being in default than an actual sound case. All estimations have an in‐sample Overall
Predictive Power higher than 76% (83% if we omit Specification 5, which has the lowest
performance, and 88% in the best specification, which includes operating cost efficiency),
whichislargelysuperiortothatofanaïvemodel(i.e.,50%).
WealsocomputetheBrierScore,whichevaluatesthequalityoftheforecastsasfollows:
(5)
where is theestimateddefaultprobabilityof thebanks(from1ton),and is theactual
outcomeof theeventofdefault (equal to1 ifobligor idefaultsand0otherwise).TheBrier
Scoremustthereforealwaysbebetweenzeroandone.TheclosertheBrierScoreistozero,
thebettertheforecastofdefaultprobabilities.Allestimatedmodelshaveanin‐sampleBrier
Scoreclosetozero.
Overall, the specifications (especially Specification 1, which includes cost efficiency)
provide sound estimates in line with (or better than) the overall performance of previous
studies(Shumway,2001;MännasooandMayes,2009;Arena,2008).
<InserthereTable5>
We also run an out‐of‐sample test by estimating hazard rates using the estimated
coefficients and data from 2009. This enables us to validate our results by tackling the
problemofsample‐specificestimation.Themeaningof thegoodness‐of‐fit test is limited,as
1
1( )
n
i ii
B pn
ip i
20
we observe only six default cases in 2009 (and no defaults in 2007 and 2008). Not
surprisingly,we find that thepredictive accuracyof themodel (Table6) is lower than that
obtainedin‐sample(Table5),butthepredictivepowerofourmodelisstillhigh.Theoverall
predictivepowerishigherthan74%forall themodels,andtheBrierScorecontinuestobe
verylow(lowestvalue0.014).
<InserthereTable6>
7.Conclusions
Ourpaperanalysesthecontributionofefficiencytotheprobabilityofbankfailureamong
cooperative banks. Cooperative banks play a key role in the European banking industry.
Despite their importance, these banks experience financial distress more frequently than
commercialbanksduringperiodsoffinancialstability.ThedefaultrateofItaliancooperative
banks was almost four times that of commercial banks in the period before the recent
financial crisis (1997–2006). Specifically, there were 44 default cases among cooperative
banks(defaultrate1.04%)and8amongcommercialbanks(defaultrate0.28%).
Ourpaperanalysesthedeterminantsoftheprobabilityofsurvivalofcooperativebanksby
focusingon the role of efficiency.Our results indicate that lower risk is related to a higher
survival time forcooperativebanks.These findingscontribute to theexisting literaturethat
investigates the direct link between efficiency and risk‐taking (Chortareas et al., 2011;
Chronopoulos et al., 2011; Fiordelisi et al., 2011) by showing that prudent and skilful
managerialabilitiesincreasethesurvivaltimeofcooperativebanks.Tosupportourview,we
testthreeresearchhypothesesthatfocusonthecontributionofefficiencytotheprobabilityof
21
bankfailure.Namely,wepositthatbanksurvival isrelatedtothemanagerialabilitytosave
costs(H1),maximizerevenues(H2),andmaximizeprofits(H3).
We study a large sample of Italian cooperative banks. We show that all efficiency
measureshavelowerhazardrates,meaningthatanincreaseinthesevariablesincreasesthe
survivaltimeofbanks.Specifically,wefindthatmoreefficientbanks(eitherintermsofcost
savingorrevenueandprofitmaximization)showahigherprobabilityofsurvival,supporting
thecost,revenue,andexcellent‐managementhypotheses.
Our findings are of interest for policymakers and supervisors. Recent developments in
banking regulations stem from the idea that efficiency is a key element in assessing the
relationship between bank risk and capital levels. There is a parallel between Basel II
prescriptions regarding internal control processes andhigher efficiency gains becauseboth
contributetoanincreaseintheresilienceofbanks.Similarly,thenewcorporategovernance
directivesforbankssupportthecost,revenue,andexcellent‐managementhypotheses,inline
with the results of the paper. Moreover, we show that higher capital levels reduce the
probabilityofdefault:thissupportstheviewthathighercapitallevelsprovideadditionalloss
absorbency and reduce a bank’smoral hazard. As such, in the cooperative banking sector,
strengthening the capital requirement as proposed in the Basel III agreementmay help to
preventbankdistress.
