The Competitiveness Impacts of Climate Change Mitigation Policies

Transcription

1 The Cometitiveness Imacts of Climate Change Mitigation Policies The Harvard community has made this article oenly available. Please share how this access benefits you. Your story matters Citation Aldy, Joseh, and William Pizer The Cometitiveness Imacts of Climate Change Mitigation Policies. HKS Faculty Research Working Paer Series RWP11-047, John F. Kennedy School of Government, Harvard University Published Version htt://web.hks.harvard.edu/ublications/workingaers/ citation.asx?pubid=8115 Citable link htt://nrs.harvard.edu/urn-3:hul.instreos: Terms of Use This article was downloaded from Harvard University s DASH reository, and is made available under the terms and conditions alicable to Other Posted Material, as set forth at htt:// nrs.harvard.edu/urn-3:hul.instreos:dash.current.terms-ofuse#laa

2 The Cometitiveness Imacts of Climate Change Mitigation Policies Faculty Research Working Paer Series Joseh E. Aldy Harvard Kennedy School William A. Pizer Duke University National Bureau of Economic Research December 2011 RWP The views exressed in the HKS Faculty Research Working Paer Series are those of the author(s) and do not necessarily reflect those of the John F. Kennedy School of Government or of Harvard University. Faculty Research Working Paers have not undergone formal review and aroval. Such aers are included in this series to elicit feedback and to encourage debate on imortant ublic olicy challenges. Coyright belongs to the author(s). Paers may be downloaded for ersonal use only.

3 THE COMPETITIVENESS IMPACTS OF CLIMATE CHANGE MITIGATION POLICIES Joseh E. Aldy and William A. Pizer* December 2011 Draft Abstract The ollution haven hyothesis suggests that unilateral domestic emission mitigation olicies could cause adverse cometitiveness imacts on domestic manufacturers as they lose market share to foreign cometitors and relocate roduction activity and emissions to unregulated economies. We construct a recise definition of cometitiveness imacts aroriate for climate change regulation that can be estimated exclusively with domestic roduction and net imort data. We use this definition and a 20+ year anel of 400+ U.S. manufacturing industries to estimate the effects of energy rices, which is in turn used to simulate the imacts of carbon ricing olicy. We find that a U.S.-only $15 er ton CO2 rice will cause cometitiveness effects on the order of a 1.0 to 1.3 ercent decline in roduction among the most energy-intensive manufacturing industries. This amounts to roughly one-third of the total imact of a carbon ricing olicy on these firms economic outut. JEL Codes: Q54, Q52, F18 * Aldy is affiliated with Harvard University, Resources for the Future, and the National Bureau of Economic Research. Joseh_Aldy@hks.harvard.edu; ; Harvard Kennedy School, 79 JFK Street, Mailbox 58, Cambridge, MA Pizer is affiliated with Duke University and the National Bureau of Economic Research. We benefited from the excellent research assistance rovided by Evan Herrnstadt. This research was suorted by the Pew Center on Global Climate Change and Resources for the Future. Elliot Diringer, Garth Heutel, Trevor Houser, Arik Levinson, Joanna Lewis, Steve Lin, Carol McAusland, and David Po, and seminar articiants at the 2007 Mannheim Climate Policy Network meeting, the 2008 Pew Center on Global Climate Change Business Environmental Leadershi Council meeting, Resources for the Future, Duke, UNC, and Harvard rovided helful comments on an earlier version of the aer. All errors and omissions remain the resonsibility of the authors.

4 I. Introduction The debate over climate change olicy has largely focused on the design of instruments that will imose a rice on the emissions of carbon dioxide (CO 2 ) and other greenhouse gases. In the context of this debate, attention has turned to the rosect that olicy instruments such as ca-and-trade and emission taxes could cause adverse cometitiveness effects for energy-intensive firms in develoed countries, such as in Western Euroe and the United States, if they move forward with mitigation efforts while major develoing countries ostone action. 1 The concerns about cometitiveness are consistent with the ollution haven hyothesis that suggests that firms relocate economic activity from high regulatory cost to low regulatory cost countries. While sometimes framed as a jobs versus the environment question with regard to conventional ollution (Morgenstern et al. 2002), this effect is esecially troubling in the context of climate change olicy. The relocation of economic activity would increase CO 2 emissions in develoing countries, thereby undermining the global environmental benefits of the develoed country's emission mitigation olicy. That is, it is a jobs and the environment roblem. In this aer, we resent evidence that the cometitiveness imacts of carbon ricing would reduce roduction by about 1 ercent, reresenting a small share erhas one-third of the overall imact of recent climate roosals on energy-intensive industries. To draw these conclusions, we begin with a formal definition of adverse cometitiveness effects, a definition that is frequently unclear in existing studies. We 1 A variety of energy and climate olicies could cause adverse cometitiveness imacts by raising the cost of using fossil fuels, including state ca-and-trade rograms (such as in California and the northeastern states), state renewable and alternative energy mandates in the ower sector, and greenhouse gas regulatory mandates under the Clean Air Act. 1

