This study aims to analyse the relationship between the gross domestic product (GDP) and the foreign direct investment (FDI) in the BRIC-T countries (Brazil, Russia, India, China and Turkey) using panel unit root tests and panel cointegration tests as FMOLS and DOLS estimators and the Granger causality test and Dumitrescu–Hurlin panel causality tests over the period from 1992 to 2013.
The findings of this paper show that the coefficient of the FMOLS estimator is 0.25, the coefficient of the DOLS estimator is 0.26 and both forecasted results are positive and reasonable at the 1% level of statistical significance. The direction of causality for these countries in the panel Granger causality tests is found to be GDP → FDI, while a GDP ↔ FDI homogeneous cause is found in the pairwise Dumitrescu–Hurlin panel causality tests. These test results indicate that the presence of FDI in the economic growth of the BRIC-T countries is very important and therefore these countries have become highly dependent on foreign-capital-exporting countries.
Keywords: FDI, Panel Data, Cointegration, Causality, BRIC-T
Introduction
Beginning in the early 1980s and especially after the fall of the Berlin Wall, the globalization of the financial sector, the abatement of controls on cross-border financial assets and the drive to open capital accounts indicated a rooted break with the post-war international policy framework, with capital flows floating better financing of international trade. Afterwards, the financial sector increased across all countries and regions. In this context the financial leverage (the ratio of debt to revenue) rose sharply and the proliferation of financial products and markets was supported by the international market makers. Thus, “shadow” financial institutions emerged with heightened speculative behaviour, and all individuals and institutions increasingly focused on quick returns from speculation on financial assets, exchange rates, real estate, and mergers and acquisitions, often fuelling asset price bubbles. This expansion in the world’s financial structure increased its command over global resources and tightened its grip on both corporate governance and policymaking (UNCTAD, 2012). Against this background the measures of economic “success” became disconnected from the pressures of making productive investments, raising productivity and creating jobs. In combination with a general shift to more open markets, this structural shift in economic activities and the changes in economic actors can be described as finance-driven globalization (FDG). In this setting transnational corporations (TNCs) or developed countries’ firms play a crucial role via FDI in finance-driven emerging markets’ economies. Many developing countries, especially those defined as emerging market economies (Brazil, Russia, India, China and Turkey), are developing successfully by opening up their economies following outward-oriented policies, albeit to varying degrees.
Thus, from 1980 to 1990, all types of private sector capital flows were limited, with an average total of only $36 billion annually, and FDI was important in percentage terms, with 53% of the total private sector capital flows, respectively. However, from 1991 to 2002, a cyclical upswing in capital flows began, with an annual average flow of $220 billion, more than 4.5 times the average level in the first period. This upswing was driven primarily by FDI and bank lending. The increasing flows of FDI caused direct investment to become the most important factor, accounting for 56% of the total private sector capital flow. Critically, FDI was also a relatively stable source of capital for developing economies from 2003 to 2007; as financialization peaked in developed countries, there were further sharp upswings in the inflows, with FDI growing from $299 billion in 2002 to $889 billion in 2007 and financial flows rising from a negative $30 billion in 2002 to a positive $1.706 trillion in 2007. Indeed, the strength of the FDI flows from the private sector was important for the rapid growth in the economies that received such investment, including China and India. Financial flows were also buoyant, as liberalized capital accounts facilitated inflows through bank lending. However, by the global financial crisis that started in 2007 and during the acute phase of the crisis, from approximately 2007 to 2009, there was a sharp contraction in inflows and a severe rise in their costs as strong risk aversion swept through the global financial markets. After this acute phase receded, the global cross-border capital flows recovered but with high levels of volatility, including those due to repeated shock events, such as the eurozone crisis and the unprecedented monetary easing in advanced economies, especially the United States. These trends were reflected in the cross-border capital flows to developing countries. The financial crisis disrupted these patterns. FDI responded relatively little to the crisis, in contrast to other flows, with a peak in 2007 of $445 billion, a slight increase in 2008 to $472 billion and a sharp decline in 2009 to $330 billion, before returning to and then exceeding the pre-crisis level from 2010 to 2013 and exceeding the pre-crisis level after 2013 (Tyson and McKinley 2014).
