Explaining Baltic Migration After EU Accession: Determinants and Consequences December 2015 Lydia Lamberty, M.Sc. University of Trier Abstract Emigration from the Baltic countries has been substantial to the extent that it is perceived as a threat to the reproduction of the population and to sustainable economic development. Integration into the EU’s single market contributed to the mobility of people and intensified the competition for young educated workforce. Based on push and pull factor analysis and human capital theory, I assess the determinants of Baltic migration and estimate bilateral migration flows from 1998 to 2013. Following a standard international migration model which focuses on relative (expected) income differences net of physical and psychological migration costs, I control for international economic linkages, indicators of the quality of governance and social environment, measures of productivity and innovative capacity as well as demographic factors. Across the Baltic countries, I find a varying significance of the determinants and derive consequences for Estonia, Latvia and Lithuania. Keywords: International migration, EU integration I Table of Contents List of Figures ......................................................................................................... II List of Tables......................................................................................................... III List of Abbreviations............................................................................................. IV List of Country Codes .............................................................................................V Introduction ............................................................................................................. 1 1 2 3 4 Theoretical Background to Analyze Migration............................................... 2 1.1 Migration Theory: Push and Pull Factors ................................................. 3 1.2 Human Capital Theory: What Happens to the Brain? .............................. 6 1.3 Labor Market Economics: Consequences of Migration ........................... 9 Describing Migration Behavior: Economic and Demographic Outlook....... 13 2.1 Migration Behavior of the Baltic Population ......................................... 14 2.2 Socio-Economic Determinants of Migration.......................................... 18 Explaining Baltic Migration After EU Accession ........................................ 23 3.1 Variables and Data to Explain Migration ............................................... 24 3.2 Estimation Results for Baltic Migration ................................................. 27 3.3 Comparing Migration by EU-Accession and Economic Crisis.............. 33 The Impact of Migration on the Baltic Countries ......................................... 35 4.1 Estimation Results for Estonia, Latvia and Lithuania ............................ 35 4.2 Human Capital and Returning Migrants ................................................ 39 Concluding Remarks ............................................................................................. 44 Appendix ............................................................................................................... 46 References ............................................................................................................. 57 II List of Figures Figure 1: EU8 emigration relative to population in per cent ................................ 13 Figure 2: Net migration plus statistical adjustment for natural change ................ 15 Figure 3: Share of the young (under 25 years) among Baltic population and emigrants .............................................................................................. 17 Figure 4: Real GDP growth rate and HICP inflation rate ..................................... 19 Figure 5: Trade volume and trade balance in % of GDP ...................................... 20 Figure 6: Unemployment rates overall and among the young (under 25 years) ... 21 Figure 7: Government expenditure and consolidated gross debt .......................... 22 Figure 8: A simple labor market equilibrium with skilled emigration.................. 48 Figure 9: Comparison of emigration and immigration figures ............................. 49 Figure 10: Immigrants to considered OECD countries by origin, sex and skilllevel ...................................................................................................... 53 III List of Tables Table 1: Estimation results for the Baltic panel fixed effects estimation, 19982013......................................................................................................... 29 Table 2: Estimation results for Estonia, Latvia and Lithuania, 2001-2011 .......... 36 Table 3: F-test results on full specification for the variable categories (I) – (V) .. 38 Table 4: Comparison of educational levels among movers and stayers ............... 42 Table 5: Estimation results for Baltic migration before EU accession, after, and during crisis ............................................................................................. 50 Table 6: Summary statistics .................................................................................. 51 Table 7: Correlation matrix ................................................................................... 52 Table 8: Shares of migrants per nationality in UK NINo numbers....................... 54 Table 9: Comparison of approximated Estonian migrant stock in the UK ........... 55 Table 10: Restricted access to labor market 2003-2011 ....................................... 55 IV List of Abbreviations APS Annual Population Survey in the UK CEEC Central and Eastern European Countries CEPII Centre d’Études prospectives et d’informations internationales CPI Corruption Perception Index EEA European Economic Area EFTA European Free Trade Association EU European Union ILO International Labor Organization LFS Labor Force Survey NINo National Insurance Number in the UK NMS New Member States of the EU OECD Organization for Economic Cooperation and Development ONS Office for National Statistics in the UK PPS Personal Public Service number in Ireland WGI Worldwide Governance Indicators V List of Country Codes CH Switzerland DK Denmark DE Germany EE Estonia EEA EU, Iceland, Liechtenstein, Norway EFTA Iceland, Liechtenstein, Norway, Switzerland ES Spain EU8 Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia, Slovenia EU15 Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, United Kingdom EU EU15, EU8, Malta, Cyprus, Bulgaria, Romania, Croatia IE Ireland IT Italy LV Latvia LT Lithuania NL Netherlands FI Finland SE Sweden UK United Kingdom NMS Cyprus, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovakia, Slovenia NO Norway 1 Introduction In 2004, the European Union (EU) accepted ten new member states among which there were eight Central and Eastern European Countries (CEEC) who had become independent in the beginning of the 1990’s. While migration in Europe after the fall of the Iron Curtain was characterized by political and social motives, migration in the new decade was more often economically motivated. The principle of free mobility of labor in the EU’s single market raised concerns of a massive influx of migrant workers from the new member states (NMS) to the old member states (EU15). The EU15 therefore chose to apply transitional arrangements in order to mitigate the potential labor supply shock by restricting work permits or allowing access only to sectors with unsatisfied labor demand. Yet, by proceeded European integration such as in tertiary education, motives for migration have become diverse, and disentangling temporary migration and permanent migration is crucial but possible only within limits. In the Baltic states Estonia, Latvia and Lithuania, net migration has been negative and substantial for over a decade. Not only do people more people emigrate, also the natural population change has turned negative. Only very few studies have considered Baltic migration separately from Eastern European new member states (EU8), e.g. Kancs and Kielyte (2002). Therefore, I ask what has determined emigration from the Baltic countries to Western European countries in the last decade? And how do the sending countries cope with emigration? Based on push and pull factors analysis, I estimate Baltic migration according to a standard migration model which focuses on relative (expected) income differences net of physical and psychological migration costs. In a cluster-robust fixed effects estimation, I control for international economic links, descriptive indicators of the social environment, measures of productivity and innovative capacity as well as demographic factors. Chapter 1 provides theoretical background to international migration such as identifying push and pull factors, human capital theory and labor market economics. Chapter 2 shortly describes the Baltic economies. Chapter 3 estimates and analyses the determinants of Baltic migration. Chapter 4 analyses the impact of emigration on the home economies in the light of brain drain concerns and derives policy recommendations before conclusion. 2 1 Theoretical Background to Analyze Migration A migrant is a person who establishes his residence in another country for a period that is, or is expected to be, of at least twelve months. An in-migrant or immigrant has previously been resident in another country, and an out-migrant or emigrant ceases to have his usual residence in that country (Eurostat, 2015). The United Nations’ definition also includes that the decision of an individual to migrate is taken freely, albeit under certain constraints, and distinguishes temporary labor migrants such as guest workers, highly skilled migrants or business migrants, irregular and undocumented migrants, forced migration such as displacement, family reunification, and return migrants (UNESCO, 2015). There are numerous economic, political, social and environmental reasons for migration. In the neoclassical framework on which standard international migration theory builds, migration is a macro-economic equilibrating mechanism of production factors where wage rates at origin and destination country are the main explanatory variables. On micro-level, migration is an individual decision of utility maximization stemming from regional disparities in expected prosperity and constrained by various migration costs. These disparities are represented as regional differentials in observed or expected income, unemployment rates or the likelihood of finding a job, the provision of public goods, and the extent of welfare benefits. The migration decision appears as optimization of the present value of expected earnings net of the cost of migration (Zaiceva and Zimmermann, 2008, p. 2; Schoorl et al., 2000, pp. 3f). A positive relationship between high wage levels and high immigration rates has often been demonstrated empirically but no equivalent negative relationship between low wage levels and emigration has been established so that the micro approach is usually preferred over the macro approach to explain migration (Borchers and Breustedt, 2008, p. 13). It has thus been extended to noneconomic motives of individuals such as cultural and linguistic proximity to the destination country or social networks in the destination country, which mitigate the psychological cost of migration (Massey et al., 1993). The individual’s decision is determined by the expected return to migration depending on characteristics such as age, education or skills, enabling the individual to benefit from migration and to cope with the adjustment into the host economy (Kahanec et al., 2010, p. 6). 3 Economic, political and social reasons to migrate can be divided into push and pull factors of migration which to some extent represent just two sides of the same coin, as will be elaborated in the next section. Individual characteristics which promote migration will be analyzed according to the human capital theory and issues related to brain drain thereafter. Finally, an outlook on the impact of migration on the home and host labor markets is given. 1.1 Migration Theory: Push and Pull Factors Push and pull factors can be categorized into economic, political, legal, social, demographic, cultural or ecological ones and often represent two sides of the same coin or, empirically, similar sized effects with opposite signs (Haas, 2008, p. 10; Mayda, 2010, p. 1251). The theory of push and pull factors traces back to Lee’s (1966) description of migration behavior in a framework of attracting and repelling factors, apart from indifference, at origin and destination, respectively. Migrants who respond to attracting (pull) factors at the destination tend to be positively selected, because they are under no pressure to leave and take informed decisions, whereas migrants who respond to repelling (push) factors at the origin tend to be negatively selected because they are not free to decide according to their preferences. With increasing obstacles to migration, positive selection becomes more distinct (Lee, 1966). Economic push factors inducing migration consist of high unemployment rates, low wages, insufficient education possibilities, poverty, or mediocre health care. Pull factors mirror the former and include greater job opportunity, higher wages, a lack of (un-)qualified workforce, higher standard of living, or the prospect of personal success (Borchers and Breustedt, 2008, p. 15). Harris’ and Todaro’s (1970) seminal paper demonstrated that migration proceeded in response to differences in expected earnings. Allowing for unemployment and introducing a minimum wage greater than the marginal product, a migrant worker chooses to maximize his expected utility by migrating regardless of unemployment. Because of the pool of unemployed, an additional job at minimum wage attracts more than one worker, and all but one worker therefore migrate to find themselves unemployed. The unemployment rate thus acts as an equilibrating mechanism whereas migration is a disequilibrium phenomenon (Harris and Todaro, 1970). 4 Political and legal push factors contain conflicts in the origin area, poor governance and corruption, while a stable political system, the rule of law, respect of human rights and protection of minorities attract migrants. Social insecurity, a lack of housing and property protection or environmental damages repel domestic citizens. Social or cultural pull factors, on the other hand, consist of a diaspora community abroad, sufficient education and training opportunities or family reunification (Borchers and Breustedt, 2008, p. 15). These objectively describable push and pull factors may deliver a convincing picture of determinants of migration in aggregate, but on an individual level, perceived differences and aspirations of quality of life exhibit a stronger explicatory power (Haas, 2008, p. 11; Schoorl et al., 2000). Furthermore, individual comparisons under peers may provoke feelings of relative deprivation and increase the propensity to migrate with second-round effects to those left behind (Stark and Bloom, 1985, p. 173). Hence, models of push and pull migration are used to put drivers of migration into a broader static picture and rather describe the propensity to migrate than explain causal links. Since they assume homogenous, perfectly informed individuals but do not restrain the analysis to an individual level, they are unable to explain why some people stay while others leave nor why in- and out-migration happen simultaneously, and why indeed no factor price equalization occurs to end migration (Haas, 2008, pp. 9f).1 It is useful to disentangle the costs of migration into different explaining factors. The network approach is often appreciated to expand the neoclassical framework because it offers an explanation why pull factors lose attractiveness over time, namely due to decreasing cost and risk of migration with an increasing number of migrants from the home country living in the host country. Networks or diaspora communities are a result of a steady flow of migrants from one country to another where some stay and become ‘stock’ migrants while others return home. Networks reduce the cost of migration physically because they help finding employment or housing, dealing with administrative requirements and so on. They also reduce the 1 A general formulation of the neoclassical migration model is given by Massey et al. (1993, p. 435) 𝑛 in continuous form: 𝐸𝑅(0) = ∫0 [𝑃1 (𝑡)𝑃2 (𝑡)𝑌𝑑 (𝑡) − 𝑃3 (𝑡)𝑌𝑜 (𝑡)]𝑒 −𝑟𝑡 𝑑𝑡 − 𝐶(0), where ER(0) is the expected net return to migration calculated before departure in t=0, P1(t) is the probability of deportation being 1 for legal migrants, P2(t) is the probability of employment abroad, P3(t) is the probability of employment at home, Yd(t) is income from employment abroad, Yo(t) is income from employment at home, C(0) is the sum of total costs, and r is the discount factor. Migration occurs if ER(0)>0 and the migrant goes to where ER is highest. 5 psychological cost of moving in terms of cultural or linguistic barriers to overcome, or ease the perceived distance to family and friends left behind. With every new migrant, the (open-access) network becomes increasingly self-perpetuating because any new migrant has a set of family members and friends to whom migration now becomes a less risky and cost-constrained option. Migration experience thus feeds back to potential migrants at home and migration becomes less selective as well as less correlated with wage differentials and unemployment rates. With intensified institutionalization of the network, the movement of people develops independently of the initial causes that drove the first migrants. As a consequence, governments retain only limited possibilities to regulate flows. In such a dynamic, self-sustaining framework, “migration alters the social context within which subsequent migration decisions are made” (Massey et al., 1993, pp. 448-451; Haas, 2008, p. 19f).2 Interpreting migration as a strategy of risk diversification, households rather than individuals appear as the interesting unit of decision-making to study. The new economics of migration theory assumes that migration decisions are taken collectively to maximize expected income and minimize risk through diversifying the allocation of household resources (Massey et al., 1993, p. 436). In the absence of social insurance schemes or constrained access to credit, the household’s optimal decision might be to send one member abroad as breadwinner or to invest in a member’s higher education abroad (Schoorl et al., 2000, p. 4). A policy changing the distribution of income, e.g. by entitlements to welfare benefits, can therefore also influence migration via other markets than labor (Massey et al., 1993, p. 440). Several studies estimated the migration potential of Eastern European countries which were about to enter the EU in 2004 relying on the aforementioned standard international migration model. Kancs and Kielyte (2002) estimated bilateral gross migration potential from the Baltic States to the EU using a neoclassical approach and found the expected impacts of relative per capita income, relative unemployment rates, bilateral distance and migrant stock, proxying migration cost 2 This approach emphasizes the need for a structural rather than reduced-form estimation of migration decisions (Stark and Bloom, 1985). Kancs (2010) rejects the reduced-form, because fixing explanatory variables such as wages a priori is likely to bias results due to endogeneity and reverse causality which is a particularly important issue in small open transition economies. He employs a new economic geography approach to explain labor migration in the enlarged EU. Kancs and Kielyte (2010a) point out that the standard reduced-form approach is the preferable choice for bilateral gross migration flows, which are the main focus of this thesis, although one should not be ignorant about the implicit assumptions of how migration affects the determinants of migration. 