MIGRATION AND FDI: COMPLEMENTS OR SUBSTITUTES? Maurice KUGLER* and Hillel RAPOPORT** *Department of Economics, Southampton University, and CID, Harvard University ** Department of Economics, Bar-Ilan University, CADRE, University of Lille II, and CReAM, University College London Paper prepared for presentation at the ESF-CEPR conference on Outsourcing and migration in a European context, Rome, September 15-17. 1
1. INTRODUCTION Globalization indicators for the 19s: World Trade/GDP: 1.5, World FDI/GDP: 3 OECD immigration stock: 1.5, higher for the skilled ( 1.7) than for the unskilled ( 1.3), due to push-pull factors, and to selective policies. General questions: Are international trade and factor flows complements or substitutes? Are migration and FDI complements or substitutes, and how is this relationship affected by the skill composition of migration? For both, standard trade models v. diaspora externality arguments. 2
Literature on trade and factor flows Standard trade models: substitutes (factor price equalization) Recent empirical studies: complements (diaspora effects, networks) - immigration and bilateral trade: Gould (1994) for the U.S., Head and Ries (1998) for Canada. Interpretation: demand for ethnic goods + networks; interregional trade and business/social networks: Combes et al. (2005) for France - ethnic networks and international trade (Rauch and Trindade, 2002, Rauch and Casella, 2003): information problems with trade in differentiated products. - cross-country comparisons: Lopez and Schiff (1998), trade liberalization episodes induce more unskilled and less skilled emigration. 3
Literature on migration and FDI Standard trade models: substitutes (factor price equalization) Sociological literature on diasporas: migrants can favor FDI by - revealing information on the characteristics of their home-country, notably the quality of its workforce (would seem to apply mainly to unskilled migrants) - taking part in business networks, thus encouraging joint ventures, technology transfers, etc. (would seem to apply mainly to skilled migrants) Empirical literature includes mainly regional (e.g., Aroca and Maloney, 2005, find substitution for US-Mexico border states) and sectoral case-studies (e.g., on the software industry in India, Ireland or Israel). 4
Only a handful of studies based on macro data: Docquier and Lodigiani (2006): cross-country comparisons using emigration by skill level as a determinant of total FDI; positive effect, stronger for democratic countries and intermediate corruption levels Kugler and Rapoport (2006): US bilateral data; dynamic complementarity between skilled migration and FDI in services, and contemporaneous substitutability between unskilled migration and manufacturing FDI) 5
What do we do in this paper? Section 2: Theoretical framework. We supplement a standard two-country model of factor flows with endogenous skills and transaction costs to derive testable predictions on the migration-fdi relationship Section 3: Data. Section 4: Empirical methodology. We rely mainly on Heckman estimation techniques to account for selection bias Section 4: Results. We estimate a selection equation and a flow equation, with FDI on the LHS and migration variables on the RHS 6
1 Theoretical framework Small open developing economy, Cobb-Douglas technology with constant returns to scale: Y t = A(H t )Kt 1 L, with: L t = N t H t, the stock of labor measured in e ciency units H t =1+P t (h 1)H t, the average number of such units/level of human capital per worker, where h>1 is the skill premium, P is the proportion of skilled, skilled and unskilled workers are perfect substitutes and total factor productivity depends on human capital externalities. Assuming competitive markets and denoting by k the capital to labor ratio, factor returns are given by: r t =(1 )A(H t )k t ;w t = A(H t )k 1 1