Acknowledgements
The authorswish to thankAlessandro Carretta, OrnellaRicci, IftekarHasan, PhilMolyneux
andMariaPaulaGerardinowhokindlygavetheircommentsonearlierversionsofthispaper
22
aswellasFedercasseforprovidingthedatanecessaryfortheestimationofthemodelandfor
thehelpfulcollaboration,especially inthepersonsofAlessandraAppennini, JuanLopezand
RobertoDi Salvo. The authors are also grateful to the conference participants at the EURO
Working Group on Efficiency & Productivity analysis (EWG‐EPA, 2010) and the Infinity
ConferenceonInternationalFinance(2011).Allremainingerrorsandomissionsrestwiththe
authors.
23
References
Altman, E.I., 1968. Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy.JournalofFinance23(4),589‐609.
Arena, M., 2008. Bank failures and bank fundamentals: A comparative analysis of Latin
America and East Asia during the nineties using bank‐level data. Journal of Banking &
Finance32(2),299‐310.
Battaglia,F.,Farina,V.,Fiordelisi,F.,Ricci,O.,2010.Theefficiencyofcooperativebanks:the
impactofenvironmentaleconomicconditions.AppliedFinancialEconomics20(17),1363‐
1376.
Battese,G.E.,Coelli,T.J.,1995.Amodelfortechnicalinefficiencyeffectsinastochasticfrontier
productionfunctionforpaneldata.EmpiricalEconomics20(2),325–332.
Beaver,W.H.,1966.Financialratiosaspredictorsoffailures.JournalofAccountingResearch4,
EmpiricalResearchinAccounting:SelectedStudies1966,71‐111.
Berger, A.N., DeYoung, R., 1997. Problem loans and cost efficiency in commercial banking.
JournalofBanking&Finance21(6),849–870.
Boyacioglu,M.A.,Kara,Y.,Baykan,O.K.,2009.Predictingbankfinancial failuresusingneural
networks, support vectormachines andmultivariate statisticalmethods:A comparative
analysis in the sample of savings deposit insurance fund (SDIF) transferred banks in
Turkey.ExpertSystemswithApplications36(2),3355‐3366.
Brunner,A.,Decressin,J.,Hardy,D.,Kudela,B.,2004.Germany’sthree‐pillarbankingsystem–
cross‐countryperspectivesinEurope.IMFOccasionalPaper233.InternationalMonetary
Fund:Washington,D.C.
24
Chortareas, G.E., Girardone, C., Ventouri, A., 2011. Financial frictions, efficiency and risk:
evidence from theEuroZone. Journal ofBusiness, Finance andAccounting38(1‐2),259‐
287.
Chronopoulos, D.K., Girardone, C., Nankervis, J.C., 2011. Are there any cost and profit
efficiencygainsinfinancialconglomeration?Evidencefromaccessioncountries.European
JournalofFinance17(8),603‐621.
Cihák,M.,Hesse,H.,2007.Cooperativebanksand financial stability. IMFWorkingPapers2.
InternationalMonetaryFund:Washington,D.C.
Cihák, M., Schaeck, K., 2013. Competition, efficiency, and stability in banking. Financial
Management,forthcoming.
Cole,R.,White,L.,2012.Déjàvualloveragain:thecausesofU.S.commercialbankfailuresthis
timearound.JournalofFinancialServicesResearch42(1‐2),5‐29.
Davis,E.P.,Karim,D.,2008a.Couldearlywarningsystemshavehelpedtopredictsub‐prime
crisis?NationalInstituteEconomicReview206(1),35–47.
Davis,E.P.,Karim,D.,2008b.Comparingearlywarningsystemsforbankingcrises.Journalof
FinancialStability4(2),89–120.
Demirguc‐Kunt,A.,Detragiache, E., 1998.Thedeterminants of banking crises in developing
and developed countries. IMF Staff papers 45(1). International Monetary Fund:
Washington,D.C.
Demyanyk, Y., Hasan, I., 2010. Financial crises and bank failures: a review of prediction
methods.Omega38(5),315‐324.
DeYoung,R.,2003.DeNovobankexit.JournalofMoney,Credit,andBanking35(5),711‐728.
25
Espahbodi, P., 1991. Identification of problem banks and binary choice models. Journal of
Banking&Finance15(1),53‐71.
Estrella,A.,Park,S.,Peristiani,S.,2000.Capitalratiosaspredictorsofbankfailure.Economic
PolicyReview6(2),33‐52.
