Estimating Pollution Abatement Costs: A Comparison of Stated and Revealed Approaches. Rolf Färe* Shawna Grosskopf** and. Carl A. Pasurka, Jr.

Transcription

1 Estimating Pollution Abatement Costs: A Comparison of Stated and Revealed Approaches by Rolf Färe* Shawna Grosskopf** and Carl A. Pasurka, Jr.*** *Department of Economics and Department of Agriculture and Resource Economics Oregon State University Corvallis, OR **Department of Economics Oregon State University Corvallis, OR ***U.S. Environmental Protection Agency (1809) Office of Policy, Economics, and Innovation 1200 Pennsylvania Ave., N.W. Washington, D.C Phone: (202) FAX: (202) PASURKA.CARL@EPA.GOV C:\ELECTRIC\ELECT-12A.WPD DRAFT - DO NOT QUOTE OR CITE WITHOUT PERMISSION OF THE AUTHORS May 3, 2002 Earlier versions of this study was presented at the January 2001 AEA meetings in New Orleans and at the U.S. Environmental Protection Agency. Gale Boyd, Scott Farrow, and Anton Steurer provided helpful comments on an earlier version of this study. We thank Curtis Carlson for providing the capital stock and employment data, and Tom McMullen for providing the U.S. EPA emission estimates. Any errors, opinions, or conclusions are those of the authors and should not be attributed to the U.S. Environmental Protection Agency.

2 Estimating Pollution Abatement Costs: A Comparison of Stated and Revealed Approaches Abstract Surveys have been the principal method used to estimate costs associated with environmental regulations in the United States. Although surveys have been widely used, there are concerns about their accuracy. These concerns have been exacerbated by increased use of change-inproduction process techniques to abate pollution. In order to investigate the accuracy of survey estimates of pollution abatement costs, a joint production model is specified and data from power plants in the United States for 1994 and 1995 are used to estimate pollution abatement costs incurred by power plants. These estimates of pollution abatement costs generated by the joint production model are then compared with survey estimates of pollution abatement costs incurred by power plants. JEL Classification Code: Q28

3 I. Introduction Surveys have been the principal method used to estimate the costs associated with environmental regulations in the United States. 1 The Pollution Abatement Cost(s) and Expenditures (PACE) survey (U.S. Department of Commerce, 1996) estimated the pollution abatement costs borne by U.S. manufacturing industries for 1973 through 1994 (excluding 1987). In addition to the PACE survey, the Form EIA-767 survey ( Steam-Electric Plant Operation and Design Report ), which is administered by the Energy Information Administration of the U.S. Department of Energy, includes questions concerning pollution abatement expenditures. These survey estimates of pollution abatement costs, which were used by the Bureau of Economic Analysis (BEA) in its discontinued annual report on pollution abatement expenditures (see Vogan 1996), can be viewed as stated costs. For 1994, 64 percent of BEA s estimates of pollution abatement expenditures were from surveys and the remaining 36 percent of BEA s estimates were derived from other sources (Vogan, 1996, p. 54). According to the System for Integrated Environmental and Economic Accounts, SEEA, (United Nations 1993) current account expenditures for pollution abatement by business establishments are classified as either external or internal pollution abatement activities. External pollution abatement activities are undertaken by establishments for which the activity is its primary or secondary activity (e.g., sewage treatment). Internal pollution abatement activities (e.g., operating a scrubber) are those undertaken by establishments emitting the pollutant. While surveys appear to be the appropriate method for estimating the extent of external pollution abatement activities, they encounter difficulties when estimating the costs associated with internal pollution abatement activities embodied in the technology used by a producer. Although surveys of pollution abatement costs have been conducted for a more than

4 twenty-five years, there are concerns about their accuracy. One of the concerns is the difficulty associated with estimating change in production process capital expenditures. The share of manufacturing air pollution abatement capital expenditures represented by change in production process techniques increased from 17.4 percent in 1973 to 48.3 percent in 1994 according to the U.S. Department of Commerce (1976, p. 47 and 1996, p. 25). As the share of the pollution abatement capital expenditures represented by change in production process techniques increases, it becomes increasingly difficult to estimate the current account (i.e., operation and maintenance) expenditures associated with pollution abatement activities. This measurement difficulty arises because as an increasing percentage of abatement activities are imbedded in production processes, it becomes increasingly difficult to determine which operating costs are associated with pollution abatement activities. Modeling pollution abatement activities is an alternative method of estimating the costs associated with pollution abatement activities. There are two approaches to modeling pollution abatement costs. One method assumes pollution abatement activities are separable from the activities associated with producing the marketed output. Martin, Braden, and Carlson (1990) and Bellas (1998) are examples of studies that estimate pollution abatement functions by assuming pollution abatement activities are separable from electricity production. The second method models the joint production of good and bad outputs, in the sense that the bad outputs are byproducts of the production of good outputs. 2 There are some advantages to estimating pollution abatement costs by modeling the joint production of good and bad outputs. First, it does not require information about pollution abatement technologies and their associated costs. Instead, the cost of pollution abatement activities is measured by the reduced production of -2-

