Market effects of environmental regulation: coal, railroads ... - CiteSeerX

Market effects of environmental regulation: coal, railroads, and the 1990 Clean Air Act∗ Meghan R. Busse†

Nathaniel O. Keohane‡

March 13, 2007

Abstract Many environmental regulations encourage the use of “clean” inputs. When the suppliers of such an input have market power, environmental regulation will affect not only the quantity of the input used, but also its price. We investigate the effect of the Title IV emissions trading program for sulfur dioxide on the market for low-sulfur coal. We find that the two railroads transporting coal were able to price discriminate on the basis of environmental regulation and geographic location. Delivered prices rose for plants in the trading program relative to other plants, and by more at plants near a low-sulfur coal source.

JEL codes: Q28, L51, L92, L94.

∗

We are particularly grateful to Erin Mansur and Florian Zettelmeyer for many comments that improved this paper. Thanks are also due to Spencer Banzhaf, Severin Borenstein, Michael Greenstone, Matthew Kotchen, Jonathan Levin, Paul Macavoy, Fiona Scott Morton, Sharon Oster, Christopher Timmins, Frank Wolak, Rob Williams, and seminar participants at Berkeley, USC, UCF, the University of Connecticut, Dartmouth, Harvard, Stanford, Yale, and the 2004 NBER Summer Institute for their comments and suggestions. We are very grateful to Catherine Wolfram and to John Bitzan who provided us with data used in this paper. Daryl Newby of the Kentucky Public Service Commission and Jim Thompson of Energy Publishers provided helpful assistance in gathering information on coal contracts. We also thank the Editor and two anonymous referees, whose comments improved the paper. † Haas School of Business, 2220 Piedmont Ave., Berkeley, CA 94720-1900; [email protected]; tel (510) 642-9589; fax (510) 643-5180. ‡ Yale School of Management, P.O. Box 208200, 135 Prospect St., New Haven, CT 06520-8200; [email protected]; tel (203) 432 6024; fax (203) 432 6974.

1

Introduction

Burning coal to generate electricity releases sulfur dioxide, a pollutant that harms human health and contributes to acid rain. Electric power plants can reduce sulfur dioxide emissions by burning lowsulfur coal rather than high-sulfur coal or by “scrubbing” their flue gases before they are released into the atmosphere. Since 1971, various forms of environmental regulation have encouraged (or mandated) power plants in the U.S. to use one of these abatement methods. Title IV of the 1990 Clean Air Act Amendments used a a novel program of emissions “allowance” trading to extend sulfur dioxide regulation to a set of power plants that had been exempted from previous federal regulation. This paper investigates how this regulatory change affected the price of low-sulfur coal from the Powder River Basin in Wyoming, the major source of low-sulfur coal in the U.S. Since coal is a commodity, one might suppose that the form of regulation should have no effect: the price of coal, before and after a change in regulation, should equal its marginal costs of extraction and transportation. This characterization turns out to be too simple. In particular, two aspects of the market for low-sulfur coal draw it away from this perfectly competitive benchmark. First, coal is transported from the Powder River Basin by only two railroads, the Burlington Northern Santa Fe (BNSF) and the Union Pacific (UP). Moreover, many power plants are served only by one of the two carriers. These facts suggest considerable potential for market power in transportation, which accounts for a large fraction of the delivered price of coal. Second, power plants vary in their willingness to pay for low-sulfur coal, both because different plants are regulated in different ways and because the geographic location of a plant determines how attractive lowsulfur coal is relative to other abatement options. This variation creates an opportunity for the railroads to price discriminate among plants on the basis of regulatory status and location. We use detailed data on shipments of Powder River Basin (PRB) coal to examine how delivered prices responded to the advent of the sulfur dioxide allowance market created by Title IV. What we find is compelling evidence of price discrimination by the railroads on the basis of environmental regulation. Controlling for a wide range of other factors, delivered prices of PRB coal rose at power plants that were affected by the new regime, relative to prices at other plants. Moreover, the relative increase in coal prices was greater at plants near to Wyoming (where few substitutes were available) than at more distant ones (where the plants had better outside options). Our identification strategy relies on the fact that only a subset of power plants – known as

1

“Table A” plants – was required to participate in the first phase of the allowance market. We measure the effect of the allowance trading regime by analyzing delivered prices before and after the allowance market took effect, comparing the difference at Table A plants with the change at plants that remained governed by conventional emissions standards. Our results imply that prices at Table A plants, relative to non-Table A plants, increased by over two dollars per ton at the power plants nearest to the PRB, but by only 9 cents per ton at the most distant plant. The comparison with non-Table A plants is important, since delivered prices at those other plants fell over the same time period (even controlling for transportation cost and other factors). When this general decline in prices is incorporated in the estimated price changes at Table A plants, we find that delivered prices rose at Table A plants within roughly 940 miles, and fell further away. Our estimates control for transportation costs, minemouth prices, and coal characteristics, while including fixed effects for power plants, coal mines, and delivery years. Our results are also robust to a variety of alternative specifications. Although we lack the necessary data to estimate the effect on railroads’ profits directly, we perform a few “back-of-the-envelope” calculations to gauge the magnitude of their gains. Those estimates suggest that the annual producer surplus enjoyed by the railroads on deliveries to Table A plants roughly tripled under the new regulation. These gains were on the order of fifteen percent of the total value of the “regulatory rents” created by the new market, where those “rents” are defined as the difference between the market value of the pollution allowances (given away freely to electric utilities by the government) and the variable costs of emissions reductions. Other studies have noted that the delivered price of PRB coal fell during the mid-1990s, as the allowance trading regime was getting underway (Ellerman and Montero, 1998; Kunce, Hamilton, and Gerking 2005; Winston, Dennis, and Maheshri 2004). Indeed, this fall in prices has been cited as a reason that the market prices of sulfur dioxide allowances were much lower than expected (Burtraw, 1996). We shall have more to say in Section 6 about how our findings relate to these other analyses. For now, we note that our conclusions are not inconsistent with a general decline in prices. Rather, we are interested in the effects of the allowance trading program at those plants that were directly affected by the new regime, controlling for contemporaneous changes in transport costs and a range of other factors. Our findings highlight an often overlooked consequence of market-based environmental policies: namely, that the form of emissions regulation may affect the prices of “clean” inputs, such as low-sulfur coal. The familiar argument in favor of emissions trading emphasizes cost-effectiveness: 2

such a program can achieve a given amount of abatement at a lower total cost than emissions standards, since an allowance market equalizes marginal abatement costs across polluting sources. We focus here on the interaction between environmental regulation and input markets. The form of emissions regulation determines not only how much of a clean input (like low-sulfur coal) polluters want to consume, but also how much they are willing to pay for it on the margin. When the suppliers of clean inputs have market power, therefore, regulation will not simply increase the quantity consumed, but rather will affect price and quantity jointly. In turn, by influencing input prices, the form of regulation can affect the cost of abating emissions. In the case of Title IV, our results suggest that railroads effectively practiced price discrimination on the basis of regulatory status and distance from the PRB. Higher prices for low-sulfur coal at Table A plants were one consequence. The next section gives an overview of the relevant environmental regulation, along with some background on railroads and low-sulfur coal mines. Section 3 uses a simple bilateral bargaining framework to consider how geography, environmental regulation, and market power in transportation interact to determine delivered coal prices. We describe the data in Section 4, and conduct preliminary analyses in Section 5 to validate our theoretical framework and empirical approach. We present our main empirical results in Section 6. Section 7 concludes.

2

Context

Regulation of sulfur dioxide emissions Title IV of the 1990 Clean Air Act Amendments introduced a novel market-based policy to control sulfur dioxide (SO2 ) emissions from fossil-fueled generating units at electric power plants.1 Each generating unit in the program is allocated a number of “allowances,” each corresponding to one ton of SO2 emissions. If the total cap on emissions is binding (as it was in the case of Title IV, which required participating plants to cut their sulfur dioxide emissions by roughly 40 percent in the first five years), allowances have a positive price. A unit that emits more sulfur dioxide than its allowance allocation can buy permits from other generating units. A generator that emits less than its allocation, either by using low-sulfur coal or by installing and operating a flue-gas desulfurization device (better known as a “scrubber”), may sell its surplus allowances or bank them for future use or sale. 1

Ellerman et al. (2000) provide a comprehensive analysis of the allowance trading program.

