Trade and Agglomeration: the Strategic use of Protection Revisited

Transcription

1 Trade and Agglomeraton: the Strategc use of Protecton Revsted Thede, Susanna 2007 Lnk to publcaton Ctaton for publshed verson (APA): Thede, S. (2007). Trade and Agglomeraton: the Strategc use of Protecton Revsted. (Workng Papers, Department of Economcs, Lund Unversty; No. 7). Department of Economcs, Lund Unversty. General rghts Copyrght and moral rghts for the publcatons made accessble n the publc portal are retaned by the authors and/or other copyrght owners and t s a condton of accessng publcatons that users recognse and abde by the legal requrements assocated wth these rghts. Users may download and prnt one copy of any publcaton from the publc portal for the purpose of prvate study or research. You may not further dstrbute the materal or use t for any proft-makng actvty or commercal gan You may freely dstrbute the URL dentfyng the publcaton n the publc portal Take down polcy If you beleve that ths document breaches copyrght please contact us provdng detals, and we wll remove access to the work mmedately and nvestgate your clam. L UNDUNI VERS I TY PO Box L und

2 Trade and Agglomeraton: the Strategc use of Protecton Revsted Susanna Thede Aprl4,2007 JEL Codes: F12, F13, R12, R13 Keywords: New economc geography, nput-output lnkages, strategc trade polcy, Nash polcy game. Abstract The purpose of ths paper s to examne whether the strategc motve for protecton present n trade and agglomeraton models, n the so-called new economc geography framework, s senstve to the standard assumpton that there s a sole agglomeraton ndustry. We frst nvestgate unlateral trade polcy effects on the nternatonal producton and trade pattern and the resultng natonal welfare levels n a new economc geography model ncludng several agglomeraton ndustres. The strategc use of trade polcy s then examned by dentfyng optmal polcy postons as well as equlbrum polcy strateges n a Nash polcy game between the trade partners. Our results show that the strategc use of protecton and the resultng tarff war outcome prevalent n standard trade and agglomeraton models s senstve to the ncluson of several agglomeraton ndustres. Specfcally, trade lberalsng polces are optmal and free trade equlbra result from the Nash game unless there s a too wde ndustry gap n agglomeraton economes. Our results show that the case for free trade can be drectly attrbuted to ether a relatvely strong drect polcy mpact on natonal real ncome (workng through rased mport prces and reduced mport volumes) or smlar agglomeraton economes n the two ndustres. The results of ths paper jontly suggest that the stark argument for the strategc use of protecton present n standard new economc geography models can be attrbuted to overemphassed gans from agglomeraton and/or the lack of ndustres wth smlar agglomeraton economes. Ths paper s an offsprng of my prevous work. I gratefully acknowledge the comments made by Fredrk Andersson, Rchard Baldwn, Rkard Forsld, Ben Ferret, Peter Neary, Hanna Norberg, Pascals Ramondos-Moller and Therry Verder. Materal ncluded n the paper has also benefted from suggestons made by semnar group partcpants at the Department of Economcs, Lund Unversty, DELTA, Pars, the School of Economcs, Unversty College Dubln and the School of Economcs, Nottngham Unversty as well as partcpants at the ZEW Economc Geography workshop n Mannhem 2003 and the Irsh Economc Assocaton n Belfast Department of Economcs, Lund Unversty, Box 8072, Lund, Sweden. E-mal: Susanna.Thede@nek.lu.se 1

3 1 Introducton The normatve trade polcy lterature provdes two man arguments for the use of protecton. The termsof-trade argument for protecton states that, when producton s charactersed by perfect competton and constant returns to scale, a country wth a non-neglgble market sze gans from usng protecton as the rased terms-of-trade effect s large enough to exceed the encountered deadweght loss. Ths outcome, whch was ntroduced by Johnson ( ), has had a profound effect on the understandng of the optmal use of protecton. It has, for nstance, been used to explan the unlateral pattern of protecton (see Broda, Lmão and Wensten (2006)) and the economc foundaton of the multlateral trade negotaton system (see Bagwell and Stager (1999)). The other man economc argument for protecton, whch was ntally proposed by Graham (1923), reles on the observaton that a country can beneft from protectng an ncreasng-returns-to-scale ndustry n order to get a cost advantage over compettors n nternatonal markets. Whle both arguments have ganed much attenton n the trade polcy feld, the strand of research focused on the strategc use of protecton n ncreasng-returns-to-scale ndustres has been generally crtcsed for ts lack of generalty. 1 In a general equlbrum context, the argument for protectng ncreasng-returns-to-scale ndustres carres over to neoclasscal trade models. As shown by Flam and Helpman (1987) and Venables (1987), a country always gans from restrctng ts trade n an ndustry charactersed by monopolstc competton of the Dxt-Stgltz (1977) form (whch s the standard form used to ncorporate monopolstc competton n general equlbrum settngs). Ths result reles on the fact that, due to free market entry n the home market or nternatonal frm moblty, the demand shft caused by unlateral protecton leads to an ncrease n the domestc number of frms and a conceptual welfare gan. The trade models that ncorporate agglomeraton economes, n the so-called new economc geography framework, are based on the Dxt-Stgltz (1977) form of monopolstc competton and nclude addtonal welfare gans from protecton stemmng from trade cost effects on the concentraton of economc actvty n the presence of agglomeraton economes. Several researchers have dentfed and dscussed the economc motve for unlateral protecton n standard new economc geography models (see e.g. Krugman and Venables (1995), Neary (2001), and Baldwn et. al. (2003)). These researchers have argued that the stark motve for protecton seemngly prevalent n these models cannot be readly nterpreted as a polcy recommendaton due to the specfcty of the underlyng theoretcal framework. In partcular, as ponted out by Neary (2001), a reason to take cauton n nterpretng trade polcy effects n the new economc geography framework s that the drect lnk between natonal welfare and the protectonst support of an agglomeraton ndustry could be senstve to the exstence of more than one agglomeraton ndustry. In ths paper, we dentfy the strategc use of trade polcy n a new economc geography model ncorporatng several agglomeraton ndustres. The purpose of ths paper s to examne whether the strategc motve for protecton present n trade and agglomeraton models s senstve to the standard assumpton that there s a sole agglomeraton ndustry. Specfcally, we nvestgate optmal trade polcy outcomes and equlbrum polces of a strategc trade polcy game played by the trade partners n a new economc geography model ncorporatng two agglomeraton ndustres. Snce agglomeraton economes rely on nput-output lnkages between frms n our nternatonal new economc geography settng, a natural consequence of allowng for several agglomeraton ndustres s that the strategc trade polcy determnaton can be related to the use of 1 See, amongst others, Brander (1995) and Leahy and Neary (2000). 2

