MARGINAL INTRA-INDUSTRY TRADE: TOWARDS A MEASURE OF NON-DISRUPTIVE TRADE EXPANSION

Transcription

1 MARGINAL INTRA-INDUSTRY TRADE: TOWARDS A MEASURE OF NON-DISRUPTIVE TRADE EXPANSION Publshed n Fronters of Research on Intra-Industry Trade, edted by P.J. Lloyd and Hyun-Hoon Lee. Palgrave-Macmllan, 2002 Marus Brülhart Ecole des HEC Unversty of Lausanne CH Lausanne Swtzerland (Marus.Brulhart@hec.unl.ch) 1. Introducton When Verdoorn (1960) found that the formaton of a customs unon among the Benelux countres had stmulated large two-way trade flows of smlar products, and Drèze (1961) dscovered the same phenomenon n the fledglng sx-naton EEC, economsts took note because of one man reason: adjustment costs. Instead of nter-sectoral specalsaton accordng to countres comparatve advantage, the natonal economes seemed to preserve ther broad ndustral structures and to specalse predomnantly at the ntra-sectoral level. A smooth adjustment hypothess (SAH) soon became frmly rooted n economc thnkng, accordng to whch ntra-ndustry trade (IIT) expanson generally entals lower adjustment costs than nter-ndustry trade. In tme, the early conjectures on the shape of emergng trade patterns were confrmed by a number of hgh-profle studes such as those of Grubel and Lloyd (1975), Greenaway and Mlner (1986), Greenaway and Hne (1991) and OECD (1994). These emprcal studes have wthout excepton uncovered a secular ncrease n the share of ntra-ndustry flows n trade among developng as well as developed economes, and many scholars have cted ths fact as

2 a powerful force for attenuatng trade-nduced economc frctons wthn and between countres durng the past half-century. The choce of IIT measure used n emprcal work, however, has untl recently been guded not so much by ts relevance for factor-market adjustment but rather by ts sgnfcance for the theory of nternatonal trade. Snce trade theory conssts to the most part of statc models, the statc Grubel-Lloyd ndex of IIT has been by far the most wdely employed measure. Hgh statc IIT s dffcult to reconcle wth the predctons of neoclasscal trade theory, and ts dscovery has therefore drven the development of the new trade theory, whch can accommodate the exstence of IIT. For that reason, measures of statc IIT can be a useful gauge of the determnants of trade flows. However, Hamlton and Knest (1991) have ponted out n a semnal contrbuton that, n the context of adjustment, dynamc measures of IIT may be more nformatve than statc measures. They proposed an ndex of margnal IIT (MIIT) to be used n studes of the SAH. A number of authors have snce proposed dfferent measures of MIIT, and some emprcal tests of the SAH have been carred out. Ths contrbuton provdes a survey of that growng lterature. The survey s organsed as follows. Secton 2 provdes some theoretcal background to the SAH. In Secton 3 I brefly revew the propertes of the most wdely used measure of IIT, the Grubel-Lloyd ndex. Secton 4 presents some alternatve measures whch capture changes n IIT but not MIIT n the strct sense. An overvew of MIIT measures s gven n Secton 5. Secton 6 s dedcated to a specal class of MIIT measures, namely those that combne nformaton on the symmetry of trade changes wth nformaton on a country s sectoral trade performance. In Secton 7 I conclude wth some generalsatons and suggestons for future research. 2. The Smooth Adjustment Hypothess Adjustment costs arse from temporary neffcences when markets fal to clear nstantaneously n the wake of a change of demand or supply condtons. More specfcally, the adjustment costs that are normally studed n the context of trade expanson are those 2

3 welfare losses that arse n labour markets from temporary unemployment due to factor-prce rgdty or from costs ncurred through job search, re-locaton and re-tranng. Trade per se cannot be called a cause for adjustment costs. The sze and pattern of trade flows are not exogenous. Rather, they are shaped by underlyng factor endowments, demand patterns, technologes, ncome levels and polcy regmes of the tradng countres. The concept of trade-nduced changes therefore mplctly alludes to ulteror causes whch are manfested n the structure of trade flows. Ths concepton s easest to grasp n a settng of trade lberalsaton. There, any change that can be tracked to the change n the trade-polcy regme s defned as trade-nduced. The models of the new trade theory are consstent wth the smooth adjustment hypothess. Krugman (1981, p. 970), n a general equlbrum model consderng only prce adjustments (changes n relatve wages), found a one-for-one relatonshp between smlarty of factor endowments and ntra- ndustry trade, whch he nterpreted as a vndcaton [...] that ntrandustry trade poses fewer adjustment problems than nter-ndustry trade. Ths result effectvely stems from the fact that all the nfluental models explanng IIT through scale economes and monopolstc competton assume the products of an ndustry to be perfectly homogenous n terms of quanttatve and qualtatve factor requrements. Intra-ndustry adjustment costs are thus elmnated smply by assumpton. Unfortunately, the ndustry concept whch underles the emprcal measurement of IIT does not contan only goods wth perfectly dentcal producton requrements. For the analyss of emprcally observed IIT, there are three concevable reasons why IIT mght ental smaller labour-market adjustment costs than nter-ndustry trade: 1. the moblty of labour across frms and occupatons mght be greater wthn ndustres than between ndustres, 2. relatve wages mght be more flexble wthn ndustres than between ndustres, or 3. other producton factors mght be more moble wthn than between ndustres. The frst hypothess has much ntutve appeal. If one defnes IIT as the exchange of goods wth smlar producton requrements, then t s mpled that labour requrements are more smlar wthn ndustres than between ndustres. If the sklls acqured by the workers and managers of a contractng frm can be appled wthout much re-tranng n an expandng frm 3

