Economic Reforms, FDI, and Economic Growth

in India: A Sector Level Analysis

CHANDANA CHAKRABORTYMontclair State University, NJ, USA

and

PETER NUNNENKAMP *

Kiel Institute for the World Economy, Kiel, Germany

Summary. — Booming foreign direct investment (FDI) in post-reform India is widely believed topromote economic growth. We assess this proposition by subjecting industry-specific FDI and out-put data to Granger causality tests within a panel cointegration framework. It turns out that thegrowth effects of FDI vary widely across sectors. FDI stocks and output are mutually reinforcingin the manufacturing sector, whereas any causal relationship is absent in the primary sector. Moststrikingly, we find only transitory effects of FDI on output in the services sector. However, FDI inthe services sector appears to have promoted growth in the manufacturing sector through cross-sec-tor spillovers.Published by Elsevier Ltd.

Key words — foreign direct investment, economic reform, growth effects, India, cointegration,causality

1. INTRODUCTION

The stock of foreign direct investment (FDI)in India soared from less than US$ 2 billion in1991, when the country undertook major re-forms to open up the economy to world mar-kets, to about US$ 45 billion in 2005(UNCTAD, online database). Policymakers at-tach high expectations to FDI. According tothe Minister of Finance, P. Chidambaram,‘‘FDI worked wonders in China and can doso in India’’ (Indian Express, November 11,2005). Various economists, including Bajpaiand Sachs (2000, p. 1), advise policymakers inIndia to throw wide open the doors to FDIwhich is supposed to bring ‘‘huge advantageswith little or no downside.’’

Yet, it is far from obvious that FDI in Indiawill have the desired growth effects. Skepticismmay be justified for several reasons. The recentboom notwithstanding, FDI inflows may stillbe too low to make a big difference (Bhat,Sundari, & Raj, 2004; Kamalakanthan &

Laurenceson, 2005). Some observers doubt thateconomic reforms went far enough to changethe character of FDI in India and, thus, resultin types of FDI that may have more favorablegrowth effects (Balasubramanyam & Maham-bare, 2003; Fischer, 2002). Others suspect thatthe type of FDI and its structural compositionmatter at least as much for economic growth ef-fects as does the overall volume of inward FDI(Agrawal & Shahani, 2005; Enderwick, 2005).All the more surprisingly, the structure andthe type of FDI are hardly considered in previ-ous empirical studies on the FDI–growth linksin India.

* We would like to thank Kai Carstensen, Marcel Fra-tzscher, Erich Gundlach, George Mavrotas, Peter Ped-roni and five anonymous referees for critical commentsand most useful advice on how to improve earlier vers-ions of this paper. The usual disclaimer applies. Finalrevision accepted: June 29, 2007.

World Development Vol. 36, No. 7, pp. 1192–1212, 2008Published by Elsevier Ltd

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2007.06.014

1192

Against this backdrop, this paper addressestwo major issues: first, we discuss in Section 2whether India’s reforms in 1991, apart fromgiving rise to FDI, have also induced changesin the structure and type of FDI which maybe relevant for its growth impact. Second, weevaluate in Section 3 whether the growth im-pact of FDI differs between the primary, sec-ondary, and tertiary sectors. We applycointegration and causality analyses on the ba-sis of industry-specific FDI stock data whichare available for the period 1987–2000. We findthat the growth impact of FDI differs signifi-cantly across sectors. Most notably, there is atbest weak evidence for a causal link betweenFDI and output growth in the services sector,which attracted the bulk of additional FDI inrecent years. By contrast, manufacturing out-put appears to have been promoted not onlyby FDI in this sector but also by FDI in the ser-vices sector through spillovers across sectors.

2. THEORETICAL BACKGROUND ANDSTYLIZED FACTS

(a) Major arguments and cross-country findings

FDI is widely regarded as a composite bun-dle of capital inflows, knowledge, and technol-ogy transfers (Balasubramanyam, Salisu, &Sapsford, 1996). Hence, the impact of FDI ongrowth is expected to be manifold (De Mello,1997). Greenfield FDI, in particular, may com-plement local investment and can thus add tothe production capacity of the host country.FDI can promote growth through productivitygains resulting from spillovers to local firms. Asnoted by Borensztein, De Gregorio, and Lee(1998), the rate of growth of a lower-incomecountry depends on the extent to which thiscountry adopts and implements advanced tech-nologies applied in higher-income countries.FDI by multinational corporations based inhigher-income countries is considered a majormechanism through which lower-income coun-tries may access advanced technologies (seealso Findlay, 1978). Likewise, managerialexpertise and knowledge about internationalmarkets may spill over to local companies inlower-income host countries of FDI. Thismay promote growth by relaxing human-capi-tal constraints in the host country and strength-ening the competitiveness of its export sector.Taken together, FDI is supposed to help over-come various bottlenecks which, according to

new growth theory, tend to constrain growthin lower-income countries such as India.

Some of the theoretically expected growthimplications of FDI are difficult to captureempirically. The controversial debate on thereasons underlying India’s recent accelerationin growth clearly reveals the problems involved.According to a skeptical view, of which De-Long (2003) is a prominent proponent, it mayeven be misleading to trace higher growth tothe whole reform program of the early1990s. 1 While DeLong’s reasoning is stronglycontested, for example, by Panagariya(2005), 2 this still leaves the problem of isolat-ing the effects of FDI, the liberalization ofwhich constituted just one, though an impor-tant element of the reform program. Accordingto DeLong (2003, p. 203), it may well be that‘‘deeper changes,’’ notably the general changein official attitudes in India and the widespreadbelief that the rules of the economic game hadbecome more favorable to entrepreneurialactivities, ‘‘had more importance for Indiangrowth than did individual policy moves.’’ Fur-thermore, DeLong clearly has a point in thatreforms in general, and FDI liberalization inparticular could have long-run effects that es-cape econometric investigations.

These arguments imply that assessments ofthe growth impact of capital inflows, includingthe present one on FDI effects in India, maysuffer from two biases working in oppositedirections. On the one hand, the impact of con-crete reforms such as FDI liberalization tendsto be overstated if general attitudes and beliefsare important but cannot be measured. Attri-bution problems of this sort appear to be insur-mountable in econometric analyses relying onmeasurable explanatory variables. The presentanalysis shares this limitation with essentiallyall empirical studies investigating the effects offinancial globalization on economic growth. 3

On the other hand, the impact of FDI (andother types of capital inflows) tends to beunderstated when focusing on relatively short-term effects. It may thus be surprising that mostof the studies surveyed by Kose et al. (2006)consider a time period of up to 5 years to assessthe growth effects of financial globalization.This also applies to prominent FDI studies,including Hermes and Lensink (2003) as wellas Carkovic and Levine (2005). 4 This restric-tion is mainly for two reasons. First, as notedby Rajan and Subramanian (2005, p. 7), empir-ical studies often ‘‘bow to fashion and examine5 year growth horizons’’ in order to have

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1193

enough observations. Data availability alsoconstrains the present analysis (see Section 3(a) for details). Second, any evaluation of a pol-icy intervention faces ‘‘an inescapable tradeoffbetween comprehensiveness and attribution’’(Clemens, Radelet, & Bhavnani, 2004, p. 10).Comprehensiveness would require that FDIstudies such as the present one take a longertime perspective, in order to capture technolog-ical spillovers, managerial learning, and otherdemonstration effects that may take consider-able time to materialize. However, a longer per-iod of observation can greatly increase noiseand impede attribution of growth effects to cau-sal events in the distant past.

Conceptual problems and limitations not-withstanding, surveys of the cross-country evi-dence claim that empirical findings are largelyin line with theoretical expectations in thatFDI promotes growth (Lim, 2001; Lipsey,2002; OECD, 2002). Recent studies finding aclearly positive nexus between FDI and growthacross host countries include Khawar (2005), aswell as Blonigen and Wang (2004) for develop-ing countries. Several studies qualify this opti-mistic view by identifying certain thresholds(e.g., in terms of human-capital endowmentor financial market development) that hostcountries would have to reach before they canreap favorable growth effects of FDI (e.g., Al-faro, Chanda, Kalemli-Ozcan, & Sayek, 2004;Borensztein et al., 1998). Moreover, the direc-tion of causality underlying the positive FDI–growth nexus is still debated (Carkovic &Levine, 2005). Chowdhury and Mavrotas(2006) corroborate the earlier finding of Nair-Reichert and Weinhold (2001) that the causalrelationship between FDI and growth is char-acterized by a considerable degree of heteroge-neity. This is why these authors call for hostcountry-specific studies.

(b) Economic reforms and the type of FDI inIndia

The case of India is of particular interest inthis context. While India is a latecomer indrawing heavily on FDI to foster growth, ithas attracted booming FDI since the economicreform program of 1991. Earlier studies on In-dia typically fail to find significantly positivegrowth effects (e.g., Agrawal, 2005; Pradhan,2002). 5 Analyses accounting for the fact thatcausation may run both ways tend to show thathigher growth leads to more FDI, rather thanvice versa (Dua & Rashid, 1998; Chakraborty

& Basu, 2002; Sahoo & Mathiyazhagan,2002). 6 Kumar and Pradhan (2002) considerthe FDI–growth relationship to be Grangerneutral in the case of India. Bhat et al. (2004)provide no evidence of causality in either direc-tion. However, even studies accounting fortwo-way causation have two major shortcom-ings in common:

• Earlier analyses capture the effects of thechanged policy environment for FDI inIndia and the possible implications for theFDI–growth nexus at best partially.• The empirical estimates are typicallybased on aggregate FDI data, even thoughthe growth effects of FDI are likely todepend on the sector in which FDI takesplace.

