Exploring the link between environmental pollution and economic growth in EU-28 countries: Is there an environmental Kuznets curve?

This study examines the Environmental Kuznets Curve hypothesis (EKC), considering the primary energy consumption among other country-specific variables, for a panel of the EU-28 countries during the period 1990–2014. By estimating pooled OLS regressions with Driscoll-Kraay standard errors in order to account for cross-sectional dependence, the results confirm the EKC hypothesis in the case of emissions of sulfur oxides and emissions of non-methane volatile organic compounds. In addition to pooled estimations, the output of fixed-effects regressions with Driscoll-Kraay standard errors support the EKC hypothesis for greenhouse gas emissions, greenhouse gas emissions intensity of energy consumption, emissions of nitrogen oxides, emissions of non-methane volatile organic compounds and emissions of ammonia. Additionally, the empirical findings from panel vector error correction model reveal a short-run unidirectional causality from GDP per capita growth to greenhouse gas emissions, as well as a bidirectional causal link between primary energy consumption and greenhouse gas emissions. Furthermore, since there occurred no causal link between economic growth and primary energy consumption, the neo-classical view was confirmed, namely the neutrality hypothesis.


Introduction
Energy is a crucial resource for any economy since all production and consumption undertakings are directly connected to energy consumption, hence ensuring the basis for economic activity and social welfare; however, the use of fossil fuels as primary source of energy caused a noteworthy upsurge in the global emissions of several potentially harmful gases [1]. Greenhouse gases (GHGs) absorb heat arriving from the Sun and retain it in the atmosphere which a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 discussions are shown in Section 4, while the final section concludes the paper and provides policy implications.
period 2000-2013 through panel generalized method of moments (GMM). [36] applied panel unit root tests and co-integration techniques and did not confirm the EKC for 12 Middle East and North African Countries (MENA) over 1981-2005, except for Jordan.
[58] used co-integration and Granger causality methods and revealed that EKC is not valid for ASEAN-5 economies, particularly for Indonesia, Malaysia, and Thailand. The results of the FMOLS panel estimator employed by [59] for 12 Middle East countries over 1990-2008 yielded evidence divergent to the EKC.
[60] estimated pooled, fixed effects, random effects, and generalized-least square regressions, but did not validate the EKC for 152 states over six years. Furthermore, contradictory evidence towards EKC hypothesis is also established in single country studies. [61] found that nuclear energy has a positive influence on environmental quality in Korea over 1971-2007 by employing the autoregressive distributed lag (ARDL) approach to co-integration. By using threshold co-integration tests, [62] revealed that EKC hypothesis is valid in India's case during 1971-2008. [63] underlined the validity of EKC in Turkey over 1961-2010 using the ARDL method. [64] used the Spatial Durbin Model and supported the presence of the EKC for per capita coal consumption in China between 1995 and 2012. [65] validated the EKC in Indonesia for the period 1971-2010 by utilizing ARDL method. In the case of Qatar, [66] employed the ARDL method over the timeframe 1980-2011 and rejected the EKC hypothesis when using carbon dioxide emissions, but confirmed the EKC when using ecological footprint. Otherwise, the pooled mean group estimates undertaken by [67] reject the EKC for the case of US over 1945-2004. As well, apart from the inverted U-shaped relationship between environmental degradation and economic growth, other findings pointed out diverse shapes. As such, monotonically decreasing curve denotes that environmental quality improves as income growths, whilst monotonically increasing curve involves the diminution of environmental quality as income rises. Moreover, the N-shaped curve unveils that environmental degradation appears again after a decline to a specific level.
Furthermore, by applying panel unit root and panel co-integration tests, [68] rejected the EKC for a sample of 21 Latin American and Caribbean nations during 1970-2007. By employing a multivariate vector error correction model (VECM), [69] did not confirm also the EKC hypothesis in Russia between 1990 and 2007 and neither did [70] for Cambodia over 1996-2012, by estimating GMM and two-stage least square regression models. [71] used the STIR-PAT empirical model, as well as panel co-integration and FMOLS, but failed to support the EKC in five African states during 1980-2011.
A brief literature review that found different patterns on this topic is described in Table 2. However, even if [72] confirmed the EKC hypothesis, there was argued that the results cannot be generalized, being fragile without performing sensitivity analysis.

