Trade Liberalization, Economic Growth, Energy Consumption and the Environment: Time Series Evidence from G-20 Economies

This study examines the dynamic interrelationships between trade, income growth, energy consumption and CO2 emissions for G-20 economies in a framework of cointegrated vector autoregression (CVAR). Johansen’s maximum likelihood procedure is used to estimate the coefficients of the cointegrated VAR. The results show that trade and income growth have a favorable effect on environmental quality for the developed G-20 member countries, while they have an adverse effect on the environment for the developing member countries. We also find that energy consumption tends to worsen environmental quality for both the developed and developing countries. Finally, it is found that trade and income to emission and energy causality holds for the developed countries; changes in degree of trade openness and income growth lead to corresponding changes in the rates of growth in emission and energy consumption. Emission and energy to trade and income causality, on the other hand, is found to hold for the developing countries; any shocks in emission and energy consumption cause corresponding fluctuations in income growth and trade openness.


I. Introduction
Over the past three decades, a plethora of studies have been conducted to empirically identify the relationship between environmental quality (e.g., sulfur dioxide (SO 2 ) and carbon dioxide (CO 2 ) emissions) and measures of economic activity such as economic (income) growth and energy consumption. Given the primary focus of empirical analysis, the literature can be classified into three categories.
The first group analyzes the (causal) relationship between income (economic) growth and environmental pollutants, referred to as the growth-environment nexus (e.g., Grossman and Krueger 1991;Shafik 1994;Agras and Chapman 1999;Heil and Selden 1999;Friedl and Getzner 2003;Dinda and Coondoo 2006;Managi and Jena 2008). These studies have typically concentrated on identifying the existence of environmental Kuznets curve (EKC). The EKC hypothesizes an inverted U-shaped relationship between (per capita) income and pollution levels; that is, environmental quality first deteriorates and then improves with per capita income (see Dinda (2004) for detailed review of EKC literature). In their seminal work, for example, Grossman and Krueger (1991) find that the EKC holds for North American countries. Agras and Chapman (1999) use a cross-section panel of countries to examine the EKC hypothesis; they find little evidence of the inverted relationship between income and the environment.
The second group turns its attention to investigate the relationship between income growth and energy consumption, referred to as the growth-energy nexus (e.g., Kraft and Kraft 1978;Yu and Choi 1985;Glasure and Lee 1997;Soytas et al. 2001;Soytas and Sari 2003;Akinlo 2008). The central issue of this group is to explore whether economic growth stimulates energy consumption or vice versa (see Ozturk (2010) for detailed literature survey on this issue). Kraft and Kraft (1978), for example, examine the causal linkages between energy consumption and economic growth in the United States; they find that the causal relationship runs from economic growth to energy, but the reverse does not hold. Soytas et al. (2001) analyze income-energy causality in Turkey; they show that economic growth depends on energy consumption, and a decrease in energy con-sumption may restrain economic growth. Finally, there has recently been a growing body of literature that has combined the first and second approaches as noted earlier in order to analyze dynamic relationships among income growth, energy consumption and the environment, referred to as the growth-energy-environment nexus (e.g., Soytas et al. 2007;Zhang and Cheng 2009;Soytas and Sari 2009;Jalil and Mahmud 2009). Zhang and Cheng (2009), for example, examine the dynamic interrelationships between energy consumption, output and carbon emission in China; they find that, in the long-run, CO 2 emissions tend to increase as income and energy consumption increase.