The ADB (2004, p. 244) expects a fundamen-tal shift in the behavior of foreign investors andin the benefits that host countries may derivefrom FDI when the policy environmentchanges as it did after India’s reform programof 1991. The New Industrial Policy markedthe departure from restrictive FDI regulationsand included the liberalization of trade barri-ers. 7 It is beyond serious doubt that India’s re-form program of 1991 has boosted FDIinflows. Annual average inflows of US$ 200million in 1987–90 pale against annual averageinflows of US$ 4.1 billion in 2001–04 (UNC-TAD, online database). Inward FDI stocks,relative to GDP, soared from less than 1% inthe late 1980s and early 1990s to almost 6%in 2004.

At the same time, the type of FDI appears tohave changed. Foreign investors increasinglyentered into technical collaboration agreements(Athreye & Kapur, 2001). As shown in Table 1,survey data compiled by the Reserve Bank ofIndia (varous issues) on the so-called FDI com-panies point to higher technological sophistica-tion of FDI, more local R&D undertaken byFDI companies, and a stronger world-marketorientation of FDI in post-reform India. 8

These changes in the type of FDI may havestrengthened the growth effects of FDI. Argu-ably, India no longer belongs to the group ofrelatively closed host countries for which,according to Basu, Chakraborty, and Reagle(2003), long-run causality is uni-directionalfrom GDP to FDI (see also Gupta, 2005).Balasubramanyam et al. (1996) refer to thehypothesis advanced by Bhagwati (1978)according to which the growth effects of FDIare stronger in host countries pursuing an out-ward oriented trade policy. A more open trade

1194 WORLD DEVELOPMENT

Table 1. FDI Characteristics, 1990–91 and 2002–03

Export, %of prod.

Ratioexports to

imports

Imports ofcapital goods,

% of totalimports

Raw materials,stores and spares,imported in % of

indigenous

Royaltypayments, %

of prod.

R&D, %of prod.

Salaries,% of prod.

Memorandum

Companiesnumber

Value ofproduction, all

industries = 100

1990–91All industries 9.3 1.3 9.0 20.0 0.11 0.09 9.0 300 100.0Tea plantations 13.7 95.7 18.4 0.5 0.00 0.00 17.0 24 6.3Textiles 16.4 3.5 19.5 18.7 0.00 0.04 14.4 6 2.0Rubber products 11.2 1.7 7.2 12.8 0.01 0.00 7.9 4 3.5Chemicals 9.5 1.2 2.9 23.3 0.02 0.06 2.0 63 29.3Engineering 7.0 0.8 12.3 26.6 0.24 0.14 9.5 126 38.7Trade 16.3 2.1 61.6 0.3 0.00 0.05 7.4 8 0.7

2002–03All industries 14.8 1.3 7.7 20.6 0.26 0.38 8.3 490 100.0Tea plantations 22.4 49.3 9.8 1.5 0.00 0.05 37.2 10 1.0Food products 8.9 2.9 5.1 4.6 0.01 0.09 5.6 16 3.3Rubber/plastic products 16.4 1.9 16.2 18.8 0.00 0.21 5.0 11 2.0Chemicals 11.8 0.9 3.4 23.6 0.28 0.39 5.7 76 28.2Engineeringa 11.1 0.9 9.2 22.7 0.49 0.65 8.7 153 26.3

machinery and tools 13.5 1.0 3.4 23.8 0.27 0.68 9.5 85 8.5electr. mach. 11.4 0.8 6.7 30.4 0.25 0.47 7.5 33 5.9transport equipment 9.2 1.0 16.9 18.6 0.76 0.72 8.8 35 11.9

Computer and related act. 12.7 5.0 74.8 0.0 0.05 0.77 31.8 23 4.4Trade 19.9 1.4 1.0 0.5 0.01 1.80 9.3 20 1.2

Source: Reserve Bank of India (various issues) a Sum of machinery and tools, electrical machinery and transport equipment.

EC

ON

OM

ICR

EF

OR

MS

,F

DI,

AN

DE

CO

NO

MIC

GR

OW

TH

ININ

DIA

1195

regime is supposed to be conducive to knowl-edge and technology spillovers, that is,growth-promoting FDI features identified bynew growth theory.

(c) Sector-specific hypotheses

While some of the empirical studies referredto above do provide first indications that theimpact of FDI in India has become more favor-able in the post-reform period, they fail to ac-count for sector-specific effects. 9 It is for tworeasons that, in contrast to previous studies,we use disaggregated FDI data in the following:(i) the sectoral composition of FDI in India haschanged dramatically and (ii) the growth effectscan be expected to differ significantly acrosssectors.

Data on inward FDI stocks for specific sec-tors and industries reveal a tremendous shiftfrom FDI in the primary and the manufactur-ing sectors to FDI in services since the mid-1990s (Figure 1). In the manufacturing sector,all previous priority areas, notably the chemicalindustry and (electrical and nonelectrical)machinery, accounted for steeply decreasingshares in overall FDI stocks. 10 The data situa-tion leaves much to be desired when it comes toFDI in services. This is mainly because boom-ing FDI stocks in the services sector are largelyconfined to the unspecified category of ‘‘otherservices.’’ 11 Presumably, FDI in this categoryis heavily concentrated in information and

communication services (Kumar, 2003; ReserveBank of India, 2005).

The changing composition of FDI in Indiamatters as various arguments suggest that thegrowth effects of FDI should be sector-specific.In particular, the potential for productivityenhancing spillovers is widely believed to differacross sectors. According to Alfaro (2003),FDI-related transfers of technology and knowl-edge primarily occur in the manufacturing sec-tor. Arguably, it is mainly in manufacturingthat foreign investors use intermediate inputsintensively which creates positive externalitiesand allows local producers to draw on a largervariety of inputs and, thereby, increase theirproductivity (Rodriguez-Clare, 1996). Like-wise, UNCTAD (2001, p. 138) argues that themanufacturing sector comprises a broad rangeof linkage-intensive activities. Aykut and Sayek(2007) suspect that technology and knowledgespillovers in manufacturing are most likely ifFDI is motivated by efficiency-seeking reasons,unless FDI is located in enclaves such as ex-port-processing zones.

By contrast, the potential for linkages is typ-ically considered limited in the primary sector(UNCTAD, 2001, p. 138). Resource-seekingFDI in this sector often takes place in economicenclaves that are largely isolated from the localeconomy. Positive growth effects of FDI in theprimary sector may be compromised in otherways, too. FDI in this sector tends to be vola-tile; it is sensitive to international commodity

0

10

20

30

40

50

60

70

80

90

100

1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000Primary Sector Secondary Sector Tertiary Sector

Percent

Figure 1. Sector-wise composition of FDI Stocks, 1987–2000 Source: UNCTAD (2000) and Central StatisticalOrganisation (various issues).

1196 WORLD DEVELOPMENT

prices and often financed through inter-com-pany loans rather than equity (Aykut & Sayek,2007; World Bank, 2005). According to Len-sink and Morrissey (2006), the volatility ofFDI has a negative impact on growth. Further-more, large-scale FDI for resource-seeking rea-sons may give rise to Dutch disease effects andencourage unproductive activities such as rentseeking (Aykut & Sayek, 2007). While all thisrenders positive growth effects rather unlikely,the typically high export orientation of FDIin the primary sector may counterbalance neg-ative factors.

Compared to the primary sector and themanufacturing sector, the growth effects ofFDI in the services sector appear to be moreambiguous a priori. Alfaro (2003) and UNC-TAD (2001) suspect that the services sectorresembles the primary sector with regard tothe limited potential of linkages and spillovers(see also World Bank, 2005, p. 96). The increas-ing tradability of services notwithstanding, thebulk of FDI in this sector still appears to bemarket-seeking. The superior market power offoreign service providers, whose entry into thehost country is often through mergers andacquisitions rather than greenfield FDI, has‘‘significant crowding-out potential’’ (Aykut &Sayek, 2007). 12 Moreover, linkages with thelocal economy may remain weak even forFDI in tradable services; Kumar (2003, p. 27)reckons that foreign companies in India’s soft-ware industry operate as ‘‘export enclaves.’’This suggests that technological spilloversplayed a minor role as a transmission mecha-nism through which FDI may have promotedthe development of IT services in post-reformIndia. This may have limited the output effectsof FDI in the services sector.

Given that the growth effects of FDI arelikely to be sector-specific it is fairly surprisingthat almost all empirical studies use aggregatedata and ignore the composition of FDI.Alfaro (2003) and Aykut and Sayek (2007)provide major exceptions. Alfaro appliescross-country panel data on sector-specificFDI flows and controls for macroeconomicand institutional factors. It turns out thatFDI has significantly positive growth effects inthe manufacturing sector only. Aykut and Say-ek consider the composition of FDI inflows, to-gether with aggregate inflows, and find positivegrowth effects only when FDI in manufacturingfigures prominently in the composition. Bothstudies provide a differentiated picture onFDI effects in the context of a heterogeneous

group of host countries. By contrast, we followChowdhury and Mavrotas (2006) advice to per-form host country-specific studies. Further-more, in contrast to Alfaro (2003) and Aykutand Sayek (2007), we focus on questions relatedto the cointegration process and the causallinks in the FDI–growth relationship.