The purpose of this study is to measure the relationship between the GDP and the FDI in the BRIC-T countries by conducting a panel data analysis. First, we will apply the cointegration tests, after which we will implement the FMOLS developed by Pedroni (2000) to gain inferences about the BRIC-T countries’ long-term relationship between GDP and FDI variables. The FMOLS and DOLS methods do not indicate the direction of causality. For this reason we perform the Granger causality test and Dumitrescu–Hurlin panel causality tests to examine the causal relationship between the foreign direct investment (FDI) and the real GDP.
2. Literature Review
There is a widespread belief among academics and policymakers that foreign direct investment (FDI) generates positive productivity effects for developing countries. According to this belief, FDI is a way to transfer technology and capital from other developing and especially developed countries. The main mechanisms for these positive productivity effects are the adoption of owned technology and its dissemination formats – for example know-how, which can be gained via licensing agreements, imitation and employee training – of developed countries, the introduction of new processes and products by firms of developed countries and the creation of linkages between developing and developed countries’ firms. The results of these main mechanisms occur together with direct capital financing and suggest that FDI can play an important role in modernizing a developing country’s economy and promoting its economic development. Although the theoretical manuscripts on FDI point to advantages, the empirical evidence on the existence of spillovers could nevertheless be small.
The macro-field of empirical literature finds weak support for an exogenous positive effect of FDI on economic growth, while the micro-field of empirical literature finds ambiguous results regarding the effect of FDI on firms’ productivity. For example, according to Carkovic and Levine (2002), FDI inflows do not exert an independent influence on economic growth. Thus, while sound economic policies may spur both growth and FDI, their results are inconsistent with the view that FDI exerts a positive impact on growth that is independent of other growth determinants. Alfaro, Chanda, Kalemli-Ozcan and Sayek (2006) show that an increase in FDI leads to higher growth rates in financially developed countries than those observed in financially poorly developed ones. They find larger growth effects when the goods produced by domestic firms and MNEs are substitutes rather than complements. On the other hand, Borensztein, De Gregorio and Lee (1998) and Xu (2000) show that FDI brings technology, which translates into higher growth only when the host country has a minimum threshold of the stock of human capital. Romer (1993) argues that important idea gaps between rich and poor countries exist and accordingly that foreign investment can ease the transfer of technological business know-how to poorer countries.
Lamine and Jang (2010) shed light on the impact of foreign direct investment (FDI) in developing the Guinea Republic on the economic growth over the 1985–2008 period. According to the study’s results, the level of FDI is still too low to promote economic growth for the Guinea Republic. Thus, the Granger causality test demonstrates that the GDP can promote the level of FDI, which means that, if the level of GDP increases in Guinea, FDI will follow. Some other factors, such as employment, can promote FDI; thus, the Guinean Government has to play the key role of employment promotion to attract investments from abroad. This study also finds that school enrolment can increase the GDP and indirectly the FDI. Hermes and Lensink (2003) claim that the development of the financial system in the recipient countries is an important precondition for FDI to have a positive impact on economic growth. They argue that a more developed financial system contributes positively to the process of technological diffusion associated with FDI. According to this paper’s empirical investigation, the development of the financial system plays a role in increasing the positive relationship between economic growth and FDI. Of the 67 countries in the data set, 37 have a sufficiently developed financial system to enable FDI to contribute positively to economic growth, and most of them are Latin American and Asian countries. Haskel, Pereira and Slaughter (2002) find positive spillovers from foreign to local firms; they use a plant-level panel covering U.K. manufacturing from 1973 to 1992. Durham (2004) investigates the effects of FDI and equity foreign portfolio investment (EFPI) on economic growth using data on 80 countries from 1979 to 1998. The results of this study largely propose that lagged FDI and EFPI do not have direct, unmitigated positive effects on growth.
