• Non ci sono risultati.

Does Prudential Regulation Contribute to Effective Measurement and Management of Interest Rate Risk? Evidence from Italian Banks

N/A
N/A
Protected

Academic year: 2021

Condividi "Does Prudential Regulation Contribute to Effective Measurement and Management of Interest Rate Risk? Evidence from Italian Banks"

Copied!
14
0
0

Testo completo

(1)

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/316770195

Does Prudential Regulation Contribute to Effective Measurement and

Management of Interest Rate Risk? Evidence from Italian Banks

Article  in  Journal of Financial Stability · May 2017

DOI: 10.1016/j.jfs.2017.05.004 CITATION 1 READS 52 4 authors:

Some of the authors of this publication are also working on these related projects: Fnancial Statement and Performance Disclosure of Pension FundView project Rosaria Cerrone

Università degli Studi di Salerno

22 PUBLICATIONS   5 CITATIONS    SEE PROFILE

Rosa Cocozza

University of Naples Federico II

49 PUBLICATIONS   115 CITATIONS    SEE PROFILE

Domenico Curcio

University of Naples Federico II

11 PUBLICATIONS   54 CITATIONS    SEE PROFILE Igor Gianfrancesco Extrabanca 8 PUBLICATIONS   15 CITATIONS    SEE PROFILE

All content following this page was uploaded by Rosa Cocozza on 13 April 2018.

(2)

ContentslistsavailableatScienceDirect

Journal

of

Financial

Stability

journalhomepage:www.elsevier.com/locate/jfstabil

Does

prudential

regulation

contribute

to

effective

measurement

and

management

of

interest

rate

risk?

Evidence

from

Italian

banks

Rosaria

Cerrone

a

,

Rosa

Cocozza

b

,

Domenico

Curcio

b,∗

,

Igor

Gianfrancesco

c

aDepartmentofBusinessStudies,ManagementandInnovationSystems,UniversityofSalerno,ViaGiovanniPaoloII,132,Fisciano,SA,84084,Italy bDepartmentofEconomics,ManagementandInstitutions,UniversityofNaples“FedericoII”,ViaCinthia,ComplessoMonteS.Angelo,Napoli,NA80126,

Italy

cRiskManagementDepartment,Extrabanca,ViaPergolesi2/A,Milano,MI20124,Italy

a

r

t

i

c

l

e

i

n

f

o

Articlehistory: Received9October2016

Receivedinrevisedform5March2017 Accepted2May2017

Availableonline6May2017 JELcodes: G21 G28 G32 Keywords: Bankregulation Interestraterisk MonteCarlosimulations Historicalsimulations Backtesting

a

b

s

t

r

a

c

t

Thispapercontributestopriorliteratureandtothecurrentdebateconcerningrecentrevisionsofthe

regulatoryapproachtomeasuringbankexposuretointerestrateriskinthebankingbookbyfocusing

onassessmentoftheappropriateamountofcapitalbanksshouldsetasideagainstthisspecificrisk.We

firstdiscusshowbanksmightdevelopinternalmeasurementsystemstomodelchangesininterestrates

andmeasuretheirexposuretointerestrateriskthataremorerefinedandeffectivethanareregulatory

methodologies.Wethendevelopabacktestingframeworktotesttheconsistencyofmethodologyresults

withactualbankriskexposure.UsingarepresentativesampleofItalianbanksbetween2006and2013,

ourempiricalanalysissupportstheneedtoimprovethestandardizedshockcurrentlyenforcedbythe

BaselCommitteeonBankingSupervision.Italsoprovidesusefulinsightsforproperlymeasuringthe

amountofcapitaltocoverinterestrateriskthatissufficienttoensurebothfinancialsystemfunctioning

andbankingstability.

©2017TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBY-NC-ND

license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Internationalbankingauthoritieshaverecentlydeveloped mod-elsandissuedguidelinestoassessbanks’exposuretointerestrate riskinthebankingbook(IRRBB), becauseofitspeculiarnature andsystemicrelevance.Followingthesavingsandloancrisisof the1980s and‘90s, theU.S.Federal ReserveBank (FED) devel-opedtheeconomicvaluemodel,whichmeasuresIRRBBthrough changesinabank’seconomicvalueofequity(EVE)viaa duration-basedestimateofinterestratesensitivity (Houptand Embersit, 1991;SierraandYeager,2004;Sierra,2009;WrightandHoupt, 1996Sierra,2009;WrightandHoupt,1996).Subsequently,in2004, theBaselCommitteeonBankingSupervision(BCBS)adoptedan accounting-baseddurationmodelthatestimatesIRRBBby

apply-夽 Thisresearchdidnotreceiveanyspecificgrantfromfundingagenciesinthe public,commercial,ornot-for-profitsectors.

∗ Correspondingauthor.

E-mailaddresses:rocerro@unisa.it(R.Cerrone),rosa.cocozza@unina.it (R.Cocozza),domenico.curcio@unina.it(D.Curcio),gianfrancescoi@extrabanca.eu (I.Gianfrancesco).

ingastandardizedshockininterestrates.Thestandardizedshock isalternativelygivenbya±200bpparallelshiftintheyieldcurve fixedforallmaturities(theparallelshiftsmethod),orbythe1st and99thpercentileofobservedinterestratechanges,usinga one-yearholdingperiodandaminimumfiveyearsofobservations(the percentilesmethod).Thesescenariosofchangesininterestrates allowcalculationofarisk indicator,whichisgiven bytheratio ofthechangeinabank’sEVEtoitsregulatorycapitalandwhose alertthresholdissetequalto20percent.Throughoutthepaper, wetermtheparallelshiftsandpercentilesmethodsas“regulatory methodologies.”

ThedrawbacksoftheBCBSframeworkhavebeenpointedout notonlybyacademicresearch,bymainlytestingitsunderlying assumptions (Abdymomunovand Gerlach,2014;Cocozza etal., 2015;Entropetal.,2008;Entropetal.,2009;FioriandIannotti, 2007),butalsobythefinancialauthorities(BCBS,2009).InMay 2015,theEuropeanBanking Authority (EBA,2015)publisheda technicaldocumenttoreviseandsupplementtheguidelines pro-posedbytheCommitteeofEuropeanBankingSupervisors(CEBS, 2006).Recently,BCBS(2016)issuednewstandardstorevisethe principlesfor IRRBB management and supervision setin 2004.

http://dx.doi.org/10.1016/j.jfs.2017.05.004

1572-3089/©2017TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4. 0/).

(3)

Thesestandardsareintendedtocomeintoforcein2018andshould provide more extensive guidanceon crucial areas, such asthe developmentofinterestrateshockscenariosandanupdated stan-dardizedframework.

Within their internal capital adequacy assessment process (BCBS, 2006;CEBS, 2006), banksusetoset asideagainstIRRBB an amount of internal capital corresponding to the numerator of the previously mentioned risk indicator, i.e., the change in banks’EVE due toan interest rateshock.Therefore, inaccurate estimates of IRRBB would lead banks to reserve anamount of internalcapitalthatmighteitherunderestimateoroverestimate abank’sactualriskiness,withpotentiallynegativeconsequences inbothcases:underestimationmightjeopardizebankingstability, whereasoverestimationmightchargebankshigheropportunity costsandeventuallyreducetheircreditsupplytotheeconomy.1

Apropermeasurementofbankriskexposureisrelevantfromthe perspectiveofsupervisoryauthoritiessince,basedontheresults ofthisassessment,theiractionsmayhaveahugeimpactonbank activity.2 Furthermore,inaccurateestimatesofIRRBBwouldalso

provide poorindications for banks’ assetand liability manage-ment(ALM)strategies:underestimationmightdrivemanagersto takeexcessiverisk,whereasoverestimationmightpreventbanks fromimplementingprofitableALMstrategies.Overall,theoutput oftheriskmodelsembeddedwithintheBaselregulationscould haveatremendousimpactontherealeconomyand,therefore,itis importanttounderstandtowhatextentbothbanksandfinancial authoritiescanrelyonthemandwhentheiruseisnotadvisable (seeDanielssonetal.,2016forageneralframeworktoquantify modelrisk).

Basedontheseissuesandthecurrentlow-interest-rate environ-ment,modelinginterestratechangesandmeasuringtheirimpact onthebankingsectorarestilltimelyandkeyresearchareas.Recent papershaveinvestigatedhowbanksmanagedtheirinterestrate riskduringthecrisisyears,whenunconventionalmonetary pol-icydecisionsdroveinterestratestounprecedentedlowlevels3,4

(Espositoetal.,2015;Chaudron,2016;Memmeletal.,2016).Our contributiontopriorliteratureandtothecurrentdebate concern-ingrecentrevisionstotheIRRBBregulatorytreatmentistwofold andcentersonanassessmentoftheappropriateamountofthe internalcapitalthatbanksshouldsetasideagainstthisspecificrisk. First, consistentwith important regulatoryconstraints com-monlyappliedtoquantifyIRRBB-relatedinternalcapital,weshow howbanksmightadoptsimulationtechniquestomodelchanges in interest rates in a more refined and effectiveway than the regulatorystandardizedshock.Simulationtechniqueshavebeen extensivelyexaminedbypriorliteraturetoestimatemarketrisk

1Theriskindicatoralsoprovidesusefulinformationwithregardtothe

equilib-riumofbanks’assetsandliabilitiesstructuresintermsoftheirrespectivematurities andrepricingdates.Fromthisperspective,itcomplementsthenetstablefunding ratiosetbytheBCBStomeasureandmanagebanks’exposuretoliquidityrisk(for ananalysisoftheimpactofthenetstablefundingratioonfinancialstabilitysee Ashrafetal.,2016).

2Kupiecetal.(2016)haverecentlyhighlightedtheimportanceofbank

examina-tionbysupervisoryauthoritieswithregardtoU.S.creditinstitutions.Theirevidence suggeststhatthebanksupervisionprocesssuccessfullyconstrainsthelending activ-itiesofbanksoperatinginanunsafeandunsoundmanner.

