The Impact of Cheap Natural Gas on Marginal Emissions from Electricity Generation and Implications for Energy Policy
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Impact of Cheap Natural Gas on Marginal Emissions from Electricity Generation and Implications for Energy Policy J. Scott Holladay∗ Jacob LaRiviere† July 2014 Abstract We use quasi-experimental variation due to the introduction of fracking to estimate the impact of a decrease in natural gas prices on marginal air pollution emissions from electricity producers. We find natural gas generation has displaced coal fired generation as the marginal fuel source even over baseload hours significantly changing the marginal emissions profile. The impact of cheap natural gas varies across U.S. regions as a function of the existing stock of electricity generation. We then simulate the effect of the natural gas price decrease on welfare from adding bulk electricity storage to the grid. We find that the non-market benefits of bulk storage have increased while the private benefits have decreased. This result highlights the uncertainty in welfare from second best policies that support specific technologies in the energy sector as opposed to pricing externalities directly. JEL Classification: Q4, L5, Q5, H4 Keywords: energy, air pollution, climate change, natural gas ∗ Department of Economics, University of Tennessee; email: jhollad3@utk.edu † Department of Economics, University of Tennessee; email: jlarivi1@utk.edu 1
1 Introduction Fossil fuels account for roughly 83% of U.S. energy production despite policies aimed at increasing the share of energy produced by renewables.1 A growing body of work evaluates emissions associated with the residential, commercial and transportation sectors (Jacobsen et al. (2012), Chong (2012), Bento et al. (2013), Kahn et al. (2014) and Allcott et al. (2014)), but electricity production both consumes the most energy and produces the most pollution of any sector in the U.S. economy.2 Further, much of the emissions from other sectors stem from their demand for electricity. Due to its reliance on coal, natural gas and oil, the electricity sector remains the single most important source of greenhouse gasses along with many other types of pollution. Recently, there has been a dramatic shift away from coal-fired electricity generation toward natural gas-fired generation due to the widespread adoption of hydraulic fracturing and horizontal drilling in natural gas extraction known as “fracking.”3 EIA (2012) reports that natural gas extracted in this way from shale deposits accounts for over a third of all natural gas produced in 2012 up from less than two percent in 2000. Total production of natural gas also increased by 25% over the same time frame with the increase coming entirely from shale gas extraction. This increase in production has been associated with a large decrease in market prices. Both NYMEX and Henry Hub natural gas spot and futures prices peaked above $14/MMbtu in late 2005 with continued high prices through 2008 when prices began to fall rapidly as the number of fracking wells increased dramatically. The current price of natural gas is roughly $4.50/MMbtu. This change in price was not predicted in commodities markets: in 2008 the futures market for natural gas prices was significantly higher than realized spot market prices. McKinsey, a global management consultancy, has suggested that the fracking boom made “a significant shift in the way we 1 As of January 2014. See Energy Information Administration data at: http://www.eia.gov/totalenergy/data/browser/xls.cfm. 2 See http://www.eia.gov/totalenergy/data/browser/xls.cfm?tbl=T02.01&freq=m. 3 The academic literature is beginning to evaluate the direct environmental impacts of fracking: Olm- stead et al. (2013) and Osborn et al. (2011) estimate the water quality impacts of fracking and Muehlen- bachs et al. (2012) estimates the hedonic impacts on well-proximate housing prices. 2
think about energy security, and the way we think about the impact of energy prices on our economy.”4 In this paper, we identify how emissions in the electricity sector have been affected by the unanticipated price change in natural gas due to fracking. To do so, we construct a large and unique data set of hourly generators level and emissions data from the U.S. Environmental Protection Agency (E.P.A.) to estimate marginal pollution emissions from electricity production in the U.S. We use the unanticipated change in natural gas prices to identify changes in the marginal emission profile over time as a result of changes in the dispatch of electricity generation. We estimate the effect on marginal emissions due to the well-known need to account for within day variation in emissions rates to evaluate the change in emissions due to electricity sector policies (Holland and Mansur (2008), Zivin et al. (2012), Cullen (2013), Kaffine et al. (2013), and Novan (2012)). We provide evidence that the natural gas price decrease was unanticipated by the markets, and therefore take our estimates for the effect of the natural gas price decrease on marginal emissions to be causal. We use a semi-parametric econometric model with a large number of fixed effects to estimate marginal emissions for eight regions throughout the U.S.5 We study the time period between 2005 to 2011 in our analysis for two important reasons. First, 2005- 2011 saw major exogenous changes in the price of natural gas from the perspective of the electricity generators due to the emergence of fracking. Second, we are primarily interested in the intensive margin response of electricity generators for our policy experiment, which we motivate and discuss below. The 2005-2011 time period is short enough to preclude any significant increase in natural gas generation capacity due to the price change in natural gas. We provide evidence that the decrease in natural gas prices did not spur a large increase in natural gas generation before the end of 2011.6 As a result, we are able to 4 McKinsey Insights and Publications interview with McKinsey partner Scott Nyquist. Available at http://www.mckinsey.com/insights/economic_studies/the_us_growth_opportunity_in_shale_oil_ and_gas. 5 A similar identification technique has been used in other contexts to identify marginal emissions given a set of input prices by Zivin et al. (2012). 6 Although, including 2012 does not change the qualitative findings of our results. 3
