The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
http://clevelandfed.org/research/workpaper Best available copy FEDERAL RESERVE BANK. CLEVEMND 90005625 Working Paper 8713 A TEST OF TWO VIEWS OF THE REGULATORY MECHANISM: AVERCH-JOHNSON AND JOSKOW by Philip Israilevich and Kim J. Kowalewski Philip Israilevich is an economist at the Federal Reserve Bank of Chicago. Kim J. Kowalewski is an economist at the Federal Reserve Bank of Cleveland. Working papers of the Federal Reserve Bank of Cleveland are preliminary materials circulated to stimulate discussion and critical comment. The views stated herein are those of the authors and not necessarily those of the Federal Reserve Bank of Cleveland, the Federal Reserve Bank of Chicago, or the Board of Governors of the Federal Reserve System. December 1987 http://clevelandfed.org/research/workpaper Best available copy A TEST OF TWO VIEWS OF THE REGULATORY MECHANISM: AVERCH-JOHNSON AND JOSKOW I. Introduction The impact of regulation on the production decisions of electric utilities was first described by Averch and Johnson (1962). They argued that rate-of- return regulation gives utilities the incentive to overcapitalize, that is, to employ a capital-labor ratio that is larger than one that minimizes costs for a given output level.' Courville (1974), Spann (1974), Petersen (1975), and Cowing (1978), for example, find evidence of an overcapitalization bias using variations of the Averch and Johnson (A- J) model. The major challenge to the A-J model concerns the nature of the regulatory environment. Implicit in the A-J model is a regulator that constantly monitors capital returns and adjusts electricity prices to keep capital returns equal to their "fair" levels. Joskow (1974) argues that regulators are more concerned with nominal electricity prices than with the rate of return on capital. As long as nominal electricity prices are not increasing, regulators will not actively enforce the rate-of-return constraint, thereby eliminating the source of the A-J bias. As evidence in favor of his view, Joskow finds a positive relationship between changes in the average cost of electricity production and the frequency of rate hearings initiated by utilities. He also argues that the implementation of fuel-cost-adjustment clauses and environmental regulations in the 1970s reflects his more general view of regulators as political entities rather than as Averch and Johnson's strict rate-of-return enforcers. The total impact of these and other constraints on electric utility production decisions was examined by Atkinson and Halvorsen (1984). They http://clevelandfed.org/research/workpaper Best available copy - 2 - developed a generalized cost model that includes the impact of additional regulatory constraints and found empirical evidence of these impacts in a cross-section sample of electric utilities. However, they did not include Joskow's view of the regulatory process in their model. The purpose of this paper is to test Joskow's view of the regulatory mechanism by estimating a modified version of the Atkinson and Halvorsen model. The modifications are of two sorts. The first allows for different regulatory impacts over time as argued'byJoskow. The second permits the use of panel data and the estimation of total factor productivity (TFP) and its returns-to-scale and technical-change components. Joskow argues that when the A-J bias occurs, utilities have less incentive not only to employ an efficient team of production inputs, but also to innovate or to maintain a high rate of technical change. Nelson and Wohar (1983) attempted to examine the impact of regulation on utility technical change, but they could not estimate a direct regulatory impact on technical change. Our procedure yields such an estimate. Our data are a panel sample of the seven major electric utilities in Ohio over the period 1965 to 1982.3 The advantage of this sample is that the technologies employed by these utilities should be fairly similar; these Ohio utilities are all privately owned, coal-burning plants and are subject to the same regulator. Thus, the estimation of a common cost structure for these utilities should yield a smaller potential for specification bias than is true of previous studies of electric utilities, whose samples include utilities that employ varying technologies or face different regulators. Our results square with Joskow's view. We find considerable circumstantial evidence in Ohio consistent with Joskow's more general regulatory mechanism. Our estimation results show that these utilities produce electricity less efficiently during the years when Joskow expects regulatory constraints to be more binding, and that regulation significantly retards the http://clevelandfed.org/research/workpaper Best available copy rate of technical change implemented by these utilities. Thus, the emphasis that regulators and economists place on efficient production using a given capital stock appears to be misplaced; the retardation of the rate of technical change implemented by these utilities appears to be an important source of bias. However, contrary to Joskow's view, we find that regulation retards the technical change implemented by these utilities to a lesser extent during the years when regulatory constraints are more tightly binding. The next section of this paper contrasts the Averch-Johnson and Joskow views of the regulatory mechanism. After that, the rate hearing experience in Ohio over the 1965 to 1982 period is discussed and is found to correspond quite well with Joskow's view of the regulatory mechanism. The fourth part presents the model and outlines the testing procedures; the fifth section describes the empirical results. The final section provides summary and concluding remarks. 11. Averch-Johnson and Joskow Views of the Regulatory Process It is useful to view the regulatory process in two parts: 1) the mechanics of setting a utility's electricity price structure, and 2) the events that initiate a rate hearing or a review of a utility's electricity price structure. There is little disagreement among economists about the first part. What brings a utility to a rate hearing and what motivates a regulator are open questions in the empirical literature. The predominant answers to these questions were influenced by Averch and Johnson. They investigated the optimal response of a cost-minimizing utility in static equilibrium t0.a "fair" rate of return on capital regulatory constraint. They showed that when the rate of return on capital constraint is binding, and when http://clevelandfed.org/research/workpaper Best available copy the "fair" rate of return is larger than the cost of capital, a utility has the incentive to overcapitalize, that is, to employ a capital-labor ratio that is larger than one that minimizes costs for the chosen output level.4 Implicit in the A-J model are two assumptions about the behavior of the regulator. One is that the motivating factor behind regulatory action is the rate of return on capital; in the A-J model, the constraint on a utility's profit-maximization actions is that the actual rate of return on capital earned by a utility is no greater than the "fair" rate. The second i-sthat an active regulator continually monitors utility returns and pounds on a utility with a "visible hand" to maintain the equality of a utility's profits with its "fair" profits. This follows from Averch and Johnson's assumption of static equilibrium. When a utility's profit is less than its "fair" level of profits, the regulator calls a rate hearing to raise the "fair" return and, hence, the utility's price of electricity. When a utility's profits are above the "fair" level, the regulator calls a rate hearing to lower its "fair" return and the price of electricity. With minor amendments, this view of regulatory behavior predominates in the economics literature, especially in empirical studies of electric utility behavior, with the exception of Joskow (1974).