Revisiting Linkages between Openness, Education and Economic Growth: System GMM

Approach
Author(s): Emiko Fukase
Source: Journal of Economic Integration , March 2010, Vol. 25, No. 1 (March 2010), pp.
193-222
Published by: Center for Economic Integration, Sejong University

Stable URL: https://www.jstor.org/stable/23000971

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Center for Economic Integration, Sejong University is collaborating with JSTOR to digitize,
preserve and extend access to Journal of Economic Integration

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
Journal of Economic Integration
25(1), March 2010; 193-222

Revisiting Linkages between Openness, Education

and Economic Growth: System GMM Approach

Emiko Fukase

The Graduate Center, the City University of New York

Abstract

For 106 countries over 1969-2004, this paper revisits the relationship between
openness, education and economic growth using the System Generalized Method
of Moment (GMM) approach. Trade and growth are usually positively correlated
and capital formation appears to be an important channel through which trade
brings growth. There generally exists a positive relationship between FDI inflow
and economic growth for a subset of non-OECD countries. The System GMM
estimator improves substantially the estimate of the impact of education on growth
relative to the models which focus on within-country changes in education, adding
information on cross-country variation in education levels.

• JEL Classification: F13, F21, 120, 024, 040

• Key Words: growth, system GMM, openness, trade, education, foreign direct
investment

I. Introduction

Many cross-country studies from the 1990s find a strong positive relationship
between outward orientation and economic growth (e.g. Dollar, 1992; Sachs and
Warner, 1995). These studies typically regress per capita growth rate on various
openness level variables. There also exist a number of studies to assess the linkages
between economic growth and foreign direct investment (FDI) and their results are

* Corresponding address: Emiko Fukase: 725 West 172nd Street, #31, New York, NY 10032, U.S.A. Tel:
212-928-1095, Fax: 212-593-5115, e-mail: [email protected].
©2010-Center for International Economics, Sejong Institution, Sejong University, All Rights Reserved.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
194 Emiko Fukase

highly heterogeneous.1 Utilizing data on FDI flows from industrial countries to 69

developing countries, Borensztein, de Gregorio and Lee (1998) find that FDI
contributes to economic growth only when the host country has a minimum
threshold stock of human capital.
Rodriguez and Rodrik (2001) re-investigate critically the conclusion of previous
cross-country studies that openness is associated with higher rates of growth. They
argue that a variety of measures of openness used in these studies are proxies for
other policy or institutional variables and that the earlier results that openness
enhances growth are not strong. Using the various authors' actual data sets in
undertaking tests of the robustness, they show that many openness variables lose
their statistical significance when these other policy and institutional variables are
included in the regressions, different data are used, or different weights are utilized.
Rodriguez and Rodrik warn that a liberal trade policy cannot guarantee higher
growth.
Some economists argue that an attempt to isolate the impact of trade on growth
in empirical analyses is futile, and emphasize the importance of combining liberal
trade policies with other good policies and institutional development (Baldwin,
2003; Winters, 2004). Their recommendations include sound macroeconomic and
fiscal policies, corruption-free institutions and an environment that encourages
physical and human capital investments. Winters (2004) points out that many of
these policies and institutions respond positively to trade liberalization (e.g.
through increasing incentives to develop better institutions, promoting transparency
and increasing the return to investments). Among these factors, Winters suggests
education as "possibly top of any a priori list" (F16) as education imparts
transferable skills; increases productivity which may lead to higher output;
promotes health; creates social and political capital; and facilitates the absorption of
new technologies.
While there is solid microeconomic evidence of a positive return of schooling on
individual wages, macroeconomic studies which focus on changes in education
frequently find an insignificant or even a negatively significant relationship
between education and economic growth (e.g., Benhabib and Spiegel, 1994;
Krueger and Lindahl, 2001; Pritchett, 2001). For instance, Pritchett (2001) calls

'in his extensive review of home- and host-country effects of FDI, Lipsey (2004) concludes that one
cannot say that there are "universal effects" of FDI on growth from the studies who find positive effects
in some periods or among some groups of countries and suggests that there are circumstances, periods,
and countries where FDI appears to have little relation to growth (p.369).

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 195

this a "micro-macro paradox of negative externalities" (p.376).2 The linkages

between openness, education and growth are not settled issues.
For 106 countries over the period 1969-2004, this paper revisits the relationship
between openness,3 education and economic growth, using a series of panel data
techniques. In particular, the paper highlights how the results obtained by the
System Generalized Method of Moment (GMM) approach (Arellano and Bover,
1995; Blundell and Bond, 1998), which solves difference and level equations as a
system, bring new insights relative to the previous literature. Section II reviews
summary statistics on economic growth, education and openness by growth
quartiles. Section III presents a series of regressions with varying specifications.
When it is not possible to isolate the partial effect of openness on growth from
other related policy and economic variables, e.g., when openness variables lose
statistical significance with the inclusion of control variables in my regressions,
potential relationships between these variables are discussed. It is also
demonstrated how the System GMM estimator may improve the estimate of the
impact of education relative to the models which focus on within-country changes
adding information on cross-country variation in education levels. Section IV
revisits the findings of Borensztein et al. (1998), using the actual dataset used in
that study. Section V summarizes my conclusions.

II. Growth, Education, and Openness - Overview

Table 1 shows the average annual growth rates of real per-capita GDP for 106
countries by growth performance for the period 1969 to 2004.
The first, second, third and fourth columns present the highest, high-middle,
low-middle, and the lowest quartile countries classified by their average annual
growth rates during the same period. The highest quartile countries include many
fast-growing East-Asian and South-East Asian economies as well as other
transitional economies. Among these, one sub-Saharan African country, Botswana,
was the second fastest growing country during the same period. Many OECD
countries belong to the high-middle growth quartile. With an exception of Papua
New Guinea, the lowest growing quartile is comprised of Sub-Saharan African,
Middle-Eastern, Latin American and Caribbean countries.

2Pritchett hypothesizes that in some countries education may have created "better-educated pirates" who
engage in "privately remunerative but socially wasteful or counterproductive activities" (p.387).
3Openness variables are measured by the volumes of trade and FDI as shares of GDP.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Emiko Fukase

Table 1. Average Annual Real Per Capita Growth Rates for the Periods 1969-2004

High-middle Low-middle
Highest Quartile Lowest Quartile
Quartile Quartile
Growth Growth Growth Growth
Country Country Country Country
Rate (%) Rate (%) Rate (%) Rate (%)
China 7.40 Hungary 2.46 Mexico 1.64 Bahrain12 0.59
Botswana 6.80 Spain 2.45 Denmark 1.60 Argentina 0.57
Korea 5.80 Uganda4 2.29 Nepal 1.55 Senegal 0.45

Singapore 5.31 Belgium 2.27 Panama 1.53 Jamaica 0.44

Malta 5.28 Brazil 2.27 Paraguay 1.51 Peru 0.41

5.21 Pakistan 2.26 Barbados9 1.51

Papua New 0.41
Cyprus1
Guinea

Hong Kong 4.82 Greece 2.23 Tanzania10 1.47 El Salvador 0.34

Syrian
Thailand 4.68 Arab 2.20 Ecuador 1.44 Iran13 0.31

Republic
Mauritius2 4.31 Italy 2.19 Yemen11 1.31 Burundi 0.31
United South
Ireland 4.23 2.14 Algeria 1.31 0.29
Kingdom Africa

Indonesia 4.17
Mozambiq 2.10 1.30 Bolivia 0.27
ue5
Uruguay

Malaysia 4.08 Turkey 2.10 Cameroon 1.29 Ghana 0.20

Lesotho 3.94 Jordan6 2.09 Bangladesh 1.27 Zimbabwe14 0.18
New
Poland3 3.47 Israel 2.04 1.22 Togo -0.05
Zealand
United Sierra
Tunisia 3.12 2.03 Philippines 1.20 -0.37
States Leone
Dominican Republic Guinea
3.10 France 2.01 1.19 -0.42
Republic of Congo Bissau15
Venezuela,
Sri Lanka 3.06 Canada 1.97 Mali 1.17 -0.65
RB
Central

