Relationships among export, FDI and growth in India: an application of auto regressive distributed lag (ARDL) bounds testing approach.

Guru-Gharana, Kishor K.

INTRODUCTION

India is an interesting and increasingly important case for the study of relationship among growth, Trade and Foreign Direct Investment. India is the second largest country in the world, with a population of over 1.123 billion (in 2007) which is more than one-sixth of all the people in the world. The output of India accounts for almost 8% of global GDP when measured appropriately. In particular, when National Income is measured using PPP (purchasing power parity) reflecting the actual purchasing power of a country's currency, India is fourth after the US, China, and Japan (with PPP GNI $3078.7 billion compared to only 2782.7 billion for Germany in the fifth position and $4,420.6 for Japan in the third position in 2007: see World Development Report 2009). Moreover, India alone has accounted for roughly one fifth of global GDP growth in the last five years. During the period between 2000 and 2007, the growth rate of Indian Economy was an impressive 7.8% per year compared to only 3.2% for the world, only 2.4% for High Income, only 4.3% for upper middle Income, and only 5.6% for low Income countries (World Development Report, 2009). India is fast catching up in overall growth with the champions, the East Asia and Pacific region countries, which have so far been ahead and above (8.9% during this period).

Relatively low wages and vast reservoir of trained manpower make India a natural destination for foreign direct investment (FDI). Until recently, however, India has attracted only a small share of global FDI, primarily due to government restrictions on foreign involvement in the economy. But beginning in 1991 and accelerating rapidly since 2000, India has liberalized its investment regulations and actively encouraged new foreign investment: a sharp reversal from decades of discouraging economic integration with the global economy. India's recent liberalizations and foreign investment de-regulations have generated strong interest by foreign investors, turning India into one of the fastest growing destinations for global foreign direct investment (FDI) inflows. Foreign firms are setting up joint ventures and wholly owned enterprises in various services and manufacturing sectors. Net foreign direct investment (FDI) flows into India reached $22.8 billion in 2007, more than five times the $4.0 billion recorded during 2001. In 2008 it is recorded to jump to $34.4 billion. India has emerged as the second most attractive destination for FDI after China and ahead of the US, Russia and Brazil. According to the News report published in October 2009 by the Trade Council of Denmark, India achieved a stunning 85.1% increase in foreign direct investment flows in 2008, the highest increase across all countries, even as global flows declined by 14.5%, says the findings (quoting a recent UNCTAD study--Assessing the impact of the current financial and economic crisis on global FDI flows). Similarly, export volume has increased from a mere $16.6 billion in 1990 to $ 163.1 billion in 2008 (an increase of over 1000 % in 18 years!). Policy makers and Research scholars have been touting this impressive export and FDI growth in recent decades as the vehicles for India's accelerated growth in the recent years and possibly in decades to come.

In the last two decades there have been several studies on such relationships investigating Export led or Foreign Investment led growth in India, but all suffer from methodological issues. Most studies ignore the time series nonstationarity properties of these macro variables which can lead to spurious Regressions and Correlations. Some do investigate the nonstationarity properties but then perform the Granger Causality Tests using simple VAR or VECM or Johansen-Juselius cointegration procedures with the exception of Shirazi and Abdul-Manap (2005), which does not include FDI. But Toda and Phillips (1993) have provided evidence that the Granger causality tests in error correction Models (ECMs) still contain the possibility of incorrect inference and suffer from nuisance parameter dependency asymptotically (Lutkephol 2004, p. 148). The underlying assumption of these studies that all variables in the system are integrated of the same order (usually order 1) makes their subsequent results doubtful, because there are several evidences that all macro variables included in the system may not be of the same order. In the present study too, we find that the variables in the system may not be of the same order of integration. It is also evident from studying the estimation results of the previous studies, because they assume arbitrarily that all variables are integrated of order I(1) so as to be able to apply the cointegration technique, even after their own tests show that some of the variables may be I(0) or I(2). Another problem with previous studies is that they have studied such relationships in bivariate contexts. However, in time series multivariate relationships may be quite different from bivariate relationship (Love & Chandra 2005a and 2005b).

Therefore, in the present study we employ the more robust technique developed by Pesaran and Shin (1995 and 1998), and Pesaran et al. (2001), and apply multivariate framework by including GDP, Export and FDI. These state-of-the-art techniques have not been employed in the Indian context to study the relationships among these variables. The rest of the paper is organized as follows: in section 2 we briefly outline the theoretical framework and the literature linking Exports and FDI to Growth; in section 3 we briefly mention the relevant empirical literature; in section 4 we describe the Econometric Methodology; in section 5 we discuss the data and the results of estimation and hypotheses testing; and in the final section we conclude with the main findings of this study.

THEORETICAL FRAMWWORK

Export Led Growth (ELG)

The notion that export leads economic growth has been subject to considerable debate in the development and growth literature for many decades (Keesing, 1967 and Krueger, 1995). Broadly, the focus of the Export led Growth (ELG) debate is on whether a country is better served by orienting trade policies to export promotion or to import substitution. The neoclassical view has been that growth can be achieved by ELG. The growth records of Asian newly industrializing countries (NICs), mainly, Hong Kong, Singapore, Korea and Taiwan, and second-generation NICs (Malaysia and Thailand)--are cited as such examples (compared to other developing regions). Over the last four decades these NICs have approximately doubled their standards of living in every successive decade. China is the newest and now the most dominant member of this group. China's experience since the 1980s tends to support the argument that trade openness is a mechanism for achieving more rapid and efficient growth and better distribution of domestic resources (Findlay and Watson, 1996, p.4). Some authors (e.g., Krueger, 1995) identify trade policy as the crucial element of economic policy.