3Cukierman(2013)describesthechangesthatoccurredintheconductand

instru-mentsofmonetarypolicyusedbymajorcentralbankswhenthecrisishit.Healso discussestheconsequentnewtradeoffsandcontroversiesandspeculatesabout additionallikelyfuturechangesinmonetarypolicyandinstitutions.

4Theextremelylowlevelofinterestrateshasraisedmanyissuesintermsof

finan-cialintermediaries’profitabilityandoverallfinancialsystemstability.Accordingto Barneaetal.(2015),monetarypolicyandfinancialstabilitypolicyarehighly inter-related.Theyshowthatthemonetarypolicytransmissionmechanismdependson financialstabilitypolicytoolsaswellasonregulatoryandinstitutionalconstraints. Inparticular,theyfindpolicytradeoffsintryingtoaccomplishbothmonetaryand financialstabilitytargetsthatcentralbanksmusttakeintoaccount.

(Jorion, 2007) and suggested by regulators tomeasure interest rateriskwithinbanks’internalmeasurementsystems(BCBS,2004; BCBS,2016;ComptrolleroftheCurrencyAdministratorofNational Banks,2012).Intheabsenceofspecificpreviouscontributions,we discusstheapplicationofhistoricalandMonteCarlosimulation techniquestomodelchangesininterestratestomeasureabank’s exposuretoIRRBBinalow-interest-rateenvironmentand hence-forthtermthem“internalmethodologies.”

Second,wedevelopabacktestingframeworktotestthe consis-tencyofamethodology’sresultswithactualbankriskexposure under ordinary operating business conditions. Our backtesting frameworkdrawsontheforecastevaluationmethodsdeveloped byLopez(1999)totestVaRmodelsbasedonpotentiallosses result-ingfromtheunderestimationofmarketrisk.However,giventhe peculiaritiesofIRRBBmeasurementandmanagement,wefocusnot onlyonitsunderestimationbutalsoontheeffectsof overestima-tion.Wecomparethedifferentmethodologiesbyassessingtheir abilitytoavoidunnecessaryreductioninbanklendingactivityand associatedopportunitycostsshouldtheestimatedriskindicator overestimateactualriskexposure,andtoreducetheriskof under-miningbankingsectorstabilityshouldunderestimationoccur.

ThisresearchexaminesexposuretoIRRBBbycollecting pro-prietarybalancesheetinformationfromasample of130Italian commercialbanksovertheperiodfrom2006to2013.Analysisof theItalianbankingsystemisespeciallyinterestingbecauseItalian banksgenerallyactasqualitativeassettransformers,meaningthat IRRBBarisesfrombasicbankingbusiness.Becauseofthecrucial roleplayedbycommercialbanksinmodernfinancialsystems,our investigationisalsorelevantfromaglobalperspective.

Overall,ItalianbanksadequatelymanageIRRBBandtheirrisk indicatorsarewellbelowthealertlevelenforcedbytheregulator. In termsofthenatureoftheirriskexposure, oursample inter-mediariesappearexposedtobothraising(asset-sensitivebanks) and,morefrequently,decreasing(liability-sensitivebanks)interest rates.WefindthathistoricalandMonteCarlosimulations (inter-nalmethodologies)produceestimatesofinternalcapitalthatare morerobustfromaprudentialadequacyperspectivewithregard tothisspecificaspect,whencomparedwithregulatory method-ologies.Infact,theinternalmethodologiesavoidaspecificissue that we term“riskneutrality.”Based onthis misleadingresult, somebanksexperienceanincreaseintheirEVEwhenthe regu-latorystandardizedshockisapplied,whethertheshockbeupor down.Consistentwiththeoveralldecreasingtrendininterestrates observedduringthesampleperiod,liability-sensitivebanksuse derivativestooffsettheiron-balancesheetriskexposuretoIRRBB, whereasasset-sensitivebanksusetheiroff-balancesheetpositions toenhancethegainsassociatedwiththesamescenario.Onaverage, derivativeshavealsoallowedbankstoreducethedistancebetween theirestimatedandactualriskexposureandthusoptimizetheir amountofinternalcapitalagainstIRRBB.Furthermore,theinternal methodologies,andtheMonteCarlosimulationsinparticular, per-formbetterthantheregulatoryonesintermsoftheirconsistency withactualbankriskiness,measuredthroughtheexpostchanges ininterestratesthatactuallyoccurredaftertheestimationdate.

Ouranalysisimprovesuponthestandardizedshocktreatment adoptedinBCBS (2004)andprovidesnewinsightsforproperly measuringbankinternalcapitalagainstIRRBB.Themethodological frameworkweproposeisapplicabletopubliclyavailabledataand canbereplicatedbythoseoutsidebankinginstitutions.Such pub-liclyavailablemeasuresofIRRBBcanprovideacrucialcontribution tothefunctioningandstabilityoftheinternationalbankingsystem, thuscallingforgreaterattentionnotonlyfromregulators, super-visorsandtheoverallbankingindustry,butalsofromacademic researchers.

Therestofthepaperisorganizedasfollows.Inordertoposition thisresearchwithintheliterature,inSection2wediscussthree

(4)

papersthatexaminebanks’exposuretoandmanagementof inter-estrateriskduringthefinancialcrisisandmeasureIRRBBthrough theeconomicvalueapproachadoptedinthispaper.Section3 pro-videsanoverviewofthecurrentregulatoryIRRBBframeworkand discusses previousliterature testing therobustness of its main assumptions. In Section 4, we describe the two methodologies thatbanksmightinternallydeveloptomodelinterestratechanges basedonhistoricalandMonteCarlosimulationtechniques. Sec-tion5presentsabacktestingframeworktotesttheeffectivenessof bothinternalandregulatorymethodologies.Section6presentsthe resultsofourempiricalanalysis,inwhichwemeasuretheIRRBB ofasampleofItalianbanksbasedonbothregulatoryandinternal methodologiesandimplementourbacktestingprocedure.Section 7concludesthepaper.

2. Theexposureandmanagementofinterestraterisk

duringthefinancialcrisis

AfterthecollapseofLehmanBrothersin2008,interest-raterisk managementbecamemoreimportantthaneverforbothEuropean andU.S.banks,asaresultofunprecedentedmonetarypolicy deci-sionstakenbyboththeEuropeanCentralBank(ECB)andtheU.S. FED.5Thepreviouslyunheardofreductioninofficialpolicyrates

andunconventionalmeasuresadoptedbythetwocentralbanks hadastrongimpactonthetermstructureofmarketinterestrates andreducedthemtoanextremelylowlevel.AsshownbyBorio etal.(2015)andBuschandMemmel(2015),currentlowinterest ratesandtheflatyieldcurvehavealsodrivendownbank prof-itability.Majorissues, therefore,especially forbanksupervisors andpolicymakers,arewhetherornotthecurrent low-interest-rateenvironmenthascreatedincentivestotakemorerisk,andhow banksmanagedtheirriskexposureunderextraordinaryfinancial andeconomicconditionsduringthecrisis.Recentliteraturehas investigatedtheexposuretoandmanagementofinterestraterisk duringthecrisisyears.Weexaminebelowthreepapersthatrefer totheItalian,GermanandDutchbankingsectors,respectively,all usinganeconomicvalueapproachtomeasureIRRBB.

Espositoetal.(2015)measuretheIRRBBexposureofasampleof 68Italianbanksovertheperiodrangingbetweenthesecondhalfof 2008andthefirsthalfof2012.Inparticular,theyassesshow inter-mediariesmanagedIRRBBeithermodifyingthedurationmismatch betweenon-balancesheetassetsandliabilities(on-balancesheet restructuring)orrelyingoninterestratederivatives(off-balance sheetadjustment).Overall,byadoptingthedurationgapapproach ofthesimplifiedmethodologyintroducedbyBankofItaly(2006), andbyusingauniquedatasetofsemi-annualobservations,they findthatItalian banks’exposurewaswellbelowthepreviously mentionedalertthreshold.Furthermore,theyobservethatbanks usedtheiron-balancesheetinterestrateriskandtheiroff-balance sheetexposuretooffseteachother,ratherthantopursueastrategy aimingtoenhancethepotentialgainsthatmighthaveoccurredin thecaseofariseininterestrates.

Memmeletal.(2016),examineasampleofGermanbanks dur-ingtheperiodfrom2005to2014,inordertostudytherelationship betweenbanks’expectedreturnsfromtermtransformation and theirexposuretothecorrespondinginterestraterisk.According tothis research, decreasinginterest rates not only drivedown bankprofitmarginsbut,ifthelow-interest-ratescenariopersists, bankmanagershavetheincentivetoactasiftheyareriskprone, whichisextremelydangerousfromafinancialstabilityperspective. Theauthorsfindthattherelationshipbetweeninterestraterisk andbankrisktakingis,asexpected,usuallypositive,butbecomes

5 SeeECB(2011)foradiscussionoftheeffectivenessofunconventionalmonetary

policymeasuresadoptedbythetwocentralbanks.

weakerwhenabank’soperatingincomedecreases.Theythenshow thattherelationshipchangesitssignandbanksstarttoincrease theirexposuretointerestraterisk,eveniftheexpectedreturns fromtermtransformationdecrease,whenoperatingincomefalls belowacertainthreshold.

