Mehdi Farsi, Massimo Filippini
Regulation and measuring cost efficiency
with panel data models : application to
electricity distribution utilities
Quaderno N. 03-05
Decanato della Facoltà di Scienze economiche
Via G. Buffi, 13 CH-6900 Lugano
REGULATION AND MEASURING COST EFFICIENCY
WITH PANEL DATA MODELS:
APPLICATION TO ELECTRICITY DISTRIBUTION
UTILITIES
Mehdi Farsi Massimo Filippini
Center for Energy Policy and Economics
Federal Institute of Technology
ETH Zentrum, WEC, 8092 Zurich, Switzerland
and
Department of Economics
University of Lugano
Via Ospedale 13, 6900 Lugano, Switzerland
January 2003
2
ABSTRACT
This paper examines the application of different parametric methods to measure cost
efficiency of electricity distribution utilities. The cost frontier model is estimated using four
methods: Displaced Ordinary Least Squares, Fixed Effects, Random Effects and Maximum
Likelihood Estimation. These methods are applied to a sample of 59 distribution utilities in
Switzerland. The data consist of an unbalanced panel over a nine-year period from 1988 to
1996. Different specifications are compared with regards to the estimation of cost frontier
characteristics and inefficiency scores. The results point to some advantages for the FE model
in the estimation of cost function’s characteristics. The mutual consistency of different
methods with regard to efficiency measures is analyzed. The results are mixed. The summary
statistics of inefficiency estimates are not sensitive to the specification. However, the ranking
changes significantly from one model to another. In particular, the least and most efficient
companies are not identical across different methods. These results suggest that a valid
benchmarking analysis should be applied with special care. It is recommended that the
regulator use several specifications and perform a (mutual) consistency analysis. Finally, the
out-of-sample prediction errors of different models are analyzed. The results suggest that
benchmarking methods can be used as a control instrument in order to narrow the information
gap between the regulator and regulated companies.
3
1. INTRODUCTION
Transmission and distribution of electricity have been considered as natural monopolies,
thus less affected by the recent waves of deregulation in power industry. However, as
competition is introduced into generation sector, regulatory reform and incentive regulation of
distribution utilities have become more common. In traditional cost-of-service regulation
systems companies recover their costs with a risk-free fixed rate of return and therefore have
little incentive to minimize costs. The incentive-based schemes on the other hand, are
designed to provide incentive for cost-efficiency by compensating the company with its
savings. A variety of methods are proposed in the literature. Main categories of incentive-
based schemes used for electricity utilities are: price or revenue cap regulation schemes,
sliding-scale rate of return, partial cost adjustment, menu of contracts, and yardstick
regulation.
1
Jamasb and Pollitt (2001) provide an extensive survey of different regulation
practices in electricity markets around the world. Virtually all the models used in practice, are
based on benchmarking that is, measuring a company’s efficiency against a reference
performance. Inefficiency is a deviation from the optimal point on the production or cost
frontier. This deviation can be resulted from two sources: technological deficiencies and
problems due to a non-optimal allocation of resources into production. Both technical and
allocative inefficiencies are included in cost-inefficiency, which is by definition, the deviation
from minimum costs to produce a given level of output with given input prices. In
benchmarking applications the regulator is generally interested in obtaining a measure of
firms’ inefficiency in order to reward (or punish) companies accordingly. The reliability of
inefficiency scores is therefore crucial for the regulator. In particular, if the estimated
inefficiency scores are sensitive to the benchmarking method, a more detailed analysis to
justify the adopted model is required. However, in most cases it is difficult to identify the
“right” model and the regulator should not use the results in a mechanical way. Rather, the
benchmarking analysis can be used as a complementary instrument in the regulation.
There are a wide variety of methods to measure cost-efficiency. These measures range
from basic indicators to more complex measures obtained from a multivariate analysis.
Simple measures such as average unit cost or average labor productivity are commonly used
in practice but fail to account for the differences among conditions and opportunities that
different companies face. Multivariate analyses however adjust the measures with respect to
factors that are beyond companies’ control. These methods can be classified into two main
1
See Jamasb and Pollitt (2000) and Joskow and Schmalensee (1986) for reviews of regulation models.
4
categories: non-parametric or deterministic methods such as data envelopment analysis, and
parametric or stochastic methods such as least squares method and stochastic frontier analysis.
Rossi and Ruzzier (2000) provide a comparative discussion of different approaches used in
cross-sectional and panel data.
One of the main advantages of parametric methods is their ability to control for
unobserved heterogeneity among companies. In particular, panel data models give a better
possibility to control for such heterogeneities. This turns out to be an important issue in
network industries like electricity distribution sector, where different companies deal with
networks with different shapes and consumer densities and various topographical conditions.
These factors as well as other potentially unobserved characteristics do affect the production
costs but are not necessarily indicative of different efficiencies. The inefficiency measures
may therefore be affected by these confounding factors. In this case companies that face more
difficult conditions may be classified as inefficient producers.
The theoretical development of stochastic frontier models in panel data has been subject
of a great body of literature.
2
Many studies compared the inefficiency scores obtained from
different models. After reviewing their previous studies, Kumbhakar and Lovell (2000)
3
conclude that different approaches are likely to generate rather similar efficiency rankings,
especially at the top and bottom of the distribution. However, using Monte Carlo simulations,
Gong and Sickles (1989) find that with complex production functions all models show a poor
performance. Their results suggest that the reliability of different models depends on the
nature of production. We argue that in industries such as electricity distribution, the
production technology is a rather complex function that depends on a variety of external
parameters associated with the production environment and demand characteristics.
In practice, regulators have used virtually all measures including simple univariate
indicators. The effect of unobserved differences among companies is often ignored. This issue
may however be crucial in network industries. The main goal of this paper is to study how
and to what extent the unobserved differences among companies affect the inefficiency
measures. Focusing on parametric methods and using different panel data models, it is shown
that the inefficiency scores and rankings are quite sensitive to whether and how the
heterogeneity of production conditions is accounted for. Moreover, there is no unique model
with a decisive advantage over all other methods.
2
See Kumbhakar and Lovell (2000) for a review and Greene (2002) and Tsionas (2002) for some recent
developments.
3
See page 107.
