Trade openness and environmental quality: International evidence

Energy Policy 92 (2016) 45–55

Contents lists available at ScienceDirect

Energy Policy journal homepage: www.elsevier.com/locate/enpol

Trade openness and environmental quality: International evidence Thai-Ha Le a, Youngho Chang b,n, Donghyun Park c a b c

RMIT University, Vietnam Nanyang Technological University, Singapore Asian Development Bank, Philippines

H I G H L I G H T S








Relation between trade and the environment is examined for 98 countries. A long-run relationship exists among pollutant emissions, openness, and growth. Openness appears to lead to environmental degradation for the global sample. The relationships differ according to the income of countries. A feedback effect between trade openness and pollutant emissions exists.

art ic l e i nf o

a b s t r a c t

Article history: Received 18 August 2015 Received in revised form 28 December 2015 Accepted 21 January 2016

We examine the relationship between trade openness and the environment in a cross-country panel, using the emission of particulate matter (PM10) as the basic indicator of environmental quality. The panel cointegration test results show a long-run relationship between particulate matter emissions, trade openness, and economic growth. We find that increased trade openness leads to environmental degradation for the global sample. However, the results differ according to the income of countries. Trade openness has a benign effect on the environment in high-income countries, but a harmful effect in middle- and low-income countries. These results are generally robust to different measures of trade openness and environmental quality. Interestingly and significantly, the results are consistent with the popular notion that rich countries dump their pollution on poor countries. Finally, we find evidence of a feedback effect between trade openness and particulate matter emissions for the global sample as well as different income groups of countries. & 2016 Elsevier Ltd. All rights reserved.

Jel classifications: Q56 Keywords: Trade openness Environment Particulate matter emissions Global panel data set

1. Introduction Reflecting widespread concerns over environmental degradation, protecting the environment has emerged as a global priority in recent decades. In this context, the impact of trade on the environment is an issue of growing importance in trade policy. There has been an increasing number of empirical studies that investigate the relationship between openness to trade and environmental quality (see, e.g., Frankel and Romer, 1999; Antweiler et al., 2001; Cole and Elliot, 2003; Boulatoff and Jenkins, 2010; Shabaz et al., 2013). The evidence from these studies is mixed, with some studies finding a positive relationship while others find a negative relationship. Both theory and evidence suggest that trade promotes growth. To name just one example, the remarkable n

Corresponding author. E-mail address: [email protected] (Y. Chang).

http://dx.doi.org/10.1016/j.enpol.2016.01.030 0301-4215/& 2016 Elsevier Ltd. All rights reserved.

rise of China was driven to a large extent by its integration into the global trading system. If trade leads to growth, and growth leads to environmental deterioration, the affected countries may impose more stringent environmental regulations. We can expect the consequent employment of more environmentally friendly production methods to improve environmental quality. This view is empirically supported by Antweiler et al. (2001) who found that trade openness is associated with reduced pollution as measured by sulphur dioxide concentrations. Baek et al. (2009) also showed that trade and income positively affected environmental quality in developed countries and China. Boulatoff and Jenkins (2010) found evidence of a negative long-term relationship between trade and oil-related carbon emissions across different income groups of countries. On the other hand, a number of studies have found that increased openness can worsen environmental quality. Conceptually, a country which has a comparative advantage in products that

46

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

require a lot of pollution may specialise in the production of that commodity. But doing so will increase pollutant emissions and adversely affect environmental quality. China’s emergence as a global manufacturing powerhouse is a good example of rapid export growth coinciding with extensive environmental deterioration. This negative view of the impact of trade on the environment is consistent with Kellenberg (2009) and Managi and Kumar (2009). So does trade openness harm or benefit the environment? More specifically, does trade openness increase or reduce pollutant emissions? In this study, we characterise environmental quality as a stock and its rate of deterioration or improvement as a flow. It therefore follows that higher flows of pollutant emissions cause a greater degradation of the environment. We present new evidence from a panel dataset of countries across different regions of the world. Using the globally representative panel dataset, we hope to contribute toward a broader and deeper understanding of the impact of trade on the environment, in particular the linkage between trade and pollutant emissions. The main findings of the study are as follows. First, we could not reject the existence of long-run relationship between trade openness, particulate matter emissions, and income for the whole panel as well as across different income groups of countries. The finding is qualitatively robust to different measures of trade openness and pollutant emissions. Second, we find that in the long run, increased openness to trade leads to environmental degradation for the global sample. However, the results differ according to country income level. For high-income countries, the environmental effect of trade openness is found to be positive while for middle- and low-income countries, this impact is significantly negative. The results are generally robust to different proxies of trade openness and pollutant emissions. Third, we find evidence of a feedback effect-i.e. bidirectional causal relationsbetween trade openness and particulate matter emissions for the global sample as well as across different income groups of countries. The rest of the paper is organised as follows. The second section reviews the relevant literature. The third section discusses the empirical framework and data. The fourth section reports and discusses the empirical results. The final section concludes the paper.

2. Trade and the environment: Some conceptual issues According to Copeland and Taylor (2013), trade openness can influence the environment through two key channels: the scale effect and the composition effect. The scale effect refers to the impact of trade on the level of economic activity. Specifically, increased openness leads to a greater economic activity, for instance, more transportation services, and more generally, more production and consumption of goods and services. Since these activities inherently entail environmental costs, one might conclude that increased economic activity stimulated by trade openness worsens environmental quality. On the other hand, the composition effect refers to the influence of trade on the composition of output across countries. Specifically, poor countries with relatively weak environmental regulations will specialise in producing dirty goods while rich countries with tough environmental policies specialise in clean goods. This leads to a shifting of polluting industry from developed to developing countries (Copeland and Taylor, 2013). This view is consistent with Baek et al. (2009) who find that trade and income favourably impact environmental quality in developed countries. However, the study found that the environmental impact of trade was negative in most developing countries. Dirty industry

migration raises a serious concern that poor and less developed countries are increasingly bearing the pollution burdens of consumption in rich and developed countries (Copeland and Taylor, 2013). Yet there are also conceptual grounds for a beneficial effect of trade on the environment. Antweiler et al. (2001) argue that increased openness may promote the environment through the technique effect. Specifically, if higher real income induced by trade liberalisation leads to a higher level of economic development, which is usually associated with greater ability and willingness to implement and enforce environmental regulations, environmental quality might improve. In addition, if the greater scale of economic activity due to increased openness encourages exploration into cleaner production techniques, this will reduce pollutant emissions. In other words, openness to trade can benefit the environment if it brings about income gains which enable some countries to specialise in relatively clean industries (Copeland and Taylor, 2013). Since theory offers grounds for both positive and negative relationship between trade and the environment, the issue must be settled through empirical analysis. In this context, we empirically analyse a large cross-country dataset of particulate matter emissions in 98 countries. Our empirical analysis of the relationship between trade openness and environmental pollution takes into account the income effect – i.e. the Environmental Kuznets Curve (EKC) hypothesis. This hypothesis posits an inverted U-shaped pattern between environmental pollution per capita and income per capita. As income increases, environmental pollution increases up to some income threshold, after which environmental pollution declines.1 The evidence on the EKC hypothesis is mixed and inconclusive. The results from some studies are consistent with the hypothesis (e.g., Dinda and Coondoo, 2006; Managi and Jena, 2008). On the other hand, many other studies (e.g., Dinda et al., 2000; Coondoo and Dinda, 2008; Akbostanci et al., 2009) refute the EKC hypothesis. Coondoo and Dinda (2002) discuss the issue of causality in the context of EKC from the standpoint of economic theory, and explain how this links up with the concept of income–emission causality underlying the Granger causality test. Their study used a dataset which covers 88 countries to examine income– emission causality patterns separately for 12 country groups across different continents. The results reveal three different types of causality relationship for different country groups. These studies that test the EKC hypothesis are, however, based on a bivariate framework which might suffer from omitted variable bias that causes spurious results. To summarise, the literature on the relationships between trade openness and environmental quality is inconclusive. Further, very few studies have used a panel framework to address possible cross-country dependence. We were only able to find Jaunky (2011), Arouri et al. (2012), Haggar (2012) and Ozcan (2013). Yet, none of these studies looks at a sample of countries from the whole world nor do they incorporate income level and the EKC hypothesis into the analysis. Our aim is to contribute to the literature by filling these gaps.

3. Empirical framework and data In this section, we described the framework and data that we use for the empirical analysis of the relationship between trade and the environment. 1 Dinda (2004) provides a thorough review of the EKC literature, including background history, conceptual insights, policy implications, and the conceptual and methodological critique.

