IT TAKES LABOR, FUEL, OPERATING CASH AND INVESTMENT capital to produce and deliver electric power. Which utilities have managed to use these resources optimally to produce and sell kilowatt-hours? How do these utilities compare with each other? Is there room for improvement?
And what about financial success? Does efficiency, as measured by a ratio of inputs to outputs, serve as a reliable predictor of market-to-book ratios or merger premiums?
Some of these questions are answerable; others not. Yet a simple observation of the range of utility expenses on the four basic inputs - fuel, capital, labor and O&M - can provide a window of which company we might choose to label as "most efficient." This method also allows a less-efficient utility to identify "peer" companies higher up on the ladder, to mark as examples to emulate.
Economists have wrestled with these questions for a long time. Several ways to provide an answer have been proposed and used, from the simple back-of-the-envelope method to complex multi-equation econometric models. The questions and the tools are becoming increasingly relevant in today's utility markets. To stay competitive in a restructured environment, utilities are searching for ways to understand productive efficiency better, to cut costs and to ensure survival in the 21st century.
Using historical data for 140 holding companies in the United States, we analyzed the relative efficiency of the top 100 using Data Envelopment Analysis (DEA), an approach for measurement of operational efficiencies and identification of "peers" to be used as best benchmarks.
Economic theory of productive efficiency is based on the comparative analysis of the best-in-class producers vis-à-vis all others. The criterion for determining the "best" producers refers to the ability to produce maximum output given a specific level of input, or conversely, the ability to use the least amount of input to produce a specific level of output. DEA is a linear programming technique first introduced in the early 1980s by Charnes, Cooper and Rhodes. It has since been used in various applications ranging from healthcare to banking to retail. Fortune magazine stated that DEA is a tool every manager must have if his business is to remain competitive. We used static and dynamic DEA methods to measure annual operational efficiencies of holding companies as well as their respective improvements over time.
Striving for Efficiency
Increases in productivity may prove the key to competitive advantage of any economic enterprise. Yet few take the necessary steps to actually measure it. The measurement of productivity by economists, for the most part, is based on comparisons between inputs and outputs. The complexity ranges from Robert Solow's econometric production functions to the Jorgenson Divisia index to simple ratios of output to input (for example, MWh per employee).
Productive efficiency can be measured in terms of input-conserving or output-increasing orientation. Choice of orientation in most cases won't impact efficiency ratings significantly and will identify the same efficient utility companies. Since holding companies are more likely to have control over input usage than over demand for output, we chose to use an input-oriented analysis. In this study, we examine each holding company as a productive unit that converts inputs to outputs. We refer to each such entity as a Decision-Making-Unit (DMU).
Traditionally, research on technical efficiency has relied on one of two approaches: 1) a parametric approach using econometric tools or, 2) a non-parametric approach using linear programming techniques, such as DEA. Econometric methods involve estimating a production function based, on average, on how various inputs are used by a group of similar producers. These techniques require that certain statistical assumptions be satisfied (e.g., that there should exist no significant relationship among various independent variables or inputs) and some knowledge of the functional form. On the other hand, DEA, being nonparametric, requires no such assumptions. DEA also optimizes each company individually (by benchmarking it against its closest peers), whereas traditional statistical methods rely on averages.
The DEA Method
Using historical production data, DEA measures how efficiently a producing unit converts inputs to output. DEA uses mathematical optimization to construct a piecewise convex production frontier based on the most efficient companies. Companies that form the production frontier are considered efficient and receive a score of 1; all other companies receive an efficiency score between 0 and 1 based on distance from the production frontier.