5 then derive an estimable exression for this definition and use data to calculate the effect for a unilateral U.S. emission mitigation olicy on the U.S. manufacturing sector. Here, we emloy a somewhat novel emirical strategy that examines the historical relationshi between energy rices and roduction and consumtion in the U.S. manufacturing sector. Taking advantage of the fact that market-based CO 2 olicy instruments such as ca-andtrade and emission taxes oerate rimarily by raising energy rices, we use this estimation to infer the cometitiveness effect of U.S.-only CO 2 regulation. Our aroach uses idiosyncratic, within-industry energy rice variation to identify the cometitiveness effect defined in our theoretical section below. 2 This is akin to estimating the various elasticities used to run comutable general equilibrium models that have yielded revious economy-wide cometitiveness and emission leakage estimates, excet that we generate results in a reduced-form regression framework of equilibrium outcomes at a much more disaggregated level (4-digit industry). In articular, through interaction terms, we allow the estimated effects to vary with the energy intensity of roduction, allowing us to differentiate imacts among more or less energy-intensive industries. Our estimated model in hand, we simulate the imacts of a U.S.-only $15 er ton CO 2 rice, translated into the likely changes in energy rices. We focus on $15 er ton CO 2 because the energy rice changes are consistent with the observed variation in our historic energy rice data; $15 er ton is also in line with rices exected under various ca-and-trade and carbon tax legislative roosals in recent years. 2 As discussed below, we focus on consumtion and roduction, rather than net imorts directly for econometric reasons. 2

6 We find that the higher energy rices associated with this carbon rice would lead to a roduction decline of between 3 and 4 ercent among key energy-intensive sectors (e.g., iron and steel, aluminum, cement, etc.). We also find, however, that this energy rice increase would cause a 2 to 3 ercent decline in consumtion (defined as roduction lus net imorts). The decline in consumtion reflects efforts to economize on the use of energy-intensive manufactured commodities in end-use roducts and substitution to lessenergy-intensive inut (more below). 3 This suggests that an emission mitigation olicy would induce a roughly 1 ercent shift in roduction overseas -- our estimated adverse cometitiveness effect. Put another way, as a share of the total 3 to 4 ercent effect on roduction, the cometitiveness comonent is only about one-third. More broadly across the manufacturing sector, there is an interesting attern. Suly declines less for less energy-intensive goods -- but always declines. Demand, however, rises for the least energy-intensive goods, reflecting the substitution noted above. For these less energy-intensive industries, the cometiveness effect can actually be larger than the suly effect (net imorts rise by more than suly declines). Put another way, the energy-intensive firms that have remained in the United States may be more resilient to energy rice shocks than some of the less energy-intensive firms who actually see demand for their roducts rise. Quantitatively, the overall results suggest the cometitiveness effects associated with a $15 er ton CO 2 rice is consistently around 1 ercent, a shift suggesting that relatively small effect (comared to annual fluctuations) to industry at the rice levels we can study. 4 3 These estimated imacts may also reflect the limits to the caacity of foreign firms to suly more of these goods at a given rice in the short run. 4 To ut the 1 ercent cometitiveness imact in context, the absolute value of the annual real ercentage change in the value of shiments in the manufacturing sector averaged 8.8 ercent during our samle. 3

7 Our work builds on a substantial literature that has examined the more general question of whether environmental regulations adversely affect the cometitive osition of American industry. Numerous theoretical analyses have suggested that environmental olicy could create so-called ollution havens in develoing countries: The conventional wisdom is that environmental regulations imose significant costs, slow roductivity growth, and thereby hinder the ability of U.S. firms to comete in international markets. This loss of cometitiveness is believed to be reflected in declining exorts, increasing imorts, and a long-term movement of manufacturing caacity from the United States to other countries, articularly in ollution-intensive industries (Jaffe et al. 1995,. 133). Emirically evaluating this conventional wisdom has roven challenging (Jaffe et al. 1995; Levinson and Taylor 2008). A variety of factors may mitigate or dominate the effect of environmental regulatory costs in determining manufacturing location decisions. First, the availability of relevant factors of roduction, such as aroriately skilled labor, natural resources, and caital, can lay a more significant role than ollution control costs (Antweiler et al. 2001). Second, transortation costs may discourage relocation to countries far from the major markets for manufactured goods (Ederington et al. 2005). Third, firms with a significant share of their investments in large, fixed hysical structures also aear to move activity less in resonse to environmental regulations (Ederington et al. 2005). Fourth, roximity to firms that roduce inuts or urchase oututs e.g., agglomeration economies also discourages relocation (Jeesen et al. 2002). Some energy-intensive industries, such as iron and steel and aluminum, exerienced annual ercentage changes on average in excess of the manufacturing sector average. Other energy-intensive industries, including chemicals, aer, cement, and bulk glass, exerienced annual ercentage changes in value of shiments in the 5.6 to 8.0 ercent range, on average. 4