Aitken and Harrison (1999) do not find any evidence of a beneficial spillover effect between foreign firms and domestic ones in Venezuela over the 1979–1989 period. Similarly, Haddad and Harrison (1993) and Mansfield and Romeo (1980) find no positive effect of FDI on the rate of economic growth in developing countries, namely in Morocco. Blomstrom et al. (1994) also show that the positive growth effect of FDI may be real when the country is insufficiently rich.
The micro-field of empirical literature finds that FDI does not encourage economic growth in developing countries; especially, according to Carkovic and Levine (2002), firm-level studies of particular countries often find that FDI inflows do not exert an independent influence on economic growth. Thus, while sound economic policies may spur both growth and FDI, Aitken and Harrison’s (1999) study related to Venezuela between 1979 and 1989 tests for spillovers from joint ventures to plants with no foreign investment. Foreign investment negatively affects the productivity of domestically owned plants. The net impact of foreign investment, considering these two offsetting effects, is quite small. FDI gains appear to be captured entirely by joint ventures. Other authors, such as Germidis (1977), Haddad and Aitken (1993) and Mansfield and Romeo (1980), obtain similar results. Briefly, the micro-field of empirical literature for developing countries does not lend much support to the view that FDI increases economic growth as a whole.
In recent research in this field, for example, Iamsiraroj (2016) investigates the linkage between FDI inflows and per capita income growth using the simultaneous system of equations approach with 124 cross-country data for the period 1971–2010, and the results from the estimation indicate that the overall effects of FDI are positively associated with growth and vice versa. Zekarias (2016) analyses the relationship between GDP and FDI for 14 Eastern African countries by employing 34 years (1980–2013) of panel data and finds that FDI has a positive and marginally significant effect on economic growth, while Agrawal (2015) investigates foreign direct investment and economic growth in the BRICS economies by conducting a panel data analysis and finds bidirectional causality between FDI and economic growth and a bidirectional positive correlation between growth and foreign investments.
3. Data Set, Econometric Methods and Evaluation of the Results
Data Set and Econometric Methods
A panel data analysis is applied to the 1992–2013 period covering the BRIC-T countries (Brazil, Russia, India, China and Turkey). The analysis of the gross domestic product (GDP) and foreign direct investment (FDI) uses the current numerical quantities from the World Bank database denominated in USD. All variables were measured in logarithms. The long-run relationship between GDP and, FDI for BRIC-T is tested using the following linear logarithmic functional form:
〖GDP〗_it=α_it+β_it 〖FDI〗_it+u_it (1)
First, we establish that both the GDP and the FDI have unit roots. Secondly, we look for cointegration using residual-based tests in the panel developed by Johansen’s (1988) Fisher panel Cointegration test. Pedroni (1999) -Engle-Granger based-, Kao(1999) -Engle-Granger based-. Finally, to estimate the ultimate drift coefficients, the FMOLS (full modified ordinary least square) method developed by Pedroni (2000) and the DOLS (dynamic ordinary least square) method developed by Pedroni (2001) are used.
Panel Unit Root Tests
In this paper we apply different seven-panel unit root tests, which include common and individual unit root tests, to the GDP and FDI flow series of the BRIC-T countries. These tests are performed in the studies by Breitung (2000), Im et al. (2003) and Levin et al. (2002), and the Fisher–ADF and Fisher–PP-type unit root tests for panel data are used by Choi (2001), Hadri (1999) and Maddala and Wu (1999). These tests are called first generation tests and if the probability value is smaller than 0.10, the panel data sets are considered to be stable as the common feature of these tests. The appropriate length of delay that addresses the problem of autocorrelation between errors is selected according to the Schwarz information criterion.
Table 1 shows the first-generation unit root test statistics and the results of the probability values applied to the data on the annual gross domestic product (GDP) and annual foreign direct investment flows (FDI) of the BRIC-T countries in the period 1992–2013. Looking at the panel unit root test results, the GDP variable is included in the unit root without a trend in the level of values, which appears to be a stable trend in the level of values; however, in their first difference, both a stagnant trend model and a stagnant without-trend model are observed. Considering the foreign direct investment variable of the panel unit root test, the BRIC-T countries show the same results.