Figure 1 is a graphical presentation of a simple one-input, one-output DEA production frontier. DMUs A, B, and C form the efficient production frontier (most efficient). Given their input levels, they are able to produce more output relative to any other DMU. All three receive an efficiency score of 1.0. D, E, and F are less efficient (fall below the efficient frontier). D, E, and F could move closer to the efficient frontier by using less input for the current level of output or increase their output given the existing inputs by using, for example, better technology. These DMUs thus can move from their current positions (D, E, and F) to the closest efficient position (D*, E*, and F*). Based on similarities in the input and output mix, DEA identifies efficient peers for each of the inefficient units. For example, unit D may end up with unit A as the peer against which it is compared and its efficiency score (the horizontal distance to the production frontier, DD*) is computed. Similarly, unit E's peers maybe utilities A and B, and F's may be B and C.
To assess changes in technical efficiency over time, we use the Malmquist productivity index. Overall change in productivity consists of not only the change in efficiency, but also change in technology. The advantage of the Malmquist productivity index is that it is comprised of these two distinct elements. For ease of interpretation, we use the natural log of the Malmquist index, thereby reporting change in productivity as a percent increase or decrease.
For each inefficient company, it is possible to calculate individual target values for labor, capital, operation and maintenance and fuel. The target values represent realistic goals for operating at peak efficiency with respect to identified peers. These are the changes necessary to move the company to an optimal position on the efficient production frontier. As written earlier, the production frontier, or "best" practice, is based on the observed performance of other utilities. Therefore, optimal performance in terms of allocation of inputs and resources is also measured in relative terms. Targets and goals set in this manner are, therefore, realistic and obtainable. In this article we present the target values results aggregated across all the utilities used in the study. We show, on average, how the inefficient utilities have "misallocated" their resources with respect to the various inputs.
Just the Facts
Data were obtained from POWERdat (c)1998 Version 2.01, a Resource Data International Inc. database. Original data sources included the Federal Energy Regulatory Commission Form 1 and the U.S. Securities and Exchange Commission 10-K and 10-Q reports for holding companies and utility operations. The data set included 140 holding companies from 1990 to 1996.
Output was defined as total physical production in megawatt-hours produced and sold to all sectors (Schedule 14). Purchased power was removed from total MWh sales. Input variables consisted of labor cost, O&M expenses (excluding depreciation), pensions and benefits, total outlays for all fuels (Schedule 14), and capital (book value of total electric plant, including production, transmission and distribution). All data were converted to 1996 dollars using the producer price index.
Table 1 lists the top 100 utilities in terms of achieved efficiency in 1996. Nineteen utilities were classified as efficient (efficiency score = 1). The inefficient utilities received a score between 0 and 1 indicating the proportionate amount of inputs they should be using. That is, an efficiency score of 0.8 would indicate that the utility is underutilizing its input resources by about 20 percent. Table 1 also lists the DEA-selected peers identified by "Holding Company Code." The efficient utilities will not have a peer since there are no other utilities that can produce as much output using less input. For the inefficient utilities, we provided the peers to which they were compared; the companies that produced proportionally the same output using less input. The peer utilities are selected based on the same mix of inputs. It is understood that the DEA-selected peers may differ with respect to production conditions such as fuel mix, geography or customer base. Identifying peers based on these factors would require a case-by-case analysis of all the utilities in the study.
The list of top performers includes a wide mix of utilities in terms of size (from North Central Power to Southern Co.) and geographic location. Table 1 shows the specific conditions of scale economies under which we believe the utility is operating (i.e., constant, decreasing, or variable returns to scale - CRS, DRS or VRS, respectively).
Further ranking of the 19 efficient utilities is possible through DEA. The analysis creates a ranking of the DMUs on the efficient frontier based on the number of instances that they have been designated in DEA as peers. From an analytical point of view, this increases the confidence in the assessment concerning the operational efficiencies of these utilities. For example, Southern Co. received an efficiency score of 1, but was never used as a peer This indicates that Southern, although efficient, was not influential in determining the efficiency of the other companies (i.e., no companies matched the criteria for comparison with Southern Co.). Furthermore, this indicates that Southern was found to be efficient, at least partially, due to its uniqueness. Idaho Power, on the hand, also received an efficiency score of 1 and was used as a peer for 55 different utilities. (Idaho Power received an efficiency score of 1.0 even when placed with 55 similar utilities, a more convincing accomplishment.)