6 and network effects, respectively. Two reports on labor mobility commissioned by the EU, namely Brucker et al. (2009) and Holland et al. (2011), analyze the impact of the EU 2004 and 2007 enlargements on labor migration in detail. Both studies rely on a model with migrant stock as dependent variable and use relative wage or capital income, unemployment relative to the EU15 benchmark, migrant stock a year earlier, and several restriction dummies for transitional arrangements, guest worker schemes and others as explanatory variables. The first study found that the income gap still was a reliable explanation for intra-EU migration since despite accelerated convergence of wage levels and labor market conditions, capital endowments in the NMS remained below EU15 level. Geographical proximity has decreased in importance because of low cost carriers and because, due to networks, transportation costs decline with more migrants, making distance an endogenous explanatory variable (Brucker et al., 2009, pp. 9ff). The second study focused on macroeconomic impacts of labor mobility, the consequences of the transitional arrangements and the financial crisis. They found that notably in the Baltic States, persistently high emigration rates leave a permanent scar on the economic development in terms of a reduced output by 3-10% (Holland et al., 2011, p. 14). Summing up, the push and pull factor analysis based on the neoclassical framework successfully suggests expected income differentials and employment opportunities or unemployment threats and their interaction term as main determinants of migration net of migration cost, and has been enlarged to migrant stocks and network effects to explain the shape of migration flows. Individual characteristics such as age, experience or schooling have so far been left out but determine all the same the benefits from and returns to migration. 1.2 Human Capital Theory: What Happens to the Brain? A migration decision is, as mentioned above, an optimization of discounted net expected earnings in order to maximize the return to mobility. Migration is also an investment decision in human capital motivated by expected returns to higher productivity through mobility. Sjaastad (1962) provided a human capital model of migration presuming migration as investment in human capital via training, experience or language skills, which has costs and renders returns. Costs pose 7 obstacles while returns depend on occupation, age and sex. These characteristics account for (remaining) differences in earnings and limit the relevant alternatives between which a migrant decides (Sjaastad, 1962). Both migrants and non-migrants are affected by social cost-benefit differences arising from market imperfections. If a positive migration decision exhibits an excess of social cost over return, then nonmigrants are relatively worse-off and brain drain might constitute a problem. Borjas (1987) renewed Sjaastad’s human capital model of migration by applying the Roy model of self-selection into migration. In Roy’s model, individuals exhibit heterogeneous skills, normally distributed among the population, which enable them to choose their form of participation in the labor market upon a given set of technologies to employ their skills efficiently. This makes labor market outcome a function of a variety of individual choices and therefore endogenous (Roy, 1951). Yet, immigrants do not constitute a randomly selected sample of foreign-born. Borjas (1987) formulates this self-selection bias into migration based on migrants’ position in the earnings distribution before and after migration, and distinguishes positive self-selection and negative self-selection. In a policy setting at destination where income distribution is cushioned by relatively higher benefits for low-income workers via taxation of high-income workers, i.e. the host is more prone to equality, the selection of migrants by schooling and earning is negative. In the opposite case, where the home country protects low-income workers from poor labor market outcomes, the selection is positive and brain-drained. If mean income rises in the origin country where income distribution is more unequal, then the marginal immigrant no longer moves. But since he is more productive than the average immigrant, the average quality of the immigrant population in the destination country declines. The same effect is generated by a decrease in cost of migration, e.g. by abolishing barriers to mobility (Borjas, 1987; 1999). In the context of freedom of labor in the EU, migration intimately relates different immigration preferences and outcomes in sending and receiving countries. Receiving EU member states have an incentive to actively recruit skilled immigrants because of a lack of adequate home workforce and a declining and ageing population (Brauner, 2010, p. 237). Due to free mobility, they cannot, however, select immigrants according to their needs so that, at the bottom line, the market decides on the allocation and composition of human capital at home and abroad. To the sending countries, the loss of well-trained individuals, the brain 8 drain, is an important issue and has severe consequences on economic development. It has been argued that the complement to brain drain is brain gain considering that emigration of high-skilled due to better earnings prospects incentivizes nonmigrants to invest in their education so that human capital formation is stimulated (Stark et al., 1997; Marchiori et al., 2009; Brauner, 2010, p. 232).3 In development economics theory, negative skilled emigration effects can be offset by remittances, foreign direct investments (FDI) from diaspora communities, transfers of knowledge and technology, and also returning migrants and brain circulation. Although intra-EU migration involves developed high-income countries, remittances sent by emigrants, invested or consumed, still increase the relative satisfaction in comparison to peer households which again induces migration of then relatively deprived households (Stark and Bloom, 1985).4 Brucker at al. (2009, pp. 89-103) do not see a general brain drain from NMS to the EU15 but find evidence for slightly positive rather than negative self-selection, because relatively more tertiary educated people left. Yet, the picture at sectoral and occupational level differs, notably in construction and health care, so that intraoccupational mobility choices explain inter-occupational mobility divergence, i.e. why some stay while others leave (Sjaastad, 1962). Particularly for the Baltic countries, however, brain drain has some relevance. The authors confirm brain gain in terms of increased investment in education even above EU15 level, but also brain waste because returns to education for NMS migrants in EU15 have been very low given that migrants tend to select into low-qualified jobs where their human capital depreciates.5 A migrant whose primary goal is to earn money will decide on absolute instead of relative wages so that a rational choice to him might include taking a deskilled job, or any readily available job because of a preference for temporary work (Brucker et al., 2009, p. 103; Holland et al., 2011, p. 36). Migrants 3 Mayr and Peri (2009) analyze brain drain from Eastern to Western Europe and find that an increased probability of migration of 1-20% may add about one year to average schooling, which they suppose likely to offset the brain drain effect. 4 Whether skilled emigrants remit more because they earn more (wage effect), or remit less because they stay longer and invite their family members (reunification effect) is a disputed topic in the literature (Faini, 2006; Brauner, 2010). 5 Brucker et al. (2009) find that Polish migrants have been positively self-selected to the UK and negatively to Germany which supports the idea that networks reduce migration cost. The authors also estimate the return to education which NMS migrants receive in the UK and find that, e.g., 36% of NMS migrants arriving after 2004 have been employed in elementary occupation despite their 15 years of education compared to 1% of UK-born, and that recent NMS migrants earn 42.5% less than natives, although wage gaps decline when controlling for occupation. 9 differently prioritize earning money, gaining skills and professional qualification, establishing networks and status. Recent intra-EU mobility has not been permanent but features short- and medium-term moves which are characterized by a high degree of brain circulation and desirable skill transfer offering a win-win situation (Holland et al., 2011, pp. 27ff; Hazans and Philips, 2010). NMS migrants have predominantly been young and relatively high-skilled. Concluding on what happens to the migrants’ ‘brain’ raises the question about transferability of qualifications in the EU (Kahanec et al., 2010). In line with the theory, migrant-sending NMS in the short run experienced lower unemployment rates and higher employment growth rates, more vacancies due to labor shortages, and higher wages as well as higher unit labor costs, while incoming capital flows such as remittances and FDI increased (Kahanec et al., 2010, pp. 3237; Brucker et al., 2009, pp. 71ff)). In the long run, the NMS have to cope with a decline in innovative capacity next to demographic challenges which result from the loss of working-age population and increased dependency ratios. To them, it is crucial that the ‘brain’ is circulating (Kancs and Kielyte, 2010b). 1.3 Labor Market Economics: Consequences of Migration It has been argued that immigration has been rather demand-driven due to a structurally unsatisfied labor demand in the EU15 despite unemployment so that, in fact, immigration is to the largest extent labor migration.6 Pull factors of migration can be considered demand-driven while push factors are supply-driven. According to the dual labor market theory based on Piore (1979), pull factors in receiving countries are more important drivers of migration than push factors in sending countries because any developed, modern economy has a structural demand for new workers “at the bottom of the social hierarchy, who will accept low wages and a lack of social mobility” (Schoorl et al., 2000, p. 4). The duality stems from a two sector approach accounting for the relation between capital and labor, i.e. jobs 6 The EU LFS ad hoc module on the labor market situation of migrants conducted in 2008 displayed that among all immigrants, work and family were the main reasons for migration, yet, less pronounced for EU27 immigrants. By cross country average, 30% of men migrated because of work and almost 50% of women because of family reasons, i.e. having a relative or spouse working abroad. Work as reason for migration was distinguished in whether migrants already had found a job or not, where the latter case was more frequent with 23% compared to 15% (Eurostat, 2015). 10 in the capital-intensive sector are more stable and require higher skills, whereas jobs in the labor-intensive sector are unstable and unskilled. Since natives accept unstable, low-paid jobs to a lesser extent, immigrant workers fill the need. As a consequence, employers’ recruitment mechanisms appear to induce immigration more than individual efforts so that the policy levy remains with employers (Massey et al., 1993, pp. 440ff). Assuming different skill-levels, direct effects of migration in the labor market depend on the substitutability and complementarity of migrant and native labor as well as low-skilled and high-skilled labor via wages (in a competitive market) and unemployment (under rigid union wages) (Kahanec et al., 2010, p. 7). At origin, skilled emigration exerts upward pressure on high-skilled wages and reduces wages or employment of the low-skilled, where the latter feeds back into reduced demand for high-skilled mitigating the upward wage pressure. Unskilled emigration increases low-skilled wages or employment, which has a negative effect on high-skilled wages. At destination, skilled immigration benefits unskilled natives and skilled natives may suffer, whereas unskilled immigration hurts unskilled natives while skilled natives may benefit.7 Indirect effects outside a static framework alter direct effects, for example, when incentives for schooling increase due to a larger skilled-unskilled wage ratio so that welfare considerations become very complex (Kahanec et al., 2010, pp. 7ff; Zimmermann, 1996, p. 113). In the EU15, NMS immigration has had little effect on wages and employment in the short run and no effect in the medium to long run and, moreover, no negative impact on the welfare system (Kahanec et al., 2010, p. 30; Holland et al., 2011, pp. 28ff). Indeed, modest negative effects of immigration in the EU15 labor markets have been outweighed by strong positive effects in the goods and capital market due to the market integration (Brucker et al., 2009, p. 57).8 Before EU enlargement, numerous studies, which were conducted via extrapolations from various historical data and surveys of migration intentions, estimated a NMS emigration potential of 1-4% of the population, including commuting up to 14% (Zaiceva and 7 Since the 1970s, a vertical labor supply curve can be assumed so that, when a strong trade union fixes real wages above equilibrium, push migration shifts the labor supply curve right, increasing unemployment and thus fiscal pressure of compensation payments, as demonstrated in the appendix (Zimmermann, 1996, p. 97). 8 On the other hand, under free trade it is likely that most immigration impacts on host labor market are observed also in the absence of immigration (Borjas, 1999, p. 48). Still, it has been pointed out that benefits from opening labor markets to migration can well outperform potential gains from goods and capital market liberalization (Baas et al., 2010, p. 48). 11 Zimmermann, 2008, pp. 5f).9 Worried that unrestrained worker inflow from NMS after accession to the single market would dampen public acceptance, not all EU15 countries opened access to their domestic labor markets at the same time. So-called transitional arrangements for up to seven years were introduced by all but three member states in 2004 and not fully lifted for the EU8 until 2011, when the maximum period expired. Mayda (2010) investigates on migration quotas as binding constraints and estimates the impact of push and pull factors in terms of supply- and demand-side effects based on the OECD’s migration database. She finds that if quotas bind, push and pull factors have no or only small effects on emigration rates, and that once restrictions are relaxed, push effects turn negative while pull effects turn more positive. Pull effects such as per worker GDP have a more significant impact, while push effects such as demography, in particular the share of young people at home and geographic distance as proxy for migration cost, only shape freely moving migration flows. Cultural variables seem rather irrelevant in the OECD context, but relative inequality and network effects appear important as both push and pull factors (Mayda, 2010, pp. 1251ff). Regardless of whether transitional arrangements eventually inhibited push (supply-side) migration or pull (demand-side) migration, there is evidence that they diverted migration flows from traditional immigration countries, notably Germany and Austria, towards open access countries, such as the UK and Ireland.10 Fihel et al. (2015) analyze the impact of the transitional arrangements on the free movement of workers in the EU and find that due to the diversion, those countries opening the labor market in 2004 benefited to a larger extent because of positively selected migrants.11 In essence, the imposition of the transitional arrangements, although still understudied, affected the allocation of human capital but not necessarily the absolute numbers of people in the destination countries (Fihel et al., 2015, p. iv). Holland et al. (2011, p. 27), on the other hand, argue that they may have encouraged some brain drain because they posed an obstacle to low-skilled migrants which may have been prohibitive to many. 9 Brucker et al. (2009, p. 37) provide a comprehensive literature review. Kancs (2010) criticizes the methodological overestimation of most studies based on reduced-form approach which imply continued migration regardless of its reverse effects on wages, income or employment. 10 In 2003, Germany and Austria hosted two thirds of EU8 migrants but after the EU8 accession, the UK and Ireland absorbed up to 70% (Baas et al., 2010, p. 51). 11 The exact form of the arrangements has been up to national policy-makers and despite denied free access, self-employment, commuting and posting were usually excluded from the restrictions. 