5 the good output that results from reducing production of the bad output. The foregone production of the good output occurs as a result of the reallocation of inputs from producing the good output to their use in pollution abatement activities. Second, modeling the joint production of good and bad outputs avoids the difficulties associated with survey efforts to estimate pollution abatement costs associated with changes in the production process. Finally, synergies among the abatement processes of two or more pollutants are automatically captured by the joint output technology. Joint production models have been used to measure the marginal abatement costs of reducing emissions from electric utilities. Turner (1995) applied the methodology developed by Färe et al. (1993) to power plant data from 1985 to 1987 in order to estimate the shadow price of SO 2 emissions. Coggins and Swinton (1996) also applied the Färe et al. (1993) methodology and estimated the marginal abatement costs of reducing sulfur dioxide emissions by fourteen power plants in Wisconsin using data from 1990 to The Coggins and Swinton study was extended by Swinton (1998) whose sample included power plants in Wisconsin, Illinois, and Minnesota for 1990 to Kolstad and Turnovsky (1998) and Carlson et al. (2000) specified joint-production cost functions to estimate the marginal abatement costs faced by electric utilities when reducing sulfur dioxide emissions. While Färe et al. (1993) and Coggins and Swinton (1996) assume fuels are homogeneous, Kolstad and Turnovsky (1998) and Carlson et al. (2000) incorporate differences in fuel quality (i.e., differences in the sulfur content) when estimating their cost functions. Kolstad and Turnovsky (1998) and Carlson et al. (2000) assume the restriction on emissions faced by each plant is binding. Hence, a plant emits the maximum amount of the -3-

6 pollutant permitted by the environmental regulation. While several studies have specified joint production models to estimate marginal abatement costs, there has been less interest in using joint production models to estimate the total cost of pollution abatement activities. 3 Thus the survey and joint production approaches have not been directly compared, which is the purpose here. This study estimates the cost of pollution abatement using the joint production approach and derives the price of electricity that would prevail if that cost of abatement were equal to the survey approach estimate, providing evidence concerning the consistency of the two approaches. Our approach models the production of good outputs (i.e., marketed goods) and bad outputs (i.e., emissions of air pollutants) within a data envelopment analysis (DEA) framework (see Färe and Grosskopf, 1983). The original Färe and Grosskopf methodology measured the costs of pollution abatement activities when the producer is restricted to maintaining its observed mix of the good output and the bad output, which we modify in this paper. Färe, Grosskopf, and Pasurka (1986) applied the Färe and Grosskopf (1983) framework to a cross section of data of 100 steam power plants in the United States for They specified particulate matter, sulfur dioxide, nitrogen oxides, heat discharge in water used by plant as the undesirable outputs and found a 1.3 percent reduction in the production of the desirable output as a result of the undesirable outputs not being freely disposable. In their study, Färe, Grosskopf, and Pasurka (1986) did not control for fuel quality (i.e., sulfur and ash content). Färe et al. (1989) proposed an alternative methodology to measure the costs of pollution abatement activities when the producer adopts a production process that allows an equiproportional increase in the good output and decrease in the bad output relative to the -4-

7 observed production levels. When estimating pollution abatement costs with a joint production model, we distinguish between two technologies. The free disposability or unregulated technology assumes the bad output can be thrown away at no cost to the producer, whereas the weak disposability or regulated technology allows for reductions in the production of the bad output via a proportional decrease in the good output. Within this framework, pollution abatement costs are determined by computing the difference between the maximum production of the good output under the unregulated and regulated technologies. Since the unregulated and regulated production possibilities frontiers are constructed from data that reflect the actual behavior of producers, the cost estimates generated by the DEA framework can be viewed as the revealed costs (i.e., lost revenue) of pollution abatement activities. The electric utility industry represents a unique case in which plant-level data for inputs, the good output, and the bad outputs are publically available. For each power plant included in this study, its pollution abatement costs reported on the Form EIA-767 survey are compared with its costs of pollution abatement activities estimated by modeling the joint production of the good and bad outputs within the DEA framework. Linear programming (LP) problems are specified in order to estimate the measurable pollution abatement costs for a panel data set of coal-fired power plants from 1994 and This study represents the first attempt to compare estimates of pollution abatement costs from a survey with pollution abatement costs estimated by a modeling approach and it allows us to determine the extent of any divergence between the survey and modeling estimates and the source(s) of any divergence. The remainder of this study is organized in the following manner. -5-

8 In Section II, a review of surveys of pollution abatement expenditures by electric utilities is presented. In Section III, the joint production model and the associated linear programming (LP) programs are specified. In Section IV, the data and results are presented. Finally, Section V summarizes this study, discusses future avenues of research, and examines the implications of the empirical results of this study. 4 II. Survey Estimates of Pollution Abatement Expenditures by Electric Utilities The Federal Power Commission (FPC) Form 67 entitled Steam-Electric Plant Air and Water Quality Control Data collected information about the operating costs associated with the pollution abatement activities of power plants for 1969 through These data were published in a series of annual reports by the U.S. Federal Power Commission for 1969 to 1973 and the Federal Energy Regulatory Commission for 1974 to 1976 (Appendix A lists these reports). The EIA-767 survey Steam-Electric Plant Operation and Design Report is the successor to FPC Form Although Form EIA-767 was administered during 1981 to 1984, the Energy Information Administration (EIA) does not consider these data to be as accurate as the data starting in In its annual report on pollution abatement expenditures, the Bureau of Economic Analysis (see Vogan, 1996) used data collected by FPC Form 67 to estimate the costs associated with the operation of air and water pollution abatement capital equipment of privately owned electric utilities for the years from 1972 through 1980, and data from the EIA-767 survey were used to estimate costs for the years from 1985 through 1994 (Farber and Rutledge 1989, pp and 16 and Vogan 1996, p. 54). Changes in related series of data were used to generate -6-

