1

SKILLED MIGRATION, FDI AND HUMAN CAPITAL INVESTMENT*

Daniele Checchi†

(Università di Milano and IZA)

Gianfranco De Simone

(Università di Milano and Centro Studi Luca d’Agliano)

Riccardo Faini‡

(Università di Roma “Tor Vergata”, CEPR and IZA)

current version: May 2008

Abstract

In a globalised world, where factors of production are increasingly mobile, the process of

domestic accumulation of human capital (HC) can be affected in several ways through

migration and capital inflows. In addition, endowment of skilled labour and foreign direct

investments (FDI) may reinforce each other through possible “complementary effects”.

Our paper aims to advance the existing empirical literature on the relationship between

international factor mobility and domestic accumulation of HC in developing countries.

We provide new evidence on how the presence of foreign firms in the domestic economy

and the emigration of skilled workers impact the domestic school enrolment. We also

investigate whether existing supply of skilled labour is an important determinant of

inward flows of foreign capital, finding no significant effects. The interdependence

between factor mobility and HC accumulation supports some simple back-of-the-envelop

calculations that tend to exclude the presence of a significant ongoing virtuous circle

between HC accumulation and FDI inflows. Less developed countries tend to gain

domestic human capital under foreign capital inflow (job opportunities for skilled

individuals), but they loose domestic human capital through the emigration of highly

educated people both directly and indirectly (“disincentive effect” on further investment

in higher education).

Keywords: Human Capital Investment, Factor Mobility, FDI, Brain Drain/Gain,

Complementarity Effects, Developing Countries

JEL Classification: F22, F23, O15

* This research has benefited from comments received at the Final Conference on “Trade, Industrialization and

Development” (Paris, October 2006), CNR International Economics Working-Group Conference (Torino, February

2007), 9th ETSG Conference (Athens, September 2007), Sus.Div. General Meeting (Athens, September 2007),

“Transnationality of Migrants” Meeting (Riga, October 2007). We would like to thank Giorgio Barba Navaretti,

Massimiliano Bratti, Jaime de Melo and Andrea Gavosto for useful remarks and suggestions. This paper is part of the

project “People and Firms” funded by Fondazione CRT (Torino) and managed by Centro Studi Luca d’Agliano

(Milano). † Corresponding author. E-mail address: daniele.checchi@unimi.it. ‡ Riccardo Faini passed away while we were working at this paper. The paper significantly suffers from his

disappearance.

2

1. Introduction

It is commonly believed that accumulation of human capital (HC) and availability of physical

and financial capitals are among the major determinants of economic growth; it is also widely

accepted that the lack of these resources (along with the inability to expand them) are potential

reasons behind the delay of many poor countries in achieving development.

In a globalised world, where factors of production are increasingly mobile, the process of

domestic accumulation of HC might be affected in several ways. In fact, while in principle the

availability of foreign capital in the form of inward foreign direct investments (FDI) and an elastic

supply of skilled (educated) workers may individually enhance growth prospects, they can also

reinforce each other through possible “complementary effects”. The presence of foreign investors in

the home economy can provide incentives to invest in education for both people and governments:

people may want to attain higher level of education in order to access better job opportunities

offered by foreign firms, and governments may want to support the accumulation of HC in order to

benefit from possible spillovers of FDI (technology and knowledge transfer). In addition, a good

HC endowment makes the investment climate more attractive for foreign investors, offering an

educated workforce which is also likely to be associated to socio-political stability.

Ideally, a virtuous circle of HC and FDI can be attained whenever «host countries experience

continuous inflow of FDI over time by increasingly attracting higher value-added MNEs, while at

the same time upgrading the skill contents of pre-existing MNEs and domestic enterprises»

[Miyamoto (2003), p.9]. Symmetrically, a Pareto inferior equilibrium is also possible: inadequate

supply of skills discourages FDI and the lack of FDI depresses the demand for skills.

But factor mobility does not concern financial and physical capitals only. Domestic workforce is

also mobile, and when international migration is considered, the domestic accumulation of HC

needs further qualification. Even if migration flows have grown less than trade and FDI flows over

the last decades [see Sapir (2000), Faini (2006)], the ongoing “brain drain”, enhanced by selective

immigration policies1 in developed economies, is one of the suspects among the forces negatively

affecting the economic performance of developing countries. According to an established view,

skilled migration causes the flee of the most talented and entrepreneurial individuals from the

countries of origin, and severely hampers its growth prospects. Thus the outflow of educated

workers is expected to negatively impact onto the domestic stock of cumulated HC.

1 In response to the growing shortage of skilled workers, most receiving countries have tried to shift the focus of their

immigration policy, favouring the recruitment of highly skilled workers. This new twist in the policy stance toward

immigration has become a source of considerable concern in traditionally sending countries, which fear the loss of their

most skilled and entrepreneurial workers.

3

In sharp contrast with this expectation, a recent but rapidly developing literature emphasizes a

possible positive effect of skilled migration on the origin country. The brain drain becomes, in this

view, a “brain gain”. Among others, three different channels can be distinguished for a beneficial

brain drain to operate: a) skilled migrants raise economic welfare at home thanks to a relatively

large flow of remittances2; b) selective immigration policies in host countries may raise the

attractiveness of migration for high skilled individuals, which in turn raises private returns to

education (due to reduced supply) and induces additional investment in education at home; c)

skilled migration may favour growth-enhancing technology transfer, trade and foreign direct

investments between the source and the host country (network effects).

Points b) and c) provide further qualifications about possible complementarities between HC

and FDI in the wake of international migration. Mountford (1997) was the first to suggest the

possibility that migration prospects create incentives to invest more in education: since not all of

those who invest in education can (or will choose to) migrate, the post-migration level of human

capital can increase. Similar results were found by Stark et al. (1998). Stark et al. (1997) add to this

literature by showing that the possibility of a brain gain might stem from the imperfect information

of destination country’s employers on the skills of the migrants and the impact of return migration.

The wage adjustment taking place once the true ability of immigrants is revealed to foreign

employers may induce a subset of individuals to return home. Under certain conditions the post-

return average level of human capital is higher than that of a closed economy. The literature on this

issue is rapidly growing, but the empirical evidence is mixed. In a cross-country regression with 50

developing countries, Beine et al. (2001), using data from Carrington and Detragiache (1998), find a

positive effect of skilled migration on human capital investment in the source country and a positive

relation between growth and the proportion of highly educated individuals at home. Applying a

different empirical approach to the same dataset, Faini (2002) found that the rate of migration

among educated individuals was weakly and negatively correlated with tertiary enrolment at home.

Using a new dataset on migration stocks and rates by country of origin and educational attainment,

developed by Docquier and Marfouk (2005), Mariani (2004) estimates a cross-country growth

regression on a large number of developing countries and finds that the relation between brain drain

and growth is non linear and high skilled migration affects positively the growth rate only if a large

proportion of individuals at home is enrolled in (or have completed) at most the secondary school;

according to the author, this result indicates that larger countries are more likely to enjoy positive

2 The underlining argument proceeds as follows: skilled migrants typically earn relatively more and, ceteris paribus,

will therefore save more and remit to relatives remaining inland. However, skilled migrants are also likely to spend a

longer span of time abroad and also are more likely to reunite with their close family in the host country. Both factors

should be associated with a relatively smaller rather than larger flow of remittances from skilled migrants. Faini (2006)

provides evidence supporting this counter-argument.

4

feedbacks from high skilled migrations. Thus, if the focus is on the accumulation of human capital,

the role of skilled migration cannot be neglected and it still represents an unsettled empirical issue.

For what concerns implications of point c), namely technology transfer through networking, it is

worth noticing that since developing countries typically lack resources to develop new technologies

on their own, what matters for growth is their ability to appropriate and adopt advanced

technologies developed elsewhere. The literature on technology diffusion/transfer has focused on

trade and foreign direct investments as the two main channels in this respect, provided that the host

country is endowed with a sufficient level of competences to make this absorption viable. Migrants

may personally be involved in trading and investing in their home country, thus boosting trade and

foreign capital inflows, thanks to their inside knowledge or their social ties. Network effects with

people still living in their country of origin can also be exploited by their foreign employers to enter

their home market (Lucas, 2004).

