Economic Modelling 29 (2012) 868–878

Contents lists available at SciVerse ScienceDirect

Economic Modelling
journal homepage: www.elsevier.com/locate/ecmod

Economic growth, exports and foreign direct investment in

Least Developed Countries: A panel Granger causality analysis☆
Rıfat Barış Tekin
Marmara University, Department of Economics, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: This paper investigates potential Granger causality among the real GDP, real exports and inward FDI in Least
Accepted 25 October 2011 Developed Countries for the period between 1970 and 2009. A new panel-data approach developed in Kónya
(2006) [Kónya (2006), Exports and growth: Granger causality analysis on OECD countries with a panel data
JEL classification:
approach, Economic Modelling, 23, 978–992] which is based on SUR systems and Wald tests with country
F14
specific bootstrap critical values has been employed. The results indicate direct, one-period-ahead,
F21
F43
unidirectional causality from exports to GDP in Haiti, Rwanda and Sierra Leone, and from GDP to exports
O10 in Angola, Chad and Zambia. Considering the FDI–Growth nexus, there is evidence of FDI Granger-causing
O11 GDP in Benin and Togo, and GDP Granger-causing FDI in Burkina Faso, Gambia, Madagascar and Malawi.
While studying EXP–FDI relations, this paper finds that the causality is from FDI to real exports in Benin,
Keywords: Chad, Haiti, Mauritania, Niger, Togo and Yemen, and from real exports to FDI in Haiti, Madagascar,
Export-led growth Mauritania, Malawi, Rwanda, Senegal and Zambia.
Growth-led exports © 2011 Elsevier B.V. All rights reserved.
Foreign direct investment
Least Developed Countries
Granger causality
Bootstrapping

1. Introduction
This paper examines causality relations among the real GDP, real ex-
ports and real net FDI inflows as they affect Least Developed Countries
(LDCs). Understanding causality relations among economic growth, ex-
ports and FDI in LDCs poses an important research question as this
group of countries includes the most vulnerable economies of the
world, suffering from drastic economic and social problems, extreme
poverty, hunger and inadequate levels of human development.1 For
quite some time, practically since the dissolution of colonial empires,
assisting development in the economically backward areas of the
world has been on the international community's agenda. Developed
☆ The author acknowledges financial support by the Marmara University Scientific
Research Committee (project no: SOS-D-070211-0021).
E-mail address: rbtekin@marmara.edu.tr.
1
Currently, (as of January 2011) 48 countries are classified by the United Nations as
“least developed countries”. The selection of LDCs depends on three criteria: a “low-in-
come” criterion, based on a three-year average estimate of the gross national income
(GNI) per capita, with a threshold of $905 for inclusion to the LDC list, and a threshold
of $1086 for graduation from the status; a “human assets weakness” criterion, involv-
ing a composite index based on indicators of nutrition, health, school enrolment, liter-
acy and an “economic vulnerability” criterion, involving a composite index based on
indicators of natural shocks, trade shocks, exposure to shocks, economic smallness
and economic remoteness (UNCTAD, 2010). The LDC list is reviewed in the light of the-
se criteria every three years by the United Nations Economic and Social Council (ECO-
SOC) in order to identify countries that might graduate from LDC status.
countries of the world, often under the leadership of the United Nations,
have pursued extensive, multifaceted aid schemes and economic devel-
opment assistance programs with the aim of encouraging sustainable
economic growth and human development in developing areas.
Recognizing the special needs of the Least Developed Countries,
the United Nations has organized three major conferences on LDCs
over the last three decades. The three “Programme of Actions” initiat-
ed by these successive UN Conferences served as the main framework
of international development assistance provided to the LDCs. The
first UN Conference on LDCs organized in Paris in 1981 initiated the
first comprehensive “Programme of Action” (PoA), with the aim of
improving the deteriorating conditions of the LDCs. The Second UN
Conference on LDCs adopted the Paris Declaration and initiated a
new “PoA” for the 1990s. The third UN Conference on LDCs was orga-
nized in Brussels in May of 2001 and initiated the “Brussels Declaration
and Programme of Action for the LDCs” (BPoA) for the period from 2001
to 2010. These three successive PoAs had the intention of reversing the
negative economic and developmental trends witnessed in LDCs, en-
hancing economic growth, and alleviating poverty; while their ultimate
aim remained the eventual graduation of targeted countries from LDC
status.2 Although the ideas, argumentation and general economic
2
Up until today only three LDCs have managed to graduate from the UN's list of
LDCs: Botswana in December 1994, Cape Verde in December 2007, and most recently
Maldives on 1 January 2011.

