Oliver Franz, Marcus Stronzik
Benchmarking-Ansätze zum Vergleich der Effizienz von Energieunternehmen
Nr. 262 / Februar 2005
Summary
Regulation in the electricity sector is mainly targeted at transmission and distribution as they represent monopolistic bottlenecks. Furthermore, no competitive alternatives do exist that might serve as efficiency benchmarks. The crucial issue concerning the regulatory process is the asymmetry of info rmation about production and cost structures between the regulator and grid companies. In this context benchmarking techniques provide for a promising tool in order to generate more objective info rmation about the efficiency performance of the individual network operators. With over 900 companies in Germany, there should be sufficient data available to make these tools viable.
This study tries to answer the question which method out of the pool of the various benchmarking approaches the regulator should concentrate on in the future. The analysis is centred around the Data Envelopment Analysis (DEA) – a linear programming technique – and the Stochastic Frontier Analysis (SFA) – a stochastic approach. Moreover, Price Index Numbers (PIN) as well as the more traditional regression analysis of Corrected Ordinary Least Squares (COLS) and Modified Ordinary Least Squares (MOLS) are tackled.
Out of the class of econometric approaches the two more traditional ones (COLS and MOLS) are systematically dominated by the Stochastic Frontier Analysis. Concerning the remaining three methods, namely PIN, SFA and DEA, no absolute statement can be derived. If it is intended to measure the productivity/efficiency on a company level, PIN can not be recommended because of the level of aggregation of the utilised data. Concerning the question of data availability, DEA as well as SFA can – at least partly – circumvent inconsistency problems by using data input only in physical terms instead of price info rmation. If the quality of data is poor, SFA seems to be superior to DEA since the former explicitly accounts for data noise whilst the latter does not. DEA results seem to be a bit more vulnerable to outliers. Furth ermore, the SFA results can be tested regarding their robustness using confidence intervals. The last argument is not an argument against the use of DEA anymore as this problem can be overcome, e.g. via bootstrapping. One clear advantage of DEA over SFA is that there is no necessity of ex ante assumptions concerning cost or production structures.
All in all, both approaches should be seen as complements and not as competitors. The outcome of DEA can be checked against SFA results and vice versa. The correlations between the results might give first hints regarding problems in the data set or systematic errors. Because of the monopolistic situation of grid companies it is recommended to use input oriented approaches. Finally, it should be stressed that the numbers resulting from DEA/SFA should not be used "mechanically" in the regulatory process, e.g. to determine X-factors under a price cap regime. Other company-specific factors also play a role in achieving certain future productivity goals and should therefore not be overlooked. [Only German language version available.]
Diskussion Paper is available for download.