Labor Market and Growth Implications of Emigration: Cross-Country Evidence

BACKGROUND PAPER FOR THE WORLD DEVELOPMENT REPORT 2013 Labor Market and Growth Implications of Emigration: Cross-Country Evidence Shoghik Hovhannisyan The World Bank

Labor Market and Growth Implications of Emigration: Cross-Country Evidence Shoghik Hovhannisyan Abstract The number of migrants residing in 30 OECD countries increased from 42 million to nearly 59 million over 1990-2000. This paper studies the impact of emigrants with different education levels on their home countries employment-population ratio, GDP per worker, and its factors obtained by a production function decomposition. It uses migration data from 195 countries of origin to 30 major destination OECD countries in 1990 and 2000 and applies two different econometric approaches to estimate this impact. The first approach discusses how changes in native population due to emigration affect the growth of macroeconomic variables and constructs an instrument based on pull factors of migration and migrants networks to correct for endogeneity bias. The second approach estimates the elasticities of variables of interest with respect to emigration rates and uses instruments widely discussed in the literature: dummy variables for being in a colonial relationship; low-income countries; whether the migrant sending country s official language is English; distance; and country size. Estimation results across two econometric approaches indicate that total emigration rates increase GDP per worker in all countries, non-high income countries, and low and lower middle income countries, primarily driven by improvements in total factor productivity (TFP). In contrast, there is no robust significant impact of emigration on the employment-population ratio across different specifications. 1

1 Introduction Along with increased flows of capital, goods, and services, international labor mobility has become an inseparable part of globalization, with enormous economic, social and cultural implications in both countries of origin and destination. The number of foreign-born individuals residing in 30 OECD countries increased from 42 million to nearly 59 million over the period of 1990-2000. This paper studies the impact of emigration for different education levels on GDP per worker, its factors obtained by a production function decomposition, and the employment-population ratio in migrant-sending countries. It uses data on emigration from 195 countries of origin to 30 OECD destination countries which account for about 70 percent of total emigration and 90 percent of skilled emigration in the world. Emigration can have both a negative and a positive impact on migrant-sending countries. On one hand, these countries face deprivation of their labor force especially the most educated, known in the literature as a brain drain phenomenon. On the other hand they benefit from emigration in several ways. Migrants remit money home and these financial flows account for a significant share of GDP in developing countries. Remittances relax financial constraints of households which have a member or relative abroad and can increase not only their consumption of goods and services, but also expenditures on health and education, thus having both short-term and long-term effects on GDP. Emigration might also promote transfer of technology and knowledge across countries by facilitating more foreign direct investment (FDI), trade, and other partnerships through established diasporas abroad and their networks. Finally, the high probability of emigration of educated labor raises returns to education and, therefore, might lead to higher investments in education. As not everyone has a chance to migrate, this increases overall human capital in the migrant-sending country. There is a large empirical literature on emigration effects in migrant-sending countries and their various transmission mechanisms. Macroeconomic studies using cross-country data provide mixed evidence on the effects of emigration and remittances on growth and its drivers, with results highly dependent on the econometric approaches and instruments used to control for endogeneity bias. Estimations by Cartinescu et al. (2009) and Acosta et al. (2008) show an increase in growth due to remittances, while Chami et al. (2009) find a negative effect of remittances on growth volatility. However, Barajas et al. (2009) conclude that remittances have no impact on economic growth 2

in their cross-country analysis. At the same time, Easterly and Nyarko (2009) find no significant impact of brain drain or outflow of high-skill migrants on GDP growth in African countries by using distance from France, U.K., and U.S., and population as instruments to address endogeneity issues. In addition, Gould (1994) shows that U.S. bilateral trade is larger with countries that send more migrants to the U.S. and Head et al. (1998) estimate similar effects for Canada. A significant fraction of the emigration literature discusses its impact on human capital of migrantsending countries. Papers by Beine et al. (2008), Docquier et al. (2008), Docquier et al. (2007), and Easterly and Nyarko (2009) use a variety of instruments such as total population size; migration stocks at the beginning of the period; geographical proximity to developed countries; dummy variables for small islands, landlocked, least-developed, and oil exporting countries; former colonial links; etc., to correct for endogeneity bias in estimating these effects. These papers find a positive, significant impact of emigration on human capital formation in countries of origin due to a higher propensity to migrate for more educated people, which increases investments in education. Studies using household, firm, or individual-level data discuss various transmission mechanisms of emigration impact on growth. Yang (2008) finds an increase in remittances to households in the Philippines at the time of the Asian financial crisis, consistent with consumption smoothing. In contrast, government transfers have no impact on remittances in Mexico, according to Teruel and Davis (2000) or in Honduras and Nicaragua, according to Nielsen and Olinto (2007). Woodruff and Zenteno (2006) find that migration is associated with higher investment levels and profits when analyzing data on self-employed workers and small firm owners in urban areas of Mexico. Also, using panel data from rural Pakistan, Adams (1998) shows that availability of remittances helps to increase investment in rural assets by raising the marginal propensity to invest for migrant households. In addition, remittances increase households school attendance in El Salvador according to Cox and Ureta (2003) and improve health outcomes in Mexico according to Hilderbrandt and McKenzie (2005). Finally, a study by Saxenian (2002) concludes that emigration of India s highskill labor to Silicon Valley increased trade with and investment from the U.S., promoting creation of local high-technology industries. In terms of labor market outcomes, Mishra (2007), Aydemir and Borjas (2007), and Hanson (2007) find a positive correlation between wages and emigration in Mexico. This paper contributes to the emigration literature by studying the growth and labor market im- 3