European Association of Co‐operative Banks, 2011. Annual report for 2010.
http://www.eurocoopbanks.coop.
Fiordelisi, F., 2007. Shareholder value efficiency in banking. Journal of Banking & Finance
31(7),2151–2171.
Fiordelisi,F.,Molyneux,P.,2010.Totalfactorproductivityandshareholderreturninbanking.
Omega38(5),241–253.
Fiordelisi,F.,Marques‐Ibanez,D.,Molyneux,P.,2011.EfficiencyandriskinEuropeanbanking.
JournalofBanking&Finance35(5),1315‐1326.
Fonteyne,W., 2007. Cooperative banks in Europe: policy issues. IMFWorking Paper (159).
InternationalMonetaryFund:Washington,D.C.
Girardone, C., Molyneux, P., Gardener, E.P.M., 2004. Analysing the determinants of bank
efficiency:thecaseofItalianbanks.AppliedEconomics36(3),215‐227.
GlennonD.,Kolari,J.,Shin,H.,Caputo,M.,2002.PredictinglargeUScommercialbankfailures.
JournalofEconomicsandBusiness54(4),361‐387.
Goodhart,C.A.E.,2004.Somenewdirectionsforfinancialstability.ThePerJacobssonLecture,
BankforInternationalSettlements:Basel.
González‐Hermosillo, B. 1999.Determinants of ex‐ante banking system distress: a macro‐
micro empirical exploration of some recent episodes. IMF Working Paper (33).
InternationalMonetaryFund:Washington,D.C.
26
Groeneveld, J.M., 2012.The cooperative bankingmodel: performance and opportunities. In:
Balling,M.,FrankLierman,VandenSpiegel,F.,Ayadi,R.,andLlewellyn,D.T.(Eds.),SUERF
Study2012/2.Larcier:Bruxelles,101‐129.
Groeneveld,J.M.,deVries,B.,2009.Europeancooperativebanks:Firstlessonsofthesubprime
crisis.InternationalJournalofCooperativeManagement4(2),8‐21.
Haq,M.,Heaney,R.,2012.FactorsdeterminingEuropeanbankrisk. Journalof International
FinancialMarkets,Institutions&Money22(4),696–718
Hughes, J.,Mester,L.,2010.Efficiency inbanking: theory,practice,andevidence. In:Berger,
A.N., Molyneux, P., Wilson, J.O.S. (Eds.), Oxford Handbook of Banking. Oxford: Oxford
UniversityPress,463–85.
Jin,J.Y.,Kanagaretnam,K.,Lobo,G.J.,2011.Abilityofaccountingandauditqualityvariablesto
predict bank failure during the financial crisis. Journal of Banking & Finance 35(11),
2811–2819.
Lane,W.R.,Looney,S.W.,Wansley,J.W.,1986.AnapplicationoftheCoxproportionalhazards
modeltobankfailure.JournalofBanking&Finance10(4),511–531.
Männasoo, K., Mayes, D.G., 2009. Explaining bank distress in Eastern European transition
economies.JournalofBanking&Finance33(2),244‐253.
Mare, D.S., 2012. Contribution of macroeconomic factors to the prediction of small bank
failures.WorkingPaper.
Martin, D., 1977. Early warning of bank failure: A logit regression approach. Journal of
Banking&Finance1(3),249‐276.
Meyer,P.A.,Pifer,H.W.,1970.Predictionofbankfailures.JournalofFinance25(4),853‐868.
27
Santomero,A.,Vinso,J.D.,1977.Estimatingtheprobabilityoffailureforcommercialbanksand
thebankingsystem.JournalofBanking&Finance1(2),185‐205.
Shumway,T.,2001.Forecastingbankruptcymoreaccurately:Asimplehazardmodel.Journal
ofBusiness74(1),101‐124.
Sinkey, J.F.,1975.Amultivariate statistical analysisof the characteristicsofproblembanks.
JournalofFinance30(1),21‐36.
Sundararajan,V.,Enoch,C., San Jose´,A.,Hilbers,P.,Krueger,R.,Moretti,M.,Slack,G.,2002.
Financial soundness indicators:analyticalaspectsandcountrypractices. IMFOccasional
Paper(212).InternationalMonetaryFund:Washington,D.C.
VanderVennet,R.,2002.Costandprofit efficiencyof financial conglomeratesanduniversal
banksinEurope.JournalofMoney,CreditandBanking34(1),254‐282.
