Why are (cid:28)rms that export cleaner? International trade, Abatement and Environmental Emissions∗ Rikard Forslid†, Toshihiro Okubo‡, and Karen Helene Ulltveit-Moe(cid:159) Revised version of CEPR Discussion paper no. 8583. This version, January 2015 Abstract This paper develops a theoretical model of trade and environmental emissions with heterogeneous (cid:28)rms, where (cid:28)rms make abatement investments and thereby a(cid:27)ect their levelofemissions. Weshowthatinvestmentsinabatementarepositivelyrelatedto(cid:28)rm productivity and (cid:28)rm exports, while emission intensity is negatively related to (cid:28)rms’ productivity and exports. The basic reason for these results is that a larger produc- tion scale supports more investments in abatement and, in turn, reduces emissions per output. We (cid:28)nd that trade liberalization weeds out the least productive and dirtiest (cid:28)rms thereby shifting production away from relatively dirty low productive local (cid:28)rms to more productive and cleaner exporters. The overall e(cid:27)ect of trade is therefore to reduce emissions. We test the empirical implications of the model on emission inten- sity, abatement and exporting using (cid:28)rm-level data from Sweden. The empirical results support our model. JEL Classi(cid:28)cation: F12, F14, F18, Q56 Keywords:heterogeneous (cid:28)rms, environmental emissions, abatement, international trade ∗We are grateful for comments from Andrew Bernard, Peter Egger, Peter Fredriksson, Beata Javorcik, Gordon Hanson, Peter Neary, Scott Taylor, Adrian Wood, and Tony Venables. Financial support from Jan Wallander and Tom Hedelius’ Research Foundation, The Swedish Research Council, Grant-in-Aid for Scienti(cid:28)c Research (JSPS) and Research Institute of Economy, Trade and Industry (RIETI) is gratefully acknowledged. †Stockholm University and CEPR; email: [email protected] ‡Keio university, e-mail: [email protected] (cid:159)University of Oslo and CEPR, e-mail: [email protected] 1 1 Introduction There is no consensus on the e(cid:27)ect of international trade on the environment, in particular on the e(cid:27)ect of trade on global emissions. Neither the theoretical nor the empirical literature provides a clean cut answer to the link between trade and environmental emissions. Hence, we do not know if international trade increases or decreases the emissions of greenhouse gases and contributes to global warming. However, this paper sets out to explain why we may expect exporter to emit less, and why trade liberalization may thus lead to cleaner industrial production. We do so by focusing on inter-(cid:28)rm productivity di(cid:27)erentials and interdependence among productivity, exporting, abatement and environmental emissions. In theoretical neoclassical models, international trade has opposing e(cid:27)ects. On the one hand, tradeincreasesincome, whichwilltendtoincreasethedemandforacleanenvironment and therefore increase investments in clean technology and abatement. On the other hand, trade liberalization may also imply an overall expansion of dirty production, because trade allows countries with low emission standards to become pollution havens. Copeland and Taylor (1995) show how trade liberalization may increase global emissions if the income di(cid:27)erencesbetweentheliberalizingcountriesarelarge, asdirtyindustriesarelikelytoexpand strongly in the poor country with low environmental standards. The empirical literature that analyses the link between trade in goods and emissions based on sector level data and Heckscher Ohlin type models is also inconclusive.1 Antweiler et al. (2001) and Frankel and Rose (2005) (cid:28)nd that trade decreases emissions. Using sector level data for the U.S., Ederington et al. (2004) do not (cid:28)nd any evidence that pollution intensive industries have been disproportionately a(cid:27)ected by tari(cid:27) changes. On the other hand, also using sector-level trade data, Levinson and Taylor (2008) (cid:28)nd evidence that higher environmental standards in the US have increased the imports from Mexico in dirty industries. Our point of departure for the analysis of trade and the environment is a model with heterogeneous (cid:28)rms and intra-industry trade, where trade gives rise to intra-industry real- locations across (cid:28)rms where we build on Melitz (2003). The choice of theoretical framework is motivated by descriptive evidence on the environmental emissions of Swedish manufac- turing (cid:28)rms. We (cid:28)nd (cid:28)rms’ emissions di(cid:27)er signi(cid:28)cantly across (cid:28)rms, even within rather narrowly de(cid:28)ned industries, and that the majority of variation in emissions can be ascribed to intra- rather than inter-industry variation. Moreover, comparing non-exporters and ex- 1Early surveys are made by Copeland and Taylor (2004) and Brunnermeier and Levinson (2004). 