isolate the causal effect of short run price fluctuations insofar as it affects the dispatch of extant generation capacity. Our estimated marginal emissions vary dramatically over both the hours of the day within each of our eight study regions and across regions. Importantly, though, we focus on the change in marginal emission as a function of the relative input prices for coal and natural gas. In several regions, the pattern of marginal emissions across the day switched dramatically due to how the decrease in natural gas prices affected the dispatch order of electricity generators. We decompose the marginal emissions profile to show that observed changes are due to natural gas displacing coal-fired generation. Further, we show that the natural gas units being used on the margin had higher emissions rates after natural gas prices decreased relative to the earlier period of higher natural gas prices.7 Put another why, even within a fuel type, the emission rate of the marginal generator has changed in many regions and often became dirtier. As a result, we show that the marginal emissions profile has changed due to a change in the dispatch order both across fuels (e.g., coal versus natural gas) and across plants within fuels. We then evaluate how fracking’s effect on the marginal emissions profile has affected the profitability and pollution intensity of bulk electricity storage. Bulk storage allows electricity to be generated at one time of day and sold during another. The private benefit of bulk storage is that it allows inexpensive electricity to be dispatched during peak demand hours when wholesale electricity prices are high. This both increases profits for electricity generators who store and decreases peak hour wholesale electricity prices. Technological advances in battery technology are predicted to reduce the cost of building bulk storage facilities (Going with the Flow (2014)). Bulk electricity storage has well-understood non- market effects since inexpensive electricity produced during low demand hours at night could be dispatched during times of high demand during the daytime, thereby trading nighttime for daytime emissions (Carson and Novan (2013)). 7 This is possibly due to either less efficient natural gas power plants being able to compete with coal due to natural gas price decreases or peaking natural gas plants being used differently. We leave the theoretical underpinning of this result, though, to future research. 4
We evaluate bulk electricity storage because government support for its research and development is an ideal example of contemporary second best U.S. electricity sector pol- icy.8 The electricity sector is responsible for negative unpriced externalities like CO2 emissions.9 The optimal way to correct market failure associated with pollution emissions from the electricity sector is to tax carbon (or any other pollutant) at a price equal to its marginal external cost. However, a policy taxing carbon is politically challenging in the U.S. We do observe, though, direct subsidies to industries which are thought to have non- market benefits such as national subsidies for wind generation and state solar subsidies in California. As a result, bulk electricity storage is gaining support as one feasible technol- ogy to increase the profitability of some renewable electricity generation techniques like on-shore wind generation.10 Using our marginal emissions estimates, we are able to isolate how the decrease in natural gas prices affected the change in CO2 emissions attributable to bulk storage. We simulate the impacts (both private and public) of widespread installation of bulk electricity storage and estimate how the private and social benefits have changed due to the natural gas price decrease. Valuing CO2 emissions at accepted levels taken from the literature, the average annual external costs of bulk storage decrease by over 50% due to the decrease in natural gas prices (e.g., bulk storage is associated with a smaller increase in emissions with inexpensive natural gas). However, this change in external costs varies tremendously across regions. At the same time, the private benefits of bulk storage that have decreased since the on-peak wholesale price of electricity has fallen due to cheaper natural gas.11 The reduction in external costs varies between four and ninety-nine percent 8 There are two types of second best policies that regulatory bodies use to ameliorate emissions in the electricity sector. First, demand side polices seek to reduce electricity usage. One popular demand side policy in the U.S. is efficiency standards for durable goods. Second, supply side policies seek to alter the composition of electricity generation. For example, wind farm subsidies have led to increases in wind generation in the U.S. We focus on supply side policies in this paper. Most generally, then, we evaluate bulk electricity storage because it is an example of a second best policy which subsidizes a technology rather than prices an externality. 9 It is important to note that in the last 20 years, some externalities like NOx and SO2 are now regulated through cap and trade in specific areas of the country. 10 Although not adopted, the STORAGE Act introduced to U.S. Congress on May 20, 2009 offered large subsidies to installed storage capacity (Kaplan (2009)). 11 Due to the fall in wholesale electricity prices, the benefit of many investments in electricity generation, 5
depending on the region. We report similar qualitative results for other pollutants (SO2 and NOx ). Using disaggregated data on hourly electricity generation, we show that this change is caused by changes in the marginal fuel source across hours of the day over our study period. In sum, inexpensive natural gas has created an even more stark tradeoff between the private and public benefits of bulk electricity storage. There are several important implications of these results. First, while the EPA and in- dustry has documented the impact of natural gas broadly on emission trends the electricity sector, we are not aware of any study which carefully evaluates the effect of fracking on marginal emissions from the electricity sector.12 Given the level of attention paid to frack- ing, these estimates are important in their own right. Second, we provide evidence that the unanticipated decrease in natural gas prices has significantly affected the non-market benefits of bulk storage. Clearly, the social benefits of other energy sector policies have also been affected. More generally, we find evidence that evaluating the robustness of second best policy rankings due to input price fluctuations is a possibly overlooked criterion for policy makers. Third, our approach highlights the uncertainty in benefits from second best policies that support specific technologies in the energy sector as opposed to pricing exter- nalities directly. This is important since second best policies supporting specific durable goods technologies (like wind or solar subsidies) lead to long-lived capital which doesn’t quickly respond to changing market conditions, even if economic agents interacting with those public goods do. Fourth, our simulation evaluating the social and private benefits of bulk storage contributes to the growing literature which finds that addressing market failure by not pricing externalities directly exposes the economy to uncertain consequences (Davis (2008), Goulder et al. (2012) and Bento et al. (2013)). The remainder of the paper is organized as follows: section two provides background on marginal emissions in the electricity sector, bulk electricity storage and the emergence of fracking. Section three describes natural gas prices and generation capacity over our study period. Section four introduces our data and econometric model. Section five presents our including nuclear has likely decreased (Davis (2012)). 12 See http://www.epa.gov/air/airtrends/aqtrends.html#emission. 6