3

Prior to Title IV, power plants were governed by “command-and-control” style regulations that imposed maximum allowable SO2 emission rates on individual generating units. These regulations varied in stringency depending on the age of a plant and its location. Units built after 1971 were subject to stringent uniform emissions standards imposed by the federal government under the 1970 Clean Air Act. Older units had to meet standards prescribed by the states in order to maintain local ambient air quality; these varied widely among and even within states. Our analysis focuses on the years 1990 to 1999, encompassing the first phase of the allowance market created by Title IV and the five years leading up to it.2 During that decade, environmental regulations divided power plants sharply into two groups. Of particular interest is the set of “Table A plants” – plants housing generating units3 that were required to participate in Phase I of the new allowance market.4 These were the largest, dirtiest generating units, built in the 1950s and 1960s, before the federal standards took effect. (They are called “Table A” units after the table that listed them in the 1990 legislation.) These plants experienced a shift in regulation over the study period. In the first part of the decade, they were governed by state-level emissions standards. Starting in 1995, they came under the newly created emissions trading program. The second set of plants were subject to command-and-control regulation over the entire decade. Some of these were built before 1971, but faced fairly stringent state-level emissions standards. (Pre1971 plants that faced lax state standards were typically the plants targeted by Table A.) Others were built after 1971 and subject to federal emissions standards.5 Importantly, both federal and state-level emissions standards were essentially constant over the period at any given plant.

Railroad market power Rail rates during the time period covered by this study were generally individually negotiated between shippers and railroads; the details of particular contracts were not publicly known. Several features of the railroad industry appear to allow ample latitude for railroads to exercise some degree of market power. While rail rates are regulated by a price cap, it was seldom binding for PRB 2

Phase II of the program, beginning in the year 2000, extended the allowance market to all fossil-fuel-based units with at least 25 MWe in generating capacity. 3 A single plant may have several units governed by different regulations. Nonetheless, we focus on Table A plants rather than Table A units because coal delivery data is available only at the plant level. 4 Some generating units that were not listed on Table A participated in the allowance market voluntarily, under the so-called “substitution and compensation” program. Because voluntary participation raises obvious issues of adverse selection (Montero, 1999), we focus on the units that were required to participate as the “affected” group. 5 The 1970 Clean Air Act Amendments required units built after August, 1971, to meet a New Source Performance Standard of 1.2 pounds of SO2 emissions per million Btus. Subsequently, the 1977 Amendments required new sources to install scrubbers. The latter units have no reason to burn low-sulfur coal, and are thus not relevant to our analysis.

4

coal transport.6 Although utilities may file complaints with the Surface Transportation Board, they have had little success in doing so, due in part to the time and expense involved (Grimm and Winston, 2000). Nor does the threat of entry appear to be an effective deterrent to BNSF and UP exercising market power in the PRB. A competing rail line proposed in 1998 remained held up as of early 2007 due to opposition from those living along the proposed route. Finally, a secondary coal market among power plants (which might otherwise constrain the railroads’ ability to set prices) is effectively ruled out: coal is delivered on specialized coal hopper cars and unloaded into vast piles adjacent to the plant. Arbitrage around the railroads would require a power plant to load the coal onto trucks, a costly means of transporting coal over long distances. Industry sources confirm that resale is rare.

Coal extraction in the Powder River Basin Figure 1 shows the locations of the power plants in our sample, along with the three major coal deposits in the United States: the Powder River Basin, the Illinois Basin, and Central Appalachia.7 Coal varies widely among the three regions. The median sulfur content (using all deliveries to power plants in the 1990s) was 0.33 percent by weight for PRB coal – much less than the medians for Central Appalachia (0.90 %) and the Illinois Basin (2.7 %). Heat content is also much lower in PRB coal: a median of 8,674 British thermal units per pound of coal, compared with 12,490 and 11,309 Btus/lb for Central Appalachia and the Illinois Basin, respectively. Our focus is on price discrimination among power plants by the BNSF and UP railroads. There appears to be much less scope for coal mines to exercise market power. At least eighteen mines were active in the PRB during our sample period, producing similar coal. Several more mines stood dormant but with estimated extraction costs only a few dollars higher than prevailing prices. In our empirical analysis, moreover, we control for the price of PRB coal at the minemouth. The minemouth price rose at the advent of the allowance market, but soon returned to its previous level. If the coal mines were able to capture substantial gains from the allowance market, they did so through advances in extraction technology, rather than through increases in coal prices. 6 The Staggers Act of 1980 gave railroads without “market dominance” free rein to set rates up to 170-190 percent of variable costs, and permitted greater scope in establishing contract rates with shippers. 7 The Powder River Basin itself extends into Montana. The vast majority (over 80%) of coal deliveries from the PRB come from the southern portion in Wyoming, however, and we confine our analysis to that region.

5

3

Regulation, location, and the price of low-sulfur coal

We wish to understand how the allowance trading program altered the delivered price of PRB coal. The important factors in the determination of delivered prices are that an individual power plant’s demand for low-sulfur coal depends on regulatory status and geographic location, and that railroads have some scope for price discriminating among plants. To analyze the interplay of these factors, we contemplate delivered coal prices as the outcome of bargaining between a power plant and the railroads. Our treatment is informal; the aim is to motivate the subsequent empirical analysis.8 The value of low-sulfur coal to a power plant depends on how its emissions are regulated. Under an emissions standard, a power plant is willing to pay a premium for low-sulfur coal – but only up to the amount necessary to comply with the regulatory limit. In contrast, an allowance market rewards marginal reductions in emissions: every ton of pollution abated saves the price of an allowance. Meanwhile, distance matters to the value a plant places on PRB coal because it helps to determine a plant’s outside options. For the power plants in our sample, the main alternatives to PRB coal are high-sulfur coal from the Illinois Basin or low- to medium-sulfur coal from Central Appalachia. Hence the cost of PRB coal relative to alternative coals increases with a plant’s distance from Wyoming. A simple stylized example illustrates how these two characteristics of a plant—environmental regulation and distance—interact to determine delivered price at that plant. Consider two hypothetical power plants, identical in all respects, except that one is located relatively near to the PRB – say, in Kansas – while the other is farther east, in Indiana. Each power plant is served by one railroad. Bargaining yields a contract specifying a quantity of coal and a total price equal to the railroad’s share of the economic surplus plus its delivered cost. Figure 2 depicts demand and delivered costs for low-sulfur coal at the Kansas (panel (a)) and Indiana (panel (b)) plants. Under an emissions standard, demand for low-sulfur coal is a step function. The marginal willingness to pay for low-sulfur coal equals the price of high-sulfur coal plus the cost of avoided abatement, up to q; that quantity is sufficient to meet the standard, and the plant is not willing to pay a premium beyond it. In contrast, under the allowance market, marginal willingness to pay for low-sulfur coal equals the price of high-sulfur coal plus the price of an allowance up to the maximum quantity of coal the plant would consume, denoted Q. In effect, the presence of an allowance market lowers the maximum amount a plant is willing to pay for 8

For a formal model of pricing with market power, see the earlier version of this paper (Busse and Keohane, 2004).

6

low-sulfur coal, but raises the minimum. The lower maximum arises because tradeable allowances give a plant another compliance option. The higher minimum comes from the positive price on emissions, which makes a plant willing to pay more on the margin for low-sulfur coal, even at high quantities. Note that the demand functions have similar shapes at the two plants. What differs is the delivered cost of PRB coal and the price of high-sulfur coal.9 The Kansas plant finds it economic to consume only PRB coal, regardless of regulation. The surplus from low-sulfur coal equals area a + b + c under the emissions standard, and a + b + d under the allowance market. If area d is greater than area c, total surplus grows. Because the plant purchases the same quantity of low-sulfur coal under both types of regulation, the average surplus—and thus the price—also rises in this case. For the Indiana plant, PRB coal is a costlier fuel, but an attractive means of abatement: its delivered cost is less than the high-sulfur coal price plus either the alternative abatement cost or the allowance price. That plant consumes q under the emissions standard, but consumes Q > q under the allowance market. At the same time, the increase in total surplus (from area e + f to area e + g) is smaller than at the Kansas plant, since the cost of PRB coal is higher. As a result, the rise in average surplus (and hence price) under the allowance market is smaller in Indiana than in Kansas. This simple example conveys the intuition we will take to the data. Plants close to the PRB consume coal from the region regardless of regulation, becaus it has low transportation cost. At those plants, the effect of the allowance market is to increase surplus without much affecting the quantity a plant near the PRB consumes. The result—as railroads with market power move to obtain some of this surplus–is to raise the price per ton. Farther away, the allowance market makes plants willing to increase their consumption of low-sulfur coal beyond what they would have consumed under an emissions standard, but it lowers the maximum amount they will pay because the allowance market has more flexible compliance options. Thus price rises by less than it does at plants near the PRB. This basic result is reinforced by two additional factors that increase the negotiating position of eastern power plants (compared to western power plants) vis-a-vis the railroads. First, eastern plants have access to coals with a wide range of sulfur contents from Appalachia, the Illinois Basin, and nearby regions. Having a variety of compliance choices is especially valuable under 9

We take the price of high-sulfur coal as given. High-sulfur coal is mined throughout the Illinois Basin, Ohio, and Northern Appalachia. Several railroads serve the region, and a large volume of barge traffic plies the Ohio and Mississippi river networks. Hence one can treat coal transportation as reasonably competitive in this region.

