Does Foreign Direct Investment Accelerate Economic Growth? · Does Foreign Direct Investment Accelerate Economic Growth? MARIA CARKOVIC and ROSS LEVINE 195 With commercial bank lending

Does Foreign Direct InvestmentAccelerate Economic Growth?MARIA CARKOVIC and ROSS LEVINE

195

With commercial bank lending to developing economies drying up in the1980s, most countries eased restrictions on foreign direct investment (FDI)and many aggressively offered tax incentives and subsidies to attract for-eign capital (Aitken and Harrison 1999; World Bank 1997a, 1997b). Alongwith these policy changes, a surge of noncommercial bank private capitalflows to developing economies in the 1990s occurred. Private capital flowsto emerging-market economies exceeded $320 billion in 1996 and reachedalmost $200 billion in 2000. Even the 2000 figure is almost four times largerthan the peak commercial bank lending years of the 1970s and early 1980s.Furthermore, FDI now accounts for over 60 percent of private capital flows.While the explosion of FDI flows is unmistakable, the growth effects remainunclear.

Theory provides conflicting predictions concerning the growth effects ofFDI. The economic rationale for offering special incentives to attract FDI fre-quently derives from the belief that foreign investment produces external-ities in the form of technology transfers and spillovers. Romer (1993), forexample, argues that important “idea gaps” between rich and poor countriesexist. He notes that foreign investment can ease the transfer of technological

8

Maria Carkovic is senior fellow in finance and Ross Levine is Curtis L. Carlson Professor of Financeat the Carlson School of Management at the University of Minnesota. We thank Norman Loayza forhelpful statistical advice and Stephen Bond for the use of his DPD program. We thank participants atthe World Bank conference, Financial Globalization: A Blessing or a Curse, in Washington (May 2002),and the Institute of International Economics conference, The Impact of Foreign Direct Investment onDevelopment: New Measurements, New Outcomes, New Policy Approaches, in Washington (April2004). We received particularly useful suggestions from Monty Graham, Marc Melitz, and Ted Moran.

2668-10_CH08.qxd 04/14/05 09:17 Page 195

and business know-how to poorer countries. According to this view, FDImay boost the productivity of all firms—not just those receiving foreigncapital. Thus, transfers of technology through FDI may have substantialspillover effects for the entire economy. In contrast, some theories predictthat FDI in the presence of preexisting trade, price, financial, and otherdistortions will hurt resource allocation and slow growth (Boyd and Smith1992). Thus, theory produces ambiguous predictions about the growtheffects of FDI, and some models suggest that FDI will promote growthonly under certain policy conditions.

Firm-level studies of particular countries often find that FDI does notboost economic growth, and these studies frequently do not find positivespillovers running between foreign-owned and domestically owned firms.Aitken and Harrison’s (1999) influential study finds no evidence of a pos-itive technology spillover from foreign firms to domestically owned onesin Venezuela between 1979 and 1989. While Blomström (1986) finds thatMexican sectors with a higher degree of foreign ownership exhibit fasterproductivity growth, Haddad and Harrison (1993) find no evidence ofgrowth-enhancing spillovers in other countries. As summarized by Lipseyand Sjöholm (in this volume), in some countries, researchers find evidenceof positive spillovers in some industries, but country-specific and industry-specific factors seem so important that the results do not support the over-all conclusion that FDI induces substantial spillover effects for the entireeconomy. In sum, firm-level studies do not imply that FDI accelerates over-all economic growth.

Unlike the microeconomic evidence, macroeconomic studies—using ag-gregate FDI flows for a broad cross section of countries—generally sug-gest a positive role for FDI in generating economic growth, especially inparticular environments. For instance, Borensztein, De Gregorio, and Lee(1998) argue that FDI has a positive growth effect when the country has ahighly educated workforce that allows it to exploit FDI spillovers. WhileBlomström, Lipsey, and Zejan (1994) find no evidence that education is crit-ical, they argue that FDI has a positive growth effect when the country issufficiently wealthy. In turn, Alfaro et al. (2003) find that FDI promoteseconomic growth in economies with sufficiently developed financial mar-kets, while Balasubramanyam, Salisu, and Sapsford (1996) stress that tradeopenness is crucial for obtaining the growth effects of FDI.

The macroeconomic findings on growth and FDI must be viewed skep-tically, however. Existing studies do not fully control for simultaneity bias,country-specific effects, and the routine use of lagged dependent variablesin growth regressions. These weaknesses can bias the coefficient estimatesas well as the coefficient standard errors. Thus, the profession needs toreassess the macroeconomic evidence with econometric procedures thateliminate these potential biases.

This study uses new statistical techniques and two new databases to re-assess the relationship between economic growth and FDI. First, based on

196 DOES FDI PROMOTE DEVELOPMENT?

2668-10_CH08.qxd 04/14/05 09:17 Page 196

a recent World Bank dataset (Kraay et al. 1999), we construct a panel datasetwith data averaged over each of the seven five-year periods between 1960and 1995. We also confirm the results using new FDI data from theInternational Monetary Fund (IMF).

Methodologically, we use the Generalized Method of Moments (GMM)panel estimator to extract consistent and efficient estimates of the impact ofFDI flows on economic growth. Unlike past work, the GMM panel estima-tor exploits the time-series variation in the data, accounts for unobservedcountry-specific effects, allows for the inclusion of lagged dependent vari-ables as regressors, and controls for endogeneity of all the explanatory vari-ables, including international capital flows. Thus, this study advances theliterature on growth and FDI by enhancing the quality and quantity of thedata and by using econometric techniques that reduce biases.

Investigating the impact of foreign capital on economic growth hasimportant policy implications. If FDI has a positive impact on economicgrowth after controlling for endogeneity and other growth determinants,then this weakens arguments for restricting foreign investment. If, how-ever, we find that FDI does not exert a positive impact on growth, then thiswould suggest a reconsideration of the rapid expansion of tax incentives,infrastructure subsidies, import duty exemptions, and other measures thatcountries have adopted to attract FDI. While no single study will resolvethese policy issues, this study contributes to these debates.

This study finds that the exogenous component of FDI does not exert arobust, positive influence on economic growth. By accounting for simul-taneity, country-specific effects, and lagged dependent variables as reg-ressors, we reconcile the microeconomic and macroeconomic evidence.Specifically, there is no reliable cross-country empirical evidence supportingthe claim that FDI per se accelerates economic growth.

This chapter’s findings are robust to

� econometric specifications that allow FDI to influence growth differ-ently depending on national income, school attainment, domestic finan-cial development, and openness to international trade;

� alternative estimation procedures;

� different conditioning information sets and samples;

� the use of portfolio inflows instead of FDI; and

� the use of alternative databases on FDI.

The data produce consistent results: there is not a robust, causal link run-ning from FDI to economic growth.

This study’s results, however, should not be viewed as suggesting thatforeign capital is irrelevant for long-run growth. Borensztein, De Gregorio,and Lee (1998) show, and this study confirms, many econometric specifi-

DOES FDI ACCELERATE ECONOMIC GROWTH? 197

2668-10_CH08.qxd 04/14/05 09:17 Page 197

cations in which FDI is positively linked with long-run growth. FDI mayeven be a positive signal of economic success as emphasized by Blomström,Lipsey, and Zejan (1994). More generally, “openness”—defined in a lessnarrow sense than FDI inflows—may be crucial for economic success, assuggested by other research (e.g., Bekaert, Harvey, and Lundblad 2001;Klein and Olivei 2000). Rather than examine these broad issues, this study’scontribution is much narrower: after controlling for the joint determinationof growth and foreign capital flows, country-specific factors, and othergrowth determinants, the data do not suggest a strong independent impactof FDI on economic growth. In terms of policy implications, this study’sanalyses do not support special tax breaks and subsidies to attract foreigncapital. Instead, the literature suggests that sound policies encourageeconomic growth and also provide an attractive environment for foreigninvestment.

