Cost Efficiency Analysis of Electricity Distribution Sector under Model Uncertainty

This paper discusses a Bayesian approach to analyzing cost efficiency of Distribution System Operators when model specification and variable selection are difficult to determine. Bayesian model selection and inference pooling techniques are adopted in a stochastic frontier analysis to mitigate the problem of model uncertainty. Adequacy of a given specification is judged by its posterior probability, which makes the benchmarking process not only more transparent but also much more objective. The proposed methodology is applied to one of Polish Distribution System Operators. We find that variable selection plays an important role and models, which are the best at describing the data, are rather parsimonious. They rely on just a few variables determining the observed cost. However, these models also show relatively high average efficiency scores among analyzed objects.