Using disaggregated FDI data for a singlehost country such as India has two advantages:It allows us to test for sector-specific effects ofFDI, and it avoids biased results due to inap-propriate pooling of heterogeneous host coun-tries (Blonigen & Wang, 2004). This is not toignore that sector-specific analyses, too, comeat a cost. Most importantly, estimated FDI ef-fects tend to be biased downwards if FDI inone specific sector creates spillovers from whichother sectors, too, may derive benefits. For in-stance, India’s manufacturing industries couldhave benefited from more efficient servicesbrought about by FDI in the services sector. 13

Several reviews of the literature, however, re-veal that empirical evidence on the importanceof FDI-induced spillovers is highly ambiguous.Gorg and Greenaway (2004) conclude that ro-bust empirical support for positive spilloversis hard to find. Almost all studies focus on in-tra-industry effects (see also Blomstrom & Kok-ko, 1998; Gorg & Strobl, 2001). This is becausethe major transmission channels (demonstra-tion and imitation; human-capital externalities;and competition effects) are supposed to oper-ate within the same industry. 14 For instance,Jenkins (1990, p. 213) argues: ‘‘Over time,where foreign and local firms are in competi-tion with each other, producing similar products,on the same scale and for the same market,there is a tendency for local firms to adopt sim-ilar production techniques to those of theMNCs’’ (emphasis added). Likewise, FDI-re-lated competition effects are most likely to oc-cur within the same industry.

If inter-industry spillovers are addressed atall, the analyses are typically confined to effectswithin the manufacturing sector of host coun-tries. Hence, the verdict of Lipsey (2002, p.42) that the inter-industry effects of FDI ‘‘havereceived a great deal of speculation but littlestatistical testing’’ probably is even moreappropriate when it comes to the question ofspillovers across sectors, for example, fromFDI in services to productivity in manufactur-ing. Obviously, this skeptical assessment doesnot preclude that spillovers across sectors haveplayed a role in specific host countries such asIndia. Therefore, we make an attempt in

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1197

Section 3 to capture such linkages at least ten-tatively, by performing pairwise tests of Gran-ger causality between FDI and output in theservices and manufacturing sectors.

3. COINTEGRATION AND CAUSALITY

(a) Approach

While a growing literature has recognized thetheoretical possibility of two-way feedbacks be-tween FDI and economic growth along withtheir long-run and short-run dynamics, empiri-cal investigations in the context of the Indianeconomy have failed to provide conclusive evi-dence in support of such two-way feedback ef-fects at the sector level. Moreover, earlierstudies are typically devoid of a test of the coin-tegrated relationship between the two variablesof interest. Given the unit root characteristicsof time series variables in general, results basedon panel regression analysis are subject to spu-rious correlation. Therefore, a better under-standing of the FDI–growth relationship inthe context of policy reform and changes inthe structure of FDI requires complementaryanalyses that rigorously explore the issue ofcointegration as well as the long-run andshort-run dimensions of the causal relationshipbetween FDI and growth. 15

To assess the causal links between the re-ferred variables, we estimate a vector errorcorrection model that emanates from the coin-tegrated relationship between the variables. 16

We apply a panel cointegration framework thatallows for heterogeneity across 15 industries inthe primary, secondary and tertiary sectors (seeAppendix A for the sample of industries). Twoquestions are of particular importance: (1) Isthere a long-run steady state relationship be-tween FDI and output for all of the 15 indus-tries included in our panel? (2) Given theexistence of a cointegrated relationship, canwe accurately identify the chronology of causaleffects between FDI and output by unravelingthe short-run dynamics of the long-run rela-tionship?

Our empirical investigation regarding theassociation between FDI stocks and economicgrowth follows the three step procedure sug-gested by Basu et al. (2003). We begin by test-ing for nonstationarity in the two variables ofFDI stocks and output in our panel of 15industries. Prompted by the existence of unitroots in the time series, we use the panel cointe-

gration technique developed by Pedroni (2004,1999) to test for a long-run cointegrated rela-tionship between the two variables in the sec-ond step of our estimation. Given theevidence of cointegration in the long-runFDI–growth relationship across the panel, weuse an error correction model to uncover Gran-ger causality in the relationship in the final stepof our estimation.

While our approach of examining causalitywithin a panel cointegration framework hasmajor advantages compared to the existing lit-erature on the FDI–growth link in India, thereare some limitations. A consistent series ofindustry-specific FDI stocks is only availablefor the period 1987–2000. As noted in Section2, the relatively short period of observation ren-ders it all but impossible to fully capture thelong-run effects of FDI. In other words, datarestrictions leave us with no choice but to optfor attribution, rather than comprehensivenessin dealing with the ‘‘inescapable tradeoff’’(Clemens et al., 2004, p. 10) between the two.Moreover, the choice of the 15 industries, eachof which covers a broad range of goods or ser-vices, is driven by data availability. A simplepanel regression with the variables defined inlevels reveals a positive association betweenFDI stocks and output; the correlation coeffi-cient is 0.89 (see Appendices B and C for thedefinition of variables, sources and summarystatistics).

Our analysis is restricted to the bivariate rela-tionship between FDI and growth. This limita-tion is fairly common in the relevant literature.The bivariate approach has been used in severalrecent studies on the causal links between FDIand growth, including Chowdhury and Mavro-tas (2006), Frimpong and Oteng-Abayie (2006),and Hansen and Rand (2006). The same appliesto related fields such as the causal links betweenexports and growth as well as those between lo-cal financial development and growth. 17 Thepreference for bivariate approaches in the rele-vant literature is to avoid the complicationsresulting from indirect causality once the so-called auxiliary variables are accounted for ina multivariate framework (Dufour & Renault,1998). For example, Konya (2004, p. 79) con-siders it ‘‘a clear advantage’’ that ‘‘in a bivariatesystem no-causality for one period ahead im-plies no-causality at, or up to, any horizon.’’Moreover, the usable sample size tends toshrink considerably when testing for causalityin a multivariate system (Konya, 2004, p. 88).Hence, we follow the standard bivariate

1198 WORLD DEVELOPMENT

approach. The important contribution to theexisting literature is rather that we accountfor the heterogeneity of the FDI–growth linkacross sectors by applying a panel cointegrationframework. In this way, we aim at identifyingthe precise direction of causality between thesetwo variables, rather than identifying the rela-tive importance of various possible determi-nants of growth.

For the specific case at hand, it is for severalreasons that the bivariate analysis offers a rea-sonable strategy, even though we omit otherpotential growth determinants. The omittedvariable bias could pose a serious problem inthe present context only if we found large andconsistently positive effects running from FDIto growth. As shown below, however, our find-ings are ambiguous and sector-specific. Nota-bly in the services sector, the weak results forFDI as a growth determinant suggest that theomission of other determinants such as humanand physical capital does not cause major dis-tortions. In other words, FDI does not appearto capture the effects of omitted variables.

More specifically, ignoring total (foreign plusdomestic) investment as a controlling variableshould not be of major concern. This is eventhough FDI stocks in India of about 45 billionUS$ in 2005 (UNCTAD, online data) paleagainst total capital stocks of almost 1700 bil-lion US$ (Reserve Bank of India, online). Con-trolling for total investment is meant to addressthe question whether FDI is more efficient thandomestic investment in promoting growth(Borensztein et al., 1998). By contrast, we arenot particularly interested in isolating this effectfrom the capital-augmenting effect of FDI. Ourmore modest ambition is to capture the totalimpact that FDI may have on growth, what-ever the specific transmission mechanisms maybe. In that sense, our approach is similar tothe basic empirical model applied in the seminalregression analysis of Borensztein et al. (1998),which does not include total investment as acontrol.

All this is not to ignore that a multivariateanalysis would still be desirable in order to as-sess in which way FDI may cause highergrowth. To the best of our knowledge, how-ever, industry-specific data on domestic capitalstocks or investment that would be required forsuch an analysis are not available for the caseof India. This lack of data also implies thatthe FDI variable cannot be defined as a sharein total capital stocks. Alternatively, FDI maybe expressed relative to (industry-specific) out-

put. In regression analyses on the FDI–growthlink, FDI is typically normalized by employingits share in GDP. The reason is that this sharevariable is stationary. In the present panel coin-tegration framework, however, the variables weuse should have unit roots. 18 Hence, unit rootsand cointegration tests help us decide on theappropriate definition of the FDI variable. Itturns out that the ratio of FDI to GDP is gen-erally stationary and no consistent cointegrat-ing relationship exists, which is in contrast tononnormalized FDI (see below). 19 This resem-bles Canning and Pedroni (2004) who foundthat it is the log level of infrastructure, ratherthan its share in GDP, that is cointegrated withGDP. 20

A final methodological question concerns theuse of Granger causality tests within a cointe-grated framework, as opposed to the proceduresuggested by Toda and Yamamoto (1995).Toda and Yamamoto’s test has been appliedin several recent contributions to the literature(e.g., Chowdhury & Mavrotas, 2006). In a timeseries framework, this test is sometimes pre-ferred over standard Granger causality testsas it does not rely so heavily on pre-testing. Ifpre-testing with respect to unit roots and coin-tegration has ambiguous results, this may ren-der the final conclusions concerning Grangercausality less reliable (Konya, 2004, p. 82).However, this potential drawback is hardly rel-evant in the present context. We refer to differ-ent methods and present various test statisticswhen pre-testing for unit roots and cointegra-tion, and results turn out to be consistent withvery few exceptions. Further, a cointegratedframework serves as a precondition for testinglong-run causality. Hence, it is appropriate torely on standard Granger causality in the pres-ent context of using disaggregated FDI andoutput data in order to differentiate betweenlong-run and short-run causality in a panelcointegration framework.

