Bankruptcy Prediction Models: Can the prediction power of the models be improved by using dynamic indicators?

The present approach to developing bankruptcy prediction models uses financial ratios related to the time of one year before bankruptcy. Some authors try to improve the prediction accuracy of the models by using averaged ratios involving several years before bankruptcy. This of course assumes that a bankruptcy can be predicted several years ahead. This idea led us to investigating the differences between the dynamics of the financial ratios developments. Here we assume that the dynamics of the values of some indicators in a group of prospering companies may be different from that of those facing bankruptcy threats. The indicators that showed a significant difference in the development dynamics were used to develop a bankruptcy prediction model. The research was carried out using data of the Czech manufacturing industries obtained from the AMADEUS database for years 2004 to 2011, with each company providing data for up to five years prior to the bankruptcy. Along with investigating the different approach to the selection of indicators for the development of a bankruptcy model, we were also concerned with the selection of a method to develop it. Researching the literature, we found that the most commonly used method is one of linear discrimination analysis, whose precision is improved if applied to normally distributed data without outliers. With financial data, however, these assumptions are difficult to meet. Therefore, a non-parametric boosted-trees method was used to select the predictors and develop the bankruptcy models. (C) 2014 Elsevier B.V.