Trade and Migration to New Zealand *

Transcription

1 Trade and Mgraton to New Zealand * John Bryant a, Murat Genç b and Davd Law a a Treasury, PO Box 3724, Wellngton, New Zealand b Department of Economcs, Unversty of Otago, Dunedn, New Zealand Paper presented at the 45th Congress of the Regonal Scence Assocaton n Amsterdam, August 2005 Emal address for correspondence: mgenc@busness.otago.ac.nz Abstract Ths paper examnes the hypothess that a greater stock of mgrants n New Zealand from a partcular country leads to more trade between that country and New Zealand. The lterature suggests that mgrants can stmulate trade by lowerng transacton costs, and by brngng wth them preferences for goods produced n ther home country. We use panel data technques wthn the framework of a standard gravty model of trade. Our sample ncludes an average of over 170 countres for the years 1981 to Prevous studes of trade and mgraton have not dealt satsfactorly wth problems of unobserved heterogenety and selecton bas. We address these problems usng correlated random effects and selecton models. Results suggest that larger mgrant stocks are assocated wth hgher trade flows. J E L C L A S S I F I C A T I O N F00 Internatonal Economcs - General F10 Trade - General F14 Country and Industry Studes of Trade F22 Internatonal Mgraton K E Y W O R D S Mgraton; Internatonal Trade; Panel Data; New Zealand * We would lke to thank to Wllam Greene, Dean Hyslop, Jacques Poot, Jm Rose, Rchard Downng, Phlp Lu, members of Treasury s Appled Econometrcs Group, and our colleagues n the Polcy Coordnaton and Development Secton for ther contrbutons to ths paper. Ths paper has also benefted from comments receved n semnars gven at Canterbury and Otago Unverstes, The New Zealand Assocaton of Economsts conference, The New Zealand Econometrcs Study Group and the Australasan Meetng of the Econometrc Socety.

2 Trade and Mgraton to New Zealand 1 I ntroducton Mgrants have language sklls, local knowledge, and access to nternatonal networks that can help overcome barrers to trade. Increasng the number of mgrants from a country mght therefore be expected to stmulate trade wth that country. Econometrc analyses n the Unted States, Canada, Brtan, and Span fnd support for ths hypothess. One study even found a lnk between mgraton and trade for nternal mgraton between dfferent regons of France (Combes, Lafourcade and Mayer 2003). Such effects are potentally mportant n explanng New Zealand s recent trade performance, as the country s mgrant populaton has become sgnfcantly larger and more dverse over the past 20 years. Ths paper presents an econometrc analyss of the effects of mgraton on trade n New Zealand. It tests the hypothess that, all else equal, a larger stock of mgrants from a gven country ncreases New Zealand s mports from and exports to that country. As wth prevous studes of mgraton and trade, the startng pont s a gravty model of trade, n whch trade s a functon of the sze of two economes and the dstance between them. As well as lookng at the overall relatonshp between mgraton and trade, the paper also examnes whether the relatonshp between mgraton and trade dffers wth the nature of the goods traded, the mgrants orgn countres, and the number of mgrants. Prevous studes have been restrcted to merchandse trade. We also look brefly at the effect of mgraton on nternatonal toursm (whch n 2001 earned New Zealand an estmated $NZ5.1 bllon). 1 The analyss draws on an unusually rch dataset panel data on more 1 Ths fgure excludes nternatonal arfares and apples to the year to June. It was obtaned from the Internatonal Vstors Survey, on the Toursm New Zealand webste 1

3 than 170 countres over 21 years whch enhances our ablty to deal wth the econometrc problems of unobserved heterogenety and selecton bas. Secton 2 of the paper gves a bref overvew of recent trends n mgraton and trade n New Zealand. Secton 3 summarses deas from the nternatonal lterature on why mgraton mght stmulate trade. Secton 4 provdes some evdence on the actvtes of mgrants n New Zealand. Secton 5 summarses the results from prevous econometrc analyses. Secton 6 descrbes our methodology, and Secton 7 our results. The fnal secton summarses our fndngs and dscusses ther mplcatons. 2 Trends n mgraton and trade Between 1981 and 2001 the number of people usually resdent n New Zealand who stated on ther census forms that they were born overseas rose from 450,000 to 698,000, an ncrease of 55%. As Table 1 shows, the sources of mgrants also became more dverse, wth partcularly large ncreases n the number of mgrants from East Asa and the Pacfc. The data n Table 1 suggest that New Zealand s startng to develop mgrant communtes from an ncreasngly wde varety of countres. Detaled examnaton bears ths out. For nstance, the number of countres for whch New Zealand had at least one thousand mgrants ncreased from 28 n 1981 to 48 n 2001, and the number of countres for whch New Zealand had at least 10 thousand mgrants ncreased from 5 to Most analysts trace the dramatc changes n New Zealand s mgrant populaton to changes n mmgraton polcy. From the md-1980s offcal preferences for tradtonal mgrant sources were ended, and decsons were based mostly on personal characterstcs such as qualfcatons and age (Ldgard, Bedford and Goodwn 1998). In the case of some East Asan countres, rapd ncome growth n the mgrants home countres also presumably played a role. Ths provdes partal reassurance that mgraton was not respondng to trade per se, whch would bas upwards our estmates of the effect of mgraton on trade. 3 New Zealand s mports and exports have also grown substantally over the perod Table 1 presents estmates based on data from the Unted Natons Commodty Trade Database (Comtrade). Trade values n Comtrade are reported n nomnal US dollars; we 2 Bryant and Law (2004) contans a more detaled analyss of trends n New Zealand s foregn-born populaton. 3 For more dscusson on the drecton of causaton between mgraton and trade see Gould (1994: 310). 2