Before continuing, it is worth emphasizing this study’s boundaries. Wedo not discuss the determinants of FDI. Instead, we extract the exogenouscomponent of FDI using system panel techniques. Also, we do not examineany particular country in depth. We use data on 72 countries from 1960 to1995. Thus, our investigation provides evidence based on a cross section ofcountries.

Econometric Framework

This section describes two econometric methods that we use to assess therelationship between FDI inflows and economic growth. We first use sim-ple ordinary least squares (OLS) regressions with one observation percountry over the 1960–95 period. Second, we use a dynamic panel proce-dure with data averaged over five-year periods, so that there are seven pos-sible observations per country between 1960 and 1995.

OLS Framework

The pure cross-sectional OLS analysis uses data averaged from 1960–95.The data include one observation per country and heteroskedasticity-consistent standard errors. The basic regression takes the form:

where the dependent variable, GROWTH, equals real per capita grossdomestic product (GDP) growth, FDI is gross private capital inflows to acountry, and CONDITIONING SET represents a vector of conditioninginformation.

GROWTH CONDITIONING SETi = + + ′[ ] +α β γ εFDIi i i ( . )8 1


2668-10_CH08.qxd 04/14/05 09:17 Page 198

Motivation for the Dynamic Panel Model

The dynamic panel approach offers advantages to OLS and also improveson previous efforts to examine the FDI-growth link using panel procedures.First, using panel data—that is, pooled cross-section and time-series data—to make estimates allows researchers to exploit the time-series nature of therelationship between FDI and growth. Thus, the panel approach includedmore information than the pure cross-country approach with positive ram-ifications on the precision of the coefficient estimates. Second, in a purecross-country instrumental variable regression, any unobserved country-specific effect becomes part of the error term, which may bias the coefficientestimates (as we explain in detail below). Thus, if there are country-specificfixed effects that are not included in the conditioning set and that helpexplain economic growth, then the OLS procedure may produce erroneousestimates on the FDI coefficient. The panel procedures control for country-specific effects. Third, unlike existing pure cross-country studies that useinstrumental variables to control for the potential endogeneity of FDI, thepanel estimator controls for the potential endogeneity of all explanatoryvariables. This distinction is important. If the other growth determinantsbesides FDI are endogenously determined with growth, which seemslikely since the other growth determinants include inflation, governmentsize, and the black market premium, among others, and if the estimationprocedure does not account for this endogeneity, then this could biasFDI’s estimated coefficient and standard error. Finally, the panel estima-tor that we employ accounts explicitly for the biases induced by includinginitial real per capita GDP in the growth regression. Since initial real percapita GDP is a component of the dependent variable, economic growth,including this variable as a regressor may bias both the coefficient esti-mates and their standard errors, potentially leading to erroneous conclusions.For these reasons, we augment the OLS regressions with panel estimates.

Detailed Presentation of the Econometric Methodology

We use the GMM estimators developed for dynamic panel data. Our panelconsists of data for a maximum of 72 countries from 1960–95, though capitalflow data do not begin until 1970 for many countries. We average data overnonoverlapping, five-year periods, so that, data permitting, seven observa-tions per country (1961–65, 1966–70, etc.) are made. Thus, we exploit the time-series, along with the cross-country, dimension of the data. Consider thefollowing regression equation:

y y y Xi t i t i t i t i i t, , , , , ( . )− = −( ) + ′ + +− −1 11 8 2α β η ε


2668-10_CH08.qxd 04/14/05 09:17 Page 199

where y is the logarithm of real per capita GDP, X represents the set ofexplanatory variables (other than lagged per capita GDP), η is an un-observed country-specific effect, ε is the error term, and the subscripts i andt represent country and five-year time period, respectively. Specifically, Xincludes FDI inflows to a country as well as other possible growth deter-minants. We also use time dummy variables for each five-year period toaccount for period-specific effects, though these are omitted from the equa-tions in the text. We can thus rewrite equation 8.2:

To eliminate the country-specific effect, take first differences of equation8.3:

Thus, this eliminates potential biases associated with unobserved fixed,country effects.

Instrument variables are required to deal with both the endogeneity of allthe explanatory variables and the problem that the new error term εi, t− εi, t−1,which is correlated with the lagged dependent variable yi, t−1 − yi, t−2, createsbecause of the routine inclusion of lagged values of the dependent variableas a regressor. Under the assumptions that the error term is not seriallycorrelated, and the explanatory variables are weakly exogenous (i.e., theexplanatory variables are uncorrelated with future realizations of the errorterm), the GMM dynamic panel estimator uses the following moment con-ditions, where s and t indicate the five-year period under evaluation:

We refer to the GMM estimator based on these conditions as the differenceestimator.

There are, however, conceptual and statistical shortcomings with thisdifference estimator. Conceptually, we would also like to study the cross-country relationship between financial development and per capita GDPgrowth, which is eliminated in the difference estimator. When the explana-tory variables are persistent over time, lagged levels make weak instru-ments for the regression equation in first differences. Instrument weaknessinfluences the asymptotic and small-sample performance of the differenceestimator. Asymptotically, the variance of the coefficients rises. In smallsamples, weak instruments can bias the coefficients.

To reduce the potential biases and imprecision associated with the usualestimator, we use a new estimator that combines in a system the regressionin differences with the regression in levels. The instruments for the regres-

E X for s t Ti t s i t i t, , , ; , . . . , ( . )− −−( ) = ≥ =� ε ε 1 0 2 3 8 5

E y for s t Ti t s i t i t, , , ; , . . . , ( . )− −−( ) = ≥ =� ε ε 1 0 2 3 8 4

y y y y X Xi t i t i t i t i t i t i t i t, , , , , , , ,− = −( ) + ′ −( ) + −( )− − − − −1 1 2 1 1α β ε ε

y y Xi t i t i t i i t, , , , ( . )= + ′ + +−α β η ε1 8 3


2668-10_CH08.qxd 04/14/05 09:17 Page 200

sion in differences are the same as above. The instruments for the regres-sion in levels are the lagged differences of the corresponding variables.These are appropriate instruments under the following additional assump-tion: although there may be correlation between the levels of the right-handvariables and the country-specific effect in equation 8.3, there is no cor-relation between the differences of these variables and the country-specific effect. The following equation specifies this more formally, wherep, q, and t indicate time periods:

The additional moment conditions for the second part of the system (theregression in levels) are:

Thus, we use the moment conditions presented in equations 8.4, 8.5, 8.7,and 8.8, use instruments lagged two periods (t − 2), and employ a GMM pro-cedure to generate consistent and efficient parameter estimates.1

Consistency of the GMM estimator depends on the validity of the instru-ments. To address this issue we consider two specification tests. The first isa Sargan test of overidentifying restrictions, which tests the overall validityof the instruments by analyzing the sample analog of the moment conditionsused in the estimation process. The second test examines the hypothesis thatthe error term εi, t is not serially correlated. In both the difference regression

E X X for si t s i t s i i t, , , ( . )− − −−( ) +( ) = =1 0 1 8 8� η ε

E y y for si t s i t s i i t, , , ( . )− − −−( ) +( ) = =1 0 1 8 7� η ε

E y E y and E X E X

for all p and q

i t p i i t q i i t p i i t q i, , , ,

( . )

+ + + + = = � � � �η η η η

8 6


1. We use a variant of the standard two-step system estimator that controls for het-eroskedasticity. Typically, the system estimator treats the moment conditions as applying toa particular time period. This provides for a more flexible variance-covariance structure of themoment conditions because the variance for a given moment condition is not assumed to bethe same across time. The drawback of this approach is that the number of overidentifyingconditions increases dramatically as the number of time periods increases. Consequently, thistypical two-step estimator tends to induce overfitting and potentially biased standard errors.To limit the number of overidentifying conditions, we follow Beck and Levine (2003) byapplying each moment condition to all available periods. This reduces the overfitting bias ofthe two-step estimator. However, applying this modified estimator reduces the number ofperiods by one. While in the standard estimator time dummies and the constant are used asinstruments for the second period, this modified estimator does not allow the use of the firstand second periods. We confirm the results using the standard system estimator.