(b) Empirical findings

(i) Test of unit rootsThe panel data framework for unit root test

has gained attractiveness in the empirical litera-ture because of its weak restrictions. It capturesthe member-specific effects and allows for het-erogeneity in the direction and magnitude ofthe parameters across the selected panel. Inaddition, it allows for a great degree of flexibil-ity in terms of model selection. The alternativesfor model choice range from a model with

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1199

heterogeneous intercepts and heterogeneoustrends to a model with no intercepts and notrends. Within each model, it is possible to testfor common time effects.

Following the methodology used in earlier re-search, we test both mean stationarity andtrend stationarity in the two variables of outputand FDI stocks. We also control for time effectscommon to all industries (t = 1987–2000) with-in each model. Consequently, the models ofinterest are: model with heterogeneous inter-cepts and no common time effect (M1); modelwith heterogeneous intercepts and commontime effect (M2); model with heterogeneousintercepts and heterogeneous trends ignoringcommon time effects (M3); and model with het-erogeneous intercepts and heterogeneous trendsallowing for common time effects (M4). We testfor the null of nonstationarity in the two re-ferred variables against the alternative of sta-tionarity by taking each of the models in turn.The test is a residual-based test that evaluatesfour different statistics for variables at their lev-els and at first differences. These four statisticsrepresent a combination of the tests used byIm, Pesaran, and Shin (2003) and Levin, Lin,

Chu, and Shang (2002). 21 While the first twotest statistics are nonparametric rho-statistics,the last two are parametric ADF t-statistics.Sets of these four statistics for each of the fourmodels are reported in Table 2.

The first two rows under each model reportthe panel unit root statistics for output andFDI stocks at levels. Given that the left tail ofthe normal distribution is used to reject the nullof nonstationarity, the positive values and thesmall negative values reported in these rowsconsistently fail to reject the null across differ-ent models. 22 The last two rows under eachmodel report the panel unit root statistics forfirst differences in output and FDI stocks. Thelarge negative values for the statistics indicaterejection of the null of nonstationarity at the1% level for all models. We may, therefore,conclude that output and FDI stocks have unitroot properties, or are integrated of order one,that is, I (1) variables for short.

(ii) Test for panel cointegrationWith confirmation on the integrated order of

the two variables of interest, the question isthat they might or might not have a common

Table 2. Full panel unit root test for GDP and FDI stocksa

H0: Variables are non-stationary

Variables Levin-Lin rho-stat Levin-Lin t-rho-stat Levin-Lin ADF-stat IPS ADF-stat Decision on H0

M1: Heterogeneous intercepts with no common time effectGDP 2.30209 3.48977 3.39246 3.66133 AcceptFDI !0.11060 1.58909 0.64754 !0.27242 AcceptGDPDIFF !14.51160 !17.36676 !21.40637 !27.51017 RejectFDIDIFF !15.50983 !14.59209 !9.36342 !13.23267 Reject

M2: Heterogeneous intercepts with common time effectGDP 1.92248 3.32841 2.94893 2.94396 AcceptFDI 1.85163 3.57200 1.52558 !1.03011 AcceptGDPDIFF !12.58931 !12.17797 !10.52295 !20.31529 RejectFDIDIFF !10.04381 !6.87181 !6.37505 !8.223068 Reject

M3: Heterogeneous intercepts and heterogeneous trends with no common time effectGDP 1.93409 0.22568 !0.29699 !0.53593 AcceptFDI !1.70406 !1.77677 !1.86162 !3.71896b AcceptGDPDIFF !17.19989 !10.18715 !14.02306 !20.43317 RejectFDIDIFF !14.86436 !9.98292 !6.64733 !12.14174 Reject

M4: Heterogeneous intercepts and heterogeneous trends with common time effectGDP !0.50786 !0.71247 !1.07342 !1.29829 AcceptFDI 2.66658 0.70148 !1.64521 !4.32851b AcceptGDPDIFF !14.62652 !8.80268 !8.54622 !15.98889 RejectFDIDIFF !12.77869 !7.33768 !5.46092 !8.41581 Reject

Source: own calculations based on RBI online database; UNCTAD (2000); CSO (various issues).a All tests are left-tail tests that follow normal distribution.b Exceptions to all other statistics.

1200 WORLD DEVELOPMENT

stochastic trend, or, they might or might not becointegrated. We resolve this question by look-ing for a long-run relationship between outputand FDI stocks using the panel cointegrationtechnique. This technique is a significantimprovement over the conventional cointegra-tion tests applied on a single series. As ex-plained in Pedroni (1999), conventionalcointegration tests usually ‘‘suffer from unac-ceptably low power’’ when applied on the dataseries of restricted length. Panel cointegrationtechnique addresses this issue by allowing topool information regarding common long-runrelationships between a set of variables fromindividual members of a panel. Further, withno requirement for exogeneity of the regressors,it allows the short-run dynamics, the fixed ef-fects, and the cointegrating vectors of thelong-run relationship to vary across the mem-bers of the panel.

The specific cointegration relationship weestimate has the following form:

FDIit ¼ ai þ dt þ biGDP it þ eit ð1Þwhere ai (i = 1, 2, . . . , 15) refers to industry-specific effects, dt refers to time effects, and eit

is the estimated residual indicating deviationsfrom the long-run steady state relationship.With a null of no cointegration, the panel coin-tegration test is essentially a test of unit roots inthe estimated residuals of the panel. If eit inEqn. (1) is found to be stationary, or consistentwith I (0), one may claim that cointegration ex-ists between FDI stocks and output. Pedroni(1999) refers to seven different statistics for test-ing unit roots in the residuals of the postulated

long-run relationship. Of these seven statistics,the first four are referred to as panel cointegra-tion statistics; the last three are known as groupmean panel cointegration statistics. In the pres-ence of a cointegrating relation, the residualsare expected to be stationary. A positive valuefor the first statistic and large negative valuesfor the remaining six statistics allows the rejec-tion of the null of no cointegration. All of theseven statistics under different model specifica-tions are reported in Table 3. Most of the statis-tics for all different model specifications suggestrejection of the null at the 1% level. We, there-fore, conclude that the two unit root variablesof output and FDI stocks are cointegrated inthe long run. Put differently, FDI and economicgrowth in India are positively associated witheach other.

(iii) Test of causality: all industriesWith the affirmation that output and FDI

stocks are cointegrated, we test for Grangercausality in the long-run relationship using anerror correction model. As proposed by Engleand Granger (1987), and demonstrated byGranger et al. (2000), the causality test itselfis a two-stage estimation process. The first steprelates to the estimation of the residual fromthe cointegrated relationship shown in Eqn.(1). Incorporating the residual eit as a righthand side variable, the dynamic error correc-tion model is estimated at the second step fordrawing inferences on Granger causality. Fol-lowing these steps, the dynamic error correc-tion model of our interest has the followingform:

Table 3. Results for panel cointegration between GDP and FDIa

H0: No cointegration vector between GDP and FDI

Statistics Model specification

M1 M2 M3 M4

Panel v-stat 2.49707 2.94133 !0.23055 !0.68771Panel rho-stat !5.64840 !5.19672 !3.67648 !2.53801Panel pp-stat !12.79293 !10.23135 !13.03658 !9.22998Panel ADF-stat !10.91080 !8.65209 !11.75143 !8.50382Group rho-stat !3.46427 !3.21411 !1.67613b !0.79810b

Group pp-stat !17.74692 !12.30255 !13.73666 !9.10667Group ADF-stat !18.89325 !11.04659 !13.63381 !9.23078Decision Reject H0 Reject H0 Reject H0 Reject H0

Source: own calculations based on RBI online database; UNCTAD (2000); CSO (various issues).a The first test is a right-tail test; all other tests are left-tail tests.b Exceptions to all other statistics in the row.

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1201

DFDIit ¼ a1i þ g1ieit!1 þX

k

b1ikDFDIi;t!k

þX

k

b2ikDGDPi;t!k þ u1it;

DGDPit ¼ a2i þ g2ieit!1 þ Rkc1ikDGDPi;t!k

þX

k

c2ikDFDIi;t!k þ u2it

ð2Þin which k refers to the optimal lag length foreach industry in the panel. The decision forthe optimal lag length for this model rests onthe comparison of regression results with alter-native lag structures. 23 Allowing for up tothree period lags, we find no noticeable changesin the significance of the estimates. Conse-quently, with the intent of having a longer timeperspective for the analysis, we keep the laglength limited to two periods.