12 In the Baltic countries, effects of emigration have been most pronounced in Lithuania, which lost up to 6% of its GDP over the period 2004-2009 due to emigration compared to 3% in Estonia and Latvia with age-adjustment for the 15 to 64 year olds (Holland et al., 2011, p. 63). These effects are much larger than in the other NMS. Short run effects on unemployment and wages have been positive and small, with 0.3% largest in Lithuania, whereas long run effects are close to zero (Brucker et al., 2009, pp. 70ff; Holland et al., 2011, p. 64). In all Baltic countries, positive short run effects by skill-level have been greatest for low-skilled unemployment, while positive effects on wages were similar across skill levels. The long run effects are close to zero. The early restrictive setting yields a negative effect on change in labor force and GDP but small positive effects on wages and unemployment compared to free movement (Brucker et al., 2009, pp. 70ff). Later on, Holland et al. (2011, pp. 87ff) deny any effect of transitional arrangements on the location decision of migrants and argue that a shift in migration patterns had already started prior to the enlargement. The financial crisis starting in 2008 has had an impact on migration via two channels, namely as an end to accession-driven migration, which slowed down by two thirds, and as re-orientation of migrants towards better faring EU15 economies (chain migration), i.e. from the UK towards Germany (Holland et al., 2011, pp. 99ff). Latvia was the only NMS to experience even larger emigration waves attributable to an exceptional economic deterioration. The financial and economic crisis inspired various studies on return migration in the EU alongside return premium considerations (Galgoczi et al., 2012).12 Return incidences differ across host and home country combinations, i.e. there was no mass return to the NMS. Baltic citizens became even more mobile not only returning but also (re-) migrating, giving rise to circular migration (Zaiceva and Zimmermann, 2012). Incentives to return include higher purchasing power at home and relatively improved employment, earnings, living or general economic conditions as well as expected gains from having worked abroad (Hazans and Philips, 2010, p. 282). 12 Jauer et al. (2014) analyze migration as equilibrating force during recession in the USA and Eurozone based on optimal currency area arguments and find that third country nationals exhibited the highest mobility. Pungas et al. (2012) find for Estonian migration (to Finland) that the intention to return and the educational level are not correlated, but that over-education on any level increases the propensity to return. Mayr and Peri (2009) find that a return premium which increases with education attracts high-skilled return migrants, which further boosts the education incentive, but a lump-sum premium results in negatively selected returnees. Iara (2008) finds a wage premium for Eastern European citizens who worked in Western Europe of up to 30%. 13 2 Describing Migration Behavior: Economic and Demographic Outlook Among the EU8, Poland, Czech Republic and Lithuania recorded the highest emigration numbers. Poland, for example, has been reporting a quarter of a million emigrants per year since 2009. Latvia and Lithuania counted the largest outflows relative to their populations, peaking in 2010 at 40,000 Latvians and 83,000 Lithuanians emigrating, while Estonia experienced a steadily rising number of emigrants from 3,000 in 2003 to 6,700 in 2013. Figure 1 illustrates EU8 emigration relative to the domestic population in the reference year. Figure 1: EU8 emigration relative to population in per cent 3,00 2,50 2,00 1,50 1,00 0,50 0,00 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Czech Republic Estonia Latvia Lithuania Hungary Poland Slovenia Slovakia Source: Data retrieved from Eurostat (2015). The pattern of Lithuanian, Latvian and Estonian migration shows some peculiarities. Lithuanian emigration seems to follow Latvian emigration with more pronounced changes to previous years and more peaks, namely in 2001, 2005 and 2010. The accession to the EU in 2004 seems to have been most important for the emigrants’ positive decision in Lithuania, affecting the 2005 emigration. In Latvia, the 2004 emigration figure is highest (until 2008), yet lower than before, so that EU accession did not trigger massive outflows.13 Estonian emigration is hardly volatile 13 In the CEEC Eurobarometer prior to EU accession, 2-3.5% of Lithuanian, Latvian and Estonian respondents indicated intentions to move abroad within the next five years, compared to less than 1.5% in the other NMS. The post-accession Eurobarometer revealed that Baltic and Polish respondents were even more prone to migrate at 5-9% of respondents. In line with the high migration intentions, 80% of Latvian and Lithuanian respondents perceived migration outflows as important as opposed to limited (Zaiceva and Zimmermann, 2008, p. 26). 14 with only small rises in 2003 and 2005 by around 50% to the previous year. It has increased rather consistently, which might be due to the preference for Finland which receives 70% of Estonian migrants (Eurostat, 2015). Hence, Baltic migration as such effectively does not exist and aggregation blurs the different migration patterns. For example, the Finnish socio-economic situation might be an important determinant of Estonian migration, since Estonian migration did not increase in 2004 but in 2006, when Finland opened its labor market. Lithuanian emigration in 2005 might not be linked to EU accession as much as it is linked to the open labor markets in the UK and Ireland. Furthermore, for Latvia and Lithuania, a similar socio-economic situation might induce similar migration behavior, and either country’s migration behavior might influence the other’s. The Baltic countries underwent similar economic and social changes which resulted in similar socio-economic environments but not so similar migration behavior as presented in the following. In order to describe Baltic migration, it is constructive to evaluate whether the countries have overcome the financial and economic crisis. 2.1 Migration Behavior of the Baltic Population The accession of the EU8 in 2004 was unprecedented in scale and scope, and opened a new chapter for European market integration, in particular accelerated labor mobility. Alvarez-Plata et al. (2003, p. 7) point out that the CEEC’s income as GDP in purchase power relative to the EU15 was half the ratio of Greece, Spain or Portugal at the point of accession. In fact, the CEEC income level was comparable to the Southern Europeans’ 1960s level of GDP. Over the three decades between these enlargement rounds, structural unemployment appeared, the demographic pyramid transformed, education levels ameliorated, and the CEEC underwent political and economic transition (Alvarez-Plata et al., 2003, p. 11). In the 1990s, in the aftermath of the dissolution of the Soviet bloc, Baltic migration was coined by ethnic belonging since many citizens of former Soviet states, notably ethnic Russians, emigrated (Sipaviciene and Stankuniene, 2013, pp. 46ff). There is still a large Russian-speaking population in Estonia and Latvia of about one third in either population. The propensity of the Russian-speaking population to migrate is estimated higher by national studies but ethnicity is not considered separately in 15 international studies. The Russian-speaking ‘minority’ is not separated in the migration data if they possess Baltic citizenship, which, however, roughly a fifth of the Russian-speaking minority does not. Hazans and Philips (2010) find that the citizenship effect is negative, i.e. minority non-citizens are much less likely to move. Yet, minority citizens were over-represented in the migrant workforce before EU accession, in Latvia at 56%, but less so after accession (45%), although still largely over-proportional (Hazans and Philips, 2010, p. 294; Hazans, 2012). Not only in the 1990s, but also in the past 15 years, emigration has been significant for the small Baltic populations. The Lithuanian population decreased from 3.5 million people in 2000 to 2.9 million in 2014. Latvia lost 380,000 inhabitants and counted 2 million people in 2014. Estonia fared comparatively well with a loss of 85,000 citizens. It still is the smallest Baltic country with 1.3 million people. Since 2004, the population loss has accelerated in Lithuania and Latvia whereas it has slowed in Estonia, resulting in an overall decrease in population of 12-13%, and 4% in Estonia (Eurostat, 2015). As figure 2 shows, net migration rates have been negative not only since 2000, and emigration is accountable for 50-80% of the population loss as opposed to the natural change in population. The natural change, too, is negative and may well be related to the outflow of the younger working age population who is also of reproductive age. The demographic development towards an ageing population in the Baltic states is in line with the Western European changes, although it is a more recent, severe phenomenon in the Baltic countries. Figure 2: Net migration plus statistical adjustment for natural change 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 10.000 0 -10.000 -20.000 -30.000 -40.000 -50.000 -60.000 -70.000 -80.000 Estonia Source: Data retrieved from Eurostat (2015). Latvia Lithuania 16 The top migration destination for Estonians is Finland, which counted more than 40,000 Estonian immigrants (flow) over the past 15 years and as many residents (stock). For Latvians, the UK was by far the most important destination country: 142,000 people over the period 2004-2013, and 89,000 residents. Germany and Ireland each host 20,000 to 24,000 Latvian residents. The UK, Ireland and Germany were also the most important destination countries for Lithuanian migrants. 250,000 Lithuanians immigrated into the UK (received insurance numbers) and 82,000 respectively 63,000 to Ireland and Germany.14 It is striking that for the UK, Ireland and Germany the flow-stock-ratio is much greater than for other important receiving countries, i.e. 2.3 for Ireland, 1.9 for Germany and 1.7 for the UK compared to an unweighted average of 1.2 for the other receiving countries. The ratio is lowest (almost 1) for Finland, Switzerland and Spain, suggesting moves of more permanent character. The distribution of sex, age and education among migrants is not always in line with the distribution among the home population over the period 2004-2013, which figure 3 illustrates. In the Baltic societies, more citizens are women, increasingly fewer citizens are under 25 years old, and education levels have improved. In Estonia, women are slightly overrepresented among emigrants, especially among the young emigrants under 25 years of age (52-60%), based on national emigration statistics (Eurostat, 2015). Since 2008/2009, almost a third of Estonian emigrants have been younger than 25 years, and nearly half have been younger than 29, suggesting that university graduates and people in early career stages are relatively prone to go abroad. Their share is also more volatile compared to the age groups of 30 to 45 years (30% of emigrants) or 45 to 60 (15%), which supports the argument of migration as investment rather than the breadwinner model. In Latvia and in particular Lithuania, the emigrant gender gap is more balanced around 50%. The age structure of Latvian emigrants seems similar to Estonia, i.e. one third is younger than 25 and half is younger than 29 years, but no age data is available prior to 2010. In Lithuania, the age distribution among emigrants is more heavily biased towards young people with the distinction that since 2008/2009, the share of the under-25 14 As will be explained in section 3.1, no immigration data is available for the UK and Ireland. Following the literature, I use individually attributed public insurance numbers. Although they are by definition unique, there is suspicion for over-estimation. Immigration figures reported by receiving countries, as opposed to emigration figures reported by sending countries, are therefore not directly comparable, but offer upper and lower bounds due to forgone (de-)registration. 17 year olds has not increased but dropped below 40%, and below 60% for the under29 year olds compared to the years before, due to a greater share of the older age groups (Eurostat, 2015). This puts forward the breadwinner approach arguing that a household’s migration decision follows a risk diversification strategy especially in times of trouble such as the economic and financial crisis. Figure 3: Share of the young (under 25 years) among Baltic population and emigrants 60% 50% 40% 30% 20% 10% 0% 2004 2005 2006 2007 2008 2009 2010 2011 U25 pop EE U25 pop LV U25 pop LT U25 emi EE U25 emi LV U25 emi LT 2012 2013 Source: Data retrieved from Eurostat (2015). The share of tertiary educated people among the working age population has increased by 7 to 10 percentage points in the Baltic countries since 2004 to above EU15 average, i.e. 32% in Estonia, 30% in Lithuania and 27% in Latvia and the EU15. The share of upper- and post-secondary educated people has been decreasing since 2004 to 52-56%, because more people went on to tertiary education, but remains above EU15 average by more than 10 percentage points (Eurostat, 2015). It is striking that among the working age population, women are 1.5 to 2 times more often tertiary educated in all Baltic countries. Among the young population of 20 to 24 years, more than 70% had completed upper- and/or post-secondary education in 2013, with a strong bias towards men in Latvia and Estonia but equal gender shares in Lithuania (Eurostat, 2015). Unemployment by educational attainment in 2013 was four times higher, in Lithuania even six times, among lower educated than higher educated persons. From the early 2000s until the financial crisis hit, unemployment was falling in the Baltic countries. In 2008, it increased and peaked in 2010 at almost 20%. So far, unemployment has by no educational level fallen to pre-crisis level (Eurostat, 2015). 18 According to the reported immigration statistics by the receiving countries, which exclude data for Germany, Ireland and the UK, registered immigrants were mostly female, however, exhibiting a certain pattern. The Scandinavian countries mostly received male immigrants whereas the Southern European countries Spain, Italy, Switzerland and also the Netherlands mostly received female immigrants by 6080%. Again, the migrant gender gap is more balanced for Lithuanians and more pronounced for Estonians and Latvians. Ignorant of the respective gender shares in the three missing countries, it appears as if smaller bilateral distances and high income favor male (work) migration.15 In terms of the immigrant age profile, Denmark and the Netherlands report that 50-60% (68% for Estonians in Denmark) are under 25 years old. In the receiving countries, the young are over-represented compared to their share in the home population of roughly 30% (Eurostat, 2015). Alongside economic growth, there is a rise in migration once a restraining threshold has been overcome, i.e. when more less wealthy and skilled people have earned enough money to migrate. If migration increases, networks abroad increase, which again induces less selective migration. However, this is only temporary if economic development of the home country continues so that a migration hump, a concave shape of migration flow is literally visible in the data (Haas, 2008, p. 16). Lithuanian and Latvian emigration graphs point towards such a hump, while the Estonian emigration curve remained flat. Whether this is related to an income threshold overcome or to socio-economic factors is subject to discussion in the next section. 2.2 Socio-Economic Determinants of Migration In 1998, while the Baltic countries were pursuing economic transition and integration into Western Europe in accordance with the guidelines of the so-called Washington Consensus, the Baltic economies were severely hit by the Russian crisis and had to re-orientate their exports towards Western Europe (Muravska, 2011). With the beginning of the 2000s, the Baltic economies experienced an economic boost, triggered by EU accession aspirations and actual market 15 Commuting is explicitly exempt from the data. It would be interesting to conduct a regression for gender specific migration with an interaction term of bilateral distance and per capita income as well as marital status and compare the results to commuting patterns, possibly using the framework of new economics of migration. Due to limitations in scope and time, this cannot be pursued. 19 integration. The ‘Baltic tigers’ attracted foreign investment, while externallyfinanced domestic demand increased, consumer prices rose, and the intensified economic activity generated GDP growth rates between 6% and 12% per year. Figure 4 illustrates the course of real GDP growth and inflation in the Baltic countries compared to the EU28 average growth rate. Income differences decreased as GDP grew from 40% of the EU28 in 2000, to 55% in 2005, and 75% in 2013 in Estonia and Lithuania, while Latvia’s slower catch-up resulted in 62% of EU28 GDP (Eurostat, 2015). Consumer prices increased but wages and salaries increased by 10 percentage points more, based on the Baltic averages of the years 2004-2008, which boosted labor costs. Productivity, however, did not increase (Eurostat, 2015; Purfield and Rosenberg, 2010). Figure 4: Real GDP growth rate and HICP inflation rate 15 10 5 0 -5 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 -10 -15 inflation EE inflation LV inflation LT GDP growth EE GDP growth LV GDP growth LT EU28 growth Source: Data retrieved from Eurostat (2015). The financial crisis in 2008 hit the Baltic economies so hard that in 2009 a loss of GDP of almost 15% was recorded. The inflationary pressure by the previous little consolidated economic growth turned into the opposite. Since the Baltic economies were on the way into the Eurozone, the decision-makers reacted to the crisis not by exchange rate adjustments, but maintained the currency peg. They resorted to severe internal devaluation, namely contractionary fiscal and nominal wage policy both in the public and private sector (Purfield and Rosenberg, 2010). As a consequence, the economies recovered quickly to positive growth rates, mitigating the post-crisis rebound, however, to below pre-crisis level. The impact of the crisis is also visible in indicators of international economic links, such as capital and trade flows. Already prior to EU accession, Estonia successfully 20 attracted foreign direct investment (FDI) until a peak at 22% of GDP in 2005 and subsequently slowly decreasing with a drop to pre-accession levels in 2008. Latvia and Lithuania attracted around 5% of their respective GDPs in the early accession years but experienced disinvestments in 2009. Recent inflows remain below precrisis level (World Bank, 2015). In general, investors seem less willing to commit by assets to the Baltic real economy. Bilateral FDI statistics for the considered migrant receiving countries show that Sweden is the most important investor in the Baltic countries, followed in Estonia by Finland, the UK and the Netherlands, in Latvia by Denmark, and in Lithuania by Germany and Denmark (UNCTAD, 2014). Considering the flow of foreign goods rather than capital, domestic demand has increased with EU accession, but Baltic exports to the EU have been decreasing since from shares of 80% to 65% so that the trade balance deteriorated, as figure 5 shows. Estonia and Latvia import 75-80% from the EU, and 20-25% from extraEU countries. Lithuania imports less, around 60%, from the EU. On country level, however, the Russian Federation is the most important trading partner for the Baltic economies (UN Comtrade, 2015). Figure 5: Trade volume and trade balance in % of GDP 25 15 150 5 100 -5 50 -15 0 Tade balance Trade volume 200 -25 1998199920002001200220032004200520062007200820092010201120122013 Balance EE Balance LV Balance LT Volume EE Volume LV Volume