Recall that we assume that the explanatory variables are “weakly exogenous.” This meansthey can be affected by current and past realizations of the growth rate but not future real-izations of the error term. Weak exogeneity does not mean that agents do not take expectedfuture growth into account in their decision to undertake FDI—rather, it means that unantic-ipated shocks to future growth do not influence current FDI. We statistically assess the valid-ity of this assumption.

2668-10_CH08.qxd 04/14/05 09:17 Page 201

and the system difference–level regression, we test whether the differencederror term is second-order serially correlated (by construction, the differ-enced error term is probably first-order serially correlated even if the origi-nal error term is not).

The panel procedure also has disadvantages and limitations. The majordisadvantage relative to a pure cross-country comparison is that this studyfocuses on economic growth and seeks to abstract from business cycles andcrises. To use panel procedures, however, the data are averaged over five-year periods, which may not eliminate higher frequency forces. Thus, toassess the robustness of the results, we employ both OLS techniques that usedata averaged over more than 35 years and panel techniques that use dataaveraged over five-year periods. Furthermore, the panel procedure has lim-itations in that it does not solve all of the problems associated with cross-country regressions. For instance, FDI may have complex dynamic effects,such that the impact of FDI is different from the short run to the long run. Weprovide some sensitivity checks along this dimension by presenting resultsbased on data averaged over both 35 years and 5 years. Nevertheless, thisstudy does not attempt to trace the potential time-varying effects of FDI ongrowth. Finally, this study provides an aggregate examination. While a mul-titude of firm-level and industry-level studies of FDI exist, in particular coun-tries that attempt to assess the effects of specific policies (see the chapters byMoran as well as Lipsey and Sjöholm in this volume), this study undertakesa general assessment of the relationship between FDI and growth.

Data

We collected FDI data from two sources. First, we use data from the WorldBank’s ongoing project to improve the accuracy, breadth, and length ofnational accounts data (Kraay et al. 1999). Second, we confirm the findingsusing the IMF’s World Economic Output (2001) data on openness. We nowdefine each variable.

� FDI equals gross FDI inflows as a share of GDP. We confirm the resultsusing FDI inflows per capita.2

� GROWTH equals the rate of real per capita GDP growth.


2. Countries in the sample: Algeria (DZA), Argentina, Australia, Austria, Belgium, Bolivia,Brazil, Cameroon, Canada, Central African Republic, Chile, Colombia, Congo, Costa Rica,Cyprus, Denmark, Dominican Republic, Ecuador, El Salvador, Egypt, Finland, France, Gambia,Germany, Ghana, Britain, Greece, Guatemala, Guyana, Haiti, Honduras, Hong Kong, India,Indonesia, Ireland, Israel, Italy, Jamaica, Japan, Kenya, Lesotho, Malaysia, Malta, Mauritius,Mexico, Netherlands, New Zealand, Nicaragua, Niger, Norway, Pakistan, Panama, PapuaNew Guinea, Paraguay, Peru, Philippines, Portugal, Republic of Korea, Rwanda, Senegal, SierraLeone, South Africa, Spain, Sri Lanka, Suriname, Sweden, Switzerland, Syria, Togo, Thailand,Trinidad and Tobago, Uruguay, United States, Venezuela, Zaire, Zimbabwe.

2668-10_CH08.qxd 04/14/05 09:17 Page 202

To assess the link between international capital flows and economicgrowth and its sources, we control for other growth determinants: Initialincome per capita equals the logarithm of real per capita GDP at the start ofeach period, so that it equals 1960 in the pure cross-country analyses and,thereafter, the first year of each five-year period in the panel estimates.Average years of schooling equals the average years of schooling of the working-age population. Inflation equals the average growth rate in the consumer priceindex. Government size equals the size of the government as a share of GDP.Openness to trade equals exports plus imports relative to GDP. Black marketpremium equals the black market premium in the foreign exchange market.Private credit equals credit by financial intermediaries to the private sectoras a share of GDP (Beck, Levine, and Loayza 2000).

Tables 8.1a and 8.1b present summary statistics and correlations usingdata averaged over the 1960–95 period, with one observation per country.There is considerable cross-country variation. For instance, the mean percapita growth rate for the sample is 1.9 percent per annum, with a standarddeviation of 1.8. The maximum growth rate was enjoyed by South Korea(7.2), while Niger and Zaire suffered with a per capita growth rate of worsethan −2.7 percent per annum. In the five-year periods, the minimum valueis −10.0 percent growth (Rwanda 1990–95), and a number of countries expe-rienced five-year growth spurts of greater than 8 percent per annum. Thedata also suggest large variation in FDI with the average of 1.1 percent ofGDP. Malaysia as well as Trinidad and Tobago had FDI inflows of morethan 3.6 percent of GDP over the entire 1960–95 time period, while Sudanessentially had no FDI over this period. In terms of five-year periods, themaximum value of FDI was 7.3 percent of GDP (in Malaysia from 1990–95).The variability over five-year periods is much larger than when using lower-frequency data. Although tables 8.1a and 8.1b do not suggest a simple, pos-itive relationship between FDI and growth, we will see that many growthregression specifications yield a positive coefficient on FDI.

Results

This study estimates the effects of FDI inflows on economic growth aftercontrolling for other growth determinants and the potential biases inducedby endogeneity, country-specific effects, and the inclusion of initial incomeas a regressor. Moreover, we examine whether the growth effects of FDIdepend on the recipient country’s level of educational attainment, eco-nomic development, financial development, and trade openness.

Findings

Table 8.2 shows that the exogenous component of FDI does not exert a reli-able, positive impact on economic growth. The table presents OLS and panel


2668-10_CH08.qxd 04/14/05 09:17 Page 203

estimates using a variety of conditioning information sets. In the OLS regres-sions, initial income and average years of schooling enter significantly andwith the signs and magnitudes found in many pure cross-country regres-sions. FDI does not enter these growth regressions significantly. When wemove to the five-year panel data, FDI enters three of the regressions signifi-cantly but not the other four. FDI enters the regressions significantly and pos-itively in the regression that includes only initial income per capita andaverage years of schooling as control variables. FDI remains significantlyand positively linked with growth when controlling for inflation or govern-ment size. However, FDI becomes insignificant once we control for tradeopenness, the black market premium, or financial development. In sum, FDIis never significant in the OLS regressions and becomes insignificant in thepanel estimation when controlling for financial development or when con-trolling for international openness as proxied by either the trade share or theblack market premium.3

Furthermore, the coefficient on FDI is unstable in the panel regressions,ranging from 323 (when controlling for initial income, schooling, and infla-tion) to −34 (when controlling for initial income, schooling, and financialdevelopment). Changes in the sample do not cause this instability. Whenthe regressions are restricted to have the same number of observations, the


Table 8.1a Summary statistics, 1960–95Standard Minimum Maximum

Mean deviation value value

Growth rate 1.89 1.81 −2.81 7.16School (years of school in 1960) 5.01 2.51 1.20 11.07Inflation rate 0.16 0.18 0.04 0.91Government size (government

consumption/GDP) 0.15 0.05 0.07 0.31Openness to trade

(exports + imports/GDP) 0.60 0.37 0.14 2.32Black market premium 0.23 0.49 0.00 2.77Private credit 0.40 0.29 0.04 1.41FDI (as a share of GDP) 0.011 0.010 0.000 0.043

3. While some may argue that it is inappropriate to control for trade openness in assessingthe relationship between FDI and growth because trade openness may be closely associatedwith FDI openness, we disagree. It is important to know whether there is an independent rela-tionship between FDI and growth or whether FDI is some general proxy for openness, ratherthan representing a specific measure of FDI’s effect on growth. Moreover, the FDI-growthresults do not hold in any of the OLS regressions and the FDI-growth results vanish in thepanel regressions even without controlling for trade openness or the black market exchangerate premium.

