Agglomeration or Dispersion? Industrial Specialization and Geographic Concentration in NAFTA

Transcription

1 Journal of Internatonal Economc Studes (2006), No.20, The Insttute of Comparatve Economc, Studes, Hose Unversty Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA Klms Vogatzoglou Unversty of Patras Abstract We analyze ndustral specalzaton and geographc concentraton patterns wthn the NAFTA regonal ntegraton area durng and examne the determnants of spatal concentraton. The results ndcate that NAFTA countres have become ncreasngly dssmlar over tme. A changng spatal structure of total NAFTA manufacturng s also evdent. Manufacturng s ncreasngly relocatng to Mexco, whch comes at the expense of the US. In addton, there s evdence of a general upward trend n the degree of relatve geographc concentraton of North Amercan ndustres. Labor-ntensve and low-technology actvtes appear to be the most spatally concentrated ndustres, exhbtng a strong ncreasng trend. Comparatve advantage factors largely explan geographc concentraton of ndustres across NAFTA countres, ndcatng the emprcal relevance of tradtonal trade theory n the NAFTA case. Keywords: Specalzaton; Geographc concentraton; Economc ntegraton; NAFTA JEL classfcaton: C23; F15; L6; R12 1. Introducton Snce the late 1980s, economc ntegraton n North Amerca has progressed steadly. The establshment of the North Amercan Free Trade Agreement (NAFTA) n 1994 between the US, Canada, and Mexco gave a further stmulus to ntra-north Amercan trade and nvestment. Generally, t s accepted that nternatonal economc ntegraton s benefcal, leadng to aggregate welfare gans. However, effcency and welfare gans resultng from resource reallocaton and ncreased ndustral specalzaton have usually a long-term horzon and come along wth at least short-run structural adustment pressures. These pressures are partcularly mportant n the labor market context. In addton to these effects, the spatal mplcatons of the ongong North Amercan economc ntegraton process are equally crucal. A hghly polcy relevant queston n the NAFTA context s whether and how ntegraton affects the spatal structure of economc actvtes across NAFTA countres. In ths connecton, one of the key concerns of the US and Canada s that of losng domestc manufacturng ndustry to Mexco, as a result of deeper economc * Unversty of Patras, Department of Busness Admnstraton, Unversty Campus, Ro - Patras, Greece. Offce Tel./Fax: , e-mal: vogatzo@upatras.gr 89

2 Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA ntegraton n North Amerca (Cremeans, 1999). Gven the crucal mportance of all these ssues, NAFTA ntegraton has gven rse to a substantal number of emprcal studes. However, most of the emprcal lterature s concerned wth ether ntra-nafta trade patterns of member countres (Brox, 2001; Brulhart and Thorpe, 2001; Sarkar and Park, 2001), or ntegraton effects on regonal patterns of ndustral specalzaton and localzaton nsde a gven NAFTA country (Bene and Coulombe, 2004; Holmes and Stevens, 2004; Hanson, 1997; 1998a). Compared to the above ssues, lttle attenton has been pad to the spatal effects of North Amercan economc ntegraton on specalzaton and ndustry locaton wthn the whole NAFTA space. In fact, most emprcal studes on specalzaton patterns wthn a regonal ntegraton area (across member countres) refer to the European Unon (Agnger and Pfaffermayr, 2004; Mdelfart-Knarvk et al., 2002; Brulhart, 2001; Haaland et al., 1999). Yet, gven the polcy and welfare relevance of the topc, the knowledge of how specalzaton and ndustry locaton wthn the NAFTA area has evolved durng the ntegraton process s partcularly valuable. The present study seeks to provde the emprcal evdence on ths ssue. In partcular, the am of ths paper s to analyze patterns of ndustral specalzaton and geographc concentraton of ndustres wthn the NAFTA space and to dentfy the forces that drve the spatal concentraton of North Amercan manufacturng ndustres. The remander of the paper s organzed as follows. Secton 2 provdes a bref overvew of the relevant theoretcal and emprcal lterature. In secton 3, we dscuss data and measurement ssues and present the descrptve emprcs. Secton 4 dscusses specfcaton and estmaton ssues of our econometrc model and presents the determnants of spatal concentraton n NAFTA. Fnally, secton 5 concludes. 2. Theoretcal and Emprcal Background 2.1. Theory Though trade and locaton theores are closely related, and ndeed consttute two sdes of the same con (Isard, 1956; p. 207), they have long been regarded as separate sub-dscplnes of economcs. In fact, t was not untl the early 1990s that economsts recovered the spatal dmenson of nternatonal trade theory. Specfcally, Krugman s (1991) poneerng work has been an mportant step to brng nternatonal economcs and regonal economcs closer together and for the development of a theoretcal framework of ndustry locaton that encompasses elements of trade theory, economc geography and urban economcs. At the present development, trade theory can be categorzed nto three dfferent strands or classes of models: tradtonal trade theory, new trade theory, and new economc geography. Here we provde only a short and broad overvew, as comprehensve surveys of recent developments n trade and locaton theory can be found elsewhere (e.g. Ottavano and Puga, 1998; Futa et al., 1999; Neary, 2001). Tradtonal trade theory s characterzed by constant returns to scale, perfect competton, homogeneous goods and dentcal-homogeneous preferences. Trade and locaton s determned exogenously by country characterstcs. Thus, nternatonal trade and specalzaton s drven by natonal dfferences n technologes and relatve factor endowments. In ths settng, economc ntegraton leads to nter-ndustry trade and thus nter-ndustry specalzaton. Neoclasscal theory mples that f there are no dfferences n comparatve advantage across countres or trade costs are extremely hgh, then economc actvtes wll be perfectly dspersed. However, comparatve advantage s not the only source of trade and specalzaton. In fact, the 90