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

3.1. The baseline model and data description To illustrate theoretical relationships among trade openness, per capita income and environmental quality, we first define environmental quality (EQ) as a function of trade openness (TO) and per capita income and the square of per capita income (GDP and GDP2) as follows:

(

)

EQ = f TO, GDP , GDP 2

(1a)

As illustrated in Eq. 1a, since income plays a strong role in determining environmental outcome, we incorporate the EKC into our analysis. Since environmental quality is a key variable in the analysis of this study, it is important to choose an appropriate indicator for it. The environmental impact of trade might take various forms of pollution. In other word, trade might entail several types of environmental effects (Frankel and Rose, 2005). For example, in modelling the effect of trade on a country’s environment for a given level of per capita GDP, Frankel and Rose (2005) look at seven measures of environmental quality, namely, SO2 (sulphur dioxide), NO2 (nitrogen dioxide), PM (total suspended particulate matter), CO2 (industrial carbon dioxide emissions), deforestation (average annual percentage change), energy depletion (the product of unit resources rents and the physical quantities of fossil fuel energy extracted which consists of coal, crude oil and natural gas, as a percentage of GDP), and rural clean water access as percentage of rural population. Nevertheless, their study focuses on the first three measures of local air pollution, since they are the most relevant. In particular, they argue that it is impossible to address CO2 by regulation at the national level since it is a purely global externality. Deforestation and energy depletion are not measures of pollution. Furthermore, while it might be worthwhile to look at these broader measures of environmental quality, Frankel and Rose (2005) assert that the measurements of deforestation, energy depletion and water access might involve serious problems of composition and data reliability. Since our study covers a global panel data set, availability of data that is commensurable for a large number of countries and over long time periods is a major problem. Data are not available for many environmental quality indicators, such as NOx (nitrogen oxides), VOC (volatile organic compound), CO (carbon monoxide), PM2.5 (particulate matter up to 2.5 μm in size), ozone, measures of water quality such as BOD (biochemical oxygen demand) and COD (chemical oxygen demand), indicators of soil degradation, deforestation, biodiversity loss, and the like. Also, we did not use composite environmental quality indices, such as the Environmental Sustainability Index or the various indices for anthropogenic ecological footprints (www.ciesin.org), because they are available only for very few (most recent) years and usually combine ecological and environmental policy components (Bernauer and Koubi, 2009). On the other hand, data for PM10 is reliable and available for a large number of countries (98 countries) during the investigation period. As such, we use PM10, an indicator of local air pollution, as the main indicator of environmental quality in this study. Air pollution is indeed a major environmental concern in most major cities across the world (Nowak et al., 2006). Specifically, particulate matter (PM) air pollution is one of the major causes of poor environmental quality in many cities all over the world, with serious effects on human health, buildings, and vegetation (Salvador et al., 2012). PM is solids or liquids that are dispersed in ambient air (Miah et al., 2011). PM10 are particulates with a diameter less than 10 μm, small enough to be inhaled. They may thus enter deep into the human respiratory tract and pulmonary system (Shandilya et al., 2007). These particles are responsible for most of the airborne particle threat to human health due to their

47

small size. They pose a human health hazard, potentially causing heart disease, lung cancer, asthma, atherosclerosis, cancer (as soot particles can contain carcinogenic material), and even death (Brunekreef and Holgate, 2002). These particulates also contribute to global warming, since increase in suspended PM directly increases greenhouse gas emissions (Miah et al., 2011). Several studies in the literature use PM10 as an indicator of environmental quality (e.g., Nowak et al., 2006; Plassmann and Khanna, 2006; Lee et al., 2009; Miah et al., 2011; Salvador et al., 2012). To conclude, there are several indicators of environmental quality in literature, namely, SO2, NOx, BOD, COD, and particulate matter (PM10 and PM2.5). In this study, due to the data availability of our global panel data set, we use particulate matter 10 μm or less in diameter (PM10) as the main measure of environmental quality, which is taken from Word Development Indicators (WDI).2 Nonetheless, we will attempt to judge the robustness of our results by examining carbon dioxide emissions (CO2, measured in metric tons per capita) (CE hereafter) as another indicator of environmental quality since data is available for our global sample. The Environmental Protection Agency (EPA) recently determined that CO2 emissions pose a threat to human health and welfare (Boulatoff and Jenkins, 2010). PM10 has local and trans-boundary impacts whereas CO2 is a greenhouse gas and has a global impact. Trade openness variable has been measured in various ways in the literature. By employing different measures of trade openness, we are able to examine the relationship between trade openness and environmental quality more thoroughly. Squalli and Wilson (2011) review several common measures of trade openness that have been used in literature, and list M/GDP, X/GDP and (X þM)/ GDP (which, respectively, express import, export and total trade as a share of GDP of a given country) as the three most widely used measures of trade openness. A large number of studies have also employed these three indicators of trade openness in their empirical investigation, for instance, Deme (2002), Kim and Lin (2009), Squalli et al. (2010), Herrerias and Orts (2011), Kim (2011), Harris et al. (2011), Hye (2012), and Liargovas and Skandalis (2012). Hence, to check for the robustness of the results, besides using the ratio of total trade to GDP as the main measure of trade openness, we also experiment with alternative measures, namely, the ratio of exports to GDP and the ratio of imports to GDP. As such, the baseline model in this study could be written in this following compact form:

PMit = f (TOit , GDPit , GDPit2)

(1b)

or, alternatively:

PMit = νi + η1iTOit + η2iGDPit + η3iGDPit2 + εit

(1c)

where i¼ 1, 2, 3, … N for each country in the panel and t¼ 1, 2, 3, … T refers to the time period. PMit is the particulate matter 10 μm or less in diameter (PM10, as measured in micrograms per cubic metre), TOit is the trade openness (measured as the ratio of the value of total trade to GDP), GDPit is per capita real GDP (constant 2005 US$), and GDPit2 is the square of per capita real GDP. The data of all required variables are taken from World Development Indicators (WDI, 2014). The data used in this study is pooled annual time series. All variables are converted into natural logarithms to reduce heteroskedasticity. The variables are entered into the model [1c] in their log level or first log difference, 2 According to WDI (2014), PM10 are “capable of penetrating deep into the respiratory tract and causing significant health damage. Data for countries and aggregates for regions and income groups are urban-population weighted PM10 levels in residential areas of cities with more than 100,000 residents. The estimates represent the average annual exposure level of the average urban resident to outdoor particulate matter. The state of a country’s technology and pollution controls is an important determinant of particulate matter concentrations”.

48

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

depending on whether they are I(0) or I(1) series. In the latter case, we can obtain the growth rate of the relevant variables by their differenced logarithms. The coefficients η1, η2, η3 correspond to the elasticities of particulate matter dioxide emissions with respect to openness to trade, real GDP per capita, squared real GDP per capita, respectively. Under the EKC hypothesis, which implies nonlinear effect of income on the environment, the elasticity estimates of particulate matter emissions per capita with respect to real GDP per capita and the square of real GDP per capita are expected to be positive and negative, respectively. The sign of η1, the elasticity estimate of particulate matter emissions per capita with respect to trade, is the main variable of interest, in light of the mixed and inconclusive effect of trade on the environment found in the literature. Our country sample includes 98 countries and our sample period spans 1990 to 2013. The choice of countries in the sample and the sample period is based on data availability. The sample countries included are at various stages of economic development. Previous studies seem to indicate that the relationship between economic growth and the environment might be different for countries at different income levels (e.g., Al-Mulali, 2011). In addition, Tayebi and Younespour (2012) showed that it is useful to examine the trade-environment relationship conditional on country characteristics. This is because besides differences in factor endowments, trade flows are also determined by differences in environmental and pollution policy. Tayebi and Younespour (2012) argue that countries with higher per capita income tend to have more stringent environmental regulations since environmental quality is a normal good. This gives rise to the pollution haven hypothesis, which implies that dirtier industries have been shifting from the developed to the developing world to escape tighter environmental standards (Esty, 2001).3 In light of this possibility, in addition to analysing the whole world panel, we divide the 98 countries into three sub-samples according to World Bank’s income classification. Specifically, the first sub-sample includes high-income countries, which have incomes of US$12,616 or more, the second sub-sample is middleincome countries (lower-middle income countries have incomes between US$1,036 and US$4,085, and upper-middle income countries have incomes between US$4,086 and US$12,615) and the third sub-sample consists of low-income countries, which have a per-capita income of US$1,035 or less. Table 1 presents the averages and the average growth rates of the key variables used in the study. The levels of most variables are higher for high-income countries than those of the low- and middle-income countries, except for particulate matter emissions. However, the pattern is different for growth rates. All the income groups have negative numbers for particulate matter emissions. High-income countries have the biggest negative growth rate while low-income countries have the lowest negative one. Middleincome countries have the highest growth for carbon dioxide emission and GDP per capita while low-income countries have the highest growth for different measures of trade openness. 3.2. Methodology To investigate the relationships between particulate matter emissions (PM) and trade openness (TO), real GDP per capita (GDP), and square of GDP per capita (GDP2) for 98 countries in the world for the period 1980 to 2013, we use a panel data model. Panel data has many advantages over cross-sectional or time series 3 Here if trade flows could be specified by income group, it would be optimal. However, this is not empirically feasible due to the lack of comprehensive data for a global panel during a sufficiently long period.