West,R.C.,1985.Afactor‐analyticapproachtobankcondition.JournalofBanking&Finance
9(2),253‐266.
Wheelock,D.C.,Wilson,P.W.,2000.Whydobanksdisappear?ThedeterminantsofUSbank
failuresandacquisitions.TheReviewofEconomicsandStatistics82(1),127‐138.
28
Appendix1
Efficiencymeasuresestimation
The set of potential explanatory variables is chosen in order to explain the
probability of failure as a consequence of the better management of bank operations
quantifiedbytheefficiencymeasures.Costefficiencyismeasuredusingstochasticfrontier
analysis and the Battese and Coelli (1995) stochastic frontiermodel.We use the following
translogfunctionalform:
(6)
where TC is the logarithm of the total production cost; yi (i=1, 2, 3, 4) are output
quantities; Pj (j=1, 2, 3) are input prices; ln E is the natural logarithm of total equity
capital;andTisthetimetrendtoaccountforpossiblechangesintechnologyduringthe
observed period. In order to guarantee linear homogeneity in factor prices, it is
necessary(andsufficient)toapplythefollowingrestrictions:1)thestandardsymmetry
and2)linearrestrictionofthecostfunction.Followingpreviousstudies(VanderVennet,
2002;Girardoneetal.,2004),weincludeequityasanetput,specifyinginteractionterms
withbothoutputquantitiesandinputprices.
Italian cooperative banks constitute a heterogeneous dataset. To overcome this
problem and take into consideration local market conditions, we adopt the technical
inefficiencyeffectsmodelproposedbyBatteseandCoelli(1995).FollowingBattagliaet
LnTC a0 bi lnYi i1
4
c j lnPj j1
3
e1 ln E t1T 1
2ij
j1
4
lnYij i1
4
ijj1
3
lnPij e11 ln E ln E t11T2
i1
3
ijj1
3
lnYi ln Pj i lnYi ln E i1
4
i lnYiT i1
4
j lnPj ln E j1
3
j lnPjTj1
3
i1
4
T ln E
29
al.,(2010),weincludealargesetofenvironmentalvariables(Table7)thatareestimated
attheregionallevel8.
<InserthereTable7>
To obtain a complete view of the performance of cooperative banks, we also
estimate revenue‐ and profit‐efficiency measures, which specify various measures of
profits(respectively,totalrevenue,andoperatingincomeandinterestmargin)giventhe
levelofoutputsratherthantheoutputprices.Thefrontierdefinitionisthesameasinthe
cost case, except for the dependent variable: total cost is replaced by the previously
mentionedprofitmeasure.
8Our frontiermodel includesenvironmentalvariablesthatareestimatedat theregional level (foreachof the20Italiancountry regions), but we only report summary statistics at the macro‐geographical‐area level. Statistics at the moredetailedregionallevelareavailableuponrequest.
30
Tables
Table1.Numberofbanksandhistoricaldefaultrates:cooperativevs.commercialbanks
CooperativeBanks CommercialBanks
YearNumberofBanks(A)
Liquidation(B)
Administration(C)
TotalNumberofBanksinDefault
(D=A+B)
DefaultRate(E=D/A)
NumberofBanks(A)
Liquidation(B)
Administration(C)
TotalNumberofBanksinDefault
(D=A+B)
DefaultRate(E=D/A)
1997 405 1 3 4 0.99% 291 ‐ 2 2 0.69%
1998 404 1 5 6 1.49% 299 ‐ ‐ ‐ 0.00%
1999 414 ‐ 4 4 0.97% 288 1 1 2 0.69%
2000 432 ‐ 5 5 1.16% 284 ‐ 1 1 0.35%
2001 436 ‐ 5 5 1.15% 296 ‐ ‐ ‐ 0.00%
2002 436 ‐ 6 6 1.38% 293 ‐ 1 1 0.34%
2003 430 1 6 7 1.63% 282 ‐ 1 1 0.35%
2004 434 ‐ 2 2 0.46% 279 ‐ 1 1 0.36%
2005 424 ‐ 4 4 0.94% 279 ‐ ‐ ‐ 0.00%
2006 432 ‐ 1 1 0.23% 283 ‐ ‐ ‐ 0.00%
Pre‐CrisisPeriod(1997‐2006)
4,247 3 41 44 1.04% 2,874 1 7 8 0.28%
2007 431 ‐ ‐ 0 0.00% 234 ‐ ‐ ‐ 0.00%
2008 431 ‐ ‐ 0 0.00% 255 ‐ 1 1 0.39%
2009 421 1 5 6 1.43% 285 ‐ 6 6 1.75%
CrisisPeriod(2007‐2009)
1,283 1 5 6 0.47% 774 0 7 7 0.90%
OverallPeriod(1997‐2009)
5,530 4 46 50 0.90% 3,648 1 14 15 0.41%