Table 1. Panel Unit Root Test Results
Unit Root Tests of the GDP of the BRIC-T Countries – The Level and First Differences (LNBRAZILGDP, LNRUSIAGDP, LNINDIAGDP, LNCHINAGDP and LNTURKEYGDP)
Method Test statistic I(0) Test statistic I(1)
Without trend With trend Without trend With trend
Levin et al. t* 0.26(0.60) -1.82(0.03)** -5.08(0.00)*** -4.32(0.00)***
Breitung t-stat. 1.04(0.85) -2.93(0.00)***
Im et al. W-stat 3.61(0.99) -2.89(0.00)*** -4.93(0.00)*** -3.26(0.00)***
ADF–Fisher chi-square 0.64(1.00) 25.49(0.00)*** 41.86(0.00)*** 28.01(0.00)***
PP–Fisher chi-square 1.15(0.99) 11.67(0.30) 41.59(0.00)*** 27.63(0.00)***
Hadri Z-stat. 7.25(0.00)*** 3.13(0.00)*** 1.76(0.03)** 4.13(0.00)***
Heteroscedastic- consistent Z-stat. 7.25(0.00)*** 2.41(0.00)*** -0.07(0.53) 2.40(0.00)***
Unit Root Tests of Foreign Direct Investment of the BRIC-T Countries – The Level and First Differences (LNBRAZILFDI, LNRUSIAFDI, LNINDIAFDI, LNCHINAFDI and LNTURKEYFDI)
Method Test statistic I(0) Test statistic I(1)
Without trend With trend Without trend With trend
Levin et al. t* -2.81(0.00)*** -3.54(0.00)*** -9.32(0.00)*** -7.01(0.00)***
Breitung t-stat. -0.67(0.24) -2.54(0.00)***
Im et al. W-stat. -1.13(0.12) -4.66(0.00)*** -8.99(0.00)*** -7.16(0.00)***
ADF–Fisher chi-square 17.70(0.06)* 39.51(0.00)*** 78.44(0.00)*** 56.54(0.00)***
PP–Fisher chi-square 15.35(0.11) 24.39(0.00)*** 308.57(0.00)*** 212.27(0.00)***
Hadri Z-stat. 6.33(0.00)*** 1.26(0.10) -0.43(0.00)*** 0.89(0.18)
Heteroscedastic-consistent Z-stat. 6.33(0.00)*** 1.01(0.15) 0.08(0.46) 1.43(0.07)*
Note: The probability values are in brackets. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels.
Panel Cointegration Tests
After looking at the unit root tests, in the second step, the parameters of the long-run relationship are estimated. In this stage the existence of a long-term relationship between the gross domestic product (GDP) and the foreign direct investment flows (FDI) is analysed using the Johansen–Fisher, Kao (1999) and Pedroni (1999) cointegration tests.
Pedroni (1995) establishes the first residual-based panel cointegration tests. Furthermore, Pedroni (1999, 2004) expands his panel cointegration testing procedure to the case of multiple regressors. In this context Pedroni (1999, 2004) suggests seven different residual-based panel cointegration tests for testing the null hypothesis of no cointegration. Four of these tests is known as panel-v, panel-ρ, semi-parametric panel-t and parametric panel-t, are within-dimension-based statistics. These tests are calculated by summing up the numerator and the denominator over N cross-sections separately. Three other tests is known as group-ρ, semi-parametric group-t and parametric group-t, are between-dimension-based statistics calculated by dividing the numerator and the denominator before summing up over N cross-sections.
Pedron’s( 1999, 2004) the panel cointegration tests are started as the computation of the residuals of the hypothesized cointegrating regression,
y_it=α_i+δ_i t+β_i X_it+e_(i,t) ; i =1,…, N; t =1,…,T (1)
in which N denotes the number of individuals and t is the number of observations over time in the panel. y_it and the K-dimensional vector of independent variables; X_it=x_(i,t-1)+v_it are assumed to be at most I(1). The cointegrating vector βi = (β1i , . . . , βKi)’, the individual-specific intercept〖 α〗_i and the trend parameter δ_i can vary over cross-sections, and these parameters allow for the possibility of member specific fixed effects and deterministic trends, respectively.