Some of the reasons cited in support of the ELG proposition are: (a) export growth represents an increase in demand for the country's output; (b) exports promote specialization in the production of export products, which in turn may boost the productivity level and the general level of skills and result in a more efficient reallocation of resources; (c) the outward oriented trade policy may also give better access to advanced technologies, learning by doing gains, and better management practices (Ben-David & Loewy, 1998) that may result in further efficiency gains; (d) exports may loosen a foreign exchange constraint (Chenery & Strout, 1966), which makes it easier to import inputs to meet domestic demand, and so enable output expansion; (e) some authors argue (Lal & Rajapatirana, 1987) that an outward-oriented strategy of development may provide greater opportunities and rewards for entrepreneurial activity, the key to extended growth. However, the support for ELG is not universal. Critics point out that the experiences in the East and Southeast Asian countries are unique in many ways and not necessarily replicable in other countries (Buffie, 1992). Other researchers question whether a reliance on exports to lead the economy will result in sustained long-term economic growth in LDCs due to the volatility and unpredictability in the world market (Jaffee, 1985). There is also a potential for no causal relationship between exports and economic growth when the time paths of the two series are determined by other, unrelated variables (e.g., investment) in the economic system (Pack, 1988). Eventually, effectiveness of export promotion is an empirical issue. Over the last two decades or so there has been a plethora of such investigations, using a number of statistical techniques, from very simple to very complex. But the results have been mixed.

Foreign Direct Investment Led Growth (FDI-LG)

It is a general belief among policy makers and academicians that foreign direct investment (FDI) can be a source of valuable technology and know-how in addition to increased capital. Some of the popularly cited potential benefits of FDI are: (a) backward and forward linkages with the rest of the economy; (b) enhanced access to advanced technologies; (c) learning of improved management practices; (d) expansion and diversifification of the production capacity of an economy; (e) transfer of best practices in corporate governance and accounting practices; (f) integration of the domestic economy with the global economy and infusion of competition in the domestic economy; and (g) relatively more stability than other forms of international capital flows because of longer-term perspective.

Both trade and FDI are also associated with growth, though their multichannel causal links remain largely unexplored especially in the case of India. Notwithstanding the strong conceptual case for a positive relationship between economic growth and FDI, the empirical evidence has been mixed (Blomstrom & Kokko (1998), Gorg & Greenaway (2004), and Alfaro & Rodriguez-Clare (2004)).

It has been recognized and well documented in the literature that there is possibility of two-way feedbacks between FDI and economic growth along with their long-run and short-run dynamics. Empirical investigations in the context of the Indian economy have generally failed to provide any conclusive evidence in support of such two-way feedback effects; causality between FDI and economic growth is either found neutral for India, or to run mainly from economic growth to FDI. Earlier studies, however, have several limitations in common. First, the period of observation is typically too short in the post-liberalization period to capture the effects of economic reforms and the subsequent boom in FDI. In the present study we show that this factor has significant influence on the results. Second, the econometric techniques employed (even in those studies which take into account the nonstationarity properties) are highly dependent on the results of testing for the cointegration relationships (Basu, et al, 2003). Third, only bivariate relationship is studied in most of the previous studies, which may involve biases (see Love and Chandra 2005b). In this paper we avoid these methodological problems.

SELECTED EMPIRICAL LITERATURE REVIEW

The empirical literature separates into three or four groups: the first group uses cross-country correlation coefficients to test the ELG hypothesis; the second group uses regression models (typically least squares based) that are again usually cross-country predicated; the third, recent group of studies applies various time series techniques to examine the exports-growth or FDI-growth nexus and the fourth group applies panel data and panel cointegration techniques. The group of cross-section research looks at rank correlation coefficients or simple OLS regressions between exports and output or FDI and output (or their growth) across a number of countries. The ELG or FDI led Growth hypotheses are supported when a positive and statistically significant correlation is observed. One issue arising from this body of work is that some of the results may involve a spurious correlation due to exports and FDI themselves being part of national product or all of them being influenced by some other variables.

Potential problems with the later time series studies are also well documented in the literature. Jung and Marshall (1985), Greenaway & Sapsford (1994), Riezman et al. (1996), and Dhananjayan & Devi (1997) provide surveys on the earlier ELG works. For a more recent and an extensive survey of empirical works on export led growth see Giles and Williams (2000a and 2000b).

Nandi & Biswas (1991) and Bhat (1995) found support for ELG hypothesis in the case of India, while Xu (1996) contradicts this finding. Similarly, Ahmed, Butt & Alam (2000), using trivariate causality framework, rejected the ELG hypothesis for all but one (Bangladesh) of the countries they studied. A recent study by Kalirajan et al. (2009) employs Multivariate VAR analysis using the VECM procedure to study relationships among FDI, Exports and Economic Growth in South Asia and selected emerging countries and find evidence in support of ELG hypothesis. Duttaray, et al. (2008) studied the role of FDI in less developed countries. Thangavelu & Rajaguru (2004) compared the roles of exports and imports on productivity growth in rapidly developing Asian countries. Makki & Somwaru (2004) studied the impact of FDI and trade on economic growth in some selected developing countries. Chandra (2003) followed and updated Dhawan & Biswal (1999) to test the export-led growth hypothesis in India in a multivariate framework but used the cointegration technique discussed above to test for causality. Moreover, this study uses data mostly overlapping with the pre-liberalization period (1950-1996) and the results may also suffer from pretesting bias as mentioned above. There have been several other empirical studies conducted for India but all are dominated by pre liberalization data.

