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

Outlines previous research on business failure prediction models and investigates the impact of serial correlation and non‐stationarity in financial variables on models based on linear discriminant analysis, logit and cumulative sums using 1974‐1991 data from a sample of failed and non‐failed US firms, plus a similar 1992 sample. Presents and discusses the time series behaviour of the explanatory variables, the estimation of the three types of models and their error rates over time. Concludes that models based on variables with strong positive serial correlation deteriorate over time in their forecasting power; and calls for research to develop stationary models.