Equation 1 is estimated by the ordinary least squares (OLS) method, separately for each cross-section as the cointegrating regression. Additionally, parametric panel-t and the panel-v statistics are calculated using the following first-differenced regression equation, which is obtained by ignoring the deterministic terms: Using the residuals from the differenced regression
∆y_it=b_1i ∆x_1it+b_2i ∆x_2it+⋯+b_Ki ∆x_Kit+ (2)
In these equations, the properties of tests for the null hypothesis; H0: “all of the individuals of the panel are not cointegrated.” For the alternative hypothesis, H1: “all of the individuals are cointegrated.”
Kao (1999) has investigated that methods for testing cointegration of panel data. These methods are more appropriate and powerful than those used by Coe and Helpman (1995). Thus, he describes two tests under the null hypothesis of no cointegration for panel data: Dickey‐Fuller(DF) test and Augmented Dickey‐Fuller(ADF) test. In this context, these tests hold out parametric residual-based panel tests for the null hypothesis; H0: “all of the individuals of the panel are not cointegrated. The tests are based on the spurious least squares dummy variable (LSDV) panel regression equation and, suppose y_it and x_it, are incorrectly estimated by least squares for all i using panel data; the spurious LSDV regression model is
y_it=α_i+x_it β+e_it ; i =1,…, N; t =1,…,T (3)
The LSDV estimator of β is
β ̂=(∑_(i=1)^N▒∑_(t=1)^T▒〖y_it (x_it-¯(x_i ) )〗)/(∑_(i=1)^N▒∑_(t=1)^T▒〖(x_it-¯(x_i ) )〗^2 )
in which e_it is I(1), let y_it= ∑_(s=1)^t▒u_is and x_it= ∑_(s=1)^t▒_is are restricted to be at most I(1) with u_i0= ε_it= O_p (1).
After all, Kao (1999) has provided an asymptotic theory for the behavior of the LSDV estimator in a model which attempts to estimate the panel regression when the dependent variable and independent variable are actually independent I(1) processes. According to Kao(1999), in particular, the t-statistic diverges in spite of the fact that the LSDV estimator converges to its true value in probability.
The long-run relationship between the GDP and the foreign direct investment of the BRIC-T countries is examined by performing a panel cointegration test. According to Pedroni’s cointegration test, “H_0: There is no cointegration between the series”, the hypothesis is rejected. Three of the panel statistics from the test results in Table 2 are statistically significant according to the different levels of significance. The group-PP statistics and group-ADF statistics from the group statistics are statistically significant at the 5% significance level. When the assessments of Pedroni’s cointegration test are viewed as a whole, the results of five of the seven tests constitute the panel and the group statistics point out a strong cointegration relationship between series.
According to Kao’s cointegration test, “H_0: There is no cointegration between the series”, the hypothesis is not rejected. Following the Johansen–Fisher cointegration test, “H_0: There is no cointegration between the series”, the hypothesis is not rejected and the alternative hypothesis, “H_1: There is cointegration between the series”, is accepted. Thus, the Johansen–Fisher cointegration test points to the existence of a long-run relationship between the GDP and the foreign direct investment in the BRIC-T countries.
Table 2: Panel Cointegration Test Results and Evaluation
〖GDP〗_it=α_it+β_it 〖FDI〗_it+u_it
Pedroni’s Panel Cointegration Test Results
(Within-Dimension)
t- Statistic Prob. Weighted t-Statistic Prob.
Panel v-statistic 0.3667 0.35 0.0695 0.47
Panel rho statistic -3.0856 0.00*** -1.3846 0.08*
Panel PP statistic -3.2975 0.00*** -1.6971 0.04**
Panel ADF statistic -3.2682 0.00*** -1.9181 0.02**
(between dimension)
t- Statistic Prob.