Previous evidence on the causal link between CO2 emissions and growth
Other part of literature examines the causal relationship between energy consumption and economic growth. The pioneering examinations with regard to the economic growth-energy consumption connection belongs to [81] which found an unidirectional causality between the gross national product and energy consumption for the United States, whilst [82] did not provide any causal link.
There are two prevailing divergent hypothetical points of view on the connection between energy consumption and economic growth [83,84]. The neo-classical view, known as the neutrality hypothesis [85,86], claims that there are other significant factor inputs in the production process than energy and no causal link occurs between energy consumption and economic growth. Thus, a nation may follow a protectionist energy strategy for lowering CO 2 emanations deprived of compromising growth. The antagonist view, referred as the nonneutrality hypothesis of energy, suggests that energy is a fundamental factor input in the production process and energy conservation policies may obstruct the economic growth. Subsequently, three theoretical models derive from the latter approach [87]. The growth hypothesis [88][89][90][91][92] advocates a uni-directional causality from energy consumption to economic growth and pretends that energy conservation policies will have negative effects on economic growth. The conservation hypothesis [93,94] supports a uni-directional causality from economic growth to energy consumption and motivates that energy conservation policies will not impair economic growth.
The feedback hypothesis [104][105][106][107] postulates a bidirectional causality between energy consumption and economic growth and asserts that energy conservation policies may weaken economic growth performance, so growth variations are reflected back to energy consumption. Overview of the studies that investigated the causal link between CO 2 emissions and economic growth is showed in Table 3.

Sample and variables
Our data sample covers the period 1990-2014 for a panel consisting of the EU-28 countries. The variables used for analysis, as well as their definition and data sources are presented in Table 4. All variables, except GDPCG, ENVTR, GFCF, RD, FFEC, EMPL, INDVA, and CPI are expressed in natural logs.
A part of preceding studies that tested the EKC hypothesis used only pollutant emissions, economic growth, and energy consumption [2,21,24,28,36,59,60,104], while other examinations furthermore considered new variables such as electricity consumption, trade openness, corruption, and government effectiveness index [70], electricity production [61], oil prices [25], population and urbanization [44,55,71,78], trade [1,56,58,62,66,75,79,80]. In line with [108][109][110] we control for the technology level via FFEC and EMPL. We employ the fossil fuel energy consumption to account for the level of use of dirty energy, which causes air pollution in the combustion process. As well, the employment in high-and medium-high-technology manufacturing sectors is included in as much as highly skilled labor contributes to the development of an efficient fabrication process that can stimulate energy savings and pollution deterrence. According to [52], an advanced level of comprehension to integrate cleaner technologies and collective responsiveness among people will drive higher ecological standards. INDVA is applied to control for the scale, whilst CPI aims to control for the selected states political state. Corrupted or unstable political regimes damage industrial progress and destroy the application of environmental policy in pollution abatement undertakings [110]. In the same vein, [111] concluded that prior communist nations had trouble introducing modern productive equipment conceived in capitalistic countries and depended on old and inefficient tools. On the contrary, [112] revealed that regulatory stringency positively influence efficiency, arguing that more rigorous emission standards lead firms to search for more efficient methods of fuel use and emissions decrease. However, further country-specific variables employed within current research are environmental tax revenues, gross inland consumption of renewable energies [51, 56], as well as gross fixed capital formation and research and development expenditure [51, 65,85]. According to [113] amongst nations variation systematically differs conditional on its richness, on institutional features, and on its whole commitment to the diffusion of renewable energy.

Econometric methods
In order to examine the EKC hypothesis we followed the approach of [35,44,51,76,77,104]. The long-run relationship between pollutant emissions, GDP per capita, environmental tax revenues, primary energy consumption, gross inland consumption of renewable energies, gross fixed capital formation, research and development expenditure, and other cross-country control variables, is given as follows: where i = 1, . . ., 28 and t = 1990, . . ., 2014 reveal the country and time, respectively, whereas PE denotes the pollutant emissions which take form of greenhouse gas emissions, greenhouse gas emissions intensity of energy consumption, emissions of Sulphur oxides, emissions of nitrogen oxides, emissions of non-methane volatile organic compounds, emissions of ammonia, and greenhouse gas emissions from transport. α it indicates the country specific fixed effect.
The parameters δ 1i -δ 11i are the long-run elasticities related to each explanatory variable of the panel. ε it describes the estimated residuals which characterize deviations from the longrun equilibrium.
Considering the inverted U-shaped EKC hypothesis, the sign of δ 1i is expected to be positive and the sign of δ 2i is expected to be negative, whilst the monetary value representing the turning point is computed by τ = exp[-β 1 /(2β 2 )] [23,49,60,76]. Furthermore, our aim was to establish the causal links between greenhouse gas emissions, economic growth, primary energy consumption, and environmental tax revenues.
Subsequently, we performed the heterogeneous panel co-integration test proposed by Pedroni [115,116] since it allows cross-section interdependence with different individual effects, alongside Kao [117] and Johansen approaches [118], as well as Westerlund [119]: where i = 1, . . ., 28 for each country in the panel and t = 1990, . . ., 2014 denotes each year of the period. Besides, the parameters α i and δ i allow country-specific fixed effects and deterministic trends. By pursuing the two-step procedure of Engle-Granger, the long-run model specified in Eq (2) is estimated in which the one period lagged residuals serve as the error correction term.
The dynamic error correction model is presented below: where Δ signifies the first-difference operator, q represents the lag length set at one according to likelihood ratio tests, and u exposes the serially uncorrelated error term.    Table 4. ÃÃ indicates the statistical significance at 5% levels. For the definition of variables, please see Table 4 Ã indicates the statistical significance at 10% levels. For the definition of variables, please see Table 4 https://doi.org/10.1371/journal.pone.0195708.t006

