国际清算银行-银行信贷损失预测：监管数据的经验教训（英）-2023.9.docx

OBISBISWorkingPapersNo1125Banks'credit1ossforecasts:1essonsfromsupervisorydatabyMartinBirn,RenzoCorrias,ChristianSchmiederandNiko1aTarashevMonetaryandEconomicDepartmentSeptember2023JE1c1assification:G21,G28,G32,G33,E44,P52.Keywords:Expected1ossforecasts;regu1atorycapita1;portfo1iocreditrisk.BISWorkingPapersarewrittenbymembersoftheMonetaryandEconomicDepartmentoftheBankforInternationa1Sett1ements,andfromtimetotimebyothereconomists,andarepub1ishedbytheBank.Thepapersareonsubjectsoftopica1interestandaretechnica1incharacter.Theviewsexpressedinthemarethoseoftheirauthorsandnotnecessari1ytheviewsoftheBIS.Thispub1icationisavai1ab1eontheBISwebsite(www.bis.org).©BankforInternationa1Sett1ements2023.A11rightsreserved.Briefexcerptsmaybereproducedortrans1atedprovidedthesourceisstated.ISSN1020-0959(print)ISSN1682-7678(on1ine)Banks7credit1ossforecasts:1essonsfromsupervisorydataByMartinBirn,RenzoCorrias,ChristianSchmiederandNiko1aTarashev1AbstractFocusingoncreditrisk,wecomparebanks'expected1oss(E1)rates,co11ectedconfidentia11ybytheBase1CommitteeonBankingSupervisionfrom2009to2023,andthecorrespondingactua11oss(A1)rates,asreportedinvendordata.Consistentwiththeuseofthrough-the-cyc1eriskestimatesforregu1atorypurposes,E1ratesrare1ymovein1inewithA1ratesovertime,whichhe1psexp1aina1argeprecautionarye1ementinBase1IIIcapita1requirements.Wea1sofindthattherank-orderofE1ratesacrossbanksmatchesc1ose1ythatoftheA1rates,in1inewithrecentandforthcomingregu1atoryeffortstoimproverisk-measurementpractices.E1ratesaremore1ike1ytobeexcessive1yoptimisticonthehee1sofhigherbankprofitabi1ityandfinancia1overheating,ascapturedbythecredit-to-GDPgap.Keywords:Expected1ossforecasts;regu1atorycapita1;portfo1iocreditrisk.JE1c1assification:G21,G28,G32,G33,E44,P52.WethankIhakiA1dasoro,IrinaBarakova,StijnC1aessens,SebastianDorr,RodrigoCoe1ho,Nei1Esho,MarcFarag,Mikae1Juse1ius,Wi11Kerry,Wo1fMu11erandEgonZakrajsekforveryhe1pfu1commentsonanear1ierdraft,andBurcuErik,RobertoOtto1iniandJhuveshSobrunforva1uab1eresearchsupport.Theviewsexpressedinthispaperareoursanddonotnecessari1yref1ecttheviewsoftheBankforInternationa1Sett1ements.Contents1. Introduction12. 1ossabsorbingresourcesforcreditrisk:basics32.1 Bui1dingb1ocksofregu1atorycapita132.2 I11ustrativeexamp1ewithahypothetica1bank63. Dataandkeymetrics73.1 Data73.2 Keymetrics8Expected1ossrate8Actua11ossrate8FirstjuxtapositionsofE1andA1rates9InconsistenciesbetweendataonE1andA1rates94. Adequacyof1ossestimatesandcapita1requirements104.1 E1vsA1:crosssection104.2 E1vsA1:timedimension114.3 Conservatisminactua1capita1requirements145. E1vsA1:pane1econometricana1ysis15Howaccuratearebanks'reportedE1saspredictorsofA1s?15Whatexp1ainstheforecasterrors?166. Conc1usion18References19Appendix:Additiona1regressionresu1ts21Fi1teredobservations21Step-oneregressions21Step-tworegressions22Fi1teredobservationsandconstantsamp1e23Step-oneregressions23Step-tworegressions241. IntroductionWithborrowingand1endingatthecoreofbankbusinessmode1s,creditassessmentsareessentia1forbanks'riskmanagement.Credit1ossesfeaturedprominent1yinthegreatfinancia1crisis(GFC)(egBernanke(2009),C1aessenseta1(2010)and60%to80%ofinternationa11yactivebanks'capita1requirementsref1ectcreditrisk(BCBS(2023b),p50-51).Today,debtservicecostsseempoisedtoriseonthebackofsignificantdebtaccumu1ationduringthe1ow-for-1ongeraandrecentinterestratehikes.Underp1ausib1escenarios,theriseofthesecostsmaydrivecredit1ossesuptoGFC1eve1s(BIS(2023).U1timate1y,accuratecredit-1ossforecastsand/orregu1atoryconservatism-ie1argerprecautionarye1ementsinthemappingfromtheforecaststocapita1requirements-wou1dbeneededtoensureenoughresourcesforabsorbingcredit1ossesthatexceed1oan1ossprovisions.Weuseanove1datasetonbanks'credit1ossforecasts-orexpected1oss(E1)rates-toassesstheiraccuracy.WejuxtaposetheseE1rateswithbanks'actua11oss(A1)ratesandidentifydriversofthediscrepancies.Attheheartoftheexerciseareexc1usiveconfidentia1supervisorydata,co11ectedbytheBase1CommitteeonBankingSupervision(BCBS).Thesedatacontainone-yearE1rates,asreportedby65internationa11yactivebankstosupervisorsfromend-2008toend-2023.CombiningE1ratesonnon-defau1tedexposureswithvendoraccountingdataonA1rates,weanswertwoquestions:2(i)howwe11doE1ratescapturetheevo1utionofA1rates?and(ii)doestherank-orderofE1ratesacrossbanksa1ignwiththatofA1rates?Wea1sore1atethediscrepanciesbetweenE1andA1ratestobank-1eve1characteristicsandmacro-financia1variab1es.E1ratesperformdifferent1ya1ongthetimeandcross-sectiondimensions.First,wefindthattheygenera11yfai1tocapturethetimeprofi1eofA1rates.AsE1ratestendtomissbothspikesandtrendsinA1rates,thecorre1ationofyear-to-yearchangesinthetwoseriesisstatistica11ysignificantforon1y15%ofthebanks.Thisimp1iesthataconservativemappingfromE1ratestocapita1requirementsisneededtoensureenough1oss-absorbingresourcesateachpointintime.Weestimatethat-overoursamp1eperiod-suchconservatismwou1dresu1tincapita1being(at1east)twiceas1a