7

an allowance market, since even marginal reductions in sulfur content reduce compliance costs. Second, relationship-specific investments play an important role. A western plant whose boilers had been tuned to burn PRB coal for economic reasons – raising its cost of switching to eastern coal – would not have been well-positioned to bargain with the railroads after Title IV took effect and raised the value of PRB coal.

4

Data

Our dataset comprises information on coal deliveries from the PRB region of Wyoming to railserved power plants between 1990 and 1999. Our primary source is the records of Form 423 of the Federal Energy Regulatory Commission (FERC), the “Monthly Report of Cost and Quality of Fuels for Electric Power Plants.” The form – a monthly report of all fuel deliveries received by a power plant – must be filed by all electricity generating plants with capacity of at least 50 megawatts. For each delivery, we know the delivered price of the coal; the quantity delivered; and selected characteristics of the coal, including its heat, sulfur, and ash contents. We supplement the Form 423 data with information on minemouth prices taken from surveys conducted every other month of coal company staff, published in the trade newsletter Coal Outlook. These data represent the average spot prices that coal mines in the PRB report receiving over a given two-week period. We use the intermediate-goods producer price index to express all monetary values in 1995 dollars. Distances from the Powder River Basin to each power plant – actual distances by rail, not “as the crow flies” – were compiled from state maps and railroad atlases, along with the Platt’s Coal Map produced by Financial Times Energy.10 We also know each plant’s transportation options: which railroads serve it, and whether it is also served by barge and/or truck. To account for changes in transportation cost, we use the Railroad Cost Adjustment Factor for the western United States, computed quarterly by the Association of American Railroads and used by the Surface Transportation Board in assessing railroad rates. This measure is designed to be a real cost index: it is an index of input prices (fuel, labor, and so on) adjusted for productivity. The index is set to unity in the first quarter of 1990, and declines smoothly to 0.72 at the end of the decade. FERC divides deliveries into two categories: “spot market” deliveries, defined as deliveries under short-term contracts less than one year in length, and “contract” deliveries under longer-duration 10 Rail distance is measured as the shortest distance from the center of the PRB (Reno, Wyoming) on routes accessible to the railroad(s) serving the plant. It is constant for each plant across all deliveries. Since mines are only tens of miles apart, measuring from a central point is highly unlikely to influence our results.

8

contracts. We make the same distinction. Although Form 423 does not identify individual contracts, it includes the source name for each delivery (e.g., the coal mine), and reports when a contract is about to expire. We use these data to link deliveries that occur under the same contract.

5

Validation of the modeling framework

Before testing our main predictions, we explore the data along a number of dimensions. First, we confirm that coal prices are correlated with coal characteristics in natural ways; this supports our basic price-hedonic empirical approach. Second, we explore how coal prices vary with measures of a power plant’s outside options, to help validate the bargaining intuition that lies behind the predictions of price discrimination we developed in Section 3. Third, we discuss the geography of PRB coal use over time at Table A and non-Table A plants. Table 2 presents results from an OLS regression of the real delivered coal price on a range of characteristics of coal shipments and power plants. One of the biggest factors in determing the price of coal is delivery distance. We estimate that the delivered price of coal rises with distance from the PRB at a rate of just under 1 cent per ton-mile (9 mills) – a figure well in line with published surveys of rail rates from the period (e.g., General Accounting Office, 2002). We find a transportation rate of similar magnitude in our estimates from the full specification, presented in Section 6. The delivered price of coal also rises with its price at the minemouth. The coefficient on minemouth price – although well over a dollar – is not significantly different from unity. In the full specifications the estimate is even closer to one, suggesting that changes in the minemouth price are passed on to the power plant dollar-for-dollar. The estimated effects of coal characteristics also accord with expectations. Higher heat content makes coal more desirable, and thus raises its price. Ash (which impedes combustion) has a negative, though insignificant, effect. We measure sulfur content by the amount of SO2 (in tons) that would be emitted by burning one ton of coal; thus the estimated regression coefficient can be interpreted as the implicit price of SO2 , in dollars per ton. This implicit price is positive in the full sample. Although somewhat surprising at first blush, this finding is readily explained. The marginal value of sulfur at non-Table A plants need not be negative, especially for coal that is already well below the emissions standard. The positive sulfur price suggests that sulfur content is correlated with other characteristics of coal which matter for prices, but which we don’t observe in this data. (Evidence from a different dataset with more comprehensive coal characteristics suggests

9

that moisture content is a likely candidate.) For plants governed by the allowance market, sulfur should have a negative implicit price, since SO2 emissions were costly. From 1995 to 1999, allowance prices ranged between a low of $70 and a high of $210 per ton of SO2 , with an average of just under $140.11 When the regression in Table 2 is run separately on the subsample of Table A plants during Title IV, the estimated implicit price of sulfur content is negative and significantly different from zero at the 1% level, with a magnitude ($235) close to the actual allowance price. The regressions reported in Table 2 also include measures of the outside options available to power plants. We first consider alternate delivery options. We distinguish four mutually exclusive categories of plants: (i) those served by one railroad only (the omitted category in the regression); (ii) those served by one railroad and at least one other form of transport (such as barge or truck); (iii) those served by multiple railroads but no other form of transport; and (iv) those served by multiple railroads as well as other transport. As expected, more transportation options translate into lower prices. The estimated price discounts are a dollar and a half per ton for plants with a single railroad but other transport options; somewhat over two dollars per ton for plants with multiple railroads; and nearly four dollars per ton for multiple railroads and multiple transport options. A second measure of a power plant’s outside options is its distance from Central Appalachia – a key determinant of the price of low-sulfur coal from that region. The results in Table 2 show that the price of PRB coal rises by nearly one cent per ton, for each mile of rail distance from Appalachia. Finally, the changing geography of PRB coal consumption provides support for our basic approach.12 The most striking change was an increase in the intensity of PRB coal usage west of the Mississippi. In 1990, PRB coal accounted for 40 percent of all coal burned at coal-fired non-Table A power plants west of the Mississippi, but less than 20 percent of the coal burned at Table A plants in the same region. By the end of the decade, the pattern was reversed. In 1999, the fraction of PRB coal at western non-Table A plants had crept up from 40 to just under 50 percent. In the same year, over 95 percent of the coal burned at Table A plants west of the Mississippi came from the PRB. This disparity strongly suggests that the allowance market – and not just a common 11

Allowance price data are from figures compiled by Cantor Fitzgerald EBS and Fieldston Publications and made available by the U.S. EPA. 12 For an illustration of PRB coal consumption over time, see the maps in the earlier version of this paper (Busse and Keohane, 2004).

10

factor such as falling transportation costs – was a driving force behind the expansion of PRB coal use over the decade.13

6

The effect of Title IV on delivered coal prices

In this section, we estimate the effect of Title IV sulfur dioxide regulation on the price of low-sulfur coal. After describing the econometric framework and discussing identification, we present results from OLS and fixed effects regressions. We then probe the robustness of our results by relaxing the assumption that price rises linearly with distance. Next, we present a back-of-the-envelope estimate of the increase in “net contribution” or “producer surplus” captured by the railroads. Finally, we discuss the results of three previous studies in light of our findings.

