STATISTICAL RECOUPLING - BREAKING THE LINK BETWEEN ELECTRIC UTILITY SALES AND REVENUES

In 1991, US electric utilities spent $1.8 billion on demand-side manag ement (DSM) programs. However, utility DSM efforts vary enormously acr oss the country, concentrated in only a few states. This concentration is partly a function of regulatory reforms that remove disincentives to utility shareholders for investments in DSM programs. A key compone nt of these reforms is recovery of the net lost revenues caused by uti lity DSM programs. These lost revenues occur between rate cases when a utility encourages its customers to improve energy efficiency and cut demand. The reduction in sales means that the utility has less revenu e to cover its fixed costs. This article describes a new method, stati stical recoupling (SR), that addresses this net-lost-revenue problem. Like other decoupling approaches, SR breaks the link between revenues and sales. Unlike other approaches, SR minimizes changes from traditio nal regulation. In particular the revenue risks associated with year-t o-year changes in weather and the economy remain with the utility unde r SR. Statistical recoupling uses statistical models that explain reta il electricity sales as functions of the number of utility customers, winter and summer weather the condition of the local economy, electric ity price, and perhaps a few other key variables. These models, along with the actual values of the explanatory variables, are used to estim ate ''allowed'' electricity sales and revenues in future years. For ex ample, a utility might use quarterly data from 1980 through 1992 to es timate the SR models. The models would then be used to determine allow ed revenues for 1993, 1994, and 1995. Five utilities provided data to use in testing this new approach. The empirical results are quite prom ising. The annual errors are almost all less than 2%. And the 3-year a verages for these utilities are less than 1.3%. The lack of patterns a cross these 3 years and five utilities suggests that statistical recou pling is a robust method.