4 of the same ndustry, then labour moblty may well be hgher wthn ndustres than between them. If specalsaton occurs nsde frms, then workers can smply be transferred from one department to another. Where ndustres are spatally concentrated, labour s lkely more moble wthn than between ndustres. The problem s that one cannot assume a pror that the statstcal product categores underlyng emprcal calculatons of IIT actually correspond to ths defnton of ndustres. The second hypothess relates to ntra-ndustry specfc-factors model where asymmetrc trade shocks across producers n one ndustry, combned wth short-term mmoblty of workers, wll result n temporary unemployment f wages are not flexble across producers. The man mpedments to wage flexblty are mnmum-wage legslaton and contractual wage agreements at the ndustry level. Snce such constrants generally apply at the level of the entre economy or of ndvdual ndustres, they mght actually be expected to allow greater wage flexblty between ndustres than wthn them. If wage nflexblty through ndustrywde centralsed barganng s the domnant cause of adjustment problems, then adjustment would be greater when trade shocks are ntra-ndustry than when trade alters the relatve compettveness among ndustres. The thrd hypothess, lke the frst, s plausble f we assume that ndustres are delneated accordng to supply-sde substtutablty of goods. In the specfc-factors model t s normally assumed that all producton factors except labour are non-transferable among products. If we relax ths assumpton and allow for ncreasng moblty of the complementary factors(s), then smaller wage adjustments wll be needed to restore labour-market equlbrum. Hence, the thrd hypothess offsets to some extent the caveat provded by the second hypothess. In sum, whle the SAH has not yet been rgorously embedded n a fully specfed theoretcal model, exstng trade theory does supply some arguments n ts favour. 1 Ultmately, however, the homogenety and adaptablty of ndustres defned n trade statstcs can only be determned through emprcal nvestgaton. Ths secton has been concerned wth the effect on an asymmetrc demand shock between or wthn ndustres on labour-market adjustment. The man focus of ths survey, however, s on how to measure the relevant degree of (a)symmetry of trade-nduced shocks n emprcal work. 4

5 3. Statc IIT: The Grubel-Lloyd Index By far the most wdely used measure of IIT s due to Grubel and Lloyd (1975), who suggested the followng formula: GL X M = 1, (1) ( X + M ) where X and M refer to a country s exports and mports of goods contaned n ndustry n one partcular year. Ths measure takes values between zero and one and ncreases n IIT. Furthermore, Grubel and Lloyd (1975) have represented the summary IIT ndex calculated over several ndustres as a trade-weghted (rather than smple arthmetc) average of the ndustry ndces: n GL = w GL = = 1 X + M ) GL ( X + M ) n = 1 ( n = 1 n = 1 = 1 = 1 n X M. (2) ( X + M ) Equvalent to ths defnton s the followng formula: GL = n = 1 ( X M ) 2* mn, n ( X + M ) = 1. (3) The GL ndex can be calculated for a country s world-wde trade or for a subset of trade partners, as well as for total merchandse trade or for a subset of ndustres. The statstcal propertes and lmtatons of the GL ndex have been carefully scrutnsed n the lterature. An authortatve survey can be found n Greenaway and Mlner (1986). I wll therefore gve only a bref summary of the four man measurement ssues. 1. Categorcal aggregaton. As seen above, the defnton of an ndustry central to the SAH and probably the most contentous ssue n appled IIT research. Grubel and Lloyd (1975, p. 86), have defned IIT as trade n dfferentated products whch are close substtutes. Over 5

6 tme, t has become generally accepted that the relevant crteron s substtutablty n producton (rather than n consumpton), snce ths s the aspect of ndustres that (a) dstngushes IIT from comparatve-advantage based trade and (b) les at the heart of the lnk between IIT and factor-market adjustment. Whlst statstcal product classfcatons are nevtably mperfect n ths respect, they are nevertheless largely guded by the correct crteron,.e. an effort to group together goods wth smlar nput requrements. 2 Ths stll leaves open the queston about the most approprate level of statstcal aggregaton for the calculaton of IIT ndces. Whlst the majorty of emprcal studes use data at the 3-dgt level, ths choce s mostly motvated by expedency rather than any a pror reason for favourng that level of aggregaton. 2. Adjustment for overall trade mbalance. The upper bound of a country s mean GL ndex s negatvely related to the share of the overall trade surplus or defct n total trade. Hence, mbalance n the trade account wll tend to bas the GL ndex downwards. Adjustment methods have therefore been suggested, wth the one brought forward by Aquno (1978) havng been most wdely used. The Aquno ndex s defned as follows: GL A Xˆ ˆ M = 1, (4) ( Xˆ + Mˆ ) where M = M * ({ X + M }/ 2 M ) and X = X * ({ X + M }/ 2 X ) ˆ = expected mports, ˆ = expected exports. The ratonale for such adjustment measures has been questoned on the grounds that vsble trade mbalances, both blateral and multlateral, may well be compatble wth balance of payments equlbrum (Greenaway and Mlner, 1986; Kol and Mennes, 1989; Vona, 1991). Gven the dffculty n estmatng equlbrum trade mbalances, the professonal consensus has been to work wth unadjusted GL ndces. 3. Scale nvarance. The GL ndex for an ndvdual ndustry s not related to the absolute sze of mports and exports n that sector, nor ndeed to the sze of the ndustry n terms of domestc producton or consumpton. What matters for studes of trade-nduced adjustment s not only the structure of trade flows, whch s captured by the GL ndex, but also the degree 6

