Serial correlation, non-stationarity, and dynamic performance of business failures prediction models

This article examines the implications of serial correlation of the financial variables on the dynamic performance and robustness of the business failure prediction models based on the linear discriminant analysis, Logit, and Cumulative Sums (CUSUM)methods. Statistical tests show that most of the financial variables included in business failure prediction models exhibit strong positive serial correlation over time and in many cases a unit root. As a result, the predictive ability of these types of models deteriorates over time.