A trivariate Gaussian copula stochastic frontier model with sample selection

Abstract

We propose a new stochastic frontier model with sample selection, in which the dependencies between the sample selection mechanism, the inefficiency term and the two-sided error in the production equation are modeled by a trivariate Gaussian copula. This model is compared to Greene's original stochastic frontier model with sample selection, and to an alternative model based on two bivariate copulas. The relative performances of the three models are analyzed using simulated data and cross-sectional data about Jasmine rice production in Thailand. We show that our trivariate Gaussian copula model has the best performance among all models, and that ignoring some correlations may cause estimation bias as well as over or underestimation of technical efficiency scores.