Portugal 3.05 Swaziland7 1.95 Switzerland 1.08 African -1.06

Rep.
Egypt 3.04 Germany8 1.94 Malawi 0.98 Zambia -1.12

Norway 2.95 Costa Rica 1.87 Kenya 0.94 Haiti -1.21

Nether
India 2.89 1.85 Guatemala 0.89 Niger -1.55
lands

(Continued)

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth

Table 1. Average Annual Real Per Capita Growth Rates for the Periods 1969-2004 (Continued)

High-middle Low-middle
Highest Quartile Lowest Quartile
Quartile Quartile
Iceland 2.88 Sweden 1.84 Guyana 0.84 Nicaragua -1.63
Chile 2.76 Australia 1.82 Rwanda 0.83 Liberia -2.80
United
Japan 2.66 Sudan 1.74 Benin 0.81 Arab -3.09
Emirates16
Trinidad
Finland 2.61 and 1.74 Gambia 0.72
Dem. Rep. -3.60
of Congo
Tobago
Austria 2.47 Colombia 1.73 Mauritania 0.68 Kuwait17 -3.83

Fiji 1.72 Honduras 0.67

Source: The World Development Indicators, the World Bank

Notes: '1976-2004; 21981-2004; 31991-2004; 41983-2004; 51981-2004; 61976-2004; 71971-2004; 81972
2004; '1969-2003; lo1989-2004; "1991-2004; ,21981-2003; l31975-2004; l41969-2002; 151971
2004; l61974-2002; "1963-1989 and 1996-2003.

Table 2 demonstrates summary statistics for GDP growth rate, GDP per capita,
total education years, trade, and FDI inflow as the averages over the period 1969 to
2004 by growth quartiles.
There is substantial variation in annual real GDP per capita growth rates among
these four groups of countries with averages over the period of 4.0 per cent, 2.0 per

Table 2. Summary Statistics by Growth Quartiles for the Period 1969-2004

Highest High-middle Low-middle Lowest

Obs. Mean Obs. Mean Obs. Mean Obs. Mean
4.0 2.0 1.1 -0.66
GDP Growth (%) 175 184 182 178
(2.6) (2.1) (2.4) (4.6)
8136.7 9449.5 3820.6 3198.8
GDP per capita ($) 198 206 206 197
(9606.9) (8592.2) (7371.4) (7418.8)
Total Education 6.1 6.4 4.5 3.4
189 176 168
Years1 (2.2) (3.0) (3.0) (1.9)
91.7 62.9 67.9 61.1
Trade (% of GDP)2 175 185 179 170
(64.6) (37.8) (37.5) (32.2)
FDI Inflow 2.3 1.8 1.6 1.9
181 187 181
(% of GDP) (2.9) (2.1) (2.4) (3.6)
Sources: Author's calculation based on the World Development Indicators; Barro and Lee (2000);
UNCTAD.

Notes: The standard deviations are between parentheses. '"Total Education Years" reflect average
schooling years for the population over age 15 (Barro and Lee, 2000). 2Trade is measured by
adding imports and exports and comparing the sum to the GDP.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
198 Emiko Fukase

cent, 1.1 per cent and -0.66 per cent for the highest, high-middle, low-middle and
lowest growth economies respectively. The average real GDP per capita are
recorded as $8,136.7, $9,449.5, $3,820.6, and $3,198.8 for these same groups.
GDP per capita of high-middle growth countries on average is higher than that of
the highest growth economies partly because the former group includes many high
income OECD economies. Within the subgroup of fast growers (the highest and
high-middle quartiles combined), lower income countries appear on average to
grow faster than higher income countries implying that the economies are
converging even unconditionally. Unfortunately, within the sub-group of slow
growers (low-middle and lowest quartiles), the lowest growing group records the
lowest average GDP per capita implying that these economies are falling further
behind.

During the same period, the high-middle quartile economies record the highest
schooling attainment with 6.4 total years of education which is slightly higher than
that of the highest quartile countries (6.1 years). The levels of education
attainments of low-middle and lowest quartile economies are clearly low relative to
the upper-half group, registering 4.5 years and 3.4 years respectively. In terms of
openness variables, trade relative to GDP reaches levels of 91.7 per cent, 62.9 per
cent, 67.9 per cent, and 61.1 per cent for the highest, high-middle, low-middle, and
lowest quartile economies respectively. FDI inflow relative to GDP accounts for

Figure 1. Time Trends of Average Years of Schooling by Four Growth Categories

0 J 1 . — ,
1970 1975 1980 1985 1990 1995 2000

Source: Author's calculation based on Barro and Lee (2000)'

Note: The data for 2000 are projections in Barro and Lee's d

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 199

2.3 per cent, 1.8 per cent, 1.6 per cent and 1.9 per cent for the same groups of
countries.

The time trends of educational achievement by growth performance are shown

in Figure 1, which plots the average years of schooling by growth category over
the 1970-2000 periods.
During the past three decades, the world as a whole has achieved a substantial
improvement in educational attainment as the countries in all four growth
categories show steady progress. The changes in educational attainment appear to
be similar across growth categories. The only noticeable exception is that the
schooling years of the countries classified into the highest growth quartile rose
slightly faster than the other groups converging towards the highest level attained
by the high-middle quartile of countries. In contrast, Figure 1 reveals a clear
positive relationship between the levels of educational attainment and economic
growth. Throughout the period, the total education years for the highest and high
middle quartile economies are higher relative to the lower groups, those attained by
low-middle growing economies are lower than the upper half growing economies,
and those for the lowest growing economies lag further behind. These
characteristics of the actual data are consistent with the previous studies which find
no or even negative association between economic growth and education growth
{e.g. Benhabib and Spiegel, 1994; Pritchett, 2001). An econometric technique
which exploits information on the levels of education is clearly needed to assess
properly the impacts of education on growth.

III. Regression Analyses

A. Model Specification

In this section, I outline 5-year period growth regressions using a series of

models: pooled Ordinary Least Squares (OLS), fixed effects, first-differenced
GMM, and System GMM. My panel data consist of observations from 106
countries during the period 1969 to 2004 at the five year intervals.41 first specify a
growth level equation of the type considered typically in the literature (see, for
instance, Caselli, Esquivel, and Lefort, Section 5, 1996):

4This time Frame is chosen since the period ending at the year 2004 is the latest five year interval for
which the initial education data are available and the 1970s are the oldest decade when the FDI data
become available for a sufficient number of countries.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
200 Emiko Fukase

yu-yiM = (iyut-i) + x'uy + £ + % + vit (i)

where yit denotes the logarithm of real per-capita GDP in country i in year t\ xit is a
vector of explanatory variable; § is the time specific effect which captures global
shocks; 7, is the country-specific effect; and vit is the error term. The education
variable is measured at the beginning of each five year period and the other
variables are measured as averages from t to t-4. Since my analysis is focused on
five year growth rates, lis 5 in my regression. Equation (1) can be re-written as a
dynamic model in the level of per capita GDP by adding yi(t.t) to both sides:

yti = ayi(t-o + x'ity+ § + Tjj + vit (1')

where a = (1+/3).
In the presence of the country specific effect //„ it is well known that the OLS
estimate of the coefficient on the lagged dependent variable a is likely to be biased
upward since the lagged dependent variable is positively correlated with 77, (see for
instance Blundell and Bond, 1998). One approach to address the country specific
effects is the fixed effects estimator. For this model, equation (I1) is transformed by
taking the deviation from the time series mean of each variable for each country,
and then the transformed equation is estimated by OLS. In this transformation
process, the country specific effects 77, can be removed. A disadvantage of using
the fixed effects model is that it uses only the variation within countries and the
cross-sectional variation is discarded. In addition, equation (1') contains a lagged
endogenous variable, namely the income term. Thus, with a small number of time
series periods, the model provides biased and inconsistent estimates even if data
from a large number of countries are considered. In contrast to the OLS estimate,
the fixed effects estimate of the coefficient on the lagged dependent variable a is
likely to be biased downward (Arellano and Bond, 1991).
Arellano and Bond (1991) suggest an alternative estimation technique that
addresses the presence of the lagged endogenous variable and permits a certain
degree of endogeneity in the other explanatory variables. Their GMM estimator
first-differences Equation (I1) in order to eliminate the country-specific effect, and
then uses all possible lagged levels as instruments. Arellano and Bond's Difference
Equation is:

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 201

Ayit = aAyi(H + Ax'it y+ A%, + Avit (2)

Since and Avit are correlated, the OLS estimate of Equation (2) is
inconsistent. Assuming that the error terms are serially uncorrelated in Equation
(1), the level of>v, lagged two periods or more can be used as valid instruments for
Equation (2). This is because Ayi(t_2,) and earlier values are correlated with Ayi(t.0,
but not with Avit. Assuming that the xit are predetermined in the sense that xit and
Vjt are uncorrelated, but xit may be correlated with and earlier errors, xit lagged
one period or more are also used as valid instruments. Thus, the relevant moment
conditions are:

E[yi(t.s)AvjJ = 0 for s>2i; t = 3,....T

E[xi(t_s)AvjJ = 0 for s> i; t = 3....T.
Using Arellano and Bond (1991)'s first-differenced GMM estimator, Dollar and
Kraay (2004) focus on the effects of within-country change in trade policy on
subsequent growth. They point out that the methodology employed by the earlier
literature which relates growth to cross-country differences in trade volumes is not
suitable to evaluate the effects of trade policy since a significant share of the
variation in trade across countries reflects countries' geographical characteristics.
By taking the difference, they argue that their results are not driven by geography
and other time-invariant country specific characteristics. Their first-differenced
GMM estimators reveal a strong positive relationship between change in trade and
growth.
Blundell and Bond (1998) point out that the first-differenced GMM estimators
are likely to perform poorly when the time series are persistent and the number of
time periods is small. This is because lagged levels of the series provide only weak
instruments for the differenced equations. Another shortcoming of using the
difference estimator is that the process of differencing to remove the country
specific effect also eliminates information on the cross-country variation in levels.
The System GMM estimator proposed by Arellano and Bover (1995) and Blundell
and Bond (1998) combines the standard set of moment conditions in first
differences with lagged levels as instruments, with an additional set of moment
conditions derived from the equation in levels. The availability of additional
moment conditions depends on assumptions made about the correlation between xit
and the country-specific effect 77,. Following Blundell and Bond (1998), it is
assumed that the difference of xit is uncorrelated with the individual effects
although xit and 77, are allowed to be correlated.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
202 Emiko Fukase

Thus, the additional moment conditions for the equation in levels are:
E[Ayi(i_ 0UiJ = 0 where uit = 77, + vit
E[AxitUjJ = 0.
Carkovic and Levine (2005) use the System GMM estimator to re-examine the
relationship between FDI and growth, but find no robust evidence supporting the
claim that FDI accelerates growth. Felbermayr (1995) employs the System GMM
estimator to revisit the relationship between trade and growth and finds a robust
and positive relationship between these variables.

B. Regression Results

Table 3 demonstrates a series of regression results using pooled OLS, fixed

effects, Arellano and Bond (1991) differenced GMM, and System GMM
estimators.5

Table 3. Regression Results

Variables Pooled OLS Fixed effects Differenced GMM System GMM

0.98"* 0.77*** 0.69*** 0.94***
Log of Lagged GDP
(0.013) (0.036) (0.050) (0.025)
1.85** 0.30 -3.07 5.49***
Total Education Years
(0.73) (1.15) (2.32) (1.41)
-0.0096*** -0.0069"* -0.0038*** -0.0081***
Inflation
(0.00080) (0.00072) (0.00080) (0.0013)
0.018 0.15"* 0.031 0.13*"
Trade as % of GDP
(0.020) (0.043) (0.098) (0.043)
1.08*** 1.25*** 0.92" 0.70
FDI as % of GDP
(0.38) (0.33) (0.43) (0.58)
m2 0.876 0.930

Hansen Test 0.357 0.141

Number of Observations 607 607 504 607

Number of Countries 102 102 99 102

Number of Instruments 96 92

Notes: The dependent variable is the log of GDP. Robust standard errors are in parentheses. For the
specification tests, p-values are reported. Time dummies are included in all the regressions (not
reported).
*, ** and *** indicate that the coefficients are significant at the 10, 5, and 1 per cent level,
respectively. Column 3 reports the results of two-step Arellano-Bond (1991) difference GMM.
Column 4 shows the results of two-step Blundell and Bond (1998) system-GMM estimator with
Windmeijer finite-sample correction.

5The Data Appendix provides details on data sources.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 203

The coefficients on the lagged dependent variable in all the models are found to
have a value of less than one and to be statistically significant at the 1 per cent
level, providing strong evidence of conditional convergence.6
The coefficient on education7 estimated by OLS is found to be positively
significant confirming the positive association between the initial level of human
capital and subsequent economic growth (Barro, 1991). In contrast, using fixed
effects and first-differenced GMM estimators, the coefficients on education lose
statistical significance perhaps because these within country models discard
information on cross-country variations in education levels. When the System
GMM estimator is employed, the coefficient on education reverts to become
significantly positive suggesting that the additional moment conditions derived
from the level equation are highly informative.
The coefficients on trade are found to be statistically significant and positive
using the fixed effects and System GMM models. The coefficients on FDI turn out
to be significantly positive using all methods except the System GMM model. The
coefficients on inflation show a significant and negative relationship in all the
models.

The validity of instruments for first-differenced and System GMM estimator can
be evaluated by a set of specification test (Arellano and Bond, 1991). The

6As expected, the magnitude of the coefficient for the lagged dependent variable estimated by OLS is
higher than that estimated using the fixed effects method and it is likely that the OLS estimate and fixed
effects estimates give upwards- and downwards-biased estimates respectively. The differenced GMM
estimate for the lagged dependent variable is found to be below the fixed effects estimate. This
downward bias in the differenced GMM estimator is consistent with the finite sample bias expected in
the case of highly persistent series (Blundell and Bond, 1998). The System GMM estimator of the
lagged dependent variable appears to provide the most reasonable results with the magnitude of
coefficient lower than the OLS and higher than the fixed effects estimates.
'The average of male and female schooling years for the population over age 15 is used as a proxy of
education. An alternative model of disaggregating education by gender was also considered. However,
the latter specification reproduced somewhat "puzzling" results in the previous studies. The results of
OLS model, a significantly positive coefficient on male education and a negative one on female
education, are consistent with the literature (Barro and Sala-i-Martin, 1995; Wacziarg, 2001). The results
of fixed effects and differenced GMM estimators reveal the reversal of the signs of male and female
eduction although they are not statistically significant or are significant only marginally. These changes
of signs are also consistent with the earlier works which use the first-differenced GMM estimators
(Caselli et al., 1996). With the System GMM estimates, the coefficients on male and female education
years both show positive signs although only the coefficient on female education is significant at the five
per cent level. A potential explanation of these "unstable" results is that the collinearity between male
and female schooling variables is inflating the standard errors. The Variance Inflation Factors (VIFs) for
female and male schooling variables are 15.7 and 13.0 respectively suggesting a high degree of
collinearity between two variables. Wetherill (1986) suggests that, as a rule of thumb, a VIF higher than
10 is of concern. Thus, the specification of including separate male and female education years was
dropped.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
204 Emiko Fukase

application of the Hansen test of over-identifying restrictions provides no ground to

reject the validity of the instruments, ml and m2 tests show a first order serial
correlation in the first-differenced equation, but detect no evidence of second order
serial correlation. Overall, there are several advantages in the System GMM
estimator since it controls for time-invariant country specific effects; deals with the
endogeneity problem of lagged dependent variable; permits a certain degree of
endogeneity in the other regressors, and optimally combines information on cross
country variation in levels with that on within-country variation in changes.

