CO₂ emissions, renewable energy, and environmental regulations in the EU countries

Abstract

This paper analyzes the factors explaining the slight decrease of CO₂ emissions in the European Union (EU), recorded during the last period. With a focus on 12 EU countries, we apply a panel data analysis over the period 1990 to 2017 and we investigate the impact of renewable energy share in energy production, and the role of EU environmental regulations, in explaining the level of CO₂ emissions. Our static and dynamic panel data analysis points to a negative impact of an increased renewable energy share on CO₂ emissions, while there is no clear evidence about the role of environmental regulations. It appears that the 2020 climate and energy package contributed to the reduction of pollution level, while the ratification of the Kyoto protocol by the EU countries had no significant influence. At the same time, our findings validate the environmental Kuznets curve (EKC) hypothesis and the pollution halo (PH) hypothesis, showing that foreign companies export eco-friendly technologies. Our results prove to be robust regarding the use of static fixed and random effects models, of two-stage least square models and the use of difference and system generalized method of moments (GMM) frameworks.

Introduction

Starting with the 1992 “Earth Summit” in Rio de Janeiro, and afterwards with the Kyoto protocol agreed in 1997, developed countries committed to act against climate change and decided to reduce the greenhouse gas emissions, including CO₂. The European Union (EU) behaved as a world leader in this area and implemented different climate and energy legislative packages and environmental regulations to reach the agreed targets in greenhouse gas emissions. Against this background, several questions remain unanswered in the environmental economics literature: Are the environmental regulations of the EU efficient enough to ensure a reduction of CO₂ emissions? Does the increase of renewable to total energy production ratio explains the slight decrease in CO₂ emissions recorded by the EU during the last period? Can the classical environmental Kuznets curve (EKC) hypothesis and the pollution haven hypothesis (PHH) be validated in the case of the EU, or is the new pollution halo (PH) hypothesis better explaining the situation? The purpose of this paper is to find answers to these questions, resorting to a panel data analysis for selected EU countries, over the period 1990 to 2017.

The theories explaining the determinants of CO₂ emissions (a proxy for the level of greenhouse gas emissions and pollution in general) can be roughly classified in theories addressing the role of economic development and theories addressing the role of international factor movements. In the first case, the EKC theory advanced by Kuznets (1955) shows that the level of pollution increases with the level of per capita income, and decreases if the per capita income is sufficiently high. Holtz-Eakin and Selden (1995) empirically demonstrated the inverted U-shaped relationship between CO₂ emissions and individual income, while Stern (2004) brings further clarifications, stating that slow progresses and efforts to reduce pollution in developed countries can overcome the scale effects of economic growth, which put into question the validation of this theory in the case of developed countries. In the second case, the PHH theory investigates the environmental impact of foreign direct investments (FDI) on the host country pollution level. According to this view, the weak environmental policies implemented by developing countries determine multinational companies to invest there and thus to obtain comparative advantages in the production pollution-intensive goods (Copeland and Taylor 1994). However, if the opposite applies and foreign companies implement eco-friendly technologies, the environmental impact of FDI is positive (Hubler and Keller 2010). In this case, we talk about the pollution halo (PH) hypothesis, recently validated by Zhang and Zhou (2016), who show that FDI have a favorable regional impact on CO₂ emissions in China. In the same line, Eskeland and Harrison (2003) argue that US firms are energy efficient and find evidence in the favor of PH hypothesis for Mexico.

Starting from the arguments by Stern (2004), the EKC might be put into question in the case of EU countries where the level of income per capita is high. This is also the case of the PHH, given the common environmental regulations implemented by EU member states. Therefore, our first contribution to the existing literature is represented by the investigation of the factors explaining the reduction in CO₂ emissions in the EU,¹ going beyond the classical theories, and looking to the role of environmental regulations and the increased share of renewable energy to total energy production. We investigate therefore the effect of two legislative packages related to climate change, namely the ratification of the Kyoto protocol in April 2002, with the subsequent legislation, and the independent EU commitment to reduce overall greenhouse gas emissions by 20% compared with 1990 levels. This commitment which generated a package of legislative measures (the 2020 climate and energy package) agreed before the Copenhagen Conference of Parties to the United Nations Framework Convention on Climate Change (UNFCCC) in December 2009. Afterwards, we look to the role of renewable energy production in explaining the decrease of CO₂ emissions in the EU. The Enerdata Global Energy Statistical Yearbook provides information about the share of renewable energy production to total energy production for 12 EU countries over the period 1990 to 2017.² Further, we explore the channels through which the environmental regulations might contribute to better climate conditions. Using a series of interaction dummy variables, we investigate to what extent the environmental regulations for the reduction of total energy consumption and the promotion of renewable energy sources influence CO₂ emissions.