Shirazi & Abdul-Manap (2005) examine the ELG hypothesis for five South Asian Countries through cointegration and multivariate Granger causality tests. No causality among exports, imports and output was found for Sri Lanka and India, although for India GDP and exports did induce imports. This study employs the relatively robust Toda & Yamamoto (1995) approach to testing Granger causality in the ELG context but does not include FDI, which is a significant omission considering the recent boom in this variable in India. Including Import instead of FDI does not seem to be appropriate in our view because of the extremely high interdependence and correlation between Exports and Imports. In simple words, Imports add little to what the Exports reveal, whereas FDI would be a significant largely independent addition to the model. Another limitation of this study is that the data for India are quite dated (only up to 2002) considering the very recent boom in Exports, FDI and Growth in India. Thus the post liberalization period is quite underrepresented while the data is dominated by periods of import-substituting-inward-looking policy regimes since 1960. Shirazi & Abdul-Manap (2005) do not even perform the subperiod analysis. Therefore, their conclusions (finding non-causality) are biased because of merging different policy regimes as if nothing important has happened in India in the 1990's onwards. This seems to be a general problem of cross country studies which lose focus on country specific events and unique characteristics. Because of this deficiency, it is now recognized that 'tests of the export-led model, must intrinsically involve country case studies' (Medina-Smith, 2001). As we have found in our own empirical results discussed below, the characteristics of Indian economy, especially in the context of Export-FDI-Growth nexus seem to have undergone significant changes following the watershed liberalization which started in the early 1990's and has become more vigorous in the last decade or so. Rahman (2009) has applied the ARDL methodology to study the effects of exports, FDI and expatriates' remittances on real GDP of Bangladesh, India, Pakistan and Sri Lanka. The results reveal close similarities of long-run and the short-run dynamics of the variables between Bangladesh and India. The same applies to Pakistan and Sri Lanka in terms of their short-run dynamics with no significant long-run causal flows.

Finally, Guru-Gharana & Adhikari (2010) apply the Toda-Yamamoto-Dolado-Lutkephol methodology to study the Granger Causality relationships among Growth, FDI and Exports in the case of China. The present study employs the alternative ARDL bounds testing methodology in the case of India.

THE ECONOMETRIC METHODOLOGY

The autoregressive distributed lag (ARDL) models were popular in energy analysis until the introduction of unit root tests and cointegration techniques which showed that the Least Squares methods could lead to spurious regressions in the presence of nonstationarity in the time series. This led to the band wagon effect of almost dismissing all methods based on OLS techniques, including ARDL and universally employing the popular Johansen-Juselius maximum likelihood techniques for studying long-run relationships and error correction models for studying Granger causality. Then came the criticisms of these methods (Toda & Phillips, 1993 and 1994, Toda & Yamamoto, 1995, and Zapata & Rambaldi, 1997) which showed that these methods also suffer from pre-testing biases, the low power of unit root tests, dependency on the accuracy of the assumed cointegration relationships, unsuitability for small samples, influences of nuisance parameters, and the need for the rank conditions to be satisfied for the validity of the results. There are added problems when the time series are of different order of integration. Parallel to these developments, there was a revival of ARDL methods in the late 1990s by a series of works, in particular, Pesaran & Shin (1999), and Pesaran, Shin & Smith (2001). Especially after Pesaran et al. (2001), the ARDL bounds testing approach has become the state-of-the-art technique for studying long-run and short-run relationships among time series variables and also for examining Granger causality. This approach is called the bounds testing approach because it involves testing whether the calculated F statistics are within or outside two bounds: the lower bound for I(0) and the upper bound for I(1).

The bounds testing approach has certain econometric advantages in comparison to the cointegration procedures of Engle & Granger (1987), Johansen (1988), Johansen & Juselius (1990), and the full Information Maximum Likelihood procedure of Johansen (1996). First, the endogeneity problems and inability to test hypotheses on the estimated coefficients in the long-run associated with the Engle-Granger methods are avoided. Second, the long-run and the short-run parameters of the model are estimated simultaneously. Third, the econometric methodology is relieved of the burden of establishing the order of integration amongst the variables and of pre-testing for unit roots, because the ARDL approach to testing the existence of long-run relationship is applicable irrespective of whether the underlying regressors are purely I(0), purely I(1) or fractionally integrated. Pre-testing is problematic in the unit-root-cointegration literature where the power of unit root tests are low and there is a switch in the distribution function of the test statistics as one or more roots of the x, process approaches unity (Pesaran and Pesaran, 1997).

Finally, the small sample properties of the bounds testing approach are superior to that of multivariate cointegration (see Mah, 2000, Narayan, 2005). The bounds testing approach modifies the ARDL framework while overcoming the inadequacies associated with the presence of a mixture of I(0) and I(1) regressors in a Johansen-Juselius type framework. There are, however, some caveats. The ARDL bounds testing of Pesaran et al. (2001) is valid only for order of integration up to 1, that is, between I(0) and I(1) inclusive. If the time series involves integration of higher order, say, I(2) then the results are not valid. Therefore, at the very outset, the unit root tests are performed (although not essential for ARDL per se) in order to establish the suitability of this approach. Moreover, there is a possibility that the results turn out to be inconclusive (when the test statistics fall within the two critical value bounds), in which case recourse to other methods of testing is required. The ARDL bounds testing approach in the context of Export-FDI-Growth nexus is outlined below.

