PRZEWIDYWANIE UPADŁOŚCI POLSKICH PRZEDSIĘBIORSTW–ZASTOSOWANIE ANALIZY DYSKRYMINACYJNEJ. WYNIKI BA-DAŃ PILOTAŻOWYCH
BANKRUPTCY PREDICTION OF POLISH COMPANIES - THE APPLICATION OF DISCRIMINANT ANALYSIS

Summary/Abstract: The aim of the article is to present preliminary research results in the field of bankruptcy prediction models and compare methods of variable selection for the model. Whether models based on variables eliminated based on mutual correlation show better prediction than models with arbitrarily selected variables.The problem of bankruptcy in the economy is particularly significant in times of economic crisis. According to the PIE report, Poland and other European Union economies are expected to experience a clear slowdown in 2023. The number of corporate bankruptcies in Poland in the fourth quarter of 2022 amounted to 112, which was 28.7% higher than in the corresponding period of the previous year, according to the GUS report.Discriminant analysis was the method used in the research. In 6 models, indicators for constructing the discriminant function were selected by eliminating the most correlated variables. In 4 models, indicators were chosen arbitrarily.The conducted research revealed that statistical methods of variable selection for models are more effective than arbitrary selection of variables. The best prediction was observed in models based on data one year before bankruptcy. In the test sample, the models correctly classified 62.5% of non-bankrupt companies and 87.5% of bankrupt companies. Additionally, a comparative analysis of misclassified entities by the models was conducted. It was found that the models misclassified the same companies. This may indicate atypical cases that, with a small database, have a high percentage share in classification errors.

• Details
• Contents