3 emprcal observaton of hgh and growng trade volume between smlar countres has been the drvng force for the development of the new theores of trade. New trade theory (Krugman, 1980; Helpman and Krugman, 1985) ncorporates ndustry-specfc characterstcs, such as economes of scale, mperfect competton, dfferentated goods, trade costs and draws attenton to domestc market sze, whch s determned by the sze of the labour force. The theory predcts the home market effect, where countres specalze n and export those products n whch they have a large domestc market. Increasng returns producton concentrates n one locaton (n the country wth the larger domestc demand) n order to realze scale economes and mnmze trade costs. In ths case, economc ntegraton can nduce nter-ndustry or ntra-ndustry specalzaton, dependng on the mportance of scale economes and trade costs. As trade costs fall towards zero, ncreasng returns ndustry wll tend to locate near the core, nducng thus more nter-ndustry trade and specalzaton between the core and the perphery. Though n the new trade theory trade and specalzaton s not determned by exogenous comparatve advantage, market sze s exogenous and thus the theory fals to explan the core-perphery structure across countres. In the new economc geography, countres (or regons) are assumed to be dentcal n all aspects and the core-perphery pattern s determned endogenously. The man story s that an ntal symmetrc equlbrum can result n a new locatonal equlbrum, where producton and demand structures across regons are no longer dentcal. Industral locaton becomes entrely endogenous, because of ether market sze spllovers (Krugman, 1991) or vertcal (nput-output) lnkages between ndustres (Venables, 1996), whch can nduce crcular processes of agglomeraton. The mplcaton of the former model s that economc ntegraton can lead two dentcal countres to become dfferentated nto an ndustral core and an agrcultural perphery, whle the latter predcts the concentraton of vertcally lnked ndustres n one locaton Emprcs for North Amerca Klms Vogatzoglou Despte the theoretcal advances that have been made n the feld, our emprcal knowledge s stll rather lmted. Thus compared to the rch and ever-growng theoretcal lterature, the emprcal lterature has made less progress and there s clearly scope for an extenson of the avalable emprcal evdence. Ths s especally true as regards drect tests of the new theores of trade and economc geography. In fact, a large number of studes refer to descrptve emprcs. Here we lmt our selectve overvew of emprcal studes to North Amerca. Krugman s (1991) work can be regarded as the startng pont of ths lne of emprcal research. Krugman (1991) calculated locatonal Gn coeffcents for 106 ndustres and found that low-tech ndustres are amongst the most concentrated US sectors. However, he argues that ths rather surprsng result could be drven by data consderatons, as some hgh-technology products are bured n meanngless aggregates, and n ths way some hghly localzed hgh-tech sectors are excluded. In addton, by comparng four large US regons (Northeast, Mdwest, South, West) wth four large European countres (France, Germany, Italy, UK), he found that regonal concentraton was more pronounced n the US rather than n Europe. However, he also found that between 1947 and 1985 US manufacturng became less geographcally concentrated. Km (1995) examned US regonal specalzaton patterns over a longer tme perod ( ) and found that regonal concentraton was hgher n 1860 than n Hs analyss ndcates a regonal de-specalzaton trend snce the late 1920s, whch has been assocated wth the shft n the US economc structure from manufacturng to the servce sector. 91