The two coefficients g1i and g2i representspeeds of adjustment along the long-run equi-librium path; while g1i can be interpreted as dis-playing the long-run effects of output on FDIstocks, g2i can be taken to imply the long-runeffects of FDI stocks on output. 24 FollowingEngle and Granger (1987), for the ith industryin the panel, the existence of cointegration be-tween the referred variables indicates causallinks among the set of variables as manifestedby Æjg1ij+Æjg2ij>0. Accordingly, failing to rejectH0: g1i = 0 for all i, i = 1, 2, . . ., 15, implies thatoutput does not Granger cause FDI stocks forany of the industries included in the panel forthe long run. Conversely, failing to reject H0:g2i = 0 for all i, i = 1, 2, . . ., 15, implies thatFDI stocks do not Granger cause output inany of the industries in the panel in the longrun.

The set of coefficients b2ik and c2ik capture in-terim effects and reflect the adjustment processbetween the associated set of variables in re-sponse to a random shock. Consequently, fail-ing to reject H0: b2ik = 0 for all i and k,

(i = 1, 2, . . ., 15, k = 1, 2, . . ., k), implies thatoutput does not Granger cause FDI stocksfor any of the industries included in the panelin the short run; and failing to reject H0:c2ik = 0 for all i and k, (i = 1, 2, . . ., 15,k = 1, 2, . . ., k), implies that FDI stocks donot Granger cause output for any of the indus-tries included in the panel in the short run. Fol-lowing conventional procedure, we use astandard F-test to test the referred sets oflong-run and short-run hypotheses. The resultsof these tests are shown in Table 4.

As is apparent from the table, the null of noshort-run causality and no long-run causality isrejected for both of the linear causal links testedwithin the cointegrated model. For the shortrun, both the hypotheses of no causality are re-jected at the 1% level indicating strong bi-direc-tional links between FDI stocks and output.For the long run, the hypothesis of no causalityfrom output to FDI stocks is rejected at the 1%level; the hypothesis of no causality from FDIstocks to output is rejected at the 5% level.Thus, though there is evidence of bi-directionalcausal links, causality running from FDI stocksto output is relatively weaker when consideringthe entire panel of 15 industries. 25

(iv) Test of causality: sector-wise disaggregationTo explore the possibility that the direction

and magnitude of causal links between the twovariables might vary between individual mem-bers of the industry panel, we repeat the Gran-ger causality tests for each of the three broadsectors. And indeed, the results reported in Ta-ble 5 reveal that the nature of the causal linksbetween FDI stocks and output are strikinglydifferent across sectors. For the primary sector,the null of no causality from output to FDIstocks and that of no causality from FDI stocksto output cannot be rejected for either the shortrun, or the long run. By contrast, the manufac-turing sector displays robust bi-directionalcausal links in the long-run relationship

Table 4. Results of full panel causality testsa

Null hypothesis Long-run Short-run

H0: Output does not Granger cause FDI 11.7506* 4.2864*

H0: FDI does not Granger cause output 2.1569** 2.7564*

Critical F-value 2.19 1.95

Source: own calculations based on RBI online database; UNCTAD (2000); CSO (various issues).a Critical F-values correspond to 1% level of significance.* Significant at 1% level.** Significant at 5% level.

1202 WORLD DEVELOPMENT

between the two variables of FDI stocks andoutput; in the short run, causality for the man-ufacturing sector is seen to be uni-directionaland running from FDI stocks to output. Mostinterestingly, there is no strong evidence oflong-run causal links between the two variablesof interest in the tertiary sector, even though thebulk of additional FDI flowing to post-reformIndia was attracted by the tertiary sector. Thereis only a weak long-run causality running fromoutput to FDI. The results for the short run,however, reflect feedback effects between thetwo variables in the tertiary sector.

The sector-specific causality tests for the caseof India are largely in line with the cross-coun-try findings of Alfaro (2003) and Aykut andSayek (2007). Moreover, it is not only in termsof the statistical significance of Granger causal-ity tests that the manufacturing sector seems tohave benefited most from FDI in post-reformIndia. In Appendix D, we report complemen-tary calculations of elasticity coefficients basedon logarithmic transformation of FDI and out-put at the level of specific industries. The elas-ticities suggest that the economic impact ofFDI on output growth is particularly high insome manufacturing industries, notably inchemicals and metals. Except for food, bever-ages and tobacco, all manufacturing industriesdisplay higher elasticities of output with respectto FDI than industries belonging to the pri-mary and tertiary sectors.

The marked differences in the short and long-run dynamics of the FDI–growth relationshipbetween major sectors of the Indian economy

may be attributed to specific characteristics ofFDI in these sectors and their capacity to ab-sorb foreign technologies and make use of spill-overs. Our findings support the reasoning inSection 2 that the scope for linkages betweenforeign and domestic firms is typically limitedin the primary sector. The enclave characterof FDI projects in the primary sector as wellas the volatility of resource-seeking FDI 26 ap-pears to have constrained growth effects in thissector. It is also in line with expectations that itwas mainly the manufacturing sector that ben-efited from trade liberalization, financial liber-alization and human-capital formation inpost-reform India, and the complementary pro-cess of technological diffusion. Several manu-facturing industries have become more closelyintegrated into world markets in terms of ex-ports and imports as well as in terms of tech-nology transfers (Table 1). Even if FDI inIndia’s manufacturing sector has remainedmarket-seeking (or: of the horizontal type) inthe first place, FDI-induced competition mayhave strengthened productivity enhancing spill-overs within the manufacturing sector.

Granger causality tests as well as the elastic-ity estimates, reflecting the economic impactof FDI on output growth, tend to support theproposition of Alfaro (2003) and UNCTAD(2001) that the tertiary sector resembles the pri-mary sector with regard to the limited potentialof linkages and spillovers. More specifically,UNCTAD (2004, pp. 169–170) argues thatFDI has not been a dominant feature in India’ssoftware development, while the benefits from

Table 5. Results of sector level causality testsa

Hypothesis and critical F-value Long-run Short-run

Agriculture and miningH0: Output does not cause FDI 0.2321 1.6847H0: FDI does not cause output 4.3275 0.9746Critical F-value 5.42 4.01

ManufacturingH0: Output does not cause FDI 3.5182* 0.9990H0: FDI does not cause output 4.0070* 3.1099*

Critical F-value 3.21 2.59

ServicesH0: Output does not cause FDI 2.4072*** 4.6138*

H0: FDI does not cause output 0.7077 3.6208*

Critical F-value 3.85 2.98

Source: own calculations based on RBI online database; UNCTAD (2000); CSO (various issues).a Critical F-values are reported at 1% level.* Significant at 1% level.*** Significant at 10% level.

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1203

offshoring of IT-enabled activities such as back-office services have been concentrated in thetechnology parks of a few metropolitan centers.Moreover, over much of our period of observa-tion (1987–2000), the offshoring of service-re-lated activities by foreign investors to Indiawas still at the lower end of the value chain(UNCTAD, 2004, p. 172). The recent upgrad-ing to higher value-added activities, whichmay trigger more pronounced spillovers andoutput effects, tends to escape our analysis.This would imply that we underestimate theoutput effects of FDI in the services sector.

(v) Test of cross-sector spilloversThe results reported so far may also underes-

timate FDI effects by not taking into accountthat booming FDI in the services sector mayhave an impact on other sectors of the Indianeconomy. As noted in Section 2, FDI in thissector could promote output growth in othersectors, notably in manufacturing, if it in-creased the efficiency of business services usedthroughout the economy. We explore the possi-bility of spillovers across sectors at least tenta-tively by performing additional Grangercausality tests. These tests are based on the fol-lowing two cross-sector pairs of aggregateddata series: (1) service sector FDI and manufac-turing output; and (2) manufacturing FDI andservice sector output. The results reported inTable 6 reveal that the null of no causality be-tween the alternative pairs of variables cannotbe rejected for the short run. In other words,we do not find evidence of any cross-sector cau-sality in the short run. For the long run, how-ever, the first group of paired variables showsevidence of causality running from service sec-tor FDI to output in the manufacturing sector.

This suggests that output growth in the manu-facturing sector has not only been promoted byFDI in this sector but also by FDI in the ser-vices sector through cross-sector spillovers. Atthe same time, the results for the second groupof paired variables point to long run causalityrunning from service sector output to manufac-turing FDI. This indicates that service sectordevelopment in India has not only stimulatedFDI in this sector but also FDI in the manufac-turing sector.

4. SUMMARY AND CONCLUSIONS

Inward FDI has boomed in post-reform In-dia. At the same time, the composition and typeof FDI has changed considerably. Even thoughmanufacturing industries, too, have attractedrising FDI, the services sector accounted for asteeply rising share of FDI stocks in India sincethe mid-1990s. While FDI in India continues tobe local-market-seeking in the first place, itsworld-market orientation has increased in theaftermath of economic reforms. It is againstthis backdrop that we assess the growth impli-cations of FDI in India. By using industry-spe-cific FDI and output data and applying a panelcointegration framework that integrates long-run and short-run dynamics of the FDI–growthrelationship, we address important gaps in theearlier literature.

For the Indian economy as a whole, we findthat FDI stocks and output are cointegratedin the long run. At the aggregate level, Grangercausality tests point to feedback effects betweenFDI and output both in the short and the longrun. However, the impact of output growth inattracting FDI is relatively stronger than that

Table 6. Results of cross-sector causality testsa

Hypothesis and critical F-value Long-run Short-run

Manufacturing GDP paired with service FDIH0: Manufacturing output does not cause service FDI 4.2348 1.2672H0: Service FDI does not cause manufacturing output 5.3341** 1.7499Critical F value 11.259 10.562

Service GDP paired with manufacturing FDIH0: Service output does not cause manufacturing FDI 17.9925* 0.0119H0: Manufacturing FDI does not cause service output 0.5140 0.9141Critical F-value 11.259 10.562

Source: own calculations based on RBI online database; UNCTAD (2000); CSO (various issues).a Critical F-values are reported at 1% level.

