CEP Discussion Paper No 908 January International Trade Integration: A Disaggregated Approach Natalie Chen and Dennis Novy

Transcription

1 CEP Discussion Pper No 908 Jnury 2009 Interntionl Trde Integrtion: A Disggregted Approch Ntlie Chen nd Dennis Novy

2 Astrct This pper investigtes the sources nd size of trde rriers t the industry level. We derive micro-founded mesure of industry-specific ilterl trde integrtion tht hs n in-uilt control for time-vrying multilterl resistnce. This trde integrtion mesure is consistent with rod rnge of recent trde models including the Anderson nd vn Wincoop (2003) frmewor, the Ricrdin model y Eton nd Kortum (2002) nd heterogeneous firms models. We use it to explore trde rriers for mnufcturing industries in Europen Union countries etween 1999 nd We find lrge degree of trde cost heterogeneity cross industries. The most importnt trde rriers re trnsporttion costs nd policy fctors such s Technicl Brriers to Trde. Trde integrtion is generlly lower for countries tht opted out of the Euro or did not olish order controls in ccordnce with the Schengen Agreement. Reductions in trde rriers explin out one-hlf of the growth in trde over the period nd re therefore mjor driving force of the EU Single Mret. JEL Clssifictions: F10, F15 Keywords: Trde Integrtion, Grvity, Trde Costs, Multilterl Resistnce, Industries, Disggregtion, Europen Union This pper ws produced s prt of the Centre s Glolistion Progrmme. The Centre for Economic Performnce is finnced y the Economic nd Socil Reserch Council. Acnowledgements We re grteful for comments y Richrd Bldwin, Pierre-Philippe Comes, Donld Dvis, Ronld Findly, Jmes Hrrign, Thierry Myer, Steve Redding, Dvid Weinstein nd prticipnts t the CEPR ERWIT 2007, Midwest Spring Meetings 2008, the Product Mret Integrtion in the EU 2008 worshop t the Universidde do Minho, Columi, Fordhm, Georgi Tech, HEI Genev, Nottinghm, Oxford, Snt Clr, UC Dvis, UC Snt Cruz, Utrecht School of Economics, Vnderilt nd Yle. Ntlie Chen is Visitor with the Glolistion Progrmme t the Centre for Economic Performnce, London School of Economics nd n Associte Professor of Economics t the University of Wrwic. Dennis Novy is n Assistnt Professor of Economics t the University of Wrwic. Pulished y Centre for Economic Performnce London School of Economics nd Politicl Science Houghton Street London WC2A 2AE All rights reserved. No prt of this puliction my e reproduced, stored in retrievl system or trnsmitted in ny form or y ny mens without the prior permission in writing of the pulisher nor e issued to the pulic or circulted in ny form other thn tht in which it is pulished. Requests for permission to reproduce ny rticle or prt of the Woring Pper should e sent to the editor t the ove ddress. N. Chen nd D. Novy, sumitted 2009 ISBN

3 1 Introduction Trde costs re stple ingredient in tody s trde literture. Brodly defined, trde costs include ny cost of engging in interntionl trde such s trnsporttion costs, triffs, non-triff rriers, informtionl costs, time costs, different product stndrds, exchnge rte costs nd locl distriution costs, mong others. They feture prominently in the vst mjority of theoreticl ppers, including the models of Eton nd Kortum (2002) nd Melitz (2003). In ddition, growing empiricl literture, surveyed y Anderson nd vn Wincoop (2004), is devoted to exploring the sources nd size of trde costs. A deeper understnding of the cuses of trde costs is of prticulr importnce ecuse it would enle etter evlution of their welfre implictions. These re suspected to e lrge: on their own, policy-relted trde costs my e worth more thn ten percent of ntionl income (Anderson nd vn Wincoop, 2002). A mjor chllenge fced y empiricl reserchers is to mesure overll trde costs since direct mesures re remrly sprse nd inccurte (Anderson nd vn Wincoop, 2004, p.692). Direct mesures re only ville for few components, for instnce trnsporttion nd insurnce costs, usully proxied y the rtio of c.i.f. nd f.o.. trde vlues (Hrrign, 1993, Hummels, 2001, 2007), 1 policy rriers such s specific triff or non-triff rriers (Chen, 2004, Hrrign, 1993, Hed nd Myer, 2000), informtionl costs (Ruch, 1999) or time costs (Evns nd Hrrign, 2005, Hrrign, 2005, Hummels, 2001). But even for those components, dt coverge is often limited to few countries nd yers, nd it cn e hrd to gther disggregted trde cost dt t the industry or product level. Given those difficulties in otining ccurte mesures of trde costs, some reserchers indirectly infer the level of trde impediments from trde flows. This pproch hs the ovious dvntge of extending the nlysis to more countries, yers nd more finely disggregted dt. One wy of doing this is to compute trde to output rtios (Hrrign, 1996); nother is to estimte order effects, which mostly reflect the extent of order-relted costs. 2 This pper is prt of the reserch effort tht ttempts to indirectly infer trde impediments from trde flows. The first contriution of the pper is to develop micro-founded mesure of ilterl trde integrtion tht cn e pplied to disggregted pnel dt. We derive this mesure y modeling disggregted trde flows t the industry level in the grvity frmewor pioneered y Anderson nd vn Wincoop (2003, 2004), llowing trde costs nd sustitution elsticities to e heterogeneous cross industries. Anderson nd vn Wincoop (2003) show tht trde flows re determined not only y ilterl trde costs etween two countries ut lso y verge trde rriers with other countries, which they refer to s multilterl resistnce. Following the pproch of Novy (2008), we derive n nlyticl solution for multilterl resistnce vriles tht vry cross industries nd over time. In turn, this enles us to derive micro-founded mesure of ilterl trde integrtion tht hs n in-uilt control for time-vrying multilterl resistnce nd cn therefore e pplied to pnel dt. 1 Moreover, Limão nd Venles (2001) use the quotes from shipping firms for stndrd continer shipped from Bltimore to severl destintions. Comes nd Lfourcde (2005) develop new methodology to compute trnsporttion costs nd pply it to rod trnsport y truc in Frnce. 2 Exmples include Anderson nd vn Wincoop (2003), Bldwin, Forslid, Mrtin, Ottvino nd Roert-Nicoud (2003), Chen (2004), Eton nd Kortum (2002), Evns (2003), Hed nd Myer (2000), Hed nd Ries (2001), McCllum (1995), Nitsch (2000) nd Wei (1996). 1

