Are failure prediction models transferable from one country to another? An empirical study using Belgian financial statements

Abstract

Faced with the question as to whether failure prediction models (multiple discriminant and logit analysis) from different countries can easily be transferred to other countries, this study examines the validity of a range of models on a dataset of Belgian company accounts, both when using the original and re-estimated coefficients. Firstly, contrary to expectations, models that show bad performance results with the original coefficients reveal an improvement after re-estimation of the coefficients, while models that perform well reveal a decrease in performance. On average, the failure prediction power of the models deteriorates after re-estimation of the coefficients. The Belgian Ooghe-Joos-De Vos and Ooghe-Verbaere models seem to be generally the best- performing models one and three year prior to failure. Furthermore, if the term of failure prediction is longer: (1) it seems more difficult to distinguish between models performing well and badly and (2) the average failure prediction abilities of the models decrease. Finally – rather than the estimation technique, the complexity and the number of variables - the type of variables included in the models appears to be an important explanatory factor for model performance. This study makes a strong case for including all different aspects of the financial situation in a failure prediction model