Determinantes de inadimplência de financiamento imobiliário: uma aplicação da modelagem hierárquica

Abstract

This paper studied real estate financing contracts for private individuals under the Minha Casa Minha Vida program, with national coverage. Apparently, this study innovated, including as explanatory variables the value of financing divided by income and the variable of receiving salary at the bank account. Another innovation was the use of the multilevel technique, with individuals at the first level and bank branches at the second level, seeking to identify which factors determine the default of real estate financing operations and the influence of the individual's context (bank branch). The conducted tests in this study indicate the superiority of the multilevel logistic model over the traditional logistic model, including a study of the area under the ROC curve, which in the multilevel logistic showed greater predictive capacity compared to traditional logistic. The final multilevel logistical model identified that the variables “financing / income” and the level of education (individuals who have up to elementary school are more likely to default in relation to individuals with high school education) have a positive influence on default. Conversely, age, years of relationship with the bank, the fact of being an employee who receives proceeds and the level of higher education (incomplete, complete or post-graduate, master's and doctorate, with less probability of default in relation to individuals with high school) negatively influence default. The variables sex (male and female) and marital status (single and others) were not significant in the analyzed sample. The work also has a robustness test, analyzing the delay (characterized as a delay of more than 1 day) instead of the default characterized by a delay of more than 180 days, maintaining the results regarding both the superiority of the multilevel model over the traditional model, as for the significant variables (with the exception of the “superior” variable). Therefore, given the superiority found in the tests performed, it is recommended to apply hierarchical modeling in the credit analysis models.