9 estimates for the years from 1981 to This study investigates the relationship between the EIA-767 survey estimates of O&M expenditures associated with abating particulate and sulfur emissions and modeling estimates of pollution abatement costs. 7 Throughout the remainder of this study, we refer to survey estimates of pollution abatement expenditures as PACS and modeling estimates of pollution abatement costs as PACM. In the next section, we introduce the theoretical model of the joint production of good and bad outputs, which underpins our empirical work. III. Modeling Pollution Abatement Costs The opportunity cost of pollution abatement activities is the foregone production of the good output resulting from the reallocation of inputs from producing the good output to pollution abatement activities. In this section, a formal model of pollution abatement costs is developed from a model of the joint production of good and bad outputs. In this study, the cost of pollution abatement activities is the value of lost potential output due to regulation. This is the cost which we will compare to the estimates of pollution abatement costs from the EIA-767 survey in order to provide an indication of the accuracy of such surveys. To derive pollution abatement costs and show that it can be interpreted as the value of lost potential output we formulate two production models, one regulated and one unregulated. In the regulated model, we explicitly recognize that good and bad outputs are jointly produced and that the bad outputs cannot be disposed of freely. On the other hand, in the unregulated model we allow bad (and good) outputs to be freely disposable. In measuring the potential output loss, we differ from Färe, Grosskopf, and Pasurka -7-

10 (1986) by not scaling all outputs including the bads, but rather scaling only on the good output. Another difference is that we use an additive directional distance function rather than a multiplicative Shephard distance function. To model the abatement cost we introduce the required production model. Denoting inputs by x = (x 1,..., x N ) 0 ú N + and outputs by y = (y 1,..., y M ) 0 ú +M, the output sets are given by (1) P(x) = {y: x can produce y}, x 0 ú N + We distinguish between good or desirable outputs y g = (y 1g,..., y Gg ) and bad or undesirable outputs y b = (y 1b,..., y Bb ), so that y = (y g, y b ) 0 ú +M. Emissions of sulfur dioxide (SO 2 ) and particulate matter less than ten microns in diameter (PM-10), which are the bad outputs, are undesirable byproducts of producing the good output - kilowatt-hours (kwh) - and therefore will be modeled as such. In particular we say that the good and bad outputs are nulljoint or byproducts if (2) (y g, y b ) 0 P(x) and y b = 0 imply y g = 0. Equation (2) means that no bad outputs are produced (y b =0) only if none of the good outputs are produced (y g =0). Equivalently, if some good outputs are produced then some bad outputs must also be produced. We impose this assumption on our regulated model, and note that if good output is produced, then some of the bad (byproducts) output is also produced. Moreover, in our regulated model we assume that outputs (y g, y b ) are weakly disposable, i.e., (3) y = (y g, y b ) 0 P(x), 0 # θ # 1 imply (θy g, θy b ) 0 P(x) This assumption states that proportional reduction of good and bad outputs is feasible, but reduction of bads alone may not be. -8-

11 In addition to assumptions (2) and (3) we impose standard properties on P(x), including: inputs and good outputs are freely disposable and P(x) is a compact, convex set (see Färe and Primont, 1995, for details). Prior to formally showing how to calculate the output loss due to regulation, we provide some intuition based on a simple diagram. In Figure 1, the regulated output set, P R (x), is bounded by the line segments 0abcd0. This output set has the properties that good and bad outputs are weakly disposable and nulljoint. The unregulated output set, P U (x), is bounded by 0ebcd0, and includes the regulated technology in our example as a proper subset. To measure the potential output loss, i.e., the difference in the two output sets, first an observation (y g, y b ) (point A in Figure 1) is projected to the boundary (point B in Figure 1) by scaling good output. The distance AB represents the reduced production of the good output resulting from technical inefficiency. Hence, this producer could increase production of its good output without increasing production of its bad output. The downward sloping segment of the frontier - bc - represents the possibility that a producer can simultaneously increase production of the good output and reduce production of the bad output. While not all frontiers have this downward sloping segment, there are two possible explanations for why we might observe this counter-intuitive result. First, observation c may represent an older technology than the other observations used to construct the frontier. While the model assumes a frontier is constructed with observations with access to similar technologies, this is not always the case. Second, observation c may represent an outlier due to measurement error. -9-

12 g y e a C B A b c 0 d b y Figure 1. Measure of Potential Output Loss -10-

13 In this study, costs associated with technical efficiency are not included in PACM. Here we assume that technical inefficiency, which is represented by the distance between an observation and the weak disposability frontier, occurs for reasons unrelated to pollution abatement activities. Hence, this study defines PACM as the difference between the production of the good output when the bad output is unregulated and the production of the good output when the bad output is regulated. In our figure, the distance between the two output sets - here BC - gives us the potential loss due to regulation. Again, we only expand the good output. Assuming that we have k = 1,..., K observations of inputs x k, fuel quality q k, and outputs y k, we may formulate the output sets as an activity analysis or Data Envelopment Analysis (DEA) model. The regulated model is R g b g g ( 4) P ( x) = {( y, y ): z y y m= 1,..., G K k = 1 K k = 1 K k = 1 K k = 1 K k = 1 k km b b zy = y i = 1,..., B k ki zx x n= 1,..., N k kn n zq = q j= 1,..., J z = 1 k = 1,..., K z 0 k = 1,..., K} k i k kj j k m -11-

