The Effect of Carbon Pricing on Firm Emissions: Evidence from the Swedish CO₂ Tax

Abstract

Anthropogenic climate change is one of the most pressing issues of our time, representing a massive market failure in need of urgent policy intervention (Stern 2008). Many economists have argued the most important policy tool to combat climate change is to price CO₂ emissions through a global carbon tax (e.g., Nordhaus 1993; Golosov et al. 2014; Rockström et al. 2017; Sterner et al. 2019), ideally in combination with subsidies for green innovation (e.g., Acemoglu et al. 2016; Aghion et al. 2016). While there is still no agreement on a global carbon tax, many countries around the world have implemented local or regional carbon pricing schemes (World Bank 2022). Despite the risk of carbon leakage, Conte, Desmet, and Rossi-Hansberg (2022) argue that such unilateral schemes, if properly designed, can still be effective.

Until just a few years ago, there was little evidence on whether existing carbon pricing schemes had actually reduced firm CO₂ emissions (Burke et al. 2016; Martin, Muûls, and Wagner 2016). While the literature has grown over the last few years, the evidence on the effectiveness of carbon pricing is still mixed (see Rafaty, Dolphin, and Pretis 2021; Timilsina 2022; Green 2021, for recent reviews). One reason results differ is that carbon pricing schemes vary greatly in their structure, coverage, and magnitude. In addition, the majority of studies examine aggregated data at the sector and/or country level, which makes it difficult to account for important heterogeneity in marginal pricing and abatement costs across firms.¹ The relatively few studies that analyze micro-data on individual firms or plants estimate average treatment effects around the introduction of a particular carbon pricing scheme. Since emission pricing differs significantly in magnitude across schemes and over time, however, it is perhaps not surprising that results vary greatly across studies.²

Using data from Sweden, we construct the longest firm-level panel to date on economic activity and CO₂ emissions for the population of manufacturing firms over 1990–2015. Our analysis aims to contribute to the existing literature in several ways.

First, we provide estimates of carbon pricing elasticities for the manufacturing sector by relating the marginal cost of emitting a unit of CO₂ to actual emissions for every firm and year. Such elasticities can be used to assess different carbon pricing schemes, for example, by calibrating macroeconomic models of optimal climate policy (such as Golosov et al. 2014; Acemoglu et al. 2016). Our data set is large in both the cross-sectional and time-series dimensions, which enables econometric identification due to numerous changes in tax rates and firm-level exemptions over time. It also allows estimation of the full dynamic response to carbon pricing, where we can account for the time it takes for firms to adapt their technologies and business models (Dessaint, Foucault, and Fresard 2022). We are thus able to provide precise estimates of carbon pricing elasticities, unlike the earlier literature, which either bases estimates on aggregated country- or sector-level time-series data, or estimates average treatment effects around the introduction of a given carbon pricing scheme.³

Second, our micro-data enable us to investigate the heterogeneity in carbon pricing elasticities across firms and subsectors. Both the incentive and the ability of firms to lower their emissions in response to carbon pricing will differ depending on, for example, production technology, abatement costs, financial constraints, mobility of production, and the competitive environment (e.g., Ederington, Levinson, and Minier 2005; Gillingham and Stock 2018; Xu and Kim 2022; Martin et al. 2014; Lyubich, Shapiro, and Walker 2018). Since the bulk of CO₂ emissions are concentrated to a few subsectors, and often to a few large firms within each subsector, such heterogeneity has large implications for the aggregate impact of carbon pricing.

Third, existing evidence on the effectiveness of carbon pricing have been mixed, and several studies have found limited impact of carbon pricing schemes on aggregate CO₂ emissions from manufacturing (see, e.g., Rafaty, Dolphin, and Pretis 2021, and the references therein). In contrast, the elasticities we uncover from micro-data imply economically significant effects of carbon pricing on manufacturing emissions. These elasticities can be applied to infer expected emission reductions across carbon pricing schemes with different price levels and industry structures, which increases the external validity compared to previous studies.