0264-9993/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.econmod.2011.10.013
 R.B. Tekin / Economic Modelling 29 (2012) 868–878
rationale behind these programs have been changing across time, the
three PoAs implemented since 1980s relied on similar policy measures.
Typically, besides direct bilateral and multilateral official development
aid flows, all the three PoAs attributed a central role to export revenues
and foreign direct investment inflows in supporting economic growth
and development in LDCs. A short look at the building blocks of the
BPoA makes it easier to understand the core economic thinking behind
these three types of implemented development assistance programs.
The BPoA foresaw a series of major policy tools to support devel-
opment and economic growth in LDCs, including (i) Official develop-
ment assistance (ODA), (ii) preferential market access and trade
related assistance, (iii) Foreign direct investment (FDI) inflows and
(iv) debt relief. The BPoA attaches a particular importance to foreign
direct investment (FDI) and export revenues to complement Official
Development Assistance (ODA) and other international transfers in
compensating for the inadequate level of capital accumulation in
LDCs. The BPoA is based on the view that sustainable expansion of
output in LDCs depended more on foreign demand for exports, rather
than domestic demand. In the BPoA, like in the previous PoAs, FDI in-
flows are seen as a vital source of finance, and a means for acquiring
new technology and know-how for LDCs to be able to compete in
international markets (UN, 2011: 3). Based on these views, the inter-
national community has been encouraging LDCs to adopt export-
oriented development strategies and more open trade and invest-
ment regimes. Export-orientation and internationalization in LDCs
have also been supported by special aid-for-trade and trade capacity
building schemes.
Despite the fact that export revenues and FDI inflows remain at
the core of the major development assistance programs for the
LDCs, only a very limited number of empirical studies have targeted
their impact on economic growth in this group of countries. Although
there exists voluminous empirical literature on both the ‘exports–
growth’ and ‘FDI–growth’ nexuses, the bulk of this literature focuses
on the case of “developing countries” in general, with no explicit
focus on LDCs. The LDCs, however, provide a particular subset of de-
veloping countries, as they share very similar developmental prob-
lems and characteristics, having been subject to the same
development assistance and poverty alleviation schemes for many
decades. Furthermore, the body of empirical evidence on the more
general case of developing countries still remains to be inconclusive,
as the results of existing studies are obviously country- and
methodology-specific. Most of the existing empirical works, on the
other hand, are taint with severe estimation biases as they employ
earlier econometric techniques which do not take into account
cross-sectional dependency and heterogeneity issues. There is there-
fore a need for further research on causality relationships among eco-
nomic growth, exports and FDI in LDCs. This is what this study strives
to do.
This paper examines the possibility of Granger causality among
real GDP, real exports and real net FDI in Least Developed Countries
(LDCs), for the period between 1970 and 2009. While studying cau-
sality relations among real economic growth, real exports and real
net FDI inflows this study employs a new panel-data approach re-
cently developed in Kónya (2006), which is based on seemingly unre-
lated regression systems (SUR) and Wald tests with country specific
bootstrap critical values. This new approach makes it possible to
test for Granger-causality on each individual panel member separate-
ly, while accounting for the problem of cross-sectional dependence in
panel data. These advantages make this approach particularly well-
suited for the purposes of this study, given the cross-sectional depen-
dency of the data at hand, and the high degree of heterogeneity exist-
ing across LDCs.
The objectives of this paper can now be stated as follows. The pri-
mary objective of the study is to examine the ‘exports-growth’ nexus
in LDCs, testing explicitly for both the “export-led growth” and
“growth-led export” hypotheses. The findings of Granger causality
869
tests implemented will shed light on the question of whether export
revenues contribute significantly to real economic growth or whether
it is domestic growth dynamics that trigger exports growth in LDCs in
the first instance. Secondly, this paper aims to investigate potential
causality directions between FDI inflows and economic growth em-
pirically. At this level, the objective is to see whether FDI inflows
from the rest of the world contribute to the growth of LDCs, or rather
LDCs with better growth performance attract more FDI. The findings
of our causality tests are expected to provide some new information
on the determinants of FDI inflows to this group of low income devel-
oping countries. In addition to studying the ‘exports-growth’ and
‘FDI-growth’ nexuses, the third objective of this paper is to examine
a less scrutinized, but equally important set of relationships between
FDI and total merchandise exports in LDCs. The methodology
employed in this research allows us to address all these three sets
of causality relations simultaneously. In an attempt to study unidirec-
tional and bidirectional causality relationships among the variables at
hand, we estimate both bivariate and trivariate systems of equations. 3
To the best of our knowledge, this is the first study that exclusively fo-
cuses on causality relationships among economic growth, exports and
FDI in the case of LDCs.
The remainder of the paper is organized as follows: the second
section provides a short survey of the related theoretical and empiri-
cal literatures; the third section presents the research methodology
employed for testing causality; the fourth section reports on the
data employed, and presents empirical findings of the research; the
final section provides concluding remarks and some policy
recommendations.
2. Literature survey
2.1. Exports–growth nexus
The relation between export growth and economic growth has
long been an area in international and development economics re-
ceiving a great deal of research attention. Relying on basic economic
theory, one can suggest that exports growth contribute to economic
growth first through what is known as the foreign trade multiplier ef-
fect (see Stolper, 1947). The foreign trade multiplier analysis asserts
that, given the spending function, an export surplus will have an ex-
pansionary effect whose magnitude depends on the marginal propen-
sity to import. Transfer of scarce resources from low-productivity
domestic industries to higher-productivity export industries results
in an increase in overall productivity, accelerating output growth.
Economic theory also suggests that a higher level of exports might
contribute to economic growth as export revenues provide an impor-
tant source of foreign exchange, which is crucial when domestic sav-
ings are inadequate for making imports of capital goods possible.
Finally, export growth might also trigger economic growth through
the expansion of the efficient market size, bringing in substantial
economies of scale that accelerate the rate of capital formation and
technical change. 4
The causality relation between exports and economic growth can
work in both directions; a reverse causality direction, from economic
growth to export growth might well exist. This idea is often referred
as the “growth-led export” hypothesis and is based on the view that
domestic economic growth dynamics is more relevant for explaining
export growth (see, for example, Jung and Marshall, 1985). Basically,
the rationale behind this hypothesis is the idea that output growth
triggers productivity growth, which in turn enhances international
competitiveness of export products, and accelerates export growth
3
Our bivariate and trivariate analyses led to very similar results. For the sake of clar-
ity, however, only the trivariate model results are detailed in the paper. Bivariate mod-
el results are available from the author upon request.
4
See Reppas and Christopoulos (2005: 930).
870 R.B. Tekin / Economic Modelling 29 (2012) 868–878
(Kaldor, 1967). In the ‘new trade theory,’ this feature is understood as
a process of ‘cumulative causation’ in which the development of pro-
ductive capacities and the growth of demand mutually work to rein-
force each other (see Markusen and Venables, 1998; UNCTAD, 2010:
104–105). There exists voluminous literature focusing on the empir-
ical relationship between exports and growth, covering diverse as-
pects of this relationship in different countries, testing for the
export-led growth or growth-led hypothesis, or both. 5 Earlier studies
in the literature that employed cross-sectional data to study the rela-
tion between exports and economic growth are generally supportive
of the export-led growth hypothesis. However, this large corpus of
earlier work does not take into account country-specific factors and
are potentially flawed with severe estimation biases. More recent
studies that employ time-series data for studying Granger causality
in single country cases fail to provide strong support neither for the
export-led nor the growth-led exports hypotheses. 6 The findings of
these studies, however, can also be misleading as they suffer from
the spurious regression problem given the fact that regressions
were based on the use of non-stationary data. Furthermore, most of
the literature on the causal link between exports and economic
growth are based on bivariate models alone, omitting other potential
explanatory variables from the analysis.
Although there exists a wide empirical literature on causality rela-
tions between exports and economic growth, there are only a few
studies that focused on the case of LDCs in particular. The empirical
evidence for the LDCs come either from single country studies or
from large panels covering the full set of low and middle income de-
veloping countries. The findings of this literature on causality rela-
tions between exports and economic growth are mixed. Earlier
studies such as Jung and Marshall (1985), Ahmad and Kwan (1991),
Bahmani-Oskooee et al. (1991) and Hutchinson and Singh (1992)
failed to provide evidence in support of neither the export-led growth
nor the growth-led exports hypotheses. Testing for Granger causality
within an error-correction framework Bahmani-Oskooee and Alse
(1993) have found evidence supporting bidirectional causality be-
tween exports and real economic growth while Kugler and Dridi
(1993) provided some support for bidirectional causality. More re-
cently, Reppas and Christopoulos (2005), focusing on a sample of 22
countries including some LDCs and middle income developing coun-
tries found evidence supporting the growth-led export and not the
export-led growth hypothesis. Focusing on long-run causality be-
tween export and economic growth in LDCs Bahmani-Oskooee and
Economidou (2009) concluded that there is no clear support for nei-
ther of the two hypotheses and the results are country-specific. 7
2.2. FDI–economic growth nexus
There are several potential ways in which FDI can contribute to
economic growth. In earlier neo-classical growth models FDI in-
creases the capital stock and promotes economic growth by financing
capital formation. In these models the impact of FDI inflows are exact-
ly the same as domestic capital investments, that is, similar to domes-
tic investments, FDI only has a ‘short-run’ growth effect, due to the
diminishing returns to capital. In the new growth theory that places
the emphasis on technological change, on the other hand, FDI is as-
sumed to have a positive impact on economic growth, both in the
short and the long-run (Herzer et al., 2008: 794). According to the
new growth theory, FDI is more productive than domestic invest-
ment, as FDI-related technological spillovers might offset the impact
5
Giles and Williams (2000) and Singh (2010) provide comprehensive reviews of re-
lated theoretical and empirical literature.
6 8
See Singh (2010).
7
For a review of earlier literature on LDCs, see Bahmani-Oskooee and Economidou
(2009).
of diminishing returns to capital and allow the economy to continue
to growth in the long run. 8
Similar to the export-growth nexus, the causality relation between
FDI and growth is not necessarily unidirectional, and causality can
work on both directions. The primary explanation brought by standard
economic theory for the possibility of a reverse causality direction (i.e.
from economic growth to FDI) is again based on the process of “cumu-
lative causation.” In this line of thinking, a long-term process of eco-
nomic growth based on the development of productive capacities
might well create new economic activities, new markets and a higher
demand for new consumer products, which will in turn attract a higher
level of FDI. Theoretical literature, however, suggests that the positive
relationship between FDI and economic growth may not come true
for many reasons. Several theoretical works have shown that the
growth accelerating effect of FDI inflows is crucially dependent on the
assumption that FDI does not ‘crowd out’ substantial amounts of invest-
ment from domestic sources (Herzer et al., 2008: 794). In the case of
substantial crowding-out, FDI inflows might well have a growth decel-
erating impact on the recipient country. Theoretically, therefore, there
might also be a negative link between FDI and economic growth in
LDCs. Generally speaking, the positive impact of FDI inflows on econom-
ic growth is argued to be conditional on a number of factors, such as the
level of per capita income, human capital, the degree of trade openness
and the depth of the financial market (Herzer et al., 2008: 795–796; also
see Aizenman and Noy, 2006). Despite these potential negative effects,
however, empirical evidence generally suggests that FDI has a positive
impact on economic growth in developing countries. Based on panel
cointegration and causality tests, Basu et al. (2003) found that there is
a bidirectional causality between economic growth and FDI in 23 devel-
oping countries over the period between 1978 and 1996. Basu et al.
(2003) further argued that for relatively open economies causality
runs in both directions, while for relatively closed economies long-run
causality mainly runs from growth to FDI. Nair-Reichert and Weinhold
(2001) have found that FDI on average has a significant and positive im-
pact on economic growth in a sample of 24 developing countries. In an-
other widely cited recent study Carkovic and Levine (2005) have found
that FDI does not exert a significant, positive impact on economic
growth in developing countries. Carkovic and Levine's (2005) study,
however, was based on the unlikely assumption of the homogeneity
on the coefficients of the lagged dependent variables. In a heteroge-
neous panel data context, Hansen and Rand (2006) tested for Granger
causality between FDI and GDP in a sample of 31 developing countries,
finding that FDI has a positive impact on GDP in the long run.9
Most of the existing empirical literature test bivariate causality re-
lations between each pair of GDP, exports and FDI. Recently, a series
of works have examined the relations among the three variables si-
multaneously. Through utilizing Instrumental variable estimation
techniques, Makki and Somwaru (2004) provided evidence for a pos-
itive impact of exports and FDI on economic growth in a sample of 66
developing countries data for the period between 1970 and 2000. In a
panel data analysis, Wang et al. (2004) have found evidence that FDI
is relatively more important for high-income countries, while inter-
national trade is more beneficial to lower income developing coun-
tries. In a panel Granger causality testing framework Hsiao and
Hsiao (2006) found evidence that FDI has unidirectional effects on
GDP, both directly and indirectly through exports, and there also ex-
ists bidirectional causality between exports and GDP for a selected set
of East and Southeast Asian economies. Ahmed et al. (2007) studied
causal links between export, FDI and output for a set of Sub-Saharan
African Countries in a panel cointegration setting. Their findings sup-
port the export-led growth hypothesis for the five Sub Saharan
Borensztein et al. (1998). See Herzer et al. (2008) for a review of theoretical expla-
nations for the FDI-growth relationship.
9
See Chowdhury and Mavrotas (2006), Hansen and Rand (2006), and Lim (2001).
 R.B. Tekin / Economic Modelling 29 (2012) 868–878 871

African countries studied. Ahmed et al. (2007) further found that panel Granger non-causality the equations in the above VAR system
there exist bidirectional Granger causality between FDI and exports are divided into two sets of equations in the following way:
in Ghana, Kenya and Nigeria, while the Granger causality runs from
FDI to exports in South Africa and from exports to FDI in Zambia. ly
X 1
lx
X 1

