The Environmental Kuznets Curve: Empirical Evidence from OECD Countries

Abstract

The Environmental Kuznets Curve (EKC) hypothesis reflects the relationship between environmental quality and GDP per capita. The environmental quality decreases in early periods of GDP growth per capita until a certain point, and after that, it begins to increase. This paper investigates the presence of the EKC curve in OECD countries over the period 1997–2015 and identifies thresholds levels of GDP per capita leading to lower emissions per capita for these countries. Also, it points out the key role of energy from renewable sources. Based on nonlinear panel data methods, findings show that CO2 emissions increase up to a certain level of income (10774$–44494$ per head) and then, they decrease. Policymakers are encouraged to consider economy, technology, and environment all together and handle the legal regulations they will implement, accordingly.

1. Introduction

One of the biggest environmental challenges impacting the planet and humanity is climate change. To handle this challenge, various international initiatives aiming to rethink the energy-energy security-climate change nexus have been taken. Among them, the 2021 OECD International Program for Action on Climate (IPAC), the Earthshot Prize, and the 2020 European Green Deal (EGD) support countries progress toward net-zero greenhouse gas (GHG) emissions and a more resilient economy by 2050. They propose a mixt of channels and ambitious goals to mitigate climate change and provide more certainty to policymakers and investors regarding their future decisions on CO₂ emission levels. These emissions must be consistent with the EU’s/OECD’s goals of being climate-neutral by 2050 such as: reduce the GHG to at least 55% below 1990 levels by 2030; increase the share for renewable energy to more than 32%, improve energy efficiency at about 33%, sustain people and companies to find solutions to help fight climate change and so on.

But these challenges will not be easy to keep given that OECD or EU countries still rely on fossil fuels (such as coal, oil, gas, and petroleum products) for about 80% of their energy supply, notably for industry and transport that are the largest emitters of GHGs. New OECD data indicate that total fossil fuel support in 44 OECD and G20 economies rose by 10% in 2019 to 178 billion USD, ending a 5-year descending trend and undermining global efforts to mitigate climate change [1, 2].

In this context, the question that rises is whether economic growth leads inevitably to more pollution. Until the 1980s, the Club of Rome view (highlighting a negative impact of economic growth on the environment and natural resources) and the Dasgupta and Heal [3]’s view (pointing out a complementary relationship between economic growth and environmental improvement) have been the main approaches of the applied economic analysis. Since 1990s, the Environmental Kuznets Curve (EKC) penciled by Baloch and Wang [4] comes to the scene by suggesting an inverse U-shaped curve between GHG and income. Four main aspects explain it. The first one is income elasticity of environment demand. This means that as income increases, people are incited to consume more healthy products and pay more attention to the environmental quality while government is expected to put in place stricter rules on environment protection to improve it. The second one is the effect of economic scale, technology, and structure. The economic scale reflects the situation in which more economic growth means more energy consumption from nonrenewable sources and, as a result, more pollution [4]. The structural effect implies that structural changes may improve environment with economic development. Finally, the technology effect suggests that old pollution machines should be replaced by clean techniques of production, which are expected to positively affect the environmental quality. Overall, the EKC indicates that, at the beginning stage of development of a country, the negative scale prevails and the pollution increases (see, [5]). Once the income reaches a certain threshold, the structural and technological effect exceeds the scale effect leading to a fall of pollution levels. The third aspect refers to the international trade, which is a key factor of EKC validity. Trade has a dual, twin influence on the environment quality: a negative one through the scale effect or a positive one via the technological effect (see, [5]). The last aspect is that of FDI, which has two possible impacts on the quality of environment, too. On the one hand, the developing countries are tempted to lower their claim on the environment protection to attract more foreign investors, which aggravates environmental pollution; on the other hand, by getting technology by FDI from developed countries, most emerging countries help improve environment.

