credit risk modeling
TRANSCRIPT
ProposalforCreditRiskModelingUsingSta6s6calTechniques
I. Interviewclient.Planstrategically:Takemanagement’svisionandconvertitintoanac6onablepathway.
II. Determinehowrecommendedbankingguidelinesfitwiththeneedsoftheclientwithintheregulatoryframework.
III. DiscusshowproposedCreditRiskModelingsta6s6caltechniquesforkeyCreditRiskparametersPD,LG,EADmaybeusedandsuggestmodelvalida6ontechniques.
IV. Reviewhowstrategicobjec6vesweremetandwhythesepar6cularsta6s6calanalysismethodswouldfitwiththerecommenda6onsofBaselII.
Author:C-GStefanita 1
EstablishKeyRequirements
DevelopaPlan
AllocateResources
TrackProgressthroughMetrics Assess
Deliverablesagainst
Requirements
Author:C-GStefanita 2
PARTI.StrategicPlanning:• DetermineClient’sNeeds• PlananAc6onablePathway
PARTII.RegulatoryFramework&BankingGuidelines
June2004-BaselCommiXeeon
BankingSupervision(BCBS)
issuesBaselII
RevisedFrameworkonInterna6onalConvergenceof
CapitalMeasurement
ResearchTaskForce
(RTF)
whichisaresultsintheforma9onofa
AccordImplementa6onGroup(AIG)
Op6onsforimplemen6ng
BaselII
ComprehensiveCapitalAnalysisandReview(CCAR)regulatory
frameworkintroducedbytheFederalReserve
asubgroupofRTFgives
onafederallevelresultsin
CCARassessessoundnessof
internalcreditriskmeasurementbutbanksdeveloptheirownmethodologies
Bankshavetoworkwithintwointernal-ra6ngsbased(IRB)approaches:a
founda6onapproachandanadvanced
approach
DependingonIRBapproachfollowed,banksareallowedto
usetheirowninternalmeasuresforkeydriversofcredit
risk
asaresult nevertheless
BCBS–BaselII–CCAR–followedbyBaselIIIthatfocusesprimarilyoncreditrisk
Author:C-GStefanita 3
PARTII.–con6nued:
ClientNeed:Bankshavetodeveloptheirowncreditriskmodelssubjecttofederalscru6ny.
DependingonIRBapproach,banksreviewandanalyzevalida6on
methodologies&classifyra6ngssystems
Banksundertakevalida6ontoensureinternalra6ngssystemissuitablefor
internaluses
Banksneedtotakeintoaccountdifferencesin
quan6fica6onapproaches
Commonra6ngssystems:Point-in-6me(PIT)
Through-the-cycle(TTC)
SubjecttoFederalReserveapprovaluponmee6ng
certaincondi6ons
ClientNeed:Valida6onofInternalRa6ngsSystems
Ra6ngssystemsareacornerstoneforthecalcula6onofbanks’regulatorycapitalchargeintheIRBapproachofBaselII.
Ra6ngssystemsarethebasisforthedetermina6onofaborrower’sPD,alsotheothercreditriskfactorsLGD,EAD.Valida6oniskey.
Author:C-GStefanita 4
PARTII.–con6nued:
ModelDesign
RiskComponents
Valida6onofra6ngs
systembyabank
Banksmustdemonstratethattheycanassesstheperformanceoftheirinternalra6ngsandtheirriskes6ma6onssystemconsistently.
Realizeddefaultrateshavetobewithin
expectedrange
Banksmustuse
differentquan6ta6vevalida6on
tools
Internalstandardsmustexistfor
significantdevia6onsofobservedvaluesofriskcomponentsfromes6matedvalues.
2.
KeyDriversofCreditRisk:ProbabilityofDefault(PD)LossGivenDefault(LGD)ExposureatDefault(EAD)
3.
1.
