Efficiencies and unobserved heterogeneity in Turkish banking: 1990-2000, unpublished working paper

Studies of bank efficiency tend to draw conclusions from pooled estimates, assuming that all banks in a sample use the same technology, or estimates based on a priori classifications of the banks. It is well known that efficiency rankings may be corrupted if banks that use different technologies are pooled together in estimating the technological frontier with respect to which inefficiency is estimated. In this paper we model unobserved heterogeneity in banking technologies as a mixture model, and investigate the efficiencies of 53 Turkish banks using likelihood-based stochastic frontier analysis for the period 1990-2000. We use the EC (Estimation-Classification) estimator to obtain data-driven identification of bank-technology-classes in our sample, and to estimate the mixture components of banking technologies therein. We have four dimensions of possible a priori heterogeneity in our sample: Large vs. small, state vs. private, foreign vs. domestic, and Islamic vs. conventional banks. In some cases, multiple a priori criteria may be confounded, e.g. most foreign banks were also small. Simple tests of homogeneity cannot disentangle those confounded effects. Our likelihood-based analysis finds no evidence of heterogeneity along the state vs. private and Islamic vs. conventional dimensions. The estimated classifications and mixture components have intuitive ex post institutional explanations