4 have converted these nto 1995 NZ dollars by multplyng by the NZ-US exchange rate, and then dvdng by Statstcs New Zealand s aggregate merchandse mport and export prce deflators. 4 Table 1 Changes n New Zealand s mgrant stock, exports, and mports, by regon, Populaton by regon of brth (thousands) Exports by regon (NZ$1995 mllons) Imports by regon (NZ$1995 mllons) Incr Incr Incr. Australa % 1,343 4, % 1,529 5, % East Asa & Pacfc % 2,688 9, % 2,611 7, % Europe & Central Asa % 1,317 2,612 98% 898 4, % Md East & N Afrca % % 661 1, % North Amerca % 1,337 4, % 1,721 4, % South Amerca % 212 1, % % South Asa % % % Sub-Saharan Afrca % % % Unted Kngdom % 1,166 1,221 5% 770 1,006 31% Unspecfed % 476 1, % % New Zealand 2,679 2,891 8% Total 3,143 3,737 19% 9,413 25, % 8,463 26, % Source Populaton data from Statstcs New Zealand unpublshed census tabulatons. Trade estmates calculated from the Unted Natons Statstcs Dvson s Comtrade database. The orgnal Comtrade data were denomnated n US dollars. See the text for a descrpton of the converson to NZ dollars. As wth mgraton, there s substantal geographc varaton n growth rates. Trade wth the Unted Kngdom, for nstance, has ncreased relatvely lttle, whle trade wth South Amerca has ncreased markedly. New Zealand has ncreased the number of countres wth whch t conducts substantal nternatonal trade. Between 1981 and 2001, the number of countres from whch New Zealand mported goods worth at least $100 mllon (n 1995 NZ dollars) ncreased from 13 to 31. Durng the same perod, the number of countres to whch New Zealand exported goods worth at least $100 mllon ncreased from 20 to We calculated annual exchange rates by averagng the Internatonal Monetary Fund s monthly rates. It would have been preferable to have used country-specfc mport and export prce deflators. However, country-specfc deflators are avalable only for New Zealand s top fve tradng partners. 3

5 3 M e c hansms through whch mgraton could stmulate trade Followng Gould (1994), most authors postulate two mechansms through whch mgraton could stmulate trade between the host and orgn countres: transacton cost effects, and mmgrant preference effects. 3.1 Transacton cost effects Mgrants are expected to stmulate trade by lowerng transacton costs. There are two related sets of reasons why mmgrants mght face lower transacton costs for trade wth ther country of orgn. The frst s that mmgrants have superor knowledge of home country markets, languages, busness practces, laws, and other matters related to trade. The second s that mgrants may be able to partcpate n nternatonal networks, as exemplfed by the networks of ethnc Chnese (Rauch and Trndade 2002). These networks can be conduts of nformaton, and can deter opportunstc behavour. Transacton cost effects are generally expected to stmulate both exports and mports. Most authors argue that mgrants nformatonal advantages are more mportant for dfferentated goods than for homogenous goods, because of the greater nformaton problems nvolved n the trade of dfferentated goods. Most authors also argue that the trade-stmulatng effect of mgraton s greatest when the host and orgn countres have very dfferent nsttutons, languages and cultures, and when alternatve sources of nformaton and contract enforcement are lackng, snce ths s when the specal sklls or mgrants are most needed. 3.2 Immgrant preference effects Immgrants are assumed to demand certan goods produced n ther home countres, or smlar to those produced n ther home countres. These preferences are expected to boost mports to the host country but not exports from the host country. The effect s assumed to be more marked for dfferentated goods than for homogenous goods. Some authors note that there may be a countervalng mmgrant substtuton effect. If there are suffcent mmgrants n a country, these mmgrants may begn to produce goods themselves rather than mportng them (Dunlevey and Hutchson 1999, Grma and Yu 2000). 4

6 4 Evdence on the actvtes of mgrants n New Zealand Case studes n many countres show that mmgrants typcally have found tradng actvtes an accessble nche to fll n the labor market (Gould 1994: 302), whch s to be expected f mgrants enjoy a transacton cost advantage. There s some survey evdence that mgrants to New Zealand also use ther backgrounds to facltate nternatonal trade, as employees or as busness owners. Watts and Trln (1999) sent questonnares to 460 companes that they dentfed as beng nvolved n foregn markets, and receved reples from 187 (41%). Of these frms, 70% employed mgrants from non-englsh speakng countres. Some, but not all, of the frms that employed mgrants from non-englsh speakng countres used ther employees lngustc sklls and local knowledge. For nstance, 58% used mgrant employees for assstng clents vstng NZ and 42% for translatng documents ; 29% ndcated that they made use of the overseas contacts and networks of ther employees (Watts and Trln 1999: 123-5). Watts and Trln also sent questonnares to 156 mmgrants from non-englsh speakng countres. Usable reples were receved from 52 people (33%). Of the 52 people who repled, 41 were employed. Of these, 14 sad that ther cultural backgrounds were relevant to ther work actvtes and seven reported use of ther busness connectons n ther employment. Over three-quarters of mmgrants surveyed were strongly of the opnon that ther natve-speaker sklls n languages other than Englsh and ther cultural background could be used to better advantage n New Zealand. Some respondents stated that ther sklls were under-used because of prejudce and dscrmnaton (Watts and Trln 1999: 124-8). In 2002, the New Zealand Immgraton Servce surveyed Long-Term Busness Vsa and Investment Category mgrants whose applcatons had been approved n 1999 and 2000 (New Zealand Immgraton Servce 2002). The Immgraton Servce attempted to locate 823 people, but made contact wth only 84. Of the 59 busnesses run by Long-Term Busness Vsa mgrants, 43% were nvolved n mportng, and 33% n exportng. 5 Thrteen out of 25 Investment Category mgrants owned busnesses. One of these busnesses was nvolved n mportng and fve n exportng. The report states that n most cases, nvestors 5