LT Source: Data retrieved from World Development Indicators (World Bank, 2015). The trade volume rose only modestly, i.e. EU accession did not shock trade relations but intensified economic links until the drop in 2009. The rebound in trade volume after the crisis exceeds previous, but the formerly negative trade balance is closer to zero since 2009, which hints at a change in trading partners away from troubled EU economies (World Bank, 2015). Again, considering the relevant migrant receiving countries in terms of bilateral trade flows, the main trading partners for 21 Estonia are Finland, Sweden and Germany, Germany and the UK for Latvia, and Germany, Sweden and the Netherlands for Lithuania (UN Comtrade, 2015). The national ramifications of the financial and economic crisis are also seizable in the unemployment rates; their development is depicted in figure 6. Until EU accession, the rates were above EU28 average although decreasing to lower-thanaverage unemployment until the dip in 2007. As a consequence of the crisis, unemployment rose to 16.7% in Estonia, 19.5% in Latvia and 17.8% in Lithuania, while the unemployment rate for the young population under 25 years amounted to 35% (Eurostat, 2015). Emigration mitigated the effect on the labor market. Unemployment is at accession level again, suggesting that the crisis has been overcome and that two-digit unemployment rates, notably in Latvia and Lithuania, are structural and linked to weak productivity. The discrepancy to youth unemployment has decreased, while their share in the population decreased and roughly 30% of under 25 year olds emigrate, which diminishes (future) labor supply but also relieves unemployment pressure (Eurostat, 2015). The flexibility of the Baltic labor markets has been appreciated by the IMF as key to a successful exit from the crisis, acknowledging the high social cost borne. Households, in particular the vulnerable and public sector employees, had to accept sharp decreases in income not only because of lower wages and higher unemployment but also because of reduced social services and benefits (Purfield and Rosenberg, 2010). Figure 6: Unemployment rates overall and among the young (under 25 years) 40,00 30,00 20,00 10,00 ,00 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 ue25 EE ue25 LV ue25 LT UE EE UE LV UE LT UE EU28 Source: Data retrieved from Eurostat (2015). The financial and economic crisis actuated the Baltic countries to increase government spending by 10 percentage points from 2007 to 2009 in order to bolster 22 the economic downturn, but government spending was restructured towards economic and financial sector activity. Latvia had to take on an IMF loan and, despite internal devaluation measures, Latvia and Lithuania ran a government deficit of 9% of GDP in 2009. Public debt, which is still far from reaching the Eurozone’s permissible 60% of GDP, doubled with the onset of the financial crisis in Latvia and Lithuania. Figure 7 illustrates government expenditure and debt. Governmental expenditure to provide public goods and services adds up to 37% of GDP across the Baltic countries, which is ten percentage points below EU28 average (Eurostat, 2015). In 2011, Estonia introduced the Euro, while the accession of Latvia and Lithuania was postponed to 2014 and 2015, respectively. The fiscal and monetary policy paved the way of successful macroeconomic stabilization as correction to the bubble that had inflated, but structural reforms are needed to make sure that labor productivity and competitiveness can ensure growth. Figure 7: Government expenditure and consolidated gross debt 50,00 40,00 30,00 20,00 10,00 ,00 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Gov EE Gov LV Gov LT Debt EE Debt LV Debt LT Source: Data retrieved from Eurostat (2015). The financial and economic crisis did not trigger massive return migration of the emigrants of the booming years since migrants rather took on a ‘wait-and-see’ strategy (Zaiceva and Zimmermann, 2012). According to the Labor Force Survey 2009/2010, roughly 80% of Latvian, 65% of Estonian and 60% of Lithuanian returnees were employed abroad, while additional 10% of Estonians were students and 10% of Lithuanians were unemployed. Returnees to Estonia and Lithuania, based on LFS figures, were less frequently high-skilled than stayers but more frequently than migrants, hinting a positive selection and brain circulation. For Latvian returnees, the selection was slightly less positive compared to both stayers 23 and migrants (Zaiceva and Zimmermann, 2012, pp. 22ff). The authors estimated a higher return probability for men, singles, older people, high- and medium-skilled, and people without children. In particular, the positive impacts of education, complemented by observed self-selection of migrants into lower qualified jobs, point towards brain circulation, although depreciated, rather than brain drain (Zaiceva and Zimmermann, 2012). The premium of having worked abroad for skilled migrants upon return is crucial to consider (Iara, 2008). Furthermore, people who have already worked abroad are more inclined to go abroad again. Returning to the above described migration model and the extensions stemming from push and pull factors, the estimation and analysis of Baltic migration is conducted in the following. 3 Explaining Baltic Migration After EU Accession Migration can be traced as emigration out of the origin country or immigration into the destination country. Emigration figures reported by national offices rely on population registers and household surveys and offer a lower bound of migration estimates, because many emigrants do not deregister when they leave.16 Emigration statistics are not as reliable as immigration statistics to the receiving countries which issue work permits or legally require immigrants to register, for example, to be entitled to benefits. In order to trace Baltic migration by immigration statistics of the receiving countries, I limit the scope of analysis to the most important destination countries. 70-90% of Estonian, Latvian and Lithuanian emigration was directed to eleven receiving countries: Denmark, Finland, Germany, Ireland, Italy, the Netherlands, Norway, Spain, Sweden, Switzerland and the UK.17 By immigration figures, they account for 78% of Estonian migrants, 93% of Latvian and 66% of Lithuanian migrants, despite substantial data gaps for UK, Ireland and The deregistration effect contributes for example to Lithuania’s peak in emigration rates in 2010 when the government introduced a statutory duty for all permanent residents to pay health insurance (Sipaviciene and Stankuniene, 2013, p. 49). 17 Applying the concept of next usual residence also shows that for three quarters of Estonian and Lithuanian migrants, the considered European countries were indicated as next usual residence over the period 2004-2013; no data is available for Latvia. 16 24 Germany (Eurostat, 2015).18 Based on the available Eurostat data, these eleven European destination countries comprise the set of relevant Baltic migration alternatives to a reliable extent. In the following, migration is thus analyzed in terms of immigration of Baltic citizens into the relevant European destination countries.19 Hence, I call an emigrant leaving a Baltic country and a Baltic immigrant arriving in a European country simply a migrant, distinguishing Estonian, Latvian and Lithuanian citizenship.20 3.1 Variables and Data to Explain Migration The dependent variable is the bilateral flow of migrants from a Baltic country to a European country relative to the Baltic home population. Although migration intentions of stay, which is by definition at least 12 months, and actual duration are not observable in the data I use, there is evidence that work is the main driver of migration. The 2008 ad hoc module on migrants in the labor market showed, that most intra-EU migration is due to work, and that also family reunification is related to relatives having found a job abroad.21 There are arguments for explaining the migrant stock instead of the migrant flow, as Brucker et al. (2009) and Holland et al. (2011) did, which comprise long-term policy objectives, social acceptance or comparability with third country nationals who move less freely (more costly) and intent to stay longer. In contrast to the stock-approach, I decided for the flowapproach because of the specific features of EU migration such as freedom of mobility and choice of work location, which calls for more flexible analysis. The response of workers, which is migration, may suggest a higher elasticity of job opportunity in the integrated European market. 18 Considering the UK data gap is important, since 37% of Lithuanians indicated the UK as country of their next usual residence over the period 2004-2013. Furthermore, the UK population survey of 2013 reports 161,000 Lithuanian residents, making it the 5th most common nationality among the non-British population; 1st is Poland since 2007 with 736,000 nationals in 2013 (British Office for National Statistics, 2013). 19 The differences in immigration and emigration numbers are presented in the appendix. 20 Eurostat data by citizenship is available since 1998, whereas data by country of birth is only available since 2008. The citizenship concept is not time-invariant, because naturalizations are not counted. However, the rate of naturalization among EU citizens in EU member states is negligible. 21 Statistics on labor mobility in the EU, which show the extent of work migration compared to total migration, are in detail provided by the EU Labor Force Survey (LFS). Unfortunately, the LFS data is only very limitedly available online, i.e. without distinction of the migrant worker’s nationality by country; the detailed microdata is not freely accessible. 25 In order to trace bilateral migrant flows, I rely on immigration statistics of the selected receiving countries by nationality of the Baltic migrants as available from Eurostat (2015). Neither the UK nor Ireland report immigration statistics so that individually appointed national insurance numbers (NINo) in the UK and personal public service numbers (PPS) in Ireland are employed as a count of immigrated workers, which is also common in the literature (Hazans and Philips, 2010, p. 260). Also for Germany, national immigration numbers were necessary for the earliest and most recent years while Eurostat provides the figures in between. The migrant stock abroad proxies network effects and is counted as foreign population of Baltic citizenship in the receiving country. Since no numbers are provided for the UK and Ireland by Eurostat, figures from Holland et al. (2011) were used as well as national population survey estimates in the case of the UK.22 The period considered is 19982013, the earliest and latest years available from Eurostat immigration data.23 Within the push and pull framework, earnings differentials and job opportunity are the most important determinants of migration. Income is proxied by GDP per capita in current Euro prices based on Eurostat. I use current prices rather than purchasing power parity because the latter may underestimate monetary incentives to migrate since income is consumed abroad and at home (Brucker et al., 2009, p. 9). Unemployment rates for the working age population (15-64 year olds) as well as the young population (15-25 year olds) are straightforward and based on Eurostat. The only exception are Swiss data which are taken from the national institute and equally follow the International Labor Organization’s (ILO) definition (Swiss Federal Statistical Office, 2015). Following the literature, bilateral distance is used as a proxy for the cost of migration and taken from CEPII’s geo-distance database which is widely employed in gravity models for trade or migration (Mayer and Zignago, 2011).24 A restriction dummy accounts for whether a transitional arrangement is in place or not (Fihel et al., 2015). Capital and trade flows are used as controls for labor flows following the standard trade theory of comparative 22 The resident population of Estonians in the UK had to be approximated from NINo accounts because the (small) numbers were not reported in the population survey, see appendix. 23 Due to data limitations, migration to the UK prior to 2002, to Ireland prior to 2000, and to Italy in 2001 is missing, hence, not considered. 24 A gravity model, e.g. by Greenwood (1985), assumes that bilateral flows occur proportional to the population size in both directions with a magnitude inversely proportional to the squared distance. As Holland et al. (2011, p. 26) point out, these assumptions insufficiently capture recent migration trends in the EU where distances are short and transport cheap. 26 advantages. Bilateral FDI inflows for the period 2001-2012 are reported by UNCTAD (2014) and set relative to the recipient country’s GDP in the reference year. Trade data provided by UN Comtrade (2015) proxy bilateral economic links in both directions and are transformed into bilateral trade volume as per cent of the Baltic country’s GDP; see appendix for detail. Government spending in per cent of GDP is employed as a proxy for the provision of public goods as well as public welfare schemes with data based on Eurostat (2015). Social and political push and pull factors are investigated via international benchmarks and chosen in the light of factors promoting a conducive environment for firms, because job creation and job opportunity are essential to retain talented people at home. As reported by the Global Competitiveness Report, inefficient government bureaucracy is the most important constraint to economic activity in Latvia and Lithuania. Additionally, corruption remains one of the most problematic factors for doing business in both countries (World Economic Forum, 2014; International Finance Corporation, 2013). Transparency International’s Corruption Perception Index (CPI) is employed for the entire period to account for any level of bribery, which I assume to affect not only firms but also the aggregate of individual migration decisions (Transparency International, 2015). An evaluation of government effectiveness is retrieved from the Worldwide Governance Indicators (WGI) in terms of the percentile ranking among all countries (Kaufmann and Kraay, 2015). Finally, I consider indicators of productivity, education and demography to analyze not only determinants of migration but to derive consequences in the discussion of the endogeneity of migration. The Penn World Tables offer data on human capital based on years of schooling and return to education as well as on total factor productivity (Feenstra et al., 2013). The comparison of one country relative to another is useful when considering migration as maximization of returns to education. The share of tertiary educated is used as a hint on brain drain. The share of the young population under 25 years, the dependency ratio and the youth unemployment rate contribute to picturing the relation of migration and demographic challenges with data taken from Eurostat (2015). The explanatory variables can therefore be categorized into five blocks which I refer to in the following: standard migration model, controls by theory, social environment, productivity and innovative capacity as well as demography. 27 3.2 Estimation Results for Baltic Migration In order to explain flows of migrants from the three Baltic countries to the eleven considered European countries over the period 1998-2013, I use a fixed effects estimator with cluster-robust standard errors and time fixed effects. The descriptive statistics of the previous chapter suggested some heterogeneity of the Baltic migration patterns which can be captured with country specific fixed effects in the Baltic panel. Additionally, separate panels for Estonia, Latvia and Lithuania are presented. I also use cluster-robust standard errors for country pairs to address heteroscedasticity and serial correlation because in their presence, the standard errors of the fixed effects estimators are often severely understated (Schmidheiny, 2015, p. 8f). Year dummies should capture impacts on the dependent variable which are not caused by any explanatory variable. I did not find evidence for nonstationarity of the variables. Reverse causality should not constitute a problem concerning European destination countries because the number of Baltic migrants is comparatively small, nevertheless, in the sending countries, a sufficiently large loss of workforce impacts on wages and unemployment rates (Kancs and Kielyte, 2002, p. 15). In order to address reverse causality and also endogeneity, I introduce lags of one year for all variables except for time-invariant and restriction dummies so that, for example, although unemployment rates are not perfectly exogenous, migration flows can only determine contemporaneous and future but not previous unemployment rates (Mayda, 2010). Potential biases due to endogeneity and simultaneity are discussed in the interpretation. Since the time-invariant variables are omitted when controlling for unobserved fixed effects, I also provide a fully specified pooled OLS estimation with year dummies and cluster-robust standard errors, ignoring the correlation with explanatory variables. The only time-invariant variable is bilateral distance, which, on the one hand, is the only proxy for migration cost, while, on the other hand, the eroding importance of distance seems apparent. The basic standard model of migration is given in Kancs and Kielyte (2002), which is analytically derived in the appendix: 𝑀𝑏𝑒𝑡 = 𝑎1 ln 𝑌𝑒𝑡−1 𝑈𝐸𝑒𝑡−1 + 𝑎2 ln + 3 ln 𝑀𝑆𝑒𝑡−1 + 4 ln 𝐵𝐷𝑏𝑒 𝑌𝑏𝑡−1 𝑈𝐸𝑏𝑡−1 (1) where Mbet is the bilateral gross migration rate from Baltic country b to European country e relative to the home population in b, Y is GDP per capita, UE is the 28 unemployment rate, MS is migrant stock of Baltic nationals in e and BD is the bilateral distance from b to e. Instead of employing absolute levels, I use the ratio of European to Baltic country for all variables and estimate the following level-log econometric model with one-year lag: 𝑟𝑖𝑚𝑚𝑖𝑡 = 𝛼 + 𝛽1 ln 𝑟𝑔𝑑𝑝𝑐𝑟𝑡−1 + 𝛽2 ln 𝑟𝑢𝑒𝑡−1 + 𝛽3 ln 𝑚𝑠𝑏𝑒𝑡−1 + 𝛽4 ln 𝑏𝑑𝑏𝑒 + 𝛽5 𝑟𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑡 + 𝛽6 𝑓𝑑𝑖𝑏𝑒𝑡−1 + 𝛽7 ln 𝑡𝑟𝑎𝑑𝑒𝑏𝑒𝑡−1 + 𝛽8 ln 𝑟𝑔𝑡−1 + 𝛽9 ln 𝑟𝑔𝑖𝑛𝑖𝑡−1 + 𝛽10 ln 𝑟𝑐𝑝𝑖𝑡−1 + 𝛽11 ln 𝑟𝑤𝑔𝑖𝑡−1 + 𝛽12 ln 𝑟ℎ𝑐𝑡−1 + 𝛽13 ln 𝑟𝑡𝑓𝑝𝑡−1 + 𝛽14 ln 𝑟𝑡𝑒𝑑𝑡−1 + 𝛽15 ln 𝑟𝑢25𝑡−1 + 𝛽16 ln 𝑟𝑑𝑟𝑡−1 + 𝛽17 ln 𝑟𝑦𝑢𝑒𝑡−1 + 𝑒𝑡 (2) where rimmit is the bilateral gross migration flow from Baltic country b to European country e relative to the home population in b in the reference year t. The standard explanatory variables include relative GDP per capita at current prices rgdpcrt-1, relative unemployment rate ruet-1, the migrant stock msbet-1 of Baltic nationals in e, the bilateral distance between b and e bdbe, and the restriction dummy restrictt, which is equal to 1 if transitional arrangements are in place. The second category includes fdibet-1 which is bilateral FDI inflow in per cent of the Baltic country’s GDP, where because of negative figures in case of disinvestment the logarithm is undefined and therefore not taken, bilateral trade volume tradebet-1 in per cent of the Baltic country’s GDP, and relative share of government spending in GDP rgt-1. The third group consists of the relative Gini coefficient of income inequality rginit-1 interpolated for the missing years, the relative corruption perception index rank rcpit-1, and relative government effectiveness rank rwgit-1 interpolated for the two missing years. The fourth group comprises relative human capital per person rhct-1, relative total factor productivity rtfpt-1, and relative share of tertiary educated among 25-64 year-olds rtedt-1. Finally, the fifth category consists of the relative share of under-25 year-olds in the population ru25t-1, the relative dependency ratio rdrt-1 of the over 65 year-olds and under 14 year-olds to the working age population, and the relative youth unemployment rate ryuet-1 of the under-25 year-olds. Table 1 depicts the estimation results of the cluster-robust fixed effects estimation of the Baltic panel with year fixed effects, which are not reported, and compares the final specification (6) to the pooled OLS results (7) which also uses clusterrobust standard errors and year fixed effects. The standard migration model variables are maintained throughout all specifications with varying variable groups. 