Numerical methods for bankruptcy prediction: A comparison between data envelopment analysis and logistic regression

Abstract

This research examines the efficacy of Data Envelopment Analysis (DEA) as a model for bankruptcy prediction compared to the logistic regression method. A sample of failed and non-failed firms from the Untied States is obtained to test each model. The models are compared on both a within and out of sample basis, using different time periods and different sized samples. Logistic regression is found to be generally better at predicting bankruptcy on a within sample basis. However, DEA's relative strength is that the model does not require an estimation sample to be used as a predictive model. It is found that DEA does predict better when the estimation sample for the logistic regression is deficient or biased (usually when the sample size is small). However, it is impossible to determine if an estimation sample is sufficient on an ex-ante basis, since the goal is bankruptcy prediction. Thus, DEA is found to be useful for bankruptcy prediction, at least as a check of logistic regression results.