The authors further provided evidence in favor of a positive causal re- y1;t ¼ α 1;1 þ ϕ1;1;k yi;t−k þ β1;1;k xi;t−k
lation from exports and FDI to income for all five Sub Saharan African k¼1 k¼1
ly lx
countries studied. X1 X 1

y2;t ¼ α 1;2 þ ϕ1;2;k yi;t−k þ β1;2;k xi;t−k ð2Þ

k¼1 k¼1
3. Methodological framework ⋮¼ ⋮
ly lx
X 1 X 1

The Granger causality test is a useful device to determine wheth- yN;t ¼ α 1;N þ ϕ1;N;k yi;t−k þ β1;N;k xi;t−k
er the lags of a variable, say, xit contribute to the better forecasting k¼1 k¼1

of yit when the lagged values of xit are introduced into the regres-
sion of yit on the lagged values of yit. In the panel data context, and
Granger non-causality can be tested by making use of a finite-
order panel VAR model, where a random variable can be expressed X
ly2
X
lx2
as a function of its own past values and past values of other vari- x1;t ¼ α 2;1 þ ϕ2;1;k yi;t−k þ β2;1;k xi;t−k
ables in the system. In a bivariate setting the VAR model takes the k¼1 k¼1

following form: X
ly2
X
lx2
x2;t ¼ α 2;2 þ ϕ2;2;k yi;t−k þ β2;2;k xi;t−k ð3Þ
k¼1 k¼1
⋮¼ ⋮
ly
X
lyi X
lxi X 2 X
lx2

yi;t ¼ α 1;t þ ϕ1;i;k yi;t−1 þ β1;i;k xi;t−1 þ ε1;i;t xN;t ¼ α 2;N þ ϕ2;N;k yi;t−k þ β2;N;k xi;t−k
k¼1 k¼1 k¼1 k¼1
ð1Þ
X
lyi X
lxi
xi;t ¼ α 2;t þ ϕ2;i;k yi;t−1 þ β2;i;k xi;t−1 þ ε 1;i;t
k¼1 k¼1 The systems above are SUR (seemingly unrelated regressions)
models which have different predetermined variables whose only
possible link is contemporaneous correlation within the systems
where index i refers to the country i = 1, …, N, t to the time (Kónya, 2006: 981). With respect to these SUR systems, in country i
period t = 1, …, T and l to the lag length. ε1, i, t and ε2, i, t are assumed there is one-way Granger causality running from X to Y if in Eq. (2)
to be white-noise errors that might be contemporaneously correlat- at least there is one β1, i, k, k = 1, 2, …, lx1that is not zero and all ϕ2, i's
ed for a given country, but not across countries (see Konya, 2006: are zero in Eq. (3). Similarly, there is one-way Granger causality
3). from Y to X if in all β1, i, k's in Eq. (2) are zero and at least there is one
With respect to this system, in country i there is unidirectional ϕ2, i, k, k = 1, 2, …, ly2 that is not zero, in Eq. (3). There is two-way
Granger causality from X to Y if in the first equation not all β1, i's are Granger causality between Y and X if neither all ϕ2, i's nor all β1, i's
zero but in the second all ϕ2, i's are zero, there is unidirectional are zero. Finally, there is no Granger causality between Y and X if all
Granger causality from Y to X if in the first equation all β1, i's are ϕ2, i's and all β1, i's are zero.
zero but in the second not all ϕ2, i's are zero, there is bidirectional The framework presented above is suitable for testing one-period
Granger causality between Y and X if neither all β1, i's nor all ϕ2, i's ahead, direct causality relations in a bivariate setting. Bivariate models
are zero, and there is no Granger causality between Y and X if all β1, for testing Granger causality, however, are tainted with the omittance
i's and ϕ2, i's are zero. of all other potential explanatory variables. We therefore add an exten-
When estimating the VAR model above for testing Granger non- sion of our bivariate models into a trivariate setting, including a third
causality between the variables, however, one should note that the variable into our analysis. We can formulate the following trivariate
correlation between lagged values of dependent variable and the dis- variant of the SUR system presented in Eqs. (2) and (3) as
turbance terms should be taken into account (see Kar et al., 2011:
688-689). In the presence of correlated disturbances with the lagged ly lx lz
X 1 X 1 X 1
values of dependent variable, estimation of Eq. (1) by within estima- y1;t ¼ α 1;1 þ ϕ1;1;k yi;t−k þ β1;1;k xi;t−k þ γ 1;1;k zi;t−k þ ε1;1;t
tor is inconsistent when T and N are small and it is consistent when T k¼1 k¼1 k¼1
ly lx lz
and N tend to infinity. If there is cross-sectional dependence in the X 1 X 1 X 1

data, a first alternative is to use Generalized Method of Moments y2;t ¼ α 1;2 þ ϕ1;2;k yi;t−k þ β1;2;k xi;t−k þ γ 1;1;k zi;t−k þ ε1;2;t ð4Þ
k¼1 k¼1 k¼1
(GMM) estimators, as long as T is big enough. However, the GMM es-
⋮¼⋮
timators can still yield inconsistent parameters, unless slope coeffi- ly lx lz
X 1 X 1 X 1
cients are homogeneous (Pesaran et al., 1999). yN;t ¼ α 1;N þ ϕ1;N;k yi;t−k þ β1;N;k xi;t−k þ γ1;1;k zi;t−k þ ε1;N;t
Homogeneity of slope parameters is a very unlikely assumption in k¼1 k¼1 k¼1
our case, as the LDCs have a high degree of heterogeneity. Again, sim-
,
ilar to many other studies working with annual data we don't work
with a necessarily large time dimension. We therefore cannot pro-
ceed with the GMM estimation methodology for testing possible ly lx lz
X 2 X 2 X 2
Granger causality in our case. x1;t ¼ α 2;1 þ ϕ2;1;k yi;t−k þ β2;1;k xi;t−k þ γ2;1;k zi;t−k þ ε2;1;t
A second alternative for testing Granger causality among exports, k¼1 k¼1 k¼1
ly lx lz
FDI, and economic growth in LDCs is to use the approach developed X 2 X 2 X 2

by Hurlin (2008). This approach can control for heterogeneity in x2;t ¼ α 2;2 þ ϕ2;2;k yi;t−k þ β2;2;k xi;t−k þ γ2;1;k zi;t−k þ ε2;2;t ð5Þ
k¼1 k¼1 k¼1
panel data, which is extremely important in our case, but cannot
⋮¼⋮
properly cope with the problem of cross-sectional dependency. ly
X 2
lx
X 2
lz
X 2
We therefore choose to employ in this paper a third alternative re- xN;t ¼ α 2;N þ ϕ2;N;k yi;t−1 þ β2;N;k yi;t−k þ γ2;1;l zi;t−1 þ ε 2;N;t
cently developed by Kónya (2006). In this new approach for testing k¼1 k¼1 k¼1
872 R.B. Tekin / Economic Modelling 29 (2012) 868–878

and, 3.2. Estimation and testing procedure

X
ly3 lx
X 3
lz
X 3
In this study we employ the SUR estimator proposed by Zellner
z1;t ¼ α 3;1 þ ϕ3;1;k yi;t−k þ β3;1;k xi;t−k þ γ 3;1;k zi;t−k þ ε3;1;t (1962) to estimate the bivariate and trivariate systems presented
k¼1 k¼1 k¼1 above. 12 In each of the bivariate and trivariate models, we introduce
lx lz
X
ly3
X 3 X 3 a linear time trend as a proxy variable that might substitute for
z2;t ¼ α 3;2 þ ϕ3;2;k yi;t−k þ β3;2;k xi;t−k þ γ 3;2;k zi;t−k þ ε3;2;t some variables that are omitted from the original regression models.
k¼1 k¼1 k¼1
⋮¼⋮ The aim is to mitigate the omitted variable bias.
ly
X3 X3
lx lz
X 3
Prior to estimation we first determine the optimal lag structure.
zN;t ¼ α 3;N þ ϕ3;N;k yi;t−k þ β3;N;k xi;t−k þ γ 3;N;k zi;t−k þ ε3;N;t : This is a crucial step as Granger-causality tests are known to be very
k¼1 k¼1 k¼1 sensitive to the choice of the lag length. For specifying the number
ð6Þ of lags in the analysis we follow the procedure proposed in Kónya
(2006: 982-983). That is, for selecting the optimal lag structure, we
Following Kónya (2006), our focus will first be on the pairwise, allow different maximal lags for Y and X (1–4 lags), without allowing
one-period-ahead causal relationships between all pairs of y, x and z them to vary across countries. Thus, in the bivariate case, for example,
in the above bivariate and trivariate systems. In our trivariate sys- we estimate Eqs. (2) and (3) for each possible pair of mly1, mlx1, and
tems, we treat the third variable in each of the equations above as mly2, mlx2, respectively, and select the combinations that minimize
an auxiliary variable that will not be directly involved in the Granger the Akaike and Bayesian Schwartz Criterion (SC).
causality analysis. As a result, our analysis can handle only direct, In the bivariate case, first we estimate Eq. (2) under the null hy-
one-period-ahead causality between variable pairs (X and Y), disre- pothesis that there is no causality from X to Y (i.e. imposing the re-
garding the possibility of indirect causality at longer time horizons. striction β1, i, l = 0 for all i = 1, …, N and l = 1, …, lyi) and by the
Next, we shall study the possibility of two variables jointly causing restricted equation
the third variable. 10
ly
X 1
^ i;j − ϕ
eH0 ;i;t ¼ yi;t −α ^ for i ¼ 1; …; N; t ¼ 1; …; T ð7Þ
1;i;k yi;t−1
3.1. Cross-sectional dependency k¼1

Before estimating the models above, we first test for cross- to obtain the N × T matrix eH0 , each row of which is the time indexed
sectional dependency. Breusch and Pagan (1980) proposes the La- residuals of Eq. (7) (see Kónya (2006: 985-986).
grange Multiplier (LM) statistics Secondly, we re-sample these residuals. In order to preserve
the contemporaneous cross-correlation structure of the error
X
N −1 X
N terms in Eq. (2), we, instead of drawing the residuals for each
CDLM ¼ T ^ 2ij
ρ country one-by-one, randomly select a full column from the eH0
i¼1 j¼iþ1
matrix at a time. We denote the bootstrap residuals as eH0, i, t,
∘
where t ¼ 1; …; T which is practically an extended version of eH0, i, t
∘
for testing cross-sectional dependence. It is well-known that LM test in time length T ≥T.
is valid for T → ∞ with N fixed. Below we assume unbalanced panel Next, we generate a bootstrap sample of Y again under the as-
with invariant size T without loss of generality. 11 sumption that it is not caused by X, i.e.:
2
The test statistic is distributed as χn(n − 1)/2. It is inappropriate in
cases where the N dimension is large. ly
X i ∘
∘ ∘ ∘
A scaled version of Breusch and Pagan (1980) test statistics appli- yi;t ¼ ˆ
α 1;i þ ˆ
ϕ1;i;k yi;t−1 þ eH0 ;1;t ; t ¼ 1; … T ð8Þ
cable if T → ∞ and N → ∞ is defined as k¼1

