Two-tier stochastic frontier analysis: heterogeneous error distributions and model selection

We develop two-tier stochastic frontier models where the two one-sided error components follow different marginal distributions. As estimation methods, we present both Simulated Maximum Likelihood and Corrected Ordinary Least Squares coupled with Generalized Method of Moments. We propose a formal statistical test as well as informal diagnostics to choose between specifications. An application illustrates the use of these new tools.