Dipartimento di Scienze Statistiche

This work aims at identifying an optimal set of features for predicting rms bankruptcy events in the current macroeconomic context. Firstly, we assess the eectiveness of Support Vector Machines (SVMs) in comparison with other commonly used methods, considering a wide set of variables known in literature and new country-specic macroeconomic factors. Secondly, we select optimal subsets of variables through a feature selection method. The results show that the conjunct use of SVMs and the proposed feature selection technique signicantly improves the accuracy of forecasts. Furthermore, we show that nowadays the proposed country-specic factors are relevant information for predicting bankruptcy events.