7 of openness of ndvdual sectors. In regresson analyss t s therefore advsable to nteract the GL ndex wth a measure of sectoral trade openness or smple trade volumes. 4. Statc nature. The GL ndex refers to the pattern of trade n one year. In the context of structural adjustment, however, t s the structure of changes n trade patterns whch s mportant. Ths nsght, attrbutable to Hamlton and Knest (1991), has motvated the development of measures of margnal IIT (MIIT) and thus provdes the key ssue for ths survey Quas-Dynamc Measures: Changes n IIT Some measures have been developed that are nether statc nor measures of MIIT n the strct sense. These quas-dynamc measures take account of trade flows n two dfferent years, but, as I shall argue below, they may not present the most approprate gauge of trade patterns n the adjustment context. Frst-Dfferenced GL Indces Pror to the ntroducton of the MIIT concept, the evaluaton of IIT changes over tme was confned to the comparson of GL ndces for dfferent tme perods, where M X M X GL = GLt GLt n = 1 1, (5) ( M + X ) M + X t ( ) t n s the frst-dfference operator, t s the end year, and n s the number of years separatng the base and end years. 4 The nature and lmtatons of ths approach to measurng IIT are best llustrated n a smple dagram. Fgure 1A plots one country s mports and exports of one partcular ndustry (). 5 All ponts along any ray from the orgn share the same GL ndex, snce they represent equal sectoral mport-export proportons. The GL ndex equals 1 along the 45 degree lne, and zero along ether of the axes. 7

8 * FIGURE 1 ABOUT HERE * Assume that P represents the sectoral trade balance n the base year (t-n). In ths ntal year, home-country mports n ndustry exceed exports by a rato of 3:1. The ndustry thus exhbts a GL ndex of Assume further that the GL ndex s hgher n the end year (t). In Fgure 1A, both ponts Q1 and Q2, ft ths scenaro, snce they both correspond to a sectoral GL ndex of 0.8. Both a move from P to Q1 and a move from P to Q2 would thus show up as an ncrease n the GL ndex from 0.5 to 0.8. However, the pattern of trade change s qute dfferent between the two scenaros. Consder, frst, a shft from P to Q1. Q1 les n a 45- degree angle to the north-east of P. Exports and mports of have thus ncreased at the same absolute rate, and both countres (assumng there are only two) have captured an equal share of the ncreased volume of trade n ths sector. If ths pattern appears for other ndustres as well, then the adjustment process s ntra-ndustry, snce all countres share equally n the growth (or declne) of all these sectors. Now consder a move from P to Q2. In ths scenaro, the amount of home country exports has declned whle mports have ncreased. If ths pattern appears also n other ndustres - wth the home country not necessarly always on the losng sde - the adjustment process s nter-ndustry. A rse n the GL ndex can thus hde both a process of ntra- and nter-ndustry trade change. Thus, the juxtaposton of correspondng GL ndces for dfferent perods conveys some nformaton on the structure of trade n each of these tme perods, but t does not allow conclusons on the structure of the change n trade flows. 7 Ths, however, s not to say that ntertemporal analyss of correspondng GL ndces s useless or msleadng n tself. If the am of the analyss s comparatve statc, meanng that what s sought s a comparson of the structure of trade at dfferent ponts (years) n tme, then the comparson of GL ndces may well be adequate. It s only when the am of the analyss s dynamc n nature, meanng that the structure of the change n tradng patterns s to be scrutnsed, that the comparson of GL ndces s nadequate. Snce smple logc suggests that the costs of adjustment depend on the latter rather than on the former, an alternatve measurement method s warranted. 8

9 The Greenaway-Hne-Mlner-Ellott Measure Greenaway et al. (1994) have suggested the followng measure: GHME = [( X + M) X M ] t [( X + M) X M ] t n, (6) or: GHME = [( X + M) X M ]. (7) The GHME measure fundamentally dffers from the GL ndex n that t reports IIT n absolute values rather than as a rato. Ths feature can be desrable because t facltates the scalng of IIT relatve to gross trade levels, producton or sales n a partcular ndustry; whch n turn s useful for the econometrc analyss of the forces that determne structural adjustment. The drawback n ths s that the unscaled GHME measure says nothng about the proporton of (margnal) ntra- relatve to nter-ndustry trade and t lacks the presentatonal appeal of a smple bounded ndex. Hence, ts rason d être rests upon the fact that t can be related to correspondng levels of gross trade or real output (Greenaway et al., 1994, p. 424). 8 The GHME measure belongs to the quas-dynamc class, snce t corresponds to the dfference n the amounts of IIT n two perods, and t therefore shares lmtatons of the GL ndex for the assessment of the structure of change n tradng patterns. Hamlton and Knest s (1991) nsght on the GL ndex thus also apples to the GHME measure. Assume, for nstance, that over the perod of nvestgaton a partcular sector experences a shft from a trade defct to balanced trade whle exports reman unchanged. The GHME measure wll show a postve value of twce the ncrease n exports, even though ths s an obvous case of nter-ndustry adjustment, because the ncrease n exports s not matched by any correspondng ncrease n mports. Ths can be seen n Fgure 1B, whch gves a mappng of so-ghme contours: the nter-ndustry trade change represented by a change from P to Q2 n Fgure 1A would yeld a postve value of the GHME measure. 9