An important point frequently overlooked in the literature, however, is that most of studies analyzing the income growth-environment nexus have used crosssection or panel data to examine the relationship between income and pollution level for multiple countries and enough attention has not been given to a countryspecific data (e.g., Shafik 1994;Agras and Chapman 1999;Heil and Selden 1999;Friedl and Getzner 2003;Dinda and Coondoo 2006); considering heterogeneity of conditions observed in social, economical and political factors, economic development trajectory for an individual country may not be the same as a pattern of a group of countries. In addition, studies evaluating the growth-energy nexus have mostly employed the standard Granger causality test with little cognizance of a cointegrating relationship among variables; if, in fact, the selected variables are cointegrated in a model, the causality test provides misleading results (Miller and Russek 1990). Further, Granger causality analysis typically focuses on shortrun dynamics rather than long-run equilibrium relationships. Although some studies (e.g., Stern 2000;Ghali and El-Sakka 2004) have examined the existence of long-run relationships among variables, they did not investigate the mechanisms through which long-run equilibria restored. Finally, the environmental consequences of trade liberalization have so far received little attention. 1) In other 1) Over the years economists have vigorously debated this issue (e.g., Copeland and Taylor 1994;Copeland 2005). Proponents of trade liberalization, for example, claim that, given the fact that environmental quality is a normal good, income growth induced by trade causes people to increase their demand for a clean environment and thus encourages firms to shifts towards cleaner techniques of production, which in turn improves both words, few studies have combined the three approaches to examine dynamic relationships among trade liberalization, income growth, energy consumption and environmental quality. To our knowledge, Baek et al. (2009) Baek et al. (2009). In its simple form, this model can be stated as: where  is the emission level;  is the trade openness; and  is the (per capita) income. If trade openness leads to economic growth through an increase in the scale of economic activity in a country, it could be hypothesized that trade has a positive relation with income ( ). Assuming that an economy follows the full trajectory of the environmental Kuznets curve (EKC), it could be hypothesized that emission level increases with growing income up to a threshold level ( ) beyond which emission level declines with ). Given the heated debate on the close link between global warming and emission of greenhouse gases (i.e., CO 2 emissions), energy consumption ( ) is thought to be of another potential factor influencing emission levels. In the empirical model used here, therefore, we extend equation (1) to represent the relationship between the environment and measures of economic activity including energy consumption as follows: To the extent that an increase in energy consumption induced by economic growth brings about a proportionate increase in emission level, it could be hypothesized that energy consumption positively affect emission level ( ).
With the modification, then, it is this model which provides the theoretical basis for the growth-energy-trade-environment nexus.
It is important to note that, since individual countries experience different levels of income and openness, and therefore energy consumption corresponding to their process of development, the true form of the pollution-income-opennessenergy relationship mainly depends on where an economy is currently located in a development trajectory (Baek et al. 2009). For example, as for individual countries that move beyond the EKC turning points with energy consumption

The Johansen cointegration approach
To estimate the long-run relationships among the selected variables in equation (2), we use the Johansen maximum likelihood estimation procedure. Following Johansen (1988), the cointegrated vector autoregression (VAR) model can be defined as follows:  is a vector of constant; and   is the white noise. Granger's representation theorem asserts that 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 (), that is Here,  is the number of cointegrating relations,  represents the speed of adjustment to equilibrium, and ′ is a matrix of long-run coefficients.
For four endogenous non-stationary variables in this analysis, for example, the term ′    in equation (3) represents up to three linearly independent cointegrating relations in the system. The number of cointegration vectors, the rank of  , in the model is determined by the likelihood ratio test (Johansen 1988 The second specification issue to be address is the determination of the lag length for the VAR model. Maddala and Kim (1998) show that the Johansen procedure is sensitive to changes in lag structure. The lag length () for the model is selected based on the likelihood ratio (LR) tests. This method compares the models of different lag lengths to see if there is a significant difference in results (Doornik and Hendry 1994 (Table 2).
3) When dealing with finite samples, especially with small numbers of observations, the power of the standard ADF test is known to be notoriously low (Maddala and Kim 1998;Harris and Sollis 2003). In other words, the ADF test has high probability of accepting the null hypothesis of nonstationarity when the true data-generation process is, in fact, stationary. 4) The results are not reported here for brevity but can be obtained from the authors upon request.  Normality of the residuals is tested with the Doornik-Hansen (1994) method.
The null hypothesis of normality is rejected for five cases at the 10% significance level; however, non-normality of residuals does not bias the results of Trade Liberalization, Economic Growth, Energy Consumption and the Environment 15 the cointegration estimation (Gonzalo 1994).