marginal emission estimates. Section six shows the results from our policy experiment. Section seven offers discussion and concluding remarks. 2 Background This section provides background on evaluating externalities in the electricity market. First, we discuss the importance of using marginal emissions in evaluating non-market effects of policy in the electricity sector. Second, we discuss the emergence of fracking technology used to extract natural gas. Finally, we discuss bulk electricity storage. Emissions from electricity generation account for roughly 33% of all emissions in the U.S (EPA (2013)). These emissions significantly impact air quality and subsequently hu- man health. In order to determine the marginal non-market effects of moving electricity production from one hour of the day to another, economists must estimate the marginal emissions at each hour. There is a robust literature highlighting the importance of esti- mating marginal emissions for analysis of electricity sector policies (Holland and Mansur (2008), Zivin et al. (2012), Cullen (2013), Novan (2012), Cullen and Mansur (2013), Kaffine et al. (2013)). We contribute to this literature by estimating how marginal emissions are influenced by changing relative input prices. Our identification technique is most similar to Zivin et al. (2012). We add to that identification strategy by incorporating additional fixed effects to account for variation in electricity demand patterns across days within a week. Our ability to identify how marginal emissions change over time is tied to quasi- experimental variation in the price of natural gas due to the rapid expansion of hori- zontal drilling and hydraulic fracturing (fracking) in the U.S. in the late 2000s. Horizontal drilling, in which a drill tip can bore thousands of feet underground and then turn at a near 90 degree angle and be driven horizontally for several thousand more feet allows a single drilling rig to drill as many as eight wells. Horizontal drilling has been combined with hydraulic fracturing, which involves pumping hydraulic fluid underground to frac- ture rock formations and release oil and natural gas. Fracking and horizontal drilling, used together, allow producers to extract small pockets of natural gas that are trapped 7
in rock formations. EIA (2012) reports that natural gas extracted in this way from shale deposits makes up just over a third of all natural gas produced in 2012, up from less than two percent in 2000. Total production increased by 25% over the same time frame with the increase coming entirely from shale gas extraction. Pennsylvania Department of Environmental Protection reports, for example, that the number of new Marcellus Shale wells drilled increased from roughly 200 in 2008 to over 2,000 in 2011. Not surprisingly this increase in production has been associated with a large decrease in market prices. Henry Hub natural gas spot prices peaked above $14 in 2005 with continued high prices through 2008 when prices began to fall rapidly. Spot market prices have been below $5 since 2009 and are forecasted to remain depressed for the foreseeable future as improved extraction technology allows more shale plays to be developed (see Figure 2 below). According to EIA data, since 1999 the electricity generation sector, along with the industrial sector, have been the largest consumers of natural gas in the U.S.13 The effects of the price shock in the electricity sector have been pronounced. Electric power plants are dispatched to meet inelastic demand from lowest marginal cost to highest marginal cost to a first order approximation.14 We show below that when gas prices were high, natural gas fired generation was used to provide peaking capacity during high demand hours while coal and nuclear fired generation served baseload.15 This led to high marginal costs during times of high demand and for existing natural gas generators to be fired for a small fraction of the year. As natural gas prices drop, the distinction between natural gas plants serving peak-load and coal plants serving base-load has blurred. We provide evidence that efficient natural gas plants now have lower marginal costs than some inefficient coal plants meaning that some of these coals plants are operating at less than peak capacity while natural gas plants are operating for a much larger fraction of the day. We show that natural gas price changes 13 see http://www.eia.gov/ “Natural Gas Consumption By End Use” data. 14 There are dynamic ramping costs for example which also affect the seasonality of marginal emissions. 15 Oil generation also provides some peaking capacity. 8
have significantly influenced the marginal emissions profile within a day and over space. Our policy simulation shows that bulk storage’s effect CO2 emissions varies dramatically due to this change in relative fuel input prices. We are primarily interested in how the private benefits and externalities associated with the implementation of bulk electricity storage would be affected by observed fossil fuel input price changes. Bulk electricity storage is an emerging technology in which large amounts of electricity is taken off the grid during times of low demand and stored. Stor- age is receiving increasing attention from government agencies: a December 2013 report from the U.S. Department of Energy (DOE) identifies the challenges and corresponding strategies so as to widely implement grid energy storage DOE (2013). The most common storage techniques include pumping water uphill (pumped storage) or as compressed air in an underground tank during times of low electricity demand when marginal costs of electricity production are low (e.g., at night).16 The stored electricity can then be sold back to the grid during higher cost hours at a profit. There are at least three attributes of electricity storage which affect welfare. First, when demand increases and the marginal cost of electricity is high (e.g., during the day), this stored energy is released by running water downhill through a generator or allowing compressed air to escape through a turbine. The resulting electricity is returned to the grid potentially decreasing the wholesale price of electricity during high demand hours. This has the potential to increase the profitability of on-shore wind facilities, which generate primarily overnight. Storage is also an attractive option for dealing with the intermittency of renewable electricity sources such as wind and solar power that may generate at times of low demand or stop production unexpectedly if a cloud bank appears or if the wind stops. Third, there is some discussion of how bulk storage may be used to lower CO2 emissions (Elkind (2010)). Storing electricity at night to be dispatched during the day involves trading emissions from nighttime electricity generation with emissions from day- time generation. This tradeoff is carefully evaluated in Carson and Novan (2013), which 16 Researchers are aggressively exploring large scale battery technologies, but they are not currently deployed to the grid in significant scale. 9