8 Since the most ollution-intensive industries tend to be relatively immobile by these measures of footlooseness, the emirical literature tyically finds quite limited imacts of environmental regulations on international cometitiveness. Recent research by Levinson and Taylor (2008) shows that U.S. ollution abatement costs in the 1970s and 1980s increased net imorts in the manufacturing sector from Mexico and Canada. The estimated increase in net imorts roughly equaled about 10 ercent of the total increase in bi-lateral trade for both Mexico and Canada, suggesting that other factors layed much more substantial roles in the evolution of trade among the North American trading artners. An extensive literature on the cometitiveness effects of variation in environmental olicies across the U.S. states has shown more significant imacts on domestic firm relocation resulting from variation in the stringency of environmental regulations (Henderson 1996; Greenstone 2002). Recent work by Kahn and Mansur (2010) finds even larger effects looking at adjacent counties. The larger domestic cometitiveness imacts may reflect the fact that labor costs and availability of caital do not vary much across the U.S. states and counties, and transortation costs are less imortant, relative to the international context. This emirical literature has focused on retrosective analyses of U.S. environmental regulations. The absence of a domestic CO 2 regulatory or taxation regime recludes us from taking exactly the same aroach. The oular alternative has been to use alied comutable general equilibrium models to simulate otential cometitiveness imacts of ricing carbon (IPCC 2001). While informative, this aroach suffers from assumed rather than estimated arameters and insufficient detail to break out differences in behavior or results among industries with different energy intensities. Indeed, it is 5

9 tyical to make a common set of assumtions that yield a common resonse across the entire manufacturing industry to a carbon ricing olicy. As our analysis shows below, this aroach can underestimate the imacts on the more energy-intensive manufacturing industries. Furthermore, these CGE models focus on aggregate estimates of emission leakage, not effects on individual industries. To motivate our emirical analysis, the next section resents an analytical model of the cometitiveness imacts of environmental regulations articularly the case of climate change, where we distinguish between the absence of foreign regulation (where global coordination on a regulatory regime is the benchmark) versus the resence of domestic regulation (where zero regulation is the benchmark). Section three resents our emirical methods and data. Then we resent the results of our emirical analyses of the relationshis between energy rices and roduction and consumtion. Section five illustrates the results of our simulation of a near-term unilateral U.S. CO 2 mitigation olicy on the U.S. manufacturing sector. The final section concludes with comments on future research and imlications for olicy design. II. Definition of the Cometitiveness Effect In order to define the cometitiveness effect, we have to consider the benchmark or counterfactual where we resume there is no such effect. Studies of the cometitiveness imacts of conventional local ollution regulation usually examine the effect of domestic regulation against a static, largely unregulated, environment. That is, the counterfactual is no regulation anywhere. However, an effective international climate change effort will require significant regulation by all countries. Recent studies show 6

10 that even reducing develoed countries greenhouse emissions to zero by 2050, will not be sufficient for global attainment of moderate mitigation goals if develoing countries take no action (Blanford et al 2009). For this reason, we emloy an assumtion of comarable regulation in all countries as our climate olicy benchmark. That is, we ask what haens when a articular country acts when everyone should be acting, versus conventional studies that ask what haens when a articular country acts against a backdro of no one acting. From a U.S. ersective, this benchmark means we need to distinguish between the effects on U.S. manufacturing associated with U.S.-only regulation versus a global CO 2 ricing regime. Global regulation would still cause a shift in roduction towards less carbon-intensive roducts and rocesses, with a corresonding decline in those roducts and rocesses with high emissions. If U.S. and foreign firms face comarable CO 2 ricing olicies -- effectively ensuring a so-called level laying field with resect to climate olicy -- then any decline in roduction in carbon-intensive U.S. firms would not be considered a cometitiveness effect. The true notion of a cometitiveness effect is therefore the difference between this outcome and what haens to U.S. firms with U.S.- only regulation. To further elaborate this definition, consider a simle model of market equilibrium: (1) D S, r NI, r US FOR 7

11 where D() is domestic demand as a function of domestic market rice, S(, r US ) is domestic suly as a function of domestic market rice and domestic regulation r US, and NI(,r FOR ) is foreign suly (net imorts) as a function of domestic rice and foreign regulation r FOR. We make the standard assumtions that D < 0, S > 0, NI > 0, S r < 0, NI r < 0 (that is demand is downward sloing, suly and net imorts are uward sloing, and regulation increases costs). Now imagine a global climate regime that increases r US to r r r for US US US domestic roducers and r FOR to r r r for foreign roducers. Taking the FOR FOR FOR total derivative of (1), we can solve for the change in rice under global regulation, 0 : (2) 0 S r D, rus r US NI r, rfor rfor S, r NI, r US FOR This imlies a corresonding change in domestic suly associated with global regulation S 0 : (3) S 0 D P Sr rus rus S, rus DP S, r NI, r, 0 US FOR S r, r US r US D P S P, rus S, r NI, r US FOR NI r, rfor rfor Note that the sign of exression (3) is ambiguous: the first term is negative and the second ositive. Unless foreign regulation has a larger effect than domestic regulation on 8

12 the domestic market, the net effect will be negative. This is shown grahically in Figure 1, where the left anel shows domestic suly and demand and the right anel shows net imorts. 0 is the vertical change in rice, and S 0 the horizontal change in suly, associated with global regulation. The horizontal lines across the two anels reflect the equilibrium rices, with and without regulation, where net imorts equal the difference between domestic demand and suly. While the general case is ambiguous, Figure 1 shows the (conventional) negative effect on domestic suly from global regulation. Assuming rʹfor and rʹus are considered aroriate resonses under a global climate agreement, we would not look at the decline S 0 as a cometitiveness effect. That is, it does not reresent an adverse effect on U.S. firms arising from the absence of regulations abroad. So where is the cometitiveness effect? Now consider what haens if there is no foreign regulation. From (2), we have a rice change from U.S.-only regulation of 1 : (4) 1 D Sr, rus rus S, r NI (, r ) US FOR and a domestic suly change from U.S.-only regulation of S 1 : (5) S 1 D P Sr rus rus S, rus DP S, r NI (, r, 1 US FOR S ) r, rus rus 9