² Capital is perfectly mobile internationally: r t = r + ¼ t,where¼ captures the extent of transaction costs as well as a variety of risks associated to internal institutional factors; ² Due to a persistent technology gap, the wage rate (per e cient unit of labor) is higher in a large, more advanced economy, but labor is imperfectly mobile internationally due to migration costs and restrictive immigration policies Therefore: h (1 )A(Ht ) i 1 k t = r +¼ t k(¼ t ;H t ) h w t = [A(H t )] 1 w(¼ t ;H t )with the derivatives k 0 1 < 0, k 0 2 > 0 and i1 1 r +¼ t w 0 1 < 0; w 0 2 > 0. 2
The equilibrium stock of capital in the economy is thus given by: K t = k t L t = A t = A(H t ); h (1 )At L t r +¼ t i 1 ; with: L t =(N t Mt U Mt S )H t ; where Mt U and Mt S denote the number of unskilled and skilled migrants, H t = N t(1 P t ) Mt U +N tp t h Mt Sh N t Mt U MS t out, is the average level of human capital after migration is netted P t = P fh; E t 1 (m S t ) E t 1 (m U t )g; where E t 1 (m i t); i= S; U; is the migration rate by skill level for a given period as expected when the education decision is made, and ¼ t = ¼fX t ;Mt 1;M U t 1g S where X t is a vector of country characteristics with ¼ 0 S ;¼0 U < 0 capture the diaspora e ects 3
Noting that: @H t @M S t @H t @M U t = h 1 (N t M U t MS t )2 [N t (1 P t ) M U t ] < 0 = h 1 (N t M U t MS t )2 [N t P t M S t ] > 0 @H t N t(h 1) @P t = > 0 (N t Mt U MS t )2 and using ln K t = 1 ln(1 )+ 1 ln A t +ln(n t Mt U Mt S )+lnh t 1 ln(r + ¼ t ); we can describe the relationship between migration and FDI and derive testable implications. Since the skill composition of the labor force is endogenous (i.e., depends on migration prospects), the relationship between migration and FDI will depend on the way people form their expectations about future migration possibilities. 4
1.1 Myopic expectations With expectations based on emigration rates observed among the previous generation, we have: E t 1 (m S t )= MS t 1 N t 1 P t 1 and E t 1 (m U t )= Mt 1 U N t 1 (1 P t 1 ; which gives: ) @ ln K t @M S t =[ 1 A 0 A t + 1 H t ] @H t @M S t 1 N t M U t MS t < 0 @ ln K t @M S t 1 =[ 1 A 0 A t + 1 H t ] @H t @P t @P t 1 1 @¼ t @Mt 1 S r +¼ t @Mt 1 S > 0 @ ln K t @M U t =[ 1 A 0 A t + 1 H t ] @H t @M U t 1 N t M U t MS t 7 0 @ ln K t @M U t 1 =[ 1 A 0 A t + 1 H t ] @H t @P t @P t 1 1 @¼ t @Mt 1 U r +¼ t @Mt 1 U 7 0 5
Predictions: ² For skilled migration: dynamic complementarity and contemporaneous substitutability ² For unskilled migration: unclear Interpretation in terms of: ² N effect (through the size of the labor force) ² H effect (through the average level of human capital) ² incentive e ect (past skilled migration encourages current education investment ² and a network or ¼ e ect 6
1.2 Rational expectations With rational expectations, we have E t 1 (m S t )= MS t N t P t and E t 1 (m U t )= The proportion of educated is now given by the following implicit function: MU t N t (1 P t ) : G(P t ;Mt S ;Mt U ) P t P [h; m] =0,wherem =( MS t N t P t MU t N t (1 P t )) is the di erential in migration probability between skilled and unskilled workers and Pm 0 > 0: This gives: @P t @M S t @P t @M U t P 0 m=n t P t ) = @G=@MS t @G=@P t = > 0 1 Pm 0 [ ms t =P t +m U t =1 P t] P 0 m=n t (1 P t ) = @G=@MU t @G=@P t = < 0 1 Pm[ m 0 S t =P t +m U t =1 P t] Note that @P t > 0 and @P @Mt S t < 0 since P @M m[ 0 mu t U t 1 P t ms t P t ] < 1 given that m S >m U and the fact that it is reasonable to assume that in developing countries the proportion of educated is lower than one half (P t < 1 P t ): 7
With rational expectations we therefore obtain the following predictions: @ ln K t @M S t =[ 1 A 0 A t + 1 H t ]( @H t @M S t + @H t @P t @P t @Mt S ) 1 N t M U t MS t 7 0 @ ln K t @M S t 1 = 1 1 @¼ t r +¼ t @Mt 1 S > 0 @ ln K t @M U t =[ 1 A 0 A t + 1 H t ]( @H t @M U t + @H t @P t @P t @Mt U ) 1 N t M U t MS t 7 0 @ ln K t @M U t 1 = 1 1 @¼ t r +¼ t @Mt 1 U > 0 Predictions: dynamic complementarity and unclear contemporaneous relationship (for both skilled and unskilled) 8