A second contribution of our paper is represented by an investigation of the environmental and climate change EU regulations, to better understand how these regulations contribute to assumed greenhouse gas emission targets. Third, we investigate the determinants of CO₂ emissions but also the determinants of CO₂ intensity and we therefore consider in our analysis the role of the business cycle. To this end, we use both static and dynamic panel data models to deal with the heteroscedasticity and potential endogeneity effect between CO₂ emissions and FDI underlined by the previous literature (i.e., Bao et al. 2011; Apergis and Ozturk 2015; Sapkota and Bastola 2017).

To preview our findings, we discover that both the EKC and PH hypotheses are validated for the selected EU member states. While the increase share of renewable energy production has a marginal but significant contribution to the reduction of CO₂ emissions, the impact of environmental regulations shows for the moment mixed evidence. It appears that the 2020 climate and energy package has a reduced contribution to the decreasing trend of CO₂ emissions, but the ratification of the Kyoto protocol in 2002 had no significant influence.

The “Related literature” section of the paper presents the literature review on the determinants of CO₂ emissions. The “EU environmental and climate change policies and regulations: a review” section analyzes the EU environmental and climate change regulations and underlines the potential channels by which the targets in greenhouse gas emissions can be reached in the EU countries. The “Data and methodology” describes the data and the methodology. The “Empirical results” section presents the main empirical findings, while the “Robustness check analyses” section addresses the robustness checks. The last section concludes our findings.

Related literature

The EKC hypothesis represents one of the key focuses of environmental economics. The reference study by Holtz-Eakin and Selden (1995) shows that the relationship between pollution and income follows an inverted U-shaped pattern. Starting from that moment, a considerable amount of papers investigates the existence of EKC, especially in the case of developing countries. Early studies document, in general, the existence of an inverted U-shaped relationship. For example, Selden and Song (1994) perform a global-level investigation considering emissions of four important air pollutants. Their cross-national panel analysis reveals the existence of an inverted U-shaped relationship between all four pollutants and the GDP per capita. Similarly, Grossman and Krueger (1995) argue that environmental deterioration is affected in the initial phase by the economic growth, but in a second phase registers improvements. Representing a contrary point of view, Perman and Stern (2003) reject the EKC hypothesis altogether.

While most of the early papers validate the EKC theory, especially in the case of developing and emerging economies, the recent econometric progress allowed for a deeper understanding of the EKC hypothesis. In this line, Atici (2009) performs a panel data analysis for the period 1980 to 2002 for Bulgaria, Hungary, Romania, and Turkey and validates the EKC hypothesis. In the same spirit, Jalil and Mahmud (2009) employ time series data from 1975 to 2005 for China and find evidence for an inverted U-shaped relationship between CO₂ emissions and economic development. Similar results are reported by Nasir and Rehman (2011) for Pakistan, Shahbaz et al. (2013) for Romania, and Tiwari et al. (2013) for India.

More recent studies resort to more sophisticated techniques to investigate the EKC theory. While Apergis and Ozturk (2015) perform a panel data analysis and use a GMM approach to validate the hypothesis for 14 Asian countries spanning the period 1990–2011, Baek (2015) uses an autoregressive distributed lag (ARDL) modeling and finds little evidence for the EKC hypothesis in the case of Arctic countries, over the period 1960–2010. Relying on a similar ARDL model, Yang et al. (2015) underline in turn the fragility of the EKC theory for China. By contrast, Sapkota and Bastola (2017) show the existence of an inverted U-shaped relationship between CO₂ emission and the GDP per capita using a panel framework for the Latin American countries. Particularly interesting investigations are performed by Youssef et al. (2016) who reported a bi-directional relationship between economic growth and pollution for a panel of fifty-six countries, and by Shahbaz et al. (2017), who performed a historical, nonparametric analysis of the EKC, for the G7 economies.

With a focus on the EU countries, Ang (2007) investigates the case of France and verifies the inverted U-shaped EKC hypothesis. Similarly, Acaravci and Ozturk (2010) investigate the EKC hypothesis for nineteen European countries, using an ARDL specification. They report a long-run relationship between CO₂ emissions per capita and the GDP per capita.

The recent literature also focuses on the investigation of PHH, showing that weak environmental policies implemented by emerging economies favorize the inward FDI given that multinational companies might obtain a comparative advantage in the production of pollution-intensive goods (Copeland and Taylor 1994; Spatareanu 2007). On the contrary, if foreign companies introduce environmentally friendly production technologies, according to the PH hypothesis, FDI might contribute to the reduction of pollution level (Hubler and Keller 2010). Early studies find mixed evidence regarding the PHH-PH trade off. On the one hand, Carrada-Bravo (1995) validates the PHH for the Latin American countries. On the other hand, Eskeland and Harrison (2003) argue that US firms are more energy efficient compared with the Mexican ones and support the PH hypothesis. Mixed evidence is documented by Blanco et al. (2013) for 18 Latin America countries, over the period 1980–2007, who showed that the PHH is validated only in pollution-intensive sectors. Similar findings are reported by Seker et al. (2015) for Turkey, showing that the impact of FDI on CO₂ emissions in the long run is reduced.

