Connexionist Approach and Corporate Distress Diagnosis

Abstract

Over the last thirty years, and in particular up to the mid-1980s, evaluation of the risk of corporate bankruptcy has been the subject of many empirical research projects, using mostly linear discriminant analysis. The methodology used in that research has generated a great deal of criticism, thus favouring the emergence of various alternative approaches. The most recent one, called the connexionist approach, allies the dynamics of complex systems and the neuro-mimetic paradigm. It applies the neural networks’ properties of learning and generalization (or self-organization) to the prediction of corporate bankruptcy. Paradoxically, given the degree of sophistication of the statistical techniques used, higher than in any other numerical induction procedure, empirical research on corporate bankruptcy remains dependent on the quality of the data and in particular their degree of completeness. This problem may be solved by employing one of the following two techniques: the elimination cf those companies whose data are incomplete or the use of econometric methods to complete the series (Hachette 1994). The removal of some companies from the sample — the most commonly used procedure — introduces a methodological bias whereby the incompleteness does not affect distressed companies and the others equally.