In this paper, we verify which qualitative banking attributes can determine the level of American state-chartered Financial Institutions (FIs) and evaluate its underlying variables. The methodology followed three procedures of analysis. First, we measured banking efficiency using a two-stage SBM network data envelopment analysis (NDEA). Subsequently, we used machine learning methods to predict efficient FIs from qualitative attributes. Finally, we tested the variables related to the attributes, using a fractionated logistic regression controlled by economic-financial variables. As main results, we found that attributes linked to political-administrative localization criteria were the more important attribute in predicting if the FI was in the efficient group; we confirmed the recent findings of the literature that state that less governmental influence (freedom) is related to more efficient institutions. Besides that, we found that a population with a higher financial education have FIs with higher levels of efficiency.
de Abreu Emmanuel Sousa, Kimura Herbert
Banking efficiency, Fractional logistic regression, Machine learning, SBM DEA network, State-chartered financial institutions