7 were exportng to ther country of brth (New Zealand Immgraton Servce 2002: 106). The low response rates and small sample szes of these surveys mean that ther results need to be treated cautously. They do, nevertheless, suggest that the transacton cost theory, and the mages of mgrants nvolvng themselves n nternatonal trade, have some emprcal groundng. 5 P r e vous econometrc tests of the effects of m graton on trade Prevous econometrc tests of the effect of mgraton on trade have, lke ours, been based on a gravty model of trade. The gravty model has been hghly successful n descrbng emprcal patterns of nternatonal trade (Frankel 1997). It can be derved n a number of dfferent ways. Rauch (1999) and Head and Res (1999) provde an ntutvely appealng dervaton, whch we summarse here. The dervaton starts from a proposton about the pattern of trade n a frctonless world. Let m be the value of New Zealand s mports from country (the expresson for New Zealand s exports to country s exactly symmetrcal). Lety NZ, y, and y w be the GDPs of New Zealand, country, and the world. In the absence of transport or transacton costs (and wth some addtonal assumptons about product dfferentaton and preferences) New Zealand consumes the output of country n proporton to New Zealand s share of world output ynz y W, so that m y NZ = yw y. (1) In practce, transport costs, tarffs, and transacton costs nduce departures from ths pattern. These effects are modelled by applyng an adjustment factor exp ( ) rght hand sde of Equaton 1, where Xβ to the X s a vector that ncludes a range of varables attemptng to capture transport and transacton costs. Takng logs yelds the equaton lnm = lny + lny lny + Xβ. (2) NZ W 5 Calculated from data gven n (New Zealand Immgraton Servce 2002: Tables 8.18, 8.19). The calculatons exclude the 11 busnesses for whch nformaton on mportng s not avalable, and the 12 for whch nformaton on exportng s not avalable. 6

8 One varable that s almost always ncluded s the dstance between the two countres. The dstance varable tres to measure transport and communcaton costs, but t s generally beleved to pck up cultural, nsttutonal, and lngustc dfferences as well. Transport and communcaton costs have fallen over tme, but t s not clear how the mpact of culture, nsttutons, and language have changed. Because the effect of dstance s not a central concern of the paper, we decded to follow tradton and assume that the effect of dstance s constant over tme (though the use of tme dummes, descrbed below, provdes some protecton aganst any bases ntroduced by changes n the effect of dstance.) Other frequently used varables nclude ol prces, real exchange rates, common languages, common borders, membershp of trade blocs, and colonal tes. It has become standard to also nclude a populaton or a GDP per capta varable, to allow for effects such as subsstence thresholds or self-suffcency. For studes of mgraton and trade, the key varable n X s one measurng the number of mgrants from each potental tradng partner lvng n the country of nterest. In prncple, a varable measurng the number of mgrants from the country of nterest lvng n each potental trade partner country should also be used. The necessary data are, however, dffcult to obtan. The only study to nclude such a varable s one on overseas Chnese (Rauch and Trndade 2002). Table 2 summarses results from the nne prevous econometrc studes of mgraton and trade that we have located. The studes cover fve host countres the Unted States, Canada, the Unted Kngdom, Span, and France and varous tradng partners, though n the case of Combes et al (2003) the trade n queston s between dfferent regons of France. Dunlevy and Hutchson (1999, 2001) use data from 1870 to 1910; all the other studes use more recent data. Some studes use data from a sngle perod, whle others use tme seres technques to combne data from several perods. Some studes ft the model n ts orgnal multplcatve form usng non-lnear statstcal models, and others take logs of both sdes and use lnear models. The export and mport elastctes n Table 2 show the extent to whch an ncrease n the sze of the mmgrant stock ncreases trade. The elastctes derved by Gould (1994), for nstance, mply that, all else equal, a 1% ncrease n the number of mmgrants resdent n a country would ncrease exports from that country by 0.02% and ncrease mports to that country by 0.01%. In cases where several specfcatons are presented, elastctes from the 7

9 authors man or preferred elastcty are cted. Wherever possble, average elastctes across all goods and all trade partners are shown. Table 2 The effect of mgraton on exports and mports, 9 studes Study Sample Export elastcty Import elastcty Gould (1994) US and 47 trade partners; Head and Res (1998) Canada and 136 trade partners; Dunlevy and Hutchnson (1999, 2001) US and 17 trade partners; Grma and Yu (2000) UK and 48 trade partners; Combes et al (2002) 95 French Departments; Rauch and Trndade (2002) 63 Countres; 1980, /0.47 a 0.21/0.47 a Wagner, Head, and Res (2002) 5 Canadan regons and 160 foregn countres; Blanes-Crstobal (2003) Span and 40 trade partners, Chng and Chen (2000) Canada and Tawan b 0.30 b a The estmate of 0.21 apples to homogenous goods, and 0.47 to dfferentated goods; nsuffcent data were ncluded n the artcle to allow the calculaton of an overall elastcty. No dstncton s made between mports and exports. b Export elastcty refers to exports from Canada to Tawan, mport elastcty refers to exports from Tawan to Canada. Notes Rows 1-6 are based on Table 1 n Wagner, Head, and Res (2002). The elastctes for Gould (1994) and Rauch and Trndade (2002) were calculated by Wagner et al. The elastctes for Grma and Yu (2000) and Chng and Chen (2000) were calculated by the present authors. As s apparent from Table 2, most studes fnd some relatonshp between mgraton and trade, n the expected drecton, though the magntudes of the estmated effects vary greatly. All the studes shown n Table 2 also nvestgate how the relatonshp between mgraton and trade vares across goods, countres or the type of mgrant. Gould (1994) and Dunlevy and Hutchnson (1999, 2001), for nstance, fnd that the effect of mgrants s stronger for consumer goods than producer goods. Rauch and Trndade (2002) and Wagner, Head, and Res (2002) fnd that the effect s stronger for dfferentated goods than for homogenous goods. Grma and Yu (2000) fnd that the effect s stronger when there are no colonal tes; Blanes-Crstobal (2003) obtans the opposte result. Chng and Chen (2000) fnd that the effect s stronger for entrepreneur rather than passve nvestment type mgrants. Gould (1994: 307) and Wagner, Head, and Res (2002: ) experment wth alternatve specfcatons n whch the elastcty of trade wth respect to mgraton changes as the number of mgrants ncreases. They fnd that the elastcty decreases wth the number of mgrants. Ths s a very strong form of dmnshng returns. Dmnshng returns n the 8