4 Arguly, the Anderson nd vn Wincoop model is one of the most prsimonious trde models of recent yers. It rests on the Armington ssumption tht countries produce differentited goods nd trde is driven y consumers love of vriety, leding to the ey grvity eqution. However, we show tht isomorphic trde integrtion mesures cn e derived from the Ricrdin trde model y Eton nd Kortum (2002), from Chney s (2008) extension of the Melitz (2003) heterogeneous firms model s well s from the heterogeneous firms model y Melitz nd Ottvino (2008) with liner non-ces demnd. This is possile ecuse ll these models led to grvity equtions tht hve similr structure. Our trde integrtion mesure is therefore consistent with rod rnge of the recent theoreticl trde literture nd is not limited to ny one prticulr frmewor. Although we regrd our pproch s complementry to other reserch tht indirectly infers trde impediments from trde flows, we discuss how our trde integrtion mesure differs from d hoc proxies commonly used in the literture such s trde to output rtios. In prticulr, we show tht d hoc proxies without micro-foundtions cn erroneously pic up chnges in multilterl trde rriers s chnges in ilterl trde rriers. We re insted le to focus on ilterl trde integrtion ecuse our micro-founded mesure is not distorted y multilterl resistnce effects. The second contriution of the pper is to ring our mesure of trde integrtion to the dt. This enles us to document nd explin the vrition of trde rriers cross 166 mnufcturing industries in 11 Europen Union countries over the period The cse of the Europen Union is ppeling since trde integrtion is expected to e strong etween its memer sttes due to two fctors. First, these countries hve succeeded in dismntling mny restrictions on trde, including triffs nd quots tht were completely eliminted y Second, the sitution hs een further reinforced y the implementtion of the Single Mret Progrmme (SMP), lunched in the mid-1980s. Consistent with the stndrd grvity literture, the vrition of trde integrtion cross country pirs cn to lrge extent e cptured y typicl grvity vriles such s distnce nd djcency ut lso y policy-relted vriles such s memership in the Eurozone or prticiption in the Schengen Agreement. But our focus lies on the sustntil degree of heterogeneity in trde integrtion cross industries. Our results suggest tht modeling trde costs s one-fits-ll impediment is clerly t odds with empiricl evidence. We investigte the role of severl sectorl chrcteristics in explining trde integrtion cross industries, with prticulr emphsis on policy-relted vriles such s the extent of Technicl Brriers to Trde (TBTs). Such rriers re predominnt concern in tody s glol trde negotitions, nd for the WTO in prticulr s it precisely sees to ensure tht technicl regultions nd stndrds, including pcging, mring nd lelling requirements [...] do not crete unnecessry ostcles to interntionl trde. 3 We find tht trde integrtion is indeed lower in countries nd industries where TBTs re strong, suggesting tht there is room left for policy ction nd tht further gins re possile through the reduction of those rriers. We lso show tht trde integrtion tends to e high for industries chrcterized y high productivity, low trnsporttion costs nd high degree of trnsprency in pulic procurement. From dynmic perspective, verge trde integrtion hs 3 Agreement on Technicl Brriers to Trde (p.117). This Agreement, negotited during the Uruguy Round, is n integrl prt of the WTO Agreement. 2

5 improved for most countries over the period , s well s individully for lrge numer of industries in our smple. We lso contrst our empiricl pproch of explining the vrition in ilterl trde integrtion to the stndrd grvity pproch of regressing ilterl trde flows. We elieve this is n importnt exercise tht yields two insights. First, our pproch llows us to identify the effects of numer of crucil explntory vriles such s weight-to-vlue tht would normlly drop out in correctly specified grvity eqution due to perfect collinerity with fixed effects. Second, we show tht misspecifying stndrd grvity eqution y omitting multilterl resistnce vriles leds to sustntil is in the estimted trde cost elsticities. Our pproch voids this prolem ecuse our trde integrtion mesure hs n in-uilt control for time-vrying multilterl resistnce. The third contriution of the pper is to explore to wht extent the increse in trde flows over recent yers cn e explined y the decrese in trde rriers. For tht purpose, we use our model to decompose the growth in trde into two min components (1) the growth of mnufcturing output nd (2) improvements in trde integrtion. We find tht on verge 58 percent of the growth in trde cn e ccounted for y chnges in ilterl trde rriers nd multilterl resistnce, while the rest cn e ttriuted to the seculr growth of mnufcturing output. This contriution of improved trde integrtion is quntittively lrger thn the contriution found y Bier nd Bergstrnd (2001) for smple of OECD countries. As we focus on the Europen Union, our results suggest tht on verge, intr-eu trde integrtion hs progressed fster thn integrtion mongst other developed countries. The pper is orgnized s follows. In Section 2 we develop generl equilirium model with industry-specific trde costs. In tht section we lso derive n nlyticl solution for time-vrying industry-level multilterl resistnce vriles nd our trde integrtion mesure. Section 3 presents our dtset. Section 4 reports our empiricl results, focusing on the determinnts nd time trend of trde integrtion. We lso present the decomposition of trde growth into output growth nd improvements in trde integrtion. Section 5 provides roustness checs nd Section 6 concludes. 2 A Model with Industry-Specific Trde Costs Our model closely follows the seminl pper y Anderson nd vn Wincoop (2003). Their generl equilirium model of trde results in grvity eqution tht incorportes trde costs. The ey insight from their grvity eqution is tht ilterl trde flows re not determined solely y the solute ilterl trde rrier etween two countries ut rther y their ilterl reltive to their verge trde rrier. Anderson nd vn Wincoop (2003) refer to the pproprite verge trde rrier s multilterl resistnce. As generliztion, Anderson nd vn Wincoop (2004) model ilterl trde for n individul industry tht is chrcterized y industry-specific ilterl trde costs nd n industry-specific elsticity of sustitution. We follow Anderson nd vn Wincoop (2004) in modeling trde flows with heterogeneous trde costs nd heterogeneous elsticities of sustitution t the industry level. The innovtion of our pproch is to derive n nlyticl solution for time-vrying industry-specific multilterl resistnce vriles tht cn e relted to oservle dt. With this solution t hnd, we re 3

