Bankruptcy prediction models: How to choose the most relevant variables?

This paper is a critical review of the variable selection methods used to build empirical bankruptcy prediction models. Recent decades have seen many papers on modeling techniques, but very few about the variable selection methods that should be used jointly or about their fit. This issue is of concern because it determines the parsimony and economy of the models and thus the accuracy of the predictions. We first analyze those variables that are considered the best bankruptcy predictors, then present variable selection and review the main variable selection techniques used to design financial failure models. Finally, we discuss the way these techniques are commonly used, and we highlight the problems that may occur with some non-linear methods.