Does immigration boost per capita income?

Economics Letters 107 (2010) 177–179

Contents lists available at ScienceDirect

Economics Letters j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c o l e t

Does immigration boost per capita income? Gabriel J. Felbermayr a, Sanne Hiller b, Davide Sala b,⁎ a b

Department of Economics, University of Stuttgart-Hohenheim, 70593 Stuttgart, Germany Department of Economics, Aarhus School of Business, Aarhus University, Hermodsvej 22, 8230 Aabyhøj, Denmark

a r t i c l e

i n f o

Article history: Received 31 October 2008 Received in revised form 16 December 2009 Accepted 19 January 2010 Available online 25 January 2010

a b s t r a c t Using a cross-section of countries, we adapt Frankel and Romer's (1999) IV strategy to international labor mobility. Controlling for institutional quality, trade, and financial openness, we establish a robust and nonnegative causal effect of immigration on real per capita income. © 2010 Elsevier B.V. All rights reserved.

JEL Codes classification: F22 F12 Keywords: Gravity model International trade International migration Cross-country income regression

1. Motivation Advocates of liberal policies argue that immigration raises the income of native factor owners (Borjas, 1994). For this immigration surplus to materialize, the aggregate production function must display complementarity between inputs and the factor-content of the immigrant inflow must have a different composition than the pre-existing stock of natives.1 In contrast, per capita income — including immigrants — does not necessarily go up. If immigrants on average are poorer than natives, it trivially falls.2 Besides being of intrinsic interest, the per capita income effect determines whether the winners of immigration may compensate the losers without excluding the immigrants from the redistribution scheme.3 In the long run, if immigrants assimilate perfectly (i.e. become indistinguishable from natives) and the capital stock adjusts, per capita income reverts to the initial level. However, even in the long run, migrants may have different propensities to accumulate financial and human capital than natives, which affects per capita income. Moreover, by fostering diversity immigration increases the value of aggregate output (Ottaviano and Peri, 2006). But if it exacerbates ethnic tensions, it

⁎ Corresponding author. Tel.: + 45 894 86132; fax: + 45 894 86197. E-mail addresses: [email protected] (G.J. Felbermayr), [email protected] (S. Hiller), [email protected] (D. Sala). 1 2 3

Other conditions such as the absence of distortions must hold too. Total GDP always increases if immigrants find productive employment. See Felbermayr and Kohler (2009).

0165-1765/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2010.01.017

may have the opposite effect. To settle these ambiguities, the paper conducts an empirical assessment on a broad number of countries.4 To deal with the endogeneity of immigration to per capita income, we construct the geographical component of migration from crosscountry data on bilateral migrant stocks and use it as an instrument — akin to Frankel and Romer (1999, henceforth F&R) for trade openness. Corroborated by non-weak instruments, our analysis establishes a robust, non-negative effect of immigration on per capita income. 2. Empirical strategy 2.1. Income regression Following a well established literature, we estimate the following cross-country income regression: ln yi = α + βM ln Mi + βT Ti + βF Fi + γ′ Γi + ιIi + ui ;

ð1Þ

where i is the country index, yi is per capita GDP (constant dollar PPP, in 2000) and Mi the stock of immigrants. The vector Γi contains geographical factors as population, land surface, a continuous measure of landlockedness — all related to the size of the domestic market — and the malaria index (see Rodrik et al., 2004).5 Ii proxies institutional quality. Our focus is on βM. Since we include population (native and foreign-born), β M does not measure the pure size effect of 4

Buch and Toubal (2008) account for labor market openness in their study of German states. The index of landlockedness is the share of land borders in total border length. Compare Sachs (2003). 5

178

G.J. Felbermayr et al. / Economics Letters 107 (2010) 177–179

Table 1 Per capita income and immigration. (A) Second-stage regressions Large sample (N = 162) (1)

(2)

Small sample (N = 63) (3)

Migration only

Migration (M)

(4)

(5)

(7)

(8)

(9)

(10)

(11)

Migration, trade and finance

Migration, trade and finance

Migration, trade and finance

OLS

IV

IV

OLS

IV

OLS

IV

OLS

IV

OLS

IV

0.261a (0.0462)

