# Oxford University Press 1999

Oxford Economic Papers 51 (1999), 133151

133

Foreign direct investment-led growth:

evidence from time series and panel data
By Luiz R. de Mello, Jr.
Department of Economics, Keynes College, University of Kent, Canterbury, Kent,
CT2 7NP
This paper estimates the impact of foreign direct investment (FDI) on capital
accumulation, and output and total factor productivity (TFP) growth in the recipient
economy. Time series and panel data evidence are provided for a sample of OECD
and non-OECD countries in the period 197090. Although FDI is expected to
boost long-run growth in the recipient economy via technological upgrading and
knowledge spillovers, it is shown that the extent to which FDI is growth-enhancing
depends on the degree of complementarity and substitution between FDI and domestic
investment.

1. Introduction
Recent developments in growth theory have been primarily theoretical, although
signicant progress has also been made in growth empirics. The former proposes
the endogenisation of technological change, and hence output growth, through
mechanisms that prevent the unbounded decline in the marginal productivity of
capital in the course of the accumulation process. The latter developments are
concerned chiey with the estimation of cross-country and time series growth
equations, and the methodology used is based on standard growth accounting,
pioneered by Solow (1957) and Denison (1962, 1967).
In general, the search for the keys to economic growth has been arduous in the
recent literature, particularly since De Long and Summers' (1991, 1992) study of
the impact of capital equipment accumulation on output growth. Empirical work
on cross-country and time series growth has been directed at dealing with two basic
problems; namely, the lack of unconditional convergence of growth rates across
countries and high estimates of the elasticity of output with respect to capital
stocks. Although conventional neo-classical growth in the Solovian tradition predicts that the elasticity of output with respect to capital should be equal to the
capital share in output, cross-country estimates point to a much higher value. High
capital elasticities can nevertheless be explained on the grounds of simultaneity and
omitted variable biases. As for the absence of unconditional convergence, in the
sense of SolowCassKoopmans, the problem has been dealt with by the advent of
conditional convergence (Barro and Sala-i-Martin, 1992), once the variables affecting individual countries' steady states have been controlled for.

134

foreign direct investment-led growth

A recent approach in growth empirics is concerned with estimating growth

equations using panel data (Islam, 1995; Blomstrom et al., 1996), on the grounds
that much of the difculties encountered in cross-country estimations can be eliminated by correcting for country-specic differences in technology, production,
institutions, culture, and socio-economic factors, which are expected to evolve
through time. Hence, the time series dimension of cross-country growth processes
can be acknowledged as much as unobservable cross-country effects that can be
deemed to affect growth in the long-run. The problem of autocorrelation in time
series analysis and heteroscedasticity in the estimation of cross-section equations
can also be minimised (Hsiao, 1986; Matyas and Sevestre, 1996).
The focus of this paper is on cross-border externality-generating investment and
knowledge transfers. The impact of cross-border externalities on growth deserves
more careful examination than the traditional closed-economy growth-theoretic
approach and international trade models have permitted (Grossman and Helpman,
1991). Primarily, the impact of foreign direct investment (FDI) on growth is
expected to be twofold.1 First, through capital accumulation in the recipient economy, FDI is expected to be growth-enhancing by encouraging the incorporation of
new inputs and foreign technologies in the production function of the recipient
economy. Second, through knowledge transfers, FDI is expected to augment the
existing stock of knowledge in the recipient economy through labour training and
skill acquisition, on the one hand, and through the introduction of alternative
management practices and organisational arrangements, on the other. Also, FDI
can be expected to promote technological upgrading, and hence diffusion, even
without signicant physical capital accumulation; in the case of, for instance,
start-up, marketing, and licensing agreements; management contracts; and joint
ventures in general (de Mello and Sinclair, 1995). As a result, foreign investors may
increase productivity in the recipient economy and FDI can be deemed to be a
catalyst for domestic investment and technological progress.
It can also be argued that the absorptive capacity of the host country affects the
volume and type of FDI inows. The latter depends on institutional factors, such as
the recipient economy's trade regime, legislation, and political stability; and scale
factors, such as balance of payments constraints and the size of the domestic market
for the goods produced via FDI. The consideration of such (sometimes unobservable) country-specic effects, given new FDI-related and existing domestic production possibilities, and their evolution over time, allows for the examination of such
FDI-driven cross-country or region-specic externalities or spillovers.
The paper is organised as follows. Section 2 summarises the expected impact of
FDI on growth. Section 3 introduces the empirical analysis by focusing on the time
series aspects of FDI and growth. Section 4 complements Section 3 by concentrating on cross-sectional evidence provided by panel data estimations. Section 5
provides further discussion on the results presented in the paper and concludes.
..........................................................................................................................................................................
1
See de Mello (1997) for further details.

