BRAIN DRAIN’ OR ‘BRAIN CIRCULATION’ EVIDENCE FROM OECD''S INTERNATIONAL MIGRATION AND R&D SPILLOVERS

As human capital is embodied in people and contains knowledge about new technologies and materials, production methods, or organizational skills, it raises the question of whether the in

‘ B R A I N D R A I N ’ O R ‘ B R A I N

C I R C U L A T I O N ’ : E V I D E N C E F R O M

O E C D ’ S I N T E R N A T I O N A L M I G R A T I O N

A N D R & D S P I L L O V E R S

Thanh Len

Abstract

This paper empirically investigates whether labour mobility can transfer technology across borders based on the panel cointegration method Estimates of specifications on a cross-section of 19 OECD countries during 1980–1990 lend strong support to this thesis Data indicate that international labour movement may help transfer technology across borders in both directions: from donor countries to host countries and vice versa This suggests that migration may more likely create

a ‘brain circulation’ rather than a ‘brain drain’ In addition, human capital has a significant impact on the research and development (R&D) diffusion process as it enhances a country’s capacity to learn from a foreign technology base

I Introduction

In studying the impact of international migration on economic development, many studies (e.g., Haque and Kim, 1995; Wong and Yip, 1999) argue that international migration negatively affects donor countries through the ‘brain drain’ of high skilled workers.1 This brain drain reduces the growth rate of effective human capital that remains in the economy Consequently, the growth rate of per capita income of those countries is retarded

However, there is another research line that suggests a ‘brain gain’ associated with that brain drain: a temporary loss of skilled workers may permanently increase the average level of productivity of the source country This is based

on the following reasoning: the possibility of migration of qualified educated people to a higher income country raises the return to education and, hence, increases the human capital formation which may be greater than the

n University of Queensland, St Lucia, QLD 4072, Australia

1

According to Beine et al (2001), ‘brain drain’ not only means the migration of engineers, physicians, scientists or other very highly skilled professionals but can also be broadly defined

as the emigration of a fraction of the population that is relatively highly educated as compared with the average.

Journal compilation r 2008 Scottish Economic Society Published by Blackwell Publishing Ltd,

9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main St, Malden, MA, 02148, USA

618

negative effect of a brain drain (e.g., Mountford, 1997; Vidal, 1998; Beine

et al., 2001).2

Results in the literature are, therefore, mixed So far, much of the debate is based on the impact of migration on the formation of the stock of human capital As human capital is embodied in people and contains knowledge about new technologies and materials, production methods, or organizational skills, it raises the question of whether the international movement of human capital with embodied technology will give rise to technology diffusion across countries With the existence of bilateral worker flows across economies, foreign workers who acquire R&D-induced technological knowledge through on-the-job training and work experience in their home country may contribute to a productivity increase in the host country In addition, people are often tied to their homeland so by maintaining close and frequent contact with people at home (even visiting home occasionally or regularly), those workers can also contribute knowledge they obtained in the host country to productivity improvement in their home country This suggests a pattern of ‘brain circulation’ rather than a draining of skills from one country to another

So far, economic research on this brain circulation issue is limited to a small number of sectoral case studies, notably within the software industry.3 These case studies show that when integrating into the business community, migrants transfer technical and institutional know-how between distant countries much faster and more flexibly than most corporations In addition, migrant participation in the labour force of the host country may reveal information about production techniques and productivity in their country of origin.4 This paper will revisit the issue of brain drain and brain gain from the aspect

of knowledge spillovers This is achieved by examining the extent to which international labour migration effectively transmits knowledge across countries International R&D spillovers on total factor productivity (TFP) due to worker flows are tested based on the cointegration method against a cross-country data set of 19 OECD countries for the period 1980–1990 The paper also empirically considers the presence of the complementarity between R&D spillovers and investment in human capital: an increase in the level of human capital improves the technological ‘absorptive capacity’ in an open economy context Empirical findings in this study indicate that worker migration can act as a significant channel for R&D spillovers More importantly, the knowledge spillovers may be bidirectional: from a donor country to a host country and vice versa

2 Other possible gains include the return migration of ex ante low-skilled workers who are now equipped with new skills learned abroad (Stark et al., 1997, 1998) and the migrants’ remittances which help alleviate liquidity constraint when financial markets are imperfect (Stark

et al., 1997; Beine et al., 2001).