data. For instance, in the case of short time series, using panel data allows for more observations by pooling the time series data across countries and results in higher power for the Granger causality test (Pao and Tsaim, 2010). In addition, in contrast to time series and cross-sectional data, by controlling for individual heterogeneity, panel data allows for “more informative data, more variability, less collinearity among the variables, more degrees of freedom, and more efficiency” (Baltagi, 2005). The first step of the analysis is to test cross-sectional dependence to decide appropriate unit root tests. Cross sectional independence states that error terms are not cross-correlated, and zero error covariance is a very important issue in panel unit root and cointegration tests. Relaxing this assumption makes the derived distributions of panel unit root and cointegration tests no longer valid. Instead, they are dependent in a very complicated way upon various nuisance parameters, which result in correlations across individual units (Chang, 2002). As noted in Cerrato (2001), cross sectional dependence can be caused by different factors such as model misspecification or common shocks. Failure to take into consideration cross-sectional dependence between the series may cause significantly biased results (Breusch and Pagan, 1980; Pesaran, 2004). In terms of econometrics, cross-sectional dependency could be explained as individuals forming panels are related to error terms in the panel data model, as indicated in Eq. 2. That is, if individuals forming a panel are affected by a shock, other individuals of the panel are affected as well.

yit = αi + βi xit + εit

(2)

cov(εit , εij ) ≠ 0. A number of Lagrange multiplier (LM) tests are available to check cross-sectional dependency (CD), namely, CDLM1, CDLM1adj, CDLM2 and CDLM. All CDLM tests come with the null hypothesis of no cross sectional dependency across units. The first two LM tests, i.e. CDLM1 and CDLM1adj tests, are developed by Breusch and Pagan (1980) and Pesaran and Yamagata (2008), respectively. These tests are useful when N is fixed and T goes to infinity (T4 N). The other two LM tests, which are CDLM2 and CDLM tests, are proposed by Pesaran (2004). However, the CDLM2 test is useful when T and N are larger enough (Guloglu and Ivrendi, 2008). Meanwhile, the CDLM test is better to use when N is larger and T is smaller (T oN), which is the case of this study. As such, this study employs the Lagrange multiplier CDLM test to check cross sectional dependency. In the second step, panel unit root tests are conducted. There are two groups of panel unit root tests developed in literature. The first group consists of first generation unit root tests that ignore cross-sectional dependence. The second group includes second generation unit root tests that allow for cross-sectional dependence (e.g., Phillips and Sul, 2003; Bai and Ng, 2004; Moon and Perron, 2004; Smith et al., 2004; Pesaran, 2007). This study employs the IPS unit root test by Im et al. (2003). The test removes cross-sectional averages from the data to control for the cross-sectional correlation (as suggested by Levin et al., 2002). IPS test is used as it is suitable for panel with N → σ and T is fixed. The third step involves investigating the long-run relationships between TO, CE, GDP and GDP2. We perform panel cointegration tests on the three following sub-samples. The first sub-sample includes all the 98 countries in the global sample. The second subsample consists of only high-income countries, the third panel includes only middle-income countries, and the fourth panel comprises low-income countries. This study uses the panel cointegration tests developed by Westerlund (2007) and Persyn and Westerlund (2008). Two different classes of tests can be used to evaluate the null hypothesis of no cointegration and the

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

49

Table 1 Summary of the Variables, 1990–2013.

In levels All countries East asia & Pacific Europe & Central asia Latin america & Caribbean Middle east & North africa North America South Asia Sub-Saharan africa Low-income countries Middle-income countries High-income countries In growth rates All countries East Asia & Pacific Europe & Central asia Latin America & caribbean Middle East & North Africa North America South Asia Sub-Saharan Africa Low-income countries Middle-income countries High-income countries

CO2 (metric tons per capita)

PM10 (micrograms per cubic meter)

GDP per capita (constant 2005 TO (TRADE/ US$) GDP)

EXPORT/GDP IMPORT/GDP

4.352 7.021 8.034 2.106 3.374 17.878 0.654 0.983 0.206 2.174 9.115

52.498 46.251 37.246 53.379 70.395 23.588 125.230 56.630 58.160 63.458 34.527

12080 15764 29001 4484 4579 36529 724 1643 363 2979 29590

84.623 125.100 88.556 69.134 92.035 45.954 43.624 78.754 53.369 79.001 105.495

41.590 64.987 42.180 39.057 44.526 21.961 18.099 35.197 21.006 40.368 51.879

44.362 57.545 40.930 45.886 54.659 21.781 23.648 42.371 34.323 44.767 48.106

0.59%  2.74%  4.88%  2.44%

 5.62%  6.04%  6.67%  5.19%

2.20% 3.40% 1.70% 2.10%

2.00% 1.60% 2.40% 1.20%

3.00% 3.10% 3.50% 3.50%

3.50% 3.00% 3.50% 4.70%

 3.12%

 5.32%

2.40%

0.90%

1.80%

0.30%

 5.16%  0.80% 11.08%  2.88% 4.46%  3.23%

 7.65%  4.38%  4.81%  4.31%  5.25%  6.70%

1.40% 3.80% 1.90% 1.50% 2.30% 2.00%

1.60% 2.30% 2.80% 3.80% 1.80% 1.50%

2.10% 3.10% 2.50% 4.10% 3.10% 2.50%

2.70% 2.30% 3.90% 4.20% 4.00% 2.70%

alternative hypothesis: group-mean tests and panel tests. Westerlund (2007) develops four panel cointegration test statistics (Ga, Gt, Pa and Pt) based on the Error Correction Model (ECM). These four test statistics are normally distributed. The two tests (Gt, Pt) are computed with the standard errors of the parameters of the Error Correction (EC) estimated in a standard way, while the other statistics (Ga, Pa) are based on Newey and West (1994) standard errors, adjusted for heteroskedasticity and autocorrelations. By applying an Error-Correction Model in which all variables are assumed to be I(1), the tests proposed by Westerlund (2007) test the absence of cointegration, by determining whether errorcorrection is present for individual panel members and for the panel as a whole. Depending on the presence of cointergration, we estimate the long-run parameters in the cointegrating vector. In cross-sectional analysis, the error variance is likely to vary across the groups, affecting the consistency of the estimators. Using the generalised least squares method (GLS) in the estimation could address this issue. However, other sources of variance variability might still exist, for example the correlation of the squared residuals with the regressors in each group. There are two sources of within-group heteroskedasticity, which could be given either by differences in the unconditional variance of the residual terms or by differences in the variance of the residual terms conditioned on the regressors. For this purpose, we use an efficient estimator which uses the generalized method of moments (GMM) to control for both heteroskedaticity sources. Considering the model:

Yit = α +

Xit′ β

+ δi + γt + εit

M

M

∑i = 1 gi(β ) = ∑i = 1 Zi′εi(β )

S(β ) = (∑

(4)

M

Zi′εi(β ))′W (∑

i=1

where

M

i=1

Zi′εi(β )) = g (β )′Wg (β )

(5)