Our data set includes comprehensive information on financials and CO₂ emissions for the universe of Swedish manufacturing firms over the period 1990–2015. Sweden serves as an ideal testing ground for analyzing the incidence and impact of carbon pricing. It was one of the first countries to introduce a carbon tax in 1991, levied on the heating emissions from manufacturing firms, and the Swedish carbon tax rate is currently the highest in the world.⁴ In addition, several subsequent changes in tax rates, various tax exemptions, and the introduction of the EU Emissions Trading System (ETS) toward the end of our sample lead to substantial variation in effective marginal tax rates across firms and over time, facilitating econometric identification.

When the carbon tax scheme was introduced in 1991, it contained various tax exemptions for the highest emitters, motivated by the desire to mitigate “carbon leakage” (ie, CO₂-emitting plants closing in Sweden and/or moving to other jurisdictions). As a result, the 10% of firms with the highest CO₂ emissions had significantly lower (sometimes even zero) marginal carbon tax rates, despite facing a high average tax rate (reducing average EBIT margins by more than 6 percentage points). Consistent with reduced marginal incentives, we find that the emission intensity of the highest-emitting firms decreased only modestly between 1990 and 2015, while the remaining 90% of firms facing higher marginal carbon tax rates experienced significantly higher reductions.

To measure short-term responses, we follow previous literature and perform difference-in-differences analysis around the introduction and subsequent changes of the carbon tax regime, utilizing the caps on total tax payments for the highest emitters. In the first test, we focus on the 10% most emitting sectors and sort firms into two groups: those qualifying for exemptions around the introduction of the carbon tax in 1991–1992 and those that did not. The results show that a rise (decline) in marginal cost is associated with decreasing (increasing) firm-level emission intensity. We also study the reintroduction of a carbon tax payment exemption in 1997 and obtain similar results.

We then examine the longer-term relationship between emission intensity and the marginal emission tax a firm faces, including both the explicit CO₂ tax and the implicit tax from the price of emission rights for firms with installations under the EU ETS. Using data from about 4,000 manufacturing firms, covering 85%–90% of Sweden’s manufacturing CO₂ emissions over 1990–2015, we find a significantly negative relationship between firm-level CO₂ emission intensity and the marginal cost of emissions. In our main specification, which includes firm and year fixed effects, we estimate that a 1% increase in the marginal emissions cost share reduces carbon emissions per unit of (PPI-deflated) sales by roughly 2% over a 3-year period. This magnitude is stable over the introduction of the EU ETS in 2005 and robust to including tighter sets of fixed effects, more lags, and various firm-level controls.

We also document significant heterogeneity in the response to carbon pricing. We first sort firms into two groups based on the ex ante costs of reducing CO₂ emissions, using data on air pollution abatement costs and expenditure (PACE; see Becker 2005). Firms in low PACE sectors, that is, where it is relatively cheaper and easier to reduce emissions, display a carbon pricing elasticity of around three, compared to an elasticity less than two in high PACE sectors. To get at carbon leakage risk, we further separate low and high PACE sectors based on the ex ante mobility of their assets (Ederington, Levinson, and Minier 2005). The smallest point estimate is for firms in high PACE and low-mobility sectors, with an elasticity of 1.7. This group, containing firms that face high abatement costs and are less able to avoid tax by moving production, comprises between 80% and 90% of aggregate manufacturing CO₂ emissions. We find similar results when we instead separate firms by whether their subsector is included on the EU “carbon leakage list” (European Commission 2009).

Since access to external financing might affect the ability of firms to invest in abatement, we explore whether financial constraints affect carbon pricing elasticities. Following the literature, we consider firms that are privately held (rather than listed), smaller, younger, and with lower dividend payout ratios as being more financially constrained (see, e.g., Saunders and Steffen 2011; Hadlock and Pierce 2010; Bartram, Hou, and Kim 2022). Less constrained firms display elasticities between two and three, whereas estimates for their more constrained counterparts are insignificant and consistently less than one. The difference in elasticities is only found among firms with high abatement costs, while financial constraints have no visible effect on firms with low abatement costs, consistent with financial constraints primarily hurting firms for which abatement requires significant investment.