Econometric framework Section 3 proposed that the allowance market altered the surplus created by low-sulfur coal and the outside options available to power plants – and that those effects varied with distance from Wyoming. In the wake of Title IV, therefore, prices at nearby power plants should have risen relative to prices farther away. The natural question for empirical analysis is: How did the price of PRB coal vary with distance, and how was that relationship affected by market-based environmental regulation? We look for the impact of regulation by estimating the difference in delivered prices to Table A plants before and after Title IV took effect, and then comparing that difference to contemporaneous prices for plants outside of the allowance market. The dependent variable in our regressions is the real delivered price of coal at the power plant.14 Our basic regression equation is the following, where pijt denotes real delivered price for delivery i 13

Note that the sharp increase in the intensity of PRB coal use at western Table A plants is also consistent with the stylized model of Section 3. 14 Price is reported in cents per million Btus on Form 423. We divide by the heat content of the delivered coal to get price in dollars per ton. Our results are essentially identical when we use average heat content (rather than delivery-specific heat content) in the conversion.

11

to plant j at time t: pijt = α0 + β 0 Constant Cost Rail Distancejt + + α1 T able Aj + β 1 T able Aj × Constant Cost Rail Distancejt + α2 T itle IVt + β 2 T itle IVt × Constant Cost Rail Distancejt + α3 T able Aj × T itle IVt + β 3 T able Aj × T itle IVt × Constant Cost Rail Distancejt + γ 1 Spotijt + γ 2 M ineP ricet + γ 3 RailroadCostIndext + θCoal Charsijt + ϕ P lantCharsj + δM onthDummiest + εijt (1) Three variables (and their interactions) are of primary interest. Constant Cost Rail Distance is the rail distance (in miles) from the PRB to the power plant, multiplied by the transportation cost index; thus it varies across plants and over time at a given plant.15 Multiplying by the index accounts for changes over time in transportation costs. (Recall that the index is set to unity in the first quarter of 1990, and captures changes in nominal input prices as well as productivity.) Thus, βˆ0 estimates how much the real delivered price per ton would increase with each additional mile if variable transportation cost were held constant at its 1990Q1 level. Similar reasoning holds for βˆ1 through βˆ3 . Note that we also include separate controls for the cost index and the raw rail distance.16 T itle IV equals zero before Title IV took effect and one afterward. We determine Title IV status by delivery date for spot-market deliveries, and by initiation date for contract deliveries.17 T able A equals one for plants listed on Table A and thus required to participate in the allowance market. (It is constant through the sample period for any given plant.) We also include an interaction term T able A × T itle IV . The coefficients on these dummy variables, and their interactions with rail distance, are the basis for our inferences about the changes in delivered prices at plants in the wake of Title IV. We can interpret the regression coefficients as tracing out price as a function of distance: the α’s represent the implicit price at zero distance, or at the “railhead”; the β’s represent the price gradient in distance. 15

As shown in Table 5, our results are robust to using the actual distance (not the constant cost distance) as the measure interacted with T able A and T itle IV . 16 The interaction of the cost index with distance captures changes in variable, i.e., per mile, cost. The linear cost index term captures changes in costs that contribute the same dollar amount to price, regardless of distance. Our results are robust to using the simple (rather than the productivity-adjusted) cost index. 17 We define the Title IV period as beginning in July 1994 to allow power plants, which typically maintain several months’ worth of coal inventories, to have anticipated the regulation in their purchase decisions. The empirical results change very little if the Title IV period is defined as beginning in October 1994 or January 1995 instead.

12

We include a dummy for Spot-market deliveries in order to control for systematic differences in spot and contract prices. M ineP rice controls for contemporaneous variation in the prevailing price of coal at the minemouth, and also accounts (as well as we can) for Wyoming state severance taxes and county-level ad valorem taxes.18 RailroadCostIndex controls for the cost of rail transportation. The vector Coal Chars includes the sulfur, heat, and ash contents of the coal. These coal characteristics are interacted with the T itle IV and T able A to allow their implicit prices to vary with the policy regime. The vector P lantChars includes the rail distance from the PRB to the power plant; the distance from the plant to the Central Appalachia coal region; and the indicators of transportation options discussed in Section 5. Throughout the analysis, we include month dummies to allow for seasonal effects. The full sample comprises 16,552 deliveries of coal from the Powder River Basin in Wyoming between January 1990 and December 1999, the end of Phase I of the Title IV program. The sample includes 130 plants, 35 of which are Table A plants. We include all rail-served power plants listed on Table A that purchased PRB coal during the sample period, along with all non-Table A rail-served power plants that bought PRB coal and are located east of the Rocky Mountain states. Table 1 presents summary statistics broken down into Table A and non-Table A plants, and into the preand post-Title IV time periods.

Estimation and identification In the econometrician’s ideal world, the Table A designation would have been randomly assigned. Of course, in the real world it was not. Indeed, the generating units that ended up on Table A were the ones of greatest regulatory concern: environmental organizations even nicknamed them the “Big Dirties.” It is not surprising, therefore, to find that the Table A and non-Table A plants differ significantly in many observable characteristics, as Table 1 shows. Our estimation strategy accommodates these differences. The difference-in-difference-style approach controls for any unobserved factors that affected prices at both Table A and non-Table A plants, such as changes in fixed costs or regulatory oversight. Our identification assumption is that no unobserved and unrelated change in transportation costs or in the demand for PRB coal, affecting Table A plants only, coincided 18 County-level taxes vary little over time, and are quite similar in the two counties comprising the PRB. State severance taxes fell during the sample period. Taxes cannot explain our results since we find that the “railhead price”rises instead of falls, and because taxes should not affect Table A and non-Table A plants differently.

13

with the allowance market.19 We will estimate equation (1) both by OLS and using plant-level fixed effects. Estimating equation (1) by OLS has two attractions: the estimates are easily interpretable, and they use information from power plants that purchased PRB coal only after 1994 to help identify the effect of the new regime. (In the fixed effects specification, plants that buy only in the post-Title IV period do not contribute to the estimation of the effect of Title IV on prices.) The railroads presumably regarded new customers as well as existing ones in setting prices; OLS estimates will incorporate data from new customers while fixed effects will not. We account for possible unobserved correlation among deliveries by computing robust standard errors clustered by contract (or by year, for spot deliveries). Nonetheless, the OLS results will be biased if prices depend on unobserved characteristics of power plants that are correlated with variables in the regression. For example, plants at greater distance from the PRB may have boilers designed to burn other types of coal; or observed coal characteristics may be correlated with unobserved characteristics, such as “grindability,” which may differ systematically by boiler type. To sweep out such unobserved heterogeneity, we include plant-level fixed effects in our main specifications.

Results Table 3 presents the results of estimating equation (1) by OLS (columns 1-2) and fixed effects (cols. 3-6).20 The estimates of greatest interest are the coefficients on T able A × T itle IV and its interaction with distance, which capture the change in delivered prices that coincided with the new regulatory regime. The discussion of price discrimination in Section 3 suggests that under the allowance market prices should have risen by more at nearby plants than at distant ones. This corresponds to a rise in the railhead price, and a decline in the gradient. This is just what happened, as the results presented in Table 3 show. In the results reported in column 1, for example, which uses data for Table A firms only, the railhead price jumps by $4.83 per ton after Title IV takes effect. At the same time, the estimated per ton-mile transportation rate falls by just over five mills per ton-mile. The estimated effects of Title IV are even more striking when Table A plants are compared to the full sample of plants (column 2). The coefficient on 19

Divergent pre-existing trends in delivered prices at Table A and non-Table A plants would also pose a problem for our analysis. Looking only within the sample of deliveries prior to 1994, we found no significant trend in prices at either Table A or non-Table A plants. (We thank a referee for suggesting this approach.) 20 Note that because our measure of constant-cost rail distance varies over time with the cost index, it can be identified by fixed effects.