14 The intensity variables, z k, are the weights assigned to each observation when constructing the production set (i.e., the production possibilities set). The inequality constraints in (4) on the good outputs, y mg, m=1,..., G imply that these outputs are freely disposable. 8 Together with the equality constraints in (4) on the bad outputs (y ib, i=1,..., B), good outputs and bad outputs are weakly disposable, i.e., they can be scaled down jointly to zero and hence they satisfy (3). The equality constraint on the undesirable qualities of the fuels consumed (q k ) specifies that the undesirable qualities of the fuel consumed by the reference technology must equal the undesirable qualities of the fuels consumed by the observation. This model satisfies the assumption of good and bad outputs being nulljoint provided K b ( 5) ( a) y > 0 i = 1,..., B k = 1 ki B ( b) y b ki > 0 k = 1,..., K i= 1 Condition (5a) states that every bad output is produced by some plant k, and (5b) states that every plant k produces at least one bad output. To further illustrate null-jointness, assume that b each y i = 0 in the expression of the output set (4). Then, due to (5) each intensity variable z k g must be zero, implying that each good output y m must be zero. In addition, the output correspondence (4) models variable returns to scale since the intensity variables sum to unity. That is, it allows for increasing, constant, and decreasing returns to scale. The unregulated model is obtained from (4) by allowing for the free disposability of bad outputs, i.e., by changing the i = 1,..., B equalities to inequalities. -12-

15 U g b g g () 6 P () x = {( y, y ): z y y m= 1,..., G K k = 1 K k = 1 K k = 1 K k = 1 K k = 1 k km b b zy y i = 1,..., B k ki zx x n= 1,..., N k kn n zq = q j= 1,..., J z = 1 k = 1,..., K z 0 k = 1,..., K} k i k kj j k m To measure the output loss due to regulation we apply a directional distance function, in G particular we choose a directional vector d 0 ú + to be d= (1,...,1) then for some observation (x kn, y kn ) we compute r R k k g b R k ( 7) D ( y, x ; 1) = max{( y + β 1, y ) P ( x )} k k In our case with one good output the efficient output relative to the regulated technology is g R k k () 8 y + D r ( y, x ;) 1 k -13-

16 y k g corresponds to point A in Figure 1. r R k k D ( y, x ; 1) corresponds to AB, and the sum g y k r D ( y, x ; 1) R k k of and corresponds to the production of the good output represented by point B. The corresponding directional distance function of the unregulated technology is r U k k g b U k ( 9) D ( y, x ; 1) = max{( y + β 1, y ) P ( x )} k k and the efficient output relative to the unregulated technology is g U k k ( 10) y + D r ( y, x ; 1) k r D ( y, x g ; 1) y k U k k where corresponds to AC, and the sum of and r U k k D ( y, x ; 1) corresponds to the production of the good output represented by point C. The revenue loss due to regulation is r r ( 11) p k y D ( y, x ; 1) p y D ( y, x ; 1) g U k k g R k k ( + ) ( + ) k k or r r k U k k R k k ( 12) PACM = p D ( y, x ; 1) D ( y, x ; 1) [ ] -14-

17 where p kn is the observed price (i.e., revenue per kwh) of the good output for producer kn. The difference inside the square brackets in (12) corresponds to the distance (BC) in Figure 1, which is our estimate of the loss in output due to regulation. We may compute the total loss of potential revenue by summing (12) over all kn: r r k U k k k R k k ( 13) ΣPACM = p D ( y, x ; 1) p D ( y, x ; 1) k k For feasible output vectors, the directional distance function is greater than or equal to zero. It equals zero if and only if the observation vector (x kn, y kn ) is on the production possibilities frontier (i.e., the observation vector is technically efficient), while a point inside the production frontier has a value greater than zero. Hence, the value of the directional distance function represents the expansion of the good output required to project an observation (x kn, y kn ) from inside the production frontier to the frontier. Next, we show how we use our estimate of lost revenue to provide a comparison to the survey estimates of pollution abatement costs. We proceed by setting the lost revenue (equation 12) equal to the PACS incurred by producer kn. Then we can solve for the implied price per kwh for kn ( 14) p$ k = k c r r U k k R k k D ( y, x ; 1) D ( y, x ; 1) -15-

18 where c kn is the PACS for producer kn. The price $p k estimates the revenue per kwh required for the value of the reduced production of the good output derived from the modeling method (i.e., PACM) to equal PACS. There are two ways to compute the mean of (14). We may compute the average of the $p k by summing (14) over all kn and dividing by the number of power plants or we can calculate the following: ( 15) p = k c k r r U k k R k k D ( y, x ; 1) D ( y, x ; 1) k k The directional distance functions can be calculated as solutions to LP problems. In order to determine PACM, two LP problems must be solved for each producer. When the bad output is regulated, the LP problems impose weak disposability. As an example, we have for observation kn: -16-