0.810a (0.153)

0.535a (0.138)

0.240a (0.043) 0.006a (0.002)

0.484b (0.228) 0.019a (0.007)

Yes

Yes

Yes

Yes

Yes

0.210a (0.043) 0.004a (0.001) 0.723a (0.097) Yes

0.100 (0.133) 0.015b (0.007) 0.582a (0.159) Yes

0.231a (0.054) 0.001 (0.001) 0.426a (0.095) Yes Yes

0.226b (0.090) 0.005 (0.003) 0.415a (0.098) Yes Yes

0.343

66.200 27.860

79.100 18.030

0.387

68.090 5.506

0.542

161.700 6.351

0.774

475.700 7.066

0.232a (0.053) 0.001 (0.001) 0.431a (0.100) Yes Yes Yes 0.773

0.221b (0.088) 0.005 (0.003) 0.420a (0.101) yes Yes Yes 488.200 6.921

M for (2)

M for (3)

M for (5)

T for (5)

M for (7)

T for (7)

M (for 9)

T for (9)

M for (11)

T for (11)

−-2.881a (0.710)

− 1.763b (0.861) 0.025a (0.009)

55.620 (39.610) 1.441b (0.584)

0.531 16.490

0.549 15.940

0.375 3.996

− 2.242b (0.921) 0.042a (0.015) − 0.609b (0.243) 0.565 12.730

57.750c (32.900) 1.369c (0.811) 2.708 (11.830) 0.372 1.658

−2.923b (1.302) 0.043b (0.020) − 0.947a (0.287) 0.311 8.882

131.600a (48.61) 2.559b (1.136) − 13.390 (16.730) 0.521 3.687

− 2.937b (1.334) 0.043b (0.021) − 0.957a (0.291) 0.300 8.345

132.0a (49.43) 2.553b (1.147) − 13.110 (17.17) 0.512 3.596

Trade (T) Finance (F) Market size controls Geography controls Institutional controls Adj R2/Wald Chi MES

(6)

Migration and trade

(B) First-stage regressions Variable instrumented M̂ ~ M

a

1.500 (0.279)

T̂ Finance (F) Adj R2 F-stat. on Excl. Instr.

0.556 28.820

Robust standard errors in brackets. a, b, c denote significance at the 1, 5, 10% levels, respectively. All regressions include a constant (not shown). Critical values for minimum eigenvalue statistics (MES) are 7.03 (5), 4.58(10), 3.95(20), and 3.63(30). Market size controls: ln area, ln population 1960, landlockedness. Geography controls: malaria index. Institutional controls: constitutional review. All first-stage regressions include all exogeneous regressors (not shown).

immigration, but rather its compositional effect, namely the one of an increase in the immigrant share. A country's attitude towards migration is likely shaped by its history and culture. Unable to use fixed effects in a cross-section, we account for unobserved heterogeneity adding to the analysis Ti — the ratio of exports and imports over GDP — and Fi — an indicator of financial market integration. We presume that a liberal attitude towards labor flows comes along with a high degree of integration on goods and financial markets. 2.2. Instrumental variables Mi and Ti are potentially endogenous. Following F&R, we exploit geographical variation to instrument Ti. We extend this strategy to Mi, using a 226 × 226 matrix of international bilateral migrant stocks for the year 2000. Migrants are defined as “foreign-born”, so that our measure is unaffected by national naturalization policies. Let Mij denote the number of individuals born in country j and residing in i, and let J be the set of all sending countries. Then, Mi = ∑j ∈ JMij is the total immigrant stock in the receiving country i. Let
 h
 i E Xij jGij ; ADJij = exp Gij γX + ADJij λX + Gij ADJij δX

ð2Þ

with X ∈ {M, T}. Gij is a vector of geographical variables (e.g. landlockedness, the logarithm of the population in 1960, of the area of country i and of country j and of the bilateral distance between country i and country j). ADJij is an adjacency dummy.6 We use 6 The specification is borrowed from F&R. Lewer and van den Berg (2008) show that bilateral migration flows are accurately predicted by geographical variables (along with income and population).