10 ordnary sense of each mgrant contrbutng less than the one before s already possble under the constant-elastcty specfcaton 6. All the studes summarsed n Table 2 looked at trade n goods rather than servces. We know of no studes that have looked at the effect of mgrant stocks on exports of servces, even though mgraton could plausbly lower transacton costs for trade n servces n the same way that t lowers costs for trade n goods. 6 M e thodology 6.1 Data We have assembled data for a large panel of countres for every year from 1981 to The mnmum number of countres ncluded n the panel n our benchmark specfcatons s 171; the maxmum s 179. We nclude substantally more observatons than any prevous studes of trade and mgraton, wth the excepton of Wagner, Head, and Res (2002). As dscussed below, the reason for assemblng a large panel dataset s to address problems of unobserved heterogenety and selecton bas. Our data on mports and exports come from the Unted Natons Statstcs Dvson s Comtrade Database. The UN obtans estmates of New Zealand mports and exports from Statstcs New Zealand. We treat the data as complete. If no trade s reported between New Zealand and a gven country n a gven year, we assume that the true value for that year was zero. Estmates of the foregn-born populaton n New Zealand come from unpublshed tabulatons prepared by Statstcs New Zealand usng data from the 1981, 1986, 1991, 1996, and 2001 Censuses. To calculate exact values for the nter-censal years t would be necessary to have data on deaths and nternatonal movements by place of brth, whch are not avalable. An alternatve would be to nterpolate. We decded, however, to use the total from the most recent Census for the whole nter-censal perod, so that, for nstance, mgrant stock n years 1981, 1982, 1983, 1984, and 1985 s set equal to the Census estmate n The advantage of ths method, besdes ts smplcty, s that t gves 6 Let m be trade and x mgraton. There are dmnshng returns when m < 0. But f lnm = βlnx, then m = x β, and m = β(β-1) x β < 0, provded β 0 and x > 0. 9

11 partal protecton aganst the possblty that mgraton s respondng (n the short term) to trade. Data on New Zealand s GDP and populaton come from the World Bank s World Development Indcators database. Data on language come from Grmes (1996), and dstance from New Zealand from the webste Great Crcle Dstances Between Captal Ctes Unobserved heterogenety 8 The varables avalable to us cannot possbly capture all nfluences on New Zealand s trade. In other words, there s lkely to be unobserved heterogenety across our sample. Applyng ordnary cross-sectonal technques n the presence of unobserved heterogenety can lead to ncorrect standard errors and based coeffcent estmates. Use of panel data, however, permts models of the form where lnm = β lny + β lny + β lny + X β + α + u (3) t ynz NZ, t y t yw W, t t t u t s a tme-varyng dosyncratc error, and α s an unobserved country-specfc effect that represents the permanent cross-country heterogenety. If the α are assumed to be uncorrelated wth the explanatory varables, then Equaton 3 can be estmated usng a Random Effects approach. The assumpton of zero correlaton s, however, dffcult to justfy n our case. No such assumpton s requred under a Fxed Effects approach. Under Fxed Effects, however, t s not possble to obtan coeffcents for varables that are constant over tme, such as Language and Dstance. Prevous econometrc studes of mgraton and trade have used ether ordnary crosssectonal technques or Fxed Effects. There s, however, an alternatve approach, referred to as Correlated Random Effects, that avods the zero correlaton assumpton and allows the ncluson of varables that are fxed over tme. Under Correlated Random Effects, the correlaton between the country-specfc fxed effect α and the explanatory varables s explctly modelled usng the expresson α X λ X λ X λ η (4) = T T + 7 Avalable at: 8 Ths secton draws heavly on unpublshed lecture notes by Dean Hyslop. 10

12 where the λ t are vectors of projecton coeffcents and η s a true random effect that s uncorrelated wth the explanatory varables. perods, so that We assgn the same weght to all tme and λ1= λ2 =... = λt = λ, (5) α = TXλ + η. (6) Substtutng ths expresson nto Equaton 3 (and absorbng T, a constant, nto λ ) gves lnm = β lny + β lny + β lny + X β+ Xλ + η + u, (7) t ynz NZ, t y t yw W, t t t whch can be estmated usng Random Effects. Some unobserved heterogenety also potentally takes the form of shocks affectng New Zealand s trade wth all countres more or less equally at the same tme. An mportant example s the trade lberalsaton that New Zealand began n the md-1980s (Evans, Grmes, Wlknson and Teece 1996). We allow for such affects by addng a tme dummy for the perod 1995 to 2001 to all equatons. 6.3 Sample selecton Equaton 7 does not allow for zero trade. In practce, however, 29% of our observatons for mports are zeros, as are 20% of our observatons for exports. Followng prevous studes of mgraton and trade, we nterpret the zeros to mean that observed trade values emerge from a two-step process. Countres n effect decde whether to trade, and then decde how much to trade (Head and Res 1998; Dunlevy and Hutchnson 1999: fn20; Wagner, Head, and Res 2002: 518). Our model s z = β lny + β lny + β lny + X β + Xλ + η + u (8a) t ynz NZ, t y t yw W, t t t z t 0, zt < 0 = 1, zt 0 (8b) lnm t 0, z= 0 = β lny + β lny + β ln y + X β + Xλ + η + u, z= ynz NZ, t y t yw W, t t t (9) 11