2 porters in Swedish manufacturing, we (cid:28)nd that in most manufacturing industries exporters do on average have a lower emission intensity. Motivated by these basic facts we build a model with international trade and environmental emissions where (cid:28)rms are heterogeneous with respect to productivity, abatement investments and emission intensity. We propose and develop a mechanism for why exporters may have a lower emission intensity when emissions are subject to an environmental tax. This mechanism runs through (cid:28)rms’ investments in abatement. Firms’ abatement investments depend on their production volumes as a larger scale allow them to spread the (cid:28)xed costs of abatement investment across more units. Pro- duction volumes are moreover determined by (cid:28)rms’ productivity and export status. More productive (cid:28)rms access international markets, have higher volumes and make higher abate- ment investments. As a consequence, (cid:28)rms’ emission intensity is negatively related to (cid:28)rms’ productivity and export status. Ourtheoreticalmodelmoreoverallowsforpredictionsontheimpactoftradeliberalization ontotalenvironmentalemissions. We(cid:28)ndthattotalemissionsfromthemanufacturingsector decreases as a result of trade and trade liberalization. Trade a(cid:27)ects the exporting and non- exporting sector in di(cid:27)erent ways. Exporters are for any level of trade costs always cleaner than non-exporters, and we show that trade liberalization may make exporters even cleaner by inducing them to invest more in abatement. But trade liberalization also implies higher production volumes for exporters, which ceteris paribus entails higher emissions. Total emissions therefore increases from the exporting sector. However, trade moreover increases local competition, which implies that the least productive, and therefore dirtiest, (cid:28)rms are forced to close down, while the remaining non-exporters are forced to scale down their productionvolume. Togetherthesedi(cid:27)erente(cid:27)ectsoftradeliberalizationserveastodecrease total emissions from the non-exporting sector. Adding up the e(cid:27)ects on exporters and non-exporters we (cid:28)nd that trade liberalization will always lead to lower total emissions. Thus, as trade weeds out some of the least productive and dirtiest (cid:28)rms, thereby shifting production away from relatively dirty low productive local (cid:28)rms to more productive and cleaner exporters, the overall e(cid:27)ect of trade liberalization is to reduce emissions. The theoretical model allows us to derive a set of empirical predictions on emissions and exporting as well as abatement investment and exporting. Access to the detailed (cid:28)rm level data set for Swedish manufacturing (cid:28)rms allow us to test these predictions. Our data set contains (cid:28)rm-level emissions and (cid:28)rm-level abatement investments as well as (cid:28)rm exports. According to our model, productivity drives the (cid:28)rm level emission intensity as well as the export status of a (cid:28)rm. However, while productivity has a continuous e(cid:27)ect on the emission intensity, the model predicts a discontinuous jump down in the emission intensity as (cid:28)rms become exporters. The same kind of relationship is predicted for abatement and exporting. 3 We exploit these features of the model as we take the model to the data. The empirical results are strongly supportive of the results derived in the theoretical model; exporters are found to invest more in abatement and to have lower emission intensity. Our theory is related to the idea presented in Levinson (2009) that trade may contribute to reduced pollution as trade liberalization may encourage technological upgrading. From a more methodological point of view, our work is also related to the literature on hetero- geneous (cid:28)rms and trade induced technological upgrading, see e.g. Bas (2012) and Bustos (2012). The majority of studies on international trade and environmental emissions are, unlike this paper, based on industry level analysis. There is, however, a rising literature focusing on (cid:28)rms rather than industries, which are thus closer in the spirit to our analysis. Holladay (2011) analyses (cid:28)rm-level data for the US, and (cid:28)nd that exporters pollute less per output. Unlike Holladay we develop a rigorous theoretical model with heterogeneous (cid:28)rms and environmental emissions, where we introduce a mechanism - economies of scale in abate- ment - motivating why exporter invest more in abatement than non-exporters. Moreover, based on our theory we do not not only make predictions on exporting and emissions but also on exporting and abatement, and are able to test both of these empirically. Cui et al (2012) analyse the relationship between exporting and emissions, but on the basis of a theoretical model distinctly di(cid:27)erent from ours. In our model exporters’ relatively lower emission inten- sity is due to their endogenous choice of abatement investment, while in their model it is due to exporters discrete choice of technology of production. Their empirical analysis focuses on emissions and exporting, while we also analyse the relationship between abatement and exporting.Batrakova and Davies (2012) examine the link between exporting and energy use employing Irish manufacturing data. Their theoretical model predicts a positive correlation between exporting and energy expenditures for low energy intensity (cid:28)rms and a smaller or even a negative correlation for high energy intensity (cid:28)rms. This asymmetry is due to the fact that trade as such requires extra energy, but on the other hand may also encourage a shift towards more energy e(cid:30)cient technologies if a (cid:28)rm is highly energy intensive. Their theoretical results are con(cid:28)rmed empirically. Girma et al (2008) studies the reported envi- ronmental e(cid:27)ects of UK (cid:28)rms innovations and the role of exporting, and (cid:28)nd that exporters are more likely to denote their innovations as having high environmental e(cid:27)ects. Tang et al (2014) examine the impact of environmental policy within a framework of heterogeneous (cid:28)rms in a closed economy. They (cid:28)nd that environmental policy reduces both consumption andpollutionemission, butthatoutputcouldbemaintainedusingsubsidiesdirectedtowards the more productive (cid:28)rms. Finally, Rodrigue and Soumonni (2014) employ Indonesian (cid:28)rm level data to investigate the impact of environmental investment on productivity dynamics and exports. While productivity dynamics do not appear to be a(cid:27)ected, growth in exports 4 does show a positive e(cid:27)ect. However, our paper is to our knowledge the (cid:28)rst to provide both a thoroughly theoretical analysis of emissions, abatement and trade that can be solved analytically and an empirical set of results that matches the theoretical (cid:28)ndings on emissions as well as abatement. The structure of the paper is as follows. In the next section we present a set of basic facts on the variation in environmental emission intensity across industries and (cid:28)rms, and exam- ine the di(cid:27)erences in emission intensity and abatement among non-exporters and exporters relying on data for Swedish manufacturing (cid:28)rms. Motivated by the descriptive evidence on emissions and (cid:28)rms, in Section 3 we develop a theoretical model on international trade, environmental emissions and heterogeneous (cid:28)rms. Based on this model we are able to derive a set of propositions and empirical implications regarding emissions, abatement and trade. In Section 4 we take the theory to the data, and test the empirical predictions on the rela- tionship between environmental emissions, export and productivity, and on the relationship between abatement, export and productivity. Finally, Section 5 concludes. 2 Data and background 2.1 Data In order to analyze the relationship between trade, emissions and abatement, we use manu- facturing census data for Sweden. The census data contains information at the (cid:28)rm level for a large number of variables such as export, employment (number of employees), capital stock , value of purchase of intermediates and value of output. Reported values are in thousand Swedish kroner (tSEK). Our (cid:28)rm level data covers the period 2000-2011 and include all (cid:28)rms with at least one employee. This leaves us with an unbalanced panel of around 23000 (cid:28)rms per year. We moreover have data for three types of environmental emissions, SO2, NOx and CO2. Information on emissions is, however, not available for the whole panel of manufacturing (cid:28)rms. As for SO2 and NOx emissions, we rely on calculations made by Statistics Sweden. Statistics Sweden calculate SO2 and NOx emissions on an annual basis for all manufacturing (cid:28)rms that uses at least 325 tons of oil equivalents the respective year. As for CO2 emissions, we rely on own calculations exploiting data on energy usage and emissions coe(cid:30)cients. Statistics Sweden collect information on the usage of energy from all manufacturing plants with 10 or more employees, and we have data for the time period 2005-2011. The energy statistics include all types of fuel use, from which CO2 emissions can be calculated by using fuel speci(cid:28)c CO2 emissions coe(cid:30)cients provided by Statistics Sweden. 