Table 2 shows the gains/losses in productivity for the top 100 holding companies utilities for each year between 1990 and 1996, as well as an overall estimate spanning the entire time period. It is interesting to note that over the 1995-96 period, the top performing companies witnessed an average of 7-percent efficiency gain, while the remaining utilities witnessed a 4-percent efficiency loss.
There appear to be outstanding achievers among the utilities in the study. For example, the top five performers in terms of efficiency improvement over the 1995-96 period (Upper Peninsula, Ameren Corp., Northwestern Wisconsin Electric, American Electric Power Co., and North Central Power ) averaged more than a 54-percent gain. Four of these companies also ranked among the most efficient in 1996. Upper Peninsula Energy Corp., at greater than 133-percent efficiency improvement in 1995-96, is especially interesting. Upon detailed data examination, its generation fuel mix showed a drop in steam from nearly $2 million to zero in 1996 without a significant change in the net generated MWh.
On the whole, the results suggest that the less efficient utilities in the study (all except the top 19) underutilized their inputs by nearly 40 percent, on average. Target values were computed to investigate discrepancies in underutilization of individual inputs. Figure 2 illustrates that, on average, for the less efficient utilities, nearly 54 percent of their capital (electric plant) is underused, representing large over-capacity; O&M is a close second at nearly 53 percent. Fuel underutilization is at the overall average for all input (40 percent).
For electric utilities, the need for efficiency improvement is likely to manifest itself both in regulated and unregulated branches of their business. The regulated segments are likely to be subject to some form of incentive ratemaking, which is most likely decided primarily based on some measure of efficiency. In the unregulated segments, efficiency is likely to be the most significant determinant of price and thus the ability to compete effectively.
Despite what one may think, the industry as a whole is not showing very strong signs of improved efficiency. Only 16 percent of the utilities studied showed positive trends in efficiency over the 1990-96 period. Among the less efficient utilities, there seems to be enormous opportunities for improvement. Based on our results, the overall underutilization in this segment of the industry for all resources in 1996 was nearly 40 percent.
We believe that efficiency measurement tools are needed and are going to be critical in the coming decade. We believe that the measurement process itself leads to improvements. As Kenneth Galbraith once said, "Things that are measured tend to improve."
Janice Forrester is a Ph.D. student in system science at Portland State University and is the director of strategic and analytical services at Cytera Systems Inc. M. Sami Khawaja, Ph.D., is president of Quantec L.L.C., an energy economics consulting firm. Hossein Haeri, Ph.D., is director of energy information services at PG&E Energy Services. Michael Carter is an analyst with Financial Times Energy's Energy Insight.
Mathematical Appendix
Following is a brief summary of the mathematical programming model used. Readers interested in more detail may contact the authors at samik@quantecllc.com.
First stage Second stage
Min ui Max 1s + 1e
S.t. S.t.
1) Yl # y1 1) Yl - s1 = y1
2) Xl - ux1 # 0 2) Xl + e1 = u1x1
3) Sili = 1
Where Y represents outputs, X represents inputs, and u represents the inverse of the proportionate reduction in inputs required to become efficient, (if u = 0.8, then a 20-percent reduction of all inputs is required for technical efficiency). The second stage equations are used to determine the target values where the s and e represent additional reduction in individual inputs in order not only to seek the shortest path to the efficient frontier but to find the optimal position. Efficiency scores for all utilities are first computed under each of different assumptions regarding returns to scales, i.e., constant (CRS), decreasing (DRS) and variable (VRS). Three technical efficiency scores are calculated: (1) CRS using the above equations as is (no restrictions on the ls); (2) VRS using the above equations with the added the restriction Sil = 1; and (3) DRS using the above equations with added the restriction Sil < 1. Since most utilities appeared to be operating under DRS conditions, decreasing returns efficiency scores were used.
