We have simplified and consolidated the financial statements to facilitate the modeling process. Part 2: Project the Revenue Expenses
Financial Modeling Mastery. – Certification Quiz Questions. Module 5 – 3-Hour 3-Statement Modeling Test and Debt vs. Equity Case Study.
Types of financial models. Financial. Models. Three. Statement. Model. DCF Model 3. Plan. 4. Integrity. 5. Model Testing. •Use test data to.
Directions: You will hear a question or statement and three responses spoken in English. They will not be printed in your test book and will be spoken only
and Wollmershäuser T.: 2012
3. Margin period of risk and cash flows. 139. 4. Collateral modelling (d) for PD models covering exposures to financial institutions: business model.
3. 2. Supplementary validation reporting on credit risk The ECB is conscious of the limitations of statistical tests and the importance of the.
11 mai 2017 3. Conclusion. 68. Chapter 6 Top-down modelling for market risk ... Financial sector stress tests have proved to be an important tool for ...
Students will build financial models in each class with the Professor. Build a three statement model including revolver and debt schedules.
identify manage and control the risks inherent in the use of stress test models. 1.2 This SS is relevant to PRA authorised banks
![Working Paper Series - Banking euro area stress test model Working Paper Series - Banking euro area stress test model](https://pdfprof.com/EN_PDFV2/Docs/PDF_2/99_2ecb_wp2469~a139d2f5cd_en.pdf.jpg)
99_2ecb_wp2469~a139d2f5cd_en.pdf
Working Paper Series
Banking euro area stress test model
Katarzyna Budnik, Mirco Balatti,
Ivan Dimitrov, Johannes
Groß,
Michael Kleemann, Tomas Reichenbachas,
Francesco Sanna, Andrei Sarychev,
Ƽ, Matjaz Volk
Disclaimer: This paper should not be reported as representing the views of the European Central Bank
(ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB.
No 2469 / September 2020
Abstract
The BankingEuro AreaStress Test (BEAST)is alarge scalese mi-s tructuralmo delde- velopedtoassess theres ilienceof theeuro areabankingsystemfrom amacropr udenti al perspective.Themodel combine sthedynamicsof ahighnumber ofeuro areabanks with that ofthe euroarea economies.It re ectsbanks' heterogeneit yb yreplicating thestructure of theirbalance sheetsand protand lossaccoun ts.In themo del,banks adjusttheirassets, interestrates, andprot distributionin linewith theeconomi cconditions theyface. Bank responsesfeedbac kto themacroeconomic environmen taecting creditsupplyconditions. When appliedto astress testof theeuro areabanking system,the model reveals higher system-wide capitaldepletion thanthe analogousconstan tbalance sheet exercise. Keywords:macro stresstest, macropruden tialpolicy,bankingsector deleveraging,real economy-nancialsector feedback loop JEL Classification:E37, E58,G21, G28ECB Working Paper Series No 2469 / September 20201
Non-technicalsummary
A macroprudentialstresstest attemptsto informregulators andthe publicab outp otential risks originating inand propagatingfrom thenancial sector.Suc ha stresstest aimsto makesenseof bothcomplex balancesheets andmost likely bankreactions tonegativ eeconomicdevelopmen ts. Ultimately,it allows totimelyalertthe regulatorand themark etab outc api taland liquidity needs ofthe bankingsector . This paperintroduces theBankingEuroAreaStressT est(BEAST) model develop edfor the purposeof macroprudential stresstestingofthe euroarea bankingsector. Themo delincludes a macroeconomicblock for19euroareaeconomiesand are pre se ntation of91 systemically importantbankswith theirindividual balancesheets andprot andloss accounts. Themo del incorporatesthe interactions between banksandtherealeconomyand,b yfollo winga semi- structural design,can provide anarrativefor systemicrisk transmissionc hannels. Twofeaturesmak ethe BEASTparticularlysuitablefor macroprudential stresstesting. Banks inthe model canadjustthesize andcomp ositionof theirassets, alongwith interest rates ordividend pay-outs. Thereby,banks behaviouralresponsefunctions areestimated using historical bank-leveldatasets.This model featuredeviates fromtheso-cal ledconstan tbalance sheet assumptioninmost ofthe microprudential stress-tests.Sec ond, themodelincorp orates a feedbacklo opbetweenthe nancialsectorandthereale conomy .Tocounterbalancethe re- alisation ofcapital shortfallsan dthe deteriorationofassetqualit y, banksadjust theircredit supply.These reactions ofindividualbanks,once aggregated ona countr ylev el,translateinto a negativecreditsupply shoc kand addtotheinitialscenarioadv ersity . Toillustrate thew orkingof themodel,the BEASTis appliedto evaluatethep erformance of theeuro areabanking sectorunder theadv ersescenario ofthe 2018EBA supervisorystress test. Theadv ersescenariofor2018{2020 considereda perio dof prolongedsystemic stress,with a substantialdropin euroarea outputand inr eside ntial houseprices, alongwith risingunem- ploymentrates.By emphasizingthe feedback bet ween thebankingsectorand therealeconomy, the modelampliesthe severit yof thescenarioandrev ealshigher bankcapital depletionthan the originalEBA stresstest results. ECB Working Paper Series No 2469 / September 20202
1 Introduction
Financial intermediation,and bankingin particular,are highl ycomplex activities.Manycon - tracting decisionsare basedon trust,making themsusceptible todrastic ch angesin theface of newinformation (Morrisand Shin2008). Such changes may occasionallygive risetosharp reversalsin inv estors'actionsresultinginnancialdistress andreduced private sectorac cessto credit. Topreven tsuchmarket disruptionistheraisond' ^etreofthe macroprudentialregulation and macroprudentialstresstesting. A stresstest canmak esense ofcomplexbal anc esheets andbusiness models,uncovervul- nerabilities ofnancial institutions,and timelyalert regulatorsand themark etsab outcapital and liquidityneedsof institutions(Bo okstaber etal. 2014).Thefocusofthe microprudential regulator willb etominimisethe costto thetax-pa yer ofbailing outinsured deposits. Thecrit- ical nuanceisthat froma microprudential pe rspectiv e,theinstitutionsare treatedasisolated entitieswhen aggregatingthis hyp othetical costacrossbanks.Incontrast, themacroprudential regulator aimsto account forstressamplicationmec hanisms, such asdelev eragingandcredit crunches.A ccordingly,macroprudentialstresstests needtoackno wledgea broaderset ofin ter- actions betweenbanksandtheirp otential disruptive in uenceon therealeconomy(Claessens and Kose2017),whic hw ewillreferto asrealeconomy-n ancial sectorfeedbac klo ophereafter. This paperintroduces theBankingEuroAreaStressT est(BEAST) model develop edfor the purposeof macroprudential stresstestingofthe euroarea bankingsector. Themo delincludes a macroeconomicbloc kforthe19euroarea economies anda representation of91 systemically importantbankswith theirindividual balancesh eets andprot andl ossaccounts.Further, it captures thedynamic interdep endenciesbetweenaggregate andbank-levelvariablesincluding cross-borderspillo versandthefeedback loop bet weenthenan cialsectorand therealeconomy. The paperfocuses onthemodel version thathas beenappl iedtothemacropruden tialstress test of theeuro areabanking systemin 2018(Budnik etal. 2019)and recallsthe maintak eaw ays from thisexercise. The BEASThas tw oingredientsthatmak eitsuitablefor macroprudentialstresstesting. First, itincorp oratesbanks'behavioural reactionsto economicconditions.Amongthose,banks can adjustthe sizeand composition oftheir assets,canresetin terestrates onb othloans and deposits,and scaleup ordo wntheir dividendp ayments.This allows deviatingfrom theso- called constantbalancesheet assumptionwhic his commonlyapplied especiallyinb ottom-up ECB Working Paper Series No 2469 / September 20203 stress tests.Second, themo delincorporatesa feedback loopbet weenthe financialsectorand the realeconom y.Amacroeconomicsce nari oimpactsthequalit yofbankassetsleadingto changesin impairment lossesandtheadjustmen tsof capitalc har ges.Accordingly, anadv erse scenario lowersbankprofitabilit yandincreasesthe riskweighted exposu re amounts, viab oth channelslo weringthesolvencyle vel. Inordertoc ounterbalancetheresu ltin gcapital shortfalls with respecttoregulatory capitaltargets, banks adjust theircredit supply(intermsof lending volumesand interest rates).Aggregatedbanks"credit supplyresp onsestranslate into ac hange in aggregatecr editconditionsaddingup tothe initialscenario adversit y. Additionalstrength ofth emo delisitsabilit ytocapturethe heterogeneity ofthe euroarea banking system.The model incorporatesdetailsof balancesheetsandprofit andloss statements of individualbanks. Furthermore, banks"reactionsdepend ontheir own solvency,asset quality and profitability.Andlast,banks" beha viouralequations in volvearangeof thresholdswhic hare either identifiedeconometrically(e.g. atendency ofb anksto deleverage proportionately stronge r once theyhit theirregulatory limits)or deriv eddirectly fromthe regulation(e.g.Maximum Distributable Amountlimits).All threeele ments generatea highdegreeofheterogeneity in bank responses. The BEASTfalls bet weenmacroeconometricsemi-structuralmodels,used fore.g.inflation and outputforecasting, andheterogeneous agent models. Regardingitsse mi-struc tural design, the modelis amixtureofstructural anddata-driv enequations. Thelatter areinspired by theory andestimated employing macro-andmicro-economicdata andiden tificationtec hniques. Regarding itspro ximitytoheterogeneousagent models, theBEAST incorporatesthein forma- tion aboutmany individualbanksandaggregates bank-lev eloutcomes toarriv eat system-wide variablesthat enter themacrobloc k. The originalcontributionof thepaperis toprop osea comprehensivesemi-structuralsetup in thefield sofar dominatedb ysequen tialuse ofmultipleuncons olidated approaches. These sequentialapproac hesallowfor acombinationof several models thatoriginallywork ond ifferent levelsof granularit yandemphasizedifferenttransmission mechanisms. Examplesof modular frameworksinclude theBan kof Canada"sMacrofinancialRiskAs sessment Framew ork(MFRAF) (Fique 2017),th eBankofEngland RiskAssessmen tMo delof SystemicInstitutions (RAMSI) (Alessandri etal. 2009),the Bankof France frameworks by Bennanietal.(2017) andCamara et al.(2015) orin theECB Stress-Te st Analyti csforMacroprudentialPurp osesintheeuro area (STAMP ¿ )(Dees etal. 2017).The corew eaknessof amo dularapproach isits ingrained ECB Working Paper Series No 2469 / September 20204 inconsistency andlimited ability todescribeamplificationmechanisms.The conv ergenceinthese approachesis ach ievedbyiteratingtheworkflo w.Ourapproach hasthe ambition toencapsulate all relevantelementsof anamplificationmechanism (inour cas ereal economy-financialsystem feedbacklo op)inone,join tlysolv ed,system. Tothisend itresemble sthe proposalofKr znar and Matheson(2017). The paperisstructured asfollo ws.Section 2pro videsahigh-level intro ductionof themodel. Section 3d escribesthemacroblock.Section 4des crib estheidentificationof core beha vioural bank-levelequations. Additionally ,theidentificationofempirical equationsdescribing thesen- sitivityof bankassets tomacro economicd ev elopmentsisp lacedinAppendixA.Allremaining modeliden titiesarelistedin Appendix B.S ection 5elaborateson selectedresults oftheeuro area macroprudentialstresstest of2018 asan exampleof model application.Secti on6 concludes.
