Article ID: | iaor2002781 |
Country: | United Kingdom |
Volume: | 5 |
Issue: | 1 |
Start Page Number: | 45 |
End Page Number: | 55 |
Publication Date: | Jan 1993 |
Journal: | IMA Journal of Mathematics Applied in Business and Industry |
Authors: | Hand D.J., Henley W.E. |
Keywords: | credit scoring |
The true good/bad status of applicants accepted for credit is ultimately known. However, the status of rejected applicants will never be known. ‘Reject inference’ is the process of inferring the status of applicants who have been rejected. This paper reviews methods of reject inference, and describes some new approaches. Three classes of method are described: (i) methods based on extrapolating a model built on the accepted applicants into the reject region; (ii) methods based on the distribution of the rejected applicants; (iii) methods using supplementary information. In particular, we conclude that the distribution of the rejected applicants cannot assist reject inference unless additional assumptions are made.