luiz de mello 135

2. The theoretical argument

FDI is conventionally dened as a form of international inter-rm co-operation
that involves signicant equity stake and effective management decision power in,
or ownership control of, foreign enterprises. FDI is also considered to encompass
other broader, heterogeneous non-equity forms of co-operation that involve the
supply of tangible and intangible assets by a foreign enterprise to a domestic rm.
Those broader collaborative associations include most types of quasiinvestment arrangements, such as licensing, leasing, and franchising; start-up and
international production sharing arrangements; joint ventures with limited foreign
equity participation; and broad R&D co-operation.
In the presence of FDI, aggregate production in the recipient economy is carried
out by combining labour and physical capital. The latter can be domestic (Kd ) or
foreign-owned (Kw ). FDI affects growth directly, by increasing the stock of physical
capital in the recipient economy, as Kw is accumulated, and indirectly, by inducing
human capital development and promoting technological upgrading. It is also
important to evaluate the extent of complementarity and substitution between
domestic investment and FDI because a simplistic Schumpeterian view of FDIrelated innovative investment that emphasises creative destruction through substitution may overlook the scope for complementarity between FDI and domestic
investment. Under complementarity, innovations embodied in foreign investment
may create, rather than reduce, rents accruing to older technologies (Young, 1993).
Also, if FDI is expected to affect growth positively, it may be argued that it requires
some degree of complementarity with domestic investment, at least in the short
run, given that the existing factor endowments in the host country act as a FDI
determinant.
The introduction of FDI in standard Ramsey models yields interesting results.
Under constant returns to domestic capital, the condition for saddle pointstability with FDI implies that negative consumption may not be avoided, and
hence FDI may be immiserising (Bhagwati, 1973; Brecher and Diaz Alejandro,
1977; Calvo et al., 1996) or dynamically inefcient.2 On the other hand, as in
the tradition of endogenous growth models, long-run growth can be achieved if
the marginal product of capital can be bounded away from the rate of time
preference as the stock of FDI increases, and the long-run growth rate depends
positively on FDI. Increases in the stock of foreign-owned capital lead to temporary
increases in the output growth rate if diminishing returns prevail in the aggregate.
However, the FDI-led increase in the growth rate is permanent under constant
returns. Long-run growth can also be shown to depend on the degree of
complementarity between capital stocks embodying domestic and foreign technologies, and the volume of FDI as a share of GDP. Under linearity, the growth
rates of the capital stock and output are constant and equal to the growth rate of
..........................................................................................................................................................................
2
The possibility of growth-enhancing FDI and saddle point stability is ensured if the open-economy
determinants of FDI are examined further (de Mello, 1996).

136

foreign direct investment-led growth

consumption, and permanent increases in FDI lead to permanent increases in the

output growth rate.

3. The time series aspects of FDI and growth

3.1 Preliminary considerations
For time series growth equations, empirical work has focused on the difculty of
lending support to the theoretical prediction that permanent changes in factor
accumulation should lead to permanent changes in output (Jones, 1995), given
the linearity hypothesis in a number of endogenous growth models. If growth rates
can be increased endogenously, then long-run growth should not in principle
admit approximations by processes with a constant mean, unless endogenous
growth-enhancing accumulation processes are offsetting. In this respect, time
series models provide a preliminary test of the linearity hypothesis, in so far as
the greater the accumulation of reproducible inputs, the greater the growth rate of
output. In other words, if a country's per capita growth rate exhibits little persistent
change over time, so should its determinants, or their persistent movements must
be offsetting. If, for instance, FDI generates signicant externalities and spillovers in
production, as suggested above, FDI-related capital accumulation should lead to an
increase in output growth.
As far as empirical validation is concerned, the impact of FDI on capital accumulation, and output and TFP growth is estimated for a sample of 32 countries in
the period 197090. The Data Appendix provides further details on data sources
and denitions of the variables used in this paper. We start by testing the hypothesis of stationarity of the output growth and investment rates using the augmented
DickeyFuller equation below
m
n
X
X
l gi t l
j
gi t j ei t
1

gi t 0
l1

j1

where gi is the growth rate of i, n and m are chosen according to the Schwartz
criterion to produce white noise disturbance terms ei , and i y; k; kw , for
k kd kw . Obviously, gi FDI, if i kw .
Country selection was primarily governed by the availability of physical capital
data in the Summers and Heston data set. It is further based on the stationarity
hypothesis, such that countries for which the output growth series are not I(0) by
eq. (1) were eliminated from the sample.3 The countries that satised both criteria
were further divided into two groups: OECD and non-OECD countries. The distinction between OECD and non-OECD countries is all the more important here,
since the impact of FDI on growth is expected to be stronger in the recipient
..........................................................................................................................................................................
3
Ben-David and Papell (1995) use Zivot and Andrews' (1992) methodology to test for structural break
for both stationary and unit root data, whereas Jones (1995) uses only unit root tests. The Phillips
Perron test was also carried out here but the results were similar to the ADF tests reports for the vast
majority of countries in the sample and therefore omitted.

luiz de mello 137

economy than in the country of origin of FDI. If the advanced economies in the
OECD group are the main net exporters of FDI, then the impact should be smaller
in those countries than elsewhere. In short, the distinction between OECD and
non-OECD countries proxies crudely for the distinction between technological
leaders (capital exporters) and followers (capital importers).
However, three points could be raised in this respect. First, there is strong
evidence that most FDI occurs across technologically advanced economies
(Lucas, 1990). Second, FDI is very sensitive to balance of payments constraints
and factors related to the macroeconomic performance and institutional features of
the recipient economy, such as the degree of openness and trade regime, political
instability and government intervention, the existence of property rights legislation,
and ensuing law enforcement institutions, etc. Finally, the incorporation of new
technologies in countries undergoing a process of industrialisation is likely to differ
a great deal from that in industrial, technologically advanced economies.
In addition to the augmented DickeyFuller test in eq. (1), stationarity can also
be assessed by testing for the existence of a deterministic time trend (t) in the series
according to the equation below
2
gi t 0 1 t vi t
where vi is a white noise term.
In principle, if a country exhibits persistent movements in output growth rates
and in capital accumulation (or FDI), then permanent changes in the former can be
deemed to result from permanent changes in the latter. This is supportive of the
linearity hypothesis, even if permanent changes in output or factor accumulation
derive from FDI-related disembodied technological change.
The results of the estimations of eqs (1) and (2) are reported in Tables 1 and 2.4
Stationarity of the relevant series can also be assessed by inspection of Figs 1 and 2
(FDI series on the secondary vertical axis). In the case of the OECD countries,
both FDI and total investment were found to follow a time trend in Italy,
Germany, UK, and USA, without a corresponding trend for output growth. For
the remaining countries, FDI alone (Australia, Belgium, Finland, Spain, and
Sweden) or total investment alone (Austria, Denmark, New Zealand, and
Switzerland) seem to follow a time trend without a counterpart in output
growth. Hence, for the OECD sample, there is no time series evidence of
linear endogenous growth derived from FDI and/or capital accumulation in the
period under examination.
For the non-OECD countries, in Table 2, in the case of Ecuador and Ivory
Coast, there is a negative time trend in the case of FDI (and capital investment
in the latter country) and a negative time trend in output growth. This nding
is suggestive of a linear endogenous relationship between output growth and
FDI. Both FDI and capital investment also appear to follow a time trend in
the case of the Dominican Republic, Mexico, and the Philippines, without
..........................................................................................................................................................................
4
Ideally, the total capital stock in the economy should be disaggregated into domestic and foreignowned capital, such that the analysis could be carried out for the two types of capital separately.

