A Probabilistic Approach for Assessing the Significance of Contextual Variables in Nonparametric Frontier Models: An Application to Brazilian Banks

Abstract

This article presents an empirical banking application illustrating the use of a nonparametric frontier model relying on a probabilistic definition of the production frontier. It investigates the statistical significance of nonperforming loans in productive efficiency for a sample of Brazilian banks using the concepts of conditional and unconditional efficiency measures in a context where it is not necessary to impose any particular distribution on the production data. The analysis is robust relative to the assumptions of convexity and separability of the underlying technology. It is concluded that, on average, a relative increase of 1% in nonperforming loans implies a statistically significant relative reduction of 5.7% in the conditional measure of efficiency.