Reject Inference
From Open Risk Manual
Definition
Reject Inference is a set of methodologies aimed at addressing the problem that credit scoring is applied to data on the accepted population of clients rather than the total population because there will typically not be any performance data available for the "rejected" population.
The problem of reject inference can be considered also as a missing data problem
Practices
- Augmentation: After the scorecard is developed on the accepted data set, rejected cases are evaluated as good or bad on the basis of a score cutoff. The scorecard is then re-estimated using the total set
- Fuzzy augmentation: Similar as above, except that good/bad assignment of rejected applications is probabilistic (fuzzy)
- Iterative Reclassification
Validation
- Construction of sample with arbitrarily assigned "rejects"
Issues and Challenges
- Actual performance data on rejected populations is considered the preferred way to address this Blind Spot
