Immigration-Trade Nexus: the Case of the EU *1

Asia-Pacific Journal of EU Studies Vol. 13 No. 2 69 Immigration-Trade Nexus: the Case of the EU *1 MATSIUK NADIIA ** AND KIM CHONG-SUP Seoul National University The empirical findings of this study concerning immigration and trade in the EU contribute to the notion of international migration and international trade being complementary phenomena. Regression analysis (PPML) of the gravity model of trade augmented with immigration variables that encompassed 22 members of the EU and their 155 trading partners for the years 2000 to 2012 was used to show that immigration to the EU tends to have a trade-creating effect. This effect was found to be stronger for the EU countries imports, than for exports and for the intra-eu migration compared to immigration to the EU from the non-eu countries. The trade-creating effect from newcomers turned out to be higher than from those living in a given EU country for more than a year. Keywords: International Migration, International Trade, EU, Gravity Model, PPML I. INTRODUCTION Despite certain controversy in the eyes of the nations and governments, international migration flows are gaining momentum. According to the UN estimates, 213 million people (or 3.2% of world population) resided out of their country of birth in 2013, 1 with Europe being the largest region of migrant concentration in absolute terms. 2 With the intensification of globalization processes, the majority of countries choose to participate in international trade and open their markets to international capital, while international migration remains generally restricted and very conditional (Hatton, 2006). According to the Hecksher-Ohlin-Samuelson theorem international trade displaces (or substitutes for) international migration. In other words, goods trade substitutes for factor trade. As the relative prices for goods converge as a result of trade, the relative prices for capital and labor converge across * The authors are grateful to Prof. Rhee Yeongseop (Seoul National University) for valuable comments and suggestions and to Mario Pasquato (Yonsei University) for consulting on the statistical aspects of this research. ** Master in International Studies, Seoul National University, Graduate School of International Studies; E-mail: nadia.matsiuk@gmail.com Professor of Economic Development, Seoul National University, Graduate School of International Studies, E-mail: chongsup@snu.ac.kr

70 Immigration-Trade Nexus trade partners. The above statement was challenged by the advocates of the complementarity-type relationship between international migration and trade. 3 The latter study the causal links of how international migration can boost trade of the migrant-receiving country. We present a concise description of some of the channels 4 through which migration from country A to country B might cause the augmentation of trade between the two countries: 234 Immigration can lead to international trade through the reduction of certain trade barriers. Language proficiency, better knowledge of native country legal system as well as networking connections that immigrants posses can lead to flourishing of the foreign-born entrepreneurship in receiving countries. Both export and import of country B are augmented in this case (with prevalence of the import increase effect). This effect is usually called the network effect. As immigration raises production and aggregate demand in country B it also leads to an increased demand for goods, including the imported ones (especially, those originating from immigrants home countries). This is often referred to as the preference effect. Subsequent demonstration effect (through ethnic shops or restaurants) might lead to even higher levels of import of goods from country A. One can expect a positive effect on import of country B from country A as a result. If immigrants tend to work in exporting sectors and drive the real marginal costs of production down (by accepting lower wages), this can result in the increases in export from B to the rest of the world, including country A. This channel can be referred to as the export sector employment effect. Are international migration and trade substitutes or complementaries? An evidence-based answer to these questions is crucial for adequate policymaking. The aim of this paper is to contribute to the above discussion through the examination of the empirical relationship between international migration and international trade in the region that is in the center of migration processes-the European Union. The issue of migration in the EU is gaining increasing public and aca- 1 In comparison, in 1990 the number of international migrants stood at 155 million (or 2.9% of the global population). UN, Department of Economic and Social Affairs, Population division. 2 In 2013 72 million international migrants resided in Europe, compared to 71 million in North America, UN, Migration Wallchart 2013. 3 Mundell (1957), Markusen (1983), Wong (1983), Krugman (1991), Puga (1999), Ottaviano et al. (2003) etc. 4 These channels were discussed among others in Gould (1994), Pang and Lim (1996), Egger et al. (2011).

MATSIUK NADIIA AND CHONG-SUP KIM 71 demic attention, as the scope of the process increases. In 2013, approximately 33 million 5 foreign-born resided in European Union countries (20 million of whom are non-eu country nationals). A conventional approach to testing the trade-migration nexus is to find a correlation between flows or stocks of migrants from country A to country B and trade volumes between the countries involved. If this relationship is confirmed, the complementarity-type relationship can be concluded. In case of the negative correlation, trade and migration are viewed as substitutes. The gravity model has been employed as a means of analyzing the relationship between trade and migration starting with Gould (1994). Gould (1994) found positive effect of net migration on the US trade (1970-1986, 47 trade partners). He found that a 1% increase in immigrant stock of a particular sending nation increases US import from that nation by 0.01% and export to that nation by 0.02%. However, the author further assumed that one could expect diminishing immigration effect on trade in the long run. Since then, the gravity model was used for evaluating the influence of immigration in studies on many separate countries and regions. Out of the extensive body of research concerning the links between international migration and trade the studies that confirm trade-enhancing effects of the former prevail. Examples include the studies for Canada (Head and Ries 1998), New Zealand (Qian, 2007), the US (White, 2010), Sweden (Garmaza, 2011), Greece (Piperakis et al., 2003), Germany (Bruder, 2004) etc. Some previous studies did not find a significant relationship between international migration and trade, for example Girma and Yu (2002) for Great Britain, Bettin and Lo Turco (2009) for the OECD, Clarke and Hillbery (2009) for Australia. The results of a meta-analysis conducted by Genc et al. (2011) suggest that among 48 studies on the trade-creating effect of immigration with similar methodology (augmented gravity model) and estimation techniques (least squares for log-linear models) 30 presented positive effects. Among the remaining 18 papers only 4 found negative results. As for the papers focusing on the EU, Cagatay et al. (2013) examined the effects of immigration from Mediterranean and Eastern Europe on the EU countries (1998-2010). The authors useda gravity model augmented with an immigrant stock variable. They found that a 1% increase in immigrant stock from a given country from regions under examination leads to 0.05% increase in exports, but migration from Eastern Europe is found to have adverse effect on exports. For imports, no significant effect for the 5 Eurostat migration databases. Immigrant stock and flow to the EU countries by citizenship of immigrants: Eurostat migration database, datasets (migr_imm5 prv) and (migr_ pop1ctz), http://ec.europa.eu/eurostat/statistics-explained/index.php/migration_and_mi grant_population_statistics.