Group rho statistic -0.6196 0.26
Group PP statistic -1.5266 0.06*
Group ADF statistic -2.6087 0.00**
Kao’s Panel Cointegration Test Results
t-statistic Prob.
ADF -0.0533 0.47
Residual variance 0.0007
HAC variance 0.0016
Johansen–Fisher Panel Cointegration Test Results
Hypothesized
no. of CE(s) Fisher stat.*
(from trace test) Prob. Fisher stat.*
(from max-eigen test) Prob.
None 24.05*** 0.0075 26.76*** 0.0028
At most 1 4.168 0.9394 4.168 0.9394
Note: *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. The appropriate lag length is chosen in accordance with the Schwarz information criterion.
3.4 The Findings of the FMOLS (Fully Modified Ordinary Least Squares) Method and the DOLS (Dynamic Ordinary Least Squares) Cointegration Coefficients
The results of the panel cointegration tests support the existence of strong long-run relationships among the model’s variables. Thus, the next step is to apply the FMOLS and DOLS methods. These methods are highlighted by Kao and Chiang (1998, 2000), Pedroni (1996) and Phillips and Moon (1999) and show the differences in the asymptotic statistical properties of the non-stationary panels. These works concern the fully modified OLS (FMOLS) and DOLS panel cointegration methods and extend the field of panel cointegration to the estimation and inference of cointegrated regressions with panel data.
The FMOLS method produces reliable estimates for a small sample size and provides a check of the robustness of the results. The FMOLS method was originally introduced and developed by Philips and Hansen (1990) for estimating a single cointegrating relationship. The FMOLS method of Pedroni (2000) and Phillips and Moon (1999) is the panel analogue of the Philips and Hansen (1990) FMOLS method of the time series literature, and these methods use non-parametric correction bias and decrease the endogeneity problems of the OLS (ordinary least squares) method.
The paper uses the FMOLS and DOLS estimators by Pedroni (2000, 2001). To facilitate the comparison with the conventional time series literature, Pedroni (2000) uses as a starting point a few Monte Carlo simulations analogous to the ones studied by Phillips and Hansen (1990) and Phillips and Loretan (1991) based on their original work on FMOLS estimators for conventional time series. Following these studies, we model the errors for the data-generating process in terms of a vector MA(1) process and consider the consequences of varying certain key parameters. In particular, for the purposes of the Monte Carlo simulations, we model this data-generating process for the cointegrated panel. In this context Pedroni provides the between-dimension “group mean” FMOLS and DOLS estimators. The advantage of using the between estimators is that the form in which the data are pooled allows for greater flexibility in the presence of heterogeneity of the cointegrating vectors. The test statistics derived from the between-dimension estimators are constructed to test the null hypothesis H_0: βi= β0 for all i against the alternative H_1: βi β0, so the values for βi are not constrained to be the same under the alternative hypotheses. Considering the following co-integrated system for a panel of L = 1, 2,…., N members, the FMOLS method is based on the following panel regression model:
y_it=α_i+βx_it+μ_it (4)
x_it=x_(it-1)+ε_it
In the equations y_it: dependent variable, x_it: independent variable; α_i ∶fixed effects and β: long-term cointegration coefficient are represented. i = 1, … , N, t = 1, … , T, for which Pedroni (2000) models the vector error process _it=( _it, ε_it) in terms of a vector moving average process given by
_it= _it-_i _(it-1); _it i.i.d.N(0,_it) (5)
where _i is a 2 × 2 coefficient matrix and _it is a 2 × 2 contemporaneous covariance matrix to accommodate the potentially heterogeneous nature of these dynamics among different members of the panel. For the panel DOLS estimation, the cointegration equation (4) is augmented as follows:
s_it=α_i+β_i p_it+∑_(k=-Ki)^Ki▒〖γ_ik 〖∆p〗_(it-k) 〗+μ_it^* (6)
From this regression we construct the group mean panel DOLS estimator as
β ̂_GD^*=[N^(-1) ∑_(i=1)^N▒〖(∑_(t=1)^T▒z_it z_it^’ )^(-1) (∑_(t=1)^T▒〖z_it s ̃_it 〗) 〗]_1 (7)
where z_it is the 2(K + 1) × 1 vector of regressors z_it=( z_it- (p_i ) ̅,∆p_(it-K),…,∆p_(it+K) ), s ̃_it=s_it-(s_i ) ̅ and the subscript 1 outside the brackets indicates that we are taking only the first element of the vector to obtain the pooled slope coefficient. Again, because the expression following the summation over the i is identical to the conventional time series DOLS estimator, the between-dimension estimator can be constructed simply as β ̂_GD^*=N^(-1) ∑_(i=1)^N▒β ̂_(D,i)^* , where β ̂_(D,i)^* is the conventional DOLS estimator, applied to the ith member of the panel. Similarly, if we let
σ_i^2= 〖lim〗_(r→∞ ) E[T^(-1) 〖(∑_(t=1)^T▒〖μ ̂_it^*) 〗〗^2 ]
be the long-run variance of the residuals from the DOLS regression (which again can be estimated using standard HAC methods), then the associated t-statistic for the between-dimension estimator can be constructed as
t_(β ̂_GD^* )= N^(-1/2) ∑_(i=1)^N▒t_(β ̂_(D,i)^* )
where
t_(β ̂_(D,i)^* )=(β ̂_(D,i)^*-β_0 ) (σ ̂_i^(-2) ∑_(t=1)^T▒(p_it-p ̅_i )^2 )^(1⁄2)
Afterall, According to Pedroni(2001), in most cases, the results of the FMOLS and DOLS are in agreement for the panel tests.
Table 3. The Findings of the FMOLS and DOLS Cointegration Coefficients
GDP_it=α_it+β_it FDI_it+u_it
FMOLS
Coefficient t-Statistic
0.2554 6.7925(0.00)
DOLS
0.2626 6.1686(0.00)
Note: The probability values are in brackets. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels.
The findings of this paper indicate that the coefficient of the FMOLS estimator is 0.25, the coefficient of the DOLS estimator is 0.26 and both forecasted results are positive and reasonable at the 1% level of statistical significance by the Pedroni (2000, 2001) estimators in Table 3. Thus, according to these results, an increase of $100 in foreign direct investment will cause an increase of approximately over $25 in long-term economic growth as the FMOLS estimator, while an increase of $100 in foreign direct investment will cause an increase of approximately over $25 in long-term economic growth as the DOLS estimator in the BRIC-T countries. In this context the findings of this paper confirm the strong and positive relationship between the foreign direct investment (FDI) and the GDP in the BRIC-T countries, which in turn yields a positive impact on economic growth.
3.5) Granger Causality Tests
The FMOLS and DOLS methods do not indicate the direction of causality. For this reason we use the Granger causality test to examine the causal relationship between the foreign direct investment (FDI) and the real GDP. The results in Table 4 show that Granger causalities are present implicitly via the FMOLS and DOLS methods.
The Granger causality test is a profitable analysis to determine whether a causal relationship exists between the current value of the dependent variable and the lagged values of the explanatory variables. In other words, if series X Granger causes series Y, approximately repeated movements are observed in series Y after the movements in series X. There are a number of different approaches to testing for Granger causality in a panel context. In general the bivariate regressions in a panel data context take the form:
y_(i,t)=α_(0,i)+α_(0,i) y_(i,t-1)+⋯+ α_(1,i) y_(i,t-1)+β_(1,i) x_(i,t-1)+⋯+β_(1,i) x_(i,t-1)+ε_(i,t)
x_(i,t)=α_(0,i)+α_(1,i) x_(i,t-1)+⋯+ α_(1,i) x_(i,t-1)+β_(1,i) y_(i,t-1)+⋯+β_(1,i) y_(i,t-1)+ε_(i,t)
where t denotes the time period dimension of the panel and i denotes the cross-sectional dimension.