shows that bulk storage can interact with renewable generation to have significant effects on CO2 and other pollutants. Increasing capacity of energy storage will require a significant investment in the grid and storage subsidies given current technologies (DOE (2013)). In the absence of a tax on CO2 (or other pollutants) which aligns private and social costs, subsidizing bulk storage and investing in the grid amounts to a subsidy for electricity generators who would benefit from it, such as on shore wind. Our policy experiment attempts to decompose how cheap natural gas has affected both the private and external effects (mainly from CO2 emissions) of adding bulk storage to the U.S. electricity grid. 3 Natural Gas Prices and Electricity Generation In this section we demonstrate that when natural gas prices fell, that the change was not forecasted in commodities markets. We also show that electricity generators did not construct a significant amount of new generating capacity between the fall in natural gas prices and the end of our study period. For ease of exposition, in this paper we divide our sample period into two regimes: a high natural gas price regime and a low natural gas price regime. This is a a natural divi- sion as we show that the drop in natural gas prices was sudden and persisted throughout the second part of our sample period. We present results for marginal emission estimates in which prices enter directly into the regression specification in the appendix. Those results are consistent with our preferred high versus low price regime specification. We begin our sample period in January 2005 for two main reasons. It balances the sample given that we end the sample period in 2011 to focus on the short term response of electricity generators. Beginning in 2005 also avoids variation in environmental regulation. The Clear Air Interstate Rule (CAIR) Act introduced new environmental regulation for power plants in our study region.17 17 The CAIR Act was litigated for nearly a decade up to and beyond the passage of the act. Discussions with industry sources suggest that generators responded before the act was implemented despite the uncertainty of its legal status. 10
To determine when natural gas prices fell, we use a Markov Switching Model to identify when the cheap natural gas era began using collected daily data on Henry Hub natural gas spot prices from Bloomberg, excluding weekends and holidays. We’ve also estimated both a QLR and Chow test which give similar results. We estimate the following simple switching model: Ps,t = µs + s,t , s,t ∼ N (0, σs2 ) s = H, L (1) In equation (1), s indexes the state and t indexes time. We estimate a two by two matrix of transition probabilities (ρss for s = H, L) as well. In order to ensure a global maximum, we implement a two dimensional solver in which we assign values for µH and µL manually then estimate the other six parameters (σs2 , ρss for s = H, L). We then choose the model with the highest likelihood as the true model.18 The results from the Markov switching model are show in Figure 1. The top panel shows Henry Hub spot prices. The second panel shows the standard deviation in the system conditional on the estimated state. The third panel shows the probability of being in each regime. Note that t=1 is January 1, 2005 and t=1000 is January 6, 2009. The model selects µH = 8.0 and µL = 4.0 as the mean natural gas prices. ρ11 and ρ22 are both precisely estimated at one. The variance across regimes are also both precisely 2 = 4.463, σ 2 = .368. All estimated parameters are highly significant. The estimated: σH L relatively lower variance during the low price regime confirms the graphical depiction of lower volatility later in the data. Starting on January 8, 2009, the model is very confident in a sustained period of low gas prices interrupted by a fifty day span around t= 1200 (e.g., late 2009). Given the low estimated variance in regime two relative to regime one, the model selects prices in this interval to reflect the high price regime. We attribute this increase to seasonal demand for 18 We employ the two dimensional solver to ensure the search algorithm does not find a local maximum. The normality assumption is motivated by the noting that the first difference of the natural gas prices looks like a white noise process. 11
natural gas for heating.19 The differences in electricity generator behavior across these two natural gas price regimes will be the source of quasi-experimental variation that allows us to identify changes in marginal emissions in electricity market and the behavior of producers therein.20 There is one main concern with using Henry Hub spot prices. There are times of natural gas price differences across region within the U.S. often due to capacity constraints in pipelines. For example, the northeast U.S. sometimes incurs prices spikes in the winter when demand is high. However, these regional prices are highly correlated with the Henry Hub price. Further, we’ve estimated our main specifications removing both three and six months of data on either side of the January 8, 2009 date found in the MSM and the qualitative results do not change.21 Those results are available from the authors upon request. Lastly, since we estimate the main econometric model at the NERC region level with a large number of time specific fixed effects, we control for these systematic differences across regions within a natural gas price region. The more important question for the current paper is that the drop in natural gas prices was not predicted by the market. If the drop in natural gas prices in this market was not predicted, it gives us quasi-experimental variation in input prices needed to identify the causal effect of input price variation on marginal emissions given a composition of electricity generation. We then use those estimated marginal emissions curves in order to simulate the difference in non-market impacts from bulk-electricity storage. One convenient way to determine if the drop in natural gas prices we predicted by the market is to look at the difference between futures prices and subsequent spot prices. If these two measures diverge it is evidence that spot prices were determined by un- forseen events. Figure 2 shows both spot and futures prices for natural gas (measured 19 We also performed a rolling Chow test on first differenced spot and futures natural gas prices. We run the first difference of the same sequence of Henry Hub natural gas spot prices on seasonal dummies and a time trend. We then create a dummy variable equal to one if the time period is after the date indicated. There is evidence of a break in prices between March and May 2009. We have estimated all of our models below using alternative break dates in that range and all results are qualitatively identical. 20 In an appendix we show the these results are robust to several other definitions pre- and post-fracking era. 21 The distinction between the two regions becomes even more pronounced. 12