Other studies document a bi-directional relationship between inward FDI and CO₂ emissions. Likewise, with a focus of different high-, middle-, and low-income countries, Shahbaz et al. (2015) prove the existence of a bi-directional causal relationship between CO₂ emissions and FDI, which is, however, sensitive to the level of income. In addition, Tang and Tan (2015) investigate the Vietnamese case for the period 1976–2009 and find a similar bi-directional relationship. The PHH is also validated by recent studies (e.g., Sapkota and Bastola 2017; Solarin et al. 2017), while other researchers (e.g., Zhang and Zhou 2016) find evidence for the PH hypothesis.

The recent studies use a large set of control variables to investigate the PHH and the EKC hypothesis, especially the panel-oriented analyses. While the energy consumption is largely used as a control variable (e.g., Acaravci and Ozturk 2010), other control variables are represented by trade openness and energy use per capita (Atici 2009), population density and human capital (Sapkota and Bastola 2017), land, industry shares in GDP, and quality of institutions (Apergis and Ozturk 2015). We include most of these variables in our investigation on the EU countries, and we pay a special attention to the role of environmental regulations and to the increased share of renewable energy in the total energy production.

Although the impact of environmental regulations on CO₂ emissions is barely investigated, there are some recent essays in this direction. Correspondingly, Zhao et al. (2015) perform a micro-level analysis on the role of environmental regulations on the efficiency and CO₂ emissions of power plants in China. They show that market-based regulations and government subsidies contribute to a reduction of CO₂ emissions. A similar study with a focus on the Chinese economy is conducted by Wenbo and Yan (2018), who posit that the nexus between environmental regulations on the one hand, and CO₂ emissions intensity on the other hand, follows a significant inverted U-shaped curve. In opposition, Niedertscheider et al. (2018) find no clear link between climate change policies in Australia and the decreasing trend of CO₂ emissions. Similar findings are reported by Ma et al. (2018) for influence of state regulations on CO₂ emissions in China’s mining industry.

The investigation of renewable energy role in influencing CO₂ emissions also gained an increased interest during the last period and the empirical papers report mixed findings. In an analysis conducted over the period 1960 to 2007, Menyah and Wolde-Rufael (2010) state that renewable energy consumption does not cause CO₂ emissions in the sense of a Granger causality. In the same line, with a focus on Italy, Aliprandi et al. (2016) state that the impact of energy produced from renewable sources on CO₂ emissions is lower than expected. While Cherni and Jouini (2017) document the existence of a long-run relationship between CO₂ emissions, economic growth, and renewable energy consumption in Tunisia, their Granger causality analysis shows no relationship between CO₂ emissions and the renewable energy consumption. On the contrary, relying on an unbalanced panel approach of 128 countries over the period 1990 to 2014, Dong et al. (2018) show that the renewable energy intensity leads to a decline in the pollution level, for all analyzed regions. Chen et al. (2019) also document a long-run relationship between CO₂ emissions, GDP per capita, foreign trade, and non-renewable energy production in China. Similar findings are reported by Zoundi (2017) and Inglesi-Lotz and Dogan (2018) for African countries, and by Sinha and Shahbaz (2018) for India. As far as we know, there is no study investigating the role of renewable energy production on CO₂ emission in the EU countries. https://ec.europa.eu/eurostat/statistics-explained/index.php/Foreign_direct_investment_-_stocks#EU-28.27s_inward_and_outward_investment.

EU environmental and climate change policies and regulations: a review

The beginnings of the EU climate policy lie in the 1990s when the issue of climate change has been brought to international attention with the United Nations (UN) Conference on Environment and Development held in Rio de Janeiro from 3 to 14 June 1992. The main result of the summit was the agreement of the United Nations Framework Convention on Climate Change (UNFCCC). This international treaty is the cornerstone of the UN environmental conventions system. With its signing, the EU Council of the EC agreed to stabilize greenhouse gas emissions of the European Community at 1990 levels by the year 2000.³ The UNFCCC was followed since by the Kyoto Protocol of 1997 and its continuation with the non-binding Copenhagen Accord of 2009, the legally binding Cancun agreement of 2010, the Durban Platform for Enhanced Action of 2011, and finally the Paris Agreement of 2015. The EU as such, beside its member states, has been a signatory of all the treaties named above, thus creating a separate and complementary commitment to the goals expressed in these conventions.

The EU climate policy truly started in 2000s with the adoption of the European Climate Change Program (ECCP) as a result of the Union’s implementation of the Kyoto Protocol (the Kyoto Protocol’s ratification by the EU member states ended in 2002). The work of experts within the ECCP resulted in the adoption in 2003 of the European Union Emissions Trading System (EU ETS)⁴ which established a scheme for greenhouse gas emission allowance trading within the Union. The first phase (trading period) of the ETS (January 2005 to December 2007) was already operational before the entry into force of the Kyoto Protocol. The second phase (January 2008 to December 2012) coincided with the first commitment period of the Kyoto Protocol. The third phase began in January 2013 and will span until December 2020. In this context, we consider the year 2003 as a turning point in the EU environmental regulations.