29 Table 1: Estimation results for the Baltic panel fixed effects estimation, 1998-2013 VARIABLES L.lrgdpcr L.lrue L.lms o.lbd restrict (1) FEcrt (2) FEcrt (3) FEcrt (4) FEcrt (5) FEcrt (6) FEcrt -0.391 (0.274) -0.133*** (0.0375) 0.0335 (0.0250) - -0.482* (0.243) -0.175*** (0.0486) 0.0741* (0.0398) - -0.510* (0.285) -0.165*** (0.0438) 0.0492 (0.0323) - -0.529* (0.297) -0.116*** (0.0379) 0.0455* (0.0269) - -0.509* (0.297) -0.202*** (0.0710) 0.0375 (0.0240) - -0.673** (0.257) -0.333*** (0.0929) 0.0652** (0.0281) - -0.0402** (0.0187) -0.0404** (0.0190) -0.00515** (0.00249) -0.0609 (0.0399) -0.0293 (0.104) -0.0390** (0.0159) -0.0542** (0.0214) -0.0369** (0.0179) L.fdi L.ltrade L.lrg L.lrgini -0.0435 (0.118) 0.128 (0.135) 0.105 (0.137) L.lrcpi L.lrwgi L.lrhc 0.287 (0.316) 0.374** (0.158) -0.00441 (0.0577) L.lrtfp L.lrted L.lru25 -0.602 (0.687) -2.744 (1.940) 0.510 (0.440) -0.829* (0.450) 0.0529 (0.0479) 1.956 (1.990) 387 0.292 33 410 0.274 33 444 0.301 33 L.lrdr L.lryue Constant 0.331 (0.216) 0.327 (0.460) Obs. 444 390 R-squared 0.256 0.320 No. bilmig 33 33 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 (7) OLScrt -0.343** (0.130) -0.289*** (0.104) 0.0948*** (0.0231) 0.00671 (0.0347) -0.00929 -0.00408 (0.0202) (0.0314) -0.00727** -0.00836** (0.00287) (0.00405) -0.144** -0.00496 (0.0567) (0.0194) 0.158 -0.0480 (0.181) (0.0841) -0.0332 0.167 (0.141) (0.146) 0.212* 0.251** (0.111) (0.112) -0.459* 0.211 (0.260) (0.143) 0.331 -0.291 (0.612) (0.234) 0.690** 0.0666 (0.253) (0.128) -0.299* -0.0784 (0.163) (0.0757) 0.0877 0.138 (0.514) (0.267) -1.748** 0.151 (0.688) (0.365) 0.131** 0.0855 (0.0629) (0.0618) 5.387 -3.138 (6.349) (2.052) 333 0.450 33 333 0.610 Source: Own estimation. Note: The dependent variable is bilateral gross migration flow from a Baltic to a European country relative to the Baltic home population in the reference year. Specifications (1) to (6) report clusterrobust fixed effects results with year fixed effects which are not reported. Variables are lagged by one year (L.) and mostly log of the ratios (lr). No. bilmig presents the number of country pairs of bilateral migration. Obs. stands for observations of a total of 528. 30 The coefficients of the standard migration model have the expected signs, except for income and distance. If relative income increases by 1%, due to an increase in European income or decrease in Baltic income ceteris paribus, then migration decreases by 0.0039%. Since migration flow is set relative to the population, this means a decrease by 3.9 individuals per 100.000 inhabitants. The coefficient becomes higher and more significant over the specifications, i.e. up to 6.7 individuals per 100.000, significant at the 5% level. The negative sign therefore suggests that a cost constraint is binding potential migrants who, for example, cannot afford to spend time abroad searching for a job or accommodation without income while they still incur expenditures at home. If the relative unemployment rate increases by 1%, due to a rise in European or a drop in Baltic unemployment, then migration decreases by 0.0013%, significant at the 1% level, or 1.3 individuals per 100.000 inhabitants. The migrant stock abroad describes the network available to the new migrant to facilitate job and accommodation search, and mitigate the psychological cost of moving. If the migrant stock abroad increases by 1%, migration increases by 0.0003%. In the final specification, the network effect is larger and significant at the 5% level. Among the subsequent pooled OLS specifications I conducted, the migrant stock generally appears more significant than income which is also reported in specification (7). The impact of the timeinvariant bilateral distance is contrary to the expectations positive, which means that the greater the distance, the higher the migration rate, although not significantly different from zero. Migration restrictions such as those imposed by transitional arrangements have the expected negative effect of a 0.0004% decrease, significant at the 5% level. This picture is slightly different to findings for NMS countries, such as in Fihel et al. (2015), who state that restrictions did not reduce migrant flows by numbers, although flows were diverted from traditional receiving countries towards open access countries. International economic links such as FDI inflows or trade volume impact on migration, too. According to the second specification, if bilateral FDI inflows increase by 1 percentage point in the share in GDP, then migration is reduced by 0.00005%, significant at the 5% level, and if bilateral trade volume increases by 1%, migration decreases by 0.0006%.25 The standard neoclassical argument is that 25 Note that bilateral FDI figures are only available for the period 2001-2012, i.e. 390 observations. 31 if capital flows in, entrepreneurial activity is fueled, the demand for labor increases, aggregate output increases and therefore also aggregate income. Increasing income enhances domestic demand also for foreign goods, for which the capital-trigger is not needed since trade volume measures exports and imports so that, consequently, potential migrants have less incentive to migrate because they expect higher wage and employment levels, regardless of any market corrections via prices. Relative government spending, on the other hand, has an ambiguous and insignificant effect on migration, which casts doubt on any notion of welfare tourism. Column (3) uses indicators to characterize influences of the social environment. If relative income inequality increases by 1%, migration decreases by 0.0004%. Since Baltic migrants move from a more unequal to a more equal country, they might be negatively selected. If the relative inequality increases either due to more inequality abroad or less inequality at home, the incentive to migrate is smaller for the costconstrained individuals at the lower end of the income distribution. The coefficient is, however, not significant. If the perceived relative corruption increases by 1% due to a better rank of a European or worse rank of a Baltic country, migration increases by 0.0012%, significant at the 10% level in the final specification with a larger coefficient. If relative government effectiveness increases by 1%, again, either if European governance effectiveness increases or the Baltic decreases, then migration increases by 0.001%. Conversely, in the full specification, the impact of governance effectiveness is negative and significant at the 10% level. Possible explanations imply a bureaucratic constraint on migration which becomes binding beyond an upper threshold in a European country where, for example, work permits are scrutinized, or beyond a lower threshold in a Baltic country where, for example, deregistration procedures are not executed. The specification based on productivity indicators in column (4) depicts an insignificant positive impact of relative human capital and a significant positive impact of relative total factor productivity.26 Accordingly, if relative human capital per person based on years of schooling and returns to education increases by 1%, migration increases by 0.0028% because migrants are attracted by a higher return to education abroad or repelled by lower returns at home. The same story holds for relative total factor productivity on whose 1% increase the migration rate responds 26 Note that no data is available for 2012 and 2013. 32 by a 0.0037% increase, even more pronounced in the final specification, while both are significant at the 5% level. Seeking higher returns to education abroad combined with relatively higher total factor productivity abroad suggests some brain drain so that it becomes crucial to know who exactly is migrating. The relative share of tertiary educated has a negative impact on migration by 0.0029% significant at the 10% level in the final specification if it increases by 1%. Since the relative increase is either due to an increase of tertiary educated among the workforce in Europe or to a decrease among the Baltic workforce, a negative impact means less incentives to migrate, although no inference about skill-levels of migrants must be made. Yet, by scale, human capital and productivity have greater coefficients so that an increase in tertiary education merely mitigates potential brain drain. A separate consideration of the Baltic countries can further illuminate this result and clarify the second round effects that brain drain has on the potential migrants at home, which partly defines the endogeneity of migration. In specification (5), demographic aspects are considered to derive hints on the longterm effects of migration on the home population. If the relative share of under 25 year-olds increases by 1%, then migration increases by 0.0051% due to a lower share of the young at home or to a higher share of the young abroad so that the demographic change is reinforced if especially young people migrate. The effect is insignificant though. On the other hand, if the relative dependency ratio increases by 1%, because working age persons are supporting more children and retired persons in Europe or because of the opposite in the Baltic countries, migration decreases significantly at the 10% level by 0.0082%, or 0.0174% significant at the 5% level, which is the highest coefficient in the final specification. Since a lower dependency ratio means a smaller burden on the working age population, a negative impact of the relative ratios means a relatively smaller burden for Baltic working age persons which makes them stay. Again, the demographic change in the Baltic countries is reinforced by migration of the younger, which increases the burden for the stayers as a whole and feeds back into higher migration rates. Interestingly, relative youth unemployment has the opposite effect on migration than overall unemployment. If the ratio increases by 1%, migration increases by 0.0005%, which means that if youth unemployment rises in Europe or falls in the Baltic countries, more Baltic citizens migrate. The effect is even greater and significant at the 5% level in the final specification. The desire for schooling might contain some 33 explanation, for example, if migrants of any age perceive a lower youth unemployment rate at home, their expectations for the future might include higher competition on the labor market which makes them invest in education, preferably abroad to enforce their comparative advantage. Due to overall optimistic expectations about future earnings, such an investment comes at relatively low (present value) opportunity cost. A policy lever with regard to retain talented young people at home by creating jobs might therefore, on the other hand, have little effects on the target group. For the five variable categories, I conducted F-Tests on the final specification, where the period considered is effectively reduced to 2001-2011 due to missing data. The F-tests show that the standard migration model explains most of the bilateral migration significantly at the 1% level. The indicators of international economic links together are significant at the 5% level, as are the productivity and innovative capacity indicators. The indicators on demography are significant at the 10% level. Only the social environment indicators on inequality, corruption and governance insignificantly influence the bilateral migration rate of Baltic citizens. The picture by sending country is somewhat different, due to the heterogeneity of the three Baltic countries, as will be described in the next chapter. 3.3 Comparing Migration by EU-Accession and Economic Crisis The explanatory variables are all expressed as log of the ratios of a European country figure to a Baltic country figure. European and Baltic figures have, however, developed differently dynamically. Baltic income and unemployment rates exhibited higher growth rates, in especially until the financial crisis when the picture was reversed so that the alignment of European and Baltic figures since accession to the EU yields a dynamic momentum in the analysis. In terms of migration before and after EU accession, Zaiceva and Zimmermann (2008), for example, found that macroeconomic determinants such as income and unemployment have become significant determinants only after enlargement. An increase in GDP per capita resulted in lower migration intentions in the NMS after EU accession, while the same increase had a positive effect on migration intentions in the EU15. Unemployment was found to have a negative significant effect only 34 after accession in both old and new member states. Furthermore, they found that migration intentions have aligned by 2005 in old and new member states (Zaiceva and Zimmermann, 2008, pp. 20f). Hazans (2012) separates Estonian and Latvian migration before EU accession, after accession, and after the crisis and finds that with accession, geographical diversity decreased and migrants were more negatively-selected due to lower cost and human capital thresholds, whereas after the crisis, migrants were again positively-selected, and push factors such as unemployment became more important (Hazans, 2012). Based on the full specification given in column (6) of table 2, I run the estimation for pre-accession (2001-2004), post-accesssion (2005-2008) and crisis years (20092011); the results are reported in the appendix. I find that the effect of relative income is insignificantly positive before the crisis but significantly negative afterwards. Relative unemployment is negative in all periods but at different significant levels which also depend on the unobserved effects. The network has a positive, highly significant impact prior to accession when the sign switches and remains negative, although insignificantly thereafter. Significant differences appear also in the determinants controlling for government spending or inequality, which exhibit positive and highly significant coefficients prior to 2004 which turn either negative or insignificant after EU accession. Differences in public goods provisions, therefore, seem to offer less incentives to migrate, while the former positive selection of migrants has turned negative. The productivity indicators exhibit the same signs over the periods but are less significant, except for relative factor productivity which has a large significant coefficient before 2004. This is also true for the relative dependency ratios. Relative youth unemployment impacts also significantly positive prior to 2004 but insignificantly negative afterwards. In general, post-accession and crisis periods are rather similar and distinct from preaccession coefficients, which also suffers from a smaller number of observations. Since these indicators are as such not perfectly exogenous to migration, the analysis for each Baltic country is conducted carefully in the following chapter. The three periods cannot be analyzed separately for the Baltic countries because of too few observations. Comparing the estimation results for the full specification across the Baltic countries further clarifies, nevertheless, the specific determinants of Estonian, Latvian and Lithuanian migration. 35 4 The Impact of Migration on the Baltic Countries The indicators mentioned above impact differently on the Baltic countries taken together than separately. The results presented in the following are, nevertheless, mostly in line with the panel results but also reveal the heterogeneity of the Baltic countries, in particular of Estonia compared to Latvia and Lithuania. Since controls for innovative capacity and demography require but do not allow inference on the characteristics of migrants, national studies will be referred to in order to analyze economic and demographic impacts of migration in the last section. Coping strategies and policy levers will be derived before concluding. 4.1 Estimation Results for Estonia, Latvia and Lithuania The results of the fixed effects and pooled OLS estimations for the full specification are reported in table 2. Results for migration by Estonians are given in columns (1) and (2), by Latvians in columns (3) and (4) and by Lithuanians in columns (5) and (6). Estimation results for Estonia frequently diverge from the panel results and are different to the Latvian and Lithuanian ones so that the panel estimation indeed disguises the heterogeneity of the Baltic countries. In the full specification, due to missing data, the period considered is effectively reduced to 2001-2011, which is, however, also the case for the Baltic panel estimation. While the combined analysis exploited the panel structure, the separate consideration may suffer from a small number of observations, i.e. a smaller N while T remains the same, although proportionately with a total of 333 observations and 111 observations for each Baltic country. Since relative variable ratios are considered, an increase in the ratio is, as mentioned above, either linked to an increase in the European country’s figure or to a decrease in the Baltic country’s figure, ceteris paribus, which facilitates intuitive understanding. Separate considerations with regard to attracting and repelling variables, i.e. push and pull factors, cannot be conducted. Most importantly, relative income differences have a larger significant negative effect compared to the combined results, especially for Latvia and Lithuania of up to 0.0178%, e.g., if the income difference relative to Latvia increases by 1%. 