ByinvestigatinghowbankshavemanagedtheirIRRBBexposure underascenarioofdecreasinginterestratesandaflatteningyield curve,Chaudron(2016)detectstheinterest-rateriskpositionof42 Dutchbanks,representingroughly90percentoftheentirebalance sheetoftheDutchbankingsectorduringtheinvestigationperiod from2008untilthemiddleof2015.Theauthorinvestigateshow bankriskpositionchangesovertime,howmuchoftheirreturns onassetsandnetinterestmarginsdependonincomefrom matu-ritytransformationandwhichfactorsinfluencetheirinterest-rate riskposition.Interestrateriskismeasuredthroughthebasispoint value,whichisthechangein theeconomicvalueofequity due toa changeintheinterest ratebyone hundredthof apercent (0.01%).Dutchbanksshowrathersmallinterestrateriskpositions, areactiveinmaturitytransformationonlytoalimiteddegree,and donotmaintainaconstantinterestrateriskposition,butadjustit tochangesintheeconomicenvironment.Basedonapanelmodel estimation,interest-rateriskpositionsarenegativelyrelatedto on-balancesheetleverage,exhibitaU-shapedrelationwithsolvability, and, contrarytotheevidencereportedin otherstudies,do not varysystematicallywiththesizeofthebanks.Finally,banksthat receivedgovernmenthelpduringthecrisistookongreaterinterest raterisk.

3. Regulatorymethodologiestoestimateinterestraterisk

inthebankingsector:overviewandrelatedliterature

BasedonBCBS(2004),BankofItaly(2013)requiresbanksto allocateon-andoff-balancesheetaccountsintothefollowing14 timebandsofamaturityladder:i)demandandrevocable,ii)upto 1month,iii)1–3months,iv)3–6months,v)6–12months,vi)1–2 years,vii)2–3years,viii)3–4years,ix)4–5years,x)5–7years,xi) 7–10years,xii)10–15years,xiii)15–20years,xiv)morethan20 years.Overall,floatingrateassetsandliabilitiesareslottedintothe timebandsbasedontheirnextrepricingday,whereasfixed-rate accountsareassignedaccordingtotheirresidualmaturity.

Byassumingthat on-and off-balancesheetaccountshave a maturityexactlycoincidingwiththemidpointofeachtimebandito whichtheyareallotted,IRRBBismeasuredthroughpredetermined sensitivitycoefficients,i.e.,modifiedduration coefficients(MDi). Assets and liabilitiesare offset to calculate netpositions (NPi), whichareweightedbythecorrespondingMDiandtheassumed interestrateshock(r).Theresultingnetweightedpositionsare thensummedupacrossthedifferenttimebandstocalculatethe changeinabank’sEVE,whichisfinallydividedbythebank regula-torycapital(RC)toobtainariskindicator(RI),whosealertthreshold issetequalto20%: RI= 14



i=1 NPi·MDi·r RC ≤20% (1)

Undertheparallelshiftsmethod,risgivenbya±200bp paral-lelshiftintheyieldcurve.Accordingtothepercentilesmethod,the interestrateshockisbasedonthe1stand99thpercentilesofthe yearlyinterestratechange,obtainedbyusingtheoverlappingdata techniquewithaone-yearholdingperiodandaminimumfiveyears ofobservations(BCBS,2004).Basedontheso-callednon-negativity constraint,negativeshocksininterestratescannotdrivetheterm structureoftheobservedinterestratesbelowthezerolevel.

WithintheBCBSframework,accounting-baseddataareusedto obtainthecashflowstructureofabank’son-andoff-balance

(5)

posi-tions.Therefore,abank’sexposuretoIRRBBdepends,amongthe otherthings,onassumptionsconcerningthedistributionofasset andliabilitymaturitiesandrepricingdateswithinthetimebands ofthematurityladder.Byusingtime-seriesaccounting-baseddata, Entropet al. (2008) develop a model toestimate the distribu-tionofthematuritiesofabank’sassetsandliabilitieswithineach timeband.Theirestimatesaremoreinlinewiththoseinternally obtainedbybanks,andtheirmodelexplainsthecross-sectional variationinbankinterestrateriskbetterthantheBCBSapproach.

Asfarasinterestrateshocksandpredeterminedsensitivity coef-ficientsareconcerned,FioriandIannotti(2007)developaValue at Risk(VaR)methodology tomodel interest ratechanges that takesintoaccountbothasymmetryandkurtosisoftheinterest-rate distribution.Theirmethodology,basedonaprincipalcomponent MonteCarlosimulation,accountsnotonlyfortheconceptof dura-tionbutalsoforthat ofconvexity,andcalibratesthesensitivity coefficientsusingmarketdata.Byanalyzingthe18major large-andmedium-sizedItalianbanks,theyshowthattheirresultsare consistentwiththeriskexposureestimatedviathe±200bp par-allelshiftiftheregulatorysensitivitycoefficientsarecalibratedon thebasisofcurrentmarketdataattheevaluationdate.

BCBS(2004)andBankofItaly(2013)setspecificcriteriatoallot specificaccountswhoselegalmaturitydiffersfromthebehavioral one.Withregardtonon-maturitydeposits,Cocozzaetal.(2015) develop a methodology that considers their actual behavior in termsofbothpricesensitivitytochangesinmarketratesand vol-umestabilityovertime.Accordingtotheirresultsfromasampleof 30Italiancommercialbanks,theuseofdifferentallocationcriteria affectsnotonlythesizeoftheriskestimatebutalsothenatureof bankriskexposure.Allotmentcriteriamayalsodeterminethe so-calledriskinversionphenomenon,i.e.,banksformerlyexposedto anincreaseininterestratescanalsobecomeexposedtoareduction ininterestrates.Furthermore,theyshowthat,whenmarketrates arelow,somebanks,definedasrisk-neutralbanks,appearto expe-rienceanincreaseintheirequityeconomicvaluewhetherinterest ratesdecreaseorincreaseundertheparallelshiftsmethod.

AbdymomunovandGerlach (2014)showthat,when market rates are low, the methods used by banks and supervisors to assessinterestrateriskunderstressedconditionscanunderstate banks’exposureandproposeanewmethodforgenerating yield-curvescenarios.TheirevidencefromalargeU.S.bankshowsthat theirmodelgeneratesyield-curvescenarioswithawidervariety ofslopesandshapesthanalternativehistoricalandhypothetical scenario-generationmethods,includingtheregulatory methodolo-gies.

ByconsideringtheaggregatedGermanuniversalbanking sys-tem, Entrop et al. (2009) analyze how a bank’s risk exposure changesifsomemajorassumptions oftheregulatorymodelare modified.Inparticular,theyconsidertheallotmentofnon-maturity depositsintothetimebands,theallotmentofassetsand liabili-tieswithinthetimebands,thenumberandboundariesofthetime bands,theamortization rateofcustomerloans, and thespread betweenthecouponandthemarketinterestrateusedto calcu-latethemodifieddurationassociatedwitheachtimeband.They provethattheresultsoftheregulatoryframework significantly dependontheunderlyingassumptionsand mustbeconsidered withcautionforbothsupervisoryandrisk-managementpurposes, sincetheycannotbeassumedtobeabovethelevelofrealbank riskiness.

4. InternalmethodologiestoestimateIRRBB

Within their internal capital adequacy-assessment process, ItalianbanksmeasureIRRBB,underbothordinaryandstressed con-ditions,byapplyingdifferenttechniques,ceterisparibus,tomodel

interestratechanges.Therefore,inlinewithbankingpractice,here wetaketherestoftheregulatoryassumptionsasgivenandfocus onmodelinginterestratechanges,inordertoaddressthelimitsof theBCBS(2004)parallelshiftsandpercentilesmethods.In particu-lar,theparallelshiftsmethodissetregardlessofactualchanges in interestrates. In this respect,thetwo scenarios correspond-ingtothe1stand99thpercentilesofthepercentilesmethodare changesthatactuallyoccurred.Nevertheless,thesechangesmight haveoccurredondifferentdays.Forexample,the1stpercentile mightrefertoJanuary22nd,2008fortheinterestrateofthefirst timeband,toDecember23rd,2010fortheinterestrateassociated withthesecondtimeband,etc.Therefore,aswithparallelshifts,the percentilesmethoddoesnotaccountforthecorrelationsactually observedamongtheannualchangesininterestrates.

In this section, we present two methods that banks can internallydevelopbymakinguseofhistoricalandMonteCarlo sim-ulationtechniques,respectively.Theshocksareappliedtoaterm structureofinterestrates(henceforth,ourkeyrates)whosenodes correspondtothemidpointsofthe14timebandsoftheregulatory maturityladder.

Thehistoricalsimulationmethodcanaccountforcorrelations amongtheannualchangesinkeyratesbecausetheriskindicatoris basedonscenariosofchangesininterestratesthatarerepresented bythejointannual changesin ourkey ratesthat haveactually occurredoverthepastfiveyears.Thesescenariosarecalculated onagivendaythroughtheoverlappingtechnique,assuggested byBCBS(2004),andareappliedtothenetpositionstoobtainthe netweightedpositions.Wethensumthenetweightedpositions tocalculatethechangeinabank’sEVEanddividethissumbythe regulatorycapitaltoobtainanempiricaldistributionof therisk indicator.Thisdistributioniscutincorrespondencewiththe per-centileassociatedwiththedesiredconfidencelevel,whichisset equalto99%followingBCBS(2004).However,itispossiblethatthe non-negativityconstraintmightpreventthismethodfrom captur-ingthecorrelationswheninterestratesareverylow.