2 Highlev elmodeldescription
This chapterintroduces themainsegmentsofthe model andexplains theirinteractions.The BEAST worksona quarterlyfrequency and,along with thelogic ofsemi-structural modelling, it consistsof aset ofestimated beha viourale quations whosespecificationisinformed bytheory, and aset ofs tru cturalrelationships.
Macroblock
Country1Country2....Country19
Bank-level block
BBBB....BB
... ... ... .............
BBBB....BB
Creditsupply
Economicconditions
Figure 1:Basic modelstructure
The model canb ethoughtof asajoint representation of19 individual euroareaeconomies (macro block)andov er90 largesteuroareabanks(bank-level bloc k).The bank-level bloc kis based onthe templatessu bmittedb ybanksinthe EBAstress -testexerc ise in2018, andfo cusses on banklending, creditrisk andnet intere st income.Less emphasisisputonmodelling ofthe ECB Working Paper Series No 2469 / September 20205 trading book,marketand operationalrisk,and bankliabilities. Theestimation ofb ehavioural equations involvesmultipledatasetsinclud ingiBSI/iMIR (individualBalance SheetItemsand MFI InterestRateStatistics), COREP/FINREPinformation, and publicdatabases such as
Bankscopeand SNL.
The twoblocks areinterlinkedasillustrated inFi gure1.Economicconditionsin the19 euro areacoun triesaffectthequalit yof bankassets andcreditdemand.Bank lendingdecisions, aggregated ona country level,affect inturnthemacroeconomicoutlo ok.
2.1 Macroblo ck
CountryA
GDP
Imports
Exp. prices
Foreign
prices ROW
Exp.price
from RO W ...
Foreign
demand
CountryC
Imports
of countryC
CountryB Imports
of countryB
Exp.prices
from coun tryCFigure 2:Cross-countrytrade spillov ers Eacheconom yisrepresented by asetofmacro-financial variables,such asGDP ,inflation , house pricesor gov ernmentbondyields.Thedynamicsof thesevariablesar emo delledin a simplified fashionwith aV ectorA utoregressionmodel(V AR).Adetaileddescription ofthe underlying identificationstrategycan be foundin chapter3. Cross-bordertrade spi lloversareintroducedviaforeigndemand andprice variablesentering country-levelVARs. AsoutlinedinFigure 2foreign demandof acoun trydep endson import volumesof itstrading partners,while fore ignprices depend onexportpricesofother countries (in bothcasesas weigh tedwith thecorrespondingexport orimp ortshares). Beyond,themodel involvescountry-specific rest-of-the-world(ROW)v ariables. 11 The methodologyofin troducing tradespillo versfollowstheprop osalbyBudnik andR unstler(2018)that
allowsfor linkingcoun trylev elmodelsin tolinearizedmulti-coun trysystemsinan easilytractablew ay .Same
specicationoftrade spillov ersen tersaswelltheStress TestElasticities usedin theEuropeanBanking Authorities
stress testexercisessince 2011.T radeshares ofall euroareacoun triesare assumedto remainconstan t.ECB Working Paper Series No 2469 / September 20206
2.2 Bank-levelbloc k
Theasset sideof each bankbalancesheetis illustratedin Table 1.The BEASTmo delsthe evolutionof exposures tothenon-nancial corporate sector(NFC),households back edbyreal estate (HHHP)and householdfor consumption purp oses(HHCC),follow edbyexposures to sovereigns(SOV)an dthenanc ialsector(FIN).Exp osuresto thenon-nancialprivatesector are splitb ythecountry ofe xposure,whileexp osuresto sovereignandnancialsectors are aggregated acrossjurisdictions. Thedynamics ofother loans,equit yexp osuresand securitized portfoliosis exogenou s.
PortfolioModellingapproac h
Loans toNF C
geographical breakdownLoans toHHHP Loans toHHCC
Loans toSO Vaggregated
Loans toFIN Other loans
exogenousEquityexp osureSecuritized portfolio Table1: Schematicillus tration ofbanks' bankingb ooks
The modelcaptures
owsb etw eenthethree IFRS9assetimpairmentstages,i.e.performing, with increasedcredit risksince initialrecognition, andcredit-impaired, for each endogenous banking bookportfolio.Changes inassetquality arere ected inresp ective owsofpro visions which,aggregated up, entertheprot andlossstatement. Adetailed descriptionof theimpair- mentstages andthe loanloss pr ovisioning canb efoundinAppendicesB.1andB.2. Eachendogenous banking bookp ortfoliohasitsassigned riskweight. Total riskw eighted amountsare derived byaddingthe endogenouscreditriskexp osureamoun tsto exogen ousas- sumptions regardingthe evolution ofcapitalcharges relatedto market andoperationalrisk (see
AppendixB.7).
Foreac hendogenousbanking bookp ortfoliot
On thebank liability side,themodel tracks separatelyequit y,sight andtermdeposits (to corporatesNF Candhouseholds HH),wholesalefundingincluding interbank liabilitiesand debt securities (Table2).It isassumed thatthe structureof debtfun ding(dep osits,mark etfunding and debtsecurities) remainsconstan twhile thetotaldebtfunding changes in proportion tothe size ofbanks' assets. ECB Working Paper Series No 2469 / September 20207
Liabilityclass Modellingapproac h
Capital
Sightdep ositsHHgeographical breakdown
Sightdep ositsNF CTermdep ositsHH
Termdep ositsNFC
Wholesale fundingexogenous
Table2: Schematicillustration ofbanks" liabilities Foreac hbank,netprofit isbrok endo wnin toimpairmen tsarisingfromcredit risk,netinterest income performance,the devaluationofassets duetomark etrisk andnet-f ee-commissionincome. Bank dividendp ay-outsofbanksfollow anestimated beha viouralequationtakingin toaccoun t regulatory payoutrestrictions(seesub chapter 4.3and AppendixB.5foradditionalinformation on theprofit andloss account).