Summary statistics, correlations and unit root examination
Exploring the link between environmental pollution and economic growth in EU-28 countries in order to fulfill the targets related to the second pledge period of the Kyoto Protocol. However, a noteworthy decrease was registered in the case of sulfur oxides determined by shifting from high-sulfur solid and liquid fuels to low sulfur fuels for power and heat fabrication aims within the energy, industry and domestic areas, developments in energy efficiency, and the establishment of flue gas desulfurization equipment in new and current industrial equipment. The results comprised in Table 7 show the presence of cross-section dependence, similar to the previous findings of [99]. Hence, in order to remove this inconvenience, we considered Driscoll-Kraay standard errors in the fixed-effects models, comparable to [2]. Such standard errors are heteroskedasticity consistent, also being robust to very general forms of cross-sectional and temporal dependence [120].
Further, in line with [121] we perform two categories of stationarity tests, according to their cross-section dependence assumption, respectively the so called first generation that assume cross-section independence (Table 8), as well as the second generation that allow for cross-section dependence between the cross-section units ( Table 9). As we can notice, the variables are non-stationary in their levels, but become stationary after taking the first difference. Hence, we conclude that all series are I(1). Table 10 provides the results of pooled OLS regressions with Driscoll-Kraay standard errors. According to F statistic, all the estimated models are statistically highly significant, and the  Table 4.

Panel regression analysis
https://doi.org/10.1371/journal.pone.0195708.g001 Exploring the link between environmental pollution and economic growth in EU-28 countries values related to R-squared reveal that it could explain between 32% and 98% of the variability in pollutant emissions. The coefficients related to GDP per capita and squared GDP per capita are statistically significant in all the estimated models, except the models 4, 6 and 7. Nevertheless, the EKC hypothesis is confirmed in case of ESOX and ENMVOC. Furthermore, the output of fixed-effects regressions with Driscoll-Kraay standard errors is showed in Table 11. The estimated regressions appear to fit the data rather well since they can explain almost 42% and 87% of the pollutant emissions variation. An inverted U-shaped curve emerges in case of all selected harming emanations, except ESOX and GGET. With regard to the environmental tax revenues, we acknowledge that the expectation concerning ecological damage reduction is not supported since almost all the estimated models show a positive influence of such taxes on pollutant emissions. While primary energy consumption drives pollution, we notice, with some exceptions, that renewable energies consumption reduces pollutant emissions, as previously found by [56]. In terms of research and development expenditure, like [51], our results reinforce the beneficial effect of innovation on environmental pollution. As regards the variables employed to control for the scale effect, technique effect and political condition, we ascertain mixed evidence. The EKC hypothesis appears to be sustained since the inversed U-shaped curve tends to fit properly at least for greenhouse gas emissions. In addition,   Table 4. https://doi.org/10.1371/journal.pone.0195708.t007 Exploring the link between environmental pollution and economic growth in EU-28 countries  Table 4. ÃÃÃ indicates the statistical significance at 1% levels. ÃÃ indicates the statistical significance at 5% levels.