36 Table 2: Estimation results for Estonia, Latvia and Lithuania, 2001-2011 VARIABLES L.lrgdpcr L.lrue L.lms o.lbd restrict L.fdi L.ltrade L.lrg L.lrgini L.lrcpi L.lrwgi L.lrhc L.lrtfp L.lrted L.lru25 L.lrdr L.lryue Constant (1) eeFEcrt (2) eeOLScrt (3) lvFEcrt (4) lvOLScrt (5) ltFEcrt (6) ltOLScrt -0.237** (0.0850) -0.0534 (0.0409) -0.0372 (0.0207) - -0.293*** (0.0922) -0.0663 (0.0500) 0.0360** (0.0137) -0.0742*** (0.0188) -0.0115 (0.0120) -0.00364 (0.00298) -0.0193 (0.0290) 0.0762 (0.0553) -0.00618 (0.0854) 0.0713 (0.104) 0.221 (0.199) -0.330** (0.115) 0.487** (0.154) -0.0254 (0.0456) -0.528* (0.285) -0.325*** (0.0996) -0.0417 (0.0248) 2.028** (0.881) -1.789*** (0.498) -0.196 (0.295) 0.119 (0.0725) - -1.154** (0.434) -0.519** (0.176) 0.167*** (0.0467) -0.166** (0.0537) 0.0461 (0.0546) 0.00792 (0.0143) -0.130 (0.0812) 0.574* (0.258) 0.204 (0.310) 0.420** (0.174) 0.528 (0.343) -0.797** (0.356) 1.584** (0.566) 0.107 (0.0889) -2.165** (0.850) -0.437 (0.262) 0.00376 (0.0609) 0.0759 (2.389) -1.730*** (0.445) -0.179 (0.218) 0.152** (0.0510) - -1.041* (0.548) -0.475* (0.236) 0.129*** (0.0401) -0.418** (0.181) 0.0228 (0.0650) 0.00192 (0.00863) -0.0189 (0.0693) -0.0592 (0.159) 0.149 (0.363) 0.412 (0.248) -0.0605 (0.246) -1.341** (0.487) 1.405** (0.568) -0.0444 (0.142) -0.567 (0.568) -1.149 (0.960) 0.0821 (0.116) 8.833 (6.973) 111 0.783 111 0.710 11 0.00800 (0.0152) -0.00296 (0.00239) -0.0164 (0.0109) 0.464 (0.283) -0.208** (0.0881) -0.0611 (0.120) -0.106 (0.190) -0.604 (0.513) 0.488** (0.183) -0.324*** (0.0921) -0.692 (0.464) -0.588** (0.236) -0.0685 (0.0419) 7.773 (4.698) Obs. 111 111 R-squared 0.621 0.818 No. e 11 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.0258 (0.0377) 0.0221 (0.0180) -0.189*** (0.0411) 0.847 (0.483) 0.397 (0.221) -0.0614 (0.344) -0.147 (0.368) 2.002 (1.335) 1.655*** (0.364) -0.156 (0.331) -0.842 (1.161) -2.696 (1.657) -0.315 (0.343) -3.721 (14.18) 111 0.703 11 -0.0148 (0.0448) 0.0384** (0.0134) -0.205*** (0.0598) 1.112* (0.603) 0.0527 (0.173) -0.0895 (0.349) -0.485 (0.302) 1.617 (1.749) 1.772*** (0.456) -0.417 (0.353) 0.510 (1.387) -2.199* (1.177) -0.257 (0.228) -7.508 (16.77) 111 0.779 Source: Own estimation. Note: The dependent variable is bilateral gross migration flow from a Baltic to a European country relative to the Baltic home population in the reference year. The specifications report cluster-robust fixed effects and pooled OLS estimations with year fixed effects which are not reported. Variables are lagged by one year (L.) and mostly log of the ratios (lr). No. e presents the number of European countries. Obs. stands for observations of a total of 176. 37 Relative income attracts migrants at different positions in the earning distribution at home differently, and since earnings at the lower end are related to lower skill levels, the cost constraint may generate a hump-shaped relation between income at origin and emigration, which is more pronounced for low-skilled liquidity constrained individuals (Djajic et al., 2015). Relative unemployment rates have a smaller, less significant, but still negative effect on migration. If no country specific factors are controlled for, the coefficients resulting from a 1% increase in relative unemployment rates are still large and significant for Latvia and Lithuania. The separate results on the importance of the migrant stock abroad are still large, positive and mostly significant. The coefficients are smaller for Estonia, where the fixed effects estimation reports an insignificant negative effect. The impact of the network might indeed be smaller for Estonians than Lithuanians and Latvians because Estonian migration is very homogenous and to 70% directed towards Finland so that migration might have become more institutionalized and independent of personal networks as well as initial causes. In the pooled OLS specification, bilateral distance now impacts significantly negative on migration, whereas above, it was positive but insignificant. Therefore, it is likely that the greater the distance, the smaller the migration flow will be because of cost constraints, which are not necessarily monetary. The impact of restrictions imposed by the transitional arrangements remains negative but insignificant if the countries are analyzed separately with fixed effects, where the sign is again different for Estonia. Altogether, the variables of the standard migration model appear less significant compared to the combined estimation, they exhibit, nevertheless, a significant impact in the common F-test conducted on the full specification at 1% for Latvia, 5% for Lithuania and 10% for Estonia, as table 3 shows. Among the category of international economic links, on the other hand, the bilateral trade volume coefficients remain significantly negative at the 1% level, although insignificant for Estonia. If bilateral trade volume with Lithuania increases by 1%, migration decreases by 0.002%, a slightly higher coefficient than the panel estimation. The former negative significant impact of capital inflows on migration is reversed for Latvia and Lithuania, and becomes significantly positive at the 5% level for Lithuania, while it is insignificantly negative for Estonia. A 1 percentage point increase of the inflowing FDI share in the domestic GDP leads to a 0.0003% increase in Lithuanian migration so that, as capital flows in, labor takes the opposite 38 direction towards absolute higher remuneration. Relative government spending yields higher positive coefficients for all three Baltic countries than the combined estimation. Moreover, it has a large positive impact on Lithuanian migration significant at the 10% level. The F-test on the combined impact of the category is only significant for Latvia and Lithuania, at the 1% and 5% level, respectively, but insignificant for Estonia. Indicators to describe the social environment in the Baltic countries reveal some more heterogeneity. While the impact of governance effectiveness on migration remains negative but insignificant, the sign of the impact of corruption switches and turns insignificant. The pooled OLS results, however, remain positive and significant at the 5% level for Latvia. The coefficient of income inequality turns insignificantly positive for Latvia and Lithuania but remains negative and significant at the 5% level for Estonia. One explanation maybe related to the relatively high Gini coefficient in the UK and the high number of Latvian and Lithuanian migrants to the UK. Overall, the social environment indicators characterize significant determinants of migration only in Estonia, at the 5% significance level, as table 3 depicts. Table 3: F-test results on full specification for the variable categories (I) – (V) Estonia Latvia Lithuania (I) std. model 0.0808 (2.86) 0.0001 (18.56) 0.0179 (4.99) (II) intl. links 0.2156 (1.77) 0.0017 (10.95) 0.0263 (4.74) (III) social environ. 0.0374 (4.16) 0.3511 (1.22) 0.3714 (1.16) (IV) productivity 0.0012 (11.91) 0.0016 (11.06) 0.0051 (8.05) (V) demography 0.0651 (3.32) 0.4293 (1.01) 0.2537 (1.59) Source: Own estimation. Note: F-statistics are reported in parentheses. The fourth category on productivity and innovative capacity as well as the final category on demography offer a similar picture compared to the Baltic panel analysis. The estimation results for relative total factor productivity depict a positive, highly significant and comparatively large impact on migration. If relative productivity increases by 1%, Lithuanian migration, for example, increases by 0.0177%. The coefficient is lower and less significant at the 5% level in Estonia. The results for relative human capital based on years of schooling and returns to 39 education are ambiguous and depend on the estimation method. Controlling for unobservable country fixed effects, which is the preferred method in order to allow for correlation, the impact on migration is positive, comparatively large but insignificant, whereas the pooled OLS estimation renders smaller, significantly negative coefficients. The case is again different for Estonia, where the impact of an increase in relative human capital on migration is negative, although insignificantly. An increase in the relative share of tertiary educated shows a negative but less significant effect on Baltic migration compared to the panel results, although it is significant at the 1% level in Estonia. The relative share of the young in the total population impacts negatively on migration, contrary to the panel results. An increase in relative youth unemployment has an insignificant negative effect on migration, when controlling for fixed effects, again, contrary to the significant positive impact in the combined analysis. Results for the relative dependency ratio are in line with the panel results. If the relative dependency ratio increases by 1%, migration decreases by 0.0219% significant at the 10% level in Lithuania, for example. The coefficient is smaller but highly significant in Estonia. The F-test on productivity indicators reveals an impact highly significantly different from zero, whereas demographic indicators are only significant at the 10% level in Estonia. 4.2 Human Capital and Returning Migrants The capacity to attract and retain talented people in the country is assessed as ordinary in the Baltic countries by the Global Competitiveness Report, since interviewees felt that the best and brightest tended to pursue opportunities in other countries (World Economic Forum, 2014). Job opportunities for talented people in the Baltic countries were assessed weakest in Lithuania and strongest in Estonia. In Estonia, inadequately educated workforce is reported as most important constraint to doing business, and also the insufficient capacity to innovate is among the most problematic factors. This points towards an innovation driven economy, while Latvia and Lithuania, both noted as transition economies, still struggle with government inefficiency and corruption (World Economic Forum, 2014). Whereas the estimation results did not provide evidence for the latter two to have impacted 40 on migration, let alone to influence individual migration decisions, becoming an innovation driven economy is essential for the Baltic countries. In the following, an attempt shall be made to identify human capital and demographic challenges linked to migration, since migration also affects the determinants of migration. For all Baltic states in the early accession period 2004-2007, Brucker et al. (2009, pp. 71ff) find positive impacts on GDP per capita in the short run, but smaller negative effects in the long run, whereas unemployment and also wages decreased in the short run without any long run effects. By skill level, wages increased with a long run effect only for the low skilled, who also profited from larger decreases in unemployment, whereas initial benefits for medium- and high-skilled dropped to zero or even smaller than zero in the long run. Effects were particularly large for Lithuania which also experienced the highest opportunity cost in terms of GDP losses of up to 6% compared to 3-3.3% for Estonia and Latvia if migrants had stayed home (Holland et al., 2011, p. 79). Remittances are also an important feature of migration and may affect, for example, the structure of domestic demand if foreign goods become affordable which in turn may have an effect on imports. Incoming remittances from diaspora communities amount to roughly 2-5% of GDP in Estonia, Lithuania and Latvia (World Bank, 2015). Such form of capital income is a desirable target for policy-makers who seek to mitigate the negative effects of migration and to promote sustainable remittance investment, for example, by issuing diaspora bonds (Engbersen and Jansen, 2013, pp. 19f). Diaspora communities, in theory, are also likely to become trading partners or to invest directly in the home economy. If the population decreases, however, not least due to increased migration, the market shrinks which may in return discourage investments (Hazans, 2013, pp. 101f). In order to reasonably identify the relation between human capital and migration, i.e. whether evidence supports brain drain, brain gain or brain circulation, detailed information on the characteristics of migrants by education or occupation are necessary. Such data is, however, not available comprehensively for the Baltic migrant flow, but some studies describe the profile of stock migrants, i.e. resident population abroad. From the receiving country perspective, Brucker et al. (2013) offer estimates of brain drain towards OECD countries which show how the Baltic resident migrant profile (stock) differs across the considered receiving countries. Analysis shows that Denmark, Germany, Spain and Norway mostly retain medium- 41 skilled Baltic migrants, with a tendency in Germany and Denmark for lower skills and in Norway for higher skills. Resident migrants in Switzerland are high-skilled, while they are low-skilled in the Netherlands. The skill level of Baltic stock migrants in the remaining countries is either bipolar, namely many low- and highskilled such as in the UK (roughly 60% and 40%), or different for the Baltic origins, e.g., Ireland receives well educated Estonians but weakly educated Lithuanians and Latvians whereas Finland and Sweden receive many low-skilled migrants and only some high-skilled Lithuanians (Brucker et al., 2013); see appendix for detail. In the case of Estonia, migration has been rather steady, homogenous and presumably more institutionalized so that traditional pull factors such as income, unemployment, or network have indeed lost attractiveness over time. Relative differences towards European countries are smaller than in Latvia and Lithuania. Young people emigrated increasingly over-proportionately since EU accession so that the demographic challenge is exacerbated by emigration which, as mentioned above, increases the social burden for the remaining. Among the Estonian migrant stock in the considered European receiving countries, the low-skilled are overrepresented compared to the home population, as table 4 shows. Since low-skilled migrants are likely to take low-skilled positions upon their return, although they might profit from a foreign experience premium, a skill transfer is questionable (Engbersen and Jansen, 2013, p. 23). Nevertheless, encouraging return migration is key to activate in particular young professionals living abroad who might experience a decreasing attachment to the home country and increasing attachment to the host country (Kaska, 2013; Engbersen and Jansen, 2013).27 The ratio of immigration by Estonians to Estonia, which exhibits a smaller variance than in Latvia and Lithuania, and emigration by Estonians from Estonia is roughly 0.3 and has increased since the early EU accession years, pointing towards more dynamic mobility (Eurostat, 2015). Summing up, Estonian migration suggests a circular pattern and offers little evidence for brain drain apart from specific sectors, such as health care. Forming competent human capital to a sufficient extent in order to match labor market needs appears as major challenge, where migration only plays a minor role (Kaska, 2013). 27 Engbersen and Jansen (2013, p. 15) distinguish four migrant profiles according to the relative attachment to host and home country: Temporary, circular and seasonal migration as opposed to settlement migration, and transnational (bi-national) migration as opposed to footloose migration. 42 Table 4: Comparison of educational levels among movers and stayers 2000 2005 2010 2000 Latvia 2005 2010 2000 Lithuania 2005 2010 Estonia Share of skill levels Share of skill levels Ratio of skill among the resident among the domestic proportions among population abroad population at home migrants relative to 28 (over 25 years) (25-64 years) non-migrants low med high low med high low med high 48 26 26 14 57 29 3.37 0.46 0.90 45 27 28 11 56 33 4.08 0.49 0.84 40 31 29 11 54 36 3.73 0.58 0.81 44 29 27 17 65 18 2.62 0.44 1.50 37 35 29 16 64 20 2.34 0.54 1.41 32 39 29 11 62 27 2.81 0.63 1.07 42 32 26 16 42 42 2.64 0.77 0.61 33 38 29 13 61 27 2.67 0.62 1.09 30 41 29 8 60 32 3.66 0.69 0.90 Source: Own calculations based on Brucker et al. (2013) and Eurostat (2015). The situation in Latvia and Lithuania is comparable but somewhat distinct from Estonia, as mentioned above. In general, the Latvian and Lithuanian migration patterns are more volatile, more heterogeneous, and still significantly determined by differences in relative income, relative unemployment and networks abroad, but also by economic links and factor productivity. Young people are over-represented among migrants by ratios of 1.4 to 1.7 and the trend is increasing, deteriorating the demographic outlook while the natural population change is also negative. In Latvia, emigration is perceived as a major threat to the economic development, social security systems and reproduction of the Latvian population (Hazans, 2013, p. 67). Among Latvian residents abroad, as table 4 depicts for the age groups who can be assumed to have completed education, low-educated but also high-educated are over-represented. This underlines the above mentioned migration hump, a reverse U-shaped relation between emigration and income if low-educated are also low-wage earners and high-educated high-wage earners (Djajic et al., 2015). The migrant profile changed from single, with temporary and low-skilled job preference, minimal language skills and return intentions during the early accession years to moving with family, seeking permanent, legal and skilled employment as well as social security in the following years prior to the financial crisis. Consequently, in 2010, high educated emigrants were less likely to return to Latvia 28 Only the considered receiving countries are taken into account. No data is provided for Italy. 