Zi′ is the instrument matrix for the i-th cross-sec-

tion, εi(β ) = (Yit − α − Xit′ β )and W is a weighting matrix. In addition, desirable properties of the GMM and system GMM estimators hold asymptotic for large N, particularly more suitable when N 4T, as in this study. Meanwhile, GLS is more suitable for samples with a small number of cross-sectional units. This gives rise to an important reason for choosing the GMM method as the main method in this study. The final step is conducting panel short-run and long-run causality tests. To determine the direction of Granger causality among the variables in both the long-run and the shot-run, a panel-based error correction model is employed, following the two steps of Engle and Granger (1987). The study first estimates the long-run parameters in Eq. 1c via the FMOLS estimator to obtain the residual. It then defines the first-lagged residual as the error correction term and estimates the following dynamic error correction models: m

m

m

∑ γ11ikΔPMit − k + ∑ γ12ikΔTOit − k + ∑ γ13ikΔGDPit − k

ΔPMit = γ1i +

k=1

k=1

k=1

m

+

γ14ikΔGDPit2− k

∑

+ ϕ1iECTit − 1 + ε1it

(6)

k=1

(3)

where i = 1, N , t = 1, T , Y is a dependent variable, α is a constant, X is a vector of explanatory variables, β represents a vector of coefficients to be estimated, εit represents the residual terms, δi and γt are the cross-section and, respectively period fixed or random effects. The GMM estimator is computed based on the following:

g (β ) =

and solves the following minimisation problem, function of β :

m

ΔTOit = γ2i +

m

m

∑ γ21ikΔPMit − k + ∑ γ22ikΔTOit − k + ∑ γ23ikΔGDPit − k k=1

k=1

k=1

m

+

∑ γ24ikΔGDPit2− k + ϕ2iECTit − 1 + ε2it k=1

(7)

50

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

Statistic Fixed effect 10.878***

Statistic Random effect 15.997***

Statistic Fixed effect 8.945***

Statistic Random effect 12.719***

Statistic Fixed effect 22.516***

Statistic Random effect 32.230***

Statistic Fixed effect 8.945***

Statistic Random effect 12.719***

deviations from long-run equilibrium. Our empirical strategy follows several previous studies which employed the EKC hypothesis in their examination (for example, Lean and Smyth, 2010; Wang et al., 2011; Hamit-Haggar, 2012; Ozcan, 2013). With respect to the estimation method, Eqs. (6) and (7) are estimated using Arellano–Bond (1991)’s linear dynamic panel-data estimation, and Eqs. (8) and (9) are estimated using Zellner (1962; 1963) and Zellner and Huang (1962)’s seemingly unrelated regression as this allows joint tests. Subsequently, the Wald test is used to determine the short-run and long-run causality.

Statistic Fixed effect 19.486***

Statistic Random effect 27.235***

Statistic Fixed effect 8.776***

Statistic Random effect 11.412***

4. Empirical results and discussions

Table 2 Results of Bias-adjusted LM Test of Error Cross-section Independence. PM10 Trade Test CDLM Export Test CDLM Import Test CDLM

CO2

H0: cross-sectional independence. ***

indicates rejection of the null hypothesis at the 1% significance level. m

ΔGDPit = γ3i +

k=1 m

+

k=1 m

∑ γ33ikΔGDPit − k + ∑ γ34ikΔGDPit2− k + ϕ3iECTit − 1 + ε3it k=1

k=1

m

ΔGDPit2 = γ4i +

(8)

m

∑ γ41ikΔPMit − k + ∑ γ42ikΔTOit − k k=1

m

+

m

∑ γ31ikΔPMit − k + ∑ γ32ikΔTOit − k

k=1 m

∑ γ43ikΔGDPit − k + ∑ γ44ikΔGDPit2− k + ϕ4iECTit − 1 + ε4it k=1

k=1

(9)

where the term Δdenotes the first difference, m is the lag length set at three, which is based on Akaike information criterion. ECT is the error-correction term, ϕj, i (j¼1, 2, 3, 4) is the adjustment coefficient, and εjit is the disturbance term presumed to be uncorrelated with zero means. The error correct term in Eqs. (6–9) can be derived from the long-run equilibrium Eq. (1c) as follows:

^ ECTit = PMit − ν^i − η^1iTOit − η^2iGDPit − η^3iGDPit2

(10)

Two sources of causation can be derived from the estimation of the dynamic error correction models as in Eqs. (6–9), including the short-run causality and long-run causality. In terms of short-run Granger causality represented in Eq. 6, causality runs from ∆TO to ∆PM if the null hypothesis: γ12ip=0, ∀ ip is rejected, while causality runs from ∆GDP and ∆GDP 2 to ∆PM if the joint null hypothesis: γ13ip = γ14ip=0, ∀ ip is rejected via a Wald test. In Eq. 7, causality runs ∆PM to ∆TO if the null hypothesis: γ21ip=0, ∀ ip is rejected, whereas causality runs from ∆GDP and ∆GDP 2 to ∆TO if the joint null hypothesis: γ23ip = γ24ip=0, ∀ ip is rejected. In Eqs. 8 and 9, due to the presence of two variables related to GDP growth (∆GDP and ∆GDP 2) in the system, we need to make cross-equation restrictions to determine the causality from either ∆PM or ∆TO to GDP growth via a likelihood ration (LR) test. Causality from ∆PM to ∆GDP and ∆GDP 2 is supported if the null hypothesis: γ31ip=0, ∀ ip and

γ41ip=0, ∀ ip is rejected. Similarly, causality from ∆TO to ∆GDP and

∆GDP 2 is supported if the null hypothesis:

γ32ip=0, ∀ ip and

γ42ip=0, ∀ ip is rejected. With regard to long-run causality in Eq. 6, if the null hypothesis: ϕ1i=0, ∀ i is rejected, then ∆PM responds to deviations from long-run equilibrium. In Eq. 7, if the null hypothesis: ϕ2i=0, ∀ i is rejected, then ∆TO responds to deviations from long-run equilibrium. Finally, in Eqs. 8 and 9, if the null hypothesis: ϕ3i = ϕ4i=0, ∀ i is rejected, then ∆GDP and ∆GDP 2 jointly respond to

In this section, we report and discuss the results of our empirical analysis. In the first stage of the analysis, we test for crosssectional dependence. Table 2 reports the results of conducting the Lagrange multiplier CDLM test. The CD test statistic strongly rejects the null hypothesis of no cross-sectional dependence. This implies the presence of crosssectional dependence under a fixed effect (FE) specification. Following Baltagi (2005), we also report the results of the model using the random effect (RE) estimator.4 The finding is the same as the case of FE estimator and qualitatively robust to different measures of trade openness and pollutant emissions. Thus, we proceed by conducting panel unit root tests that take cross-sectional dependence into account. The results of conducting IPS (2003)’s panel unit root test are reported in Table 3. The unit root statistics reported are for the logged variables in level and in first difference. For the variables in level, the results reveal that we can accept the null hypothesis of a unit root for all the variables at conventional significance levels. Meanwhile, when taking the first difference, the test rejects the null hypothesis at the 1% level for all the variables. The third step of the empirical study involves investigating the long-run relationship between TO, PM, GDP and GDP2, using the panel cointegration tests developed by Westerlund (2007) and Persyn and Westerlund (2008). As these results in Table 2 strongly indicate the presence of common factors affecting the cross-sectional units, the study bootstraps robust critical values for the test statistics. Table 4 reports the within and between dimension results of the panel cointegration tests. The results indicate that we can reject the null hypothesis of no cointegration at conventional significance levels. We thus conclude that the variables TO, PM, GDP and GDP2 move together in the long run. This result holds across different income groups of countries. The results are robust to different measures of trade openness and environmental quality. The finding of a long-term relationship among the variables could be explained due to the important channels through which trade can affect the environment, including scale, technique, and composition effects proposed by Copeland and Taylor (2004) and Antweiler et al. (2001), reviewed thoroughly in Section 2. The overall environmental effect of trade on economic growth depends on the net result of all three effects. Our finding is consistent with previous studies in literature which also found evidence of a longterm relationship between trade, income and the environment 4 We conducted the Hausman test (with random effects is preferred under the null hypothesis due to higher efficiency, while under the alternative, fixed effects is at least consistent and thus preferred) for all models in our study. The results suggested fixed effects are preferred for all the models, regardless of the different measures of trade openness and environmental quality. Nevertheless, we also conducted and present the results for random effects in Table 2 for robustness checks. The Hausman test results are available upon request.

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

Table 3 Panel Unit Root Test Results. Intercept case Test

Level IPS Statistic

1st difference IPS Statistic

TO GDP GDP2 IM/GDP EX/GDP CE PM

 1.169 4.8483 8.286  0.449 0.723 2.985 1.444

 8.329***  4.6323***  4.942***  7.267***  6.276***  3.933***  6.745***

Note: Ho: All panels contain unit roots. *** indicates rejection of the null hypothesis at the 1% significance level. Crosssectional correlation is controlled by removing cross-sectional means (Levin et al., 2002). The number of lags for each panel is chosen by minimising the AIC, subject to a maximum of 4 lags.

