A Dynamic Approach to the FDI-Environment Nexus: The Case of China and India

The cointegration analysis and a vector error-correction (VEC) model are applied to examine the shortand long-run relationships among foreign direct investment (FDI), economic growth, and the environment in China and India. The results show that FDI inflow plays a pivotal role in determining the shortand long-run movement of economic growth through capital accumulation and technical spillovers in the two countries. However, FDI inflow in both countries is found to have a detrimental effect on environmental quality in both the shortand long-run, supporting pollution haven hypothesis. Finally, it is found that, in the short-run, there exists a unidirectional causality from FDI inflow to economic growth and the environment in China and India ─ a change in FDI inflow causes a consequence change in environmental quality and economic growth, but the reverse does not hold.


INTRODUCTION
Since the economic reform and opening up to the outside world in the late 1970s and the early 1980s, China and India have been the fastest growing economies in the world.
Between 1992 and 2005, for example, the Chinese and Indian economies have grown on average by approximately 10% and 7% annually ( Figure 1). Accordingly, foreign direct investment (FDI) inflows to the two countries have grown rapidly during the same period.
Between 2000 and 2005, for example, the average annual inflows of FDI in China and India have reached $54.5 billion and $5.2 billion, respectively, more than double the amount of the 1992-1999 period ( Table 1) A plethora of studies has been conducted to deal with the economics of FDI in developing countries over the last three decades. Theoretical research in this area can be roughly categorized into two groups. The first group of studies has provided the theoretical rationale of the effect of FDI inflows on economic growth, which is known as 4 the FDI-growth nexus (e.g., Romer 1986, Lucas 1988, Rebelo 1991, Helphman and Grossman 1991. For example, the modern endogenous growth theory shows that longrun economic growth of the economy can result from more open liberalized government policies conductive for FDI inflows. More specifically, if capital is considered as knowledge rather than just plant and equipment, then the inflow of foreign capital can itself result in technological change and spillovers of ideas across countries (Grossman and Helphman 1991). With the capital exhibiting such increasing returns to scale, therefore, changes in FDI inflows can be an important vehicle for long-run economic growth in developing countries. The second group of studies has attempted to relate theoretical consideration to the impact of FDI on the environment in developing countries, which is referred to as the FDI-environment nexus (e.g., Pethig 1976, Copeland and Taylor 1994, Porter and van der Linde 1995. For example, the pollution haven model asserts that, under globalization circumstance, the relatively lax environmental standards in developing countries become attractive comparative advantage to the pollution-intensive foreign capital seeking for weaker regulations to avoid paying costly pollution control compliance expenditure domestically (Copeland and Taylor 2003). On the other hand, the Porter hypothesis claims that, since environmental quality is a normal good, as income increases with FDI inflows, developing countries tend to adopt more strict environmental regulations (Porter and van der Linde 1995).
To date, on the other hand, empirical studies have mostly concentrated on how the inflow of FDI affects economic growth in developing countries (e.g., Tsai 1991, Wang and Swain 1997, Liu et al. 1997, Sun and Parikh 2001, Bende-Nabende et al. 2001, Liu et al. 2002, Shan 2002, Chakraborty and Basu 2002, Yao 2006, and Chang 2007. For 5 example, Wang and Swain (1997) employ a single equation model (i.e., ordinary least squares) to analyze factors affecting foreign capital inflows into China and Hungary; they show a positive relation between changes in the level of GDP and the inflow of FDI in those countries. Sun and Parikh (2001) use a structural model (i.e., three least squares) to examine the relationship between inward FDI, exports and economic growth in China; they find that an increase in FDI (and exports) has a positive and significant impact on Chinese economic growth. Chakraborty and Basu (2002) adopt a non-structural time series model (i.e., vector error-correction) to explore the dynamic interaction between FDI and economic growth in India; they discover evidence that GDP has a significant positive effect on inflows of FDI for the Indian economy in both short-and long-run.
Accordingly, empirical analyses of the FDI-environment nexus in developing countries have received little attention. To the best of our knowledge, Smarzynska and Wei (2001), Xing and Kolstad (2002), Eskeland and Harrison (2003), and He (2006) are the only four empirical studies that have attempted to address this issue. For example, Xing and Kolstad (2002) examine the effect of the U.S. FDI on environmental quality in both developed and developing countries; they find that developing countries tend to utilize lenient environmental regulations as a strategy to attract dirty industries from developed countries. He (2006) explores the relationship between FDI and the environment in China; he discovers evidence that an increase in FDI inflow results in deterioration of environmental quality. However, these studies implicitly assume a oneway causality from measures of environmental quality/regulations (SO 2 and CO 2 emissions or pollution abatement cost) and/or economic growth (GDP) to FDI and adopt a structural model (i.e., reduced-form equations) to estimate the impacts of FDI based on 6 such causality. As such, previous studies have neglected the endogenous nature as well as the possible causal relationships between FDI (and economic growth) and environmental quality in a multivariate framework; that is, whether an increase in FDI in developing countries caused by their weaker regulations deteriorates environmental quality or, alternatively FDI related spillover of knowledge tends to improve environmental quality via economic growth. In other words, no study has dealt with dynamic movements of FDI (and economic growth) and environmental quality. 1 The contribution of this study, therefore, is to examine the FDI inflowenvironment nexus in a dynamic framework of multivariate time-series. For this purpose, we assess the short-and long-run relationships among FDI, sulfur dioxide (SO 2 ) emissions and GDP in China and India using the Johansen cointegration analysis and vector error-correction (VEC) model. The Johansen approach features multivariate autoregression and maximum likelihood estimation; this method is well suited to address the issue of endogeneity and causal mechanisms when variables used in the model are non-stationary and cointegrated. In addition, the cointegration test is used to find the long-run equilibrium relationships among the selected variables. Finally, the VEC model provides information on the short-run dynamic adjustment to changes in the variables within the model. This analysis will shed new light on dynamic interrelationships between FDI inflows, economic growth and the environment, and contribute to the empirical literature on FDI-environment nexus.
In the next section, the theoretical and empirical modeling of FDI-environment nexus is presented. This is followed by a description of data used in the analysis and a 7 discussion of unit root tests. The empirical results are discussed followed by some conclusions.

