Using Non-Parametric Count Model for Credit Scoring

Abstract

The purpose of this paper is to apply count data models to predict the number of times a credit applicant will not pay the amount awarded to repay the credit. Poisson models and negative binomial distribution models, taking into account the observed heterogeneity, are generally used in situations where the dependent variable is discrete. Alternatively, we propose to use non parametric model where the relationship form between conditional mean and the explanatory variables is unknown. The empirical results found suggest that the nonparametric poisson model regression has the best prediction of the number of default payment.