The empirical literature on the relationship between economic growth and environmental quality is considerable (e.g., [6, 7, 8, 9, 10, 11, 12, 13, 14, 15]) and concentrates three main strands of studies. The first group of studies focuses on the growth-pollution nexus by testing the EKC validity with different time series and panel data models (e.g., [16, 17, 18]). The second group of studies focuses on the link between the economic growth and energy consumption from (non)renewable sources given that the GHG emissions are produced by fossil fuels (e.g., [19, 20, 21, 22, 23]). The last group of studies combines the two approaches by analyzing the link between energy consumption from renewable and nonrenewable sources, economic growth, and environmental quality [24]. Compared with the previous one, this wave of studies addresses concerns regarding biasness caused by omission of variables. By integrating variables on international trade, urbanization degree, foreign direct investment, and so on, one can assess the linear and nonlinear relationship between economic growth, pollution, and energy consumption with time-series and/or panel data. Overall, empirical studies conducted until now show diverging results on the growth-pollution nexus and explain this gap in findings by the use of various methods (with time series or panel data), length of periods, and/or the nature of country sample.

Regarding the OECD studies, some of them seek to empirically estimate the nonlinear relationships between income level and environmental degradation and dynamic causalities between environmental pollution and economic growth, along with other key explanatory macroeconomic indicators (e.g., [11, 25, 26, 27]). But the attempt to validate the EKC hypothesis still gives ambiguous outcomes. Frodyma [26] shows that the EKC models fail to explain the relationship between income and production-based emissions (or consumption-based emissions) when using an ARDL testing bound approach (due to the heterogeneity of the panel sample considered) over the 1970–2017 period. At the opposite side, Benjebli et al. [25] identifies an inverted U-shaped EKC curve for 25 OECD for the period 1980–2010 based on fully modified ordinary least squares (FMOLS) and dynamic ordinary least squares (DOLS) estimates. Differently, Madaleno and Moutinho [11] focuses on 15 EU countries (old members of OECD) and 12 Emerging EU countries over the 2008–2018 period. With dynamic fixed effects (DFE) and DOLS estimates, they identify a U-shaped EKC curve. Likewise, Churchill et al. [28] finds evidence that supports the EKC hypothesis in nine of the 20 OECD countries over the period 1870–2014 (a long-time horizon of 144 years). The EKC hypothesis is validated for Australia, Canada, Denmark, Finland, France, Japan, Spain, United Kingdom, and the United States, albeit with different turning points for income per capita. These findings are generally comparable to other empirical studies, which identified an EKC curve for OECD with panel data over shorter periods. For example, Dijkgraaf and Vollebergh [29] finds that 11 out of 22 OECD countries have a statistically significant turning point and verify an inverted U-shaped EKC pattern. Likewise, Apergis [30] and Shafiei and Salim [31] find evidence supporting an inverted EKC relationship for some OECD countries as well.

Another group of studies go further and compute the turning points in income per capita leading to lower emissions per capita. Galeotti [32] finds for OECD over the period 1960–1997 a turning point ranging between 15599.9$ and 21185.83$ (in 1990 US dollars). In the same vein, Churchill et al. [28] identifies greater turning points ranging from $19 978 to $84001, depending on whether or not financial development variable is included in the equation’s model. Based on a panel smooth transition approach over the 2000–2018 period, for 75 countries (including some OCDE countries), Tatoglu and Polat [14] finds a turning point between $42053 and $42057. Also, Sun et al. [27] uses the Common Correlated Effect Mean Group (CCEMG) and Augmented Mean Group (AMG) methods for the period 1992–2015 and shows that growth output has significant positive effect on environmental pollution. The estimated turning points are greater ranging from 1479029.30 to 3277677.02. As we can observe, the EKC empirical evidence is still controversial, and there is no consensus on the income level at which environmental degradation begins diminishing.

This paper will contribute to the last strand of the empirical literature by providing evidence on the validity of the EKC curve for 34 OECD countries [33] and by identifying threshold levels of GDP per capita leading to lower emissions per capita for these countries. But, unlike previous studies, it will point out the key role of energy from renewable sources.

