Spatial statistical methods applied to the 2015 Brazilian energy distribution benchmarking model: Accounting for unobserved determinants of inefficiencies

- Statistical analysis between DEA efficiency scores and environmental variables is proposed.
- Spatial statistic methods are applied to the first stage efficiency scores.
- Second stage model accounting for a spatial latent structure is presented.
- Results show major differences between original and corrected efficiency scores.

In 2015 the Brazilian regulator presented a DEA benchmarking model to set the regulatory operational cost goals, to be reached in four years for 61 electricity distribution utilities. The DEA model uses: adjusted operational cost as the input variable, seven output variables and weight restrictions. Although non-discretionary variables or environmental variables are available in the dataset, the regulator argued that no statistically significant correlation was found between the DEA efficiency scores and the non-discretionary variables. This study evaluates the statistical correlation between the DEA efficiency scores and the available environmental variables. Spatial statistic methods are used to show that the efficiency scores are geographically correlated. Furthermore, due to Brazil's environmental diversity and large territory it is unlikely that only one environmental component is sufficient to adjust inefficiencies across the Brazilian territory. Thus, a new combined environmental variable is proposed. Finally, a second stage model using the proposed environmental variable and accounting for a spatial latent structure is presented. Results show major differences between original and corrected efficiency scores, mainly for utilities located in harsh environments and which originally achieved lower efficiency scores.