5 CO2 emissions are accurately calculated from fuel inputs since a technology for capturing CO2 at the pipe is not yet operational.2 The calculated plant level emissions are aggregated to the (cid:28)rm level. We match the (cid:28)rm level emission data with the census data.3 Note that the data at hand allows us to estimate emission intensity as tonnes of emissions relative to value added - rather than relative to sales or the value of output. Hence, unlike other studies we implicitly take into account di(cid:27)erences across (cid:28)rms with respect to outsourcing when measuring emission intensity. Wealsohaveaccessto(cid:28)rmleveldataonabatementovertheperiod2000-2011. Theabate- ment data is collected based on an annual survey where (cid:28)rms are asked about abatement investments(tSEK)aswellasvariableabatementcosts(tSEK).The(cid:28)rmsareaskedtoreport not only investments in machines and equipment speci(cid:28)cally aimed at reducing emissions, but in addition to report extra expenses related to investment in relative more environment friendly machines and technology. Hence, investments which allow for fuel-switching or in- creased energy e(cid:30)ciency are also counted. Firms are asked to report abatement related to air, water and waste. The abatement data is based on a semi-random sample of manufac- turing (cid:28)rms, and include all manufacturing (cid:28)rms with more than 250 employees, 50 percent of the (cid:28)rms with 100-249 employees, and 20 percent of the (cid:28)rms with 50-99 employees. In total, around 1500 manufacturing (cid:28)rms are surveyed over the time period 2000-2011. Swedish (cid:28)rms face uniform SO2 and a NOx taxes which were introduced in 1990 and 1992 respectively. Swedish manufacturing (cid:28)rms also face a CO2 tax. Sweden enacted a tax on carbon emissions in 1991 which has applied throughout our period of observation. The tax is a general one, and applies to all sectors, but manufacturing industries have from the introduction of the tax been granted a tax credit. The tax credit is uni(cid:28)ed and identical across industries and (cid:28)rms. Moreover, in 2005 the European Union Emissions Trading System (EU-ETS) was set up, of which Sweden is a member. The EU-ETS mainly applies to (cid:28)rms in the energy intensive industries,4 but also to some energy intensive (cid:28)rms outside these industries. The (cid:28)rms included in the EU-ETS face the quota regime but are on the other hand exempted from the national CO2 tax. Our dataset identi(cid:28)es all (cid:28)rms with an EU-ETS quota. 2A few large power plants are experimenting with capturing CO2 under ground, but as we are focusing on manufacturing, these are not included in our data. 3We are left with a (cid:28)rm level panel with CO2 emission information of around 3700 manufacturing (cid:28)rms per year for the period 2005-2011, and a (cid:28)rm level panel with SO2 and NOx emission information of around 550 (cid:28)rms for the period 2000-2011. 4The energy intensive industries are paper and pulp (17), coke and re(cid:28)ned petroleum products (19), chemicals (20), non-metallic mineral products (23), and basic metals (24). 6 2.2 Basic Facts on Swedish Firms’ Environmental Emissions and Trade The manufacturing sector was in 2010 responsible for 28 percent of the CO2 emissions, 24 percentoftheSO2emissionsand11percentoftheNOxemissionsinSweden. Butneedlessto say there are huge di(cid:27)erences in environmental intensities across individual industries within the manufacturing sector. The energy intensive industries have much higher emissions as well as emission intensities than the other industries. So far these inter-industry variations have got the most attention from academics and policy makers. Hence, also analyses of envi- ronmentalemissionsandinternationaltradehaveuntilrecentlymainlyfocusedondi(cid:27)erences in emissions across sectors and industries as surveyed by