3

See, for example, Saxenian (2002, 2005).

4 The role of migrant networks in promoting bilateral international trade is also recognized due to the work of Rauch and Trinidade (2002), Rauch and Casella (2003) among others For

an analysis of the relationship between migration and foreign direct investment, see for example, Kugler and Rapoport (2007) However, these issues are beyond the scope of this paper.

‘ B R A I N D R A I N ’ O R ‘ B R A I N C I R C U L A T I O N ’ 619

r 2008 The Author

The results of this study provide novel contributions to the literature on international R&D spillovers and economic growth Recent empirical studies have focused on identifying potential transmission channels of R&D spillovers The main channel is international trade as stated by Coe and Helpman (1995) and many subsequent papers such as Engelbrecht (1997), Lichtenberg and van Pottelsberghe (1998), Keller (1999, 2002), and Frantzen (2000, 2002) These papers find that in the current world of international trade, domestic productivity of one country can benefit from R&D activities occurring in that country’s trading partners Other identified channels include direct foreign technology transfer (Soete and Patel, 1985), foreign direct investment (e.g., van Pottelsberghe and Lichtenberg, 2001), international student flows (e.g., Park, 2004), or pure proximity in a technological space (e.g., Park, 1995; Guellec and van Pottelsberghe, 2001) This paper, therefore, adds a potentially new conduit

of technological diffusion to the literature: the international labour movement The remainder of this paper is structured as follows Section II briefly discusses the theoretical and empirical framework based on which econometric estimates of the impact of foreign R&D embodied in imports and the international labour movement on national productivity growth are performed

A brief data description is given in Section III The main empirical findings and their economic interpretation are reported in Section IV Section V concludes the paper with some closing comments and suggestions for further research

II Theoretical and Empirical Framework

Empirical regressions in this paper are based on some recent theoretical models

of R&D-based growth such as those of Romer (1990), Grossman and Helpman (1991), and Aghion and Howitt (1992, 1998) The production function for a final consumption good Y using labour L and capital K as production inputs is assumed to take the following form5:

Yit¼ FitKa

itL1
ait ; 8i; 0 < a < 1;

where i is a country index (i 5 1,2, ), t is the time index, and F represents the technical efficiency or TFP The specified production function exhibits constant returns to scale to both production factors but diminishing returns to each production factor employed This implies that an index of TFP is defined in the following way:

log Fit¼ log Yit
a log Kit
ð1
aÞ log Lit:

In addition, the growth accounting method indicates that:

gY ¼ gFþ agKþ ð1
aÞgL; where gF, gY, gK, and gL are the rate of growth of TFP, final output, capital stock, and labour force, respectively This implies a causal relationship between TFP growth and output growth: TFP growth can be translated into output 5

The derivation of the estimating equation in this paper is based on the work of Keller (1998) For more details, see Keller (1997, 1998).

r 2008 The Author

growth This result is important for calculating international output elasticities

of domestic R&D capital stocks later in the study

TFP is positively related to the number of differentiated intermediate goods used:

log Fit¼ biþ Z log zit; 8i;

where biis country i’s specific efficiency factor, and zitis the range of intermediate goods used in country i’s production Intermediate goods can be interpreted largely

to include not only physical production inputs but also ideas, know-how, and production knowledge

With international flows of goods, services, and labour, both domestic, zd

it, and foreign intermediate goods, zfit, can be employed for country i’s production.6 As R&D investment leads to the expansion of product varieties, thus by an appropriate choice of unit normalization, zd

it is identical to the cumulative stock of R&D expenditure, SDit, and zfit is captured by the foreign knowledge stock variable, SFit This means that TFP in country i may grow either as a result of domestic innovation or international technological spillovers from foreign countries