4 ntermedate nputs between as well as wthn ndustres. Emprcal observaton shows that the use of nter-ndustry nputs s large at the narrowly defned level of ndustry aggregaton sutable to examne frm agglomeratons. 2 For example, on the bass of 4-dgt level 1997 nput-output data for the USA, 75 percent of manufactured nputs used n manufacturng producton were produced outsde the own ndustry. 3 In the same data, typcal agglomeraton ndustres such as semconductor, electronc component, pharmaceutcal and medcne manufacturng share an average nput use of 53 percent from other manufacturng ndustres. We use a Krugman and Venables (1996) model modfed to allow for asymmetrc ndustry condtons and trade cost asymmetres. In dentfyng the strategc use of trade polcy whle elaboratng on the specfc role played by nput-output lnkages for the formaton of ndustry agglomeratons, ths paper s closely related to the work by Krugman and Venables (1995) and Puga and Venables (1999). 4 In ther 1995 contrbuton, Krugman and Venables depcted the postve protecton effects on the localsaton of ndustral producton and natonal welfare. In turn, Puga and Venables (1999) showed that ths effect contnues to preval when the Krugman and Venables (1995) model s modfed nto a developng country context. The man polcy conclusons n Puga and Venable s (1999) model were that unlateral protecton can be used to create an economc foundaton for ndustralsaton and that a trade-lberalsng polcy may consttute an alternatve and optmal nstrument for ndustralsaton. As wll be further descrbed below, ths trade lberalsng result s not found n standard new economc geography models charactersng the trade and producton pattern between rch, ndustralsed countres. Snce prevous researchers have descrbed that strategc polcy nteractons between countres would lead to a tarff war n standard new economc geography models wth one agglomeraton ndustry (see Baldwn et al. (2003) and Neary (2001)), we examne both polcy optmalty per se and the polcy outcomes resultng n a strategc trade polcy game between the trade partners. Our am s to dentfy polcy outcomes consstent wth nternatonal dstrbutons of economc actvty that are ndependently targeted as natonally welfare-maxmsng by the trade partners. The polcy game therefore takes the form of a non-cooperatve Nash game n whch trade partners smultaneously choose the polcy strateges maxmsng ther natonal welfare levels. Equlbrum polces of the game thereby reflect the mutually consstent strateges of the trade partners. The rest of the paper s structured as follows. The modfed Krugman and Venables (1996) model s presented n the next secton together wth a detaled descrpton of ts equlbrum outcomes. In secton 3, the natonal welfare effects of unlateral protecton are depcted and the welfare-maxmsng trade polces are dentfed. Secton 4 contans a descrpton of the Nash polcy game and ts equlbrum polcy strateges. In the last secton, a summary of the paper s man approach and conclusons s provded together wth some concludng remarks. 2 See Davs and Wensten (2003). 3 Ths nput-output table, whch s based on the NAICS ndustry classfcaton, can be downloaded from the US Bureau of Economc Analyss webste 4 Welfare effects of unlateral protecton n the new economc geography framework has also been dentfed by Baldwn and Robert-Ncoud (2000) and Baldwn et. al. (2003). 3

5 2 The model We use a modfed Krugman and Venables (1996) model, whch s a two-country model that ntutvely can be nterpreted as descrbng a country s trade vs-à-vs the rest of the world. There are two countres that share the same endowments, preferences and technologes and there are two ndustres charactersed by monopolstc competton of the Dxt-Stgltz (1977) form. A fnal dfferentated good s produced n each ndustry usng labour and ntermedate nputs from both ndustres. Labour s the only prmary producton factor and the labour force n each country s normalsed to one for smplcty. The same basc condtons govern producer behavour n the two ndustres and the countres share the same nherent economc condtons. We therefore restrct the model descrpton to nclude condtons facng producers n one sole ndustry, referred to as ndustry, and economc condtons for the home country only. Foregn country varables are denoted *. We follow prevous trade-polcy researchers of new economc geography settngs n keepng the standard model assumpton that trade costs are of the Samuelson ceberg form. Ths mples that a share of exports melts away n transt and that the domestc protecton level n sector, τ & 1, equals the number of good- unts that has to be exported from the foregn country for one unt of good to reach domestc consumers. In order to smplfy the dentfcaton of trade polcy effects, t s assumed that there are nfntesmall natural restrctons to trade (that cannot be nfluenced by polcy). Wthout a lmtng effect of borders, the polcy outcomes would be more complcated to dentfy whle no addtonal nformaton would be ganed of the strategc use of trade polcy n practce. In addton to ths small alteraton of the orgnal model assumptons, three man model modfcatons are made to enable the nvestgaton of unlateral trade polcy effects. Two man of the modfcatons affect the trade cost specfcaton n allowng for trade cost asymmetres on one hand and n formalsng that trade barrer revenues are generated on the other hand. We follow Baldwn and Robert-Ncoud (2000) n modellng trade barrer revenues consstent wth the ceberg trade cost specfcaton by lettng the government collect goods lost due to domestc protecton. In turn, these goods are transformed nto a publc good Z that s costlessly provded to the populaton. Consumers share an addtve utlty functon u consstng of a Cobb-Douglas part depctng the marketed goods consumpton and an addtve term depctng the publc good s consumpton: 5 u = M 0.5 Mj θz, 0 < θ 0.5, (1) where M and M j are consumpton ndces of good- and good-j varetes and θ s a parameter capturng the relatve preference for publc good consumpton. As prevously noted by Baldwn and Robert-Ncoud (2000), the welfare mpact of trade barrer revenues s neglgable n ndustralsed (OECD) countres. Ths suggests that θ should have a value close to zero n our ndustralsed country settng. To provde general evdence of the revenue ncentve for protecton, we nevertheless choose an upper boundary of θ equal to 0.5. For example, ths could capture the relatvely stronger dependency on trade barrer revenues n newly ndustralsed countres. The consumpton ndex for each dfferentated good takes the form of a constant-elastcty-of-substtuton (CES) functon across a large number of produced varetes. The good- consumpton ndex, M,equals: 5 Baldwn and Robert-Ncoud (2000) uses a Cobb-Douglas utlty functon to descrbe the consumpton of all goods. In ths paper, publc good consumpton enters welfare through an addtve term as, n dentfyng optmal protecton levels, the analytcal result that no utlty at all s obtaned at free trade would be msleadng. 4