Step1. The Examination of Lon-run Cointegration

After determining the suitability of the ARDL approach though unit root tests, a system of Unrestricted Error correction Model (UCEM) is estimated. The UCEM is a system of equations using each variable in turn as the dependent variable. Following Jayaraman and Singh (2007) the UCEM for this study can be represented (representing log of GDP as G, log of Export as E, and log of FDI as F) as the following:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)

Here, the [lambda]'s are the long run multipliers, [alpha]'s are the drift terms, [beta]'s are the short term dynamic coefficients, [DELTA] is the difference operator and p is the optimal lag length selected by some suitable criteria(on), and [epsilon]'s are white noise error terms, that is, i. i. d. with zero mean, Homoscedasticity and no autocorrelation. The optimal order of the lags on the first-difference variables in equations (1) to (3) can be obtained from unrestricted vector autoregression (UVAR) by means of Akaike and/or Schwarz criteria. Bahmani-Oskooee & Bohl (2000) and Bahmani-Oskooee and Ng (2002) have, however, shown that the results of this stage are sensitive to the order of VAR. We allow maximum lag order of 3 considering that our model consist of annual data. The Wald (F) test is performed to determine whether along run relationship exists between the variables through testing the joint significance of the lagged levels of the variables. The log-run relationship test is equivalent to the cointegration test. The Null Hypotheses for No Cointegration are:

[H.sub.o]: [[lambda].sub.gg] = [[lambda].sub.ge] = [[lambda].sub.gf] =0, against [H.sub.1]: [[lambda].sub.gg] [not equal to] 0, [[lambda].sub.ge] [not equal to] 0, [[lambda].sub.gf] [not equal to] 0 denoted as [F.sub.g](G | E, F) in equation (1).... (h1)

[H.sub.o]: [[lambda].sub.eg] = [[lambda].sub.ee] = [[lambda].sub.ef] =0, against [H.sub.1]: [[lambda].sub.eg] [not equal to] 0, [[lambda].sub.ee] [not equal to] 0, [[lambda].sub.ef] [not equal to] 0 denoted as [F.sub.e](E G, F) in equation (2).... (h2)

[H.sub.o]: [[lambda].sub.fg] = [[lambda].sub.fe] = [[lambda].sub.ff] =0, against [H.sub.1]: [[lambda].sub.fg] [not equal to] 0, [[lambda].sub.fe] [not equal to] 0, [[lambda].sub.ff] [not equal to] 0 denoted as [F.sub.f](F G, E) in equation (3).... (h3)

The F test has a nonstandard distribution which depends upon: (i) whether variables included in the ARDL model are I(0) or I(1); (ii) the number of regressors; (iii) whether the ARDL model contains an intercept and/or a trend; and (iv) the sample size. Two sets of critical values (CVs) or asymptotic critical value bounds assuming that the independent variables are I(d) (where 0[less than or equal to] d [less than or equal to] 1, that is integrated of order 1 or less) are reported in Pesaran et al. (2001) and Narayan (2005) for various sample sizes. The upper bounds are derived assuming all variables to be I(1) and the lower bounds are derived assuming all variables to be I(0). Given the relatively small sample size in the present study (up to 38 observations), we extract appropriate CVs from Narayan (2005) instead of using the tables reported in Pesaran et al. (2001), because the latter correspond to much larger sample sizes. If the calculated F statistics fall outside the two bounds a conclusive inference can be drawn. If the calculated value exceeds the upper bound of the CV then the Null hypothesis of no cointegrating relationship is rejected. If the calculated values fall below the lower bound of the CV then the null hypothesis of no cointegration cannot be rejected. If, however, the calculated F falls within the two bounds then the test is inconclusive and further examination of the accurate order of cointegration has to be performed.

Step 2: Estimation of the Long-run ARDL (p, q, r) Model

If cointegration is established in the above step, then the next step is to estimate the conditional ARDL (p, q, r) long-run model. Treating GDP as the Dependent Variable in accordance with prevalent growth theory, this can be shown as the following:

[G.sub.t] = [gamma] + [[summation].sup.p.sub.i=1] [[lambda].sub.g,i] [G.sub.t-i]+ [[summation].sup.q.sub.i=0] [[lambda].sub.e,i] [E.sub.t-i]+ [[summation].sup.r.sub.i=o] [[lambda].sub.f,i] [F.sub.t-i] + [[eta].sub.t], (4)

where all variables are previously defined, [gamma] is the constant term and [[eta].sub.t] is a white noise error term. This step involves selecting the orders of the ARDL (p,q,r) model in the three variables using Akaike or Schwarz criteria before the model is estimated using Ordinary Least Squares Technique. Pesaran and Shin (1999) have shown that valid asymptotic inferences on short-run and long-run parameters can be made under least squares estimates of an ARDL model, provided the order of the ARDL model is appropriately augmented to allow for contemporaneous correlations between the stochastic components of the data-generating processes included in the estimation (Narayan, 2004). Pesaran and Shin (1999) demonstrate that the Schwarz criterion is superior over Akaike in the context of ARDL model. With the use of annual data they recommend choosing a maximum of two lags. This involves trials and experiments with all possible combinations of the lag orders for the variables in the system. The vector of lag orders which minimizes the Akaike or Schwarz statistic is selected.

Step 3: The Short-run Dynamics and the Granger Causality

In the third and the final step, we obtain the short-run dynamic parameters by estimating an error correction model associated with the long-run estimates. Moreover, this equation can also be used for testing short-run and long-run Granger causality. This model is specified as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)

The error correction term is derived as: [ECT.sub.t] = [G.sub.t] - [ [gamma] + [[summation].sup.p.sub.i=1] [[lambda].sub.gg,i] [G.sub.t-i]+ [[summation].sup.q.sub.i=0] [[lambda].sub.ge,i] [E.sub.t-i]+ [[summation].sup.r.sub.i=1] [[lambda].sub.gf,i] [F.sub.t-I]] after estimating equation (4).The coefficient of the lagged error correction term, [sigma] is the speed of adjustment, and the [beta]'s are the shot-run dynamic coefficients of the model's convergence to equilibrium. A joint hypothesis testing of the [beta]'s can be used for short-run Granger causality test for each regressor in turn. The significance of the coefficient of the error correction term ([sigma]) can be used for testing Long-run Granger causality. A combined hypothesis test for the [beta]'s and [sigma] can be used for strong form of Granger causality test. It has also become customary to test the stability of the model through recursive regression residuals using the Brown et al. (1975) technique, also known as the cumulative sum (CUSUM) and cumulative sum of squares (CUSUMSQ) tests. If the plots of these statistics fall inside the critical bounds of 5% significance then we assume that the coefficients of a given regression are stable.

