Stochastic frontier models with flexible random coefficients

Abstract

We propose a stochastic frontier model with random coefficients having a flexible distribution. The distribution is modelled non-parametrically. It is shown that maximum likelihood estimation reduces to a fixed-point problem. A fixed-point iteration is proposed and we show that there is a unique regular fixed point. The fixed-point iteration is used in the context of MCMC to perform inferences for all unknown parameters including the optimal support of the distribution of random coefficients.