Modified logistic regression using the EM algorithm for reject inference

Abstract

Reject inference is one of the key processes required to build relevant credit scorecard models. Reject inference is used to infer the good or bad loan status to credit applicants that were rejected by the financial institution. If rejected applicants data is not used in the updating of the credit scoring model, the model is biased because it will not be representative of the entire applicant population. Many reject inference techniques perform an extrapolation method to infer the good or bad loan status of the rejected applicants. The issues with extrapolation are discussed, and this study provides a novel reject inference technique in which the rejected applicants are included in the model estimation process. The extrapolation problem is avoided using the methodology in this paper. The newly proposed reject inference technique is shown to outperform the standard extrapolation technique using a simulation study.