19 r ( 16) D ( x, y ) = maxβ g g k st.. z y y + β m= 1,..., G k = 1 K k = 1 K k = 1 K k = 1 K k = 1 z K k R k k k k km km b b zy = y i = 1,..., B k ki ki zx x n= 1,..., N k kn k n zq = q j= 1,..., J k kj k j z = 1 k = 1,..., K k 0 k = 1,..., K The weak disposability reference technology relative to which (x kn, y kn ) is evaluated is constructed from the observed production processes, i.e., the constraints are consistent with P R (x) in (4). The solution to this LP problem gives the distance AB in Figure 1. The value of the objective function represents the difference between the observed production of the good output and the maximum potential production of the good output for a given input vector and technology. The first constraint of the LP problem represents the constraint imposed on the good output. There is a separate constraint for each of the G good outputs of producer kn. The right- -17-

20 hand side of the constraint represents the actual production of the good outputs for producer kn. The left-hand side represents the production of the good output of the theoretical efficient producer. The greater than or equal to sign imposes the restriction that the production of good outputs by the theoretical producer must be greater than or equal to the observed production of the good output of producer kn. The second constraint of the LP problem represents the constraint imposed on the bad output. There is a separate constraint for each of the B bad outputs produced by producer kn. The equality sign associated with the constraint on the bad outputs imposes weak disposability on the bad outputs. The right-hand side of the constraint represents the observed generation of the bad outputs of producer kn. The left-hand side represents the level of the bad output generated by the theoretical efficient producer. The difference between the LP problems for the regulated and unregulated technologies are the constraints associated with bad outputs. The equal to sign imposes the assumption of weak disposability on the bad outputs. For the unregulated technology, the constraint is written as less than or equal to. Since β kn is excluded from the constraints associated with the bad outputs, the decline in production of the good output associated with environmental regulations assumes production of the bad output remains at its observed level. The third constraint of the LP problem represents the constraint imposed on input use. There is a separate constraint for each of the N inputs employed by a producer. The right-hand side of the constraint represents the observed input use of producer kn. The left-hand side represents the inputs employed by the theoretical efficient producer. The inequality sign means the theoretical producer cannot employ more inputs than producer kn. -18-

21 The fourth constraint of the LP problem represents the constraint imposed on the undesirable qualities of the fuels consumed by producer kn. There is a separate constraint for each of the J undesirable attributes of the fuels. The undesirable qualities of the fuels are the sulfur content of coal and oil and the ash content of coal. A higher sulfur or ash content of a fuel represents more undesirable attributes of that fuel. The right-hand side of the constraint represents the observed quality of the fuel consumed by producer kn. The left-hand side represents the undesirable quality of the fuel consumed by the theoretical efficient producer. The equality sign means the undesirable qualities of the fuel consumed by the theoretical producer must equal those of the fuels consumed by producer kn. A non-negativity constraint is imposed on the z k. The z k are the weights assigned to each of the available production processes when constructing the production frontier. Since the summation of the intensity parameters (i.e., the z k ) is constrained to equal unity, variable returns to scale is assumed for all of the LP problems. 9 IV. Data and Results The technology modeled in this study consists of one good output, net electrical generation (kwh), and two bad outputs - emissions of sulfur dioxide (SO 2 ) and particulate matter less than ten microns in diameter (PM-10). 10 The inputs consist of the capital stock, the number of employees, and the heat content (in Btu) of the coal, oil, and natural gas consumed at the plant. Undesirable fuel qualities consist of the ash content of coal and the sulfur content of coal and oil. Carlson et al. (2000, pp ) discusses the derivation of the estimates of the capital stock and number of employees. The Form EIA-767 survey is the source of -19-

22 information about fuel consumption, fuel quality, and net generation of electricity. The U.S. EPA is the source of emission estimates for PM-10 and SO 2. In order to model a homogeneous production technology, the sample consists of 237 power plants for 1994 and 232 power plants for Although a power plant may consume coal, oil, or natural gas, coal must provide at least 95 percent of the Btu of fuels consumed by it. 11 Table 1 presents summary statistics of the data and Appendix A contains a detailed discussion of the data. The Form EIA- 861 survey provides information on sales of electricity and its associated revenue from sales to ultimate consumers and sales for resale by each utility. In this study, the revenue per kwh is identical for each power plant operated by a utility. When a power plant is owned by more than one utility, it is assigned the revenue per kwh of its principal owner. The EIA-767 survey requests information on operation and maintenance (O&M) expenditures associated with both the collection and disposal of fly ash, bottom ash, and flue gas desulfurization (FGD). Hence, six categories of expenditures in the EIA-767 survey are relevant for this study. For the purposes of the PACS estimates used in this study, a nonresponse or a response of estimate not available is treated as a zero. The instructions for the EIA-767 survey (U.S. Department of Energy, 2001a, Plant Information -- Financial Information) state that operation and maintenance (O&M) expenditures... should exclude depreciation expense, cost of electricity consumed, and fuel differential expense (i.e., extra costs of cleaner, thus more expense fuel). 12 Appendix B contains a discussion of BEA s use of the EIA-767 and how it estimated the costs associated with consuming cleaner fuels. Collection activities can be viewed as internal pollution abatement activities, while disposal activities can be viewed as external pollution abatement activities. Only expenditures associated with collection activities are -20-

