Why are firms that export cleaner? International trade and CO 2 Emissions

Transcription

1 Why are firms ha expor cleaner? Inernaional rade and CO 2 Emissions Rikard Forslid, Toshihiro Okubo and Karen Helene Ullvei-Moe

2 Why are firms ha expor cleaner? Inernaional rade and CO2 emissions Rikard Forslid, Toshihiro Okubo, and Karen Helene Ullvei-Moe This version, May 204 Absrac This paper develops a model of rade and CO2 emissions wih heerogenous firms, where firms make abaemen invesmens and hereby have an impac on heir level of emissions. The model shows ha invesmens in abaemens are posiively relaed o firm produciviy and firm expors. Emission inensiy is, however, negaively relaed o firms produciviy and expors. The basic reason for hese resuls is ha a larger producion scale suppors more invesmens in abaemen and, in urn, lower emissions per oupu. We show ha he overall effec of rade is o reduce emissions. Trade weeds ou some of he leas producive and diries firms hereby shifing producion away from relaively diry low producive local firms o more producive and cleaner exporers. The overall effec of rade is herefore o reduce emissions. We es empirical implicaions of he model using unique Swedish firm-level daa. The empirical resuls suppor our model. JEL Classificaion: F2, F4, F8, Q56 Keywords:heerogeneous firms, CO2-emissions, inernaional rade We are graeful for commens from Andrew Bernard, Peer Egger, Peer Fredriksson, Beaa Javorcik, Gordon Hanson, Peer Neary, Sco Taylor, Adrian Wood, and Tony Venables. Financial suppor from Jan Wallander and Tom Hedelius Research Foundaion, The Swedish Research Council, Gran-in-Aid for Scienific Research (JSPS) and Research Insiue of Economy, Trade and Indusry (RIETI) is graefully acknowledged. Sockholm Universiy and CEPR; rf@ne.su.se Keio universiy, okubo@rieb.kobe-u.ac.jp Universiy of Oslo and CEPR, k.h.ullvei-moe@econ.uio.no

3 Inroducion Thereisnoconsensusonheeffec of inernaional rade on he environmen, in paricular on he effec of rade on global emissions. Neiher he heoreical nor he empirical lieraure provides a clean cu answer o he link beween rade and CO2 emissions. Hence, we do no know if inernaional rade increases or decreases he emissions of greenhouse gases and conribues o global warming. However, his paper ses ou explain why we may expec exporer o emi less CO2, and why rade liberalizaion may hus lead o cleaner indusrial producion. We do so by focusing on iner-firm produciviy differenials and inerdependence among produciviy, exporing, abaemen and CO2 emissions. In heoreical neoclassical models, inernaional rade has opposing effecs. On he one hand, rade increases income, which will end o increase he demand for a clean environmen and herefore increase invesmens in clean echnology and abaemen. On he oher hand, rade liberalizaion may also imply an overall expansion of diry producion, because rade allows counries wih low emission sandards o become polluion havens. Copeland and Taylor (995) show how rade liberalizaion may increase global emissions ifheincomedifferences beween he liberalizing counries are large, as diry indusries are likely o expand srongly in he poor counry wih low environmenal sandards. The empirical lieraure ha analyses he link beween rade in goods and emissions based on secor level daa is also inconclusive. Anweiler e al. (200) and Frankel and Rose (2005) find ha rade decreases emissions. Using U.S. daa, Ederingon e al. (2004) do no find any evidence ha polluion inensive indusries have been disproporionaely affeced by ariff changes. On he oher hand, also using secor-level rade daa, Levinson and Taylor (2008) find evidence ha higher environmenal sandards in he US have increased he impors from Mexico in diry indusries. We employ a unique firm level daa based on he Swedish Manufacuring Census which do no only provide informaion on sandard firm characerisics like oupu, value added ec, bu also on heir inernaional rade as well as CO2 emissions. Based on hese daa, we sar by providing evidence ha firms emissions differ significanly across firms, even wihin raher narrowly defined indusries. Moreover, comparing non-exporers and exporers in Swedish manufacuring, we find ha in mos manufacuring indusries exporers do on average have a lower emission inensiy. Moivaed by hese basic facs we build a heoreical model on inernaional rade and CO2 emissions where firms are heerogeneous wih respec o produciviy, abaemen invesmens and emission inensiy. We propose and develop a mechanism for why exporers may have a lower emission inensiy. This mechanism runs hrough firms invesmens in abaemen. According o he heory firms abaemen invesmens depend on heir producion volumes, as a larger scale allow hem o spread he fixed coss of abaemen invesmen across more unis. Early surveys are made by Copeland and Taylor (2004) and Brunnermeier and Levinson (2004). 2

4 Producion volumes are moreover deermined by firms produciviy and expor saus. More producive firms access inernaional markes, have higher volumes and make higher abaemen invesmens. As a consequence, firms emission inensiy is negaively relaed o firms produciviy and expor saus. Our heoreical model also allows for predicion on he impac of rade liberalizaion on oal CO2 emissions. We find ha oal emissions from he manufacuring secor decreases as a resul of rade and rade liberalizaion. Trade affecs he exporing and non-exporing secor in differen ways. Exporers are for any level of rade coss always cleaner han non-exporers, and we show ha rade liberalizaion may make exporers even cleaner by inducing hem o inves more in abaemen. Bu rade liberalizaion also implies higher producion volumes for exporers, which c.p. enails higher emissions. Toal emissions herefore increases from he exporing secor. However, rade moreover increases local compeiion, which implies ha he leas producive, and herefore diries, firms are forced o close down, while he remaining non-exporers are forced o scale down heir producion volume. Togeher hese differen effecs of rade liberalizaion serve as o decrease oal emissions from he non-exporing secor. Adding up he effecs on exporers and non-exporers we find ha rade liberalizaion will always lead o lower oal emissions. Thus, as rade weeds ou some of he leas producive and diries firms, hereby shifing producion away from relaively diry low producive local firms o more producive and cleaner exporers, he overall effec of rade liberalizaion is o reduce emissions. The heoreical model allows us o derive a se of empirical predicions on emissions and exporing as well as abaemen invesmen and exporing. Access o he deailed firm level daa se for Swedish manufacuring firms allow us o es hese. Our daa se conains firm-level emissions and firm-level abaemen invesmens as well as firm expors. According o our model, produciviy drives he firm level CO2 emission inensiy as well as he expor saus of a firm. However, while produciviy has a coninuos effec on he emission inensiy, he model predics a disconinuous jump down in he emission inensiy as firms become exporers. The same kind of relaionship is prediced for abaemen and exporing. We exploi hese feaures of he model as we ake he model o he daa. The empirical resuls are srongly supporive of he resuls derived in he heoreical model; exporers are found o inves more in abaemen and o have lower emission inensiy. Our heory is relaed o he idea presened in Levinson (2009) ha rade may conribue o reduced polluion as rade liberalizaion may encourage echnological upgrading. From a more mehodological poin of view, our work is also relaed o he lieraure on heerogeneous firms and rade induced echnological upgrading, see e.g. Bas (202) and Busos (202). The majoriy of empirical analyses of environmenal emissions are, unlike our sudy, based on indusry level daa. There are, however, a few excepions, which are hus closer in he spiri o our analysis, see e.g. Holladay (20), Barakova 3