14

T itle IV and its interaction with distance indicate that delivered prices did not change significantly at plants where the regulatory regime did not change. Relative to contemporaneous prices at those non-Table A plants, however, the railhead price of coal at Table A plants rose by about six dollars per ton after Title IV took effect, while the transportation rate fell by more than five mills per ton-mile. These changes in the railhead price and price gradient remain substantial and significant in the fixed-effects specifications, and are robust to the inclusion of dummy variables for coal mines (column 5) and month-year pairs (column 6). Controlling for plant-level fixed effects, along with the other covariates in equation (1), the estimates in column 4 indicate that delivered prices at Table A plants rose by more than four dollars per ton relative to prices at non-Table A plants – an increase of roughly the magnitude of the minemouth price. At the same time, the transportation rate – that is, the price gradient in distance – fell by 2 mills per ton-mile. These estimates imply higher prices within about 2000 miles of the PRB – a radius that includes all the Table A plants in our sample. Hence the coefficient estimates in column 4 imply that delivered prices rose at all Table A plants in our sample after Title IV took effect – relative to prices at non-Table A plants at similar distances, and controlling for coal characteristics, minemouth price, transportation costs, and underlying price variation across power plants (accounted for in the fixed effects). Moreover, the large increase in the railhead price, combined with the fall in the transportation rate, imply that prices increased by more the closer a plant was to Wyoming. The estimated effect is virtually unchanged when mine dummies are added in column 5 and attenuates somewhat when month × year dummies are added in column 6. As a robustness check, we estimated the railhead price and distance gradient separately by 6-month intervals. (We did this by running fixed-effects regressions identical to those in Table 3, except with indicators for each six-month time interval included and interacted with the Table A dummy and with the distance measures.) The results are presented in Figure 3. (The railhead price is plotted in the top panel; the distance gradient is plotted in the bottom panel.) These graphs show two things. First, there does not appear to be a significant difference in the time trends of either the railhead price or the distance gradient during the pre-Title IV period, supporting our differences-in-differences identification strategy. Second, the differences between Table A and non-Table A plants in both the railhead price and the distance gradient is most pronounced from mid-1997 on. This suggests that a substantial part of the Title IV effect estimated in Table 3 is attributable to movements in price that begin in mid-1997. While Table 3 estimates a large and 15

statistically significant increase in railhead prices (i.e. a mean shift in prices) for Table A plants after Title IV, in Figure 3 that mean shift is observable primarily in the latter part of the Title IV period. We use the results from the specification in column 5 of Table 3 to compute predicted values for the delivered price at each plant before and after Title IV, holding all other variables (transportation cost, coal characteristics, etc.) fixed. These simulated prices isolate the estimated effects of the new regulatory regime. That is, they answer the question: “How did the allowance market affect delivered prices as a function of distance, for Table A versus non-Table A plants, independently of other contemporaneous changes?” They are best thought of as “constant-cost, constant-coal” prices rather than predictions of what prices actually were. Because transportation costs fell over the period, actual prices were lower for most of the decade than these constant-cost prices. Figure 4 illustrates the results. The lines in the figure plot constant-cost prices against distance from the PRB. The bars along the bottom of figure show the corresponding estimates of the change in delivered price that coincided with the allowance market, at each power plant in the sample. The relative change in prices at Table A plants after Title IV is measured by the difference between the height of the bar at any Table A plant, and the height of the bar at the nearest non-Table A plant. While the dark bars show that the change in delivered price was positive for some Table A plants and negative for others, the lighter bars show that the change in delivered price for non-Table A plants was always a larger negative value than the change for Table A plants. This illustrates the result – evident in the regression estimates – that prices rose at all Table A plants relative to non-Table A plants, with the size of the relative increase declining with distance from the PRB. Figure 4 also illustrates what happened to constant-cost prices at Table A plants when they are considered on their own, rather than in comparison to the non-Table A plants. Controlling for transportation costs, coal characteristics, and so on, delivered prices rose at Table A plants near to the PRB, and fell further away. This is reflected in the constant-cost price lines, which “pivot” at a distance of 940 miles; and in the bars representing the change in estimated prices at Table A plants, which start out above the horizontal axis, but end up below it. (The median distance, among all deliveries to Table A plants in our sample, was 1112 miles; the mean was 1107 miles.) To gauge the economic significance of these results, consider how the estimated price changes play out at two pairs of power plants. For two plants located about 700 miles from the PRB, Title IV coincided with an increase in the constant-cost coal price of 88 cents per ton at the Table A plant and a decrease of $1.61 per ton at the non-Table A plant. Hence the relative price increase 16

at the Table A plant was $2.49. At a distance of about 1700 miles from the PRB, the estimated change in price for the Table A plant was a fall of $3.02 per ton. This decline was just smaller (by about 9 cents per ton) than that at a nearby non-Table A plant, where the constant-cost price was estimated to have fallen by $3.11. While these price differences are on the order of a dollar or two per ton, they translate into substantial sums: the median annual quantity of PRB coal delivered to plants in this dataset is 1.4 million tons per year. These empirical results are consistent with discussion of price discrimination in Section 3. The response of delivered coal prices to the allowance market was markedly different for Table A plants than for plants that were not directly affected by the new regime. Controlling for costs and a range of other factors, delivered prices rose at Table A plants within several hundred miles of the Wyoming coal mines – a region with limited alternatives to low-sulfur PRB coal. Farther away, where plants have more attractive outside options, prices fell. Even there, however, prices remained higher at Table A plants than at other plants located at similar distances from the PRB. Hence the railroads charged different prices to different power plants on the basis of the environmental regulations governing the plants and the attractiveness of the plants’ alternative abatement options. Their ability to price discriminate in this way depended on two key factors: the market power afforded by the railroads’ control of transportation out of the Powder River Basin, and the differences in demand for low-sulfur coal under the two emissions regulations.

Robustness We probe the robustness of our results in two ways. First, we take into account contract length. Since prices generally declined over time, one might be concerned that our results merely reflect the “price overhang” from long-term contracts signed early in the decade by western plants. This concern proves to be unfounded, however. Although prices were indeed higher for contracts signed in early years, the correlation between distance and contract length is weak. Table 4 presents direct evidence that variations in contract length are not driving our results. This table reports the results of a regression that adds dummy variables for contract length to the specification reported in column 5 of Table 3.21 The results are quite similar to those in Table 3.22 21 We include dummy variables for contracts less than two years in length, 2-3 years, 3-4 years, 4-5 years, 5-6 years, 6-7 years, and 7-plus years. We have data through 2005, and contract expiration dates are not reliably reported until the final year of a contract, so that we cannot distinguish contract lengths beyond 7 years. 22 The coefficients on the contract length dummies, suppressed for the sake of space, are all positive; they imply a premium over spot-market deliveries of somewhat over two dollars per ton for short-term contracts (under three years) and between three-and-a-half and four-and-a-half dollars per ton on longer contracts.

17

As a second check on robustness, we address the suitability of the way we have controlled for cost changes so far, namely by multiplying actual distance by the transportation cost index to create our constant cost distance measure. To do this check, we modify equation (1) by dropping the ConstantCostRailDistance term and substituting the actual rail distance for ConstantCostRailDistance in the interaction terms with T itle IV and T able A × T itle IV . We allow the plant fixed effects to absorb all of the price variation due to distance, while still identifying the effect of Title IV off of the interactions of distance with T itle IV and T able A × T itle IV . Dummies for delivery years are included to provide a crude control for changes in costs and other variables. This approach sacrifices a direct accounting of transportation cost for the sake of greater transparency. These results appear in Table 5. They are remarkably similar to the results in column 5 of Table 3.23

Estimating the gains to the railroads from Title IV If the railroads were able to exercise market power, then they may well have benefited from the regulation, even while lowering the price of delivered coal to faraway plants. We cannot calculate profits directly, because we lack detailed cost data. Moreover, since we have not estimated a full demand model, we are unable to model what demand would have been under an alternative set of delivered prices. Thus we cannot directly estimate the gains to the railroads from the observed rotation in price as a function of distance, given the change in the regulatory regime. Instead, we derive “back-of-the-envelope” estimates for changes in revenues and producer surplus from deliveries to Table A plants, comparing annual figures for two periods: early in the decade (1990-1992) and towards the end of Phase I (1997-1999). Revenues to the railroads, net of minemouth prices, can be computed directly from the reported price and quantity data, along with the data on minemouth prices. The results are reported in Table 6. (In calculating revenue, we use real delivered prices – not predicted prices obtained from regression estimates.) Although the average delivered price per ton-mile fell after the advent of the allowance market, delivered volume increased dramatically. The net effect is that annual revenues to the railroads from deliveries to Table A plants after Title IV took effect were more than triple what they had been in the first few years of the decade. 23

We also estimated specifications that used polynomials of distance, and that used categorical variables to specify distance by sorting plants into different bins based on distance. The resulting predicted prices echo the linear predictions plotted in Figure 4.