C. Models with Alternative Specifications

In order to test the sensitivity of the results to the conditioning variables

included, Table 4A gives the results of the models estimated by System GMM with
a series of different specifications.8
The first six columns report the regression results when the openness variables
are included separately When only trade and FDI variables are included along with
income terms, the coefficients on trade and FDI are positively significant at the one
per cent level in regression (1) and in regression (2) respectively. With the latter
specifications, the coefficients on the lagged dependent variable turn out to be
greater than one implying that these economies would not be converging
unconditionally. This result is consistent with Barro's seminal paper (1991) which
finds that countries' incomes converge only when controlling for initial human
capital and some other variables. The next two regressions introduce education and
inflation as control variables: the coefficients of trade and FDI remain positively
significant in regression (3) and in regression (4) at the one per cent level. The next
two regressions include gross capital formation as a control variable. As expected,
the coefficients on capital are found to be highly significant at the one per cent
level. However, adding the capital variable, the coefficients on trade and FDI lose
statistical significance in regression (5) and regression (6) respectively.
The final three columns show the results when trade and FDI variables are

simultaneously included in the regressions. The results of regression (7)9 reveal

that, including both openness variables, only the coefficient on trade demonstrates
a statistically significant positive relationship. Controlling for government
consumption expenditure as per cent of GDP in regression (8), the results of other

sAlthough the results of System GMM models are highlighted in this paper, the outcomes of the pooled
OLS and fixed effect models are reported in Appendix Table A for the purpose of comparison.
'This model is the same as that in Column 4 in Table 3.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth

/-s 00
-a os

oo * ^ m p $: * fz? © in £2
oo «n £l
V 3 *<N — p © O ort ^ ^
m O w
cn T—< O O d. -a
oo

o2^^wOO 1 W '—' w ©
. vo
2 § s i
w
3 «

CO c9
o £ 11
fc -S
c —
.2 (u o 8
co t3
2 a
J
*

P *
g§
00 r 00
&3
*
• cn * in OO m cn
"V (N
so
r-;
t^ o m 00 OS SO cn
00
o o
^ a.
OS p in o o p o o O o
00
0) <l>
o w a
2^
s
^ /
w i
o in o o
o ^
OT
i
c w
*~'. rti
= o
a ^
■s tt 1.S
o Q
"S S, CX O
•s a p o
*
m T3 d
m
(N
*
cn
o S? o
cn r 0 c
CO

* § o
p
o
o
in
os o
V©
•S s ra
-O S
co o
r

o 0.13 o © 2 — U M -H
©
« 8 M V, S
CO _

■- ^3 >5u lis
E^ s.
» 00 * ^ *_^ © o
1-
3 0) a g "
M .3 U
vo cn
V)
Os P 2® \o
o ® o o

<r""5 P. rri ^11.

S £ u

■Ifl g 2
; vo o ^ o c «
S rt £ o oo
^ m (N r- o o <n in
MD 00 CN
-H O D, 3> 3
os P ^
p
o o o o CN co ^
P O © w ^
© S Tt
o o ' w
•S -S ^
s3§
* /—s % ^t" | a
. J- <u
* r- t O O OO * ^J
(N
cn cn o 1—1 bb
O

*£ s s - g !
<N t} °0 CN -H O
^ o © © <L>
C3
b o »i5
° *3
a>
3 o
os 2 »m o r
m oo oo cn
9- 6S Z
S ©
|SS
O ^f
OS . r*- •
O os "2 w
o a)
o
© o
d S ° 3 2
j=) «a
c S|
* r
cn
* in
cn in
cn r
n' co
^ Im:^ c^ o

os cn o
e I #S
cn o cn
<N r-;
^ ©
CL
o o a
o
CO

C
11 •

tJ) *Sm
Q ^ c
O
11 § jy
i CN o
cn cn <m o m vo m
O ooo ^ B "
'I
cn o O oo
CO
P V?
ox
© .3 S O *o
= o w
C/3 ^ 'S O 45 (U
"S O
aS
g bp X
o.
Q
C .§ ^ £ .S w
I«
o '•m co -y
co 3 C
£ O i^i "E
Cv
a> 3

5§8
<M
Oh a PL, C .2
Q Q
o 53 -a
O
<M
.o
"cd O
v° o
Ph
X5 g
c .5
(U *
?^f
O QD <u
o Cm D-I
o =3
Td
O <D
-o "O E -S I
OD N°
£ o - c
*3 3 I]
j3 W ox
o §■ SJ § ^ CO M
-o
<D
_o
c
u fS: ft> w c
CO S-i .^h

<u
OX)
§P QJ £ co
03 <U

8P
53
S3 "S s o
Table4A.RegrsionResultwihAlternaiveSpcifatons-FulSample i—i 3 a
Ph a

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Emiko Fukase

oo
1 r~. 2 .£a sm
«■>
- §^o ®
"8 o fn.
-$■ vi p; _ s 1 % 9 2
lt) on
r- r

© s ^ w « o =? © 2 s

(n
i f?°o ^ m ^o

*8 Si vo n f: o (n
g s » ^ on ^ m on
^ 5 r~~ r~~
© o ,^d © ~ s ® s o ©

rt :
*
r *
*no *
no no ^ .
(n
O
V 00 oo
o
ir) o
on On t [ .
oo <n o
p p p ^ rn ^ ^
d o
d <n © ©
©
d d^

p: rv,2 1 ^t ^8 ° vrs
sj o^ t- pi £ o (n in ^t" (n n >n r

S 3 "
o

^11.
on .
^ r-; NO
© £ cn

*
*
m *
on ** o
m r
<n
v no r o o
m
*00 <n On ^ i/-) t^
o
on t n £
/-"■n

in o
© o p p d
cn
d <rl ° ^
in
d d^ d

m
p o 2: • <n ° 00 ^
oo g
00 p ^ g <=> m r oo rt £2
no in ^
m m
r- in
(n O ^ <n ^
do
w © s i

*
t—< ^ £ : (o ^ 00 ^
g *>^ © o oo £ on m 9 j (n in
^ in
(n O
o in £7} t— i/-)
p, oo d d

m — oo
on
(n
o <n tt ^ on on
d 2 2^
o o

Oh
q
* oo
a r-- r- „
£- cn g oo as
ri™ s
h
o o 2 2°^^
cu c
a3
O o

£ o
a
.. a
4) 4)
Ph c Oh
q .o q o
£ 5
a a 4> s
td uh co p
<+h
o
o —

£ °
o
"3 o u
00 *6
w
\®
cn £ <+h
jd o ■& o o
0) c qh
t3 u c 4> 4> 4)
4>
00
c3
2 4> x) -o x>
*-+—» 4)
00 vh
1)
as t3
s £ b S
gp 53 03
5 o
(n
3
•-H 3 3 ^
Table4B.RegrsionResultwihAlternaiveSpcifatons-NonOECD ountries j
g
ph a £ ffi 2 z z

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 207

variables remain essentially unchanged. Finally, adjusting for capital in regression

(9), the FDI variable turns out to be significant whereas the trade variable loses its
statistical significance.
Several findings emerge as the results of these regressions. First, without gross
capital formation as a control variable, the coefficients on trade are positively
significant in all the System GMM models. In contrast, when the regressions
include capital formation, the coefficients on trade lose their explanatory power.
These results are consistent with the view that trade might be contributing to
economic growth largely through investment (for instance, in the form of increase
in imports of capital goods or investment related to the expansion in exports).
Wacziarg (2001) employs a three-stage least squares (3SLS) model in order to
evaluate simultaneously the mechanism through which trade would bring growth.
He considers six channel variables, namely improved macroeconomic policies,
reduced distortions, government consumption, technological transmissions (proxied
by manufactured exports), investment rate and FDI, through which trade affects
growth. Wacziarg (2001)'s model confirms that trade has a strong positive impact
on economic growth, and that the accelerated accumulation of physical capital
accounts for more than half of the total effect of trade policy on growth. Levine
and Renelt (1992) also identify a positive, robust correlation between growth and
the share of investment in GDP and between the investment share and the ratio of
trade to GDP.