Malmquist productivity index (M) can be decomposed into two distinct pieces M=E*P:
(E) D efficiency = Efficiencyt,t/Efficiencyt+1,t+1
(P) D technology = [(Efficiencyt,t+1/Efficiencyt+1,t+1)*( Efficiencyt,t/Efficiencyt+1,t)]0.5
Therefore
Mit = [(Efficiencyt,t+1/Efficiencyt,t)*( Efficiencyt+1,t+1/Efficiencyt+1,t)]0.5
Where
Efficiencyt,t = input technical efficiency under CRS for a given utility in year t with respect to all other
utilities at year t.
Efficiencyt+1,t+1 = input technical efficiency under CRS for a given utility in year t+1 with respect to all
other utilities at year t+1.
Efficiencyt+1,t = input technical efficiency under CRS for a given utility in year t+1 with respect to all other
utilities at year t.
Efficiencyt,t+1 = input technical efficiency under CRS for a given utility in year t with respect to all other
utilities at year t+1.
Note: All systems of linear equations are solved using the commercial solver XA from Sunset Technologies Inc. that employs the Primal and Dual Simplex solving method.
1 We would have preferred to conduct the analysis at the utility level, and we set out to do so. However, as we conducted the analysis, it became apparent that the data, while accurate at the holding company level, were not very useful at the utility level.
2 For a report on the application of the econometric technique used by the authors, see, H. Haeri, M. S. Khawaja, and M. Perussi "Competitive Efficiency: A Ranking of U.S. Electric Utilities," Public Utilities Fortnightly, June 15, 1997,
p. 26.
3 For a study using a simplified DEA approach, see, D. Thomas Taylor and Russel G. Thompson, "The Efficient Utility: Labor, Capital and Profit," Sept. 1, 1995, Public Utilities Fortnightly, p. 25. Certain weaknesses in that study were later noted by Matthew Morey and L. Dean Hiebert in the Jan. 1, 1996 issue,
p. 8.
4 With output being defined as net MWh and inputs including those associated with power trading, we may have introduced some bias against companies with large amounts of purchased power. However, we believe that the amount of resources dedicated to power trading is insignificant compared with those dedicated to power generation.
5 It is important to note that the issue of returns to scale and efficiency ranking are not directly related. In other words, a company may rank high on the efficiency scale and still operate under DRS conditions. In the case of this study, more than 60 percent of the utilities were found operating at DRS, with the remaining operating under CRS. CRS simply indicates that doubling your inputs doubles your output. DRS on the other hand, means that doubling your inputs leads to less than doubling of your output. Allowing efficiency scores to be determined under various return scale conditions allows for consideration to be given for relative size. Consider three utilities A, B, and C. Utility A uses 2 units of input and produces 2 units of output; utility B uses 40 units of input and produces 30 units of output; utility C uses 35 units of input and produces 30 units of output. Assuming CRS conditions, only A would be considered efficient. Under CRS, B should be able to produce 40 units of output given the use of 40 units of input; similarly C should be able to produce 35 units of output. Now, considering DRS, Utility C would be considered efficient as well as A since no other utility can produce as much output as C using any less input. In short, it does not allow for utilities such as C to be directly compared to a virtual utility A* (utility A scaled 20 times larger).
6 Furthermore, based on another analysis conducted by the authors (Fortnightly, June 15, 1997), Upper Peninsula ranked among the least efficient utilities. That analysis, however, covered 1990-95, which did not include the period during which Upper Peninsula showed the largest efficiency gains, and used a different modeling approach.
7 DEA computes overall underutilization, or the closest distance to the efficiency frontier. This measure provides an estimate of the overall proportional reduction in all inputs for the DMU to become efficient (produce the same level of output using less input). In the case of the companies in this study, the overall proportional reduction in all inputs required to make inefficient companies efficient, is nearly 40 percent. The fuel target value is also 40 percent. This indicates that, while fuel is underutilized, other inputs (e.g., capital), have a higher potential for improvement (beyond the overall proportionate reduction).
Articles found on this page are available to Internet subscribers only. For more information about obtaining a username and password, please call our Customer Service Department at 1-800-368-5001.