13 We assume that 1 2 t u ~ N(0, σ ) and are non-autocorrelated, and that u 0 t ~ N (0,1), wth a correlaton, ρ, between the two that may not equal zero. Equatons 8a and 8b together make up the selecton equaton, whle Equaton 9 s the trade equaton. If the correlaton between the two error terms s not zero, then smply usng Equaton 7 on the sub-sample wth non-zero trade wll lead to based estmates. As explaned n LIMDEP (Verson 8.0), ths sample selecton model can be estmated as a 1 0 random parameters model by treatng η and η as random coeffcents. We adopt ths approach and estmate the model n two steps by maxmum smulated lkelhood method. We frst ft a random parameters probt model and store the results for use n the next phase where we ft the trade equaton. Followng prevous studes, we use the log of mgrant numbers n X t. In some cases, however, the number of mgrants equals zero, so that the log s undefned. Smply omttng these cases could potentally create a selecton bas. We therefore adopt an approach used by Wagner, Head, and Res (2002). We ntroduce a dummy varable called Zero Mgrants that takes a value of one when there are no mgrants, and zero otherwse. We set our Mgrant Stock varable equal to zero when there are no mgrants, and the log of the number of mgrants otherwse. The Zero Mgrants varable shows the change n trade that occurs when New Zealand has exactly one mgrant from a country rather than none. In prncple, t should be close to zero. 6.4 Dfferences n the characterstcs of goods and countres Theory suggests, and emprcal studes largely confrm, that the effect of mgraton on trade vares wth the goods beng traded and the countres nvolved. We ntend to carry out detaled analyses of dfferences between types of goods n future work. In the present paper, we smply look at the effect of excludng certan goods for whch the effect of mgrants s lkely to be small. We re-estmate our results usng exports excludng agrculture. The reason for excludng agrculture s that most of New Zealand s nternatonal agrcultural exports are channelled through a few large companes, whch presumably are large enough to recrut nternatonally f they cannot draw on local mgrants. We also re-estmate our results usng mports excludng ol, on the grounds that mports are channelled through a few large companes, and also because petroleum products are homogeneous goods whch pose fewer of the transactonal dffcultes that 12

14 mgrants are expected to allevate. Our expectaton s that the coeffcents on the mgraton varables should be larger n the specfcatons excludng agrculture and ol than they are n the benchmark specfcatons. As wth prevous studes, we hypothesse that mgrants have a stronger effect on trade when they come from a non-englsh-speakng country, because the mgrants language sklls are then needed, and because language proxes for cultural and nsttutonal dfferences from New Zealand. We test for such effects by nteractng the mgrant varable wth the language varable. We also hypothesse that mgrants have a stronger effect when they come from a low-ncome country (havng controlled for the sze of the countres GDPs), snce low ncome proxes for cultural and nsttutonal dfferences, and for dffcultes n obtanng nformaton and enforcng contracts. We test for ths by nteractng the mgrant varable wth a low-ncome varable. 6.5 Changes n elastcty wth the sze of the mgraton stock We examne how the sze of the mgrant stock affects the elastcty of trade wth respect to mgraton. We do ths by addng the square of our mgrant varable to the regressons. Ths s equvalent to assumng that the elastcty of trade wth respect to mgraton declnes lnearly wth the log of the number of mgrants. Ths assumpton s somewhat arbtrary and has the unrealstc mplcaton that the elastcty wll eventually become negatve n many cases. As dscussed n Secton 4, Gould (1994: 307) and Wagner, Head, and Res (2002: ) use alternatve, more complcated, expressons. Gould s coeffcents are, however, dffcult to nterpret. The coeffcents n Wagner et al s expresson have a clear nterpretaton, but t s not feasble to estmate them as part of a selecton model. 6.6 Mgrant stocks and toursm As dscussed n Secton 4, prevous studes of the effect of mgraton on trade have looked exclusvely at merchandse trade. We examne the effect of stocks on an mportant component of the nternatonal servces trade: toursm. Ideally, we would lke to use data on expendture by overseas vstors. Unfortunately, such data are only avalable for a small subset of orgn countres. Comprehensve data are, however, avalable on the number of vstors arrvng from each country. We therefore use vstor numbers to proxy for expendtures. Most vsts to New Zealand are for toursm or 13

15 smlar purposes: n the year to March 2004, 51% of vstors stated that ther reason for vstng was toursm/holday and a further 28% stated that t was to vst frends/relatves. 9 Census respondents are recorded as usually resdent, and hence ncluded n our foregnborn measures, only f they answer yes to a queston askng whether they lve n New Zealand. Some short-term vstors may, however, msnterpret the queston and be napproprately ncluded. Ths would bas upwards our estmates of the effect of mgraton on vstor flows. The help sheets accompanyng the 1996 and 2001 Census forms explctly stated people should not answer yes to the resdency queston unless they were n New Zealand for more than a year. We re-run our model usng data from the perod alone to see whether ths affects our results. 6.7 Addtonal robustness testng To examne the senstvty of our results to the sample chosen, we run the model on several dfferent sub-samples of countres. We splt the sample nto Englsh and non-englsh speakng countres and hgh ncome and low ncome countres. We also examne the effect of smply omttng countres wth no mgrants, rather than usng the Zero Mgrant varable. We have not ncluded a real exchange rate varable n most of our regressons, as the necessary data are avalable for only about half of our sample. To assess whether the omsson of the exchange rate varable s lkely to have affected our results, we apply our benchmark specfcaton to the sub-sample, and then recalculate usng the real exchange rate varable. 6.8 Varables Table 3 summarses the varables. The sources of our data are dscussed n Secton Estmates taken from Statstcs New Zealand External Mgraton (March 2004) - Hot Off The Press, from the Statstcs New Zealand webste 14

16 Table 3 Varables used n the models Varable name Mgrant Stock Zero Mgrants Foregn GDP New Zealand GDP World GDP Populaton Dstance Non-Englsh Average Mgrant Stock Average Foregn GDP Average Populaton Defnton Log of the number of mgrants at the tme of the most recent Census. Dummy varable takng a value of one f there are no mgrants from the country Log of a foregn country s GDP (n 1995 $US) Log of New Zealand s GDP (n 1995 $US) Log of World s GDP Log of a foregn country s populaton Log of the dstance between the foregn country s captal and Wellngton A dummy varable takng a value of one f Englsh s not wdely spoken n the country. The average value over tme of the Mgrant Stock varable The average value over tme of the Foregn GDP varable The average value over tme of the Populaton varable 1995 Dummy A dummy varable takng a value of one f the year s between 1995 and 2001 Square of Mgrant Stock Mgrant Stock x Low ncome Mgrant Stock x Non- Englsh The square of the Mgrant Stock varable. Note that the varable s squared after takng logs, not before. Equal to Mgrant Stock f the country s classfed as low or mddle ncome by the World Bank, and zero otherwse Equal to Mgrant Stock f Non-Englsh equals one, and zero otherwse 7 R e sults All results presented throughout ths secton are generated usng Correlated Random Effects models. A constant term and country specfc effects are ncluded n all specfcatons. We do not, however, present the coeffcents from these. We use one, two and three stars (*) to denote sgnfcance at the 10%, 5% and 1% level respectvely. P- values on many of our varables fluctuate between 0.01 and Benchmark results In Table 4 we report our benchmark results. As descrbed n Secton 6, the varables Average Mgrant Stock, Average GDP and Average Populaton capture correlatons between the explanatory varables and country-specfc effects. In the selecton equaton for exports the estmated coeffcents on Zero Mgrants, Foregn GDP, New Zealand GDP, World GDP and Non-Englsh are all postve and hghly sgnfcant, ndcatng that, all else equal, hgher values for these varables mply a hgher 15

