Analisi e previsione del rischio di default nelle piccole e medie imprese italiane: un approccio modellistico

Rapidly identifying signs of deterioration in the creditworthiness of small and medium-sized enterprises (SMEs) represents, for the Italian banking system, a strategic lever to reduce the incidence of non-performing loans, strengthen the effectiveness of credit granting and monitoring processes, contributing to the achievement of a sustainable competitive advantage for the banking institution. Predictive models of default risk, despite integrating economic and financial variables, are insufficient in capturing the effects of the qualitative size of the company being assessed. This is especially true for SMEs, which are characterised by less information transparency, a marked dependence on bank credit and different ownership and management structures compared to larger companies. This thesis analyzes the role and effects of economic and financial variables in the prediction of default risk and also tests the contribution of qualitative governance variables. The aim is to verify whether the integration of the latter can improve the predictive capacity of the models in question. In line with the objective of the research, the empirical analysis conducted develops a logistic regression model, applied to a sample of Italian SMEs composed of defaulted firms and solvent firms, in which the dependent variable identifies default as an observable ex-post event, while the explanatory variables include financial ratios, control variables by sector and geographical area in Italy and governance variables. To assess the specific contribution of the block of governance variables to the classification of defaulted firms, the performance of the two models is compared. The results confirm the central role of economic and financial variables in the prediction of default and highlight that the integration of governance variables into the model significantly improves predictive performance, increasing its accuracy, sensitivity, specificity and AUC compared with the baseline model.