In 2007, the European Council endorses “an EU objective of a 30% reduction in greenhouse gas emissions by 2020 compared with 1990.”⁵ The same Council adopted also accompanying targets for renewable energies and energy efficiency. The 2007 commitment led to the adoption, in 2009, of a legislation package, the so-called “Climate and Energy Package 2020,” that consists of several Directives on emissions trading, carbon capture and storage, fuel quality, and renewable energy, as well as the regulation on CO₂ emission performance standards for cars. The 2009 package sets as goals the 20% cut in greenhouse gas emissions (from 1990 levels), obtaining 20% of EU energy from renewable sources and a 20% improvement in energy efficiency by 2020. Therefore, we consider in our analysis the year 2010 as an important point for the EU environmental regulations.⁶

Other recent steps were implemented by the EU for the reduction greenhouse gas emissions. Likewise, in October 2014, the European Council adopted the 2030 Climate and Energy Framework and endorsed a binding EU target of an at least 40% domestic reduction in greenhouse gas emissions by 2030 compared with 1990. Following the 2016 signing by the EU of the Paris Agreement (which replaces the Kyoto Protocol), the European Parliament and the Council adopted in 2018 a legislation package to implement the 2030 Climate and Energy Framework, relying on three pillars. The first pillar consists of Regulation (EU) 2018/842 on binding annual greenhouse gas emission reductions by the member states from 2021 to 2030.⁷ For the second pillar, the Parliament and the Council decided on the inclusion of greenhouse gas emissions and removals from land use, land use change, and forestry in the 2030 climate and energy framework.⁸ Under this regulation, EU member states have to ensure that greenhouse gas emissions from land use, land use change, or forestry (LULUCF) are offset by at least an equivalent removal of CO₂ from the atmosphere in the period 2021 to 2030. As a third pillar of the 2030 framework, the ETS Directive was reviewed⁹ in order to include and implement the 4th ETS trading phase from 2021 to 2030. At the end of 2018, the European Commission published a climate strategy proposal up to the year 2050.¹⁰ The commitment to a climate neutral EU is the main goal for 2050, the Commission proposing that up to 25% of its yearly budget be allocated to this task.

We notice a sustained EU effort to reduce the level of pollution and to improve environmental and energy policies. One of the objectives of this paper is to see how effective these policies are in reduction of the CO₂ emissions.

Data and methodology

Data

The annual data for 12 EU countries¹¹ is extracted from the Enerdata database (2018 Global Energy Statistical Yearbook) and covers the period from 1990 to 2017 (324 observations). Our dependent variables are the CO₂ emissions (MtCO₂) from fuel combustion expressed in natural log (co₂em) and the CO₂ intensity (please refer to our third robustness check analysis), computed as the ratio between CO₂ emissions and GDP (co₂in). According to the EU Climate Policy¹², the CO₂ emissions represent more than 80% from the global greenhouse gas emissions, which shows the interest for investigating the CO₂ emission determinants.

The share of renewable in total electricity production (renewable) represents the first interest variable¹³. Second, we investigate the role of EU environmental regulations, whose effects are captured using two regulation dummy variables namely dummy2003, which takes value 1 after the ratification of the Kyoto protocol by the EU member states, and dummy2010 that takes value 1 starting with 2010, when the EU started the implementation of the 2020 climate and energy package. For robustness purposes, we also use a dummy2005 variable (instead of dummy2003). The reason is represented by the existence of a potential time delay for the effective entry into force of the Kyoto protocol, after the EU ETS adoption in 2003. In fact, the first phase of the ETS called “trading period” started only in 2005.

In a subsequent analysis, we explore the channels through which these regulations might contribute to a reduction in CO₂ emissions. On the one hand, the regulatory effect may stimulate the renewable energy production and we capture this effect using interaction dummy variables (dummy × rnw) between regulations dummy (dummy2003/dummy2005,-dummy2010) and renewable to total energy production ratio (renewable). On the other hand, the regulation may act in the favor of the reduction of the total energy consumption and we investigate this channel resorting to a second category of interaction dummy variables (dummy × tec) between regulations dummy and total energy consumption, and between regulations dummy and energy intensity (dummy × ei).¹⁴

