How Good Is "Good"? : Making Better Use of Subjective Information in Bank Internal Credit Scoring Systems

摘要: Lenders experience positive net revenue impacts from lending if they increase the classification power of their credit scoring systems. If loan officers’ subjective assessments otherwise intangible borrower characteristics contain additional information about a borrower, lender may improve default forecast quality his internal systems by utilizing this information. The Basel II regulatory framework requires lenders to use all available both and nonsubjective, but at same time produce consistent objectified ratings. However, soft is often laden with inconsistencies due lack comparability different raters’ existence incentives manipulate rating. These leave expensive acquire only limited lenders’ It objective thesis introduce empirical methods that allow analyze in more sophisticated way, treat data facts. Instead using total scores scorecards as an indicator customer’s probability default, we rating patterns applying latent trait models borrowed psychometrics. We set 20,000 SME (Small Medium Enterprises) observations, including hard (financials, account behavior) (scorecard responses). Applying Mixed Rasch Model, six response pattern classes are identified our such that, within each class, item responses independent there no redundancies. interpretation analysis provide managers usage scorecard, them develop monitoring tools, mitigate adverse rater behavior. A new score constructed classes’ individual rates power. To compare alternative ROC (Receiver Operating Curve) inspection related measures. find making better already existing information, lender’s system can be significantly increased without affecting front end processes.