Electric power distribution: economies of scale, mergers, and restructuring

Abstract

Electricity distribution is generally viewed as a natural monopoly and therefore as having the least potential for the kinds of reforms that have swept the electric power sector in many countries. Mergers among distribution companies and efforts at retail competition have nonetheless altered the operation of the distribution stage. This research into US electric utilities uses a much larger and less selective data base than previously available to examine the scale properties of distribution with respect to output, distance, and customer numbers, and for different functions within distribution. It finds significant economies at low output levels, holding system size and customer density constant, but the cost gradient is otherwise modest. It also finds that geographic size and customer numbers are quite important and that economies are significantly stronger for the infrastructure or ‘wires’ business than for the marketing function performed by distribution utilities. These results lend credence to efforts at retail competition that separates these functions, but cast doubt on the benefits of mergers between distribution systems.

Acknowledgements

Gratitude is expressed to the American Public Power Association for assistance in compiling the data used in this study, and to Charlene Kalenkoski and Kamen Madjarov for excellent research assistance.

Notes

¹ Transmission and distribution are commonly combined, both to conserve degrees of freedom and because of their rough similarity. The present study will separate the two.

² Discussion of various specifications of cost functions for the purpose of estimating efficiency can be found, inter alia, in Berndt (1991), Kumbhakar (1997), and Sengupta (1998).

³ It has often been argued that distribution is likely to be separable from generation and transmission, whereas generation and transmission are related. The reason for the latter is that size and siting decisions for generation capacity directly affect transmission costs. Local distribution, by contrast, is structured for end users and therefore would appear independent of upstream operations. See, for example, Burns and Weyman-Jones, pp. 42–43. Studies of economies of joint operation come to mixed conclusions. See Kaserman and Mayo (1991), Gilsdorf (1994), and Hayashi et al. (1997).

⁴ Indeed, since there is often a customer charge for local distribution, there is a factual basis for this view.

⁵ This is the equivalent of taking the total derivative and imposing proportionality as a side condition on the partials with respect to the two variables.

⁶ We later test for and correct remaining heteroskedasticity. While both heteroskedasticity and multicollinearity could be addressed in any form of this equation, we take advantage of this specification opportunity to reduce its effects.

⁷ The relevant part of the underlying total cost function can be written as follows:

Dividing by Q^D gives text Equation 7.

⁸ Although there have been a number of important structural changes in the industry since 1989, most of those have occurred in the generation and transmission stages. The economics of the distribution stage itself are driven by much the same forces as before.

⁹ Put in terms previously employed, distribution costs are [C(·) − C^{G, T}(·)]. See Equation 5.

¹⁰ The cost of common stock is measured by dividend yield, while the cost of preferred is preferred dividends divided by outstanding preferred stock. The cost of long-term debt is percent interest paid.

¹¹ While a utility-specific labor cost would be preferable, such data are missing for too large a number of utilities in this data base. Use of this manufacturing wage proxy has precedent in, e.g., Nelson and Primeaux (1988).

¹² The Cook-Weisberg test on the residuals of the OLS version of this regression results in a chi-square of 25.76, indicating remaining heteroskedasticity.

¹³ Both R² and especially R² are calculated differently and require careful interpretation in the case of heteroskedasticity-corrected estimates, since the standard errors have changed. R ² continues to be a valid measure of overall goodness of fit, however, and is therefore the superior measure.

¹⁴ Similar negative results have been reported by Mayo (1984) and Sing (1987). Fraquelli et al. (2004), however, find economies of scope for multi-utilities below median size.

¹⁵ There is starkest possible proof of this distribution: While unit cost is minimized at 82.1 million mwh, the actual value of unit cost is negative at that point. Greater mwh of output are being used by a constant number of customers, and the increased usage-really, beyond plausible values – drives unit costs into infeasible regions.

¹⁶ These estimates overstate economies in supply to the extent that all ‘general plant’ is attributed to supply. Previous discussion noted its multiple functions, which cannot be disentangled in these cost data directly, but we do include controls for those in the empirical estimation.

¹⁷ Use of the same model facilitates comparison across cost categories and aids in understanding what components are most responsible for the effects found in the overall cost regression.

¹⁸ This possibility that competition might raise unit costs rather than lowering them has often been found in advertising and marketing studies (e.g., Scherer and Ross 1990, p. 598).