43 (Hazans, 2013, pp. 82f). Emigrants’ labor market outcomes by employment status are significantly better than those of non-migrants. Additionally, an employed return migrant can expect a 13% wage premium due to foreign work experience (Hazans, 2012, p. 89). Emigration intentions based on economic motives decrease with education but increase with education based on non-economic motives (Hazans, 2012, p. 199). Nevertheless, deskilling and brain waste while working abroad are frequent and result in a depreciation of human capital. Latvian migration can be characterized, due to the increasingly permanent character, as settlement migration that suffers brain drain and increases demographic pressure at home. For Lithuanians, important determinants of migration next to income, employment opportunities and networks also include a recent phenomenon which stems from the financial and economic crisis. Some people were unable to pay back a bank loan taken during the booming years, or lost their (small) firm and suffered (personal) bankruptcy, social insecurity or injustice, and migrated (Sipaviciene and Stankuniene, 2013, p. 50; Hazans, 2013, p. 96). Among Lithuanian migrants residing abroad, low-educated are over-represented to a greater extent than among Latvians but smaller extent than among Estonians. An inadequate match between high education and labor demand repels especially young people, for whom prolonged education may already have constituted a strategy to circumvent labor market entry and disappointing outcome, who then rather invest time in finding a job abroad than in Lithuania. Given that more low-skilled emigrated, the higher skilled have also become relatively abundant. Furthermore, emigration potential is not decreasing since almost 60% of under 30 year-olds declare intentions to move abroad (Sipaviciene and Stankuniene, 2013, pp. 55f). The ratio of Lithuanians immigrating and emigration to and from Lithuania is smaller than in Estonia around 0.2, exhibiting a much higher variance, which suggests a less circular, less dynamic migration pattern more prone to settlement. Research on Lithuanian emigration is scarce and also policy-makers, although aware of the depopulation, did not take substantial measures yet. Given the low return to education in Lithuania and in the absence of significant return migration, the outlook for a boost of productive and innovative capacity on top of increased demographic pressure is gloomy. The situation might indeed constitute an example of immiserizing migration (Zakharenko, 2012), but in the opposite direction since contrary to brain drain, the ‘brain’ remains undemanded and unrewarded. 44 Concluding Remarks Emigration from the Baltic countries has been substantial over the last decade. In Latvia and Lithuania, the level is perceived as a threat to the reproduction of the population and to sustainable economic development, since especially young people emigrate and/or have intentions to emigrate. The demographic development has taken on Western European features but over a shorter period of time. While Estonian migration is characterized by the geographic and cultural proximity to Finland, Latvian and Lithuanian migration patterns are more heterogeneous and subject to changes in the economic environment. Economic motives remain the most important reasons for migration, because family reasons are often linked to a relative working abroad, too. The combined Baltic panel estimation of the determinants of migration and also the separate country estimations show that income differentials, relative unemployment and networks abroad still explain a great deal of bilateral migration. The restrictions imposed in the EU15 upon accession of the EU8 from 2004 to 2011 were not binding to migration, although they might have deterred some migrants and otherwise redirected flows to openaccess countries. Controlling for international economic links and measures of productivity and innovative capacity contributes to the explanatory power of the standard migration model. Descriptive indicators of the social environment such as inequality or corruption also help picturing reasons for migration. Overall, detailed information about the features of migrants such as age, sex, education or occupation is needed for a more profound analysis. Further indicators such as relative private sector debt should be introduced because they are likely to have affected Baltic migration decisions. Additional dummies for language skills and also ethnicity, possibly interacting with Baltic language skills, which ethnic minorities in the Baltics have to prove, complement the description of migrant characteristics, too. EU accession in general meant access to the integrated market with freely moving goods, services, capital and labor, and more intense competition for outputs and inputs, such as skilled workforce. Notably in Latvia and Lithuania, EU accession led to a change in the migrant profile, characterized by lower costs and thresholds, and thus more independent of income and associated skill levels. The turmoil of the financial and economic crisis was severe in the Baltic states but did not so much impact on the migrant profile, merely on the size of migration flows. It did, 45 however, not trigger massive return migration either. Returning migrants are at the core of considerations on positive emigration effects because they are expected to transfer skills to the home population who then benefit from brain circulation. A conclusion for a general brain drain cannot be drawn, and without more information on the migrant characteristic results are ambiguous. Low-skilled remain over-represented among resident migrants abroad. Yet, studies suggest that the share of tertiary educated among migrants is increasing. A break-down by occupation and age should be more revealing. Both sides contribute, nevertheless, to an outlook on productivity that is not that optimistic. If low-skilled become increasingly scarce at home, the labor market position of the high-skilled, who complement low-skilled, deteriorates and returns to education decrease. Lower returns to years of schooling lead to lower incentives to invest in schooling, and thus to frustration and pessimism about future development. Disentangling labor migration from general migration is necessary for more concrete conclusions, which is also helpful for national or EU-wide policy-makers who take into account the consequences of reduced invectives for schooling. Furthermore, the geographic impact of migration on the distribution of and among the home population is likely to reveal regions which suffer considerably from emigration. Competitiveness, productivity and innovative capacity are crucial for the economic development of the Baltic countries. A thorough understanding of the determinants and consequences of migration for Estonia, Latvia and Lithuania is necessary and requires further research in order to target and design policies to cope with emigration. 46 Appendix A.1 Formal Description of the Model Following the human capital theory (Sjaastad, 1962), migration is an investment undertaken if the net utility differential between origin and destination country is positive. Kancs and Kielyte (2002) offer a simple formal approach, assuming the probability that an individual i (i ϵ 1, …, n) migrates from origin country o to destination country d as: ∆𝑈𝑜𝑑 = 𝑈(𝑦𝑑 ) − 𝑈(𝑦𝑜 ) + 𝑡𝑜𝑑 (3) where ∆Uod is the net utility differential, yo is net income at home, yd is net income in the destination region, and tod is migration cost from o to d (Kancs and Kielyte, 2002, p. 23). Employing a concave utility function increasing in income yields: ∆𝑈𝑜𝑑 = 𝐸 ln(𝑦𝑑 ) − 𝐸 ln(𝑦𝑜 ) + 𝑡𝑜𝑑 , (4) and by expanding E ln(yd) around E(yd) using second order Taylor expansion: 𝐸 ln(𝑦𝑑 ) = ln(E𝑦𝑑 ) + 𝐸(𝑦𝑑 −𝐸𝑦𝑑 ) (𝐸𝑦𝑑 ) − 𝐸(𝑦𝑑 −𝐸𝑦𝑑 )2 2(𝐸𝑦𝑑 )2 = ln(𝐸𝑦𝑑 ) − 𝑉𝑎𝑟(𝑦𝑑 ) . 2(𝐸𝑦𝑑 )2 (5) Employment is the only source of workers’ income yd, namely the product of real wage wd and the probability of employment ed: 𝐸𝑦𝑑 = 𝑤𝑑 𝑒𝑑 . The probability of (uncertain) employment increases with the employment rate, respectively one minus the unemployment rate, which follows a binominal distribution with expected value e and variance e (1 - e), so that: 𝐸 ln(𝑦𝑑 ) = ln(𝐸𝑦𝑑 ) − 𝑤𝑑2 𝑒𝑑 (1−𝑒𝑑 ) 2𝑤𝑑2 𝑒𝑑2 1 ≈ ln(𝐸𝑦𝑑 ) + 2 ln(𝑒𝑑 ). (6) Hence, expected income at destination generating utility is a function of the wage rate wd and the unemployment rate ed (Kancs and Kielyte, 2002, p. 24): 1 3 𝐸 ln(𝑦𝑑 ) = ln(𝑤𝑑 ) + ln(𝑒𝑑 ) + 2 ln(𝑒𝑑 ) = ln(𝑤𝑑 ) + 2 ln(𝑒𝑑 ). (7) Expected income at home features smaller uncertainty of employment, which abroad weighs greater on employment than on wage so that the parameter γ captures the differences in employment uncertainty (γ < 1): 3 𝐸 ln(𝑦𝑜 ) = ln(𝑤𝑜 ) + 2 γ ln(𝑒𝑜 ). (8) The net utility differential based on income abroad and at home weighted by uncertainty of employment can thus be rewritten (Kancs and Kielyte, 2002, p. 24): 47 3 3 ∆𝑈𝑜𝑑 = ln(𝑤𝑑 ) + 2 ln(𝑒𝑑 ) − ln(𝑤𝑑 ) − 2 ln(𝑒𝑜 ) − 𝑡𝑜𝑑 . (9) Accounting for the importance of expected future income for a migration decision, the net present value of moving in period t stems from the utility differences in the following periods so that individual i’s migration probability is given by: Pr(𝑚𝑖𝑡 = 1) = Pr(∆𝑈𝑖𝑡 + ∆𝑈𝑖𝑡+1 ) > 0 ∩ 𝑈𝑖𝑡 > 0. (10) If ∆𝑈𝑖𝑡 < 0, the migrant might postpone the decision, potentially expecting a higher net present value after the next period. For all individuals, the migration probability yields the emigration rate Mt: 𝑀𝑡 = 𝛼(𝛿∆𝑈𝑖𝑡 + ∆𝑈𝑖𝑡+1 ), with a larger weight δ on current conditions and where the slope parameter captures the impact of utility differences between periods on emigration (Kancs and Kielyte, 2002, p. 25). Assuming for simplicity that migration probability in t only depends on the utility difference in t, the probability and emigration equations simplify to: Pr(𝑚𝑖𝑡 = 1) = Pr(∆𝑈𝑖𝑡 ) > 0, and (11) 𝑀𝑡 = 𝛼∆𝑈𝑖𝑡 . (12) The cost of migration tod is determined by the stock of previous emigrants Sd who constitute the migrant network, and the distance along which cost of migration increase (Kancs and Kielyte, 2002, p. 25): ̅ = 𝜃0 + 𝜃1 ln 𝑆𝑑 + 𝜃2 ln 𝐷𝑜𝑑 𝑡𝑜𝑑 (13) where 𝑡̅od depicts the mean cost over all i, Sd is the stock of migrants from o in d, and Dod is bilateral distance between o and d. By combination of equations (8), (11) and (12), gross migration rate Modt from origin country o to destination country d can be expressed as follows (Kancs and Kielyte, 2002, p. 25): 𝑤 3𝑒 𝑀𝑜𝑑𝑡 = 𝛼 ln 𝑤𝑑𝑡 + 𝛼 ln 2𝑒𝑑𝑡 + 𝛼𝜃1 ln 𝑆𝑑𝑡 + 𝛼𝜃2 ln 𝐷𝑜𝑑 + 𝛼𝜃0 . 𝑜𝑡 𝑜𝑡 (14) The gross migration rate is therefore a function of relative wages, relative unemployment, resident migrant population and bilateral distance. 48 A.2 Figures and Tables Figure 8: A simple labor market equilibrium with skilled emigration Wunskilled Wskilled m1 mr m0 nr n0 n1 S1 S0 Lskilled L‘r Lr L Lunskilled Source: Adapted for skilled emigration impacts on origin country from Zimmermann (1996, p. 113) and Kahanec et al. (2010, p. 21). Figure 8 depicts a simple labor market equilibrium in case of skilled emigration from the origin country. Emigration by skilled workers is a shock to the workforce supply and shifts the vertical supply curve from S0 to S1. Wages for skilled workers increase from m0 to m1 as a consequence of the decrease in supply. Since unskilled workers are complementary to skilled workers, the supply shock also reduces demand for unskilled workers. In a competitive market, a reduced demand for unskilled work while unskilled labor supply remains unchanged leads to a reduction of unskilled wage from n0 to n1 and a new equilibrium. Under a rigid union wage regime, which is present in most European countries, the wage level is fixed at nr. Decreased unskilled labor demand will thus reduce employment from Lr to Lr’. The reduced unskilled employment feeds back into the skilled labor market due to the complementarity. This mitigates the previous wage level increase from m1 to mr. Skilled emigration thus leads to unsatisfied skilled labor demand and upwardpressure on wages. In case of a competitive labor market, unskilled wages end employment respond via a reduction in wages due to decreased demand. In case of rigid union wage, the wage level remains fixed and labor markets respond with higher unemployment, and an additional reduction of skilled wage. 49 Figure 9: Comparison of emigration and immigration figures Estonia 14.000 12.000 10.000 8.000 6.000 4.000 2.000 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Emigration Immigration without UK Latvia 50.000 40.000 30.000 20.000 10.000 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Emigration Immigration without UK Lithuania 100.000 80.000 60.000 40.000 20.000 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Emigration Immigration without UK Source: Data retrieved from Eurostat (2015) and national statistics for the UK (British Department for Work and Pensions, 2015), Ireland (Irish Department of Social Protection, 2015) and Germany (German Federal Ministry of the Interior, 2005; DeStatis, 2015). Note: Emigration is entirely based on Eurostat data for the three sending countries. Immigration is based on Eurostat data for the eleven receiving countries and partly on national statistics where necessary. Because of the potential over-estimation of immigration to the UK, the UK is exempt in the immigration figures of the last indicator. Especially in Latvia and Lithuania, the difference is large recently, at 50-60%. 50 Table 5: Estimation results for Baltic migration before EU accession, after, and during crisis VARIABLES L.lrgdpcr L.lrue L.lms o.lbd restrict L.fdi L.ltrade L.lrg L.lrgini L.lrcpi L.lrwgi L.lrhc L.lrtfp L.lrted L.lru25 L.lrdr L.lryue Constant (1) preFErct (2) postFErct (3) crisFErct (4) preOLSrct (5) postOLSrct (6) crisOLSrct 0.552 (0.401) -0.336* (0.167) 0.141*** (0.0424) - 0.0976 (0.150) -0.161*** (0.0527) -0.0667*** (0.0197) - -0.393** (0.181) -0.103 (0.106) -0.1000 (0.121) - -0.0773 (0.0659) 0.00352 (0.00877) -0.0740 (0.0628) 1.573*** (0.331) 0.599*** (0.210) 0.245 (0.248) 0.658 (0.434) 0.199 (1.107) 1.164** (0.475) -0.325 (0.448) -0.104 (1.765) 3.868*** (0.859) 0.254* (0.140) -39.05** (15.75) -0.0214 (0.0134) 0.000666 (0.000960) -0.0259 (0.0308) -0.176** (0.0715) 0.0194 (0.0758) 0.166 (0.242) -0.260** (0.106) 0.289 (1.027) 0.124 (0.165) -0.0793 (0.108) 0.718 (0.716) 0.215 (0.416) -0.0763 (0.0668) -4.371 (7.405) -0.0413 (0.0322) 0.00350 (0.00308) 0.0226 (0.0295) 0.181 (0.241) -0.416 (0.314) 0.219 (0.198) -1.275 (1.082) 0.539 (3.057) 0.172 (0.247) -0.141 (0.392) 2.490 (2.466) 0.510 (0.866) -0.0945 (0.127) -9.215 (21.59) 0.000857 (0.0844) -0.0158 (0.0593) 0.0319*** (0.00926) -0.0284 (0.0199) -0.0971** (0.0375) -0.00460 (0.0111) -0.0137 (0.0157) 0.0733 (0.101) 0.0837 (0.0954) 0.0270 (0.0535) -0.224 (0.259) 0.0331 (0.0739) 0.113 (0.0858) 0.0323 (0.0653) -0.174 (0.138) -0.553** (0.261) -0.00289 (0.0446) 2.827* (1.643) -0.251* (0.125) -0.189** (0.0769) 0.0687*** (0.0147) 0.0167 (0.0339) -0.0449*** (0.0163) -0.00367 (0.00292) 0.00996 (0.0265) -0.221** (0.101) 0.360** (0.146) 0.436* (0.217) -0.0471 (0.169) -0.114 (0.203) -0.202 (0.228) -0.200 (0.138) 0.849* (0.425) 0.902 (0.678) -0.0102 (0.0448) -8.283* (4.474) -0.295* (0.150) -0.308** (0.125) 0.124*** (0.0221) 0.104 (0.0627) -0.0615 (0.0400) -0.00289 (0.00448) 0.0488* (0.0261) -0.234 (0.174) 0.163 (0.393) 0.209 (0.175) -0.329 (0.460) -0.993* (0.561) -0.00211 (0.441) 0.353 (0.258) 0.150 (0.655) 1.340 (0.963) 0.0634 (0.0754) -4.553 (6.111) 78 0.764 123 0.804 132 0.750 Obs. 78 123 R-squared 0.833 0.762 No. bilmig 30 33 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 132 0.294 33 Source: Own estimation. Note: Pre-, post-accession and crisis estimation results for cluster-robust fixed effects and clusterrobust pooled OLS are reported, both with fixed year effects, which are not reported. The considered time period is effectively 2001-2011 so that the pre-accession period consists of 2001-2004, the postaccession period of 2005-2008, and the crisis period of 2009-2011. 51 Table 6: Summary statistics Variable Label Obs Mean Std. Dev. Min Max rimmi lrgdpcr lrue lms lbd restrict fdi ltrade lrg lrigini lrcpi lriwgi lrhc lrtfp lrted lru25 lrdr lryue Immi/Pop Income Unemp Network BilatDist Restrict FDI Trade GovSpend Inequality Corruption GovEffic HumanCap TFP TertEduc YoungPop DependR YouthUE 507 528 495 519 528 528 396 528 528 432 517 528 462 462 519 528 528 495 0.0768 1.5844 -0.5504 7.4447 7.0867 0.5625 0.3331 0.9025 4.6226 4.4025 5.0862 4.8369 4.5904 5.0717 4.6028 4.5786 4.6352 -0.4048 0.1686 0.4420 0.5510 1.6525 0.5879 0.4965 1.2365 1.0474 0.2583 0.1359 0.2994 0.1381 0.0823 0.1784 0.3640 0.1073 0.0611 0.5892 0.0003 0.4283 -1.9810 3.0910 5.1648 0 -4.9117 -1.5054 3.7474 4.0897 4.0943 4.3215 4.4253 4.6706 3.0664 4.3898 4.4891 -1.8536 1.3519 2.6122 1.1102 11.9892 7.9676 1 16.6552 3.4861 5.0685 4.7265 5.9145 5.1111 4.7521 5.5734 5.2553 4.8052 4.7538 1.0879 Source: Own estimation. 52 Table 7: Correlation matrix rimmi rimmi lrgdpcr lrue lms lbd restrict fdi ltrade lrg lrigini lrcpi lriwgi lrhc lrtfp lrted lru25 lrdr lryue rimmi lrgdpcr lrue lms lbd restrict fdi ltrade lrg lrigini lrcpi lriwgi lrhc lrtfp lrted lru25 lrdr lryue lrgdpcr lrue lms lbd restrict fdi ltrade lrg 1 -0.112 -0.037 0.648 0.095 -0.278 -0.032 0.100 -0.016 0.232 0.051 0.051 -0.063 0.103 0.109 0.241 0.029 -0.036 1 -0.612 -0.239 -0.262 0.478 -0.023 -0.127 -0.100 -0.471 0.648 0.623 0.508 0.462 0.505 0.349 0.056 -0.554 1 0.183 0.089 -0.122 0.080 0.095 0.101 0.239 -0.335 -0.404 -0.230 -0.420 -0.218 -0.336 -0.132 0.888 1 0.011 -0.342 0.052 0.227 -0.017 0.194 -0.034 -0.052 0.093 -0.037 0.034 0.184 -0.020 0.073 1 -0.064 -0.327 -0.556 -0.455 0.658 -0.428 -0.436 -0.125 0.089 -0.288 -0.084 -0.495 0.017 1 -0.080 0.072 -0.245 -0.169 0.238 0.296 0.150 0.022 0.108 -0.265 -0.086 -0.243 1 0.299 0.231 -0.196 0.016 0.043 -0.003 -0.002 -0.014 0.031 0.189 0.128 1 0.529 -0.315 0.073 0.058 -0.046 -0.371 -0.121 -0.374 0.500 0.062 1 -0.466 0.103 0.052 -0.018 -0.187 0.008 0.107 0.518 0.197 lrgini lrcpi lrwgi lrhc lrtfp lrted lru25 lrdr lryue 1 -0.568 -0.537 -0.481 -0.055 -0.409 -0.124 -0.392 0.212 1 0.899 0.527 0.155 0.860 0.328 0.108 -0.335 1 0.471 0.198 0.830 0.392 0.023 -0.429 1 0.334 0.471 0.269 0.046 -0.253 1 0.322 0.673 0.142 -0.299 1 0.506 -0.042 -0.223 1 -0.025 -0.270 1 0.079 Source: Own estimation. 1 53 Figure 10: Immigrants to considered OECD countries by origin, sex and skill-level Estonia 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 1,00 0,80 0,60 0,40 0,20 0,00 DK FI DE IE low skilled NL NO ES high skilled SE CH UK females Latvia 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 1,00 0,80 0,60 0,40 0,20 0,00 DK FI DE IE low skilled NL NO ES high skilled SE CH UK females Lithuania 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010 1,00 0,80 0,60 0,40 0,20 0,00 DK FI DE IE low skilled NL NO high skilled ES SE CH UK females Source: Data retrieved from Brucker et al. (2013). Note: The shares of low-skilled, high-skilled and females are calculated based on the total number. Medium skill-levels are not reported but account for the remaining share. 54 A.3 Additional Data Sources A.3.1 Migrant Stock in the UK The resident population of Estonians in the UK (migrant stock) 2004-2013 has been approximated based on the Annual Population Survey (APS) and NINo. The APS by the national statistical office reports only the 60 most common countries by nationality in the reference year. Almost all NMS are among the 60 most common origin countries except for Estonia and Slovenia over the period 2006-2013. In 2004, next to Estonia and Slovenia, Latvian citizenship is not on the list, and in 2005, the Hungarian is not listed. I use NINo figures to describe the migrant flow and assume that they reflect the migrant stock, too, because residents have greater interest to use national insurance provisions. Although NINo are attributed only once and remain fluxes, adding the numbers produces very high figures in ignorance of return migration. For 2004, I treated Estonia, Slovenia and Latvia as one country and added their NINo in order to calculate the countries’ share in the numbers issued. The same was done in 2005 for Estonia, Slovenia and Hungary. For the period 2006-2013, only Estonian and Slovenian NINo shares were calculated. These shares are reported in table 5. Table 8: Shares of migrants per nationality in UK NINo numbers 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 EE 0,20 0,27 0,83 0,74 0,72 0,84 0,87 0,88 0,85 0,81 SL 0,12 0,05 0,17 0,26 0,28 0,16 0,13 0,12 0,15 0,19 LV 0,68 HU 0,68 Source: Own calculations based on British Office for National Statistics (2013). Since the APS reports the aggregate EU8 nationals and because five or six of EU8 nationalities are single-listed, the missing countries account for the residual number of EU8 nationals. Therefore, I applied the shares in NINo on the residual figures, in order to receive the number of Estonian citizens in the UK. The results are in fact close to Holland et al. (2011) estimates for an earlier period, as table 6 shows. The correlation of immigration and migrant stock is weaker for Estonian (50%) than for Latvian and Lithuanian citizenship. 