Artificial Factors Within the Logit Bankruptcy Model with a Moved Threshold

Bankruptcy models are by default based on financial indicators derived from corporate financial reports. However, based on these reports, it is possible to define tens or hundreds of indicators, which makes the process of building a model more demanding. This article demonstrates the process of estimating a logistic regression model via artificial factors, which allows information from a large number of variables to be retained, but at the same time prevents the very common problem of the multicollinearity of variables. This article also pays attention to the optimisation of the classification threshold, which allows the error rate in the identification of both majority active and minority bankruptcy companies to be balanced. The empirical results of this article show that for the area of bankruptcy prediction, it is appropriate to set a limit of at least 85% of the explained variability when identifying a suitable number of artificial factors. The involvement of artificial factors together with the threshold optimisation has a positive effect on the classification capabilities of the model and also on its applicability in practice. The results also show that the proposed procedure is applicable at least to related manufacturing sectors.