in $/MMBtu) from January 2005 - January 2014. These data are taken from Bloomberg and show Henry Hub prices for the U.S. traded on the New York Mercantile Exchange (NYMEX). The “n” month future price is plotted on the day it was traded (rather than the date it was scheduled to be delivered). As a result, Figure 2 should be interpreted as follows: when the “n” month futures price lies directly on top of a spot price it means that the market expects the price of natural gas to be exactly the current spot price in “n” months. As a result, if the 6-month futures price was above spot price on a given date, it means that the market expects the price of natural gas to be higher than the current level in six months. Figure 2 shows that in the second half of 2008 and through 2009, futures prices were significantly higher than spot prices for natural gas. Put another way, the markets ex- pected that natural gas prices were going to be higher in the future rather than lower. Specifically, in late 2008, the 18 month future contract of natural gas was not significantly lower than the spot price, which ranged from $9 to $5.50. We take figure 2 as evidence that electricity generators could not have predicted the sustained period of low natural gas prices and built new capacity within our study period. Figure 2 provides convincing evidence that the decrease in natural gas prices was not predicted in commodities markets. It is still possible, though, that electricity generators could quickly respond to decreases in natural gas prices once they do fall. If natural gas ca- pacity expanded, this would be problematic for our study: it would mean that determining non-market impacts of electricity storage are due to a mix of intensive and extensive mar- gin responses. We isolate the intensive margin effect since investments associated with second best policies, like subsidizing bulk electricity storage, is often slow moving and long-lived and doesn’t quickly respond to changing market conditions, even if economic agents interacting with those public goods do. As a result, isolating the intensive margin response is important given our research question.22 22 Another identifying assumption we need is that generation supplied by hydroelectric and renewable sources does not systematically change with natural gas prices. This is very unlikely given that hydroelectric dispatch responds to electricity demand, which is inelastic, and renewable generation is a function largely of weather, which is exogenous. 13
Figure 3 summarizes changes in electricity generation capacity from 2005-2011. Nat- ural gas capacity increased by roughly 6.5% over this time period and Figure 4 shows that much of this increase was in Florida. The figure shows that increases in natural gas capacity were occurring during the study period but the rate of change in natural gas generation was roughly constant when futures markets began to predict decreases in natural gas prices in 2009.23 If natural gas capacity where to have increased in response to decreased natural gas prices observed in 2009, then there would be a significant increase in the growth rate of installed capacity in 2010 or 2011. Instead we observe no significant deviation from the pre-trend growth rate. Coal and nuclear fueled electricity generating capacity is constant and oil fueled capacity drops by twelve percent. Total generating capacity increases by nearly four percent from 2005-2011 (compared to nine percent from 2002-2005). To put Figure 3 into context, total electricity demand was flat to decreasing over this time period. Figure 4 shows the spatial distribution of the increase in natural gas capacity between January 2007 and December 2011 taken from the EIA 860 and 923 forms. We limit the increase in capacity from 2007-2011 since it is entirely implausible that futures market predictions could have affect increases in natural gas capacity in 2005 or 2006. Units are displayed in 1000s of MWs.24 Figure 4 shows that much of the increase in capacity took place in Florida (FRCC) with more modest increases elsewhere. We show below that Florida had the largest share of oil-fired generation throughout the mid-2000s in the U.S. While beyond the scope of this paper, as oil prices rose in the mid-2000s the incentive to build new more efficient natural gas price may have been sufficient to motivate new investment. In sum, then, Figure 3 and Figure 4 show that there doesn’t appear to be any large increase in installed natural gas fired capacity before December 2011. There is no evidence that there was any systematic increase in coal power plant retirements over 23 While beyond the scope of this paper, the increase in natural gas capacity may be due to the imple- mentation of air quality trading rules such as CAIR in the northeast. 24 According to census and DOE data, per capita electricity use is .0014 MWs/hour. 1,000 MWs implies that there is enough power to supply 714,285 people with their average electricity or a bit more than .5% of the U.S. population. 14
the same period; even in 2012 after the industry had an extra year to respond to cheap natural gas net coal plant retirements were less than 8,000 MWs in capacity for the entire nation.25 We take this as evidence that our empirical strategy isolates the intensive margin response of the electricity sector to decreased natural gas prices. Figure 5 illustrates the interconnections which are essentially isolated from each other electrically. The Eastern Interconnection is further divided into six sub-regions across which electricity flows are small but nontrivial. The Western Interconnection (WECC), Texas Interconnection (TRE) and the six sub-regions of the Eastern Interconnection make up the eight NERC regions. Table 1 describes the Interconnections and NERC regions in more detail. Following Zivin et al. (2012), we use these NERC regions as our unit of analysis throughout the paper. The EIA data also contains information on nameplate capacity by fuel type and gen- erator type. Importantly, there is significant variation in electricity generation capacity by source. Figure 6 shows total installed capacity as of 2006 by fuel type as a percent of total capacity by National Electricity Reliability Council (NERC) region. The top panel demonstrates that different areas had significant heterogeneity in capacity for generation of different fossil fuel generation. For example, MRO (upper midwest) and RFC (Ohio Valley to New Jersey) had a large amount of coal relative to FRCC (Florida) and TRE (Texas) which had significant natural gas capacity.26 The bottom panel shows breaks down 2006 natural gas capacity by generator type. Combined cycle (CC) capacity is sig- nificantly more efficient than gas turbine (GT) generation (EIA (2012) Table 8.2). As a result, it is reasonable to expect areas with significant CC natural gas capacity (FRCC and NPCC) to response significantly more to the change in natural gas prices than less combined cycle capacity (e.g., MRO). Lastly, Figure 7 shows the monthly generation electricity by fuel type by NERC region aggregated from EPA’s hourly CEMS generation data. To construct Figure 7 we merge in 25 See http://www.eia.gov/ Annual Electricity Summary Table 4.6 “Table 4.6. Capacity Additions, Retirements and Changes by Energy Source, 2012 (Count, Megawatts)”. 26 The west region WECC has significant hydroelectric power capacity as indicated by the “other” cate- gory. 15
