Credit Risk Measurement and Ethical Issues: some Evidences from the Italian Banks

Rating the “not-for-profit” sector is a complex issue due to its specific purpose and form of activity. Moreover, the challenge for financial institutions, working with these type of organizations, of assigning them to the appropri-ate rating scale is a relevant topic nowadays due to the their sharp increase over the last decades. The main discriminating criterion, in comparison to those of conventional profit oriented companies, relies on a different inter-pretation of the classical indicators, which contribute to the organization’s debt service capacity, combined with qualitative factors. This paper gives a contribution in variable identification within credit scoring models using Random Forest, when the observed units are divided into a number of groups, according to some categorical variables. We present an application to a real dataset provided by Banca Popolare Etica, an Institution focused on the “not-for-profit” sector and a pioneer in credit risk measurement techniques which integrate traditional quantitative factors and qualitative ones, related to the social-environment of customers. The standard algorithm is modified in the selection procedure in order to assess the impact of the two grouping variables (“not-for-profit” – “for-profit”; bankrupt – non bankrupt) on the variable importance measure. The technique is applied separately to “for-profit” and “not-for-profit” enterprises in order to extract foremost important variables in the framework of the tree-based learning ensembles and compare results obtained for the two types of organizations considered. Classical classification techniques (logistic, discriminant, neural networks) are compared in terms of discriminatory power and probability of ”concordance”.