4 Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA Drawng on state and frm-level employment data, Dumas et al. (1997) found that geographc concentraton of ndustres n the US declned slghtly between 1972 and Furthermore, t s argued that the observed de-agglomeraton process seems to be drven by new frms locatng at relatvely perpheral regons. In addton to the declnng degree of ndustry concentraton n the US, there s also evdence that the economc geography n the sense of regonal structure of economc actvty of the US has changed substantally over tme. Intally, US manufacturng was heavly concentrated n the Northeast, the Mdwest, and n the Great Lakes regon (n the so-called manufacturng belt), but snce the 1950s manufacturng actvty has moved out of ths regon and nto the southern and western regons (Hanson, 1998b; Holmes and Stevens, 2004). Wth respect to Canada, the country s economc geography and degree of spatal concentraton to a large extent resembles the US stuaton, wth the excepton however that regonal concentraton patterns n Canada have remaned largely unchanged over tme (Hanson, 1998b). The most specalzed Canadan regon n manufacturng seems to be Quebec, followed by Ontaro (Holmes and Stevens, 2004). On the other hand, there have been sgnfcant developments n concentraton patterns n Mexco. Hanson s (1997; 1998a) work ndcates that Mexcan manufacturng has been hghly clustered around the Mexco Cty regon, and that snce US-Mexco trade lberalzaton, a process that started n the md 1980s, Mexcan regons close to the US-Mexco border have ganed mportance. Ths delocalzaton process resulted n the creaton of new manufacturng centers n the border regon of Mexco. 3. Patterns of Industral Specalzaton and Concentraton n NAFTA 3.1. Data and measurement The descrptve analyss of ndustral specalzaton and concentraton trends s based on producton (gross value of output) data from the World Matrx of Sectoral Economc Data, where OECD and UN data are provded n a consstent way (n US dollars for all ndustral and trade statstcs), sutable for nternatonal comparsons and studes. Domestc and nternatonal data are provded for 23 2-dgt ISIC (Revson 3) manufacturng ndustres, coverng the perod. We also use data on output, value-added, employment, exports, and mports from the same source for our econometrc analyss, conducted n secton 4. Due to mssng values for some ndustres or NAFTA countres n specfc tme ponts, the analyss ncludes 18 2-dgt ISIC sectors. In general, the ndustral structure of a gven country (or regon) s sad to be specalzed, f a small number of ndustres account for a large share n the country s total producton. On the other hand, a gven ndustry s sad to be geographcally concentrated, clustered, or localzed, f a large part of the ndustry s total producton s carred out n a few locatons (countres or regons). There are several ways to assess the extent of ndustral specalzaton of countres (regons) and geographc concentraton of ndustres. In the emprcal lterature, varous absolute and relatve measures have been used, each havng certan advantages and dsadvantages. In ths paper, we employ what has now become known as the Krugman specalzaton ndex, whch represents a standard relatve measure of regonal specalzaton. Though our analyss s not a study of regonal specalzaton, ths ndex proves to be sutable for the knd of analyss we whsh to pursue. Specfcally, we are nterested n the evoluton of the degree of specalzaton of NAFTA countres wthn the NAFTA area. In other words, the specalzaton structure of every NAFTA member s compared to a benchmark dstrbuton (that of the whole 92

5 NAFTA area). In that sense, NAFTA consttutes the country-level and ndvdual NAFTA members consttute the regons. Ths allows us to examne how North Amercan economc ntegraton has affected relatve specalzaton, concentraton, and ndustry locaton among NAFTA members (wthn the whole NAFTA area) rather than regonal specalzaton patterns wthn a specfc country. We defne the Krugman specalzaton ndex as: Q () t SPEC () t = - Q () t The specalzaton ndex of any country n tme (t) s the sum of the absolute dfferences of the producton shares of and NAFTA as a whole, summed over all ndustres. Ths ndex ranges between zero and two, where an ndex value of zero ndcates that country has the exact ndustral structure as the entre NAFTA area, and a value of two ndcates that country s ndustral structure has nothng n common wth that of NAFTA. In Mdelfart-Knarvk et al. (2002), the Krugman specalzaton ndex s calculated as the sum of the absolute dfferences between a country s producton shares and an average value calculated for all other countres. In our case, t s the average of the whole NAFTA area. In addton, we calculate blateral dfferences, where the ndustral structure of one country s compared to that of the other country. Ths amounts to a standard blateral dssmlarty ndex. Wth respect to geographc concentraton of ndustres wthn NAFTA, we employ the concentraton verson of the Krugman dssmlarty ndex: Q () t Q () t CONC () t = - (2) Q () t Q () t Ths concentraton ndex s calculated annually over the whole sample perod for each ISIC ndustry n our sample Specalzaton trends of NAFTA countres Q () t Q () t Klms Vogatzoglou In ths secton, we brefly examne specalzaton trends wthn NAFTA over the perod. Fgure 1, whch reports specalzaton and blateral dssmlarty ndces, ndcates that Mexco s ndustral structure s the most dssmlar compared to NAFTA s structure, whlst the US seems to be the least specalzed country compared to the producton dstrbuton of the entre NAFTA area. However, gven our defnton of specalzaton wthn NAFTA, ths s not surprsng. The US exhbts a very low ndex value, because US manufacturng producton domnates wthn the NAFTA area. Ths almost amounts to compare the US wth the US producton structure. However, our man nterest les n the dynamcs, where of partcular mportance s the evoluton of specalzaton durng North Amercan economc ntegraton. It s evdent that ndustral specalzaton ncreased n all member states, suggestng that these locatons have become more dssmlar. Ths result s confrmed, f we look at blateral dssmlarty ndces. A sgnfcant dvergence of producton structures s apparent n all country-pars. Though the most dssmlar country-par s US-Mexco, t seems that there has been a perod ( ) of ndustral convergence between those two countres. Notably, the dvergence of producton structures s partcularly pronounced between Canada and Mexco. Insofar as ths dvergence s a result of blateral trade expanson, t reveals that there have been ncreasng nter-ndustry specalzaton trends and severe trade-nduced adustment pressures. However, there s a plethora of other factors that affect the ndustral structures n those coun- (1) 93