Therefore, the paper contributes to the existing empirical literature in the following ways. First, contrary to the most part of studies focusing on the role of energy consumption from fossil fuels in validating the EKC curve, the paper embodies indicators on energy from renewable sources, since OECD countries are considered as regions with a less energy-intensive economic structure and a more advanced development in renewable sources than emerging ones (e.g., [34]). Second, the paper includes a heterogeneous sample of countries with respect to their energy consumption mix over the period 1997–2015, which allow capturing disparities, if any, in the value of the threshold above and below which the effect of economic activity on environmental pollution may differ within this group of countries. The threshold is identified with a recent empirical model by Fouquau et al. [35] and González et al. [36] a panel smooth transition model (PSTR), which is the best model for tackling the heterogeneity in the effect of economic activity on the environmental degradation. Third, the paper includes two types of indicators for energy from non-fossil fuels: the renewable energy consumption and the electricity production from renewable sources per capita, excluding hydroelectric (kW h), the last being used for robustness purposes. Fourth, the findings provide further support to economic approaches investigating the renewable EKC curve (R-EKC curve). However, they go further by pointing out the non-monotonous relationship between environmental degradation and economic activity in combination (or not) with the renewable energy use, FDI, and urbanization degree. To my best knowledge, this aspect of the R-EKC curve has never been explored before with a PSTR approach. Finally, the paper provides useful insights for policymakers in supporting green economy to achieve the 2030 Agenda for Sustainable Development (via the IPAC and the EGD international initiatives).

The results show evidence for an inverted U-shaped relation between income and environmental pollution for OECD-34. However, the overall effect of output on the environmental degradation (through carbon dioxide emissions) is positive meaning that technology, preference, and environmental investments in clean energy are not sufficient developed in the energy consumption mix. The turning point or the income level in inflexion of R-EKC ranges between 10774.1 and 44967.7 dollars per head when the indicator of gas pollutant is the carbon dioxide emissions (such as in [17, 18, 27]).

The rest of the paper is structured as follows. Section 2 describes the data and presents the PSTR methodology. Section 3 provides the empirical results and discussion. The final section concludes.

2. Data and methodology

2.1 Data and descriptive statistics

The dataset contains 34 advanced countries of OECD and covers the yearly period 1997–2015. The following countries are included: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Slovak Rep., Spain, United Kingdom, Japan, United States, Australia, Canada, Denmark, Iceland, Luxembourg, New Zeeland, Norway, Switzerland, Turkey, Chile, Czech Rep., Estonia, Hungary, Israel, Korea Rep., Mexico, Poland, Slovenia, and Sweden.

In the models, the dependent variable is the natural logarithm of CO₂ emissions (metric ton per capita) for country i in year t, in line with prior literature (e.g., [31]). The vector of explanatory variables includes the following variables: real GDP per capita as a measure of output (GDP), ratio of exports and imports to GDP (TO) for openness to trade, the urban population (as % of total population-UR), foreign direct investment (as % of GDP-FDI), renewable energy consumption (% of total final energy consumption-RE), and the renewable electricity output (as % of total electricity output-REL), which is the share of electricity generated by renewable power plants in total electricity generated by all types of plants. The first three variables could have alternative impacts (they should be positively or negatively associated with CO2 emissions) while the last two variables are likely to be negatively linked to the environmental quality. All data come from the World Bank Development Indicators database and are in natural logarithms (i.e., each estimated coefficient is a constant elasticity of the dependent variable with respect to the independent variable).

The descriptive statistics on the selected variables for each country are displayed in Table 1. It can be observed that the lowest level of renewable energy consumption is 0.7% of total final energy consumption (United Kingdom) while the highest level is 77.8% of total final energy use (Island). The highest value of GDP per capita is up to 112418$ (Luxembourg) while the lowest level of GDP per capita corresponds to 6075$ (Poland). Regarding the CO2 emission per capita, the highest level is equal to 25.6% (Luxembourg) while the lower level of CO₂ emissions is for Chile, Mexico, and Turkey.