23 included in the PACS-1 estimates reported in this study, whereas PACS-2 includes expenditures associated with collection and disposal activities. While Yaisawarng and Klein (1994) interpret the sulfur content of fuels as an bad input, we view the sulfur content as a quality of the fuel accounted for by the model. Accounting for the sulfur and ash content of the fuel allows us to model the fuels as a homogeneous inputs. By assuming no change in the sulfur and ash content of the coal and oil consumed and no change in the ash content of the coal consumed by the power plant, we exclude the costs associated with switching to fuels with fewer undesirable qualities (e.g., coal with a lower sulfur level). Since the estimates of pollution abatement costs reported in the EIA-767 survey exclude the costs associated with fuel switching, the constraint on the ash and sulfur content of the fuels forces the reference technology to consume the same quality of fuel as the observation. This allows us to focus solely on comparing the estimates from our model with the stated costs of environmental protection activities reported in the EIA-767 survey. 13 Separate LP problems are solved for each coal-fired power plants in 1994 and Table 2 presents results for each power plant in 1995 and Appendix C reports the results for Column (1) lists the reduced production of electricity (in kwh) which is the following r r U k k R k k component of equation (12): D ( y, x ; 1) D ( y, x ; 1). Column (2) lists the p kn [ ] observed for each power plant. Column (3), which is calculated using equation (12), is the product of columns (1) and (2). Column (4) is the ratio of the reduced production of electricity, which is reported in column (1), to the observed production of electricity. Column (5) reports PACS-1, which is estimate of PACS for producer kn - c kn - which includes only collection -21-

24 expenditures. Column (6), which is estimated using equation 14, lists the estimated price $p k associated with column (5). Column (7) reports PACS-2, c kn, which includes collection and disposal expenditures. Column (8), which is estimated using equation 14, lists the estimated price $p k associated with column (7). The results in Table 2 show the total ΣPACM estimates exceed the ΣPACS estimates. 14 For 1995, ΣPACM is $2,364 million, while ΣPACS-1 is $524 million and ΣPACS-2 is $697 million. In 1994, ΣPACM is $2,538 million, while ΣPACS-1 is $360 million and ΣPACS-2 is $518 million. If the 10 power plants with the highest ΣPACM (i.e., lost output in excess of $75 million) in 1995 are excluded, ΣPACM declines to $983 million. If the 10 power plants with the highest ΣPACM (i.e., lost output in excess of $100 million) in 1994 are excluded, ΣPACM declines to $774 million. Finally, it is worth noting that the reduced production of electricity (in KWh) associated due to environmental regulations is 4.22 percent in 1994 and 3.87 percent in 1995 of the observed electric generation of all power plants in our sample. The finding that ΣPACM exceeds ΣPACS is surprising for several reasons. Five factors lead to the expectation that the PACS estimates would exceed the PACM estimates. First, respondents might have an incentive to overstate the costs associated with pollution abatement activities. 15 Second, respondent to the EIA-767 survey may perceive environmental regulations as more binding than the joint production model used to generate the PACM estimates. Third, the technology specified in this study is assumed to be noncumulative (i.e., the technology available to a producer consists solely of the processes used in that year). Since pollution abatement activities have been undertaken by power plants for several decades (see -22-

25 U.S. Department of Commerce, 1982), the unregulated technology based solely on data from 1994 or 1995 is unlikely to represent the true unregulated technology. If a process (i.e., observation) from an earlier period allows a power plant to produce more electricity than can be produced with the same input vector in period t, then the true unregulated technology is not accurately modeled. Instead of an unregulated technology, it is more accurate to depict it as the least regulated technology available in the current year. The consequence of the failure to depict the true unregulated technology is a downward bias in the revealed estimates of measurable pollution abatement costs generated by the data used in this study. Fourth, if a power plant operates a pollution abatement device (e.g., a scrubber) and the plant produces more of the desirable output with a given input vector than any other plant, the DEA model will determine there are no pollution abatement costs - PACM - even though PACS reports expenditures associated with the operation of the pollution abatement device. Since some of the O&M disposal expenditures in the EIA-767 survey may represent external pollution abatement activities and expenditures for materials not included as inputs in the production technology modeled in this study, the PACS estimates may exceed the PACM estimates. 16 However, there are several explanations for the finding that PACM is greater than PACS. One explanation is associated with the expenditure categories in the EIA-767 survey. The PACM estimates may capture opportunity costs of pollution abatement activities excluded from the PACS estimates (e.g., paperwork costs associated with environmental regulations). A second explanation is the PACM estimates include the costs of electricity consumption associated with pollution abatement activities, while the EIA-767 survey excludes the cost of electricity associated with pollution abatement activities. 17 Since pollution abatement activities -23-

26 are one of the uses of the electricity consumed at the plant, some of the fuels consumed and the labor employed by the plant are used to generate the electricity consumed for pollution abatement activities. As a result, PACM estimates include the costs of electricity consumed for pollution abatement activities. A third explanation is respondents to the EIA-767 survey may perceive environmental regulations as less binding constraints than the DEA model used to generate the PACM estimates. The specification of the regulated and unregulated technologies reflect assumptions about how to determine the costs associated with pollution abatement activities. When answering the EIA-767 survey, the respondents may perceive a different baseline technology than the unregulated technology specified by the DEA methodology used to derive the PACM estimates. Alternatively, lower PACS estimates may reflect the perception of respondents that the options available to electric utilities in an unregulated world are more limited than assumed by economic models. A fourth explanation for the discrepancy is the treatment of nonreponses to questions regarding O&M expenditures for pollution abatement activities associated with reducing sulfur dioxide and PM-10 emissions. Do respondents perceive no O&M expenditures or are these instances of respondents failing to report O&M expenditures when in fact there are pollution abatement activities? The electronic files containing the results of the EIA-767 survey do not indicate whether the zeros represent nonresponses or zeros on the actual survey form. Those cases in which nonreponses mask pollution abatement expenditures provide a downward bias to the estimates from the EIA-767 survey. 18 A fifth explanation for why PACM estimates exceed the PACS estimates is the PACM -24-