55 Table 9: Comparison of approximated Estonian migrant stock in the UK 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 NINo 1060 2999 2160 1676 1434 1955 2227 2335 1643 1708 Results 1200 2700 4980 4440 3600 6720 6960 8800 5950 8100 Holland 3577 4618 5346 7681 3667 14100 . . . . residual EU8 nationals 6000 10000 6000 6000 5000 8000 8000 10000 7000 10000 Source: NINo; own calculations based on NINo and APS (British Office for National Statistics, 2013); Holland et al. (2011). A.3.2 Transitional Arrangements The restriction dummy is equal to one if a transitional arrangement is in place and zero for open labor market access from the year following the end of the transitional arrangement onwards (Fihel at al., 2015). Accordingly, all non-EU15 migration prior to 2004 is restricted. In 2004, only the UK, Ireland and Sweden lifted the restrictions. In 2006, the first transitional phase ended in Finland, Spain and Italy. In 2007, the Netherlands opened access, and in 2009, Denmark ended the second transitional phase. Germany was one of the countries taking advantage of all three phases and the entire period of 7 years of restricted labor mobility until 2011. Table 10: Restricted access to labor market 2003-2011 UK IE SE FI IT ES NL DK NO DE CH 2003 1 1 1 1 1 1 1 1 1 1 1 2004 0 0 0 1 1 1 1 1 1 1 1 2005 0 0 0 1 1 1 1 1 1 1 1 2006 0 0 0 0 0 0 1 1 1 1 1 2007 0 0 0 0 0 0 0 1 1 1 1 2008 0 0 0 0 0 0 0 1 1 1 1 2009 0 0 0 0 0 0 0 0 0 1 1 2010 0 0 0 0 0 0 0 0 0 1 1 2011 0 0 0 0 0 0 0 0 0 0 0 Source: Based on Fihel et al. (2015, p. 16). Norway and Switzerland are both EFTA members but only Norway is part of the EEA and Schengen area and therefore bound to apply EU rules on labor mobility 56 including the option for a transitional arrangement. Overall, its immigration policy was close to the Danish, opening access in 2009 (Dolvik and Eldring, 2006, pp. 16ff). Switzerland conducted a restrictive immigration policy towards EU8 migrants over the entire period and lifted restrictions only in 2011, as in Germany (OECD, 2012, pp. 55f). Table 7 depicts the dummy variable for restricted labor mobility for the period of 2003-2011. A.3.3 Trade and Capital Flows First, in order to use trade as a control, I calculated the bilateral trade volume as per cent of GDP. In the World Bank’s WDI (2015), aggregate trade volume is calculated as sum of imports and exports relative to the reporting country’s GDP in the reference year. The bilateral export and import figures are retrieved from the UN’s Comtrade (2015) database, setting the Baltic countries as reporter and the European countries as partners. The mirror figures differ and have not been chosen in order to maintain the focus on the migrant sending countries. Since the trade figures are denoted in US dollars (at market prices), I used the UN’s data for the Baltic countries’ GDP in current US dollars (UN Statistics Division, 2015) in order to calculate bilateral trade volume over all commodities i: 𝑇𝑏𝑒𝑡 = 𝑗 𝑗 ∑𝑖 𝑋𝑏𝑒𝑡 +∑𝑖 𝑀𝑏𝑒𝑡 𝑌𝑏𝑡 . Second, in theory, capital flows occur in the opposite direction of labor flows. Therefore, I use FDI inflows as a control for workforce outflows, but ignore FDI outflows, which are also much smaller and suffer from more lacks. UNCTAD (2014) reports comprehensive bilateral FDI statistics from which I use the inflow data. Net of costs increases in liabilities (investments) and also net decreases in liabilities (disinvestments) are reported in the FDI inflows so that the inflow figure may turn negative if the disinvestments are greater than the investments (UNCTAD, 2014). Since FDI is reported in US dollars (at market value), I calculate the share of inward FDI as per cent of the recipient country’s GDP in US dollars in the reference year, based on the data by UN Statistics Division (2015): 𝐾𝑏𝑒𝑡 = 𝐹𝐷𝐼𝑏𝑒𝑡 𝑌𝑏𝑡 . 57 References Alvarez-Plata, Patricia; Brücker, Herbert; Siliverstovs, Boriss (2003): Potential Migration from Central and Eastern Europe into the EU15. An Update. European Commission, DG Employment and Social Affairs (ed.). Retrieved 2015-08-28, from http://www.diw.de/documents/ dokumentenarchiv/17/41211/report_european_commission_20040218.pdf. Baas, Timo; Brücker, Herbert; Hauptmann, Andreas (2010): Labor Mobility in the Enlarged EU. Who Wins, Who Loses? In: Martin Kahanec und Klaus F. Zimmermann (ed.): EU Labor Markets after Post-Enlargement Migration. Heidelberg, pp. 47–71. Borchers, Kevin; Breustedt, Wiebke (2008): Die Datenlage im Bereich der internationalen Migration. Europa und seine Nachbarregionen. Bundesamt für Migration und Flüchtlinge (ed.). Retrieved 2015-10-20, from https://www.bamf.de/SharedDocs/Anlagen/DE/Publikationen/ WorkingPapers/wp18-internationale-migration.pdf. Borjas, George J. (1987): Self-Selection and the Earnings of Immigrants. In: The American Economic Review 77 (4), pp. 531–553. Retrieved 2015-08-12, from http://www.jstor.org/stable/1814529. Borjas, George J. (1999): The Economic Analysis of Immigration. Retrieved 201508-26, from http://www.ppge.ufrgs.br/giacomo/arquivos/eco02268/borjas1999.pdf. Brauner, Yariv (2010): Brain Drain Taxation as Development Policy. Retrieved 2015-11-01, from http://scholarship.law.ufl.edu/facultypub/169/. British Department for Work and Pensions (2015): NINo Registrations. Retrieved 2015-11-13, from https://sw.stat-xplore.dwp.gov.uk/webapi/jsf/ login.xhtml. British Office for National Statistics (2015): Datasets and reference tables. Retrieved 2015-11-13, from http://www.ons.gov.uk/ons/taxonomy/ index.html?nscl=Migration#tab-data-tables. 58 Brucker, Herbert; Baas, Timo; Boeri, Tito et al. (2009): Labor Mobility within the EU in the Context of Enlargement and the Functioning of the Transitional Arrangements. Final Report. European Commission, DG Employment, Social Affairs and Equal Opportunities (ed.). Retrieved 2015-08-11, from http://www.frdb.org/upload/file/Final_Report.pdf. Brucker, Herbert; Capuano, Stella; Marfouk, Abdeslam (2013): Education, Gender and International Migration. Insights from a Panel-Dataset 19802010. Retrieved 2015-11-17, from http://www.iab.de/en/daten/iab-braindrain-data.aspx. DeStatis (2015): Genesis-Online Datenbank. Wanderungsstatistik. Retrieved 2015-10-11, from https://www.genesis.destatis.de/genesis/online. Djajic, Slobodan; Kirdar, Murat G.; Vinogradova, Alexandra (2015): SourceCountry Earnings and Emigration. In: Journal of International Economics. Retrieved 2015-12-23, from http://www.sciencedirect.com/science/article/ pii/S0022199615001750. Dolvik, Jon Erik; Eldring, Line (2006): The Nordic Labor Market Two Years after the EU Enlargement. Mobility, Effects and Changes. Retrieved 201511-09, from https://books.google.de/books?id=NviFy4ZGCd8C&lpg=PP1& hl=de&pg=PA4#v=onepage&q&f=false. Engbersen, Godfried; Jansen, Joost (2013): Emigration from the Baltic States: Economic Impact and Policy Implications. In: OECD (ed.): Coping with Emigration in Baltic and East European Countries, pp. 13–28. Eurostat (2015): Database. Retrieved 2015-12-22, from http://ec.europa.eu/ eurostat/data/database. Faini, Riccardo (2006): Remittances and the Brain Drain. Retrieved 2015-11-01, from http://ftp.iza.org/dp2155.pdf. Feenstra, Robert C.; Inklaar, Robert; Timmer, Marcel P. (2013): The Next Generation of the Penn World Table. V 8.1. Retrieved 2015-11-17, from http://febpwt.webhosting.rug.nl/Dmn/AggregateXs/VariableCodeSelect. Fihel, Agnieszka; Janicka, Anna; Kaczmarczyk, Pawel; Nestorowicz, Joanna (2015): Free Movement of Workers and Transitional Arrangements. Lessons 59 from the 2004 and 2007 Enlargements. Retrieved 2015-07-08, from http://ec.europa.eu/social/BlobServlet?docId=14000&langId=en. Galgóczi, Béla; Leschke, Janine; Watt, Andrew (2012): EU Labor Migration in Troubled Times. Skills Mismatch, Return and Policy Responses. Farnham. German Federal Ministry of the Interior (2005): Migrationsbericht 2005. Migrationsbericht der Beauftragten der Bundesregierung für Migration, Flüchtlinge und Integration im Auftrag der Bundesregierung. Retrieved 2015-09-10, from https://www.bmi.bund.de/SharedDocs/Downloads/DE/ Themen/MigrationIntegration/migrationsbericht_2003.pdf?__blob=publicati onFile. Greenwood, Michael J. (1985): Human Migration. Theory, Models and Empirical Studies. In: Journal of Regional Science 25 (4). Retrieved 2015-10-11, from http://onlinelibrary.wiley.com/doi/10.1111/j.1467-9787.1985.tb00321.x/ abstract. Haas, Hein de (2008): Migration and Development. A Theoretical Perspective. In: International Migration Review 44 (1), pp. 227–264. Retrieved 2015-10-20, from http://www.imi.ox.ac.uk/pdfs/wp/wp-09-08.pdf. Harris, John R.; Todaro, Michael P. (1970): Migration, Unemployment and Development. A Two-Sector Analysis. In: The American Economic Review 60 (1), pp. 126–142. Retrieved 2015-08-14, from http://www.jstor.org/ stable/1807860. Hazans, Mihails (2012): Selectivity of Migrants from Baltic Countries Before and After Enlargement and Responses to the Crisis. In: Béla Galgóczi, Janine Leschke and Andrew Watt (ed.): EU Labor Migration in Troubled Times. Skills Mismatch, Return and Policy Responses. Farnham, pp. 169–210. Hazans, Mihails (2013): Emigration from Latvia, Recent Trends and Economic Impact. In: OECD (ed.): Coping with Emigration in Baltic and East European Countries, pp. 65–110. Hazans, Mihails; Philips, Kaia (2010): The Post-Enlargement Migration Experience in the Baltic Labor Markets. In: Martin Kahanec und Klaus F. 60 Zimmermann (ed.): EU Labor Markets after Post-Enlargement Migration. Heidelberg, pp. 255–304. Holland, Dawn; Fic, Tatiana; Rincon-Aznar, Ana; Stokes, Lucy; Paluchowski, Pawel (2011): Labor Mobility within the EU. The Impact of Enlargement and the Functioning of the Transitional Arrangements. Final Report. European Commission, DG Employment, Social Affairs and Equal Opportunities (ed.). Retrieved 2015-08-11, from http://ec.europa.eu/social/BlobServlet?docId= 7120&langId=en. Iara, Anna (2008): Skill Diffusion by Temporary Migration? Returns to Western European Work Experience in Central and East European Countries. Retrieved 2015-09-18, from http://wiiw.ac.at/skill-diffusion-by-temporarymigration-returns-to-western-european-work-experience-in-central-andeast-european-countries-dlp-546.pdf. International Finance Corporation (2013): Entreprise Surveys. Country Profiles. Retrieved 2015-11-15, from http://www.enterprisesurveys.org/. Irish Department of Social Protection (2015): Statistics on Personal Public Service Numbers Issued. Retrieved 2015-11-13, from http://www.welfare.ie/en/Pages/Personal-Public-Service-Number-Statisticson-Numbers-Issued.aspx. Jauer, Julia; Liebig, Thomas; Martin, John P.; Puhani, Patrick (2014): Migration as an Adjustment Mechanism in the Crisis? A Comparison of Europe and the United States. Retrieved 2015-08-09, from http://www.oecd.org/eu/Adjustment-mechanism.pdf. Kahanec, Martin; Zaiceva, Anzelika; Zimmermann, Klaus F. (2010): Lessons from Migration after EU Enlargement. In: Martin Kahanec und Klaus F. Zimmermann (ed.): EU Labor Markets after Post-Enlargement Migration. Heidelberg, pp. 3–46. Kancs, d'Artis (2010): Labor Migration in the Enlarged EU. A New Economic Geography Approach. Retrieved 2015-08-09, from https://www.lse.ac.uk/geographyAndEnvironment/research/Researchpapers/ 131%20Kancs.pdf. 61 Kancs, d'Artis; Kielyte, Julda (2002): Migration in the Enlarged European Union. A Perspective from the Baltic States. Retrieved 2015-08-10, from http://www.eeri.eu/documents/wp/EERI_RP_2002_04.pdf. Kancs, d'Artis; Kielyte, Julda (2008): Does Talent Migration Increase the Gap Between East and West? Retrieved 2015-08-09, from http://www.ssoar.info/ssoar/bitstream/handle/document/1642/ssoar-2008kancs_et_al-does_talent_migration_increase_the.pdf?sequence=1. Kancs, d'Artis; Kielyte, Julda (2010a): European Integration and Labor Migration. Retrieved 2015-08-09, from http://eiop.or.at/eiop/pdf/2010016.pdf. Kancs, d'Artis; Kielyte, Julda (2010b): Educating in the East, Emigrating to the West? Retrieved 2015-09-08, from http://www.eeri.eu/documents/wp/ EERI_RP_2010_01.pdf. Kaska, Veronika (2013): Emigration from Estonia, Recent Trends and Economic Impact. In: OECD (ed.): Coping with Emigration in Baltic and East European Countries, pp. 29–44. Kaufmann, Daniel; Kraay, Aart (2015): Worldwide Governance Indicators. Retrieved 2015-11-16, from http://info.worldbank.org/governance/wgi/ index.aspx#home. Lee, Everett S. (1966): A Theory of Migration. In: Demography 3 (1), pp. 47–57. Retrieved 2015-08-12, from http://www.jstor.org/stable/2060063. Marchiori, Luca; Shen, I-Ling; Docquier, Frederic (2009): Brain Drain in Globalization. A General Equilibrium Analysis from the Sending Countries' Perspective. Retrieved 2015-11-01, from http://ftp.iza.org/dp4207.pdf. Massey, Douglas S.; Arango, Joaquin; Hugo, Graeme; Kouaouci, Ali; Pellegrino, Adela; Taylor, J. Edward (1993): Theories of International Migration. A Review and Appraisal. In: Population and Development Review 19 (3), pp. 431-466. Retrieved 2015-10-20, from http://www.jstor.org/ stable/2938462. Mayda, Anna Maria (2010): International Migration. A Panel Data Analysis of the Determinants of Bilateral Flows. In: Journal of Population Economics 23 62 (4), pp. 1249–1274. Retrieved 2015-08-26, from http://www.jstor.org/ stable/40925859. Mayer, Thierry; Zignago, Soledad (2011): Notes on CEPII's Distances Measures. The GeoDist Database. Retrieved 2015-08-20, from http://www.cepii.fr/PDF_PUB/wp/2011/wp2011-25.pdf. Mayr, Karin; Peri, Giovanni (2009): Brain Drain and Brain Return. Theory and Application to Eastern-Western Europe. Retrieved 2015-07-27, from http://homepage.univie.ac.at/Papers.Econ/RePEc/vie/viennp/vie0907.pdf. Muravska, Tatjana (2011): Crisis in Latvia 2008-2010. Responsible Factors. Retrieved 2015-11-21, from http://www.hs-bremen.de/internet/forschung/ einrichtungen/itd/projekte/baltic/paper_-_muravska.pdf. OECD (2012): Free Movement of Workers and Labor Market Adjustment. Recent Experiences from OECD Countries and the European Union. Retrieved 2015-11-09, from https://books.google.de/books?id=IR82f8O3BBsC. Piore, Michael J. (1979): Birds of Passage. Migrant Labor and Industrial Societies. Cambridge. Pungas, Enel; Toomet, Ott; Tammaru, Tiit; Anniste, Kristi (2012): Are Better Educated Migrants Returning? Evidence from Multi-Dimensional Education Data. Retrieved 2015-08-09, from http://norface-migration.org/publ_ uploads/NDP_18_12.pdf. Purfield, Catriona; Rosenberg, Christoph B. (2010): Adjustment Under a Currency Peg. Estonia, Latvia and Lithuania during the Global Financial Crisis 2008-09. International Monetary Fund (ed.). Retrieved 2015-11-22, from http://www.imf.org/external/pubs/ft/wp/2010/wp10213.pdf. Roy, Arthur. D. (1951): Some Thoughts on the Distribution of Earnings. In: Oxford Economic Papers 3 (2), p. 135–146. Retrieved 2015-08-12, from http://www.jstor.org/stable/2662082. Schmidheiny, Kurt (2015): Short Guides to Microeconometrics. Panel Data, Fixed and Random Effects. Retrieved http://kurt.schmidheiny.name/teaching/panelmf.pdf. 2015-12-14, from 63 Schoorl, Jeannette; Heering, Lisbeth; Esveldt, Ingrid; Groenewold, George; et alia (2000): Push and Pull Factors of International Migration. A Comparative Report. Retrieved 2015-10-20, from https://www.nidi.nl/ shared/content/output/2000/eurostat-2000-theme1-pushpull.pdf. Sipaviciene, Andra; Stankuniene, Vlada (2013): The Social and Economic Impact of Emigration on Lithuania. In: OECD (ed.): Coping with Emigration in Baltic and East European Countries, pp. 45–64. Sjaastad, Larry A. (1962): The Costs and Returns of Human Migration. In: Journal of Political Economy 70 (5), p. 80–93. Retrieved 2015-08-12, from http://www.jstor.org/stable/1829105. Stark, Oded; Bloom, David E. (1985): The New Economics of Labor Migration. In: The American Economic Review 75 (2), pp. 173–178. Retrieved 201510-26, from http://www.jstor.org/stable/1805591. Stark, Oded; Helmenstein, Christian; Prskawetz, Alexia (1997): A Brain Gain With a Brain Drain. Retrieved 2015-11-01, from http://www.ihs.ac.at/ publications/eco/es-45.pdf. Swiss Federal Statistical Office (2015): Erwerbslosenstatistik gemäß ILO. Retrieved 2015-11-13, from http://www.bfs.admin.ch/bfs/portal/de/index/ themen/03/03/blank/data/01.html. Transparency International (2015): Corruption Perceptions Index. Open Knowledge Foundation (ed.). Retrieved 2015-11-16, from http://data.okfn.org/data/core/corruption-perceptions-index#resource-cpi. UN Comtrade (2015): Comtrade Database. Retrieved 2015-11-15, from http://comtrade.un.org/data/. UNCTAD (2014): Bilateral FDI Statistics 2014. Retrieved 2015-11-15, from http://unctad.org/en/Pages/DIAE/FDI%20Statistics/FDI-StatisticsBilateral.aspx. UNESCO (2015): Glossary of Migration Related Terms. Migrant and Migration. Retrieved 2015-10-13, from http://www.unesco.org/shs/migration/glossary. UN Statistics Division (2015): Data. National Accounts Estimates of Main Aggregates. Retrieved 2015-11-16, from http://data.un.org/. 64 World Bank (2015): World Development Indicators. Retrieved 2015-11-28, from http://databank.worldbank.org/data/reports.aspx?source=worlddevelopment-indicators. World Economic Forum (2014): The Global Competitiveness Report 2014-2015. Retrieved 2015-11-15, from http://www3.weforum.org/docs/WEF_ GlobalCompetitivenessReport_2014-15.pdf. Zaiceva, Anzelika; Zimmermann, Klaus F. (2008): Scale, Diversity, and Determinants of Labor Migration in Europe. Retrieved 2015-10-13, from http://ftp.iza.org/dp3595.pdf. Zaiceva, Anzelika; Zimmermann, Klaus F. (2012): Returning Home at Times of Trouble? Return Migration of EU Enlargement Migrants during the Crisis. Retrieved 2015-07-08, from http://ftp.iza.org/dp7111.pdf. Zakharenko, Roman (2012): Human Capital Acquisition and International Migration in a Model of Educational Market. In: Regional Science and Urban Economics 42 (5), pp. 808–816. Retrieved 2015-10-28, from http://www.sciencedirect.com/science/article/pii/S0166046211001323. Zimmermann, Klaus F. (1996): European Migration. Push and Pull. In: International Regional Science Review 19 (1 & 2), pp. 95–128. Retrieved 2015-10-12, from http://irx.sagepub.com/content/19/1-2/95.full.pdf+html.
© Copyright 2026 Paperzz