∘ ∘
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where y1;t is the new series obtained by Eq. (8) and T ≥T is the new
T X
N −1 X
N
CDSLM ¼ ^ 2
ρ time length to be utilized in bootstrapping. Notice that Eqs. (7) and
NðN−1Þ i¼1 j¼iþ1 ij (8) are distinct deterministic difference equations and no estimation
is conducted.
Next, we estimate the following SUR system
which is distributed as standard normal.
Alternatively, Pesaran (2004) proposes the following statistics ly lx
∘ X 1
∘ X 1

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi y1;t ¼ α 1;1 þ ϕ1;1;k yi;t−k þ β1;1;k xi;t−k

2T X
N −1 X
N k¼1 k¼1
CDP ¼ ^
ρ ly lx
N ðN−1Þ i¼1 j¼iþ1 ij ∘ X2
∘ X 2

y2;t ¼ α 1;2 þ ϕ1;2;k yi;t−k þ β1;2;k xi;t−k ð9Þ

k¼1 k¼1

^ ij s rather than ρ
^ 2ij . Under the null hypothesis of ⋮¼ ⋮
based on sums of ρ ly lx
∘ X N
∘ X N
no cross-sectional dependence, the test statistic is asymptotically yN;t ¼ α 1;N þ ϕ1;N;k yi;t−k þ β1;N;k xi;t−k
distributed as standard normal. Pesaran's approach has remarkable k¼1 k¼1
positive qualities in samples of practically all relevant sizes and ∘
remains robust in a variety of settings (see Pesaran, 2004). where t ¼ 1; …; T, which is exactly the same equation system with
Eq. (2), except that we employ the generated variable y∘ instead of y.
10
The concept of causality in the sense of Granger is typically defined in terms of pre- The Wald test statistics implied by the non-causality hypotheses in
dictability of a vector of variables one period ahead (Dufour and Renault, 1998: 1099). Eq. (9) for each single equation are calculated.
In a bivariate system, non-causality for one period ahead implies non-causality at, or By repeating the above procedure 10,000 times and stacking the
up to, longer horizons (see Dufour and Renault, 1998).Also see Kónya (2004) for fur-
Wald test statistics of each single equation in a different vector at
ther implications of this definition of causality.
11
In the case of an unbalanced panel, only complete observations are included; i.e.
12
Tij = min(Ti,Tj), where Ti is the number of observations for individual i. If the panel is The estimations were performed by the SUR routine of TSP4.5. We are grateful to
balanced, Tij = T for each i, j. László Kónya for sharing his TSP routines with us.
 R.B. Tekin / Economic Modelling 29 (2012) 868–878 873

each repetition, we obtain N vectors of dimension 10,000. By selecting country to country. Considering FDI as the auxiliary variable in the
relevant percentiles of these vectors, i.e., the sampling distributions, model, and focusing solely on EXP–GDP relations, we first test for
we get the bootstrap critical values for Wald test statistics. the export-led growth hypothesis. The Granger causality test results
for the null hypothesis that EXP does not cause GDP are shown in
3.2.1. Testing procedure in the trivariate case the first column of Table 1.
This bootstrapping procedure can be extended to a trivariate case by At the 10% level of significance there is no sufficient evidence
including a third variable zi, t.The restricted Eq. (7) in a trivariate case is against the null hypothesis in the case of 16 countries, namely, Ango-
la, Benin, Burkina Faso, Central African Republic, Chad, Gambia, Libe-
ly
X 1 X
lx1 ria, Madagascar, Malawi, Mauritania, Niger, Rwanda, Senegal, Togo,
^ i;j − ϕ
eH0 ;i;t ¼ yi;t −α ^ ^ 1;i;k zi;t−1 for i ¼ 1; …; N; t ¼ 1; …; T
1;i;k yi;t−1 þ γ Yemen and Zambia. We therefore fail to provide empirical support
k¼1 k¼1
for the exports-led growth hypothesis for these countries. For only
ð10Þ
in 2 out of 18 LDCs in the sample – Haiti and Sierra Leone – do we
find evidence that EXP Granger-causes GDP at the 5% significance
under the restriction β1, i, l = 0 for all i = 1,…, N and l = 1, …, lyi.
∘ level. This finding suggests that in these two LDCs there is direct,
Accordingly, yi;t is generated by the equation system of Eq. (4) and
one-period-ahead Granger causality from growth of real merchandise
the analogous of Eq. (8) in trivariate case is
exports to economic growth. It should be noted that UNCTAD classi-
ly
fied Haiti as a manufacturing exporter, while Sierra Leone is a mineral
∘ X i
∘ X
lx1
∘ (diamonds and rutile) exporter (UNCTAD, 2010: xv). 14
^ 1;i þ
yi;t ¼ α ^
ϕ ^ 1;i;k zi;t−1 þ eH ;1;t ; t ¼ 1; … T ∘ : ð11Þ
1;i;k yi;t−1 þ γ
k¼1 k¼1
0
At this level of the analysis it should be noted that the signs of the
regression coefficients involved in the causality tests are crucial for
In order to get the bootstrapped Wald test statistics for the hy- claiming support for the export-led growth hypothesis. In fact, one
pothesis that there is no causality from X and Z jointly to Y, we follow cannot claim support for the export-led growth or the growth-led ex-
the procedure adopted in the bivariate case except that instead of port hypothesis unless the regression coefficients involved in the cau-
Eq. (9) we estimate the following SUR system sality tests are positive as both hypotheses imply positive parameters.
Both in our trivariate and bivariate specifications regression coeffi-
ly lx cients are found to be significant and positive in all cases when cau-
∘ X 1
∘ X 1 X
lx1
y1;t ¼ α 1;1 þ ϕ1;1;k yi;t−k þ β1;1;k xi;t−k þ γ 1;i;k zi;t−1 sality is detected from exports to GDP. We can therefore conclude
k¼1 k¼1 k¼1 that in Haiti and Sierra Leone (and Rwanda) there is evidence for
∘ Xly2
∘
lx
X 2 X
lz2 the export-led growth hypothesis. 15
y2;t ¼ α 1;2 þ ϕ1;2;k yi;t−k þ β1;2;k xi;t−k þ γ 1;i;k zi;t−1 ð12Þ Here, it is remarkable that in Haiti, the only LDC in our sample that
k¼1 k¼1 k¼1
is classified as a manufacturing exporter, the export-led growth hy-
⋮¼ ⋮
ly
X N
lx
X N
lz
X N
pothesis seems to hold true. This might reflect that real exports
∘ ∘
yN;t ¼ α 1;N þ ϕ1;N;k yi;t−k þ β1;N;k xi;t−k þ γ1;i;k zi;t−1 growth is capable of generating real output growth when the case is
k¼1 k¼1 k¼1 that exports are composed mainly of apparel and manufacturing
products. UN reports evaluating the BPoA as well as the country re-
for obtaining the Wald test statistics. ports by the Haiti government to the UN, reflect that this country
has succeeded in establishing a manufacturing exports base over the
4. Data and findings last few decades as well as managing to increase manufacturing ex-
ports considerably during the same time period. More than 70% of
The data set comprises annual measures on 18 least developed the country's exports are composed of apparel and manufactures.
countries: Angola, Benin, Burkina Faso, Central African Republic, The country has been enjoying in the last few decades increasing tar-
Chad, Gambia, Haiti, Liberia, Madagascar, Malawi, Mauritania, Niger, iff free access to the USA which is its primary export destination.
Rwanda, Senegal, Sierra Leone, Togo, Yemen, and Zambia. The selec- Again it should be mentioned that Rwanda, for which we have
tion of this sample of LDCs was made due to data availability. found evidence supporting the export-led growth hypothesis in our
The variables employed in estimations are as follows: real Gross Do- bivariate models, is one of the two services-exporting LDCs in our
mestic Product (GDP), real exports of merchandise (EXP) and real net sample. The Rwandan case is particularly interesting in the sense
foreign direct investment inflows (FDI). Following the previous works, that in recent UN reports on the outcomes of BPoA, Rwanda was
the data are employed in constant US Dollars (year 2000) and constant framed as a success story, as it succeeded in establishing a well-
exchange rates. All data are taken from the UNCTAD statistical database, developed services industry (tourism, information and communica-
available on http://unctadstat.unctad.org. All variables are used in natu- tion technologies) in the last decade. Rwanda is also known to have
ral logarithms. 13 The sample period is 1970–2009 for all countries. made considerable efforts towards the establishment of a well-
functioning liberal market economy and is often shown as an exam-
4.1. Empirical results ple of good governance in LDCs. The third country for which we
have found evidence in support of the export-led growth hypothesis
4.1.1. GDP–EXP and FDI
The results for our first trivariate model, where GDP is the inde- 14
UNCTAD has classified the LDCs into six export specialization categories, namely:
pendent variable, EXP and FDI are the dependent variables, are pro- agriculture, manufacture, mineral, mixed, oil and services. Agricultural exporters are
vided in Table 1. Our first observation is that the bootstrap critical Afghanistan, Benin, Burkina Faso, Guinea-Bissau, Kiribati, Liberia, Malawi, Solomon
values are considerably higher than the chi-square critical values usu- Islands, Somalia, Tuvalu and Uganda. Manufactures exporters are Bangladesh, Bhutan,
Cambodia, Haiti, Lesotho and Nepal. Mineral exporters are Burundi, Central African Re-
ally applied with the Wald test, and that they vary considerably from
public, Democratic Republic of the Congo, Guinea, Mali, Mauritania, Mozambique, Ni-
ger, Sierra Leone and Zambia. Mixed exporters are Lao People's Democratic Republic,
13
Our FDI inflows data includes some negative observations (less than 3% of the total Madagascar, Myanmar, Senegal and Togo. Oil exporters are Angola, Chad, Equatorial
observations) as the result of large profit remittances, particularly in the 2000s. In or- Guinea, Sudan, Timor-Leste and Yemen. Services exporters are Comoros, Djibouti, Eri-
der to include as many observations as possible, we truncate non-positive observations trea, Ethiopia, Gambia, Maldives, Rwanda, Samoa, Sao Tome and Principe, United Re-
to a number very close to zero, and then take the natural logarithm of that value. This is public of Tanzania and Vanuatu.
15
the approach taken by Berger et al. (2010), and Busse et al. (2010) for negative values In our bivariate models, in addition to Haiti and Sierra Leone, we find evidence for
of FDI inflows. export-driven growth in a third LDC, Rwanda, which is classified as a services exporter.
874 R.B. Tekin / Economic Modelling 29 (2012) 868–878