The set of control variables starts with the income (gdp), calculated in terms of GDP per capita in 2010 constant US$ (World Bank Development Indicators). If the gdp-estimated coefficient is positive and significant, while the coefficient of gdp² is negative and significant, then the EKC hypothesis is validated. The second control variable is the inward foreign direct investment (data is extracted from the UNCTAD database). When we test the determinants of CO₂ emissions, we refer to the log of FDI stock in US$ (fdi1). When we test the determinants of CO₂ intensity for robustness purposes, we consider the FDI stock to GDP ratio (fdi2). In line with Sapkota and Bastola (2017) and Soytas et al. (2007), the remaining control variables refer to the following: (i) the total energy consumption (tec) and the energy intensity of GDP at constant purchasing power parities (ei), whose effect on CO₂ emissions and CO₂ intensity, respectively, is expected to be positive; (ii) the investment measured in terms of gross fixed capital formation in 2010 constant US$ (gfcf1) and as GDP percentage (gfcf2) whose impact on CO₂ emissions/intensity is expected to be positive; (iii) population density calculated in terms of people per km² of land (popdens), with a positive impact on CO₂ emissions; and (iv) the unemployment rate (unemp), which may have either a positive impact on pollution if the governments allocate more financial resources for social adds and neglect the environmental issues (Lan et al. 2012), or a negative impact, if unemployment is associated with smaller revenues and smaller investments and consumption. These data are extracted from the World Bank database. Given the downward trend in the CO₂ emissions at EU level, our analysis also considers a linear time trend as explanatory variable (time). The general statistics are presented in Table 1¹⁵ and show noteworthy disparities between countries, in terms of the share of renewable energy or population density. At the same time, the differences in terms of GDP per capita are much smaller.

Table 1 Summary statistics

In Table 2, we perform a series of panel unit root tests. In a preliminary analysis, we apply cross-sectional dependence tests to see in the first-generation or the second-generation panel unit root tests are recommended (first-generation tests are more powerful in the presence of cross-sectional independence).

Table 2 Cross-sectional dependence and panel unit root tests

However, because it is no clear evidence in the favor of cross-sectional independence (Frideman’s and Frees’s tests point in the favor of cross-sectional dependence), we present the results of the classic LLC test and Pesaran’s CIPS test. Except for gross fixed capital formation to GDP ratio (gfcf2), population density (popdens), and unemployment rate (unemp), the pooled CIPS test of Pesaran (2007) shows that our variables are stationary. For these variables, however, the LLC test indicates the absence of unit root processes. Therefore, we conclude that our variables are stationary.

Methodology

To analyze the determinants of CO₂ emissions, we first resort to a fixed effect model that allows to avoid the omitted variable bias. We also run a random effect model to control for all stable covariates and we chose between these models performing a Hausman test. The general equations for the fixed (Eq. (1)) and the random (Eq. (2)) models are:

$$ {\mathrm{Y}}_{\mathrm{i},\mathrm{t}}={\upbeta}_0+{\upbeta}_1{\mathrm{X}}_{\mathrm{i},\mathrm{t}}+{\upbeta}_2{\mathrm{Z}}_{\mathrm{i},\mathrm{t}}+{\upalpha}_{\mathrm{i}}+{\upvarepsilon}_{\mathrm{i},\mathrm{t}} $$

(1)

where Y_it is the dependent variable (CO₂ emissions, CO₂ intensity); X_it represents the vector of interest variables (renewable and environmental dummy variables); Z_it represents the vector of control variables; β₀ is the intercept; α_i represents all the stable characteristics of countries; β₁ and β₂ are the coefficients; ε_i,t is the error term.

$$ {\mathrm{Y}}_{\mathrm{i},\mathrm{t}}={\upbeta}_0+{\upbeta}_1{\mathrm{X}}_{\mathrm{i},\mathrm{t}}+{\upbeta}_2{\mathrm{Z}}_{\mathrm{i},\mathrm{t}}+{\upalpha}_{\mathrm{i}}+{\upmu}_{\mathrm{i},\mathrm{t}}+{\upvarepsilon}_{\mathrm{i},\mathrm{t}} $$

(2)

where μ represents between-entity errors; ε_i,t is the within-entity error.

However, an endogeneity bias may appear between CO₂ emissions and FDI. Foreign investors may not want to invest in a country that is already polluted because they risk paying higher environmental taxes in the future, or because their business model promotes a reduced carbon intensity (see for example Bao et al. 2011; Sapkota and Bastola 2017).¹⁶ Therefore, in a second step, we perform an instrumental variable regression, verifying the presence of endogeneity (Wu-Hausman and Durbin-Wu-Hausman tests), and of disturbance heteroscedasticity (Pagan-Hall general test). If heteroscedasticity is present, we correct the bias using robust standard errors (see Baum et al. 2007). However, in this case, GMM models may perform better (Baum et al. 2003). Therefore, in the third step, we resort to a dynamic GMM estimator, comparing the difference-GMM estimator proposed by Arellano and Bond (1991) with the system-GMM estimator advanced by Blundell and Bond (1998). Equation (3) describes the difference-GMM model:

$$ {\Delta \mathrm{Y}}_{\mathrm{i},\mathrm{t}}={\upbeta}_0+{\sum}_{\mathrm{j}=\mathrm{t}-\mathrm{p}}^{\mathrm{t}-1}{\uptheta}_{\mathrm{j}}\Delta {\mathrm{Y}}_{\mathrm{i},\mathrm{j}}+{\upbeta}_1{\Delta \mathrm{X}}_{\mathrm{i},\mathrm{t}}+{\upbeta}_2{\Delta \mathrm{Z}}_{\mathrm{i},\mathrm{t}}+\Delta {\upmu}_{\mathrm{i},\mathrm{t}}+{\Delta \mathrm{u}}_{\mathrm{i},\mathrm{t}} $$

(3)

where θ is the lagged-variable parameter and p is usually established to 1 (first lag); u_i,t is the error term which varies over both country and time.