NERC region identifiers from the EIA 923.27 It is clear that in areas with significant coal generation, the share of coal generation relative to natural gas decreased toward the end of the sample once natural gas price dropped. The figure shows that natural gas generation increased over the same time period. This is consistent with natural gas generation moving further down in the dispatch order. The goal of the remainder of the paper is to show how this relative input price change affected marginal emissions and how those changes affect the welfare effects of second best policies. 4 Data and Econometric Model In order to estimate hourly marginal emissions by U.S. region, we need hourly data on electricity generation and emissions. Fortunately, hourly data on fuel consumption, elec- tricity production and pollution emissions for power plants are available through the EPA Clean Air Markets program. As a result, we observe hourly generation and emissions for almost all electricity generating units in the U.S. over our sample period. This allows us to estimate marginal hourly emissions and marginal fuel source. This section discusses our data and econometric specifications for estimating marginal emissions by hour and marginal fuel type by natural gas price regime. 4.1 Data Electricity generating units at every fossil fuel burning power plant with a capacity of greater than twenty-five MWs must install a Continuous Emissions Monitoring System (CEMS). The systems sample power plant smokestack air frequently to calculate the amount of SO2 , NOx and CO2 emissions.28 The data is primarily used by EPA to con- firm that plants are complying with their obligation to purchase pollution permits in the SO2 and NOx markets to cover all their emissions. The CEMS data also includes the primary and secondary fuel type of the plant along with a variety of other plant attributes 27 In the few cases where matching was impossible due to mismatched or incomplete ORISPL codes, we matched plants in the CEMS data to NERC regions manually. 28 CO2 emissions are imputed for fossil fuel fired power plants based on fuel inputs. 16
useful in identifying the location and ownership of the facility. The data set does not include nuclear, hydroelectric or renewable generators, but these producers have low or zero marginal costs and no air pollution emissions so we exclude them from the analysis for the main analysis.29 From the perspective of our policy simulation, though, this is precisely the marginal emissions estimate we wish to consider. The dataset also excludes a portion of gross generation from combined cycle units. We are not aware of any work which addresses the magnitude of this missing generation. The data, though, do contain the appropriate level of emissions from these units. As a result we record the “effective” emission rates of these units which is the appropriate variable from a policy perspective. The combined U.S. data set consists of approximately 4,600 generating units (the number varies slightly over the sample period) at over 1,200 facilities. Each unit is observed hourly over the sample period 2005-2011. To avoid duplicate date and time observations we average the repeated hours at the end of daylight savings time in the fall when the clock “falls back.” This produces approximately 61,300 observations for each unit in the data set. We follow Zivin et al. (2012) by aggregating data to conduct our analysis at the National Electricity Reliability Council (NERC) region level to ensure that we are able to identify the marginal fuel for changes in demand within regions. We estimate the model at both the NERC and interconnection level. In our interconnection level specifications we implicitly allow trading within an interconnection as in Zivin et al. (2012). 4.2 Econometric Specification We estimate a fixed effects model to identify marginal emissions over all eight NERC regions across both the high and low natural gas price regimes. The specification employs a semi-parametric approach to estimate marginal emissions over hours of the day for each NERC region. We call this a semi-parametric approach because we estimate marginal emissions separately for each hour of the day. In order to remove noise from our estimates we include a wide variety of fixed effects to control from ramping effects, day of week 29 In the robustness checks section we will evaluate the potential impact of renewable generation on the relationship between generation and pollution. 17
effects and month by year variation. To that end, we estimate the following model: Eh,n = βh,r,n (loadh,n ∗ hourh ∗ regimer ) + γn (month ∗ year ∗ hour ∗ dow) + h,n (2) Eh,n represents aggregate hourly CO 2 emissions measured in tons for all generators in the NERC region. We also report results for SO2 and NOx in the appendix. The loadh,n variable describes the total hourly fossil fuel generation in the region, h indexes hour, n denotes NERC region and regimer is a variable indexing whether the observation occurs within the high or low natural gas price regime. The γ vector is a set of 14,112 month- by-year-by-hour-by-day of week fixed effects (e.g., 14,112 = 7*12*24*7). To address serial correlation concerns we report standard errors clustered by date. Although any systematic differences would be removed by our fixed effects, we verify in the appendix that average generation within each NERC region-hour stays roughly constant over the entire study window. While some differences do exist, they are of second order importance for the reasons we discuss there. Gas fired power plants purchase much of their fuel on spot markets, but some employ longer term contracts. Most coal is purchased using long term contracts. Even though the spot market price of natural gas decreased starting in early 2009, the average relative price paid by electricity generators may have taken longer to adjust.30 Using the high versus low price regime, as opposed to including prices directly in the estimating equation, avoids the the issue. The coefficients of interest in equation (2), βh,r , represent the average hourly marginal emissions in an hour for a natural gas price regime for a particular region. Comparing coefficients in an hour across regimes reveals the changes in marginal emissions between the high and low natural gas price regimes. Using the complete set of fixed effects in equation (2) is important: our coefficient estimates are identified from variation in emissions for 30 The timing of spot input price change pass through is an important question in its own right but that we leave to future work. 18