13 This is shown grahically in Figure 2, where the right anel again shows net imorts and the left anel shows domestic suly. 1 is the vertical change in rice, and S 1 the horizontal change in suly, now associated with U.S.-only regulation. Given the second term in Equation (3) is ositive, we know that S 1 < S 0 and the difference is negative: (6) P SP, rus S, r NI, r S1 S0 NI r, D US FOR rfor r FOR This is what we define as the cometitiveness effect (CE) the equilibrium difference in domestic suly owing to the absence of foreign regulation. In Figure 2, where a small circle on the x-axis indicates suly with global regulation from Figure 1, this difference is labeled CE. Given this relates to the absence of foreign regulation, not surrisingly, we can understand this exression as the negative of the effect of foreign-only regulation on domestic suly (e.g., comare to Equation (3)). This is the missing element when the U.S. acts alone to regulate a global ollutant associated with goods that comete in a global market, absent comarable actions by other nations. While this measure deends on events abroad notably the vertical shift in net imorts, NI r (,r FOR ) r FOR it is fundamentally about the effect on U.S. firms. In the context of emirical analysis, if our data allowed us to construct a roxy measure of foreign regulation, we could quantify this reduced-form effect directly. We could estimate the coefficient on foreign regulation in a regression with domestic 10

14 roduction as the regressand and both domestic and foreign regulation as the regressors (e.g., Equation (3)). The roduct of this foreign regulation coefficient estimate and a value of the absent foreign regulation would yield our measure of a cometitiveness effect from comarable U.S.-only regulation. Unfortunately, such data are not available and, instead, we are confronted with the question of how we might estimate this effect in (6) with rimarily domestic data. One ossibility is to focus on the effect of U.S.-only regulation on net imorts. This has intuitive aeal it aears to be the shift overseas of roduction, emissions, and jobs, arising from U.S. regulation that fuels the rhetoric over cometitiveness in the first lace even if we know it is not exactly correct. From above, this measure equals: (7) NI, rfor S, r NI, r NI1 Sr, D US FOR rus r US How does this relate to the true cometitiveness effect in (6)? We can rewrite the two exressions (6) and (7) as: (8) S 1 S 0 D P S P, rus NI, rfor S, r NI, r US FOR NI r, NI r FOR rfor, rfor and (9) NI 1 D S, rus NI, rfor S, r NI, r US FOR Sr S, r US rus, rus 11

15 where the first term is oosite in sign but otherwise the same in both exressions and the second term equals the (negative of the) vertical suly shift associated with regulation in foreign (8) and domestic (9) markets (the change in suly divided by the dq/d sloe). Therefore, the effect of domestic regulation on net imorts will reflect the true cometitiveness effect to the extent the marginal cost increase is the same for domestic and foreign roducers. A larger cost increase for foreign roducers means we underestimate the cometitiveness effect; a smaller increase for foreign roducers means we overestimate the cometitiveness effect. Figures 1 and 2 show the case where these vertical shifts are the same and the change in net imorts with U.S.-only regulation equals the true cometitiveness effect. We believe it is reasonable aroximation to assume that domestic and foreign climate change regulation should have comarable imacts on the marginal costs of roduction for domestic and foreign manufacturers. First, it is likely that governments will imlement olicies that deliver comarable carbon rices. This may reflect a harmonized carbon tax, as some economists have advocated (Cooer 2007, Nordhaus 2007). It could reflect the linkage of domestic emission mitigation olicies that result in a common clearing rice in tradable allowance markets (Jaffe and Stavins 2010). It could also reflect imlicit rice coordination among nations as they develo and imlement their domestic emission mitigation olicies (Pizer 2007). The threat of imosing a carbon tax on imorts from unregulated foreign roducers may also induce regulatory convergence across nations. 12

16 Second, comarable carbon rices would likely yield comarable increases in the marginal cost of roduction in the manufacturing sector. The energy-intensity of manufacturing is fairly similar by industrial activity across develoed countries. Given the extensive investment in new manufacturing caacity in China over the ast decade, the characteristics of the roduction technology in China are aroaching those of the develoed world articularly those destined for cometitive exort markets. For examle, the energy intensity of advanced cement manufacturing in China exceeds the average international advanced cement manufacturing intensity by less than 6 ercent (Tsinghua University of China 2008). The energy intensity of blast-oven furnace steel manufacturing in China is a few ercentage oints better than that of the United States, although U.S. electric arc furnace technology still requires less energy than Chinese lants (Hasanbeigi et al. 2011). In the event that cost increases from a notion of equitable global regulation do differ significantly across nations, then our measures would yield a biased estimate. If foreign roducers have lower marginal comliance costs, then the rice increase from foreign regulation would be lower and the exression (7) would rovide an overestimate. If foreign roducers have higher marginal comliance costs (e.g., suose an identical carbon rice across nations raises roduction costs more in Chinese manufacturing because of higher energy intensities), then the larger foreign cost increase suggests (7) is an under-estimate of the cometitiveness effect in (5). In the end, we believe that, to a first order, cost increases are likely to be similar, and our use of net imort effects from U.S.-only regulation should rovide a reasonable estimate of the true cometitiveness effect. 13