Nevertheless, as in the case of the instrumental variable approach, for the difference-GMM estimator, the lagged levels of the explanatory variables might be considered poor instruments when the autoregressive process is significantly persistent (Albulescu and Tămășilă 2016). If ∆Y_{i, j} is still correlated with ∆u_{i, t}, the difference-GMM estimator cannot reach the best estimation. Therefore, to overcome this limitation, the Blundell and Bond’s (1998) system-GMM estimator relies on a system of two simultaneous equations, one in level (with lagged first differences as instruments) and the other in first difference (with lagged levels as instruments).

Empirical results

We first present the results of the static analysis, comparing a fixed and a random effects model (Table 3). For each estimator, we test two models. Model 1 considers the environmental dummy variables. If we admit the impact of the environmental and climate change regulations on the CO₂ emissions, we explore that the channels throughout these regulations influence the pollution level, using the interaction dummy variables described above (Model 2).¹⁷

Table 3 Determinants of CO₂ emissions: static panel data analysis

For the first model, we notice that the effect of renewable electricity share on the CO₂ emissions is significant and has the expected sign, but is marginal (result in agreement with that reported for Italy by Aliprandi et al. 2016). At the same time, it appears that the second set of environmental regulations, implemented starting with 2010, contributes in a significant way to the reduction of pollution. This result is documented for both fixed and random effect estimations, although according to the Hausman statistics, the random model performs better. In addition, the EKC hypothesis is validated, but also the PH hypothesis, meaning that foreign investors avoid the countries with a high level of CO₂ emissions given the polluter pays principle.¹⁸ This result can be explained by the fact that multinational companies use eco-friendly technologies. The coefficients of the control variables are significant and have the expected sign, except for the investment, which has a negative contribution to pollution.

Although this result is in contrast with several previous finings reported in literature (i.e., Sapkota and Bastola 2017), it can be explained by the increased share of green investment. Consequently, in the presence of green technology, the effect on investment on CO₂ emissions might become positive. The increase of energy consumption has an important impact on the pollution level; the results are in agreement with previous findings.

These results remain unchanged in the case of Model 2. In addition, it seems that the second set of environmental regulations contributed to a decrease of CO₂ emissions by generating a contraction in total energy consumption. However, the ratification of the Kyoto protocol has no significant impact on the level of pollution in the EU countries.

We continue our analysis with the instrumental variable analysis, where the first lag of FDI stock is used as instrument (the results are presented in Table 4). In this case, we also have the two models (Models 1 and 2) which consider the impact of environmental regulations and the channels throughout the regulations might influence the pollution level. In the first step, we use GMM errors. While the endogenity tests (Wu-Hausman and Durbin-Wu-Hausman) show no endogeneity bias, on the contrary, the Pagan-Hall test clearly points a heteroscedasticity. In this case, robust errors should be used. However, the Sargan test becomes inconsistent in the presence of robust errors and a Hansen test statistic must be used to identify over-identification problems (Roodman 2009). Table 4 shows that the instruments are exactly identified.

Table 4 Determinants of CO₂ emissions: instrumental variable analysis

Compared with the previous analysis (Table 3), we notice that the results remain the same. In terms of environmental regulations, we notice on the one hand that the first set of regulations has no significant effect, while the second set contributed to a decrease of CO₂ emissions. These findings agree to those reported for the random effect model. On the other hand, the coefficients of interaction dummies are significant for the same 2010 regulatory package and the same total energy consumption channel.

Both the EKC and the PH hypotheses are validated, while the control variable coefficients are in general significant and have the expected sign. However, the coefficient of the GDP per capita and the intercept becomes very high with the instrumental variable approach. Therefore, given the presence of heteroskedasticity (Pagan-Hall test) which may introduce a bias in our results, we proceed to a GMM analysis, which according to Baum et al. (2003) provides better estimates in the presence of heteroscedasticity.

Table 5 presents the results of the GMM approach, comparing a difference- and a system-GMM model. To avoid the over-proliferation of instruments, we have restricted the maximum lag length of the lagged instruments to one. Given the presence of heteroscedasticity, we report the results for robust errors.¹⁹

Table 5 Determinants of CO₂ emissions: dynamic panel data analysis

The results of the difference-GMM estimator confirm the first set of results and show the negative impact of the share of renewable electricity production and environmental regulations (e.g., the 2020 climate and energy package) on the level of CO₂ emissions. When we look to the transmission channels, we notice that the total energy consumption channel is significant for the second regulatory package. However, in this case, the renewable energy channel associated with the implementation of the Kyoto protocol (dummy2003 × rnw) became also significant; the results are confirmed by the system-GMM specification.