138

foreign direct investment-led growth

Table 1 Time series analysis, OECD sample, 1970/90

Countries
Australia
Austria
Belgium
Denmark
Finland
France
Germany
Italy
Luxembourg
New Zealand
Netherlands
Spain
Sweden
Switzerland
UK
USA

Output
growth
70.09
(72.8862){
70.15
(72.0254)*
0.08
(72.1365)*
70.18
(72.5963)*
70.01
(73.0477){
70.003
(72.2660)*
70.19
(72.2770)*
70.48
(72.1885)*
0.09
(72.2811)*
0.10
(73.2841){
0.22
(72.0752)*
0.61
(72.2814)*
0.14
(72.4511)*
70.01
(73.2121){
0.08
(72.9344){
70.20
(73.2862){

Trend
70.0004
(70.463)
70.001
(71.442)
70.0005
(70.663)
70.0002
(70.280)
70.003
(71.939)
70.0005
(71.056)
70.000001
(70.130)
70.0007
(70.780)
0.0006
(0.575)
70.001
(71.120)
70.0001
(70.178)
70.0001
(70.0113)
70.001
(71.503)
0.0003
(0.262)
70.0006
(70.591)
70.0005
(70.497)

Capital
investment
0.27
(72.1219)*
0.48
(72.8086){
0.12
(72.0348)*
0.52
(71.6133)
0.50
(72.9027){
70.02
(72.3914)*
0.28
(72.3174)*
0.23
(71.1141)
70.71
(75.1017){
0.07
(72.2134)*
0.65
(71.3903)
0.29
(71.4875)
70.71
(73.3179){
0.81
(71.8110)
0.63
(72.4388)*
0.55
(71.7975)

Trend
0.0003
(0.702)
70.003{
(79.116)
70.00003
(70.045)
0.002{
(3.408)
0.0005
(70.862)
0.0001
(0.238)
0.0008*
(2.043)
0.001{
(3.001)
70.0008
(70.807)
0.002{
(2.645)
0.001
(1.953)
0.0008
(0.792)
0.0004
(0.492)
0.002{
(4.139)
0.001{
(4.848)
0.001*
(2.413)

FDI
0.86
(0.1833)
0.71
(70.636)
0.70
(0.911)
0.22
(70.5158)
0.37
(70.0311)

Trend
0.063{
(6.892)
70.016
(71.143)
0.039*
(2.230)
70.018
(71.089)
0.094{
(2.839)

0.56
70.053*
(70.3812) (72.046)
0.93
70.067{
(72.3796)* (73.226)
0.49
(0.6583)
70.43
(70.3196)
0.61
(0.4561)
0.73
(2.4807)

70.012
(70.623)
70.058
(70.892)
70.198{
(73.108)
0.087{
(2.693)

0.30
(0.8711)
0.93
(6.7763)

0.163{
(3.791)
0.163{
(16.573)

Notes: Numbers in parentheses are ADF(1) statistics (calculated without a constant or a time trend) and
the coefcient reported is the 1 coefcient in eq. (1). For the time trends, the numbers in parentheses
are t-statistics and the coefcients reported are the 1 coefcients in eq. (2). Capital stocks refer to
producer capital (durables minus transport). Germany refers to former West Germany only. FDI data for
Belgium also includes Luxembourg. In the case of France and Switzerland, there were too few observations in the FDI series for tests to be carried out. In the case of Denmark, the DF statistic is reported
instead of ADF(1) for the FDI series. (*) signicant at the 5% level, and ({) signicant at the 1% level.

nevertheless a corresponding trend in output growth. In the case of

Mexico, the positive trend in FDI may be offsetting the negative trend in capital
formation, despite the fact that, in oil-exporting countries, the dynamics of FDI are
expected to differ from the case of their non-oil-exporting counterparts. Important
explanatory factors for the surge of FDI inows in developing countries in recent
years are foreign acquisition of domestic rms in reform-driven privatisation

luiz de mello 139

Table 2 Time series analysis, non-OECD sample, 197090
Countries
Dom. Rep.
Honduras
Mexico
Panama
Bolivia
Brazil
Chile
Ecuador
Paraguay
Peru
Venezuela
Ivory Coast
Kenya
Nigeria
Sierra Leone
Zimbabwe
Philippines

Output
growth
70.14
(73.7842){
0.24
(73.2605){
0.17
(73.0991){
0.15
(72.5313)*
0.51
(72.0775)*
0.50
(72.7002){
0.18
(72.6934)*
0.68
(72.4762)*
0.08
(72.7406){
0.0006
(73.8209){
0.11
(72.5763)*
0.39
(72.0156)*
70.14
(72.6613)*
0.42
(72.6166)*
70.38
(72.4507)*
70.14
(73.2551){
0.41
(73.1402){

