Data reduction and univariate splitting — Do they together provide better corporate bankruptcy prediction?

Abstract

Discussion on methodological problems of corporate survival and solvency prediction is enjoying a renaissance in the era of financial and economic crisis. Within the framework of this article, the most frequently applied bankruptcy prediction methods are competed on a Hungarian corporate database. Model reliability is evaluated by Receiver Operating Characteristic (ROC) curve analysis. The article attempts to answer the question of whether the simultaneous application of data reduction and univariate splitting (or just one of them) improves model performance, and for which methods it is worth applying such transformations.