To address multicollinearity concerns, Table 2 shows the matrix correlation among explanatory variables. The explanatory variables are not highly correlated so that they can be safely integrated into the model (except for the energy use from fossil fuels, which will be dropped from the final model).

2.2 PSTR model

The R-EKC curve is examined by using the empirical approach proposed by González et al. [36]. This recent method focuses on heterogeneous panels that allow estimated coefficients to vary both across countries and over time. The specification supposes the existence of an infinite number of intermediary regimes, and the coefficients depend upon these regimes. As it is considered a nonlinear panel model, it serves to capture a rise in the level of income that does not affect the income-pollution nexus linearly, but conditionally on the position in the income distribution.

Considering the level of income (GDP per capita) as a transition variable $q_{i,t}$, the PSTR model with two regimes and a single transition function can be written as follows:

$$\mathit{CO}{2}_{\mathit{it}}={\mu }_i+{\beta }_0{\mathit{GDP}}_{i,t}+{\beta }_1{\mathit{GDP}}_{\mathit{it}}f\left(q_{i,t};\gamma ,c\right)+\zeta X_{i,t}+{\epsilon }_{i,t}$$E1

where $\mathit{CO}{2}_{\mathit{it}}$ is the dependent variable (the carbon dioxide emissions per capita for country i at time t), ${\mu }_i$ the individual fixed-effects, ${\mathit{GDP}}_{i,t}$ is the GDP per capita of the country i, at time t, the $f\left(q_{i,t},\gamma ,c\right)$ is the transition function, and ${\epsilon }_{i,t}$, the error term, which is i.i.d (0, ${\sigma }_{\epsilon }^2$). $X_{i,t}$corresponds to the vector of the control variables including renewable energy use or renewable electricity output (${\mathit{RE}}_{\mathit{it}}\mathit{\text{or}}{\mathit{RE}}_{\mathit{Lit}}$), urban population (${\mathit{UR}}_{\mathit{it}}),$ trade openness (${\mathit{\text{TO}}}_{\mathit{it}})$, foreign direct investment (${\mathit{FDI}}_{\mathit{it}})$. The transition function is continuous and integrable on the interval [0,1] and depends on three parameters: $q_{i,t},$which is the transition variable, $\gamma$ – the slope of the transition function, and c is the vector of location parameters such as c =(c₁,...,c_m)’, with m as the vector dimension.

There is no specific rule regarding the optimal number of thresholds in studying the income-pollution nexus. To identify the optimal number of thresholds, two tests are performed in the next section (the Lagrange Multiplier Wald test and the Lagrange Multiplier Fisher test), which indicate that m=2¹. This means that there are two thresholds of “income” around which the effect of income on pollution is a nonlinear one. However, even in this case (m = 2), there are still a continuum of regimes that lie between the extremes (high pollution/low income and low pollution/high income). Therefore, as the transition variable $q_{i,t}i$ncreases, the effect of “income” progresses from ${\beta }_0$ in the first regime corresponding to $f\left(.\right)=0$to ${\beta }_0+{\beta }_1$ in the second extreme regime corresponding to $f\left(.\right)=1$ and so on. Between two extreme cases $f\left(.\right)=0$ and $f\left(.\right)=1$, the sensitivity/elasticity of pollution to income is computed by differentiating CO₂ emissions with respect to the level of income such as:

$$q_{i,t}=\frac{\partial \mathit{CO}{2}_{i,t}}{\partial {\mathit{GDP}}_{i,t}}={\beta }_0+{\beta }_1\ast f\left(q_{i,t};\gamma ,c\right)$$E2

Furthermore, the sensitivity of CO₂ emissions to GDP per capita can vary under the two extreme regimes ${\beta }_0$ and ${\beta }_0$+${\beta }_1$ and is a weighted average of parameters ${\beta }_0$ and ${\beta }_1$. Note that literature indicates that it is difficult to directly interpret the values of these parameters, but easier to interpret their sign as an increase or decrease in the elasticity, depending on the value of the transition variable, and the individual and time dimension given by the previous equation.