To assess the economic importance of these findings, we relate the estimated elasticities to changes in aggregate manufacturing emissions of CO₂ in our sample period. Following Grossman and Krueger (1993) and Levinson (2009), we decompose the change in aggregate emissions into scale, composition, and technique effects. CO₂ (heating) emissions from the Swedish manufacturing sector decreased by 31% over 1990–2015. The decomposition attributes 3 percentage points of this decrease to lower aggregate output (“scale”) and 10 percentage points to the changing composition of Swedish manufacturing toward lower-emitting subsectors. By definition, the remaining 18 percentage points (58% of the total reduction) is attributed to changes in technology (“technique”). We then use our estimated carbon elasticities to calculate the contribution from carbon pricing on these aggregate reductions. Our calculations suggest that carbon pricing, through its effect on reduced emission intensities, can account for between one-third up to almost all of the total decrease in CO₂ emissions from manufacturing over our sample period.

In terms of implications, we believe that our findings are relevant for discussions on optimal carbon taxation more generally (e.g., Nordhaus 1993; Bovenberg and De Mooij 1994; Bovenberg and Golder 1996; Pindyck 2013; Gillingham and Stock 2018; Stock 2020). While our reduced-form estimates ignore potentially important general equilibrium effects, they confirm that even in a unilateral carbon pricing scheme (as in Conte, Desmet, and Rossi-Hansberg 2022), firms do respond to the marginal cost of emitting CO₂ in a way consistent with economic theory. Our results also imply that Sweden could have achieved significantly larger reductions in CO₂ without the various tax exemptions that reduced marginal incentives to reduce emissions for the highest-emitting firms.

We also contribute to the literature examining the effects of environmental policy on firms (e.g., Fowlie 2010; Greenstone, List, and Syverson 2012; Fowlie, Reguant, and Ryan 2016; He, Wang, and Zhang 2020; Brown, Martinsson, and Thomann 2022; Bartram, Hou, and Kim 2022; Hartzmark and Shue 2023).⁵ By estimating elasticities rather than average treatment effects, we uncover several sources of heterogeneity in firms’ responses to carbon pricing and show that they are of economic importance.

Finally, our paper is part of a growing literature documenting the connections between finance and the environment (e.g., Bolton and Kacperczyk 2021; Giglio et al. 2021; Hong, Li, and Xu 2019; Ilhan, Sautner, and Vilkov 2021; Bartram, Hou, and Kim 2022; Giannetti et al. 2023).⁶ Specifically, our study adds to the work examining the legal and financial determinants of environmental behavior (e.g., Akey and Appel 2021; Bartram, Hou, and Kim 2022; Brown, Martinsson, and Thomann 2022; Xu and Kim 2022) by showing that financial constraints play an important role in determining the response of firms’ CO₂ emissions to carbon pricing.

1 Carbon Pricing in Sweden

Sweden introduced its carbon tax in 1991 alongside a handful of countries.⁷ The tax was part of a comprehensive reform and included elements of so-called (”green tax shifting”) with the idea to increase costs on polluting and lower, for instance, labor taxes (see, e.g., Jonsson, Ydstedt, and Asen 2020).⁸ Some reasons for Sweden being an early adopter of carbon pricing are the lack of significant fossil fuel resources and (perhaps as a result) the absence of significant anticlimate lobbying (e.g., Meckling, Sterner, and Wagner 2017; Sterner 2020). Internet Appendix A provides further background on the introduction and evolution of Swedish carbon taxation.

The Swedish carbon tax is levied on fossil fuels used either in combustion engines (“mobile emissions”) or for heating (“stationary emissions”). The tax on mobile emissions primarily affects road transportation (and is included in the after-tax price of fuel “at the pump”), while the tax on stationary emissions is levied on power plants and manufacturing firms. This study focuses on the stationary emissions tax on manufacturing firms, summarized in Figure 1. Manufacturing production releases heating and process CO₂ emissions and the carbon tax on stationary emissions is levied on emissions from heating only, while process CO₂ emissions are exempt. A plant must declare the use of its fossil fuel separately for production and heating and the tax is levied on heating fuel inputs in proportion to the implied emissions of CO₂ during combustion.⁹ Since the Swedish manufacturing sector uses about one-third of its fossil fuel for production and the remainder for heating, about two-thirds of the sector’s stationary CO₂ emissions are subject to carbon taxation.

