Environmental factors in frontier estimation - A Monte Carlo analysis

We compare three recently developed frontier estimators, namely the conditional DEA (Daraio and Simar, 2005; 2007b), the latent class SFA (Greene, 2005; Orea and Kumbhakar, 2004), and the StoNEZD approach (Johnson and Kuosmanen, 2011) by means of Monte Carlo simulation. We focus on their ability to identify production frontiers and efficiency rankings in the presence of environmental factors. Our simulations match features of real life datasets and cover a wide range of scenarios with variations in sample size, distribution of noise and inefficiency, as well as in distributions, intensity, and number of environmental variables. Our results provide insight in the finite sample properties of the estimators, while also identifying estimator-specific characteristics. Overall, the latent class approach is found to perform best, although in many cases StoNEZD shows a similar performance. Performance of cDEA is most often inferior.