17 III. Methods and Data for Emirical Analysis We are ultimately interested in a reduced-form estimate of the imact of U.S. regulation on net imorts in Equation (9), e.g. the coefficient on regulation in a regression with net imorts on the left-hand side. We do not directly estimate a net imorts regression, however, because of the large variation in industry size and the variation in the sign of net imorts in our data (which revents a direct log transformation). We considered two ossible alternatives: (1) estimating searate regressions in logarithms for domestic suly and demand, then looking at differences in relevant coefficients, and (2) estimating one regression using net imorts as a share of domestic suly (which has been the traditional aroach in the literature). Given NI = D S, the relationshi among these various aroaches and the imlied derivative of net imorts (with resect to regulation R) can be exressed as: (10) where NI/ R is the derivative of net imorts what we really care about, ( lnd/ R lns/ R) is the difference between the derivatives of logged demand and logged suly the estimate using aroach 1, and (NI/S)/ R is the derivative of net imorts as a share of domestic roduction the estimate using aroach 2. Both aroaches, examining NI/ R via consideration of ( lnd/ R lns/ R) or (NI/S)/ R, slightly misreresent what we really care about, NI/ R. This error is small when net imorts as share of domestic roduction (NI/S) is small, something true for 75 ercent of the industries in the samle (where we define small as ±15 ercent). While the results for these industries are similar using either aroach, the first aroach using 14

18 ( lnd/ R lns/ R) can be corrected easily as we have an estimate of lnd/ R. In addition, the second aroach breaks down when there are observations with S very small comared to NI, leading to unusually large swings in NI/S for small changes in domestic roduction, something that arises for 5 industries (where NI>2S) in our samle and requires those industries to be droed with the second aroach. For that reason, we focus on the first aroach using ( lnd/ R lns/ R) and consider the corrected estimates in Table 2 that resents the results of our carbon ricing simulation below. Having chosen the basic aroach, we estimate a two-equation system of regressions using a samle of more than 400 U.S. industries at the 4-digit industry (SIC 1972) level of disaggregation over the eriod. The basic regression secification takes this form of reduced-form estimates of a system of domestic suly and demand: (11) Yitk ik tk f ( rus it k, ; ) k X it itk where Y itk reresents the measure for outcome k the natural logarithm of suly and demand measures (S and D in equations (1-10)) for industry i and year t; the s are fixed effects for industries (i), and years (t); r, reresents the level of U.S. regulation US it the natural logarithm of the average electricity cost in 1987 dollars as discussed below; X it is a vector of additional determinants of the industry outcome measures, including average industry tariffs and factor intensity variables (to estimate the returns to human caital and hysical caital). 15

19 The two-equation system of regressions ermits correlation in the residuals, a factor that must be included when we calculate our net imort effect with arameters from both equations using a seemingly unrelated regression framework. In addition, we correct the standard error estimates to control for industry-secific heteroskedasticity. Energy rices serve as a roxy for the imact of a carbon ricing regime because ca-and-trade rograms and carbon taxes both would raise energy rices. In turn, we use electricity rices as our rimary measure of energy rices because electricity exenditures reresented a majority of energy exenditures for 88 ercent of all manufacturing industries in our samle. It is also an informative index of fossil fuel rices, since all three tyes of fossil fuels are used to generate electricity in our samle. In any case, we were unable to construct industry-secific rice measures for other fuels. 5 It is also worth noting that our use of energy rices as a roxy for regulatory stringency circumvents a number of roblems noted in the emirical ollution haven literature, which tyically use the ratio of regulatory comliance costs to value added as a roxy for the stringency of environmental regulations. Levinson and Taylor (2008) note that changes in roduction levels can affect this ratio of ollution abatement cost exenditures (PACE) to outut and create an endogeneity roblem. Production levels change this regulatory cost burden measure directly, as roduction or a related variable is the denominator of the PACE share. Production levels can also change the numerator of the PACE share indirectly, as changes in roduction affect lant turnover, scale economies, and the difficulty in meeting regulatory standards all of which affect 5 The NBER-CES manufacturing industry database rovides data on electricity exenditures and quantity of electricity consumed that allows us to construct an annual average electricity rice by industry. The Annual Survey of Manufactures collected only energy exenditures data, not quantities or rices of energy, for all other fuels. 16

20 regulatory comliance costs. In contrast, energy rices are unlikely to be endogenous to individual industry roduction decisions. Finally in our secification, industry fixed effects cature time-invariant characteristics of industries that may affect these measures of cometitiveness and year fixed effects account for common shocks, such as those from monetary olicy, world oil rices, etc. that affect all industries in a given eriod of time. Thus, identification is remised on idiosyncratic, within-industry electricity rice shocks, tyically driven by utility- and region- secific changes over time related to where industries are located. We consider various forms for the relationshi between U.S. regulation (e.g., electricity rices) and our left-hand side variables, ranging from a simle linear function of energy rices to flexible functions that allow the energy rice elasticities across industries to vary based on each industry s average energy intensity over the relevant samle eriod. We ultimately settle on a flexible cubic-sline aroach, although we introduce the results in the next section with simler aroaches to rovide context and motivation for the cubic sline. Intuitively, higher energy intensities imly larger cost imacts from rising energy rices. Viewed through the lens of carbon dioxide regulation, the very high ositive correlation between energy consumtion and carbon dioxide emissions imlies that energy intensity is effectively a carbon ollution intensity measure. Thus, a carbon ricing regime that imoses a common marginal cost on emissions will result in heterogeneity in the comliance costs er unit of outut across the manufacturing sector. Flexible estimation of the suly and demand elasticities as a function of this ollution intensity allows us to cature this effective comliance cost imact. 17