Poisson Pseudo Maximum Likelihood (PPML) to estimate (2) for the year 2000 (see Santos Silva and Tenreyro, 2006). Our instruments are computed as Xî = ∑jXiĵ , Xiĵ = Gij γ̂X + ADJij λ̂X + (GijADJij)δ ̂X is the insample linear prediction of Eq. (2).7 Three remarks are in order. First, PPML estimation yields stronger instruments compared to OLS since it accounts for cases with Xij = 0, representing a non-negligible share of country-pairs (6(3)% of Mij(Tij) in our sample). Second, Xî is a generated measure of multilateral remoteness determined exclusively by geography. Hence, no concerns about consistency of (2) as a gravity equation arise. Third, M̂ i and Tî prove to be collinear in our regression analysis and become weak instruments when used simultaneously. However, exploiting that a country's remoteness affects the economic and the psychological costs of migration, we compute – from the bilateral distance between two countries i and j (D ij) — geographical 1 ∑s∈Jnfjg Dis and use M̃ i = ∑j ∈ JGRij as an remoteness as GRij = N−1 alternative instrument for Mi. Fi and Ii may well be endogenous, but simultaneous instrumentation of many endogenous regressors exacerbates the concern for weak instruments. Since good bilateral data for financial flows are rare, we abstain from instrumenting Fi like Xi, but rather proxy Fi with the logarithm of the distance to the closest major financial center (Rose and Spiegel, 2008). Ii is constitutional review (La Porta et al., 2004). Compared to alternative common measures (e.g. expropriation risk or government effectiveness), it reflects permanent constraints on the executive authority, rather than election outcomes or temporary policies (Glaeser et al., 2004). By definition, endogeneity concerns are attenuated. 7 Using only in-sample predictions increases precision, compare Noguer and Siscart (2005).

G.J. Felbermayr et al. / Economics Letters 107 (2010) 177–179

3. Results The results of our analysis presented in Table 1 reveal a positive and significant elasticity of y with respect to the migrant stock. Most importantly, weakness of instruments is of no concern (see the minimum Eigenvalue statistics (MES) — Stock and Motohiro, 2002). We proceed commenting each column of Table 1. 3.1. Effect of immigration Columns (2) and (3) provide IV estimates based on M̂ and M̃ , respectively. The different performance in (2) relative to (3) is ascribed to the correlation of 0.52 between M̂ and T (and, therefore with the residual), whereas M̂ is free of this problem. Overall, OLS underestimates βM and several factors weigh upon this discrepancy: measurement error in the migrant stock causes attenuation bias in the OLS coefficient, while unobserved heterogeneity (T, F) causes overestimation of the OLS coefficient. Moreover, the direction of the simultaneity bias (reverse causality of M and y) is ambiguous. 3.2. Trade openness In (5), we add trade openness. To avoid multicollinearity between T̂ and M̂ , we instrument M with M̃ and T with T̂.8 Multicollinear instruments would hinder disentangling the growth effect of labor assimilation from the one of goods market integration. 3.3. Financial openness Columns (6) and (7) additionally control for financial openness. The IV estimates reveal that only βT is positive and significant while βM, although positive, is insignificant. Replacing F with the great circle distance gives similar results, likely the consequence of the high correlation among the two variables (ρ = 0.6059).

179

income gain of 2.2%. We also find that trade and financial integration positively affect per capita income. Acknowledgments We thank Wilhelm Kohler and the participants at the Workshop “Migration and Labor Market Integration” at Tübingen University for comments. We are grateful to the Development Research Center on Migration, Globalization and Poverty, at the University of Sussex, and to Francesco Trebbi for providing us with, respectively, the migration data and the malaria index. The three authors have mostly developed this research project while employed at the Chair of International Economics at Tübingen University. The project was supported by grant no. 10.06.1.111 of the Thyssen Foundation. All remaining errors are ours. A replication archive is available on www.asb.dk/staff/dsala. Appendix A. Data sources – Population, GDP per capita (2000 PPP USD) (yi): World Development Indicators 2007. – Bilateral migration stocks (M ij): World Bank, Development Research Center on Migration, Globalization and Poverty.9 – Bilateral trade data (Tij = (exportij + importij) / GDPi): Direction of Trade Statistics, IMF, Sept. 2006 CD-ROM. – Trade openness (Ti): Penn World Tables 6.2. – Adjacency (ADJij), Area, bilateral distance: CEPII, Paris.10 – Financial openness (Fi): negative of Rose and Spiegel (2008). – Constitutional review (Ii): La Porta et al. (2004). – Landlockedness: CIA World Factbook 2008. – Malaria index: Rodrik et al. (2004). References