17 probablty that trade between New Zealand and a gven country takes place. The sze of the ncrement n the probablty depends on the country s characterstcs. As dscussed n Secton 6.3, we had expected the coeffcent on Zero Mgrants to be close to zero. However, the fact that t s not probably reveals more about the dosyncrases of the countres wth zero reported mgrants than t does about the relatonshp between mgraton and trade. The estmated coeffcents on Populaton, Dstance and the 1995 Dummy are all negatve and hghly sgnfcant ndcatng that, all else equal, hgher values for these varables would on average result n a lower probablty that trade between New Zealand and a gven country takes place. The coeffcent estmate on Mgrant Stock, the varable of most nterest to us n ths study, s postve and hghly sgnfcant. In the trade equaton for exports the estmated coeffcents on Foregn GDP, New Zealand GDP, World GDP and the 1995 Dummy are postve whle the estmated coeffcents on Zero Mgrants, Populaton, Dstance and Non-Englsh are negatve. As Foregn GDP, New Zealand GDP, World GDP, Populaton and Dstance are n logs the estmated coeffcents assocated wth these varables are smple elastctes. The coeffcent on Foregn GDP of for example mples that, all else equal, ncreasng a country s GDP by 1% would lead to a 0.95% ncrease n exports to that country. For dummy varables such as our 1995 Dummy, a coeffcent value of β mples that, all else equal, exports to that country wll be β % hgher when the dummy varable equals one. 10 The coeffcent of on our 1995 Dummy mples that, all else equal, New Zealand would have exported approxmately 14% more to any gven country n the perod between 1995 and 2001 than n the perod between 1981 and The estmated coeffcent on Mgrant Stock s hghly sgnfcant. It mples that on average a 1% ncrease n the stock of mgrants from a gven country would result n an ncrease n exports to that country of around 0.09%. In the selecton equaton for mports the estmated coeffcent on Mgrant stock suggests that ncreasng the number of mgrants from a gven country wll, all else equal, ncrease the probablty that New Zealand mports from that country. Mgrant Stock s statstcally sgnfcant at the 10 percent level. 10 Let m1 be predcted exports when the dummy varable equals 1, and m0 predcted exports when the dummy varable equals 0. Then lnm1 - lnm0 = β, m1 / m1 1 = e β 1 (1 + β) 1 = β. Ths approxmaton ceases to be accurate f the absolute value of β s large. 16

18 Table 4 Benchmark results Varable Exports Imports Selecton Trade Selecton Trade Mgrant Stock *** (0.0365) *** (0.0165) * (0.0358) *** (0.023) Zero Mgrants *** (0.0882) (0.0477) *** (0.0875) (0.0767) Foregn GDP *** (0.0935) *** (0.0496) *** (0.0894) *** (0.0725) New Zealand GDP *** (0.8156) (0.3518) *** (0.7234) 1.97*** (0.4778) World GDP *** (0.4859) * (0.2103) (0.4217) ** (0.2812) Populaton *** (0.2574) *** (0.109) (0.2269) *** (0.1358) Dstance *** (0.1483) *** (0.0345) *** (0.129) *** (0.0429) Non-Englsh *** (0.0672) *** (0.03) 1.23*** (0.0628) *** (0.0393) Average Mgrant Stock *** (0.0359) *** (0.0174) *** (0.0389) *** (0.0243) Average Foregn GDP *** (0.0995) *** (0.0509) *** (0.0953) * (0.0735) Average Populaton *** (0.2552) *** (0.1092) (0.2294) * (0.1368) 1995 Dummy *** (0.119) *** (0.0479) (0.1097) (0.0632) Log Lkelhood Observatons Countres Notes For defntons of the varables refer to Table 3. A constant term and country specfc effects are ncluded n all regressons. Dependent varables are n 1995 New Zealand dollars. Three stars (***) ndcates that the coeffcent s sgnfcantly dfferent from zero at the 1% sgnfcance level, two stars (**) ndcates that t s sgnfcant at the 5% level, and one star (*) ndcates that t s sgnfcant at the 10% level. In the trade equaton for mports the estmated coeffcent on Mgrant Stock s hghly sgnfcant and mples that, on average, a 1% ncrease n the stock of mgrants from a gven country would result n an ncrease n mports from that country of around 0.15%. In Secton 6.4 we hypothesse that the effect of mgrants on trade may vary dependng on the type of good traded. To nvestgate ths we present n Table 5 our results when excludng agrculture from exports and ol from mports. 17