(e.g., Halicioglu, 2009; Boulatoff and Jenkins, 2010; Shahbaz et al., 2013; Mehrara et al., 2014). However, most of these studies are individual country studies while our paper covers a global sample of 98 countries. The implication of the cointegration test results is that there is a long-run relationship between PM, TO, GDP and GDP2 for a cross section of the countries. This result is robust to different measures of trade openness as well as environmental quality. Given the presence of cointergration, this study estimates the long-run

51

parameters in the cointegrating vector using GMM panel estimation techniques that allow for estimating heterogeneous cointegrated vectors.5 The results are shown in Table 5. Since all the variables are expressed in natural logarithms, the coefficients can be interpreted as long-run elasticity estimates. The results show that, increased openness to trade causes environmental degradation for the global sample. For the panel of high-income countries, however, increased openness to trade benefits the environment. One possible interpretation is that increased foreign trade results in income gains which enable some countries to specialise in relatively clean industries (Copeland and Taylor, 2013). Trade is beneficial primarily because of its effect on wealth. Trade increases wealth and therefore environmental quality through a technique effect that improves production efficiency. Furthermore, greater wealth increases investment in environmental protection (Bhagwati, 2000). This result is consistent with the finding in Baek et al. (2009) that trade has a positive effect on environmental quality in developed countries. However, Baek et al. (2009) use a time series dataset, while in this study we use a global panel dataset which includes twice as many countries. On the other hand, for middle and low-income countries, we find that the environmental impact of trade was significantly negative. The results are similar to Baek et al. (2009), which finds that trade has detrimental effects on environmental quality in most developing countries. Overall, our results could be explained by the theoretical framework for the trade-environment nexus proposed by Tayebi and

Table 4 Westerlund (2007) Panel Cointegration Test Results. Trade/GDP as TO measure Statistic

All countries Intercept PM10 Gt  4.133*** Ga 9.415 Pt  26.096*** Pa 0.715*** CO2 Gt  0.709** Ga 9.855 Pt  1.829** Pa 4.534* Export/GDP as TO measure Statistic All countries Intercept PM10 Gt  3.065*** Ga 9.802 Pt  14.049*** Pa 2.912** CO2 Gt  1.164** Ga 9.457** Pt  1.311** Pa 4.001** Import/GDP as TO measure Statistic All countries Intercept PM10 Gt  3.824*** Ga 9.655* Pt  16.679*** Pa 2.913*** CO2 Gt 0.134* Ga 9.728* Pt  1.594** Pa 4.39**

Intercept and trend  0.598* 15.259*  13.689*** 10.128**  0.462* 14.754*  0.242* 11.052**

High income Intercept Intercept and trend  0.34 1.643** 4.065 6.594 20.293*** 23.954*** 1.977*** 7.045*  0.34 1.549** 4.065 6.378 10.293** 11.751** *** 1.977 5.146*

Middle income Intercept Intercept and trend  4.607***  2.606** * 6.2 10.45  31.735***  20.291***  1.846*** 5.873** 1.832 2.385** 7.142 10.609*  11.517*** 11.705*** 2.779* 8.139*

Low income Intercept  1.298** 5.834 2.232 4.87**  2.974*** 5.426  11.025*** 12.854***

Intercept and trend  2.56**  3.331  15.74***  11.157**  3.298***  4.392  20.127***  16.007***

Intercept and trend  0.345* 15.033*  12.77*** 8.615**  0.084* 14.78** 0.895* 10.975**

High income Intercept Intercept and trend 0.112 1.719** 4.392 6.679 10.557*** 11.554*** 2.412** 5.332**  0.702  0.55 4.179* 6.387* 2.12* 1.473* 2.567* 15.101**

Middle income Intercept Intercept and trend  3.927***  1.528* 6.782 10.452*  17.072***  14.193*** 0.326*** 4.354*** 1.495 3.204 7.086** 10.776**  0.764 3.394* 2.966* 8.198**

Low income Intercept  0.603* 5.587 12.131** 4.547**  3.277*** 4.787**  3.384** 1.142**

Intercept and trend  2.691**  3.92  10.603**  2.172  3.584** 7.927  7.368*** 4.58***

High income Intercept 1.252* 4.575  10.16** 2.356** 1.473* 4.537 1.869* 2.783*

Middle income Intercept  5.847*** 6.183*  20.992***  0.668*** 0.654* 6.818  2.067** 2.907*

Low income Intercept  0.371* 5.921 2.656* 5.101**  0.371* 5.921 2.656* 5.101**

Intercept and trend 1.335* 8.984* 7.235** 8.537**  3.536*** 7.84  9.843*** 4.027***

Intercept and trend  0.78* 15.341*  8.221** 10.803* 0.283 14.88* 0.513* 11.311***

Intercept and trend 3.568** 6.763 12.404** 15.134** 1.652** 6.296 3.523* 4.958**

Intercept and trend  4.249*** 10.469*  12.117*** 6.146** 2.455 11.042* 3.169* 8.675**

Note: H0: no cointegration. * indicate rejection of the null hypothesis at the 10% significance level. Using the bootstrap approach of Westerlund (2007) to account for cross-sectional dependence, the number of replications is 400. ** indicate rejection of the null hypothesis at the 5% significance level. Using the bootstrap approach of Westerlund (2007) to account for cross-sectional dependence, the number of replications is 400. *** indicate rejection of the null hypothesis at the 1% significance level. Using the bootstrap approach of Westerlund (2007) to account for cross-sectional dependence, the number of replications is 400.

52

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

Table 5 Estimation Results of Models with Different Measures of Trade Openness and Environmental Quality. All

(2) PM

(3) PM

ΔGDP ΔGDP2

0.739*** (15.00)

0.140* (2.16)  0.171 (  1.32) 0.006 (0.48)

0.148* (2.26)  0.168 (  1.77)

ΔTO ΔGDP ΔGDP2

4.018 (5.46) 1937 GMM

4.158 (5.69) 1925

(1) PM

(2) PM

(3) PM

ΔTO ΔGDP ΔGDP2

0.800*** (20.41)  0.201** (  2.90) 0.095 (1.73)  0.009 (  1.08)

0.789*** (19.39)

0.790*** (19.74)

0.273* (2.21) 0.00821 (0.12)

 2.684* (  2.46) 282

 0.108* (  2.58)  2.335* (  2.42) 282

(1) PM

(2) PM

(3) PM

(5) CE 0.313** (2.77)

(6) CE 0.312** (2.74)

0.678*** (11.41) 0.109** (2.92) 0.104 (1.45)  0.123* (  1.90)

0.682*** (11.73)

0.678*** (12.14)

0.696 (1.70)  0.260 (  0.92) 0.305* (1.98)

0.752 (1.71)  0.279 (  0.97)

0.140 (1.81)  0.001 (  0.86)

3.616*** (5.23) 960 GMM

4.067*** (5.09) 954

0.161** (2.98) 3.989*** (4.63) 954

(1) PM

(2) PM

(3) PM

0.842*** (33.21) 0.053***

0.817*** (30.90)

0.820*** (28.71)

ΔCE(  1) ΔPM(  1) ΔTO

(4) CE 0.315** (2.77)

0.309* (2.16) 0.728 (1.84)  0.279 (  1.01)

**

1.646 (0.39) 912

1.278 (0.31) 908

0.203* (2.33) 1.498 (0.36) 908

(4) CE 0.622*** (4.68)

(5) CE 0.600*** (4.05)

(6) CE 0.600*** (4.07)

0.139**

(3) PM

(4) CE

0.039 (1.21)  0.063 (  1.00) 0.078*** (4.08)

0.044 (1.17)  0.035 (  1.68)

(2.73) 1.793* (2.12)  1.218* (  2.00)

_cons N

6.294*** (4.86) 679

6.538*** (5.05) 678

0.060*** (4.37) 6.285*** (5.10) 678

15.62 (1.94) 664

(5) CE

(6) CE

1.827* (2.33)  1.165* (  2.18) 0.218** (2.83)

1.710* (2.18)  1.118* (  2.00)

14.11* (2.01) 664

0.148* (2.07) 13.78 (1.85) 664

Note:

0.236 (1.90) 0.0355 (0.49)  0.040* (  2.40)

ΔIM/GDP

N Low-income countries

0.204 (1.71) 0.117 (1.22)