5 and avies (202), and Rodrigue and Soumonni (204). Holladay analyses firm-level daa for he US, and find ha exporers pollue less per oupu. However, his analysis is based on a sample of firms wih large emissions, he sudies environmenal emissions no CO2 emission and he does no invesigae he mechanism for why exporing and emission inensiy is relaed. The srucure of he paper is as follows. In he nex secion we presen a daa se on Swedish manufacuring firms and heir CO2 emissions. Based on hese we develop a se of basic facs on he variaion in CO2 emission inensiy across indusries and firms, and examine he differences in emission inensiy beween non-exporers and exporers. We le he descripive evidence on emissions and how hey vary, guide our heoreical model on inernaional rade, CO2 emissions and heerogeneous firms. We presen his model in Secion 3, and i allows us o derive a se of proposiions and empirical implicaions regarding CO2 emissions, abaemen and rade. In Secion 4 we ake he heory o he daa, and es he empirical predicions on he relaionship beween CO2 emissions, expor and produciviy, and on he relaionship beween abaemen, expor and produciviy. Finally, Secion 5 concludes. 2 aa and background 2. aa In order o analyze he relaionship beween rade, emissions and abaemen, we use manufacuring census daa for Sweden. The census daa conains informaion on he firm level for a large number of variables such as expor (SEK), employmen (number of employees), capial sock (SEK), use of inermediaes (SEK) and value of oupu (SEK). Our firm level daa cover he period Saisics Sweden collec informaion on he usage of energy from all manufacuring plans wih 0 or more employees, and we have access o hese for he ime period The energy saisics include all ypes of fuel use, from which CO2 emissions (kg) can be calculaed by using fuel specific CO2 emissions coefficiens provided by Saisics Sweden. CO2 emissions are accuraely calculaed from fuel inpus since a echnology for capuring CO2 a he pipe is no ye operaional. 2 The calculaed plan level emissions are aggregaed o he firm level. This provides us wih CO2 emission daa for around 9500 manufacuring firms for he years , which we mach wih he census daa. 3 2 A few large powerplans are experimening wih capuring CO2 under ground, bu as we are focusing on manufacuring, hese are no included in our daa. 3 Noe, ha as he census provides daa for all firms, limiing he he number o firms wih a leas one employee, we sar wih a number of firms for he period However, due o he fac ha energy saisics only are colleced for plans wih 0 or more employees, his reduces our sample wih close o 50 percen. 4

6 We also have access o firm level daa on abaemen over he period The abaemen daa is colleced based on an annual survey where firms are asked abou abaemen invesmens (SEK) as well as variable abaemen coss (SEK). As for abaemen invesmen he firms are asked o repor any invesmen in machines and equipmen specifically aimed a reducing emissions, bu also o repor expenses relaed o invesmen in cleaner machines and echnology. In he laer case hey are specifically asked o repor he exra expenses relaed o he choice of invesing in cleaner relaive o less clean machines and echnology. The abaemen daa is based on a semi-random sample of manufacuring firms, and include all manufacuring firms wih more han 250 employees, 50 percen of he firms wih employees, and 20 percen of he firms wih employees. In oal, around 500 manufacuring firms are surveyed over he ime period Swedish manufacuring firms face a CO2 ax. Sweden enaced a ax on carbon emissions in 99 which has applied hroughou our period of observaion. The ax is a general one, and applies o all secors, bu manufacuring indusries have from he inroducion of he ax been graned a ax credi. The ax credi is unified and idenical across indusries. Bu in addiion o he general ax credi, he mos energy inensive, and hus he mos emission inensive indusries, defined as hose wih a CO2 ax bills exceeding 0.8 percen of heir producion value, ge a furher ax credi. 4 Sweden is also par of European Union Emissions Trading Sysem (EU-ETS) which was se up in 2005 o reduce CO2 emissions. The EU-ETS applies only o firms in he energy inensive secors. Our period of observaion coincide parly wih he so called firs and second rading periods of EU-ETS ( and ). Noe ha quoas were in general disribued for free during hese rading periods, and he price of quoas in he second hand marke has been very low due o he recession in Europe during he second rading period. 2.2 BasicFacsonCO2EmissionsandTrade The manufacuring secor is responsible for around 40 percen of he CO2 emissions in Sweden. Bu needless o say here are huge differences in CO2 inensiies across individual indusries. The energy inensive indusries have much higher emissions as well as emission inensiies (CO2 emissions relaive o oupu) han he oher indusries. So far he inerindusry variaions have go he mos aenion from academics and policy makers. Hence, also analyses of CO2 emissions and inernaional rade have unil recenly mainly focused on differences in emissions across secors and indusries as surveyed by Copeland and Taylor (2004) and Brunnermeier and Levinson (2004). However, a simple decomposiion of he variaion in CO2 emission inensiy of Swedish manufacuring firms ino (i) variaion 4 The energy inensive secors are paper and pulp (7), coke and refined peroleum producs (9), chemicals (20), non-mealic mineral producs (23), and basic meals (24). 5