Our paper aims to address empirically some of these open questions guided by theoretical

considerations at the basis of a simple conceptual framework (see Appendix B). In Section 2 we

provide further evidence on the relationships between international factor mobility (FDI and

migration) and domestic accumulation of HC in developing countries. In Section 3 we explore

potential complementarities between FDI and HC by investigating whether existing supply of

skilled labour is a significant determinant of inward flows of foreign capital. The interdependence

between factor mobility and HC accumulation supports some back-of-the-envelop calculations on

the impact of migration on domestic HC accumulation in Section 4. Section 5 concludes.

2. Do migration and inward FDI impact enrolments?

We start by focussing on the consequences of factor mobility onto educational choices in

developing countries. A simple equation (which corresponds to equation (5) in Appendix B) relates

enrolment rate j

ite in educational level j ( j = secondary, tertiary) in country i and year t to the

presence of foreign firms (proxied by the cumulated stock of FDI ) in the domestic economy and to

migration trends of educated workers ( MIG )

ittitiit

T

ik

kti

j

i

j

it CMIGFDIe ε+τ+⋅δ+⋅β+

⋅β+µ= ∑

=− 2,1 log (1)

where itC is a set of country specific factors affecting educational choices (control variables), jiµ is

a country fixed effect, tτ is a specific time effect and itε is an error term.

5

On the basis of our theoretical considerations (see Appendix B), one would expect the

presence of foreign firms providing incentives to enrol in higher education programs ( 01 >β ) . As

far as the migration of skilled workers is concerned, a negative impact on domestic enrolment

( 02 <β ) can be taken as evidence of the discouraging effects of “brain drain”, whereas a positive

impact ( 02 >β ) can be taken as evidence of “brain gain”. Relevant control variables for this

specification are related to the stage of development of the economy (presence of liquidity

constraint / endemic poverty), to the quality of the educational system and to other supply side

factors3.

2.1 Dataset and variables definition

Our dependent variables are extracted by data on educational enrolment on quinquennial base

collected by Barro and Lee (2000) integrated by data on emigration rates by educational level

collected in Doquier and Marfouk (2005). The intersection of these two datasets containing non

missing information in at least one of the two points in time (1990 and 2000) is non empty for 147

developing countries. When we consider a balanced panel version of this sample of countries, their

number reduces significantly.

We expand this dataset with information on the existing stock of foreign direct investment

(referred to the two relevant years or in their proximity) from UNCTAD, with quality of the

education indicators from World Bank4, and additional control variables

5 related to the level of

country development. We have considered alternative measures of education quality, including

public spending on education as percentage of GDP, the pupil-teacher ratio at primary school (the

corresponding measure for secondary school is only available for recent years) and the repetition

rate at primary school, but the only variable showing statistical significance in some regression is

the pupil/teacher ratio at primary level, that we have retained in the text. As a complement to this

indicator we employ the population density in order to capture per-capita availability of school

resources.6 In previous version of the paper, we have also considered the inclusion of urban/rural

3 Theoretical considerations would suggest to add a variable accounting for remittances among regressors. This would

capture a potential poverty relief feedback effect of migration. But an improvement in financial conditions of a family

could enhance both investment in education and further migration (i.e. migration cost becoming affordable, family

reunion, etc.). Hence, the impact of such a variable is not univocal from a conceptual point of view. Unfortunately, data

coverage of series on international remittances is not complete for many countries in our sample. Furthermore, since not

all financial flows related to migration follow official channels, data on remittances are not entirely reliable and they

could represent an inappropriate proxy for what is called “diaspora externality”. For these reasons we decided to

exclude remittances from our regressions. 4 EdStat on-line service provided by the World Bank.

5 Control variables are drawn from the World Bank’s World Development Indicators database. 6 In principle one would expect highly concentrated population decreasing the cost of providing schooling services

(positive correlation with enrolments), unless the country is characterized by saturation effects and/or lack of school

resources leading to possible supply-side constraints to education (negative correlation with enrolments).

6

population shares (to further account for the supply side of educational resources), the fuel exports

share in GDP (to proxy possible disincentive effects on educational investment) as well as measures

of income inequality (to capture potential liquidity constraints in educational choices), but given the

weak statistical significance and the sample reduction forced by these regressors, at the end we have

decided to leave all of them out of the present analysis7. Thus in addition to GDP per capita we have

considered two other indicators of life conditions and possible poverty constraints in the country of

origin: credit to the private sector (from Beck et al. 2000) and life expectancy at birth. The choice

for the latter variable is suggested by two considerations: first, a short life prospect is usually highly

correlated with endemic poverty, and second, it might be also correlated with educational decisions

since the perspective of a longer life increases the number of years over which the returns from

investment in knowledge can be collected8. The private credit by (deposit money) banks over GDP

accounts for financial market imperfections that render liquidity constraints more stringent for poor

families.

2.2 Results

We have selected gross enrolment rates by educational level (secondary and tertiary – primary

enrolment is compulsory almost everywhere, and attendance rates in recent decades tend to reach

100%) over almost two decades (1985-2000); but available data on migration rates by skill levels

restrict us to two points in time only (1990, 2000). Taking into account missing information on

regressors, in its largest version we have 196 observations covering 113 countries for secondary

enrolment, and 182 observations for 109 countries in case of tertiary education. Given its structure,

our dataset consists of an unbalanced sample, with a very limited temporal dimension which

consequently does not offer large variation over time. This implies that when we try to account for

unobserved individual heterogeneity at the country level by estimating a specific fixed effect

parameter iµ , we might end up capturing too much of it, with the estimated individual intercept

washing out part of the effects that are supposed to be explained by the regressors.9 Thus we have

chosen a less demanding specification in which we include macro-regional controls instead of

7 We experimented with data on inequality of income distribution drawn from World Bank (Deininger and Squire)

dataset as well as from UNU-Wider dataset, but in both cases the sample size was almost halved and the variable was

always non significant. Checchi (2003) finds a significant negative correlation between Gini index and secondary

educational attainment, in a larger dataset of low-middle income countries. 8 See Castelló-Climent and Doménech (2008) on the role played by life expectancy onto educational investments.

Complementary considerations on mortality rates are in Grossman (2005). 9 A way to partially relax this assumption and to allow our independent variables to be correlated with the individual

random effect ( iµ ) would be to proceed with the estimator proposed by Hausman and Taylor. We have followed this

strategy in Checchi-DeSimone-Faini (2007) finding similar results.

7

specific country effects.10

This approach is consistent with a peculiar aspect of factor mobility in the

globalising world: Miyamoto (2003) shows that FDI inflows to developing country exhibit highly

differentiation by sector at the regional level. Differently from other regions, the share of inward

investment in the natural resource sector in Africa is high and constant over time. On the contrary,

the Latin American and the Caribbean regions show a large drop in the manufacturing sector with a

corresponding increase in the services sector. The Asian region and the Central Eastern European

Countries exhibit a large and stable share of investment in the manufacturing sector. Since different

types of investment require different types of workers (in terms of skilled/unskilled composition),

they also provide different incentives to invest in education. Thus a specification with regional fixed

effects should enable us to capture possible differentiated effects of inward FDI.

Descriptive statistics are reported in table A1 in the Appendix A.

In Table 1 we report our estimates for secondary enrolment, while Table 2 contains the

corresponding estimates for tertiary enrolment; tables A2 and A3 in the Appendix A repeat the same

analysis by cross sections in single years (1990, 2000). The first two columns of both tables reports

OLS correlations with regional controls, while columns 3 and 4 deal with the potential endogeneity

of migration rates by means of an IV estimator.

We start with secondary enrolment in Table 1. Secondary educational attainment is associated

with the stage of development of a country, possibly reflecting the availability of resources to

families which are necessary to undertake educational investments. While correlations with GDP

per capita and life expectancy at birth are highly significant, point estimates for the credit to the

private sector bear the expected sign, but they are weakly significant. Previous enrolment in primary

education is an important determinant of current enrolment in secondary schools, consistently with

the idea of schooling being a vertically integrated process. The additional control provided by the

pupil/teacher ratio at primary level as a proxy the (average) quality of education received is

negative and statistically significant; similarly the negative sign associated with the (log of)

population density would confirm the lack of school resources and possible supply-side constraints,

but it is statistically insignificant.

