INFLUENCE OF THE SAMPLING METHOD ON BUSINESS FAILURE PREDICTION MODELS

Since Beaver's and Altman's work, business failure prediction has become an important topic in corporate finance literature and numerous studies have been developed, using a variety of statistical methodologies for predictive purposes. Most previous models have used a paired sample with the same number of failed and non-failed firms, achieving good prediction results. Nevertheless, this kind of sample has the drawback of not being representative of the population from which it is chosen, since this sampling method does not respect the population proportions on the sample. In order to prove if the predictive power of the previously developed models is due to the kind of sample they have used, we develop a failure prediction model on the small and medium-sized firms with head offices located in the region of Castilla y León (Spain), using both a paired sample and a random sample based on the population size and composition. Applying a logistic regression analysis, we try to identify the financial ratios, used as independent variables in the respective models, which best explain and predict failure in the two samples. The obtained results show that there are differences in the variables which become significant in each sample, as well as in the classification results. They are not as high in the random sample as those achieved in the paired one, especially with regard to failed firms, due to its low proportion compared to the non-failed ones.