A systematic approach to construct credit risk forecast models

Abstract

Due to the recent growth in the consumer credit market and the consequent increase in default indices, companies are seeking to improve their credit analysis by incorporating objective procedures. Multivariate techniques have been used as an alternative to construct quantitative models for credit forecast. These techniques are based on consumer profile data and allow the identification of standards concerning default behavior. This paper presents a methodology for forecasting credit risk by using three multivariate techniques: discriminant analysis, logistic regression and neural networks. The proposed method (deemed the CRF Model) consists of six steps and is illustrated by means of a real application. An important contribution of this paper is the organization of the methodological procedures and the discussion of the decisions that should be made during the application of the model. The feasibility of the approach proposed was tested in a program for granting credit offered by a network of pharmacies. The use of the models for forecasting credit risk greatly reduces the subjectivity of the analysis, by establishing a standardized procedure that speeds up and qualifies credit analysis.