As far as factor mobility is concerned we observe that the migration rate of people with

tertiary educational attainment exhibits positive correlation with secondary school enrolments under

all specifications, but the impact is statistically significant only when we do not tackle the

simultaneity issue.11

When we consider an instrumental variable estimator the coefficient on

10

Regional controls are: East Asia and the Pacific, Europe and Central Asia, Latin America and the Caribbean, Middle

East and North Africa, South Asia and Sub-Saharan Africa. See Checchi-De Simone-Faini (2007) for estimates with

country fixed effects and an extended discussion of their implications. 11

Ideally, when considering the impact of migration trends onto enrolment at secondary school one would include

among regressors not only the migration rate at the corresponding educational level, but also at the tertiary level. In fact,

8

migration rate at tertiary level decreases in size and looses statistical significance, suggesting an

upward bias for OLS estimates. In our first-stage migration equation we included the following

instruments:

i) the (log of) stocks of national migrants in major destination areas (US and EU)12

. One may

reasonably argue that the stock of previous migration may attract additional migrants, without

necessarily affect educational choices at home if not indirectly by the perspective of easier

migration and reunion;

ii) an interaction between the (log of) distance from major destination areas (US and EU) and

the domestic population. Empirical literature on migration has indicated the cost of migrating

(proxied by distance from the destination) as one of the major restraints to the movement of

workers. Domestic population accounts for different effects. Small countries tend to be more

prone to migration; furthermore, with immigration systems in developed economies becoming

increasingly selective and/or based on quotas by country (like the US one), the higher the

domestic population and supply of skilled workers is, the lower the probability to migrate13

.

Thus both distance and domestic population are supposed to be negatively correlated with the

migration rate. Taking their interaction allows us to use the distance variable, which is time-

invariant on a panel dimension, and reduces the risk of population not being orthogonal with

respect to our dependent variable.14

Given the evidence of potential endogeneity of tertiary migration with respect to secondary

enrolment, we are inclined to consider skilled migration as having no impact on schooling decisions

at secondary level, just like the presence of foreign investors in the domestic economy (measured by

the log of inward FDI stock) which is never significant in these estimation. Even interacting this

variable with regional controls (in column 4) does not allow us to find evidence of possible regional

patterns. We then conclude that, regardless of the kind of inward foreign investment, the

accumulation of secondary level human capital seems not affected by the presence of foreign firms.

emigration of graduate workers could affect the decision to invest in HC also at the previous level. The high correlation

between the migration rates at secondary and tertiary level (0.74) poses serious collinearity problems on such

specification. So we restrict ourselves to one migration rate only. For the sake of comparability we present here results

obtained by employing the “rate of migration at tertiary level” as independent variable for both (secondary and tertiary)

enrolment rates. We have run regressions including an average of migration rates at secondary and tertiary level; results

are quantitatively and qualitatively consistent with those reported here. 12

See Javorcik et al. (2006) and Beine et al. (2006). 13 See Beine et al. (2006) 14

We have tested for both possible weakness of our instruments and over identifying restrictions. Instruments proved to

be strongly correlated with tertiary migration in the first-stage (partial r-squared = 0.51) and the associated Cragg-

Donald statistics is such to reject the hypothesis of weakness (F=27.15). Sargan statistic of joint over identification of

all instruments point to the direction of a correct choice of instrument at 5% degree of confidence.

9

Table 1 – Gross enrolment rate – Secondary Education (1990-2000) – Unbalanced Panel

1 2 3 4

OLS+RC OLS+RC IV-OLS+RC IV-OLS+RC Log GDP per capita 0.082 0.058 0.063 0.057

[4.70]*** [3.16]*** [3.68]*** [3.24]***

Log of Life Expectancy at Birth, years 0.218 0.231 0.215 0.188

[2.13]** [2.10]** [2.04]** [1.75]*

Private Credit by Deposit Money Banks / GDP 0.125 0.115 0.1 0.118

[1.74]* [1.57] [1.52] [1.74]*

Log inward FDI stock -0.005 -0.002 -0.006

[0.69] [0.23] [1.10]

Migration rate tertiary educated 0.11 0.127 0.019 0.066

[1.81]* [1.91]* [0.25] [0.73]

Enrolment rate primary 5 years before 0.233 0.243 0.24 0.261

[5.28]*** [5.17]*** [4.44]*** [4.57]***

Log pupil/teacher primary -0.131 -0.159 -0.163

[2.85]*** [3.46]*** [3.57]***

Log Population density (people per sq. km) -0.011 -0.006 -0.008

[1.05] [0.65] [0.85]

Log inward FDI stock × East Asia and Pacific -0.007

[0.71]

Log inward FDI stock × Eur. and Central Asia -0.007

[0.93]

Log inward FDI stock × Latin Am. and Carib. 0.012

[0.82]

Log inward FDI stock × Mid. East and N. Africa -0.032

[0.99]

Log inward FDI stock × South Asia 0.004

[0.17]

Log inward FDI stock × Sub-Saharan Africa -0.011

[0.92]

Observations 196 175 173 173

R² 0.8 0.81 0.81 0.82

Robust t statistics in brackets - * significant at 10%; ** significant at 5%; *** significant at 1%

Regional controls and year dummy included

IV for FE: log of stock of own migrants in US and in EU, log of (population*distance from EU), log of

(population*distance from US)

We now consider tertiary enrolment, as reported in Table 2. Most of stage-of-development

controls do not work in this case: credit to the private sector and life expectancy are not statistically

significant here, while there is a negligible negative impact of GDP per capita. This may not be

surprising, since people attending university in developing countries are typically self-selected from

the upper tail of the income distribution, and they are relatively unaffected by what happen in the

lower tail.15

As in the previous stage of education, we find a positive contribution of enrolment rates

at secondary level, while the negative and significant sign on pupil/teacher ratio suggests supply-

side constraints.

15 One additional control that has been introduced in the previous literature is the share of fuel and raw materials in total

exports. The rational is that whenever a country is abundant in natural resource, its population has fewer incentives to

get educated. We tried to include this indicator in our regressions for both secondary and tertiary enrolment, finding a

limited negative impact, as expected. Once again, the presence of missing data has suggested to stick to our current

specification that excludes this variable.

10

Table 2 – Gross enrolment rate – Tertiary Education (1990-2000) – Unbalanced Panel

1 2 3 4

OLS+RC OLS+RC IV-OLS+RC IV-OLS+RC Log GDP per capita -0.01 -0.021 -0.02 -0.017

[0.94] [1.96]* [2.15]** [1.80]*

Log of Life Expectancy at Birth, years 0.045 0.051 0.05 0.025

[1.03] [1.05] [0.89] [0.45]

Private Credit by Deposit Money Banks / GDP 0.037 0.054 0.055 0.046

[0.80] [1.18] [1.56] [1.31]

Log inward FDI stock 0.009 0.011 0.012

[2.77]*** [3.74]*** [3.79]***

Migration rate tertiary educated -0.199 -0.185 -0.149 -0.151

[6.68]*** [6.24]*** [2.92]*** [2.67]***

Enrolment rate secondary 5 years before 0.241 0.202 0.2 0.225

[6.26]*** [5.62]*** [5.37]*** [6.08]***

Log pupil/teacher primary -0.061 -0.058 -0.051

[2.39]** [2.22]** [2.01]**

Log Population density (people per sq. km) -0.002 -0.004 -0.004

[0.49] [0.77] [0.75]

Log inward FDI stock × East Asia and Pacific 0.012

[2.13]**

Log inward FDI stock × Eur. and Central Asia 0.02

[4.62]***

Log inward FDI stock × Latin Am. and Carib. 0.014

[1.81]*

Log inward FDI stock × Mid. East and N. Africa 0.018

[1.05]

Log inward FDI stock × South Asia 0.003

[0.26]

Log inward FDI stock × Sub-Saharan Africa -0.006

[0.95]

Observations 182 163 163 163

R² 0.76 0.77 0.77 0.79

Robust t statistics in brackets - * significant at 10%; ** significant at 5%; *** significant at 1%

Regional controls and Year dummy included

IV for FE: log of stock of own migrants in US and in EU, log of (population*distance from EU), log of

(population*distance from US)

When considering factor mobility, we find that migration rate at tertiary level has a negative

and statistically significant correlation with enrolment under all specifications (column 1-4); the

same effect declines just slightly when we take care of potential endogeneity.16

In this case we also

find that the presence of foreign firms in the domestic market (proxied by the stock of inward FDI)

exerts a significant positive impact17

. We interpret it as evidence that inward FDI creates job

opportunities for skilled workers, thus providing an incentive to enrol in a higher education

program. Looking at compositional effects (column 4) we observe that it is mostly driven by

16 Instruments explain up to 46% of migration rate of tertiary educated workers in the first-stage and the associated

Cragg-Donald statistics is such to reject the hypothesis of weakness (F=30.64). Sargan statistic of joint over

identification of all instruments point to the direction of a correct choice of instrument at all degrees of confidence. 17

The impact of capital inflows onto higher educational attainment has been studied by Hegger et al. (2005), finding

positive correlation in the Barro-Lee dataset.