While both specifications validate the EKC hypothesis, the PH hypothesis is not validated by the system-GMM approach. Compared with the difference-GMM approach, the system-GMM has better performance for large N and small T samples. However, the system-GMM might suffer from the over-proliferation of instruments. Thus, we have applied the Hansen test for instruments over-proliferation (consistent with robust errors) and the Arellano-Bond tests for autocorrelation (AR (1) and AR (2)). While the Hansen test shows that the instruments are correctly identified, the Arellano-Bond test (AR (2)) indicates no autocorrelation bias.²⁰

All in all, our results show consistence between the static and dynamic specifications. To prove the robustness of our findings, in what follows we perform three robustness checks.

Robustness check analyses

Re-sampling considering old EU members

For the first robustness check, we retain in the sample nine EU countries for the same period, 1990 to 2017. We exclude the Central and Eastern European (CEE) countries from the sample. These countries became EU members in 2004 (the Czech Republic and Poland) and in 2007 respectively (Romania), only after the ratification of the Kyoto protocol by the old EU member states.

Table 6 shows the results for the static analysis. We notice that the negative impact of the share of renewable energy is marginal, and significant as in the main analysis. As for the main results, we observe a consensus between the fixed and the random effect models. First, the EKC and PH hypotheses are both validated. Second, the coefficients of the control variables are significant and have the expected sign. Third, the dummy2010 variable negatively influences the level of CO₂ emissions in old EU member states, meaning that the environmental regulation which agreed in 2010 contributed to a reduction in the pollution level. Finally, the channel through which the regulation impacts the CO₂ emission is once again the total energy consumption channel.

Table 6 Determinants of CO₂ emissions: static panel data analysis (nine EU countries)

While the production of renewable energy marginally contributed to the reduction in CO₂ emissions, the environmental regulations from 2010 affected the CO₂ emissions by generating a reduction of energy consumption. The results are consistent with those reported in Table 4.

We continue with the dynamic analysis and we present the GMM results with robust errors.²¹ Table 7 indicates the negative impact of the share of renewable electricity production to the CO₂ emissions, showing the robustness of the previous findings. However, in this case, the EKC hypothesis holds only for the difference-GMM specification, while the PH hypothesis is validated by both approaches. Different from the main results, the impact of environmental regulations is unclear. Against this background, we perform a second set of robustness checks, considering the delay associated with the effective entry into force of EU environmental regulations.

Table 7 Determinants of CO₂ emissions: dynamic panel data analysis (nine EU countries)

Entry into force of Kyoto protocol through ETF trading period (dummy2005)

The implementation of the Kyoto protocol requirements adopted in 2003 might be made with some delay. In fact, only in 2005 starts the first phase of ECCP. Therefore, in the second robustness check analysis, we consider the effective entry into force of the Kyoto protocol that started in 2005, and we use a dummy2005 variable instead of dummy2003. Table 8 shows the static panel data results.

Table 8 Determinants of CO₂ emissions: static panel data analysis (dummy2005)

The results presented in Table 8 are very consistent with those reported in Table 4, showing the negative effect of the increase renewable electricity share in total electricity production, on CO₂ emissions, and validating the EKC and PH hypotheses. Moreover, the second environmental regulation package from 2010 significantly contributed to a reduction in CO₂ emissions, while the channel is represented by the total energy consumption as before.

We continue with the dynamic analysis. Table 15 (Appendix) presents the results of the instrumental variable approach, while Table 9 highlights the findings of the GMM models with robust errors.

Table 9 Determinants of CO₂ emissions: dynamic panel data analysis (dummy 2005)

While the role of renewable energy in the reduction of CO₂ emissions remains the same and both EKC and PH hypotheses are validated, an important difference appears in the case of environmental regulations. According to the GMM results (both difference- and system-GMM approaches), the first set of regulations has a negative and significant impact on CO₂ emissions, while the second set releases in 2010 has no significant impact. We observe that the effects of environmental regulations produce with a considerable delay, results that put into questions the robustness of the analysis. Our findings may be, however, considerably influenced by the EU business cycle.

Renewable energy, environmental regulation, and CO₂ intensity

Finally, we want to control for the effect of the business cycle on our results. To this end, we look to the CO₂ intensity (computed as the ratio between CO₂ emissions and GDP) and we consider as explanatory variable the GDP energy intensity, the FDI to GDP ratio, and the gross fixed capital formation to GDP ratio. This represents another way of looking to the pollution level and to the role of environmental regulations. The determinants of the CO₂ intensity are investigated for the initial sample of 12 EU countries. Table 10 presents the findings for the static analysis.