The different forms of panel causality test differ in the assumptions made about the homogeneity of the coefficients across cross-sections. It offers two of the simplest approaches to causality testing in panels. The first is to treat the panel data as one large stacked set of data and then perform the Granger causality test in the standard way, except for not letting data from one cross-section enter the lagged values of data from the next cross-section. This method assumes that all the coefficients are the same across all the cross-sections. The second approach, adopted by Dumitrescu and Hurlin (2012), makes the extreme opposite assumption, allowing all the coefficients to be different across the cross-sections:
α_(0,i)≠α_(0,i),α_(1,j),…α_(j,i)≠ α_(1,j),∀i,j
β_(1,i)≠β_(1,j),…β_(1,j)≠ β_(1,j),∀i,j
This test is calculated by simply running standard Granger causality regressions for each cross-section individually.
Table 4: Panel Granger Causality Tests
Pairwise Granger Causality Tests
Null Hypothesis: F-Statistic Prob.
FDI_ does not Granger cause GDP_ 0.06588 0.7980
GDP_ does not Granger cause FDI_ 3.11131 0.0807***
Pairwise Dumitrescu–Hurlin Panel Causality Tests
Null Hypothesis: W-Stat. Zbar-Stat. Prob.
FDI_ does not homogeneously cause GDP_ 3.96158 3.61786 0.0003*
GDP_ does not homogeneously cause FDI_ 9.56969 10.7706 0.0003*
Note: (1) The probability values are in brackets. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels; (2) Lags: 1.
According to the pairwise Granger causality test results, we cannot reject the hypothesis that the FDI does not Granger cause the GDP, but we can reject the hypothesis that the GDP does not Granger cause the FDI, while, in the Dumitrescu–Hurlin panel causality tests, we reject the null hypothesis that the FDI does not homogeneously cause the GDP and the GDP does not homogeneously cause the FDI. In this context the direction of causality for these countries in the panel Granger causality tests is found to be GDP → FDI, while a GDP ↔ FDI homogeneous cause is found in the pairwise Dumitrescu–Hurlin panel causality tests.
4. Conclusion
This study aimed to analyse the relationship between the gross domestic product (GDP) and the foreign direct investment (FDI) in the BRIC-T countries (Brazil, Russia, India, China and Turkey) by conducting panel unit root tests and panel cointegration tests as FMOLS and DOLS estimators and the Granger causality test and Dumitrescu–Hurlin panel causality tests over the period from 1992 to 2013.
First, using seven different panel unit root tests, the stability of the variables was investigated. According to the panel unit root test results, the level values of the variables of GDP and FDI are non-stationarity; however, in their first difference, both a stagnant trend model and a stagnant without-trend model are observed.
Second, the results of the panel cointegration tests supported the existence of long-run strong relationships between the gross domestic product (GDP) and the foreign direct investment (FDI) as the model’s variables.
This paper estimated that the coefficient of the FMOLS estimator is 0.25, the coefficient of the DOLS estimator is 0.26 and both forecasted results are positive and reasonable at the 1% level of statistical significance by Pedroni (2000, 2001). According to the FMOLS and DOLS test results, the presence of FDI in the economic growth of the BRIC-T countries is very important and therefore these countries have become highly dependent on foreign-capital-exporting countries.
The direction of causality for these countries in the panel Granger causality tests is found to be GDP → FDI, while a homogeneous GDP ↔ FDI cause is found in the pairwise Dumitrescu–Hurlin panel causality tests.
There is evidence to support the relationship between the gross domestic product (GDP) and the foreign direct investment (FDI) in the BRIC-T countries (Brazil, Russia, India, China and Turkey). Considering these findings, FDI policies aimed at improving the FDI infrastructure and increasing the FDI supply are the appropriate options for these countries, since FDI increases the income level. FDI conservation policies could hamper social and economic progress. As a result, the results highlight the importance of the FDI policy for economic growth, economic development and welfare. The current FDI policy and the FDI restructuring process should be designed to meet this goal for these countries.