Trend

Capital
investment

Trend

70.002
0.76
70.006{
(71.024)
(72.3968)* (73.981)
70.002
70.12
70.0008
(71.291)
(73.5383){ (71.149)
70.0001
0.52
70.003{
(70.477)
(71.3389) (72.758)
70.0006
71.13
0.001
(70.218)
(74.7711){
(0.355)
70.002
0.18
70.001
(71.541)
(72.2640)* (70.525)
70.002{
0.21
0.002
(72.783)
(72.2952)*
(0.435)
0.005
0.16
70.04
(1.731) (772.3780)* (71.927)
70.005{
0.16
70.004
(73.728)
(72.6348)* (71.659)
70.003
0.17
70.0008
(71.234)
(73.3298){ (70.967)
70.003
0.26
70.002{
(71.369)
(73.6484){ (73.033)
0.001
0.48
70.004{
(0.569)
(71.7348) (72.926)
70.006{
0.79
70.003{
(72.398)
(70.8391) (73.850)
0.0004
0.27
70.0009
(0.273)
(72.2280)* (71.158)
70.005
0.09
70.01{
(71.694)
(72.3106)* (74.852)
70.004
70.70
0.0001
(71.363)
(73.0149){
(0.156)
70.001
0.37
70.005*
(70.475)
(71.7668) (72.517)
70.002
0.71
70.004{
(71.474)
(72.0382)* (72.828)

FDI
0.26
(70.8999)
0.54
(70.3210)
0.66
(70.1414)
0.21
(72.1521)*
0.36
(72.0600)*
0.63
(70.7069)
0.61
(70.6819)
0.34
(72.266)*
0.62
(70.4992)
0.37
(71.7131)
70.01
(72.309)*
0.31
(70.8750)
0.46
(71.0845)
70.09
(70.6899)
70.01
(72.2323)*
0.19
(71.5547)
0.61
(70.805)

Trend
70.021*
(72.831)
0.057{
(3.185)
0.102{
(4.333)
70.035*
(72.434)
70.013
(71.429)
70.053
(71.710)
70.254{
(74.354)
70.039{
(73.537)
70.056
(70.759)
0.019
(0.592)
0.048
(1.355)
70.028*
(72.011)
70.166
(70.968)
70.013
(70.393)
0.12
(1.311)
70.142
(70.584)
70.277{
(73.402)

Notes: As in Table 1. In the case of Panama, Bolivia, and Ivory Coast, FDI stock data are unavailable or
available for short time spans.

programmes, globalisation and internationalisation trends in production, operations and investment, and increased economic and nancial integration, among
others.
The possibility that the variables under examination may be stationary around a
trend is also taken into account by including a deterministic time trend in eq. (1),
in addition to the drift term. The estimations (not reported but available upon
request) show that the results reported in Tables 1 and 2 are robust to the exclusion

140

foreign direct investment-led growth

Fig. 1
of these trends in eq. (1), for the vast majority of countries in the sample.5 The
exceptions are Italy, for which the hypothesis of a unit root in the FDI series is
accepted when a deterministic trend is incorporated; and Finland, the Dominican
Republic, Bolivia, Paraguay, and Kenya, for which unit roots are found in the
capital investment series. In the case of The Netherlands, the capital investment
series becomes stationary when eq. (1) is estimated with a deterministic trend.
..........................................................................................................................................................................
5
Robustness of the stationarity results presented in Tables 1 and 2 was also assessed using additional test
statistics, such as the DickeyFuller  ,  , '3 , and '2 statistics (see Enders, 1995, for denitions of the
test statistics). The results are nevertheless not reported in the paper due to space limitations.

luiz de mello 141

Fig. 2

142

foreign direct investment-led growth

Table 3 Cointegration analysis, 1970/90. Dependent variable: yt

Italy
kw
ML
Test
Trace
Test

Coeff.
Resid.

r0
r < 1
r0
r < 1
y
kw
y
kw

70.05
24.78{
10.15{
34.93{
10.15{
70.203
73.725
74.3039{
73.6809{

Panama

Bolivia

Ecuador

Venezuela

0.57
10.6
4.625*
15.24
4.625*
70.0313
1.115
72.4741*
73.3617{

70.36
20.47{
4.772*
25.24{
4.772*
0.2099
71.536
73.0278{
73.1235{

74.34
0.01
21.16*
17.9*
9.335{
6.138*
30.5{
24.04{
9.335{
6.138*
0.015
71.091
0.193
758.94
72.6999* 76.0589{
74.7874{ 73.7777{

Sierra Leone
0.01
19.81*
6.077*
25.89{
6.077*
70.919
20.63
72.4692*
72.6862*

Note: r denotes the rank of   0 in the estimating equation

zt
zt 1 zt 2 Dt ut;
where z y; kw , D is a vector of deterministic elements (constant and time trend), and u is an error
term. Variables are included in levels. (*) signicant at the 5% level, ({) signicant at the 1% level.
ADF(1) statistics are reported to test whether the residuals are I(0).

3.2 Further time series evidence

To extend the analysis of the time-series aspects of the growth-FDI nexus, the longrun impact of FDI on capital investment, and output and TFP growth is estimated.6
The bivariate VAR models were estimated for the countries in Tables 1 and 2 for
which the output growth, FDI, and capital accumulation series were found to be
I(0). If the relevant variables are found to cointegrate, the impact of growth-enhancing accumulation on output and TFP growth can therefore be deemed to be
permanent. The result of the estimations are reported in Tables 3, 4, and 5.
The objective here is also to test the conjecture put forward above that the
growth-FDI nexus involves the impact of FDI on capital accumulation and
technological change itself. The former assesses the scope for complementarity or
substitutability between capital stocks embodying domestic and more efcient
FDI-related technologies. The latter assesses the scope for knowledge or technology
transfers between foreign investor and the recipient economy.
As for the results of the long-run time series analysis, all countries in Table 3
exhibit a time trend in their FDI series with the exception of Sierra Leone,
Venezuela and Bolivia. The long-run FDI coefcients vary from 0.08 in Venezuela
to 0.15 in Ecuador, which provides evidence that FDI can be growth-enhancing
or depressing in the long-run. The negative sign of the FDI coefcient in most nonOECD countries can be attributed in principle to the macroeconomic instability
..........................................................................................................................................................................
6
We did not attempt to estimate a short-run VAR model for the countries in Tables 3, 4, and 5,
given that short-run deviations from the long-run equilibrium depend on the transitional dynamics
of the growth process. Under linearity, there is no transitional dynamics. As a result, in the presence
of FDI inows, output growth in the recipient economy adjusts automatically to its steady-state
path.