plications of emigration across different education groups of population, using a new econometric approach. First, to address endogeneity and simultaneity bias in Ordinary Least Squares (OLS) estimation, it applies an Instrumental Variable (IV) approach with the following instruments adopted from the literature: (i) a dummy variable for ever being in a colonial relationship, (ii) a dummy variable for low-income countries, (iii) the average distance from migrant destination countries with an exception of selective countries: Australia, Canada, and the U.S., (iv) a minimum distance from selective countries, (v) country size in terms of population, including both residents and emigrants, and (vi) a dummy variable if the migrant-sending country s primary language is English. These instruments are used to estimate elasticities of the variables of interest with respect to emigration rates for different education groups. In addition, this study suggests a new instrument constructed based on pull factors of migration and migrants networks to estimate how changes in population due to emigration affect the growth of dependent variables. An increase in total immigration stocks of destination countries is primarily driven by either changes in immigration policies or labor demand, and is taken as exogenous to developments in countries of origin. At the same time, this higher demand for immigrants in destination countries would be distributed proportionally across countries of origin based on the size of their diasporas due to the importance of migrants networks in the cross-country mobility of population. Estimation results of emigration impact on different country groups based on their income levels indicate that total emigration rates increase GDP per worker in all countries, all non-high income countries, and all low and lower middle income countries. These results remain robust to the inclusion of different control variables and across different econometric specifications. The growth in GDP per worker is primarily driven by improvements in TFP. In contrast, emigration rates of secondary and tertiary educated individuals have no consistent significant effects on the variables of interest. Finally, there is no impact of emigration on the employment-population ratio. The rest of the paper is organized as follows. Section 2 introduces the theoretical framework for growth accounting. Section 3 discusses the estimation approach including two IV methods. Section 4 describes the data and construction of variables. Empirical results are presented in Section 5. Finally, Section 6 concludes. 4

2 Theoretical Framework This paper uses a growth accounting framework to analyze the impact of emigration on GDP per worker in migrant-sending countries. To study the channels of emigration impact it decomposes the GDP into three factors using the following Cobb-Douglas production function as in Caselli (2005): Y it = A it K α it(l it h it ) 1 α (1) where A it is TFP, K it is an aggregate capital stock, α is a capital share in GDP, and (L it h it ) is a quality adjusted workforce, with the number of workers L it multiplied by their average human capital h it, in country i and period t. In per-worker terms the production function can be written as: y it = A it k α it(h it ) 1 α (2) where k it is the capital-labor ratio (K it /L it ). K it is constructed using the perpetual inventory method: K it = I it + δk it 1 (3) where I it is investment in country i and period t and δ is a depreciation rate. The initial capital stock K 0 it is obtained from the steady-state expression for capital stock in the Solow model: K 0 it = I0 i g i + δ (4) where Ii 0 is a value of the investment series in the first available year and g i is an average geometric growth rate for the investment series between the first available year and 2000 for country i. To compute the time series for K it, investment in respective years is added to the initial capital stock. The average human capital h it is a function of average years of schooling in the population as expressed in the following equation: h it = e φ(s it) (5) 5

where s it is average years of schooling in country i and period t and φ(s it ) is a piecewise linear function with slope 0.13 for s it 4, 0.10 for 4 < s it 8, and 0.07 for 8 < s it. This function resembles the log-linear functional relationship between wages and years of education in the Mincerian approach, where wages are assumed to be proportional to human capital given the production function and perfect competition. Since international data on education and wages suggest that there are some differences in marginal rates of return across countries, those differences are introduced with the convexity. Finally, TFP, A it, is constructed as a residual. The empirical strategy consists of two econometric approaches which estimate the impact of different education groups of emigrants on the employment-population ratio and GDP per worker and its components as obtained above. In the first approach (IV1), following Easterly and Nyarko (2008), population is defined as: L = L D + L F (6) where L is a total native population which includes both residents L D and emigrants L F. The percentage change in native population can be expressed as: dl L = dl D L + dl F L (7) The second component in Equation (7) captures the effect of a change in population due to emigration. The IV1 approach estimates reduced form equations of the impact of a change in population due to emigration on variables of interest: employment-population ratio, GDP per worker, capital-worker ratio and average human capital as in the following equation: db i b i = α k + β k dlk F i L k i + ɛ i (8) where db i b i is a growth rate of each variable of interest in country i, dlk F i L k i is the change in native population due to emigration for education group k in country i, and ɛ i is a zero-mean random shock. The second approach, IV2, uses data as a pooled cross-section and estimates elasticities of variables of interest (b it ) with respect to emigration rates ( Lk F it ) by controlling for a year-fixed effect (η L k t ) as it 6