* Significant at 1% level.** Significant at 5% level.

1204 WORLD DEVELOPMENT

of FDI in inducing economic growth. In otherwords, causation is mainly running from out-put growth to FDI stocks.

At the sector level, it turns out that favorablegrowth effects of FDI in India are largely re-stricted to the manufacturing sector, whereFDI stocks and output are mutually reinforcingin the long run. By contrast, there is no evi-dence at all of any causal relationship betweenthe two variables in the primary sector. Mostinterestingly, feedback effects between FDIand output turn out to be transitory in the ser-vices sector. If at all, long-run causality in theservices sector runs from output to FDI. Inaddition to Granger causality tests, we estimatethe elasticity of output with respect to FDI atthe level of industries. Results on the economicimpact of FDI on output growth corroboratethe finding that it was mainly the manufactur-ing sector that benefited from FDI in post-re-form India.

Yet, it may be premature to conclude thatbooming FDI in the services sector has failedto promote India’s economic growth. Grangercausality tests performed across sectors indicatethat output growth in manufacturing has beenstimulated not only by FDI in this sector butalso by FDI in the services sector. Moreover,it cannot be ruled out that we underestimatethe positive output effects within the servicessector. Offshoring of higher value-added ser-vices by foreign investors to India is a fairly re-cent phenomenon, the output effects of whichmay take considerable time to materialize. Inaddition, even sector-specific analyses may stillsuffer from aggregation bias (Aykut & Sayek,2007). Data limitations prevent us from assess-ing in more detail the output effects of FDI inIT-related business services. Hence, it seemsto be a promising avenue of future research tocomplement our sector-specific analysis withdetailed case studies, providing a more accurateaccount of the mechanisms through which FDIcan contribute to greater efficiency of India’sservices industries.

It would also be desirable to extend our anal-ysis by assessing the direction of causality be-tween FDI and growth in a multivariateframework. In this way, additional insightsmay be gained on indirect causality runningfrom FDI through auxiliary variables to eco-nomic growth. However, a multivariate analy-sis would be fairly demanding in terms ofdata requirements. Unless industry-specific

data on physical and human-capital formationare available, scholars have to choose betweena multivariate approach relying on highlyaggregate data and a bivariate approach usingdisaggregated FDI and output data. The caseof India clearly reveals the limitations of theformer approach, which completely ignoresthe sector-specific effects of FDI, whereas it israther unlikely that FDI captures the effectsof omitted variables in the Indian context.

Our findings qualify the optimistic view ofsome economists who have advised Indian pol-icymakers to throw wide open the doors to FDI(e.g., Bajpai & Sachs, 2000). Whether and towhat extent FDI translates into higher growthdoes not only depend on the overall amountof FDI that India attracts, but also on the typeof FDI and its structural composition. Resultson India are in line with the findings of cross-country studies according to which the growthimplications of FDI are shaped by various fac-tors, including absorptive capacity and localskills, technological spillovers and linkages be-tween foreign and local firms, and export orien-tation – all of which differ across industries andsectors in the host economy.

This raises the question whether India’s poli-cymakers should return to the highly selectiveapproval procedures of the pre-reform era thatwere meant to promote technologically ad-vanced and export-oriented FDI projects inmanufacturing industries and to discourageFDI in the tertiary sector where foreign inves-tors might replace local service providers. Inour view, the finding that the growth effects ofFDI differ across sectors does not speak in fa-vor of selective FDI policies and policymakersattempting to target preferred types of FDI inspecific industries. For such an approach tobe successful in attracting growth-promotingFDI, policymakers would have to know exactlyabout the quality of each FDI project and itseffects on the local economy. This appears tobe an overly heroic assumption.

It seems to be a more reasonable option forpolicymakers to help maximize the benefits ofFDI in India by improving local conditionsthat may render FDI more effective. Opennessto trade appears to be important for strength-ening linkages between foreign and local com-panies, notably in the manufacturing sector.The promotion of local entrepreneurship andhuman-capital development could help fosterlinkages within and across sectors. The avail-

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1205

ability of sufficiently skilled labor as well asadequate infrastructure, particularly telecom-munications, would support the process ofoffshoring higher value-added services to Indiaand the dissemination of the benefits of IT-re-lated services throughout the Indian economy(UNCTAD, 2004, pp. 207–208; World Bank,

2005, p. 76). Moreover, it may help strongerspillovers from more efficient services to othersectors if IT-related FDI were more widelyspread throughout India, rather than beingconcentrated in a few clusters and ‘‘export en-claves.’’

NOTES

1. DeLong (2003, p. 198) credits earlier, though rela-tively modest, reforms to have had ‘‘an enormous effecton India’s long-run economic destiny.’’

2. Panagariya (2005, p. 18) concludes: ‘‘DeLong’scontention that we lack hard evidence to support theview that the rapid growth of the second half of the1980s could not be sustained without the second wave ofreforms in the 1990s is untenable.’’

3. See Kose, Prasad, Rogoff, and Wei (2006) for anextensive review of the relevant literature.

4. By contrast, Borensztein et al. (1998) performed apanel analysis with decade averages of the FDI andgrowth variables.

5. Agrawal (2005) estimates a fixed effects model basedon pooled data for five South Asian host countries,among which India figures prominently.

6. In a later version of their paper, however, Sahoo andMathiyazhagan (2003) come to a different conclusionand claim that FDI inflows have played a vital role inthe Indian economy.

7. For a detailed account of India’s reform program,see Agrawal (2005), Balasubramanyam and Mahambare(2003), Gupta (2005), and Kumar (2003). As noted byBalasubramanyam and Mahambare, the relaxation ofthe dirigiste trade and FDI regime started in the mid-1980s already.

8. These changes are presented in some more detail in aprevious working paper version of this article, which isavailable upon request.

9. For example, Pradhan (2002) reports more favorableresults when restricting the period of observation to1986–97 (instead of 1969–97).

10. Yet FDI stocks in nominal terms multiplied even inthese industries. For example, the share of the chemicalindustry in overall FDI stocks dwindled from almost

30% in 1987 to 3.4% in 2000, even though FDI stocks inthis industry increased fivefold to Rs. 26.2 billion in2000.

11. In addition, FDI in financial services gainedconsiderable importance. By contrast, FDI stocks inservices such as ‘‘electricity and water distribution,’’‘‘trade,’’ and ‘‘transport and storage’’ continued to be ofminor importance.

12. See also UNCTAD (2004, pp. 111, 124).

13. The point that other sectors will be positivelyaffected if FDI improves the quality of services in thehost country is made by Aykut and Sayek (2007) as wellas UNCTAD (2004, pp.123,123,169). We would alsolike to thank several anonymous referees who stressedthe importance of linkages across sectors.

14. According to the survey by Lipsey (2002, p. 41),‘‘most studies of productivity spillovers from foreigninvestment assume that they occur mainly in theindustry in which the foreign firm operates.’’

15. Being introduced in the econometric literature byGranger (1981), the concept of cointegration was furtherextended and formalized by Engle and Granger (1987).The concept refers to the idea that, although economictime series may exhibit nonstationary behavior, anappropriate linear combination between trending vari-ables could remove the common trend component and,hence, produce a stationary relationship between thevariables.

16. The link between the cointegration technique andthe error correction model is formalized in Granger andWeiss (1983). Following the works of Granger (1986,1988), Engle and Granger (1987), Granger, Huang, andYang (2000), the use of vector error correction modelshas gained prominence in the recent literature.

17. For recent bivariate approaches with respect tocausality between exports and growth, see Clarke andRalhan (2005), Konya (2006), and Sharma and

1206 WORLD DEVELOPMENT

Panagiotidis (2005); for comprehensive reviews oncausality between local financial development andgrowth, see Arestis and Demetriades (1997) as well asWachtel (2004).

18. The combination of a stationary variable with aunit-root variable may not yield a cointegrated relation-ship and, thus, fails to provide an adequate frameworkfor assessing long-run and short-run causality.

19. Unit root and cointegration tests for normalizedFDI are not shown here, but are available from theauthors upon request. Alternatively, one can take firstdifferences of all variables under consideration and makethem I (0). However, this would imply losing a lot ofvaluable information, rendering this approach inferiorto using nonnormalized variables. We are grateful toMarcel Fratzscher for alerting us to this point.

20. We would like to thank Peter Pedroni for advisingus on this point.

21. Since each test statistic has its own weaknesses, it isnow a standard practice to use a combination of teststatistics for the unit root test.

22. The only exceptions are the ADF statistics of Imet al. (2003) in models M3 and M4.

23. Outlined in Pindyck and Rubinfeld (1991), thismethodology for choice of optimum lag structure is astandard practice in the empirical literature.

24. The long-run effects reflect movements along thepath of a steady state equilibrium relationship betweenoutput and FDI stocks and, hence, are consideredpermanent.

25. Alfaro’s (2003) observation of an insignificantgrowth effect of cross-country FDI flows offers aninteresting reference point for this result.

26. The coefficient of variation for FDI in mining andquarrying (0.91) and FDI in petroleum (1.12) wassubstantially higher than the coefficient of variation forFDI in major manufacturing industries such as chem-icals (0.36) and machinery (0.54).