Figure 2. Relationship between Trade, FDI Inflow and Gross Capital Formation
Figure 2.A. Trade and Gross Capital Formation

100 200 300 400

Trade as % of GDP

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Emiko Fukase

Figure 2.B. Trade and FDI Inflow

Figure 2.C. Gross Capital Formation and FDI Inflow

o
+ OECD ° Non-OECD
□ - OECD Non-OECD

o -
—T

-10 30

In order to investigate how trade and capital formation are correlated in my

sample, Figure. 2.A plots the relationship between these variables for the
observations in my dataset.
Overall, trade is shown to be positively associated with capital formation.
Evaluating the data by the subsets of OECD10 and non-OECD economies, it is

,0The OECD countries are defined as the current member countries for which the data are available,
namely, Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hungary,
Iceland, Ireland, Italy, Japan, Korea, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal,
Spain, Sweden, Switzerland, Turkey, United Kingdom, and United States. Czech Republic,
Luxembourg, and Slovak Republic are not included due to the unavailability of the data.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 209

revealed that there is a positive robust link between trade and capital in non-OECD
countries, but that the correlation is close to zero for OECD countries.
Second, when both trade and FDI variables are included in the regressions, one
of these variables tends to lose statistical significance. Trade and FDI are linked in
many different ways. First, both variables might reflect the general openness of a
country (e.g., if a country "opens up" and liberalizes trade and investment regimes
simultaneously). Second, increases in trade may be directly caused by increases in
FDI, due to a much greater propensity to trade of foreign firms relative to domestic
firms.11 Third, the association might reflect features of FDI, e.g. so-called
"horizontal" FDI to serve a protected host-country market vs. "vertical" FDI that is
integrated into the global production network of the parent multinationals. With
"horizontal" FDI, as FDI may occur to save trade costs, FDI acts mainly as a
substitute for trade. In contrast, in the context of "vertical" integration strategies in
which multinationals locate their production activities across different countries
taking advantage of international factor price differences, trade and FDI tend to be
complements. Figure 2.B reveals a positive correlation between trade and FDI for
both OECD and non-OECD economies.

Third, the coefficients on FDI are sensitive to the inclusion of the capital
variable. The effects of FDI on capital formation can be either positive or negative.
On the positive side, foreign firms, especially FDI in the form of greenfield
investments,12 may contribute directly to domestic capital formation. As foreign
firms are more involved in trade relative to domestic firms, FDI might promote
trade-induced investments. They might also indirectly stimulate domestic
investment through backward and forward linkages. On the other hand, FDI might
be harmful to domestic capital formation by crowding out domestic investments
and forcing inefficient firms out of business. Policies offering preferential tax
treatment and other incentives to attract FDI may result in forgone public revenues
and introduce distortions affecting domestic investments.
Figure 2.C plots the relationship between FDI inflow and gross capital

"According to the Enterprise Surveys of 96 countries conducted by World Bank/IFC (www.

enterprisesurveys.org), foreign firms are engaged in trade much more than domestic firms. The shares of
foreign firms (domestic firms) who directly import and export are 68.0 per cent (43.3 per cent) and 43.3
per cent (18.1 per cent) respectively (author's calculation based on the Enterprise Surveys).
l2In the short-run, greenfield FDI, or investment in new projects, usually contributes directly to the stock
of productive capital in the host country (by setting up a new production base for instance). In contrast,
a merger or acquisition (M&A) generally adds to "foreign-owned" capital stock, but does not represent
an immediate addition to the capital stock of countries. In the long-run, the impacts of FDI on capital
formation are difficult to distinguish by mode of entry (UNCTAD, 2006).

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
210 Emiko Fukase

formation. Overall, there exists a positive correlation between these variables.

Whereas the link is found to be positive for non-OECD economies, the association
appears to be negative for OECD economies. Some studies suggest that FDI may
play a greater role in the capital formation in developing countries since the latter
countries rely more on foreign capital as sources of funds and more FDI takes the
form of greenfield investments. In contrast, developed countries have access to
their own capital markets and most FDI in these countries represents the purchase
of existing assets and companies. De Mello (1997) argues that the extent to which
FDI is growth-enhancing is influenced by the degree of complementarity and
substitution between FDI and domestic investment in the short run. In developing
countries (technological followers), according to de Mello, FDI may be promoted
as a means to foster capital accumulation as complementarity of technology would
lead to a more diversified production base. In contrast, in more technologically
advanced economies, a "Shumpeterian view of FDI-related innovative investment"
which implies creative destruction through substitution (de Mello, p.20), might be
more relevant since more efficient technologies embodied in FDI may lead to
faster technological obsolescence. Based on data for 50 countries, Wang (2003)
finds a negative contemporaneous effect of FDI inflows on domestic investment in
OECD countries. She also demonstrates that the cumulative long-run effect of FDI
inflows on a host country's domestic investment tends to be positive in non-OECD
countries and neutral in OECD countries. Table 4.B demonstrates the regression
results for a subset of non-OECD economies.13 The results demonstrate a robust
positive relationship between FDI inflow and growth with the coefficients on FDI
of all the models becoming positively significant.14
Finally, the positive association between education and growth in System GMM
models is robust to the inclusion of various control variables with the coefficients

on education of all the models revealing a significantly positive relationship. The

magnitudes of the effect of schooling on growth vary depending on the control
variables. For instance, controlling for physical capital and other variables in
regression (9), an increase by one year in the average education years of the

l3System GMM estimator is designed for situations with "Small T, large N" panels meaning few time
periods and many individuals (Roodman, 2006). For the subset of OECD countries, it became
impossible to run the System GMM regressions due to the reduced sample size. See Appendix Table
A for the results of OLS and fixed effect models.
14However, further robustness analyses reveal that the coefficients of FDI become insignificant when
inflation is not included in the regressions. Thus, it should not be assumed that FDI is always growth
enhancing in developing countries.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 211

population over age 15 is associated with about .67 per cent faster annual growth in
per capita GDP. These findings contrast with the results obtained from the within
country fixed effects models where none of the coefficients on education reveals
any statistically significant relationship (see Appendix Table A). This exercise
suggests that the results of the coefficients on education obtained from within
country panel data models need to be interpreted with caution since the latter
models wipe out the information on cross-country variation in the levels of
educational attainment.

A similar insight would be relevant for the models derived from growth
accounting exercises since the growth rates of human capital also fail to
incorporate cross-country variation in education levels. For instance, Benhabib and
Spiegel (1994) estimate growth accounting regressions relating the growth rate of
GDP to the log change in years of schooling and find insignificant and usually
negative coefficients on human capital growth rates. However, specifying an
alternative model in which the growth rate of technological progress depends on a
country's human capital stock level, Benhabib and Spiegel find a positive
relationship between human capital and economic growth. They in turn conclude
that education is not directly important as a factor of production and that human
capital affects growth by promoting technological innovation and facilitating
adoption of technology from abroad. However, I cast some doubt on Benhabib and
Spiegel's conclusion that human capital is not a productive input in the production
function since, when the log level of GDP is regressed on the log level of
education, the coefficient on education turns out to be positive and highly
significant.15

l5Assuming a simplified version of the production function in which per capita income (7,) is a function
of physical capital (K,) and human capital (Hi), Y,=g(K,, H,), and taking log differences, I estimate the
following relationship for my sample: Alog Y, = 0.087*** + 0.22*** Alog K, - 0.16*** Alog H,
(0.0095) (0.031) (0.044)
(R2= 0.12; number of observations=560; robust standard errors are in parentheses)
Similar to the findings of Benhabib and Spiegel and others, the coefficient on log differe
capital is found to be significantly negative.
However, when the log level of income per capita is regressed on the log level of hum
coefficient on the log level of education turns out to be highly positively significant as sh
log Y, = 3.70*** + OAS'log K, + 1.71 "'log H,
(0.79) (0.27) (0.13)
(R2= 0.54; number of observations=667; robust standard errors are in parentheses