18

At the same time, the railroad cost index declined by nearly thirty percent between 1990 and 1999, which suggests that revenues net of variable cost – that is, “contribution” or producer surplus – may have risen. To obtain concrete figures, we use a range of plausible estimates for the variable cost per ton-mile during the period. The figures we use are based on econometric estimates of the marginal cost of unit train service for BNSF and UP in 1997, derived from a detailed study of railroad costs that draws on actual (confidential) waybill data.24 We use a central estimate for variable cost of 1 cent per ton-mile, along with high and low estimates of 1.1 and 0.9 cents per ton-mile. We take each cost estimate as corresponding to June 1997, and then apply the transportation cost index to extrapolate costs in other months. Multiplying these cost estimates by actual distances and quantities for each delivery and summing over observations, we recover an estimate of total variable delivered cost; subtracting this from revenues yields producer surplus. The results of this exercise are shown in Table 6. Using the central cost estimate of 1 cent/tonmile, estimated producer surplus from deliveries to Table A plants rose from $22 million per year in the early period to $64 million per year in the later period. The higher cost yields estimates of $11 million and $28 million per year in the early and late periods; using the low-cost estimate, annual producer surplus balloons to $32 million and $98 million. To put these rough estimates of producer surplus in context, the market value of allowances during Phase I was roughly $850 million per year (6.3 million allowances allocated each year, multiplied by the average price of $135). Annual variable abatement cost among Table A plants in each year of Phase I was on the order of $560 million, leaving an economic surplus (over variable cost) of $290 million.25 Our central estimate of an annual increase of $42 million in the railroads’ producer surplus represents nearly 15% of the entire surplus created by the allowance market. This strikes us as substantial but still of plausible magnitude.

Relationship to previous studies Three other studies have explored changes in coal prices during roughly the same period, emphasizing the overall decline in the delivered price of PRB coal. Ellerman and Montero (1998), seeking to explain lower-than-expected prices of sulfur dioxide allowances at the start of the emissions trading program, conjecture that railroad deregulation in 1980 led to lower transportation rates and thus 24

We thank John Bitzan, who graciously provided us with these estimates in private communication. For the study that they are based on, see Bitzan (2000). 25 This estimate of total variable abatement costs is derived by the authors from data on abatement and abatement costs described in Keohane (2007).

19

greater use of PRB coal in the 1990s. Winston, Dennis, and Maheshri (2004) examine delivered prices from 1985 to 1998 and argue that prices were driven down over this time period by the entry of the UP into markets that had previously been served only by the BNSF. Finally, a recent study by Kunce, Hamilton, and Gerking (2005), using confidential data on transportation rates, finds that the two railroads exercised considerable market power, but argues that changes in low-sulfur coal consumption during Phase I were driven by changes in the costs of extracting and transporting coal, rather than by the allowance trading regime. Our key findings do not contradict the main empirical fact discussed by these three studies: that the average delivered price of PRB coal fell over the 1990s. Indeed, the general decline in prices during the second half of the 1990s is apparent in our own econometric results. But our econometric results also demonstrate that conditional on cost changes (and a variety of other covariates), prices rose at Table A plants within roughly 940 miles of the PRB, after Title IV took effect. Moreover, relative to prices at non-Table A plants, prices rose at all Table A plants. Changes in costs cannot easily explain our results. First, our regressions include direct measures of transportation costs and minemouth prices, as well as time dummies. Moreover, any unobserved changes in costs would presumably have similar effects on coal prices at similar locations, regardless of regulatory status and would be accounted for in our difference-in-differences approach. But our econometric results demonstrate that the allowance trading regime coincided with a statistically and economically significant change in delivered prices at the plants governed by the new regime, relative to prices at other plants. Indeed, it is difficult to see how a simple change in the cost of transporting coal, even if it were confined to Table A plants, would produce the price changes we find. A broad-based change in variable costs (due to unobserved labor or fuel costs or productivity) would not increase costs by more at nearby plants than at more distant ones. Nor would an increase in fixed costs explain why the transportation rate fell. For similar reasons, our results cannot be explained by increasing price competition between railroads, cited by both Ellerman and Montero and Winston et al. as the reason for falling prices. Greater price competition would have shifted prices everywhere, rather than differentiating between plants under different emissions regulations. To put the same point another way, our finding of price discrimination on the basis of regulatory status does not depend on any assumption about market power, but rather is evidence that some market power existed. The crucial contribution of our paper, relative to these other studies, is our focus on the incremental effect of Title IV rather than the overall change in average prices. Railroad deregulation in 20

the 1980s may well have led to lower prices for PRB coal in the 1990s. But our results demonstrate that those prices were also affected by the creation of the sulfur dioxide allowance market. Title IV altered the demand for low-sulfur coal. The railroads capitalized on the change.

7

Conclusion

The aim of this paper is to understand how the institution of an allowance trading system for SO2 emissions affected the market for low-sulfur coal from the Powder River Basin in Wyoming. Our central finding is that the effect of the allowance market on delivered coal prices differed dramatically among power plants. We estimate that the “railhead price” of PRB coal rose by roughly $4 per ton at Table A plants, relative to non-Table A plants; at the same time, the transportation rate fell by just under 2 mills per ton-mile. As a result, delivered prices rose at the plants that were directly affected by the new regulatory regime, relative to nearby plants whose regulation did not change. Moreover, the relative price increase was greater at plants located closer to the PRB, whose outside options were sparse. These estimated changes control for variable transportation costs, minemouth prices, coal characteristics, and even unobservable characteristics of the power plants, coal mines, and delivery years. In effect, the railroads were able to price discriminate among power plants, on the basis of the environmental regulations governing the plants and the attractiveness of the plants’ alternative abatement options. The railroads’ ability to price discriminate in this way depended on two key factors: the market power afforded by the railroads’ control of transportation out of the Powder River Basin, and the differences in demand for low-sulfur coal under the two emissions regulations. Our results highlight the interaction between environmental regulation and input markets. Because the form of regulation affects the demand for “clean” inputs, such as low-sulfur coal, it will also affect the prices of those inputs when there is market power in their supply. Market-based environmental policies, like the allowance trading program studied here, do not only affect the allocation of pollution abatement among firms in a regulated industry; they can also alter the underlying costs of abatement.

21

Data description appendix In this appendix, we provide additional information about our data to augment the summary statistics presented in Table 1.

Plant types Our main estimation sample contains 16,522 observations. These observations comprise data on 130 plants. Of these, 35 plants are classified as Table A plants and 95 are classified as non-Table A plants. Of the Table A plants, 5 bought PRB coal only in the pre-Title IV period, 9 bought only in the post-Title IV period, and the remaining 21 bought in both periods. Of the non-Table A plants, 10 bought only in the pre-Title IV period, 19 bought only in the post-Title IV period, and the remaining plants bought in both periods. We thus drop sixteen non-Table A plants (in OR, WA, MT, WY, CO, UT, and AZ) from the sample. We do this in order to better match the “control” group with the Table A plants – the westernmost of which is in Kansas.

Delivery frequency and size On average, a plant receives 19.3 deliveries per year (15 median). Table A plants averaged 15.9 deliveries per year (median 12.5); non-Table A plants received 20.4 deliveries per year on average (median 17). The average delivery size is 108,064 tons, with a standard deviation of 121,154. The median delivery size is 65,547 tons. For Table A plants, the average delivery sizes are 81,858 tons and 90,645 tons in the pre- and post-Title IV periods (medians 53,900 and 56,000, respectively). For non-Table A plants, the average delivery size is 126,342 tons in the pre-Title IV period and 99,462 tons post-Title IV (medians 82,800 and 57,000, respectively). The “between plant” standard deviation of delivery quantity (namely, the standard deviation of the plant level averages of delivery quantity) is 85,313. The “within plant” standard deviation of delivery quantity is 88,202. (The within plant measure is the standard deviation of qit − q¯i + q¯.)

Contract type Sixty-eight percent of the deliveries in our sample are made under contract; the remainder are declared as spot market deliveries on Form 423. Contract deliveries account for 80.4% of the total volume of PRB coal in our sample. Half of the contracts spanned were longer than five years. The

22

proportion of deliveries under contract is similar at non-Table A plants and Table A plants (70% and 60%, respectively).

Mines The data allow individual identification of the 18 major coal mines in the PRB: Antelope, Belle Ayr, Big Horn, Black Thunder, Buckskin, Caballo, Clovis Point, Coal Creek, Cordero Rojo, Dave Johnston, Dry Fork, Eagle Butte, Fort Union, Jacobs Ranch, North Antelope, Rochelle, Rawhide, and Wyodak. The names of these mines are subjected to a range of typographical assaults by the filers of Form 423; we re-coded each observation in the data to ensure consistency. (Thus EAGEL BUTE and EAGLE BUTTE are assigned to the same source.) Together, the eighteen listed coal mines account for over 90% of reported deliveries from the PRB during the 1990s. The remainder name either a coal company or a broker as the source.

