An empirical likelihood estimator of stochastic frontier model

Abstract

We consider a stochastic frontier model, with independent observation errors identically distributed with an unknown probability density function. Instead of maximizing the parametric version of the likelihood function, which requires knowledge about the error distribution, we replace the parametric likelihood with an empirical likelihood. A simulation and experiment study are presented to illustrate the finite-sample of this estimator in terms of its accuracy and robustness. Our proposed estimation is competitive and allows for better analysis of datasets than existing parametric methods.