This study employs the Lichtenberg and van Pottelsberghe’s (1998) and van Pottelsberghe and Lichtenberg’s (2001) methods to construct three different R&D capital stocks (measured in level rather than in index) The first one, the imported-embodied foreign R&D capital stock, is constructed as:

SFitm¼X j6¼i

mijt

yjt

SDjt

where mijtis the value of imported goods and services of country i from country

j, and yjtis country j ’s GDP at time t This variable is equivalent to the trade-weighted foreign R&D capital stock computed by Coe and Helpman (1995).7 The focus of this study is to investigate the hypothesis that the international labour movement can serve as a channel for international technology diffusion To this end, this paper proposes two new measures of foreign R&D capital stock They are based on the assumption that flows of foreign workers can effectively transfer knowledge across borders The first new R&D capital stock, the foreign R&D capital stock embodied in the inward labour movement, is calculated by the following:

SFitg¼X

j 6¼i

gijt

njt

SDjt

6 In reality, domestically produced intermediate goods and foreign produced intermediate goods can be similar However, in this paper, for simplicity, they are assumed to be two disjointed sets so that both of them can be utilized for a country’s production.

7

In Coe and Helpman (1995), the stock of foreign R&D capital is computed as

zitf¼ SF it ¼ P

j 6¼i

m ijt

m it  SD jt , where m it is total imports of country i at time t, and measured as

an index number (1985 5 1) However, this has been shown by Lichtenberg and van Pottelsberghe (1998) to lead to a misspecified regression equation In addition, the Coe and Helpman’s method is also challenged by Keller (1998) who claims that regressions using counterfactual (randomly created) international trade patterns produce even more positive R&D spillovers and explain more of the variation in productivity than if actual bilateral trade patterns are used.

‘ B R A I N D R A I N ’ O R ‘ B R A I N C I R C U L A T I O N ’ 621

r 2008 The Author

where gijtis the stock of country j’s citizens living in country i and njtis country j’s population at time t The reason why the stock of people is used rather than flows is that the stock is less volatile than flows In addition, people with embodied knowledge continue their learning process by maintaining their communication with people back home As a result, they continue to convey their knowledge to the country of destination for as long as they stay there

The second new foreign R&D capital stockis created to test the hypothesis that people living overseas can also be a channel for transferring knowledge back to their home country It is called the outward labour movement foreign R&D capital stock and is computed as follows:

SFitk¼X j6¼i

kijt

njt

SDjt

where kijtis the stock of country i’s citizens living in country j Through the on-the-job learning process in a host country, foreign workers will learn and contribute to the development of knowledge and technology of that country In addition, people tend to be tied to their homeland so if a number of them return home or maintain close and frequent contact with people at home, their obtained knowledge will, to some extent, contribute to productivity improve-ment in their home country

In order to examine the degree of international R&D spillovers on TFP where labour movement is considered as a significant conduit, this paper extends the original Coe and Helpman’s equation to the following:

Fit¼ SDit; SFm

it; SFl

it;mit

yit

;git

nit

;kit

nit

; Hit

; where SFl

it (l 5 g, k) denotes alternative foreign R&D capital stocks based on stocks of foreigners by country of origin, mit/yitis the ratio of imports of goods and services to GDP, git/nit is the ratio of total foreigners to domestic population; kit/nitis the fraction of population living and working overseas, and

Hitis the average number of years of schooling used as a proxy for the country’s stock of human capital The reason for adding human capital to this specification is to investigate the effect of foreign R&D capital stock on productivity when the domestic labour force becomes more educated (the higher

‘absorptive capacity’)8and the effect of education itself on productivity.9To this end, the foreign R&D capital stocks are interacted with marginal propensity to import, inward/outward migration intensity, and stock of human capital This is important for checking how the regression results reported in this paper are robust to the inclusion of other variables

8

The ‘absorptive capacity’ is defined by Benhabib and Spiegel (1994) and Bils and Klenow (2000).

