Dionne, Georges, Manuel Artís, and Montserrat Guillén, "Count Data Models for a Credit Scoring System", Journal of Empirical Finance, Vol. 3, No. 3, (September 1996), pp. 303-325.

Abstract: In the recent literature on credit scoring, the emphasis was made on the estimation of default probabilities without any real effort of considering the different costs and benefits of the loans. However, an accepted loan may become a bad loan after some costly reminders and may even introduce collection and other bad debt costs. In this paper, we extend the hurdle model defined by Mullahy (1986) to estimate jointly the default probability and the two conditional truncated distributions of non-payments of good and bad loans respectively. We first show that the significant variables that affect the three distributions are not the same. Moreover, the two truncated non-payments distributions (before and after the identification of a bad loan) do not follow the same distribution. These results imply that limiting the analysis to the estimation of the default probabilities to evaluate credit scoring is not sufficient to obtain an appropriate evaluation of the ex-ante bank funding.

JEL Classification: G2, C35.

Keywords: Personal loans, Hurdle count data model, Truncated Poisson, Default probability, Credit scoring.

Books Referenced in this Paper:  (what is this?)

[Home] [Credit Scoring Papers]

[ Home ]

Please contact me with problems or suggestions.
Copyright © 2000-2008 DefaultRisk.com
Last modified: July 24, 2008