6 le to derive micro-founded mesure of industry-specific ilterl trde integrtion tht nets out multilterl resistnce effects. Bergstrnd (1989, 1990) lso derives grvity equtions for industry-level trde flows ut does not focus on multilterl resistnce. 2.1 The Bsic Frmewor Denote x ij s nominl exports from country i to country j in goods ssocited with industry. Suppose tht consumers in country j llocte expenditure x j on industry- goods nd tht their preferences over these goods cn e descried ystndrdcesutilityfunctions C j à JX i=1 ³ c ij! σ σ 1 σ 1 σ (1) where c ij is rel consumption of industry- goods from country i y country-j consumers nd where the elsticity of sustitution σ is specific to industry nd ssumed to exceed unity, σ > 1. Furthermore suppose tht the fctory gte price of industry- goods from country i is denoted y p i nd tht trde costs ssocited with the trde cost fctor t ij 1 re incurred when these goods re shipped to country j such tht the price fced y country-j consumers, denoted y p ij, cn e written s p ij = t ij p i. The demnd function for exports x ij = p ij c ij then follows s x ij = à p ij P j! 1 σ x j = à t ij p i P j! 1 σ x j (2) where the price index P j cn e derived s P j = à JX i=1! ³ 1 1 σ p 1 σ ij (3) 2.2 The Grvity Eqution Denote output of industry- goods y country-i firms s y i nd impose mret-clering s JX yi = x ij (4) Sustituting the demnd function (2) into the mret-clering condition (4) nd rerrnging yields j=1 Ã! JX t 1 σ p i = ij x j j=1 P j y i 1 σ 1 (5) 4

7 Plug eqution (5) c into the demnd function (2) nd, s in Anderson nd vn Wincoop (2004), define outwrd multilterl resistnce for industry- goods from country i s Ã! JX t 1 σ Π i ij x j j=1 P j y 1 1 σ (6) where y is world output in industry. Rerrnge to rrive t grvity eqution for industry x ij = y i x j y à t ij Π i P j! 1 σ (7) Trde flows x ij depend on supply y i of the -good from country i nd expenditure x j for the good in country j. Lrge ilterl trde costs t ij reduce ilterl trde, wheres lrge verge outwrd trde rriers of country i (i.e., lrge Π i ) nd lrge verge inwrd trde rriers of country j (i.e., lrge Pj ) led to more ilterl trde. Sustituting the solution for p i in(5)ndthedefinition of Π i in (6) into the price index (3) yields Ã! JX t 1 σ Pj = ij yi which is inwrd multilterl resistnce for industry- goods entering country j. i=1 2.3 Solving for Multilterl Resistnce Π i y 1 1 σ (8) A prolem tht rises in empiricl wor is tht we do not hve dt for the multilterl resistnce terms P j nd Π i in grvity eqution (7). The method we employ here is to solve for these terms nlyticlly s function of oservle trde flows. We exploit the fct tht multilterl resistnce is relted to the mount of trde country conducts with itself (see Novy, 2008). Intuitively, if country s trde rriers with the rest of the world re high (i.e., if the country s multilterl resistnce is high), the country will trde lot domesticlly. To see this formlly, use grvity eqution (7) nd consider domestic trde flows for industry- goods µ x ii = y i x i t 1 σ ii y Π i P i (9) where t ii re domestic trde costs for industry- goods, for exmple domestic trnsporttion costs. Eqution (9) cn e solved for the product of outwrd nd inwrd multilterl resistnce s µ y Π i Pi = i x 1 1 σ i t x ii (10) ii y Note tht we do not impose zero domestic trde costs since the trde cost fctor t ii my exceed unity. Eqution (10) implies tht for given t ii nd σ, it is esy to mesure multilterl resistnce since the quntities on the right-hnd side re oservle. 5

8 2.4 A Micro-Founded Mesure of Industry-Specific Trde Integrtion The solution for multilterl resistnce cn e exploited to solve the model. Grvity eqution (7) contins the product of outwrd multilterl resistnce of country i nd inwrd multilterl resistnce of country j, Π i P j, wheres eqution (10) provides solution for Π i P i. It is therefore useful to multiply grvity eqution (7) y the corresponding grvity eqution for trde flows in the opposite direction, x ji, to otin idirectionl grvity eqution tht contins oth countries outwrd nd inwrd multilterl resistnce vriles. This yields x ijx ji = y i y j x i x j y y Ã! t ij t 1 σ ji Π i P i Π j P j (11) Sustitute the solution for multilterl resistnce given in eqution (10) to otin! 1 σ (12) x ijx ji = x iix jj à t ij t ji t ii t jj From eqution (12) it is esy to solve for the trde cost fctors s t ij t ji t ii t jj = à x ii x jj x ij x ji! 1 σ 1 (13) It is only possile to infer reltive trde costs, in this cse ilterl trde costs t ij t ji reltive to intrntionl trde costs t ii t jj.4 We do not impose trde cost symmetry so tht t ij nd t ji on the left-hnd side of eqution (13) my e symmetric (t ij 6= t ji ). As Anderson nd vn Wincoop (2003, footnote 11) point out, it is prolemtic to infer the degree of trde rrier symmetry from trde dt ecuse there re multiple comintions of t ij nd t ji tht cn give rise to the sme trde flows x ij nd x ji. We therefore te the squre root to get n expression for the verge ilterl trde rrier. Thus, the verge reltive trde rrier cn e expressed s θ ij à t ij t ji t ii t jj! 1 2 = à x ii x jj x ij x ji! 1 2(σ 1) (14) We interpret θ ij s micro-founded mesure of ilterl industry-specific trde frictions, or the inverse of ilterl trde integrtion. The more two countries trde with ech other (i.e., the higher x ij x ji ), the lower is our mesure of reltive trde frictions ceteris prius. Conversely, if the two countries strt trding more domesticlly (i.e., the higher x ii x jj ), the higher is our mesure of reltive trde frictions ceteris prius. For the interprettion of θ ij it is lso helpful to thin of two opposite extreme cses frictionless world with no trde costs on the one hnd, nd closed economy on the other. In frictionless world, ll trde cost fctors equl unity (t ij = t ji = t ii = t jj =1)ndθ ij would e one, implying tht ilterl nd domestic rriers re the sme. In the cse pproching closed economy, ilterl 4 On this point lso see Anderson nd vn Wincoop (2004, p.709). 6