2668-10_CH08.qxd 04/14/05 09:17 Page 204

205

Tab

le 8

.1b

Co

rrel

atio

n m

atri

x, 1

960–

95O

pen

nes

s B

lack

-mar

ket

Pri

vate

G

row

thS

cho

ola

Infl

atio

nb

Go

vern

men

t si

zea

to t

rad

ebp

rem

ium

bcr

edit

aF

DI

Gro

wth

1

Ave

rage

yea

rs o

f sch

oolin

ga0.

45*

1

Infla

tionb

−0.2

8*−0

.08

1

Gov

ernm

ent s

izea

0.24

*0.

42*

−0.2

8*1

Ope

nnes

s to

trad

ea0.

210.

04−0

.36*

0.33

*1

Bla

ck-m

arke

t pre

miu

mb

−0.4

3*−0

.40*

0.38

*−0

.20

0.07

1

Priv

ate

cred

ita0.

55*

0.68

*−0

.43*

0.39

*0.

03−0

.60*

1

FD

I0.

170.

12−0

.21

0.23

0.56

*−0

.01

0.05

1

* =

indi

cate

s si

gnifi

canc

e at

the

0.05

leve

l.

Not

e: T

his

tabl

e is

bas

ed o

n a

com

mon

sam

ple

of 6

4 co

untr

ies

usin

g th

e av

erag

e be

twee

n 19

60 a

nd 1

995,

with

one

obs

erva

tion

per

coun

try.

a. In

the

corr

elat

ions

, thi

s va

riabl

e is

incl

uded

as

Ln(v

aria

ble)

.b.

In th

e co

rrel

atio

ns, t

his

varia

ble

is in

clud

ed a

s Ln

(1 +

varia

ble)

.

2668-10_CH08.qxd 04/14/05 09:17 Page 205


coefficient on FDI remains unstable.4 Note that the Sargan and serial cor-relation tests do not reject the econometric specification. The table 8.2regressions do not reject the null hypothesis that FDI does not exert anindependent influence on economic growth.

We also assess whether the impact of FDI on growth depends heavily onthe stock of human capital (table 8.3). Borensztein, De Gregorio, and Lee(1998) find that in countries with low levels of human capital the directeffect of FDI on growth is negative, though sometimes insignificant. Butonce human capital passes a threshold, they find that FDI has a positivegrowth effect. The rationale is that only countries with sufficiently highlevels of human capital can exploit the technological spillovers associatedwith FDI. Thus, we include the interaction term FDI*School, which equalsthe product of FDI and the average years of schooling of the working-agepopulation.

Table 8.3 shows that the lack of FDI impact on growth does not dependon the stock of human capital. In the OLS regressions, FDI and the interac-tion term do not enter significantly in any of the six regressions. In the panelregressions, FDI and the interaction term occasionally enter significantly,but even here the results do not conform to theory. Namely, when FDI andthe interaction term do enter significantly, the term on FDI is significant andthe coefficient on the interaction term is negative. This suggests that FDI isonly growth enhancing in countries with low educational attainment. Thesecounterintuitive results may result from including schooling, FDI, and theinteraction term simultaneously.5 When excluding schooling, however,the regressions do not yield robust results with a positive coefficient onthe interaction term.

Finally, we also examined the importance of human capital using an alter-native specification. Instead of including the interaction term FDI*School,we created a dummy variable, D, that takes on the value 1 if the countryhas greater than average schooling and 0 otherwise. We then included theterm FDI*D. This specification also indicated that FDI’s impact on growth

4. Also, note that the coefficient on FDI is frequently, though not always, an order of magni-tude larger in the panel than the OLS regressions. We speculate that this occurs because ofmore volatile data. When we restrict the sample to wealthier countries (which are also coun-tries with less volatile growth rates), the panel coefficient on FDI is similar to the OLS regres-sion coefficients. Similarly, when we use the IMF’s World Economic Outlook data, whichcontains fewer and very poor, highly volatile countries than the World Bank data, the panelcoefficients are closer to the coefficients from the OLS regressions. These estimates are con-sistent with the view that short-run fluctuations in the investment environment, and henceFDI, are associated with large, though temporary, booms and busts in economic performance.Thus, the use of higher frequency data produces larger (though still insignificant) coefficientson FDI than pure cross-country regressions with data averaged over the 1960–95 period.

5. This conjecture is supported by the observation that no country passes the inflection point.For instance, from the panel results in regression six, 351 divided by 108.6 equals 3.23, but thehighest level of school attainment is 2.4 in Denmark.

2668-10_CH08.qxd 04/14/05 09:17 Page 206

Tab

le 8

.2G

row

th a

nd

FD

I reg

ress

ion

s, 5

-yea

r p

erio

ds

1960

–95

12

34

56

7O

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

el

Con

stan

t6.

797

−0.7

237.

732

9.32

47.

363

−10.

640

6.22

25.

646

7.10

32.

391

11.5

795.

256

11.7

022.

701

(0.0

09)

(0.8

96)

(0.0

02)

(0.3

14)

(0.0

15)

(0.3

03)

(0.0

74)

(0.2

59)

(0.0

06)

(0.7

16)

(0.0

00)

(0.3

32)

(0.0

00)

(0.6

68)

Initi

al in

com

e −1

.175

−0.2

52−1

.226

−3.0

26−1

.274

−1.5

22−1

.236

0.23

3−1

.191

−0.6

67−1

.414

0.72

0−1

.643

−0.5

08pe

r ca

pita

a(0

.008

)(0

.854

)(0

.003

)(0

.254

)(0

.005

)(0

.500

)(0

.006

)(0

.822

)(0

.007

)(0

.708

)(0

.000

)(0

.415

)(0

.000

)(0

.679

)

Ave

rage

yea

rs

2.75

22.

551

2.77

48.

629

2.97

96.

770

2.93

40.

096

2.66

12.

480

1.84

0−2

.576

2.11

51.

617

of s

choo

lingb

(0.0

00)

(0.4

07)

(0.0

00)

(0.1

82)

(0.0

00)

(0.1

95)

(0.0

00)

(0.9

67)

(0.0

01)

(0.5

56)

(0.0

03)

(0.2

30)

(0.0

01)

(0.6

96)

Infla

tionb

−3.3

77−0

.887

1.39

8−0

.161

(0.0

34)

(0.8

39)

(0.3

55)

(0.9

49)

Gov

ernm

ent s

izea

−0.0

83−6

.461

−0.8

54−2

.796

(0.8

78)

(0.0

60)

(0.1

27)

(0.1

65)

Ope

nnes

s to

trad

ea0.

193

4.83

00.

427

1.66

4(0

.650

)(0

.000

)(0

.329

)(0

.375

)

Bla

ck m

arke

t pre

miu

mb

−0.2

92−0

.590

−1.0

28−1

.505

(0.7

92)

(0.6

45)

(0.2

72)

(0.2

85)

Priv

ate-

sect

or c

redi

tb1.

397

2.26

21.

714

1.25

0(0

.000

)(0

.027

)(0

.001

)(0

.333

)

FD

I 12

.553

202.

167

2.85

232

2.93

316

.598

215.

245

10.6

7717

.045

12.5

5822

0.85

414

.854

−34.