= Joskow agrees that rate- of-return regulation will give a utility the incentive to employ an inefficient mix of input factors, but he argues that the A-J bias may not always occur in a dynamic world. In Joskow's view, regulators are political institutions whose objective is to minimize "conflict and criticism," not to keep the rate of return on capital equal to the "fair" rate. One important source of conflict and criticism is an increase in the nominal price of electricity. Consumers will agitate against increases in electricity prices because they typically view these increases as pricegouging. If electricity prices are not increasing, and especially if they are http://clevelandfed.org/research/workpaper Best available copy - 5 - falling, consumers are indifferent to the profits earned by a utility. Thus, Joskow argues that utilities that are able to adjust their production and investment decisions to raise their earned rates of return without raising electricity prices will not be thwarted by the regulator. In this case, there may be little A-J bias. On the other hand, Joskow argues that regulators do not initiate any actions to raise the rate of return on a utility's capital when it is below the "fair" rate unless requested to do so by the utility. Before a rate increase is granted, the utility will earn a return on capital below the "fair" return. In this case, an A-J bias may appear. Thus, in contrast to the active A-J regulator, the Joskow regulator is passive, adjusting the rate of return on a utility's capital only when requested to do so by a utility or by a consumer advocate. Earned profits may deviate from "fair" profits over time if input prices, electricity demand, and other factors change, but the regulator does not institute a price change to re-equate earned profits with "fair" profits until the next rate hearing. In the meantime, a utility can alter its production and investment decisions in ways opposite to those predicted by the A-J model; The "fair" rate of return in Joskow's view is a means to an end (uncontroversial electricity prices), not an end in itself. After reviewing the regulatory experience across the U.S. between the 1950s and early 1970s, Joskow concludes that: Contrary to the popular view, it does not appear that regulatory agencies have been concerned with regulating rates of return per se. The primary concern of regulatory commissions has been to keep nominal prices from increasing. Firms which can increase their earned rates of return without raising prices or by lowering prices (depending on changing cost and demand characteristics) have been permitted to earn virtually any rate of return that they can. Formal regulatory action in the form of rate of return review is primarily triggered by firms attempting to raise the level of their rates or to make major changes in the structure of their rates. The rate of return is then used to establish a new set of ceiling prices which the firm must live with until another regulatory hearing is triggered. General price reductions do not trigger regulatory review, but are routinely approved without formal rate of return review. http://clevelandfed.org/research/workpaper Best available copy This regulatory process is therefore extremely passive. Regulators take no action regarding prices unless major increases or structural changes are initiated by the firms under its jurisdiction. In short, it is the firms themselves which trigger a regulatory rate of return review. There is no "allowed" rate of return that regulatory commissions are continuously monitoring and at some specified point enforcing. (Joskow, 1974, p. 298) Because they work in a political environment, public utility commissions face other sources of conflict and criticism, which have resulted in two additional constraints on utility behavior. First, in the mid-1970s, when energy costs increased rapidly, utilities requested rate hearings in greater numbers than in the past. This increased caseload put a large burden on these regulatory agencies, who were accustomed to only a few hearings in a year. The time lag between the request for a rate hearing and a change in electricity prices increased, and many utilities were forced to request another rate hearing immediately after their previous hearing. In order to shorten this lag and to appease utilities, regulators instituted fuel-costadjustment clauses that permitted utilities to pass higher fuel costs to consumers without the need for a formal rate hearing. Second, environmental advocates successfully agitated public utility commissions to establish limits on the amount of pollution emitted by fossil-fueled utilities. These two constraints complicate the analysis of the impact of a rate-of-return constraint on utility behavior. 111. Rate Hearings and Average Costs of Ohio Utilities: 1965 to 1982 Some evidence consistent with Joskow's view of the regulatory mechanism is found in the history of rate hearings in Ohio between 1965 and 1982. To put this evidence into perspective, refer to the figure on page 26, which shows the behavior of the average price per kilowatt-hour of electricity charged, and the quantity of kilowatt-hours sold, by the seven major Ohio electric utilities. http://clevelandfed.org/research/workpaper Best available copy For the purposes of t h i s discussion, three d i s t i n c t periods of- d i f f e r e n t nominal e l e c t r i c i t y price and consumption behavior can be seen: 1965 t o 1968, 1969 t o 1975, and 1976 to 1982.6 Within each period, the directions of change i n price and quantity were the same f o r each u t i l i t y i n the sample. During the 1965 t o 1968 period, the average p r i c e of e l e c t r i c i t y changed very l i t t l e and e l e c t r i c i t y s a l e s rose considerably. During the 1969 t o 1975 period, theaverage annual growth r a t e of e l e c t r i c i t y sales slowed, while t h a t of prices increased greatly. Between 1976 and 1982, e l e c t r i c i t y s a l e s declined f o r the f i r s t time i n Ohio's h i s t o r y , while prices increased a t t h e i r f a s t e s t average annual percentage r a t e . The figure also shows the percentage of the seven u t i l i t i e s requesting r a t e hearings i n each year. In -the f i r s t period, u t i l i t i e s rarely requested r a t e hearings, and t h e i r average costs were f a l l i n g . This behavior corresponds with Joskow's f i r s t proposition: "During periods of f a l l i n g average cost we expect t o observe v i r t u a l l y no regulatory r a t e of return reviews" (p. 299). The average price of e l e c t r i c i t y also was f a l l i n g during t h i s period, consistent with Joskow's second proposition: "During periods of f a l l i n g average costs we expect to obsewe constant or f a l l i n g prices charged by regulated firms" (p. 299). Given t h a t there were few r a t e hearings i n t h i s period, it i s plausible t h a t u t i l i t y returns on capital were greater than or equal t o what the " f a i r " returns the Public U t i l i t i e s Commission of Ohio (PUCO) would have defined'had they been requested t o do so.' According t o Joskow, i f actual returns were lower