6 N 1/ρ Z M = m (k ) ρ dk, ρ = σ 1, σ > 1, k =0,..., n,...n, (2) σ 0 where m(k ) s the consumpton of the kth good- varety, n s the number (mass) of domestcally produced good- varetes, N s the total number (mass) of good- varetes, ρ s a parameter capturng the ntensty of preference for varety of good and σ s the elastcty of substtuton between any par of ndustry- varetes. To smplfy the followng exposton, good- varetes are ordered wth the n domestc ones before the n mported varetes. As prevously descrbed, the total amount of goods ncurred from domestc protecton s transformed nto a publc good Z: Z Z =(τ 1) N n m (k ) dk +(τ j 1) ZN j n j m (k j ) dk j,k j =0,..., n j,...n j, (3) where τ j s the domestc ndustry-j protecton level, m(k j ) s the consumpton of the kth goodj varety, n j s the number (mass) of domestcally produced ndustry-j varetes and N j s the total number (mass) of good-j varetes. Good-j varetes are ordered wth the n j domestc varetes before the n j mported ones. Each frm produces ts own varety and s nonstrategc n the sense that t perceves the elastcty of substtuton between varetes as ts elastcty of demand. 6 Frms n the same ndustry and locaton are symmetrc snce they face dentcal producer condtons and each varety of a good enters symmetrcally nto demand. The focus s therefore placed on representatve varetes/frms n the rest of the paper. Each good s produced from a Cobb-Douglas composte of labour and ntermedate nputs. To smplfy, each produced varety s assumed to be used as an ntermedate nput n the producton of fnal goods. 7 The ntermedate nput use n producton s assumed to replcate that of fnal goods n consumpton, so that the demand structure for ntermedate nputs produced n ndustry s specfed by varey ndex (2). The nput unt cost n ndustry- producton, C,equals: C = w 1 µ υ Q µ Q υ j,µ > υ, (4) where w s the ndustry- wage, Q s the prce ndex of M, Q j s the prce ndex of M j, µ s the ndustry- cost share placed on ntermedate nputs from the same ndustry and υ s the ndustry- cost share placed on other ntermedate nputs. Snce the nput composte requrement ncludes M and M j, all produced varetes are requred as nputs n the producton of each fnal good. The thrd man modfcaton of the orgnal model s to allow for ndustry varaton n the preferences for varety, the elastcty of substtuton between varetes and n the nput cost shares. We follow Krugman and Venables (1996) n assumng that the cost share placed on nputs from the same ndustry always exceeds that placed on goods from the other ndustry. 6 Ths assumpton can be consdered reasonable gven the large number of frms. See Helpman and Krugman (1985) for a thorough exposton of ths assumpton and other features of the Dxt-Stgltz (1977) setup. 7 See Venables (1996) for a formal exposton of the dvson of ntermedate and fnal good producers n a new economc geography settng. 5

7 In equlbrum, the domestc ndustry- prce ndex equals: Q = n p 1 σ + n (p τ ) 1 σ 1/(1 σ ) (5) where p and p s the domestc and foregn ndustry- varety prce. The trade cost weght n (5) reflects the fact that there s mll prcng so that the domestc market prce on mported good- varetes are τ tmes larger than ther foregn producer prce. The protecton effect on mport prces mples that domestc fnal good producers have a cost advantage for domestcally produced ntermedate nputs. In general, ths mples that cost lnkages are stronger between frms n the same locaton. The total cost of an ndustry- frm, TC,equals: TC (x )=C (α + β x ) (6) where α and β s the fxed and margnal nput requrement n ndustry- producton and x s the output level of a representatve ndustry- producer. The average cost s decreasng n the output level due to the fxed nput requrement, mplyng that there are ncreasng returns to scale n producton. To smplfy wthout loss of generalty, unts are chosen n such a way that α =1/σ and β =(σ 1)/σ. Snce goods markets are charactersed by monopolstc competton, a representatve producer s margnal revenue equals ts margnal cost n equlbrum. Usng ths fact together wth the β normalsaton and the nput unt cost equvalence n (4) mples that the proft-maxmsng good- varety prce can be expressed as: p = w 1 µ υ Q µ Q υ j. (7) The market clearng condton for an ndustry- varety equals: x = p σ E Q σ 1 + E Q σ 1 τ 1 σ where E and E s the domestc and foregn good- expendture, Q s the prce ndex of the foregn good- consumpton ndex and τ s the foregn ndustry- protecton level. As can be seen from (8), the foregn demand for a domestc ndustry- varety s decreasng n the foregn good- protecton level and the mpact of ths protecton effect (.e. the ncurred mport loss) s ncreasng n the substtuton elastcty between good- varetes. Because the domestc demand for foregn varetes s reduced by foregn protecton, domestc (consumers and) frms have a demand advantage for domestcally produced goods. In general, ths mples that demand lnkages are stronger between producers n the same locaton. There s free market entry and ext, mplyng that equlbrum profts are zero. Usng ths fact together wth (6), (7) and the α and β normalsatons, t can be shown that the output level equals one n equlbrum. Insertng (7) nto (8), usng that there s an equlbrum unt output level and rearrangng the terms yelds: (8) w σ (1 µ υ ) Q σ µ Q συ j = E Q σ 1 A domestc equlbrum s charactersed by seven equlbrum equatons. ndustry-j counterpart are two of these equatons. + E Q σ 1 τ 1 σ. (9) Expresson (9) and ts 6