EMPIRICAL RESULTS

Data and Econometric software

Annual time series data for Export and FDI from 1971 to 2008 were collected from Handbook of Statistics of Reserve Bank of India. The GDP data were collected from the latest issue of World Development Indicators of the World Development Report. The econometric software used is the EVIEWS 7 version

Unit Root Test for the Integration Properties of the Data Series

In stationary time series, shocks will be temporary and over the time their effects will decay as the series revert to their long run mean values. Nonstationary series will contain permanent components and may show false relationships. Granger and Newbold (1974) and Phillips (1986) have demonstrated that high [R.sup.2] and low DW are typical characteristics of spurious regressions. It has been well demonstrated that most of the economic variables are found to be nonstationary. The present study employs the Augmented Dickey -Fuller (ADF) Test for test of presence of unit roots (that is nonstationarity) of the individual series. The ADF Test includes extra lagged terms of the dependent variables in order to eliminate autocorrelation. The lag length on these extract term is determined by the Akaike Information Criterion (AIC). The ADF Test results are shown in the following table, where the results clearly indicate that GDP and Export series are nonstationary when the variables are defined at levels with or without constant and trend. Looking at the Table the FDI series appears to be trend stationary in both levels and first difference. The two series, GDP and Export are clearly stationary in the first difference of their logarithms. Thus we conclude that the maximum order of integration of the series in the system is I(1), that is, the time series of the system under study are integrated of order d such that 0[less than or equal to] d [less than or equal to] 1, although they may not be of the same order of integration (FDI is likely I(0)). This provides further justification for the use of the bounds testing ARDL approach in this study.

Determination of the optimal Lag order in the UECM

Before estimating equations (1) to (3) we determined the optimal order of lag for the first-differenced variables using all popular criteria allowing maximum lag length of 3. The results are shown in Table 2 below.

Thus lag order 1 is selected by all criteria except Schwarz Information criterion. We accept the verdict of the overwhelming majority and determine the optimal lag length as p= 1 for equations (1) to (3)

The Results of Bound tests of Wald (F) Statistics

Using lag order of one for the first differenced variables, equations (1) to (3) were estimated using OLS and Wald (F) test was performed for the set of hypotheses (h1) to (h3) for the entire sample period of 1971-2008 as well as for the subperiod 1991-2008. The results are shown in Tables 3 below.

Table 3 shows some interesting results. If we consider the entire sample period, the calculated F values for all equations fall below the lower bounds of critical values indicating lack of cointegrating or long-run relationship in all cases regardless of which variable is treated as the dependent variable. On the other hand, if we focus only on the subperiod 1991-2008, that is the post liberalization period, the F value with log of GDP as the dependent variable is well above the upper bound even for 1% level test. The very large value of calculated F leaves no doubt that there is strong evidence of long-run relationship if the post-liberalization period is considered and GDP is treated as the dependent variable. However, in the case of Export and FDI as the dependent variables, the calculated F exceeds the upper bound only at 10% level even during the post-liberalization period. Thus, there is some evidence of long-run cointegrating relation with Export or FDI as the dependent variable, but it is not highly significant according to the bounds test. In short, we conclude from this step of the ARDL bounds testing methodology that there is strong evidence of long-run equilibrating relationship between GDP as the dependent variable and Exports and FDI as the regressors, after the avalanche of the liberalization efforts in India. Therefore, in our subsequent analysis we will focus on the 1991-2008 period and treat GDP (or its logarithm) as the dependent variable. The existence of a cointegrating relationship suggests that there must be Granger causality in at least one direction, but it does not clearly specify the direction of temporal causality (although from the results so far it seems very plausible to expect it from Exports and FDI towards GDP).

Estimation of the ARDL (p, q, r) Model- the Log-run and the Short-run Dynamics

For the long-run ARDL(p,q,r) model we estimated 18 equations with all possible combinations of the lag orders up to order 2 following the recommendation of Pesaran et al. (2001) for annual data. Akaike criterion selected ARDL (2, 2, 0), while Schwarz criterion selected ARDL (1, 0, 0). Following the recommendation of Pesaran et al. (2001) on this point and considering the small sample size during the post-liberalization period, we accepted ARDL (1, 0, 0) for estimating equation (4). The results were used to calculate the error correction term (ECT) to be used in the estimation of equation (5) for the short-run dynamics. We applied the Schwarz criterion again to establish the optimal lag order for Equation (5) with the lagged error correction term. This time the selected model was ARDL (1,1,1) and the results are shown in Table 4 below.

The very high significance of the error correction term clearly shows long-term Granger causality (Bannerjee, et al. 1998). The sign is correct showing movement towards equilibrium following shocks and the speed of adjustment is very high, showing a very fast adjustment (within a fraction of a year towards equilibrium). Export is a highly significant positive determinant of GDP. The F-tests on the joint deletion of the corresponding coefficients show strong evidences of short-run Granger causality from Export towards GDP, but fail to show the same from FDI towards GDP. The FDI does not seem to have significant short-run impact on GDP. The model passes all the usual diagnostic tests, which are not reported here. We will however, discuss the tests of stability of the parameters in the next subsection.

Examination Of Parameter Constancy Of the Cointegration Space

In order to test for the stability of the estimated parameters, Pesaran and Pesaran (1997) suggest applying the cumulative sum of recursive residuals (CUSUM) and the CUSUM of squared residuals (CUSUMSQ) proposed by Brown et al. (1975). The results displayed in Diagrams 1 and 2 below indicate no instability in the coefficients as the plot of CUSUM and CUSUMSQ are confined well within the 5% bounds of parameter stability.