72 Immigration-Trade Nexus Mediterranean was found, but a significant positive effect for Eastern Europe was confirmed. Casi (2011) examined the effect on immigration to 17 EU member states from 10 non-eu countries (major immigrant sending states) over the years 1997-2006 to find a significant positive effect on EU countries export. In general, the discrepancies of empirical results of the previous studies on the relationship between immigration and trade as well as the opportunity to conduct an inclusive study for the EU countries using the Eurostat data were the among the drivers of this research. Our study encompassed the relevant data for all the available migrant-sending countries, as well as the EU member-states over a timeframe of 12 years. This researchutilized the Eurostat database of the immigrant flow to the EU countries as well as the data base of the relative immigrant stocks. As a result, we were able to compare the trade-enhancing effect of the newly arrived immigrants to the effect of those who lived in a host country for longer period of time within the same study. Furthermore, we examined the differences in the trade-enhancing effect of international migration in the case of the intra-eu migration and immigration to the EU from the third countries. The empirical differences allowed us to juxtapose different migration policy approaches. The paper is organized as follows. It opens with an overview of our hypotheses and the gravity model used in this study as well as the estimation techniques (Section 2). More detail on the data and the relevant sources is provided in Section 3. We present our results in Section 4. Some considerations on the implications of our findings are provided in Section 5. II. HYPOTHESES AND MODEL We started our analysis with an expectation to find a positive relationship between trade and immigrant stock, with network effect combined to preference effect 6 producing a bigger effect on the home country exports than on the host country export (which is mostly subject to network effect only). Furthermore, we expected to observe a bigger trade-creating effect of the intra-eu migration than migration to the EU from non-eu countries. Since 2004 as a part of single market formation, the EU introduced the free 6 The possible links through which migration can enhance trade were discussed in the introduction. In gravity model-based research only two of the effects are usually accounted for: preference effect and network effect. Preference effect stands for the fact that immigrants consumption preferences might induce an increase in imports from their home country to the destination country. Network effect explains for the increase in both exports from the host country and imports from the home country, as newly arrived residents might effectively exploit their language and legal knowledge as well as personal connections with the sending countries in order to establish trading businesses. By comparing the extent to which immigration affects import and export one can assume which type of effect prevails.

MATSIUK NADIIA AND CHONG-SUP KIM 73 movement of people. 7 The features of the Eurostat data make it possible to distinguish between immigrants from other European countries and non- EU countries. Taking into account the freedom to move, settle and work that EU member-states enjoy in any other member-state and a rather restricted situation of the non-eu migrants, this empirical study can help answer the question: Which type of migration has a more distinct trade-creating effect, the liberalized EU-type one or the more restricted one applied to the third country immigrants? We expect the intra-eu migration to have a stronger trade-creating effect, as there no formal obstacles to secure employment or other income-generating activity for the intra-eu migrants as opposed to the third-country nationals, who do not enjoy such freedoms. The trade-creating effect from newcomers was expected to be stronger (immigration flow) than from those living in a country for a longer time (stock). Diminishing trade-enhancing effects of immigration have long been discussed in the literature. 8 The trade-creating effect might wither away with time as the immigrant stock from a particular country increases. As a result, the local producers adapt to immigrant s consumption demands or immigrants themselves become similar to locals in their consumption preferences. This development might diminish the preference effect of immigration on trade (boosting of the export of a sending country). Furthermore, the opportunities for conducting business with the home country might be exhausted. To test the assumption on the different trade-enhancing impact of newcomers and long-term foreign residents we compared the results for two models, the one with the immigrant flow variable (foreign-born, that moved to an EU member-state within a year) and the one with the immigrant stock variable (foreign-born, that have resided in an EU country for more than a year). We adopted the conventional gravity model methodology used in the studies on the potential trade-creating effect of international migration. 9 7 Directive 2004/38/EC on the right to move and reside freely. 8 Gould (1994), Egger et al. (2011), Foad (2009) elaborate on this topic. 9 Genc et al. (2011) outline the standard gravity model equations used for evaluating it: K k ij = α0 + α1 ij + α2 i + α3 j + α4 ij + α k 0 k ij + ε = ij ln M lni lny lny lnd lnz K k ij = β0 + β1 ij + β2 i + β3 j + β4 ij + β k 0 k ij + δ = ij ln X lni lny lny lnd lnz where M ij is import from migrant source country i to host country j; X ij is exports from migrant host country j into migrant source country; I ij s the number of immigrants from country i living in country j; Yi is the GDP of a migrant-sending country; Yj is the GDP of a migrant receiving country; D ij is the distance between countries i and j; k Z ij stands for k other explanatory variables; ε ij and δ ij are the error terms; α and β are the parameters of the model.

