Does the development context affect bankruptcy prediction models' general accuracy? A comparative analysis of four multivariate discriminant models in the Italian context.

This research starts from the study by Grice & Dugan (2001) which verified the sensitivity of bankruptcy prediction models to changings in the features of the investigated sample. The study showed that the investigated models’ accuracy was affected by the time-frame, the industry and the size of the firms which composed the investigated samples. Given these premises, we hypothesized that models applied to samples similar to the one used in their development should reach higher degrees of accuracy than models developed within different contexts. In order to verify this hypotesis, we tested four multivariate discriminant models, one developed within the American context and the others developed in Italy: Altman’s Z’-Score model (1993); Alberici’s model (1975); Bottani, Cipriani and Serao’s model (2004) and Luerti’s model (1992). We tested the four models twice: first on a sample of firms gone bankrupted within 2012 and 2014. Then on a sample equally composed by bankrupt and operating firms. Both samples were composed by firms located in the Emilia-Romagna region, in Northern Italy. The first phase of the analysis aimed at verifying the models’ predictive capacity, while the second phase aimed at verifying theirs discriminant capacity. The accuracy of the models was then assessed comparing the results of their application with the real status of each firm. The results show that even if the Italian models were developed using samples more similar to the one investigated in this research, Altman’s model reaches the highest degree of accuracy.