than the " f a i r " return, then the u t i l i t i e s would have asked f o r price increases. Hence Joskow's t h i r d proposition: "During periods of f a l l i n g average costs we expect t o observe r i s i n g or constant ( p r o f i t maximizing) r a t e s of return" (p. 299). - During the 1969 t o 1975 period, average costs increased s l i g h t l y , t r i g gering a modest increase i n the frequency of hearings, while during the 1976 http://clevelandfed.org/research/workpaper Best available copy to 1982 period, the average costs increased tremendously. Production costs increased in the late 1960s because of inflation stimulated by economic policies; they increased very quickly and unexpectedly in the mid-1970s because of inflation engendered by worldwide food shortages and by the Arab oil embargo. For a given electricity price, such increases in operating costs drove utility profits below their "fair" levels. Utilities promptly responded to these cost increases by requesting electricity price increases that, in most cases, were granted by the PUCO. The frequency of hearings increased sharply as utilities had trouble keeping up with the effects of the rapid rise in costs. Viewing the 1969 to 1975 period as a transition from a period of falling average costs to one of rising average costs, the modest increase in rate hearings during this period is consistent with Joskow's fifth proposition: The transition from a period of falling average costs to one of rising average costs for a particular regulated industry will at first yield no observable increase in the number of rate of return reviews filed by the regulatory agency, but as cost increases continue more and more rate of return reviews are triggered as firms seek price increases to keep their earned rates of return at least at the level that they expect the commission will allow in a formal regulatory hearing. (p. 300) For estimation purposes, the 1965 to 1982 interval was divided into two periods: 1965 to 1973 and 1974 to 1982. Testable hypotheses of the A-J and Joskow views deal with the absolute and relative production inefficiencies of the utilities in these two periods. The near absence of regulatory hearings in the first period would suggest, to both Joskow and A-J, that earned rates of return of these utilities were at least as great as "fair" rates of return. Averch and Johnson would argue that earned rates of return were lower than monopoly rates of return and, hence, that the A-J bias should exist in the first period. On the other hand, Joskow would argue that earned rates of return may have been close to monopoly rates. If this were true, then because monopoly rates are consistent with efficient production, there may have been http://clevelandfed.org/research/workpaper Best available copy very little A-J bias in the first period. Indeed, as Joskow argues in his seventh proposition, production may have been very efficient in the first period because reducing costs would have contributed to higher earned rates of return that were not taken away by regulators: During periods of falling or constant nominal average cost firms have an incentive to produce efficiently since all profits may be kept as long as prices stay below the level established by the regulatory commission in the last formal rate of return review. (p. 303) The high frequency of hearings in the 1974 to 1982 period suggests that earned rates of return for these utilities were lower than "fair" rates of return for most of the period. Because these earned rates were even further away from monopolistic rates of return, Joskow would argue that it is more likely that there are inefficiencies of the A-J type in the second period. His proposition eight says: "During periods of rising average cost A-J type biases may begin to become important" (p. 304). He does not exclude the possibility that firms may continue to try to be as efficient as they were in the first period in order to earn greater than "fair" rates of return. However, he argues that: Unless the direction of the cost path can be changed, however, the continuous interaction of firms and regulators in formal regulatory hearings, resulting from the necessity to raise output prices, is exactly the situation for which the A-J type model (with some modifications) would hold. I would therefore expect that it is under this situation of continuously rising output prices, triggering rate of return reviews that the A-J type models and the associated results are most useful. (p. 304) Thus, Joskow would ar.gue that utilities would try to organize their production more efficiently in the first period than in the second period. His concept of production efficiency includes the static notion of employing currently available production inputs in the least-cost way for any given level of output (that is, employing the least-cost combination of inputs along a given isoquant) and the dynamic notion of investing in more productive capital and management techniques over time (to push the family of isoquants http://clevelandfed.org/research/workpaper Best available copy toward the origin). Averch and Johnson deal only with the static notion of productive inefficiency because their model analyzes a static equilibrium. They would argue that the amounts of this static inefficiency are the same in both periods because they assume a regulator who maintains the earned rate of return on capital at its "fair" rate. The distinction between the static and dynamic notions of production efficiency is important. When a public utility commission conducts a rate hearing, it pays attention only to the static notion of production efficiency. Indeed, most models of regulatory impact deal only with the static notion. However, it is conceivable that regulation also affects the rate of technical change implemented by utilities; if regulation biases the amount of capital employed by a utility, it also may bias the type of capital employed. Regulatory impacts on overall inefficiency and on the rate of technical change are estimated below. IV. Empirical Model A. The Generalized or Shadow Cost Model The A-J and Joskow views are examined using a modified version of the Atkinson and Halvorsen(1984) generalized long-run cost function approach with capital (K) , labor (L) , and fuel (F) as inputs. Atkinson and Halvorsen argued that the long-run neoclassical cost-function approach is incorrect for a regulated firm because it assumes the firm is minimizing cost in a perfectly competitive world constrained only to produce a given level of output.g When the firm is subject to a number of regulatory constraints, the marginal product of each input does not equal the market price of the input, but the market price of the input plus the marginal changes in the additional constraints weighted by their Lagrange multipliers. Atkinson and Halvorsen use the term "shadow" prices to refer to these modified market input prices. The - http://clevelandfed.org/research/workpaper Best available copy - 11 - exact specification of these shadow prices depends on the exact form of the additional constraints. Atkinson and Halvorsen approximated these shadow prices by simple proportional relationships with market prices; that is, the shadow price of input i P: = kipi,where Pi is its market price and ki is a constant. The generalized or shadow cost function is simply the neoclassical cost function with P: substituted for Pi: where CS is the shadow total cost of electricity production; pf is the shadow price of input factor i, i Q is output of electricity; and T is