Fig. 1

Carbon and energy taxation of an industrial plant

This figure illustrates the carbon and energy taxation for a manufacturing plant in Sweden in 2019. $Heating CO2emissions$ refers to the emissions released from the combustion of fossil fuels. $Process CO2emissions$ refers to the carbon dioxide emissions released in the actual manufacturing process (ie, not combustion of fossil fuels). $Utility$ is the power plant that produces heat and/or electricity, and $Plant$ is the industrial manufacturing plant.

Open in new tabDownload slide

Figure 2 plots the evolution of the Swedish carbon tax rate over time. When it was introduced in 1991, the tax was levied at a rate of 0.25 Swedish krona (SEK) per kilogram (kg) of emitted CO₂ across all sectors in the economy.¹⁰ Already at this point, however, Swedish carbon taxation incorporated various caps and exemptions (summarized in Table 1) for the highest-emitting firms. We discuss these in greater detail in Section 2.2.

Fig. 2

Carbon tax rate, in nominal values

This figure displays the nominal carbon tax rates (Swedish krona per kilogram of emitted carbon dioxide) for Sweden from 1991 to 2017. Manufacturing tax rate refers to the tax rate for the manufacturing sector (SNI 10-33 in the SNI2007 nomenclature), while General tax rate refers to the tax rate for nonindustrial firms and households.

Open in new tabDownload slide

In 2005, the European Union introduced a cap-and-trade scheme for CO₂ emissions, the European Union Emissions Trading System (EU ETS), which had major implications for Swedish carbon taxation. Installations covered by the EU ETS were gradually phased out of the Swedish carbon tax regulation over 2008–2011 (Government Bill 2007/2008:1 2007). Emission allowances were allocated for free to the participating plants (or “installations”) in the pilot phase (ie, 2005–2007), and the bulk of emission rights were distributed for free in the second trading phase (2008–2012) as well. In the third phase, starting in 2013, auctions of emission rights were introduced, although for manufacturing plants most emission rights were continued to be distributed for free, motivated by carbon leakage concerns.¹¹

2 Carbon Pricing across Firms, Sectors, and Over Time

2.1 Data and sample construction

Our sample is constructed by matching plant- and firm-level registry data (including accounting variables, number of workers, sector classifications) with CO₂ emissions for the time period 1990–2015. The Swedish Environmental Protection Agency (SEPA) provided data on CO₂ emissions at plant- and firm-level (including emissions under the EU ETS). We obtain registry data for listed and unlisted Swedish corporations from Upplysningscentralen (UC) for the period 1990–1997 and Bisnode Serrano for 1998–2015.

To compute emission intensities, our sample firms need to have data on both sales and CO₂ emissions. The number of firms with CO₂ emissions data changes during our sample period (see Table B.1 in the Internet Appendix), most notably in 1997–1999 and 2003–2006, when only emissions by larger plants were collected by SEPA. Since the largest emitters are always sampled, our data consistently covers between 80% and 95% of aggregate manufacturing CO₂ emissions in any given year (Figure A.1 in the Internet Appendix). Over the entire period 1990–2015, our sample covers 85% of aggregate CO₂ heating emissions and 87% of total (process plus heating) CO₂ emissions (Figure A.2 in the Internet Appendix) from the Swedish manufacturing sector.