Theoretical Framework
In examining the dynamic relationship between FDI, GDP and SO 2 emissions in China and India, we rely on a FDI-environmental policy model developed by Xing and Kolstad (2002). More specifically, in its simplest form the foreign direct investment ( FDI ) in the host country can be specified as follows: where GDP is the gross domestic product of the host country, which is used as a proxy for the strength of the economy; 1 Z is a vector of exogenous variables affecting FDI inflows such as cost structures (i.e., labor costs) and differentials in rewards of factor services; and * R is the environmental regulatory laxity. The relationship between GDP and FDI is expected to be positive, implying that economic growth is the most important determinant for FDI inflow to the host country ( ). The positive relationship between FDI and * R indicates that lax environmental policy is more attractive to pollution-intensive FDI, thereby increasing polluting industries in the host country.
Similarly, the pollution ( E ) such as SO 2 emissions in the host country can be specified as follows: where 2 Z is a vector of exogenous variables affecting pollution levels such as energy consumption and prices. In general, the relationship between GDP and pollution emissions is expected to be positive, indicating that an increase in the scale of economic activity through income growth necessarily brings about a proportionate increase in ). Defining environmental quality as a normal good, however, it is further hypothesized that pollution emissions decrease as rising income passes beyond ). Economists call this relationship as the Environmental Kuznets Curve (EKC) Krueger 1991 and1993). The relationship between pollution and * R is expected to be positive, implying that lenient environmental regulations result in an increase in environmental degradation.
Assuming that 2 f is invertible in * R 2 , equation (2) can be solved for * R as a function of the other variables as follows: Finally, we substitute equation (3) into equation (1), which yields the following relationship: The estimation of equation (4) is the basic approach of this study. It should be emphasized that the relationship between FDI and pollution emissions (or environmental regulatory laxity) in developing countries is ambiguous and uncertain. More specifically, if pollution-intensive foreign capitals move to developing countries with weaker regulations, then the inflow of FDI deteriorates environmental quality ( On the other hand, if developing countries rely on technology transfer through FDI from developed countries as a primary means of technology acquisition, the inflow of FDI tends to enforce environmental regulations via economic growth, thereby improving ).