TheMonteCarlosimulationmethodallowsustogenerate sce-nariosthat bothtakeintoaccountthecorrelationsbetweenthe annualchangesinourkeyratesandmeetthenon-negativity con-straint.Wecarryoutasmanysimulationsasarerequiredtoobtain thedesirednumberKofscenarios,whichwesetequalto10,000, andrejectthosesimulationsleadingthetermstructureofourkey ratesunderthezerolevelinoneormorenodes.Inthisway,we obtainadistributionoftherisk indicatorthatis cutatthe per-centilecorrespondingtoa99percentconfidencelevel.Specifically, wedevelopthemethodasfollows:

i) Select the joint probability density function that guaran-teesthebestapproximationoftheactualdistributionsofannual changesinthekeyrates.Theapplicationoftheoverlappingdata techniquesupportstheuseofanormaljointprobabilitydensity function,alreadyadoptedbyFioriandIannotti(2007).

ii) Estimate meansand variances ofthe distributions of the annual changes in the key rates and theirvariance-covariance matrix().Distributionsofannualchangesarenotadjustedon thebasisofthenon-negativityconstraintinordertoaccountfor actualcorrelationsamongtheannualchangesinkeyrates.

iii)Generatearandomnumberui(i=1,...14)rangingfrom0to 1ateachnodeofourkeyratestermstructure.

iv)Converteachuiintoavaluezi(i=1,...14)distributed accord-ingtoastandardnormal.Insymbols:

zi=F−1(ui) (2)

whereF−1istheinverseofthedistributionfunctionofthe proba-bilitydensityfunctionoftheannualchangesoftheithkeyrate.

(6)

v)Usethealgorithm ofCholeskyinordertodecomposethe matrixintwomatricesQandQ’suchthat:

Q·Q =˝ (3)

vi)Calculatethevectorx,whoseelementsarethejoint sim-ulated annual changes in the key rates through the following formula:

x=Q·z+ (4)

wherezisthevectorofthevalues calculatedinstepiv)and␮isthe vectorofthe14meansofthedistributionsofthekeyratesannual changescalculatedinstepii).Eachvectorxrepresentsasimulated scenariothatwillbeusedtocalculatetheriskindicator.

vii)Repeatstepsiii)-vi)untilwereachanumberKofscenarios thatmeetthenon-negativityconstraint.

viii)ApplytheKsimulated scenarios tothenetpositionsto calculatethenetweightedpositions.For eachscenario,thenet weightedpositionsaresummedinordertocalculatethechange inabank’seconomicvalue,whichisfinallydividedbythe regula-torycapital.Theempiricaldistributionoftheriskindicatorisfinally cuttoidentifythe99thpercentile.

Fromamethodologicalperspective,itisinterestingtonotethat, undertheparallelshiftsandpercentilesmethods,theriskindicator isobtainedbyassumingthatallthekeyratesmovetogetherinthe samedirection.Underboththeinternalmethodologiespresented here,however,keyratesareallowedtomoveindifferentdirections andtheonlyrestrictionappliedtothechangesinthekeyratesis thenon-negativityconstraintsetbyBCBS.Weallowthekeyratesto changebecauseepisodesintherecentandmoredistantpast,such asthemeasuresofmonetarypolicythathavedrivenmarketrates toextremelylowlevelsandtheU.S.savingsandloancrisis,have shownthattheirtermstructurecanactuallyassumecharacteristics anddynamicswhich,exante,wouldhavebeenjudgedunrealistic.

5. Abacktestingframework

Based onthe related literature (for a review, see Campbell, 2006),mainlytwoapproacheshavebeenemployedtoimplement abacktesting procedure. Abacktestcouldbe basedeither ona hitfunction(see,forexample,thebacktestsdevelopedbyKupiec, 1995;Christoffersen,1998;ChristoffersenandPelletier,2004)or onalossfunction(Lopez,1999).Regardingtheformer, backtest-ingproceduresbasedonalossfunctionhavegreaterflexibility, sincethelossfunctioncanbetailoredtoaddressspecificconcerns, eveniftheincreasedflexibilitygeneratesasubstantialincreasein theinformationalburdenassociatedwithassessingtheaccuracy ofacertainmodelunderconsideration.Infact,inordertoassess whethertheaveragelossis“toolargerelativetowhatwouldbe expected,”itisnecessary tounderstandthestochasticbehavior ofthelossfunction.Therefore,backtestingproceduresbasedona lossfunctionapproachrequireanexplicitassumptionaboutthe distributionofprofitsandlosses.

Givenourobjectiveofcomparingtheperformanceofthe dif-ferentmethodologiesusedtomeasureIRRBB,however,wedecide toadoptaloss-function-basedapproach,sincepriorliterature rec-ognizesthatloss-function-basedbacktestsmaybeveryusefulfor determiningwhetheronemodelprovidesabetterriskassessment than a competing model. From this perspective, loss-function-based backtests may be more suited to discriminating among competingmodelsrather than judging theaccuracy ofa single model,whichisexactlytheaimofourbacktestingframework.

Tosetupourbacktestingframework,weadoptthelogic under-lyingtheforecastevaluationmethodsproposedbyLopez(1999), whichtestsVaRmodelsbyfocusingonthepotentiallosses

asso-ciatedwiththeunderestimationofmarketriskandgivestoeach Table

1 Key rate term structures. t (evaluation date) Demand and revocable Up to 1 month From 1 to 3 months From 3 to 6 months From 6 months to 1 year From 1 to 2 years From 2 to 3 years From 3 to 4 years From 4 to 5 years From 5 to 7 years From 7 to 10 years From 10 to 15 years From 15 to 20 years Over 20 years December 31st 2006 3.69% 3.62% 3.66% 3.79% 3.95% 4.10% 4.13% 4.13% 4.13% 4.14% 4.17% 4.24% 4.29% 4.31% December 31st 2007 3.92% 4.18% 4.49% 4.70% 4.73% 4.63% 4.54% 4.53% 4.54% 4.58% 4.67% 4.80% 4.89% 4.91% December 31st 2008 2.35% 2.45% 2.79% 2.93% 3.02% 2.72% 2.86% 3.04% 3.18% 3.36% 3.61% 3.85% 3.88% 3.76% December 31st 2009 0.41% 0.40% 0.56% 0.84% 1.13% 1.59% 2.06% 2.41% 2.68% 2.99% 3.43% 3.82% 4.02% 4.05% December 31st 2010 0.82% 0.65% 0.89% 1.10% 1.37% 1.45% 1.75% 2.08% 2.34% 2.70% 3.12% 3.50% 3.67% 3.68% December 31st 2011 0.63% 0.76% 1.18% 1.48% 1.79% 1.37% 1.35% 1.45% 1.63% 1.91% 2.24% 2.57% 2.68% 2.65% December 31st 2012 0.13% 0.09% 0.15% 0.25% 0.43% 0.35% 0.42% 0.54% 0.69% 0.95% 1.37% 1.83% 2.09% 2.20% December 31st 2013 0.45% 0.20% 0.26% 0.34% 0.48% 0.48% 0.66% 0.89% 1.13% 1.49% 1.95% 2.41% 2.65% 2.73% This table shows the term structure of the key rates referred to each time band of the regulatory maturity ladder observed at the end of the years included in the 2006–2008 period. We use data from Datastream, and, in particular, we use the EONIA (Euro Overnight Index Average) rate for the maturity corresponding to the demand and revocable time band, the Euribor rates for maturities shorter than 12 months, and interest rate swap (IRS) rates for maturities longer than, or equal to, 1 year. Key rates referred to maturities not included in the market term structure are calculated through linear interpolation.

(7)

Table 2 Descriptive statistics of the annual changes in the key rates observed over the 2009–2013 period with December 31st, 2013 as the evaluation date. Demand and revocable Up to 1 month From 1 to 3 months From 3 to 6 months From 6 to 1 year From 1 to 2 years From 2 to 3 years From 3 to 4 years From 4 to 5 years From 5 to 7 years From 7 to 10 years From 10 to 15 years From 15 to 20 years Over 20 years Mean − 0.76 − 0.80 − 0.86 − 0.88 − 0.87 − 0.79 − 0.74 − 0.70 − 0.67 − 0.61 − 0.55 − 0.49 − 0.45 − 0.43 Std. Dev. 1.32 1.41 1.45 1.41 1.38 1.16 0.97 0.86 0.79 0.71 0.61 0.56 0.56 0.57 Max. 1.22 0.94 0.85 0.77 0.83 1.00 1.09 1.05 0.93 0.78 0.59 0.60 0.62 0.66 Min. − 4.27 − 4.67 − 4.67 − 4.54 − 4.34 − 3.89 − 3.41 − 2.94 − 2.60 − 2.14 − 1.69 − 1.76 − 1.86 − 1.91 Kurtosis 0.29 0.16 0.08 0.11 0.07 0.41 0.16 − 0.33 − 0.68 − 0.99 − 1.13 − 1.07 − 0.92 − 0.93 Asimmetry − 1.21 − 1.12 − 1.07 − 1.05 − 0.99 − 1.00 − 0.67 − 0.30 − 0.08 0.11 0.17 0.12 − 0.02 − 0.11 We use data from Datastream, and, in particular, we use the EONIA (Euro Overnight Index Average) rate for the maturity corresponding to the demand and revocable time band, the Euribor rates for maturities shorter than 12 months, and interest rate swap (IRS) rates for maturities longer than, or equal to, 1 year. Key rates referred to maturities not included in the market term structure are calculated through linear interpolation.

methodologyascorebasedonacertainlossfunction:Thelower thescore, thebetteris themethodology performance. Forecast evaluationmethodsareparticularlysuitableforbacktestingwith datasetsassmallasoursbecausetheydonotsufferfromthelow powerofastandardtestsincetheyarenotfrequency-based statis-ticaltests.Furthermore,asmentionedabove,forecastevaluation methodsallowustotailorthelossfunctiontothespecific objec-tivesofthebacktestinganalysis.Therefore,giventhepeculiarities ofIRRBB,weadaptLopez’s(1999)backtestingframeworkto specifi-callyaccountforbothregulatorandindustryconcernsinestimating IRRBB,especiallyintermsofbankingstabilityandcreditsupplyto theeconomy.