Economicconditions
Asset qualityFunding costsCredit demand
Credit
supply
Bankbehaviouralreactions
Rentention
of earningsVolume adjustments Price adjustmen ts
Non-linear
resp onse
Linear
resp onse
Regulatory
conditions
Capital
requiremen ts +
Combined
buerFigure 3:Schematici llustration ofmo deldynamicsfoc using onbankreactions The importantelementof theBEASTisthe relaxationof thestatic balancesheet assumption. As illustratedin Figure3, banksadjust the irloan volumes andinterestratesinresp onseto a givenscenario. The sebehaviouralreactions takeintoaccountchangesinbank assetqualit y, profitability,solvency ratewithrespec tto theirregulatory requirements,and theev olutionof sector andcoun tryspecificcredit demand.Inparticularloan volume adjustments inv olve a linear andnon-linear response tocapitalshortfalland changes inasset quality (seec hapter4). ECB Working Paper Series No 2469 / September 20208
2.3 Thefeedbac kloopEconomicconditions
& lastp eriodbank- leveloutcomes
Macroeconomic
shocks
Creditsupply shock
triggered by excessivedelev eraging
Reactionof banks:
deleveraging, profit accumulation, lendingmargins
Detoriationof bank
balance sheetsFigure 4:Schematicillustration ofmo del dynamicsfo cusingonthereal economy -banking sector feedbackloop Next tothe relaxationof thestatic balancesheet assumption, themo delincorp oratesthe impact ofbank lendingdeci sions ontherealeconomy. Itis achiev edb yaggregatingthenon-linear elementsof bankcredit supply responses, interpretedasexcessiv edeleveraging,and mapping these intoacredit supply shoc kaffectingthereal economyasillustratedinFigure 4. The mainmec hanismofthereal economy-financial sectorfeedbac klo opcanbe described in severalsteps asi llustr atedinFigure4.Atfirst,selected macroeconomic shoc ksaffect thereal economy.Ther esulti ngeconomicconditionsinfluencebankasse tqualit yandcreditdemand conditions. Followingthechanges ontheir balancesheets, banks areallo wedtorebalancetheir assets inorder torestore theirsolv encylev els. Theresultingcreditsupply shocksaddtothe initial macro-financialsho cksandchangethe economicoutlo okinthefollo wingp eriod s. ECB Working Paper Series No 2469 / September 20209
3 Macroeconomicmodule
The macroeconomicmo duleoftheBE ASTcan be describedasa reduced-formm ulti-country setup. Thedynamics ofs in gleeuroareeconomiesareestimatedwitha structuralpanel vector autoregressivemo del.TheVAR equationsare laterinterlinked viatrade spillov ers.
3.1 Empiricalapproac h
The estimationof themacro economic blockrestsona structuralpanelvectorautoregressive model(SPV AR)for19euro areacoun tries.F oreac heconom yi, themo delincludes11endoge- nous variablesinv ectorYi;tand 2exogenous variables invectorXi;t. Thereduced formpanel
VARhas thefollo wingform:
Y i;t=ci+pX j=1A i;jYi;t j+BiXi;t+i;t(1) whereciis av ectorofcountry-sp ecicin tercepts,Ai;jis acoun try-specicmatrixofautore- gressiveco ecientsforlagj,pcorrespondsto then umb eroflags(itisassumedthatp= 2)and i;tis av ectorofredu ced{f ormresiduals. The estimationrelies onthe Bay esianestimator proposedby Jaroc inski(2010).Theestima- tor allowsfordieren tV ARcoecients acrossunits(Ai;jandBi;j) butas sumesthattheyare drawnfrom adistribution withsimilar meanand variance . The structuralrepresen tationofthepanel VAR model isderiv edcombini ngzeroand sign impact restrictions.F ollowingthemethodologydev eloped byAriaset al.(2018)weidentify 8 structural shocksbasedonthe identication sc hemein Table3. Aggregate demand,aggre gatesupplyandmonetary policy shoc ksare identiedusinga standard setof res trictionsasinHristovetal. (2012).An aggregatedemand shockis described as asho ckthatmoves in ationand GDPinthesamedirection(and triggersan increasein short-term interestrates).An aggregatesupply shoc kmo ves in ationandGDPinoppos ite directions. Anaccommo dativemonetarypolicysho ck isre ectedinlo werinterestratesand an increase inin ation andoutp ut. The identicationofcredit supplys hoc ksfollo wsHristovetal.(2012),Barnettand Thomas (2013) andDuc hiandElbourne (2016).The identicationscheme assumesthat acredit supply shockmov eslendingratesandvolumes inopp ositedir ections.Suc hacreditsu pplysho ckis ECB Working Paper Series No 2469 / September 202010
Aggregate
supplyMonetaryp olicyAggregatedemandCreditsupplyLong-terminterestResidentialpricesUnem-ploymentStock prices
RealGDP ++ ++00
HICP{++ 00 0
Unemp. rate0{
Short-term int.
rate{+ 00 0
Long-term
interestrate 0 +0
Importv olumes+
Exportprices 0
Residential
propertyprices0+0
Bank lending
rates {
Bank loan
volumes +
Equityprice
index ++Table3: Summary ofiden tifyingrestrictions inSPV AR consistentwith either adeclineinbanks capital(Gerali etal. 2010)or deteriorationof bank asset quality(Gertler andKaradi2011).The identification isstrengthened by imposing zero contemporaneousresponse restrictionsoninflation, unemploymentrate, short- andlong-term interestrates. Theresiden tialprice shocki de ntificationfollowsBuch etal.(2014).Allremaining shocks,thereinsto ck pricesshock,unemploymen tsho ckandlong-terminterest rateshock,are delineated byimposing therelevant zeror estrictions. Macroeconomicscenarios arereplicated withinthe model usingstructural shoc kdecompo- sition. Thereby,themacro economicmodul enotonlyallo wsintroducingsecond-roundeffects originating inthe bankingsec torvia thecreditsupplysho ck butalso pr ovides aclearnarrative of eachscenario.
3.2 Data
The quarterlydata from2002Q1 to2017Q4 aresourced fromthe ECBStatistical DataW are- house (SDW)macroeconomic projectiondatabase.
2Endogenous variablesconsistof realgross
domestic product,unemploymen trate,consumerpriceindex,nominal residential propert y prices, long-termnominal in terestrates
3, equitypriceindex ,imp ortv olumesand exportprices,
bank lendingrates, bankloan volumes, short-termmoney marketrates.T wo exogenousv ariables2 If theinformation inthe SDW wasmissing,t hedataset wascomple tedusing datafromnationalsources.
310-yeargo vernmentbondrates.ECB Working Paper Series No 2469 / September 202011
include country-specificforeign demandand competitor sprices.Allvariablesexcept unemploy- mentand interest ratesenterthe VAR modelinlog lev els.
3.3 Results Note: Unitsare inp ercentage pointdeviation,exceptthe interes trates andunemplo ymentrate,whichare interms ofp ercentages.Thesolid
red linedepicts themean ofcoun trysp ecificresp onseateach horizon,whilethedashed linesrepresentmean +/-one standarddeviation.
Figure 5:Impulse responsestocredit supplysho ck
Impulse responsesofthemodel"s endogenousv ariablesto aonestandard deviationadv erse credit supplyshoc karepresentedin Figure5. Afavourablecreditsupply sho ck leadsto acredit boomthat causesbankloanv olumesto increaseand banklending ratestodecrease.On av erage, favourablecreditsu pplyshocks increaseloanvolumes byabout1.1%on impactand decreases lending ratesby 5basispoin tson impact.Af terthecreditsupplysho ck, GDPincreases by about0.25% onimpact before slowly returningtobaseline.The effectofthecredit supplysho ck on otherv ariablesisrelatively limited.
4 Bank-levelbeha viouralequations
This sectionpresen tsthemainequations describingbanks" beha viouralreactions. Itstarts with the descriptionof adjustments inbanklendingv olumes.Then, itsummarises theestimates of ECB Working Paper Series No 2469 / September 202012 lending anddep ositmarginequations.Last, itla ysdownthe dynamicsof dividendp ay-outs. In eachcase,and similarto thesection onthe macro bloc k,w efirst describe thespecification, then thedata, andfinally theresults ofthe estimations.
4.1 Loanev olution
The evolutionofbanklendingshould relateb othto themacro economicen vironment (astress test scenario)andbanks"own situationsuc has theirprofitability, solvency orestimatesof credit risk.The desiredsp ecificationof lendingequationsshouldencapsulate allthis factors in acommon setup.Ho wev er,theestimationofsuchequationswith bank-level informationis hindered bydataa vailabilit y. Too vercomedatashortages,thepro cessof estimatinglending equationsseparates theimpact of creditdemand andcredit supply factors.Loan demandwilldepe nd onaggregate economic conditions, likeGDPgro wth,unemplo ymentand interestrates.Loansupply,inturn, will dependon individualbank characteri stic s,likebank"sprofitability,solvencyand assetquality. The demand-sideequationis estimatedusing iBSI/iMIRdata collected fora sufficiently long timeto allow identificationofbank lendinginertiaandits dependency onmacro economic aggregates. Thesupply-side equationis estimatedusing thesup ervisoryrep ortingdatasets COREP/FINREP thatco versonlyathree year horizonbut includesallrelevan tvariablesre- flecting banks"in ternalsituation.Thedemand-side equationis laterrein terpretedas anau- toregressivecomp onentandthesupply-side equ ation asamedium-runrelationship. Bothare merged inthe finalequation ina wa yanalogous toan errorcorrectionmodel.