10 The Dxon-Menon Measures Dxon and Menon (1997) have developed two alternatve quas-dynamc measures. The frst measure captures base-year weghted percentage growth of IIT: IIT [( X + M ) X M ] DM = GL t n * 100, (8) [( X + M ) X M ] t n and the second measure captures the base-year weghted percentage growth of net trade: X M DM NT = ( 1 GL t n ) * 100. (9) X M t n These measures can take values from 100 to nfnty. An appealng feature s that these two measures add up to the percentage growth n total trade of the relevant ndustry. However, the Dxon-Menon measures belong to the quas-dynamc class, because they cannot consstently separate MIIT from margnal nter-ndustry trade. For llustraton, suppose agan that over the perod of nvestgaton a partcular sector experences a shft from a trade defct to balanced trade whle exports reman unchanged. DM IIT (DM NT ) wll yeld a postve (negatve) value, even though ths s an obvous case of nter-ndustry adjustment. A mappng of so-dm IIT contours looks dentcal to the one derved from the GHME measure (Fgure 1B): the nter-ndustry trade change represented by a change from P to Q2 n Fgure 1A would yeld a postve value of the DM IIT. 9 To summarse, the quas-dynamc measures are representatons of the change n the share or the amount of matched trade between two years, usng dfferent scalng yardstcks. They can thus be useful for an analyss of the evoluton of IIT over tme. However, these measures do not consstently relate to the degree of matchedness n trade changes,.e. they do not capture MIIT n the strct sense. 5. Margnal IIT: Matched Trade Changes Measures of MIIT quantfy the degree of ntra-sectoral symmetry of trade changes. Hence, they are computed from frst dfferences n exports and mports, X and M, that s they can 10

11 be unrelated to the level of trade or of IIT n ether the base of end perod. In a nutshell, MIIT s about the mportance of IIT n trade changes, and not about the change n IIT. The Hamlton-Knest Index The frst measure of MIIT was proposed by Hamlton and Knest (1991): HK = { X M M X for M > X = 0 for X > M = 0 (10) 1 for X = M > 0 undefned for M < 0 or X < 0. Ths measure s related strctly to the structure of the change n tradng patterns nformaton on levels of exports or mports s not requred. Hence, the HK ndex can be mapped onto a plane that s defned by X and M (Fgure 2A). The possblty of such a mappng s what dstngushes MIIT measures from IIT and quas-miit. * FIGURE 2 ABOUT HERE * Ths ndex, however, has some mportant lmtatons. Greenaway et al. (1994) have hghlghted that the fact of the HK ndex beng undefned when ether X or M has decreased can lead to a non-random omsson of a sgnfcant number of statstcal observatons and therefore to potentally msleadng results. Furthermore, Hamlton and Knest (1991) have nterpreted any stuaton where ther ndex s undefned as representng an ncrease n exports and a decrease n mports (or vce versa), whch ndcates nter-ndustry trade. Yet, the HK ndex s also undefned where both mports and exports have decreased (the bottom left quadrant of Fgure 2A), a stuaton n whch the matched decreases should be recorded as MIIT. 11

12 A Grubel-Lloyd Style Measure of MIIT The followng ndex s derved n Brülhart (1994): B A X M =1. (11) X + M Ths ndex, lke the GL coeffcent, vares between 0 and 1, where 0 ndcates margnal trade n the partcular ndustry to be completely of the nter-ndustry type, and 1 represents margnal trade to be entrely of the ntra-ndustry type. The man appeal of the B A ndex les n the fact that t reveals the structure of the change n mport and export flows, smlar to the HK ndex. Yet, unlke the latter measure, the B A coeffcent s defned n all cases and shares many famlar statstcal propertes of the GL ndex. Ths can be seen n the mappng gven n Fgure 2B. Olveras and Terra (1997) have shown that the statstcal propertes of the B A ndex dffer from those of the GL ndex n two partcular respects. Frst, ths ndex s not subject to an rsng downward bas as the level of statstcal dsaggregaton s ncreased. Second, there s no functonal relatonshp between the B A ndex for a certan perod and the B A ndces of consttuent sub-perods. Note that B A can be summed, lke the GL ndex, across ndustres of the same level of statstcal dsaggregaton by applyng the followng formula for a weghted average: B A tot = k = 1 w B A, where w = k = 1 X + M ( X + M ) and where B A tot s the weghted average of MIIT over all ndustres of the economy or over all the sub-ndustres of an ndustry, denoted by...k. (12) Extensons to the MIIT Index Several authors have put forward amended versons of the B A ndex, whch are talored to partcular underlyng assumptons on the nature of the adjustment problem. 12

13 Lloyd (1998) has argued that t may be useful n certan contexts to ncorporate local sales of foregn afflates ( nternatonal producton ) n an analyss of trade flows. He suggested that the B A ndex could be computed for ˆ j X = X and M ˆ = j= 1,2 j M j= 1,2, where agan denotes the ndustry, and j stands for the mode of supply. If a partcular flow s a cross-border mport or export n the tradtonal sense then j=1; and when we look at local sales of foregn afflates then j=2. Ths ndex could be useful for a study of the adjustment mplcatons of globalsaton n a broader sense, snce t can be decomposed nto the separate contrbutons to MIIT of changes n nternatonal goods trade and of changes n the pattern of nternatonal producton. 10 Another varant of the MIIT ndex was developed by Thom and McDowell (1999) to take account of the ncreasng fragmentaton of nternatonal producton: TM X M 1, (13) L X + M = L l= 1 l l= 1 l where l denotes sub-ndustres of. The TM ndex s bounded between zero and one, but t dffers from B A tot aggregated over sub-ndustres j of (equaton 12). The ratonale underlyng the TM ndex s that offsettng net trade changes across subsectors should be counted as MIIT f those subsectors are vertcally lnked. Ths s best llustrated wth a smple example. Assume that country A ncreases ts exports and reduces ts mports of fnshed watches vs-à-vs country B, and that A smultaneously reduces ts exports and ncreases ts mports of watch components. Furthermore, suppose that the two trade changes are of equal sze. If we apply the B A ndex to the ndustres fnshed watches and components separately, we dagnose zero MIIT. On the other hand, f we defne fnshed watches plus components as an ndustry, then the B A ndex s undefned, snce we observe zero aggregate trade change. The TM ndex, however, wll return a value of 1,.e. perfect MIIT, snce the two net changes at sub-ndustry level offset each other perfectly. In an applcaton to data on trade between the EU and some Eastern European countres, Thom and McDowell (1999) have found ther ndex to return sgnfcantly hgher values on average than B A. 13

