A NEW LOOK AT BANKRUPTCY MODELS

Abstract

New models for bankruptcy prediction are constantly being formulated and tested against the current ones and current ones are tested to assess their current accuracy and to allow users to determine the reliability of the results when using the model. These models use accounting information as input data. Accounting systems, for example, US GAAP, or IFRS, contain rules that may be applied differently from one company to another without being breached. This leads to input data uncertainty. Likewise, uncertainties may arise due to errors in recording and transcribing input data or in translating the values of assets, equity or liabilities in foreign currencies. This research was focused on the effect of entry data uncertainty on models’ ability to accurately predict bankruptcy. The initial assumption was that raising the number of input values would increase the error rate probability in entry data, thus also heightening the uncertainty of the results in the given bankruptcy prediction model. The data set of tested companies contained 1,220 non-bankrupt and 285 bankrupt Czech companies. The tested models – Z’ score, Model 1, and Ycz – were applied to this sample, and in all cases, the resulting accuracy was lower than the accuracy declared by their authors. A procedure was created for the inclusion of entry data uncertainty in the practical application of a model. This procedure consists of changing the limit value of the model that separates bankrupt and non-bankrupt companies to an interval that “absorbs” such uncertainties. The model cannot classify the companies in this interval. The research shows that the inclusion of uncertainties in entry data further reduces their accuracy. However, the reduction in accuracy between the individual models varies significantly from 2.2% to 39.4% for bankrupt companies, and from 3.5% to 91.8% for non-bankrupt companies, respectively. The analysis of the entry data uncertainty effect shows the need to create models with high precision and minimum of input values because the model error rate grows the higher their number. The findings of this research can be applied in the creation of new models for predicting bankruptcy not only in the Central Europe but globally.