time. = K, L, F; Instead of minimizing long-run actual costs, a utility is assumed to minimize long-run shadow costs by equating the shadow marginal cost of each input with the amount of the input used. If the additional constraints are not binding, then the ki equal one and minimizing shadow costs is equivalent to minimizing actual costs. If the ki do not equal one, then the firm is not operating at the lowest point of its long-run average cost curve. An observable cost function based on the shadow cost function can be derived as follows. First, recall the accounting identity for actual cost: where Xi is the quantity of input i used in production. Similarly, the accounting identity for shadow cost is: The shadow cost share equations: PQ.xi Mi = - c for i = K, L, F can be rewritten as: (2 C"M; pixi = ki for i = K, L, F http://clevelandfed.org/research/workpaper Best available copy - 12 - and summed over all i to obtain: Taking logarithms of both sides of (3a) yields: That is, the logarithm of actual cost equals the logarithm of shadow cost plus the logarithm of the sum of the shadow cost shares each weighted by the inverse of its respective ki. To express each shadow cost share as a function of its corresponding actual cost share, first divide both sides of (2) by ca: and substitute (3a) into (4) : for i = K, L, F. Atkinson and Halvorsen estimate a system comprising (3b) and two of the three equations in (5) but without a time trend because they use cross-section data. We add the appropriate time variables to the shadow cost equation and add a shadow TFP (TFP') equation to our system in order to improve the efficiency of the shadow cost equation coefficient estimates. Actual TFP is measured as the change in the average cost of production that is not due to changes in input prices. It reflects the overall productivity of all inputs rather than the productivity of a single input such as labor. The neoclassical approach to the measurement of TFP assumes an optimal distribution of production resources in a firm, which may be an inappropriate assumption for regulated electric utilities. The generalized cost-function approach yields a shadow estimate of TFP that is consistent with regulated behavior. The most important variable for the purposes of examining Joskow's view on productivity http://clevelandfed.org/research/workpaper Best available copy - 13 - behavior is the pure technical change component of TFP. Gollop and Roberts (1981), among others, argue that this component is a better measure of productivity than TFP. The TFP' equation is derived as follows. First, take the time derivative of (la): According to Shephard's Lemma, the elasticity of actual total cost with respect to the market price of input i is equal to the share of input i in total cost: A modified Shephard's Lemma for.the shadow cost function is: Dividing both sides of (6) by cS and using(8) yields a functional relationship between the percentage change in shadow cost and the percentage changes in the P:, Q, and T: where a dot over a variable indicates the rate of change, v; is the elasticity of shadow cost with respect to output (alncS/aln~), and 9 is the rate of change in shadow cost, holding all other variables constant (BlncS/a~). scale, and -4is the (1-v8) is a measure of shadow returns to measure of shadow technical change of interest in this paper. Next, following the traditional definition of actual TFP as a Divisia index of factor inputs, the rate of change in TFP' (us)can be defined as: http://clevelandfed.org/research/workpaper Best available copy . - 14 - Totally differentiating the accounting identity (lb) with respect to time and using Shephard 's Lemma yields: Equations (10) and (11) imply: and using (9): Finally, because Ci M ~ 1, the above expression can be = rewritten in terms of one of the xi,say x=: Equation (12a) cannot be used for estimation purposes because WS is not observed. It can be used to obtain an equation explaining the actual rate of change in TFP as a function of w', but (12b) is easier to estimate. The general specification of our estimation model includes the total cost equation (3b) , the TFP equation (12b). and M; share equations from (5) , and the The estimation model is based upon the translog functional form. The translog shadow cost function is: 7gT(lnQ)T with P: = + 1 %T2, kip, and the usual linear homogeneity restrictions: http://clevelandfed.org/research/workpaper Best available copy 7 = 0 and 9 iJ foralliandj. y..=y.. 1J J1 Using Shephard's Lemma (a), the shadow cost share equations are: Substituting (13), (14), and the definition of P; into (3b) yields an estimable cost equation. Substituting (14) and the definition of P: into (5) yields estimable cost share equations. Finally, an estimable TFP equation is obtained by substituting(14) and the following v; and 9 expressions into (12b): Two modifications are made to these equations. First, separate values for the ki coefficients were estimated for the 1965 to 1973 and 1974 to 1982 periods in order to estimate a shift in regulatory impact. The ki coefficient estimates for the 1974 to 1982 period are denoted with a subscript " S" Second, % and hswere normalized to one because the shadow cost system is homogeneous of degree zero in the ki. This means that only relative price efficiency can be examined using the ki,by testing ki= kj=l and kis=kjs=lfor i,j = K,F. Differences in absolute price efficiency between the two periods, relevant for a test of Averch-Johnson versus Joskow, cannot be tested using differences between ki and kis. http://clevelandfed.org/research/workpaper Best available copy Nevertheless, the model can serve to test Averch-Johnson against Joskow, as described below.lo These translog equations form a nonlinear, seemingly unrelated regression system. It is similar to that of Gollop and Roberts (1981), only generalized to allow for the impact of all types of regulation on utility behavior. The maximum likelihood LSQ option of TSP, version 4.OE, was used to estimate this translog system. B. Data Data for labor input and the price of labor are taken from Financial Statistics of Selected Electric Utilities, 1982, Department of Energy (DOE/EIA-0437(82)), February 1984. The quantity of labor is the number of electric department employees, with a part-time worker counted as one-half of a full-time worker. The labor price is defined as the ratio of labor expense to the quantity of labor, where labor expense is total salaries and wages charged to electric operation. The fuel price data come from Standard and Poor's Compustat Services, Inc., Utility Compustat 11. Fuel operation expense is the total cost of fuel used exclusively for the production of electricity. The price of fuel is the average cost of fuel per million Btu, which is the total cost of fuel used for electricity production divided by its total Btu content in millions. The quantity of fuel input is millions of Btu, defined as the ratio of fuel operating expenses to the average cost per million Btu. The data for the capital price and capital stock come from various issues of Statistics of Privately Owned Electric Utilities in the United States, U.S. Federal Power Commission. The capital price measure is the conventional market price of capital, which is a function of the long-term debt interest rate, the required return on equity capital, the preferred stock dividend rate, the depreciation rate, and the Handy-Whitman index. The http://clevelandfed.org/research/workpaper Best available copy capital stock is computed using a perpetual inventory method.ll The depreciation rate