9 trde x ij x ji tends towrds zero nd thus θ ij tends towrds infinity, implying tht ilterl rriers re prohiitive reltive to domestic rriers. Lst ut not lest, we stress tht our trde integrtion mesure is vlid more generlly eyond the prticulr frmewor presented ove. We hve derived θ ij in eqution (14) from model sed on CES demnd system in comintion with the Armington ssumption tht goods re differentited y country of origin. But s we show in the technicl ppendix, we cn lso derive our trde integrtion mesure from rnge of other leding trde theories. These include the Ricrdin trde model y Eton nd Kortum (2002), Chney s (2008) extension of the Melitz (2003) heterogeneous firms model to symmetric countries s well s the heterogeneous firms model y Melitz nd Ottvino (2008) with liner non-ces demnd nd endogenous mrups. The reson why our trde integrtion mesure θ ij is consistent with rod rnge of trde models is relted to the fct tht they ll led to grvity equtions tht hve similr structure s eqution (7). 5 Intuitively, the grvity eqution is n expenditure eqution tht indictes how consumers llocte their expenditure cross countries suject to trde frictions (Bldwin nd Tglioni, 2006). Grvity equtions rise regrdless of why consumers wnt to uy goods from foreign countries. In n Armington world, consumers uy foreign goods ecuse those goods re inherently different nd consumers prefer vriety. In Ricrdin world, countries produce goods ccording to comprtive dvntge nd consumers uy foreign goods ecuse they re cheper. It turns out tht the prticulr motivtion ehind foreign trde is not crucil to understnd the role of trde frictions A Comprison to Alterntive Mesures We now contrst θ ij with lterntive mesures of trde integrtion. First, compre θ ij with ilterl ³ ³ trde to output rtio, x ij x ji / yi y j.vriousformsofsuchrtioshveeenusedsmesuresof trde openness (Hrrign, 1996). Dividing grvity eqution (11) y yi y j nd sustituting eqution (10) yields x ij x ji y i y j = x ii x jj y i y j Ã! t ij t 1 σ ji (15) t ii t jj Theory thus predicts tht prt from the trde cost prmeters, ³ the ilterl ³ trde to output rtio is lso determined y the two countries domestic trde shres, /. x ii x jj y i y j To compre the trde to output rtio with θ ij, te the trde cost prmeters t ij, t ji, t ii nd t jj s given. This pins down θ ij in eqution (14). However, the trde to output rtio cn still move. For 5 Feenstr, Mrusen nd Rose (2001) nd Evenett nd Keller (2002) lso show tht grvity equtions re consistent with vrious competing trde models. 6 Derdorff (1998) rgues tht in Hecscher-Ohlin frmewor, trde frictions prevent fctor price equliztion so tht for the lrge mjority of goods, only one country is the lowest-cost producer. In the presence of trde frictions, trde in Hecscher-Ohlin world therefore resemles trde in n Armington world. Grossmn (1998) provides the following intuition for the grvity eqution: Speciliztion lies ehind the explntory power [of the grvity eqution], nd of course some degree of speciliztion is t the hert of ny model of trde. Thus, the derivtion of the grvity eqution need not me reference to ny prticulr trde model t ll [...] This is true no mtter wht supply-side considertions give rise to the speciliztion, e they incresing returns to scle in world of differentited products, technology differences in world of Ricrdin trde, lrge fctor endowment differences in world of Hecscher-Ohlin trde, or (smll) trnsport costs in world of ny type of endowment-sed trde. (Emphsis in the originl.) 7

10 exmple, suppose tht trde costs with third country increse, sy, t im with m 6= i, j. The increse in t im pushes up the domestic trde rtio x ii /y i nd therefore lso the ilterl trde to output rtio in eqution (15). 7 But this is multilterl resistnce movement nd does not imply tht trde integrtion etween i nd j hs improved. In contrst, θ ij does not rect to multilterl resistnce movements ecuse θ ij only reflects chnges in ilterl nd domestic trde costs, not chnges in third-prty trde costs. In conclusion, trde to output rtios re distorted y multilterl resistnce effects. Second, compre θ ij with the phi-ness of trde, which is given y the simple rtio of ilterl to domestic trde flows nd is commonly used to estimte sectorl order effects (Bldwin et l., 2003, Hed nd Myer, 2004, Hed nd Ries, 2001). 8 Theory shows tht the order effect cn rise ecuse high degree of sustitution etween domestic nd imported goods my led to high responsiveness of trde flows even in the cse of very modest trde rriers. However, in n industry with high elsticity σ, consumers re so price-sensitive tht high rtio of domestic over ilterl trde does not necessrily reflect high ilterl trde rriers ut rther high degree of competition. 9 The definition of θ ij comines the phi-ness mesure with n exponent tht involves the industry-specific elsticity of sustitution σ so in contrst to phi-ness, θ ij is le to seprte this competition effect from the trde rrier effect. As eqution (14) shows, higher elsticity of sustitution implies lower trde frictions θ ij. Liewise, s we show in the technicl ppendix, the theoreticl models y Eton nd Kortum (2002), Chney (2008) nd Melitz nd Ottvino (2008) lso require tht the trde rtio e scled y n industry-specific exponent. We therefore rgue tht θ ij improves on the phi-ness mesure ecuse it ccounts for differences in mret power nd competition cross industries Dt nd Descriptive Sttistics To compute our mesure of trde integrtion θ ij cross industries, countries nd time, we need the domestic trde of countries i nd j in industry, x ii nd x jj, s well s their ilterl exports, x ij nd x ji,ttimet. As in previous literture (for instnce, Chen, 2004, Evns, 2003, Hed nd Myer, 2000, Nitsch, 2000, Wei, 1996), domestic trde for country i is given y its gross output in industry, yi, minus totl exports of country i to the rest of the world in tht industry. The elsticities of sustitution σ re ten from Hummels (2001). 11 Our smple includes trde flows for 166 mnufcturing industries cross 11 EU countries t the 4-digit Nce rev.1 level etween 1999 nd The smple is lnced over time. The 11 countries re Austri, Denmr, Finlnd, Frnce, Germny, Itly, Irelnd, the Netherlnds, Spin, Portugl nd the United Kingdom. In the dt ppendix we provide detiled description of the dt nd their sources Formlly, the increse in multilterl resistnce follows from eqution (6), implying n increse in x ii through eqution (10). 8 Alterntively, order effects cn e estimted from grvity eqution tht comines domestic nd interntionl trde flows s the dependent vrile (Chen, 2004, Evns, 2003, McCllum, 1996, Wei, 1996, Wolf, 2000). 9 The stndrd CES mrup is given y σ /(σ 1) nd thus inversely relted to σ. 10 We show in Section 5 tht regressions sed on the phi-ness mesure yield considerly lower R 2 thn regressions of θ ij. 11 Hnson nd Xing (2004) lso use the elsticities from Hummels (2001). 12 Due to dt limittions not ll possile trde flow comintions cross industries nd countries re ville. Although it would in principle e possile to spn the longer period from 1997 to 2003, this would come t the cost of losing ll oservtions for Germny, core EU country, ecuse Germn sectorl output dt re missing prior to