8 In equlbrum, the fact that frms make zero profts mples that total ndustry- wages are equal to the labour cost share of total ndustry- revenues: w L =(1 µ υ )n p (10) where L s the domestc ndustry- labour force. By solvng (10) for n and usng the resultng expresson and (7) n (5), the good- prce ndex can be wrtten as: Q =((1 µ υ ) 1 (L w 1 σ (1 µ υ ) Q σ µ Q σ υ j +L w 1 σ (1 µ υ ) Q σ µ Q σ υ j τ 1 σ )) 1/(1 σ) (11) where L s the foregn ndustry- labour force, w s the foregn ndustry- wage and Q j s the prce ndex of the foregn good-j consumpton ndex. Expresson (11) and ts ndustry-j counterpart are ncluded amongst the seven domestc equlbrum equatons. The total domestc expendture on a good comes from consumer demand for fnal products and producer demand for ntermedate nputs. Combned wth the ndustry- revenue equvalence obtaned from (10), ths mples that the domestc good- expendture equals: µ E =0.5Y + υ w L + w j (1 L ), (12) 1 µ υ 1 µ υ where the frst term s the fnal good- expendture and the second and thrd term s the ndustry and ndustry-j expendture placed on ntermedate nputs produced n ndustry. The expendture equaton (12) and ts ndustry-j counterpart are two of the domestc equlbrum equatons. The natonal ncome conssts of total labour earnngs: Y = w L + w j (1 L ). (13) The natonal ncome equaton s the last of the equatons charactersng a domestc equlbrum. Labour s assumed to move nto the ndustry offerng the hghest wage, mplyng that the whole labour force earns the same wage n a stable domestc equlbrum. The home country can therefore be actve n both ndustres n a stable domestc equlbrum only f domestc wages are equalsed. In addton, the complete domestc specalsaton n one ndustry s consstent wth domestc equlbrum stablty f the domestc labour force earns a wage that s at least as large as the (theoretcal) domestc wage n the other ndustry. Throughout the rest of the paper, all equlbra referred to are stable unless explctly stated otherwse. A global equlbrum s charactersed by the seven domestc equlbrum equatons and ther foregn counterparts. To summarse, the exogenous parameters of the equlbrum system are σ = σ, σ j = σ j, µ = µ, µ j = µ j, υ = υ and υ j = υ j whle the endogenous parameters are Y, Y, E, E j, E, E j, Q, Q j, Q, Q j, w, w j, w and w j. The domestc and foregn protecton levels τ, τ j, τ and τ j are kept exogenous n solvng for global equlbrum and the dentfed trade polcy effects are based on alteratons n global equlbrum outcomes. To smplfy the further exposton of global equlbra, the case when each country s actve n both ndustres s referred to as a dspersed equlbrum, wthasymmetrc equlbrum consttutng the specal case of dentcal domestc and foregn producton and trade structures, and an agglomerated equlbrum refers to the case when at least one country s completely specalsed n one ndustry. We follow prevous researchers n restrctng net agglomeraton forces to be suffcently weak 7

9 for dspersed equlbra to exst at some trade cost levels that are symmetrc across countres. In the modfed Krugman and Venables (1996) model, ths so-called no-black-hole condton s ndustry-specfc and corresponds to σ (1 µ + υ ) > 1 for ndustry. 2.1 The equlbrum structure The varous equlbrum structures of the model are presented n ths subsecton. To facltate the presentaton, the consequences of each modfcaton of the orgnal model are descrbed separately. The equlbrum structure s found usng the same basc smulaton technques as Krugman and Venables (1996). 8 The smulaton results throughout the rest of the paper are based on numerous smulatons n the followng large parameter ntervals: 1 < σ,σ, σ j, σ j 4, 0.1 µ, µ, µ j, µ j 0.6, 0.05 υ, υ, υ j, υ j 0.4 and 1.01 τ, τ j, τ,τ j 10.9 Fgure 1 dsplays the equlbrum structure obtaned when the orgnal model s modfed to allow forndustryasymmetresntradecostlevels. Thebreakpontscheduledepctsthelowertradecost boundary of symmetrc equlbra and the sustan pont schedule depcts the upper trade cost boundary of agglomerated equlbra. As dscussed n detal by prevous researchers, ths pattern reflects that trade lberalsaton eventually leads to strong enough agglomeraton economes for symmetrc equlbra to be replaced by agglomerated equlbra. As trade costs are reduced, t becomes less proftable for frms to be located close to fnal goods consumers and more proftabletobelocatedclosetoother frms (and share the demand and cost benefts of ndustry concentraton). In the fgure, the equlbrum structure of the orgnal model s the specal case of ndustry symmetry marked by the 45-degree lne. FIGURE 1 ABOUT HERE As can be seen from fgure 1, break and sustan ponts, whch are unque n the orgnal model, become numerous when allowng for ndustry asymmetres n trade cost levels. Wth trade cost asymmetry between ndustres, one-dmensonal sustan and break ponts can be specfed as thresholds n one ndustry s trade cost level at whch the equlbrum structure s altered gven the trade cost level n the other ndustry. As can be seen from fgure 1, these one-dmensonal sustan and break ponts may however not be unque. Frst, there can be two sustan ponts markng an agglomerated equlbrum nterval. In ths case, there s a lower sustan pont below whch no agglomerated equlbra are sustanable due to a relatvely large trade cost gap between ndustres. Specfcally, ndustry concentraton becomes unproftable n the low trade-cost ndustry due to hgh mport costs on ntermedate nputs produced n the hgh trade-cost ndustry. Second, there can be two break ponts depctng an nterval outsde whch there are symmetrc equlbra. Ths mples that there may be a lower break pont below whch symmetrc equlbra exst, reflectng the fact that a dsperson of economc actvty becomes vable when the ndustry gap n trade cost levels grows suffcently large. In fgure 2 and 3, the equlbrum structure s dsplayed when the orgnal model s modfed further to allow for ndustry asymmetres n substtuton elastctes and nput cost shares. These fgures are drectly comparable to fgure 1 n usng the same parameter values except for a devaton n one ndustry 8 The nterested reader s referred to ther paper for detals and to Fujta, Krugman and Venables (1999) for a general descrpton of smulaton approaches used to dentfy equlbra n new economc geography models. 9 Wthn the range of trade cost levels underlyng the smulatons, between 10 and around 100 percent of shpped goods arrve at ther destnaton. See Anderson and Wncoop (2003) for a survey of trade cost estmates. 8