74 Immigration-Trade Nexus A total of 4 models were run to test the hypotheses: gravity models of trade for export and import were augmented with variable accounting for immigration (namely, immigrant stock and immigrant flow). The summary of the basic models used is presented in Table 1: TABLE 1. GRAVITY MODELS USED IN THIS PAPER i is a migrant-receiving EU-country; j is migrant-sending EU or non-eu country. E ij is the amount of export from i to j; I ij is the amount of import from j to i; Immigrant stock ji is the stock of immigrants from j residing in i; Immigrant flow ji -is the flow of immigrants from j toi; Y i and Y j are the respective GDPs; D ij is a measure of distance between the countries. Export Import The gravity model augmented with the immigrant stock variable β1 β Y 2 i Yj Eij = Immigrant stock ji β D 3 Y I = Immigrant stock ij ji ij β1 β2 i Yj β D 3 ij The gravity model augmented with the immigrant flow variable β1 β Y 2 i Yj Eij = Immigrant flowji β D 3 Y I = Immigrant flow ij ji ij β1 β2 i Yj β D 3 ij We used four methods to estimate the gravity models: Ordinary Least Squares (OLS) for pooled data, OLS for panel data with fixed time effects, OLS for panel data with random effects, Poisson Pseudo Maximum Likelihood (PPML) estimation. However, we interpret the result of the PPML estimation only, as all the OLS models were found to have a heteroscedasticty 10 issue. Nevertheless, all the estimations provided consistent results. 11 PPML estimation was suggested by Santos Silva, Tenreyro (2006) 12 as an alternative estimator for gravity models of trade in the presence of heteroscedasticity. The method has become widely used in international trade research. Santos Silva and Tenreyro argued that the Poisson Pseudo Maximum Likelihood estimation is a more reliable (thus less biased) and a 10 Heteroscedasticity is a characteristic of the variability of a dependent variable that applies when the variance of the latter is not uniform and its pattern depends at the variance of some of the independent variables. If the variance of the dependent variable remains the same regardless of the values of independent variables, they are referred to as homoscedastic. Heteroscedasticity is very likely to occur in most of the cases of econometric research. The main concern about OLS in the presence of heteroscedasticity is that OLS regression does not estimate all the range of values of dependent variable with the same accuracy. For example, it can predict smaller values of dependent variable better than the bigger ones. After all, OLS assumes that the variables are homoscedastic. From Barreto H, F. M. Howland. Introduction to Econometrics. Chapter 19. Cambridge University Press. 2001. 11 Detailed OLS estimation results are provided in the Appendix. 12 Santos Silva, J., and S. Tenreyro, The Log of Gravity, Review of Economics and Statistics, Vol.88, No.4, 2006, pp. 641-658.

MATSIUK NADIIA AND CHONG-SUP KIM 75 more efficient estimation method, especially in the presence of heteroscedasticity (as a conditional variance is assumed to be proportional to the conditional mean, and the error s variance is accounted for more efficiently). Under PPML method gravity models are estimated in the non-linear form (actually, it is the same as non-linear least squares 13 but with specific assumptions about the error terms) and the variables do not need to have a Poisson distribution. Though the PPML estimation technique is most often used for count data the dependent variable does not need to be an integer. In the case of PPML estimation the gravity model takes the following forms. For the immigrant stock model: ( α β1 β2 β3 β4 ε ) ( α β1 β2 β3 β4 ε ) Export = exp + lnimmigrant stock + lny + lny + lnd + ij ij ij i j ij ij Import = exp + lnimmigrant stock + lny + lny + lnd + ij ij ij i j ij ij For the immigrant flow model: ( α β1 β2 β3 β4 ε ) ( α β1 β2 β3 β4 ε ) Export = exp + lnimmigrant flow + lny + lny + lnd + ij ij ij i j ij ij Import = exp + lnimmigrant flow + lny + lny + lnd + ij ij ij i j ij ij Usually the results of the PPML estimation of gravity models differ from that of the log-linear OLS estimation in the following ways: 14 more observations are analyzed; the coefficients for GDPs and distance are smaller (this might result from accounting for heteroscedasticity). As PPML estimation possesses a lot of benefits especially for gravity models of trade, there is a strong argument in the literature to use it. 15 III. DATA The information on the variables that were used for the statistical model as well as their sources is summarized in Table 2. 13 The method is based on approximation of a model to a linear one, but then the parameters are refined by successive iterations. 14 Shepherd B., The Gravity model of International Trade: A User Guide. Chapter 4: Alternative gravity model estimators. UNESCAP, United Nations Publication, 2013. http://www.unescap.org/sites/default/files/tipub2645.pdf. 15 Same as above.