Since historical firm-level records of actual carbon taxes paid could not be provided by the Swedish tax authority, we calculate both the effective marginal tax and the overall carbon tax payments from the actual CO₂ heating emissions for each plant and firm each year, using the carbon tax schedule (including possible exemptions) that was in place for the corresponding year.¹² For the EU ETS period, we infer the emissions subject to the Swedish carbon tax as the difference between emissions reported in SEPA and emissions according to the European Union Transaction Log (the official registry of EU ETS). For the regression analysis, we require sample firms to have at least four consecutive yearly observations to allow for lagged independent variables. A detailed description of the sample construction are provided in subsection B.1 and subsection B.2 of the Internet Appendix and in Sajtos (2020). Table 2 provides summary statistics for the key variables. We then sort our sample firms into different manufacturing subsectors based on CO₂ emissions intensity in 1990. Specifically, we sum up all (heating) CO₂ emissions as well as PPI-deflated sales across all firms in each four-digit industry each year. For deflating sales, we use 2010 as the base year and deflate using the Swedish Producer Price Index at the four-digit NACE code level. Removing the effect of sector-level price variation on sales is important to ensure that changes in emission intensities are not completely driven by changing output prices (e.g., resulting from producers passing on carbon tax to their customers). We then rank the industries depending on the ratio between aggregate emissions divided by aggregate sales in 1990 (the year before the introduction of the carbon tax) from highest to lowest and divide them into deciles. This results in 10 bins of about 20 four-digit industries each. We describe how Swedish manufacturing CO₂ emissions have evolved over our sample period in the Internet Appendix C.

Table B.2 in the Internet Appendix shows the distribution of emissions across two-digit NACE sectors. The most emission intensive manufacturing firms are found in nonmetallic mineral products, (such as cement, plaster, mortar, and glass production), coke and refined petroleum products, paper and paper products, textiles, basic metals (particularly iron and steel production), chemicals, and food (particularly sugar production). These seven sectors contain 82% of the four-digit subsectors in deciles 9 and 10 and jointly account for almost 88% of aggregate manufacturing CO₂ emissions in 1990.

2.2 The effect of changing carbon tax regimes

Our identification relies on cross-sectional variation in firms’ marginal tax rates, allowing us to control for time, firm, and year times sector dummies to isolate the effect of carbon pricing on emissions. This variation is due to the various exemptions that high-emitting firms enjoyed at various times during our sample period, summarized in Table 1. Figure 3 illustrates the tax rates a hypothetical firm would face across different regimes.

Fig. 3

Changes to the carbon tax: Emissions and carbon tax payments by regime

This figure compares the carbon tax payments under the different regimes through a representative manufacturing firm. The hypothetical firm earns 50,000 SEK each year, and assumed to burn only coal in 1991 and 1992. All carbon tax payments with the exception of 2015 are shown on the vertical axis on the left side. Carbon tax payments in 2015 are shown on the vertical axis on the right side.

Open in new tabDownload slide

The numerous changes in the tax schedule lead to substantial variation in carbon taxation in both the time-series and the cross-section. Figure 4 shows how the average effective tax rate, computed as total carbon taxes paid divided by total CO₂ (heating) emissions (Average tax), and the marginal tax rate for the next emitted unit of CO₂ (Marginal tax) evolves over time for two types of firms. The first type is a firm with emissions consistently below the thresholds for tax exemptions and that does not have any plants included in the EU ETS. For such firms, the average tax rate equals the marginal tax rate throughout the sample period. The second type is a firm whose emissions consistently lie above the carbon tax exemption thresholds and whose plants eventually transition into the EU ETS. Before EU ETS, the average tax rate exceeded the marginal tax rate for high-emitting firms, except for 1993–1996, when exemptions had been removed. After EU ETS was introduced in 2007, the implicit marginal tax rate, reflected in the price of emission rights, increased considerably, while the average tax rate stayed roughly constant due to free allocations of emission rights.¹³

Fig. 4

Average and marginal tax rates (1990–2015)

This figure displays the average and marginal tax rates depending on whether the firm is eligible for carbon tax exemptions and covered by the EU ETS. $no exemption/no EU ETS$ denotes firms that are not regulated by the EU ETS and are not entitled to carbon tax exemption, $exemption/EU ETS$ refers to the firms with available exemptions until they enter the emission trading scheme. Marginal tax rates for EU ETS are the price for emission rights.

