Análise de classes latentes: classificação de perfis na concessão de crédito

Abstract

Credit granting analysis is a very studied and important branch of finance. This is because it no longer uses purely subjective criteria such as human judgment, but employs faster, more accurate and reliable statistical and computational techniques. As a primary and preliminary way to grant credit, a careful analysis of the type of client (good and bad payers is necessary. Therefore, the latent class analysis (LCA) emerges as a new proposal for the evaluation of clients’ profile subject to the granting of bank credit. In this context, this paper aims to present which variables determine the conjecture of the characteristics of good and bad payers, through LCA, as a way of helping creditor institutions in decision-making. In addition, it was also the objective of this paper to study techniques and procedures on credit risk analysis, as well as to know the database and analyze its characteristics in an exploratory manner. For that, descriptive statistics, simple and multiple logistic regressions, and stepwise regression were used, in order to pre-select the LCA variables, which were then used in the poLCA package of the R software. From the analysis of data, it was possible to perceive a distinction between two latent classes, even with only a few observed variables: the good and the bad payers. However, this work suggests future studies associated with other methodologies, such as cross-validation, discriminant analysis and cluster analysis.