10 varable. Comparng fgure 1 and 2 shows that a reduced substtuton elastcty n ndustry j leads to an outward shft of the break and sustan pont schedules. These shfts reflect that the trade cost mpact on producton costs s enhanced by the fact that varetes are poorer substtutes, whch ncreases the frm benefts of ndustry concentraton. The shfts are larger along the ndustry-j trade cost axs as a hgher cost share s placed on ntermedate nputs from the same ndustry. FIGURE 2 ABOUT HERE FIGURE 3 ABOUT HERE In comparson, fgure 2 and 3 shows that a reducton n the own nput cost share n ndustry j leads to a downward shft of the break pont schedule and an nward shft of the sustan pont schedule. These shfts dsplay that a reducton of the own nput cost share ncreases the relatve frm gans of locatng close to consumers, thereby reducng the sustanablty of agglomeraton per se. In addton, the nward shft of the sustan pont schedule s larger along the ndustry-j trade cost axs as the negatve mpact on agglomeraton economes n the own ndustry exceeds the negatve mpact (transferred through nter-ndustry nput-output lnkages) on agglomeraton economes n the other ndustry. Allowng for nternatonal trade cost asymmetres does not change the general outcome that economc actvty becomes concentrated when trade s suffcently lberalsed. Unlateral trade polcy does nfluence the net benefts of ndustry concentraton, however, whch mples that a country can use ts trade polcy stand to alter the current producton structure. Ths general result has been shown to preval by prevous researchers n new economc geography models ncludng one agglomeraton ndustry. To present the general smulaton results of unlateral trade polcy effects on the producton pattern, we use an addtonal graphcal representaton of the equlbrum structure provded by Krugman and Venables (1996). In fgure 4 to 7, the domestc and foregn ndustry- employment shares are measured along the two axes and domestc and foregn equlbrum curves trace out the employment combnatons consstent wth natonal wage equalsatons between ndustres. The wage n ndustry j exceeds (s exceeded by) that n ndustry above (below) an equlbrum curve. Snce workers move nto the ndustry offerng the hghest wage, a country that s actve n both ndustres automatcally moves towards ts equlbrum curve. In contrast, a country that s actve n only one ndustry may not move towards ts equlbrum curve as a naton s complete specalsaton s consstent wth a (theoretcal) ndustry gap n wages. A global dspersed equlbrum s charactersed by wage equalsatons wthn countres, mplyng that t occurs at the ntersecton of the domestc and foregn equlbrum curve. In turn, a global agglomerated equlbrum can be located n between equlbrum curve ntersectons wth an axs. If domestc and foregn wage gaps n the neghbourhood of the equlbrum create a worker movement between ndustres towards the equlbrum pont, the global equlbrum s stable. To smplfy the exposton of unlateral trade polcy effects on the equlbrum structure, there s an ntal trade cost symmetry between countres n the fgures below. The dsplayed polcy effects are based on general movements of the equlbrum curves shown to preval n numerous smulatons (ncludng those based on an ntal trade cost asymmetry between countres). FIGURE 4 ABOUT HERE FIGURE 5 ABOUT HERE 9

11 FIGURE 6 ABOUT HERE FIGURE 7 ABOUT HERE The above fgures dsplay protecton effects on equlbrum structures ncorporatng a symmetrc equlbrum (n fgure 4), a combnaton of symmetrc and agglomerated equlbra (jontly shown n fgure 5 and 6) and agglomerated equlbra (n fgure 7). 10 As can be seen from these fgures, a protectonst ndustry- polcy leads to an outward shft of the domestc equlbrum curve and an nward shft of the foregn equlbrum curve. Due to the symmetry of the model, a protectonst ndustry-j polcy has the opposte effect on the equlbrum curves. These protecton effects are systematc and prevalent for all underlyng parameter combnatons n the smulatons and, as can be seen from the fgures, lead to a replacement of equlbra for large enough unlateral trade cost devatons (unless the home country already s completely specalsed n the targeted ndustry). 11 As can be seen from the graphs, the systematc shfts of the equlbrum curves yeld the same overarchng effect on the producton pattern, no matter whch the ntal equlbrum structure s, provdng that the unlateral polcy devaton s suffcently large. 12 Moreover, ths effect s establshed even f the polcy devaton leads to a replacement of the equlbrum structure tself. Specfcally, the smulaton results reveal that a country always can replace an equlbrum wth one charactersed by a larger domestc producton n the targeted ndustry. The result that a country can use protecton to rase ts producton n an ndustry replcates the one obtaned by prevous researchers n new economc geography models ncorporatng one agglomeraton ndustry. 13 The strategc role played by protecton n the new economc geography framework can be attrbuted to the mpact of trade cost asymmetres on frms relatve proftablty of settng up producton or remanng n a partcular locaton. Frst, the unlateral protectonst support of an ndustry rases domestc frm profts n the targeted ndustry. Second, the rased relatve proftablty of domestc producton n the targeted ndustry attracts foregn producers nto the home country. Thrd, the domestc labour market competton s ntensfed n the process and domestc wages are rased, whch leads to reduced frm profts n both ndustres. Fourth, the adverse effect on domestc frm profts lowers the relatve proftablty of domestc producton, whch trggers a frm movement out of the country n the non-targeted ndustry. The frm movements and labour market adjustments trggered by the protectonst polcy dssolves the ntal global equlbrum and contnue untl a global equlbrum s re-establshed. The symmetry of the model mples that the results for trade-lberalsng polces readly can be nterpreted as opposte to those trggered by protecton. Ths mples that a country has trade-lberalsng alternatves to the protectonst polces used as nstruments n targetng a preferred producton pattern (unless ntal trade costs are too low for free trade polces to be potent nstruments for replacng equlbra). For nstance, a country can become specalsed n ndustry by use of a domestc protectonst ndustry- polcy and/or a trade-lberalsng ndustry-j polcy. Ths result can be attrbuted to general 10 In addton to the prevously descrbed stable equlbrum structure, unstable equlbra are dsplayed n the fgures. Snce an unstable global equlbrum s unsustanable, we abstan from gvng any further reference to these equlbra n the rest of the paper. 11 Ths effect becomes neglgble at very hgh trade cost levels. 12 The polcy devaton requred to replace equlbra range from a margnal change n the case of replacng dspersed equlbra at relatvely low trade cost levels to an excessve change needed to replace dspersed equlbra at very hgh trade cost levels. 13 See Krugman and Venables, 1995, Puga and Venables, 1999, and Baldwn et. al,

