Operational conditions in regulatory benchmarking models: A Monte Carlo analysis

Benchmarking methods are widely used in the regulation of firms in network industries working under heterogeneous exogenous environments. In this paper we compare three recently developed estimators, namely conditional DEA (Daraio and Simar, 2005, 2007b), latent class SFA (Orea and Kumbhakar, 2004; Greene, 2005), and the StoNEZD approach (Johnson and Kuosmanen, 2011) by means of Monte Carlo simulation focusing on their ability to identify production frontiers in the presence of environmental factors. Data generation replicates regulatory data from the energy sector in terms of sample size, sample dispersion and distribution, and correlations of variables. Although results show strengths of each of the three estimators in particular settings, latent class SFA perform best in nearly all simulations. Further, results indicate that the accuracy of the estimators is less sensitive against different distributions of environmental factors, their correlations with inputs, and their impact on the production process, but performance of all approaches deteriorates with increasing noise. For regulators this study provides orientation to adopt new benchmarking methods given industry characteristics.