Although there is no express requirement in the regulations of the National Bank of Romania, the design, the implementation and the use of rating systems in the management of risks related to exposures of non-banking financial institutions (NFIs) on the basis of good banking practice, is a valuable and indispensable in the efficient administration of risks, especially in the current crisis. In this framework, this study describes the synthetic process of reviewing the scoring system used by IFN. The mechanism for evaluation of scoring system performance combines the indicators of discrimination power with econometric results of studying the functional relationship between risk variables integrated in the model and the credit status (default/ reimbursement) based on logit model. The empirical analysis provides clear evidences that while the discriminatory power related to qualitative component (reputational scoring) respects the requirements of a good ex-ante identification of non reimbursement cases, the power of discrimination of quantitative scoring is only marginally superior to a random model. The added value of the study is completed by highlighting the importance of technical conditions of financing facility for anticipating events of non reimbursement. At the same time, the study tries to highlight the importance of scientific calibration of such a scoring system to perform effectively in the early detection of cases of default.
non-banking financial institutions credit risk scoring model univariate analysis multivariate analysis ROC curve discriminating power