7 across firms wihin secors and (ii) variaion beween secors, shows ha he majoriy of he variaion in emission inensiy can be ascribed o firm heerogeneiy wihin acually raher narrowly secors. According o Table, almos 70 percen of he variaion in CO2 emissionsisdueodifferences beween firms raher han iner-secoral differences. Table : ecomposision of CO2 emissions Wihin secor (5 digi) Beween secors CO2 emission inensiy 67% 33% Noe: CO2 emission inensiy is measured as CO2 emissions relaive o oupu. Our hypohesis is ha he iner-firm differences in emission inensiies may be linked o oher heerogeneous characerisics of he firms and in paricular o heir inernaionalizaion. Analyses of various counries (see e.g. Bernard e al, 2007) have shown ha exporers are bigger, more producive and more capial inensive. As shown in Table 2, our daa for Swedish manufacuring confirms hese sylized facs. Table 2: Firm characerisics: Exporers versus Non-Exporers Exporers Non-Exporers obs. mean sd.dev. obs. mean sd.dev. Produciviy (TFP) Capial/labour (SEK/employee) Employees Turning o he relaionship beween exporing and emission inensiy Table 3 repor average CO2 emission inensiy for exporers versus non-exporers for all manufacuring secors as well as for energy inensive and non-energy inensive secors separaely. 5 The picure is no quie clear. Bu we noe ha in he non-energy inensive secors, which accoun for more han 80 percen of manufacuring employmen exporers emission inensiy is on average lower. oing a coun of indusries, we also find ha in 7 ou of 24 manufacuring indusries (2 digi level) exporers CO2 emission inensiy is lower han ha of non-exporers. Table 3: ecomposision of CO2 emissions CO2 emission inensiy All secors Energy inensive secors Non-energy inensive secors Exporers Non-Exporers Noe: CO2 emission inensiy is measured as CO2 emissions relaive o oupu. Ou of 24 manufacuring indusries, here are 5 energy inensive and 9 non-energy inensive. 5 The energy inensive secors are paper and pulp (7), coke and refined peroleum producs (9), chemicals (20), non-mealic mineral producs (23), and basic meals (24). 6

8 Moivaed by he facs on CO2 emissions and heir variaion across firms, we proceed by developing a simple heory of heerogeneous firms where firms wihin an indusry differ in heir emissions. In paricular, we propose and develop a mechanism for why emissions may differ across firms, and why expor performance may have an impac on firms emissions. Moreover, we le he heory be guided by he environmenal ax regime facing Swedish manufacuring firms. 3 The Model We develop a model wih inernaional rade and heerogeneous firms (see Meliz (2003)) whose producion enails emiing CO2. Firms ha are producive enough o se up producion make wo disinc decisions, wheher o ener he expor marke and how much o inves in abaemen o reduce emissions. Firms make hese decision subjec o rade coss and emission axes. We consider he case of wo counries, Home and F oreign (denoed by ). Each economy is acive in he producion in wo indusries: a monopolisic compeiive indusry (M)wherefirms produce differeniaed goods under increasing reurns and subjec o CO2 emissions, and a background indusry (A) characerized by perfec compeiion and which produces homogenous goods subjec o consan reurns o scale. To make hings simple, we shall assume ha here is jus one facor of producion. This may be a composie facor, bu for he sake of simpliciy we shall refer o i as labor. We presen he equaions describing Homes consumers and firms, and noe ha corresponding equaions apply o F oreign. The heoreical model allows us o derive analyical expressions for equilibrium emission inensiy and equilibrium abaemen invesmens, and o analyze he relaionship beween emission inensiy, abaemen invesmen and rade. Our analysis delivers predicions expor performance, emission and abaemen. In Secion 4 we proceed by esing empirically hese heoreical predicions using he Swedish manufacuring firm level daa. 3. emand Consumers preferences are given by a wo-ier uiliy funcion wih he upper ier (Cobb- ouglas) deermining he represenaive consumer s division of expendiure beween goods produced in secors A and M, and he second ier (CES), giving he consumer s preferences over he coninuum of differeniaed varieies produced wihin he manufacuring secor. Hence, all individuals in Home have he uiliy funcion U = C μ M C μ A, () 7

9 where μ (0, ) and C A is consumpion of he homogenous good. Goods produced in he A secor can be coslessly raded inernaionally and are produced under consan reurns o scale and perfec compeiion. The A-good ischosenashenumeraire, soha he world marke price of he agriculural good, p A, is equal o uniy. By choice of scale, he labor requiremen in he A-secor is one, which gives p A = w = (2) and hus, wages are normalized o one across boh counries and secors. We assume ha demand for A goodsissufficienly large o guaranee ha he A secor is acive in boh counries. The consumpion of goods from he M secor is definedasanaggregaec M, C M = i I c (i) (σ )/σ di σ/(σ ), (3) where c(i) represens consumpion of each variey wih elasiciy of subsiuion beween any pair of differeniaed goods being σ >. The measure of he se I represens he mass of varieies consumed in he Home counry. Each consumer spends a share μ of his income on goods from indusry M, and he demand for each single variey produced locally and in he foreign counry is herefore given by respecively where p is he consumer price, Y is income, and P x d = p σ μy (4) P σ x e = τ σ (p ) σ μy, P σ i I p (i) σ σ di he price index of M goods consumed in he Home counry. Producs from Foreign sold in Home incur an iceberg rade cos τ, i.e. for each uni of a good from F oreign o arrive in Home, τ > unis mus be shipped. I is assumed ha rade coss are equal in boh direcions. 3.2 Enry, Exi and Producion Coss in he M Secor To ener he M secor in counry j, afirm bears he fixed coss of enry f E measured in labour unis. Afer having sunk f E, an enran draws a labour-per-uni-oupu coefficien a from a cumulaive disribuion G(a). We follow Helpman e al. (2004) in assuming he k probabiliy disribuion o be a Pareo disribuion, 6 a i.e. G(a) = a 0, where k is he shape parameer, and we normalize he scale parameer o uniy, a 0. Since a is uni labour requiremen, /a depics labour produciviy. Upon observing his draw, a firm 6 This assumpion is consisen wih he empirical findings by e.g. Axell (200). 8