511

21.9

31−9

.434

(0.5

82)

(0.0

06)

(0.8

97)

(0.0

51)

(0.4

69)

(0.0

49)

(0.6

31)

(0.7

48)

(0.5

79)

(0.1

60)

(0.4

14)

(0.6

09)

(0.2

38)

(0.9

17)

Co

nd

itio

nin

g

info

rmat

ion

set

(tab

le c

ontin

ues

next

pag

e)

207

2668-10_CH08.qxd 04/14/05 09:17 Page 207

Tab

le 8

.2G

row

th a

nd

FD

I reg

ress

ion

s (c

ontin

ued)

12

34

56

7O

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

elO

LS

Pan

el

Num

ber

of o

bser

vatio

nsc

6827

968

270

6827

367

277

6626

067

246

6424

2

R2

(adj

uste

d)0.

238

0.28

70.

238

0.25

80.

209

0.43

70.

510

Sar

gan

test

(p-

valu

e)d

0.09

80.

770

0.75

60.

299

0.30

20.

304

0.19

1

Ser

ial c

orre

latio

n te

st (

p-va

lue)

e0.

939

0.92

20.

897

0.58

00.

805

0.23

40.

256

OLS

= o

rdin

ary

leas

t squ

ares

a. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(var

iabl

e).

b. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(1 +

varia

ble)

.c.

Pan

el e

stim

atio

ns u

se fi

ve-y

ear

perio

ds.

d. T

he n

ull h

ypot

hesi

s is

that

the

inst

rum

ents

are

not

cor

rela

ted

with

the

resi

dual

s.e.

The

nul

l hyp

othe

sis

is th

at th

e er

rors

in th

e fir

st-d

iffer

ence

reg

ress

ion

exhi

bit n

o se

cond

-ord

er s

eria

l cor

rela

tion.

Not

es: D

epen

dent

var

iabl

e is

rea

l per

cap

ita G

DP

gro

wth

. P-v

alue

s ar

e in

par

enth

eses

bel

ow e

stim

ates

’ coe

ffici

ent v

alue

s.

Co

nd

itio

nin

g

info

rmat

ion

set

208

2668-10_CH08.qxd 04/14/05 09:17 Page 208

Tab

le 8

.3G

row

th, F

DI,

and

ed

uca

tio

n r

egre

ssio

ns

12

34

56

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

Con

stan

t6.

841

1.50

47.

727

11.7

657.

312

−21.

189

6.05

06.

882

7.25

0−3

.460

6.81

2−4

.611

(0.0

11)

(0.8

57)

(0.0

03)

(0.2

52)

(0.0

17)

(0.1

20)

(0.0

93)

(0.1

79)

(0.0

07)

(0.6

51)

(0.0

29)

(0.5

13)

Initi

al in

com

e−1

.175

−1.4

84−1

.226

−4.7

18−1

.281

−2.3

46−1

.238

−0.6

25−1

.190

−0.6

31−1

.391

−3.8

43

per

capi

taa

(0.0

08)

(0.4

51)

(0.0

03)

(0.0

91)

(0.0

05)

(0.2

95)

(0.0

07)

(0.5

93)

(0.0

07)

(0.7

38)

(0.0

02)

(0.0

12)

Ave

rage

yea

rs

2.72

17.

025

2.77

815

.183

3.12

012

.607

3.05

22.

612

2.55

75.

520

3.41

514

.161

of s

choo

lingb

(0.0

01)

(0.1

11)

(0.0

00)

(0.0

26)

(0.0

02)

(0.0

15)

(0.0

00)

(0.3

41)

(0.0

06)

(0.1

91)

(0.0

01)

(0.0

00)

Infla

tionb

−3.3

78−2

.783

−3.8

12−6

.959

(0.0

35)

(0.5

86)

(0.0

52)

(0.0

26)

Gov

ernm

ent s

izea

−0.1

22−1

0.23

3−0

.555

−7.2

42(0

.837

)(0

.015

)(0

.388

)(0

.013

)

Ope

nnes

s to

trad

ea0.

199

4.01

2−0

.078

1.70

6(0

.644

)(0

.005

)(0

.871

)(0

.440

)

Bla

ck m

arke

t −0

.314

0.69

00.

037

2.25

6pr

emiu

mb

(0.7

82)

(0.5

49)

(0.9

77)

(0.0

14)

FD

I7.

585

471.

575

3.46

056

7.93

535

.139

588.

334

28.2

8415

5.47

8−2

.463

681.

882

46.0

7835

1.00

0(0

.901

)(0

.010

)(0

.953

)(0

.028

)(0

.604

)(0

.004

)(0

.618

)(0

.040

)(0

.970

)(0

.000

)(0

.485

)(0

.000

)

FD

I*S

choo

l3.

350

−183

.992

−0.4

11−1

61.5

01−1

2.17

9−2

50.2

33−1

1.90

5−4

8.64

010

.084

−243

.945

−23.

042

−108

.606

(0.9

35)

(0.0

36)

(0.9

92)

(0.1

98)

(0.7

85)

(0.0

63)

(0.7

56)

(0.2

32)

(0.8

17)

(0.0

00)

(0.6

06)

(0.0

14)

Co

nd

itio

nin

g

info

rmat

ion

set

(tab

le c

ontin

ues

next

pag

e)

209

2668-10_CH08.qxd 04/14/05 09:17 Page 209

Tab

le 8

.3G

row

th, F

DI,

and

ed

uca

tio

n r

egre

ssio

ns

(con

tinue

d)

12

34

56

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

Num

ber

of

obse

rvat

ions

c68

279

6827

066

273

6727

766

260

6524

8

R2

(adj

uste

d)0.

226

0.27

50.

226

0.24

70.

197

0.25

8

Sar

gan

test

(p-

valu

e)d

0.34

00.

690

0.82

80.

286

0.32

40.

144

Ser

ial c

orre

latio

n te

st

(p-v

alue

)e0.

332

0.50

60.

273

0.28

30.

158

0.22

1

OLS

= o

rdin

ary

leas

t squ

ares

a. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(var

iabl

e).

b. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(1 +

varia

ble)

.c.

Pan

el e

stim

atio

ns u

se fi

ve-y

ear

perio

ds.

d. T

he n

ull h

ypot

hesi

s is

that

the

inst

rum

ents

are

not

cor

rela

ted

with

the

resi

dual

s.e.

The

nul

l hyp

othe

sis

is th

at th

e er

rors

in th

e fir

st-d

iffer

ence

reg

ress

ion

exhi

bit n

o se

cond

-ord

er s

eria

l cor

rela

tion.

Not

es: D

epen

dent

var

iabl

e is

rea

l per

cap

ita G

DP

gro

wth

. P-v

alue

s ar

e in

par

enth

eses

bel

ow e

stim

ates

’ coe

ffici

ent v

alue

s.

Co

nd

itio

nin

g

info

rmat

ion

set

210

2668-10_CH08.qxd 04/14/05 09:17 Page 210


does not robustly vary with the level of educational attainment. Whilesome may interpret the results in table 8.3 as suggesting that the coefficienton FDI becomes significant and positive in the panel regressions when con-trolling for the interaction with schooling, we note that (1) the interactionterms are frequently insignificant, (2) the signs do not conform with theory,and (3) the OLS regressions suggest a fragile relationship.

Since Blomström, Lipsey, and Zejan (1994) argue that very poor coun-tries—countries that are extremely technologically backward—are unable toexploit FDI, we reran the regressions using the interaction term, FDI*Incomeper capita. Table 8.4 shows, however, that a reliable link between growth andFDI when allowing for FDI’s impact on growth to depend on the level ofincome per capita does not exist.6

Table 8.5 assesses whether the level of financial development in the recip-ient country influences the growth-FDI relationship. Better-developedfinancial systems improve capital allocation and stimulate growth (Beck,Levine, and Loayza 2000). Capital inflows to a country with a well-developedfinancial system may, therefore, produce substantial growth effects. Thus,we reran the regressions using the interaction term FDI*Credit.