2.3 Pre-tests: test for linearity (homogeneity) hypothesis

Before estimating the PSTR model, I investigate homogeneity’s model against the PSTR alternative. The Lagrange multiplier (LM) test of homogeneity based on the asymptotic χ2 distributions, their F-versions and the HAC versions are applied to each of the explanatory variables, they being potential “candidate” transition variables in the PSTR. The LM test looks at the null hypothesis of linearity (homogeneity) against the alternative logistic (m=1) or exponent (m = 2) PSTR model. The optimal number of transition functions is obtained by doing tests of no-remaining nonlinearity. The linearity supposes testing H0:$\gamma$ = 0 or H0:$\alpha =\beta$, but in both cases the test is non-standard, since under H0 the PSTR model has unidentified nuisance parameters. Usually, it is proposed to replace the transition function $f\left(q_{i,t};\gamma ,c\right)$by its first-order Taylor expansion around γ = 0 and to test an equivalent hypothesis in an auxiliary regression such as:

$$c_{i,t}={\omega }_i+\mathit{\alpha CO}{2}_{i,t}+{\theta }_1\mathit{CO}{2}_{i,t}^2+{\epsilon }_{i,t}$$E3

In this first-order Taylor expansion, the parameter ${\theta }_1$ is proportional to the slope parameter γ. Thus, testing the linearity against the PSTR model simply means testing: H0: ${\theta }_1=0$ in this linear panel model. This can be done by using standard tests such as the F-statistic.

2.4 Pre-tests: selecting the number of transition functions

The next step aims to identify the optimal number of thresholds (m) of the logistic transition function by using the LM homogeneity test. The procedure is like when testing the number of transition functions in the model. In the PSTR framework, it is assumed that the linearity hypothesis is rejected. Then the idea is to test whether there is one transition function (H0: m = 1) or whether there are at least two transition functions (H0: m = 2). Let now us consider the model with m = 2 regimes:

$$\mathit{CO}{2}_{\mathit{it}}={\mu }_i+{{\beta }_0}^{\prime }{\mathit{GDP}}_{i,t}+{{\beta }_1}^{\prime }{\mathit{GDP}}_{\mathit{it}}f\left(q_{1i,t};{\gamma }_1,c_1\right)+{{\beta }_2}^{\prime }{\mathit{GDP}}_{\mathit{it}}f\left(q_{2i,t};{\gamma }_2,c_2\right)+\zeta X_{i,t}+{\epsilon }_{i,t}$$E4

If we replace the second transition function $\left(q_{2i,t};{\gamma }_2,c_2\right)$ by the first-order Taylor expansion around ${\gamma }_2$ = 0 and test linear constraints on the parameters, the model becomes:

$$\mathit{CO}{2}_{\mathit{it}}={\mu }_i+{{\beta }_0}^{\prime }{\mathit{GDP}}_{i,t}+{{\beta }_1}^{\prime }\mathit{\text{GDPf}}\left(q_{1i,t};{\gamma }_1,c_1\right)+\theta {\mathit{GDP}}_{\mathit{it}}\ast q_{i,t}+\epsilon {\prime }_{i,t}$$E5

The test of no-remaining nonlinearity is defined by H0: θ = 0; consider SSR₀ the panel sum of squared residuals under H₀, i.e., in a PSTR model with one transition function. SSR₁ is the sum of squared residuals of the transformed model in (5). The testing procedure shows that given a PSTR model with m = m^∗, the null hypothesis will be tested such as: H₀: m = m^∗ against H₁: m = m^∗ + 1. If H₀ is not rejected, then the procedure ends. Otherwise, the null hypothesis H₀: m = m^∗ + 1 is tested against H₁: m = m^∗ + 2. The testing procedure continues until the first acceptance of H₀.