14 The valdty of the TA ndex hnges on the approprate defnton of ndustres and subndustres. If one had a classfcaton wth sub-ndustres defned accordng to the stages of producton of the ndustry s fnal product, then the TA ndex provdes an elegant measure of the nternatonal fragmentaton process of producton. In the face of the untdy exstng statstcal classfcaton schemes, however, t s dffcult to state a pror whch mght be the approprate level of aggregaton, and whether one should prefer the TM or B A ndces. Annccharco and Qunter (2000) have suggested a thrd extenson of the MIIT ndex. They propose that the ndex should take a negatve sgn when the matched trade change s negatve, so that the ndex would range from 1 to 1: AQ = { -B A f M < 0 and X <0, B A otherwse. (14) Ths touches on an mportant pont. Underlyng the B A ndex s the mplct assumpton that the quantty of producton factors dsplaced by a one unt ncrease n mports (decrease n mports) s dentcal to the quantty of producton factors requred for a one unt ncrease n exports (decrease n mports). One corollary s that a matched ncrease n mports and exports has a zero net effect on factor demand at the ndustry level, and lkewse for a matched decrease n mports and exports. 11 Unless we are n the context of multple regresson, where one can control for trade-ndependent changes n sectoral demand and productvty, we may plausbly assume that matched expanson of trade wll be assocated wth growng sectors, whlst matched contracton of trade would be ndcatve of sectors that are n general declne. Hence, unless we can control for non-trade determnants of structural change, t appears plausble that the adjustment mplcatons of matched trade expanson dffer from those entaled by matched trade contracton, and the transformed ndex suggested by Annccharco and Qunter (2000) may well be nformatve n descrptve studes. Unscaled MIIT Measures There are undenable advantages n a smple bounded ndex for presentaton and nterpretaton. Yet, as ponted out by Greenaway et al. (1994), t can be useful n certan applcatons to have gross measures of MIIT, or to scale MIIT to producton varables. For that reason, the followng measure has been suggested by Brülhart (1994): 14

15 B C ( X + M ) X M =, (15) whch s strctly non-negatve and can be scaled even at the dsaggregated ndustry level, lke the GHME measure: B C V C B =, V (16) where V s any relevant scalng varable. Fgure 2C presents a mappng of so-b C contours n the trade-change space. Menon and Dxon (1997) have proposed a smlar measure. Instead of capturng absolute values of sectorally matched trade changes, lke B C, thers s a measure of unmatched changes n trade : MD UMCIT MD UMCIT MD UMCIT = X M. (17) and B C are closely related, as B C shows the absolute magntude of MIIT and shows the absolute magntude of margnal nter-ndustry trade. The relatve propertes of MD UMCIT and B C are easly grasped through a comparson of ther respectve mappngs n Fgures 2D and 2C. Absolute values of MIIT, such as B C and MD UMCIT, are dffcult to nterpret n solaton, snce they gve no ndcaton of the proporton between ntra- and nter-ndustry trade, whch, after all, s central to the defnton of the very concept of IIT and MIIT. Therefore, t mght be approprate for studes nvestgatng MIIT and adjustment to use a two-stage approach, where MIIT s expressed both n relaton to margnal nter-ndustry trade and n relaton to other scalng varables. 6. Margnal IIT and Sectoral Performance As noted above, a latent assumpton underlyng the basc MIIT ndex s that the adjustment costs of a net mprovement n a sectoral trade balance are dentcal to those of a net deteroraton n that sectoral trade balance. In other words, an addtonal mllon dollars of net 15

16 exports wll create a number of jobs that s equal to the number of jobs that would have been destroyed by an addtonal mllon dollars of net mports n that ndustry, and the adjustment costs of a job lost and of a job created are equal. In a labour market wth unemployment and on-the-job learnng, the assumed symmetry of adjustment costs between job creaton and job destructon s clearly unrealstc. MIIT ndces have therefore been developed to take account of the asymmetry between net mport growth and net export growth. For those reasons, the followng ndex was put forward n Brülhart (1994): B B where X M =, (18) X + M B B B A = 1. (19) Ths coeffcent can take values rangng between -1 and 1. It s two-dmensonal, contanng nformaton about both the proporton of MIIT and country-specfc sectoral performance. Frst, the closer B s to zero, the hgher s MIIT. B s equal to zero where margnal trade n the partcular ndustry s entrely of the ntra-ndustry type, whereas at both -1 and 1 t represents margnal trade to be entrely of the nter-ndustry type. Second, sectoral performance s defned as the change n exports and mports n relaton to each other, wth exports representng good domestc performance and mports reflectng weak domestc performance n a partcular sector. Thus defned, B B s drectly related to sectoral performance. When B > 0, X was > M. The opposte holds for B < 0. A mappng of B nto the trade-change plane s gven n Fgure 4A. * FIGURE 4 ABOUT HERE * Unlke the B A ndex, B cannot be aggregated meanngfully across ndustres, except where the Bs of all sub-ndustres have the same sgn. Snce hgh margnal nter-ndustry trade s expressed by values close to ether -1 or 1, the weghted average of two sub-ndustres mght yeld a value close to zero (hgh MIIT) even where hgh margnal nter-ndustry trade prevals n both of them. Therefore, B B cannot be used for summary statstcs resultng from calculatons at a dsaggregated level. 16

