A statistical model for credit scoring

Abstract

Acknowledgements: I am grateful to Terry Seaks for valuable comments on an earlier draft of this paper and to Jingbin Cao for his able research assistance. The provider of the data and support for this project has requested anonymity, so I must thank them as such. Their help and support are gratefully acknowledged. Participants in the applied econometrics workshop at New York University also provided useful commentary. This chapter is based on the author's working paper ‘A Statistical Model for Credit Scoring’, Stern School of Business, Department of Economics, Working Paper 92–29, 1992. Introduction Prediction of loan default has an obvious practical utility. Indeed, the identification of default risk appears to be of paramount interest to issuers of credit cards. In this study, we will argue that default risk is overemphasized in the assessment of credit card applications. In an empirical application, we find that a model which incorporates the expected profit from issuance of a credit card in the approval decision leads to a substantially higher acceptance rate than is present in the observed data and, by implication, acceptance of a greater average level of default risk. A major credit card vendor must evaluate tens or even hundreds of thousands of credit card applications every year. These obviously cannot be subjected to the scrutiny of a loan committee in the way that, say, a real estate loan might. Thus, statistical methods and automated procedures are essential. Banks and credit card issuers typically use ‘credit scoring models’.