11

formerly planned economies18

and East Asian countries, which typically receive inward investments

in the manufacturing sector, and by Latin American countries, which experienced significant inflow

of foreign capital in the service sector due to privatisation. As expected, investments in the primary

sector (exploitation of natural resources) do not provide incentives to accumulate human capital in

developing countries.

Overall, migration rate of skilled workers seems to discourage enrolments (that per se would

imply a discouraging effect of the brain drain), but the presence of foreign firms on domestic

market provides offsetting positive incentives to higher education enrolment. Thus, we would be in

the presence of a peculiar form of brain waste. Natives would be attracted into tertiary education by

existing job opportunity created by foreign firms in the local economy (stock of inward FDI), but

the outflow of tertiary graduates through migration would dampen this incentive.19

Evaluated at

sample averages,20

the opening to international factor flows would imply an increase in tertiary

enrolment, since the impact of the inflow of foreign capital would dominate the corresponding

negative impact created by the outflow of skilled workers. But the migration of skilled worker also

directly reduces the existing stock of tertiary educated workers, the overall effect being negative, as

it is discussed in section 4.

Our results are in line with those obtained by Groizard and Llull (2006) but we model the

stock-flow relationship in a more consistent way. In fact, they study the impact of skilled migration

on the cumulated stock of human capital in the country, which almost by construction yields a

negative impact (since there is a one-to-one correspondence between a migration of a graduate

worker and a (marginal) decline in the average years of education in the working population of the

source country). On the contrary, if there are disincentive effects of migration, these should work

through the accumulation of new human capital, namely the enrolment (and, if available,

completion) rates, as we have done in our regressions. In addition, they neglect other factors that

may affect the educational attainment in the country, out of the initial level, while we have provided

a richer analysis of the process.

18 Bulgaria, Hungary, Poland and Romania in our sample. 19

Data on migration rates provided by Doquier and Marfouk (2004) define skilled migrants by education level without

distinguishing whether they acquired their education in the home or the host country. This might suggest that

coefficients on the migration rate of tertiary educated migrants obtained in our estimates could be biased. As an

additional robustness check we replicated our two sets of regressions (i.e. secondary and tertiary enrolments) employing

a migration rate of skilled workers which accounts only for foreign-born adults arrived in the destination country after

age 22. The assumption is that immigrants' age of entry is a good proxy for where they acquired their education (see

Beine et al. (2007)). Results obtained are not dissimilar from those shown here. This it is not surprising, since the

correlation rate of the two series exceeds 90% in our sample 20

The relevant elasticities are rather different at sample averages: by considering an estimated coefficient between -0.13

and -0.20, the migration elasticity lies in an interval comprised between 0.18 and 0.28, while the elasticity of the inward

FDI stock (using an estimated coefficient of 0.01) is equal to 0.42.

12

Thus our overall conclusion of this section casts doubt on the presumed beneficial effect of

factor mobility onto domestic accumulation of HC: international flows of capital and labour

generate conflicting incentives to the accumulation of HC in developing countries.. On one side

there is evidence that skilled labour migration plays a disincentive effect on enrolment decision at

the corresponding level of education. On the other side there is evidence that inward FDI modify

the relative incentives to acquire tertiary education (possibly through the adjustment of relative

returns to educational attainment).

3. Are FDI attracted by the availability of human capital?

It has been argued that foreign firms determine the choice of location looking at the

availability of human capital. Thus, along with other possible determinants, relative endowment of

human capital should affect the attractiveness of certain locations. Related questions concern the

type of human capital (education and skills) that foreign investors are seeking for, and whether

different types of firms seek different sets of skills.

Our second equation aims to model the dynamics of physical capital accumulation through

domestic inflow of foreign capitals. A linear version of equation (6) in Appendix B describes the

determinants of FDI inflows, including the domestic endowment of human capital

ittitij

it

T

ik

ktiiit ZHCFDIFDI λ+τ+⋅φ+⋅θ+

⋅θ+γ= ∑

=− 2,1 log (2)

where the cumulated sum of past FDI proxies the current stock of foreign capitals, Z is a set of

country specific factors affecting investment decision choices (control variables), iγ is a country

(area) fixed effect, tτ is a time fixed effect and itλ is an error term.

In order to fully account for possible feedback effects due to factor mobility, it would be

desirable to include in this specification the impact of return migration on the inflows of FDI.

Unfortunately comparable cross-country series on return migration rates are not available. Including

alternative measures of the stock of national migrants living abroad in the investing countries is a

method adopted in recent contributions to account for possible network effects21

. Ideally, this

approach would rely on a strict bilateral setting, otherwise it would be impossible to ascertain

whether largest flows of FDI to the domestic economy actually come from countries hosting larger

share of own migrants. Unfortunately, data on FDI flows available for developing countries are

rarely collected on a bilateral basis. In order to maintain the cross-country dimension of our analysis

21

See Docquier and Lodigiani (2006) and Javorcik et al. (2006).

13

we are therefore forced to employ data on total inflows and stocks of FDI regardless of the country

of origin. Data on stock of immigrants by country of origin in developed economies are provided by

Defoort (2006) on a large panel (1975-2000) for 6 major OECD destinations only: Australia,

Canada, France, Germany, United Kingdom, US. Considering that outward flows of migrants from

developing countries are mostly directed toward these few destinations22

which are also among the

major foreign direct investors at the world level23

, we can include in our specification the stock of

own migrants in these countries in the previous period to account for possible network effects in a

supposedly pseudo-bilateral setting.

3.1 Dataset and variables definition

We have created a second dataset integrating series of average net inflows of FDI computed

on a five-year basis (1990-1994, 1995-1999, 2000-2004) from the UNCTAD database, with series

of alternative proxies for human capital stocks obtained merging data from Barro and Lee (2000)

and Cohen and Soto (2007)24

, so to cover the highest possible number of countries in our sample.

Taking the average of FDI flows allows us to smooth the intrinsic volatility of capital flows aimed

to buy existing activities and allow for the non-instantaneous setting-up of newly started business.

We have also added some standard controls introduced in the literature on the determinants of

foreign direct investment (market-seeking, efficiency-seeking):

i) (log of) inward stock of FDI, to take into account the effects of reinvested profits and scale

economies;

ii) (log of) GDP per capita, to proxy the stage of development;

iii) (log of) population, to capture “market size” effects;

iv) price inflation, measured by consumer price index annual percent changes, averaged over 5-year

intervals (1988-1992, 1993-1997, 1998-2002), to account for economic stability;

v) to account for political stability and other determinants of institutional quality we relied on six

different indicators collected by Kaufmann et al. (2004): Voice and Accountability, Political

Stability, Government Effectiveness, Regulatory Quality, Rule of law, Control of Corruption.

Since all these measures (which are obtained by aggregating different opinion surveys

worldwide) are highly correlated among them, we summarise them by extracting a common

22

Defoort (2006) shows that the listed 6 major destinations collect up to 77% of world immigration. 23

Canada, France, Germany, United Kingdom, and US ranked among the top ten investors at the world level (Outward

FDI stocks) in the 1990-2002 period. Australia ranked 14th

. See UNCTAD (2003). 24

The overall series for HC measures are built by integrating the Cohen and Soto (2007) (C-S) data with information in

Barro-Lee (2000) (B-L) for missing countries (retaining the C-S observations in case of overlap). C-S collect data on

educational attainments in total population on a 10-year basis, while data in BL are available with 5-years intervals.

Thus, we built the series for 1995 for countries in C-S by projecting the 1990 observations using the growth rates

obtained from B-L.

14

factor from the series using factor analysis (principal component method). The first common

factor obtained, which is used in our analysis, summarizes up to the 78% of original series

variations25

;

vi) we also include trade openness (proxied by the (Import + Export) share in GDP) to consider the

exposure to globalisation forces in a country. Following our previous work (Faini 2004), we

were to include telephone mainlines (per 1000 people), to account for the endowment of

infrastructures at country level, but this variable restricts the sample significantly without ever

being significant; as a consequence we have left it out of our best models.