12 equlbrum effects n the presence of several agglomeraton ndustres and s novel n a new economc geography context. The result that trade lberalsaton can be used to rase the domestc producton n an agglomeraton ndustry s consstent wth the result obtaned by Puga and Venables (1999). As prevously descrbed, they show that a developng country can lberalse ts trade n an agglomeraton ndustry to establsh an economc foundaton for ts producton n the targeted ndustry. Ther result hnges on the fact that the wage gap between countres becomes large enough for producers n the agglomeraton ndustry to gan more from low labour costs than from a proxmty to the agglomeraton of ther ndustry and a drect access to the larger market. The focus on smlar countres n our model mples that nternatonal labour cost dfferences play a smaller role n determnng the locaton of producton. In fact, the strong country smlarty mples that relatve labour costs always mpose a weaker effect on the locaton choce of frms compared to relatve market szes and the relatve openness of markets. The strongest effect of a trade-lberalsng polcy on the locaton choce of frms n our model s therefore to rase the relatve beneft of foregn producton n the ndustry by ncreasng the foregn country s attractveness as an export base. 3 Natonal welfare effects of unlateral trade polcy In ths secton, the natonal welfare effects of unlateral trade polcy are descrbed and optmal polces are dentfed gven the trade polcy poston of the trade partner. To elucdate the presentaton, these effects are examned separately for changes n real ncome and publc good provson. 14 As prevous trade polcy research performed n the new economc geography framework, the man focus n ths paper s placed on therealncomeeffect that captures the combned welfare effect of alteratons n market prces and changes n the producton and trade pattern. In addton, trade polcy effects on the publc good provson are dentfed wth reference to changes n the amount of goods collected by the government. The reported welfare comparsons n ths secton represent general results obtaned for numerous smulatons n the prevously depcted parameter ntervals. To smplfy the presentaton, welfare consequences of polcy alteratons are descrbed from ntal observatons based on trade cost symmetry between countres. The results can thereby be drectly related to the prevously descrbed underlyng equlbrum structures. To elucdate the result descrpton throughout the rest of the paper, t s assumed that agglomeraton economes are at least as strong n ndustry as n ndustry j. The smulaton results reveal that the drect lnk between the natonal real ncome and the domestc specalsatonnanndustrythathasbeenshowntopreval n new economc geography models based on one agglomeraton ndustry contnues to preval n our settng only n the presence of a relatvely large ndustry gap n agglomeraton economes. In ths case, a rased domestc specalsaton n the ndustry wth stronger agglomeraton economes leads to a domestc real ncome gan. Ths result s due to the fact that an ncreased domestc producton n the ndustry wth stronger net agglomeraton forces leads to larger consumer gans from ndustry concentraton based on reduced domestc market prces. As descrbed n the prevous secton, agglomeraton economes are stronger n an ndustry wth a lower elastcty of substtuton between product varetes, a hgher nput cost share placed on nputs from the own ndustry and a lower nput cost share placed on other nputs. The nherent agglomeraton 14 ThedomestcrealncomeequalsRI = wq 0.5 Q 0.5 j. 11