11 Our dtset comprises totl of 15,040 domestic nd ilterl trde flow oservtions. Only five ilterl trde flows re equl to zero, which is not surprising given the huge volume of intr-industry trde in the EU. 13 Those five cses would normlly feture s missing vlues of θ ij. But zero trde flows my contin vlule informtion s to why such low levels of trde re oserved. It therefore seems more pproprite to ssocite these cses with lrge trde frictions. The pproch we dopt is to replce the zeros with vlue of one Euro, thus ssociting them with lrge vlues of θ ij. 14 Tle 1 contins summry sttistics for individul industries. Due to spce constrints we list industries t the more ggregted 3-digit level, reporting verges of θ ij over the lower-level 4-digit clssifictions s well s their mximum nd minimum vlues. Industries re ordered y decresing vlue of θ, which rnges from to We lso report the verge elsticities of sustitution σ nd the verge weight-to-vlue rtios, mesured s ilogrms per Euro exported. A first cursory glnce t Tle 1 ttests the intuitive nture of the trde integrtion mesure θ ij. Trde integrtion is lowest for Brics, followed y Cement, lime, plster. The ltter is lso the industry with the lrgest verge weight-to-vlue rtio (over 13 g/euro), indicting very low trnsportility of the goods. Another relted industry is Articles of concrete. Note tht the geogrphic mret for cement nd concrete is very locl since the perishle nture of such wet products constrins the distnce over which they cn e delivered. Printing nd Pulishing re trded very little, too, which is hrdly surprising given the relince of such products on specific lnguges. This finding is consistent with erlier studies showing tht trde in such sectors is sudued. Hrrign (1996) shows tht the volume of trde reltive to output in the OECD in 1985 is the lowest in Printing nd Pulishing. Finlly, some of the sectors with high vlues of θ elong to the food industry, for which the perishility of mny goods is most liely n importnt deterrent to trde. For exmple, Fruit nd vegetles re rned in fourth plce, nd Other food products disply the iggest individul vlue of θ equl to 1,056. Tle 2 reports the verge level of θ i for the 11 individul countries in our smple. Irelnd ppers s the lest integrted country, followed y Spin, Denmr, Finlnd nd Itly, wheres Frnce, the Netherlnds nd Germny re the most trde integrted. Finlly, Figure 1 plots the time series evolution of θ t, verged cross countries nd industries. It is interesting to oserve tht despite fluctutions from one yer to the next, trde frictions disply downwrd trend, suggesting tht the countries nd industries we consider hve on verge ecome more ilterlly integrted over time. 4 Empiricl Results In the first prt of this section, we nlyze the determinnts of trde integrtion cross country pirs, industries nd yers. In the second prt, we explin how our pproch differs from stndrd 13 The three industries with zero trde flows re Pulishing of newsppers, Builders crpentry nd joinery nd Mttresses. We consider those oservtions s true zeros ecuse the corresponding output vlues re positive ut exports re zero. We exclude the cses where output is zero ut exports re positive s well s the cses where oth output nd exports re zero. 14 Our tretment of the zero trde flows does not distort the overll results, see the roustness chec in Section 5. 9