is based on a 30-year average service life.12 The product of capital price and capital stock is the total capital costs. Total cost is the sum of labor, fuel, and capital costs. V. Empirical Evidence A. Model Characteristics The results of estimating the model over the 1965 to 1982 period are shown in table 1. Before testing the regulation bias hypotheses, it is useful to examine the sense of the estimated model. A quick glance at the t-statistics - shows that the explanatory variables are just that - - only two of the 25 estimated coefficients have t-statistics less than 2 in absolute value. Apart from the ki,the statistical significance of the coefficients does not necessarily provide strong evidence about the adequacy of the estimated model. Instead, characteristics of the production technology implied by the coefficients provide better clues of model plausibility. The estimated returns to scale are a good check of model adequacy for utilities because utilities ought to display increasing returns to scale given the large fixed costs required to supply electricity over an extensive geographic market. Table 2 reports the estimates of the elasticity of cost with respect to output averaged over all firms for each year. The shadow estimate is the elasticity of shadow cost with respect to output from (15). The actual estimate is the shadow elasticity adjusted for the difference between actual and shadow costs: If returns to scale are increasing, then the cost elasticity is less than one. As shown in table 2, the cost elasticities averaged over firms indicate http://clevelandfed.org/research/workpaper Best available copy - 18 - increasing returns to scale over the whole sample period. Both the shadow and the actual elasticities behave similarly over time: the size of the increasing returns to scale grows moderately over the first period and shrinks over the second, and returns to scale are greater on average in the second period. These results are consistent with the behavior of output over these periods. In the first period, as output was increasing, these utilities were operating on lower portions of their average cost curves, where returns to scale are lower. In the second period, as output grew more slowly and eventually fell, these utilities operated on higher portions of their average cost curves, where returns to scale are higher. These results are the opposite of those of Gollop and Roberts (1981), who do not allow for a regulatory bias. They find an increase in returns to scale in the first period and a drop in returns to scale in 1974-75, the last years of their sample. As further evidence, constant returns to scale and homogeneity of the cost function are tested. Homogeneity means that scale economies are the same for firms of all sizes in all years, and constant returns to scale means that there are no cost savings to increasing plant size. Homogeneity requires that -y =-y =-y =-y =O; LQ FQ TQ QQ constant returns requires homogeneity plus PQ=l. Both homogeneity and constant returns to scale are rejected at better than the 0.5 percent significance level. . The estimated actual and shadow cost shares for the inputs are shown in tables 3a and 3b, respectively. The actual cost shares show that capital was the largest component of actual cost in the first period, and that labor became the largest cost component in the second period; fuel was the smallest - cost component in both periods. The shadow cost shares show that capital and labor were the largest and smallest cost components, respectively, in both periods. The difference between the actual and shadow cost shares is rather http://clevelandfed.org/research/workpaper Best available copy - 19 - dramatic, and again reflects the ratio of shadow to actual cost from (4) and (5). The large difference suggests that looking at the actual cost shares will give a misleading picture of the reaction of these utilities to changes in regulated prices. Table 4 shows the decomposition of the growth rate of actual average cost into its components. This decomposition is similar to that for the growth rate of shadow cost (11): The first column of table 4 shows the average growth rate of actual average cost for each year. The next three columns are the M;P~ terms for the three inputs; the fifth column shows the contribution of the returns to scale term ($-1)~using vt from (16) ; the sixth shows the contribution of the technical change term v;: The last column is simply the difference between the first column and the sum of the next five. This remainder is not zero, because the five components on the right-hand side of (17) are estimated. Note that this remainder is not derived from any of the estimated regression equations. Every cost component except scale economies on average added to the growth of average costs in both periods. Capital and fuel were the largest contributors to average cost growth in both periods, and capital and technical change accounted for much of the increase in the growth rate of average costs between the two periods. The remainder is about one-sixth the size of the average growth rate of average costs in the first period, but it is very small in the second. This suggests that the shadow cost model fits the second period much better than the first. http://clevelandfed.org/research/workpaper Best available copy - 20 - Estimated values for actual and shadow TFP and its components are shown in tables 5a and 5b. The shadow TFP measure is the partial derivative of shadow average cost with respect to time, which is the sum of two terms, the first reflecting scale economies and the second representing technical change: (19) TFP~= (1-V;)Q - v;. Actual TFP (TFP~)comes from (19) but with v; from (16) replacing v6, and 9 from (18) replacing v;. The results in table 5a show that scale economies have boosted T F P ~growth in every year except 1980, though the gain was significantly less in the second period. However, technical change was negative in every year but 1965, pulling the growth of TFP~down, especially in the second period. The results for TFP', shown in table 5b, are qualitatively similar to those of TFPa and its components, though it is interesting that 9 was slightly positive on average in the first period. The most notable characteristic about both technical change estimates is their strong downward trend.13 This rather uniform decline is due to the strong estimated time trend yTT. That shadow input prices have little influence on technical change is not surprising, because electricity production offers little input substitutability in the short and medium runs. B. Regulatory Impact The estimation results in table 1 show that all of the log(ki) are individually significantly different from zero at better than the 0.5% significance level. The joint test of the statistical insignificance of all four of the log(ki) is rejected at better than the 0.5 percent significance level. Thus, relative price efficiency is rejected over the whole sample, and the neoclassical cost function approach for regulated firms employed by Gollop and Roberts (1981) and others is inappropriate for this sample.l4 A test of the A-J view and a test of the implications of Joskow's view is whether production inefficiencies due to regulation differ in the 1965 to 1973 http://clevelandfed.org/research/workpaper Best available copy - 21 - and the 1974 to 1982 periods. The A-J view is that the inefficiencies should be the same in each period, while the Joskow view is that there should be greater inefficiencies in the second period than in the first. The