SECTION 3: DATA We use two main data sources: OECD data on FDI between 19 and 1998 (23 source and 45 host-countries); Immigration data to OECD countries (Docquier and Marfouk, 2005), by origin country and education level (tertiary, secondary and tertiary), 19 and 2000 Main controls: FDI-host countries: GDP, GDP per capita, ethnic fractionalization, and standard indices of economic, financial and institutional development. FDI-source country: GDP per capita. 7
Bilateral variables: imports and exports in 19 and 2000, FDI in 19, gaps in human capital levels, linguistic, geographic and longitudinal distance Migration variables are interacted with their share of the immigration stock/flow of migrants from county i to country j. We expect the coefficient of this interaction term to take the opposite sign with respect to the relevant migration variable (elite network and revelation effect) Fixed effects host and source countries. 8
Section 4. Empirical methodology Gravity equation specification, with adjustments for selection bias. Justification: gravitational forces can be very small, but not zero, whereas trade or FDI flows between pairs of countries is quite often zero. OLS ignores this (assumes measurement errors or true zeros) while Tobit techniques recognizes the truncation but assumes it is exogenous Good theoretical reasons to believe selection is endogenous (networks, information and transaction costs, as in this paper, or presence of set-up costs in FDI deployment, as in Razin et al. (2004), so we use Heckman techniques 9
5. RESULTS We present the results for the whole sample and two sub-samples: North-South, and intra-eu15, and two alternative classifications for skill levels: primarysecondary-tertiary, and skilled-unskilled. For each specification, we estimate: A flow equation where we regress the (log of the) change in FDI 19-2000 on migration stocks and flows and interactions with their respective shares A selection equation where we regress the probability of having FDI taking place between i and j on the same variables except for the flow of migrants. Focusing on robust results only (i.e., where the classification used does not affect the sign and significance of the effects), the main results are: 10
e x p i, j, t 1 Table 2: Heckman Estimation - Primary/Secondary/Tertiary Classification Dependent Variable: FDI (prices adjusted - in logs) Whole Sample North-South EU15 sample Flow Equation Selection Equation Flow Equation Selection Equation Flow Equation Selection Equation 0.363*** -0.758*** -0.404*** M PRIM - - (0.113 ) (0.218) (0.145) 1.068** -1.972* 1.878*** M SEC - - (0.451) (1.178) (0.473) M TER -0.982*** 0.399*** -0.173 - - (0.377) (0.863) (0.427) -0.167 1.767* -1.897** -1.279-1.272*** 0.0001** M PRIM (0.434) (1.072) (0.962) (1.110) (0.507) (0.0001) 0.682* -2.743*** -0.954*** -2.597*** 1.314*** 0.226 M SEC (0.402) (1.119) (1.131) (0.933) (0.453) (0.417) 0.721*** 3.957*** 3.040*** 5.870*** 0.5*** 0.939*** M TER (0.464) (1.053) (0.999) (1.234) (0.527) (0.133) -0.143-1.0003*** 0.739 1.239** 1.005*** -2.344** PRIM *(M PRIM/TOT) (0.219) (0.551) (0.527) (0.556) (0.062) (0.924) -0.143-0.056 1.103* -0.139-0.659*** -0.853 SEC *(M SEC/TOT) (0.179) (0.503) (0.566) (0.453) (0.217) (1.200) -0.662*** -2.124*** -1.465*** -3.098*** -0.629** -0.477 TER *(M TER/TOT) (0.239) (0.551) (0.536) (0.654) (0.282) (1.033) No of Observations 7249 2741 4401 Definition: (M e/tot) is the ratio of M e to total migrants in 19 where e= primary, secondary and tertiary. *) significant at the 10%, **) significant at the 5%, ***) significant at the 1% 11