13 economes n ndustry can be measured by ae =(σ (1 µ + υ )) 1, whch replcates the measure of agglomeraton economes underlyng the no-black-hole condton. Wth country symmetry n trade costs, agglomeraton economes are strengthened by hgher trade costs n the own ndustry and lower trade costs n the other ndustry as relatvely stronger demand and cost lnkages are created between frms n an ndustry recevng rased protectonst support. The ndustry- agglomeraton economes are therefore ncreasng n the ndustry- trade cost gap τ τ j. In general, stronger net agglomeraton forces enhance the mpact of a gven ndustry gap n net agglomeraton forces on the allocaton of producton. Ths mples that the ndustry gap n net agglomeraton forces requred to create a drect lnk between natonal real ncome and the domestc producton n the ndustry wth stronger agglomeraton forces s smaller, the stronger are the agglomeraton economes n the other ndustry. Based on the descrbed measures of agglomeraton economes, a two-dmensonal estmate of the ndustry- gap n net agglomeraton forces s gven by (gap ae,gap τ )=((σ (1 µ + υ )) 1 1 σ j 1 µj + υ j, τ τ j ). TABLE 1 ABOUT HERE In table 1, natonal welfare effects of unlateral trade polcy are reported for observatons wth relatvely large ndustry gaps n net agglomeraton forces. The table contans exogenous parameter values, trade cost levels, the ntal equlbrum structure, measures of nherent agglomeraton economes, the two-dmensonal ndustry gap n net agglomeraton forces, equlbrum producton structures as depcted by the domestc and foregn ndustry- employment share and the polcy mpacts on real ncome, publc good provson and natonal welfare. As prevously descrbed, the welfare mpact of publc good provson depends on the preference parameter θ. To provde general evdence of the revenue ncentve for protecton, the maxmum welfare effect of publc good provson (based on θ =0.5) sreportedntable 1 together wth the nterval of possble natonal welfare outcomes (wth contrbutons of publc good provson rangng from θ =0to θ =0.5). The real ncome gan of usng trade polcy to rase the domestc ndustry- producton can be verfed from observatons n the table above. Ths table also provdes evdence of the general smulaton result that the real ncome gan from rasng the domestc ndustry- producton s larger f done by use of a trade-lberalsng ndustry-j polcy nstead of ts protectonst ndustry- polcy equvalent. Ths result hnges on the fact that rased trade costs ncrease market prces and reduce mport volumes. Polcy effects on welfare n undssolved agglomerated equlbra are also reported n the table. Snce the ntal specalsaton pattern contnues to preval wth non-dssolvng polces, ther real ncome effect can be entrely attrbuted to the drect polcy effect on domestc market prces and mport volumes. Ths mples that non-dssolvng protectonst (trade-lberalsng) polces have a negatve (postve) effect on natonal real ncome by reducng (ncreasng) the domestc consumpton possbltes of marketed goods. 15 Smulaton results show that the publc good provson always s hgher f a targeted specalsaton pattern s establshed by use of a protectonst polcy nstead of ts trade-lberalsng polcy equvalent. Ths general result, whch can be seen from observatons n table 1, contrasts to the rankng of polcy alternatves on the grounds of real ncome motves as the publc good provson s fnanced by (marketed) goods collected by the government as trade barrer revenues. The effect of non-dssolvng polces on the publc good provson s entrely attrbuted to the drect polcy mpact on the amount of goods collected by the government. Snce ths amount s ncreasng n the domestc protecton levels for a gven 15 Ths result s vald except n the case when all producton n the targeted ndustry takes place n the home country. 12

14 domestc producton structure, the polcy effect on the publc good provson s postve (negatve) wth a protectonst (trade-lberalsng) non-dssolvng polcy. 16 Wth a relatvely large ndustry gap n net agglomeraton forces, smulatons show that the welfare orderng of polcy alternatves ranks the real ncome effect above that of publc good provson. Ths mples that a country gans from usng trade lberalsaton nstead of protectonsm to attan a preferred domestc producton structure. Due to the strong gans from ndustry- specalsaton, ths result ndcates that the optmal polcy combnaton s that whch maxmses the domestc ndustry- producton by use of mnmal protecton. The welfare-maxmsng unlateral polcy stand s therefore a free trade ndustryj polcy combned wth the mnmum ndustry- protecton level requred to maxmse the domestc ndustry- producton. 17 Wthout a relatvely large ndustry gap n net agglomeraton forces, the natonal welfare s no longer drectly lnked to the domestc specalsaton n one ndustry. Instead, the real ncome effect of ndustry specalsaton s reduced by the fact that the two ndustres have agglomeraton economes of smlar strength. Indeed, smulaton results show that ths effect can be small enough to be exceeded by the drect polcy effect on real ncome. Smulatons also show that, n exceptonal cases, ths polcy combnaton may yeld a total real ncome effect that s suffcently small to be exceeded by the welfare effect of publc good provson. TABLE 2 ABOUT HERE In table 2, welfare effects of unlateral trade polcy are presented for observatons wthout a relatvely large ndustry gap n net agglomeraton forces. For comparablty, table 2 reports nput values and estmates n the same parameter categores as table 1. The general result that no drect lnk prevals between natonal welfare and the domestc ndustry- specalsaton can be seen from observatons n the table. Frst, there can be real ncome gans from ndustry-j specalsaton. Second, the use of protecton to establsh a domestc specalsaton n ndustry may lead to a real ncome loss. Thrd, the effect of publc good provson may (n exceptonal cases) exceed the total real ncome effect. Smulaton results show that there may be real ncome gans from specalsaton per se wthout a relatvely large ndustry gap n net agglomeraton forces. In table 2, ths result can be seen from observatons for whch real ncome gans are made from polces replacng a completely dspersed (symmetrc) equlbrum wth equlbra charactersed by a domestc specalsaton n any ndustry. The largest real ncome beneft from specalsaton s nevertheless attaned wth a maxmsed domestc specalsaton n the ndustry wth stronger agglomeraton economes and, f ndustres are symmetrc, equvalent real ncome gans are obtaned from the maxmsed specalsaton n any ndustry. The smulatons reveal that, when the real ncome effect manly s attrbuted to the ndrect specalsaton effect, the prevously descrbed orderng of polcy effects contnues to preval. In table 2, ths result can be recognsed from observatons for whch an ncreased specalsaton leads to welfare gans no matter whether establshed wth a protectonst polcy or ts trade-lberalsng polcy equvalent. Snce 16 Ths result s vald except n the case when all producton n the targeted ndustry takes place n the home country. 17 No welfare effect s ncurred from rasng the ndustry- protecton level further once all ndustry- producton takes place n the home country, whch mples that there s no upward lmt to the optmal use of ndustry- protecton. We abstan from gvng further reference to ths case on the grounds of realsm snce the mere exstence of mplementaton costs for protecton would rule out ths case. 13

