Is there a trade-off between the predictive power and the interpretability of bankruptcy models? The case of the first Hungarian bankruptcy prediction model

In our work, we compare the predictive power of different bankruptcy prediction models built on financial indicators calculable from businesses’ accounting data on the database of the first Hungarian bankruptcy model. For modelling, we use data-mining methods often applied in bankruptcy prediction: neural networks (NN), support vector machines (SVM) and the rough set theory (RST) capable of rule-based classification. The point of departure for our comparative analysis is the practical finding that black-box-type data-mining methods typically show better classification performance than models whose results are easy to interpret, i.e. there seems to be a kind of trade-off between the interpretability and predictive power of bankruptcy models. Empirical results lead us to conclude that the RST approach can be a competitive alternative to black-box-type SVM and NN models. In our research, we did not find any major trade-off between the interpretability and predictive performance of bankruptcy models on the database of the first Hungarian bankruptcy model.