0.0885 0.108 (1.55) (1.83)  0.0775*  0.0966** (  2.33) (  2.94)  0.602* (  2.31)  0.105 (  2.89)** *** 2.132 2.384***  2.553** (4.08) (4.76) (  2.79) 293 293 284

**

 0.109* (  2.40) 0.238* (1.98) 0.0258 (0.40)

0.140 (1.86)  0.195* (  2.00) 0.209* (2.50)

ΔGDP2

(3.82) 0.053 (1.38)  0.042 (  1.37)

(2) PM

ΔIM/GDP 0.167 (1.38) 0.114 (1.18) 0.013 (0.37)

(6) CE 0.744*** (16.87)

2.345*** (4.83) 298 GMM

ΔGDP

ΔEX/GDP

0.607** (3.55) 0.169 (1.44) 0.127 (1.32)

(5) CE 0.740*** (16.98)

ΔEX/GDP

_cons

0.611 (23.00)

(4) CE 0.740*** (17.07)

***

ΔCE(  1) ΔPM(  1)

0.628 (24.62)

 4.057 (  2.62) 1854

ΔIM/GDP

N Middle-income countries

***

 4.104 (  2.67) 1860

***

ΔEX/GDP

_cons

***

0.552** (2.76)  4.192** (  2.73) 1854

ΔCE(-1) ΔPM(-1)

(6) CE

0.729*** (16.25)

0.203** (2.60) 4.121*** (5.44) 1925

ΔIM/GDP

N High-income countries

(5) CE

0.621 (23.82)

ΔEX/GDP

_cons

(4) CE ***

0.736*** (15.47) 0.105** (2.97) 0.131* (2.07)  0.016 (  1.73)

GMM (1) PM

ΔCE(  1)

ΔTO

All

GMM (1) PM

ΔPM(  1)

Table 5 (continued )

* indicate rejection of the null hypothesis at the 10% significance level. t statistics are in parentheses. ** indicate rejection of the null hypothesis at the 5% significance level. t statistics are in parentheses. *** indicate rejection of the null hypothesis at the 1% significance level. t statistics are in parentheses.

Younespour (2012). The framework posits that international trade has an impact on environmental quality that varies with the comparative advantage of a country. They argue that capital abundance plays a critical role in determining comparative advantage. Specifically, all else equal, if a country is relatively capital abundant (sufficiently rich), it should export the capital intensive (polluting) good. Meanwhile, a poor country is normally relatively labour abundant, and hence it should export the labour intensive (clean) good. In this context, our results support the argument by Copeland and Taylor (2013) that dirty industry migration transfers the pollution burdens of developed countries’ consumption to developing countries. These findings are robust across different measures of trade openness. Further, for the global sample as well as different income groups of countries, we could not find evidence of an inverted U-shaped pattern between particulate matter emissions and income per capita at 5% significance level. The results are robust to different measures of trade openness as well as environmental quality and hence refute the EKC hypothesis. In fact, Suri and Chapman (1998) contend that the developed countries shift the production of pollution-intensive goods to developing countries, thus reducing their own emissions. Therefore, factoring in international trade weakens the EKC. One reason EKC may not be observed in some cases is that even though there may exist an underlying relationship between pollution levels and income, the observable indicators of environmental quality may continue to worsen due to stock effects. Furthermore, the existence of EKC may depend upon the type of environmental indicators chosen. This explains the conflicting nature of observations made by studies in the past. A consequential implication of this explanation is that perhaps one must not put too much emphasis on proving the existence of EKC, because it may give misleading policy implications. Pollution may indeed decline over time with technological advances or increasing preferences for environmental qualities, but there may be 5 Before the GMM estimator, the Di Iorio and Fachin (2007)’s test for breaks in cointegrated panels is performed to examine the stability of the relationship between the variables of interest. The results show that we can accept the null hypothesis of no break. That is, the relationship among the investigated variables is stable and not subject to structural breaks during the investigation period. To conserve space, the results are not presented here but they are available upon request.

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

Table 6 Long-run and Short-run Causality Test Results. All Environmental quality indicator PM10

Dependent variables

Sources of causation

2

ΔGDP, ΔGDP ΔPM ΔTO CO2

ΔGDP, ΔGDP2 – 7.87* 32.86*** ΔGDP, ΔGDP2 – 34.52*** 35.46***

Shortrun ΔPM 5.39 – 11.16** Shortrun ΔCE 11.97** – 6.07**

ΔGDP, ΔGDP2 ΔCE ΔTO High-income countries Environmental Dependent Sources of causation quality indicator variables ShortPM10 ΔGDP, run ΔPM ΔGDP2 2 ΔGDP, ΔGDP – 2.23 ΔPM 7.29 – ΔTO 5.45 13.22*** CO2 ΔGDP, ShortΔGDP2 run ΔCE ΔGDP, ΔGDP2 – 2.47 ΔCE 2.75 – ΔTO 4.60 12.49*** Middle-income countries Environmental Dependent Sources of causation quality indicator variables ShortPM10 ΔGDP, run ΔPM ΔGDP2 ΔGDP, ΔGDP2 – 4.79 *** ΔPM 21.14 – ΔTO 22.13*** 11.04*** CO2 ΔGDP, ShortΔGDP2 run ΔCE ΔGDP, ΔGDP2 – 8.75* ΔCE 5.68 – ΔTO 25.55*** 10.07*** Low-income countries Environmental Dependent Sources of causation quality indicator variables PM10 ΔGDP, ShortΔGDP2 run ΔPM ΔGDP, ΔGDP2 – 3.99 ΔPM 12.81*** – ΔTO 34.32*** 10.77*** CO2 ΔGDP, ShortΔGDP2 run ΔCE ΔGDP, ΔGDP2 – 45.29*** ΔCE 18.57*** – ΔTO 25.52*** 10.80***

ΔTO ***

39.32 14.48*** – ΔTO

39.98*** 16.69*** –

Long run ECT(  1) 10.95*** 144.99*** 1.84* Long run ECT(-1) 6.52** 935.61*** 5.82**

53

relationship running from ΔGDP and ΔGDP2 to ΔPM and ΔTO. The results also suggest that there is a feedback effect-i.e. bidirectional causal relations-between trade openness and particulate matter emissions for the whole sample as well as different income groups of countries. This finding is maintained using carbon emissions (CE) as an indicator of environmental quality. In addition, the presence of a long-run causality is indicated by the statistical significance of the estimated coefficient for one period of lagged error correction term. The consistent finding across the world sample and sub-samples of different income groups suggests long-rum causality running from ΔGDP and ΔGDP2 and ΔTO to ΔPM.

5. Policy implications ΔTO 15.10*** 9.56*** – ΔTO 14.41*** 6.02** –

ΔTO 12.15** 6.72* – ΔTO 11.20** 5.97* –

ΔTO 40.41*** 12.74*** – ΔTO 51.17*** 16.32*** –

Long run ECT(-1) 12.01*** 62.55*** 2.47** Long run ECT(-1) 5.92* 58.70*** 1.03*

Long run ECT(-1) 8.00** 495.68*** 10.96*** Long run ECT(  1) 6.20** 453.95*** 2.08*

Long run ECT(  1) 36.85*** 28.72*** 30.19*** Long run ECT(  1) 20.57*** 162.06*** 13.21***

Note: χ2 statistics are reported for the Wald test. *

indicate rejection of the null hypothesis at the 10% significance level. indicate rejection of the null hypothesis at the 5% significance level. *** indicate rejection of the null hypothesis at the 1% significance level. **

stock effects from emissions which are irreversible and subject to hysteresis (Ranjan et al., 2003). The existence of a long-run cointegration vector necessitates the exploration of Granger causality. Table 6 summarises the causality estimates for the global sample and separately for lowincome, middle-income, and high-income countries. The optimal lag structure is set to three and the significance of the causality tests is determined by the Wald F-test. For the global sample, the results indicate no short-run relationship running from particulate matter emissions (ΔPM) to economic growth (ΔGDP and ΔGDP2). Based on the statistically significant and positive coefficients of ΔGDP and ΔGDP2 in the ΔPM and ΔTO equations of the world and low-income country sub-samples, one may conclude that there is a short-run transitory