true cost of regulation, and hence the magnitude of the inefficiencies created by regulation, cannot be estimated, because there is no evidence to suggest how the utilities would have organized their production had regulation not existed over the sample period. For example, the activities of production and distribution might have been separated, different amounts of capital might have been employed, and different technologies might have been chosen.l5 Hence, it is impossible to know what these firms' cost functions and associated returns to scale and productivities would have been. However, "instantaneous" tatal and dynamic inefficiency estimates can be computed. The total measure compares actual utility costs predicted by the estimated model with the actual costs predicted by the model, but with and & % set equal to one in both periods. That is, current production costs for actual levels of output, which are generated by current production techniques and regulatory constraints, are compared with the costs generated with the same production techniques and for the same actual output levels, but without the regulatory constraints. This estimate, also examined by Atkinson and Halvorsen, measures movement along the isoquant to the efficient input mix. An estimate of the dynamic notion of inefficiency can be obtained by examining the technical change experienced by these utilities with and without regulation. As above, technical change with regulation is that implied by the estimated model; technical change without regulation is that implied by the estimated model, but with all ki set equal to one. The difference does not have a real-world counterpart or explanation, but it does indicate the direction of regulatory bias. http://clevelandfed.org/research/workpaper Best available copy - 22 - Note that our measure of the regulatory impact on technical change is different from that of Nelson and Wohar (1983). In their model, TFP is the sum of the technical-change term, the returns-to-scale term, and a separate regulatory impact. Without regulation, TFP is the sum of only the returns-toscale and technical-change terms. This naturally begs the question of how regulation affects TFP if it does not affect the components of TFP. Obviously, Nelson and Wohar cannot test for a regulatory impact on technical change. Their measures of a regulatory impact on technical change are purely hypothetical, based on the difference between different TFP values calculated using assumed, not estimated, values for the regulatory impact coefficient, and their returns-to-scale and regulatory impact terms. The reader is left to wonder why the authors believed that regulation does not affect the returns-to-scale term. Two sets of measures can be examined for a regulatory impact: actual and shadow. As shown in Israilevich and Kowalewski (1987), the actual cost and the actual and shadow returns-to-scale and technical-change equations are homogeneous of degree zero in the ki,while the shadow cost equation is not. Thus, either the actual or the shadow returns-to-scale and technical-change measures can be used to examine the regulatory bias. The regulatory bias to the translog shadow measures is a constant for each variable in each period. This can be seen by subtracting the translog shadow equation for any of these variables from the same equation, but with the ki set equal to one. The reason is that the cost-minimization model is set in a static equilibrium framework. The regulatory biases to the actual variables are not constant because they differ from the shadow measures by a proportional function of the ratio of shadow to actual cost. This ratio, and hence the degree of - http://clevelandfed.org/research/workpaper Best available copy regulatory bias, varies over time. We prefer to use the actual measures to examine the regulatory bias for this reason and because the shadow measures have no real-world meaning. Our inefficiency estimates reject Averch and Johnson's view and do not reject Joskow's view. As shown in table 6, the total inefficiency measure differs between the two periods, contrary to Averch and Johnson's view. Moreover, the direction of change between the two periods is what Joskow would expect - - total inefficiency is about 16 percentage points greater in the second period. In the first period, total inefficiency steadily increases from about 61.5 percent to 73.8 percent and averages about 66.6 percent. In the second period, it steadily increases from 74.9 percent to 87.4 percent and averages about 82.6 percent. These total inefficiency estimates give the appearance of being overly large in magnitude. Atkinson and Halvorsen find much smaller inefficiency losses (9.0 percent) in their cross-section sample of 1970 firms, which includes two of our firms.16 However, the Atkinson and Halvorsen result captures only the static portion of total inefficiency costs because they do not use time variables in their cost equation. Our estimates include the dynamic inefficiency costs, and hence are more representative of the total costs of regulation. The difference between the Atkinson and Halvorsen result and ours suggests that the dynamic inefficiency may be quite large. Indeed, as shown in table 7, we find that regulation may have retarded the growth of technical change on average by about 0.64 percentage point per year in the first period and by 0.44 percentage point per year in the second. This an important result, and one that has been neglected by economists and regulators alike. Regulation not only affects the efficient utilization of existing production inputs, but it also affects the implementation of efficient capital and http://clevelandfed.org/research/workpaper Best available copy - 24 - management techniques over time. Unlike our total inefficiency estimates, the dynamic portion of our total inefficiency estimate rejects Joskow's view of greater regulatory bias in the second period. This regulatory bias on technical change is opposite to the casual impression given by the trends in actual and shadow technical change shown in tables 5a and 5b. The strong downward trends in both technical change measures, especially given the total inefficiency cost estimates shown in table 6, might lead some analysts to infer that tighter regulatory constraints contributed to the slowdown in technical change in the second period. However, table 7 shows that the regulatory bias on technical change was less in the second period. Finally, an interesting result in table 7 is that regulation biased returns to scale upward on average in both periods. Contrary to Joskow's view, the regulatory bias on returns to scale is smaller in the second period. Netting out the two components, TFP was biased down by 0.46 percentage point per year in the first period and by about 0.33 percentage point per year in the second. This result also rejects Joskow's view of greater dynamic inefficiency in the second period. VI. Swnmary and Conclusions Electric utility regulators attempt to maintain a competitive price for electricity by adjusting the rate of return on a utility's capital. At first blush, this price-setting scheme appears sensible. It seems reasonably efficient to allow utilities to pass along operating costs and to cover their cost of capital. However, there are potentially serious problems with this type of regulation related to consumer reactions to price increases and to the types of incentives given to utilities. First, price increases may lower the consumption of electricity, which may reduce earned rates of return below