Specification of Time-Series Models
To estimate the long-run relationship among FDI, GDP and SO 2 emissions, we use the maximum likelihood estimation procedure developed by Johansen (1988) and Johansen and Juselius (1992). More specifically, given a vector having up to k lags as follows: are the matrix of long-run coefficients, where I is the identity matrix;  is a vector of constant; and t u is a vector of normally and independently distributed error terms, or white noise. If the coefficient matrix  has reduced rank ─ i.e., there are ) 1 (   n r cointegration vectors present, then the  can be decomposed into a matrix of loading vectors,  , and a matrix of cointegrating vectors,  , such as where r is the number of cointegrating relations,  represents the speed of adjustment to equilibrium, and '  is a matrix of long-run coefficients. For three endogenous nonstationary variables in our analysis, for example, the term in equation (5) represents up to two linearly independent cointegrating relations in the system. The 10 number of cointegration vectors, the rank of  , in the model is determined by the likelihood ratio test (Johansen 1988).
If all variable in a vector of stochastic process t Y are cointegrated, an errorcorrection representation captures the short-run dynamics while restricting the long-run behavior of variables to converge to their cointegrating relationships (Engle and Granger 1987). This can be done by estimating an error-correction model in which residuals from the equilibrium cointegrating regression are used as an error-correcting regressor. For this purpose, equation (5) can be reformulated as a short-run dynamic model as follows: Y is a measure of the error or deviation from the equilibrium, which is obtained from lagged residuals from the cointegrating vectors. Since the series are cointegrated, equation (6) incorporates both short-and long-run effects. That is, if the long-run equilibrium holds, then the term 1 '   t Y is equal to zero. During periods of disequilibrium, on the other hand, this term is non-zero and measures the distance of the system from equilibrium during time t . Thus, an estimate of  provides information on the speed-of-adjustment, which implies how the variable t Y changes in response to disequilibrium.

Data
It is worth noting that among principal air pollutants, sulfur dioxide (SO 2 ) and carbon dioxide (CO 2 ) are the major measures of air pollution that have been widely used in the empirical studies. Of those, SO 2 represents the measure of local air pollution, whereas 11 CO 2 represents a global pollutant (externality), which individual countries are unable to regulate without international cooperation (Frankel andRose 2005, He 2006). Given our individual country-specific approach, therefore, it is more appropriate to select SO 2 emissions as a proxy for the measure of environmental quality in China and India.

Testing For Unit Roots
When dealing with time-series data, the possibility of unit roots in a series raises issues about parameter inference and spurious regression (Wooldridge 2000). For example, OLS regression involving non-stationary series no longer provides the valid interpretations of the standard statistics such as t -statistics, F -statistics, and confidence intervals. To avoid this problem, non-stationary variables should be differentiated to make them stationary. However, Engle and Granger (1987) show that, even in the case that all the variables in a model are non-stationary, it is possible for a linear combination of integrated variables to be stationary. In this case, the variables are said to be cointegrated and the problem of spurious regression does not arise. As a result, the first requirement for cointegration analysis is that the selected variables must be non-stationary.
To determine the existence of a unit root in the series, we examine the integration order of individual time-series ( ) for China and India using the Dickey-Fuller generalized least squares (DF-GLS) test (Elliot et al. 1996). This test optimizes the power of the conventional augmented Dickey-Fuller (ADF) test using a form of detrending. The DF-GLS test works well in small samples and has substantially improved power when an unknown mean or trend is present (Elliot et al. 1996). The results show that the levels of all the series are non-stationary, while the first differences are stationary (

Johansen Cointegration Test
Before implementing the cointegration test, the important specification issue to be addressed is the determination of the lag length for the VAR model, because the Johansen procedure is quite sensitive to changes in lag structure (Maddala and Kim 1998 (Table 3). 3 More specifically, in the residual serial correlation and heteroskedasticity tests using the F -form of the Lagrange Multiplier (LM) procedure, the null hypotheses of no serial correlation and no heteroskedasticity cannot be rejected at the 5% significance level. In the residual normality test using the Doornik-Hansen method (Doornik and Hansen 1994), on the other hand, the null hypothesis of normality can be rejected for 3 individual series and the 14 system for China at the 5% significance level. However, non-normality of residuals does not bias the results of the cointegration estimation (Gonzalo 1994 which in turn leads to higher economic growth. When determining the existence of cointegration relationship, the cointegration vectors ( j  ) estimated from equation (5)  of the three eigenvectors is most highly correlated with the stationary part of the process t Y  when corrected for the lagged values of the differences. As such, 1  represents the cointegration vector determined by the cointegrated VAR model (Johansen 1988). After normalizing the coefficients of FDI, for example, the long-run equilibrium relation ( 1  ) between the three variables in China and India can be represented as the reduced forms of equations (7) and (8) (7) and (8) show that economic growth in China and India has a positive longrun relationship with FDI, indicating that economic growth tends to attract more FDI inflow. In addition, a positive long-run relationship between SO 2 emissions and FDI in both countries implies that an increase in SO 2 emissions (or relaxation of environmental 16 regulations) tend to an increase in FDI inflow. This finding provides supportive evidence for the so-called pollution haven hypothesis; that is, such developing countries as China and India tend to utilize lenient environmental regulations in an effort to attract multinational corporations, particularly those engaged in highly polluting activities from developed countries.