10 may decide o exi and no produce. If i chooses o say, i bears he addiional fixed overhead coss, f.ifhefirm does no only wan o serve he domesic marke bu also wans o expor, i has o bear he addiional fixed coss, f. Hence, firm echnology is represened by a cos funcion ha exhibis a variable cos and a fixed overhead cos. In he absence of emissions and abaemen invesmen, labour is used as a linear funcion of oupu according o l = f + ax (5) wih f = f for firms only serving he domesic marke and f = f + f for exporers. We make he simplifying assumpion ha no jus variable coss bu also all ypes of fixed coss are incurred in labor. However, since we do no focus on issues relaed o facor markes or comparaive advanage, his only serves as means o simplify he analysis, wihou having any impac on he resuls. Indusrial aciviy in secor M enails polluion in erms of emission of CO2. We follow Copeland and Taylor (2003) and assume ha each firm produces wo oupus: an indusrial good (x) and emissions (e). In order o reduce emissions, a firm can diver a fracion θ of he primary facor, labour, away from he producion of x. Wemayhinkof θ as a variable abaemen expendiure ha is chosen opimally by each firm. The join producion of indusrial goods and emissions is given by x =( θ) l a (6) e = ϕ(θ) l a (7) wih 0 θ <. Emissions are deermined by he abaemen funcion ϕ(θ) = ( θ)/α h (f A ) (8) which is characerized by ϕ(0) =, ϕ() = 0, ϕ (.) < 0 and 0 < α <. The abaemen funcion reflecs ha firms may reduce heir emission inensiy hrough wo ypes of abaemen aciviies ha incur variable and fixed coss respecively. As already noed, θ deermines he variable abaemen coss, while f A represen invesmens in abaemen, e.g. machines and equipmen ha allow for reduced emissions. 7 A given reducion of emissions may be reached eiher hrough increased θ or hrough increased f A, since we assume h (f A ) > 0. We proceed by using (7) o subsiue for in (8), and in urn (8) o subsiue for θ in (6), which gives us an inegraed expression for he join producion of goods and 7 We depar from he sandard formulaion of he abaemen funcion in he lieraure on rade and emissions by assuming ha firms can have an impac on emission inensiy hrough fixed abaemen invesmens (f A ). 9

11 emission, and explois he fac ha alhough polluion is an oupu, i can equivalenly alsobereaedasaninpu: 8 x =(h (f A ) e) α l a α. (9) Hence, wih such an inerpreaion, producion implies he use of labor as well as emission. Noe ha while firms are heerogeneous wih respec o labour produciviy and abaemen, hey are idenical wih respecs o he srucure of heir basic producion echnology and face he same ax rae on emissions. Firms minimize coss subjec o he producion funcion (9), aking wages (w =) andemissionaxes( > 0) asgiven. isregarding he sunk enry cos (f E )wecanderivefirms oal cos funcion using (5) and (9). α C = f + f A + κ a ( α) x (0) h (f A ) wih κ α α ( α) α and where f = f for firms only serving he domesic marke, and f = f + f for exporers, i.e. firms serving boh he domesic and he foreign marke. The cos funcion reflecs ha emissions are no for free, raher hey incur a ax >0. Bu hrough increasing heir invesmens in abaemen, firms can reduce heir emissions as well as heir ax bill. Hence, in conras o he oher fixed coss, invesmen in abaemen is an endogenous variable. Our analysis focuses on seady-sae equilibria and ineremporal discouning is ignored. The presen value of firmsiskepfinie by assuming ha firms face a consan Poissonhazardraeδ of deah independenly of produciviy. An enering firm wih produciviy a will immediaely exi if is profi levelπ (a) is negaive, or will produce and earn π (a) 0 in every period unil i is hi by a bad shock and forced o exi. 3.3 Profi Maximizaion Having drawn heir produciviy, firms follow a wo-sep decision process. Firs, hey decide on abaemen invesmen, and second, aking abaemen as given, hey maximize profis. We solve heir problem using backwards inducion: Firms firs calculae heir opimal pricing rule given abaemen invesmens, second hey make heir decision on abaemen invesmen given he opimal pricing rule. Implicily hey hen also decide on emission inensiy and on share of inpu facor o diver away from producion and owards abaemen, i.e. he variable coss of abaemen. Each producer operaes under increasing reurns o scale a he plan level and in line wih ixi and Sigliz (977), we assume here o be large group monopolisic compeiion beween he producers in he M secor. Thus, he perceived elasiciy of demand equals 8 See Copeland and Taylor (2003) for a discussion of his feaure of he model. 0

12 he elasiciy of subsiuion beween any pair of differeniaed goods and is equal o σ. Regardless of is produciviy, each firm hen chooses he same profi maximizing markup over marginal coss (MC)equaloσ/(σ ). This yields a pricing rule p = σ MC () σ for each producer. Using (4) and (0) we can formulae he expression for firms profis. We le super- and subscrip and denoe non-exporers and exporers respecively. Firms only serving he domesic marke earn profis α σ π = a α B f f A, (2) h(f A ) while he exporing firms, serving boh he local and he foreign marke, earn profis α σ π = a α (B + φb ) f f f A, (3) h(f A ) where B κ σ σ σ (σ ) σ μl in an index of he marke poenial of he home counry, P σ and B κ σ σ σ (σ ) σ μl depics he marke poenial of he foreign counry, and (P ) σ φ = τ σ 0, ] depics he freeness of rade. 3.4 Fixed Cos Invesmens in Abaemen Having solved he second sage of firms decision problem, we proceed o he firs sage. In order o be able o derive explici analyical expression for abaemen invesmens we employ he specific funcional form h (f A ) = f ρ A, wih ρ > 0. Since firms profis depend on wheher hey are exporers or non-exporers, abaemen invesmens will differ beween he wo groups of firms. Maximizing non-exporing firms profis wih respec o abaemen invesmens f A using (2) gives: fa = ΩB α(σ ) a ( α)(σ ), (4) wih αρ(σ ) and Ω (αρ (σ )), while he opimal invesmen in abaemen for exporers is found using (3): f A = Ω (B + φb ) α(σ ) a ( α)(σ ). (5) From (4) and (5) follow ha firms abaemen invesmens depend on heir exogenously given marginal produciviy, axes, and he marke poenial. 9 An inernal soluion o he 9 Noe ha he effecs of rade liberalisaion (a higher φ) canno be seen from his equaion since B and B are funcions of φ.