The Johansen cointegration procedure is applied to identify the number of cointegrating vectors among the selected variables. The results show one cointegration vector (  ) for 10 countries and two cointegration vectors (  ) for 3 countries at the 10% significance level ( Table 2) The cointegration vectors (  ) estimated from equation (3) explain the long-run 5) The trace test leads to a consistent test procedure, but the maximum eigenvalue test does not (Doornik and Hendry 2001, p. 175); hence, we use the trace statistic to test the null hypothesis. 6) It is worth mentioning that structural change (break) is an important issue in time-series analysis and affects all inferential procedures related to unit roots and cointegration (Maddala and Kim 1998). In fact, 7 developed countries (except Australia) show that their CO2 emissions levels tend to increase first, reach a peak and then start declining after a threshold point. For completeness, therefore, we employ the most recent Johansen cointegration technique that allows for structural breaks at known points in time (Johansen et al. 2000). For the United States, Japan and Korea, for example, the plausible structural breaks appear to occur at 1972, 1973 and 1997, respectively, for the CO2 emissions levels. We find the same results as those derived from the standard Johansen method, indicating that the plausible structural breaks in the series do not affect the long-run relationship among the selected variables for those 7 developed countries. We thank an anonymous referee for raising this issue.  four variables are cointegrated with one vector, the first eigenvector (  ) of the four eigenvectors is most highly correlated with the stationary part of the process when corrected for the lagged values of the differences. Hence,   represents the cointegration vector determined by the cointegrated vector autoregression (CVAR) model (Johansen 1988). 7) After normalizing the coefficient of CO 2 7) Since two cointegrating relationships are found with Canada, Italy and South Africa (Table 2), an identification problem may arise because of the stationarity caused by the linear combination of the two cointegration relations (Harris and Sollis 2003). To solve this problem, we impose restrictions on the cointegration spaces () to identify unique cointegrating vectors. As such, the long-run relationships for the three countries are explained using the relevant long-run coefficients (  and   ).
Trade Liberalization, Economic Growth, Energy Consumption and the Environment 17 emissions, for example, the long-run equilibrium relation (  ) among the four variables in the U.S. can be represented as the following reduced form (Table 3); In equation (4) A possible criticism of our efforts to examine the growth-energy-tradeenvironment nexus is that the sample size is relatively small for the cointegration analysis, because finite sample analyses could bias the cointegration test towards finding the long-run relationship either too often or too infrequently; our findings should thus be viewed with caution. As Hakkio and Rush (1991) note, however, the power of a cointegration test depends more on the span of the data rather than on the number of observations. Further, their Monte Carlo studies show that increasing the number of observations, particularly using monthly or quarterly data, does not add any robustness to the cointegration results. Combined with our relatively long enough time span (36~47 years) to reflect the long-run relationship among the variables, this should somewhat mitigate our concern with the relatively small sample size and may not undermine the credibility of our findings.
8) It is important to note that interpreting the coefficients in this relation as long-run elasticities are ambiguous since such an interpretation ignores the dynamics of the system (Lütkepohl 1991). A 1% increase in the U.S. income, for example, may not cause a long-term decline in CO2 emissions by 3.36% because an increase in the U.S. income is likely to have an effect on other variables as well that may interact in the long-run.

Results of long-run analysis
The cointegration vectors () estimated from equation (3) is used to explain the long-run relationship among CO 2 emissions, income, energy consumption and openness after normalizing the coefficients of CO 2 emissions and rearranging in reduced forms ( emissions are found to have a positive long-run relationship with income, suggesting that economic growth causes significant environmental degradation. These findings could be explained using the term emission intensity (ratio), which is defined as the ratio of per capita CO 2 emissions to per capita income (Baek et al. 2009 Figure 3). Notice that these 6 countries   (Table 1). 9) Of the 5 countries in which CO 2 emissions have a 9) According to the World Bank's main criterion of classifying economies, individual countries are divided based on 2008 gross national income (GNI) per capita as follows: (1) low income (or least developed countries; $995 or less); (2) middle income (or less developed/developing countries; $996~12,195); and (3) high income economies (or industrialized/developed countries; $12,195 or more). From the classification, therefore, G-20 economies can be divided into developing countries ($12,196 or less) and developed countries ($12,196 or more) since they do not include any low income countries; hence, developing countries include Argentina, Brazil, China, India, Indonesia, Mexico, Turkey and South Africa, while developed countries are Australia, Canada, France, Japan, Korea, Italy, Saudi Arabia, UK and USA (Table 1).