76 Immigration-Trade Nexus TABLE 2. THE LIST OF VARIABLES AND THEIR SOURCES (DATA FOR THE YEARS 2000-2012) Export Import Name Definition Source Notes Immigrant stock (immigrants from EU or non- EU country to a EU country) Immigrant flow (from a sending EU or non-eu country to a receiving EU country) GDP (Y i, Y j ) Distance (D ij ) From a EU country to a migrant-sending country UN Comtrade data Total From a migrant-sending UN Comtrade data country to a EU country Number of immigrants (on the basis of country of birth) living in a EU member state in a given year (an estimate based on population censuses). Number of immigrants (persons that arrive to a member-state with a view of permanent residence for longer than 12 month) that arrive to a EU country in a given year. Nominal GDP of 200 countries in a respective year Distances between country pairs are measured as distances between their capital cities. Eurostat migration database Dataset: Population on 1 January by five year age group, sex and citizenship (migr_pop1ctz) Eurostat migration database Dataset: Immigration by five year age group, sex and country of previous residence [migr_imm5prv] UN data (Original source: World Development Indicators, World Bank) The database of University of Essex professor K. Gleditsch, file capdist.csv 16 Total commodities, HS as reported. Extracted on 20.03.2015 commodities, HS as reported. Extracted on 20.03.2015 27 EU countries and 200 partner countries. Though prior to 2009 the relevant data was collected based on a gentleman s agreement, its submission was mandatory there from. Extracted on 29.02.2015 27 EU countries and 200 partners (migrant-sending countries). Though prior to 2008 the data on immigrant flows was collected based on a gentleman s agreement, since 2008 EU members were required to submit it. Extracted on 17.03.2015 Current US dollars Extracted: 17.03.2015 Accessed on18.03.2015 Examination of an extensive time period and a big number of countries (28 EU member-states vs. all the countries of the world) resulted in a total of more than 70,000 observations (for each of the models), out of which 27,347 were actually used in the analysis of the Flow model and 28,943 for the Stock model (as they did not have missing or zero values). As zero values posed a problem for log-linear estimation, they were eliminated. In the end, 25 EU countries were represented in the Stock model (together with 158 partners), 22 EU countries were represented in the Flow model (as well as 155 partners). 17 The discrepancy in the number of coun- 16 Link to the source: http://privatewww.essex.ac.uk/~ksg/data-5.html. 17 The complete list of EU member states represented in the analysis is attached as Appendix A.

MATSIUK NADIIA AND CHONG-SUP KIM 77 tries in the model arose due to the fact that we construed two models based on two separate databases (Eurostat immigrant stock and immigrant flow databases) that differed in country and year coverage. Data cleaning and harmonization across the different sources werecentral to our research. IV. RESULTS The results of OLS for pooled data as well for panel data with random effects and fixed time effect are presented only in the Appendix (as heteroscedasticity was found to be a problem in all the OLS estimations). PPML estimation technique being less biased and more efficient 18 compared to OLS, we present the results of the PPML estimation in detail. Independent variables TABLE 3. THE RESULTS OF PPML ESTIMATION Dependent variables Flow model Stock model Import Export Import Export ln(gdp EU ) 0.59 0.60 0.60 0.63 (0.017) (0.015) (0.015) (0.013) ln(gdp Migrant-sending country ) 0.65 0.65 0.68 0.68 (0.008) (0.009) (0.008) (0.008) ln(distance) -0.63-0.76-0.64-0.76 (0.014) (0.014) (0.013) (0.012) ln(immigrant flow) 0.2098 0.2066 (0.010) (0.008) - - ln(immigrant stock) - - 0.1703 0.1624 (0.008) (0.007) Constant -8.47-7.60-9.34-9.38 (0.519) (0.503) (0.484) (0.456) No of obs 21956 21956 27072 27072 Pseudo-R 2 19 0.89 0.92 0.90 0.93 Note: * 10% significance level, ** 5% significance level, 1% significance level. 18 Santos Silva, J. and S. Tenreyro, The Log of Gravity, Review of Economics and Statistics, Vol.88, No.4, 2006, pp. 641-658. 19 Pseudo-R 2 in PPML is McFadden s pseudo-r 2 that takes into account maximum likelihood estimates. It is calculated as one minus the ratio of log maximum likelihoods of full (fitted) model to the log maximum likelihood of intercept model. Pseudo-R 2 cannot be interpreted as goodness of it and thus cannot be directly compared to the OLS R 2. It is interpreted as an improvement from null model to fitted model. The closer its value is to one, the better the fitted model is over the intercept model. It is called pseudo-r 2 as it was designed to look similar to OLS R 2. From IDRE UCLA website, FAQ: What are pseudo-r-squareds? http://www.ats.ucla.edu/stat/mult_pkg/faq/general/psuedo_rsquareds.htm.

