Mixture Models Applied to Reject Inference

• Activity Number: 68
• Type: Contributed
• Date/Time: Sunday, August 6, 2006 : 4:00 PM to 5:50 PM
• Sponsor: Section on Risk Analysis
• Author(s): Billie Anderson*+ and J. Michael Hardin and Ana Landeros and Michael Conerly
• Companies: The University of Alabama and The University of Alabama and The University of Alabama and The University of Alabama
• Keywords: credit scoring ; reject inference ; mixture models ; EM algorithm

Abstract

Credit Scorecards are commonly built using data available within an organizations transactional database. Such data, however, will only contain information for those applicants who were 'accepted' or awarded credit by the organization; data will not be available for those applicants who were 'rejected'. The use of reject inference to adjust credit scorecard models for the missing data represented by rejected loan applications is common practice and several approaches are used in today's financial industry. In this paper, we investigate the use of mixture models as an alternative approach for reject inference. To estimate the parameters of the mixture model, we use the EM algorithm (Dempster, Laird, and Rubin, 1977) wherein the data associated with the 'rejected' applicants is treated as missing completely at random (Rubin, 1976).