Per capita CO2 emission (metric ton)   Figure   3). 10) Overall, these findings provide indirect evidence to support the existence 10) The remaining countries also show the similar patterns. For brevity, however, the figures for those countries are not shown here but will be available upon request.
Trade Liberalization, Economic Growth, Energy Consumption and the Environment 21  1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005  The country that warrants our immediate attention is China, which shows a negative long-run relationship between CO 2 emissions and income, indicating that economic growth appears to reduce the emission level. Give the fact that China has been the largest carbon dioxide emitter since 2006, this finding may be peculiar. However, emission intensity by definition can keep improving as long as the growth rate of real income is faster than that of per capita CO 2 emissions. Indeed, since the beginning of economic reforms and opening-up to international markets in the late 1970s and the early 1980s, China has experienced much faster growth in real income than CO 2 emissions; for example, CO 2 emissions 11) It should be noted here that empirical studies have typically used a quadratic term of income in their models in order to test for the existence of an EKC. As discussed in the Introduction, however, the main purpose of this study is to examine dynamic interrelationships among trade, energy consumption, income and CO2 emissions, rather than a U-shaped relationship only between income and pollution level (i.e., CO2 emissions). For this reason, we do not incorporate the squared income term in the model; hence, our interpretation here should be viewed as one of possible explanations for the findings.  1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005  Trade Liberalization, Economic Growth, Energy Consumption and the Environment 23 show a negative long-run relationship between CO 2 emissions and openness, indicating that trade liberalization tends to improve environmental quality (Table 3). Finally, all 13 countries found to be cointegrated are shown to have a positive long-run relationship between CO 2 emissions and energy consumption, indicating that air pollution tends to increase as a country's energy consumption increases (   Table 4). This finding indicates that, for the developed member countries, income and/or trade openness are generally the driving variables in the system; they significantly affect CO 2 emissions and energy consumption in the long-run, but are not influenced by CO 2 emissions and energy consumption. In other words, changes in income and/or trade openness would cause changes in CO 2 emissions and energy consumption. This further suggests that, since environmental quality is a normal good, economic growth induced by trade liberalization allows for the possibility that people in developed countries demand for a clean environment, which in turn pushes firms to shift towards clearer techniques of production, thereby contributing to the environmental quality improvement. In addition, this finding supports the so-called conservation hypothesis associated with the relationship between energy consumption and economic growth/openness; that is, since economic growth induced by trade liberalization causes an increase in energy consumption, the policy of conserving energy consumption may be implemented with little or no adverse effect on economic growth and expansion of trade.
Of the 5 developing countries in which all four variables are cointegrated, on the other hand, the null hypothesis of weak exogeneity cannot be rejected for CO 2 emissions and energy consumption at the 10% level in almost all cases (8 out of the 10 cases). These results indicate that, for the developing member countries, CO 2 emissions and energy consumption are generally weakly exogenous to the long-run parameters in the system; hence, these two variables do not adjust to deviations from any equilibrium state defined by the cointegration relation. In other words, any shock in CO 2 emissions and energy consumption would cause fluctuations in income and trade openness. This finding further suggests that as high regulation countries (i.e., developed economies) implement tighter environmental regulations, multinational firms, particularly those engaged in highly polluting activities, tend to relocate to developing countries with lower environmental standards, which in turn worsen environmental quality with an increase in openness.
This further supports the so-called growth hypothesis; that is, since energy consumption stimulates economic growth/trade, restrictions on the use of energy may adversely affect economic growth in developing countries, while increase in energy may contribute to economic growth.

V. Concluding Remarks
While numerous studies have analyzed the relationship between environmental quality, income growth and energy consumption, relatively little attention has been paid to the environmental consequences of trade liberalization. We examine the dynamic effects of trade, income, energy consumption on CO 2 emissions for G-20 economies in a cointegration framework. To achieve this goal, the Johansen multivariate cointegration method is used.
The results of the cointegration analysis show that there is a long-run relationship (s) between environmental quality and measures of economic activity; that is,