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
212 Emiko Fukase

IV. Revisiting Borensztein, de Gregorio and Lee (1998)

Borensztein et al. (1998) test the impact of FDI on economic growth via transfer
of technology dealing with data on FDI flows from industrial countries to 69
developing countries. They find a positively significant coefficient for the
interaction term between FDI and schooling and suggest that the beneficial effects
of FDI in terms of growth hold only when the host country has a minimum
threshold stock of human capital. However, a recent study by Carkovic and Levine
(2005) finds little evidence that FDI would affect economic growth even allowing
for the interaction between FDI and human capital.
Borensztein et a/.'s findings lead to a question whether some minimum level of
human capital is a prerequisite for a country to reap any benefit from FDI. Based
on case study evidence involving poorer developing countries, Moran (2005)
suggests that allowing foreign firms to invest in economic sectors which employ
the low-skilled workers in the early stages of export-led growth16 might provide
positive impacts to the least developed countries in terms of generating
employment opportunities and export revenues. As foreign firms have an incentive
to locate labor-intensive activities in low-wage countries either in the context of
vertical integration strategy17 or "export platform" operations, FDI may play a key
role in utilizing efficiently low-skilled labor in which the least developed
economies possess a comparative advantage. Applying the Smith-Myint model of
"vent for surplus" to China, Fu and Balasubramanyam (2005) find that the
expansion of exports from labor-intensive manufacturing, assisted by FDI, has
promoted the growth of industrial output and accelerated the transfer of surplus
labor from the agricultural to the export sector.
Slaughter (2002) suggests that FDI might contribute to a host country's good
macro environment in which fiscal policy drives productive investments, through
for instance contributing to host-country revenues. Botswana's FDI experience can
be cited as an example that even a least developed country with a minimal initial
stock of human capital might reap such benefits if a country has strong institutions.

l6Balasubramanyam, Salisu, and Sapsford (1996) find that the beneficial effect of FDI on economic
growth is stronger when developing countries pursue the export promoting (EP) strategy rather than the
import substituting (IS) strategy.
l7Using firm-level data on U.S. multinationals, Hanson, Mataloni and Slaughter (2005) examine the
behavior of multinational firms locating input processing in their foreign affiliates. They find that the
demand for imported inputs by U.S. multinationals is higher when affiliates face lower wages for less
skilled labor.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 213

Botswana is among the fastest growing economies (the second fastest after China
in my sample) with an average annual real per capita GDP growth rate of 6.8 per
cent for the past 35 years. Starting as a very poor nation at independence in 1966
with per capita income of merely $312,18 Botswana now belongs to the upper
middle income countries with real per capita GDP of $3,683.9 in 2004 (The World
Bank). Several studies document the presence of good institutions among the
causes of its high economic growth (Acemoglu, Johnson, and Robinson, 2001;
Mehlum, Moene, and Torvik, 2006; Olsson, 2006). Soon after the Kimberlite
diamond pipes were found in the early 1970s, the Botswanan government formed a
joint venture with the South African firm De Beers, which has a dominant position
in the global diamond market.19 Botswana in turn invested its revenues from
diamonds in a socially efficient way in public goods such as infrastructure, health
and education (Acemoglu et al., 2001). For instance, Botswana accumulated
human capital at a much faster rate than the world average with the average
education years of its population increasing from 2.0 years in 1970 to 6.3 years in
2000 (Barro and Lee, 2000).
Blonigen and Wang (2005) argue that it is inappropriate to pool developing and
developed countries in empirical analyses since there is a systematic difference in
the determinants and effects of FDI between these two groups of countries. They
test a hypothesis that the contrasting results between Borensztein et al. (1998) and
Carkovic and Levine (2005) might be attributed to the different coverage of host
countries included in their samples.20 Extending Borensztein et al. 's dataset adding
developed countries as host countries, and observing that the interaction term loses
its significance, Blonigen and Wang argue that Borensztein et al. 's conclusions are
supported only for developing host countries. In the next subsection, I extend
Borensztein et a/./Blonigen and Wang's analyses and investigate what happens if
the source countries are extended to both OECD and non-OECD countries using
their actual dataset.

Column 1 on Table 5 reproduces Borensztein et al. 's results of the impact of FDI
from OECD to developing countries (Regression 1.3 in Table 1, p. 124) using the

182000 constant price.

19See Olsson (2006) for the crucial role of Botswana's joint venture with De Beers in securing revenue
from diamond production.
20Carkovic and Levine (2005) deal with both developing and developed countries as source and
destination of FD1. Borensztein et al. investigate only the FDI originated in OECD countries to
developing countries assuming that FDI flows from industrialized to developing countries to "close the
technological gap" (Borensztein et al. , 1998, p. 122).

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Emiko Fukase

Table 5. Replication and Extension of Borensztein et al. (1998)

(1) (2) (3)
-0.12*** -0.0099*** -0.014***
Log of initial GDP
(0.0038) (0.0032) (0.0040)
0.012*** 0.0077** 0.021***
Schooling1
(0.0045) (0.0034) (0.0058)
-0.080** -0.10*** -0.11***
Government Consumption
(0.033) (0.031) (0.034)
-0.018*** -0.019*** -0.016***
Log(l+black market premium)
(0.0054) (0.0051) (0.0055)
FDI from OECD to 69 Developing Countries
-0.83
FDI
(0.72)
1.61***
FDI*Schooling
(0.61)
FDI from OECD to both OECD and non-OECD Countries
0.21
FDI
(0.59)
0.14
FDI* Schooling
(0.36)
FDI from both OECD and non-OECD to 69 Developing Countries
0.55**
FDI
(0.26)
-0.23
FDI* Schooling
(0.22)
Number of Countries 69 88 69
R2- 1970-79 0.39 0.28 0.36
R2 - 1980-89 0.16 0.12 0.17
50.13 42.00 45.55
Goodness of Fit - ^(p- value)
(0.000) (0.000) (0.000)
Notes: The dependent variable is the per capita GDP growth rate. Standard errors are in parentheses.1
Schooling represents average years of the male secondary schooling aged over 25. Seemingly
Unrelated Regression (SUR) estimation was done constraining all the coefficients (except the
constant) to be the same across the two decades.

Seemingly Unrelated Regression (SUR) techniques for the periods 1970s and
1980s.

The results demonstrate that whereas the coefficient on FDI is negative but
insignificant, the interaction term is positively significant at the one per cent level.
The values of these coefficients imply that the countries with male secondary
school attainment above .52 will benefit from FDI, which is the identical result
with Borensztein et al. 's (p. 125). Column 2 of Table 5 provides the results of the

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 215

effects of FDI from OECD to both developed and developing economies as host
countries using Blonigen and Wang's (2005) extended dataset. The results show
much reduced and statistically insignificant coefficient estimates on FDI measures
supporting Blonigen and Wang (2005)'s conclusion that FDI from developed
countries affects growth conditioned on a sufficient level of human capital only in
developing host countries.
Finally, Column 3 of Table 5 reports the results replacing the FDI variable with
those from all the source countries to developing host countries. Whereas the
coefficient on FDI is found to be positively significant at the five per cent level, the
coefficient on the interaction term becomes insignificant. These results imply that,
once all the source countries are taken into consideration, the growth of developing
countries is generally associated positively with FDI and that no evidence is found
to support the existence of threshold level of schooling below which developing

Table 6. Regression Results with the Interaction Term

(2') (41) (6') (7') (8') (9')

1.05*** 1.01*** 0.96*** 0.97*** 0.98*** 0.95***
Lagged log of GDP
(0.012) (0.027) (0.023) (0.028) (0.030) (0.019)
2.60* 2.36* 4.64*** 4.34*** 2.81***
Average Education Years
(1.50) (1.38) (1.56) (1.58) (1.07)
-0.0082*** -0.0067*** -0.0073*** -0.0068*** -0.0069***
Inflation
(0.00077) (0.0014) (0.0014) (0.0015) (0.0013)
0.11** 0.13** -0.024
Trade as % of GDP
(0.049) (0.057) (0.026)
FDI Measures
2.24* 5.64*** 1.28 5.70*** 5.46*** 1.39
FDI
(1.16) (1.11) (0.95) (1.31) (1.10) (0.87)
-0.29* -0.67*** -0.10 -0.81*** -0.76*** -0.082
FDI*Schooling
(0.16) (0.15) (0.12) (0.18) (0.15) (0.11)
-0.75**
Government Consumption
(0.32)
1.46*** 1.42***
Gross Capital Formation
(0.19) (0.18)
M2 0.372 0.920 0.676 0.744 0.882 0.816
Hansen 0.352 0.260 0.226 0.156 0.421 0.338
Number of Observations 656 613 606 607 605 605
Number of Countries 104 102 102 102 102 102
Number of Instruments 88 92 79 79 91 91

Notes: The models in Table 4A were re-estimated including the interaction term between schooling and
FDI.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
216 Emiko Fukase

host countries cease to benefit.

