Credit scoring is fundamentally an exercise in predicting future repayment behaviour from currently available information. Reject inference addresses a specific weakness in that exercise: the model learns mainly from applicants who were approved in the past, even though it will be used to judge the wider application population. By estimating the possible performance of rejected applicants, reject inference attempts to reduce this sample-selection problem. However, because the rejected applicants’ true outcomes were never observed, it should be treated as a modelling technique based on assumptions—not as a way of discovering historical facts. Click Here to read
Other Research & R&DReview
Reject Inference Technique
Reject inference addresses a specific weakness in that exercise: the model learns mainly from applicants who were approved in the past, even though it will be used to judge the wider application population.