6 Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA tres Locaton and geographc concentraton of ndustres wthn NAFTA Our fndng of ncreased specalzaton mples that ndustres must have become more geographcally concentrated wthn the NAFTA space. 1 In fact, the spatal concentraton of economc actvtes s of partcular mportance. Questons such as, what s the extent of localzaton of ndustres and how s t evolvng over tme, whch ndustres are becomng ncreasngly clustered, and where s manufacturng actvty relocatng, are hghly polcy-relevant n the context of North Amercan economc ntegraton. Frst, we turn to the ssue of manufacturng locaton wthn the NAFTA space. As t can be seen from Table 1, whch shows the spatal structure of total NAFTA manufacturng, the bulk of NAFTA manufacturng s, as expected, located n the US. Ths fact s more pronounced n producton rather than n employment terms. However, the evoluton of ths dstrbuton durng North Amercan economc ntegraton consttutes the nterestng part. It s evdent that NAFTA manufacturng s ncreasngly relocatng to Mexco at the expense of the US locaton. Canada s share of total NAFTA manufacturng has remaned unchanged. Though these favorable developments for Mexco s manufacturng sector could be due to varous other factors, t seems that Mexco has ganed from deeper economc ntegraton n North Amerca. Wth respect to the degree of geographc concentraton of NAFTA ndustres, some nterestng fndngs emerge. From Table 2, whch reports concentraton ndces and annual average growth rates for each ndustry, t s evdent that the extent of localzaton vares substantally across ndustres. In 1988, the leather and footwear ndustry was the most geographcally concentrated ndustry, followed by offce and computng machnery, and other transport equpment. In 2000, leather and footwear, tobacco, and motor vehcles were the most concentrated ndustres wthn NAFTA. Notably, some ndustres that are conventonally not assumed to be localzed, exhbt the hghest concentraton ndces. In addton, our fndngs ndcate a general upward trend n spatal concentraton. Most ndustres have become more concentrated over the perod. In order to dentfy what characterzes localzed ndustres wthn NAFTA, we examne some polcy-relevant ndustry characterstcs, assocated wth the observed concentraton trends. For the purpose of our analyss we draw on two OECD (1987, 2001) ndustry classfcatons that contan nformaton on factor ntenstes and technology levels. The frst classfcaton dstngushes fve categores (labor-ntensve, resource-ntensve, scale-ntensve, dfferentated goods, and scence-based ndustres) and the second four categores (hgh-technology, medum-hgh technology, medum-low technology, and low-technology ndustres). We apply the OECD classfcatons to our ndustry sample and calculate average levels of concentraton for four factor-ntensty categores (one category s dropped out, snce only one ndustry n our sample falls nto ths group) and the four technology categores over the sample perod. 2 The results, whch are shown n Fgure 2, ndcate that labor-ntensve ndustres are the most geographcally concentrated ndustres wthn the NAFTA space, followed by scale-ntensve and resource-ntensve ndustres. All categores exhbt an upward trend n concentraton, but ths trend s partcularly pronounced n labor-ntensve ndustres. Lookng 1 However, ths must not be always the case. For a dscusson on ths ssue, see, for nstance, Agnger and Daves (2004). 94

