Bayesian and non-Bayesian analysis of gamma stochastic frontier models by Markov chain Monte Carlo methods (Q2488426)

A stochastic production frontier model is considered of the form \[ y_i=f(x_i,\beta)+u_i-z_i, \] where \(f\) is a known deterministic frontier function, \(y_i\) are dependent variables, \(x_i\) are vectors of explanatory variables, \(u_i\) are random ``shocks'' with Gaussian distribution and \(z_i\) are ``inefficiencies'' with gamma distribution, and \(u_i\) and \(z_i\) are independent. A MCMC technique is developed for Bayesian inference on the parameters of the model in which the auxiliary variable method is used for inefficiencies simulation and the acceptance-rejection Metropolis-Hastings algorithm is used for the shape parameter. It is proposed to use MCMC together with a stochastic approximation procedure for maximum likelihood inference. An application to economic data is considered.