Foreachsamplebank,wecomparetheexanteriskindicators withameasureofactualbankriskexposure,termedtheexpostrisk indicator.TheexpostriskindicatorisobtainedbysettingrinEq. (1)equaltothejointannualchangesinthekeyratesthatactually occurredovertheone-yeartimehorizonfollowingtheevaluation datet.Foranyevaluationdatet,weassigntoeachmethodologym ascore,calculatedthroughascorefunction,Sm,t,takingasinputs theresultsofanaccuracyfunction,Ai,t,thatisappliedtoeachbank i(i=1,2,...N,withN=130)ateachevaluationdatet.Thegeneric scorefunction,foramethodologymandanevaluationdatet,is formalizedasfollows: Sm,t= N



i=1 Ai,t N∗ (5) where:

-Ai,tisdefinedsuchthattheoutputsofthescorefunctioncannot takenegativevaluesandbettermethodologiesarecharacterized bylowerscores.

-N*isanintegerwhosevaluedependsonthespecificationofthe accuracyfunction.

Thegenericaccuracyfunctioncanbewrittenasfollows: Ai,t=



f



RIi,tpost,RIante i,t



if RIi,tpost>RIante i,t g



RIposti,t ,RIante

i,t



if RIi,tpost≤RIante i,t

(6)

-where:RIi,tanteandRIi,tpostaretheexanteandexpostriskindicators, respectively.BothRIi,tanteandRIposti,t refertothetermstructureofthe ithbank’snetpositionsobservedintheevaluationdatet. -fandgdefinethevaluesoftheaccuracyfunctioniftheexpost

riskindicatorishigherorlowerthan(equalto)theexanteone. Thefirsttwospecificationsoftheaccuracyfunctionrefertothe caseofanunderestimationofactualriskexposure,whentheexpost riskindicatorishigherthan(equalto)theexanteone.Therefore, consistentlywithLopez(1999),thefirstandsecondspecifications bothsatisfythefollowingcondition:

f



RIposti,t ,RIante i,t



≥g



RIposti,t ,RIante i,t



(7) Inparticular,accordingtothefirstspecificationshowninthe followingEq.(8),theaccuracyfunctionequals1iftheexpostis higherthantheexanteriskindicator,and0otherwise:

Ai,t=



1 if RIi,tpost>RIante i,t 0 if RIi,tpost≤RIante

i,t

(8) BysettingN*equalto1inthescorefunction(5),thefinalscore isthenumberoftimesinwhichanunderestimationoccurs,fora givenmethodologymatanevaluationdatet,andisdefinedasa fre-quencyscore.UnderthesecondspecificationofEq.(9),theaccuracy

(8)

functionprovidesameasureoftheseverityoftheunderestimation errorsinceitequalsthedifferencebetweentheexpostandtheex anteriskindicators,iftheformerishigherthanthelatter,and0 otherwise.Thelargerthedifference,thegreateristhe underesti-mationofactualriskexposureandthegreateristhepotentialthreat tobankingstability.Insymbolswehave:

Ai,t=



RIi,tpost−RIante i,t if RI

post i,t >RIantei,t 0 if RIposti,t ≤RIante

i,t

(9) Inthiscase,N∗inthescorefunction(6)issetequaltothenumber ofbanksforwhichRIposti,t >RIantei,t andthescorefunctioncaptures theaveragemagnitudeofthemeasurementerrorforeach method-ologymonacertainevaluationdatet,providingwhatweterma severityscorefortheunderestimationcase.

ByremovingtheconstraintofEq.(7),thethirdspecificationof theaccuracyfunction,showninEq.(10),accountsforan overesti-mationofactualriskexposure.Inparticular,itequalstheabsolute valueofthedifferencebetweentheexpostandtheexanterisk indicators,whentheformerislowerthanthelatter,and0 other-wise.Thelargerthedifference,thegreateristhepotentialreduction ofthecreditsupplytotheeconomy,sincetheamountofinternal capitalthatabankunnecessarilysetsasideishigher.Insymbolswe have:

Ai,t=



0 if RIposti,t ≥RIante i,t |RIposti,t −RIi,tante|if RI

post i,t <RIantei,t

(10) Inthiscase,N∗inthescorefunction(5)isthenumberofbanks forwhichRIi,tpost<RIi,tanteandthefinalscorecapturestheaverage differencebetweenthemforagivenmethodologymandan evalu-ationdatet,providingameasureoftheaverageerror,i.e.,aseverity scorefortheoverestimationcase.

Finally,weadoptafourthspecificationoftheaccuracyfunction, whichisgivenbythecombinationofthepreviousaccuracy func-tions(9)and(10),thusconsideringthedistancebetweentheex anteandtheexpostriskindicatorsinbothcasesofunderestimation andoverestimation.Insymbols:

Ai,t=



RIi,tpost−RIante i,t if RI

post i,t >RIantei,t |RIposti,t −RIi,tante|if RI

post i,t <RIantei,t

(11) Inthiscase,N∗ofthescorefunction(5)isthetotalnumberof banksandthefinalscoreisameasureoftheoverallproximityof theexantetotheexpostriskindicator,andisdefinedsimplyas aproximityscore.Theaccuracyfunction(11)makesno distinc-tionbetweencasesofunderestimationandoverestimation.Inany event,inordertosetaspecificorderofpriority,Eq.(11)canbeeasily adjustedtotakethekeyissuesassociatedwithoverestimationand underestimationofIRRBBintoaccountbydifferentlyweightingthe twocases.

6. Empiricalevidence

6.1. Data

Weestimateoursamplebanks’riskindicatorsasofDecember 31st for each of the years included in the 2006–2013 period. December31stisalsothedateonwhichweestimateIRRBBand towhichbankspecificbalancesheetdatarefer.Inconsistencywith theprevioussection,theseevaluationdatesareindicatedwitht (t=December31st,2006,December31st,2007, ...December31st, 2013).KeyratesareobservedonDecember31stforeachoftheyears includedintheinvestigationperiod.Tobuildthetermstructureof thekeyrates,weusetheEONIA(EuroOvernightIndexAverage) rateforthenodecorrespondingtothedemandandrevocabletime

band,theEuriborrateformaturitiesshorterthan12months,and interestrateswap(IRS)ratesformaturitieslongerthan,orequal to,1year,inlinewithcurrentbankingpractices.

Thecharacteristicsanddynamicsofthekeyratetermstructure overtimehaveastrongimpactonbankriskexposure.Table1shows thatthetermstructureofourkeyratesbecomessteeperand expe-riencesadownwardshiftovertime,whichmakestheapplication ofthenon-negativityconstraintmorelikelytooccur.

Table 2 shows the main descriptive statistics of the key ratesannualchangesobservedoverthe2009–2013period,with December31st,2013asevaluationdatet.Asexpected,short-term aremorevolatilethanlong-termkeyrates.Furthermore,the neg-ativekurtosiscoefficientsconfirmthatthedistributionsofinterest ratechangesgeneratedthroughtheoverlappingtechniquedonot sufferfromthefattailissue.Theseresultsareconfirmedfortherest oftheevaluationdatesandareavailableuponrequest.

Weusebothregulatoryandinternalmethodologiestomeasure IRRBBof130Italiancommercial,savings,andcooperativebanks. Onaverage,oursampleincludesbothbankinggroupsand inde-pendentbanks.Table3showstheaveragevaluesofoursample banks’cashassetsandoff-balancesheetlongpositionsinPanelA andcashliabilitiesandoff-balancesheetshortpositionsinpanelB, respectively.Itisworthconsideringthattheoff-balancesheet posi-tionsincludehedgingderivatives,suchasinterestrateswaps,and theoptionalitiesembeddedinsomefinancialcontracts,namelythe floorsandcapsassociatedwithfloating-rateloans.

Averagecustomerloansrepresentapproximately75percentof totalon-andoff-balancesheetpositions;alargeproportionofthese loans,equalto65percentoftotalon-andoff-balancesheet posi-tions,are floating-rateloans andare slottedin short-termtime bandsbecause,following theregulatoryallotmentcriteria, they aretreatedasshort-termloans.Non-maturitydepositsrepresent around37percentoftotalon-andoff-balancesheetpositions, con-firmingtheirimportancefromarisk-managementstandpointand therelevanceoftheItalianbankmaturitytransformationfunction. Overall,debtsecuritiesconstituteabout33percentoftotalon-and off-balancesheetpositions.Specifically,securitieswithamaturity shorterthanoneyearrepresentalmost22percentoftotalon-and off-balancesheetpositionsandmainlyconsistoffloating-rate secu-rities.Overtheentireinvestigationperiod,bothlongandshort on-andoff-balancesheetpositionsaccountonaverageforlessthan10 percentofthetotal,beingequalto9.35percentand9.85percent, respectively.

Oursamplemostlyincludesbanksactiveatprovincial, inter-provincial, and regional levels. Consistent with their limited geographical scopesof activity,Table4shows that theaverage totalassetsofoursampleintermediaries,calculatedovertheentire investigationperiod,amounttoalmostD2,378million,withaclear increasingtrendfromD1,661millionin2006toD2,909millionin 2013.Furthermore,itisinterestingtonotethat,onaverage,about 88percentoftheintermediariesthatweexamine(115banksoutof 130)havetotalassetslowerthanD3.5billion,whichisthe thresh-oldbelowwhichtheyareidentifiedbyprudentialsupervisionas so-called“minorintermediaries.”Intermsofregulatorycapital,we alsofindevidenceofagrowingpath,evenifataslowerpacerelative tototalassets,fromavalueofD179millionasofDecember2006 toapproximatelyD254millionattheendof2013.Finally,despite theiroverallsmallsize,ourbanksmakeheavyuseofderivatives; onaverage,overtheentiresampleperiod,almost87percent(114 banksoutof130)arederivativeusers.Nevertheless,mostlikely becauseoftheirlimitedsize,theirhedgingpracticesappeartobe relativelyhomogeneous,consistingmainlyinusinginterestrate swapsagainsttheriskassociatedwithissuedbondsand amortiz-inginterestrateswapsspecificallytohedgetheriskpotentially arisingfromfixed-ratemortgageloans.