in the following equation: ln b it = α k + η t + γ k ln Lk F it L k it + ɛ i (9) 3 Estimation Approach This study estimates the impact of emigration on GDP per worker and its factors, obtained by a production function decomposition and the employment-population ratio in migrant-sending countries using cross-country data over the period of 1990-2000. It analyzes the impact of emigration for three different education groups of the native population: for all levels of education, those with secondary and tertiary education, and those with tertiary education. Distinguishing across these groups is important in understanding to what extent education of emigrants matters for development of their home countries. Low-skill emigration can simply lead to a decline in labor or influence countries of origin through remittances, promotion of FDI and trade, etc. In addition to these channels, high-skill emigration directly reduces the level of human capital in the migrantsending countries but might contribute to investments in education, given a higher likelihood to emigrate for individuals with more education, as emphasized in the literature. Estimating these effects in reduced form equations in an Ordinary Least Squares (OLS) might generate bias in the coefficients due to reverse causality or endogeneity. For example, emigration of highly educated people might decrease GDP per worker in the source countries, given a higher marginal productivity of high-skill labor compared to low-skill labor. At the same time, a low level of GDP per worker might induce migration of more people both high-skilled and low-skilled to higher-income countries with better standards of living. In terms of endogeneity, there might be other factors driving both emigration and GDP per worker such as civil wars, weak institutions, etc., which might reduce GDP growth and increase emigration to countries with better opportunities. To address these econometric problems this paper applies two different IV approaches. The first approach introduces a new instrument, while the second approach uses conventional instruments from the emigration literature. Comparing estimation results of these two approaches allows us to test their robustness. The first approach, IV1, studies how changes in native population due to emigration affect the 7

growth of the employment-population ratio, GDP per worker, capital-worker ratio, and average human capital. Using this regressor helps to separate the impact of emigration from changes in the structure of the domestic population. Migration to 30 OECD countries increased by 45 percent over the period of 1990-2000 with similar trends observed across all education groups: primary (27 percent), secondary (51 percent), and tertiary (68 percent). However, these substantial changes in absolute number of migrants had an insignificant impact on emigration rates in migrant sending countries due to a rise in their population and education levels (Table 1). There was a 24.3 percent increase in the total population of migrant-sending countries with 19.6, 25.1, and 52.5 percent growth, respectively, in the number of primary, secondary, and tertiary educated people. In addition, this approach estimates growth equations consisting of only time-variant variables, thus eliminating countries fixed effects, omission of which can cause endogeneity bias. Table 1: Emigration Rates in 1990 and 2000 by Education Groups 1 Variable Mean Confidence Interval Total Emigration Rate, 1990.062 (0.047,0.077) Total Emigration Rate, 2000.067 (0.051,0.082) Emigration Rate of Secondary and Tertiary Educated, 1990.099 (0.078,0.120) Emigration Rate of Secondary and Tertiary Educated, 2000.101 (0.08,0.122) Emigration Rate of Tertiary Educated, 1990.208 (0.173,0.243) Emigration Rate of Tertiary Educated, 2000.194 (0.163,0.227) The mean computed in the table is non-weighted arithmetic mean. As OLS estimates of the reduced form equation (8) might be prone to simultaneity or omitted variable bias, the IV technique is used to correct for this. Migrants networks and pull factors of migration provide variations in emigration exogenous to migrant-sending countries conditions and, therefore, can serve as a basis for constructing an instrument. There are economic incentives for labor mobility between OECD countries and the rest of the world given a huge gap in income levels. In these circumstances, migrants networks stimulate migration flows, as having individuals from the same countries of origin provides access to jobs and other information, substantially reducing migration costs. Figure 1 in the Appendix depicts these network externalities, indicating that countries with high emigration rates or with large diasporas in 1990 tend to have high emi- 1 Emigration rates are computed for 195 migrant-sending countries in each year. 8

gration rates in 2000 as well. Each point on these graphs shows a share of emigrants in the total native population in each migrant-sending country in 1990 and 2000 for three education levels: all, secondary and tertiary, and tertiary, thus highlighting the key role of networks in a choice to emigrate. In addition, Figure 2 illustrates the network effects for total emigration from India and Philippines where distribution of migrants across major destination countries remains relatively stable over time. Literature mostly discusses networks as a decisive factor in migrants location choices in the context of subnational data and this study expands the existing literature by using network effects for country level analysis. The growth in the total number of immigrants in each of 30 destination countries, which might be a combination of different factors such as increases in overall labor demand and changes in immigration policies, is also used to construct the instrument. Assuming there are economic incentives for emigration from developing to developed countries, a higher demand from destination countries triggers more emigration. At the same time, as migrants networks or diasporas play an important role in migrants destination choices, an increase in the number of immigrants in the destination country from different countries of origin is likely to be proportional to the sizes of their diasporas. The IV1 approach consists of the following steps. First, the growth in the total number of immigrants in 2000 relative to 1990 is computed for each of 30 OECD destination countries using the actual number of immigrants: G k ij = Ek ij,2000 Ek j,1990 E k j,1990 (10) where G k ij is a growth rate in the total number of immigrants with education level k in destination country j in 2000 relative to 1990 to be used for country of origin i, E k ij,2000 is the actual number of immigrants in country j with education level k excluding immigrants from country i in 2000, and E k j,1990 is the actual number of immigrants in destination country j and education level k in 1990. Excluding the number of migrants from country i in the total number of immigrants in destination countries in 2000 eliminates any impact of country of origin i on an increase of immigration in destination countries. Therefore, this measurement of an immigration growth in destination countries is purely demand driven which ensures the exogeneity of the constructed instrument. 2 Next, these destination countries growth rates are applied to the number of migrants 2 The estimations results are similar when Eij,2000 k includes immigrants from country i as well. 9

