Effectiveness of bankruptcy prediction models constructed for differently selected diagnostic variables
Effectiveness of bankruptcy prediction models constructed for differently selected diagnostic variables

Summary/Abstract: The purpose of the article is to compare the effectiveness of discriminant models for predicting corporate bankruptcy, constructed using different methods of diagnostic variables selection. We compared several methods, such as arbitrary selection, the forward stepwise method applied after the initial selection of variables by means of a significance test of differences between group averages (further called the two-step method), the Hellwig method, the tstatistics and the backward stepwise method. We also assessed the models’ accuracy in terms of the synthetic measure. It was constructed by applying eight measurements of the classification effectiveness, such as the values of Wilks' lambda statistic and AUC together with the percentage of correctly identified companies, i.e. total, bankrupts and non-bankrupts in training and testing sets. The results show that the backward stepwise method and the two-step method generate models with the highest accuracy of classification. In addition, the study found that Wilks' lambda statistic is not a good approximation of the classification abilities of bankruptcy models. The contribution of our paper is a comparative methodological study, focusing on the impact of alternative diagnostic variable selection techniques on the linear discriminant function accuracy used to bankruptcy prediction.

• Details
• Contents