9

According to Bils and Klenow (2000), if workers need human capital to use advanced technology then growth in human capital can help to improve technology.

r 2008 The Author

III Data Description

The annual data set on business sector activity for 19 OECD countries during 1980–1990 is taken from Coe and Helpman (1995) This data set includes TFP indices with 1985 5 1 (for every country) and domestic R&D capital stocks (see Coe and Helpman (1995) for a detailed description of the data sources) Average years of education of the labour force, by de la Fuente and Domenech (2001) for the period 1960–1990, are linearly interpolated from 5-yearly data and used as a proxy for human capital While GDP and population for each country come from the OECD National Account Database, bilateral import flows are from the OECD Trade Database National stocks of foreign population by country

of origin are sourced from a number of databases including the OECD International Migration Database, the International Labour Organization’s International Labour Migration Database, the Global Data Centre’s Database, the Council of Europe’s Database, as well as from national statistics offices’ databases of the 19 countries There are no complete time series of stocks of foreign population by country of origin for every country during the period 1980–1990 so data for this study are combined from different sources.10Missing values are estimated using a linear interpolation method Matrices of inward and outward migration shares are computed for every country for each year over the period 1980–1990 These weighting matrices are then used to calculate alternative foreign R&D capital stocks as described above in the text

IV Empirical Findings

The goal of this study is to estimate the long-run relationship between TFP and the domestic and foreign R&D capital stocks when foreign labour movement is considered as a channel for technological transmission The main econometric technique employed in this paper is a pooled cointegrating method in which the relationship between dependent variable and explanatory variables is estimated

in log level terms This method has an attractive econometric property It allows

us to test for international R&D spillovers in a panel of countries where every single country has a relatively small number of time-series observations

As discussed in Coe and Helpman (1995) and applied in many other TFP research studies, when estimating clearly trended variables in level, the estimated equations should reflect cointegration This means that there exists a long-run relationship between trended variables in these equations A stationary error term is a criterion for judging if an equation is cointegrating If the error term is not stationary, the regression may be spurious A common way to avoid the spurious regression problem is to estimate change specifications, rather than level specifications, by differencing data before running any regressions However, differencing has a disadvantage of removing all relevant information

10 There are some discrepancies in the way foreign population is counted in OECD countries Countries like Australia, Canada, and the USA calculate foreign population based on people who are foreign-born while other countries focus on those with foreign citizenship.

‘ B R A I N D R A I N ’ O R ‘ B R A I N C I R C U L A T I O N ’ 623

r 2008 The Author

about common trends shared by level variables and only rendering information

on the short-run relationship

This paper will first test whether the data series are nonstationary by performing unit root tests The test results based on Im et al (2003) for log levels

of TFP, domestic R&D capital stocks, different specifications of foreign R&D capital stocks, and their interaction terms with import ratio, inward/outward migration intensity, and human capital are given in Table 1 Im et al.’s group mean panel unit root test allows each member of the cross-section to have a different autoregressive root and different autocorrelation structures under the alternative hypothesis Its test statistic possesses an asymptotic normal distribution in small-sized panels A brief discussion of the test is provided in Appendix A

The study is then extended to different tests for cointegration based on those

of Pedroni (1999) Pedroni’s panel ADF test allows for considerable hetero-geneity in the panel The test statistics have standard normal distribution where significantly negative statistics indicate rejection of the null hypothesis of no cointegration (see Appendix B for a short discussion about the mechanics of this test).11The regression results are represented in Table 2

Table 1

Group mean panel unit root tests (annual data 1980–1990 for 19 countries – Im et al., 2003)

Variable t N;Ta p  b

Adjusted mean c

Adjusted variance d

Group mean statistic e Decision f

log F
2.566 1.421
2.046 1.708
1.768 I(1) log SD
2.165 2.053
1.784 1.979
1.179 I(1) log SF
0.207 2.105
1.318 1.457 4.011 I(1) log SFm
2.640 2.316
1.997 2.153
1.980 I(0) log SFg
2.301 1.842
2.025 1.947
0.860 I(1) log SFk
1.678 1.000
2.075 1.483 1.423 I(1)

m

y log SF m
1.528 1.053
2.062 1.492 1.903 I(1)

g log SF g 0.418 2.368
1.355 1.591 6.129 I(1)

k log SF k
0.600 1.053
1.448 1.292 3.251 I(1) log H
2.852 0.895
2.102 1.409
2.753 I(0) log H log SF g
2.011 1.421
2.064 1.723 0.176 I(1) log H log SF k
1.605 1.474
2.037 1.696 1.446 I(1)