19 Table 5 Benchmark results, excludng agrculture and ol Varable Exports (Excludng Agrculture) Imports (Excludng Ol) Selecton Trade Selecton Trade Mgrant Stock *** (0.0343) *** (0.0251) ** (0.0368) *** (0.0187) Zero Mgrants *** (0.0796) *** (0.0829) *** (0.0875) (0.0666) Foregn GDP *** (0.0732) *** (0.054) *** (0.0897) 1.497*** (0.0666) New Zealand GDP (0.7669) *** (0.458) *** (0.7083) *** (0.4265) World GDP *** (0.4419) *** (0.2457) (0.4242) ** (0.2395) Populaton *** (0.194) *** (0.1405) (0.2346) *** (0.1237) Dstance *** (0.1373) *** (0.042) *** (0.1256) *** (0.0374) Non-Englsh (0.0622) *** (0.0375) *** (0.0608) *** (0.0352) Average Mgrant Stock (0.0342) 0.425*** (0.0268) *** (0.0394) *** (0.0203) Average Foregn GDP *** (0.0808) *** (0.0557) *** (0.0952) *** (0.0673) Average Populaton (0.1964) *** (0.1408) (0.2368) *** (0.1248) 1995 Dummy (0.1166) *** (0.0678) (0.107) ** (0.0573) Log Lkelhood Observatons Countres Notes For defntons of the varables refer to Table 3. A constant term and country specfc effects are ncluded n all regressons. Dependent varables are n 1995 New Zealand dollars. Three stars (***) ndcates that the coeffcent s sgnfcantly dfferent from zero at the 1% sgnfcance level, two stars (**) ndcates that t s sgnfcant at the 5% level, and one star (*) ndcates that t s sgnfcant at the 10% level. As expected the effect of these exclusons s to ncrease the coeffcent estmates on Mgrant Stock n the trade equatons for both exports (excludng agrculture) and mports (excludng ol). 7.2 Extensons We now examne whether the effect of mgraton vares by type of country, and we test for the effect of mgraton on toursm exports. 18

20 Table 6 Allowng the effect of mgrants to vary by language Varable Exports Imports Selecton Trade Selecton Trade Mgrant Stock *** (0.0378) *** (0.0171) (0.0381) *** (0.0251) Mgrant Stock x Non-Englsh ** (0.0267) *** (0.0099) *** (0.0245) *** (0.0138) Zero Mgrants *** (0.0922) *** (0.049) *** (0.0914) * (0.08) Foregn GDP *** (0.0921) *** (0.0507) *** (0.0896) *** (0.073) New Zealand GDP 3.048*** (0.8134) (0.354) *** (0.7271) *** (0.4815) World GDP *** (0.4852) ** (0.21) (0.4252) ** (0.2818) Populaton *** (0.2545) *** (0.1092) (0.2272) *** (0.1363) Dstance *** (0.1587) *** (0.0357) *** (0.131) *** (0.0428) Non-Englsh * (0.1253) *** (0.057) *** (0.1186) (0.0832) Average Mgrant Stock *** (0.0354) *** (0.0174) *** (0.0388) *** (0.0248) Average Foregn GDP *** (0.0975) (0.0521) *** (0.0959) (0.0741) Average Populaton 3.023*** (0.2517) 0.375*** (0.1092) (0.2293) * (0.1371) 1995 Dummy *** (0.1188) *** (0.0482) (0.1096) (0.0635) Log Lkelhood Observatons Countres Notes For defntons of the varables refer to Table 3. A constant term and country specfc effects are ncluded n all regressons. Dependent varables are n 1995 New Zealand dollars. Three stars (***) ndcates that the coeffcent s sgnfcantly dfferent from zero at the 1% sgnfcance level, two stars (**) ndcates that t s sgnfcant at the 5% level, and one star (*) ndcates that t s sgnfcant at the 10% level. Table 6 presents our results for both export and mport equatons when we allow nteracton between our Mgrant Stock and Non-Englsh varables. If mgrants from Non- Englsh speakng countres have a greater effect on trade than mgrants from Englsh speakng countres then we should see postve coeffcent estmates on our nteracton varable, Mgrant Stock x Non-Englsh. In the selecton equatons for both exports and mports, mgrants from Non-Englsh speakng countres ncrease the probablty that trade wll take place between New Zealand and a gven country by more than mgrants from Englsh speakng countres. The effect of mgrants from Non-Englsh speakng countres 19

21 n both trade equatons s however less than that of mgrants from Englsh speakng countres. The overall mpact on trade of Englsh versus Non-Englsh speakng mgrants s therefore unclear. Ths may n large part be an artefact of the choce of our language varable whch classes over 80 countres as beng Englsh speakng. Table 7 Allowng elastcty to change wth the number of mgrants Varable Exports Imports Selecton Trade Selecton Trade Mgrant Stock *** (0.0545) Square of Mgrant Stock *** (0.0057) Zero Mgrants *** (0.1106) Foregn GDP *** (0.0912) New Zealand GDP *** (0.8099) World GDP *** (0.4825) Populaton *** (0.2576) Dstance *** (0.1451) Non-Englsh *** (0.0662) Average Mgrant Stock (0.035) Average Foregn GDP (0.0961) Average Populaton *** (0.2558) 1995 Dummy *** (0.1187) *** (0.0225) * (0.0016) (0.0574) *** (0.0507) * (0.3593) 0.33 (0.2168) *** (0.1097) *** (0.0358) *** (0.0305) *** (0.0169) ** (0.0528) * (0.1098) *** (0.0483) 0.346*** (0.0533) *** (0.005) *** (0.1107) *** (0.0889) *** (0.463) (0.3915) (0.231) *** (0.1233) *** (0.0617) *** (0.0388) *** (0.0953) (0.2335) *** (0.0296) *** (0.0021) *** (0.0864) *** (0.0727) *** (0.2914) * (0.2455) *** (0.1387) *** (0.0426) *** (0.039) *** (0.0243) *** (0.0738) *** (0.1397) Log Lkelhood Observatons Countres Notes For defntons of the varables refer to Table 3. A constant term and country specfc effects are ncluded n all regressons. Dependent varables are n 1995 New Zealand dollars. Three stars (***) ndcates that the coeffcent s sgnfcantly dfferent from zero at the 1% sgnfcance level, two stars (**) ndcates that t s sgnfcant at the 5% level, and one star (*) ndcates that t s sgnfcant at the 10% level. We repeat our analyss excludng agrculture exports and ol mports. The effect of excludng these goods s to ncrease the coeffcent estmates on the nteracton term n both the trade equatons to the pont where the coeffcent estmate for mports s postve and sgnfcant. The nteracton term n the selecton equaton for exports s however no 20