12 grvity. In the third prt,we focus on the time series ptterns of trde integrtion. In the fourth prt, we decompose the growth in trde over the smple period into two components the growth in mnufcturing output nd improvements in trde integrtion. 4.1 The Determinnts of Trde Integrtion We first nlyze the determinnts of ilterl EU trde integrtion. We estimte ln θ ij,t = ψ t + λ K + βgeo ij,t + ζpolicy ij,t + αcosts ij,t + γ Controls ij,t + ij,t (16) where Geo ij,t is set of vriles relted to geogrphy nd trnsporttion costs, Policy ij,t is set of policy-relted fctors, Costs ij,t includes other types of costs such s fixed costs of exporting nd productivity, nd Controls ij,t includes vriles controlling for mesurement issues. Given tht some of the explntory vriles vry cross sectors only, we include industry fixed effects λ K t the more ggregted level of 3-digit industries, ssuming tht the 4-digit groupings re different vrieties of the corresponding, more ggregted 3-digit sector K (Hummels, 2001). The inclusion of other explntory vriles tht only vry cross country pirs precludes us from controlling for country pir fixed effects. 15 ψ t denotes yer intercepts. β, ζ, α nd γ re vectors of coefficients to e estimted nd ij,t isnerrorterm. Ahighervlueofθ ij should e interpreted s lower degree of trde integrtion. We estimte our seline results with OLS. To control for possile utocorreltion in ech of the individul series of the pnel, roust stndrd errors re djusted for clustering t the 4-digit Nce rev.1 level in ech country pir (15,040/5 = 3,008 clusters). As roustness chec, we lso report results otined y Poisson Pseudo-Mximum Lielihood (PPML) estimtion, see Section 5. Geogrphy/Trnsporttion Costs Vriles Tle 3 reports our min results. Column (1) only includes vriles tht re relted to geogrphy nd trnsporttion costs. Across countries, stndrd grvity vriles perform well in explining the vrition in trde integrtion. Trde integrtion decreses with interntionl distnces D ij nd increses with domestic distnces D ii D jj.itishigher etween countries tht shre common order Adj ij nd common lnguge Lng ij. We lso consider proxies of trnsporttion costs tht vry cross industries. First, we consider the rtio of c.i.f. nd f.o.. trde vlues for ech industry nd country pir, verged in ech yer cross ll EU prtners in order to minimize mesurement error, cfo t (Hrrign, 1993). 16 Second, we consider the weight-to-vlue rtio of exports for ech industry nd country pir, lso verged in ech yer cross ll prtners, wv t. We do not consider ilterl weight-to-vlue rtios s Hillerry nd Hummels (2000) show tht ilterl weight-to-vlue significntly flls with distnce. This suggests tht the commodity composition of trde is sensitive to ilterl costs nd tht weight-to-vlue is endogenous. 17 Overll, since the freight component of trde costs is higher for uly, high weight-tovlue rw mterils thn for mnufctures, weight-to-vlue should e ssocited with higher vlues for θ ij. As expected, higher trnsporttion costs s proxied y cfo t nd wv t decrese trde integrtion. 15 As roustness chec, we show tht our results hold up with country pir fixed effects, see Section Aprt from insurnce nd other possile discrepncies etween prtners, the rtio of c.i.f. nd f.o.. for intr-eu trde vlues mostly cptures trnsporttion costs y rod. In 1998, 57.8 percent of totl intr-eu15 trde went y rod, 22.8 percent y se, 3.9 percent y ril, 3.9 percent y ir, 0.9 percent y inlnd wterwys nd 0.8 percent y pipeline (Europen Commission, 2000). 17 Hrrign (2005) nd Bldwin nd Hrrign (2007) lso find tht export unit vlues re positively relted to distnce. 10

13 Policy Vriles In Column (2) we consider the role of severl policy-relted fctors in ffecting trde integrtion. Across countries, we include dummy vrile for Finlnd nd Austri, FI,AT ij, s these two countries were the lst in our smple to join the EU (in 1995), nd they pper less integrted compred to the other countries. We lso explore the effect of not dopting the Euro y including dummy vrile, noeuro ij,t, which is equl to one for Denmr nd the United Kingdom from 2002 onwrds when the common currency ws introduced. 18 As expected, non-doption of the Euro hs lowered the extent of trde integrtion for these two countries since We lso dd vrile to cpture the effects of the Schengen Agreement which olishes physicl order controls mong prticipting EU countries. The dte of the first implementtion of the greement differs cross countries. Irelnd nd the United Kingdom hve not strted implementtion yet. The coefficient on this vrile, Schengen ij,t, is negtive nd highly significnt in Column (2), suggesting tht the olition of order controls mong the prticipting countries hs helped to foster trde integrtion, most proly through the elimintion of time delys nd dministrtive urdens tht were previously experienced t orders. We lso consider policy vriles tht vry cross industries. First, we ddress the role of Technicl Brriers to Trde (TBTs). TBTs result from norms (regultions nd stndrds) tht ffect the sle of goods in some mrets y requiring specific product chrcteristics or production processes. Bldwin (2000) stresses the importnce of TBTs in shping trde flows etween countries nd industries. He rgues tht in the cse of Europe, such rriers hve ecome more nd more visile over time, especilly since triff rriers were completely eliminted y In ddition, TBTs re predominnt concern in tody s glol trde negotitions nd for the WTO in prticulr. Dt on TBTs re hrd to find, however, so our pproch to mesuring TBTs uses two different sources of informtion: cross-country survey of EU mngers who revel whether they consider TBTs prolem for trde, nd rning of industries ccording to the relevnce of TBTs. The intercted vrile, TBTij, increses with the extent of TBTs cross countries nd industries. The dt ppendix provides detils. In Column (2) TBTs re found to lower trde integrtion, suggesting tht some room is left for policy ction nd tht the removl of such rriers might promote trde integrtion. This result is consistent with Chen (2004) who finds tht TBTs in Europe re ssocited with lrger order effects in interntionl trde. As is often the cse, results sed on survey dt should e interpreted with cution. In prticulr, the cross-country mesurement of TBTs might e prolemtic. For instnce, if mnger sttes tht TBTs re not prolem, this does not necessrily men tht TBTs re sent or unimportnt. Insted, if tht mnger s compny or industry is protected y TBTs initited y his country, he might ctully prefer them. This possiility might is our cross-country mesurement of TBTs. 20 In Column (3) we therefore report estimtes when only including our industry-specific vrile on TBTs, 18 The Euro exchnge rtes were fixed in 1999 ut Euro notes nd coins nd thus greter degree of trnsprency only followed in The huge literture on the trde-creting effects of currency unions rises the issue of endogeneity of the common currency dummy (Rose, 2000, Bldwin, 2006). In this pper we do not ttempt to ddress this issue. 19 As explined y Bldwin (2000, p.255), Europe s first lierliztion efforts focused on the esy rriers, triffs nd quots. With these eliminted y 1968, lierliztion ttention turned to TBTs. 20 If this is indeed the cse, we will underestimte the impct of TBTs on trde integrtion so tht our coefficients cn e considered lower-ound estimtes. 11