Notes: log X is logarithm of X F is total factor productivity, SD is domestic R&D capital stock; SF is unweighted foreign R&D capital stock; SF m

is foreign R&D capital stock embodied in imports; SF g

is foreign R&D capital stock embodied in inward foreign population; SF k

is foreign R&D capital stock embodied in outward foreign population; m/y is the ratio of imports of goods and services to GDP; g/n is the ratio of total foreigners to domestic population; k/n is the fraction of population working and living overseas; and H is the average number of years of education.

a

Cross-sectional average of individual Dickey-Fuller  t N;T statistics.

b

Cross-sectional average of individual number of lagged differenced terms in ADF(p i ) regression.

c

Cross-sectional average of E ½t i;T ðp i ; y i Þ.

d

Cross-sectional average of Var ½t i;T ðp i ; y i Þ.

e

The test statistic W  t which has standard normal distribution.

f

Test of the null hypothesis of common unit autoregressive root at 5% level (the critical value is
1.96).

11

It is true that when performed separately on the time series for each country, given that each country has only 11 annual observations, the power of the tests is really low The panel

r 2008 The Author

Table 2

Total factor productivity estimation results (pooled data 1980–1990 for 19 countries, 209 observations – in level)

(0.030)

(0.135)

(0.270)

Cointegration tests

Notes: The dependent variable is log F (log of total factor productivity, indexed as 1985 5 1) All equations include unreported country-specific constants Time dummies are omitted

is foreign R&D capital stock

is foreign R&D capital stock embodied in outward foreign population; m/y is the ratio of imports of goods and services to GDP; g/n is the ratio of total foreigners to domestic population; k/n is the fraction of population living overseas; H is the average number of years of education; G7 is dummy variable equal to 1 for the seven major countries and equal to 0 for the other twelve countries.

n

indicates that parameters are statistically significant at the 5% probability level.

a

Pedroni (1999)’s Panel ADF statistic allows dynamics and cointegrating vector to vary across individuals.

b

As shown in Table 1, a unit root test on the pooled data indicates that most

of the variables are nonstationary The null hypothesis that the panel has a unit root can only be rejected for log SFmand log H According to Edmond (2001), regressions using those variables do not fulfil a necessary condition for cointegration and should be treated with some doubt.12

Table 2 describes all pooled least squares regressions following by standardized test statistics of Pedroni’s panel cointegration tests at the bottom This paper concentrates first on equations that relate the log of TFP to the logs

of domestic and alternative foreign R&D capital stocks and then extends the equations to consider the log of human capital All of the equations include unreported country-specific constants to allow for missing country-specific fixed factors such as the influence of institutional variables In every equation, the impact of domestic R&D is allowed to differ between the seven largest countries and the other 12 countries by including an interaction term between domestic R&D capital stock and a dummy variable, G7, which takes the value 1 for the seven largest economies Except for equations (3), (5), and (6), all the other 12 models are confirmed to be cointegrating by cointegration tests In each cointegrating regression, the estimated elasticity of the domestic R&D capital stock is positive and significant.13 The test results reveal that there is a significantly different effect of domestic R&D capital stocks for the G7.14

As shown in Table 2, regressions (1)–(8) show the estimated productivity elasticities of domestic R&D and each of the foreign R&D capital stock (or its corresponding interaction term with import intensity, fraction of population working overseas, fraction of foreign population, or human capital) incorpo-rated into one of the three different channels of technological diffusion With regard to the impact of outside R&D embodied in the movement of workers across borders, equations (1) and (2) show that there may be significant international R&D spillovers and the migration of workers may induce

unit root test provided by Im et al (2003) and the panel cointegration test provided by Pedroni (1999) help overcome this disadvantage by deriving the limiting distribution The power of these tests increases dramatically as the cross-sectional dimension rises For example, in Pedroni’s panel ADF tests, as long as the time dimension is greater than five, the test statistic is shown to

be distributed as standard normal and the small sample performance of the test is reasonably satisfactory.