(9)

Table3

Cashassetsandliabilitiestermstructureandoff-balancesheetpositions.

PanelA:cashassetstermstructureandoff-balancesheetlongpositions

2006 2007 2008 2009 2010 2011 2012 2013 Avg.peryear

Debtsecuritieswithmaturity shorterthan1year

8.56% 8.28% 10.95% 11.45% 10.99% 10.48% 12.10% 14.81% 10.95%

Debtsecuritieswithmaturity between1yearand5years

3.32% 2.48% 1.23% 1.74% 1.67% 2.61% 5.55% 7.06% 3.21%

Debtsecuritieswithmaturity longerthan5years

1.24% 1.04% 0.60% 1.23% 1.46% 1.77% 3.32% 4.43% 1.89%

Loanswithmaturityshorter than1year

65.58% 70.76% 70.64% 66.93% 67.07% 63.47% 59.54% 56.17% 65.02%

Loanswithmaturitybetween 1yearand5years

6.15% 3.86% 4.15% 5.64% 4.93% 6.34% 5.61% 6.27% 5.37%

Loanswithmaturitylonger than5years

5.52% 3.66% 4.53% 5.76% 3.47% 4.20% 2.93% 2.92% 4.12%

Totalcashassets 90.37% 90.80% 92.10% 92.75% 89.59% 88.87% 89.05% 91.66% 90.65%

Off-balancesheetlong positions

9.63% 9.20% 7.90% 7.25% 10.41% 11.13% 10.95% 8.34% 9.35%

PanelB:cashliabilitiestermstructureandoff-balancesheetshortpositions

2006 2007 2008 2009 2010 2011 2012 2013 Avg.peryear

Non-maturitydeposits 38.38% 37.02% 37.18% 40.36% 38.84% 36.07% 32.12% 34.21% 36.77%

Othercurrentaccounts 8.70% 8.40% 7.77% 8.20% 8.97% 8.79% 9.84% 7.85% 8.56%

Termdepositsandotherfunds withmaturityshorterthan 1year

11.86% 9.92% 7.52% 4.99% 5.52% 8.61% 14.99% 18.45% 10.23%

Termdepositsandotherfunds withmaturitybetween1year and5years

1.93% 1.14% 0.43% 0.37% 0.34% 1.39% 1.65% 2.17% 1.18%

Termdepositsandotherfunds withmaturitylongerthan5 years

0.35% 0.20% 0.17% 0.17% 0.22% 0.18% 0.18% 0.19% 0.21%

Debtsecuritieswithmaturity shorterthan1year

20.47% 25.47% 29.74% 28.33% 21.69% 19.40% 15.17% 13.96% 21.78%

Debtsecuritieswithmaturity between1yearand5years

7.40% 7.64% 8.60% 9.46% 12.81% 13.39% 14.07% 12.96% 10.79%

Debtsecuritieswithmaturity longerthan5years

0.66% 0.53% 0.39% 0.45% 0.78% 0.76% 0.67% 0.72% 0.62%

Totalcashliabilities 89.76% 90.32% 91.81% 92.33% 89.18% 88.59% 88.69% 90.51% 90.15%

Off-balancesheetshort positions

10.24% 9.68% 8.19% 7.67% 10.82% 11.41% 11.31% 9.49% 9.85%

Thistableshowstheaveragevaluesofoursamplebanks’cashassetsandoff-balancesheetlongpositionsinPanelAandcashliabilitiesandoff-balancesheetshortpositions inPanelB,respectively.Dataaretakenfrombanks’balancesheets,andrefertoDecember31stofeachyearincludedinthe2006–2013period.Resultsareexpressedin percentageoftotalon-andoff-balancesheetpositions.

Table4

Samplebanks’sizeandderivativesusage.

2006 2007 2008 2009 2010 2011 2012 2013 Avg.peryear

Avg.totalassets(Dmillion) 1,660.98 1,999.51 2,044.52 2,209.73 2,505.39 2,704.34 2,994.21 2,908.90 2,378.45

Avg.regulatorycapital(D million)

179.03 195.12 218.82 234.42 247.18 262.35 266.62 254.23 232.22

%ofbankswithtotalassets ≤D3.5bn.

91.54% 90.00% 90.00% 87.69% 87.69% 87.69% 86.15% 86.15% 88.37%

%ofderivativeusersbanks 79.23% 84.62% 85.38% 85.38% 87.69% 91.54% 92.31% 93.08% 87.40%

Inthefirstandsecondrows,thistableshowstheaveragetotalassetsandregulatorycapitalofoursamplebanks,respectively(bothdenotedinmillioneuros).Inthethird andfourthrows,thetableshowsthepercentageofbankswithtotalassetslowerthanD3.5billionandthepercentageofbanksusingderivatives.DatarefertoDecember 31stofeachyearincludedinthe2006–2013periodandaretakenfrombanks’balancesheets.

6.2. Thenatureofbanks’interestrateriskexposure

Italiancommercialbankstypicallyfundlong-termassetswith short- and medium-term liabilities. Therefore, if interest rates moveup(down),theireconomicvalueshoulddecrease(increase) because,ceterisparibus,thereduction(increase)intheeconomic valueoftheirlong-termassetsshouldbelarger,inabsolutevalue, than the decrease observed for theirshort- and medium-term liabilities.AdoptingBCBS terminology, ifinterest rates increase (decrease),thesumofthepositivenetweightedpositionsishigher (lower) than the absolute value of the negative net weighted positions.Wedefineintermediariescharacterizedbythistypeof exposureasasset-sensitivebanks.

Nevertheless,bankriskexposurederivesfromtheinteraction of:i)thetermstructureofthenetpositions,whichistheresultof banks’ALMandcommercialstrategies;ii)theregulatoryallotment criteria,and iii)theactuallyappliedscenarioofannualchanges inthekeyrates,whichdependsontheadoptedmethodologyand ontheapplicationofthenon-negativityconstraint.Therefore,we alsofindbanksexposedtodecreasinginterestrates(alsotermed “liability-sensitive”)andrisk-neutralbanks,i.e.,thosethat expe-rience anincreasein theireconomic valuewhetherthecurrent standardizedshockisappliedupordown.

If interest rates decrease (increase), liability-sensitive banks experienceadecrease(growth)intheireconomicvaluebecause, ceteris paribus,theincrease (decrease)intheeconomicvalueof theirlong-termassetsislower,inabsolutevalue,thantheincrease

(10)

(decrease)observedfortheirshort-andmedium-termliabilities. Shouldinterestratesdecrease(increase),thesumofthepositive netweightedpositionswouldbehigher(lower)thantheabsolute valueofthenegativenetweightedpositions.

Based onthecalculation mechanism of therisk indicator, a bankisriskneutralif,inthecase ofapositive(negative) inter-estrateshock,thedecrease(increase)initsassetseconomicvalue islower(higher),inabsolutevalue,thanthedecrease(increase)in itsliabilities,thusdetermininganoverallincreaseinEVE.Inother words,usingBCBSterminology,whetherinterestratesincreaseor decrease,thesumofthepositivenetweightedpositionsislower thanthatofthenegativenetweightedpositionstakeninabsolute value.

Table5showsthat,duringourinvestigationperiod,outof130 institutions,onaverage40banksperyearareasset-sensitiveunder theparallelshiftsmethodandthatnumberincreasesto42ifwe measureIRRBBthroughtheMonteCarlosimulations.Ifweusethe percentilesmethodandhistoricalsimulations,thenumberofbanks exposedtoanincreaseininterestratesisaround25and27, respec-tively.Asfarasexposuretodecreasinginterestratesisconcerned, theaveragenumberofbanksperyeargoesfromapproximately76 undertheparallelshiftsmethod,toaround103forhistorical sim-ulations.Roughly13and21banksperyeararerisk-neutralunder theparallelshiftsandpercentilesmethods,respectively;however, wefindnoinstanceofrisk-neutralityforeithertheMonteCarloor thehistoricalsimulations.

Theapplicationofthenon-negativityconstraint,whichismore likelywheninterestratesarelow,isthemainfactorunderlying therisk-neutralityphenomenonforbothparallelshiftsand per-centilesmethods.Becauseofthenon-negativityconstraint,undera scenarioofdecreasinginterestrates,thereductioninabank’sEVE associatedwiththenegativenetpositionsofmedium-termtime bandsisnotsufficienttooffsettheincreaseintheEVEattributable tothepositivenetpositionsoflong-termtime bands.6 The

sig-nificantincreaseinthenumberofrisk-neutralbanksinthelast two years of oursample period,when interest rates are lower thanbefore,confirmsthisinterpretation.Nevertheless,underthe percentilesmethod,thefewcasesofriskneutralityobservedon December31stoftheyears2006–2008donotresultfrom appli-cationofthenon-negativityconstraint,whichdoesnotcomeinto play.Inthosecases,riskneutralitystemsfromthecombinedeffect ofpeculiarscenariosofchangesinthekeyrates,ontheonehand, andspecifictermstructuresofbanks’netpositionsacrossthetime bandsoftheregulatorymaturityladder,ontheotherhand.