4.1.1 Loandemand
4.1.1.1 Methodology
Loan demandequations inv olveasetofmacroeconomi cv ariablesand anindex capturing time variationincredit supplyfactors. Thelatter indexis laterreplaced by sector-specific loan supply equations.Ho wever,forestimationpurposesit is necessaryto deriveameasure ofc hanges in creditsupply factorsthat would be availablefor alongertimehorizon. Tothisend,we use semi-consolidated informationon banklending volumes andin terestratemargin sand identify a structuralcredit supplysho ck inaseriesofbank-level VARs. Moreprecisely ,w eestimate a ECB Working Paper Series No 2469 / September 202013 series ofbank-lev elVARmodelssimilarly asin Altavilla etal. (2016)foreach ultimateparen t bank: Y i;t=ci+pX j=1A i;jYi;t j+pX j=1B i;jXi;t+i;t(2) whereYi;tis av ectoroftheendogenous vari able sfor bankiincluding totallending volumes to the non-financialprivate sectorasw ellasthea verage interestrateon loansto thenon-financial privatesector. ciis av ectorofbank-specific intercepts. Xi;tis av ectorofexogenousv ariables with naturallogarithm ofexp osurew eightedrealGDP andchangesinshort-ter mrates. Ai;jis a bank-specificmatrixof autoregressive coefficie ntsforlag j,pis then umberoflags(andis set top= 1)and i;tis av ectorofreduced-formresiduals. Relyingon thesebanking groupsp ecific VARsw eidentifya creditsupplysho ck indexvia sign restrictionsasdescribedinTable 4:
Creditsupply
Loanvolumes{
Avg.in terest rate+ Table4: Summary ofiden tifyingrestrictions Lending demandequation isestimated viaa fixed-effectpanel regressionof bank-level loan growthrates oncoun try-specific aggregateeconomicconditions.These ctorssconsidered are: financial institutions(FIN), non-financialcorp orates(NF C)andhouseholds(HH). Wepostulate the followingfunctionalrelationship forquarterly logc hangeof bankiloan exposurestosector sin countryjat timetdenoted byΔLoanssij;t: Δ Loans sij;t=ci+pX j=1pΔLoanssij;t j+pX j=1pXj;t j+Supplyshocki;t+ij;t(3) whereXj;tis av ectorofmacroeconomic variables includingGDPgrowth, inflation(basedon HICP), short-termrate,unemploymen trate, andthespreadbetw eenlong andshort-term rate in countryj.p,pandstand forregression coe fficients.Beforeapplyingageneral-to-specific procedureof excludinginsignifican tlags ofdependent variables, thepanel includestwolags ( p= 2)of allco variates. Thepanelisestimatedsubject todynamic homogeneity restrictions (see Jensen(1994)) toensure astable long-r un relationshipb etweennominalGDPgrowthand ECB Working Paper Series No 2469 / September 202014 loan growth.Thesp ecificationincludes thesupp lyshoc kindexSupplyshocki;tfrom eq.2 and is estimatedwith banklev elclustered standard-errors.
4.1.1.2 Data
The firststageof theanalysis isbased onsemi-consolidated iBSI/iMIRloan andin terest rate data,while thesecond stageuses individualrep ortingand differentiates bet weenloansand interestrates tohouseholds, non-financialand financialcorp orationsas well asthe publicsector. The loanand interest ratedatawere seasonallyand outlier-adjustedusing theX-12-ARIMA algorithm bothonindividu albranc handconsolidatedbank level.The monthly iMIR/iBSId ata are thentransformed toquarterly timeseries inline withthe frequencyof themo delframew ork.
The macroeconomicvariables aresourcedfromSD W.
4.1.1.3 Results Note: Thered lineillustrates themedian cumulativeshock indexofall70semi-consolidatedbanking groupsrep ortingin iBSI/iMIRwhile
the greyarea represents theinterquartile- range. Figure 6:Aggregate timev ariationincred it supplyshockseries Figure 6plots theev olutionof creditsupplyshoc kseries. Itapp earsthat theproxyfor omitted bank-specificcreditsupply factorsp erformsr easonably wellincapturing theimpactof the recentfinancialcrisis in2007-2008 andlate rof theE uropean debtcrisisin2010. Table5 presents theestimationresultsfrom forthe fixed-effectpanel. Thefinal specifications ECB Working Paper Series No 2469 / September 202015 SectorNon-financial corporatesFinancialcorporates SovereignHouseholds RegressorCoefficientp-valueCo efficientp-valu eCoefficientp-value Coefficientp-v alue
Loangrowth( t1) 0:193
(0.000)0:375(0.000)
Loan growth(t2) 0:200(0.000)0:294(0.000)
GDP growth(t1) 0:317(0.000) 0:595(0.000) 0.567(0.428) 0:189(0.000) GDP growth(t2) 0:290(0.000) 0:405(0.010) 0.308(0.627) 0:142(0.000)
GDP growth(t3)0.126 (0.843)
Inflation (
t1) 0.0870(0.376) 0.404(0.546) 0:331(0.000)
Inflation (
t2) 0:520(0.000) 0.596(0.372) 0.862(0.613)
Inflation (
t3)0.138 (0.936)
Δ STN(
t1)-0.229 (0.105)1:168(0.097)3:904(0.014)0:664(0.000)
Spread (
t1)-0.0141 (0.660)-0.130 (0.579)
Unemp. rate( t1)0:031(0.077)
Δ Unemp.rate (
t1)1:430(0.066)-0.00230 (0.970) Supplyshock(t)0:0005(0.033)0:007(0.000)0:005(0.069) 0.00001(0.898) Constant0:0006(0.042) 0.002(0.147) -0.001(0.731) 0:0005(0.000)
Bankxed eectsY esYesYesYes
Banks84918186
Obs.2925317227552994Table5: Loan demandmo del
differbetweensectors.Forinstance, theregressions forfinancialandso vereign exposures donot include laggeddep endentvariables,wholehousehold loansexhibithigh inertia.The procycli- calityof corporate lendingisstrongerthan thepro cyclicality oflending inan yother market segmentsas reflectedin therelativ eimpact ofoutput growthoncorresp ondingloan volumes.
4.1.2 Loansupply
4.1.2.1 Methodology
In orderto uncov ertherelationshipbetw eenbank lendingand bankcharacteristics,werely on ap ooledbank-levelregressioninspired byKhw aja andMian (2008).Theoriginalw orkof KhwajaandMian(2008) (se ealso Jim´ enezetal.(2017))usesloan level informationonfirms, whichare indebtedto atleast tw obanks ,to establishhow two(or more)banks, whichare differentlyaffected by apolicyc hange,adjust theirlendingtofirms. Inordertoc ontrol forfirm levelc haracteristicsthatcanaffect theirloan holdings(suc has loandemand andb orro werrisk) the authorssaturate theregressions withfirm-lev elfixed effects. In ourcase, theric hset offixed-effectswillcon trolthe regressionsfor credit demandfactors. A counterpartyofabank isa specific lendingsegmen tsuc has acorporatesector inone of ECB Working Paper Series No 2469 / September 202016 the jurisdictions.
4By usingdata forcoun terpartiesthat borro wfromatleasttw obanks, we
can identifycou nterparty-timefixedeffects,whichcapture cou nterpart y-timevariationlik ely related tothe evolu tionofthemacroeconomicenvironmen t.The sal ientassumptions behind this methodologyisthat allen titieswi thin thesamesector-coun terpartyclass facethe same demand forloans and thatloandemandis notbank specific, i.e.it isequal acrossall banksthat lend toa counterpart y.Inthemodel,fixed-effectestimates aredropp edand effectivelyreplaced bythe loandemand equationestimated earlier.
The generalmo delspecificationis thefollowing:
Δ Loans sij;t=f(CET1Shortfalli;t 1; NPLsij;t 1; ROAi;t 1; ffisjt);(4) where Δ Loans sij;tis quarterlylog change ofsectorsloans toc ounterpartyjbybank iin time t ,CETShortfalli;tis ameasure ofCET1 capitalsurplus orshortfall (seesubsection 4.1.2.2for its definition)
5,NPLsij;tis sector-counterpartyspecificshareof non-performingloans, ROAi;t
is returnon assetsan dffisjtare counterparty-timefixedeffects. In additionto thelinear effectof capitalsurplus orshortfall we arein terestedin threet ypes of non-linearities.First, abank closeto itsregulatory requirements may be moreprone to deleverage.T ocapture thiseffectweinteract CETShortfalli;twith adumm yvariableequal to oneif bank iexperiencescapital shortfallin timetdenoted byI(CETShortfalli;t). Sec- ond, banksma ybemore likelytodelev erage onnon-domesticexposures. To pindownthis effect weinteract CETShortfalli;twith dummyvariablesfor domesticI(Domestic) andforeign I(Foreign) exposures.Finally, wedistinguishb etweenthe caseswhere theshareofbankNPLs increased inthe las tyearandcaseswhere suchashare decreased,and intro ducethe interactions with thecorresp ondingdummiesI(NPLincrease ) andI(NPLdecrease ).