http://clevelandfed.org/research/workpaper Best available copy - 25 - "fair" rates and trigger a price increase, which in turn may lower consumption and trigger another price increase, and so on. That is, the proper response to falling utility profits because of lower demand may not be to raise prices. Second, utilities may be able to effect price increases by using "too much" capital, that is, by overcapitalizing, which inflates their rate base. Indeed, rate increases lower the risk of capital investment below the risk level of unregulated industries, clearly giving utilities the incentive to overcapitalize. This potential bias was recognized by Averch and Johnson, and many empirical studies that adopted their model found an overcapitalization bias. Finally, the ability to pass along operating cost increases that originated from productivity declines suggests that utilities may not have the incentive to raise productivity. This dynamic source of inefficiency was recognized by Joskow, who also argued that the regulatory mechanism is more complicated than that assumed by Averch and Johnson. This paper is the first, to our knowledge, to explicitly test the AverchJohnson view against Joskow's more general view. Using a modified version of the generalized long-run cost function derived by Atkinson and Halvorsen and a sample of the seven major electric utilities in Ohio over the 1965 to 1982 period, substantial evidence is found against the A-J view. Our total inefficiency measure shows that regulatory constraints were more binding during the years in which Joskow expects.them to be more binding. We also find that regulation substantially retards the rate of technical change experienced by these utilities. However, the retardation in technical change is greater during the years when Joskow expects regulation to be less binding. This is the first demonstration of a regulatory impact on technical change. - It clearly suggests that regulators ought to pay closer attention to the incentives they give utilities to innovate.l7 http://clevelandfed.org/research/workpaper Best available copy - 26 - PRICE, AVERAGE COST, AND RATE HEARING FREQUENCY Electricity peY' 30% 25% 20% 15% 10% 5% 0% - 5% =I5 1 66 I I I I I Average % 65 I I I I I I Cost da11ar- ~ 1 I I I I I I I I I I I I I I I 67 69 71 73 75 77 79 81 68 70 72 74 76 78 80 82 66 year Rate H i year SOURCE: The authors. I 6'7 69 71 7'3'75 77 79 81 68 70 72 74 76 78 80 82 year ourrent 30% 25% 2 0% 15% 10% 5% 0% - 5 Price k1loVVECtt-he-r I Frequency http://clevelandfed.org/research/workpaper Best available copy TABLE 1 COEFFICIENT ESTIMATES Coefficient ----------- Estimate --------- 2.309321 2.408440 2.504495 1.756207 -7.593048 .I308763 log(h) log(&) log($) log(bs) Q SL SF ST PQ TLK YFK YE, YLQ YFQ Std. Error -----------.I192982 .I239441 .2273822 .2103573 1.780928 .2288987D-01 .7871482D-01 .2978847D-01 .3934191 .5171896D-02 .1416839D-01 .4092725D-02 .1919976D-02 .9527101D-02 ~ L T VET ~ Q T TQQ YTT COEFFICIENTS COMPUTED FROM PARAMETER RESTRICTIONS Coefficient ----------SK %K ~ F E r~~ ~ K Q YKT Estimate Std. Error T-Statistic 1.4586 6.46143-02 1.06233-01 8.03743-02 1.69563-02 1.40533-02 18.1472 3.8108 7.5588 -3.42733-02 -8.61253-02 1.14703-03 5.43513-03 9.56773-03 1.05543-03 -6.3058 -9.0017 1.0869 ---------- ---------- ----------- http://clevelandfed.org/research/workpaper Best available copy TABLE 2 ESTIMATED ELASTICITY OF COST WITH RESPECT TO OUTPUT (averaged over firms) Year NOTE: Actual Shadow The e l a s t i c i t y of shadow cost with respect t o output is computed using % from equation (15). The e l a s t i c i t y of actual c o s t i s computed from equation (16). http://clevelandfed.org/research/workpaper Best available copy TABLE 3a ESTIMATED ACTUAL COST SHARES (averaged over firms) Year Capital ------- ------- Labor ------ 1965 1966 1967 1968 1969 1970 1971 1972 1973 0.5494 0.5381 0.5345 0.5245 0.5154 0.4981 0.4790 0.4668 0.4536 0.2824 0.2938 0.2932 0.3047 0.3085 0.3203 0.3433 0.3572 0.3694 0.1682 0.1681 0.1723 0.1707 0.1761 0.1816 0.1777 0.1760 0.1770 ---- Fuel NOTE: The actual cost shares are computed using equation (7). http://clevelandfed.org/research/workpaper Best available copy TABLE 3b ESTIMATED SHADOW COST SHARES (averaged o v e r f i r m s ) Fue 1 Capital ------- ------ ------ 1965 1966 1967 1968 1969 1970 1971 1972 1973 0.7076 0.7021 0.6957 0.6915 0.6815 0.6665 0.6605 0.6569 0.6483 0.0372 0.0389 0.0386 0.0405 0.0410 0.0430 0.0476 0.0508 0.0532 0.2552 0.2590 0.2657 0.2680 0.2774 0.2905 0.2919 0.2923 0.2985 ---- - Labor Year NOTE: The shadow c o s t s h a r e s are computed u s i n g e q u a t i o n ( 1 4 ) . http://clevelandfed.org/research/workpaper Best available copy TABLE 4 ESTIMATED COMPONENTS OF THE RATE OF CHANGE IN ACTUAL AVERAGE COST (percentage change, averaged over firms) Year ---1966 1967 1968 1969 1970 1971 1972 1973 NOTE: Average Cost Capital ------- ------- -2.435% 3.210% 0.217 - 0.066 5.040 2.337 2.412 1.974 6.503 11.563 19.067 10.720 9.920 8.679 8.069 9.714 Labor Fuel Scale Econ. Tech. Change Remainder 0.733 2.046 6.117 4.822 2.512 3.689 -2.963 - 2.418 -1.353 -0.742 -2.250 -2.857 0.719 1.054 1.327 1.587 1.912 2.241 -1.384 -1.296 -1.326 2.397 -1.253 -1.833 ------ ------- ------- --------0.516% - 2.888% 0.038% -3.413% 1.083 - 2.012 0.129 0.362 These figures are computed using equation (17). The sum of the capital, labor, fuel, scale economies, and technical change columns is the estimated percentage change in actual average cost. The difference between the average cost column and this estimated percentage change is the remainder. http://clevelandfed.org/research/workpaper Best available copy TABLE 5a ESTIMATED ACTUAL TOTAL FACTOR PRODUCTIVITY AND COMPONENTS ( p e r c e n t a g e change, a v e r a g e d o v e r f i r m s ) Year Total Factor Produc . 1965 1966 1967 1968 1969 1970 1971 1972 1973 2.5797% 2.8500 1.6501 2.2433 1.3636 0.0256 -0.8451 0.3383 0.6156 ---- -------- Scale Econ. Tech. Change .------- 0.312% -0.038 -0.362 -0.719 - 1.054 -1.327 -1.587 -1.912 - 2.241 NOTE: A c t u a l t o t a l f a c t o r p r o d u c t i v i t y and i t s two components are computed from e q u a t i o n (19) b u t w i t h $ from (16) r e p l a c i n g vc a n d v; from (18) r e p l a c i n g v;. http://clevelandfed.org/research/workpaper Best available copy TABLE 5b ESTIMATED SHADOW TOTAL FACTOR PRODUCTIVITY AND COMPONENTS (percentage change) Year ---- 1965 1966 1967 1968 1969 1970 1971 1972 1973 NOTE: Total Factor Produc . Scale Econ. Tech. Change -------1.3841% 1.0200 0.6974 0.3259 - 0.0121 -0.2976 -0.5876 -0.9304 - 1.2743 Shadow total factor productivity and its components are computed using equation (19). http://clevelandfed.org/research/workpaper Best available copy TABLE 6 TOTAL REGULATORY IMPACT ON ACTUAL COST Year Estimated Actual Cost No Regulation Actual Cost Regulatory Impact 1965 1966 1967 1968 1969 1970 1971 1972 1973 102.614 114.387 121.560 140.683 156.215 185.397 223.909 267.667 329.468 63.958 70.675 75.208 86.073 95.322 111.908 132.645 157.020 191.387 61.529% 62.980 62.872 64.390 64.921 66.579 69.875 71.987 73.815 ---- -------- --------- ---------- http://clevelandfed.org/research/workpaper Best available copy TABLE 7 REGULATORY IMPACT ON ACTUAL TOTAL FACTOR PRODUCTIVITY (percentage change) Year Total Factor Produc . -------- -0.4696% - 0.4120 -0.5156 -0.4020 -0.4514 -0.5337 -0.5463 -0.4313 -0.3850 Scale Econ. Tech. Change --------0.6837% -0.6691 -0.6713 -0.6573 -0.6539 - 0.6409 -0.6113 -0.5936 -0.5789 NOTE: The columns show t h e d i f f e r e n c e between t h e estimated a c t u a l measures and t h e estimated a c t u a l measures w i t h t h e ki a l l s e t equal t o . o n e . http://clevelandfed.org/research/workpaper Best available copy FOOTNOTES lThis interpretation of the Averch-Johnson result is due to Baumol and Klevorick (1970). 