Research On Sample Bias Correction Of Housing Loan Credit Scoring Based On Case-Based Reasoning

With the rapid development of the real estate industry and the house mortgage loan market, the credit risk cumulated in house mortgage loan is gradually emerging. The academia and the banks now realize the importance of credit risk management in house mortgage loan, especially the significant role of credit scoring in reducing default risk and improving the asset quality. However, previous researches focus mainly on the improving of predicting accuracy of personal credit scoring models, while an essential problem– Rejection Bias, the problem of sample selection bias- is ignored. In credit scoring of house mortade loan, the main cause for sample selection bias is that the credit scoring models are based on the data of accepted appliers, rather than all the appliers.In view of the Rejection Bias, we first analyzed the sample selection bias in credit scoring models, and conducted research on sample selection bias correction techniques. Case-Based Reasoning was brought forward to perform credit scoring and correct the sample selection bias, and basic theory and steps were expatiated thoroughly. In order to form the Case-Based Reasoning of house mortage loan credit scoring, we then analyzed the concept, characteristics and inherented risks of house mortage loan in China. Drawn on the research of influencing factors on house mortage loan credit scoring, we built an index system in the four aspects of applicant features, loan features, house features and derived variables, and set forth the procedures of Case-Based Reasoning of house mortage loan credit scoring based on sample bias correction. At last, with house mortage loan data from a commercial bank, empirical study was carried out. The result of Case-Based Reasoning of house mortage loan credit scoring shows that sample selection bias was corrected effectively, and classification results were more satisfying. Meanwhile, because of the difference in credit scoring mechanism between Case-Based Reasoning and traditional methods, this study provides broader clues for further research.