It is not the purpose of this exercise to argue that human capital is not important
interactively with FDI. Perhaps, Borensztein et al. 's conclusions may be valid for
the technology transfer channel from developed countries but developing source
countries may contribute to growth in other developing countries probably due to
the use of technology that is more compatible with host-country educational levels.
In other words, this paper suggests that there appears to be a variety of channels
through which FDI can affect growth and that the roles of human capital are likely
to differ depending on different channels, industries, and host country policies and
institutional capabilities.
Table 6 demonstrates the results of System GMM models re-estimating the
regressions in Table 4.A including the interaction term between schooling and FDI.
Surprisingly, the coefficients on interaction terms turn out to be negatively
significant in four out of the six models implying that the growth effect of FDI is
greater for countries with lower human capital.21 These findings may appear
puzzling, but are consistent with Carkovic and Levine's (20 05).22 A potential
explanation is that the interaction term may have captured an overall tendency for
the impacts of FDI to be greater in developing countries (with less human capital)
than in developed countries (with more human capital). The results of the impacts
of FDI vary depending on the periods included. Excluding the most recent three
periods, namely 1989-2004, the coefficients on the interaction term turn out to be
insignificant (not reported). Perhaps, a variety of changing characteristics of FDI -
including a greater level of cross-border mergers and acquisitions (M&As) relative
to the earlier periods, and increasing participation of developing economies as
sources of FDI23 (UNCTAD, 2006) - may have been altering the impacts of FDI in
host countries over time.

2'The qualitative results are similar when I use OLS, SUR and fixed effects models.
^Investigating 70 countries over the periods 1960 to 1995, Carkovic and Levine (2005) find that the
coefficients on FDPeducation are either insignificant or negatively significant in their System GMM
models.

23For low-income developing countries, FDI from developing countries accounts for a large share of total
FDI flows. FDI by developing countries might have a greater employment-generating potential relative
to those by developed countries since they may be oriented more towards labor-intensive industries and
may use simpler technologies. The example of such FDI is the emergence of East Asian developing
countries as foreign investors in labor-intensive sectors in Asia (UNCTAD, 2006).

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth 217

V. Conclusion

This paper revisits the relationship between openness, education and economic
growth for 106 countries over the periods 1969-2004 using the System GMM
approach.
In terms of the linkages between openness and growth, the system GMM
estimators provide support for some of the earlier findings. Whereas there are
usually positive ties between trade and growth, the coefficient on trade loses
statistical significance when the regressions include capital formation. The latter
result is consistent with the earlier research findings that trade is largely linked to
higher growth via investment (Wacziarg, 2001; Levine and Renelt, 1992).
There generally exists a positive relationship between FDI inflow and economic
growth for a subset of non-OECD countries. This may be partly because FDI has
on average positive impacts on host-countries' capital formation in developing
countries (de Mello, 1997; Wang, 2003). However, the paper does not reveal clear
insights on how FDI and education interact to affect economic growth. In my
regressions, the coefficients for the interaction term between FDI and education are
found to be sensitive to the coverage of source and host countries and the periods
included. The paper suggests that there appears to be a variety of channels through
which FDI can affect growth and that the roles of human capital are likely to differ
depending on different channels, industries, and host country policies and
institutional capabilities.
Finally, the System GMM approach is found to improve substantially the
estimate of the impacts of education on growth relative to the models which focus
on the changes in educational attainment. The effect of education on economic
growth has been debated in the literature. Whereas earlier studies identified the
positive impacts of the initial level of education on economic growth (e.g. Barro,
1991), some influential papers which relate economic growth and education
growth find insignificant and even negative coefficients on education (e.g.
Benhabib and Spiegel, 1994; Pritchett, 2001). This paper hypothesizes that these
mixed results are attributed to the characteristics of the data, i.e., whereas there
exists a positive relationship between the levels in educational attainment and
economic growth, there is little systematic variation in the changes in education
across countries (see Figure 1). The System GMM approach is shown to be better
suited to the actual data: once information on cross country variation in education
levels is added to data on within-country changes in education, the System GMM

This content downloaded from

f:ffff:ffff:ffff:ffff:ffff:ffff on Thu, 01 Jan 1976 12:34:56 UTC
All use subject to https://about.jstor.org/terms
218 Emiko Fukase

estimator reveals a strong positive relationship between education and subsequent

growth. These exercises have important policy implications. Given fiscal budget
constraints, there exists a potential danger that a belief that education has little or
even a negative effect upon economic growth may discourage public investment in
education. This paper emphasizes that there indeed exists a robust positive
association between education and subsequent growth and suggests that investment
in schooling should be among the top priorities in countries' development
strategies.

Acknowledgements

I am grateful to Terence Agbeyegbe for suggesting using the System GMM

approach, Miao Wang for sharing the FDI data, Jonathan Nelson, Wim Vijverberg
and anonymous referees for very useful comments. Any remaining errors are mine.

Received 13 March 2008, Revised 26 October 2009, Accepted 2 November 2009

Data Appendix

The following data are extracted from the World Development Indicators (the
World Bank) from 1969 to 2004: real per capita GDP in constant 2000 prices;
inflation rate in consumer prices, trade (imports plus exports) as per cent of GDP;
government consumption as per cent of GDP; and gross capital formation as per
cent of GDP. The data for FDI inflow as per cent of GDP are taken from Foreign
Direct Investment (FDI) Database of UNCTAD. The averages of male and female
schooling years for the population over age 15 are computed from Barro and Lee's
dataset (2000). The FDI data used for Borensztein et al. (1998) and Blonigen and
Wang (2005)'s studies were provided by Miao Wang."

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth

Table A. Results of Education and Openness Variables Estimated by OLS and Fixed Effects
Models

Variables Included Pooled OLS Fixed Effects

Lagged Educa Educa

Model Trade FDI Infl. Capital Gov FDI Trade Education FDI Trade
GDP tion tion

Full Sample
0.059"* 0.19*"
(1) X X
(0.015) (0.039)
0.66 2.17***
(2) X X
(0.52) (0.64)
1.95"* 0.050*" 0.40 0.17***
(3) X X X X
(0.74) (0.016) (1.09) (0.046)
1.84" 1.26*** 0.31 1.50"*
(4) X X X X
(0.73) (0.28) (1.10) (0.32)
1.72"* 0.0087 0.76 0.12***
(5) X X X X X
(0.57) (0.014) (1.06) (0.044)
1.66*" 0.71*" 0.65 1.23*"
(6) X X X X X
(0.58) (0.23) (1.03) (0.31)
1.85" 1.08*" 0.017 0.30 1.25*** 0.15*"
(7) X X X XX
(0.73) (0.38) (0.019) (1.15) (0.33) (0.043)
1.89" 1.02** 0.022 0.33 1.23'** 0.15***
(8) X X X XX X
(0.74) (0.40) (0.019) (1.16) (0.33) (0.043)
1.64"* 0.98*** -0.023 0.68 1.07*** 0.10"
(9) X X X XX X
(0.56) (0.30) (0.016) (1.07) (0.29) (0.040)
Non-OECD Countries
0.087*" 0.19***
(1) X X
(0.019) (0.040)
0.87 2.41***
(2) X X
(0.69) (0.70)
2.83*** 0.064*** -0.86 0.16"*
(3) X X X X
(0.90) (0.020) (1.06) (0.047)
2.82*** 1.85*** -0.64 1.70***
(4) X X X X
(0.34)
(0.78) (1.19) (0.38)
2.57"* 0.015 -0.088 0.093**
(5) X X X X X
(0.67) (0.018) (1.16) (0.039)
2.47**' 1.15"* -0.22 1.33***
(6) X X X X X
(0.61) (0.30) (1.13) (0.38)
2.80"* 1.75*** 0.013 -0.80 1.58**' 0.14***
(7) X X X XX
(0.81) (0.45) (0.022) (1.15) (0.42) (0.047)
2.81*** 1.69*** 0.019 -0.79 1.55*** 0.14"*
(8) X X X XX X
(0.82) (0.47) (0.022) (1.15) (0.42) (0.047)
2.51*** 1.45*** -0.027 -0.20 1.24*** 0.083"
(9) X X X XX X
(0.60) (0.35) (0.018) (1.17) (0.38) (0.041)
OECD Countries
0.056 0.23"
(1) X X
(0.033) (0.093)
0.17 0.64
(2) X X
(0.73) (0.72)
-0.20 0.046 1.49 0.25"
(3) X X X X
(0.57) (0.034) (1.45) (0.095)
(Continued)