Although the OLS regressions in table 8.5 suggest that FDI has a positivegrowth effect, especially in financially developed economies, the panel evi-dence does not confirm this finding. The panel regressions never demon-strated a significant coefficient on the FDI-financial development interactionterm. On net, these results do not provide much support for the view thatFDI flows to financially developed economies exert an exogenous impacton growth.

Table 8.6 assesses whether the relationship between FDI and growthvaries with the degree of trade openness. Balasubramanyam, Salisu, andSapsford (1996, 1999) find evidence that FDI is particularly good for eco-nomic growth in countries with open trade regimes. Thus, we include aninteraction term of FDI and openness to trade in the table 8.6 regressions.The FDI*Trade interaction term does not enter significantly in any of theOLS regressions. While the FDI*Trade interaction term enters significantlyat the 0.10 level in three of the panel regressions, it enters insignificantly inthe other three. In sum, we do not find a robust link between FDI andgrowth even when allowing this relationship to vary with trade openness.

While FDI flows may go hand in hand with economic success, they do nottend to exert an independent growth effect. Thus, by correcting statistical

6. The only regression where the interaction enters significantly is the regression controllingonly for the black market premium. Even here, however, the interaction term enters nega-tively, and does not alter the relationship for hardly any country in the sample because thecutoff is so high, e.g., the logarithm of real per capita GDP would have to be greater than1114.7 divided by 110.4 equals 10.1, which is the case for only a handful of countries duringthe end of the sample.

2668-10_CH08.qxd 04/14/05 09:17 Page 211

Tab

le 8

.4G

row

th, F

DI,

and

inco

me

leve

l reg

ress

ion

s

12

34

56

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

Con

stan

t4.

609

−5.2

545.

623

8.40

05.

263

−15.

806

4.49

34.

792

5.02

9−4

.906

3.76

5−3

.550

(0.2

09)

(0.4

59)

(0.1

02)

(0.4

46)

(0.1

67)

(0.2

13)

(0.2

93)

(0.4

10)

(0.1

78)

(0.5

62)

(0.3

68)

(0.6

75)

Initi

al in

com

e −0

.880

0.32

0−0

.942

−3.3

56−0

.939

−1.6

38−0

.961

−0.2

47−0

.918

−0.1

13−1

.002

−1.3

40pe

r ca

pita

a(0

.115

)(0

.837

)(0

.071

)(0

.225

)(0

.101

)(0

.457

)(0

.090

)(0

.829

)(0

.100

)(0

.952

)(0

.072

)(0

.315

)

Ave

rage

yea

rs

2.69

82.

731

2.72

310

.933

2.99

88.

922

2.90

12.

240

2.63

54.

043

3.20

56.

488

of s

choo

lingb

(0.0

00)

(0.3

77)

(0.0

00)

(0.0

75)

(0.0

00)

(0.0

57)

(0.0

00)

(0.3

91)

(0.0

01)

(0.3

27)

(0.0

00)

(0.0

18)

Infla

tionb

−3.3

54−2

.248

−4.0

78−4

.433

(0.0

34)

(0.6

09)

(0.0

34)

(0.1

24)

Gov

ernm

ent s

izea

−0.2

82−7

.663

−0.6

62−4

.512

(0.6

27)

(0.0

29)

(0.2

88)

(0.0

90)

Ope

nnes

s to

trad

ea0.

100

4.03

4−0

.239

2.91

8(0

.813

)(0

.005

)(0

.618

)(0

.173

)

Bla

ck m

arke

t −0

.232

0.89

30.

127

1.10

5pr

emiu

mb

(0.8

40)

(0.5

72)

(0.9

20)

(0.2

57)

Co

nd

itio

nin

g

info

rmat

ion

set

212

2668-10_CH08.qxd 04/14/05 09:17 Page 212

FD

I22

4.57

661

0.12

320

6.63

866

4.20

226

8.11

166

9.82

222

6.79

125

4.81

020

9.55

011

14.6

5532

2.87

931

1.72

9(0

.265

)(0

.055

)(0

.289

)(0

.149

)(0

.219

)(0

.178

)(0

.245

)(0

.421

)(0

.312

)(0

.030

)(0

.131

)(0

.137

)

FD

I*In

com

e −2

7.39

8−5

3.44

3−2

6.32

5−4

6.45

7−3

2.29

4−5

6.91

0−2

7.56

7−2

2.90

0−2

5.43

8−1

10.3

59−3

9.59

1−3

0.88

8pe

r ca

pita

(0.2

57)

(0.2

02)

(0.2

62)

(0.4

63)

(0.2

19)

(0.3

85)

(0.2

41)

(0.6

07)

(0.3

07)

(0.0

43)

(0.1

25)

(0.3

12)

Num

ber

of

obse

rvat

ions

c68

279

6827

065

273

6727

766

260

6524

8

R2

(adj

uste

d)0.

237

0.28

60.

240

0.25

70.

206

0.36

7

Sar

gan

test

(p-

valu

e)d

0.19

10.

745

0.82

10.

322

0.44

00.

082

Ser

ial c

orre

latio

n te

st (

p-va

lue)

e0.

553

0.87

10.

935

0.68

00.

405

0.58

7

OLS

= o

rdin

ary

leas

t squ

ares

a. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(var

iabl

e).

b. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(1 +

varia

ble)

.c.

Pan

el e

stim

atio

ns u

se fi

ve-y

ear

perio

ds.

d. T

he n

ull h

ypot

hesi

s is

that

the

inst

rum

ents

are

not

cor

rela

ted

with

the

resi

dual

s.e.

The

nul

l hyp

othe

sis

is th

at th

e er

rors

in th

e fir

st-d

iffer

ence

reg

ress

ion

exhi

bit n

o se

cond

-ord

er s

eria

l cor

rela

tion.

Not

es: D

epen

dent

var

iabl

e is

rea

l per

cap

ita G

DP

gro

wth

. P-v

alue

s ar

e in

par

enth

eses

bel

ow e

stim

ates

’ coe

ffici

ent v

alue

s.

213

2668-10_CH08.qxd 04/14/05 09:17 Page 213

Tab

le 8

.5G

row

th, F

DI,

and

fin

ance

reg

ress

ion

s

12

34

56

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

Con

stan

t9.

236

4.45

39.

380

−7.6

519.

609

−4.3

378.

887

8.21

79.

454

0.38

39.

119

−4.0

88(0

.000

)(0

.592

)(0

.000

)(0

.146

)(0

.001

)(0

.508

)(0

.007

)(0

.094

)(0

.000

)(0

.935

)(0

.001

)(0

.627

)

Initi

al in

com

e −1

.407

−0.7

24−1

.401

1.49

8−1

.479

1.78

0−1

.460

−0.7

43−1

.397

0.62

4−1

.465

−0.6

50pe

r ca

pita

a(0

.000

)(0

.712

)(0

.000

)(0

.215

)(0

.000

)(0

.235

)(0

.001

)(0

.453

)(0

.001

)(0

.620

)(0

.001

)(0

.723

)

Ave

rage

yea

rs

2.29

42.

087

2.35

8−0

.596

2.48

3−1

.910

2.47

72.

637

2.16

2−1

.030

2.50

33.

060

of s

choo

lingb

(0.0

00)

(0.6

30)

(0.0

00)

(0.8

13)

(0.0

01)

(0.5

50)

(0.0

00)

(0.2

40)

(0.0

02)

(0.7

46)

(0.0

01)

(0.4

58)

Infla

tionb

−1.7

30−2

.584

−1.1

18−2

.123

(0.2

22)

(0.1

97)

(0.4

64)

(0.4

75)

Gov

ernm

ent s

izea

−0.0

611.

600

−0.3

25−4

.397

(0.9

11)

(0.3

26)

(0.5

73)

(0.0

71)

Ope

nnes

s to

trad

ea0.