Holding in due consideration the sheer population size effect and using geography-based instruments, we find robust evidence for immigration to be positively and causally related to per capita income. Hence, immigration gives rise to a gain that can — in principle — be used to make the native population better off without excluding the immigrants from the redistribution scheme. Our preferred specification — columns (10) and (11) in Table 1 — implies that a 10% increase in the migrant stock leads to a per capita

Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., Wacziarg, R., 2003. Fractionalization. Journal of Economic Growth 8 (2), 155–194. Borjas, G.J., 1994. The economics of immigration. Journal of Economic Literature 32 (4), 1667–1717. Buch, C.M., Toubal, F., 2008. Openness and growth: the long shadow of the Berlin Wall. Journal of Macroeconomics 31 (3), 409–422. Felbermayr, G., Kohler, W., 2009. Can International Migration Ever Be Made a Pareto Improvement? In: Novotny, E., Mooslechner, P., Ritzberger-Grünwald, D. (Eds.), The Integration of European Labour Markets. Edgar Elgar, Cheltenham, UK, Northampton, MA, pp. 32–50. Frankel, J.A., Romer, D., 1999. Does trade cause growth? American Economic Review 89 (3), 379–399. Glaeser, E.L., La Porta, R.F.L., Shleifer, A., 2004. Do institutions cause growth? Journal of Economic Growth 9 (4), 271–303. La Porta, R.F.L., Lopez-de-Silanes, F., Pop-Eleches, C., Shleifer, A., 2004. Judicial checks and balances. Journal of Political Economy 112 (2), 445–470. Lewer, J.J., Van den Berg, H., 2008. A gravity model of immigration. Economics Letters 99 (1), 164–167. Noguer, M., Siscart, M., 2005. Trade raises income: a precise and robust result. Journal of International Economics 65 (2), 447–460. Ottaviano, G., Peri, G., 2006. Rethinking the effects of immigration on wages. NBER Working Paper 12496. Rodrik, D., Subramanian, A., Trebbi, F., 2004. Institutions rule: the primacy of institutions over geography and integration in economic development. Journal of Economic Growth 9 (2), 271–293. Rose, A.K., Spiegel, M.M., 2008. International Financial Remotness and Macroeconomic Volatility. Mimeo University of California Berkeley. Sachs, J.D., 2003. Institutions don't rule: direct effects of geography on per capita income. NBER Working Paper 9490. Santos Silva, J.M.C., Tenreyro, S., 2006. The log of gravity. Review of Economics and Statistics 88 (4), 641–658. Stock, J., Motohiro, Y., 2002. Testing for weak instruments in linear IV regression. NBER Technical Working Paper 0284.

8 An F-test on the first stage (regressing T on T̂ and M̂ ) reveals that both instruments are jointly but not individually significant. In this case, weakness of instruments cannot be rejected. Multicollinearity persists even if (2) is not identically specified for Mij and Tij.

9 http://www.migrationdrc.org/research/typesofmigration/global_migrant_origin_database.html. 10 www.cepii.fr/anglaisgraph/bdd/distances.htm.

3.4. Geography and institutions Columns (1) to (7) draw on the largest possible sample, while columns (8) to (11) use a smaller sample for which institutional quality data is available (63 countries). Specifications (8) and (9) control for a direct effect of geography — see Sachs (2003) — adding a malaria index to the list of regressors in columns (6) and (7). We include institutional controls in columns (10) and (11) (coefficients not displayed) and all coefficients are nearly unaltered. The IV and OLS estimates become considerably close. Finally, the inclusion of ethnic fragmentation (Alesina et al., 2003) leaves estimates qualitatively unchanged (not reported). 4. Summary