17 A related measure has been proposed by Azhar and Ellott (2001): X M AE =. (20) 2(max[ X, M ]) Ths ndex also ranges from 1 to 1, and t s negatve (postve) f the sectoral trade balance has deterorated (mproved) over the relevant tme nterval. The dfference les n the data range where X and M have opposte sgns,.e. where there s no MIIT. The B B ndex returns a value of 1 or 1 for all confguratons wthn that data range. The AE ndex, however, dfferentates between the relatve szes of opposng net trade changes. Ths means that the AE ndex provdes further detal n a data range where B always returns one of ts polar values. Ths addtonal nformaton conveyed by the AE ndex does not come at a cost, snce the nformaton contaned n B (-1,1), whch s the same data range as that for whch B A (0,1], s fully contaned n AE (-0.5,0.5). Hence, the nformaton conveyed by B s a subset of that gven by the AE ndex. The apparent advantage of the AE ndex also presents ts man dffculty: t s not clear how one should nterpret dfferent confguratons of pure margnal nter- ndustry trade,.e. what to nfer from dfferent ndex values n the ranges (-1,- 0.5) and (0.5, 1). There s no ready nterpretaton for ndex values n those ntervals. 7. Some Conjectures: Whch Measure s Best? The SAH has undenable ntutve appeal and s frmly establshed n the canon of nternatonal economcs. 12 As ths survey of measurement ssues shows, however, the seemngly straghtforward hypothess s mred n ambguty once one tres to defne t rgorously. I have argued that a measure of MIIT,.e. one that reflects the degree of ntrasectoral (a)symmetry n trade changes, should be preferred to statc or quas-dynamc measures of IIT when one seeks nformaton on the lkely mplcatons of trade changes for factor-market adjustment. Unfortunately, the ambguty does not stop here. There are now a number of dfferent measures whch capture dfferent aspects of the structure of trade changes. The problem s, of 17

18 course, that no one-dmensonal measure can ever fully descrbe the three-dmensonal dstrbuton of adjustment costs over the trade-change plane. Take, for example, the hypothetcal mappng of trade-nduced adjustment costs gven n Fgure 5. That partcular map of so-adjustment contours assumes that adjustment costs rse monotoncally as one moves away from the orgn of the Cartesan space,.e. as the combned sze of trade changes ncrease; but that ths rse does not occur at the same rate dependng on the drecton n whch one departs from the orgn. The debate about whch MIIT measure to use bols down to the queston about whch s the most mportant drecton of skewness n ths dstrbuton, departng from one that s symmetrc around the orgn. * FIGURE 5 ABOUT HERE * Fgure 5 s based on the followng lnear model of adjustment costs (AC) n a certan ndustry ( subscrpts mpled): AC= α X M + β( X M) + γ( X + M ), wth (21a) α>0, β<0, (21b) (21c) α > β, (21d) γ>0. (21e) The model (21a) s farly general, ts man restrcton beng that of lnearty - whch could be easly relaxed by addng non-lnear terms. The four restrctons that are then mposed on ths model to generate the mappng of Fgure 5 are rooted n assumptons that have been made, mostly mplctly, n the MIIT lterature. Restrcton (21b) reflects the assumpton that adjustment costs ncrease n the absolute amount of unmatched trade change, (21c) that export expanson (contracton) causes lower (hgher) adjustment costs than mport expanson 18

19 (contracton), (21d) that for gven volumes of trade changes adjustment costs are mnmsed where changes n mports and exports are of equal sze, and (21e) that a adjustment costs ncrease n the absolute amount of total trade change. Estmaton of a model lke (21a) mght shed some lght on the debate about the most approprate measure of low-adjustment-cost trade change. One could assess the valdty of the SAH assumptons by testng restrctons (21b-e). In partcular, the relevance of MIIT would be confrmed f the estmated α were sgnfcant, snce ths would ndcate that the degree of matchedness of trade changes wthn sectors matters for adjustment costs, as the varable X M s the Menon-Dxon (1997) measure of margnal nter-ndustry trade (MD UMCIT ). The relevance of the Grubel-Lloyd-style MIIT ndex (B A ) would be confrmed f α were large relatve to γ. Smlarly, a sgnfcant estmated β would ndcate that sectoral trade performance matters for adjustment costs, and that ndces such as B B of Brülhart (1994) or the Azhar-Ellott (2001) ndex are mportant. There s some emprcal evdence to support the clam that MIIT s relevant to labour-market adjustment costs, derved from a specfcaton that s smlar that n equaton (21a) (Brülhart, 2000). 13 However, ths queston stll deserves to be explored further. Two major challenges need to be addressed. Frst, explct estmaton of factor-market adjustment costs, rather than measures of structural change that are merely assumed to relate to the adjustment costs, has only recently begun to be appled to ths context (Haynes, Upward and Wrght, 2000; and Wrght, Haynes and Upward, 2001). Second, there stll does not exst a formal theoretcal model that can generate margnal ntra- and nter-ndustry trade, and thus serve as a base for the specfcaton of emprcal models. The choce of control varables n such exercses therefore stll lacks a coherent theoretcal base. 19