Copeland and Taylor (2004) and Brunnermeier and Levinson (2004). However, we conduct a simple decomposition of the variation in environmental emission intensity (measured as tonnes of emissions relative to value added) of Swedish manufacturing (cid:28)rms, splitting the variation in emission intensity into (i) variation across (cid:28)rms within sectors and (ii) variation between sectors. Table 1 shows that the majority of the variation in emission intensity can be ascribed to (cid:28)rm heterogeneity within rather narrowly sectors.5 Table 1: Decomposition of Environmental Emissions Within sectors (5 digit) Between sectors CO2 emission intensity 85% 15% SO2 emission intensity 64% 36% NOx emission intensity 94% 6% Note: Environmental emission intensity is measured as tonnes of emissions relative to value added.6 Our hypothesis is that the inter-(cid:28)rm di(cid:27)erences in emission intensities may be linked to otherheterogeneouscharacteristicsofthe(cid:28)rmsandinparticulartotheirinternationalization. Analyses of various countries (see e.g. Bernard et al., 2007) have shown that exporters are bigger, more productive and more capital intensive. As shown in Table 10 in the Appendix, our data for Swedish manufacturing (cid:28)rms con(cid:28)rms these stylized facts. Swedish exporters employ more people, have relatively higher investment in capital and have higher total factor productivity.7 Hence, we proceed by comparing the emission intensity of exporters and non- exporters. We do this for CO2, SO2 and NOx emissions. We report the ratio of average emission intensity of exporters relative to non-exporters for all sectors, for energy intensive 5See the Appendix for details on the decomposition calculation. 7An exporter is de(cid:28)ned as a (cid:28)rm with foreign sales of any amount, but we have also run our regressions with exporters de(cid:28)ned as (cid:28)rms with sales above ten or hundred thousand dollars. This does not a(cid:27)ect the results. 7 sectors and for non-energy intensive sectors, see Table 2.8 The picture is not quite clear. But we note that in the non-energy intensive sectors, which account for more than 80 percent of manufacturing employment exporters’ emission intensity is on average much lower than that of non-exporters. Doing a count of industries, we also (cid:28)nd that in 13 out of 24 manufacturing industries (NACE 2 digit level) exporters’ CO2 and SO2 environmental emission intensity is lower than that of non-exporters, while the number of industries where exporters have lower NOx emission intensity than non-exporters, is 14 out of 24. Table 2: Environmental emission intensities All sectors Emission intensity: Exporters vs. Non-Exporters All Sectors Energy intensive Non-Energy intensive CO2 emission intensity 2.37 2.15 0.42 SO2 emission intensity 0.85 1.06 0.45 NOx emission intensity 1.20 2.41 0.38 Note: Environmentalemissionintensityismeasuredasenvironmentalemissionsrelativetovalueadded. 9 Inlinewiththede(cid:28)nitionappliedbySwedishauthorities,5industriesarecategorizedasenergyintensive,and19asnon-energyintensive. Motivated by the facts on environmental emissions and their variation across (cid:28)rms, we proceed by developing a simple theory of heterogeneous (cid:28)rms where (cid:28)rms within an industry di(cid:27)er in their emissions. In particular, we propose and develop a mechanism for why emis- sions may di(cid:27)er across (cid:28)rms, and why export performance may have an impact on (cid:28)rms’ emissions. 3 The Model We develop a model with international trade and heterogeneous (cid:28)rms (see Melitz (2003)) whose production entails environmental emissions.. Firms that are productive enough to set up production make two distinct decisions, whether to enter the export market and how much to invest in abatement to reduce emissions. Firms make these decision subject to trade costs and emission taxes. We consider the case of two countries, Home and Foreign (denoted by ”∗”). Each econ- omy is active in the production in two industries: a monopolistic competitive industry (M) where (cid:28)rms produce di(cid:27)erentiated goods under increasing returns and subject to environ- mental emissions, and a background industry (A) characterized by perfect competition and which produces homogenous goods subject to constant returns to scale. To make things simple, we shall assume that there is just one factor of production. This may be a composite 8The energy intensive sectors are paper and pulp (17), coke and re(cid:28)ned petroleum products (19), chem- icals (20), non-metallic mineral products (23), and basic metals (24). 