4.1.2.2 Data
The estimationrests onCOREP/FINREP reporti ng templates.Theseincluderelev ant bank characteristicssuc haslendingat bank-counterpart y-sectorlev el,capital, profitability,shareof4 Similar approachisused by MesonnierandMonks(2015)in theirstudy ofthe eectof 2011'sEBA capital exercise.
5Severalstudies hav eshownanimp ortanceofthelinkbet ween bankcapitalandlending activity (seefor
example GambacortaandMistrulli (2004),Jonghe etal. (2016)and Aiyar etal. (2016)).A generalnding of
these studiesisthat lesscapitalised bankspro videless fundingsources tothe realeconomy. ECB Working Paper Series No 2469 / September 202017
non-performingloans. Thedata areavailablefor theperiod 2014-2018,at quarterlyfrequency. These datais supplemented withtheinformationon capitalrequiremen tsfrom which we derive our measureof CET1Shortfall:
CET1Shortfalli;t=CET1REAi;t TCET1REAi;t(5)
whereCET1REAis theactual CET1ratio definedas inApp endixB.6 andTCET1REAis a capitaltarget definedas:
TCET1REAi;t=P1CR+P2CRi;t+ComBi;t+P2Gi;t(6)
whereP1CRandP2CRdenote PillarI andPillar II minimum CET1capi talrequirements. ComBis thecom binedbufferrequirement following thedefinitionasin CRDandincluding capital conservationbuffer,coun tercyclicalcapital buffer,SystemicRiskBuffer, O-SII andG- SIIb uffers.Forestimationpurposes, weassumethat Pillar2 guidance(P2G) equals2% ofRisk
WeightedAmounts.
4.1.2.3 Results
RegressorCoefficientp-v alue
Eectof CET1surplus/shortfall ondomestic/foreignexposures
CET1ShortfallI(Foreign)0:098 0:006
CET1ShortfallI(Domestic)0:038 0:660
I(CET1Shortfall) 0:044 0:000
CET1ShortfallI(CET1Shortfall)0:227 0:644
Effect ofNPLs depending ontheirincrease/decreasein lasty ear
Sector countryNPLI(NPLdecrease)0:005 0:828
Sector countryNPLI(NPLincrease) 0:068 0:001
ROA0:428 0:108
Constant0:000 0:953
Counterparty-timexedeectsYes
R 20.10
Obs.9208Table6: Loan supplymodel fornon-fi nancialcorporates Tables6-8 present theestimati onresults forthecorporate,householdandsov ereignsector, respectively.Fewcommen tsareinorder. First,thereisa differencein bankr esponses bet ween ECB Working Paper Series No 2469 / September 202018
RegressorCoefficientp-v alue
Eect ofC ET1surplus/shortfall ondomestic/for eignandconsumer/mortgage exposur es
CET1ShortfallI(Foreign)I(Consumer)0.179 0.128
CET1ShortfallI(Foreign)I(Mortgage)0.110 0.132
CET1ShortfallI(Domestic)I(Consumer)0.135 0.086
CET1ShortfallI(Domestic)I(Mortgage)-0.043 0.480
CET1ShortfallI(CET1Shortfall)0.065 0.772
Eectof NPLs dependingontheir increase/decr ease inlast year
Sector countryNPLI(NPLdecrease)-0.058 0.019
Sector countryNPLI(NPLincrease)-0.072 0.000
ROA0.074 0.549
Constant0.006 0.000
Counterparty-timexedeectsYes
R 20.21
Obs.3428Table7: Loan supplymodel forhouseholds
RegressorCoecientp-v alue
Eect ofC ET1sur./short.on domestic/foreignexposur es
CET1ShortfallI(Foreign)-0.088 0.010
CET1ShortfallI(Domestic)-0.033 0.529
Eectof NPLs dependingontheir increase/decr ease inlast year
TotalNPLI(NPLdecrease)-0.085 0.039
TotalNPLI(NPLincrease)-0.091 0.086
ROA0.134 0.647
Constant0.001 0.661
Counterparty-timexedeects
Yes R 20.15
Obs.3535Table8: Loan supplymodel forso vereigns
the threesectors. Theeffect ofCET1 surplus/shortfall isthe strongestfor corporatesand consumer loans,whereas itb ecomesnegativ eforsov ereignexposuresand ford omesticmortgage loans. Thistells usthat bankswith CET1shortfall shifttheir lendingfrom corporates and consumer creditto sov ereignsandpartiallyalsotomortgage loans.This likely reflectshigher risk weightchargeson theformerexposures comparedto thelatter. Second,the lendingeffect of CET1surplus/shortfallis amplifiedwhen aban kexp eriencesa capitalshortfall. Asexpected, this non-linearityismore pronouncedfor riskiercorp orateloans. Third,the effectof capital surplus/shortfall isstronger forforeign exposures. Bankswith capitalshortfallwillfirst red uce ECB Working Paper Series No 2469 / September 202019 their loansupply abroadand onlylater inthe domesticmar ket. Lastly, negativeeffectofNPLs is strongerfor banksthat recently experienced anincreaseinthe shareofNPLs.
4.2 Lendingand deposit interestratemargins
The intentionofinclu dingendogenous equationsforbank-levelinterest ratemargins inthe modelis tob ettercapture theevolutionof netin terest incomeunderdifferent macroeconomic scenarios. Bank-levelinterest ratesconsistoft wo componen ts,the referenceratelinkedto3 monthEURIBOR (STN) inthemacroblo ck, andmargins. Thissplit isaimedtorecover tw o risks affectingbank netin terestincome unders tress :(i)theriskrelatedto achangeinthe general "risk-free"curves;(ii)the riskrelatedtoa change inthe "premium"that themark et requires orthe banksets fordifferen tt ypes ofinstrumentsand counterparties.
4.2.1 Methodology
Lending marginson newbusiness EIRAssetNeware modelledseparatelyfor threeloan seg- ments:loans toNF C,loans forhouseholdsforhouse purchases andconsumer credit.F ors im- plicitythey areeac htime definedastehd ifference bet ween thebank specificlendinginterest rate fornew businessand there fe rencerate equalto3monthEURIBOR.Thegeneral dynamic panel modelspecification forthesectorslending marginshas thef ollo wingform:
EIRAssetNew
si;j;t=f(EIRAssetNewsi;j;t 1; SovSpreadj;t;ΔSTNj;t; Xj;t p; si) (7) where asearlier iis abank index,ssector indexandja countryindex.STNj;tis the reference rate,SovSpreadj;tis thedifferenceb etw een10YGovernmentbond yieldsfor country jand theGerman bundat timet,Xj;t pare contemporaneousandlaggedmacro economic variablesincluding GDPgro wth,residen tialhousepricegro wth,sovere ignspread andthe credit supply shockdefined asinsection4.1.1.siare bankfixed effect.Including thereference rate among theco variatesallowsthepass-through ofthemoneymark etrate into bank-level lending rates todiffer fromone. Theequations includeup tofour lagsof depend en tv ariablesand are estimates withthe Arellano-Bondestimator usingrobust standarderrors. Empirical modelsfordep ositmargins EIRLiabNewfollowanalogous specificati onwiththe breakdownin fours egmen ts:sightandtermdeposi tsbothfor householdsand non-financial ECB Working Paper Series No 2469 / September 202020 corporates.
4.2.2 Data
The interestratemodelis basedon inte restrate informationprovidediniMIR/iBSI.The data are seasonallyand outl ieradjustedusingtheX-12-ARIMA algorithman dcomplementedwith macroeconomicv ariablesfromSDW.The datasetre sultsinan unbalanced panelspanning from
2007Q4 to2017Q4.