114

4.44

80.

155

0.50

6(0

.753

)(0

.001

)(0

.714

)(0

.824

)

Bla

ck m

arke

t −0

.732

−4.5

89−1

.162

−3.9

00pr

emiu

mb

(0.3

36)

(0.0

62)

(0.1

00)

(0.0

34)

Co

nd

itio

nin

g

info

rmat

ion

set

214

2668-10_CH08.qxd 04/14/05 09:17 Page 214

FD

I 15

2.32

3−3

40.1

0613

3.01

671

.044

152.

237

−107

.266

147.

760

−40.

957

141.

844

−237

.720

119.

251

−300

.341

(0.0

00)

(0.2

22)

(0.0

00)

(0.6

24)

(0.0

00)

(0.4

31)

(0.0

00)

(0.7

75)

(0.0

01)

(0.2

63)

(0.0

00)

(0.0

46)

FD

I*C

redi

t12

3.54

113

6.39

811

0.61

5−8

.229

120.

562

41.4

6911

9.49

533

.787

113.

364

62.6

7593

.643

84.2

42(0

.000

)(0

.100

)(0

.000

)(0

.855

)(0

.000

)(0

.347

)(0

.000

)(0

.429

)(0

.000

)(0

.218

)(0

.001

)(0

.133

)

Num

ber

of

obse

rvat

ions

c67

269

6726

465

263

6626

765

250

6424

2

R2

(adj

uste

d)0.

441

0.44

70.

442

0.45

60.

432

0.45

1

Sar

gan

test

(p-

valu

e)d

0.04

30.

012

0.03

40.

116

0.07

00.

306

Ser

ial c

orre

latio

n te

st (

p-va

lue)

e0.

787

0.99

20.

206

0.35

60.

213

0.14

5

OLS

= o

rdin

ary

leas

t squ

ares

a. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(var

iabl

e).

b. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(1 +

varia

ble)

.c.

Pan

el e

stim

atio

ns u

se fi

ve-y

ear

perio

ds.

d. T

he n

ull h

ypot

hesi

s is

that

the

inst

rum

ents

are

not

cor

rela

ted

with

the

resi

dual

s.e.

The

nul

l hyp

othe

sis

is th

at th

e er

rors

in th

e fir

st-d

iffer

ence

reg

ress

ion

exhi

bit n

o se

cond

-ord

er s

eria

l cor

rela

tion.

Not

es: D

epen

dent

var

iabl

e is

rea

l per

cap

ita G

DP

gro

wth

. P-v

alue

s ar

e in

par

enth

eses

bel

ow e

stim

ates

’ coe

ffici

ent v

alue

s.

215

2668-10_CH08.qxd 04/14/05 09:17 Page 215

Tab

le 8

.6G

row

th, F

DI,

and

tra

de

op

enn

ess

reg

ress

ion

s

12

34

56

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

OL

SP

anel

Con

stan

t6.

462

4.53

17.

563

10.9

715.

700

−0.8

766.

366

5.41

96.

935

6.62

06.

336

2.52

4(0

.018

)(0

.478

)(0

.004

)(0

.255

)(0

.055

)(0

.918

)(0

.020

)(0

.330

)(0

.011

)(0

.376

)(0

.027

)(0

.706

)

Initi

al in

com

e −1

.135

−1.1

20−1

.230

−3.1

68−1

.114

−2.6

98−1

.137

−0.2

16−1

.151

−1.3

93−1

.270

−5.6

37pe

r ca

pita

a(0

.013

)(0

.482

)(0

.004

)(0

.242

)(0

.018

)(0

.257

)(0

.014

)(0

.863

)(0

.012

)(0

.504

)(0

.005

)(0

.005

)

Ave

rage

yea

rs

2.81

25.

182

2.87

89.

036

2.84

79.

223

2.80

62.

519

2.65

94.

603

2.99

116

.644

of s

choo

lingb

(0.0

00)

(0.1

55)

(0.0

00)

(0.1

87)

(0.0

00)

(0.1

00)

(0.0

00)

(0.4

13)

(0.0

01)

(0.3

73)

(0.0

00)

(0.0

01)

Infla

tionb

−3.0

57−2

.353

−3.6

09−9

.122

(0.0

61)

(0.5

29)

(0.0

65)

(0.0

14)

Gov

ernm

ent s

izea

−0.2

81−4

.762

−0.5

52−6

.782

(0.5

98)

(0.0

84)

(0.3

54)

(0.0

05)

Ope

nnes

s to

trad

ea−0

.152

4.86

9−0

.442

−3.5

53(0

.734

)(0

.001

)(0

.369

)(0

.068

)

Bla

ck m

arke

t −0

.605

−1.8

23−0

.139

0.55

5pr

emiu

mb

(0.6

54)

(0.1

76)

(0.9

19)

(0.6

25)

Co

nd

itio

nin

g

info

rmat

ion

set

216

2668-10_CH08.qxd 04/14/05 09:17 Page 216

FD

I16

.430

150.

596

7.31

023

4.04

817

.881

201.

450

20.8

5075

.550

16.8

9499

.801

22.9

6123

6.67

1(0

.458

)(0

.041

)(0

.746

)(0

.106

)(0

.435

)(0

.037

)(0

.473

)(0

.109

)(0

.417

)(0

.504

)(0

.424

)(0

.009

)

FD

I*T

rade

29.2

4125

9.74

817

.771

56.6

0533

.007

217.

435

35.4

5689

.843

33.8

8014

8.27

939

.920

324.

020

(0.4

91)

(0.0

01)

(0.6

70)

(0.6

26)

(0.4

45)

(0.0

53)

(0.4

79)

(0.1

62)

(0.3

70)

(0.2

37)

(0.3

61)

(0.0

08)

Num

ber

of

obse

rvat

ions

c67

276

6726

766

270

6727

565

257

6524

5

R2

(adj

uste

d)0.

269

0.30

50.

241

0.25

80.

249

0.27

0

Sar

gan

test

(p-

valu

e)d

0.65

50.

825

0.93

10.

589

0.38

70.

876

Ser

ial c

orre

latio

n te

st (

p-va

lue)

e0.

318

0.94

00.

996

0.44

30.

985

0.66

7

OLS

= o

rdin

ary

leas

t squ

ares

a. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(var

iabl

e).

b. In

the

regr

essi

on, t

his

varia

ble

is in

clud

ed a

s Ln

(1 +

varia

ble)

.c.

Pan

el e

stim

atio

ns u

se fi

ve-y

ear

perio

ds.

d. T

he n

ull h

ypot

hesi

s is

that

the

inst

rum

ents

are

not

cor

rela

ted

with

the

resi

dual

s.e.

The

nul

l hyp

othe

sis

is th

at th

e er

rors

in th

e fir

st-d

iffer

ence

reg

ress

ion

exhi

bit n

o se

cond

-ord

er s

eria

l cor

rela

tion.

Not

es: D

epen

dent

var

iabl

e is

rea

l per

cap

ita G

DP

gro

wth

. P-v

alue

s ar

e in

par

enth

eses

bel

ow e

stim

ates

’ coe

ffici

ent v

alue

s.

217

2668-10_CH08.qxd 04/14/05 09:17 Page 217

shortcomings with past work this study reconciles the broad cross-countryevidence with microeconomic studies.

Sensitivity Analyses

We conduct a number of sensitivity analyses to assess the robustness of theresults. First, we use a standard instrumental variable estimator in a purecross-country context (one observation per country) and reexamine whethercross-country variations in the exogenous component of FDI explain cross-country variations in the rate of economic growth. We use GMM.7 We alsouse linear moment conditions, which amounts to the requirement that theinstrumental variables (Z) are uncorrelated with the error term in the growthregression in equation 8.1. The economic meaning of these conditions is thatthe instrumental variables can only affect growth through FDI and the othervariables in the conditioning information set. To test this condition, we testthe overidentifying restrictions, and we cannot reject the given moment con-ditions. The GMM results confirm this study’s results.