Source:owncalculationsusingdatafromFedercasse.Forcommercialbanks,datawereobtainedfromtheSupervisionBulletinoftheBankofItaly.Datawerenotavailableforalltheactivebanksintheperiod.Notes:ThistablepresentsthehistoricaltimeseriesofItalianbanks,subjecttotemporaryconservatorship(Administration)orclosed(Liquidation)bytheItalianbankingsupervisor(BankofItaly).Dataarepresentedseparatelyforcooperativeandcommercialbanks.Cooperativebanks'defaultrateismorethantwicethefigureforcommercialbanksduringtheoverallperiod(1997–2009)andalmostfourtimeshigherduringthepre‐crisisperiod(1997–2006).Duringthe2007–2009financialturmoil,thistendencywasreversed,withthecommercialdefaultraterisingtoalmosttwiceashighasthefailurerateforcooperativebanks.
31
Table2.Variabledefinitions
Variable Definition Representativestudies
Cost Thisisthecostefficiency,estimatedusingthemethodillustratedinAppendix1.
Variousstudies,seeareviewinHughesandMester(2010)
RevenueThisistherevenueefficiency,estimatedusingthemethodillustratedinAppendix1. FiordelisiandMolyneux(2010)
Operating Thisistheprofitefficiency,estimatedusingthemethodillustratedinAppendix1.
Variousstudies,seeareviewinHughesandMester(2010)
InterestThisistheinterestmarginefficiency,estimatedusingthemethodillustratedinAppendix1.
Cost‐Income Thisistheratiobetweenoperatingexpensesandoperatingincome.
Laneetal.(1986);MännasooandMayes(2009)
CapitalAdequacyThisistheratiobetweencapitalinexcessofregulatoryrequirementsovertheminimumcapitalrequirements. Regulatoryrequirements
CreditOrientation Thisistheratiobetweentotalloansandtotalassets. Laneetal.(1986);MännasooandMayes(2009)
AssetQualityThisistheratiobetweenloanlossprovisionsandtotalloans.
MännasooandMayes(2009);Arena(2008)
Liquidity Thisistheratiobetweenbankdepositsandcustomerdeposits.
MännasooandMayes(2009)
Size Thisisthenaturallogarithmoftotalassets. ColeandWhite(2012);Arena(2008)
EmploymentGrowthThisistheregionalworkforceemploymentgrowthoveratwo‐yearperiod. DeYoung(2003)
GDPGrowth ThisistheannualgrowthrateintheItalianGDPpercapita(atcurrentprices).
Arena(2008);Sundararajanetal.(2002)
Notes:Thistablereportsthenamesanddefinitionsofthevariablesemployedintheestimation.Thecolumn“representativestudies”listssomestudiesthathaveusedtheexplanatoryvariables.
32
Table3.Descriptivestatistics
All Non‐Failed Failed T‐testforMeans
Variables Mean S.E. Mean S.E. Mean S.E. Difference t‐statistic
CostEfficiency
0.630 0.269 0.632 0.269 0.487 0.210
0.144 3.543***
RevenueEfficiency
0.630 0.263 0.629 0.263 0.702 0.268
‐0.073 ‐1.843*
OperatingEfficiency
0.611 0.271 0.614 0.269 0.329 0.231
0.284 6.978***
InterestEfficiency 0.729 0.242 0.731 0.241 0.542 0.230 0.189 5.172***
Cost‐Income 0.780 0.127 0.779 0.123 0.900 0.310 ‐0.122 ‐6.380***
CapitalAdequacy 2.631 2.230 2.629 2.223 2.768 2.851 ‐0.139 ‐0.4107
CreditOrientation 0.642 0.122 0.641 0.122 0.716 0.128 ‐0.075 ‐4.038***
AssetQuality 0.003 0.005 0.003 0.005 0.010 0.011 ‐0.007 ‐9.083***
Liquidity 0.026 0.045 0.026 0.044 0.031 0.059 ‐0.005 ‐0.779
Size 11.663 0.983 11.675 0.971 10.490 1.324 1.185 8.016***
EmploymentGrowth 0.021 0.020 ‐ ‐ ‐ ‐ ‐ ‐
GDPGrowth 3.877 1.248 ‐ ‐ ‐ ‐ ‐ ‐
Notes:Thistablereportsthedescriptivestatisticsforthevariablesemployedtoexplainbankfailuresovertheperiodfrom1997–2006.Dataarepresentedforallbanksandseparatelyforsurvivingbanksanddefaultedbanks.Inthelasttwocolumns,wepresentthedifferenceinmeansbetweenthetwogroups(Non‐FailedandFailed)andthet‐statisticunderthehypothesisofequalityofvariancesforthetwopopulations.Thet‐statisticiscomputedastheratioofthedifferenceinmeanstothedifferenceinstandarderrors.
