信用风险模型

IssuesinCreditRiskModellingRiskManagementSymposiumSeptember2,2000BankofThailandChotibhakJotikasthiraBankofThailandRiskManagementSymposium-September2000Page2Overview•BISregulatorymodelVsCreditriskmodels•CurrentIssuesinCreditRiskModelling•Briefintroductiontocreditriskmodels–Purposeofacreditriskmodel–Commoncomponents–Modelfrominsurance(CreditRisk+)–CreditMetrics–KMV•ModelcomparisonBankofThailandRiskManagementSymposium-September2000Page3BISRegulatoryModelVsCreditRiskModelsBISRisk-BasedCapitalRequirementsAllprivate-sectorloans(uncollateralized)aresubjectedtoan8percentcapitalreserverequirement,irrespectiveofthesizeoftheloan,itsmaturity,andthecreditqualityoftheborrowingcounterparty.Note:Someadjustmentsaremadetocollateralized/guaranteedloanstoOECDgovernments,banks,andsecuritiesdealers.BankofThailandRiskManagementSymposium-September2000Page4CreditRiskModels-CreditRisk+-CreditMetrics-KMV-OthersimilarmodelsBISRegulatoryModelVsCreditRiskModelsBankofThailandRiskManagementSymposium-September2000Page5DisadvantagesofBISRegulatoryModel1.Doesnotcapturecredit-qualitydifferencesamongprivate-sectorborrowers2.IgnoresthepotentialforcreditriskreductionvialoandiversificationThesepotentiallyresultintoolargeacapitalrequirement!!!!!BISRegulatoryModelVsCreditRiskModelsBankofThailandRiskManagementSymposium-September2000Page6BISRegulatoryModelVsCreditRiskModelsBigdifferenceinprobabilityofdefaultexistsacrossdifferentcreditqualities.CreditRatingProbabilityofDefaultAAA0.00%AA0.00%A0.06%CreditRatingProbabilityofDefaultBBB0.18%BB1.06%B5.20%CCC19.79%Note:1.Probabilityofdefaultisbasedon1-yearhorizon.2.HistoricalstatisticsfromStandard&Poor’sCreditWeekApril15,1996.BankofThailandRiskManagementSymposium-September2000Page7BISRegulatoryModelVsCreditRiskModelsDefaultcorrelationscanhavesignificantimpactonportfoliopotentialloss.KMVfindsthatcorrelationstypicallylieintherange0.002to0.15.8%8%BISmodelrequires8%oftotal.8%8%Correlation=1Correlation=0.15Actualexposureisonly6%oftotal.BankofThailandRiskManagementSymposium-September2000Page8BISRegulatoryModelVsCreditRiskModelsThecapitalrequirementtocoverunexpectedlossdecreasesrapidlyasthenumberofcounterpartiesbecomeslarger.Unexpectedloss#ofcounterparties1168%3.54%Assumption:Allloansareofequalsize,andcorrelationsbetweendifferentcounterpartiesare0.15.BankofThailandRiskManagementSymposium-September2000Page9CurrentIssuesinCreditRiskModellingAdaptedfrom“CreditRiskModelling:CurrentPracticesandApplications”,April1999,byBasleCommitteeonBankingSupervisionTopicConceptualIssues/ConcernsDefinitionofriskShouldcreditriskincludeonlydefaultorbothdefaultandratingmigrations?Isthereamaterialdifferencebetweenthedefaultmodeandthemark-to-marketmodemodels?RiskdriversWhendoesdefaultactuallyoccur?Inthethresholdmodels,whatobservablevariableshouldbeusedtorepresentabilitytopay?ModelconceptIsthemodelthatstartsfromapoolofsimilarloansorobligorsrealistic?Pooleddatausuallyhidecreditspecificrisks.ProbabilitydensityfunctionNoagreementonthefamilyofdistributionstouse.Lossdistributionisnotnormal;itempiricallyhasfattertails.CorrelationofcrediteventsHowshouldco-movementamongratingmigrationsanddefaultsbemodeled?Implicitorexplicit?ConditionalVsUnconditionalCurrently,mostmodelsareunconditional(independentfromthestateofeconomy).Usingthesemodels,riskcanunderstatedoroverstateddependingonthelocationwithinthebusinesscycle?BankofThailandRiskManagementSymposium-September2000Page10CurrentIssuesinCreditRiskModellingAdaptedfrom“CreditRiskModelling:CurrentPracticesandApplications”,April1999,byBasleCommitteeonBankingSupervisionTopicParameterSpecificationIssues/ConcernsLossgivendefault(LGD)LGDisrandom;hence,adistributionisneededtorepresentLGD.LackofsensitivityanalysiswithrespecttoLGD.Lackofhistoricaldatatovalidatecurrentlyusedmodels.Riskratings,expecteddefaultfrequency(EDF),andmigrationprobabilitiesIndeterminingEDFandmigrationprobabilities,Internalratingsystemsmaynotbeaccurateorhaveenoughhistory.EDFandmigrationprobabilitiesofpubliclytradedbondsmaynotbeaccurateforbankcredits.MostsystemscombineEDFandLGD.MigrationanddefaultcorrelationsIsitreasonabletouseequityinformationtoestimatecorrelationsforbankcredits?Lackofhistoricaldatatovalidatemodelsusedtoestimatethisparameter.BankofThailandRiskManagementSymposium-September2000Page11CurrentIssuesinCreditRiskModellingAdaptedfrom“CreditRiskModelling:CurrentPracticesandApplications”,April1999,byBasleCommitteeonBankingSupervisionTopicParameterSpecificationIssues/ConcernsCreditspreadsForMark-to-Marketmodels,howmuchspreadshouldbeusedtovalueloansateachcreditrating?Aretheforwardspreads(basedontodayyieldcurve)agoodapproximationofthefuturespreads?Howisliquidityelementofcreditspreadstakenintoaccount?ExposurelevelsDifferentinstruments(especiallymarketdriveninstruments)havedifferentlevelsofriskexposure(e.g.swapsvsloans).Estimatesaremadetomakedifferentinstrumentscomparable.Theaccuracyofestimatesisquestionable.ComputationalrequirementSomemodelsarecomputationallyintensive.BankofThailandRiskManagementSymposium-September2000Page12CurrentIssuesinC