from each country of origin i in the respective destination country j in 1990 in order to impute the number of migrants in 2000: [ ] Êi,j,2000 k = Ei,j,1990 k 1 + G k ij (11) where Êk i,j,2000 is the imputed number of migrants from country of origin i in destination country j with education level k in 2000, and Ei,j,1990 k is the actual number of migrants from country of origin i in destination country j with education level k in 1990. The imputed total number of emigrants in each country of origin i is obtained by summing across the destination countries: L k F,i,2000 = j Ê k i,j,2000 (12) Finally, the instrument for change in population due to emigration for each country i during the period of 1990-2000 is constructed as: dl k F,i L k i = L k F,i,2000 Lk F,i,1990 L k i,1990 (13) where L k F,i,1990 and Lk i,1990 are respectively the actual number of emigrants and population in country i in 1990 with education level k. This instrument has a strong explanatory power for all education levels of emigrants, as shown in Table 2. Table 2: IV1, First-Stage Regression Results for Changes in Population due to Emigration by Different Education Groups Dependent Variable Coefficient t-stat R-squared Observations All Emigrants.940 16.06 0.572 195 Secondary and Tertiary Educated Emigrants.988 17.59 0.616 195 Tertiary Educated Emigrants 1.0 13.23 0.475 195 The second approach, IV2, uses data as a pooled cross-section for the years 1990 and 2000 and studies the impact of emigration rates on the variables of interest for different education groups as shown in equation (9). To address possible reverse causality and endogeneity issues, it applies instruments selected from the emigration literature. The instruments for total emigration rates 10

include dummy variables for ever having been in a colonial relationship (Colony) and low-income countries (Low Income); the average distance from destination countries, with the exception of selective countries: Australia, Canada, and the U.S. (Distance); a minimum distance from selective countries (Minimum Distance); and a country size, in terms of population including both residents and emigrants (Population). Migrants are likely to face lower adjustment costs in the destination country if the home country was its former colony due to similarity in institutions, language, and stronger political ties. A dummy for low income countries is used as an instrument to capture financial constraints of potential migrants which reduce their costly cross-country mobility. Also, the physical distance between migrant-sending and receiving countries affects the travel costs for the initial move and visits home. In addition, migrants are also better informed about neighboring countries than distant ones. Distinguishing between selective and other countries is important in the analysis of emigration by education groups, as around 63 percent of migrants with secondary and tertiary education and 72 percent of migrants with tertiary education to 30 OECD countries were hosted by selected countries in 2000. All destination countries with the exception of Australia, Canada, the U.S., New Zealand and Mexico, are on the same continent and the average distance is more indicative of migration costs. However, the minimum distance is more informative for Australia, Canada and the U.S. given their highly dispersed locations. Finally, small countries tend to be more open to emigration due to universal or nearly equal immigration quotas based on countries of origin in migrant-receiving countries. The instruments for emigration rates of secondary and tertiary educated groups vary. They include a dummy variable if the migrant-sending country s primary language is English (English), as it increases the transferability of migrants skills for these education levels in the selective countries attracting most of them where English is an official language. The dummy for low-income countries is dropped for these education groups since emigrants with secondary and tertiary education are less financially constrained compared to the emigrants with no formal education or only a primary education. Tables 3 and 4 show the first-stage regression results for emigration rates by different education groups with instruments discussed above for all countries and non-high-income countries. All instruments have the expected sign and significance. Emigration increases for all education groups 3 In this and following tables the numbers in parentheses are standard errors of the coefficients and (*) indicates significance level at 10 percent, (**) at 5 percent, and (***) at 1 percent. 11

Table 3: IV2, First-Stage Regression Results for Emigration Rates by Education Groups: All Countries Instruments All Secondary and Tertiary Tertiary Colony 0.582** (0.203) 3 0.723*** (0.21) 0.633** (0.203) Low Income -0.932*** (0.195) -0.107 (0.176) 0.112 (0.169) Distance -1.168*** (0.143) -0.680*** (0.117) -0.419*** (0.107) Minimum Distance -1.252*** (0.237) -0.952*** (0.227) -0.583** (0.197) Population -0.260*** (0.041) -0.254*** (0.037) -0.214*** (0.033) English 0.431*** (0.125) 0.674*** (0.115) Common Language 0.432** (0.15) Observations 341 341 341 R-squared 0.458 0.383 0.354 Table 4: IV2, First-Stage Regression Results for Emigration Rates by Education Groups: Non-High Income Countries Instruments All Secondary and Tertiary Tertiary Colony 0.827*** (0.246) 0.913*** (0.241) 0.738** (0.232) Low Income -0.666*** (0.189) 0.136 (0.162) 0.326* (0.158) Distance -1.140*** (0.241) -0.796*** (0.176) -0.531*** (0.157) Minimum Distance -2.007*** (0.2) -1.730*** (0.165) -1.216*** (0.142) Population -0.195*** (0.046) -0.151*** (0.037) -0.120*** (0.034) English 0.540*** (0.124) 0.828*** (0.114) Common Language 0.076 (0.216) Observations 247 247 247 R-squared 0.522 0.526 0.471 if a country has ever been in a colonial relationship (Colony). A dummy variable for low income countries (Low Income) is significant and negative only for all emigrants, while it has no or low explanatory power for higher education groups. The average distance to destination countries with an exception of selected countries (Distance) and the minimum distance to selected emigration countries (Minimum Distance) negatively affect emigration rates of all education groups. The results indicate that Mimium Distance is more important than Average Distance, given a high share of migrants moving to selected destination countries and these estimates have overall lower magnitudes for higher education levels. As expected, small countries are more open and emigration rates decline with population (Population). To capture linguistic proximity and, therefore, lower 12