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Emiko Fukase

Table A. Results of Education and Openness Variables Estimated by OLS and Fixed Effects
Models (Continued)
Variables Included Pooled OLS Fixed Effects

Lagged Edu Educa

Model FDI
Trade FDI Infl. CapitalGov Education Trade FDI Trade
GDP cation tion
-0.23 0.16 1.05 0.78
(4) X X X X
(0.67) (0.76) (1.31) (0.69)
-0.33 0.042 1.04 0.18
(5) X X X X X
(1.48) (0.11)
(0.54) (0.034)
-0.34 0.33 0.91 0.73
(6) X X X X X
(0.57) (0.63) (1.39) (0.52)
-0.14 -0.50 0.079" 0.85 -0.0043 0.27"*
(7) X X X X X
(1.47) (0.38) (0.094)
(0.62) (0.71) (0.037)
-0.16 -0.51 0.091" 1.41 -0.15 0.23"
(8) X X X X X X
(0.63) (0.036) (1.40) (0.37) (0.087)
(0.59)
-0.26 -0.19 0.062* 0.81 0.25 0.17
(9) X X X X X X
(0.55) (0.50) (0.034) (1.59) (0.34) (0.12)

Notes: The dependent variable is the log of GDP. Robust standard errors are in parentheses. The model numbers (1)
through (9) correspond to those in Table 4.A.B. *, "* indicate that the coefficients are significant at the 10,
5, and 1 per cent level respectively.

References

Acemoglu, D., Johnson, S., and Robinson, J.(2001), "An African Success Story:
Botswana", MIT Department of Economics Working Paper, pp. 01-37.
Arellano, M. and Bond, S.(1991), "Some Tests of Specification for Panel Data: Monte
Carlo Evidence and an Application to Employment Equations", Review of Economic
Studies, 58(2), pp. 277-97.
Arellano, M. and Bover, 0.(1995), "Another Look at the Instrumental Variable Estimation
of Error-Components Models", Journal of Econometrics, 68, pp. 29-51.
Balasubramanyam, V. N., Salisu, M. and Sapsford, D.(1996), "Foreign Direct Investment
and Growth in EP and IS Countries", the Economic Journal, 106(1), pp. 92-105.
Baldwin, R.(2003), "Openness and Growth: What's the Empirical Relationship?", NBER
Working Paper, 9578.
Barro, R.(1991), "Economic Growth in a Cross Section of Countries", Quarterly Journal
of Economics, 106(2), pp. 407-443.
Barro, R. and Sala-i-Martin, X.(1995), Economic Growth. McGraw-Hill.
Barro, R., and Lee, J-W.(2000), "International Data on Educational Attainment: Updates
and Implications", Working Paper, No.42, Center for International Development
(CID).
Benhabib, J. and Spiegel, M.(1994), "The Role of Human Capital in Economic
Development: Evidence from Aggregate Cross-Country Data", Journal of Monetary
Economics, 34, pp. 143-173.
Blonigen, B. A., and Wang, M.(2005), "Inappropriate Pooling of Wealthy and Poor

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Revisiting Linkages between Openness, Education and Economic Growth

Countries in Empirical FDI Studies", Does Foreign Direct Investment Promote

Development? pp. 221-243, Institute for International Economics: Center for Global
Development.
Blundell, R. and Bond, S.(1998), "Initial Conditions and Moment Restrictions in Dynamic
Panel Data Models", Journal of Econometrics, 87, pp. 115-143.
Borensztein, E., de Gregorio J. and Lee. J.(1998), "How Does Foreign Direct Investment
Affect Economic Growth?", Journal of International Economics, 45, pp. 115-35.
Carkovic, M. and Levine, R.(2005), "Does Foreign Direct Investment Accelerate
Economic Growth? Does Foreign Direct Investment Promote Development?", 195
220, Institute for International Economics: Center for Global Development.
Caselli, F., Esquivel, G. and Lefort, F.(1996), "Reopening the Convergence Debate: a New
Look at Cross-Country Growth Empirics", Journal of Economic Growth, I, pp. 363
389.

de Mello, L. R., Jr.(1997), "Foreign Direct Investment in Developing Countries and

Growth: a Selective Survey", Journal of Development Studies, 34, pp. 1-34.
Dollar, D. and Kraay A.(2004), "Trade, Growth, and Poverty", Economic Journal, 114,
F22-F49.

Dollar, D.(1992), "Outward-Oriented Developing Economies Really Do Grow More

Rapidly: Evidence from 95 LDCs, 1976-85", Economic Development and Cultural
Change, 40, pp. 523-544.
Felbermayr, G.(2005), "Dynamic Panel Data Evidence on the Trade-Income Relation",
Review of World Economics, 141(4), pp. 583-611.
Fu, X. and Balasubramanyam, V. N.(2005), "Exports, Foreign Direct Investment and
Employment: the Case of China", The World Economy, 28(4), pp. 607-625.
Hanson, G. H., Mataloni Jr. R. J., and Slaughter, M.(2005), "Vertical Production Networks
in Multinational Firms", Review of Economics and Statistics, November 87, pp. 664
678.

Krueger, A. B. and Lindahl, M.(2001), "Education for Growth: Why and for Whom?",
Journal of Economic Literature, 39, pp. 1101-1136.
Levine, R. and Renelt, D.(1992), "A Sensitivity Analysis of Cross-Country Growth
Regressions", American Economic Review, 82(4), pp. 942-963.
Lipsey, R. E.(2004), "Home and Host Country Effects of Foreign Direct Investment", in
Baldwin R. E. and Winters L. A. (Eds.), Challenges to Globalization: Analyzing the
Economics, pp. 333-379, Chicago and London: University of Chicago Press.
Mehlum, H., Moene, K., and Torvik, R.(2006), "Cursed by Resources or Institutions?",
the World Economy, 29(8), pp. 1117-1131.
Moran, T. H.(2005), "How Does FDI Affect Host Country Development? Using Industry
Case Studies to Make Reliable Generalizations", Does Foreign Direct Investment
Promote Development? pp. 281-313, Institute for International Economics: Center
for Global Development.
Olsson, 0.(2006), "Diamonds Are a Rebel's Best Friend", The World Economy, 29(8), pp.
1133-1150.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms
 Emiko Fukase

Pritchett, L.(2001), "Where Has All the Education Gone?", The World Bank Economic
Review, 15(3), pp. 367-391.
Rodriguez, F. and Rodrik, D.(2001), "Trade Policy and Economic Growth: a Skeptic's
Guide to the Cross-National Evidence", Macroeconomic Annual 2000, 261-324,
Cambridge, MA: MIT Press for NBER.
Roodman, D.(2006), "How to Do Xtabond2: an Introduction to "Difference" and
"System" GMM in Stata", Center for Global Development Working Paper, 103.
Sachs, J. D. and Warner, A.(1995), "Economic Reform and the Process of Global
Integration", Brookings Papers on Economic Activity, No. 1.
Slaughter, M.(2002), "Does Inward Foreign Direct Investment Contribute to Skill
Upgrading in Developing Countries?", CEPA Working Paper, 2002-08, New School
University.
UNCTAD(2006), World Investment Report 2006: FDI from Developing and Transition
Economies: Implications for Development.
Wacziarg, R.(2001), "Measuring the Dynamic Gains from Trade", World Bank Economic
Review, 15(3), pp. 393-429.
Wang, M.(2003), Essays on Foreign Direct Investment, Ph.D. Dissertation, University of
Oregon.
Wetherill, G.(1986), Regression Analysis with Applications. New York: Chapman and
Hall.

Winters, L. A.(2004), "Trade Liberalization and Economic Performance: an Overview",

Economic Journal, 114 (February), F4-F21.
World Bank, World Development Indicators, Washington, D.C.

This content downloaded from

39.40.73.247 on Sun, 09 Jan 2022 18:10:38 UTC
All use subject to https://about.jstor.org/terms