e x p i, j, t 1 Table 3: Heckman Estimation - Skilled/Unskilled Classification Dependent Variable: FDI (prices adjusted - in logs) Flow Equation -0.0*** M unskilled (0.250) 0.677* M skilled (0.370) M unskilled 2.442*** (0.586) 2.457*** M skilled (0.635) -2.731*** M unskilled *(M unskilled/tot) (0.409) -1.793*** M skilled *(M skilled/tot) (0.358) -0.0002*** host GDP (0.0001) -0.1002 source GDP (0.126) -0.106 host GDP/Cap (0.252) 1.222*** source GDP/Cap (0.356) -0.008 Financial (0.007) 0.001 Economic (0.008) 0.105 Corruption (0.106) Political Whole Sample North-South Sample EU15 sample Selection Flow Selection Flow Equation Equation Equation Equation 0.004 (0.012) - - 3.400*** (0.883) 1.286*** (0.915) -3.496*** (0.661) -1.027* (0.516) 0.0003*** (0.0001) 0.098 (0.286) 0.880 (0.064) -3.715*** (0.536) -0.051*** (0.011) 0.040*** (0.010) 0.058*** (0.003) -0.322*** (0.028) 2.823*** (0.367) -6.406*** (0.815) 1.080*** (0.336) 2.827** (1.320) -2.968*** (0.739) -1.540*** (0.786) 0.001*** (0.002) 0.796*** (0.045) 0.655*** (0.068) 0.134 (0.166) -0.011 (0.013) -0.058*** (0.012) 0.548** (0.256) -0.071*** (0.024) - - 2.439** (0.967) 1.000 (0.971) -2.329*** (0.679) -1.270*** (0.547) 0.334*** (0.058) 0.001*** (0.0002) -0.1 (0.123) 0.423*** (0.037) 0.041*** (0.011) -0.036*** (0.011) 0.024*** (0.001) -0.230*** (0.021) -2.408*** (0.326) 2.682*** (0.433) 1.646*** (0.653) 3.129*** (0.855) -1.820*** (0.535) -2.340*** (0.491) -0.0001 (0.286) -0.083 (0.142) 0.847*** (0.286) -0.128 (0.395) -0.0168** (0.007) -0.002 (0.010) 0.031 (0.113) 0.028** (0.012) Selection Equation - - 9.079*** (1.583) 7.442*** (2.085) -10.676*** (1.741) -6.035*** (1.334) 70.261*** (9.570) -15.847*** (3.718) 0.0003*** (0.001) 0.040 (0.434) 1.235*** (0.102) -0.797 (0.956) 0.078*** (0.016) -0.102*** (0.018) 12
1. Dynamic relationships There is strong evidence of dynamic complementarity between skilled migration and FDI. Coefficient positive and very significant in all specifications, with an elasticity around 3 for North-South and for the whole sample and the EU15 with the 2-level classification (lower otherwise) In 5 out of 6 specifications, past skilled migration significantly increases (1%) the likelihood of having FDI in the subsequent period; insignificant for N-S In contrast, no robust evidence for unskilled migration and FDI, but past unskilled migration significantly increases the occurrence of FDI in all specifications but for the N-S sub-sample with the 3-level classification. 13
2. Contemporaneous relationships There is strong evidence of contemporaneous substitutability between unskilled migration and FDI in the EU15 sub-sample, the estimated elasticity being higher with the skilled/unskilled classification (-2.4 against -.4). For the other sub-samples the results for unskilled migration depend on the skill classification adopted. Likewise, no robust results between current skilled migration and FDI 14
3. Interaction terms The interaction terms always take a sign opposite to that of the corresponding migration variable. In particular, the coefficient on the interaction term between the stock of skilled migrants and its share in total stock is negative and highly significant in all specifications, both in the flow equation and in the selection equation. Interpretation: the effect of past skilled migration on FDI is magnified when skilled emigrants represent only a fraction of total migration, providing supportive evidence of an elite network effect. 15
SUMMARY OF RESULTS Skilled migration Unskilled migration Past Current Past Current THEORY Myopic >0 <0?? Rational >0? >0? RESULTS Whole Sample 3-level classification >0++ <0 -+ >0 2-level classification >0++ - >0++ <0 North-South 3-level classification >0++ >0 <0 <0 2-level classification >0 <0 >0++ >0 EU15 3-level classification >0++ - <0+ <0 2-level classification >0++ >0 >0++ <0 16