4.2.3 Results
Non-financialHouseholdsHouseholds
corporatesmortgageconsumption RegressorCoefficientp-value Coefficientp-valueCo efficientp-value
EIRAssetNew(t 1)0.629 0.0000.8240.000 0.8030.000
EIRAssetNew (
t 2)0.203 0.0000.0610.022
SovSpread( t)0.033 0.0010.019 0.0100.013 0.297
SovSpread( t 1)-0.014 0.050
Δ STN(
t )-0.472 0.000-0.730 0.000-0.562 0.000
Supplyshock(t)-0.004 0.6500.002 0.6640.029 0.012
GDP growth(t 2)-4.344 0.016
GDP growth(t 3)-4.067 0.037
Resid.prop.price growth(t)-1.931 0.028
Inflation (
t )10.958 0.0029.099 0.000
Inflation (
t 2)10.531 0.000
Constant0.341 0.0000.208 0.011
Bankxed eectsYesYesYes
Obs.138512661349Table9: Lending marginregressions Table9 presentstheb est-fittingestimation resultsforlendingmargins onnew loansto the non-financialprivate sector.Duringp eriodsofincreasing economicactivit y,asreflectedin high outputand propert ypricegrowth,lendingmargins typically decrease.This islik elyto reflect fluctuationsin collateral valueandbanks" riskperception.High priceinflation increases lending margins.So vereignspreadincreaseslendingmargins (thoughthis effectis insignificant for consumptionloans). Finally, thepass-throughof referencer atesappearsstrongest forloans to households. Tables10 and11 present thee stimationresults forterm andsightdepositmargins,respec- tively.Bothsho wa positiveasso ciationwith outputgrowthand sovereignspread. ECB Working Paper Series No 2469 / September 202021
Non-financialcorporatesHouseholds
Termdep
ositsTermdep osits
RegressorCoefficientp-valueCoefficientp-value
EIRLiabNew(t1)0.7630.0000.811 0.000
EIRLiabNew (t2)0.1500.0020.110 0.000
SovSpread( t)0.0360.0010.034 0.000
Δ STN(
t )-0.6780.000-0.704 0.000
GDP growth(t)1.8850.0041.934 0.664
GDP growth(t2)1.6460.001
Inflation (
t )8.9470.0006.660 0.000
Inflation (
t1)4.187 0.000
Inflation (
t2)-4.8380.018
Constant-0.0840.000-0.068 0.000
Bankxed eectsYesYes
Obs.78947879Table10: Termdep ositmargin regressions
Non-financialcorporatesHouseholds
Sightdep ositsSight deposits
RegressorCoefficientp-valueCoefficientp-value
EIRLiabNew(t1)0.9200.0000.747 0.000
EIRLiabNew
(t2)0.167 0.000
SovSpread( t)0.0120.0000.005 0.014
Δ STN(
t )-0.5280.000-0.721 0.000
GDP growth(t)1.758 0.016
GDP growth(t1)2.3190.000
Constant-0.0350.000-0.017 0.001
Bankxed eectsYESYes
Obs.14271318Table11: Sightdep ositmargin regressions
4.3 Distributionof dividends
Dividend distributionp oliciesareanimp ortant channel thatimpactsbankcapitalandtherefore also theirresilience. Dividendpa y-outsin themodelar ego verned bytheendogenous ratio of dividendpa y-outstotheafter taxprofit andthe Maximum DistributableAmoun t(MD A) restrictions. Thelatter become bindingwhenbanks"o wnfunds dropb elow theircombi ne d buffer requirement.
4.3.1 Methodology
The ratioof dividendpa y-outsto theaftertax profit isassumedtofollo w: ECB Working Paper Series No 2469 / September 202022
DivPayoutRatio
i;t=Φ(DivPayoutRatio i;t
1; CET1REAi;t 1; NPLi;t 1;(8)
RWAtoTA
i;t
1; CTIRi;t 1; Loang rowthi;t 1; GDPgr owthj;t 1; ffii)
whereDivPayoutRatioi;tis bankiratio betweendividendpayments andprofit aftertaxin timet,CET1ratioi;t 1is CET1capital adequacyratio, NPLi;t 1is shareof non-performi ng loans,RWAtoTAi;t 1is theratio bet weenriskweightedassetsand totalassets, CTIRi;t 1is cost-to-income ratio,Loan growthi;t 1is growthoftotal bankloans, GDPg rowthj;t 1is GDP growthin acoun try jwhere bankiis headquarteredand ffiiare fixedeffects controls described below.Φ( : ) denotesthe standardnormal cumulativ edistribution function. The ratioof thedividend pay-out tothe aftertaxprofitis bound bet ween zeroand one. Accordingly,we useafractional responsemo delin troduced byPapkeand Wooldridge(1996) and PapkeandWo oldridge(2008) toestimateequation8. Thetime dependency ofdividend paymentsinthe non-linearsetup presents ac hallenge.Unlik einlinearmo dels,where unobserved effects canb eeliminatedwithcertain transformations,there isin general nosuc htransformation for non-linearmo dels.Weadopt themethodology proposed byWo oldridge(2005)andinclude in the regressionmeans andin itial valuesofallright- handsidevariables asw ellas theinitialvalue of thedep endentvariable.Theseare fixed-effectscontrolsdesigned totak ea wa ythecorrelation betweenthelaggeddep endedv ariableand theerrorterm. 6
4.3.2 Data
The dataserving theestimation ofthe dividendpa y-outratio equationstem fromBlo omberg and SNL.These datasetsallo wus toconstructtimeseri esof dividendpa y-outratios, andrigh t- hand sidev ariables,foralarge sampleof euroarea banksfor the perio d2005-2017. We set dividend paymentstozeroin cas es ,when profitaftertaxisnegative,andsetdividendpa y-out ratio toone, whenbanks pay outdividends inexcessof profit.6
The ideais thatinstead ofeliminating thexed eects,one shouldcon trolfor them.This largelyfollows the
logic ofC hamberlain(1984),whoproposestomo delconditional expectation oftheunobserved eect asalinear
function ofthe exogenousv ariables.ECB Working Paper Series No 2469 / September 202023
4.3.3 Results
Table12 reports the estimatedco ecientsofthefractionalrespons emo del.Lagged dividend pay-outratio, CET1ratio, theratio bet ween riskw eightedassetsandtotalassets,loangro wth and GDPgro wthhave apositiveimpacton dividendpa yments,whereastheshareof non- performingloans andcost-to-income ratioha ve anegativ eeect.
RegressorCoecientp-v alue
DivPayoutRatio(t-1)0.3150.005
CET1REA(t-1)4.221 0.003
NPL (t-1)-4.031 0.000
RWA-to-TAratio(t-1)1.243 0.019
Cost-to-income ratio(t-1) -0.755 0.065
Loan growth(t-1)0.795 0.007
GDP growth(t-1)1.856 0.076
Constant-1.242 0.017
Numberof observations554Table12: Estimated coecientsof thedividend distributionmo del Final dividendpa ymentDivPaidis denedastheminim umb etw eenregression-predicted dividend paymentsandtheMD A:
DivPaid
i;t=min(DivPayoutRatioitProfBDivit; MDAit)(9) whereProfBDivis protaftertaxthatis av ailablefor protdistribution denedas in
AppendixB.5.
5 Modelapplicationin 2018macropruden tialstre sstest exer-
cise Toillustrate anapp lication oftheBEASTweev aluatethe performance ofthe euroareabanking sector underthe adverse scenarioofthe2018 EBAsup ervisorys tress testfor 2018- 2020. The adversescenarioconsidered ap eriod ofprolonged systemicstress,witheuroarea output contractingb y2.4%incum ulative terms.A tthesametimetheresiden tialhouse priceswere assumed todrop by 16.5%andtheeuro area unemploymen trate reac hed10.3%by 2020.The complete descriptionof theanalysis canb efound inBudnik etal.(2019),while thissection ECB Working Paper Series No 2469 / September 202024 emphasises theroleof themodel"sbeha viouralequationsingeneratingthere sults.
5.1 Assetqualit y
Adverseeconomicconditions arefirst reflectedin thedepreciation ofasset prices.Mo vemen ts in marketpricestranslate into imminent lossesonfinancialassets heldatfairv aluein banks" trading booksthatenter themo delasanexogenous variable. Marketlossesen tereither profit and lossor directlyreduce banks"o wnfunds viaother comprehens iv eincome (seeApp endix
B.6.1).
Deteriorating economicc onditionsleadasw elltoanincrease incredit losses.Lossgiven de- fault, lossrates andthe transitionprobabilities tostage 2(assets withdeteriorated quality) and stage 3(non -performingassets)underIRFS9 goup. Themain fac torsaffecting thetransition rates aredeclining GDPand houseprices asw ell aselev atedlong-term interestratesandun- employmentrates(see Appendix A.2).The shareofnon-p erformingloans tothe non-financial privatesector inthe bankingb ooks startsincreasing alreadyin2018andby 2020the share almost doublescomparedto2017 reaching 13%(see Figure7). Source: Budniket al.(2019)
Figure 7:NPLratio fornon-financial pr ivatesector
5.2 Protabilityandsolv ency
The return-on-assets(R OA)remainsnegativein every yearfrom2018 to2020 (seeFigure8.In
2018 bankp rofitabilityisnegativelyaffectedby both marketandcredit risklosses. From2019
onwards,ROA becomeslessnegativeb yaround-0.15%,whic his mainlyduetolo wercreditrisk ECB Working Paper Series No 2469 / September 202025 losses anddue tothe stabilizingeffect ofnet feeand commissionincome (NFCI). Thisreflects the countercyclicalnatureofNF CI(see Appendix A.1).Source: Budniket al.(2019)
Figure 8:ROAdecomp osition
In linewith deterioratingprofitabilit y, banks"CET1ratiosfallon av erageb y3.2pprelativ e to 2017.As aconsequence, about halfof thebanksexhibitcapital shortfalls,i.e. theirCET1
capital ratiosfall belo wtheirrespectiv eregulatory targetvalues(Figure9).Source: Budniket al.(2019)
Figure 9:Distributionof CET1surplus/shortfall in2020
5.3 Lending
Banks withlo wprofitabilityand capitalshortfallsrestraintheir dividendpa y-outs. Dividends are distributedonly whenban ksgenerate profitandmaintain capitalisationab ov etheir MDA ECB Working Paper Series No 2469 / September 202026 trigger point(seeAppendix B.5).As many banksexperiencelosses andtheir CET1ratios becomeimpaired, theirdividend pay-outs remainclose tozero. Another adjustmentmechanism preventingfurtherdeteriorationof CET1capitalratiosis deleveraging.Figure 10demonstrates thatbanks withcapital shortfallcon tractlending signif- icantlymore comparedto bankswith capitalsurplus. Asthe fractionof banks withcapital shortfall increaseso verthescenariohorizon,the deleveraging intensifies inline withthe loan supply equationsinsection4.1.2.Source: Budniket al.(2019) Figure 10:Changein banklending versus capitalshortfall Weakenedloandemand (seesection 4.1.1)ad ditionallymo deratesloan expansion.Figure 11 showscum ulativeloangrowthacross sectors,distinguishing between theimpact ofsu pplyand demand factors.The lendingto non-financial corporates contractssignificantly morethanto households withthe pronouncednegativ econ tributionofcreditsupply factorsforthe corporate sector (seeT able6). Beyond,lending evolution differsacrossdomesticand foreignmark ets.Fi gure 12sho wsthat banks deleveragemostlyonexposures toforeign markets andlesssoon dome sti cassets.