78 Immigration-Trade Nexus The coefficients of independent variables proved smaller than those returned by the OLS regressions (Appendix B). Nevertheless, their signs are consistent with our general expectations. A 1% increase in GDP of a EU member state would result in 0.6% increase of import (or export) from migrant sending country to a EU country for both the Flow and Stock models. A 1% increase in partner s GDP would lead to 0.65% increase in import (or export) for the Flow model, and a 0.68% increase in import (or export) flow for the Stock model. The elasticity of trade flows for distance is different for import and export (but similar across the models): a 1% increase in distance tends to decrease import by 0.63-0.64%, and export by 0.76%. When it comes to the immigrant flow, it is positively and significantly related to trade flows. A 1% increase in the immigrant flow would result in 0.2098% increase in import of a EU member from a migrant sending country and 0.2066% increase of exports from a EU state to migrant sending country. The elasticity coefficients are smaller for the immigrant stock variable: a 1% increase in the stock of immigrants from a given country to a EU country would lead to 0.1703% increase of imports from the former, and to 0.1624% increase in export from the EU country to the sending country. According to our findings, immigrant flow indeed tends to have a bigger trade-creating effect then the immigrant stock. This goes in line with our expectations. The consumption patterns from newly arrived foreigners are likely to be more inclined towards the goods produced in their home countries. The overseas demand would have a stimulating effect on the home-countries export. On the contrary, as a larger single-country immigrant stock builds up in a host country several developments take place that potentially diminish the stimuli to trade: the consumption patterns of immigrant population tend to converge to those of the host country population; local host-country producers start to cater for the specific needs of the new large groups of consumers etc. While we observe different trade-creating effects of immigrant stock and flow, in both cases the effect for import turns out to be bigger. More specifically, network effect of immigration on trade arises due to the immigrant s country-specific knowledge and home contacts that tend to decreasethe costs of establishing international businesses. It can have a boosting effect on both receiving country s import and export. Preference effect appears due to increased demand for sending country s consumer goods in a given receiving country when the immigrant stock of the sending country increases. However, this effect leads to an increase only in imports from the immigrant-sending country to the immigrant-receiving one. Our findings support the assumption on the preference and network

MATSIUK NADIIA AND CHONG-SUP KIM 79 effects combined having a more pronounced effect on the home country import from rather than its export to an immigrant-sending counterpart. In order to test the hypothesis on the trade-creating effect of immigration to a EU member from another EU states vs. immigration from a non- EU members, we divided the data into two parts: the EU-EU countries and the EU-non-EU countries. Two resulting groups of models were analyzed. TABLE 4. PPML ESTIMATION OF THE MODEL FOR EU-EU COUNTRIES TRADE AND MIGRATION Independent variables Dependent variables Flow model Stock model Import Export Import Export ln (GDP EU ) 0.545 0.547 0.505 0.523 (0.023) (0.018) (0.019) (0.025) ln (GDP Migrant-sending country ) 0.624 0.600 0.577 0.575 (0.014) (0.016) (0.013) (0.013) ln (Distance) -0.700-0.768-0.673-0.716 (0.024) (0.025) (0.020) (0.020) ln (Immigrant flow) 0.216 0.224 (0.016) (0.013) - - ln (Immigrant stock) - - 0.208 0.214 (0.012) (0.010) Constant -5.991-4.948-4.09-4.22 (0.792) (0.717) (0.678) (0.567) No of obs 3884 3884 4978 4978 Pseudo-R 2 0.903 0.905 0.89 0.91 Note: * 10% significance level, ** 5% significance level, 1% significance level. The EU-EU model (immigration and trade takes place between EU member states) has 3884 observations and all the independent variables are significant. The elasticities of trade for GDPs and distance are similar to those of the model with all observations. The elasticities of trade depending on changes in the immigrant flow are still higher than those for stock. However, in this case the trade-creating effect of immigration (both flows and stocks) is bigger for export than import (both models). In case of EU-non EU countries model, the trade-creating effects of both immigration flow of citizens from non-eu countries to a EU country and their migrant stock is bigger for import than for export. By comparing the elasticities of trade on immigration variables for both models, we can conclude that the effects are stronger in the EU-EU trade and migration case.

80 Immigration-Trade Nexus TABLE 5. PPML ESTIMATION OF THE MODEL FOR EU-NON EU COUNTRIES TRADE AND MIGRATION Independent variables Flow model Dependent variables Stock model Import Export Import Export ln (GDP EU ) 0.676 0.673 0.746 0.848 (0.021) (0.0004) (0.020) (0.018) ln (GDP Migrant-sending country ) 0.663 0.678 0.741 0.779 (0.009) (0.0002) (0.010) (0.008) ln (Distance) -0.541-0.717-0.633-0.802 (0.021) (0.0002) (0.020) (0.015) ln (Immigrant flow) 0.194 0.187 (0.013) (0.0002) - - ln (Immigrant stock) - - 0.132 0.092 (0.010) (0.008) Constant -11.803-24.593-14.87-17.090 (0.635) (0.0109) (0.650) (0.570) No of obs 18072 18072 22094 22094 Pseudo-R 2 0.87 0.92 0.86 0.91 Note: * 10% significance level, ** 5% significance level, 1% significance level. We interpret this finding as the evidence of a more liberalized migration regime leading to more pronounced trade augmentation. For instance, absence of the bureaucratic barriers for intra-eu immigrants contributes to the establishment of trade businesses. However, the bigger elasticity of trade on immigration inside the EU could be in part explained by generally very close economic integration in the region. The liberalization of the movement of people inside the Union is just another aspect of the lengthy socio-economic integration process. Nevertheless, our findings suggest that a more liberal regime of international migration might result in trade increase, for both the receiving and sending country. V. CONCLUSIONS Though previous research on the relationship between migration and trade produced mixed results, our results confirm the trade-creating effect of international migration. Regression analysis of the gravity model of trade augmented with immigration variables that encompassed EU countries and their trading partners for the years 2000 to 2012 showed that immigration from one country to another tends to have a trade-creating effect. According to our