Open in new tabDownload slide

The significant differences in marginal and average tax rates across groups have important economic implications, as a firm’s incentive to reduce CO₂ emissions depends on the former while its effective tax payments depend on the latter. Before the introduction of the EU ETS, high-emitting firms thus had relatively low marginal incentives to reduce emissions, despite paying a large fraction of their profits in carbon tax. After the introduction of EU ETS, marginal emission costs for high emitters increased substantially, while overall carbon tax payments decreased due to the free allowance of emission rights.

2.3 Decomposing Sweden’s manufacturing CO₂ emissions

In Figure 5, we decompose the change in aggregate CO₂ emissions from Swedish manufacturing using the framework developed in Grossman and Krueger (1991) and Grossman and Krueger (1993).¹⁴ The actual level of CO₂ emissions in 2015 was 31% lower than in 1990, representing the combined scale, composition and technique effects. Holding sector composition and emission intensities constant from 1990 and onward, CO₂ emissions would only have decreased by 3% in 2015, reflecting a 3% decrease in manufacturing output. Keeping emission intensities constant, but also allowing sector composition to change as in the data, CO₂ emissions would have been 13% lower. The 10% difference is the composition effect, indicating that a move toward less carbon-intensive manufacturing subsectors accounts for almost a third of the aggregate decrease in emissions. The residual technique effect is 18%, or almost two-thirds of the reduction in CO₂ emissions 1990–2015.

Fig. 5

Carbon dioxide emissions from Swedish manufacturing (1990–2015)

This figure displays the decomposition of the Swedish carbon dioxide emission reduction. $Scale$ captures how emissions would have evolved without tangible technological progress and structural changes in the manufacturing sector. $Composition$ refers to the change in industry composition, and $Technique$ captures the technological progress in the industrial sector.

Open in new tabDownload slide

The technique effect incorporates the impact of carbon pricing on emission intensities that is the focus of our study. A few caveats are in order, however. First, at least part of the decrease attributed to scale and composition was also likely affected by carbon pricing. Since it is difficult to estimate carbon pricing elasticities of output reliably using our reduced-form approach, we choose to focus on emission intensities in this study.¹⁵ To the extent carbon pricing also decreased the consumption of goods produced by higher-emitting firms, we will be underestimating the total effect of carbon pricing on aggregate emissions. Second, our emission intensity measure normalizes emissions with sales (e.g., revenues of a steel company) rather than actual output (e.g., tons of steel produced). While we deflate sales with PPI at the four-digit NACE level, our estimates can still be affected by relative price changes across firms within each four-digit subsector (see Foster, Haltiwanger, and Syverson 2008). As a result, the carbon pricing elasticities we estimate may also capture other strategic responses beyond changes in production technology, such as within-sector differences in pricing power (as in De Loecker and Warzynski 2012), and/or product mix across firms, and therefore should be interpreted in a broader sense.¹⁶

3 Short-Term Effects of Carbon Pricing

Before estimating carbon pricing elasticities, we first examine the short-run responses of firm-level emission intensities to the introduction and subsequent changes in the Swedish carbon taxation scheme. Similar to Colmer et al. (2022), we use a difference-in-differences approach that compares the changes in emission intensities between otherwise similar firms whose marginal cost of emissions is affected differently by a change in taxation.

As is typical in event studies, the interpretation of the results implicitly depends on assumptions regarding the expectations and rationality of relevant decision makers. To the extent the subsequent changes in tax rates are anticipated by firms, this would affect their response to a current tax change. While it is plausible that the initial introduction of the carbon tax in 1991 was at least partly anticipated (which biases against finding a short-term response), the bipartisan political commitment to environmental taxation in Sweden in the wake of the major tax reform of 1991 (Government Bill 1989/90:111 1989) and the strong reliance of the government on environmental tax revenues (Tax Shift Commission 1997) make it less likely that firms anticipated subsequent changes in carbon taxation, at least until the introduction of the EU ETS.¹⁷