5.4 Therole ofdynamic balancesheet
A largeshareof supervisory stresstests, includingthebi-annual EBAstress testexerc ise,assume a constantbalancesheet. Underthis assumptionthe sizeand structureof bankassets and liabilities remainsconstant. Thepopularity ofthe assumptionrelatestothefactthat itensures larger comparabilityofthe resultswhen astress testexercise isconducted ina bottom-up fashion ECB Working Paper Series No 2469 / September 202027
Source: Budniket al.(2019)
Figure 11:Cumulativeloan growth
Source: Budniket al.(2019)
Figure 12:Foreign anddomestic lendingb ygeograph y (relies onbank own predictions).Howev er, theassumptionclearlycompromisestherealismof stress testsb yignoringbanks"most likely reactionto stressconditions. The dynamicbalance sheetassumption ingrainedin theBEAST leadsto highercapital depletion thanthe analogousconstan tbalance sheetexercise.Comparedto theoriginal EBA stress testresults, theBEAST rev eals35 bnEURhighercapitaldepletion in2020. Thisresult is illustratedin Figure13 which contrasts CET1capitaldepletionexpressed asashareof ris k exposureamoun tatthestarting poin tfrom adynamic andconstantbalancesheetapproac h. Atthe sametime theend-of-p eriod bank-level CET1ratiostendtobe higherwhenforecasted with theBEAST ascompared tothe EBAstress testresults. Bank-l evel CET1ratios in2020 are significantlylow eronaveragecompared tothe 2017startingvaluesunderb othapproac hes. ECB Working Paper Series No 2469 / September 202028
Source: Budniket al.(2019)
Figure 13:CET1capital depletion
However,thecapitalratios witha dynami cb alancesheet aresligh tlylessadversecompared to thosederiv edunderastatic balancesheet (Figure14). Thisis anoutcome ofat leastt wo counterbalancingeffects. Bylinking thein tensity ofdele veragingtobanks"solvency ,the model introducesacapital ratiorestoring mechanism. Onthe otherhand, netinterestincome is negativelyaffected by decliningasset volumes, especiallyincombinationwith araisingshareof non-performingassets thatb earno orverylittle interest. For thescenarioconsi de redhere,the former effecttend stodominate.Source: ECBcalculations inMST 2018exercise. Figure 14:CET1 ratioECB Working Paper Series No 2469 / September 202029
5.5 Therole ofthe financialsector-real economyfeedback loop
Next tothe dynamic balancesheetpropert y, themo delcanservethedescriptionofthe nancial sector-real economyfeedback loop.To thisend, weassume thatthe originaladversescenario doesnot yet re ecttheimpactof excessive banks'delev eragingin adverse economiccondition s. More precisely,we assumethatc hangesin banklendingthatcan beattributedto credit de- mand factors,deterioration inprots, orlinear eectsof assetqualit yand capitalisationare accommodatedin theor iginalscenario. However, thenon-linear partofthelendingequations involvingcapitalshortfall andthe shareof non-performing loans,represen tsbanks' excessive delevaragingand canb eadded totheoriginalscenario afterb eingtr anslated incoun try-level structural creditsupply shoc ks.Source: Budniket al.(2019) Figure 15:GDPin theadv ersescenariowith andwithoutthefeedbac klo op Emphasising thefee dbackbetweenthe bankingsectorandtherealeconomyin creases the severityofthe adverse scen ario.Inaggregate, theeuroareaoutputcontractsab out1.6 ppmore by2020 inaddition tothe cumulativ econ tractionof 2.4%GDPintheoriginaladv ersescenari o. In thecross-coun tryperspectiv e,GDPcontractsbyadditional0.2 ppto 3.5ppdepending on the jurisdiction.
6 Conclusions
The primaryob jectiveofmacroprudentialstress testingis measuring,monitoringandunder- standing ofsystemic risk.T od eliverup onthismandate,thepaperpresents anewBanking Euro AreaStress Test (BEAST)modelfor assessingsystem-wide resiliencewhileaccounting ECB Working Paper Series No 2469 / September 202030 for interactionsbet weenthe financialsector andth ereal econom y.Themo dellinks macroand bank levelinformationand de sc ribesindetailpropagationofmacroeconomicconditionsinto bank balancesheets andthe realeconom y-financialsec torfeedbac kloop.Thanks toitssemi- structural designthe model canprovidea comprehensive narrativefors omeof thetransmission channelsof systemicrisk. A bigstrength ofthe model isthat itcapturesmany aspects ofbank heterogeneity .This includes differentstructuresof bankbalance sheetsbut alsotheir diverse reactionsto economic conditions dependingontheir individualsolv en cy situation,assetqualityandprofitabilityper- formance. Assuc hthemodel canb eprospective lyused notonlyforstress testingbutalsoto trackth eheterogeneousimpactof regulatoryor macroprudential policies onbanks. However,thereare alsonumerouslimitationsofthe described model version.Dueto data challenges,most ofthe bankreaction functionsare estimatedwithin panelsp ecificationsand representthe av erageratherthanindividualdecisionsb ybanks. Furthermore, we es timate separately bankresp onsesforvolume, priceor dividendadjustments ratherthan afull system of equationsr epresentingthejointoptimization problemof banks.Inthecurren tmo delv ersion, bank liabilitiesfollo wsimplifieddynamicsan dfu ndingcosts arelargelyexogenous.There isalso a scopeforincluding anadditional lay erof endogenouspropagation ofsystemicrisk,suchas interbanklending withfire salesor inte rbank contagion risk. Hence, manychallenges remainandourfuture work willaim toimpro ve theexistingframe- work,b othinterms ofthe identificationofbank response functionsas wellas extendingen- dogenous mechanismsinthe model. Thisw orkwillfocus onimpro vingtheidentification ofloan supply anddemand drivers, onmodellingban ksfunding costsandliability structure.Thelatter elementscould thenb eused tointroduce thesolv ency-fundingc ostfeedbackloop mec hanism. ECB Working Paper Series No 2469 / September 202031
References
Aiyar,S., Calomiris,C. W.and Wieladek,T.: 2016,Ho wdo escreditsupplyresp ondtomone tary policyand bankminim umcapital requirements?,EuropeanEconomicR eview82(C), 142{ 165.
Alessandri, P.,Gai,P .,Kapadia, S.,Mora,N.and Puhr,C.: 2009,T ow ardsa Framew orkfor QuantifyingSystemic Stability, International Journalof Central Banking5(3), 47{81. Altavilla,C., Canov a,F.andCiccarelli,M.:2016, Mendingthe broken link:heteroge ne ous bank lending andmonetary policy pass-through,WorkingP aper Series1978,European Central Bank. Arias, J.E., Rub io-Ramrez,J.F.andWaggoner,D.F.:2018, Inferencebased onstructural vectorautoregressions identied withsignandzero restriction s:Theory andapplicati ons,
Econometrica86(2), 685{720.
Barnett, A.and Thomas,R.: 2013,Has weak lendingand activity intheUnitedKingdom been drivenb ycreditsupply shoc ks?,Bank ofEngland workingp apers 482, Bankof England. Bennani, T.,Couaillier, C.,Devulder, A.,Gabrieli, S.,Idier, J.,Lop ez,P .,Piquard, T.and Scalone, V.:2017, Ananalytical framework tocalibrate macroprudentialpolicy, Working papers648 , Banquede France.
URL:https://ideas.repec.org/p/bfr/banfra/648.html
Blundell, R.and Bond,S.: 1998,Initial conditionsand moment restrictionsin dynamicpanel data models,Journal ofEc onometrics87(1), 115{143. Bookstaber,R.,Cetina, J.,F eldberg, G.,Flo od,M.andGlasserman,P .:2014, Stressteststo promote nancialstabilit y:Assessingprogress andlookingto thefuture, Journal ofR isk
Management inFi nancialInstitutions7(1), 16{25.