27 estimates may be influenced by outliers in the sample which creates an upward bias in the PACM estimates. There are two ways to address this concern. A simple approach is to eliminate a certain percentage of the outliers. Although there is no statistical theory justifying such a procedure, it provides insights into the effect of outliers on the results. A more sophisticated approach is using a bootstrap technique, which tests the sensitivity of the results to outliers in the data, to add a stochastic element to the analysis. Finally, the regulated technology specified in this study is valid if producers are engaged in pollution abatement activities. If the free disposability is the correct technology, then the observations used to construct the regulated technology are simply inefficient producers relative to the unregulated production frontier. In this case observations used to construct the regulated frontier are in fact inefficient, and the PACM estimates are biased in an upward direction. The accuracy of the results of the modeling approach can be validated in two ways. First, the data can be used to estimate the marginal abatement cost of reducing a ton of SO 2 emissions. Since previous modeling efforts have yielded reasonable estimates of the marginal abatement costs of reducing SO 2 emissions, these calculations would indicate if the data and model used in this study yield atypical results. Since the EIA-767 survey provides data on the sulfur content of the coal and oil, it is possible to implement a materials balance analysis of sulfur in order to determine the average cost of abating a ton of sulfur emissions as a second method of validating the results of this study. This calculation would provide insights into whether the data and model yield reasonable estimates of the average cost of abating SO 2 emissions. -25-

28 V. Conclusions This study investigated the relationship between stated cost estimates of pollution abatement activities and the costs of pollution abatement activities revealed by the actual behavior of the regulated entities through a comparison of PACS and PACM estimates for U.S. coal-fired power plants. The latter views the costs of pollution abatement activities as the value of the reduced production of the good output due to environmental regulations. This alternative method is based on a DEA model, which allows us to model joint production with and without regulations and estimate pollution abatement costs as the difference in production in the two models. We compare these estimates with the survey estimates of the pollution abatement costs borne by power plants in 1994 and In estimating pollution abatement costs using our DEA approach, we model the unregulated and regulated technologies using notions of free and weak disposability, respectively. Hence, the joint production model represents an example of the advantage of establishing the link between pollution abatement costs and production technologies. This study illustrates the potential of using a joint production model to assess the costs of reducing air pollutants emitted into the atmosphere. This model could be estimated parametrically-- either a parametric cost or distance function can be specified and estimated as a frontier model (see for example Färe et al., 1993). This involves estimating one regulated and one unregulated function for all observations. The use of joint production models to estimate the costs associated with pollution abatement activities follows in the tradition of using economic models to estimate the costs of regulations. The costs now depend on the specification of the production technology (i.e., the -26-

29 functional form and the associated elasticities of substitution) which is comparable to efforts that estimate the costs of other types of economic policies. In fact, the joint production of good and bad outputs has been specified in CGE models such as Jensen and Rasmussen (2000) to estimate the costs associated with proposed reductions in CO 2 emissions. We believe production models provide a useful complement to survey methods used to identify pollution abatement costs. If internal pollution abatement activities consist primarily of end-of-pipe technologies, then surveys should provide an adequate means of estimating the costs of these activities. However, as an increasing share of the internal activities associated with abating air pollutants involve integrated technologies, surveys become an exercise in stated costs. In that case, economic models, which are more closely tied to production theory, represent a means of estimating the costs associated with pollution abatement activities. Since the EIA-767 survey excludes expenditures associated with fuel switching, the expenditures reported in the EIA-767 survey are associated with end-of-pipe pollution abatement activities. Survey estimates of the costs of these activities are likely to be more accurate than cost estimates associated with change in production process abatement techniques. Hence, the divergence between the stated and revealed costs estimates reported in this study should be smaller than a study comparing model estimates of pollution abatement costs with survey estimates of the costs associated with change in process abatement technologies. Future investigations using the joint production model specified in this study might include additional bad outputs, incorporate the revenue from the sale of byproducts, and expand the sample to include observations from earlier years in order to obtain a more accurate estimate of the unregulated technology. -27-

30 Although this study is concerned with the costs of pollution abatement activities, it is possible to speculate on whether the results of this study are relevant to the discussion about the stated vs. revealed methods used to estimate the benefits of environmental controls. It seems reasonable to assume the individuals responsible for completing the EIA-767 survey are more familiar with the costs of pollution abatement activities than the typical respondent to a contingent valuation survey. Hence, the divergence between the state and revealed costs of this study is likely to be less than the divergence found by a comparable study of benefits. -28-