[FIGURE 1 OMITTED]

[GRAPHIC OMITTED]

CONCLUSIONS

Using the recently introduced and more suitable ARDL bounds testing approach this study finds that the post liberalization period in India exhibits significantly different characteristics than the pre- liberalization period. If both periods are combined, there is lack of evidence for long-run cointegration. In the post liberalization period, however, there is strong evidence of long-run relationship with GDP as the dependent variable. The analysis of error correction model shows that Export is a highly significant determining factor for explaining changes in GDP. Moreover Export has short-run as well as long run Granger causality towards GDP. In contrast FDI does not show short-run Granger causality towards GDP.

Thus there is strong support for Export-led-Growth hypothesis in the post-liberalization India, while the inclusion of pre-liberalization period weakens this evidence. Several earlier studies claim lack of evidence to support ELG hypothesis. The conclusion of this paper is that those earlier findings may be the results of pretest biases introduced by the methodologies used and/or the underrepresentation of the post-liberalization period in the sample. Finally it is also found that the speed of adjustment following a shock is very fast.

REFERENCES

Ahmed, Q. M., Butt, M.S. Butt & S. Alam (2000). Economic Growth, Export, and External Debt Causality: The Case of Asian Countries. Pakistan Development Review, 39(4), 591-608.

Alfaro, L. & A. Rodriguez-Clare (2004). Multinationals and Linkages: Evidence from Latin America. Economica, 4, 113-170.

Bahmani-O.skooee M. & J. Alse (1993). Export growth and economic growth: An application of cointegration and error-correction Modeling. Journal of Developing Areas, 27, 535-542.

Bahmani-Oskooee, M. & R. Ng (2002). Long-Run Demand for Money in Hong Kong: An Application of the ARDL Model. International Journal of Business and Economics, 1, 147-155.

Bannerjee, A., J.J. Dolado & R. Mestre (1998). Error-correction mechanism tests for cointegration in single equation framework. Journal of Time Series Analysis, 19, 267-283.

Ben-David, D. & M. Loewy (1998). Free-trade, growth, and convergence. Journal of Economic Growth, 3, 143-70. Bhat, S. K. (1995). Export and Economic Growth in India. Artha Vijnana, 37(4), 350-58.

Blomstrom, M , & A. Kokko (1998). Multinational Corporations and Spillovers. Journal of Economic Surveys, 12, 247-77.

Brown, R. L., J. Durbin & J. M. Evans (1975). Techniques for testing the constancy of regression relations over time. Journal of the Royal Statistics Society, 37, 149-192.

Buffie, E.F. (1992). On the condition for export-led growth. Canadian Journal of Economics, 25, 211-25.

Chenery, H. & A. Strout (1966). Foreign assistance and economic development. American Economic Review, 56(4), 679-733.

Dhawan, U. & B. Biswal (1999). Reexaminining Export-led-Growth Hypothesis: a Multivariate cointegration analysis for India. Applied Economics, 31, 525-30

Dhananjayan, R.S. & N.S. Devi (1997). Exports and economic growth: a study of selected nations in Asia and Europe during 1980-81 to 1993-94. Indian Journal of Applied Economics, 6, 41-63.

Dickey, D.A. & Fuller, W.A. Fuller (1981). Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root. Econometrica, 49(4), 1057--1072.

Duttaray, M., A.K. Dutt &, K. Mukhopadhyay (2008). Foreign direct investment and economic growth in less developed countries: an empirical study of causality and mechanisms. Applied Economics, 40, 1927-39.

Engle, R.F. & C.W.J. Granger (1987). Cointegration and Error Correction: Representation, Estimation and Testing. Econometrica, 55, 251-76

Findlay, C. & A. Watson (1996). Economic growth and trade dependency in China. DP #96/5, Chinese Economies Research Centre, University of Adelaide.

Giles, J. A. & C. L. Williams (2000a). Export-led growth: a survey of the empirical literature and some non causality results Part 1. Journal of International Trade & Economic Development, 9, 261-337.

Giles, J. A. & C. L. Williams (2000b). Export-led growth: a survey of the empirical literature and some non causality results. Part 2. Journal of International Trade & Economic Development, 9, 445-470.

Gorg, H. & D. Greenaway (2004). Much Ado about Nothing? Do Domestic Firms Really Benefit from Foreign Direct Investment?. World Bank Research Observer, 19(2), 171-97.

Granger, C. W. J. (1986). Developments in the Study of the Cointegrated Economic Variables. Oxford Bulletin of Economics and Statistics, 48, 201-212.

Granger, C. (1988). Some Recent Developments in a Concept of Causality. Journal of Econometrics, 39, 199-211.

Granger, C. & P. Newbold (1974). Spurious Regressions in Econometrics. Journal of Econometrics, 2, 111-20.

Granger, C. W. J. & A. A. Weiss (1983). Time Series Analysis of Error Correction Models. In S. Karlin, T. Amemiya & L.A. Goodman (Eds.), Studies in Econometrics: Time Series and Multivariate Statistics. New York (Academic Press), 255-278.

Greenaway, D. & D. Sapsford (1994). Exports, growth, and liberalization: an evaluation. Journal of Policy Modeling, 16, 165-86.

Guru-Gharana, K. K. & D. R. Adhikari (2010). Econometric investigation of relationships among Export, FDI and Growth in China: an application of Toda-Yamamoto-Dolado-Lutkephol Granger causality test. Journal of International Business Research (forthcoming), Allied Academies.

Jaffee, D. (1985). Export dependence and economic growth: a reformulation and respecification. Social Forces, 64, 102-18.

Jayaraman, T.K. & B. Singh (2007). Foreign direct Investment Creation in Pacific Island Countries: an empirical study of Fiji. Asia- Pacific Research and Training Network on Trade Working Paper Series, 35.