assimilation barriers, two variables are used: (i) a dummy variable if official or national languages and languages spoken by at least 20 percent of the population of the country are spoken in the destination country (Common Language) and (ii) a dummy variable if English is the official or national language and language spoken at least by 20 percent of the population of the country (English). Using a dummy variable English instead of Common Language for more educated emigrants is more relevant, as the majority of these emigrants are in three English-language destination countries: Australia, Canada and the U.S. and language is essential for skill transferability at higher education levels. The results indicate that having a common language (Common Language) loses significance when high-income countries are excluded from the sample, while the impact of variable English remains positive and significant for both groups of countries and for both high education levels. Based on these estimates, a dummy for low income countries (Low Income) is dropped from the analysis of secondary and tertiary educated emigrants and tertiary educated emigrants, and a dummy for commonly spoken language (Common Language) is removed from the list of instruments for all emigrants. The remaining instruments have high explanatory power as can be seen from the reported R 2. 4 Data Description This study uses the migration dataset by Docquier, Lowell, and Marfouk (2008) which provides the number of migrants from 195 migrant-sending countries to 30 main destination OECD countries. These emigration stocks account for about 70 percent of total emigration and 90 percent of skilled emigration in the world. The dataset classifies emigrants into three groups based on education: high-skill, medium-skill, and low-skill emigrants with respectively a post-secondary, an upper secondary, and a primary or no formal education. It also provides emigration rates for each education group defined as a share of emigrants in the total native population including residents and emigrants in the same education category. Country-level aggregate variables including the employment-population ratio, GDP per worker, capital per worker, and labor inputs are obtained from the Penn World Tables (PWT) by Heston, Summers and Bettina (PWT 7.0). First, the number of workers in each country i and year t is computed as (rgdpch it pop it /rgdpwok it ), where rgdpch it is a PPP converted GDP per capita 13

(Chain Series) at 2005 constant prices, pop it is a population, and rgdpwok it is a PPP Converted GDP Chain per worker at 2005 constant prices. To construct the employment-population ratio the number of workers is divided by the population. The capital-worker rat io k is computed using the perpetual inventory method: K it = I it + δk it 1 (14) where I it is investment and δ is a depreciation rate. I it is computed as (rgdpl it pop it ki it ), where rgdpl it is a PPP converted GDP per capita (Laspeyres) at 2005 constant prices, pop it is population, and ki it is an investment share of PPP converted GDP per capita at 2005 constant prices in country i and year t. The depreciation rate δ equals 0.06, which is a conventional value used in the literature. In addition, PWT 7.0 provides data on several control variables discussed below such as government size (kg it ) and openness of the economy (openk it ) measured respectively as the shares of government expenditures and trade, including exports and imports, in GDP. The average human capital h it is constructed using average years of schooling in the population over 25 years old from the Barro - Lee dataset. As in Docquier and Marfouk (2006), human capital indicators are replaced with those from De La Fuente and Domenech (2002) for OECD countries. For countries where Barro and Lee measures are missing, the proportion of educated individuals is predicted using the Cohen and Soto (2007) measures. In the result, there are 25 missing observations for 1990 and 35 for 2000 accounting respectively for 15 and 20 percent of total observations, which are imputed using the GDP per worker. Finally, TFP is constructed as a residual. To obtain instruments for the IV2 approach, this study uses the GeoDist database by Mayer and Zignago (2011), which provides information on colonial relationships, distance between countries and countries spoken languages. Colonial relationship is defined as ever having been in a colonial relationship. The distances are calculated following the great circle formula, which uses latitudes and longitudes of the most important cities or agglomerations in terms of population. Finally, spoken language is an official or national languages and languages spoken by at least 20 percent of the population. The control variables on legal origins of countries and political stability as described below are 14

taken from the Levine, Loayza and Beck (2000) dataset. The measurement of legal origins are dummy variables for British, French, German and Scandinavian legal origins. The variables on political stability include revolution and coups; assassinations; and ethnic fractionalization. A revolution is defined as any illegal or forced change in the top governmental elite, any attempt at such a change, or any successful or unsuccessful armed rebellion whose aim is independence from the central government. Coup d Etat is an extraconstitutional or forced change in the top government elite and/or its effective control of the nation s power structure in a given year. This excludes unsuccessful coups, with data averaged over 1960-1990. The measurement of assassinations is given by the average number of assassinations per thousand inhabitants over 1960-1990. Ethnic fractionalization represents an average value of five indices of ethnolinguistic fractionalization with values ranging from 0 to 1, where higher values denote higher levels of fractionalization. 5 Estimation Results This paper studies emigrants impact on the GDP per worker and its production factors: capitalworker ratio, average human capital, and TFP, and employment-population ratio. It estimates equations (8) and (9) using the IV1 and IV2 econometric approaches described above for six country groups based on the World Bank (WB) classification: (1) low income, (2) lower middle income, (3) upper middle income, (4) all, (5) all non-high income, and (6) all low income and lower middle income countries. To test the robustness of results, IV1 uses control variables adopted from Levine et al. (2000) such as logarithms of the initial levels of the dependent variable and average human capital in 1990. Next, it augments this list of regressors with growths in government size, measured as a share of government expenditures in GDP and openness of the economy, computed as a trade share in GDP. The IV2 approach first controls for logarithms of shares of government expenditures and trade in GDP and then adds variables on financial development and political stability. Dummy variables for British, French, German and Scandinavian legal origins exogenously drive differences in the legal rules covering secured creditors, the efficiency of contract enforcement, and the quality of accounting standards: hence, financial development. The number of revolutions and coups and the number of assassinations per thousand of inhabitants averaged over 1960-1990 and an index of ethnic fractionalization capture the variations in political stability. 15