21 We use the value of shiments by industry from the NBER-CES manufacturing dataset develoed by Bartlesman et al. (2000) as our measure of domestic suly. 6 We define demand (consumtion) as domestic suly (roduction) lus net imorts, which we construct from the NBER trade database develoed by Feenstra (1996). As noted above, we undertake our analysis with these suly and demand measures in logarithms because of the significant variation in size of U.S. manufacturing industries. We define energy intensity as the ratio of all energy exenditures to value of shiments (constructed from the Annual Survey of Manufactures, multile years and Bartlesman et al. 2000). For each industry, we calculate the average intensity over 1974 to 1994, as well as subsamles for and as discussed in the next section. Figure 3 resents the cumulative distribution function for industry average energy intensity over We constructed electricity rices from NBER-CES data on electricity exenditures and quantity of electricity urchased. 7 We also control for average industry tariff rates, the hysical caital share of value added, and the human caital share of value added, consistent with Ederington et al. s (2005) analysis of the imacts of domestic environmental regulation on net imorts. The average tariff is exressed in ercentage oints, and reresents the average industry-level tariff based on the total duties collected multilied by 100 scaled by total customs value (constructed from data rovided by Magee and Feenstra et al. 2002). The hysical caital share is reresented by one minus the ratio of total ayroll to value added (constructed from data rovided by Bartlesman et al. 2000). The human caital share is calculated as total ayroll minus 6 All measures of outut, net imorts, and rices have been deflated to constant 1987 dollars. 7 We thank Wayne Gray for roviding data for

22 ayments to unskilled labor, scaled by industry value added. Payments of unskilled labor are estimated from the Current Poulation Survey as the number of workers, multilied by average annual income of workers with less than a high school diloma (constructed from U.S. Bureau of the Census, and Bartlesman et al. 2000). For constructing a consistent dataset, we emloyed several concordances made available by Jon Haveman. Let us exlain why we abridge our samle at Our imort data comes from Feenstra (1996), which rovides us with U.S. bilateral trade by 4-digit SIC through These data require transformation due to differences between the imort-based SIC codes (MSIC) and domestic-based SIC codes. Essentially, a number of SIC codes are defined by rocessing methods, and this information is unknown for imorts. Feenstra overcomes the differences in SIC and MSIC using a weighting matrix derived from data in the U.S. Census Bureau s U.S. Commodity Imorts and Exorts as Related to Outut. The Census Bureau notes that this ublication was discontinued because of a significant decrease in the Census Bureau's budget in 1996 and the conversion of the SIC to the new North American Industry Classification System (NAICS) starting with the 1997 roduction data. Since our average tariff rate and consumtion data are derived in art from these imort data, we cannot reasonably extend our samle beyond IV. Emirical Estimates of the Effects of Electricity Prices on Domestic Suly and Demand To rovide context and motivation for our referred flexible regression secification, we first resent simlified results for the domestic suly and demand models with and without linear interactions between energy rice and the historic energy 19

23 intensity of the industry. Without the interaction, this is akin to revious aers that regress domestic suly and/or net imorts on the level of environmental comliance costs or on the ratio of environmental comliance costs to the value of shiments. In both cases, we allow for the effect of the electricity rice to vary between the and the time eriods. We establish this distinction to account for the imacts of the eriod of higher energy rices ( ) on fuel switching (as the utility sector switched from etroleum to coal in ower generation in the late 1970s and early 1980s) and on investments in more energy efficient caital in the manufacturing sector. The more recent eriod may also better characterize the otential imacts of a carbon ricing regime on the manufacturing sector. The left half of Table 1 shows results without including energy intensity and is comarable to revious work, for examle Levinson and Taylor (2008), Ederington et al. (2005), and Grossman and Krueger (1991). In each of these three aers, the ratio of net imorts to value of shiments is regressed on the ratio of ollution abatement costs to value of shiments (or value added), as well as other controls that enter the regression equation linearly. The estimated suly and demand elasticities with resect to electricity rices are quite small, with the suly elasticities about We cannot statistically distinguish the suly elasticity from the suly elasticity. The demand elasticities, interestingly, are statistically significant but the latter eriod has a ositive sign. Our more flexible regression secifications (both the right half of Table 1 and Figure 5 below) reveal a attern where, in resonse to higher energy rices, demand rises for less energy-intensive roducts while demand for more energyintensive roducts declines. A model that restricts the resonse to be the same across all 20

