VALIDÁCIA PREDIKČNÝCH BANKROTOVÝCH MODELOV V PODMIENKACH SR

The prediction of bankruptcy has been the major subject of many studies since first study in this area, carried out by Fitz Patrick (1931). Many economists from all over the world have been trying to find company’s bankruptcy forecasting model using different methods with the aim to achieve the best results. For this purpose discriminant analysis, probit and logit analysis have been usually used. As mentioned statistical methods needed to meet assumptions as linearity, normality and independence among predictor variables, new methods with nonlinear nonparametric properties such as neural networks have been developed and applied. Despite of many advantages of neural networks, still the above mentioned – classical statistical methods have been mostly used. In the presented paper we provide a review of bankruptcy prediction studies divided into two time periods: before and after year 1966. Three of the bankruptcy prediction models: Altman model, Beerman discriminatory function, Index IN05 have been chosen for the validation on the real data of companies established in Slovakia. We have developed new modified model while using regression analysis to get higher predictive performance on analysed sample than chosen models. To validate selected bankruptcy prediction models performance we have chosen approach based on the data minig validation methods. Hence, our study is focused on the performance evaluation at two levels: precision – proportion of correctly predicted bankruptcy of totally predicted bankruptcy; and recall – proportion of correctly predicted bankruptcy of really bankrupted companies. Based on the matched sample of 1560 firms from the time period 1993–2007, our findings based on precision and recall indicate, that chosen models are inappropriate for Slovak economy and new quest for new models should be undertaken.