Table 1
^ GDP i;t−1 þ β
^i þ ϕ
Panel Granger causality test results based on bootstrapped Wald statistics for GDP i;t ¼ α ^ EXP i;t−1 þ γ^ FDI i;t−1 þ ^i;t .
i i i

H0, i : βi = 0 H0, i : γi = 0 H0, i : γi = 0, βi = 0

Bootstrap critical values Bootstrap critical values Bootstrap critical values

Test Stat. 10% 5% 0% Test Stat. 10% 5% 0% Test Stat. 10% 5% 0%

Angola 1.906 8.634 12.408 20.977 0.298 6.294 9.948 28.501 1.853 9.791 14.430 27.608
Benin 4.865 4.190 6.068 10.788 10.213*** 2.865 4.245 7.658 6.190 7.251 9.908 17.342
Burkina Faso 1.075 4.059 5.902 10.691 0.362 11.131 16.496 35.046 0.999 3.937 5.819 10.725
C. African R. 1.435 10.781 14.028 22.159 0.029 4.328 5.930 10.956 1.412 10.922 14.561 23.316
Chad 0.326 5.633 8.198 14.477 1.079 6.590 9.524 17.322 0.251 5.337 7.913 14.349
Gambia 0.370 5.765 7.752 12.426 0.065 3.423 4.865 8.134 0.375 5.249 7.164 11.986
Haiti 9.859** 5.059 7.946 15.701 2.063 6.023 8.803 15.754 10.058** 5.132 7.485 13.518
Liberia 6.240 8.055 11.752 21.848 2.019 6.072 8.831 18.443 6.543 7.319 10.567 20.044
Madagascar 0.082 11.835 15.675 24.453 0.951 8.492 11.866 21.237 0.066 11.086 14.537 23.549
Malawi 4.436 4.753 7.022 12.776 0.043 5.739 8.340 16.229 4.407 4.621 6.754 12.265
Mauritania 0.462 5.378 7.907 14.167 1.620 5.681 8.428 16.368 0.381 5.374 7.688 14.090
Niger 2.083 6.906 9.678 17.191 3.021 7.833 11.285 19.495 2.422 5.810 8.335 15.495
Rwanda 11.015 17.085 22.852 38.565 1.632 23.115 32.041 59.912 9.306 14.148 17.935 28.373
Senegal 0.020 9.489 12.475 20.631 0.007 8.878 11.655 19.052 0.019 11.172 14.401 22.280
Sierra Leone 11.984** 8.064 11.643 21.317 0.009 5.599 8.386 16.023 12.182** 6.609 9.720 17.853
Togo 0.608 15.350 19.591 30.883 13.542** 4.512 6.531 13.549 0.869 15.136 19.857 31.305
Yemen 7.490 7.503 9.923 16.153 0.053 17.714 22.770 35.179 7.282 7.579 10.236 16.087
Zambia 1.827 6.279 8.803 16.199 3.476 6.705 9.771 18.109 2.212 5.626 8.243 14.992

Note: ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively.

is Sierra Leone, which is a country with an extremely low degree of
export diversification. More than 60% of exports of Sierra Leone are
composed of diamonds and rutile, and the country's export revenues
crucially depend on prices of these two commodities in international
markets. It should however, be noted that Sierra Leone has been in-
creasing efforts for establishing domestic diamond cutting and design
industries, and for attracting FDI in these industries so that it may be
able to keep as much of the value added within the country.
The second column of Table 1 presents bootstrap critical values
and test statistics for the same trivariate model, when EXP is the aux-
iliary variable in the model, and the focus is on FDI–GDP relations.
When we test for the hypothesis that FDI does not cause GDP we
fail to reject the null hypothesis in all LDCs in the sample, with the ex-
ception of Benin and Togo. We therefore conclude that in these two
countries there is evidence for FDI Granger-causing GDP. Given that
the estimated parameters are positive for both countries, this finding
suggests that net FDI inflows contribute significantly to economic
growth in these LDCS (Here again, it should be noted that there is a
perfect match between bivariate and trivariate model results). For
the remaining 16 LDCs in our sample, we conclude that no Granger
causality exists from FDI inflows to real economic growth, not even
at the 10% significance level.
This finding might be due to the very low levels of FDI inflows as
share of national output in LDCs in general. Of these two countries
for which we have found evidence for Granger-causality from FDI to
GDP, Benin is an agricultural exporter (cotton), and Togo is a mixed
exporter. The shares of FDI in national output in both countries are
typically low, but still higher than the LDC average. Benin's economic
growth performance is above LDC average, while Togo's performance
is below the average during the period of analysis. FDI inflows in
Benin are directed mostly into agricultural products and food proces-
sing sector while the Togon economy is based on export-oriented
production of agricultural commodities such as cotton, coffee, cocoa,
and phosphate mining. The Government of Togo has been exerting ef-
forts to encourage foreign direct investment by increasing the degree
of the country's financial openness in addition to World Bank and IMF
guided economic reforms. Country reports to the UN by Togon gov-
ernment also show that FDI to Togo is mainly directed towards sea-
food and agricultural processing, textile and mining industries.
Finally, we test for the joint hypothesis that EXP and FDI together
do not Granger-cause GDP in the same trivariate system. The third
column of Table 1 presents the Wald statistics and bootstrapped
critical values. Based on these test results we reject the null hypothe-
sis of Granger non-causality for only two countries, namely, Haiti and
Sierra Leone.
4.1.2. EXP–GDP and FDI
The Granger causality test results for our second trivariate model
are given in Table 2. In this model EXP is the independent and FDI
and GDP are the dependent variables
Considering FDI as the auxiliary variable in the system, and focus-
ing solely on the GDP–EXP relations, we can now test for the null hy-
pothesis that GDP does not cause EXP. Wald test statistics and
bootstrapped critical values are provided in the first column of
Table 2. We reject the null hypothesis that GDP does not cause EXP
for 4 LDCs. These countries are Angola, Benin, Chad and Zambia. For
the remaining 14 LDCs in our sample, there is no sufficient evidence
against the null hypothesis of Granger non-causality, not even at the
10% significance level. In all cases, when causality is detected from
GDP to exports, the estimated regression coefficients are positive as
required for claiming support for the growth-led export hypothesis,
except in the case of Benin, for which we have found a negative coef-
ficient. We can therefore conclude that in Angola, Chad and Zambia,
there is evidence in support of the growth-led export hypothesis. It
should be noted that our findings are robust to model specification;
both the bivariate and trivariate model test results suggest that the
growth-led exports hypothesis cannot be rejected in Angola, Chad
and Zambia.
Interestingly, two (Angola and Chad) of the three countries for
which we have found evidence in support of the growth-led export
hypothesis are classified as oil exporters by UNCTAD. The third coun-
try, Zambia, is also a natural resource (copper) exporting LDC, which
is classified as a mineral exporter by UNCTAD. This finding might re-
flect the fact that the economic growth of these natural resource rich
LDCs is strongly correlated with the world growth rates, and the rise
in exports is associated with the higher world demand for oil and key
mineral resources during boom periods. Again, in this group of
countries, the cumulative causation process could be working in a
stronger and more efficient way. The fact that Angola and Chad are
the largest oil exporting countries in the group of LDCs whose domes-
tic economies are almost totally dependent on a single (energy) sec-
tor is supportive of the above potential explanations. In the case of the
Zambian economy, which is almost totally dependent on copper and
minerals extraction industry, the two explanations noted above also
 R.B. Tekin / Economic Modelling 29 (2012) 868–878 875

Table 2
^ EXP i;t−1 þ β
^i þ ϕ
Panel Granger causality test results based on bootstrapped Wald statistics for EXP i;t ¼ α ^ GDP i;t−1 þ γ
^ i FDIi;t−1 þ ^i;t .
i i