7 at concentraton trends accordng to technology level, t evdent that low-technology ndustres appear to be the most localzed NAFTA ndustres, whlst medum-hgh tech sectors exhbt the second-hghest concentraton level. Notably, hgh-tech sectors dsplay a sgnfcant downward trend, suggestng spatal dsperson. However, ths result s far from beng representatve for ths category, snce we have ncluded only two sectors n ths group, and essentally reflects the sgnfcant negatve trend n the concentraton ndex of ISIC sector 30 (offce and computng machnery). It s nterestng to note that ths ndustry was one of the most concentrated sectors n Overall, these results are not what one would expect, and n general are n lne wth Krugman s (1991) fndng of hgh geographc concentraton of tradtonal low-tech ndustres n the US. 3 Yet, t has to be stressed that some of our results mght be drven by data lmtatons. 4. Determnants of Geographc Concentraton of Industres wthn NAFTA 4.1. Explanatory varables Klms Vogatzoglou Our analyss conducted so far has provded some nsghts on how North Amercan economc ntegraton has affected ndustral specalzaton and geographc concentraton wthn NAFTA. In ths secton, our am s to shed some lght on the drvng forces of the observed localzaton process of North Amercan manufacturng ndustres. In order to explan the observed pattern of geographc concentraton and examne the determnants of ndustral localzaton wthn the NAFTA space, we estmate an econometrc model of relatve spatal concentraton. Though all trade theores (tradtonal comparatve advantage trade theory, new trade theory, and new economc geography) predct that economc ntegraton leads to ncreased spatal concentraton, each theory attrbutes ths concentraton to dfferent determnstc factors. Thus, we construct explanatory varables that proxy for those factors. Although our econometrc model s not an explct and drect test of dfferent trade theores, we could, however, gan some apprecaton of the relatve mportance of dfferent theoretcal frameworks n the case of NAFTA. In order to capture factors assocated wth comparatve advantage trade theory, we construct and nclude two proxes n the explanatory model. The frst varable reflects relatve technologcal dfferences n the Rcardan sense, and s proxed by dfferences n labor productvty, defned as value added per worker. We use a smlar measure of technologcal dfferences as n Haaland et al (1999). Our technologcal dfferences varable s calculated as follows: VA E TECHDIFF = - VA E where VA and E stand for value added and employment, respectvely, and an denote ndus- VA E VA E 2 We have classfed the ISIC ndustres as follows: ISIC 17, 18, 19, 28 (labor-ntensve), ISIC 15, 16, 20, 23, 26 (resource-ntensve), ISIC 24, 25, 27, 34, 35 (scale-ntensve), ISIC 29, 31, 32 (dfferentated goods); ISIC 15, 16, 17, 18, 19, 20 (low-technology), ISIC 23, 25, 26, 27, 28 (medum-low technology) ISIC 24, 29, 31, 34, 35 (medum-hgh technology), ISIC 30, 32 (hgh-technology). 3 Brulhart (2001) found smlar results for the European Unon. 95

8 Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA tres and NAFTA countres, respectvely. The frst term measures labor productvty n ndustry, whlst the second term measures average (of total manufacturng) labor productvty n country relatve to average (of total manufacturng) labor productvty n NAFTA. The greater the dfferences n labor productvty, the hgher the value of ths ndex. Rcardan trade theory mples that the more sgnfcant relatve labor productvty dfferences are, the hgher wll the extent of spatal concentraton be. The second varable assocated wth tradtonal trade theory reflects dfferences n relatve factor endowments, and s proxed by dfferences n labor-ntenstes (defned as employment over value added). Followng Haaland et al (1999), we construct our factor-ntensty varable as follows: The frst term measures labor-ntensty n NAFTA ndustry, and the second term measures average (of total manufacturng) labor-ntensty n NAFTA. The varable takes a hgh value for ndustres that dffer consderably n ther labor-ntensty from the average value (ndustry). Industres that dsplay large dfferences n relatve factor-ntenstes are expected to be relatvely more concentrated. In order to capture new trade theory factors of ndustral localzaton, we nclude n the model a varable that accounts for the mportance of scale economes as well as a varable that reflects the home market effect. Snce one of the predomnant features of the new trade theory are ncreasng returns to scale, many emprcal studes have focused on and used proxes for economes of scale n order to account for the theory s effects. However, the emprcal measurement of scale economes s problematc and usually the varous proxes employed n emprcal analyses devate from the theoretcal defnton. In ths paper due to data lmtatons, we measure scale economes as follows: For a gven ndustry, SCALE s defned as NAFTA value added over NAFTA employment. Industres subect to hgh scale economes are expected to be more spatally concentrated. In order to account for the home market effect, we use a demand bas measure proposed by Haaland et al. (1999). Accordng to new trade theory, a large relatve demand for a gven product results n a large domestc market for ths product, and ts producton concentrates n the home market. Thus, cross-country dfferences n demand structure mght determne ndustry locaton. The demand bas measure employed s an ndex of relatve expendture concentraton, gven by: where EX s expendture n NAFTA country and ndustry, defned as apparent consumpton (output plus mports less exports). Ths ndex s essentally the Krugman concentraton ndex for expendtures. Industres that exhbt a hgher extent of expendture bas are expected to be more clustered. 96 E FACTOR = - VA SCALE = VA E EX EXPEND = - EX E VA EX EX

