Robustness, outliers and Mavericks in network regulation

Benchmarking methods, primarily non-parametric techniques such as Data Envelopment Analysis, have become well-established and informative tools for economic regulation, in particular in energy infrastructure regulation. The axiomatic features of the non-parametric methods correspond closely to the procedural and economic criteria for good practice network regulation. However, critique has been voiced against the robustness of best-practice regulation in presence of uncertainty regarding model specification, data definition and collection. This paper investigates the foundation of the critique both conceptually and by describing the actual state-of-the-art used in energy network regulation using frontier analysis models in Sweden (2000-2003) and in Germany (2007-). A principal component of the applied frontier regulation is the systematic use of outlier detection models to define homogeneous reference sets and to exclude maverick reports. We review two families of outlier detection methods in terms of their function and application using a data set from Swedish electricity distribution, illustrating the different types of outliers. Finally, the paper concludes on the role of outlier detection as a mean to implement regulation with higher robustness.