H0, i : βi = 0 H0, i : γi = 0 H0, i : γi = 0, βi = 0

Bootstrap critical values Bootstrap critical values Bootstrap critical values

Test Stat. 10% 5% 0% Test Stat. 10% 5% 0% Test Stat. 10% 5% 0%

Angola 13.447** 5.927 8.552 15.172 2.900 5.844 8.730 16.652 13.674** 5.742 8.332 15.706
Benin 10.765** 6.885 9.938 18.973 19.363*** 6.643 9.725 18.996 10.902** 6.286 9.118 17.115
Burkina Faso 4.650 4.872 7.082 12.338 0.030 7.506 11.556 23.327 4.627* 4.432 6.465 11.072
C. African R. 0.394 6.530 9.280 17.983 0.234 6.414 9.262 17.216 0.398 6.611 9.625 17.507
Chad 12.342** 7.639 11.030 20.019 2.824 5.855 8.446 15.788 12.321** 8.037 11.876 22.021
Gambia 4.467 6.938 10.422 20.086 0.393 8.127 11.487 19.312 4.451 7.215 10.621 19.521
Haiti 4.261 5.819 8.609 16.272 6.416** 3.018 4.388 8.019 4.077 9.271 13.398 24.141
Liberia 0.146 5.800 8.628 16.755 2.950 7.199 10.771 19.547 0.081 5.859 8.529 15.664
Madagascar 4.478 6.080 8.440 14.325 0.396 4.819 7.011 12.888 4.491 5.938 8.335 14.783
Malawi 13.572 20.078 26.083 41.191 0.000 3.397 4.948 8.675 13.586 19.739 25.268 39.597
Mauritania 0.033 5.740 8.312 15.913 8.631* 7.859 11.964 24.608 0.021 5.991 8.883 16.358
Niger 5.970 6.012 8.648 15.147 10.227* 7.348 10.627 19.772 6.201* 5.913 8.716 15.640
Rwanda 1.402 7.573 11.284 20.885 0.029 22.410 28.962 46.989 1.492 7.845 11.091 20.898
Senegal 0.858 3.224 4.693 8.560 5.309 7.170 10.620 20.404 0.834 3.363 4.898 8.552
SierraLeone 0.200 9.581 13.653 24.360 1.700 6.791 9.914 18.849 0.214 8.276 12.098 22.889
Togo 0.381 8.119 11.804 21.532 5.925* 5.387 8.005 16.319 0.359 7.932 11.210 21.159
Yemen 1.779 7.414 10.513 19.086 6.977 9.889 13.708 25.873 1.931 7.416 10.908 20.290
Zambia 30.702*** 6.354 9.152 17.491 0.025 5.403 7.970 15.752 30.634*** 6.202 9.209 16.418

Note: ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively.

seem to be appropriate. 16 The second column of Table 2 summarizes
the test statistics and bootstrapped critical values for the same trivari-
ate model when GDP is the auxiliary variable in the system and the
focus is exclusively on FDI–EXP relations. The Granger causality test
results for the null hypothesis “FDI does not cause EXP” suggest that
in the 5 LDCs in our sample there is evidence for direct, one-period-
ahead, unidirectional Granger-causality from FDI to EXP. These coun-
tries are Benin, Haiti, Mauritania, Niger and Togo. At the 10% signifi-
cance level we fail to find sufficient evidence against the null
hypothesis for the remaining 13 countries in our sample. We there-
fore conclude that in the remaining LDCs, growth in real net FDI in-
flows do not Granger-cause real growth in exports of merchandise,
suggesting that real net FDI inflows do not considerably add to (or
hinder) the export performance of these countries. This finding
might reflect that the international competitiveness of these coun-
tries' export products does not increase following FDI, contrary to
the expectations of the UN PoAs.
Theoretical literature suggests that trade and FDI can be related
positively or negatively; the sign of the relation is dependent on the
characteristics of the recipient country, types of industries, the nature
of FDI, and objectives of foreign investors. Institutional factors such as
political and economic stability, transparency regulations and corrup-
tion are amongst the factors that determine foreign direct investment
inflows in developing countries (Ucal et al., 2010: 21). What is crucial
is whether FDI substitutes trade, or complements trade, facilitating
exports of the recipient country. According to Kojima (1973, 1975),
FDI inflows might be “trade-oriented” or “anti-trade oriented”,
depending on the objectives of foreign investors in the first instance.
As long as FDI is “market-seeking”, trade and FDI are expected to be
negatively related; while, on the other hand, the two variables are
positively related if FDI is of “efficiency-seeking” nature (Gray,
1998). In addition to the nature and objectives of FDI inflows, the
characteristics (similarity and differences between the recipient and
investing countries) also matter. Markusen and Venables (1998), for
example, argue that trade and FDI are positively related among asym-
metric countries, and negatively among symmetric countries. We
should therefore consider the signs of the estimated coefficients
16
About 64% of Zambian exports consist of copper and cobalt. Privatization of
government-owned copper mines in Zambia during the 1990s and foreign direct in-
vestment have greatly increased productivity and output in copper mining. As a result
of soaring copper prices in the last decade, Zambia enjoyed a steady economic growth.
carefully in order to have a more complete picture of the FDI-
exports nexus in LDCs.
When the findings of our causality tests are considered in light of the
signs of coefficient parameters, we can make the following observations
on the FDI-exports nexus. The regression coefficients are positive in all
cases we considered, except Benin, indicating that there exists a positive
relation between FDI and EXP in Haiti, Mauritania, Niger and Togo. Our
causality tests, therefore, suggests that in these three LDCs FDI Granger-
causes EXP, suggesting that real net FDI inflows generate real growth in
total exports of merchandise. We can therefore conclude that only in
the case of Benin, (and in Chad and Yemen, according to bivariate
model results) FDI is market-seeking, and has a trade-substituting ef-
fect. In the case of Haiti, which is a manufacturing exporter, our test re-
sults suggest that growth in real net FDI inflows accelerate real export
growth, probably as a result of a productivity rise, increasing interna-
tional competitiveness of Haiti's manufacturing exports through the
transfer of technology and know-how, as expected by the three succes-
sive PoAs of the United Nations.17
In the case of Mauritania, which is a natural resource-rich, mineral
exporting country, on the other hand, our findings suggest that FDI is
trade-oriented. Mauritania has rich deposits of iron ore that account
for about 40% of its total exports, in addition to gold and copper.
The positive causality relation found between FDI inflows and exports
might therefore suggest that FDI is primarily oriented towards the
extraction and expropriation of the recipient country's mineral
resources. Increased productivity in mining industries and strength-
ening of the country's competitive power in global markets might
be other alternative plausible explanations. The same reasoning also
seems to be appropriate in the case of Niger, which is an extremely
poor, landlocked nation with one of the world's richest uranium
deposits.
When we test for the hypothesis that FDI and GDP jointly do not
Granger-cause EXP, we reject the hypothesis for a total of 6 LDCs:
Angola, Benin, Burkina Faso, Chad, Niger and Zambia. The Wald test
statistics and bootstrapped critical values are provided in the third
column of Table 2. The regression coefficients are all found to be pos-
itive, suggesting that rising real net FDI inflows and real growth in do-
mestic output might together contribute positively to the export
performance of these LDCs.
17
See Frobel (1980) and Dunning (1981), among others, for alternative theoretical
explanations of export-oriented FDI flows.
876 R.B. Tekin / Economic Modelling 29 (2012) 868–878

Table 3
^ FDI i;t−1 þ β
^i þ ϕ
Panel Granger causality test results based on bootstrapped Wald statistics for FDI i;t ¼ α ^ EXP i;t−1 þ γ^ GDP i;t−1 þ ^i;t .
i i i

H0, i : βi = 0 H0, i : γi = 0 H0, i : γi = 0, βi = 0

Bootstrap critical values Bootstrap critical values Bootstrap critical values

Test Stat. 10% 5% 0% Test Stat. 10% 5% 0% Test Stat. 10% 5% 0%

Angola 2.325 6.689 9.852 18.737 1.163 7.064 9.875 17.048 0.164 7.337 11.079 20.056
Benin 2.002 3.621 5.246 9.856 5.972 7.512 10.714 19.801 5.276 8.465 12.409 22.627
Burkina Faso 1.556 3.674 5.381 10.086 3.594 5.422 7.670 13.061 5.086 6.555 9.459 16.331
C. African R. 3.064 3.849 5.476 9.982 5.963 7.821 10.973 20.317 4.759 8.479 12.425 21.1
Chad 3.024 5.623 7.794 14.244 0.268 8.045 11.491 20.876 0.004 8.433 12.024 20.825
Gambia 0.573 5.806 7.883 12.931 8.591* 7.763 11.488 21.272 8.715* 8.530 12.286 23.316
Haiti 7.644** 3.999 5.768 10.604 2.981 7.691 11.130 20.830 1.035 10.002 13.832 24.498
Liberia 0.240 6.705 9.709 17.725 5.536 7.896 11.444 23.070 2.618 7.290 10.497 19.830
Madagascar 26.520*** 5.495 7.739 14.062 11.675** 4.815 6.896 12.128 4.459 9.072 13.109 22.946
Malawi 4.407 5.665 7.540 12.137 0.876 20.116 25.842 40.790 4.324 7.149 10.066 17.674
Mauritania 10.546** 4.777 6.752 11.708 0.801 4.097 5.977 11.361 0.001 7.698 11.840 23.263
Niger 4.930 5.155 7.435 12.982 1.843 6.608 9.833 16.889 3.602 7.765 11.220 18.679
Rwanda 11.107** 4.364 6.257 11.213 0.413 5.893 8.873 18.307 4.062 7.466 11.661 29.312
Senegal 13.547** 4.724 7.187 13.824 1.043 3.493 4.880 8.464 0.032 7.167 10.188 19.269
SierraLeone 0.174 6.146 8.957 16.020 0.058 8.010 11.569 21.606 0.031 9.025 13.288 24.764
Togo 5.200 7.314 10.150 18.704 4.271 6.803 9.828 18.047 3.183 9.368 13.627 25.670
Yemen 0.079 5.755 8.536 16.134 6.412 7.850 11.557 20.991 4.050 8.548 12.475 23.348
Zambia 2.421 5.931 8.256 14.575 3.303 6.342 9.346 16.728 3.081 6.452 9.454 18.579

Note: ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively.