3. Empirical results and discussion

3.1 Pre-tests discussion

Before estimating the PSTR model, I performed a homogeneity test in which the income per capita is a transition variable (q). The results of this test (Table 3 of Appendix) indicate whether a PSTR model can assess the effect of income on environmental pollution. The three statistics (LR, LMF, and LRT – the Wald, Fisher, and Likelihood ratio test statistics) indicate that the hypothesis of a homogeneous effect of income on environmental pollution is rejected. Furthermore, LM, LMF, and LRT p-values are smaller for the exponential model meaning that it is favored over the logistic one.

Considering these results, income influences environmental degradation in different ways. The test of remaining regimes (Table 4 of Appendix) concludes on the existence of a two-threshold level of income. The LM and LMF statistics of these two tests are statistically significant at 1% and 5% levels, which indicate a rejection of the null hypothesis and the acceptance of at least two transition functions. Table 5 of the Appendix complements the previous finding by indicating that three extreme regimes (m = 2) are enough to capture the nonlinearity of the income-pollution nexus based on the modulating effect of the income level. The optimal number of location parameters is given by the Akaike and Schwartz criterion and their values are also shown.

Table 6 of the Appendix gives thresholds of income per capita for 34 OECD economies. The threshold of income remains quite similar among different specifications ranging from 9.29 to 10.71. Because data are in natural logarithms, the exact value of the threshold is computed by applying an exponential function to the constant C. This transformation gives us the level of income per head around which income exerts a negative or a positive effect on the CO₂ emissions. Hence, the threshold of income ranges between about 10 774$ and 44968$.

For the advanced economies, the threshold of 25245$ means that below this level of income, each group of countries will have higher levels of pollution. Conversely, it is only above this high threshold that these countries can improve the environmental degradation. Therefore, less income means more environmental pollution for these economies. An examination of the possibility of each country having more income based on the estimated threshold value of PSTR 1 and PSTR 2 (25245$ and 26132$ in the first regime) found that about a third of the advanced-economies countries (13) can respond to this threshold. These countries are Turkey and the new EU members. The rest of OECD countries have income per head superior to 25000$.

3.2 PSTR discussion

3.2.1 The R-EKC curve discussion

In Table 7, the results show that the direct effect of income on CO₂ emissions, measured by the b₀ coefficient, is positive and significant in all the PSTR estimates. This means that in early stage of development, more income means more pollution (such as in [31]). In the second and third regimes, the coefficients become negative and statistically significant for the PSTR 1 and PSTR 2 (in the second regime) and PSTR 3 and PSTR 4 (in the third regime), respectively.

The next rows in Table 6 provide interesting insights: here the impact of income on pollution is conditional on the level of GDP per capita. More specifically, the b₁ coefficient, associated with the nonlinear component of the model, is always negative and significant at the 1% level (PSTR 1 and PSTR 2 including renewable energy consumption), with values ranging between −0.343 and −0.307. According to the exponential function, this implies that the elasticity of CO₂ emissions with respect to income changes from 0.34 or 0.42 (as b₀ takes these values in PSTR 1 and PSTR 2) to b₁, i.e., as pollution goes from high to low values. The shift between these two extreme regimes occurs around the associated endogenous threshold parameter c (shown in rows 6 and 7), which equals 25243$–44494$ (as in [14]). In the low pollution regime (indices ≥ 25243), the effect of income on environmental degradation is negative and statistically significant at the 1% level; in the extreme case (when g (s_it; γ, c) = 1), all other things being equal, a 1% increase in the income gives a 0.31–0.34% reduction in CO₂ emissions (the coefficient in the second regime being equal to β_{0 +} β₁). Given that there is a continuum of possible points between these two extreme regimes, the elasticity is a weighted average of the parameters β₀ and β_1. This result implies that without a higher income level, advanced economies cannot optimally benefit from a sound environmental quality, and that any policy would be not so much effective.