Buch,C. M.,Eic kmeier,S. andPrieto, E.:2014, Macroeconomic FactorsandMicrolevelBank Behavior,Journal ofMoney, Cre ditandBanking46(4), 715{751. Budnik, K.,Balatti, M.,Dimitro v,I., Gross,J.,Hansen,I., diIasio, G.,Kle emann, M.,Reic hen- bachas,T., Sanna,F., Sarychev, A.and Sinenko,N.:2019, Macroprudential stresstest of the euroarea bankingsystem, ECBOccasional Pap erSeries No226. ECB Working Paper Series No 2469 / September 202032 Camara, B.,Castellani, F.-D.,F raisse,H., Frey,L.,H´ eam,C., Labonne,L.andMartin,V.:
2015, MERCURE:AMacropruden tialStress Testing Modeldevelop edat theACPR,
Debatseconomiques etnanciers19, Banquede France.
URL:https://ideas.repec.org/p/bfr/decn/19.html
Chamberlain,G.:1984, Panel data,inZ. Griliches"and M.D. Intr iligator(eds),Handbookof
Econometrics
, Vol.2of HandbookofEc onometrics, Elsevier,c hapter22,pp.1247-1318. Claessens, S.and Kose, M.A.:2017,Macro economicimplications offinancial imperfections: a survey. Dees, S.,Henry ,J.,Martin,R., Grod zic ki,M., Gaiduchevici,G.,Gross,M.,Maliszewski,K., Rancoita, E.,Silva,R., Testi,S.,V enditti,F., Volk,M.,Georgescu, O.,Hilb erg,B.,Pancaro, C., Laliotis,D., Mehta, W.,Mirza,H.,Mo ccero,D. and Poblaci´ on,J.:2017,Stamp ¿: Stress-test analyticsfor macroprudential purposesinthe euroarea. Duchi,F. andElb ourne,A.: 2016,Credit suppl ysho cksintheNetherlands,Journal ofMacr oe- conomics50(C), 51-71. Fique, J.:2017, Themacrofinancial riskassess men tframew ork(mfraf),version2.0,Bank of
Canada TechnicalReport No.111.
Gambacorta,L. andMistrulli, P. E.:2004, Doesbankcapital affectlending beha vior?,Journal of FinancialI ntermediation13(4), 436-457. Gerali, A.,Neri, S.,Sessa, L.an dSignoretti, F.M.: 2010,CreditandBanking ina DSGEMo del of theEuro Area,Journal ofMoney ,Credit andBanking42(s1), 107-141. Gertler, M.and Karadi,P .:2011, Amodelof unconv entional monetarypolicy,Journal of
Monetary Economics58(1), 17-34.
Hristov,N., H¨ ulsewig,O.andWollmersh¨au ser, T.:2012, Loansupplyshocksduringthefinancial crisis: Evidencefor theEuro area,Journal ofInternational Moneyand Finance 31(3), 569- 592.
Jarocinski,M.: 2010,Resp onsesto monetarypolicysho cks inthe eastandthew estofEurope:
a comparison,Journal ofA ppliedEconometrics25(5), 833-868.ECB Working Paper Series No 2469 / September 202033
Jensen, B.:1994, The dynamicsystems ofbasice conomicgrowthmodels , KluwerAcademic
Publishing, Dordrecht.
Jim´enez,G.,Ongena, S.,P eydr´o, J.-L.and Saurina,J.:2017,Macroprudentialp olicy, coun- tercyclical bankcapital buffe rs,andcreditsupply:Evidence fromth espanishdynamic provisioningexp eriments,Journal ofPolitic alEconomy125(6), 2126-2177. Jonghe, O.D., Dewac hter,H.andOngena,S.:2016,Bank capital(requiremen ts)and credit supply: Evidencefrom pillar2 decisions,WorkingPap erResear ch303, NationalBank of
Belgium.
Khwaja,A.I.and Mian,A.: 2008,T racingthe Impactof BankLiquidit ySho cks:Evidencefrom an EmergingMark et,AmericanEc onomicReview98(4), 1413-1442. Kok,C., Mirza,H. andP ancaro,C.: 2019,Macro stresstestingeuroarea banksfees and commissions,Journal ofI nternationalFinancialMarkets,InstitutionsandMoney. Krznar, I.and Matheson,T.: 2017,T ow ardsmacropruden tialstresstesting:Incorporating macro-feedbackeffects, IMF WorkingPapers 17, 1. Morris, S.and Shin,H. S.:2008, Financialregulation ina systemcon text, Brookingspap erson economicactivity 2008(2), 229-274. M´esonnier,J.-S.and Monks,A.: 2015,Did theEBA CapitalExercise Cause aCredit Crunch in theEuro Area?,International Journalof Central Banking11(3), 75-117. Niepmann, F.and Stebunovs, V.:2018,Modeling your stressa way, BoardofGov ernorsof the FederalReserv eSystem,International FinanceDiscussion Papers,Num ber 1232. Papke,L.E. andW ooldridge, J.M.: 1996,EconometricMethodsforF ractionalResp onseV ari- ables withan Applicationto 401(K)Plan Participation Rates,Journal ofA ppliedEcono- metrics11(6), 619-632. Papke,L.E. and Wo oldridge,J.M.:2008, Paneldatamethodsfor fractionalresp onsev ariables with anapplication totest passrates, Journal ofEc onometrics145(1-2), 121-133. Wooldridge,J.M.: 2005,Simple solutionsto theinitial conditionsproblem indynamic, non- linear paneldata models withunobservedheterogeneit y, Journal ofA ppliedEconometrics 20 (1), 39-54. ECB Working Paper Series No 2469 / September 202034
A Estimatedbank sensitivities
The Appendixincludes thedescription oft wo modelsusedtomapmacro economicscenarios intobank balanc esheetsandprotandloss statements. Incon trastto equ ationsdiscussed in the maintext, theseempirical models re ectbanks' sensitivitytoa scenariorather thantheir reactions. Thismo delversionincludes onlyarelatively narrow subsetof endogenoussensitivit y equations, thenet-fee-commission income,and transitionrates bet ween IRFS9states.
A.1 Netfee andcommission income
Net feeand commissionincome togetherwith interest andtrading incomecons titutethethree most importantsourcesofincome formost euroarea banks.F urthermore,K ok,Mirza and Pancaro(2019) documen tthatfeeandcommis sionincome issubstan tiallyv aryingwithchanges in macroeconomicandnancialv ariablessuc has short-terminterestrate, stockmark etreturns and realGDP growth.
A.1.1 Dataand methodology
Net feeand commissionincome isgiv enb y:
NFCI i,t=FCIi,t+FCEi,t(10) FollowingKok etal.(2019)fee andcommission incomeFCIis projectedusinga dynamic panel regressionmo del.Thedependen tv ariableistheratioofbank i's feeand commission incomeFCIi,tto totalassets TAi,t. FCI=TAi,t=f(FCI=TAi,t 1,STNj,t 1,LTNj,t 1,Xj,t,LOANASSET i,t,i) whereSTNjstandsfor thec hangeinshortterm interest rate(EURIBOR) inthe domestic countryjof banki,LTNjrespectivelyfor thechangein thelong-term interestrate.Xj,tcontains additional macrofactors such asrealGDPgro wth,sto ck market growth,in ationand residential propertypricegr owth ofcountryj.LOANASSET icorrespondsto the loan-to-assetratio ofbank iandiare bankxed eects. Theincomemodel's variable selectionis basedonav ariable selection procedureusingthe LARSalgorithm. The nalmo delis estimatedusingthedyn amic ECB Working Paper Series No 2469 / September 202035 panel biascorrection intro ducedbyBlundell andBond (1998)where theestimates canb efound in Table13. Since coverageoffeeand commissionexp ensedata inpublic sourcesis scarce,thechange in expensesFCEof netfee andcommission incomeis estimatedusing animplied elasticity =0.867 based onsup ervisoryreportinginformation: FCE i,t=FCIi,t(11)
A.1.2 Data
The incomemo delisestimatedon apub lic datasetfrom Bloom bergcoveringann ualinformation from 1995un til2017for98 banks.Based onFINREP informationw eestimate theimplied elasticityfor FCE.
A.1.3 Results
The estimatedc oefficientsintheselectedspecification indi catea relatively highinertiainfee- and commissionincome. Theselected explanatoryv ariablesto explainthe remainingdynamics in FCEaregiv enb yGDPgro wth,short-termandlong-term rates,sto ckmarketand propert y price growthasw ellas loan-to-assetratioasindicator forbanks businessmo del.
RegressorCoefficientp-v alue
FCI=TA(t 1)0.7918 0.000
GDPgrowth(t)0.0048 0.006
Inflation(t)-0.0043 0.085
STN(t 1)-0.0032 0.332 LTN(t)0.0031 0.140 LTN(t 1)0.0013 0.517
Stockmarket growth(t)0.0006 0.002
Prop. pricegrowth (t)-0.0011 0.195
Loan-to-asset ratio(t)0.0027 0.000
Bankxed eectsYES
Number
ofobservat