13 profi maximizing choice of fa and f A, requires ha > 0. However, as his condiion is a necessary condiion for profi maximizaion, we assume i always o hold, see secion A. in he Appendix for deails. Having examined (4) and (5) we can formulae he following proposiions on he relaionship beween abaemen invesmens and firm characerisics. Proposiion More producive firms inves more in abaemen. Proof : The saemen follows direcly from (4) and (5). The logic behind his resul is ha more producive firms have higher sales. Hence, he exploiing of scale economies makes i profiable for hem o make a higher invesmen in order o reduce marginal coss. Proposiion 2 For any given level of produciviy, exporers inves more in abaemen han non-exporers. Proof: Since B+φB B > i follows from (4) and (5) ha fa >f A produciviy level (/a). for any given 3.5 Cu off Condiions and Free Enry Finally, basedonequilibriumabaemen invesmens, we can now deermine he cu off condiions he wo ypes of acive firms. The cu off produciviy level for firms only serving he domesic marke (/a )idenifies he lowes produciviy level of producing firms. From (2) and (3), we see ha profis are increasing in firms produciviy. The leas producive firms expec negaive profis and herefore exi he indusry. This applies o all firms wih a uni labor inpu coefficien above a, he poin a which profis from domesic sales equal zero, and is deermined by a α h(f A ) α σ B = f + f A. (6) Wih σ > i follows ha a ( α)( σ) increases along wih produciviy and can hus be used as a produciviy index. Exporers abaemen invesmens affec he producion and profisearnedboh inhehomemarkeandheforeignmarke. Thecu-off produciviy level for exporers (a )idenifies he lowes produciviy level of exporing firms, and is given by he produciviy level where he expor profis plus he ne exra profi inhe home marke from he higher abaemen invesmens equals he exra fixed coss incurred by exporing and he incremenal invesmen in abaemen: a α h(f A ) α σ φb + a α h(f A ) α σ B a α h(f A ) α σ B = f +f A f A, (7) 2

14 We noe ha since abaemen invesmens have an impac on firms marginal coss, i also affecs he profiabiliy of being a domesic versus an exporing firm. 0 The model is closed by he free-enry condiion 3.6 CO2 Emissions a f E = π dg(a)+ 0 a 0 π dg(a). (8) Taking abaemen invesmen as given, firms decide on heir use of labour as well as on emissions. As we are primarily ineresed in emissions, we shall focus on hese. Firms paricipaion in rade affecs heir invesmen in abaemen and herefore he emission inensiy (emissions relaive o oupu) of firms. The general expression for emission inensiy is found by using Shepard s lemma on he cos funcion (0) as we exploi ha due o he special feaures of he model, emissions appear no only as an oupu of producion, bu also as an inpu o producion: e x = ακα f ρα A a α (9) We see ha here is a simple relaionship beween abaemen invesmen and emission inensiy. The more a firm inves in abaemen, he lower is emission inensiy. Using (4) and (5), o subsiue in (9) gives he emission inensiy of non-exporers and exporers respecively: e x α = ακ B ρα ρα α a, (20) e ρα α = ακ (B + φb ) ρα α a (2) x A se of resuls on he relaionship beween emissions, firm characerisics, axes and rade emerge direcly from equaions (20) and (2): Proposiion 3 More producive firms have a lower emission inensiy. Proof: The saemen follows direcly from equaions (20) and (2). Proposiion 4 For any given level of produciviy, an exporer would have a lower emission inensiy han a non-exporer. Proof: The saemen follows from (20), (2), and he fac ha (B +φb ) ρα <B ρα. 0 The paper is in his sense relaed o he lieraure on rade induced echnological upgrading. See e.g. Bas (2008) and Busos (20). 3

15 Noe ha he laer proposiion is based on a hough experimen, since according o he model, depending on produciviy level a firm is eiher an exporer or anonexporer. There is no such produciviy level a which some firms are exporers and some are non-exporers. 3.7 Trade Liberalizaion, Abaemen and CO2 Emissions Evenually we wan o invesigae he relaionship beween rade liberalizaion, abaemen and emissions. In order o analyze he effecs of rade liberalizaion we need o solve he model. This requires ha we make addiional assumpions wih respec o marke size. We proceed by assuming ha he wo economies are idenical regarding ax regime and marke size. Hence, we solve he model for = and B = B. ue o symmery i suffices o solve for equilibrium in he home counry. Equaions (4), (5), (6), (7), and (8) deermine he endogenous variables f A, f A, a, a, and B, where we use upper bar o denoe equilibrium values derived based on he symmery assumpion. This gives us he following wo expressions for he cu-off produciviies: a k = a k = k k f E f (φ +) k f + f E k f (φ +) k, (22) k f + k f f k, (23) wih 0 < αρ(σ ) <, and ( α)(σ ) > 0. Noe ha he equilibrium expressions reduce o he sandard Meliz (2003) cu-off condiions for α =0, in which case producion does no enail any emissions. Exporers are more producive han nonexporers, i.e. a <a,as long as >, and we assume his o hold. 2 f We f (+φ) also assume ha k >, which guaranies ha he cu off produciviies are posiive. 3 From (22) and (23) follow ha rade liberalizaion will make he domesic cu-off ougher, i.e. a decreases, and he expor cu-off easier, i.e. a increases, which is line wih he resuls in he sandard Meliz model. Trade liberalizaion and abaemen invesmens Using (6) and subsiuing for he cu off produciviy employing (22) we can calculae B. Subsiuing his ino (4) and (5) we derive he abaemen invesmens for nonexporers (f A) and exporers ( f A ) for he symmeric equilibrium case: See he Appendix Secion A.7 for deails on calculaion. 2 The corresponding condiion in he sandard Meliz model is f f φ >. 3 k The condiion may be wrien: σ > α + αkρ, which reduces o he sandard condiion ha k σ > for α =0. 4

