What is at stake in the construction and use of credit scores?

Abstract

Most of statistical studies on credit scoring focus on scores construction. It is more unusual that they link the statistical technics with a detailed analysis of the users' requirements regarding the properties of these tools. Concerning companies' failure the users are financial analysis experts or bankers in credit risk departments or banking supervisors. The increasing need for better control of credit risk by banks has led to a stepping-up of research concerning credit scoring. In the context of the Basel II agreement, the International Banking Committee has stressed the importance of forecasting the expected loss (EL) and, using extreme quantiles, the unexpected loss (UL) for a population of companies, in particular for customers of each commercial bank. In order to do so, it is necessary to estimate the default probability of each company at a given time horizon (PD). The objective of an accurate forecasting gives rise to several needed properties and questions that are presented in Sect. 1. We stress what is at stake in the construction and the use of credit scores. The experience of Banque de France in prudential supervision and the importance of its data files on companies give the possibility to developp a scoring system able to fullfil these needed properties, at least partially. Some principles of credit scoring construction in order to increase the quality of the tool and the accuracy of default probability are presented in Sect. 2. Without leading a complete debate on models-choice we discuss some arguments regarding this choice and we concentrate on comparison between Fisher linear discriminant analysis (LDA) and logistic regression (LOGIT) in Sect. 3. In relation with the early detection of companies default, two pratical uses of a credit scoring system are presented in Sect. 4. Research under way on Banque de France data concentrates on informations that can be extracted from these data on purpose to study how to increase the quality of tools needed by the Basel II agreement. A short overview of this research is given in Sect. 5. © Springer Science+Business Media, LLC 2007.