The paper finds also that the impact of income on CO₂ emissions is nonlinear. Based on the three statistics (LR, LMF, and LRT) reported in Table 3 of Appendix, the hypothesis of a heterogeneous influence of income on pollution is accepted. Thus, income affects CO₂ emissions in various ways. The test of remaining regimes (Table 4 of Appendix) complements this finding and concludes on the existence of two thresholds for GDP per capita. Another argument in support of the nonlinear hypothesis is that the slope of the transition function differs between different regimes and the four PSTR. The higher the γ, the sharper the change from one extreme regime to another. For the PSTR 2, for example, results show that any effort in terms of decreasing pollution by a country just below the threshold value of 26132$ is likely to result in a sharp decrease in the elasticity of pollution with respect to income (from 0.42 to −0.34). However, for a country that is far below this threshold, the same effort will have little effect on the elasticity of pollution. Conversely, in the PSTR 1 it is found a smooth transition. This means that, unlike with the sharp transition previously defined, any effort to combat pollution, even by a country far below the threshold value, will always be recompensated (by a gradual fall in the marginal effect of pollution). Similar features are found for PSTR 3 and PSTR 4 including renewable electricity output among control variables.

3.2.2 The explanatory variables impact

Next, consistent with previous empirical literature, control variables are integrated such as trade openness, FDI, urbanization, and renewable energy consumption or renewable electricity output. These explanatory variables were almost significant and had the expected signs (except for urbanization).

Trade openness exerts a negative and statistically significant effect on pollution, which resonates with the technological effect of trade on environmental quality (contrary to [31] where a positive significant influence of trade openness on carbon emissions is found, validating rather a scale effect of trade on environmental pollution). Foreign direct investments are essential for improving productivity and increasing the competitiveness of an economy. This coefficient is negative or positive but statistically insignificant in all PSTR models. A counterfactual result is found for the urbanization. Indeed, an increase in the urban population should increase the level of emissions, particularly, carbon dioxide emissions; thus, countries that have higher urban populations are expected to pollute the environment more than other countries.

The coefficients of renewable energy consumption or renewable energy output (used for robustness purposes) are always negative and statistically significant at 1% level. This means that clean sources of energy reduce greenhouse emissions and protect environment. Therefore, governments must support the development of this burgeoning energy sector as well as the implementation of carbon taxes to discourage the use of fossil fuels (as a conventional energy source) and to protect environment. Furthermore, such “new-generation” energy policies are expected to be key for a complete decarbonization of the energy sector to achieve sustainable development goals of the IPAC or EU Green Deal by 2030 (which is in line with findings of [22]).

4. Conclusions

Despite a growing empirical literature, the income-pollution nexus remains a controversial topic. The puzzling results of previous studies incite to investigate it in-depth to better assess this issue. I examine the EKC curve in the case of OECD countries and tested the nonlinearity of the income-pollution nexus. To this end, I used the PSTR model of González et al. [36] to capture the heterogeneous effect of income across 34 advanced economies by using yearly data from the World Bank between 1997 and 2015. The overall impact of income on environmental degradation is slightly positive for the 34 OECD countries. This means that technological and structural effects do not exceed the scale effect regarding this relationship. Based on these results, I suggest that policymakers should consider supporting the development of this new energy sector by improving the quality of the existing clean technologies, by financing R&D investments in other promising renewable technologies and related infrastructure network to make renewable energy sources more competitive than fossil fuels and, to reduce, in this way, CO₂ emissions and enhance the energy security. Promoting regional cooperation and developing clean energy technologies will improve the environmental quality of countries, too.

Appendix

Variable	OBS	Mean	Std. dev.	Min	Max
CO2	646	9,1	4,1	3,1	25,6
GDP	646	34621,5	20844,8	6076,0	112417,9
TO	646	88,6	51,0	18,1	351,1
UR	646	88,6	11,3	50,7	97,9
REL	646	76,4	27,2	0,0	100,0
RE	646	27,3	15,6	0,7	77,8
FDI	646	5,6	11,5	−27,7	138,2

Table 1.

Summary statistics for all variables in this study.

