Genetic algorithm approach to design covariates of binomial Logit model for estimation of default probability

Credit risk management is one of the most important tasks of financial institutes. Default probability is the probability that a company will go into default, or be unable to fulfill an obligation, and it is a critical information for credit administration. Binomial logit model is widely used for default probability estimation. The formulas for computing covariates used in the model are designed by human experts in trial-and-error way, based on their experience. In this paper, we propose a method to design covariates. Integer-coded GA is employed and a representation of the chromosome is proposed for the purpose of optimizing the covariate. The method optimizes the covariates using the GA and estimates the coefficient of binomial logit model using Broyden-Fietcher-Goldfarb-Shanno method. The method is tested on an actual data provided for evaluation by a bank. The result of the experiment shows the method outperformed the human design.