8 factor, but for the sake of simplicity we shall refer to it as labour. We present the equations describing Home(cid:48)s consumers and (cid:28)rms, and note that corresponding equations apply to Foreign. The theoretical model allows us to derive analytical expressions for equilibrium emission intensity and equilibrium abatement investments, and to analyze the relationship between emission intensity, abatement investment and trade. Our analysis delivers predictions on export performance, emission and abatement. In Section 4 we proceed by testing empirically these theoretical predictions using the Swedish manufacturing (cid:28)rm level data. 3.1 Demand Consumers preferences are given by a two-tier utility function with the upper tier (Cobb- Douglas) determining the representative consumer’s division of expenditure between goods produced in sectors A and M, and the second tier (CES), giving the consumer’s preferences over the continuum of di(cid:27)erentiated varieties produced within the manufacturing sector. Hence, all individuals in Home have the utility function U = Cµ C1−µ, (1) M A where µ ∈ (0,1) and C is consumption of the homogenous good. Goods produced in the A A sector can be costlessly traded internationally and are produced under constant returns to scale and perfect competition. The A-good is chosen as the numeraire, so that the world market price of the agricultural good, p , is equal to unity. By choice of scale, the labour A requirement in the A-sector is one, which gives p = w = 1 (2) A and thus, wages are normalized to one across both countries and sectors. We assume that demand for A goods is su(cid:30)ciently large to guarantee that the A sector is active in both countries. The consumption of goods from the M sector is de(cid:28)ned as an aggregate C , M ˆ σ/(σ−1) CM =  c(i)(σ−1)/σdi , (3) i∈I where c(i) represents consumption of each variety with elasticity of substitution between any pair of di(cid:27)erentiated goods being σ > 1. The measure of the set I represents the mass of varieties consumed in the Home country. Each consumer spends a share µ of his income on 9 goods from industry M, and the demand for each single variety produced locally and in the foreign country is therefore given by respectively p−σ x = µL (4) d P1−σ τ1−σ(p∗)−σ x = µL, e P1−σ (cid:18)´ (cid:19) 1 1−σ where p is the consumer price, is income, and P ≡ p(i)1−σdi the price index of M i∈I goods consumed in the Home country. Products from Foreign sold in Home incur an iceberg trade cost τ, i.e. for each unit of a good from Foreign to arrive in Home, τ > 1 units must be shipped. It is assumed that trade costs are equal in both directions. 3.2 Entry, Exit and Production Costs in the M Sector To enter the M sector in country j, a (cid:28)rm bears the (cid:28)xed costs of entry f measured in E labour units. After having sunk f , an entrant draws a labour-per-unit-output coe(cid:30)cient a E from a cumulative distribution functionG(a). We follow Helpman et al. (2004) in assuming (cid:16) (cid:17)k the probability distribution to be a Pareto distribution,10 i.e. G(a) = a , where k is the a0 shape parameter, and we normalize the scale parameter to unity, a ≡ 1. Since a is unit 0 labour requirement, 1/a depicts labour productivity. Upon observing this draw, a (cid:28)rm may decide to exit and not produce. If it chooses to stay, it bears the additional (cid:28)xed overhead costs, f . If the (cid:28)rm does not only want to serve the domestic market but also wants to D export, it has to bear the additional (cid:28)xed costs, f . Hence, (cid:28)rm technology is represented X by a cost function that exhibits a variable cost and a (cid:28)xed overhead cost. In the absence of emissions and abatement investment, labour is used as a linear function of output according to l = f +ax (5) with f = f for (cid:28)rms only serving the domestic market and f = f +f for exporters. We D D X make the simplifying assumption that not just variable costs but also all types of (cid:28)xed costs are incurred in labour. However, since we do not focus on issues related to factor markets or comparative advantage, this only serves as means to simplify the analysis, without having any impact on the results. Industrial activity in sector M entails pollution in terms of environmental emissions. We follow Copeland and Taylor (2003) and assume that each (cid:28)rm produces two outputs: an 10This assumption is consistent with the empirical (cid:28)ndings by e.g. Axtell (2001). 10