Johansen, S. (1988). Statistical analysis of cointegrating vectors. Journal of Economic Dynamics and Control, 12, 231-254.

Johansen, S. (1991). Estimation and Hypothesis testing of Cointegration Vectors in Gaussian Vector Autoregressive Models. Econometrica, 59, 1551-1580.

Johansen, S. & K. Juselius (1990). Maximum Likelihood Estimation and Inferences on Cointegration-with applications to the demand for money. Oxford Bulletin of Economics and Statistics, 52, 169-210.

Jung, W.S. & P.J. Marshall (1985). Exports growth and causality in developing countries. Journal of Development Economics, 18, 1-12.

Kalirajan, K., A. K. Miankhel & S. M. Thangavelu (2009). Foreign Direct Investment, Exports, and Economic Growth in Selected Emerging Countries: Multivariate VAR Analysis. Working Paper at SSRN: http://ssrn. com/ abstract=1526387

Keesing, D.B. (1967). Outward-looking policies and economic development. Economic Journal, 77, 303-20. Krueger, A.O. (1995). Trade Policies and Developing Nations. Washington: Brookings Institution.

Lal, D. & S. Rajapatirana (1987). Foreign trade regimes and economic growth in developing countries. World Bank Research Observer, 2, 189-217.

Love, J. & R. Chandra (2005a). Testing export-led growth in South Asia. Journal of Economic Studies, 32, 132-145.

Love, J. & R. Chandra (2005b). Testing export-led growth in Bangladesh in a multivariate VAR framework. Journal of Asian Economics, 15, 1155-1168.

Lutkephol, H. (2004). Vector Autoregressive and Vector error correction Models. In Lutkephol, H. and Kratzig, M. (Ed.) Applied time series econometrics, Cambridge University Press, 86-158

MacKinnon, J. (1996). Numerical Distribution Functions for the Unit Root and Cointegration Tests. Journal of Applied Econometrics, 11, 601-618.

Mah, J. S. (2000). An Empirical Examination of the Disaggregate Import Demand of Korea--the case information technology products. Journal of Asian Economics, 11, 237-44.

Makki, S. S. & A. Somwaru (2004). Impact of foreign direct investment and trade on economic growth: Evidence from developing countries. American Journal of Agricultural Economics, 86(3), 795-801.

Medina-Smith, E. (2001). Is the export-led growth hypothesis valid for developing countries? a case study of Costa Rica. Paper No.7, UNCTAD/ITCD/TAB/8, UNCTAD, Geneva.

Nandi, S. & B. Biswas (1991). Exports and economic growth in India: empirical evidence. Indian Economic Journal, 38, 53-9.

Narayan, P.K. (2005). The Saving and Investment Nexus for China: Evidence from Cointegration tests. Applied Economics, 37, 1979-1990.

Narayan, P. K. & S. Narayan (2005). Estimating Income and Price Elasticities of Imports for Fiji in a Cointegration Framework. Economic Modeling. 22, 423-438.

Narayan, P. & R. Smyth (2004). Temporal Causality and the Dynamics of Exports, Human Capital and Real Income in China. International Journal of Applied Economics, 1(1), 24-45

Narayan, P. & R. Smyth (2005). Trade Liberalization and Economic Growth in Fiji: An Empirical Assessment Using the ARDL Approach. Journal of the Asia Pacific Economy, 10(1), 96-115.

Pack, H. (1988). Industrialization and trade. in Chenery, H. and T.N. Srinivasan(Eds.) Handbook of Development Economics, 1, Amsterdam: Elsevier.

Pesaran, M.& B. Pesaran (1997), Microfit 4.0: Interactive Econometric Analysis. Oxford: Oxford University Press.

Pesaran, M. H. & Y. Shin (1995). Autoregressive distributed lag modeling approach to cointegration analysis. DAE Working Paper Series, No. 9514, Department of Economics, University of Cambridge.

Pesaran, M. H. & Y. Shin (1998). An Autoregressive distributed lag modeling approach to cointegration analysis. In Storm, S. (Ed.) Econometrics and Economic Theory in the 20th Century: The Ragnar Frisch Centennial Symposium, Cambridge University Press. 1-31.

Pesaran, M.& R. Smith (1998). Structural Analysis of Cointegration, VARs. Journal of Economic Survey, 12, 471-505.

Pesaran, M. H., Y. Shin and R. Smith (1996). Testing for the existence of a long-run relationship. DAE Working Papers 9622, Department of Applied Economics, University of Cambridge.

Pesaran, M. H., Y. Shin & R. J. Smith (2001). Bounds Testing Approach to The Analysis of Level Relationship. Journal of Applied Econometrics, 16, 289-326.

Phillips, P. (1986). Understanding spurious regressions in econometrics. Journal of Econometrics, 33(3), 311-40

Rambaldi, A.N. & H.E Doran (1996). Testing for Granger Non-causality in Cointegrated Systems Made Easy. Working Paper in Econometrics and Applied Statistics, 88, University of New England.

Rahman, M. (2009). Contributions of Exports, FDI and Expatriates' Remittances to Real GDP of Bangladesh, India, Pakistan and Sri lanka. Southwestern Economic Review, 36(1), 141-54

Reserve Bank of India (online). Database on Indian Economy. ( https://reservebank.org.in/cdbmsi/servlet/login/).

Riezman, R.G., P.M. Summers & C.H. Whiteman (1996). The engine of growth or its handmaiden? A time series assessment of export-led growth. Empirical Economics, 21, 77-113.

Sahoo, D. & M.K. Mathiyazhagan (2003). Economic Growth in India: Does Foreign Direct Investment Inflow Matter?. Singapore Economic Review, 48(2), 151-171.

