Uma análise robusta de benchmarking utilizando o método de fronteira estocástica bayesiano aplicado às empresas brasileiras de distribuição de energia

Abstract

AData Envelopment Analysis (DEA) model is applied by the Brazilian regulator to set regulatory operational costs for 61 electricity distribution utilities or DSOs, since 2015. The current DEA model comprises non-decreasing returns to scale, one input, seven outputs and weight restrictions. Regulatory costs were estimated using average values from 2011 to 2013. In 2017, new regulatory costs were estimated using an updated data set and the previous DEA model. Recent results are similar to results achieved in 2015 and show evidence that the current benchmarking model still requires improvements. In short, some DSOs have inconsistent low efficiencies, close to 25%, and standard statistical analysis shows the presence of outliers in the data base. Furthermore, the model still lacks environmental adjustments. This study evaluates the use of Stochastic Frontier Analysis (SFA) as an alternative model to set regulatory operational costs. Pros and cons of the SFA model are highlighted. Results show that the SFA is more flexible to deal with outliers. However, the SFA has major convergence problems if applied to limited samples. Convergence issues can be overcome using Bayesian computation or penalized likelihood methods. In particular, a Bayesian SFA model is proposed that is robust to convergence problems. This study advocates the use of both DEA and SFA as the best alternatives to set regulatory operational costs for Brazilian electricity distribution companies, as indicated by European regulators.