Prices Prices vary tremendously with distance. In a simple univariate regression of real delivered price per ton (measured in 1995 dollars) on distance, the delivered price per ton rises by 1.8 cents per mile of delivery distance from the Powder River Basin. The average delivered price per ton in the entire estimation sample is $20.25 (standard deviation $8.11). Nonparametric kernel estimates of the relationship between prices and distance show the basic (and unsurprising) result that delivered price rises with distance. They also show the primary empirical result of the paper, namely that prices rise less steeply with distance for the most distant Table A firms after the implementation of Title IV. In addition to price variation with distance, there is also variation in prices across transactions for a single plant. The within-plant standard deviation of prices for the whole sample is $4.70. The within-plant variation in prices is much less for Table A plants, at $1.93, than it is for non-Table A plants, at $5.14. For both types of plants, the within-firm standard deviation of prices decreases between the pre-Title IV period and the post-Title IV period; from $1.97 to $1.51 for Table A plants and from $5.03 to $4.28 for non-Table A plants. While there is clearly a lot of variance in price, there is some evidence even in the raw data of the regression results. For example, Table A plants located furthest from the PRB paid prices that were, for the most part above $20 per ton in the Pre-Title IV period. In the post-Title IV period, those plants paid prices that tend to be around $20 per ton or below. This is consistent with the 23

rotation in prices that we find for Table A firms in the post-Title IV period. As another example, Table A plants also generally appear to pay lower prices than do non-Table A plants; there is a single Table A plant that pays prices above $40 per ton, but a number of non-Table A plants that have a significant share of their price observations above $40 per ton. This is consistent with our contention in Section 3 that non-Table A plants – who are required to meet an emissions standard and do not have the option of covering their emissions with allowances – will pay higher prices because they have fewer outside options. In order to understand how much of the variation in prices is due to differences in coal characteristics, we regressed the real price per ton on railroad delivery distances and the three coal characteristics we measure (sulfur dioxide content, heat content and ash content). We then calculated the residuals from this regression, and analyzed the variance of the residuals. Overall in the sample, the standard deviation of the residuals is $7.58. The within-plant standard deviation is $4.73. (As with the overall standard deviation of prices, the within-plant standard deviation is much larger within the sample of non-Table A firms ($5.16) than it is among Table A firms ($2.05).)

24

References [1] Bitzan, J.D. “Railroad Cost Conditions – Implications for Policy.” Report prepared for the U.S. Department of Transportation, Federal Railroad Administration, 2000. [2] Burtraw, D. The SO2 Emissions Trading Program: Cost Savings Without Allowance Trades. Contemporary Economic Policy, Vol. 14 (1996), pp. 79-94. [3] Busse, M.R., and N.O. Keohane. Market Effects of Environmental Regulation: Coal, Railroads, and the 1990 Clean Air Act. Working Paper no. 137, Center for the Study of Energy Markets, University of California, 2004. [4] Ellerman, A. D., and J.-P. Montero. “The Declining Trend in Sulfur Dioxide Emissions: Implications for Allowance Prices.” Journal of Environmental Economics and Management, Vol. 36 (1998), pp. 24-45. [5] —–, P.L. Joskow, R. Schmalensee, —–, and E.M. Bailey. Markets for Clean Air: The U. S. Acid Rain Program. Cambridge, UK: Cambridge University Press, 2000. [6] General Accounting Office. “Railroad Regulation: Changes in Freight Railroad Rates from 1997 through 2000,” GAO-02-524, 2002. [7] Grimm, C., and C. Winston. “Competition in the Deregulated Railroad Industry: Sources, Effects, and Policy Issues.” In S. Peltzman and C. Winston, eds. Deregulation of Network Industries: What’s Next? Washington, D.C.: Brookings Institution, 2000. [8] Keohane, N.O. “Cost Savings from Allowance Trading in the 1990 Clean Air Act: Estimates from a Choice-Based Model.” In C.E. Kolstad and J. Freeman, eds. Moving to Markets in Environmental Regulation: Lessons from Twenty Years of Experience. New York: Oxford University Press, 2007. [9] Kunce, M., S. Hamilton, and S. Gerking. “Marketable Permits, Low-Sulfur Coal and the Behavior of Railroads.” Mimeo, University of Wyoming, 2005. [10] Montero, J.-P. “Voluntary Compliance with Market-Based Environmental Policy: Evidence from the U. S. Acid Rain Program.” Journal of Political Economy, Vol. 107 (1999), pp. 9981033.

25

[11] Winston, C., S.M. Dennis, and V. Maheshri. “Duopoly in the Railroad Industry: Bertrand, Cournot, or Collusive?” Mimeo, Brookings Institution, 2004.

26

Table 1: Summary statistics Mean

Standard deviation

Minimum

Maximum

PANEL A – TABLE A PLANTS Pre-Title IV (N = 1517) Delivered coal price (1995 $/ton) Spot (N = 506) Contract (N = 1011) SO2 content (tons per ton coal) Heat content (Btus/lb) Ash content (% by weight) Rail distance to PRB (miles) Rail service only, multiple RRs Single RR, other options Multiple RRs, other options Rail distance to Central App.

18.47 19.80 17.80 .01 8653.48 4.98 1120.35 .13 .15 .26 719.41

4.13 4.82 3.56 .00 166.40 .74 241.77 .33 .36 .44 215.32

9.22 11.79 9.22 .00 8099 3.90 711 0 0 0 88

34.89 34.89 27.90 .01 8954 25.58 1706 1 1 1 1074

Title IV (N = 1604) Delivered coal price (1995 $/ton) Spot (N = 727) Contract (N = 877) SO2 content (tons per ton coal) Heat content (Btus/lb) Ash content (% by weight) Rail distance to PRB (miles) Rail service only, multiple RRs Single RR, other options Multiple RRs, other options Rail distance to Central App.

16.61 16.59 16.63 .01 8655.78 5.08 1093.67 .22 .21 .11 695.80

3.59 2.50 4.28 .00 205.84 .60 201.16 .42 .40 .32 171.21

8.84 9.22 8.84 .00 6409 3.20 711 0 0 0 251

58.34 28.34 58.34 .01 9003 7.50 1694 1 1 1 1074

PANEL B – NON-TABLE A PLANTS Pre-Title IV (N = 6816) Delivered coal price (1995 $/ton) Spot (N = 1397) Contract (N = 5419) SO2 content (tons per ton coal) Heat content (Btus/lb) Ash content (% by weight) Rail distance to PRB (miles) Rail service only, multiple RRs Single RR, other options Multiple RRs, other options Rail distance to Central App.

22.47 19.20 23.31 .01 8615.63 5.10 1092.80 .31 .19 .05 920.42

9.11 5.03 9.72 .00 213.98 .61 293.42 .46 .40 .22 301.46

4.13 4.13 8.92 .00 5969 3.26 367 0 0 0 269

80.68 80.68 72.10 .05 9461 17.90 1797 1 1 1 1629

Title IV (N = 6616) Delivered coal price (1995 $/ton) Spot (N = 2682) Contract (N = 3934) SO2 content (tons per ton coal) Heat content (Btus/lb) Ash content (% by weight) Rail distance to PRB (miles) Rail service only, multiple RRs Single RR, other options Multiple RRs, other options Rail distance to Central App.

19.41 17.91 20.44 .01 8611.94 5.12 1133.42 .21 .20 .08 815.07

8.14 5.20 9.51 .00 192.69 .55 304.16 .41 .40 .27 309.58

2.66 2.66 5.13 .00 6274 .56 367 0 0 0 219

64.34 50.80 64.34 .02 9565 12.24 2243 1 1 1 1629

Variable

Table 2: Model validation estimates Dependent variable: Delivered coal price (1995 $/ton) SO2 content (tons per ton coal) 385.32* (189.47) Heat content (Btus/lb) .01** (.00) Ash content (% by weight) -0.36 (.38) Single RR, other options -1.56+ (.81) Rail service only, multiple RRs -2.17** (.80) Multiple RRs, other options -3.95** (.76) Rail distance to Central App. coal price .009** (.001) Utility size (GW) 2.05** (.31) Utility PRB purchases in past 12 mos. -0.41** (.11) > 75% PRB coal 3.71** (.75) Minemouth price 1.59** (.42) Rail distance to PRB (miles) .009** (.001)

January February March April May June July August September October November Constant

1.12** (.21) 1.21** (.21) 1.21** (.20) 1.11** (.21) .91** (.20) .78** (.18) .63** (.16) .45** (.16) .28 (.15) .29* (.15) .24 (.14) -71.10** (12.55)

Notes: OLS estimation. Robust standard errors clustered by contract in parentheses. 15,803 observations. R2 = 0.3. + denotes significance at 10 % level; * at 5 % level; ** at 1 % level.