Second, we confirm this study’s findings using two alternative estima-tors. Instead of using Calderon, Chong, and Loayza’s (2000) method of lim-iting the possibility of overfitting by restricting the dimensionality of theinstrument set (described above), we use the standard system estimator.In addition, although the standard estimator and Calderon, Chong, andLoayza’s (2000) modification are two-step estimators where the variance-covariance matrix is constructed from the first-stage residuals to allow fornonspherical distributions of the error term (and thereby get more efficientestimates in the second stage), these two-step GMM estimators sometimesconverge to their asymptotic distributions slowly. This tends to bias thet-statistics upward. Nonetheless, we reran the regressions using the first-stage results, which assume homoskedasticity and independence of theerror terms.

Third, we used a variety of alternative samples and specifications. Asnoted by Blonigen and Wang (in this volume), there may be concerns aboutmixing rich and poor countries in empirical studies of FDI and growth.Nonetheless, limiting the sample to developing countries—i.e., countriesnot classified by the World Bank as high-income economies—does not alterthe findings. Also, when using a common sample across all of the regres-sions, the results do not change. Similarly, using the natural logarithm ofFDI does not alter the conclusions. We also considered exchange rate volatil-ity, changes in the terms of trade in the regression, and various combinationsof the conditioning information set (Levine and Renelt 1992). Including thesefactors did not alter the conclusions. This study does not prove that FDIis unimportant. Rather, this cross-country analysis—in conjunction with


7. Two-stage instrumental variable procedures produce the same conclusions.

2668-10_CH08.qxd 04/14/05 09:17 Page 218

microeconomic evidence—reduces confidence in the belief that FDI acceler-ates GDP growth.

Fourth, we examined whether FDI affects productivity growth using theEasterly and Levine (2001) measure of total factor productivity (TFP). Wefound that FDI does not exert a robust impact on TFP.

Fifth, we examined portfolio inflows and found that they do not have apositive impact on growth.

Finally, we repeated the analyses using the IMF’s World Economic Outlook2001 new database on international capital flows. The IMF cleaned the dataand extended the findings through the end of 2000. The results are verysimilar to those reported above, so we do not report them.

Conclusion

FDI has increased dramatically since the 1980s. Furthermore, many coun-tries have offered special tax incentives and subsidies to attract foreign cap-ital. An influential economic rationale for treating foreign capital favorablyis that FDI and portfolio inflows encourage technology transfers that accel-erate overall economic growth in recipient countries. While microeconomicstudies generally, though not uniformly, shed pessimistic evidence on thegrowth effects of foreign capital, many macroeconomic studies find a posi-tive link between FDI and growth. Previous macroeconomic studies, how-ever, do not fully control for endogeneity, country-specific effects, and theinclusion of lagged dependent variables in the growth regression.

After resolving many of the statistical problems plaguing past macroeco-nomic studies and confirming our results using two new databases on inter-national capital flows, we find that FDI inflows do not exert an independentinfluence on economic growth. Thus, while sound economic policies mayspur both growth and FDI, the results are inconsistent with the view thatFDI exerts a positive impact on growth that is independent of other growthdeterminants.

References

Aitken, Brian, and Ann Harrison. 1999. Do Domestic Firms Benefit from Foreign Direct In-vestment? Evidence from Venezuela. American Economic Review 89, no. 3 (June): 605–18.

Alfaro, Laura, Chanda Areendam, Sebnem Kalemli-Ozcan, and Sayek Selin. 2003. FDI andEconomic Growth: The Role of Local Financial Markets. Journal of International Economics61, no. 1 (October): 512–33.

Balasubramanyam, V. N., Mohammed Salisu, and David Sapsford. 1996. Foreign DirectInvestment and Growth in EP and IS Countries. Economic Journal 106, no. 434 (January):92–105.

Balasubramanyam, V. N., Mohammed Salisu, and David Sapsford. 1999. Foreign Direct Invest-ment as an Engine of Growth. Journal of International Trade and Economic Development 8,no. 1: 27–40.


2668-10_CH08.qxd 04/14/05 09:17 Page 219

Beck, Thorsten, and Ross Levine. 2003. Stock Markets, Banks, and Growth: Panel Evidence.Journal of Banking and Finance 28, no. 3 (March): 423–42.

Beck, Thorsten, Ross Levine, and Norman Loayza. 2000. Finance and the Sources of Growth.Journal of Financial Economics 58, no. 1-2: 261–300.

Bekaert, Geert, Campbell R. Harvey, and Christian Lundblad. 2001. Does Financial LiberalizationSpur Growth? NBER Working Paper 8245. Cambridge, MA: National Bureau of EconomicResearch.

Blomström, Magnus. 1986. Foreign Investment and Productive Efficiency: The Case ofMexico. The Journal of Industrial Economics 35, no. 1 (April): 97–110.

Blomström, Magnus, Robert E. Lipsey, and Mario Zejan. 1994. What Explains DevelopingCountry Growth? In Convergence and Productivity: Gross-National Studies and HistoricalEvidence, ed. William Baumol, Richard Nelson, and Edward Wolff. Oxford: OxfordUniversity Press.

Borensztein, E., J. De Gregorio, and J. W. Lee. 1998. How Does Foreign Investment AffectGrowth? Journal of International Economics 45, no. 1: 115–72.

Boyd, John H., and Bruce D. Smith. 1992. Intermediation and the Equilibrium Allocation ofInvestment Capital: Implications for Economic Development. Journal of MonetaryEconomics 30: 409–32.

Calderon, Cesar, Alberto Chong, and Norman Loayza. 2000. Determinants of Current AccountDeficits in Developing Countries. World Bank Research Policy Working Paper 2398 (July).Washington: World Bank.

Easterly, William, and Ross Levine. 2001. It’s Not Factor Accumulation: Stylized Facts andGrowth Models. World Bank Economic Review 15: 177–219.

Haddad, Mona, and Ann Harrison. 1993. Are There Positive Spillovers from Direct ForeignInvestment? Evidence from Panel Data for Morocco. Journal of Development Economics 42(October): 51–74.

Klein, Michael, and Giovanni Olivei. 2000. Capital Account Liberalization, Financial Depth,and Economic Growth. Somerville, MA: Tufts University. Photocopy.

Kraay, Aart, Norman Loayza, Luis Serven, and Jaime Ventura. 1999. Country Portfolios.Washington: World Bank. Photocopy.

La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert W. Vishny. 1999.The Quality of Government. Journal of Law, Economics, and Organization 15, no. 1: 222–79.

Levine, Ross, and Maria Carkovic. 2001. How Much Bang for the Buck: Mexico andDollarization. Journal of Money, Credit, and Banking 33, no. 2 (May): 339–63.

Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. Financial Intermediation andGrowth: Causality and Causes. Journal of Monetary Economics 46: 31–77.

Levine, Ross, and David Renelt. 1992. A Sensitivity Analysis of Cross-Country GrowthRegressions. American Economic Review 82, no. 4 (September): 942–63.

Romer, Paul. 1993. Idea Gaps and Object Gaps in Economic Development. Journal of MonetaryEconomics 32, no. 3 (December): 543–73.

World Bank. 1997a. Private Capital Flows to Developing Countries: The Road to Financial Integration.Washington.

World Bank. 1997b. Global Development Finance 1997. Washington.


2668-10_CH08.qxd 04/14/05 09:17 Page 220

Does Foreign Direct Investment Accelerate Economic Growth? · Does Foreign Direct Investment Accelerate Economic Growth? MARIA CARKOVIC and ROSS LEVINE 195 With commercial bank lending

Documents