MATSIUK NADIIA AND CHONG-SUP KIM 81 findings, the trade-creating effect of immigration tends to be stronger for import from a migrant-sending country to a migrant receiving country, than for export from the receiving to the sending country. This can be explained by the combination of network and preference effects. The immigration elasticity of trade turned out to be higher for the EU- EU trade than for the EU-non-EU one. Probably, the liberalized movement of people has a stronger effect of intra-regional trade. As for the immigrant flow and immigrant stock variables, the trade-creating effect from the former was found to be bigger. This confirms the assumption on the diminishing effects of immigration on trade. Concerning the implications of this paper for the debate on international migration, it provides an argument for the notion of migration bearing reciprocal effects for migrant-receiving countries. For the latter the effects are channeled through the augmented trade flows. Furthermore, migration and trade can be viewed as complementary phenomena: increased migration does not inhibit trade, as it would happen in the factorprice equalization theorem world. The findings of this paper should be viewed as a part of a much broader discourse on economic effects of international migration prior to giving any policy advice. Though migrant-receiving countries may experience trade augmentation as a consequence of the international migration (as in the case of the EU), other ways migration influences native economy and society should be examined to make final conclusions. Global migration processes are intensifying. Even though global economic crisis had suppressing effect on the movement of people, it is revealed by 2013 data that the migration flows are gaining momentum. By trying to curb international migration countries might be loosing in terms of trade, as all the forms of globalization tend to be interconnected. REFERENCES Barreto, Humberto and Frank M. Howland, Introduction to Econometrics, Chapter 19, Cambridge University Press, 2001. Bettin, Giulia and Alessia Lo Turco, A cross country view on South-North migration and trade: dissecting the channels, Emerging Markets Finance and Trade, M.E. Sharpe, Inc., Vol.48, No.4, 2012, pp. 4-29. Bruder, Jana, Are Trade and Migration Substitutes or Complements?-The Case of Germany, 1970-1998, University of Rostock, 2004 (http://www.etsg. org/etsg2004/papers/bruder.pdf). Cagatay Selim, Murat Genc, and Onur A. Koska, The Impact of Immi-

82 Immigration-Trade Nexus gration on International Trade in Europe: The Case of the EU-Mediterranean-Eastern Europe Zone, ERSA conference papers, European Regional Science Association, 2013 (http://www-sre.wu.ac.at/ersa/ersaco nfs/ersa13/ersa2013_paper_00376.pdf). Casi, Laura, Enhancing Trade Through Migration: A Gravity Model of the Network Effect, ISLA-Bocconi, Mimeo, 2011. Clarke, Andrew J. and Russel H. Hillberry, Immigration and Trade in Recent Australian History, University of Melbourne, 2009 (http://spot.colora do.edu/~kellerw/courses/9999f09/hillberry.pdf). Egger Peter, Maximilian Von Ehrlich, Douglas R. Nelson, Migration and Trade, The World economy, Wiley Blackwell, Vol.35, No.2, 2012, pp. 216-241. Foad, Hisham S., A Threshold Model for the Migration-Trade Link, 2009. Available at SSRN (http://ssrn.com/abstract=1424608). Garmaza Volga, The Impact of Immigration on Trade: the Case of Sweden, (Master s diss. Södertörns Högskola, Department of Economics) 2011 (http://www.diva-portal.org/smash/get/diva2: 482601/FULLTEXT01. pdf). Genc Murat, Masood Gheasi, Peter Nijkamp, and Jacques Poot, The impact of immigration on international trade: a meta-analysis, NORFACE Migration, Discussion Paper No.2011-20, 2011 (http://www.norfa ce- migration.org/publ_uploads/ndp_20_11.pdf). Ghatak Subrata, Monica I. Pop-Silaghi, Vince Daly, Trade and migration flows between some CEE countries and the UK, The Journal of International Trade & Economic Development: An International and Comparative Review, Vol.18, No.1, 2009. Girma, Sourafel and Zhihao Yu, The link between immigration and trade: Evidence from the United Kingdom, Review of World Economics, Weltwirtschaftliches Archiv, Springer, Vol.138, No.1, 2002, pp. 115-130. Gould, David M., Immigrant links to the home country: empirical implications for US bilateral trade flows, Review of Economics and Statistics, Vol.76, No.2, 1994, pp. 302-316. Hatton, Timothy J., Trade policy and migration policy: why the difference? Department of Economics, University of Essex, 2006 (https://www.tcd. ie/iiis/documents/archive/seminar%20papers/hattontrademigversio n2.pdf). Head, Keith and John Ries, Immigration and trade creation: econometric evidence from Canada, Canadian Journal of Economics, Vol.31, No.1, 1998, pp. 47-62. Krugman, Paul, Increasing returns and economic geography, Journal of Political Economy, Vol.99, No.3, 1991, pp. 483-499.