Co-integration and causality investigation
The outcome of Pedroni panel co-integration test [115,116] is presented in Table 13. According to [88,89,104], the statistics based on the dimension approach pools the autoregressive coefficients across different states for the unit root tests on the estimated residuals considering common time factors and heterogeneity across countries. To examine the null hypothesis of no co-integration, ρ i = 1, the following unit root test towards the residuals [88,89,104] is performed: As shown in panel A, panel PP and panel ADF statistics strongly reject the null hypothesis of no co-integration. Additionally, the statistics based on the between-dimension are based on the mean values of the individual autoregressive coefficients related to the unit root tests of the residuals for each member state. The results from panel B reinforce that panel PP and panel ADF statistics reject the null hypothesis of no co-integration at the significance level of 1%.
The second test of panel co-integration employed within current examination is the Kao test (the outcome is reported in Table 14). [117] suggests an ADF panel co-integration test Source: Authors' computations. Numbers in the parentheses represent t-stat values. For the definition of variables, please see Table 4. ÃÃÃ indicates the statistical significance at 1% levels.
ÃÃ indicates the statistical significance at 5% levels.
Ã indicates the statistical significance at 10% levels. https://doi.org/10.1371/journal.pone.0195708.t010 Exploring the link between environmental pollution and economic growth in EU-28 countries where the vectors of co-integration are homogeneous. However, according to [123], this test is based on a panel version of the ADF test on the residual (ε it ): Further, the following ADF statistic is produced:  Table 4. ÃÃÃ indicates the statistical significance at 1% levels.
ÃÃ indicates the statistical significance at 5% levels.
Ã indicates the statistical significance at 10% levels. https://doi.org/10.1371/journal.pone.0195708.t011 Exploring the link between environmental pollution and economic growth in EU-28 countries where b s 2 v depicts the estimated variance, b s 2 0v describes the estimated long-run variance of the error term which follows the standard normal distribution, and τ ADF reveals the ADF statistic for Eq (5). Thus, since the probability of ADF is 0.0274, the results provide support for the hypothesis of co-integration among all variables.
The third co-integration approach used is depicted by Fisher-type panel co-integration test [118] showed in Table 15. The Johansen Fisher panel co-integration test aggregates the  Exploring the link between environmental pollution and economic growth in EU-28 countries p-values of individual Johansen maximum eigen-value and trace statistics [124]; this test also rejects the null hypothesis of no co-integration.
Further, the Westerlund test [119] is employed (the outcome is showed in Table 16). Contrary to [121], the null hypothesis of no cointegration between greenhouse gas emissions, economic growth, primary energy consumption, and environmental tax revenues is rejected. Therefore, when the cross-dependence is considered in the panel, there exists a long-run equilibrium relationship between the variables.
Onward, since the variables are co-integrated, a panel vector error correction model is estimated in order to perform Granger-causality tests presented in Table 17. The panel error Exploring the link between environmental pollution and economic growth in EU-28 countries  Exploring the link between environmental pollution and economic growth in EU-28 countries  Table 4. ÃÃÃ indicates the statistical significance at 1% levels. ÃÃ indicates the statistical significance at 5% levels.
correction model, Eqs (3a)-(3d), allows two sources of causality, namely short-run causality via the lagged difference terms, as well as long-run causality by mean values of the error correction terms. With respect to Eq (3a), environmental tax revenues show a statistically significant influence on the short-run on GDP per capita growth. Concerning Eq (3b), there ensues a unidirectional causality running from economic growth to greenhouse gas emissions as in [96,102]. In addition, primary energy consumption has a statistically significant effect on GHG emissions. In Eq (3c), we noticed the statistically significant influence of GHG emissions and primary energy consumption on environmental tax revenues. Eq (3d) shows that GHG emissions and environmental tax revenues have a statistically significant impact on primary energy consumption. However, in as much as there is no causal link between GDPCG and PEC, the neo-classical view is confirmed, respectively the neutrality assumption. Moreover, the error correction term is statistically significant, but reveals a relatively slow speed of adjustment towards equilibrium.

Conclusions
The objective of the present study was to firstly examine the EKC hypothesis and subsequently the causal relationships between greenhouse gas emissions, economic growth, primary energy consumption, and environmental tax revenues, for a panel consisting of the EU-28 countries over the period 1990-2014. Since we noticed cross-sectional dependence in each of the variables in the panel, we employed Driscoll-Kraay standard errors. The results of pooled OLS regressions confirmed the EKC hypothesis for emissions of sulfur oxides and emissions of non-methane volatile organic compounds, whereas the outcome of fixed-effects estimations validated furthermore the EKC approach for greenhouse gas emissions, greenhouse gas emissions intensity of energy consumption, emissions of nitrogen oxides, emissions of non-methane volatile organic compounds and emissions of ammonia. Additionally, the estimation of a panel vector error correction model shows the presence of a short-run unidirectional causality from GDP per capita growth to greenhouse gas emissions, as well as a bidirectional causal link between primary energy consumption and greenhouse gas emissions. Besides, the neo-classical view was endorsed, respectively the neutrality hypothesis. The main policy implication deriving from our research can be formulated as follows: EU-28 states should promote the use of renewable energies that are constantly replenished and which will never end. Accordingly, the use of renewable energy will contribute to the decrease of GHGs emissions, while also reducing the reliance on fossil fuel markets. Besides, EU-28 may benefit from enhanced employment opportunities due to jobs occurrence in new cleaner technologies. Finally, as future research endeavors, our aim is to extend the empirical analysis in order to test the EKC hypothesis utilizing a composite index of environmental performance.