luiz de mello 143

Table 4 Cointegration analysis, 1970/90. Dependent variable: kt

kw
ML
Test
Trace
Test

Coeff.
Resid.

r0
r < 1
r0
r < 1
k
kw
k
kw

Panama

Bolivia

0.02
10.18
8.408{
18.59*
8.408{
71.170
73.079
73.5039{
73.1064{

70.18
9.804
0.5019
10.31
0.5019
70.396
70.947
73.1313{
73.8024{

Ecuador

Sierra Leone

76.70
13
1.248
14.25
1.248
0.019
70.089
73.3623{
76.3920{

70.01
10.95
5.61*
16.56
5.61*
70.339
750.55
73.8590{
73.9153{

Note: As in Table 3, with z k; kw .

Table 5 Cointegration analysis, 1970/90. Dependent variable:

TFPt
Italy
kw
ML
Test
Trace
Test

Coeff.
Resid.

r0
r < 1
r0
r < 1
TFP
kw
TFP
kw

70.19
25.76{
6.294*
32.05{
6.294*
70.032
71.393
73.7818{
73.2648{

Venezuela
0.02
13.27
5.341*
18.61*
5.341*
0.296
41.57
73.1808{
73.5121{

Note: As in Table 3 with z TFP; kw .

and severe international credit (and hence balance of payments) constraints that
characterised most of the period under examination, particularly the 1980s (Cohen,
1994; van der Ploeg and Tang, 1994). By Table 4, FDI has a positive long-run
impact on capital accumulation in Panama and Sierra Leone. No cointegration
relationship was found for Bolivia and Ecuador. By Table 5, there is a positive longrun relationship between FDI and TFP growth in Venezuela, and a negative one in
Italy.

4. Panel data evidence

4.1 Preliminary considerations
To complement the time series analysis above, we now turn to panel data
estimations. This is because it is well known that, in the case of cross-country
and time-series estimations, the correlation between the error term and the
regressors in standard growth accounting-based, time-series production function

144

foreign direct investment-led growth

estimations leads to simultaneity and omitted variables biases. The correlation

between the per capita capital stock and the error term leads to capital elasticity
estimates that are well above the capital share in output (Young, 1992, 1995).
With regard to the high estimates of the capital elasticity coefcient, they have
been interpreted as evidence for the linear endogenous growth model (Romer,
1990) and explained on the grounds that capital should be understood in a
broad sense to incorporate additional inputs (for instance, human capital) without
diminishing returns (Mankiw et al., 1992). This line of argument sustains that high
capital elasticities incorporate the externalities generated by the use of additional
inputs, such as knowledge and human capital accumulation (Benhabib and Jovanovic, 1991; Benhabib and Spiegel, 1994). In fact, cross-country regressions reveal
very little of the role played by policy and institutions (Crafts, 1995).
In panel data estimations, however, the existence of unobservable growth determinants that are country-specic, such as the ones discussed above, can be
acknowledged and taken into account in the estimation procedure. This can
explain a great deal of the distinct time series patterns in Tables 3, 4, and 5 on
the grounds of country-specic effects. The country-specic determinants of FDI
may provide further insight into ulterior growth determinants that are unobservable in the time series analysis.
Let country-specic factors affect primarily TFP, such that, in the presence of
FDI, TFP can be decomposed into a country-specic term (), and a term incor
porating foreign knowledge transfers or technological change embodied in FDI (),
which is exogenous to the recipient economy. If different countries are assumed to
 On the other hand, if
have the same technology and institutions, then TFP  .
country-specic growth determinants matter, which seems to be a more realistic
 where h is a recipient country index. Countryconjecture, then TFPh h ,
specic factors may identify what Abramovitz (1986) referred to as a country's
`social capability'.
Rather than engaging in a growth accounting exercise, the impact of FDI on
output and TFP growth, and capital accumulation can be estimated using the
following equation
A
xh t #0 #1 FDIh t #2 xh t 1 "h t
where x y; k; TFP and "t is an error term.
If unobservable country-specic growth determinants are to be taken into
account, then eq. (A) can be estimated as follows
B
xh t #h;0 #1 FDIh t #2 xh t 1 "h t
where #h;0 is a time-invariant individual country effect term.
Equations (A) and (B) are dynamic panel models, which are estimated by IV,
due to the likely correlation between the regressors and the disturbance terms. As
for the choice of instruments, the difculty lies in the requirement that they should
not be country-specic but correlated with the regressors. Nevertheless, to reduce
the expected sensitivity of the results to the instruments chosen, we take a conservative approach and use the lagged dependent variables as the main instruments,

luiz de mello 145

together with current and lagged values of the recipient economy's per capita
income as a share of the US per capita income. The latter variable is expected to
proxy for the differences in marginal capital productivities, given the differences in
capital endowments across countries and the technological gap between technological leaders and followers. The latter also proxies for the scope for technological
transfers in FDI. Accounting for such technological gaps, or `idea' gaps in the sense
of Romer (1993), is expected to correct for excessively high elasticities in the case of
technological laggards.