MATSIUK NADIIA AND CHONG-SUP KIM 83 Markusen, James R., Factor movements and commodity trade as complements, Journal of International Economics, Vol.14, 1983, pp. 341-356. Mundell, Robert A., International Trade and Factor Mobility, American Economic Review, Vol.47, 1957, pp. 321-335. Ottaviano, Gianmarco and Jacques-Francois Thisse, Agglomeration and economic geography, Handbook of Regional and Urban economics, in: J. V. Henderson and J. F. Thisse (ed.), Handbook of Regional and Urban Economics, Vol.4, 2004, pp. 2563-2608. Qian, Mingming, Economic Relationship between Trade and Immigration in New Zealand, 178.799 Research Report, Massey University, Albany, 2007. Santos Silva, Joao, and Silvana Tenreyro, The Log of Gravity, Review of Economics and Statistics, Vol.88, No.4, 2006, pp. 641-658. Shepherd Ben, The Gravity model of International Trade: A User Guide, Chapter 4: Alternative gravity model estimators, UNECAP, United Nations Publication, 2013 (http://www.unescap.org/sites/default/files/ tipub2645.pdf). White, Roger, Migration and International Trade: The US Experience since 1945, Cheltenham UK and Northhampton MA USA, Edward Elgar Publishing, Inc., 2010. Wong, Kar-Yiu, On Choosing Among Trade in Goods and International Capital and Labor Mobility: A Theoretical Analysis, Journal of International Economics, Vol.14, 1983, pp. 223-235.

84 Immigration-Trade Nexus APPENDIX 1. EU Member-States the Data for which is Represented in Statistical Models A. Stock model 25 EU countries are represented in the Stock model (together with 158 partners): TABLE 6. THE EU MEMBER-STATES REPRESENTED IN THE IMMIGRANT STOCK MODEL Country Frequency of observations Percent from total number of observations 1 Austria 1,504 5.20 2 Belgium 1,360 4.70 3 Bulgaria 822 2.84 4 Czech Rep. 1,857 6.42 5 Germany 1,984 6.85 6 Denmark 1,958 6.77 7 Spain 1,976 6.83 8 Estonia 108 0.37 9 Finland 1,881 6.50 10 France 304 1.05 11 UK 401 1.39 12 Greece 138 0.48 13 Hungary 1,699 5.87 14 Ireland 916 3.16 15 Italy 1,670 5.77 16 Lithuania 111 0.38 17 Luxembourg 231 0.80 18 Latvia 900 3.11 19 Malta 194 0.67 20 Netherlands 1,981 6.84 21 Poland 561 1.94 22 Portugal 1,428 4.93 23 Slovakia 1,281 4.43 24 Slovenia 1,754 6.06 25 Sweden 1,924 6.65 Total 28,943 100

MATSIUK NADIIA AND CHONG-SUP KIM 85 B. Flow model 22 EU countries are represented in the Flow model (as well as 155 partners): TABLE 7. THE EU MEMBER-STATES REPRESENTED IN THE IMMIGRANT FLOW MODEL Country Frequency of observations Percent from total number of observations 1 Austria 1,606 5.87 2 Belgium 457 1.67 3 Bulgaria 270 0.99 4 Cyprus 932 3.41 5 Czech Rep. 1,037 3.79 6 Germany 1,076 3.93 7 Denmark 1,947 7.12 8 Spain 1,967 7.19 9 Estonia 1,083 3.96 10 Finland 1,877 6.86 11 UK 1,024 3.74 12 Croatia 1,005 3.67 13 Ireland 1,054 3.85 14 Italy 1,817 6.64 15 Lithuania 1,473 5.39 16 Luxembourg 468 1.71 17 Malta 25 0.09 18 Netherlands 1,969 7.20 19 Poland 925 3.38 20 Slovakia 1,663 6.08 21 Slovenia 1,756 6.42 22 Sweden 1,916 7.01 Total 27,347 100.00 2. OLS estimation results OLS estimations took the following forms: Immigrant-stock augmented gravity model was as:

86 Immigration-Trade Nexus Export lnimmigrant stock lny lny lnd ln ij = αij + β1 ij + β2 i + β3 j + β4 ij + εij Import lnimmigrant stock lny lny lnd ln ij = αij + β1 ij + β2 i + β3 j + β4 ij + εij In the second specification of the model immigrant flow is used instead of immigrant stock in order to test Hypothesis 3 on the effects newcomers have on trade: Export lnimmigrant flow lny lny lnd ln ij = αij + β1 ij + β2 i + β3 j + β4 ij + εij Import lnimmigrant flow lny lny lnd ln ij = αij + β1 ij + β2 i + β3 j + β4 ij + εij In order to choose between the fixed time effect and random effects OLS models, a Hausman test was conducted. The test checks whether between entity errors are correlated with independent variables (null hypothesis is that they are not correlated, i.e. that the model is with random effects). In most of the situations of econometric analysis the error terms and independent variables are correlated (there is some unobserved variance caused by factors others then independent variables), so fixed effect models are preferred. In our case the Hausman test showed that fixed effect model was preferred. A. Pooled Data OLS Estimation Result This estimation having been made for rather exploratory purposes, all the data points were pooled together and considered as referring to the same point of time. The resulting coefficients had the expected signs and the model fitted the data quite well. However, heteroskedasticity was found to be present (homoscedasticity hypothesis was rejected by Breusch-Pagan/Cook-Weisberg test). Consequently, though coefficients of this regression are quite reliable, the confidence intervals obtained from this OLS estimation are rather dubious. Nevertheless, even looking at the results of this exercise, we find significant results. Both immigrant stock and immigrant flow are significant in explaining trade patterns. What is more, the immigration coefficients (elasticities) are bigger for import, than export. Comparing the Flow and the Stock model, we see that the immigrant flow tends to have a bigger trade creating effect than the immigrant stock. We used the Breusch-Pagan/Cook-Weisberg test for the H 0 of homoscedasticity (constant covariance). A large χ 2 value corresponds to highconfidence rejection of H 0.