9 Fnally n order to capture some of the new economc geography effects, we construct a measure that accounts for vertcal lnkages between ( upstream and downstream ) ndustres. Vertcal lnkages are proxed by the proporton of ntermedate goods n the producton of fnal goods. Thus, our ntermedate goods-ntensty varable ncluded n the explanatory model s defned as: INTERM = where Q and VA denote output and value added, respectvely. Accordng to new economc geography theory, the hgher the proporton of ntermedate goods n fnal producton, the stronger the vertcal lnkages, whch determne ndustral agglomeraton. Hence, ndustres wth a hgh ntermedate goods-ntensty are expected to be more agglomerated Estmaton and regresson results Klms Vogatzoglou The econometrc model to be estmated ncludes the followng fve explanatory varables that account for dfferent factors of spatal concentraton: Of course, all ndependent varables are expected to relate postvely to geographc concentraton. The dependent varable s the Krugman concentraton ndex equaton (2) calculated for each NAFTA ndustry and year ncluded n our sample. We perform a smlar estmaton procedure as n Amt (1999), where OLS estmaton wth ndustry and tme dummes s used. The regresson results, whch are reported n Table 3, ndcate that all explanatory varables are statstcally sgnfcant at ether the 1 percent or 5 percent level. The factor-ntensty, technologcal dfferences, and expendture bas varables are hghly sgnfcant and have, as expected, a postve effect on geographc concentraton of ndustres wthn the NAFTA space. These fndngs are n accordance wth the theoretcal predctons of tradtonal as well as new trade theory, and thus both theoretcal frameworks appear to be relevant n explanng ndustral concentraton patterns n NAFTA. On the other hand, the scale economes and ntermedate goods-ntensty varables exercse a sgnfcantly negatve effect on spatal concentraton. These results are rather surprsng and unexpected, and run aganst the theoretcal predctons of the new trade theory and the new economc geography. As regards the unexpected effect of the scale economes varable, t has to be stressed that the negatve coeffcent could be due to our measure we use. However, Haaland et al. (1999), who also fnd a negatve mpact of scale economes n ther study, argue that scale economes have an ambguous mpact on relatve concentraton and relate more to the concept of absolute concentraton. Snce we have conducted an emprcal analyss of relatve concentraton, ths could explan the negatve coeffcent. Wth respect to vertcal lnkages, the model surprsngly ndcates that these lnkages are a determnant of ndustral dsperson rather than concentraton. The negatve coeffcent of our ntermedate goods-ntensty varable may be compatble wth theory f nterpreted n the followng way. Though vertcal lnkages are an mportant factor of ndustral agglomeraton, n a model by Krugman and Lvas Elzondo (1996) these lnkages become weaker as a closed or nward-lookng economy lberalzes and ntegrates nto the nternatonal economy, resultng n de-agglomeraton. The lnkages (and thus agglomeraton) become weaker because for frms Q - Q CONC = f ( TECHDIFF, FACTOR, SCALE, EXPEND, INTERM ) VA 97

10 Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA there s no need anymore to be located close to domestc supplers of ntermedate nputs (and domestc market), snce nputs can be obtaned from nternatonal markets (and products can be exported). Snce our measure does not dstngush between domestc and nternatonal ntermedate goods, an ndustry may exhbt a hgh degree of ntermedate goods-ntensty and at the same tme be dspersed, f ntermedate nputs are obtaned from abroad. Thus f NAFTA members (and hence the NAFTA economy as a whole) have ncreasngly opened-up durng our sample perod (whch s surely the case), the negatve coeffcent of the ntermedate goods-ntensty varable mght be nterpreted n ths way. However, such a scenaro can be verfed only f data on domestc and nternatonal ntermedate nputs are avalable. Comparng the standardzed coeffcents of the regresson model, we dentfy factorntensty dfferences as the most mportant determnant of relatve geographc concentraton of ndustres wthn the NAFTA space, followed by expendture bas and technologcal dfferences. These results ndcate that Heckscher-Ohln (relatve factor endowments) and Rcardan (relatve technologcal dfferences) factors largely explan ndustral agglomeraton across NAFTA countres, provdng support and emprcal relevance for tradtonal comparatve advantage theory. Though the results for the new trade theory are less clear-cut, the hghly statstcally and economcally sgnfcant effect of demand bas on spatal concentraton can be nterpreted as support for the home market effect and thus for the new theory of trade. 5. Conclusons Our analyss reveals that producton structures of NAFTA economes have ncreasngly become more dssmlar over the perod. Thus, the paper s fndngs suggest that North Amercan economc ntegraton nduced ncreasng specalzaton and growng dvergence of ndustral structures between NAFTA countres. Furthermore, there s evdence of a changng pattern of the locaton of NAFTA manufacturng. In partcular, North Amercan manufacturng actvtes are ncreasngly relocatng to Mexco. Ths development comes at the expense of the US manufacturng locaton. Thus, a process of spatal dsperson of the North Amercan manufacturng sector, as a whole s evdent, snce total NAFTA manufacturng becomes less concentrated over tme across NAFTA countres. On the other hand, the evoluton of spatal concentraton (n relatve terms) of ndvdual ndustres wthn NAFTA suggests a process of ndustral agglomeraton, and s n lne wth the fndng of growng country specalzaton. The degree of geographc concentraton has ncreased sgnfcantly n almost all ndustres. Notably, tradtonal labor-ntensve and lowtechnology actvtes appear to be the most spatally concentrated manufacturng ndustres. These ndustres exhbt also a partcularly strong upward trend n ndustral localzaton. Our econometrc analyss sheds some lght on the drvng forces of relatve spatal concentraton of ndustres across NAFTA countres. The regresson results ndcate that several forces are at work, and confrm the fndngs of the descrptve analyss that comparatve advantage factors drve to a large extent the observed localzaton process n NAFTA. In vew of the paper s fndngs a number of polcy-relevant questons arse: * Is the process of growng dssmlarty n the NAFTA space lkely to contnue n the future? * How wll growng specalzaton and localzaton affect the spatal structure of NAFTA manufacturng? * What are the mplcatons of ncreasng specalzaton and geographc concentraton for NAFTA countres, as regards structural adustment, employment, and economc growth 98