Table 10 Determinants of CO₂ intensity: static panel data analysis

In this case, the EKC is validated only for the Model 1, while the PH hypothesis is not validated. The role of renewable energy in reducing CO₂ intensity is significant only for the Model 2. The energy intensity of GDP has the most important explanatory power for the decreasing trend of CO₂ intensity in the EU. The influence of environmental regulations is ambiguous. It appears that the regulations contributed to the reduction of CO₂ intensity, considering the renewable energy channel. This result is valid, however, only for the fixed effect estimation (Model 2).

Given the presence of heteroscedasticity (see the results of the Pagan-Hall test in Table 16, Appendix), robust standard errors are used, and we present, for consistency reasons, the results of the GMM models. Table 11 confirms the static analysis results and shows that the role of environmental regulations in the reduction of CO₂ intensity is unclear in this case.

Table 11 Determinants of CO₂ intensity: dynamic panel data analysis

To sum up our findings, we posit that both the income and the FDI influence the level of CO₂ emission, validating thus both the EKC and the PH hypotheses. In addition, the increasing share of the renewable electricity to total electricity production negatively influences the CO₂ emissions level, showing thus the importance of financial stimulus for renewable energy (Table 12).

Table 12 Centralized results

However, when we consider the role of environmental regulation, we discover mixed evidence. First, the main results show that the 2020 climate and energy package adopted by the EU in 2009 contributed to the reduction of CO₂ emissions, and the total energy consumption channel was at work. At the same time, the ratification of the Kyoto protocol in 2002 and implemented in 2003 had no significant influence. These findings are confirmed by our robustness checks. Second, if we consider the time delay in the implementation of the Kyoto protocol, according to the GMM models estimations, this set of environmental regulations (EU ETS) also contributed to the reduction of CO₂ emissions. However, this result lacks in robustness. We conclude that the regulations might had a negative impact on the CO₂ emission throughout the total energy consumption. At the same time, the regulations regarding the production and consumption of renewable energy that have established clear goals are important for the reduction in CO₂ emission levels.

Conclusions

This paper investigates the impact of renewable energy share in energy production, and the role of EU environmental regulations on CO₂ emissions in 12 EU countries for the period 1990 to 2017. Static and dynamic panel data results support the EKC and PH hypotheses, showing an inverted U-shaped relationship between CO₂ emissions and income level, and that foreign companies export eco-friendly technologies inside the EU.

According to the static fixed and random effect models, the impact of renewable electricity share on the CO₂ emissions is significant and has the expected sign, but is marginal (similar results are reported by Aliprandi et al. 2016), while the second set of environmental regulations implemented starting with 2010 contributes in a significant way to the reduction of CO₂ emissions. This result can be explained by the fact that the Kyoto protocol is an exponent of international, multilateral agreements, without disposing of real constraints for the signing parties. Further, its entry into force happened with some delays. On the contrary, the 2020 climate and energy regulations represent an internal EU set of rules, which are more effective in terms of constraints, sanctions, or financial stimulus applied to the EU member states.

The results of the dynamic GMM models (in particular the difference-GMM model) validate the static models’ findings, including the impact of environmental regulation dummies. We notice that the first set of regulations has no clear contribution to a decrease of CO₂ emissions, while the second set has a significant effect. However, in the case of dynamic estimations, these results lack in robustness. The unclear effect of environmental regulation on the reduction of CO₂ emissions was also underlined by Niedertscheider et al. (2018).

To check the robustness of our findings, we first excluded from the sample the CEE countries. In this case, the static analysis reveals that the negative impact of the share of renewable energy is marginal, as in the case of the main findings. At the same time, the environmental regulations which agreed in 2009 contributed to a reduction in the pollution level. Second, we have considered the time delay regarding the entry into force of the Kyoto protocol and the results are consistent with the main findings, except for the dynamic specifications. Third, we have focused on the determinants of the CO₂ intensity for the initial sample of 12 EU countries. In this case also, the influence of environmental regulations looks ambiguous. Overall, we conclude that both the income and the FDI influence the level of CO₂ emission, validating thus the EKC and PH hypothesis. Additionally, an increasing share of the renewable electricity to total electricity production negatively influences the pollution level. However, the overall impact of environmental regulations is unclear.

Two policy implications result from our findings. First, our findings highlight the importance of financial stimulus for renewable energy production. Precise policy goals which sustain the use of renewable energy (e.g., obtaining 20% of EU energy from renewable sources by 2020), as those established in 2009, are effective in stimulating the reduction of CO₂ emissions. Second, it appears that the EU ETS system is not so efficient for the reduction of CO₂ emissions. On the contrary, regulations that constraint the total energy consumption and sustain the energy efficiency are more effective for controlling the EU pollution level.

Appendix

Fig. 1

Dynamic of CO₂ emissions and CO₂ intensity in the EU (source: World Bank Development Indicators)

Table 13 Explanatory variables

Table 14 Determinants of CO₂ emissions: instrumental variable analysis (nine EU countries)

Table 15 Determinants of CO₂ emissions: instrumental variable analysis (dummy 2005)

Table 16 Determinants of CO₂ intensity: instrumental variable analysis