This set of control variables includes what the current empirical literature recognizes as

major determinants. Nevertheless, the focus of our analysis is the identification of a potential role of

human capital endowment in attracting FDI. Given the fact that we try to capture possible fixed-

effect (group/country specific) with appropriate estimation techniques, the possible risk of omitted

variables does not seem to be a major impediment. As far as our measure of the stock of domestic

human capital is concerned, we have considered alternative measures, either based on the average

years of education in the population or on the distribution of the educational attainment in the same

population. We have selected the second alternative, because it allows us to distinguish between

different levels of skill (associated to different level of educational attainments).

Descriptive statistics of this dataset are reported in Table A4. Since in this equation we do

not rely on migration-related information, our dataset is not anymore restricted to two points in

time: when considering the unbalanced version we have 243 observations from 85 countries,

whereas the balanced panel is composed of 180 observations for 60 countries, referred to 1990,

1995 and 2000. There is only a partial overlap with the dataset used in the previous section (63

countries when considering both balanced versions), because some countries26

report information

on migration, but do not give account of FDI inflows, while some other countries27

attract funds

from abroad, but do not seem to be sending migrants out of the country.

25

Data collected in Kaufmann et al. (2004) go back to mid-1990s only. We use the first available observation for 1990

and the proper one for 2000. An average of the two is assumed to be the corresponding value for 1995. 26 Countries included in the enrolment model of section 2 and not in the FDI model of section 3 are Albania, Armenia,

Bahamas, Belize, Cambodia, Cape Verde, Chad, Croatia, Czech Republic, Estonia, Ethiopia, Georgia, Guinea-Bissau,

Kazakhstan, Kyrgyzstan, Laos, Latvia, Libya, Lithuania, Macedonia, Moldova, Mongolia, Namibia, Oman, Romania,

Rwanda, Samoa, Slovakia, Slovenia, Vanuatu, Vietnam, Yemen, (average GDP per capita in 2000 equivalent to 2161

US dollars). 27

Countries included in the FDI model of section 3 and not in the enrolment model of section 2 are Afghanistan,

Barbados, China, Cuba, Ethiopia, Fiji, Liberia, Myanmar, Rwanda (average GDP per capita in 2000 equivalent to 1034

US dollars).

15

3.2 Results

Our results for the unbalanced panel using alternative measure for HC (secondary and

tertiary attainment) are reported in Table 3. Results for the balanced panel are in Tables A5 in the

Appendix A. Here we start with the two key variables in our conceptual framework under OLS

estimator and regional controls (column 1 and 4); then we add regressors accounting for motives of

FDI (columns 2 and 5), “market-seeking” and/or “efficiency-seeking”; finally, we include all other

controls regarding the business environment and economic and political stability plus the stock of

own migrants in major OECD destinations to account for possible network effects (columns 3 and

6). In column 7 we replicate the regression of the largest specification jointly considering the

secondary and the tertiary attainment in the population.

An elastic supply of HC seems not to attract foreign investors: neither the availability of

secondary educated workers nor that of tertiary educated workers are significant determinants of the

inward flows of FDI in our sample. A substantial part of inflows is due to reinvested profits or

expansion of existing investments (as witnessed by the positive correlation with existing FDI

stock). The level of development, as proxied by GDP per capita, seems to make the location more

attractive for foreign investors. Given the positive and significant correlation with the log of

population (which proxies for the size of the market) one might think that “market-seeking” motives

for FDI are still relevant in inducing foreign investment in the country.

The openness to trade is positively associated with the inward flows of FDI but its impact is

rather limited. Economic instability (here proxied by the average inflation rate) bears the correct

sign but it is never significant, while political stability seems to have a significant positive impact

on the location attractiveness. There is some collinearity between the GDP per capita and the

government factor (correlation in the unbalanced sample is equal to 0.65). Since more developed

countries experience more stable governments (the executives are more effective, the rule of law is

more frequently enforced, the level of corruption is lower, the regulatory quality is more valuable),

the effect of GDP per capita that we measure once we introduce this factor (from column 4 onward)

is basically the net effect of the stage of development. Finally, we find no evidence of possible

network effects. Having large communities of own migrants in the 6 major OECD destinations does

not seem to give any advantage at the aggregate level in terms of incoming FDI.

As argued above, when analysing the relationship between existing supply of HC and the

inflows of capital from abroad, it is crucial to discuss the nature of foreign investment. If FDI is

aimed to the exploitation of natural resources, the local availability of educated labour force could

be less relevant than in the case of investments in manufacturing or services. This has been already

16

highlighted in the previous section, where we found a differential effect in the impact of inward FDI

on educational decisions, depending on the level of education.

Table 3 – 5-years Average Inflows of FDI – 1990-1995-2000 – Unbalanced Sample

1 2 3 4 5 6 7

OLS+RC OLS+RC OLS+RC OLS+RC OLS+RC OLS+RC OLS+RC

population with secondary attained 0.314 0.773 0.366 0.329

[0.53] [1.20] [0.54] [0.48]

population with tertiary attained 1.387 1.003 0.627 0.466

[0.73] [0.56] [0.32] [0.24]

log stock of inward FDI 0.91 0.573 0.481 0.91 0.588 0.49 0.482

[16.29]*** [7.49]*** [5.43]*** [16.44]*** [8.13]*** [5.85]*** [5.39]***

log gdp per capita 0.611 0.536 0.632 0.535 0.531

[5.29]*** [3.74]*** [5.49]*** [3.54]*** [3.54]***

log population 0.412 0.628 0.39 0.608 0.621

[5.20]*** [5.80]*** [5.33]*** [5.93]*** [5.47]***

Factor extracted from political variables 0.289 0.299 0.289

[2.26]** [2.44]** [2.25]**

Trade (% GDP) 0.005 0.005 0.005

[2.27]** [2.29]** [2.23]**

Inflation, consumer prices (annual %) -0.017 -0.016 -0.017

[0.92] [0.91] [0.93]

Stock of own migrants in 6 OECD at (t-1) -0.008 0.004 -0.004

[0.16] [0.07] [0.07]

Observations 243 231 206 243 231 206 206

R-squared 0.77 0.80 0.82 0.77 0.80 0.82 0.82

Number of Countries 85 81 77 85 81 77 77

Robust t statistics in brackets - * significant at 10%; ** significant at 5%; *** significant at 1%

Year dummies and RC (regional controls) included and not reported

4. Policy implications

Is there any evidence of a virtuous circle of human capital formation and increased inflow of

FDI? What are the implications of our estimates? In order to adapt our estimates to our theoretical

framework (see Appendix B), we need to clarify the relationship between human capital and

enrolment rates. If we approximate the total human capital stock H by the average years of

education in the population, it is defined as TlSlPlH tsp ⋅+⋅+⋅= where tsp lll ,, are respectively

the school length of primary, secondary and tertiary education, while TSP ,, are the corresponding

population shares. Taking tsp lll ,, as fixed, we have that TlSlPlH tsp&&&& ⋅+⋅+⋅= . If we consider a

stationary population with a fixed life length λ , then λ/1 is the relative size of each age cohort, and

λ/il is the reference population for educational level i .28

The share of population with a given

educational attainment increases whenever the corresponding enrolment rate is greater than the

existing share. For example the variation of the population share with secondary education can be

28

One could argue that life expectancy varies with the educational attainment. But then changes in educational

attainment would translate in changes in total population, making the analytics unmanageable.

17

described by the following expression λ

−≅

SenrolmentS S& .

29 Thus the variation of human capital

stock is given by

( ) ( ) ( )Tenrolmentl

Senrolmentl

Penrolmentl

H TT

SS

PP −⋅

λ+−⋅

λ+−⋅

λ=& (3)

Thus equation (3) implies that the overall effect of FDI on the accumulation of human

capital is given by K

enrolmentl

K

enrolmentl

K

enrolmentl

K

H TTSSPP

∂

∂⋅

λ+

∂

∂⋅

λ+

∂

∂⋅

λ=

∂

∂ &

. If we neglect

the impact of FDI on primary enrolment, where it is statistically insignificant in any specification,

and we take the estimate reported in third column of Tables 1 and 2, we obtain that

( ) ( ) ( ) 001.001.0006.060

5 *** +=+−⋅≅∂

∂⋅

λ+

∂

∂⋅

λ=

∂

∂

K

enrolmentl

K

enrolmentl

K

H TTSS&

which is positive

but very small. Since K is measured in logs, it implies that doubling the stock of FDI ( 1+=∆K )

would (dynamically) increase the human capital stock by 0.001 years of education in the

population, while changing the skill composition in the labour force in favour of tertiary educated

workers.