24 industries, however, ends u being weighted toward the (more numerous) less energyintensive sectors; hence the ositive sign. Following the suly and demand estimates, we resent the difference, reresenting the effect on net imorts and, in turn, the otential cometitiveness effect. Note that the standard error of the difference is considerably smaller than the standard errors of the searate estimates; this reflects significant ositive error correlation across the suly and demand equations. During the eriod, this secification suggests the reverse of a cometitiveness effect; higher domestic energy rices lead to a modest decline in net imorts. In the eriod, we see a more conventional estimate suggesting net imorts rise with higher domestic energy rices. As noted, we exect the resonse to vary across industries based on energy intensities. The right half of Table 2 shows a simle attemt to cature this with the average energy intensity of each industry (calculated searately for the two sub-eriods) interacted with the electricity rice. The result is as we would exect: industries with higher energy intensity see more negative suly and demand resonses. Both interactions are highly significant. When we take their difference to comute the cometitiveness effect, however, the significance vanishes. A simle linear relationshi between the rice elasticity and energy intensity is inadequate. This motivates our use of a cubic sline secification for the relationshi between rice elasticity and energy intensity. We secify that the deendency of the energy-rice coefficient on energy intensity follow a restricted cubic sline with 5 knots at the 5 th, 27.5 th, 50 th, 72.5 th, and 95 th quantiles of energy intensity, as suggested by Harrell (2001). A restricted cubic sline has linear segments on either end, is connected by cubic 21

25 segments in the middle, and is twice differentiable everywhere. Given the high skewness of the data, we fit the sline in terms of the log of energy intensity. We resent the results of the flexible regression secifications grahically in lieu of a table of regression coefficients because of the difficulty of interreting the sline coefficients. 8 We focus our resentation in the aer on the results. 9 Figures 4 and 5 resent the energy rice elasticities from our domestic suly (roduction) and demand (consumtion) regression models. The horizontal axis shows the energy intensity as measured by the ratio of energy costs to the value of shiments (as in Figure 3), with the 50 th and 95 th ercentiles of the energy intensity distribution identified by vertical lines. The domestic suly-energy rice elasticities resented in Figure 4 reveal a clear trend in increasing sensitivity to electricity rice changes for the most energyintensive industries. The median industry, in terms of energy intensity, has an estimated elasticity of about -0.16, more than twice the estimate of in the simle linear regression model (left half of Table 1), and a 95 ercent confidence interval that does not include the estimate from the simle linear regression model. The estimated elasticity at the 90 th ercentile of the energy intensity distribution is about -0.35, roughly five times the value estimates in the simle linear regression model. Interestingly, a similar, but vertically shifted attern is evident in the demand (consumtion) results resented in Figure 5. Ten of the least energy-intensive industries, reresenting about 2 ercent of the manufacturing sector, exerience a statistically significant and ositive imact from an increase in energy rices based on these 8 A table of regression coefficients is available from the authors uon request. 9 Figures for the eriod and for secifications that do not distinguish between two eriods within the eriod are available from the authors uon request. These results look similar to the eriod for suly and demand, with slight vertical shifts, but unlike the eriod show no statistically significant effects of electricity rices on net imorts at higher energy intensities. 22

26 estimates, but 98 ercent of industries exerience a statistically insignificant or a negative change. This is consistent with a substitution effect into less energy-intensive goods. The median industry does not exerience a change in demand that is statistically different from zero. The magnitude of the demand elasticity increases substantially again for the more energy-intensive industries. The estimated elasticity at the 90 th ercentile of the energy intensity distribution is about -0.25, with the uer end of the distribution aroaching These two figures show that demand and domestic suly both decline with higher energy rices for the most energy-intensive firms, but that the demand resonse is less than the domestic suly resonse, suggesting some increase in net imorts when energy rices increase. Figure 6 shows this more recisely, that is, the net imort imact (demand minus suly elasticity) of an energy rice increase (the difference between Figures 4 and 5). Here, a 10 ercent energy rice increase would result in a 1 to 1.5 ercent increase in net imorts for most manufacturing industries, with some ranging below 1 ercent and some, articularly those with energy intensity above 10 ercent, exceeding 1.5 ercent. As noted reviously, the 95 th ercentile confidence interval resented in the figure reflects the correlation in the residuals of the suly and demand regression equations that are accounted for in our seemingly unrelated regression modeling framework. In the revious section, we discussed the fact that measuring the net imort effect as the difference between the demand and suly elasticities can misreresent the true effect when net imorts are a substantial fraction of domestic suly. Correcting this is straightforward, but requires us to look at individual industries and their articular ratio 23

27 of net imorts to domestic suly. We now turn to that calculation in the context of a roosed CO 2 mitigation olicy. V. Simulation of Near-term Effects of a CO 2 Mitigation Policy We can use these statistically-estimated relationshis to simulate the effects of a $15 er ton CO 2 rice from a unilateral U.S. climate change olicy. Based on the Energy Information Administration (2008) modeling of an economy-wide ca-and-trade rogram, such an allowance rice would increase industrial sector electricity rices by about 8 ercent, which is aroximately equal to a one standard deviation increase in energy rices in our samle. 10 This carbon rice is similar to allowance rices exected at the start of ca-and-trade rograms roosed in recent legislation, including EPA s (2009) estimate of a $13 er ton CO2 rice under the Waxman-Markey Bill (H.R. 2454, 111 th Congress), EPA s (2010) estimate of a $17 er ton CO2 rice under the American Power Act (draft legislation from Senators Kerry and Lieberman) as well as the first year carbon tax of $15 er ton CO2 in a 2009 Reublican-sonsored carbon tax bill (H.R. 2380, 111 th Congress). 11 Based on these estimated model arameters, this energy rice increase then drives the domestic suly, demand, and cometitiveness imacts in our simulation. Aroximating, Figure 7 resents the estimated cometitiveness effects of a carbon ricing olicy that raised energy rices reflecting $15 10 Extraolating imacts for higher CO 2 rices is beyond the scoe of this analysis since it would reflect an out-of-samle rediction. 11 The simulation focuses only on carbon dioxide emissions from fossil fuels. Since this reresents 98 ercent of all carbon dioxide emissions, and more than 80 ercent of all greenhouse gas emissions in the United States, this should serve as a sufficient simulation of the imact of climate olicy on U.S. manufacturing industries cometitiveness. The key excetion may be the cement industry, which has substantial rocess emissions of carbon dioxide. 24