The introduction of carbon taxation in 1991 and its first revision in 1993 provide relatively clean events to analyze. In 1991 and 1992, caps were in place so that a firm’s carbon tax never exceeded a certain percentage of sales. In 1993, these caps were removed in conjunction with an overall reduction in the statutory carbon tax rate. Accordingly, we divide firms into those who qualified for an exemption in 1991-1992 and those that did not. Table 3 reports average marginal costs (panel A) and emissions-to-sales (panel B) for the periods 1990, 1991–1992, and 1993–1996. We focus on firms from decile 10 as most emissions are concentrated there and the vast majority of firms are present in the sample for all years. We exclude firms in the cement, lime and glass sectors from this test as the tax changes did not apply to them.¹⁸

In 1990, before the tax was introduced, both groups of firms face a zero marginal cost of emitting carbon. After the introduction in 1991–1992, firms not qualifying for exemptions experienced a marginal tax increase of 0.203 Swedish krona (SEK) per kg of emitted CO₂, while firms with exemptions, despite paying substantial amounts of carbon taxes, still faced a zero marginal tax rate. In the 1993–1996 period, following the first tax change, both groups were taxed at 0.084 SEK/kg with no exemptions. This led to a marginal tax increase of 0.084 for firms with exemptions in the 1993–1996 period and a marginal tax decrease for the nonexemption group of –0.119. The difference-in-differences changes in marginal tax across groups are highly significant around both the introduction (–0.203) and the subsequent change 1993–1996 (0.203).

By construction, firms in the exemption group have higher emissions-to-sales than the nonexemption group (0.087 vs. 0.011 in 1990). After the introduction of the tax in 1991–1992, firms in the nonexempt group display similar emissions-to-sales ratios as they did in 1990. In contrast, firms in the exempt group, who still faced a zero marginal tax rate, increase their emissions-to-sales by about 18% (from 0.087 to 0.103). The diff-in-diff estimate of 0.016 is significant at the 1% level, consistent with higher marginal carbon taxes leading to lower emission intensities. In the 1993–1996 period, following the tax change, nonexempt firms (whose marginal tax was cut), experience a statistically significant increase in their emissions-to-sales of about 45%. Firms in the exempt group (whose marginal tax rate increased) instead saw a slight 3% decrease emissions-to-sales ratios. The diff-in-diff estimate between groups between 1991–1992 and 1993–1996 is a negative -0.007. While the diff-in-diff is not statistically significant at conventional levels, the change is again consistent with firms responding to carbon taxes by reducing emissions. Both diff-in-diff estimates are robust to including four-digit industry dummies (column 4).

In panels C and D of Table 3, we consider the subsequent tax change in 1997. The preceding years (1993-1996) was the only time during our sample period when all manufacturing firms (except for cement, lime and glass) faced the same marginal carbon tax rate. The tax change in 1997 more than doubled the marginal tax rate (from around 0.09 to 0.19 SEK/kg) but reintroduced an exemption for firms with tax payments above 0.8% of sales, for whom the marginal tax was reduced to one quarter of the statutory rate. We again focus on decile 10 subsectors, and divide firms into those qualifying for the exemption and those that did not. We require firms to be present during the entire period 1993–2000, to avoid capturing effects due to changes in the sample composition.

Panel C reports changes in marginal taxes. The two groups of firms face the same marginal cost of emitting CO₂ in the pre-period 1993–1996. The 1997 changes led exempt firms to experience a small marginal tax reduction of 0.009, while nonexempt firms faced a large increase in marginal tax of 0.087. Panel D shows the corresponding changes in emission intensities around this event. Again, by construction, exemption firms have higher emissions-to-sales ratios in the pre-period. While the differences within each group are not statistically significant, they both evolve as predicted: average emissions intensities increased by about 11% for exemption firms (whose marginal tax decreased) whereas it fell by about 11% for nonexempt firms (whose marginal tax increased). The diff-in-diff estimate of 0.0098 is positive and statistically significant, again indicating a negative short-run relationship between marginal carbon pricing and emission intensities.

The results from these event studies suggest that carbon pricing can have a visible impact on emissions even in the short run. But since significant adjustments of technology and/or strategy probably take several years for firms to fully implement, it is more relevant to evaluate the longer-term effects of such policies.

4 Estimation of Carbon Pricing Elasticities

4.1 Main specification

We now turn to the longer-term impact of carbon pricing on CO₂ emissions. As firms’ incentives should depend on the marginal (rather than the average) cost (e.g., Cropper and Oates 1988), we let emission intensity be a function of the marginal carbon tax rate.