the same hour of the day for the same month-year in the same day of the week within a natural gas price regime.31 For example, variation in emissions on Mondays in January at 8am in 2007 in NERC region n identify the coefficient on β8,1,n .32 If marginal emissions are indeed affected by input price changes, it should mean that the composition of fossil fuel electricity generation also changed. As a result, we would expect the fuel source on the margin to vary over the natural gas price regime. We therefore decompose the changes in our estimated marginal emissions rates into two components: the change in the marginal emissions rate of a particular fuel and the change in the fuel sources on the margin. The results allow us to pinpoint the source of the observed change in marginal emissions rates. If changes in fuel price have driven changes in marginal emissions rates, it should be reflected in the fuel type on the margin. Alternatively, emission rates could change due to newly installed abatement technologies or, in the case of coal, plants could switch their fuel inputs to lower sulfur coal types.33 This strategy, then, allows us to determine the change in fuel which occurs exclusively to the fuel type used for generation. We first estimate the the marginal emissions from each fuel source in each NERC region across the high and low natural gas price regimes: Eh,n,f = βh,r,n,f (loadh,n ∗ hourh ∗ regimer ) + γn,f (month ∗ year ∗ hour ∗ dow) + h,n,f (3) The β coefficients represent the marginal CO2 emissions from a fuel source, in a region, for a natural gas price regime in a given hour for fuel type f . Changes in these marginal emissions rates would suggest that abatement technologies or within fuel changes in relative 31 Our specification also has that differences in load for the same hour will identify our coefficient βh,r . As a result, the level of load in a particular hour is not important for our identifying our coefficient. This alleviates concerns about possible bias emerging from changing levels of generation within an hour over time (e.g., due to the recession) influences coefficient estimates in any significant way. To further show this recession effect is not a concern, in the appendix we estimate the same specification without July 2008-June 2009 and find almost identical results. 32 In an appendix we estimate the marginal emissions separately for each month of the year across regimes for one NERC region. The estimates show significant variation in monthly marginal emissions rates. When evaluating energy policies that have a differential impact across seasons these differences may prove important. 33 Even here, though, there could be an effect due to changing input prices: if emissions rates are heterogenous across fuels due to variation in technical efficiency, then changing the order in which plants of a particular fuel type could also affect the marginal emissions rates from that fuel. 19
operating costs have altered the marginal emissions rate from a particular fuel type. Such changes cannot be directly attributed to changes in fuel price. We then estimate the fuel source at the margin for each NERC region across the high and low natural gas price regimes. We estimate the following specification for each fuel type in each NERC region: Loadh,n,f = βh,r,n,f (loadh,n ∗hourh ∗regimer )+γn,f (month∗year∗hour∗dow)+h,n,f (4) where f denotes fuel type (oil, natural gas or oil). Loadh,n,f represents the generation from a particular fuel type during a given hour in a NERC region whereas Loadh,n rep- resents the total generation during a given hour in that NERC region. The coefficients of interest, βh,r,n,f , therefore represents each fuel’s share of marginal generation during hour h. Changes the composition of marginal fuel across natural gas price regimes can be directly attributed to changes in operating costs of the various fuel types since changes in input prices directly affect the marginal costs of generation and therefore the dispatch order of electricity generation by fuel type.34 5 Results In our results section, we report almost all of our regression output as figures due to the large number of coefficients and fixed effects. Each model includes seven years of hourly data and thousands of fixed effects ensuring that there is very good explanatory power and all of our coefficient estimates are highly significant by traditional measures. Regression output is available from the authors upon request. Figure 8 reports the estimated marginal emissions rate by hour for both the high and low natural gas price periods. Each panel on the graph represents a single NERC region. 34 The marginal costs of operating a fossil fuel plants could also be affected by environmental regulation. In an appendix we should that the costs of complying with the Clean Air Act Amendments emissions trading programs dropped during the low natural gas price regime. That suggests that the relative change in input prices across regimes actually overstates the relative costs changes and our estimates should be considered a lower bound on the impact of the natural gas price change alone. 20
In each panel, there are two lines that represent the marginal emissions rates in each hour for a given natural gas price regime. On each line there are twenty-four point estimates of the marginal emissions rates with error bars representing the 95% confidence interval range. Each panel was created from separately estimating equation 2 for each NERC region clustering standard errors clustered at the day level. There is variation in marginal emissions rates across hours of the day within a NERC region, within hours and natural gas price regimes within a NERC region and across NERC regions more generally. The largest source of variation in marginal emissions rate is geographical. Regions with significant coal fired generation capacity (the Upper Midwest) tend to have higher marginal emissions rates than those with more natural gas capacity (Texas and the Western Interconnection). In the high natural gas price regime, most regions’ marginal emissions rates follow the same hourly pattern: emissions are highest during the overnight hours when demand is low and inexpensive coal generation is on the margin. Marginal emissions rates then drop through the peak demand hours in the mid-afternoon as more expensive natural gas generation is brought online. The relative difference in emissions rates between overnight and peak hours are largest in the southeast (SERC) and mid-Atlantic (RFC) regions. In both of these regions, there is significant coal generation capacity. The drop in natural gas prices reversed the hourly pattern of marginal emissions in several regions during the high natural gas price period. In four of the six eastern NERC regions marginal emissions are lower during the overnight hours than they were during the high natural gas price regime. It seems possible that the most efficient natural gas has displaced the least efficient coal production and is more likely to be on the margin. In these regions, marginal emissions during peak hours are higher under the low gas price regime. It is possible that coal plants (and in the case of the Northeastern states oil fired plants) are more frequently on the margin during peak hours. In Florida, Texas and the Western Interconnection (FRCC, TRE and WECC) a similar pattern holds although the marginal emissions curves do not cross. Cheap natural gas has led to a level shift in marginal emissions rates, although in all three regions the difference in emissions rates between the highest and lowest demand hours has decreased. Cheap 21