MATSIUK NADIIA AND CHONG-SUP KIM 87 Independent variables ln (GDP EU ) ln (GDP Migrant-sending country ) ln (Distance) ln (Immigrant flow) TABLE 8. POOLED DATA OLS REGRESSION RESULTS Flow model Dependent variables Stock model ln (Import) ln (Export) ln (Import) ln (Export) 0.77 (0.013) 1.04 (0.007) -0.91 (0.0162) 0.30 (0.008) 0.88 (0.008) 0.78 (0.004) -1.04 (0.009) 0.21 (0.005) ln (Immigrant stock) - - Constant -22.78 (0.439) -17.00 (0.0025) 0.83 (0.013) 1.07 (0.008) -0.86 (0.016) 0.25 (0.007) 0.87 (0.007) 0.81 (0.004) -1.04 (0.009) - - -25.22 (0.413) 0.19 (0.004) -17.4 (0.229) No of obs 21956 21956 27072 27072 R 2 0.70 0.84 0.69 0.84 Heteroscedasticity test 3005.66 3392.90 3351.27 4196.48 Note: * 10% significance level, ** 5% significance level, 1% significance level. B. OLS for Panel Data with a Fixed Time Effect This estimation technique allowed us to look at the model accounting for the time trend. Under this model (the equivalent to the introduction of time dummies) the data was divided into 13 groups and within, between and overall R 2 were calculated. Though the data shows the presence of heteroscedasticity (except for the case of the Stock model for export), the general results of this regression are as follows: The coefficients having the expected signs and values, the effects of immigration on import from the sending country to the migrant-receiving country are stronger than those on export for both the Stock and Flow models. However, the effects of the immigrant stock are bigger than those of the immigrant flow (that goes against expectations). Heteroscedasticity test in this case was a modified Wald test for group-wise heteroskedasticity suited for fixed effect regression models. Ho: σ (i)^2 = σ ^2 for all i (homoscedasticity). As χ 2 was large, Ho was rejected with high confidence.

88 Immigration-Trade Nexus TABLE 9. FIXED TIME EFFECT REGRESSION RESULTS Dependent variables Independent variables Flow model Stock model ln (Import) ln (Export) ln (Import) ln (Export) ln (GDP EU ) 0.87 (0.014) 0.92 (008) 0.98 (0.014) 0.93 (0.008) ln (GDP Migrant-sending country ) 1.11 (0.008) 0.81 (0.005) 1.15 (0.008) 0.84 (0.004) ln (Distance) -0.93 (0.016) -1.05 (0.009) -0.90 (0.016) -1.06 (0.009) ln (Immigrant flow) 0.25 (0.008) 0.19 (0.005) - - ln (Immigrant stock) - - 0.19 (0.07) 0.17 (0.004) Constant -26.35 (0.456) -18.4 (0.258) -30.59 (0.433) -19.75 (0.242) No of obs 21956 21956 27072 27072 No of groups 13 13 13 13 R 2 Within Between Overall Heteroscedasticity test(χ 2 ) 0.71 0.78 0.70 0.84 0.97 0.84 0.70 0.76 0.69 54.64 44.89 118.78 Note: * 10% significance level, ** 5% significance level, 1% significance level. C. Results of OLS for panel data with random effects TABLE 10. RANDOM EFFECT OLS REGRESSION RESULTS 0.84 0.97 0.84 18.85 (0.128) Dependent variables Independent variables Flow model Stock model ln (Import) ln (Export) ln (Import) ln (Export) ln (GDP EU ) 0.53 0.78 0.45 0.71 (0.021) (0.012) (0.020) (0.011) ln (GDP Migrant-sending country ) 0.768 0.66 0.71 0.70 (0.014) (0.008) (0.013) (0.01) ln (Distance) -1.09-1.22-1.06-1.20 (0.044) (0.025) (0.043) (0.026) ln (Immigrant flow) 0.17 0.11 (0.010) (0.006) - - ln (Immigrant stock) - - 0.23 0.15 (0.012) (0.007) Constant -7.75-9.70-5.02-8.58 (0.610) (0.354) (0.576) (0.333) No of obs 21956 21956 27072 27072 No of groups 2818 1818 3321 3321 R 2 Within Between Overall 0.13 0.70 0.68 0.38 0.84 0.82 0.10 0.71 0.67 0.39 0.84 0.83 Note: * 10% significance level, ** 5% significance level, 1% significance level.