Setting network tariffs with heterogeneous firms: The case of natural gas distribution

Abstract

The appropriate treatment of firm heterogeneity plays a crucial role in the application of benchmarking analyses for regulatory purposes. Within the realm of two-step approaches, this paper challenges the widespread adoption of single-variable clustering: heterogeneity has often multiple sources, which calls for more sophisticated clustering methodologies. In fact, reliable cluster-specific rankings provide firms’ management with more realistic objectives as well as freedom to identify the appropriate strategies to improve efficiency. In order to provide regulatory guidance on this issue, we use a unique dataset of detailed accounting data and unbundled network-related costs for a panel of Italian gas distributors and we test two alternative methods: a hybrid clustering procedure (HCP) and a latent class model (LCM). Our results show that HCP and LCM perform better than size segmentation in the identification of classes, thereby leading to more reliable production frontiers, but do not support a conclusive preference for one or the other method. While both methods are sensitive to outliers, LCMs seem to provide deeper insights on the drivers of firm inefficiency. However, they also present stationarity and convergence issues, which might favour the implementation of HCP methods. Furthermore, the degree of discretionary judgement in the modelling decisions (e.g., model specification and choice of the partition) is slightly higher with LCMs than with HCP. In this respect, the HCP, with its lower modelling and analytical complexity, may feature as a more appealing option, facilitating the interactions between regulator and firm managers.