Interestingly and significantly, our central finding that trade benefits the environment in high-income countries but harms the environment in low- and middle-income countries is consistent with the popular notion that rich countries dump the pollution associated with their consumption on poor countries. The former have outsourced a lot of their environmentally dirty manufacturing activities to the latter but often remain the major market for the manufacturing output of the latter. For example, foreign direct investment from advanced economies has helped to transform China into a highly polluted factory of the world, which exports much of what it produces back to the advanced economies. Therefore, our evidence lends some support to calls for rich countries to provide assistance for the efforts of rich countries to tackle pollution. To the extent that rich countries are outsourcing pollution, there is a case for their contributing to the clean-up of the outsourced pollution. International environmental cooperation is vital for mitigating trans-boundary pollution and other negative spill-over effects. Furthermore, cooperation based on technology sharing can promote economically beneficial efficiency gains and innovations (Beghin et al., 1994). Some countries view incorporating environmental provisions into trade agreement as the most effective way to protect the global environment. These countries include Canada, the US, the European Union, and New Zealand. Among developing countries, the efforts of Chile to include comprehensive environmental chapters in its trade agreements are noteworthy (OECD, 2007). Trade agreements can strengthen the capacity for governments to address environmental issues. In particular, the reduction of trade barriers on environmental goods can lead to increased access to green technologies at lower cost. For instance, the Trans-Pacific Partnership (TPP) agreement is expected to help developing countries shift into cleaner industries and transition to low-carbon pathways by providing access to green goods, services, and investments (Meltzer and Tania Voon, 2014). On the other hand, some economists believe that trade agreements are unsuitable for tackling environmental issues. The problem is that not all countries face the same incentives to join trade agreements with environmental provisions. As such, environmental issues may hinder agreement between the negotiating parties. This is particularly relevant to trade negotiations between high-income countries on one hand and low- and middle-income countries on the other. Low- and middle-income countries often face many practical constraints in incorporating the environment into trade negotiations. These include a lack of environmental knowledge among negotiators, opposition from higher levels of governments, insufficient coordination among trade and environment ministries, and underdeveloped national environmental legislation. In such scenarios, the outcome of the negotiations many depend heavily on economic size and power, which is likely to be highly asymmetric. Furthermore, strong political

54

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

will, which is often lacking, is required to negotiate environmental provisions in trade agreements. In principle, policies can be designed as if the world was a single country, using a common charge per unit of emissions that reflects the cost of global warming. Though such international agreements might be fraught with free-riding and monitoring problems, these could be solved by trade sanctions (Beghin et al., 1994). However, a more effective way is to provide stronger incentives to cooperate through international transfers, which introduce a tangible cost for non-compliance. For instance, the TPP parties have adopted a range of renewable energy subsidies (Meltzer and Tania Voon, 2014). The Australian government provides support for the development of renewable energy such as solar power in order to meet the target of generating 20% of its electricity from renewable energy sources by 2020. The TPP also allows for subsidies for research and development on green energy. However, green subsidies might distort resource allocation so it is important to take such inefficiencies into account. In sum, successful negotiation of environmental provisions in trade agreements requires extensive preparation, close coordination among trade and environmental actors, setting of priorities, and reconciling conflicting interests. Environmental provisions are not a one-off but instead require continuous efforts to ensure effective integration of trade and environmental issues throughout the trade agreement. In this connection, developing countries would benefit from external support, especially in terms of financial resources and capacity building, either from their developed country trade partner or from other institutions such as development co-operation agencies (OECD, 2007).

6. Conclusion and policy implications Trade is an engine of economic growth. However, precisely for that reason, it is often argued that increased trade can have detrimental effects on the environment. Global warming, climate change and more generally, deterioration of environmental quality, have emerged as world-wide concerns over the past few decades. But does trade actually harm the environment? Our paper seeks to empirically explore this critical issue, using particulate matter emissions as the measure of environmental quality. More specifically, we present new and comprehensive evidence from a panel dataset of 98 countries encompassing all regions of the world. The period of investigation spans from 1990 to 2013. The use of a global panel data set, along with the incorporation of income level and thus the Environmental Kuznets Curve (EKC) into the analysis, are our main contributions to the literature. We hope that the contributions will enrich our understanding of the tradeenvironment nexus. Our main findings are as follows. First, our evidence could not reject the existence of long-run relationship between trade openness, particulate matter emissions, and income for the global sample or for different income groups of countries. The finding is qualitatively robust to different measures of trade openness. Second, we find that in the long run, increased trade causes environmental degradation for the global sample. However, the results differ depending on country income level. For high-income countries, we find the environmental effect of trade openness to be benign but for middle- and low-income countries, the effect is harmful. The results are generally robust to different proxies of trade openness and environmental quality. Third, we find evidence of a feedback effect between trade openness and particulate matter emissions for the global sample and for different income groups of countries. This finding remains when we use carbon emissions as an alternative indicator of environmental quality.

References Akbostanci, E., Turut-Asik, S., Tunc, G.I., 2009. The relationship between income and environment in Turkey: is there an environmental Kuznets curve? Energy Policy 37, 861–867. Al-Mulali, U., 2011. Oil consumption, CO2 emission and economic growth in MENA countries. Energy 36 (10), 6165–6171. Antweiler, W., Copeland, B., Taylor, S., 2001. Is free trade good for the environment? Am. Econ. Rev. 91, 877–908. Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58, 277–297. Arouri, M.E.H., Youssef, A.B., Mhenni, H., Rault, C., 2012. Energy consumption, economic growth and CO2 emissions in Middle East and North African countries. Energy Policy 45, 342–349. Baek, J., Cho, Y., Koo, W.W., 2009. The environmental consequences of globalization: a country specific time-series analysis. Ecol. Econ. 68, 2255–2264. Bai, J., Ng, S., 2004. A panic attack on unit roots and cointegration. Econometrica 72 (4): 1127-1177Baltagi BH (2005) Econometric Analysis of Panel Data. John Wiley & Sons, New York. Baltagi, B.H., 2005. Econometric analysis of Panel data, 3rd edition. John Wiley & sons. Beghin, J., Roland-Holst, D., Mensbrugghe, D., 1994. A survey of the trade and environment nexus: global dimensions. OECD J. : Econ. Stud. 23, 167–192, Winter 1994. Bernauer, T., Koubi, V., 2009. Effects of political institutions on air quality. Ecol. Econ. 68 (5), 1355–1365. Bhagwati, J., 2000. On thinking clearly about the linkage between trade and the environment. Environ. Dev. Econ. 5 (4), 485–496. Breusch, T.S., Pagan, A.R., 1980. The lagrange multiplier test and its applications to model specification tests in econometrics. Rev. Econ. Stud. 47, 239–253. Brunekreef, B., Holgate, S.T., 2002. Air pollution and health. Lancet 360 (9341), 1233–1242. Boulatoff, C., Jenkins, M., 2010. Long-term nexus between openness, income and environmental quality. Int. Adv. Econ. Res. 16, 410–418. Cerrato, M., 2001. Econometric Approaches to Testing PPP. Mimeo, Department of Economics. London Guildhall University, United Kingdom. Chang, Y., 2002. Nonlinear IV unit root tests in panels with cross-sectional dependency. J. Econom. 100, 261–292. Cole, M.A., Elliot, R.J.R., 2003. Determining the trade environment composition effect: the role of capital, labor and environmental regulations. J. Environ. Econ. Manag. 46, 363–383. Coondoo, D., Dinda, S., 2002. Causality between income and emission: A countrygroup specific econometric analysis. Ecol. Econom. 40 (3), 351–367. Coondoo, D., Dinda, S., 2008. The carbon dioxide emission and income: a temporal analysis of cross-country distributional patterns. Ecol. Econom. 65, 375–385. Copeland, B.R., Taylor, M.S., 2004. Trade, growth, and the environment. J. Econ. Lit. 42 (1), 7–71. Copeland, B.R., Taylor, M.S., 2013. Trade and the Environment: Theory and Evidence. Princeton University Press, New Jersey. Deme, M., 2002. An examination of the trade-led growth hypothesis in Nigeria: A co-integration, causality, and impulse response analysis. J. Develop. Areas 36 (1), 1–15. Di Iorio, F., Fachin, S., 2007. Testing for breaks in cointegrated panels-with an application to the feldstein-horioka puzzle. economics: The open-access. OpenAssess. E-J. 1 (14), 1–23. Dinda, S., Coondoo, D., Pal, M., 2000. Air quality and economic growth: An empirical study. Ecol. Econ. 34 (3), 409–423. Dinda, S., 2004. Environmental kuznets curve hypothesis: A survey. Ecol. Econ. 49 (4), 431–455. Dinda, S., Coondoo, D., 2006. Income and emission: a panel-data based cointegration analysis. Ecol. Econ. 57, 167–181. Engle, R.F., Granger, C.W.J., 1987. Co-integration and error correction: Representation, estimation and testing. Econometrica 55, 251–276. Esty, D.C., 2001. Bridging the trade-environment divide. J. Econ. Perspect. 15 (3), 113–130. Frankel, J., Rose, A., 2005. Is trade good or bad for the environment? Sorting out the causality. R. Econ. Stat. 87 (1), 85–91. Frankel, J., Romer, D., 1999. Does trade cause growth? Am. Econ. Rev. 89, 379–399. Guloglu, B., Ivrendi, M., 2008. Output fluctuations: Transitory or permanent? The case of Latin America. Appl. Econ. Lett. 17 (4), 381–386. Haggar, M.H., 2012. Greenhouse gas emissions, energy consumption and economic growth: a panel cointegration analysis from Canadian industrial sector perspective. Energy Econ. 34, 358–364. Halicioglu, F., 2009. An econometric study of CO2 emissions, energy consumption, income and foreign trade in Turkey. Energy Policy 37, 1156–1164. Hamit-Haggar, M., 2012. Greenhouse gas emissions, energy consumption and economic growth: A panel cointegration analysis from Canadian industrial sector perspective. Energy Econ. 34 (1), 358–364. Harris, R.G., Robertson, P.E., Xu, J.Y., 2011. The international effects of China’s growth, trade and education booms. World Econ. 34 (10), 1703–1725. Herrerias, M.J., Orts, V., 2011. The driving forces behind China’s growth. Econ. Transit. 19 (1), 79–124. Hye, Q.M.A., 2012. Long term effect of trade openness on economic growth in case of Pakistan. Qual. Quant. 46 (4), 1137–1149.