11 Klms Vogatzoglou ssues at the regonal and natonal level? These questons reman to be answered and are ndeed subect of further research. However, our analyss has provded some mportant nsghts on the lkely trends. Snce economc ntegraton wthn the NAFTA area s progressng rapdly, the degree of ndustral specalzaton and geographc concentraton s expected to ncrease further. References Agnger, K. and Daves, S. (2004), Industral Specalzaton and Geographc Concentraton: Two sdes of the same con? Not for the European Unon. Journal of Appled Economcs, vol. 7, pp Agnger, K. and Pfaffermayr, M. (2004), The sngle market and geographc concentraton n Europe. Revew of Internatonal Economcs, vol. 12, No. 1, pp Amt, M. (1999), Specalsaton Patterns n Europe. Weltwrtschaftlches Archv, vol. 135, pp Bene, M. and Coulombe, S. (2004), Economc Integraton and Regonal Industral Specalzaton: Evdence from the Canadan?U.S. FTA Experence. Paper prepared for the Canadan Economcs Assocaton Annual Meetng, June 4-6, 2004, Toronto, Canada. Brox, J. (2001), Changng Patterns of Regonal and Internatonal Trade: The Case of Canada under NAFTA. Internatonal Trade Journal, vol. 15, No. 4, pp Brulhart, M. (2001), Evolvng Geographcal Concentraton of European Manufacturng Industres. Weltwrtschaftlches Archv, vol. 137, No. 2, pp Brulhart, M. and Thorpe, M. (2001), Export growth of NAFTA members, ntra-ndustry trade and adustment. Global Busness and Economcs Revew, vol. 3, No. 1, pp Cremeans, J. (1999), Handbook of North Amercan Industry, Bernan Press, Washngton DC. Dumas, G.; Ellson, G. and Glaeser, E.L. (1997), Geographc Concentraton as a Dynamc Process. NBER Workng Paper, No Futa, M.; Krugman, P. and Venables, A.J. (1999), The Spatal Economy: Ctes, Regons and Internatonal Trade, MIT Press, Cambrdge, MA. Haaland, J.I.; Knd, H.J.; Mdelfart-Knarvk, K.H. and Torstensson, J. (1999), What determnes the economc geography of Europe?. CEPR dscusson paper, No Hanson, G. (1997), Increasng Returns, Trade, and the Regonal Structure of Wages. Economc Journal, vol. 107, pp Hanson, G. (1998a), Regonal adustment to trade lberalzaton. Regonal Scence and Urban Economcs, vol. 28, No. 4, pp Hanson, G. (1998b), North Amercan Economc Integraton and Industry Locaton. Oxford Revew of Economc Polcy, vol. 14, No. 2, pp Helpman, E. and Krugman, P. (1985), Market Structure and Foregn Trade, MIT Press, Cambrdge, MA. Holmes, T. and Stevens, J. (2004), Spatal dstrbuton of economc actvtes n North Amerca. Forthcomng n: Henderson, V. and J.F. Thsse (eds.), Handbook of Urban and Regonal Economcs, Amsterdam: North Holland. Isard, W. (1956), Locaton and Space-Economy, MIT Press, Cambrdge, MA. Km, S. (1995), Expanson of Markets and the Geographc Dstrbuton of Economc Actvtes: The Trends n U.S. Regonal Manufacturng Structure, Quarterly Journal of Economcs, vol. 110, pp Krugman, P. (1980), Scale Economes, Product Dfferentaton, and the Pattern of Trade. Amercan Economc Revew, vol. 70, pp