Table 3: Regression results

Dependent variable: Delivered coal price (1995 $/ton) Table A × Title IV Table A × Title IV × constant-cost distance

Table A (1) 4.83** (1.60)

OLS All plants (2) 6.06* (2.72)

Table A (3) 3.04** (.45)

-0.0051** (.0017)

-0.0053* (.0027)

-0.0036** (.0004)

Title IV period Title IV × constant-cost distance Table A plant Table A × constant-cost distance

Fixed effects All plants All plants (4) (5) 4.14** 4.17** (.91) (.91) -0.0021* (.0009)

-0.0024** (.0009)

All plants (6) 3.53** (.91) -0.0017+ (.0009)

-0.27 (2.05)

-0.79* (.38)

-0.56 (.38)

-1.19** (.38)

-0.0002 (.0019)

-0.0015** (.0004)

-0.0015** (.0004)

-0.0004 (.0004)

-0.0010 (.0012)

-0.0003 (.0012)

-0.0002 (.0012)

0.0059** (.0016)

0.0057** (.0016)

0.092** (.0017)

-5.10* (2.17) 0.0101 (.0062)

Constant-cost rail distance

0.0019 (.0020)

-0.0046* (.0022)

0.0043 (.0059)

Spot-market delivery

-0.21 (.32)

-3.75** (.42)

-0.55** (.08)

-2.92** (.09)

-2.98** (.09)

-3.07** (.09)

Minemouth price

0.74** (.28)

0.86+ (.48)

0.77** (.10)

1.10** (.10)

1.10** (.10)

1.38** (.46)

11.86+ (7.02)

16.24** (2.64)

8.87** (2.02)

9.49** (2.02)

4.81** (.65)

3.66** (.64)

3.33** (.66)

7.73** (1.98)

Railroad cost index Rail distance to PRB (miles) Constant Coal characteristicsa Plant characteristicsb Month dummies Fixed effects (plants) Mine dummies Month-by-year dummies Observations R-squared Number of plants

4.16 (7.66) 0.0060 (.0064)

0.0073 (.0057)

-11.00 (8.85)

-4.20 (7.29)

Yes Yes Yes No No No

Yes Yes Yes No No No

Yes No Yes Yes No No

Yes No Yes Yes No No

Yes No Yes Yes Yes No

Yes No No Yes Yes Yes

3,121 0.48

16,552 0.28

3,121 0.28 35

16,552 0.25 130

16,552 0.27 130

16,552 0.29 130

Notes: OLS (columns 1-2) and fixed effects (columns 3-6) estimates. Standard errors in parentheses; cols. 1 and 2 report robust standard errors clustered by contract. + denotes significance at 10 % level; * at 5 % level; ** at 1 % level. a Coal characteristics include sulfur, heat, and ash contents, interacted with T itle IV and T able A dummies. b Plant characteristics include rail distances to Central Appalachia and transport options.

Table 4: Further regression results: Contract length Dependent variable: Delivered coal price (1995 $/ton) Table A × Title IV Table A × Title IV × constant-cost distance Title IV period Title IV × constant-cost distance Table A × constant-cost distance Constant-cost rail distance Railroad cost index Minemouth price Constant Coal characteristics Month dummies Fixed effects (plants) Mine dummies Contract length dummies Observations Number of plants R-squared

3.65** (.91) -0.0021* (.0009) 0.42 (.39) -0.0019** (.0004) -0.0000 (.0012) 0.0056** (.0016) 10.49** (2.01) 1.14** (.10) -1.02 (.69) Yes Yes Yes Yes Yes 16,552 130 0.28

Notes: Fixed effects estimates. Standard errors in parentheses. + denotes significance at 10 % level; * at 5 % level; ** at 1 % level.

Table 5: Further regression results: Assessing robustness Dependent variable: Delivered coal price (1995 $/ton) Table A × Title IV Table A × Title IV × distance Title IV period Title IV × distance Minemouth price Spot-market delivery Constant Coal characteristics Month dummies Year dummies Fixed effects (plants) Mine dummies Observations Number of plants R-squared

4.47** (.9683) -0.0022* (.0009) 0.02 (.33) -0.0014** (.0003) 0.11 (.21) -3.08** (.09) 19.85** (.76) Yes Yes Yes Yes Yes 16,552 130 0.28

Notes: Fixed effects estimates. Standard errors in parentheses. + denotes significance at 10 % level; * at 5 % level; ** at 1 % level.

Table 6: Quantities, revenues, and estimated producer surplus: Deliveries to Table A plants

Number of deliveries

Quantity (million ton-miles/year)

Total revenues ($million/year)

Pre-Title IV (1990-1992)

164

9,016

125

32.3

21.9

11.6

Title IV (1997-1999)

367

38,353

415

98.8

63.7

28.6

Difference

203

29,337

290

66.5

41.8

27.0

Net revenues ($million/year) (low cost) (central estimate) (high cost)

Notes: All figures represent annual averages. Quantities and revenues are based on reported prices and quantities. Producer surplus is a back-of-the-envelope estimate using estimated costs from data used in Bitzan (2000). The low-, central, and high-cost estimates correspond to costs of .9, 1.0, and 1.1 cents/ton-mile in 1997, respectively.

Illinois Basin

Central Appalachia

PRB 30’

25 oN 127 o W

o 67 W

Table A plants Non Table A plants

Figure 1: Map of power plants in sample, with major U. S. coal regions.

b

c

q

a

demand under emissions standard

d

Q

quantity of low-sulfur coal

delivered cost of low-sulfur coal

demand under alllowance market

e

f

q

demand under emissions standard

g

Q

quantity of low-sulfur coal

delivered cost of low-sulfur coal

demand under alllowance market

(b) – Distant plant (the delivered cost of low-sulfur coal is greater than the price of high-sulfur coal). The plant buys q tons of low-sulfur coal under the emissions standard, but increases purchases to Q tons under the allowance market. Total surplus increases from area e+f under the standard to area e+g under the market. The increase in average surplus (and therefore price) is less than at the nearby plant depicted in panel (a).

price of HS coal

price of HS coal + allowance price

price of HS coal + abatement cost

price of low-sulfur coal

Figure 2 – Stylized model of coal demand and delivered cost at two power plants under two regulatory regimes.

(a) – Nearby plant (the delivered cost of low-sulfur coal is less than the price of high-sulfur coal). The plant purchases quantity Q of low-sulfur coal under both the emissions standard and the allowance market. Under the emissions market, surplus increases from area a+b+c to area a+b+d. Thus the average surplus, and hence price per ton, rises.

price of HS coal

price of HS coal + allowance price

price of HS coal + abatement cost

price of low-sulfur coal

2

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

3&4(5)&6*,748)*9:;-"#< $/ 0 / 1

%&'()*+*,(&#-.

!"#$%&'()*+ ,(&#-.

@4.-)*C7&64)#-*98)#-.;-"#$D4()< 0 A/ A1 A2 A? B BA/ BA1

=&#*>0

=&#*>/

=&#*>1

=&#*>2

=&#*>?

@)8*>>

=&#*>1

=&#*>2

=&#*>?

@)8*>>

!"#$%&'()*+*,(&#-.

=&#*>0

%&'()*+*,(&#-.

=&#*>/

Figure TJV 3: Estimated and S./;#( W+3.*%3(-railhead #%.'4(%-price 6#.=(and %,-distance -.+3%,=(gradient /#%-.(,3by &B six-month +.P5*",34 intervals .,3(#@%'+ for !"# Table $%&'( A ) %,non-Table ) A 6'%,3+> plants. ,",5$%&'(

-!.&/0.! #/ -!1!%!! 2 3 4)5! 2

%&'()*&(+,&-&*).-+/&)0123456

$"

$!

#"

#!

"

!

7!!

8!!

9!!

#!!!

##!! #$!! #:!! %'+()*+=3'5/&)>-4?)@%A

#;!!

#"!!

#
B45CD'E(&)FG)@-&CD+3(&)HI

D'E(&)FG)@-&CD+3(&)HI

B45CD'E(&)FG)D+3(&)HI

D'E(&)FG)D+3(&)HI

JK'5L&)0B45CD'E(&)F6

JK'5L&)0D'E(&)F6

#7!!

Figure 4: Estimated “constant-cost” delivered prices as a function of distance, for both Table A and non-Table A plants, before and after the advent of the allowance market.