28 er ton CO2 as it varies with energy intensity; it is exactly a rescaled version of Figure 6. The cometitiveness effect is on the order of about 1 ercent but rises to more than 1.5 ercent for the most energy intensive industries. This effect, however, is aroximate because it ignores the term in Equation (10). Table 2 shows the corrected results in the context for all manufacturing and for secific sectors of the most energy-intensive industries, with the results weighted by industry-secific value of shiments (Column 5). 12 This table also rovides the aroximated cometitiveness imacts from Figure 7 for comarison (Column 4), which we can see are quite close. The energy-intensive industries of iron and steel, aluminum, ul and aer, cement, glass, and industrial chemicals would bear total ercentage declines in domestic suly, on the order of -3.2 to -4.4 ercent, in considerable excess of the manufacturing sector average of -1.4 ercent (Column 2). Most of the lower domestic suly reflects lower demand, however, not an influx of net imorts; the demand declines range from about -1.9 to -2.8 ercent. Correcting for the relative size of suly and demand, the cometitiveness effect is only 1.0 to 1.3 ercent. That is, in these industries about one-third of the decline in domestic suly results from an increase in net imorts. Even more narrowly defined industries could exerience cometitiveness imacts outside this range. The largest imact among energy-intensive industries in our simulation is alkalies and chlorine, a subset of chemicals, with an estimated cometitiveness effect of 2.2 ercent. 12 In constructing the grou aggregates, we estimate each of the comonent-industry ercentage change based on that industry s energy intensity, and then add u these changes based on the comonent-industry s share of domestic suly within the industry grou. 25

29 Some non-energy intensive industries exerienced larger imacts where domestic consumtion is much greater than domestic suly, and where domestic demand rises from a substitution effect. For examle, both dolls and leather-lined clothing have effects above 3 ercent. This suggests an interesting henomenon: among the energy-intensive industries that remain in the United States, they may be somewhat more resilient to higher energy rices than less energy-intensive industries that comete with large volumes of net imorts. Given the emirical model s structure that yields common suly and demand elasticities with resect to energy rices for all industries with a comarable energy intensity, the simulation roduces similar outcomes for industries with a similar energy intensity. Therefore, we cannot rule out that some individual industries with a articular energy intensity may face a larger or smaller imact than the average that we calculate. VI. Policy Imlications and Further Research These results suggest that consumers of energy-intensive goods do not resond to higher energy rices by consuming a lot more imorts. To a large art, they economize on their use of these higher-riced manufactured goods, erhas by using less of the good in the manufacture of their finished roducts or by substituting with other, less energyintensive materials. Consumers aear to ursue only a limited substitution with imorts, suggesting that the imorted versions of domestically-roduced goods may be imerfect substitutes. Other determinants of trade flows such as transort costs, tariffs, etc. may limit the substitution ossibilities. Quantitatively, cometitiveness effects are small in the sense that they amount for around 1 ercent of suly even among energy-intensive 26

30 industries. A 1 ercent change in suly due to carbon ricing induced cometitiveness imacts is small relative to annual fluctuations in suly that average 6 to 10 ercent for energy-intensive industries. Comared to the overall effect on suly from roosed olicies, this still counts for roughly one-third of the suly effect among energyintensive domestic suliers; in fact, it accounts for a larger ortion among some nonenergy-intensive industries. This aears to reflect a substitution across goods, from energy-intensive to non-energy-intensive, and then to non-energy-intensive imorts, rather than from energy-intensive domestic roduction to energy-intensive imorts. Based on our findings, attemting to rotect energy-intensive U.S. manufacturing firms from international cometitive ressures through various olicies may have only a limited imact on these firms. The estimated cometitiveness imacts, while fairly modest at $15 er ton CO 2, suggest the need to target olicies to those most likely to face adverse imacts, such as some narrowly defined industries that may face cometitive ressures from abroad as their energy costs rise with a greenhouse gas mitigation olicy. Indeed, given the modest magnitude of the cometitiveness imacts on climate olicy in our simulation, the otential economic and dilomatic costs of such olicies may outweigh the benefits and commend no action. Regardless, energy-intensive firms oerating under the EU Emission Trading Scheme, a CO 2 ca-and-trade rogram, have lobbied extensively to receive free allowances in the ost-2012 ETS. Similar firms in the U.S. have echoed this request as they have lobbied Congress during its deliberations of a U.S. ca-and-trade rogram in 2009 and 2010 (see Interagency Cometitiveness Analysis Team 2009). The estimated cometitiveness imacts in this analysis could rovide a basis for the amount of the gratis 27