Regardinghistorical andMonteCarlosimulations,wedo not observerisk-neutralitycases.Thesesimulationsmakeuseof mul-tiplescenarios,namelyaround1200scenariosfor thehistorical simulations,givenbythesumoftheapproximately240 scenar-iosperyearobservedoverthe5-yearperiodacrosswhich,based onthecurrentregulatoryframework,annualchangesinthekey rateshavetobemeasured, and10,000scenariosfor theMonte Carlosimulations.Withregard tothelatter,theprocessto gen-eratethe10,000scenariosofchangesinthekeyratesrequiresus toselectscenariosfromthosethatdonotdrivethekeyrateterm structurebelowzero,thusfulfillingthenon-negativityconstraint

6 Intheabsenceofthenon-negativityconstraint,withregardtotheparallelshift,

byapplyinga−200bpparallelshock,abank,formerlyexposedtodecreasinginterest rates,wouldexperienceareductioninitseconomicvalueequal,inabsolutevalue,to theincreaseassociatedwitha+200bpparallelshock.Nevertheless,duetothe non-negativityconstraint,themagnitudeofthereductioninthebankeconomicvalue associatedwithnegativenetpositionswouldbelowerandnotenoughtooffsetthe increaseintheeconomicvaluegeneratedbythepositivenetpositions.Underthe percentilesmethod,wewouldnothaveasymmetricriskexposurebyapplyingthe 1stand99thpercentilesinabsenceofthenon-negativityconstraint,but,inessence, itsfinaleffectwouldbethesameasthatjustdescribedfortheparallelshiftsmethod.

byconstruction.Inthehistoricalsimulations,despiteapplication ofthenon-negativityconstraint,contrarytowhatoccurswiththe regulatorymethodologies,thekeyratesofthetermstructureare freetomoveindifferentdirectionswithinthesamescenario.The highnumberandthedifferentnature,intermsofsignand magni-tudeofinterestratechanges,ofthescenariosusedtogeneratethe distributionsoftheriskindicatorsminimizestheprobabilitythata bankwillberiskneutral.

6.3. Theon-andoff-balancesheetcomponentsofbankinterest rateriskexposure

Oursamplebankscarryoutahomogeneoushedgingactivity throughbasicinterestratederivatives,suchasinterestrateswaps. BasedontheapproachofEspositoetal.(2015),inordertodetect whetherandtowhatextentItalianbanksusederivativestomanage IRRBB,wehavedecomposedtheriskindicatorobtainedthrough Eq.(1) intotwo components,basedontheon-and off-balance sheetitems,respectively.However,itisimportanttopointoutthat, ashighlightedaboveinSection6.1,theoff-balancesheet compo-nentincludesnotonlyhedgingderivatives,butalsofloorsandcaps embeddedinfloating-rateloans.

Basedonthecurrentregulatorytreatmentoftheseoff-balance sheetitems,theireffectonbankriskexposureisasfollows. Amor-tizinginterest rateswapstransform a fixed-rate mortgageinto a floating-ratemortgage.Therefore, ceteris paribus,theyreduce theaveragematurityofabank’sassetsandaccordinglydecrease (increase)theexposureofanoriginallyasset-sensitive (liability-sensitive)creditinstitution.Interestrateswapschangeafixed-rate issuedbondintoafloating-ratebondand,therefore,reducethe averagematurityofabank’sliabilities.Consequently,theydecrease (increase)theexposureofacreditinstitutionformerlyexposedto decreasing(increasing)interestrates.Finally,concerningthecaps andfloorsembeddedintofloating-ratemortgageloans,aportionof theseloans,basedontheoption’sdelta,becomesfixed-rateandan increaseintheaverageassetmaturityoccurs.Thus,ceterisparibus, theriskexposureofaformerlyasset-sensitive(liability-sensitive) bankincreases(decreases).

Foreach ofthefourmethodologiesdiscussedabove,Table6 shows our sample banks’ average overall risk exposure (see columnslabeled“Overall”),measuredthroughtheriskindicatorof Eq.(1),anditsbreakdownintheon-andoff-balancesheet com-ponents(see thecolumnslabeled“On” and“Off,”respectively). Overall,IRRBB is adequatelymanaged byItalian banks.Table6 showsthat the20 percentcriticalthresholdsetbytheBCBSis, onaverage,byfarhigherthanoursamplebanks’riskindicators.7

Theaverageriskindicatorcalculatedovertheentiresampleperiod forasset-sensitivebanksis10.37percentundertheparallelshifts method,versus6.87percentfortheMonteCarlosimulations,4.73 percentforthepercentilesmethodand5.98percentforthe histor-icalsimulations.Banksexposedtoincreasinginterestratesshow higherriskindicatorsundertheparallelshiftsmethodbecauseof theimpactofthe+200-bpshockonthenetpositionsoflong-term timebands.Thedistanceintermsofriskindicatorsamongthe dif-ferentmethodsshrinksforbanksexposedtodecreasinginterest rates,whoseaverageriskindicatorsgofrom5.32percentforthe MonteCarlosimulationsto7.54percentforthehistorical simula-tions.

Whenweconsiderthebreakdownoftheoverallriskexposure betweentheon-andoff-balancesheetcomponents,wefind evi-denceofdifferentbehaviorbyoursamplebanks,dependingonthe

7Thehighestsinglevalue(notshownintheTable)is15.20%,whichisobserved

onDecember31st,2006fortheparallelshiftsmethodandreferstoasset-sensitive banks.Datareferringtoeachsingleyearareavailableuponrequest.

(11)

Table5

Samplebanks’typesofinterestrateriskexposure:numberofbanksbymeasurementmethodology.

Parallelshiftsmethod Percentilesmethod Historicalsimulations MonteCarlosimulations

t(evaluationdate) A L RN A L RN A L A L December31st2006 47 83 0 41 87 2 40 90 40 90 December31st2007 32 98 0 23 105 2 25 105 23 107 December31st2008 23 107 0 15 107 8 20 110 23 107 December31st2009 31 94 5 18 111 1 17 113 43 87 December31st2010 21 105 4 11 119 0 10 120 20 110 December31st2011 38 78 14 27 96 7 18 112 35 95 December31st2012 60 20 50 32 20 78 43 87 73 57 December31st2013 71 25 34 36 27 67 42 88 77 53

Averagen.ofbanksperyear 40.38 76.25 13.38 25.38 84.00 20.63 26.88 103.13 41.75 88.25

Thistableshowsoursamplebanks’typeofriskexposurebasedonthemethodologyusedtomodelinterestrateshocks.Columns1–3refertotheparallelshiftsmethod, columns4–6tothepercentilesmethod,columns7–8tothehistoricalsimulationsandcolumns9–11totheMonteCarlosimulations.Foradescriptionoftheparallelshifts andpercentilesapproaches,seeSection3.ForadescriptionofthehistoricalandMonteCarlosimulationsapproaches,seeSection4.

Aindicatesbanksexposedtoincreasinginterestrates(asset-sensitivebanks),Lindicatesbanksexposedtodecreasinginterestrates(liability-sensitivebanks)andRNindicates risk-neutralcreditinstitutions.

InterestrateriskexposureiscalculatedaccordingtotheBankofItaly’sdurationgapapproachandisexpressedaspercentageofregulatorycapital,asshowninEq.(1). Non-maturitydepositshavebeenallottedwithin5yearsaccordingtothecriteriondefinedinBankofItaly(2013)anddataaretakenfromDatastreamandbanks’balance sheets.

Table6

Samplebanks’averageriskindicators:breakdownbymeasurementmethodologyandtypeofinterestrateriskexposure.

Parallelshiftsmethod Percentilesmethod Historicalsimulations MonteCarlosimulations

Overall On Off Overall On Off Overall On Off Overall On Off

A 10.37% 6.82% 3.55% 4.73% 2.95% 1.78% 5.98% 4.66% 1.32% 6.87% 5.38% 1.49%

L 6.62% 7.93% −1.31% 7.11% 8.61% −1.51% 7.54% 8.93% −1.39% 5.32% 6.22% −0.91%

Thistableshowstheaverageriskindicatorsofoursamplebanksoverthe2006–2013period,basedonthemethodologyusedtomodelinterestrateshocks.

Columns1–3refertotheparallelshiftsmethod,columns4–6tothepercentilesmethod,columns7–9tothehistoricalsimulationsandcolumns10–12totheMonteCarlo simulations.Foradescriptionoftheparallelshiftsandpercentilesmethods,seeSection3.ForadescriptionofthehistoricalandMonteCarlosimulationapproaches,see Section4.

Aindicatesbanksexposedtoincreasinginterestrates(asset-sensitivebanks)andLindicatesbanksexposedtodecreasinginterestrates(liability-sensitivebanks).Overall indicatestheoverallriskindicatorbasedonboththeon-andoff-balancesheetitems,Onindicatestheriskindicatorbasedontheonlyon-balancesheetitems,andOff indicatestheriskindicatorbasedontheonlyoff-balancesheetitems.Overallriskindicatorshavebeendecomposedintheon-andoff-balancesheetcomponentsfollowing Espositoetal.(2015).

InterestrateriskexposureiscalculatedaccordingtotheBankofItaly’sdurationgapapproachandisexpressedaspercentageofregulatorycapital,asshowninEq.(1). Non-maturitydepositshavebeenallottedwithin5yearsaccordingtothecriteriondefinedinBankofItaly(2013)anddataaretakenfromDatastreamandbanks’balance sheets.

natureoftheirriskexposure.Onaverage,fortheliability-sensitive banks,theon-andoff-balancesheetcomponentsshowapositive andanegativesign,respectively,whereas, thetwocomponents havethesamepositivesignfortheasset-sensitivebanks.Basedon theseresults,liability-sensitivebanksseemtousederivativesasa hedgingtooltooffseton-balancesheetinterestraterisk,whichis consistentwiththeoveralldecreasingtrendexperiencedby mar-ketinterestratesovertheinvestigationperiod.Ontheotherhand, asset-sensitivebanksseemtousetheiroff-balancesheetpositions toenhancethegainsassociatedwiththedecreasinginterest-rate trend.8

Thispreliminaryevidencedisregardsa substantial degreeof heterogeneity among intermediaries concerning their exposure toIRRBB. Therefore, in Table7 we splitour sample into three groups, according to the sign of their risk exposure over the entire2006–2013sampleperiod:i)banksthatwerealways asset-sensitive;ii)bankscontinuouslyexposed todecreasinginterest rates;iii)banksthatvariedthesignoftheirexposureatleastonce, alsotermed“otherbanks.”