Table 5 in the Appendix reports regression results of emigration impact on GDP per worker for different education groups as in equations (8) and (9). In IV1 estimates, a one percent change in native population due to total emigration increases GDP per worker for all countries by 2.1 percent at the 10 percent significance level; for all non-high income countries by 1.67 percent; and for all low income and lower middle income countries by 1.88 percent at the 5 percent significance level. These coefficients retain their significance and signs when adding control variables to the regressions. Moreover, the signs and significance of these coefficients is consistent with the IV2 estimates, although their magnitudes differ. In particular, the GDP per worker grows in response to an increase in total emigration rates for all countries by 0.69 percent; for all n on-high income countries by 0.46 percent; and for all low income and lower middle income countries by 0.45 percent. The sensitivity analysis shows that this impact remains robust when control variables are included in the regressions. Decomposing the GDP per worker into capital per worker, human capital, and TFP helps to understand the main channels of GDP growth driven by emigration. There is no robust significant estimate of the impact of emigration on capital per worker across the IV1 and IV2 approaches (Table 6). In the IV1 regressions, a change in population due to emigration of secondary and tertiary educated individuals consistently raises capital per worker in all non-high income countries with a magnitude in the range of 0.18-0.24 at the 10 percent significance level across different specifications. In contrast, the results for IV2 indicate that there is an increase in capital per worker in response to a one percent change in total emigration rates for all countries and all nonhigh income countries across all estimates with a coefficient in the range of 0.1-0.22. There is no consistent significant estimate of emigration s impact on human capital across different econometric specifications in the IV1 approach (Table 7). In IV2, the only robust result is a positive impact of total emigration rates on all countries with a coefficient in the range of 0.05-0.09. Table 8 reports results for an impact of emigration rates for different education groups on the last component, TFP. Similar to GDP per worker, TFP increases in all countries, all non-high income countries, and all low and lower middle income countries in both IV1 and IV2 estimates. Thus, despite the differences in IV1 and IV 2 estimation techniques and the inclusion of various control variables, some results remain robust across all specifications. They indicate that migrant-sending countries GDP per worker benefits from total emigration, mainly through improvements in TFP. These changes in 16

TFP might be a result of trade, FDI, and other cross-country partnerships facilitated by established diasporas abroad which lead to a transfer of knowledge and technology. This paper also studies the labor market implications of emigration for different education groups by using both the IV1 and IV2 approaches to estimate the impact of emigration on the employmentpopulation ratio. Overall, there is no consistent significant estimate of this impact across different education groups and specifications. The IV1 estimation results in Table 9 indicate that the employment-population ratio declines by 0.17-0.21 percent in response to a one percent change in population of secondary and tertiary educated due to emigration for upper middle income countries. However, there is no significant change in the employment-population ratio of other income groups across different education levels. The IV2 estimates produce no consistent significant results across different specifications. In addition, the results for GDP per worker can serve as a basis for wage analysis. In the conditions of perfect competition and constant returns to scale, wages are equal to the marginal product of labor, expressed as (1 α) Y it L it. Therefore, wages rise in response to an increase in total emigration rates in all countries, all non-high income countries, and all low and lower middle income countries. 6 Conclusions This paper studies the impact of emigration on several macroeconomic variables of migrant-sending countries using 1990 and 2000 emigration data from 195 source countries to 30 OECD destination countries. It applies two econometric approaches varying by the choice of instruments and specifications. The first approach studies the impact of changes in the native population due to emigration on the growth of employment-population ratio, GDP per worker, and its components. To overcome the endogeneity bias, it uses instruments based on migration pull factors and migrants networks. The second approach estimates the elasticities of variables of interest with respect to emigration rates, using conventional instruments from the literature such as colonial relationship, distance, country size, country s development level, and English as a primary language. Estimation results indicate that total emigration rates increase GDP per worker in all countries, non-high income countries, and low and lower middle income countries: an effect primarily driven by improvements in TFP. These results are robust to both econometric approaches and inclusion of various con- 17

trol variables in the regressions. In addition, emigration has no consistent significant impact on migrant-sending countries employment-population ratio across different specifications. 18