Author:C-GStefanita 5
PARTIII-PlanSta6s6calTechniquesforKeyFactorsofCreditRisk
Author:C-GStefanita 6
1.Uselogis6cregression(logit)basedonanAltmanscoringmodel(oraprobitmodel)todeterminetheweightsofthefactors
contribu6ngtoPD;involvesMaximumLikelihoodEs6ma6on(MLE)–seeseparatedocumentfora
solvedexample.
2.Testthevalidityofthemodelbyperformingaone-samplet-testforthees6matedbcoefficients.
Usemul$variateANOVAtocomparedifferentPD‘buckets’.Considerdifferentscenarioswhereonly
somebcoefficientsarezero.Thisisamoreadvancedanalysisthatdiffersfromthecommonlyusedtextbook
ANOVA.ThegoalistodeterminewhichfactorscontributemoretoPD.Analysisinvolvesusingmatrix
algebratoimplementANOVAtheory.
PartIII–con6nued:
1.Calculateobligor-specificPDswithina‘bucket’byes6ma6ngweightsofexplanatoryvariables.
2.Validatemodelifes6matedweightsaresta6s6callymeaningful.A. WithinaPD‘Bucket’
LogitorProbit,MLE,t-test
B.BetweenPD‘Buckets’Mul6variateANOVA
1.
ProbabilityofDefault(PD)–Non-StructuralModel
Obligorsareassignedtoa‘bucket’dependingonobligor-specificPDs.
PooledPDsreflectwithin-’bucket’averagesofobligor-specificPDs.
Author:C-GStefanita 7
CreditPorgolioRiskModel
ProbabilityDistribu6onofLosses
AssetValueorLatentValueApproach
producesa
arisesfromaporMolioofcreditriskyinstruments
wecanuse
answersques9onofprobabilityoflossesonloanporMolioexceedingforexample$100Minayear
ObtainedthroughMonteCarloSimula6on
StartbyUsingSimplifiedApproach:ConsiderLossesfromDefaultOnly,notChangesinMarketValue
SpecifyPDsofindividualcredit
events
SpecifyLGDas%ofEADlostat
default
Specifycorrela6onsofindividualcredit
events
ObtainPorgolioValueDistribu6onthroughModeling
PartIII–con6nued:
MeasuringCreditPorgolioRiskwiththeAssetValueApproach
Author:C-GStefanita
LossGivenDefault(LGD)2.
MarketLGD
ImpliedMarketLGD
ImpliedHistorical
LGD
WorkoutLGD
FourModels
• Mostcommonlyused• Basedondiscountedcashflowsajerdefault
Basedonpricesoftradeddefaultedloans
Derivedfromnon-defaultedbondpricesthroughassetpricingmodel
• Forretailporgolio• BasedonexperienceoftotallossesandPDes6mates
Cri6calissues: • Howrecoveriesaremeasured• Howworkoutcostsareallocated• Howappropriatediscountfactorsareallocated
8
PartIII–con6nued:
PDes6ma6onstronglydependsonquan6fica6ontechniques,andvalida6onhastwostages
PARTIV–Review:FitwithBaselIIRecommenda6ons?
PDes6ma6onfromhistoricaldefaultrates(scoringmodel)ismostmeaningful
whenpooledPDsareunstressed
Theseareunbiasedes6matesofthelikelihoodofdefaultinthefollowingyearfor‘bucket’PDs
Along-runTTC‘bucket’ofPDswillnotprovidegoodes6mates
ofunstressedpooledPDs
UnstressedpooledPDstendtobelowerthanlong-runaveragedefaultfrequencyduringcyclicalpeaksandhigherduring
cyclicaltroughs
Sta6s6calmodelmaybepreferredthatusesahybridbetweenstructural
(Merton)andnon-structural(Altman)modelswithupdatedmarketdata
Valida6onofdiscriminatory
powerofra6ngsystem
Valida6onofaccuracyof
quan6fica6on(calibra6on)
Author:C-GStefanita 9
PDEs6matesDetermineLGD,EAD