20 Bblography Annccharco, Barbara and Qunter, Benamno (2000) Aggregated Measures of Intra- Industry Trade: A Crtcal Comparson. Mmeo, CEIS Unversty of Rome Tor Vergata. Aquno, Antono (1978) Intra-Industry Trade and Intra-Industry Specalsaton as Concurrent Sources of Internatonal Trade n Manufactures. Weltwrtschaftlches Archv, vol. 114, pp Azhar, Abdul K., Ellott, Robert J.R. and Mlner, Chrs (1998) Statc and Dynamc Measurement of IIT and Adjustment: A Geometrc Reapprasal. Weltwrtschaftlches Archv, vol. 134, pp Azhar, Abdul K. and Ellott, Robert J.R. (2001) A Note on the Measurement of Trade- Induced Adjustment. Mmeo, Unversty of Manchester. Balassa, Bela (1985) Intra-Industry Specalsaton. European Economc Revew, vol. 30, pp Balassa, Bela and Bauwens, Luc (1987) Intra-Industry Specalsaton n a Mult-Country and Mult-Industry Framework. Economc Journal, vol. 97, pp Brülhart, Marus (1994) Margnal Intra-Industry Trade: Measurement and Relevance for the Pattern of Industral Adjustment. Weltwrtschaftlches Archv, vol. 130, pp Brülhart, Marus (1999) Margnal Intra-Industry Trade and Trade-Induced Adjustment: A Survey. In: Brülhart, M. and Hne, R.C., Intra-Industry Trade and Adjustment: The European Experence. Macmllan, London. Brülhart, Marus (2000) Dynamcs of Intrandustry Trade and Labor-Market Adjustment. Revew of Internatonal Economcs, vol. 8, pp Brülhart, Marus and Ellott, Robert (1998) Adjustment to the European Sngle Market: Inferences from Intra-Industry Trade Patterns. Journal of Economc Studes, vol. 25., pp Brülhart, Marus and Hne, Robert C. (1998) Intra-Industry Trade and Adjustment: The European Experence. Macmllan, London. Chrstodoulou, Mara (1992) Intra-Industry Trade n Agrofood Sectors: The Case of the EEC Meat Market. Appled Economcs, vol. 24, pp Dxon, Peter B. and Menon, Jayant (1995) Measures of Intra-Industry Trade as Indcators of Factor Market Dsrupton. Economc Record, vol. 73, pp

21 Drèze, Jacques (1961) Les exportatons ntra-c.e.e. en 1958 et la poston Belge. Recherches Économques de Louvan, vol. 27, pp Ellott, Robert J.R.; Greenaway, Davd and Hne, Robert C. (2000) Tests for Factor Homogenety and Industry Classfcaton. Weltwrtschaftlches Archv, forthcomng. Ether, Wlfred (1982) Natonal and Internatonal Returns to Scale n the Modern Theory of Internatonal Trade. Amercan Economc Revew, vol. 72, pp Greenaway, Davd and Hne, Robert C. (1991) Intra-Industry Specalzaton, Trade Expanson and Adjustment n the European Economc Space. Journal of Common Market Studes, vol. 24, pp Greenaway, Davd; Hne, Robert C.; Mlner, Chrs and Ellott, Robert (1994) Adjustment and the Measurement of Margnal Intra-Industry Trade. Weltwrtschaftlches Archv, vol. 130, pp Greenaway, Davd; Lloyd, Peter and Mlner, Chrs (1998) Intra-ndustry FDI and Trade Flows: New Measures of Globalsaton of Producton. GLM Research Paper, No. 98/5, Centre for Research on Globalsaton and Labour Markets, Unversty of Nottngham. Greenaway, Davd and Mlner, Chrs (1986) The Economcs of Intra-Industry Trade. Oxford, Basl Blackwell. Grubel, Herbert and Lloyd, Peter J. (1975) Intra-Industry Trade. Macmllan, London. Hamlton, Clve and Knest, Paul (1991) Trade Lberalsaton, Structural Adjustment and Intra-Industry Trade: A Note. Weltwrtschaftlches Archv, vol. 12, pp Haynes, Mchelle; Upward, Rchard and Wrght, Peter (2000) Smooth and Stcky Adjustment: A Comparatve Analyss of the US and UK. Revew of Internatonal Economcs, vol. 8, pp Kol, Jacob and Mennes, L.B.M. (1989) Correctons for Trade Imbalance: A Survey. Weltwrtschaftlches Archv, vol. 125, pp Krugman, Paul (1981) Intrandustry Specalzaton and the Gans from Trade. Journal of Poltcal Economy, vol. 89, pp Lttle, Jane Sneddon (1996) U.S. Regonal Trade wth Canada durng the Transton to Free Trade. New England Economc Revew, January 1996, pp Lloyd, Peter J. (1998) Globalsaton, Internatonal Factor Movements and Market Adjustments. CREDIT Research Paper, No. 98/7, Unversty of Nottngham. Lovely, Mary and Nelson, Doug (2000) On the Economc Relatonshp Between Margnal Intra-Industry Trade and Labour Adjustment n a Dvson of Labour Model. Revew of Internatonal Economcs, vol. 8, pp