natural gas has flattened the marginal emissions profile in all regions reducing the intra- day variation in emissions rates. Controlling for input prices is at least as important as controlling for time of day when evaluating the environmental impacts of energy policy. Lastly, in almost every region during the low natural gas price regime between hours nine and ten, marginal emission rates jump up. Between hours 23 and zero marginal emission rates jump down. This is an artifact of estimating marginal emissions at the yearly level; in the appendix we show that when marginal emissions are estimated within a month, this peculiarity is eliminated (see Figures 3 and 4 in appendix). We attribute this aggregation issue to a systematic relationship between marginal emission levels varying with total generation levels, how consistent generation levels are in high generation months versus low generation months, and ramping.35 We address this directly in the policy experiment below and do perform a generalized policy experiment with monthly marginal emission estimates in the appendix. We now decompose the changes in estimated marginal emissions rates into the changes in marginal emissions rates of each fuel type to determine if coal generation has in fact been displaced by relatively cheaper natural gas during the low natural gas price regime. We then estimate the frequency with which each fuel is on the margin. Table 2 summarizes es- timated marginal emissions rates (βh,n,f coefficients) for the three NERC interconnections (Eastern, Texas and Western) by fuel type when we re-estimate our main specification aggregating to the interconnection level.36 Aside from the aggregation technique, all other estimating procedures are the same as 2. Comparing the two left most columns in a row of Table 2 compares the marginal emissions rates from natural gas generation across input price regimes. Marginal emissions 35 Marginal emission point estimates are much more precise in summer months since overall fossil fuel generation is more consistent in summer months than in “shoulder” months. However, in those summer months total generation is also higher which affects the level of estimated marginal emissions. The reason is that fuel mix and efficiency of marginal generators changes across regimes. After natural gas prices fell, more natural gas is inframarginal, especially in summer months when generation is ramping up and down (e.g., between hours 9 and 10 in the morning and midnight and 1am in the evening). This systematic rela- tionship increases the OLS estimates away from the smooth downward marginal emissions trend observed in summer months within days in order to minimize squared errors. Therefore, when we aggregate to the year level there is a discrete increase and decrease as generation ramps up and down each day. 36 To conserve space we aggregate the six regions in the Eastern Interconnection into a single region. 22
rates from natural gas increased significantly from the high to low natural gas price regimes in the Eastern Interconnection. Each hour’s marginal emissions rate is very precisely estimated and statistically significantly different from the same hour and fuel’s emissions rate. Emissions rates for natural gas fired generation in the Eastern Interconnection increased by between fifty and sixty percent. Differences within fuel type across input price levels are much smaller for each of the other Interconnections.37 This is possibly due to much higher proportion of relatively inefficient gas turbine natural gas capacity being on the margin during the low natural gas price period, although further inquiry along this line is needed. Finally, we estimate equation 4 for each NERC region and find the fraction of time each fuel is on the margin by hour for each input price regime. The results are reported in Figure 9. The panels display the point estimate of each type of fossil fuel. We restrict the sum of the coefficient point estimates to be one in the estimation procedure, although relaxing that assumption doesn’t significantly change the qualitative results. All point estimates are highly significant and they are very often significantly different across hours within a price regime. Each panel represent a single NERC region. Within each panel there are two graphs, each with twenty-four stacked bars that sum to 1. The left graph is displays the fraction of time each fuel is on the margin in the high natural gas price regime and the right graph presents that fraction for low natural gas regimes. In every NERC region coal is less frequently on the margin during the overnight hours during the low natural gas price regime. During peak demand hours coal is more fre- quently on the margin. In the northeast (NPCC) and Florida (FRCC), both of which have significant oil fired capacity, oil generation is less likely to be on the margin during the low natural gas price regime. As natural gas fired generation becomes less peak fol- lowing in each region the variation in the fuels on the margin within a region, by hour, drops and the marginal emissions profile across the day flattens. As a result, we conclude 37 The Western states (WECC) had some oil fired generation on the margin during the high natural gas price regime, but during the low gas price regime oil is no longer on the margin. Texas has no oil fired generation (TRE). 23
that the changes in marginal fuel appear to be driving the changes in observed marginal emissions. 6 Policy Experiment The marginal emissions estimates provide evidence that the environmental impacts of elec- tricity prices are sensitive to input prices. As a result, any energy policy that differentially impacts generation across hours of the day will have different non-market effects as rela- tive input prices changes. To illustrate this point we simulate introducing bulk electricity storage to the grid in each NERC region in both the high and low natural gas price eras. The prospects of subsidizing bulk electricity storage is an ideal example of contempo- rary U.S. electricity sector policy. Currently it is not profitable for private firms to build bulk electricity storage (Going with the Flow (2014)). However, bulk electricity storage would increase the profitability of renewable electricity sources like onshore wind genera- tion by increasing the revenue firms would earn by selling their electricity during the day when wholesale prices are higher. As a result, subsidizing bulk storage would act as a subsidy for wind generation. Directly subsidizing particular technologies with desirable non-market features (like wind or solar) is a defining characteristic of contemporary U.S. energy policy. However, subsidizing bulk storage would also create the opportunity for non-wind generators who produce over night hours to increase their profitability as well. As a result, we calculate both the private viability and the the environmental impact of implementing bulk electricity storage and illustrate the variation in both across regions as a function of natural gas prices. We simulate installing a small amount of high capacity electricity storage in each NERC region. We follow Carson and Novan (2013) and assume that the storage can charge and discharge completely in one hour.38 As stated above, private electricity generators efficiently utilize bulk storage by producing electricity when both demand and wholesale price are low during overnight hours. Bulk storage allows them to save that electricity and 38 This is a reasonable assumption given the levels we consider here. 24
You can also read