Building Credit Scoring Models with SAS Enterprise

�����������������BuildingCreditScoringModelswithSAS®EnterpriseMinerTableofContentsIntroduction....................................................................................................................1Buildingcreditmodelsin-house..................................................................................2BuildingcreditmodelswithSAS®EnterpriseMiner..................................................2SAS®EnterpriseMinerprocessflowtemplates.........................................................2Thelargercreditscoringprocess................................................................................3Choosingtherightmodel..............................................................................................3Scorecards....................................................................................................................4Decisiontrees...............................................................................................................4Neuralnetworks............................................................................................................5Casestudy......................................................................................................................5Scenario........................................................................................................................5SAS®EnterpriseMinerprocessflow............................................................................6Developmentsample....................................................................................................6Classing........................................................................................................................7Scorepointsscaling....................................................................................................10Scorecardassessment...............................................................................................11Decisiontree...............................................................................................................14Modelcomparison......................................................................................................15Rejectinference..........................................................................................................16Summary.......................................................................................................................17References....................................................................................................................18Recommendedreading................................................................................................18CreditScoring.............................................................................................................18Datamining.................................................................................................................19BuildingCreditScoringModelswithSAS®EnterpriseMiner1IntroductionOverthepast30years,growingdemand,strongercompetitionandadvancesincomputertechnologyhavemeantthatthetraditionalmethodsofmakingcreditdecisionsthatreliedmostlyonhumanjudgmenthavebeenreplacedbymethodsthatemploystatisticalmodels.Statisticalmodelstodayareusednotonlyfordecidingwhetherornottoacceptanapplicant(applicationscoring),butalsoforpredictingthelikelihoodofdefaultsamongcustomerswhohavealreadybeenaccepted(behavioralscoring)andforpredictingthelikelyamountofdebtthatthelendercanexpecttorecover(collectionscoring).Thetermcreditscoringcanbedefinedonseveralconceptuallevels.Mostfundamentally,creditscoringmeansapplyingastatisticalmodeltoassignariskscoretoacreditapplicationortoanexistingcreditaccount.Onahigherlevel,creditscoringalsomeanstheprocessofdevelopingsuchastatisticalmodelfromhistoricaldata.Onyetahigherlevel,thetermalsoreferstomonitoringtheaccuracyofoneormanysuchstatisticalmodelsandmonitoringtheeffectthatscore-baseddecisionshaveonkeybusinessperformanceindicators.Creditscoringisperformedbecauseitprovidesanumberofimportantbusinessbenefits,allofthembasedontheabilitytoquicklyandefficientlyobtainfact-basedandaccuratepredictionsofthecreditriskofindividualapplicantsorcustomers.Forexample,inapplicationscoring,creditscoresareusedforoptimizingtheapprovalrateofcreditapplications.Applicationscoresenableanorganizationtochooseanoptimalcut-offscoreforacceptance,suchthatmarketsharecanbegainedwhileretainingmaximumprofitability.Theapprovalprocessandthemarketingofcreditproductscanbestreamlinedbasedoncreditscores.Forexample,high-riskapplicationscanbegiventomoreexperiencedstaff,orpre-approvedcreditproductscanbeofferedtoselectlow-riskcustomersviavariouschannels,includingdirectmarketingandtheWeb.Creditscores,bothofprospectsandexistingcustomers,areessentialinthecustomizationofcreditproducts.Theyareusedtodeterminecustomercreditlimits,downpayments,depositsandinterestrates.Behavioralcreditscoresofexistingcustomersareusedintheearlydetectionofhigh-riskaccounts,andtheyenableorganizationst