Finanças e Regionalidade: um modelo de Credit Scoring com uso da Regressão Logística Geograficamente Ponderada no Programa Minha Casa Minha Vida em Minas Gerais

Abstract

In Brazil, due to the historical context and its continental dimensions, regional issues have always been present, and studies in finance characterized a space searched with regionality research regarding regional development. In this sense, the objective of this dissertation was to carry out a theoretical and an empirical study involving the theme of finance and regionality. For this, a systematic analysis of the literature was carried out through the Proknow-C process involving the themes of finance and regionality in order to list opportunities for new research, among which was based on the empirical study carried out in order to analyze whether the geographical weighting of borrowers makes it possible to build a better credit scoring model. For this, at first they were approached as the four phases of the Proknow-C process, in which it is observed in the articles of the Bibliographic Portfolio (BP) that regionality is treated at different levels. In Finance, he investigated a way in which BP studies are integrated or dissociate in (sub) tests, which provides opportunities for future research. From the articles of the BP, the majority (51%) deal with public finances, followed by the credit issues (15%). And in order to empirically address the regionality-finance link using the Geographical Weighted Logistic Regression (GWLR) methodology for modeling credit scores in a sample of the Minha Casa Minha Vida Program (PMCMV) in Minas Gerais (MG). Housing is intrinsically related to regionality and the state of Minas Gerais has internal differences that would justify the approach. The credit scoring model via GWLR presents differences in the parameters of the variables, demonstrating a certain variation by region, but it is statistically observed that the GWLR model does not present better accuracy than Logistic Regression, which in practice does not bring improvements in predictive capacity of model.