16 f a A = f, (24) a f a A = ( + φ) f, (25) a We can now formulae he following proposiion on he effec of rade liberalizaion on abaemen invesmens: Proposiion 5 Trade liberalizaion (higher φ) will decrease non-exporing firms abaemen invesmens. Trade liberalizaion will always increase exporers abaemen invesmens for sufficienly high rade coss. Proof : See Secion A.2 in he Appendix. Trade liberalizaion increases compeiion, and leads o lower sales for he non-exporers. This implies ha he leas producive firms close down and he remaining firms lower heir abaemen invesmens. Exporers also face increased compeiion in he domesic marke, bu on he oher hand hey also experience higher sales in he foreign marke as rade is liberalized. For high level of rade coss, he laer effecs dominae and leads o increased invesmens in abaemen. However, as rade coss reach a low level, he former effec ges sronger and as a consequence, invesmens in abaemen may be reduced. Trade liberalizaion and emission inensiy Nex, we urn o he effec of rade liberalizaion. Again we use (6) and (22) o calculae B, and subsiue his ino (20) and (2) in order o derive emission inensiies for domesic firms and exporers for he symmeric case: e x = ακα f ρα ( ) ρα a ( α) a ( α), (26) e x = ακα f ρα ρα ( + φ) ( ) a ( α) Making use of (22), gives us he following proposiions: ρα a ( α). (27) Proposiion 6 Trade liberalizaion (a higher φ) leads o a higher emission inensiy among non-exporers. Trade liberalizaion leads o a lower emission inensiy among exporers if k > (φ +). Proof: See Secion A.3 in he Appendix. We observe ha he higher he iniial level of rade coss prior o liberalizaion, he more likely is i ha rade liberalizaion will have a benign impac on emissions. Trade liberalizaion and oal emissions 5

17 Trade liberalizaion affecs emissions by weeding ou some of he leas producive firms wih low abaemen invesmens and accordingly high emission inensiies. For relaively high levels of iniial rade coss rade liberalizaion moreover induces exporers o inves more in abaemen, which in urn lowers heir emission inensiy. However, rade liberalizaion also implies lower abaemen invesmens by non-exporers and larger producion volumes as such for he exporers, boh of which conribue o higher oal emissions. The overall effec of rade liberalizaion depends on he ne effec of his se of effecs. We proceed by analyzing oal emissions by he non-exporers and he exporers separaely. Toal emissions are finally given by he sum of hese. Toal emissions by non-exporers and exporers are given by he inegrals and E = n a a edg(a a ), (28) a E = n edg(a a ). (29) 0 Solving hese inegrals condiional on firm enry gives he expressions for emissions of non-exporers and exporers respecively. The derivaion of hese expressions is found in he Appendix Secion A.7. Toal emissions of non-exporers are given by E = σ α (σ ) +φ ( + φ) (φ +) (φ +) The expression leads o he following proposiion: k f f k μl. (30) f f Proposiion 7 Trade liberalizaion decreases oal emissions of non-exporing firms. Proof: The proposiion follows direcly from (30), given our assumpion ha k >. The weeding ou of he leas producive and diries firms ogeher wih lower producion volumes for hose remaining lead o falling emissions by non-exporers. Trade liberalizaion also leads o a lower mass of firms, which also conribues lower emissions. These benign effecs on emissions overshadow he fac ha all non-exporers decrease heir abaemen emissions (see Proposiion 5). Toal emissions by exporers are given by E = α (σ ) k f σ ( + φ) ( + φ) k f μl. (3) + φ (+φ) 6

18 Even if he condiion in Proposiion 6 holds, so ha he emission inensiy of exporers decrease, his group of firmsalwaysincreasesisemissionsbecauseofincreasedoal producion volume: Proposiion 8 Trade liberalizaion increases oal emissions from exporers. Proof: See secion A.4 in he Appendix. The quesion now is wha he overall effec of on emissions is. Adding emissions by exporers and non-exporers give E = α (σ ) σ + k f ( + φ) ( + φ) k f k f φ ( + φ) ( + φ) f k μl, (32) which leads o he following proposiion: Proposiion 9 Trade liberalizaion, i.e. higher φ, decreases oal emissions. Proof: The proposiion follows direcly from (32). Hence, we find ha in he case wih symmeric counries, he overall effec of rade liberalizaion is o decrease emissions. Trade increases producion volumes bu he combined effec of he weeding ou of low producive and diry firms, he higher abaemen invesmens by exporers, and he shif of producion from diry non-exporers o relaively clean exporers, leads o lower overall emissions. Noe ha emissions implied by ransporaion are accouned for in he analysis due o how hey are modelled. Iceberg ransporaion coss imply ha ransporaion coss are in incurred erms of he good ranspored, and emissions relaed o he producion of he quaniy ha is absorbed by ransporaion are hus accouned for in equaion (32). 4 Empirical esign and Resuls Our heoreical model suggess ha exporers have relaively lower emission inensiy han non-exporers. We have proposed a mechanism hrough which firms expor saus affecs heir abaemen invesmen and ulimaely heir emission inensiy, which may explain he relaionship beween emission inensiy and expor saus. We proceed by aking he model s predicions of he relaionship on firms produciviy, exporing and emission inensiy, as well as on produciviy, exporing and abaemen invesmens o he daa. The heoreical model suggess log linear specificaions for boh relaionships (see equaions (4), (5), (20) and (2)). Hence, we proceed by regressing he firm s emission inensiy and abaemen respecively on produciviy and exporing saus, 7