4.1.3. FDI–GDP and EXP
The empirical results for our third trivariate model that takes FDI
as the independent variable and GDP and EXP as the dependent vari-
ables are provided in Table 3 above.
We first take the GDP as the auxiliary variable in the system in order
to be able to test the reverse causality direction in the FDI–export nexus.
The Wald test statistics and bootstrapped critical values for the null hy-
pothesis that EXP does not cause FDI are provided in the first column of
Table 3. Test results suggest that when GDP is the auxiliary variable in
the system we reject the null hypothesis that EXP does not cause FDI
for only 5 of the LDCs. These countries are Haiti, Madagascar, Maurita-
nia, Rwanda and Senegal. For the remaining countries in our data set
we thus fail to provide empirical support for Granger causality from
EXP to FDI. This finding is similar to our bivariate model results in
which we have also found evidence supporting EXP Granger-causing
FDI for the same LDCs with the exception of Haiti (Madagascar,
Mauritania, Rwanda and Senegal) in addition to two other countries,
(Malawi and Zambia).
The signs of estimated coefficients are found to be positive for
Haiti and Rwanda, and negative for Madagascar, Mauritania and Sen-
egal (this finding is again in parallel with our bivariate model results).
We can therefore conclude that in the case of Rwanda and Haiti, EXP
Granger-cause FDI, suggesting that real growth in exports of mer-
chandise attract more FDI in these particular LDCs (which are services
and manufacturing exporters, respectively).
Next, in this third trivariate model we take EXP as the auxiliary
variable in the system, and focus exclusively on GDP–FDI relations.
The Granger causality test results for the null hypothesis that GDP
does not cause FDI are shown in the second column of Table 3. Test
results show that there is one-way, direct, one-period-ahead Granger
causality from GDP to FDI in only 2 countries; namely, Gambia and
Madagascar, at the 10 and 5% significance levels, respectively. For
the remaining countries in our sample, we fail to find sufficient evi-
dence against the null hypothesis of no-Granger causality. It should
be noted that Gambia is classified as a services exporter by the
UNCTAD while Madagascar is a mixed exporter. Gambia is one of
the LDCs that received the highest FDI inflows as share of national in-
come during the last two decades. 18
18
In addition to these two countries, in our bivariate models we reject the null of no
causality for Burkina Faso and Malawi. The regression parameter is positive for Burkina
Faso and negative for Malawi. Both Burkina Faso and Malawi are agricultural exporters.
FDI into Madagascar has been on rise since the early 1990s. Mad-
agascar has a very open investment regime, with very low corporate
and property tax rates, no restrictions on profit remittances or foreign
ownership of land. The Government of Madagascar has recently re-
vised the Investment Code and found an investment promotion agen-
cy in order to further improve the country's investment climate. Most
of the FDI goes into sectors such as textile and apparel, agro-
industries and wood products.
This finding might be the result of the very low levels of FDI in
LDCs in general. However, this finding might also equally reflect
that FDI inflows to LDCs do not target the domestic markets in the
first instance, as countries in this group have too narrow domestic
markets. This might be the reason behind our failure to provide
strong evidence in support of a significant causality link that goes
from economic growth to FDI inflows. This observation is in confor-
mance with theoretical expectations and previous empirical works
on FDI towards developing countries and LDCs. These works claim
that FDI towards this group of underdeveloped countries has different
objectives, such as access to, or exploitation of, strategic mineral and
energy resources. However, it should be noted that the signs of the
estimated coefficients are found to be negative for Gambia, and posi-
tive for Madagascar (as it was in our bivariate models). This finding
suggests that real economic growth enables Madagascar to attract
more real net FDI inflows; while, in Gambia FDI inflows accelerate
during periods of economic slowdown.
Finally, we test for the joint hypothesis that FDI and GDP together
do not Granger-cause EXP in the same trivariate model. According to
the Wald test statistics and critical values provided in the third col-
umn of Table 3, we cannot deny the null hypothesis of no-Granger
causality for only one country, Gambia.
5. Concluding remarks
This study examined causality relations among real GDP, real ex-
ports and real net FDI inflows among Least Developed Countries by
making use of a newly developed Granger causality testing procedure
for panel data sets. We have found general evidence for a multiplicity
of causality relations among the variables considered. In only two of
the LDCs in our sample, namely the Central African Republic and Libe-
ria, we failed to find evidence for causality in any directions (Table 4).
All of the causality relations identified in this study are found to be
 R.B. Tekin / Economic Modelling 29 (2012) 868–878
unidirectional, that is, causality is from one variable to the other, and
not vice versa, with the exception of two cases in which we have
found a bidirectional causality.
The findings of our empirical test of the export–growth nexus are
equally supportive of the export-led growth and the growth-led export
hypotheses in the case of Least Developed Countries. We identified only
three cases in this study in which there is evidence of real export growth
Granger-causing real growth in GDP, while, similarly, there are three
cases in which real GDP growth is found to be Granger-causing real ex-
port growth. Furthermore, we observed that the economic structure of
LDCs seems to be the primary determinant of the direction of causality
between exports and economic growth. Countries for which we have
found evidence supporting the growth-led exports hypothesis are rich
in oil or other natural resources. Interestingly, two of the three cases
for which we have found evidence in support of the export-led growth
hypothesis are manufacturing or services exporters. For the only
manufacturing exporter in our sample (Haiti) and one of the two ser-
vices exporters (Rwanda), we have found evidence in support of the
export-led growth hypothesis. When these two findings are considered
simultaneously, it can be argued that there is a clear divide between dif-
ferent types of LDCs. This observation suggests important policy pre-
scriptions; for those countries which are capable of competing in
international markets for manufacturing products or services, exports
seems to be a true source of economic growth. The international com-
munity, therefore, should focus more on supporting export-
orientation through aid-for-trade, trade-capacity building schemes
and other types of policies in order to support the development of
export-oriented manufacturing industries in such LDCs. Again, Official
Development Assistance and other sources of financial support in the
case of these countries might be directed at supporting manufacturing
or services exports. Supporting the internationalization of manufactur-
ing firms and the services sector in these LDCs can also be an effective
policy in enhancing their economic growth. For agricultural exporters,
mineral exporters, natural resources, and particularly oil-rich LDCs, on
the other hand, export revenues do not seem to contribute significantly
to (and Granger-cause) real economic growth. In this group of LDCs, it
seems plausible that domestic growth mechanisms are more significant
in explaining export growth, rather than the reverse. The international
development assistance programs should therefore more address the do-
mestic growth mechanisms (including investment in human capital, pri-
mary and secondary education, good governance and macroeconomic
Table 4
Summary for the directions of causality.
Countries Granger causality directions
Angola GDP↠EXP, GDP|FDI→EXP, GDP&FDI→EXP
Benin FDI⇀EXP, GDP|FDI→EXP, FDI|GDP↠EXP, GDP&FDI→EXP,
FDI→GDP
Burkina Faso GDP&FDI→EXP, GDP⇀FDI
C. African R.
Chad GDP⇀EXP, FDI⇀EXP, GDP|FDI→EXP, GDP&FDI→EXP
Gambia GDP⇀FDI, GDP|EXP⇀FDI, GDP&EXP→FDI
Haiti FDI→EXP, FDI|GDP→EXP, EXP⇀GDP, EXP|FDI→GDP,
EXP&FDI→GDP, EXP|GDP→FDI
Liberia
Madacasgar GDP|EXP→FDI, EXP|GDP↠FDI, EXP → FDI
Malawi GDP→FDI, EXP→FDI
Mauritania FDI⇀EXP, EXP⇀FDI, FDI|GDP⇀EXP, EXP|GDP→FDI
Nijer FDI|GDP⇀EXP, GDP&FDI→EXP
Rwanda EXP→GDP, EXP↠FDI, EXP|GDP→FDI, EXP|GDP→FDI
Senegal EXP→FDI, EXP|GDP→FDI
Sierra Leone EXP→GDP, EXP|FDI→GDP, EXP&FDI→GDP
Togo FDI⇀EXP, FDI|GDP⇀EXP, FDI→GDP, FDI|EXP→GDP,
Yemen EXP⇀FDI
Zambia GDP↠EXP, GDP|FDI↠EXP, GDP&FDI→EXP, EXP→FDI
Notes: ↠, →, and ⇀ stand for 1%, 5%, and 10% levels of significances consequently. X|Y →Z
means “X Granger causes Z while Y is the auxiliary variable”, X & Y → Z means “X and Y
jointly cause Z”.
877
policies etc.), as opposed to promoting trade orientation in this group of
LDCs.
When it comes to the second nexus examined in this study, our
major finding is that real net Foreign Direct Investment inflows con-
tribute significantly to real economic growth in only a very few num-
ber of LDCs. A probable reason for this might be the very low levels of
FDI in national incomes of this group of countries. Although FDI in-
flows to LDCs have been constantly on rise during the last two de-
cades, the shares of FDI in GDP remain insignificant in these
countries. Furthermore, FDI inflows to LDCs are very highly concen-
trated in a couple of countries, mainly the largest LDCs, as well as
the resource-rich, mineral or oil exporting countries. Despite the