4.2 The results

By Table 6, there is a positive impact of FDI on output growth in all panels, with
and without country-specic terms. The nding is suggestive of a dominant complementarity effect between FDI and domestic investment. Contrary to the case of
Tables 3 and 5, the ndings in Table 6 with respect to output growth are consistent
with those for TFP growth, by which FDI appears to have a positive impact on
technological change, in the OECD panel. In the non-OECD panel, however, there
seems to be a negative relationship between FDI and TFP growth, when group
Table 6 Panel data estimations (xed-effect estimations)
yt
kt
TFPt
equation
equation
equation
................................................................................................................................................................
Samples
A
B
A
B
A
B
All countries* (672 obs.)
#1
0.01
0.006
(0.0025)
(0.0029)
#2
0.19
0.14
(0.0160)
(0.0150)
R2
0.20
0.36

0.02
(0.0049)
70.57
(0.0247)
0.07

0.01
(0.0060)
70.45
(0.0276)
0.53

70.004
(0.0041)
0.58
(0.0208)
0.54

0.001
7(0.0049)
0.41
(0.0221)
0.64

OECD (273 obs.)

#1
0.02
(0.0031)
#2
0.04
(0.0139)
R2
0.09

0.03
(0.0036)
0.01
(0.0146)
0.10

0.003
(0.0023)
70.003
(0.0112)
0.01

70.001
(0.0015)
0.01
(0.0065)
0.71

0.01
(0.0037)
0.008
(0.0178)
0.12

0.03
(0.0036)
0.004
(0.0154)
0.44

Non-OECD (357 obs.)

#1
0.03
(0.0060)
#2
0.16
(0.0198)
R2
0.17

0.01
(0.0059)
0.09
(0.0134)
0.66

70.00
(0.0121)
70.56
(0.027)
0.53

0.02
(0.0130
0.30
(0.0243)
0.7

0.02
(0.0112)
0.55
(0.0288)
0.51

0.01
7(0.0123)
0.30
(0.0256)
0.74

Note: Estimations carried out by IV, columns (A) and (B) refer to estimates without and with group
dummy variables, respectively. The common intercept in column (A) is not reported. Instruments
include the lagged dependent variables and per capita income as a share of the USA per capita
income. The numbers in parentheses are standard errors. * Includes France and Switzerland, and
dummy variables for African and Latin American countries separately.

146

foreign direct investment-led growth

dummy variables are incorporated in the equation. It can be inferred that in

technological followers, FDI reduces TFP growth by fostering producer capital
accumulation, given the complementarity effect, although the converse cannot be
inferred in the case of technological leaders.
Also, by Table 6, FDI seems to have a positive impact on producer capital
accumulation in the broad panel, but not in the OECD (without group dummies)
and non-OECD panels. In the latter, however, after the introduction of country
effects, the relationship between FDI and capital accumulation becomes positive.
The results lend support to the hypothesis of some degree of substitutability
between FDI and domestic investment, whereby in more advanced economies,
the more efcient technologies embodied in FDI may lead to a higher rate of
technological obsolescence of the capital stocks embodying older technologies. In
the same vein, complementarity seems to prevail in technological laggards. In this
respect, it can be argued that the degree of complementarity between old and new
technologies found in technological followers, after country-specic effects were
accounted for, suggests that those economies are either less efcient in the use of
the new technologies embodied in FDI-related capital accumulation, or that the
latter are not much more modern or productive than the ones existing in the
recipient economy. Those country-specic effects may include, for instance,
protectionist trade and investment policies implemented to safeguard indigenous
industries from foreign competition, which would distort social and private returns
to capital investment and hence the efciency of FDI.7 Foreign investors can then
be deemed to select the technologies embodied in FDI-related capital accumulation
depending on the specic productive and institutional characteristics of the recipient economy, in which case, FDI may be a less important vehicle for cross-border
knowledge transfers than previously thought.
The xed-effect estimates above rely on the assumption of homogeneity
of the different panel groups for a common slope to be imposed in pooled
regressions. The xed-effect estimator also reduces biases due to omitted variables,
particularly when group dummies are incorporated in the regression (Matyas
and Sevestre, 1996). Nevertheless, in dynamic panels with a large time dimension,
ignoring group heterogeneity produces serially correlated disturbances and
hence inconsistent parameter estimates (see Pesaran and Smith, 1995, for
further details). The biases in this case tend to overestimate the average shortrun effects and underestimate the average long-run effects in xed-effect
estimations, and cannot be eliminated by using the IV method or by considering
autocorrelation correction. Instead, the average effects in dynamic panels with
changing slopes across groups can be consistently estimated using the mean
group procedure.
..........................................................................................................................................................................
7
Kawai (1994) points out that FDI may not be a good determinant of TFP growth due to the fact that
foreign production may occur in oligopolistic sectors or that there may be a time lag before FDI-induced
productivity gains creep in.

luiz de mello 147

Table 7 Mean group estimations
Means of country-specic short-run FDI impact on the growth rate of:
................................................................................................................................................................
Samples
yt
kt
TFPt
All countries*
(32 countries)

0.004
(0.0227)

0.02
(0.0295)

70.016
(0.0034)

OECD
(15 countries)

0.026
(0.0067)

0.002
(0.0031)

0.012
(0.0060)

Non-OECD
(17 countries)

70.016
(0.0037)

0.03
(0.0538)

70.040
(0.0582)

Note: The coefcients reported are based on separate regressions for each country in the sample,
197090. The numbers in parentheses are standard errors, calculated assuming cross-country independently distributed regression coefcients. The means are unweighted. * Includes France and Switzerland.