14 TBT. The results remin consistent, i.e., when TBTs re high, trde integrtion is low. 21 Furthermore, it is well-estlished tht ntionl governments often fvor domestic over foreign firms for some of their purchses, even if foreign suppliers could ctully offer them etter del (Dvies nd Lyons, 1996). Firms in such pulic procurement mrets re hence protected from foreign imports, sometimes to such n extent tht trde my e completely suppressed. In the lst few yers, competition hs een stedily incresing in those mrets, with the proportion of EU15 pulic procurement contrcts openly dvertised in the Officil Journl of the EU stedily incresing from 8.4 percent in 1995 to 16.8 percent in 2005 (Eurostt). To investigte whether this opening of the mrets hs helped to lower trde frictions, we rely on time-vrying cross-country dt on the shre of pulic procurement contrcts dvertised in the Officil Journl nd compute the verge cross prtners. As similr dt cross industries re not lwys ville, we compute n indictor vrile for high pulic procurement sectors. 22 We interct this indictor vrile with the time-vrying dt for ech country pir, denoting the vrile so otined y Proc ij,t. Column (2) revels tht the opening of pulic procurement mrets to foreign competition hs indeed een successful in fostering trde integrtion cross countries nd industries nd over time. Next, Vlue-Added Txes (VATs), which differ extensively cross EU countries nd goods, re chrcterized y the so-clled destintion principle : VAT for good is pid in the country where it is sold, not where it is mnufctured, implying tht VATs uniformly ffect domestic trde nd imports. In our model, the imposition of VATs y country i thus increses oth domestic costs t ii nd importing costs t ji y the sme proportion, leving ilterl trde integrtion θ ij unchnged. In Column (2) we include the log of one plus the verge VAT rte cross prtners t the sectorl level, VAT ij,ndinterestinglyfind tht higher VATs re ssocited with significntly lower trde frictions. One possile interprettion reltes to VAT frud. Bldwin (2006) documents tht VAT frud is very serious prolem in the EU. Since the removl of Europe s internl orders in 1993, trde sttistics re now collected y VAT uthorities, creting direct correltion etween trde sttistics nd VAT frud: EU firms hve n incentive to over-report their exports (to get the VAT rete) nd to under-report their imports (which re suject to VAT). This infltes export sttistics reltive to import sttistics. As result, in the countries nd industries where VATs re high, the vlue of θ ij might e lower simply ecuse intr-eu export sttistics re inflted due to VAT frud. 23 Other Costs The new trde literture on heterogeneous firms rtionlizes why some firms export nd others do not. In prticulr, the models y Melitz (2003), Melitz nd Ottvino (2008) nd others show tht only the most productive firms prticipte in foreign mrets. Only those firms re productive enough to cover the fixed costs of exporting or to withstnd tough competition on foreign mrets. Due to this selection effect we should oserve tht ceteris prius, more productive industries trde more on foreign mrets thn less productive industries, i.e., we should oserve negtive 21 The use of this sectorl vrile on TBTs might e criticized on the grounds tht it cptures chnges in TBTs rther thn levels (see the dt ppendix). As in Chen (2004), we experimented using dummy vrile only, which is equl to one when TBTs re present. The results remin unffected. 22 Exmples of high pulic procurement industries re Shipuilding, Ril stoc, Phrmceuticls, Aerospce, Telecoms. 23 If the intr-eu exports of country i re inflted ecuse of VAT frud, then x ij will e high reltive to x ii, leding tolowervlueofθ ij. 12

15 reltionship etween productivity nd trde frictions θ ij. In Column (4) of Tle 3 we therefore include verge rel lor productivity cross industries nd countries. Moreover, growing empiricl litertureshowsthtstrdecostsfll,lessefficient firms exit from the mret nd verge industry productivity increses (for exmple, see Bernrd, Jensen nd Schott, 2006, for the U.S. nd Greenwy nd Kneller, 2008, for the UK). To mitigte potentil endogeneity prolems we lg productivity y one yer, Prod ij,t 1, reducing the numer of oservtions y one cross section. The negtive reltionship etween productivity nd θ ij confirms the prediction tht more productive industries trde more on foreign mrets. 24 We lso include dummy vrile in Column (4) tht controls for the five zero trde oservtions in our smple, Zeros ij,t. We find tht trde frictions etween two countries in given industry re stronger whenever either of them does not export. As models of heterogeneous firms typiclly predict tht firms only export if they re le to cover the fixed costs of exporting, the result might reflect theroleofthesefixed costs. Controls Finlly, we control for some mesurement issues nd the possiility tht the mgnitude of θ ij my e ffected y the nture of trde. In prticulr, the vlue of θ ij computed under intr-industry trde will e lower thn under comprtive dvntge driven y technology or fctor endowment differences cross countries, even if the ctul friction is the sme. 25 To control for this potentil is we compute the (solute) difference in cpitl shres etween countries nd industries in order to proxy for differences in fctor endowments. For lrger differentil in cpitl shres, trde is more liely sed on comprtive dvntge, leding to n overestimted θ ij. This vrile, KSij,t, displys the expected positive (ut insignificnt) coefficient in Column (5) ut its inclusion leves the coefficients on the other vriles virtully unchnged. The previous resoning presumes tht differences in fctor endowments cn e fully cptured y differences in cpitl shres, the ltter eing computed t the industry level. One ovious limittion stems from the use of dt disggregted t the level of industries nd not t the level of products. Industry clssifictions inevitly ggregte mny different types of products into one single ctegory so tht the volume of intr-industry trde cn pper more importnt thn it ctully is. For instnce, mny different types of steel re produced, from flt-rolled to specilty steels, nd it my e tht the production of some types of steel requires resources or technologies in which one country hs comprtive dvntge. However, since ll these types of steel re ggregted into one ctegory, it ppers s if the countries export nd import lmost identicl products while in relity they export one type of steel nd import nother type. Thus, the lrger the numer of vrieties in ech industry ctegory, the more liely the industry ggregtes trde flows with different comprtive dvntges in different vrieties, ut in the dt this will show up s intr-industry trde, i.e., more lnced trde flows etween the two countries. 24 We lso tried to use dt on productivity for country tht is not included in our smple, i.e., the US. But those dt re ggregted t the 2-digit level nd proved insignificnt in explining trde integrtion cross EU countries, possily due to the higher level of ggregtion. 25 With intr-industry trde, the denomintor of (14) will e given y the product of two lnced trde flows. But with comprtive dvntge, the denomintor of (14) will e given y the product of two unlnced trde flows, leding to reltively smll mgnitude in the denomintor nd thus overestimted trde frictions. 13