33
Table4.Estimationresultsofthediscrete‐timehazardregressionmodelincooperative
bankingbetween1997and2006
Models
Variables (1) (2) (3) (4) (5)Cost ‐4.34***
(1.017) Revenue ‐1.643*
(0.927) Operating ‐3.559***
(0.962) Interest ‐1.959***
(0.662)Cost‐Income 0.060
(0.051)CapitalAdequacy ‐0.752*** ‐0.548*** ‐0.493** ‐0.524*** ‐0.653***
(0.202) (0.203) (0.195) (0.191) (0.225)CreditOrientation 0.896*** 0.472** 0.425** 0.501*** 0.623***
(0.164) (0.207) (0.178) (0.178) (0.166)AssetQuality 0.453*** 0.583*** 0.489*** 0.497*** 0.544***
(0.076) (0.08) (0.077) (0.077) (0.073)Liquidity 0.322*** 0.254** 0.357*** 0.323*** 0.249**
(0.102) (0.105) (0.095) (0.099) (0.106)Size ‐0.678*** ‐1.082*** ‐0.728*** ‐0.85*** ‐0.949***
(0.208) (0.17) (0.167) (0.164) (0.183)EmploymentGrowth ‐0.311 9.325 2.743 4.064 6.631
(8.224) (8.59) (7.78) (7.913) (8.398)GDPGrowth ‐0.484*** ‐0.531*** ‐0.451*** ‐0.534*** ‐0.738*** (0.098) (0.164) (0.094) (0.103) (0.08)Observations 4215 4215 4215 4215 4215NumberofFailures 44 44 44 44 44NumberofBanks 476 476 476 476 476Notes:Thistablepresentstheresultsofthesurvivalmodelestimatedindiscretetime.Thefivemodelsincludethesamesetofvariablesapartfromtheefficiencymeasure.InModel(1),weincludecostefficiency;inModel(2)weconsiderrevenueefficiency;Model(3)comprisesoperatingincomeefficiency;inModel(4)weevaluatethecontributionofinterestmarginefficiency;andinModel(5)weconsiderthecost‐to‐incomeratio.Standarderrorsareinparenthesis.*,**,***indicatesignificanceatthe10%,5%,and1%levels,respectively,withrobuststandarderrors.
34
Table5:Predictiveaccuracy:in‐samplechecks
PanelA:Goodness‐of‐fitindicators
Measure Models1 2 3 4 5
Sensitivity 0.795 0.705 0.773 0.750 0.545Specificity 0.833 0.801 0.814 0.813 0.748OverallPredictive
0.832 0.800 0.813 0.812 0.745
ROCArea 0.879 0.832 0.866 0.845 0.765AccuracyRatio 0.757 0.664 0.733 0.689 0.531BrierScore* 0.947 0.939 0.908 0.926 1.108Source:owncalculations.*BrierScoreismultipliedby100.Notes:Thistablereportstheresultsofthemeasuresofpredictivepowerofthemodelin‐sample.
PanelB:Probabilityrankingsversusactualbankruptcies;percent*
Measure Models1 2 3 4 5
1‐5 0.068 0.136 0.045 0.114 0.0916 0.023 0.000 0.023 0.000 0.1147 0.023 0.068 0.023 0.068 0.1368 0.091 0.091 0.068 0.068 0.1599 0.114 0.091 0.091 0.182 0.13610 0.682 0.614 0.750 0.568 0.364Source:owncalculations.*Probabilityrankingsversusactualbankruptcies;percentclassifiedoutof44possible.**Decilesofthedistributionoftheestimatedhazardrate.Notes:Thistablereportstherankingofthebanksusingtheestimatedhazardrate.Noticethatthesumisnotnecessarilyequaltooneduetorounding.