REFERENCES

ADB (2004). Asian Development Outlook 2004. Part 3:Foreign Direct Investment in Developing Asia.Manila, Philippines: Asian Development Bank.

Agrawal, P. (2005). Foreign direct investment in SouthAsia: Impact on economic growth and local invest-ment. In E. M. Graham (Ed.), Multinationals andforeign investment in economic development(pp. 94–118). Basingstoke, UK: Palgrave Macmillan.

Agrawal, R., & Shahani, R. (2005). Foreign investmentin India: Issues and implications for globalisation. InC. Tisdell (Ed.), Globalisation and world economicpolicies: Effects and policy responses of nations andtheir groupings (pp. 644–658). New Delhi, India:Serials Publication.

Alfaro, L. (2003). Foreign direct investment and growth:Does the sector matter?. Boston, MA: HarvardBusiness School, Mimeo.

Alfaro, L., Chanda, A., Kalemli-Ozcan, S., & Sayek, S.(2004). FDI and economic growth: The role of localfinancial markets. Journal of International Econom-ics, 64(1), 89–112.

Arestis, P., & Demetriades, P. (1997). Financial devel-opment and economic growth: Assessing the evi-dence. Economic Journal, 107(442), 783–799.

Athreye, S., & Kapur, S. (2001). Private foreign invest-ment in India: Pain or panacea?. The World Econ-omy, 24(3), 399–424.

Aykut, D., & Sayek, S. (2007). The role of the sectoralcomposition of foreign direct investment on growth.In L. Piscitello, & G. Santangelo (Eds.), Do multi-

nationals feed local development and growth?(pp. 35–62). London: Elsevier.

Bajpai, N., & Sachs, J. D. (2000). Foreign directinvestment in India: Issues and problems. Develop-ment Discussion Paper 759. Harvard Institute forInternational Development, Harvard University,Cambridge, MA.

Balasubramanyam, V. N., & Mahambare, V. (2003).FDI in India. Transnational Corporations, 12(2),45–72.

Balasubramanyam, V. N., Salisu, M. A., & Sapsford, D.(1996). Foreign direct investment and growth in EPand IS countries. Economic Journal, 106(434),92–105.

Basu, P., Chakraborty, C., & Reagle, D. (2003). Liber-alization, FDI, and growth in developing countries:A panel cointegration approach. Economic Inquiry,41(3), 510–516.

Bhagwati, J. N. (1978). Anatomy and consequences ofexchange control regimes. New York: NationalBureau of Economic Research.

Bhat, K. S., Sundari, T. C. U., & Raj, K. D. (2004).Causal Nexus between foreign direct investment andeconomic growth in India. Indian Journal of Eco-nomics, 85(337), 171–185.

Blomstrom, M., & Kokko, A. (1998). Multinationalcorporations and spillovers. Journal of EconomicSurveys, 12(2), 1–31.

Blonigen, B. A., & Wang, M. (2004). Inappropriatepooling of wealthy and poor countries in empirical

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1207

FDI studies. NBER Working Paper 10378. NationalBureau of Economic Research, Cambridge, MA.

Borensztein, E., De Gregorio, J., & Lee, J. W. (1998).How does foreign direct investment affect economicgrowth?. Journal of International Economics, 45(1),115–135.

Canning, D., & Pedroni, P. (2004). The effect ofinfrastructure on long run economic growth. Williams-town, MA: Williams College, Mimeo.

Carkovic, M. V., & Levine, R. (2005). Does foreigndirect investment accelerate economic growth?. In T.H. Moran, E. M. Graham, & M. Blomstrom (Eds.),Does foreign direct investment promote development?(pp. 195–220). Washington, DC: Institute for Inter-national Economics.

Central Statistical Organisation (various issues). Statis-tical pocket book India. Government of India, NewDelhi, India.

Chakraborty, C., & Basu, P. (2002). Foreign directinvestment and growth in India: a cointegrationapproach. Applied Economics, 34(9), 1061–1073.

Chowdhury, A., & Mavrotas, G. (2006). FDI andgrowth: What causes what?. The World Economy,29(1), 9–19.

Clarke, J.A., & Ralhan, M. (2005). Direct and indirectcausality between exports and economic output forBangladesh and Sri Lanka: Horizon matters. Eco-nometrica Working Paper EWP05012. University ofVictoria, Victoria, BC, Canada.

Clemens, M.A., Radelet, S., & Bhavnani, R. (2004).Counting chicken when they hatch: The short-termeffect of aid on growth. Working Paper 44. Centerfor Global Development, Washington, DC.

DeLong, J. B. (2003). India since independence: Ananalytic growth narrative. In D. Rodrik (Ed.). Searchof prosperity: Analytical narratives on economicGrowth (pp. 184–204). Princeton and Oxford: Prince-ton University Press.

De Mello, L. R. Jr., (1997). Foreign direct investment indeveloping countries and growth: A selective survey.Journal of Development Studies, 34(1), 1–34.

Dua, P., & Rashid, A. I. (1998). FDI and economicactivity in India. Indian Economic Review, 33(2),153–168.

Dufour, J.-M., & Renault, E. (1998). Short run and longrun causality in time series: Theory. Econometrica,66(5), 1099–1125.

Enderwick, P. (2005). Attracting ‘‘desirable’’ FDI:Theory and evidence. Transnational Corporations,14(2), 93–119.

Engle, R. F., & Granger, C. W. J. (1987). Cointegrationand error correction: Representation, estimation andtesting. Econometrica, 55(2), 277–304.

Findlay, R. (1978). Relative backwardness, direct for-eign investment, and the transfer of technology: Asimple dynamic model. Quarterly Journal of Eco-nomics, 92(1), 1–16.

Fischer, S. (2002). Breaking out of the third world:India’s economic imperative. <http://www.imf.org/external/np/speeches/2002/012202.htm>.

Frimpong, J. M., & Oteng-Abayie, E. F. (2006).Bivariate causality analysis between FDI inflowsand economic growth in Ghana. MPRA Paper 351.

University of Munich, Munich. <http://mpra.ub.uni-muenchen.de/351>.

Gorg, H., & Greenaway, D. (2004). Much ado aboutnothing? Do domestic firms really benefit fromforeign direct investment?. The World Bank ResearchObserver, 19(2), 171–197.

Gorg, H., & Strobl, E. (2001). Multinational companiesand productivity spillovers: A meta-analysis. Eco-nomic Journal, 111(475), F723–F739.

Granger, C. W. J. (1981). Some properties of timeseries data and their use in econometric modelspecification. Journal of Econometrics, 16(1), 121–130.

Granger, C. W. J. (1986). Developments in the study ofthe cointegrated economic variables. Oxford Bulletinof Economics and Statistics, 48(3), 201–212.

Granger, C. W. J. (1988). Some recent developments in aconcept of causality. Journal of Econometrics, 39(1–2), 199–211.

Granger, C. W. J., & Weiss, A. A. (1983). Time seriesanalysis of error correction models. In S. Karlin, T.Amemiya, & L. A. Goodman (Eds.), Studies ineconometrics: Time series and multivariate statistics(pp. 255–278). New York, NY: Academic Press.

Granger, C. W. J., Huang, B.-N., & Yang, C.-W. (2000).A bivariate causality between stock prices andexchange rates: Evidence from recent Asian Flu.The Quarterly Review of Economics and Finance,40(3), 337–354.

Gupta, S. (2005). Multinationals and foreign directinvestment in India and China. In E. M. Graham(Ed.), Multinationals and foreign investment in eco-nomic development (pp. 198–211). Basingstoke, UK:Palgrave Macmillan.

Hansen, H., & Rand, J. (2006). On the causal linksbetween FDI and growth in developing countries.The World Economy, 29(1), 21–41.

Hermes, N., & Lensink, R. (2003). Foreign directinvestment, financial development and economicgrowth. Journal of Development Studies, 40(1),142–163.

Im, K., Pesaran, M. H., & Shin, Y. (2003). Testing forunit roots in heterogeneous panels. Journal ofEconometrics, 115(1), 53–74.

Jenkins, R. (1990). Comparing foreign subsidiaries andlocal firms in LDCs: Theoretical issues and empiricalevidence. Journal of Development Studies, 26(2),205–228.

Kamalakanthan, A., & Laurenceson, J. (2005). Howimportant is foreign capital to income growth inChina and India? ABERU Discussion Paper 14.Melbourne: Asian Business and Economic ResearchUnit, Monash University. (<http://www.bus-eco.monash.edu.au/units/aberu/papers/1405Abby.pdf>).

Khawar, M. (2005). Foreign direct investment andeconomic growth: A cross-country analysis. GlobalEconomic Journal, 5, Article 8. (<http://www.be-press.com/gej>).

Konya, L. (2004). Unit-root, cointegration and Grangercausality test results for export and growth in OECDcountries. International Journal of Applied Economet-rics and Quantitative Studies, 1–2, 67–94.

1208 WORLD DEVELOPMENT

Konya, L. (2006). Exports and growth: Granger causal-ity analysis on OECD countries with a panel dataapproach. Economic Modelling, 23(6), 978–992.

Kose, M. A., Prasad, E., Rogoff, K., & Wei, S.-J. (2006).Financial globalization: A reappraisal. IMF Work-ing Paper WP/06/189. International Monetary Fund,Washington, DC.