22 Menon, Jayant and Dxon, Peter B. (1997) Intra-Industry versus Inter-Industry Trade: Relevance for Adjustment Costs. Weltwrtschaftlches Archv, vol. 133, pp OECD (1994) The OECD Jobs Study (Part I: Labour Market Trends and Underlyng Forces of Change). Pars. Olveras, Joaquín and Terra, Inés (1997) Margnal Intra-Industry Trade Index: The Perod and Aggregaton Choce. Weltwrtschaftlches Archv, vol. 133., pp Shelburne, Robert L. (1993) Changng Trade Patterns and the Intra-Industry Trade Index: A Note. Weltwrtschaftlches Archv, vol. 129, pp Thom, Rodney and McDowell, Moore (1999) Measurng Margnal Intra-Industry Trade. Weltwrtschaftlches Archv, vol. 135, pp Verdoorn, P.J. (1960) The Intra-Block Trade of Benelux. In: Robnson, E.A.G. (ed.) Economc Consequences of the Sze of Natons. Macmllan, London. Vona, Stefano (1991) On the Measurement of Intra-Industry Trade. Weltwrtschaftlches Archv, vol. 127, pp Wrght, Peter; Haynes, Mchelle and Upward, Rchard (2001) Estmatng the Wage Costs of Inter- and Intra-Sectoral Adjustment. CEPR Dscusson Paper, No

23 Fgure 1: Measures of IIT and Quas-MIIT (A) Grubel-Lloyd (1975) (B) Greenaway, Hne, Mlner, Ellott (1994) Dxon and Menon (1997): DM IIT Note: contour values depend on startng pont 23

24 Fgure 2: MIIT Measures (A) Hamlton and Knest (1991) (B) Brülhart (1994): A ndex (C) Brülhart (1994): C measure (D) Menon and Dxon (1997): UMCIT 24

25 Fgure 3: Measures of MIIT and Sectoral Performance (A) Brülhart (1994): B ndex (B) Azhar and Ellott (2001) Fgure 4: A Hypothetcal Map of Adjustment Costs and Trade Changes 25

26 Abstract: Economsts conventonally assume that ntra-ndustry trade (IIT) entals relatvely smooth factor-market adjustment to trade lberalsaton. However, the approprate defnton of IIT n the adjustment context contnues to be a matter of debate. A consensus s emergng that, n the context of adjustment, one should use measures that are based on margnal IIT, and a range of measures have been developed. These measures capture the degree of symmetry of changes n exports and mports at the sector level. In ths paper I gve a crtcal overvew of these measures. 1 Lovely and Nelson (2000) have explored changes n IIT n an Ether (1982) trade model. In that model, trade lberalsaton can yeld both changes n countres relatve specalsaton and changes n the sze of ndustres at the world level. In the conventonal understandng of the SAH, the latter effect s subsumed nto the ceters parbus assumpton, but the Lovely-Nelson (2000) analyss hghlghts the mportance of controllng for world - wde structural change - be t nduced by trade lberalsaton, technology or taste changes n emprcal analyses of the relatonshp between (M)IIT and factor-market adjustment. 2 In the lst of fve smlarty crtera used by the experts n charge of the thrd revson of the SITC code, for nstance, the frst prncple was the nature of the merchandse and the materals used n ts producton, whle the uses of the product only ranked thrd (Unted Natons, 1986, p. v). Evdence n favour of reasonable homogenety of statstcal sectors n terms of factor requrements has been found by Ellott, Greenaway and Hne (2000). Some researchers, ncludng Aquno (1978), Balassa (1985), Balassa and Bauwens (1987) and Chrstodoulou (1992), have re-arranged trade data nto groups that would seem more approprate n the IIT context. 3 A note on termnology. I refer to the GL ndex as a statc measure, and to MIIT as a dynamc concept. The GL ndex s calculated on the bass of cross-border flows of goods and s thus not a statc measure n the strctest sense. Yet, statc IIT n the sense of the GL ndex contrasts wth dynamc measures of MIIT snce the latter relate to the change n these flows between two dfferent years. 4 Industry subscrpts are mpled. Ths wll also be the case for all subsequent equatons, unless stated otherwse. 5 Ths graphcal representaton s orgnally due to Shelburne (1993) and has been developed as the trade box by Azhar, Mlner and Ellott (1998). 6 If the (constant) slope of a ray s defned as S = M / X, then the GL ndex on any pont along a partcular ray s gven by: GL = ( 1 S /{ 1 + S} ) 1. 7 These consderatons are supported by emprcal evdence n Lttle (1996, p. 16), who observed that regons wth rsng IIT tended to experence a relatvely large shft n the composton of ther exports, mports, or both. By contrast, regons wth declnes n IIT faced somewhat less structural change. These results suggest the need to re-examne the conventonal wsdom that ncreasng IIT automatcally smoothes adjustment to trade lberalsaton. 8 Greenaway et al. (1994) also ponted out the mportance of usng deflated trade values for the calculaton of M and X. Ths s true for all measures whch use frst dfferences of trade flows, hence t apples to all the MIIT measures dscussed below. 9 Note also that the frst measure s undefned f base-year IIT s zero, and the second measure s undefned where base-year net trade s zero. 10 Greenaway, Lloyd and Mlner (1998) have appled ths defnton of extended trade to compute GL ndces for the US and fve of ts major tradng partners. They found that two-way foregn-owned producton s sgnfcantly larger than arms -length IIT. 11 Another mplcaton s that the magntude of adjustment costs s unaffected by whether net trade changes are postve or negatve. Ths assumpton s relaxed n Secton For a lst of references to the SAH n the recent lterature, see Brülhart (1999). 13 The exercse of Brülhart (2000), whch s conducted on Irsh data and where adjustment costs are proxed by plant-level job turnover rates, ncludes among the ndependent varables the MIIT ndex (B A ) and a measure of trade ntensty ((X+M)/output). It does not consder measures of sectoral trade performance.