many efforts by LDC governments and the emphasis successive UN
PoAs have placed on the role of FDI in promoting growth, it seems
that in any of LDCs, real net FDI reached beyond a significant thresh-
old. What seems to be noteworthy in our empirical study on FDI-
growth nexus is that growth in real net FDI inflows is found to have
a positive impact on real economic growth in all cases in which we
detected a causality relation in the Granger sense. We did not identify
any case in which FDI inflows had a significant negative impact on
real GDP. This finding is important in the sense that it suggests that
inward FDI does not have a “crowding out” effect on national output
in the case of LDCs. The international community, therefore, should
accelerate efforts to direct more FDI to LDCs and make FDI reach to
a significant share of GDP in this group of countries.
This study has also provided many insights on the third set of cau-
sality relations considered, that is, the FDI-exports nexus. Test results
suggest that only in two countries (Haiti and Rwanda) real exports
growth enhances (and Granger causes) growth in real net FDI in-
flows. We can therefore claim that in the case of manufacturing and
services exporters, better export performance is a way to attract
more FDI in conformance with theoretical expectations. In the
remaining LDCs however, there is no evidence in support of real ex-
port growth having a positive impact on real net FDI received. This
study has also found evidence that in the case of five LDCs (Benin,
Haiti, Mauritania, Niger and Togo) the reverse causation direction
(from FDI to exports growth) seems to hold true. This finding sug-
gests that FDI has a trade-complementing nature in the case of
LDCs; FDI inflows are not market-seeking and do not have a trade-
substituting impact. In the view of these empirical observations, the
international community and domestic governments should continue
to search for ways to encourage more FDI towards LDCs, particularly
in the case of manufacturing and services exporters. It should howev-
er be noted that the overall evidence in support of FDI/EXP relations,
is weak, potentially due to the very low levels of FDI to LDCs.
This study however, shows that in the case of natural resource rich, oil
and mineral exporters, although FDI might enhance export performance,
the growth in real exports is not accompanied by economic growth. The
famous resource curse seems to be striking a blow on these LDCs.
References
Ahmad, J., Kwan, A.C., 1991. Causality between exports and economic growth: empir-
ical evidence from Africa. Economics Letters 37, 243–248.
Ahmed, A., Cheng, E., Messinis, G., 2007. Causal links between export, FDI and output:
evidence from sub-Saharan African countries. Working Paper No. 35, Centre for
Strategic Economic Studies Victoria University.
Aizenman, J., Noy, I., 2006. FDI and trade — two way linkages? The Quarterly Review of
Economics and Finance 46 (3), 317–337.
Bahmani-Oskooee, M., Alse, J., 1993. Export growth and economic growth: an applica-
tion of cointegration and error correction modelling. Journal of Developing Areas
27, 535–542.
Bahmani-Oskooee, M., Economidou, C., 2009. Export led growth vs. growth led ex-
ports: LDCs experience. The Journal of Developing Areas 42 (2), 179–209.
Bahmani-Oskooee, M., Mohtadi, H., Shabsigh, G., 1991. Exports, growth and causality in
LDCs: a re-examination. Journal of Development Economics 36, 405–415.
Basu, P., Chakraborty, C., Reagle, D., 2003. Liberalisation, FDI, and growth in developing
countries: a panel cointegration approach. Economic Inquiry 51, 510–516.
Berger, A., Busse, M., Nunnenkamp, P., Roy, M., 2010. More stringent BITs, less ambig-
uous effects on FDI? Not a bit! WTO Staff Working Paper.
878 R.B. Tekin / Economic Modelling 29 (2012) 868–878
Borensztein, E.J., De Gregorio, J., Lee, J.W., 1998. How does foreign direct investment af-
fect economic growth. Journal of International Economics 45, 115–135.
Breusch, T., Pagan, A., 1980. The LM test and its applications to model specification in
econometrics. Review of Economic Studies 47, 239–254.
Busse, M., Königer, J., Nunnenkamp, P., 2010. FDI promotion through bilateral invest-
ment treaties: more than a bit? Review of World Economics 146 (1), 147–177.
Carkovic, M., Levine, R., 2005. Does foreign direct investment accelerate economic
growth? In: Moran, H., Graham, E.M. (Eds.), Does Foreign Direct Investment Pro-
mote Development? Institute for International Economics, Washington, D.C.
Chowdhury, A., Mavrotas, G., 2006. FDI and growth: what causes what? World Econo-
my 29 (1), 9–19.
Dufour, J.M., Renault, E., 1998. Short run and long run causality in time series: theory.
Econometrica 66 (5), 1099–1125.
Dunning, J.H., 1981. International Production and the Multinational Enterprise. Allen
and Unwin, London.
Frobel, F., 1980. The New International Division of Labor: Structural Unemployment in
Industrialized Countries and Industrialization in Developing Countries. Cambridge
University Press, New York.
Giles, J.A., Williams, C.L., 2000. Export-led growth: A survey of the empirical literature
and some non-causality results. Part 1. Journal of International Trade and Econom-
ic Development 9 (3), 261–337.
Gray, H.P., 1998. International trade and foreign direct investment: the interface. In:
Dunning, J.H. (Ed.), Globalization, Trade and Foreign Direct Investment. Elsevier,
Oxford, pp. 19–27.
Hansen, H., Rand, J., 2006. On the causal links between FDI and growth in developing
countries. The World Economy 29, 21–41.
Herzer, D., Klasen, S., Nowak-Lehmann, F., 2008. In search of FDI-led growth in devel-
oping countries: the way forward. Economic Modelling 25, 793–810.
Hsiao, F., Hsiao, M.C., 2006. FDI, exports, and GDP in East and Southeast Asia — panel
data versus time-series causality analyses. Journal of Asian Economics 17 (2006),
1082–1106.
Hurlin, C., 2008. Testing for Granger Non Causality in Heterogeneous Panels, Mimeo.
Department of Economics: University of Orleans.
Hutchinson, M., Singh, N., 1992. Exports, non-exports and externalities: a Granger cau-
sality approach. International Economic Journal 6, 79–94.
Jung, W., Marshall, P., 1985. Exports, growth and causality in developing countries.
Journal of Development Economics 18, 1–12.
Kaldor, N., 1967. Strategic Factors in Economic Development, New York State School of
Industrial and Labour Relations. Cornell University, Ithaca, NY.
Kar, M., Nazlıoğlu, Ş., Ağır, H., 2011. Financial development and economic growth
nexus in the MENA countries: bootstrap panel granger causality analysis. Econom-
ic Modelling 28 (1-2), 685–693.
Kojima, K., 1973. A macroeconomic approach to foreign direct investment. Hitotsubashi
Journal of Economics 14 (1), 1–21.
Kojima, K., 1975. International trade and foreign investment: substitutes or comple-
ments. Hitotsubashi Journal of Economics 16, 1–12.
Kónya, L., 2004. Export-led growth, growth-driven export, both or None? Granger cau-
sality analysis on OECD countries. Applied econometrics and international devel-
opment,. Euro-American Association of Economic Development 4 (1), 73–94.
Kónya, L., 2006. Exports and growth: Granger causality analysis on OECD Countries
with a panel data approach. Economic Modelling 23, 978–992.
Kugler, P., Dridi, J., 1993. Growth and exports in LDCs: a multivariate time series study.
International Review of Economics and Business 40, 759–767.
Lim, E.-G., 2001. Determinants of, and the relation between, foreign direct investment
and growth: a summary of the recent literature. IMF Working Paper No. 175.
Makki, S.S., Somwaru, A., 2004. Impact of foreign direct investment and trade on eco-
nomic growth: evidence from developing countries. American Journal of Agricul-
tural Economics 86, 795–801.
Markusen, J., Venables, J., 1998. Multinational firms and the new trade theory. Journal
of International Economics 46, 183–203.
Nair-Reichert, U., Weinhold, D., 2001. Causality tests for cross-country panels: a new
look at FDI and economic growth in developing countries. Oxford Bulletin of Eco-
nomics and Statistics 63, 153–171.
Pesaran, M.H., 2004. General diagnostic tests for cross section dependence in panels.
CESifo Working Paper 1229; IZA Discussion Paper 1240. [970].
Pesaran, M.H., Shin, Y., Smith, R.J., 1999. Pooled mean group estimation of dynamic het-
erogeneous panels. Journal of the American Statistical Association 94 (446),
621–634.
Reppas, P., Christopoulos, D., 2005. The export–output growth nexus: evidence from
African and Asian countries. Journal of Policy Modeling 27, 929–940.
Singh, T., 2010. Does international trade cause economic growth? A survey. The World
Economy. doi:10.1111/j.1467-9701.2010.01243.
Stolper, W.F., 1947. The volume of foreign trade and the level of income. Quarterly
Journal of Economics 61 (2), 285–310.
Ucal, M., Metin Özcan, K., Bilgin, M.H., Mungo, J., 2010. Relationship between financial
crisis and foreign direct investment in developing countries using semiparametric
regression approach. Journal of Business Economics and Management 11 (1),
2029–4433.
UN, 2011. Ten-year appraisal and review of the implementation of the Brussels Pro-
gramme of Action for the Least Developed Countries for the Decade 2001-2010,
A/66/66, Report of the Secretary-General, http://www.un.org/wcm/webdav/site/
ldc/shared/documents/Ten-Year%20appraisal.pdf (01.09.2011).
UNCTAD, 2010. The Least Developed Countries Report 2010: Towards a New Interna-
tional Development Architecture for LDCs. United Nations, Geneva and New York.
Wang, C., Liu, S., Wei, Y., 2004. Impact of openness on growth in different country
group. World Economy 27 (4), 567–585.
Zellner, A., 1962. An efficient method of estimating seemingly unrelated regressions
and tests for aggregation bias. Journal of the American Statistical Association 57,
348–368.