Shirazi, N.S. and T.A. Abdul-Manap (2005). Export-led growth hypothesis: further econometric evidence from South Asia. The Developing Economies. 63(4), 472-88.

Thangavelu, S.M. & G. Rajaguru (2004). Is there an export or import-led productivity growth in rapidly developing Asian countries? A Multivariate VAR analysis. Applied Economics, 36, 1083-1093.

Toda, H.Y. & P.C.B. Phillips (1993). Vector autoregressions and causality. Econometrica. 61, 1367-93.

Toda, H.Y. and P.C.B. Phillips (1994). Vector autoregression and causality: a theoretical overview and simulation study. Econometric Reviews. 13, 259-85.

Toda, H. Y. & T. Yamamoto (1995). Statistical inference in vector autoregression with possibly integrated processes. Journal of Econometrics. 66, 225-250.

World Development Indicators (2009), World Bank

Xu, Z. (1996). On the causality between export growth and GDP growth: an empirical reinvestigation. Review of International Economics. 4, 172-84.

Zapata, H.O. and, A.N. Rambaldi (1997). Monte Carlo evidence on cointegration and causation. Oxford Bulletin of Economics and Statistics. 59, 285-98.

Kishor K. Guru-Gharana, Texas A & M University-Commerce

Table 1: Augmented Dickey-Fuller (ADF) Test Results (Max lag 12)

Variable No Intercept or Trend

Log(GDP) 5.845 (1.000)
Log(Export) 7.168 (1.000)
Log(FDI) -1.772 * (0.073)
[DELTA]Log(GDP) -2.884 *** (0.005)
[DELTA]Log(Export) -1.728 * (0.08)
[DELTA]Log(FDI) -0.242 (0.589)

Test critical values 1% -2.629
 5% -1.950
 10% -1.611

Variable Intercept

Log(GDP) -0.006 (0.952)
Log(Export) 0.779 (0.992)
Log(FDI) -1.638 (0.453)
[DELTA]Log(GDP) -4.467 *** (0.001)
[DELTA]Log(Export) -3.623 *** (0.010)
[DELTA]Log(FDI) -3.538 ** (0.015)

Test critical values 1% -3.621
 5% -2.943
 10% -2.610

Variable Intercept and Trend

Log(GDP) -1.307 (0.871)
Log(Export) -0.511 (0.979)
Log(FDI) -4.083 ** (0.015)
[DELTA]Log(GDP) -4.400 *** (0.007)
[DELTA]Log(Export) -3.619 ** (0.042)
[DELTA]Log(FDI) -3.686 ** (0.042)

Test critical values 1% -4.227
 5% -3.537
 10% -3.200

MacKinnon (1996) one-sided p-values are shown inside parentheses.
Values are rounded to three decimal places. Significance at 10% if one
*, significance at 5% if two **, and significance at 1% if three ***

Table 2: VAR Lag order Selection Criteria and Results
Endogenous Variables: [DELTA]G [DELTA]E [DELTA]F; Exogenous:
C G(-1) E(-1) F(-1)
Sample: 1971-2008; included observations: 34

Lag LR (1) FPE (2) AIC (3) SC (4) HQ (5)

0 NA 6.30e-05 -1.162752 -0.624036 * -0.979034
1 18.81314 * 5.40e-05 * -1.330123 * -0.387371 -1.008617 *
2 7.684302 0.000069 -1.120890 0.225898 -0.661597
3 10.32361 0.0000768 -1.083079 0.667746 -0.485998

* indicates lag order selected by the criterion

(1.) LR: sequential modified LR test statistic (each test at 5% level)

(2.) FPE: Final Prediction Error

(3.) AIC: Akaike Information Criterion

(4.) SC: Schwarz Information Criterion

(5.) Hannan-Quinn Information Criterion

Table 3: Calculated F-Statistics for Bounds Test (Lag order
1); Intercept but no Trend

Hypotheses Calculated F Calculated F
 (1971-2008) (1991-2008)

[F.sub.g](G | E, F) 2.3304 8.3804 *
[F.sub.e](E | G, F) 1.3340 3.6769
[F.sub.f](F | G, E) 2.7703 3.9559

Critical values are calculated by Narayan and Smith (2004), Table 2,
using stochastic simulations for T=40 and two regressors based on
40,000 replications. They are [2.835 3.585] for 10% and [3.435 4.260]
for 5%. But Narayan and Smith (2005) Table 4 calculate with T= 34 and
40, 000 replications as [3.990 4.538] for 5% and [4.943 6.128] for 1%.
The first values inside brackets are for I(0) and the second for I(1).
Corresponding values in Pesaran, et al (2001) Table CI (iii), Case III
are very slightly different from these values. Considering our small
sample size we will use the values reported by Narayan and Smith (2004
and 2005). 3.478 and 4.335 for 5%?

Table 4: Dependent Variable [DELTA]G(1,1,1); Sample (1991-2008)

Variable coefficient t-statistic Prob.

C 0.0086 0.3786 0.7129
[DELTA]G(-1) 0.6474 2.0946 0.0626
[DELTA]E 0.7249 3.2731 0.0084
[DELTA]E(-1) -0.3575 -2.1069 0.0614
[DELTA]F 0.0237 0.5608 0.5873
[DELTA]F(-1) -0.0740 -1.6966 0.1206
ECT(-1) -2.0592 -4.5821 0.0010
R-squared 0.788410 Mean dependent var 0.089138
Adjusted R-squared 0.661457 S.D. dependent var 0.077306
S.E. of regression 0.044980 Akaike info criterion -3.072287
Sum squared resid 0.020232 Schwarz criterion -2.729200
Log likelihood 33.11444 Hannan-Quinn criter. -3.038184
F-statistic 6.210218 Durbin-Watson stat 2.110210
Prob(F-statistic) 0.006050