16 The resulting θ ij is therefore expected to e smller ecuse the denomintor in eqution (14) is the product of two more or less lnced (intr-industry) trde flows t the industry level, msing imlnces of (inter-industry) trde flows t the product level. 26 To control for this ggregtion is, we include the numer of product ctegories within ech industry, Goods. InColumn(5)itisindeed the cse tht lrger numer of vrieties is ssocited with lower vlue for θ ij. But the inclusion of this control hrdly ffects other estimtes. In Column (6) we replicte the specifiction of Column (5) ut replce the country-vrying vrile on TBTs y the sectorl vrile, TBT. It is ressuring tht ll results remin virtully unchnged, providing some evidence tht they re not ised y the use of cross-country survey dt on TBTs. Finlly, Tle 4 reports vrince decomposition to illustrte the quntittive contriution of ech fctor in explining the vrition of θ ij. Following Fields (2003), the contriution of ech explntory vrile x m to the totl vrince of θ ij,t is clculted s c m = β m cov x m,θ ij,t /vr θ ij,t where βm is the prtil regression coefficient of θ ij,t on the explntory vrile x m (holding ll other explntory vriles constnt). Columns (1) nd (2) correspond to regressions (5) nd (6) in Tle 3, respectively. Column (1) shows tht regression (5) in Tle 3 explins 80.8 percent of the vrition in θ ij,t percent is ttriutle to the 3-digit industry fixed effects nd 0.01 percent to the yer fixed effects. Of the 38 percent tht the remining regressors explin, geogrphy nd trnsport costs lone re responsile for out 25 percent of the vrition in θ ij, the most importnt fctor eing weight-tovlue (16.7 percent), followed y ilterl distnce (5.4 percent). Policy vriles explin 7.6 percent of the vrition in θ ij, which is fr from negligile. TBTs re the most importnt fctor (4.5 percent), while pulic procurement, Schengen nd opting out of the Euro only ply minor roles. The numer of product ctegories within ech industry explins up to 5.4 percent of the vrition in θ ij. Finlly,the decomposition reported in Column (2), which corresponds to regression (6) in Tle 3, gives similr results. 4.2 How Our Approch Differs from Stndrd Grvity We now compre our specifiction to the stndrd grvity pproch nd illustrte the empiricl pitflls of not ting the theory seriously. We elieve this is n importnt nd necessry exercise. Although specifiction (16) is relted to stndrd grvity estimtion, there re three importnt differences. First, stndrd grvity uses output yi y j s the size vrile. In contrst, the trde integrtion mesure θ ij includes domestic trde x ii x jj ecuse this nets out the multilterl resistnce vriles. Second, correctly specified grvity equtions hve to include time-vrying industry fixed effects for ech country pir to control for multilterl resistnce vriles (Bldwin nd Tglioni, 2006). But prcticl prolem is tht these fixed effects re perfectly colliner with mny explntory vriles of interest such s cfo t nd wvt in Tle 3. We void this prolem since θ ij nets out the time-vrying multilterl resistnce vriles. Third, our theory clls for trde cost elsticities tht vry cross industries. Those re emodied in θ ij. 26 The product of two lnced trde flows yields reltively lrge mgnitude in the denomintor nd thus reltively low θ ij. 14

17 We now evlute the is tht rises in stndrd grvity equtions due to the omission of multilterl resistnce nd the ssumption of coefficient homogeneity cross industries. In its simplest form (henceforth the nive model), the trditionl grvity model reltes ilterl trde to output nd ilterl trde costs where the trde cost elsticity α 2 is ssumed to e the sme cross industries: ln ³ x ijx ji = α 1 ln ³ y i y j + α 2 ln ³ t ijt ji + ij (17) We drop time suscripts for simplicity. Compre the nive specifiction in eqution (17) to the log-linerized version of micro-founded grvity eqution (12): ³ ³ ³ ³ ln x ijx ji = β 1 ln x iix jj + β 2 (1 σ )ln t ijt ji β 3 (1 σ )ln t iit jj + ij (18) The size vrile is domestic trde insted of output. Moreover, eqution (18) includes domestic trde costs nd the trde cost coefficients now vry cross industries. 27 An lterntive micro-founded grvity specifiction cn e otined y log-linerizing eqution (11): Ã! ³ ³ x ln x ijx ji =ln yi yj i x ³ ³ j +ln y y +(1 σ )ln t ijt ji +(σ 1) Π i Pi Π j Pj To control for the unoservle multilterl resistnce vriles Π i P i Π j P j, we include fixed effects for ech industry (t the 4-digit level) in ech yer nd seprtely for countries i nd j, denoted y γ i nd γ j. Such fixed effects lso wipe out the domestic demnd terms (if included seprtely for ech country) nd world output in industry. To ese the comprison with the nive specifiction, we eep the domestic output terms so tht we estimte ln ³ x ijx ji = δ 1 ln ³ y i y j + δ 2 (1 σ )ln ³ t ijt ji + γ i + γ j + ij (19) To eep our exposition s simple s possile, we ssume tht trde costs (domestic nd interntionl) re proxied y the log of distnce only. We then estimte the nive specifiction (17) nd compre its distnce elsticity to the ones estimted in the micro-founded specifictions (18) nd (19). Tle 5 reports our results. The upper pnel refers to nive grvity while the two lower pnels refer to the micro-founded models. For ll pnels, Column (1) ssumes homogeneity on the trde cost coefficient cross industries (ilterl nd domestic). We relx this ssumption in Column (2) where we report the verge of the elsticities estimted for individul industries. In Columns (3) nd (4), we replicte the specifictions in Columns (1) nd (2) ut we llow the output vriles, domestic distnces nd domestic trde to enter seprtely for ech country. As cn e seen in Column (1) of the upper pnel, the nive grvity eqution yields n solute distnce elsticity of 3.589, which increses slightly to once we llow for coefficient heterogeneity in Column (2). 28 The sectorl output regressor displys coefficient slightly lrger thn the unity vlue predicted y theory. All elsticities remin similr in Columns (3) nd (4). 27 Comes, Lfourcde nd Myer (2005, eqution 13) estimte form of eqution (18) where they constrin β 1 to unity. But they use ggregte trde dt so tht y construction the trde cost elsticity cnnot vry cross industries. 28 The mgnitude of this coefficient is lrger thn tht typiclly otined in the literture ecuse the dependent vrile is the product of ilterl exports nd not unidirectionl ilterl exports. 15