References [1] Acosta, P., C. Calderon, P. Fajnzylber and H. Lopez (2008): What is the Impact of International Remittances on Poverty and Inequality in Latin America?, World Development, 36, 1: 89-114. [2] Adams, R. H. Jr. (1998): Investment, and Rural Asset Accumulation in Pakistan, Economic Development and Cultural Change, 47, 1: 155-173. [3] Aydemir, A., and G. J. Borjas (2007): A comparative analysis of the labor market impact of international migration: Canada, Mexico, and the United States, Journal of the European Economic Association, 5, 4: 663-08. [4] Barajas, A., R. Chami, C. Fullenkamp, M. Gapen and P. Montiel (2009): Do Workers Remittances Promote Economic Growth?, IMF Working Paper, WP/09/153. [5] Beine, M., F. Docquier and C. Oden-Defoort (2011): A Panel Data Analysis of The Brain Gain, World Development, forthcoming. [6] Beine, M., F. Docquier and C. Ozden (2011): Diasporas, Journal of Development Economics, 95, 1: 30-41. [7] Beine, M., F. Docquier and H. Rapoport (2001): Brain drain and economic growth: theory and evidence, Journal of Development Economics, 64, 1: 275-89. [8] Beine, M., F. Docquier and H. Rapoport (2007): Measuring international skilled migration: new estimates controlling for age of entry, World Bank Economic Review, 21, 2: 249-54. [9] Beine, M., F. Docquier and H. Rapoport (2008): Brain drain and human capital formation in developing countries: winners and losers, Economic Journal, 118: 631-652. [10] Card, D. (2001): Immigrant Inflows, Native Outflows, and the Local Market Impacts of Higher Immigration, Journal of Labor Economics 19: 22-64. [11] Catrinescu, N., M. Leon-Ledesma, M. Piracha and B. Quillin (2009): Remittances, Institutions, and Economic Growth, World Development 37, 1: 81-92. 19

[12] Caselli, F. (2005): Accounting for Cross-Country Income Differences, Handbook of Economic Growth, in: Philippe Aghion and Steven Durlauf (ed.), Handbook of Economic Growth, edition 1, volume 1, chapter 9: 679-741. Elsevier. [13] Chami, R., D. Hakura and P. Montiel (2009): Remittances: An Automatic Output Stabilizer?, IMF Working Paper, WP/09/91. [14] Cox, A. and M. Ureta (2003): International migration, remittances and schooling: evidence from El Salvador, Journal of Development Economics, 72, 2: 429-461. [15] Docquier, F., O. Faye and P. Pestieau (2008): Is migration a good substitute for subsidies, Journal of Development Economics, 86, 2: 263-76. [16] Docquier, F., O. Lohest and A. Marfouk (2007): Brain drain in developing countries, World Bank Economic Review, 21, 2: 193-218. [17] Docquier, F., B. L. Lowell and A. Marfouk (2009): A gendered assessment of the brain drain, Population and Development Review, 35, 2: 297-321. [18] Docquier, F. and A. Marfouk (2006): International migration by educational attainment (1990-2000), in C. Ozden and M. Schiff (eds): International Migration, Remittances and Development, Palgrave Macmillan: New York. [19] Easterly, W. and Y. Nyarko (2009): Is the Brain Drain Good for Africa? Chapter 11 in J. Bhagwati and G. Hanson (eds): Skilled immigration: problems, prospects and policies, Oxford University Press: 316-60. [20] Gould, D. M. (1994): Immigration links to the home country: Empirical implications for U.S. bilateral trade flows, Review of Economics and Statistics, 76: 302-316. [21] Grogger, J. and G. Hanson (2011): Income maximization and the selection and sorting of international migrants, Journal of Development Economics, 95, 1: 42-57. [22] Grubel, H. and A. Scott (1966): The international flow of human capital, American Economic Review, 56: 268-74. [23] Hanson, G. H. (2007): International Migration and the Developing World, Mimeo, UCSD. 20

[24] Head, K., J. Ries and D. Swenson (1998): Immigration and trade creation: Econometric evidence from Canada, Canadian Journal of Economics, 31, 1: 47-62. [25] Hildebrandt, N. and D. McKenzie (2005): The effects of migration on child health in Mexico, World Bank Policy Research Working Paper, 3573. [26] Levine, R., N. Loayza and T. Beck (2000): Financial intermediation and growth: Causality and causes, Journal of Monetary Economics, 46: 31-77. [27] Mayda, A.M. (2010): International migration: a panel data analysis of the determinants of bilateral flows, Journal of Population Economics, 23, 4: 1249-74. [28] Mishra, P. (2007): Emigration and wages in source countries: Evidence from Mexico, Journal of Development Economics, 82, 1: 180-199. [29] Nielsen, M. E., andp. Olinto (2007): Do Conditional Cash Transfers Crowd Out Private Transfers? in P. Fajnzylber and H. Lopez (eds.): Remittances and Development: Lessons from Latin America, The World Bank, Washington, DC: 253-298. [30] Saxenian, A. (1999): Silicon Valley s new immigrant entrepreneurs, Public Policy Institute of California. [31] Teruel, G. and B. Davis (2000): Final report: An evaluation of the impact of PROGRESA cash payments on private inter-household transfers, International Food Policy Research Institute. [32] Woodruff, C., and R. Zenteno (2007): Migration networks and microenterprises in Mexico, Journal of Development Economics, 82, 2: 509-528. [33] Yang, D. (2008): International migration, remittances, and household investment: Evidence from Philippine migrants exchange rate shocks, Economic Journal, 118, 528: 591-630. notes://notes303/85257726004c7554/38d46bf5e8f08834852564b500129b2c/2d7d81b0dd8aca4f85257b9 7 Appendix 21

2000 2000 Figure 1: Share of Emigrants in Native Population across Countries by Different Education Groups in 1990 and 2000. 60% All Emigrants 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 1990 90% Emigrants with Secondary and Tertiary Education 80% 70% 60% 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 1990 22