If we take the migration decision as exogenous, the Jacobian corresponding to system (8) in

Appendix B is therefore given by30

29

This expression can be derived as follows. If we define totPOP as the total population and secPOP the population

with secondary educational attainment, then totPOPPOPS /sec= as well as sec

sec

sec

sec

POP

POP

POP

POP

POP

POP

S

S

tot

tot&&&&

=−= under the

assumption of constant population. Given the fact that secsecsec deadgraduate −=POP & , that graduates are assumed to be

a constant fraction α of (current) enrolment

λ⋅⋅α=⋅⋅α=⋅α= totPOP

secsecsecsec rateenrolment population referencerateenrolment enrolledgraduate

and that the outflow of secondary educated population is a constant fraction of the existing stock λ

= secsecdead

POP,

then ( )SSPOP

POP

POP

POP

S

S−⋅α

λ=⇔

λ−

⋅λ

⋅⋅α== sec

sec

totsec

sec

sec rateenrolment 11rateenrolment

&&&

which corresponds

to what shown in equation (3) for 1=α .

30

The figures reported in the second row of the Jacobian (4) are obtained by OLS regression of average FDI flow onto

FDI stock and average years of education in the population, which replaces the population shares with different

educational attainment (primary, secondary and tertiary). In the unbalanced sample it yields

[ ] [ ]controls regionallog619.0127.0

67.763.1++⋅+⋅= POPKHK&

Results are however unstable depending on whether GDP per capita is included/excluded and the data source for years

of education. Finally, the coefficient of HH ∂∂ /& is derived under the assumption of identical duration for each

18

×

−=

K

H

K

H

619.0127.0

001.008.0

&

&

(4)

which is saddle-path stable.

Going to the debate over brain gain/drain, let us consider an exogenous increase in migration

of skilled (tertiary educated) workers, in the order of 100%. At sample mean of the balanced panel,

this implies a passage from 0.145 to 0.290. Looking at Tables 1 and 2 (column 4) this entails an

increase in secondary enrolment of 0.9 (corresponding to an impact of +0.066×+0.145, though non

statistically different from zero) accompanied by a reduction in tertiary enrolment of 2.1

(corresponding to an impact of –0.151**×+0.145). If we are available to assume that the average

school length at secondary and tertiary level is approximately 5 years, we obtain a reduction in the

average years of education of 0.06 (which becomes 0.10 if we consider only statistically significant

effects), approximately one twentieth (one tenth respectively) of a year of schooling in the

population. We now know from previous results, that this produces a reduction in capital inflow:

since our dependent variable in the estimation of Table 3 is the log of the ratio between capital

inflow and GDP, a variation of –0.037 (corresponding to +0.627×–0.06, though not statistically

significant) implies a significant drop of capital inflows, in the order of one fifth of pre-existing

flows (equal to 0.18 at sample mean of the balanced panel). In the long run, this reduction

cumulates in lower stock, yielding lower enrolment and lower human capital.

educational level ( 5=== TSP lll ) and an identical life expectancy for any educational attainment, equal to 60 years

( 08.060

5−=− ) (sample average is 61.43).

19

Figure 1

EastAsia Pacific

European/central Asia

Lat.America Carribbean

mid.East North Africa

south Asia

subSaharan Africa

24

68

ave

rag

e y

ea

rs o

f e

du

ca

tio

n

4 5 6 7 8log inflow FDI

Change over 1990-2000

When we graph the region position according to these state variables (human capital stock

H , proxied by the average years of education in the population, and physical capital stock, K ,

proxied by the log of foreign investment) we observe that regions tend to align along a downward

sloping ray, and they shift rightward over the decade at different speeds (see figure 1). Qualitative

analysis of the dynamic properties of system described by equation (4) indicates that the stable

branch of the saddle path will exhibit a negative slope comprised between 85.2217.0

619.0−≅− and −∞ ,

while the unstable branch can be either positively or negatively sloped, but with a slope lower than

01.008.0

001.0+≅ . Thus, according to our model, most of the countries in the regions would be

positioned in the proximity of the stable path, leading to a convergence in long run to given values

in both H and K . In this case both human capital and physical capital move together, exhibiting a

negative correlation (given the negative slope of the stable branch of the saddle-path).

20

5. Concluding remarks

Two main results are obtained in this paper, which can be summarized as follows:

1. we do not find strong evidence of the existence of a virtuous cycle between human capital

accumulation and foreign direct investment. In our estimates, FDI does not affect secondary

enrolment while favouring tertiary enrolment, and the overall effect is weakly positive. On

the other side, in our data FDI are weakly affected by the existing endowment of human

capital, mostly depending on the proxy adopted for HC.. If the underlying dynamics of the

dynamical system in ( )KH , space is analysed by dynamical system analysis, we would

identify a saddle-path stable system, which implies the existence of a unique combination of

stocks of human capital and foreign capital leading to a stable equilibrium, all other

combinations leading to unstable trajectories;

2. in addition to direct reduction of domestic human capital, we find evidence of a sort of brain

drain through skilled (tertiary educated) worker migration. We interpret this result as a

disincentive effect: when the domestic population observes that a large share of university

graduate migrating abroad, it takes this as evidence of lack of adequate local job

opportunities, and reduces the corresponding investment in higher education.

On both grounds, less developed countries are not necessarily benefited by factor mobility: they

gain domestic human capital under foreign capital inflow and but they loose domestic human

capital through skilled worker migration. Unfortunately we do not possess data on the type of FDI

involved in this analysis. Looking at their geographical distribution, we suspect that our story

involves natural resource exploitation (like mining and oil extraction) rather than Greenfield

investment. In such a case the local endowment of human capital is less relevant, as well as the

incentive to further human capital accumulation.

21

APPENDIX A – Additional tables

Table A1 – Descriptive statistics (1990-2000)

Variable Obs Mean

Std.

Dev. Min Max Obs Mean

Std.

Dev. Min Max

unbalanced panel balanced panel

gross enrolment rate secondary 283 0.55 0.30 0.05 1.15 120 0.48 0.28 0.05 0.05

gross enrolment rate tertiary 255 0.15 0.14 0.00 0.59 120 0.11 0.11 0.00 0.51

Enrolment rate primary 5 years before 254 0.91 0.25 0.10 1.47 120 0.91 0.22 0.24 0.47

Enrolment rate secondary 5 years before 254 0.49 0.30 0.03 1.19 120 0.43 0.26 0.03 0.01

log GDP per capita 277 6.95 1.18 4.45 9.69 120 6.79 0.23 0.45 0.38

log inward stock of FDI 283 6.15 2.65 -4.60 12.17 120 6.96 0.45 0.60 0.54

Log of Life Expectancy at Birth, years 299 4.12 0.19 -0.57 12.88 120 4.07 0.20 0.43 0.35

Private Credit by Deposit Money Banks / GDP 250 0.23 0.19 0.00 1.04 120 0.23 0.19 0.00 0.04

log pupil/teacher primary 249 7.94 0.41 6.97 8.95 120 8.06 0.39 0.96 0.84

Migration rate secondary educ 281 0.08 0.14 0.00 0.70 120 0.04 0.06 0.00 0.30

Migration rate tertiary educ 281 0.21 0.23 0.00 0.92 120 0.15 0.16 0.00 0.84

log Population density (people per sq. km) 295 3.87 1.31 0.30 6.90 120 3.83 0.29 0.29 0.85

Table A2 – Gross enrolment rate – Secondary Education – Cross Section 1990 / 2000

1990 1990 1990 2000 2000 2000 OLS+RC OLS+RC IV-OLS+RC OLS+RC OLS+RC IV-OLS+RC Log GDP per capita 0.095 0.07 0.072 0.078 0.052 0.057

[3.21]*** [2.32]** [2.67]*** [3.59]*** [2.60]** [2.83]***

Log of Life Expectancy at Birth,

years 0.225 0.316 0.296 0.142 0.087 0.113

[1.49] [2.10]** [1.96]** [0.97] [0.63] [0.75]

Private Credit by Deposit Money

Banks / GDP 0.035 0.03 0.02 0.149 0.122 0.112

[0.32] [0.27] [0.17] [1.54] [1.32] [1.46]

Log inward FDI stock -0.014 -0.017 -0.019 0.004 0.014 0.006

[1.45] [1.87]* [2.73]*** [0.40] [1.48] [0.69]

Migration rate tertiary educated 0.109 0.118 0.043 0.134 0.177 0.032

[1.11] [1.06] [0.40] [1.75]* [2.34]** [0.29]

Enrolment rate primary 5 years

before 0.218 0.249 0.235 0.278 0.245 0.252

[3.26]*** [3.45]*** [3.05]*** [4.64]*** [3.75]*** [3.61]***

Log pupil/teacher primary -0.055 -0.081 -0.182 -0.197

[0.89] [1.27] [3.00]*** [3.29]***

Log Population density (people per

sq. km) 0.006 0.007 -0.019 -0.011

[0.36] [0.60] [1.54] [0.97]

Observations 87 78 77 109 97 96

R² 0.74 0.75 0.76 0.83 0.86 0.85

Robust t statistics in brackets - * significant at 10%; ** significant at 5%; *** significant at 1%

Regional controls included.