12

There may not be any shared trends among variables when I(0) variables are included in equations with existing I(1) variables In addition, Pedroni (1999) does not specify cointegration tests for this type of regression equation For these reasons, this paper opts not to consider estimation equations including those I(0) variables.

13

According to Coe and Helpman (1995), OLS estimates of a cointegrating equation are

‘super consistent’ because they converge to true parameter values much faster than the case where variables are stationary when the number of observations increases and their distribution does not necessarily follow standard t-distribution Because the specific distribution associated with those estimates is unknown, this study follows a large number of research works in the existing literature (e.g., van Pottelsberghe and Lichtenberg, 2001; Park, 2004) using the standard t-distribution to draw inference about their significance as a limiting case.

14

In fact, this paper also tries to include a time trend in every regression equation However, the results are not supportive because the coefficients of most variables, including that of domestic R&D capital stock, are negative (incompatible with reality) Therefore, the empirical estimation is carried out without any time trend.

r 2008 The Author

substantial technology transfers In equation (1), the elasticity of the foreign R&D capital stock embodied in inward labour movement is positive and highly significant That is, the hypothesis of knowledge gain from attracting highly skilled workers can be confirmed by the estimates By allowing foreign workers

to immigrate, host countries seem to be able to enhance their stock of knowledge, thereby increasing their productivity In equation (2), a positive and significant estimate for the elasticity of foreign R&D capital stock embodied in outward labour movement is obtained This is evidence that people working overseas can make contributions to the enhancement of productivity back home through knowledge transfer Contrary to the conventional assumption of ‘brain drain’ associated with emigration, this result is consistent with the emerging theory of ‘brain gain’ within that ‘brain drain’ literature such as that reported by Mountford (1997), Vidal (1998), and Beine et al (2001)

In equation (3), the unweighted foreign R&D capital stock is included The main purpose of this regression is to check whether unweighted foreign R&D capital stocks make any difference in explaining the variation in productivity across countries compared with the migration-weighted patterns.15 It can be seen that the inclusion of this variable makes domestic R&D capital stock insignificant, which is economically implausible In addition, the cointegration test cannot reject the null hypothesis of no cointegration which implies a potential spurious regression in the equation Together, these results refute the explanatory power of the unweighted measure of foreign R&D and support the measures which combine both migration and R&D in affecting TFP growth as found in equations (1) and (2)

Equation (4) is Coe and Helpman’s preferred model for this study’s data sample where import-weighted foreign R&D capital stocks are expressed in levels and calculated using Lichtenberg and van Pottelsberghe’s (1998) method

As discussed in their paper, the equation is suggestive of the role of trade in the international transmission of R&D benefits The estimated coefficient for the interaction between import ratio and import-weighted R&D capital stock is slightly higher than the original one This is probably because this study uses time-varying import ratios while those of Coe and Helpman are static.16 Equations (1) and (2) are modified to become equations (5) and (6) Although each foreign knowledge stock in these equations consists of migration-weighted foreign R&D capital stocks, these weights may not perfectly capture the level

of migration, either inward or outward It might be expected that when two countries have the same composition of migration and face the same composition of R&D capital stocks among economic partners, the country that has more inward and outward migration relative to its population may benefit more from foreign R&D.17 For these reasons, equations (5) and (6)

15 The author is grateful to an anonymous referee for this useful comment.

16

To achieve sustainable development, it is necessary that the import-GDP ratios are not high An investigation into the data used in this paper reveals that these ratios are actually mean reverting (the highest value is 0.578 by Belgium in 1984).

17 I would like to thank the same referee for this interesting comment.

‘ B R A I N D R A I N ’ O R ‘ B R A I N C I R C U L A T I O N ’ 627

r 2008 The Author