2000 100% Emigrants with Tertiary Education 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1990 23

People Figure 2: Total Number of Emmigrants in India and Philippines in 1990 and 2000 by Major Destination Countries. 1,600,000 1,400,000 1,200,000 1,000,000 800,000 600,000 400,000 UK Germany USA Canada Australia Italy Japan 200,000 0 Philippines, 1990 Philippines, 2000 India, 1990 India, 2000 24

Table 5: Estimation Results for GDP per Worker Estimation approach Low Income Lower Middle Income Upper Middle Income All All Non-High Income Low and Lower Middle Income (1): All, IV1 0.07(2.45) 1.651*(0.66) 0.84(0.89) 2.1*(0.81) 1.67**(0.55) 1.88** (0.6) (2):(1)+ initial GDP per Worker and Human Capital (3): (2)+ Changes in Gov. Expenditures and Trade Shares in GDP (4): Secondary and Tertiary, IV1 (5):(4)+ initial GDP per Worker and Human Capital (6): (5)+ Changes in Gov. Expenditures and Trade Shares in GDP 2.5 (2.24) 2.33***(0.62) 0.91(0.96) 2.15*(0.96) 1.66**(0.6) 2.15**(0.67) 1.29(1.62) 2.35***(0.64) 0.37(0.9) 2.17*(1) 1.58**(0.59) 2.17**(0.7) -0.57(0.43) 0.9*(0.44) 0.01(0.3) 1.22(0.87) 0.49(0.26) 0.8(0.4) -0.19(0.53) 1.15*(0.43) 0.2(0.34) 1.27(0.89) 0.47(0.25) 0.81(0.43) -0.29(0.47) 1.16*(0.45) 0.01(0.32) 1.19(0.84) 0.42(0.25) 0.8(0.45) (7): Tertiary, IV1-0.07(0.19) 0.58*(0.28) 0.14(0.22) 0.27(0.18) 0.22(0.16) 0.25(0.23) (8):(7)+ initial GDP per Worker and Human Capital (9): (8)+ Changes in Gov. Expenditures and Trade Shares in GDP 0.04(0.21) 0.66*(0.31) 0.34(0.3) 0.25(0.15) 0.19(0.15) 0.21(0.22) 0.09(0.22) 0.66*(0.32) 0.2(0.29) 0.29(0.18) 0.18(0.15) 0.23(0.23) Number of Observations 30 44 42 164 117 74 (10): All, IV2 0.16(0.08) 0.17***(0.04) 0.09(0.05) 0.69***(0.07) 0.46***(0.05) 0.45***(0.06) (11): (10)+ Gov. Expenditures and Trade Shares in GDP (12): (11)+ Financial Development and Political Stabiity (13): Secondary and Tertiary, IV2 (14): (13)+ Gov. Expenditures and Trade Shares in GDP (15): (14)+ Financial Development and Political Stabiity 0.18(0.09) 0.15**(0.04) 0.11(0.07) 0.67***(0.08) 0.44***(0.05) 0.39***(0.06) 0.21(0.11) 0.25(0.2) 0.03(0.06) 0.51***(0.13) 0.37*(0.14) 0.44*(0.19) 0.02(0.1) 0.16***(0.05) 0.1(0.06) 0.25**(0.08) 0.27***(0.05) 0.31***(0.08) 0.03(0.1) 0.13**(0.04) 0.14(0.07) 0.2*(0.09) 0.22***(0.06) 0.22** (0.08) 0.33**(0.1) 0.28(0.17) 0.05(0.06) -0.06(0.14) 0.12(0.14) 0.17(0.18) (16): Tertiary, IV2-0.03(0.09) 0.15*(0.06) 0.1(0.07) -0.12(0.09) 0.14(0.07) 0.227*(0.1) (17): (16)+ Gov. Expenditures and Trade Shares in GDP -0.03(0.09) 0.11*(0.05) 0.15(0.09) -0.22*(0.1) 0.07(0.07) 0.15(0.09) Number of Observations 63 93 89 341 247 156 (18): (17)+ Financial Development and Political Stabiity 0.37**(0.1) 0.14(0.23) 0.05(0.06) -0.38**(0.13) -0.08(0.16) 0.08(0.22) Number of Observations 20 30 34 140 86 50 The dependent variable for IV 1 approach is the growth of GDP per worker and the explanatory variable is the change in labor force due to emigration. In IV 2 approach the dependent variable is a logarithm of GDP per worker and the explanatory variable is logarithm of emigrants as a share of total labor force. Each cell is the result of a separate regression. The units of observations are migrant receiving countries in 1990 and 2000. In IV 2 approach each regression includes year fixed effects. The method of estimation is Instrumental Variable approach. Instrument for IV 1 is a change in labor force due to the imputed number of emigrants. Instruments for IV 2 are dummy variables for colonial relationship and low-income countries; the average distance from destination countries, with the exception of selective countries: Australia, Canada, and the U.S.; a minimum distance from selective countries; and a country size, in terms of population including both residents and emigrants. The numbers in parentheses are heteroskedasticity robust standard errors of the coefficients and (*) indicates significance level at 10, (**) at 5, and (***) at 1 percent. All shows regression results for all emigrants, Secondary and Tertiary includes only emigrants and labor force with secondary and tertiary educations, and Tertiary includes only emigrants and labor force with tertiary education. 25