Choosing the Best Set of Bankruptcy Predictors

Discriminant analysis and logit analysis are traditionally used to predict company bankruptcies. More recently, neural networks have been shown to outperform these statistical methods for the problem. The selection of financial indicators as independent variables for all these methods is a major problem. In this paper, we let a genetic algorithm operate on a set of variables. The goal is to find the optimal set of financial indicators for the bankruptcy prediction problem. Keywords: Genetic algorithms, financial indicators, bankruptcy prediction, accounting. 1 Introduction Bankruptcy prediction is as interesting today as it was in 1930's when Fitzpatrick 's study was published [Fit32] and during the last 60 years an impressive body of both theoretical and empirical research concerning this topic has evolved. The research focus has been on both finding the best prediction methods and the best failure indicators. From 1966 until 1980's discriminant analysis was the dominant method in ..