35
Table6:Predictiveaccuracy:out‐of‐samplechecks
PanelA:Goodness‐of‐fitindicators
Measure Models1 2 3 4 5
Sensitivity 0.500 0.167 0.167 0.167 0.000Specificity 0.793 0.894 0.878 0.749 0.998OverallPredictive
0.789 0.884 0.868 0.741 0.984
ROCArea 0.715 0.657 0.680 0.567 0.626AccuracyRatio 0.430 0.313 0.359 0.134 0.252BrierScore 0.016 0.015 0.014 0.015 0.014Source:owncalculations.Notes:Thistablereportstheresultsofthemeasuresofthepredictivepowerofthemodelout‐of‐sample.Thecoefficientsestimatedthroughthemodelaremultipliedbythevaluesoftheexplanatoryvariablesin2008toobtaintheestimatedhazardratefor2009(see§4foradetailedexplanationoftheestimationofthediscrete‐survivalmodel).
PanelB:Probabilityrankingsversusactualbankruptcies;percent*
Measure Models1 2 3 4 5
1‐5 0.167 0.167 0.333 0.500 0.3336 0.167 0.333 0.000 0.000 0.0007 0.000 0.167 0.000 0.333 0.1678 0.167 0.000 0.000 0.000 0.3339 0.167 0.167 0.500 0.000 0.00010 0.333 0.167 0.167 0.167 0.167Source:owncalculations.*Probabilityrankingsversusactualbankruptcies;percentclassifiedoutof6possible.**Decilesofthedistributionoftheestimatedhazardrate.Notes:Thistablereportstherankingofthebanksusingtheestimatedhazardrate.Noticethatthesumisnotnecessarilyequaltooneduetorounding.
36
Table7:Environmentalvariablesincludedintheefficiencyestimation(meanvaluesby
macro‐geographicalarea)
Centre NorthEast NorthWest South Total
Z1 Population density (number of inhabitants persquarekilometre)(A)
194.00 132.74 337.80 214.53 192.17
Z2Index of concentration in the territory, (percentageratio between people resident in the main city of theregionandthoseresidentinthetowns)(A)
64.76 33.22 32.24 33.77 39.13
Z3Grossdomesticproductperhead(A) 24.75 27.73 28.81 15.18 24.20
Z4Entrepreneurialliveliness(ratioofthenetnumberofincorporationsintheRegistrarofCompanies)(A)
2.11 1.77 1.79 2.82 2.10
Z5 Incidence of nonperforming loans (incidence ofprecarious loans, overdue bills, groundings, andrestructured loanson the totalamountofbankassets)(B)
7.90 6.50 5.77 16.43 9.13
Z6 Number of cash points (ATM and POS) owned bycooperativebanksoverthetotalexistingintheterritory(B)
10.23 19.13 6.52 6.05 12.52
Z7 Number of bank branches owned by cooperativebanksoverthetotalexistingintheterritory(B)
9.66 37.80 8.58 9.12 21.47
Z8NumberofATMandPOS(ownedbycooperativesandotherbanks)per1,000inhabitants(B) 18.45 25.51 15.50 8.63 18.65
Z9 Number of branches (owned by cooperatives andotherbanks)per1,000inhabitants(B) 0.58 0.85 0.62 0.32 0.64
Z10 Index of firm weakness (number of bankruptciesdeclaredper1,000firms)(A)
2.97 1.75 2.53 2.66 2.31
Z11Level of criminality(number of bank robberies per1,000branches)(C)
77.06 41.40 93.44 91.45 67.52
Z12 Index of solidarity (number of blood donors per1,000inhabitants)(D)
0.22 0.35 0.34 0.12 0.27
(A)Sourceofdata:ISTAT(ItalianNationalInstituteofStatistics)(B)Sourceofdata:StatisticalBulletinsattachedtothemagazineCooperazionediCredito(C)Sourceofdata:Ministerodell’Interno(MinistryofHomeAffairs)(D)Sourceofdata:AVIS(ItalianAssociationofBloodDonors)