VEC Model
The VEC model is estimated to identify the short-run adjustment to long-run steady states as well as the short-run dynamics among FDI, GDP and SO 2 emissions in China and India. For this purpose, we estimate the short-run VAR model in equation (6), with the identified cointegration relationships in equations (7) and (8). We adopt a general-tospecific procedure to estimate the VEC model (Hendry 1995). In the case of China, for example, since FDI and SO 2 emissions are found to be weakly exogenous to the system, the VEC model is first estimated conditional on the two variables. By eliminating all the insignificant variables based on an F -test, the parsimonious VEC (PVEC) model is then estimated using OLS (Harris and Sollis 2003). Likewise, the VEC model for India is estimated conditional on FDI. The number of lags used in the PVEC model is the same as that in the cointegration analysis. The multivariate diagnostic tests on the estimated model as a system show no serious problems with serial correlation, heteroskedasticity, and normality (Table 6). This suggests that the PVEC specifications do not violate any of the standard assumptions.
The results show that the error-correction terms ( 1  t EC ) for China and India are negative and significant at the 5% significance level (Table 6). More specifically, the negative coefficients of 1  t EC ensure that the long-run equilibrium can be achieved. The absolute value of 1  t EC indicates the speed of adjustment to equilibrium. As such, the results indicate that, with a shock to the Chinese and Indian economies, GDP and SO 2 emissions tend to recover to their long-run equilibrium position. However, the adjustment toward equilibrium is not instantaneous. For example, the coefficients of  2. Since the environmental regulatory laxity is not directly observed, Xing and Kolstad (2002) solve this latent variable problem by using pollutant emissions to infer laxity.
For example, SO 2 emissions can be used as a yardstick to characterize the change of environmental regulation laxity; that is, relaxation (enforcement) of environmental regulation leads to an increase (decrease) in SO 2 emissions. Accordingly, pollution emissions ( E ) and environmental regulatory laxity ( * R ) is interchangeable in this model.
3. The sample size could be another issue of concern for the Johansen procedure, because finite-sample analyses can bias the cointegration test toward finding the longrun relationship either too often or too infrequently. In fact, the number of observations used in this study seems to be a bit small; our findings should thus be viewed with caution. However, Hakkio and Rush (1991) note: "Our Monte Carlo studies show that the power of a cointegration test depends more on the span of the data rather than on the number of observations. Furthermore, increasing the number of observations, particularly by using monthly or quarterly data, does not add any robustness to the results in tests of cointegration." Following these authors, the annual data used in this study (23 years) can be considered to be long enough to reflect the long-run relationship between FDI, GDP and SO 2 emissions, which should somewhat mitigate our concern with the relatively small sample size. 20 4. Doornik and Hendry (2001) note: "The sequence of trace tests leads to a consistent test procedure, but no such result is available for the maximum eigenvalue test.
Therefore current practice is to only consider the former." Following these authors, we only depend on the former to test the null hypothesis. Note:  denotes the first differences of the variables. p -values are given in parentheses.
** and * indicate rejection of null hypothesis of non-stationarity at the 5% and 10% significance levels, respectively. Serial correlation of the residuals of individual equations and a whole system was examined using the F -form of the Lagrange-Multiplier (LM) test, which is valid for systems with lagged independent variables. Heteroskedasticity was tested using the F -form of the LM test. Normality of the residuals was tested with the Doornik-Hansen test (Doornik and Hendry 1994). Note: ** indicates rejection of the null hypothesis at the 5% significance level.
Parentheses are p -values. The trace test leads to a consistent test procedure, but the maximum eigenvalue test does not (Doornik and Hendry 2001, p. 175). For this reason, we only report the former to test the null hypotheses.  1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Year