19 ln Emission inensiy i = α 0 + f(log produciviy i )+α 2 Exporer i + ε i (33) ln Abaemen invesmen i = α 0 + f(log produciviy i )+α 2 Exporer i + ε i (34) where f is a polynomial funcion of firm i s produciviy in year and Exporer i equals one if he firm expors in year, and zero oherwise. 4 According o our heory produciviy is he forcing variable. I drives he expor saus of a firm as well as he firm level emission inensiy (CO2 emissions/oupu). However, while produciviy has a coninuos effec on he emission inensiy, he model predics a disconinuous jump down in he emission inensiy when we compare an exporer o a non-exporer. We exploi his by including firm produciviy in a very flexible manner using a coninuos polynomial up o he fourh order as refleced by f(). 4. CO2 Emission Inensiy and Expors We sar by esimaing equaion (33). Emission inensiy is measured as firm-level CO2 emissions per oupu. Firms produciviy is measured by oal facor produciviy, and is calculaed from esimaes of produciviy funcions using he mehod by Levinsohn andperin(2003). 5 To accoun for secorial variaions in emissions we include indusry dummies based on 5-digi indusries, while year dummies pick up rends as well as he sligh changes over ime in he emission ax facing Swedish firms. We repor regression resuls where errors are clusered a he firm level, while noing ha clusering a he secor level gives very similar resuls. In Table 4 we repor he OLS resuls for esimaions based on he enire sample. We repor resuls for five differen specificaions wih respec o he modelling of he produciviy variable. In line wih wha our heory would predic, we find ha he coefficiens for he exporer dummy are negaive and significan a he one percen level in all specificaions. Exporers emi on average around 2 percen less per uni of oupu han non-exporers acive in he same indusry. 6 There are obviously huge variaions in emission inensiy across indusries. Hence, including indusry dummies increases he R- square subsanially. We have also explored he impac of using indusry dummies based on a more aggregae indusry classificaion (2 digi level). No surprisingly, he resuls are roughly he same as wih he finer classificaion, bu he fi of he model as suggesed by he R- square is reduced. As for year dummies, we have also run he regressions wihou hem, bu he exclusion of hese dummies does no affec he resuls in any significan way. 4 Expor saus is defined by exporing income > 0. 5 Producion funcions are esimaed a he wo-digi secor level, where we use value added as measure of firm oupu. Explanaory variables are labour (measured by he wage bill) and capial. Finally we use raw maerials as proxy for conemporaneous produciviy shocks. All variables are in logs. 6 Which is found using ha 00 (exp( 0.3) ) = 2. 8

20 Table 4: CO2 emission inensiy, produciviy and exporing (OLS), I ependen variable: ln CO2 emission inensiy () (2) (3) (4) (5) Exporer a -.28 a -.3 a -.33 a -.33 a (.036) (.037) (.037) (.037) (.037) ln TFP none linear 2nd order 3rd order 4h order polynom. polynom. polynom. Indusry dummies Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes R-squared No. of obs Noe: OLS esimaes are based on he panel Errors are clusered a he firm level. CO2 emission inensiy gives he raio of emissions o oupu. Indusry dummies are based on 5 digi indusries. a significan a % level, b significan a 5% level, c significan a 0% level. Since emission ax raes, in general, do no vary beween firms we conrol for he sligh changes in he emission ax over ime by including ime fixed effecs. However, he mos energy inensive indusries enjoy a ax credi for he par of heir CO2 ax bill which exceeds 0.8 percen of heir producion value. The same group of firms has since 2005 gradually become included in he EU Emission Trading Sysem. Boh hese feaures may have an impac on firms behavior and hus on heir CO2 emissions. Hence, o allow for variaion beween he energy and non-energy inensive indusries, we proceed by spliing he sample ino energy inensive indusries and non energy inensive indusries. The resuls are repored for boh groups of indusries in Table 5. The general message from above is confirmed: exporing firms has a lower emission inensiy. Comparing he resuls for he wo groups of firms, he coefficien for expor saus for he energy inensive firms is much larger bu no as srongly significan as in he regressions for low energy inensive secors. The energy inensive group conains many of he large exporers in he heavy processing indusry (paper, pulp, seel ec.) which are also he mos subsanial emiers of CO2. 7 As for he resuls for he non-energy inensive indusries hese are very similar o hose for he complee sample. 7 The energyinensive secors are Paper and pulp (7), Coke and refined perolium prod. (9), Chemicals (20), Non-mealic mineral producs (23), and Basic meals (24): 9

21 Table 5: CO2 emission inensiy, produciviy and exporing (OLS), II ependen variable: ln CO2 emission inensiy Energy inensive indusries Non-energy inensive indusries () (2) (3) (4) (5) (6) (7) (8) Exporer b -.37 b c c a -.00 a -.0 a a (.46) (.44) (.45) (.46) (.037) (.037) (.037) (.037) ln TFP linear 2nd order 3rd order 4h order linear 2nd order 3rd order 4h order polynom. polynom. polynom. polynom. polynom. polynom. Indusry dummies Yes Yes Yes Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes Yes Yes Yes R-squared No. obs Noe: Esimaes are based on he panel Errors are clusered a he firm level. CO2 emission inensiy gives he raio of emissions o oupu. Indusry dummies are based on 5 digi indusries. a significan a % level, b significan a 5% level, c significan a 0% level. One may, however, argue ha calculaing emission inensiy based on oupu may gives inaccurae measures of he firms emission inensiy. Our choice of value of producion oupu as denominaor is guided by our heory, bu if firms ousource subsanial shares of heir producion, using value added would provide more correc measures of he firms emission inensiy. Reviewing he daa, his is neverheless no an obvious choice since anumberoffirms run deficis and appear wih negaive value added. We also observe ha he value added flucuaes much more over he years han oupu does. Hence, our baseline regression are all based on emission inensiy calculaed using oupu. Sill, in secion A.5 in he Appendix we repor resuls boh for he enire sample as well as for he energy and non-energy inensive indusries using value added raher han value of oupu o calculae emission inensiy. The resuls are in line wih hose based on oupu. Exporing firms do in general have a lower emission inensiy. This is also rue for he non-energy inensive firms, while he resuls for he energy inensive firms are weaker. 4.2 Abaemen and Expors Nex, we urn o he relaionship beween exporing, produciviy and firm level abaemen being he proposed heoreical mechanism for why expor saus affecs firms CO2 emission inensiy. The model again suggess a log-linear specificaion and we esimae amodelbasedon(34). 8 Abaemen daa are based on a more limied survey han he emission inensiy daa. The survey is described in secion 2. and is biased owards larger firms. This leaves us wih a more limied sample han wha was he case when we analyzed emission. Hence, idenificaion of he impac of exporing as such is more 8 Yearly firm level abaemen invesmens vary beween 0 and more han 400 mio SEK. In order no o exclude he firms wih zero abaemen invesmens from he sample as we use logs, we le he dependen variable be ln(abaemen invesmens +). 20