There are a few specification issues that need to be dealt with. First, some of the responses to carbon pricing involve significant investments and possibly other strategic changes that take time to implement. In addition, expectations about future tax changes may affect the speed of the response to current taxes. Since it is not theoretically clear at what time lag carbon pricing should affect firms’ CO₂ emissions, we allow for the lag length to differ across specifications.¹⁹ Second, the elasticity of emissions to carbon pricing is likely to be heterogeneous across firms and/or industries, since it depends on factors, such as the ability of firms to pass the tax on to their customers through higher prices,²⁰ the ability to move production to other jurisdictions, costs of emissions abatement, and the ability to finance such abatement. In addition to including fixed effects, we will address such heterogeneity by estimating elasticities for relevant subsamples.

4.2 Baseline results

Table 4 presents baseline results from estimating Equation (1) with q=1 up to q=3. In columns 1 and 2, we display results with the marginal cost share of sales at the beginning of the year without and with firm fixed effects. The change in the marginal cost of CO₂ emissions is strongly related to changes in firm-level carbon emissions intensity. The result implies that a change in the marginal cost of emissions to sales is associated with a change in carbon intensity by about a factor of one. (We will further discuss the economic magnitude of these estimates in Section 5.) Adding $Δln(1−Ci,t-2)$ (column 3) yields a larger estimate of the t—1 coefficient and a joint impact of around 1.6. In column 4 we also include a third lag $Δln(1−Ci,t-3)$⁠, which increases the estimated joint impact to a total elasticity of 2.1 (with all three lag coefficients being statistically significant). In unreported regressions, we show that additional lags have small and a statistically insignificant coefficients and leave the joint estimated magnitude largely unchanged. We therefore choose the specification with q = 3 as our baseline model.²² In column 5 we include a full set of four-digit industry by year fixed effects to control for differences in industry trends and shocks over time, which results in a slightly higher estimated total elasticity of 2.2.

Since the top deciles of emitters account for a disproportionately large fraction of total CO₂ emissions, it is particularly relevant to investigate possible heterogeneity in this dimension. The remaining three columns in Table 4 show results for three subsamples split according to 1990 subsector emission deciles as in Tables B.3 and B.4 in the Internet Appendix.

Column 6 shows estimates for firms with low emission intensities in 1990 (the bottom 40% of four-digit sectors in terms of CO₂ emissions to sales in 1990). The joint effect is three times larger for this subsample. Recall from Table B.3 in the Internet Appendix that firms from these sectors comprise under 6% of CO₂ emissions. In column 7 we consider the group of firms from deciles 5 to 8. The estimated joint carbon pricing effect is 2.7 and highly statistically significant. In column 8, we consider firms from deciles 9 and 10, which have significantly higher emissions-to-sales ratios compared to firms in other manufacturing subsectors and jointly account over 80% of CO₂ emissions in 1990. The joint carbon pricing effect in this subsample is considerably lower than in the other deciles (reported in columns 6 and 7) and also lower than the one estimated for the full sample in column 4. The results in the final three columns in Table 4 suggest that firms in subsectors with production technologies associated with higher CO₂ emissions also have the highest cost of abatement. We carry out additional tests to shed more light on this possible mechanism below.

Table 5 reports the results from a set of additional robustness tests. To account for the possibility that EU’s cap-and-trade system leads to different carbon pricing elasticities, we interact our marginal cost variable with an indicator variable taking on the value one if the firm-year is regulated under EU ETS and zero otherwise. We report results both for the full sample and for the subsample of firms in deciles 9 and 10, where almost all EU ETS regulated firms belong. We also control for the impact of broader changes to corporate taxation (by including the cost share of income taxes to sales) and for green tax shifting (by including labor taxes to sales), respectively, on firm CO₂ emissions.²³ Finally, following Brännlund, Lundgren, and Marklund (2014), we include firm size and capital intensity as control variables. The estimated elasticities are essentially unchanged across these alternative specifications.