Does The Development Context Affect Bankruptcy Prediction Models’ General Accuracy? A Comparative Analysis Of Four Multivariate Discriminant Models In The Italian Context

Abstract

This research starts from the work by Madonna and Cestari (2015) that aimed at assessing the usability of three bankruptcy prediction models applied in contexts other than the ones of their elaboration, in order to evaluate their generalizability and the possibility to apply them in wide-scale investigations. We took the cue from that study to assess the usability of four bankruptcy prediction models, when applied to a sample with characteristics other than the ones related to their elaboration. We aimed at verifying the predictive accuracy and the discriminant capacity of the four models, basing on the assumption that the performances displayed by bankruptcy prediction models are usually better when they are applied in contexts similar to the one of their elaboration. Given this premise, we hypothesized that Italian models should perform better than the American one. In order to verify this hypothesis, we tested the four multivariate discriminant models twice: the predictive accuracy was tested applying the models on a sample of firms gone bankrupt within 2012 and 2014; the discriminant capacity on a sample equally composed by bankrupt and operating firms. Both samples were composed by firms located in Italy and operating in recent years. Hence the sample provided and the context of application were different from the ones of the models‘ elaboration. The results show that even if the Italian models were elaborated basing on contexts more similar to the one of the present application, the best performance is reached by the American Altman’s Z‘-Score model.