Overall,Table7showsthatthevastmajorityofour130 sam-plebanksexperiencedachangeinthenatureoftheirinterestrate riskexposureatleastonceoverthetimehorizonweconsider.If

8Itisworthunderliningthat,tosomeextent,thetreatmentofthecapsandfloors

optionsembeddedintofloating-rateloanscanhelpexplaintheevidence refer-ringtotheasset-sensitiveintermediaries,sinceitdeterminesgreatersensitivity toincreasinginterestrates.

IRRBBismeasuredthroughthehistoricalandMonteCarlo simula-tions,weobserveahighernumberofliability-sensitivebanks,i.e., 41fortheformerand20forthelatter,bothhigherthanthefigure observedfortheparallelshiftsandpercentilesmethods. Concern-ingthebreakdownoftheoverallriskexposureintotheon-and off-balancesheetcomponents,onaverage,theresultsreportedin Table7areconsistentwiththoseshowninTable6;theon-and off-balancesheetcomponentshavethesamepositivesignforthe asset-sensitivebanksandanoppositesignfortheintermediaries exposedtodecreasinginterestrates.

7. Backtestingresults

Table 8 shows theaverage resultsacrossthe wholesample period of our backtesting framework applied tothe regulatory andinternalmethodologies.PanelAreportstheaveragefrequency scoresbasedontheaccuracyfunction(8),whereasPanelsBand Cshowtheaverageseverityscoresinthecaseofunderestimation andoverestimationoftheactualexpostriskexposure,thatare cal-culatedthroughtheaccuracyfunctions(9)and(10),respectively. Finally,PanelDpresentstheaverageproximityscoresbasedon theaccuracyfunction(11).Theexanteriskexposureis alterna-tivelyassessedthroughtheparallelshiftsmethod(Columns2and 3),thepercentilesmethod(Columns4and5),thehistorical simula-tions(Columns6and7)andtheMonteCarlosimulations(Columns 8and9).Ineachpanel,ourresultsseparatelyrefertothewhole sampleandtothetwosub-groupsofasset-sensitiveand liability-sensitivebanks.Table8alsodistinguishesbetweenthescoresbased

(12)

Table7

Samplebanks’averageriskindicators:breakdownbymethodologyandconstanttypeofinterestrateriskexposure.

Parallelshiftsmethod Percentilesmethod Historicalsimulations MonteCarlosimulations

N.ofbanks Overall On Off N.ofbanks Overall On Off N.ofbanks Overall On Off N.ofbanks Overall On Off

A* 6 13.22% 11.35% 1.87% 3 5.58% 5.47% 0.11% 2 5.97% 5.83% 0.14% 5 8.21% 7.00% 1.21%

L* 9 10.13% 10.83% −0.70% 9 11.05% 11.95% −0.90% 41 8.22% 9.48% −1.25% 20 6.18% 7.09% −0.91%

O 115 7.12% 6.78% 0.34% 118 5.77% 6.57% −0.80% 87 6.87% 7.61% −0.74% 105 5.62% 5.73% −0.10%

Thistableshowstheaverageriskindicatorsofoursamplebanksoverthe2006–2013period,basedonthemethodologyusedtomodelinterestrateshocks.

Columns2–4refertotheparallelshiftsmethod,columns6–8tothepercentilesmethod,columns10–12tothehistoricalsimulationsandcolumns14–16totheMonteCarlo simulations.Foradescriptionoftheparallelshiftsandpercentilesmethods,seeSection3.ForadescriptionofthehistoricalandMonteCarlosimulationapproaches,see Section4.

A*indicatesbanksexposedtoincreasinginterestrates(asset-sensitivebanks),overtheentiresampleperiod,L*indicatesbanksexposedtodecreasinginterestrates (liability-sensitivebanks)overtheentiresampleperiod,andOindicatesbankswhosenatureofinterestrateriskexposurechangedatleastonceoverthesampleperiod.

Overallindicatestheoverallriskindicatorbasedonboththeon-andoff-balancesheetitems,Onindicatestheriskindicatorbasedontheonlyon-balancesheetitems,andOff indicatestheriskindicatorbasedontheonlyoff-balancesheetitems.Overallriskindicatorshavebeendecomposedintheon-andoff-balancesheetcomponentsfollowing Espositoetal.(2015).

InterestrateriskexposureiscalculatedaccordingtotheBankofItaly’sdurationgapapproachandisexpressedaspercentageofregulatorycapital,asshowninEq.(1). Non-maturitydepositshavebeenallottedwithin5yearsaccordingtothecriteriondefinedinBankofItaly(2013)anddataaretakenfromDatastreamandbanks’balance sheets.

Table8

Backtestingresults.

PanelA:averagefrequencyscoresb

ParallelShifts Percentilesmethod Historicalsimulationsa MonteCarlosimulationsa

Overall On Overall On Overall On Overall On

W 19.75 17.50 22.50 18.63 11.13 8.38 11.00 10.13

A 3.25 5.75 5.75 7.13 6.50 5.25 3.25 3.38

L 10.88 7.38 7.25 5.13 4.63 3.13 7.75 6.50

PanelB:averageseverityscoresintheunderestimationcasec

ParallelShifts Percentilesmethod Historicalsimulationsa MonteCarlosimulationsa

Overall On Overall On Overall On Overall On

W 0.68% 0.76% 0.76% 0.83% 0.42% 0.50% 0.27% 0.23%

A 0.70% 0.89% 0.89% 0.82% 0.46% 0.38% 0.25% 0.29%

L 0.53% 0.59% 0.53% 0.64% 0.31% 0.34% 0.24% 0.19%

PanelC:averageseverityscoresintheoverestimationcased

ParallelShifts Percentilesmethod Historicalsimulationsa MonteCarlosimulationsa

Overall On Overall On Overall On Overall On

W 6.60% 6.73% 4.72% 5.41% 5.22% 6.06% 3.67% 3.93%

A 9.73% 9.54% 3.82% 4.23% 4.78% 4.86% 5.47% 5.19%

L 4.34% 5.39% 4.81% 5.79% 5.34% 6.35% 3.06% 3.59%

PanelD:averageproximityscorese

ParallelShifts Percentilesmethod Historicalsimulationsa MonteCarlosimulationsa

Overall On Overall On Overall On Overall On

W 6.02% 6.29% 4.62% 5.28% 5.17% 5.96% 3.55% 3.82%

A 9.08% 8.28% 4.07% 3.98% 4.65% 4.57% 5.28% 4.93%

L 4.19% 5.22% 4.74% 5.72% 5.27% 6.31% 3.01% 3.55%

Thistableshowstheaverageresultsoverthe2006–2013sampleperiodofourbacktestingprocedureappliedtothemethodologiesusedtomodelinterestrateshocks. Columns2–3refertotheparallelshiftsmethod,columns4–5tothepercentilesmethod,columns6–7tothehistoricalsimulationsandcolumns8–9totheMonteCarlo simulations.Foradescriptionoftheparallelshiftsandpercentilesmethods,seeSection3.ForadescriptionofthehistoricalandMonteCarlosimulationapproaches,see Section4.

Windicateswholesamplebanks,Aindicatesbanksexposedtoincreasinginterestrates(asset-sensitivebanks),andLindicatesbanksexposedtodecreasinginterestrates (liability-sensitivebanks).

Overallindicatestheoverallriskindicatorbasedonboththeon-andoff-balancesheetitems,Onindicatestheriskindicatorbasedontheonlyon-balancesheetitems,andOff indicatestheriskindicatorbasedontheonlyoff-balancesheetitems.Overallriskindicatorshavebeendecomposedintheon-andoff-balancesheetcomponentsfollowing Espositoetal.(2015).

InterestrateriskexposureiscalculatedaccordingtotheBankofItaly’sdurationgapapproachandisexpressedaspercentageofregulatorycapital,asshowninEq.(1). Non-maturitydepositshavebeenallottedwithin5yearsaccordingtothecriteriondefinedinBankofItaly(2013)anddataaretakenfromDatastreamandbanks’balance sheets.

aThescoreiscalculatedbycomparingthe99thpercentileoftheexanteriskindicatordistributionwiththeexpostriskindicator. b Scoresarecalculatedbyapplyingthescorefunction(5)andtheaccuracyfunction(8).

c Scoresarecalculatedbyapplyingthescorefunction(5)andtheaccuracyfunction(10). d Scoresarecalculatedbyapplyingthescorefunction(5)andtheaccuracyfunction(11). eScoresarecalculatedbyapplyingthescorefunction(5)andtheaccuracyfunction(12).

onoverallriskexposureandthoseassociatedwithriskexposure basedononlytheon-balancesheetcomponents.Inthisway,we

cancapturetheeffectofderivativesonbankriskexposurewithin ourbacktestingframework.

Riferimenti

Documenti correlati

This review aims to: (1) examine developmental trajectories of multisensory processes both in animal and human models; (2) highlight effects of a multisensory-based

Tra i diversi metodi di valutazione oggi più usati (v anche Grabe &amp; Low, 2002 e Rouas &amp; Farinas, 2004), quello di Ramus et alii (1999) ha proposto il ricorso a tre

Holistic descriptors of body movements have been proposed by a few authors. Among the most notable solutions, Bobick et al. [ 17 ] proposed motion his- tory images and their

The San Precario network, in particular, has been able to elaborate a collective identity and political action using irony as a form of postmodern resistance and proposing a

To do this, we use Spanish data to calibrate the general equilibrium model with banks introduced in Nuño and Thomas (2013). The latter work presents a model in which unregulated

Genes differentially expressed in the normal and Dlx5-/- developing murine olfactory system, prioritized using human KS-causing genes and the coexpression network.. regulator

We find that under certain conditions, the resulting current can excite a non-resonant instability that forces particles with energy below some critical value to be confined within

We combine constraints from a catalog of 909 weakly lensed galaxies and 39 multiply-imaged sources comprised of 114 multiple images, including a system of multiply-imaged candidates