12 Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA Krugman, P. (1991), Geography and Trade, MIT Press, Cambrdge, MA. Krugamn, P. and Lvas Elzondo, R. (1996), Trade Polcy and the Thrd World Metropols. Journal of Development Economcs, vol. 49, No. 1, pp Mdelfart-Knarvk, K.H.; Overman, H.; Reddng, S. and Venables, A. (2002), The Locaton of European Industry. European Economy, Specal Report, No. 2/2002: European ntegraton and the functonng of product markets, Luxembourg. Neary, J.P. (2001), Of hype and hyperbolas: Introducng the New Economc Geography, Journal of Economc Lterature, vol. 39, No. 2, pp OECD (1987), Structural Adustment and Economc Performance. Organsaton for Economc Co-operaton and Development, Pars. OECD (2001), Scence, Technology and Industry Scoreboard: Towards a knowledge-based economy. Organsaton for Economc Co-operaton and Development, Pars. Ottavano, G. and Puga, D. (1998), Agglomeraton n the Global Economy: A Survey of the New Economc Geography. World Economy, vol 21, No. 6, pp Sarkar, S. and Park, H.Y. (2001), Impact of the North Amercan Free Trade Agreement on the US Trade wth Mexco. Internatonal Trade Journal, vol. 15, No. 3, pp Venables, A.J. (1996), Equlbrum Locatons of Vertcally Lnked Industres. Internatonal Economc Revew, vol. 37, pp Fgure1. Specalzaton and structural dssmlarty n NAFTA 100 Fgure 2. Geographc concentraton patterns accordng to ndustry categores

13 Klms Vogatzoglou Table 1. Spatal structure (dstrbuton) of total NAFTA manufacturng, producton Unted States 87,3% 86,3% 86,9% 84,2% Canada 8,5% 7,5% 7,7% 8,4% Mexco 4,2% 6,2% 5,4% 7,4% NAFTA 100% 100% 100% 100% employment Unted States 79,7% 78,4% 78,8% 74,7% Canada 7,6% 6,8% 7,1% 7,8% Mexco 12,7% 14,8% 14,1% 17,5% NAFTA 100% 100% 100% 100% Note: Due to data lmtatons, the shares wth producton (output) data have been calculated wthout the ISIC sectors 21, 22, and 33, whle the correspondng shares wth employment data do not nclude the ISIC sectors 15, 16, 21, 22, and 33. Table 2. Geographc concentraton trends wthn NAFTA by ndustry, ISIC-Industry descrpton Food products and beverages 0,061 0,094 0,087 0,012 0,110 (a) 0, Tobacco products 0,124 0,121 0,211 0,021 0,012 0,106 (a) 17-Textles 0,068 0,077 0,088 0,012 (a) 0,020 (b) 0, Wearng apparel 0,024 0,025 0,126 0,097 (b) -0,062 0,333 (a) 19-Leather products and footwear 0,260 0,340 0,389 0,024 (a) 0,063 (b) 0,045 (b) 20-Wood / products of wood and cork 0,094 0,095 0,140 0,033 (a) -0,016 0,040 (b) 23-Coke & refned petroleum products 0,010 0,034 0,074 0,090 (a) 0,216 (c) 0,155 (a) 24-Chemcals and chemcal products 0,033 0,058 0,096 0,072 (a) 0,122 (a) 0,111 (a) 25-Rubber and plastc products 0,027 0,055 0,067 0,057 (a) 0,166 (a) 0,046 (c) 26-Other non-metallc mneral products 0,068 0,150 0,081-0,025 0,170 (a) -0, Basc metals 0,096 0,067 0,072-0,016 (b) -0,074 (b) 0,012 (b) 28-Fabrcated metal products 0,054 0,084 0,081 0,041 (a) 0,106 (a) 0, Machnery and equpment 0,102 0,128 0,143 0,015 (b) 0,047 (a) 0, Offce and computng machnery 0,168 0,159 0,094-0,056 (a) -0,009-0, Electrcal machnery and apparatus 0,026 0,048 0,079 0,099 (a) 0,136 (a) 0,053 (b) 32-Rado, TV and communcaton 0,051 0,056 0,042 0,033 0,037-0, Motor vehcles 0,106 0,150 0,195 0,033 (a) 0,082 (b) 0,049 (a) 35-Other transport equpment 0,157 0,174 0,142-0,018 0,033 (b) 0,025 Note: 1, 2, and 3 denote annual average growth rate (estmated wth a lnear trend model, ln(conc) t = + *t+u t ) for the tme perods , , and , respectvely; (a), (b), and (c) denote sgnfcance at the 1%, 5%, and 10% level, respectvely. 101

14 Agglomeraton or Dsperson? Industral Specalzaton and Geographc Concentraton n NAFTA Table 3. Determnants of geographc concentraton patterns n NAFTA Independent Varable Coeffcent Standardzed Coeffcent t-statstc p-value Constant 0,034 0,893 0,373 Factor-ntensty ,532 4,490 0,000 Technologcal dfferences 0,041 0,120 3,079 0,002 Scale economes -3,7E-07-0,189-2,055 0,042 Expendture bas 0,345 0,181 4,415 0,000 Intermedate goods-ntensty -0,132-0,144-2,353 0,020 Ad. R2 0,95 F-value 115,8 0,000 Number of observatons