Predicting corporate failure : how useful are multi-discriminant analysis models?

Abstract

The aim of the paper is to present how multi-discriminant models (MDA) perform in practice and to measure these models' effectiveness in bankruptcy prediction. For this purpose an ex-ante approach is adopted to emulate the way in which the models are used in practice. Thus two commercially applied models, Altman's and Datastream's, are presented and examined on independent samples of companies. The findings are that these two models have a very similar predictive ability and that the prior probability of failure is an important feature in determining this ability. The general conclusion of the paper is that the use of MDA models as predictors of bankruptcy can involve major understatement of classification errors. Therefore the robustness of these models as well as the acceptability of using the models as the sole means of assessing potential bankruptcy of companies could be doubtful. The paper fills a gap in the literature on independent testing of the developed MDA models. We stress the importance of shifting the threshold and consequently we show the impact of the choice of the threshold in a practical setting. (original abstract)