IV for FE: log of stock of own migrants in US and in EU, log of (population*distance from EU), log of

(population*distance from US)

22

Table A3 – Gross enrolment rate – Tertiary Education – Cross Section 1990 / 2000

1990 1990 1990 2000 2000 2000

OLS+RC OLS+RC IV-OLS+RC OLS+RC OLS+RC IV-OLS+RC

Log GDP per capita -0.004 -0.008 -0.007 -0.013 -0.026 -0.026

[0.42] [0.86] [0.74] [0.81] [1.52] [1.82]*

Log of Life Expectancy at Birth,

years -0.027 -0.024 -0.015 0.102 0.1 0.117

[0.68] [0.57] [0.27] [1.35] [1.20] [1.12]

Private Credit by Deposit Money

Banks / GDP 0.036 0.045 0.038 0.051 0.079 0.075

[0.76] [0.95] [0.95] [0.71] [1.04] [1.48]

Log inward FDI stock 0.007 0.006 0.007 0.003 0.007 0.006

[2.89]*** [2.62]** [2.56]** [0.57] [1.38] [1.03]

Migration rate tertiary educated -0.212 -0.211 -0.175 -0.181 -0.153 -0.191

[7.08]*** [6.68]*** [4.13]*** [3.54]*** [2.87]*** [2.10]**

Enrolment rate primary 5 years

before 0.185 0.166 0.16 0.31 0.254 0.254

[4.09]*** [3.84]*** [4.73]*** [4.88]*** [3.94]*** [3.77]***

Log pupil/teacher primary -0.028 -0.025 -0.064 -0.064

[1.27] [1.08] [1.47] [1.46]

Log Population density (people per

sq. km) 0.005 0.003 -0.005 -0.003

[1.33] [0.76] [0.74] [0.40]

Observations 83 75 75 99 88 88

R² 0.77 0.76 0.75 0.77 0.79 0.79

Robust t statistics in brackets - * significant at 10%; ** significant at 5%; *** significant at 1%

Regional controls included.

IV for FE: log of stock of own migrants in US and in EU, log of (population*distance from EU), log of

(population*distance from US)

Table A4 – Descriptive statistics – 1990-1995-2000

Variable Obs Mean

Std.

Dev. Min Max Obs Mean Std. Dev. Min Max

unbalanced panel balanced panel

log of inflow over GDP - average over 3 years 439 4.49 2.38 -6.21 10.84 180 5.44 2.23 0.67 10.84

population share with secondary 259 0.19 0.14 0.01 0.79 180 0.20 0.15 0.01 0.79

population share with tertiary completed 259 0.05 0.05 0.01 0.32 180 0.06 0.05 0.01 0.32

log gdp per capita 418 6.92 1.18 4.03 9.69 180 6.87 1.20 4.45 9.38

log stock of inward FDI 428 6.17 2.52 -4.61 12.17 180 7.42 1.95 1.79 12.17

factor extracted from political variables 362 0.00 1.00 -2.57 3.06 180 0.16 0.89 2.45 3.06

Trade (% GDP) 411 80.96 40.34 3.15 228.88 180 69.08 38.38 14.41 228.88

Inflation, consumer prices (annual %) 373 0.88 4.44 -0.03 53.99 180 1.17 5.98 0.03 53.99

log population 448 15.38 2.01 10.62 20.96 180 16.50 1.48 13.11 20.96

23

Table A5 – 5-years Average Inflows of FDI – 1990-2000 – Balanced Panel –

1 2 3 4 5 6 7

OLS+RC OLS+RC OLS+RC OLS+RC OLS+RC OLS+RC OLS+RC

population with secondary attained 0.645 1.206 0.748 0.764

[1.03] [1.86]* [1.11] [1.13]

population with tertiary attained 1.173 0.653 0.233 -0.182

[0.58] [0.35] [0.12] [0.09]

log stock of inward FDI 0.839 0.411 0.416 0.85 0.445 0.436 0.415

[12.9]*** [4.63]*** [4.55]*** [13.7]*** [5.27]*** [5.05]*** [4.48]***

log gdp per capita 0.775 0.535 0.819 0.544 0.537

[5.54]*** [3.33]*** [5.73]*** [3.23]*** [3.23]***

log population 0.572 0.65 0.53 0.622 0.653

[6.21]*** [5.90]*** [6.29]*** [5.94]*** [5.65]***

factor extracted from political variables 0.331 0.357 0.331

[2.31]** [2.64]*** [2.30]**

Trade (% GDP) 0.005 0.005 0.005

[1.63] [1.78]* [1.67]*

Inflation, consumer prices (annual %) -0.018 -0.016 -0.018

[0.97] [0.93] [0.96]

Stock of own migrants in 6 OECD at (t-1) 0.022 0.035 0.02

[0.37] [0.56] [0.32]

Observations 180 180 180 180 180 180 180

R-squared 0.77 0.82 0.83 0.77 0.82 0.83 0.83

Number of Countries 60 60 60 60 60 60 60

Robust t statistics in brackets - * significant at 10%; ** significant at 5%; *** significant at 1%

Year dummies included – RC (regional controls) included in OLS .

Countries included: Algeria, Argentina, Bahrain, Bangladesh, Bolivia, Brazil, Bulgaria, Burkina Faso, Chile, China, Colombia,

Congo, Dem. Rep. of the, Congo, Rep. of the, Costa Rica, Ecuador, Egypt, El Salvador, Ethiopia, Gambia, Ghana, Guatemala, Haiti,

Honduras, Hungary, India, Jamaica, Jordan, Madagascar, Malawi, Malaysia, Mali, Mauritius, Mexico, Morocco, Nepal, Nicaragua,

Niger, Nigeria, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Senegal, Sierra Leone, Sri Lanka, Syria,

Tanzania, Thailand, Togo, Trinidad and Tobago, Tunisia, Turkey, Uganda, Uruguay, Venezuela, Zambia, Zimbabwe.

24

APPENDIX B – Theoretical considerations

We are interested in analysing the long run consequences of factor mobility on human

capital investment in developing countries, when feedbacks from capitals and workers mobility

onto educational choices of the population are taken into account. While in the empirical analysis

we will distinguish between different types of educational attainment (as proxy for different levels

of skills in the workforce), here let us define M as the migration rate (defined as the fraction of

nationals leaving the domestic country, which is assumed to take the role of “less developed”

economy), H as the domestic human capital stock and K as the domestic physical capital stock.

While in principle an economy could be either exporter or importer of both workers and

capitals, developing countries are typically net exporter of workers and net importer of (foreign)

capital in the form of foreign direct investment. In addition, the low level of domestic production

and/or the high level of domestic absorption make it rather difficult to obtain domestic

accumulation of physical capital. For this reason we assume that immigration of foreign workers

and domestic investment are set to zero.

Domestic human capital can be augmented through (domestic) school attendance and

decreased through migration of educated workers (the so-called “brain drain”). Furthermore, brain

drain can discourage further investment in education if those who are still in school take the

emigration of skilled individuals as a signal of lack of opportunities for educated workers. However,

some recent literature has drawn attention on the potential existence of a sort of “brain gain”

originated by the increased returns on education due to the increasingly selective immigration

policies in developed countries. Better educated people have higher chances to succeed in

emigrating. This provides an incentive to acquire education in the country of origin yielding an

overall positive balance onto domestic human capital accumulation, under the hypothesis that not

all perspective migrants will succeed in their purpose.