31 References Bellas, Allen S. (1998), Empirical Evidence of Advances in Scrubber Technology, Resource and Energy Economics, 20, No. 4 (December), Brännland, Runar, Rolf Färe, and Shawna Grosskopf (1995), Environmental Regulation and Profitability: An Application to Swedish Pulp and Paper Mills, Environmental and Resource Economics, 6, No. 1, Carlson, Curtis, and Dallas Burtraw, Maureen Cropper, and Karen Palmer (2000), Sulfur Dioxide Control by Electric Utilities: What are the Gains from Trade? Journal of Political Economy, 108, No. 6 (December), Coggins, Jay and John Swinton (1996), The Price of Pollution: A Dual Approach to Valuing SO 2 Allowances, Journal of Environmental Economics and Management, 30, No.1 (January), Farber, Kitt and Gary Rutledge (1989), Pollution Abatement and Control Expenditures: Methods and Sources for Current-Dollar Estimates, mimeo. Färe, Rolf and Shawna Grosskopf (1983), Measuring Output Efficiency, European Journal of Operational Research, 13, Färe, Rolf, Shawna Grosskopf, C.A. Knox Lovell and Carl Pasurka (1989), Multilateral Productivity Comparisons When Some Outputs are Undesirable: A Nonparametric Approach, Review of Economics and Statistics, LXXI, No. 1 (February), Färe, Rolf, Shawna Grosskopf, C.A. Knox Lovell, and Suthathip Yaisawarng (1993), Derivation of Shadow Prices for Undesirable Outputs: A Distance Function Approach, Review of Economics and Statistics, 75, No. 2 (May), Färe, Rolf, Shawna Grosskopf, and Carl Pasurka (1986), Effects on Relative Efficiency in Electric Power Generation Due to Environmental Controls, Resources and Energy, 8, No. 2 (June), Färe, Rolf and Daniel Primont (1995), Multi-Output and Duality: Theory and Application, Boston: Kluwer-Nijhoff Publishing. Gollop, Frank M. and Mark J. Roberts (1983), Environmental Regulations and Productivity Growth: The Case of Fossil-fueled Electric Power Generation, Journal of Political Economy, 91, No. 4,

32 Jensen, Jesper and Tobias N. Rasmussen (2000), Allocation of CO2 Emissions Permits: A General Equilibrium Analysis of Policy Instruments, Journal of Environmental Economics and Management, 40, No. 2 (September), Kolstad, Charles D. and Michelle H.L. Turnovsky (1998), Cost Functions and Nonlinear Process: Estimating a Technology with Quality-Differentiated Inputs, Review of Economics and Statistics, 80, No. 3 (August) Martin, David W., John B. Braden, and J. Lon Carlson (1990), Estimation of Process Change for Industrial Pollution Abatement, Journal of the Air and Waste Management Association, 40, Streitwieser, Mary L. (1996), Evaluation and Use of the Pollution Abatement Costs and Expenditures Survey Micro Data, Center for Economic Studies, CES 96-1, ( Streitwieser, Mary L. (1997), Using the Pollution Abatement Costs and Expenditures Micro Data for Descriptive and Analytic Research, Journal of Economic and Social Measurement, 23, No. 1, Swinton, John B. (1998), At What Cost Do We Reduce Pollution? Shadow Prices of SO 2 Emissions, Energy Journal, 19, No. 4, Tran, Ngoc-Bich and V. Kerry Smith (1983), The Role of Air and Water Residuals for Steam Electric Power Generation, Journal of Environmental Economics and Management, 10, No. 1 (March), Turner, Judi A. (1995), Measuring the Cost of Pollution Abatement in the U.S. Electric Utility Industry: A Production Frontier Approach, Ph.D. Dissertation, University of North Carolina, Chapel Hill, NC. United Nations (1993), Handbook of National Accounting, Studies in Methods, Series F, No. 61, Integrated Environmental and Economic Accounting, Interim Version, United Nations: New York. (ST/ESA/STAT/SER.F/61, UN publication, Sales No. E.93.XVII.12). U.S. Department of Commerce, Bureau of Economic Analysis (1982), Stock of Plant and Equipment for Air and Water Pollution Abatement in the United States, , Survey of Current Business, 62, No. 11 (November), U.S. Department of Commerce, Bureau of Economic Analysis (1994), Integrated and Environmental Satellite Accounts, Survey of Current Business, 74, No. 4 (April), (see NIPA Related Articles for PDF and HTML versions of the article at: ) -30-

33 U.S. Department of Commerce, Bureau of the Census (1976), Pollution Abatement Costs and Expenditures: 1973, Current Industrial Reports, MA200, Washington, D.C.: U.S. Government Printing Office. U.S. Department of Commerce, Bureau of the Census (1996), Pollution Abatement Costs and Expenditures: 1994, Current Industrial Reports, MA200, Washington, D.C.: U.S. Government Printing Office. ( U.S. Department of Energy, Energy Informational Administration (2001), FORM EIA-767 Steam-Electric Plant Operation and Design Report, and its accompanying instructions are located at: ( U.S. Department of Energy, Energy Informational Administration (2001), FORM EIA-861 Annual Electric Utility Report, and its accompanying instructions are located at:( U.S. Department of Energy, Energy Informational Administration (1999), information collected by FORM EIA-861 Annual Report of Public Utilities is located at: U.S. Department of Energy, Energy Informational Administration (1999), Electric Power Annual (Volume II). ( Vogan, Christine (1996), Pollution Abatement and Control Expenditures, , Survey of Current Business, 76, No. 9 (September), (See NIPA Related Articles at: for PDF and HTML versions of the article) Yaisawarng, Suthathip and J. Douglass Klein (1994), The Effects of Sulfur Dioxide Controls on Productivity Change in the U.S. Electric Power Industry, Review of Economics and Statistics, 76,