On the sensitivity of US electric utilities' efficiency estimates – a distance function approach

Abstract

Previous applications of different benchmarking techniques, both in academia and regulation practice, have shown substantial differences among the models' results. To analyse the sensitivity of efficiency estimates of a sample of US electricity distribution utilities, we compare the results of the generalized least squares frontier model proposed by Schmidt and Sickles (1984) and the maximum likelihood estimation frontier model of Pitt and Lee (1981) with the true random effects frontier model introduced by Greene (2004, 2005). We find substantially higher efficiency scores for the Greene model, indicating that the other formulations underestimate firms' efficiency due to an insufficient consideration of firm specific heterogeneity. In contrast to other studies, the efficiency estimates in this article do not differ considerably.

Acknowledgements

The authors would like thank Annika Frohloff and Christian von Hirschhausen for their helpful comments.

Notes

¹ For a discussion of modelling inefficiency as a time-variant stochastic term, see Farsi et al. (2006).

² For an excellent overview of parametric estimation techniques in efficiency measurement, the interested reader may be referred to Kumbhakar and Lovell (2000).

³ A detailed description of the construction of the data set can be found in Hess (2006).

⁴ For a few missing observations we approximated OPEX and the number of employees for a single year averaging over the variable values of the previous and the following year.

⁵ Note that in our distance functions setting, the first order coefficients satisfy the monotonicity requirements and can be interpreted as partial production elasticities with regard to output.

Table 2 Distance function parameters: panel data (1994–2005)