22 challenging. In Table 6 we repor he resuls from esimaing equaion (34) using OLS. Again, we repor resuls for five differen specificaions wih respec o he modelling of he produciviy variable. We include ime and indusry (5 digi level) dummies in all regressions. Our resuls sugges ha exporers make significanly higher invesmen in abaemen han non-exporers. Conrolling for indusry, an exporer invess on average 65 percen more in abaemen han a non-exporer. 9 Table 6: Abaemen, produciviy and exporing (OLS), I ependen variable: ln Abaemen Invesmens () (2) (3) (4) (5) Exporer.070 a.505 b.528 a.482 b.494 b (.25) (.203) (.203) (.202) (.203) ln TFP none linear 2nd order 3rd order 4h order polynom. polynom. polynom. Indusry dummies Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes R-squared No. obs Noe: OLS esimaes are based on he panel Errors are clusered a he firm level. Indusry dummies are based on 5 digi indusries. a significan a % level, b significan a 5% level, c significan a 0% level. As noed above, we also need o ake ino consideraion he fac ha firms in he energy inensive indusries faces a hreshold for heir ax bill and have since 2005 gradually become included in he EU Emission Trading Sysem. Boh hese feaures may have an impac on firms behavior and hus on heir CO2 emissions. Hence, we proceed as we did when analyzing emission inensiies by spliing he sample ino energy and non-energy inensive indusries. Resuls are repored in Table 7. When i comes o he energy inensive indusries, we noe ha 98 percen of hese firms are exporers. This enhances he idenificaion problem furher, and no surprisingly exporer saus does no have any significan impac on abaemen invesmens. Bu as we urn o he non-energy inensive indusries, we find ha, in line wih wha our heory would predic, exporing firms make higher invesmens in abaemen. 9 Which is found using ha 00 (exp(0.5) ) = 65. 2

23 Table 7: Abaemen, produciviy and exporing (OLS), II ependen variable: ln Abaemen Invesmens Energy inensive indusries Non-energy inensive indusries () (2) (3) (4) (5) (6) (7) (8) Exporer a.653 a.623 a.606 a (.724) (.724) (.747) (.746) (.200) (.203) (.20) (.200) ln TFP linear 2nd order 3rd order 4h order linear 2nd order 3rd order 4h order polynom. polynom. polynom. polynom. polynom. polynom. Indusry dummies Yes Yes Yes Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes Yes Yes Yes R-squared No. obs Noe: Esimaes are based on he panel Errors are clusered a he firm level. Indusry dummies are based on 5 digi indusries. a significan a % level, b significan a 5% level, c significan a 0% level. 4.3 Robusness One concern is ha we esimae he produciviy polynomial for a range of he produciviy disribuion ha lacks common suppor. We herefore reesimae (33) dropping all observaions ouside he region of common suppor for exporers and non-exporers wihin each 5-digi secor. The resuls wih CO2-emission inensiy as dependen variable are shown in Table 8. Reassuringly resuls are almos idenical o he esimaes wih he full sample in spie of dropping abou 3 percen of he sample. Table8:CO2emissioninensiy;OLSoncommonsuppor ependen variable: ln CO2 emission inensiy () (2) (3) (4) (5) Exporer a -.33 a -.3 a -.22 a -.20 a (.036) (.036) (.036) (.036) (.036) ln TFP none linear 2nd order 3rd order 4h order polynom. polynom. polynom. Indusry dummies Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes R-squared No. obs Noe: OLS esimaes are based on he panel Errors are clusered a he firm level. Observaions ouside he common suppor in each 5-digi secor are dropped. CO2 emission inensiy gives he raio of emissions o oupu. Indusry dummies are based on 5 digi indusries. a significan a % level, b significan a 5% level, c significan a 0% level. We also reesimae equaion (34) wih abaemen as dependen variable on a sample wih common suppor for exporers and non-exporers. Esimaing wih common suppor 22

24 implies in his case dropping 36 percen of he sample. As shown in Table 9, he resuls are neverheless very similar o he regressions resuls based on he full sample. Table 9: Abaemen; OLS on common suppor ependen variable: ln Abaemen invesmens () (2) (3) (4) (5) Exporer a.479 b.492 b.46 b.465 b (.200) (.95) (.95) (.96) (.95) ln TFP none linear 2nd order 3rd order 4h order polynom. polynom. polynom. Indusry dummies Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes R-squared No. obs Noe: OLS esimaes are based on he panel Errors are clusered a he firm level. Observaions ouside he common suppor in each 5-digi secor are dropped. Indusry dummies are based on 5-digi indusries. a significan a % level, b significan a 5% level, c significan a 0% level. 5 Conclusion This paper analyses CO2 emissions and exporers wihin a framework wih heerogeneous firms and rade. We develop a heoreical model ha proposes a mechanism for why exporers may be expeced o have lower emission inensiies han non-exporers: In line wih he sandard heory on heerogeneous firms and rade, he mos producive firms become exporers. Exporers larger producion scale suppors higher fixed invesmens in abaemen which in urn reduces boh heir emission ax bills and heir emission inensiy. Hence, according o our model we would expec emission inensiy o be negaively relaed o firm-level produciviy and expor saus. Solving he model for symmeric counries we find ha rade liberalizaion allows for a higher producion volume and make new exporers cleaner as hey are induced o inves more in abaemen. Bu rade liberalizaion also makes non-exporers dirier as hese firms are forced o downsize and reduce heir invesmens in abaemen. We show ha he overall effec of rade liberalizaion is o reduce emissions. This is a resul of he less producive and diries firms being weeded ou, and of producion being shifed from relaively diry non-exporers o more efficien and cleaner exporing firms. A unique daa se for Swedish manufacuring firms ha includes informaion on firmlevel abaemen invesmens and firm-level emissions of CO2 allows us o es he empirical predicions of he model on he relaionship beween produciviy, expor saus and emission inensiy as well as on produciviy, expor saus and abaemen invesmens. Produciviy is he forcing variable in our model. I drives he firm level emission inensiy 23