Kumar, N. (2003). Liberalization, foreign direct invest-ment flows and economic development: The Indianexperience in the 1990s. RIS Discussion Paper 65.Research and Information System for the Non-Aligned and Other Developing Countries, NewDelhi, India.

Kumar, N., & Pradhan, J. P. (2002). Foreign directinvestment, externalities and economic growth indeveloping countries: Some empirical explorationsand implications for WTO negotiations on invest-ment. RIS Discussion Paper 27. Research andInformation System for the Non-Aligned and OtherDeveloping Countries, New Delhi, India.

Lensink, R., & Morrissey, O. (2006). Foreign directinvestment: Flows, volatility and the impact ongrowth. Review of International Economics, 19(3),478–493.

Levin, A., Lin, C., Chu, J., & Shang, C. (2002). Unitroot tests in panel data: Asymptotic and finite-sample properties. Journal of Econometrics, 108(1),1–24.

Lim, E.-G. (2001). Determinants of, and the relationbetween, foreign direct investment and growth: Asummary of the recent literature. Working Paper 01/175. International Monetary Fund, Washington,DC.

Lipsey, R. E. (2002). Home and host country effects ofFDI. NBER Working Paper 9293. National Bureauof Economic Research, Cambridge, MA.

Nair-Reichert, U., & Weinhold, D. (2001). Causalitytests for cross-country panels: A new look at FDIand economic growth in developing countries.Oxford Bulletin of Economics and Statistics, 63(2),153–171.

OECD (2002). Foreign Direct Investment for Develop-ment – Maximising Benefits, Minimising Costs.OECD, Paris.

Panagariya, A. (2005). The triumph of India’s marketreforms: The record of the 1980s and 1990s. PolicyAnalysis (Cato Institute), 554, 1–22.

Pedroni, P. (1999). Critical values for cointegration testsin heterogeneous panels with multiple regressors.Oxford Bulletin of Economics and Statistics, 61(Sup-plement 1), 653–670.

Pedroni, P. (2004). Panel cointegration: asymptotic andfinite sample properties of pooled time series testswith an application to the PPP hypothesis. Econo-metric Theory, 20(3), 325–597.

Pindyck, R. S., & Rubinfeld, D. L. (1991). Econometricmodels and economic forecasts. New York, NY:McGraw-Hill.

Pradhan, J. P. (2002). Foreign direct investment andeconomic growth in India: A production functionanalysis. Indian Journal of Economics, 82(327),582–586.

Rajan, R. G., & Subramanian, A. (2005). Aid andgrowth: What does the cross-country evidence reallyshow? IMF Working Paper WP/05/127. Interna-tional Monetary Fund, Washington, DC.

Reserve Bank of India (2005). Annual Report 2004–05.New Delhi, India. <http://rbidocs.rbi.org.in/rdocs/AnnualReport/PDFs/65516.pdf>.

Reserve Bank of India (various issues). Finances offoreign direct investment companies, New Delhi,India. <http://rbidocs.rbi.org.in/rdocs/Bulletin/DOCs/62203.doc>.

Reserve Bank of India (online). Database on Indianeconomy (<http://reservebank.org.in/cdbmsi/serv-let/login/>.

Rodriguez-Clare, A. (1996). Multinationals, linkagesand economic development. American EconomicReview, 86(4), 852–873.

Sahoo, D., & Mathiyazhagan, M. K. (2002). Economicgrowth in India: Does foreign direct investmentinflow matter? Working Paper 115. Institute forSocial and Economic Change, Bangalore, India.

Sahoo, D., & Mathiyazhagan, M. K. (2003). Economicgrowth in India: Does foreign direct investmentinflow matter? Singapore Economic Review, 48(2),151–171.

Sharma, A., & Panagiotidis, T. (2005). An analysis ofexports and growth in India: Cointegration andcausality evidence (1971–2001). Review of Develop-ment Economics, 9(2), 232–248.

Toda, H., & Yamamoto, T. (1995). Statistical inferencein vector autoregressions with possibly integratedprocesses. Journal of Econometrics, 66(1–2), 225–250.

UNCTAD (2000). World Investment Directory, VolumeVII – Part 1: Asia and the Pacific. United Nations,New York and Geneva.

UNCTAD (2001). World Investment Report 2001.United Nations, New York and Geneva.

UNCTAD (2004). World Investment Report 2004.United Nations, New York and Geneva.

Wachtel, P. (2004). How much do we really know aboutgrowth and finance?. In M. Bagella (Ed.). Monetaryintegration markets and regulations. research inbanking and finance (Vol. 4, pp. 91–113). Amsterdam:Elsevier.

World Bank (2005). Global development finance. Mobi-lizing finance and managing vulnerability. Part I.World Bank, Washington, DC.

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1209

APPENDIX A. INDUSTRIES INCLUDED IN PANEL

Broad sector Included industries

Primary sector Agriculture, hunting, forestry and fishingMining and quarryingPetroleum.

Secondary sector Food, beverages and tobaccoTextiles, leather and clothingChemicals and chemical productsBasic metals and metal productsMachinery equipment and electrical machineryMotor vehicles and other transport equipment

Tertiary sector Electricity and water distributionConstructionDistributive tradeTransport and storageFinanceOther services

APPENDIX B. VARIABLE DESCRIPTION AND SOURCES

Variable Source Definition

FDI stocks(FDI)

Data on nominal stocks aspresented by UNCTAD (2000)for 1987–95 and CentralStatistical Organisation(various issues) for more recentyears. Both UNCTAD andCSO refer to the Reserve Bankof India as the ultimate sourceof data.

Industry-specific FDI stocks inconstant prices; nominal stocksin Indian Rupees convertedinto constant prices byapplying the deflator for netcapital stocks (all institutions;1993 = 1)

Output (GDP) Series in constant pricesavailable from RBI, OnlineDatabase on Indian Economy,for industries in the primary(except petroleum) and tertiarysectors; for petroleum andindustries in the secondarysector, indices of industrialproduction are available fromthe same source; industry-specific weights on theircontribution to GDP are givenby Central StatisticalOrganisation (various issues)

Industry-specific contributionto GDP in constant prices of1993; series in constant pricesavailable for industries in theprimary (except petroleum) andtertiary sectors; for petroleumand industries in the secondarysector, indices of industrialproduction are converted intoconstant Rupee series byapplying industry-specificweights on their contribution toGDP in 1993. These weightsare also used to align the finerdisaggregation of industrieswith respect to output with theindustry aggregation availablefor FDI stocks

1210 WORLD DEVELOPMENT

APPENDIX C. SUMMARY STATISTICS: 1987–2000

FDI,constant prices

Output, constantprices

FDI inpercent of

output

Mean(mill Rs)

2000over 1987

Mean(mill Rs)

2000over 1987

Mean Std.dev.

Primary sectorAgriculture, hunting, forestryand fishing

3006 0.69 245480 1.57 1.25 0.33

Mining and quarrying 340 9.24 21504 2.02 1.42 1.00Petroleum 1400 130.20 11048 3.46 10.36 9.43

Secondary sectorFood, beverages and tobacco 6551 10.53 21289 2.01 27.29 22.21Textiles, leather and clothing 3396 3.73 20477 2.06 14.78 8.62Chemicals and chemical products 14420 1.93 24847 2.62 58.06 10.48Basic metals and metal products 2792 1.66 17095 2.33 16.14 4.54Machinery equipment and electricalmachinery

15185 2.80 18953 2.68 77.68 26.53

Motor vehicles and other transportequipment

9842 8.86 7781 2.65 111.36 49.93

Tertiary sectorElectricity and water distribution 2512 7.94 20687 2.43 9.79 17.83Construction 519 3.13 43994 2.10 1.16 0.48Distributive trade 1100 25.05 108093 2.39 0.86 0.80Transport and storage 117 0.56 45809 2.12 0.30 0.28Finance 10207 3468.29a 46679 4.02 14.17 20.36Other services 75006 450.77 174317 2.53 30.86 45.23

Source: UNCTAD (2000); Central Statistical Organisation (various issues); Reserve Bank of India (Database onIndian Economy).a 2000 over 1991.

APPENDIX D. FDI ELASTICITY OF OUTPUT BY SECTORS: 1987–2000

Elasticity t-Statistic

Primary sectorAgriculture, hunting, forestry and fishing !0.13699 !1.0025Mining and quarrying 0.22483 6.480608Petroleum 0.157552 5.487986

Secondary sectorFood, beverages and tobacco 0.201614 5.304724Textiles, leather and clothing 0.281226 6.866871Chemicals and chemical products 0.718907 5.35094Basic metals and metal products 0.512982 3.655454Machinery equipment and electrical machinery 0.489263 4.973232Motor vehicles and other transport equipment 0.432267 15.78618

Tertiary sectorElectricity and water distribution 0.094868 3.529823

(continued on next page)

ECONOMIC REFORMS, FDI, AND ECONOMIC GROWTH IN INDIA 1211

Available online at www.sciencedirect.com

Appendix D. – continued

Elasticity t-Statistic

Construction 0.194779 2.064932Distributive trade 0.25656 8.705797Transport and storage !0.10766 !1.85666Finance 0.086496 17.33855Other services 0.101826 8.911131

Source: Authors’ calculation based on regression of log transformed variables of FDI and Output.

1212 WORLD DEVELOPMENT