22 longer sgnfcant. We also allow the effect of mgrants to vary wth the ncome of the country from whch they were born n. These results are not ncluded here due to space lmtatons, but they are avalable upon request. In Table 7 we allow for the possblty that the effect of mgrants on trade vares wth the number of mgrants from a gven country. We do ths by ncludng the square of our Mgrant Stock varable. The estmated coeffcent on ths varable s negatve n all four equatons ndcatng that there are decreasng returns to mgrants. The effects are weaker when we exclude agrcultural goods and ol from exports and mports respectvely. Table 8 The effect of mgraton on toursm exports Varable Toursm Exports (Model One) Toursm Exports (Model Two) Selecton Trade Selecton Trade Mgrant Stock (0.0315) *** (0.0069) ** (0.0554) Square of Mgrant Stock ** (0.0063) Zero Mgrants * (0.0906) Foregn GDP (0.081) New Zealand GDP *** (0.6395) World GDP * (0.3517) Populaton *** (0.2077) Dstance *** (0.119) Non-Englsh *** (0.0595) Average Mgrant Stock *** (0.0326) Average Foregn GDP 0.662*** (0.0885) Average Populaton *** (0.2092) 1995 Dummy *** (0.0964) *** (0.0289) *** (0.0181) *** (0.2071) *** (0.1164) *** (0.0504) *** (0.0183) ** (0.016) *** (0.0079) *** (0.0191) *** (0.0506) *** (0.0286) (0.1238) (0.0808) *** (0.6386) * (0.3515) *** (0.2081) *** (0.1184) *** (0.0597) *** (0.033) 0.645*** (0.0884) *** (0.2095) 0.33*** (0.0963) *** (0.0126) *** (0.0009) *** (0.0368) *** (0.0192) -0.78*** (0.2068) *** (0.1154) *** (0.0505) *** (0.0185) *** (0.0158) *** (0.0085) *** (0.0203) *** (0.0508) *** (0.0285) Log Lkelhood Observatons Countres Notes For defntons of the varables refer to Table 3. A constant term and country specfc effects are ncluded n all regressons. Dependent varables are short term vstor flows to New Zealand durng any gven year. Three stars (***) ndcates that 21

23 the coeffcent s sgnfcantly dfferent from zero at the 1% sgnfcance level, two stars (**) ndcates that t s sgnfcant at the 5% level, and one star (*) ndcates that t s sgnfcant at the 10% level. In Table 8 we use short term vstor flows to proxy for toursm exports. It s evdent from the table that mgrants have a strong postve effect on toursm exports. There s evdence to suggest that ths effect ncreases as the number of mgrants from a gven country ncreases. Because there was a change to the way n whch mgrants were counted n the 1996 census that may nfluence these results, we repeat our analyss for the perod 1996 to The estmated coeffcents on Mgrant Stock for ths perod are somewhat larger. The tables wth full results are agan omtted due to space lmtatons, but are avalable upon request. 7.3 Addtonal robustness testng To nvestgate the robustness of our results we have tred splttng up our sample by Hgh and Low ncome countres and by Englsh and Non-Englsh speakng countres. The effect of splttng our sample by language s that the estmated coeffcents on Mgrant Stock fall as compared to our benchmark results, and vary markedly between Englsh and Non- Englsh speakng countres (wth the estmated coeffcents on Mgrant Stock actually beng hgher for Non-Englsh speakng countres). When we splt our sample by ncome the estmated coeffcents are smlar to those we present n Table 4. We have also tred excludng from our regressons all countres from whch New Zealand has no mgrants. The effect of ths on our estmated coeffcents for Mgrant Stock s neglgble. We also nclude the log of the real exchange rate n our regressons aganst the same sample when the real exchange rate varable s excluded. The effect of excludng the real exchange rate term on our coeffcent estmates on Mgrant Stock s neglgble. The coeffcents are, however, consderably hgher than our benchmark results presented n Table 4. Ths agan shows the senstvty of our results to sample selecton and may help account for the varety n the predcted effects of mgrants on trade n other studes. The tables from these regressons are avalable upon request. 22

24 8 Dscusson Results from our benchmark specfcaton strongly suggest that the more mgrants New Zealand receves from a gven country, the more lkely t s that New Zealand exports to that country, controllng for plausble confoundng factors such as GDP, language, populaton, and dstance. Smlarly, the results strongly suggest that f New Zealand does export to a country, then an ncrease n mgrants from that country s assocated wth an ncrease n exports. Results for mports are slghtly dfferent. There s some support for the dea that more mgrants are assocated wth a greater probablty of mportng, and strong support for the dea that more mgraton s assocated wth hgher mports. Taken at face value the results mply that f New Zealand does mport from a country, a 1% ncrease n mgrants s assocated wth a 0.15% ncrease n mports. Followng prevous studes, we attrbute any relatonshp between mgraton and exports to the ablty of mgrants to reduce transacton costs. We attrbute the relatonshp between mgraton and mports to some unknown mx of transacton cost effects and preferences by mgrants for goods from ther country of orgn. In most of our specfcatons t appears as though the relatonshp between mgraton and trade s strongest for mports. Ths suggests that the preferences of mgrants may well be mportant, but could also be due to the goods that New Zealand mports beng more dfferentated n general than those New Zealand exports. Usng nternatonal vstor numbers as a proxy for earnngs, we use the same specfcaton to test for a relatonshp between mgraton and toursm exports. The relatonshp appears to be strong. However, some of ths apparent strength s probably an artefact of some short-term vstors beng recorded as long-term mgrants. Some varants on the benchmark case for merchandse exports and mports gve results that are consstent wth the hypothess that mgraton stmulates trade. We expected that mgraton would have less effect on transacton costs and hence trade for agrcultural exports and ol mports. It turns out that excludng agrculture and ol do ndeed yeld stronger relatonshps. A possble objecton to our benchmark specfcaton s that t does not nclude a real exchange rate varable. We appled our benchmark specfcaton on the sub-sample for whch the necessary data were avalable, and then added n the exchange rate varable. Addng the exchange rate varable made lttle dfference to the results. 23