2That is, Courville (1974), Spann (1974), Petersen (1975), Cowing (1978), and Nelson and Wohar (1983), for example, test only for an overcapitalization bias against an alternative hypothesis of no bias. Of these papers, only Nelson and Wohar do not find an overcapitalization bias. 3~he seven major electric utilities in Ohio are Ohio Power; Cincinnati Gas and Electric; Cleveland Electric Illuminating; Columbus and Southern Ohio Electric; Dayton Power and Light; Ohio Edison; and Toledo Edison. Over the 1965 to 1982 period, they accounted for about 90 percent of electric power sales in Ohio. 4~ctually,Baumol and Klevorick (1970) argue that Averch and Johnson did not prove this as a general-result. Note that if there are additional production factors, then the amount of capital relative to these other inputs also will be higher than for the cost-minimizing firm. 5~ slight modification to the Averch-Johnson regulatory process was the introduction of a "regulatory lag"; see, for example, Bailey and Coleman (1971) and Baumol and Klevorick (1970). 6~he average price shown in the figure is not the regulated price, but the ratio of average total revenue for the seven utilities to their average total sales. In general, different consumers face different regulated price schedules, and utilities serving different geographic markets may be allowed to charge different prices for the same category of consumer. 7 ~ can t never be known whether earned returns were greater than "fair1' returns because there were no rate hearings for all firms during these years. 80ther production inputs, such as materials, managerial skills, and available infrastructure, for example, are excluded because there are no reliable data for these factors. g~evertheless, some authors, for example Gollop and Roberts (1981,1983) , use the neoclassical approach to study electric utilities. strict test of Averch and Johnson's view using the ki is % not equal to 1 and hSnot equal to one, because Averch and Johnson consider only a rate-of-return regulatory constraint. If these hypotheses cannot be rejected, then Nelson and Wohar (1983) and other papers that test only this constraint are potentially incorrect. llsee Cowing, Small, and Stevenson(1981) for the equations used to compute the capital stock and capital price variables. 12capital Stock Estimates for Input -Output Industries: Methods and Department of Labor, Bureau of Labor Statistics, 1979. Data, Bulletin 2034, U.S. http://clevelandfed.org/research/workpaper Best available copy - 37 - FOOTNOTES 1 3 strong ~ downward trend in the rates of technical change experienced by utilities also was found by Nelson and Wohar (1983), Gollop and Roberts (1981), and Gollop and Jorgenson (1980), all of whom used samples that ended in the 1970s. Thus, the results reported here confirm these earlier findings for the late 1970s and early 1980s. 14The strict test of Averch and Johnson's view is rejected; % and kFs are jointly statistically different from zero at better than a 0.5 percent significance level. ''under the current regulatory environment, the production and distribution of electricity must be handled by each utility. Moreover, the transferal of electric power across state lines also is impeded. 161t is likely that our estimates are more accurate for Ohio because our sample includes only Ohio firms, which are fairly similar in a number of important respects, as mentioned earlier. he poor technical-change performance also may be due to increased investment in nuclear power plants over this period, which drew funds away from conventional power-generation capital investments. http://clevelandfed.org/research/workpaper Best available copy REFERENCES Atkinson, Scott E., and Robert Halvorsen. "Parametric Efficiency Tests, Economies of Scale, and Input Demand in U.S. Electric Power Generation," International Economic Review, vol. 25, no. 3 (October 1984), 647-62. Averch, Harvey, and Leland L. Johnson. "Behavior of the Firm under Regulatory Constraint," American Economic Review, vol. 52, no. 5 (December 1962), 1052 - 69. Bailey, E.E., and R.D. Coleman. "The Effect of Lagged Regulation in an Averch-Johnson Model," The Bell Journal of Economics and Management Science, vol. 2, no. 1 (Spring 1971),, 278-92. Baumol, William J., and Alvin K. Klevorick. "Input Choices and Rate-of-Return Regulation: An Overview of the Discussion," The Bell Journal of Economics and Mana=ement science, vol. 1, no. 2 (Autumn 1970), 162 -90. Berndt, Ernst R., and Melvyn A. Fuss. "Productivity Measurement with Adjustments for Variations in Capacity Utilization and Other Forms of Temporary Equilibrium," Journal of Econometrics, vol. 33, no. 1/2 (October/November 1986), 7-30. Christensen, L.R., and Dale W. Jorgenson. "U.S. Real Product and Real Factor Input, 1929-1967," The Review of Income and Wealth, series 16, no. 1 (March 1970), 19-50. Courville, Leon. "Regulation and Efficiency in the Electric Utility Industry," The Bell Journal of Economics and Management Science, vol. 5, no. 1 (Spring 1974), 53-74. Cowing, Thomas G. "The Effectiveness of Rate-of-Return Regulation: An Empirical Test Using Profit Functions," in M. Fuss and D. McFadden, Eds., Production Economics: A Dual Avvroach to Theory and Avvlication, Amsterdam: North Holland Publishing Company, 1978. Cowing, Thomas G., Jeffery Small, and Rodney E. Stevenson. "Comparative Measures of Total Factor Productivity in the Regulated Sector: The Electric Utility Industry," in T.G. Cowing and R.E. Stevenson, Eds., Productivity Measurement in Regulated Industries, New York: Academic Press, 1981. Fare, Rolf, and James Logan. "Shephard's Lemma and Rate of Return Regulation," Economics Letters, vol. 12 (1983), 121-25. Gollop, Frank M., and Dale Jorgenson. "U.S. Productivity Growth by Industry, 1947-1973," in J.W. Kendrick and B.N. Vaccara, Eds., Studies in Income and Wealth, National Bureau of Economic Research, Chicago: University of Chicago Press, 1980. http://clevelandfed.org/research/workpaper Best available copy Gollop, Frank M., and Mark J. Roberts. "The Sources of Economic Growth in the U.S. Electric Power Industry," in T.G. Cowing and R.E. Stevenson, Eds., Productivity Measurement in Regulated Industries, New York: Academic Press, 1981. Gollop, Frank M., and Mark J. Roberts. "Environmental Regulations and Productivity Growth: The Case of Fossil-Fueled Electric Power Generation," Journal of Political Economy, vol. 91, no. 4 (August 1983), 654-74. Israilevich, Philip and K.J. Kowalewski. "Estimating Total Factor Productivity in a Generalized Cost System," Federal Reserve Bank of Cleveland Working; P a ~ e r8702, March 1987. Joskow, Paul L. "Inflation and Environmental Concern: Structural Change in the Process of Public Utility Price Regulation," The Journal of Law and Economics, vol. 17 (October 1974), 291-327. Nelson, Randy A., and Mark E. Wohar. "Regulation, Scale Economies, and Productivity in Steam-Electric Generation," International Economic Review, vol. 24, no. 1 (February 1983), 57-79. Petersen, H. Craig. "An ~ m ~ i r i c aTest l of Regulatory Effects," The Bell Journal of Economics and Management Science, vol. 6, no. 1 (Spring 1975), 111-26. Spann, R.M. "Rate of Return Regulation and Efficiency in Production: An Empirical Test of the Averch-Johnson Thesis," The Bell Journal of Economics and Management Science, vol. 5, no. 1 (Spring 1974), 38-52.