Parameter estimates using the mean group estimator for heterogeneous panels
are reported in Table 7. It is noticeable that the aggregate parameter estimates using
the xed-effect estimator are fairly similar to the means of the country-specic
regressions in the case of the OECD sample. In the case of the non-OECD sample,
aggregate estimates differ from the average of individual country coefcients.
Although this discrepancy suggests the existence of a heterogeneity-related bias,
the nding is not surprising, given that heterogeneity is likely to prevail in the latter
sample, given the diversity of countries pooled together in the aggregate xed-effect
estimations. In the full sample, the homogeneity of the OECD countries is likely to
have reduced the bias, given that the parameter estimates reported in Tables 6 and
7 are fairly similar.

5. Discussion and Conclusion

General problems in modern growth empirics offer new challenges to the growth
theorist. The problem of simultaneity and omitted variable biases in growth
equations has been investigated more thoroughly in recent years, in particular
as far as the high estimates of the capital elasticity in cross-country equations
are concerned. Nevertheless, the problem of sensitivity of the estimates to the
instruments chosen to reduce the simultaneity bias in growth equations has not
been addressed to the same degree of depth and deserves more careful analyses in
the near future. In fact, the solution to these problems lies in the revision of the
whole methodology that permeates growth empirics, which is based on 1960s
growth accounting or national accounts-type intertemporal investment equations,
such as the ones used in Ramsey-type dynamic optimisation problems. The latter
too are sensitive to the assumptions regarding depreciation and savings behaviour.
Country-specic effects are not commonly accounted for in growth empirics and,
as seen above, can play an important role in acknowledging unobservable growth
determinants. In addition, attention should be focused on the biases in xed-effect

148

foreign direct investment-led growth

estimations of heterogeneous panels with a large time dimension which cannot

be reduced or eliminated by using instrumental variables or autocorrelation
correction.
Whether FDI can be deemed to be a catalyst for output growth, capital accumulation, and technological progress seems to be a less controversial hypothesis in
theory than in practice. For most of the countries for which our methodology could
be applied, it seems that the growthFDI nexus is sensitive to country-specic
factors that are unobservable in time series analysis. The results reported in this
paper suggest that, if FDI is growth-enhancing in the long run, via both knowledge
transfers and the accumulation of capital stocks embodying newer technologies,
then this impact is likely to be lower in technological leaders than laggards
(Grossman and Helpman, 1991). As a result, the impact of FDI on growth seems
to depend inversely on the technological gap between leaders and followers, even
though there is evidence that the bulk of FDI occurs across technologically
advanced economies.
The degree of substitutability between capital stocks embodying old (domestic)
and new (FDI-related) technologies seems to be higher in technologically
advanced, rather than developing recipient, economies. The rate of technological
obsolescence of the capital stock embodying old technologies is possibly
increased in technologically advanced economies in the presence of FDI.
Alternatively, it can be argued that the degree of complementarity between old
and new technologies found in developing economies, after country-specic
effects were accounted for, suggests that those economies may: (i) be less efcient
in the use of the new technologies embodied in FDI-related capital accumulation;
(ii) have difculty to assimilate capital and technology-intensive improvements;
or (iii) that the latter are not much more modern or productive than the ones
existing in the recipient economy. Foreign investors can then be deemed to
select the technologies embodied in FDI-related capital accumulation depending
on specic productive and institutional characteristics of the recipient economy.
In this case, FDI may be a less important vehicle for cross-border knowledge
transfers and the elimination of technological gaps between leaders and followers
than previously thought.
Country-specic factors (institutions, trade regime, political risk, policy, etc.)
may have inhibited the elimination of technological gaps between leaders and
followers, such that success in technology or knowledge transfers seems to be
affected by institutions.8 When heterogeneity is taken into account, the conclusions
are less clear-cut in the case of non-OECD countries, which are more likely to
exhibit signicant cross-country diversity, than in the case of their more
homogeneous OECD counterparts.
..........................................................................................................................................................................
8
See Romer (1993) for an interesting discussion of the role played by FDI-related knowledge transfers in
closing the idea gap between developed and developing countries.

luiz de mello 149

Acknowledgements
The author is indebted to Alan Carruth, Tony Thirlwall, Geoff Harcourt, Andy Dickerson,
Peter Sanfey, Francisco Carneiro, and Joao Ricardo Faria for helpful comments and discussions. I also thank Simon Cowan, Peter Sinclair, and two anonymous referees for their
comments. The usual disclaimer nevertheless applies.

References
Abramovitz, M. (1986). `Catching-up, Forging Ahead and Falling Behind', Journal of
Economic History, 36, 385406.
Bhagwati, J.N. (1973). `The Theory of Immiserising Growth: Further Applications', in M.B.
Connolly and A.K. Swododa (eds), International Trade and Money, Toronto University Press,
Toronto.
Barro, R.J. and Sala-i-Martin, X. (1992). `Convergence', Journal of Political Economy, 100,
22351.
Benhabib, J. and Jovanovic, B. (1991). `Externalities and Growth Accounting', American
Economic Review, 81, 82113.
Benhabib, J. and Spiegel, M. (1994). `The Role of Human Capital in Economic
Development. Evidence from Aggregate and Cross-Country Data', Journal of Monetary
Economics, 34, 14373.
Ben-David, D. and Papell, D.H. (1995). `The Great Wars, the Great Crash, and Steady-State
Growth: some New Evidence about an Old Stylised Fact', Journal of Monetary Economics, 36,
45375.
Blomstrom, M., Lipsey, R.E. and Zejan, M. (1996). `Is Fixed Investment the Key to
Economic Growth?', Quarterly Journal of Economics, 111, 26976.
Brecher, R.A. and Diaz Alexandro, C.F. (1977). `Tariffs, Foreign Capital, and Immiserising
Growth', Journal of International Economics, 7, 31722.
Calvo, G., Leiderman, L. and Reinhart, C. (1996). `Inows of Capital to Developing
Countries in the 1990s', Journal of Economic Perspectives, 10, 12340.
Cohen, D. (1994). `Growth and External Debt', in F. van der Ploeg (ed.), The Handbook of
International Macroeconomics, Blackwell, Oxford.
Crafts, N. (1995). `Endogenous Growth: Lessons for and from Economic History', mimeo,
London School of Economics, London.
De Long, B. and Summers, L.H. (1991). `Equipment Spending and Economic Growth',
Quarterly Journal of Economics, 106, 445502.
De Long, B.J. and Summers, L.H.. (1992). `Equipment Spending and Economic Growth:
How Strong is the Nexus', Brookings Papers on Economic Activity, 15799.
Denison, E.F. (1962). `Sources of Growth in the United States and the Alternatives before
Us', Supplement Paper 13, Committee for Economic Development, New York, NY.
Denison, E.F. (1967). Why Growth Rates Differ, The Brookings Institution, Washington, DC.
Enders, W. (1995). Applied Econometric Time Series, John Wiley Sons, Inc., New York, NY.
Grossman, G.M. and Helpman, E. (1991). Innovation and Growth in the Global Economy,
The MIT Press, Cambridge, MA.