Since both possibility are equally likely, and we are agnostic on this issue, we leave the data

speak. Therefore our first equation is given by

( ) MXMKeH e −=±+

,,& (5)

where dtdHH /=& (the Newtonian derivative), ( )eXMKe ,, summarises school enrolment (with a

supposedly positive impact of foreign investment K in the domestic economy, an ambiguous effect

of migration M and country specific factors eX affecting educational choices – like income

inequality, poverty, school resources and so on). Equation (5) indicates that domestic human capital

stock is increased by school attendance and decreased by migration of educated workers (even if in

the long run the incentive created by migration may rise enrolment and therefore its long run stock).

25

The sign of 0/ >dMde is taken as indicator of the occurrence of “brain gain” (incentive effect),

whereas 0/ <dMde is interpreted as evidence of the possible discouraging effect due to the

ongoing “brain drain”.

Our second equation aims to model the dynamics of physical capital accumulation through

domestic inflow of foreign capitals. We know from the literature that FDI tend to be attracted by the

existence of local favourable conditions31

(like infrastructure, political stability) as well as by the

local availability of skilled labour [Lucas (1990), Zhang and Markusen (1999)], which is positively

correlated with the educational attainment in the population. In addition, we also consider the

possibility of economies of scale and/or of technology/knowledge linkages: both make a new

investment more likely in countries where other investments have already taken place. We also

consider the possibility of a decline in the relative profitability of domestic investment (due to

decreasing marginal productivity, exhaustion of raw materials, shortage of adequate skill): in such a

case, the impact of current stock on new investment would be obviously negative. Thus our second

equation takes the form

( )hXHKkK ,,

+±=& (6)

where hX indicates country specific factors affecting FDI inflows (i.e. infrastructures, degree of

openness, country size, political stability, etc.).

Finally, our third equation takes into account the determinants of outward migration. Here

again the literature is quite substantial. We do consider two aspects: the first one is the impact of the

availability of skilled jobs on the decision to migrate, which is correlated with the prevailing

technology in the country; if the technological progress is embodied in the newly invested physical

capital, then migration should report a negative correlation with foreign direct investment. The

second aspect is the internal competition for skilled jobs, since the greater is the unemployment in

the educated labour force, the longer will be the unemployment spell, and the more likely becomes

the migration. Our assumptions are then summarised in the following

( )mXHKmM ,,

±−= (7)

where mX include the identifying restrictions for this equation, like language facilities, distance,

former colony status, and so on.

Equations (5)-(6)-(7) describe a dynamical system in 2R . In facts, by replacing equation (7)

into (5) we obtain the following system

31

See Blonigen (2005) for a comprehensive review of the literature on FDI determinants. Faini (2004) provides

evidence of a positive effect of domestic human capital stock (proxied by average years of education in the population)

as well as domestic infrastructure (proxied by telephone lines) onto FDI.

26

( )( )

=

=

±+

±±

h

em

XKHkK

XXKHhH

,,

,,,

&

&

(8)

In the case of “brain gain” the Jacobian associated to the system (8) takes the form

±+

−±, while it

exhibits the following signs

±+

+m when negative effects of the “brain drain” prevail. The system

incorporates a feed-back mechanism that contributes to its stabilisation. In facts, when capital stock

increases, it tends to reduce (skilled) workers migration, thus favouring domestic accumulation of

human capital (through the reduced outflow of skilled migrants as well as through an incentive

effect on enrolled students to proceed further on in education). In its turn, an increase in human

capital stock makes additional inflows of new capitals more likely. In both cases, global stability

can be assessed only when the sign and the size of the elements on the Jacobian main diagonal are

known.

27

REFERENCES

Barro R. J. and J. Lee (2000), International Data on Educational Attainment: Updates and

Implications, CID Working Paper No. 42

Beck, Thorsten, Asli Demirgüç-Kunt and Ross Levine, (2000), A New Database on Financial

Development and Structure, World Bank Economic Review 14, pp. 597-605.

Beine M., F. Docquier and H. Rapoport (2001), Brain Drain and Economic Growth: Theory and

Evidence, Journal of Development Economics, vol. 64, pp. 275-289.

Beine M., F. Docquier and. H. Rapoport (2006), Brain Drain and Human Capital Formation in

Developing Countries: Winners and Losers, IRES (Louvain) Discussion Paper No.23

Beine M., F. Docquier and. H. Rapoport (2007), Measuring international skilled migration: new

estimates controlling for age of entry, World Bank Economic Review 21, pp. 249-254.

Blonigen B.A. (2005), A Review of the Empirical Literature on FDI Determinants, NBER Working

Paper No. 11299

Carrington W. and E. Detragiache (1998), How big is the brain drain?, IMF Working Paper

n.98/102.

Castelló-Climent A. and R. Doménech (2008), Human Capital Inequality, Life Expectancy And

Economic Growth, The Economic Journal 118, pp. 653–677

Checchi D. (2003), Inequality in incomes and access to education. A cross-country analysis (1960-

95), Labour 17(2), pp. 153-201

Checchi D., G.DeSimone and R.Faini (2007), Migration, FDI and Human Capital Investment. IZA

Discussion Paper No. 2795/2007

Cohen D. and M. Soto (2007), Growth and Human Capital: Good Data, Good Results, Journal of

Economic Growth 12(1), pp. 51-76

Defoort C. (2006), Tendances de long terme en migrations internationales: analyse à partir de 6

pays receveurs, Université Catholique de Louvain, mimeo

Docquier F. and E. Lodigiani (2006), Skilled Migration and Business Networks, IRES (Louvain)

Discussion Paper No.36

Docquier F. and A. Marfouk (2005), Measuring the International Mobility of Skilled Workers

(1990-2000) - Release 1.1, World Bank, mimeo

Faini R. (2002), Migration, Remittances and Growth, paper presented at the Conference on Poverty,

International Migration and Asylum (Helsinki)

Faini R. (2004). Trade liberalization in a globalizing world. CEPR Discussion Paper n.4665

Faini R. (2006), Remittances and the Brain Drain, Centro Studi d'Agliano Development Working

Paper No. 214

28

Groizard, José Luis and J. Llull (2006). Skilled Migration and Growth. Testing Brain Drain and

Brain Gain Theories. DEA Working Paper No. 20, December 2006

Grossman M. (2005), Education and Nonmarket Outcomes, NBER Working Paper no. 11582

Hegger H., P. Egger, J. Falkinger and V. Grossman (2005), International Capital Market

Integration, Educational Choice and Economic Growth, IZA discussion paper n.1863

Javorcik B.S., C. Ozden, M. Spatareanu and C. Neagu (2006), Migrant Networks and Foreign

Direct Investments, World Bank Policy Research Working Paper No. 4046

Kaufmann D., A. Kraay, M. Mastruzzi (2004), Covernance Matters IV: Governance Indicators for

1996-2004, World Bank Policy Research Working Paper No. 3630

Lucas B.E. (2004), International Migration to the High Income Countries: Some Consequences for

Economic Development in the Sending Countries, Boston University, mimeo

Lucas R.E., Jr. (1990), Why doesn’t Capital Flow from Rich to Poor Countries, American

Economic Review 80, pp. 92-96

Mariani F. (2004), Brain Drain, Inequality and Growth, ESPO/ECON Working Paper No. 33, UCL

Miyamoto K. (2003), Human Capital Formation and Foreign direct investment in Developing

Countries, OECD Development Center Working Paper No. 211

Mountford A. (1997), Can a Brain Drain be Good for the Growth in the Source Economy?, Journal

of Development Economics, 53, pp. 287-303

Sapir A. (2000), Who is Afraid of Globalization? The Challenge of Domestic Adjustment in Europe

and America, CEPR Discussion Paper No. 2595

Stark O., C. Helmenstein and A. Prskawetz (1997), A Brain Gain with a Brain Drain, Economic

Letters, 55, pp. 227-234

Stark O., C. Helmenstein and A. Prskawetz (1998), Human capital depletion, human capital

formation, and migration: a blessing or a “curse”?, Economic Letters, 60, pp. 363-367

UNCTAD (2003), World Investment Report “FDI Policies for Development: National and

International Perspectives”, Geneva

Zhang K. and J. Markusen (1999), Vertical Multinationals and Host-Country Characteristics,

Journal of Development Economics 59, pp. 233-252