The Economic Value of Reject Inference in Credit Scoring

We use data with complete information on both rejected and accepted bank loan applicants to estimate the value of sample bias correction using Heckman’s two-stage model with partial observability. In the credit scoring domain such correction is called reject inference. We validate the model performances with and without the correction of sample bias by various measurements. Results show that it is prohibitively costly not to control for sample selection bias due to the accept/reject decision. However, we also find that the Heckman procedure is unable to appropriately control for the selection bias. † Data contained in this study were produced on site at the Carnegie-Mellon Census Research Data Center. Research results and conclusions are those of the authors and do not necessarily indicate concurrence by the Bureau of the Census or the Carnegie-Mellon Census Research Data Center. Åstebro acknowledges financial support from the Natural Sciences and Engineering Research Council of Canada and the Social Sciences and Humanities Research Council of Canada’s joint program in Management of Technological Change as well as support from the Canadian Imperial Bank of Commerce.