150

foreign direct investment-led growth

Hofman, A.A. (1992). `Capital Accumulation in Latin America: a Six Country Comparison
for 195089', Journal of Income and Wealth, 38, 365401.
Hsiao, C. (1986). Analysis of Panel Data, Cambridge University Press, Cambridge.
Islam, N. (1995). `Growth Empirics: a Panel Data Approach', Quarterly Journal of Economics,
110, 112770.
Jones, C.I. (1995). `Time Series Tests of Endogenous Growth Models', Quarterly Journal of
Economics, 110, 495525.
Kawai, H. (1994). `International Comparative Analysis of Economic Growth: Trade
Liberalisation and Productivity', Developing Economies, 32, 37297.
Lucas, R. (1990). `Why Doesn't Capital Flow from Rich to Poor Countries?', American
Economic Review, 80, 926.
Mankiw, N.G., Romer, D. and Weil, D. (1992). `A Contribution to the Empirics of
Economic Growth', Quarterly Journal of Economics, 107, 40737.
Matyas, L. and Sevestre, P. (1996). The Econometrics of Panel Data, Kluwer Academic
Publishers, London.
de Mello, L.R., Jr. (1995). `Vintage Capital Accumulation: Endogenous Growth Conditions',
Journal of Macroeconomics, 17, 70316.
de Mello, L.R., Jr. (1996). `Foreign Direct Investment, International Knowledge Transfers,
and Endogenous Growth: Time Series Evidence', Discussion Paper, Department of
Economics, University of Kent.
de Mello, L.R., Jr. (1997). `Foreign Direct Investment in Developing Countries and Growth:
a Selective Survey', Journal of Development Studies, 34, 134.
de Mello, L.R., Jr. and Sinclair, M. T. (1995). `Foreign Direct Investment, Joint Ventures,
and Endogenous Growth', Discussion Paper, Department of Economics, University of
Kent.
Pesaran, M.H. and Smith, R. (1995). `Estimating Long-run Relationships from Dynamic
Heterogeneous Panels', Journal of Econometrics, 68, 79113.
van der Ploeg, F. and Tang, P. (1994). `Growth, Decits, and R&D in the Global Economy',
in F. van der Ploeg (ed.). The Handbook of International Macroeconomics, Blackwell,
Oxford.
Romer, P.M. (1990). `Endogenous Technological Change', Journal of Political Economy, 98,
S71-102.
Romer, P.M. (1993). `Idea Gaps and Object Gaps in Economic Development', Journal of
Monetary Economics, 32, 54373.
Solow, R.M. (1957). `Technical Change and the Aggregate Production Function', Review of
Economics and Statistics, 39, 31220.
Young, A. (1992). `A Tale of Two Cities: Factor Accumulation and Technical Change in
Hong Kong and Singapore', NBER Macroeconomics Annual.
Young, A. (1993). `Substitution and Complementarity in Endogenous Innovation', Quarterly
Journal of Economics, 108, 775807.
Young, A. (1995). `The Tyranny of Numbers: Confronting the Statistical Realities of the East
Asian Growth Experience', Quarterly Journal of Economics, 110, 64180.
Zivot, E. and Andrews, D. (1992). `Further Evidence on the Great Crash, the Oil Price Crisis
and the Unit Root Hypothesis', Journal of Business and Economic Statistics, 10, 25170.

luiz de mello 151

Data Appendix
Output: Summers and Heston gross national income series in the period 197090. Per capita
GDP as a share of US per capita GDP is also available in the Summers and Heston data set.
Capital Stock: Summers and Heston producer capital series (durables minus transport equipment stocks) in the period 197090. In the case of Brazil, the producer capital series is
available from Hofman (1992).
Foreign Direct Investment (FDI): net FDI ows (inows minus outows) are available from
IMF's Balance of Payments statistics. When disaggregated series are available, gross FDI
inows are used. Unfortunately, the availability of FDI data is limited and time series for
most countries start in the late 1960s and early 1970s, which prevents the consideration of a
longer time span. For most developing countries, data on FDI stocks is not available. When
needed, the FDI stock series were constructed using the perpetual inventory method with a
constant annual depreciation rate of 10%. Also, it can be argued that the FDI measures
available from individual countries' national accounts can only be taken to be a crude proxy
for the impact of foreign technologies and spillovers on growth, given that they are nancial
ows that may capture (or obscure) operations of foreign investors (typically MNCs) in the
recipient economy. Also, FDI ows are sensitive to cross-country differences in the treatment
of re-invested earnings and uctuations in intra-rm transactions. The role of tax havens
and offshore banking centres also pose additional measurement complications. However,
such ow variables are the ones currently available and most widely used in this line of
research.
Total Factor Productivity (TFP) growth measured as the difference between per capita output
growth and the per capita capital accumulation, both domestic and foreign-owned.