15 therelatvempactofthedfferent polcy effects on welfare replcate those dentfed for a relatvely large ndustry gap n agglomeraton economes n these cases, the same optmal polcy s dentfed unless ndustres are symmetrc. In that case, the optmal polcy poston s nstead a free trade poston n each ndustry snce t combnes a maxmsed domestc specalsaton wth the largest consumpton possbltes of marketed goods. 18 In other words, whle agglomeraton economes are strong enough for the ndrect real ncome effect of specalsaton to exceed other polcy effects on welfare, there s no ncentve to use protecton as long as agglomeraton economes are as strong n the two ndustres. If the drect polcy effect s the predomnant real ncome effect, there are real ncome gans from trade lberalsaton and real ncome losses from protecton no matter whch the establshed specalsaton pattern s. 19 In table 2, ths case can be seen from observatons for whch protecton (trade lberalsaton) that ncreases (reduces) the domestc ndustry- specalsaton leads to a real ncome loss (gan). In these cases, the smulaton results show that the real ncome effect always exceeds the welfare effect of publc good provson f a trade lberalsng polcy s used (so that the two real ncome effects renforce each other) and, so that the optmal trade polcy stand s a free trade poston. Ths orderng of welfare effects s also vald for non-dssolvng polces as can be seen from the example provded n the table. Onthebassoftheseandtheprevousresults,therevenue motve for protecton can be ruled out as a determnant of the optmal polcy stand, whch s a result coherent wth the emprcal observaton that trade barrer revenues mpose a neglgble mpact on natonal ncome n ndustralsed countres. 3.1 Strategc trade-polcy nteractons Ths secton descrbes the strategc trade-polcy nteractons between countres, whch ntutvely can be nterpreted as the home country s unlateral polcy stand vs-à-vs a polcy poston summarsng the multlateral polces of ts trade partners. In practce, the strategc trade polces could correspond to polcy stands that the home country and ts trade partners brng to the table n trade negotatons. Snce our man objectve s to dentfy strategc polcy postons consstent wth global equlbrum n the trade model, the polcy game takes the form of a non-cooperatve game n whch the trade partners smultaneously choose polcy strateges that maxmse ther natonal welfare levels. Equlbrum polces thereby consttute polcy strateges that are ndependently targeted as natonally welfare-maxmsng when global equlbrum effects on the trade and producton pattern are taken nto account. The polcy game s assumed to be a Nash game where equlbrum outcomes are defned by the fact that none of the trade partners want to devate from ther chosen polcy strateges gven the polcy poston of the other country. The countres are assumed to be perfectly nformed of the trade partner s natonally welfare-maxmsng objectve, global economc condtons and the consequences of each domestc and foregn polcy choce. A smple one-shot game s used to formalse strategc trade-polcy nteractons n our model n coherence wth the fact that the tme dmenson s absent n the new economc geography framework. Snce the optmal polces n the prevous secton were dentfed gven the trade partner s polcy poston, t s suffcent to establsh that these polces are mutually consstent to dentfy a Nash equlbrum. As prevously descrbed, wth a relatvely large ndustry gap n net agglomeraton forces, t s optmal for 18 Due to the smplfyng assumpton that mnscule natural trade costs exst at free trade, a maxmsed specalsaton also occurs f the trade partner uses a free trade poston. 19 Ths result s vald except for non-dssolvng polces used n an ndustry n whch all the producton s domestc. 14

16 a country to use a free-trade ndustry-j poston combned wth the ndustry- protecton level requred to attan a maxmsed domestc ndustry- specalsaton. Ths mples that one of the countres always gans from rasng ts ndustry- protecton level above that of ts trade partner to establsh a locatonal equlbrum charactersed by ts maxmsed ndustry- producton. Ths result ndcates that no Nash equlbrum exsts and reflects the tarff war outcome prevously dentfed for new economc geography models wth one agglomeraton ndustry. 20 Wthout a relatvely large ndustry gap n agglomeraton economes, the outcomes of the polcy game can be categorsed nto two groups based on the dentfed polcy optmalty. When the optmal polcy choce replcates the one attaned wth a relatvely large ndustry gap, so does the outcome of the strategc polcy game. Ths result hnges on the fact that a maxmsed domestc ndustry- specalsaton can be optmal even f other polcy effects on welfare are large enough to undercut the drect lnk between natonal welfare and domestc ndustry- producton. Ths replcaton of the tarff war outcome can be drectly attrbuted to a relatvely large welfare mpact of ndustry- specalsaton. The optmal polces were dentfed as free trade polces for all remanng underlyng parameter combnatons, mplyng that the mutually coherent equlbrum polces of the Nash game are free trade postons. Ths polcy equlbrum s consstent wth a maxmsed nternatonal specalsaton due to the smplfyng assumpton that very small natural trade costs exst so that an agglomerated equlbrum structure prevals at free trade. Notably, each country s welfare s maxmsed wth respect to ts own polcy poston as well as that of ts trade partner n ths Nash equlbrum. Ths s the consequence of that the largest domestc consumpton possbltes of marketed goods are attaned at free trade n combnaton wth the fact that the opportunty cost of marketed good consumpton s to hgh for publc good consumpton to occur n equlbrum. The free trade polcy equlbrum s based on ndustry symmetry and/or weak agglomeraton economes (partally due to the nter-ndustry use of ntermedate nputs). 4 Summary and Concludng Remarks In ths paper, the strategc use of protecton n agglomeraton and trade models (wthn the so-called new economc geography framework) has been revsted. Specfcally, we have examned whether the welfare argument to protect agglomeraton ndustres carres over to a new economc geography model ncludng more than one agglomeraton ndustry. In so dong, partcular focus has been placed on the role played by nput-output lnkages between ndustres wth reference to evdence that these lnkages matter for producton n agglomeraton ndustres. The paper s approach has been to examne natonal welfare effects of unlateral trade polcy n a modfed Krugman and Venables (1996) model n order to dentfy the polcy optmalty. Ths was done on the bass of polcy effects on the relatve szes of the drect real ncome effect, the ndrect real ncome effect caused by specalsaton and the effect on the domestc publc good provson (fnanced by trade barrer revenues). Our results showed that ndustry asymmetres n agglomeraton economes play a vtal role n determnng the polcy optmalty. If the gans from agglomeraton, as measured n terms of net agglomeraton forces, are suffcently larger n one ndustry compared to the other, the welfaremaxmsng polcy s smlar to that dentfed by prevous researchers n that t s optmal to protect 20 See Baldwn et. al. (2003). 15