T.-H. Le et al. / Energy Policy 92 (2016) 45–55

Im, K.S., Pesaran, M.H., Shin, Y., 2003. Testing for unit roots in heterogeneous panels. J. Econ. 115, 53–74. Jaunky, V.C., 2011. The CO2 emissions-income nexus: evidence from rich countries. Energy Policy 39, 1228–1240. Kellenberg, D.K., 2009. An empirical investigation of the pollution haven effect with strategic environment and trade policy. J. Int. Econ. 78, 242–255. Kim, D., 2011. Trade, growth and income. J. Int. Trade Econ. Dev. 20, 677–709. Kim, D., Lin, S., 2009. Trade and growth at different stages of economic growth. J. Dev. Stud. 45, 1211–1224. Lean, H.H., Smyth, R., 2010. CO2 emissions, electricity consumption and output in ASEAN. Appl. Energy 87 (6), 1858–1864. Lee, H., Verances, J.B., Song, W., 2009. The Tourism-Environment Causality. Int. J. Tour. Sci. 9 (3), 39–48. Levin, A., Lin, C.F., Chu, C., 2002. Unit root tests in panel data: Asymptotic and finitesample properties. J. Econom. 108, 1–24. Liargovas, P.G., Skandalis, K.S., 2012. Foreign direct investment and trade openness: The case of developing economies. Soc. Indic. Res. 106 (2), 323–331. Managi, S., Jena, P.R., 2008. Environmental productivity and Kuznets curve in India. Ecol. Econ. 65, 432–440. Managi, S., Kumar, S., 2009. Trade-induced technological change: analyzing economic and environmental outcomes. Econ. Model. 26, 721–732. Mehrara, M., Karsalari, A.R., Musai, M., 2014. The relationship among Income, openness and environmental quality: An ARDL approach for the case of Iran. Int. J. Acad. Res. Bus. Soc.Sci. 4 (5), 32–40. Meltzer, J.P., 2014. The Trans-Pacific Partnership Agreement, the environment and climate change. In: Tania Voon (Ed.), Trade Liberalisation and International Cooperation: A Legal Analysis of the Trans-Pacific Partnership Agreement. Edward Elgar. Miah, M.D., Masum, M.F.H., Koike, M., Akther, S., Muhammed, N., 2011. Environmental Kuznets Curve: the case of Bangladesh for waste emission and suspended particulate matter. Environmentalist 31 (1), 59–66. Moon, H.R., Perron, B., 2004. Testing for a unit root in panels with dynamic factors. J. Econom. 122 (1), 81–126. Newey, W.K., West, K.D., 1994. Automatic lag selection in covariance matrix estimation. Rev. Econ. Stud. 61, 631–654. Nowak, D.J., Crane, D.E., Stevens, J.C., 2006. Air pollution removal by urban trees and shrubs in the United States. Urban For. Urban Green. 4 (3), 115–123. OECD, 2007. Environment and Regional Trade Agreements. OECD Publishing, Paris. Ozcan, B., 2013. The nexus between carbon emissions, energy consumption and economic growth in Middle East countries: A panel data analysis. Energy Policy 62, 1138–1147. Pao, H.T., Tsaim, C.M., 2010. CO2 emissions, energy consumption and economic growth in BRIC countries. Energy Policy 38, 7850–7860. Persyn, D., Westerlund, J., 2008. Error Correction Based Cointegration Tests for Panel Data. Stata J. 8, 232–241. Pesaran, M.H., 2004. General diagnostic tests for cross section dependence in panels. University of Cambridge, England, p. 0435. Pesaran, M.H., 2007. A simple panel unit root test in the presence of cross-section

55

dependence. J. Appl. Econom. 22, 265–312. Pesaran, U., Yamagata, T., 2008. A bias-adjusted LM test of error cross-section independence. Econom. J. 11, 105–127. Phillips, P.C.B., Sul, D., 2003. Dynamic panel estimation and homogeneity testing under cross section dependence. Econom. J. 6, 217–259. Plassmann, F., Khanna, N., 2006. Preferences, technology, and the environment: understanding the environmental Kuznets curve hypothesis. Am. J. Agric. Econ. 88, 632–643. Ranjan R., Shortle J.S., Marshall E.P. (2003) The Relevance and Implications of the Environmental Kuznets Curve Under Stock Effects and Non-Linearities: A Hysteresis Based Approach. In: 2003 Annual meeting, July 27-30, Montreal, Canada (No. 22147). American Agricultural Economics Association (New Name 2008: Agricultural and Applied Economics Association). Salvador, P., Artíñano, B., Viana, M., Alastuey, A., Querol, X., 2012. Evaluation of the changes in the Madrid metropolitan area influencing air quality: Analysis of 1999–2008 temporal trend of particulate matter. Atmos. Environ. 57, 175–185. Shabaz, M., Tiwari, A.K., Nasir, M., 2013. The effects of financial development, economic growth, coal consumption and trade openness on CO2 emissions in South Africa. Energy Policy 61, 1452–1459. Shandilya, K.K., Khare, M., Gupta, A.B., 2007. Suspended particulate matter distribution in rural-industrial Satna and in urban-industrial South Delhi. Environ. Monit. Assess. 128 (1–3), 431–445. Smith, L.V., Leybourne, S., Kim, T., Newbold, P., 2004. More powerful panel data unit root tests with an application to mean reversion in real exchange rates. J. Appl. Econom. 19, 147–170. Squalli, J., Wilson, K., 2011. A new measure of trade openness. World Econ. 34 (10), 1745–1770. Squalli, J., Wilson, K., Hugo, S., 2010. An analysis of market access. Appl. Econ. 42 (14), 1833–1844. Suri, V., Chapman, D., 1998. Economic growth, trade and the energy: implications for the environmental Kuznets curve. Ecol. Econ. 25, 195–208. Tayebi, S.K., Younespour, S., 2012. The effect of trade openness on environmental quality: Evidence from Iran’s trade relations with the selected countries of the different blocks. Iran. Econ. Rev. 16 (32), 19–40. Wang, S.S., Zhou, D.Q., Zhou, P., Wang, Q.W., 2011. CO2 emissions, energy consumption and economic growth in China: a panel data analysis. Energy Policy 39 (9), 4870–4875. Westerlund, J., 2007. Testing for error correction in panel data. Oxford Bull. Econ. Stat. 69, 709–748. World Development Indicators (WDI) (2014) Retrieved on 10 January 2015. Available at: data.worldbank.org/sites/default/files/wdi-2014-book.pdf. Zellner, A., 1962. An efficient method of estimating seemingly unrelated regressions and tests for aggregation bias. J. Am. Stat. Assoc. 57, 348–368. Zellner, A., 1963. Estimators for seemingly unrelated regression equations: Some exact finite sample results. J. Am. Stat. Assoc. 58, 977–992. Zellner, A., Huang, D.S., 1962. Further properties of efficient estimators for seemingly unrelated regression equations. Int. Econ. Rev. 3, 300–313.