How should we measure bank efficiency? A comparison of classic and recent techniques based on simulated data

In this paper, a cost function is used to generate the data of six samples of banks producing three outputs by means of two factors; unlike previous studies, the data-generation process used here is designed to reflect some structural characteristics of the banking sector (e.g., big producers are less frequent than small ones, the production levels of loans, deposits and services are highly correlated). A known amount of inefficiency and random noise is then added to each production plan. Finally we compare the "true" inefficiency levels to those estimated through the following techniques: stochastic frontiers, D.e.a., and several models of stochastic D.e.a. (two original models - multiplicative and heteroskedastic stochastic D.e.a. are also developed). All the "classic" techniques perform well. The stochastic D.e.a. models can outperform the "classics" in some specific situations, but on average they cannot compete with older techniques; however, the two new stochastic D.e.a. models perform better than the standard one.