	GDP	NE	RE	UR	TO	FDI	REL
GDP	1
NE	0,69	1
RE	0,03	0,08	1
UR	0,43	0,41	−0,04	1
TO	0,08	0,13	−0,02	−0,19	1
FDI	0,40	0,26	−0,15	0,17	0,33	1
REL	0,26	0,15	0,68	−0,07	−0,07	−0,02	1

Table 2.

The matrix correlation between the explanatory variables for OECD-34.

Note: GDP – real GDP per cap, NE – energy use per capita, RE - renewable energy consumption; UR–Urbanization, TO-trade openness, FDI- foreign direct investment, REL - renewable electricity output.

Results of the test of linearity

Stats	Thresholds
	m = 1		m = 2		m = 3
PSTR 1 : Co2 = f(GDP, GDP2, RE, UR)
LM	34.068***	(0.000)	10.609***	(0.014)	30.185***	(0.000)
LMF	11.301***	(0.000)	3.356**	(0.019)	9.803***	(0.000)
LRT	34.999***	(0.000)	10.697***	(0.013)	30.913**	(0.000)
PSTR 2: Co2 = f(GDP, GDP2, RE, UR, TO, FDI)
LM	40.503***	(0.000)	34.724***	(0.000)	38.901***	(0.000)
LMF	8.121***	(0.000)	6.783***	(0.000)	7.587***	(0.000)
LRT	41.128***	(0.000)	35.692***	(0.000)	40.121***	(0.000)
Robustness Checks
PSTR 3: Co2 = f(GDP, GDP2, REL, UR)
LM	117.434***	(0.000)	33.100***	(0.000)	18.071***	(0.000)
LMF	45.101***	(0.000)	10.855***	(0.000)	5.756***	(0.000)
LRT	129.608***	(0.000)	33.978***	(0.000)	18.328***	(0.000)
PSTR 4: Co2 = f(GDP, GDP2, REL, UR, TO, FDI)
LM	125.056***	(0.000)	37.713***	(0.000)	24.099***	(0.000)
LMF	29.143***	(0.000)	7.403***	(0.000)	4.588***	(0.000)
LRT	138.991***	(0.000)	38.859***	(0.000)	24.560***	(0.000)

Table 3.

LM, LMF, and LRT tests of linearity (p-values) for OECD-34.

^***

statistical significance at 1%

^**

statistical significance at 5% level

Results of test for the number of regimes

Institut. Var.	Hypotheses	Test	Stat	(p-value)
PSTR with Pub. Debt
PSTR 1	H0 : m = 0 vs H1 : m = 1	LM	34.068***	(0.000)
		LMF	11.301***	(0.000)
	H0 : m = 1 vs H1 : m = 2	LM	10.609***	(0.014)
		LMF	3.356**	(0.019)
	H0 : m = 2 vs H1 : m = 3	LM	30.185***	(0.000)
		LMF	9.803***	(0.000)
PSTR 2	H0 : m = 0 vs H1 : m = 1	LM	40.503***	(0.000)
		LMF	8.121***	(0.000)
	H0 : m = 1 vs H1 : m = 2	LM	34.724***	(0.000)
	LMF	6.783***	(0.000)
	H0 : m = 2 vs H1 : m = 3	LM	38.901***	(0.000)
		LMF	7.587***	(0.000)
Robustness Checks
PSTR 3	H0 : m = 0 vs H1 : m = 1	LM	117.434***	(0.000)
		LMF	45.101***	(0.000)
	H0 : m = 1 vs H1 : m = 2	LM	33.100***	(0.000)
		LMF	10.855***	(0.000)
	H0 : m = 2 vs H1 : m = 3	LM	18.071***	(0.000)
		LMF	5.756***	(0.000)
PSTR 4	H0 : m = 0 vs H1 : m = 1	LM	125.056***	(0.000)
		LMF	29.143***	(0.000)
	H0 : m = 1 vs H1 : m = 2	LM	37.713***	(0.000)
		LMF	7.403***	(0.000)
	H0 : m = 2 vs H1 : m = 3	LM	24.099***	(0.000)
		LMF	4.588***	(0.000)

Table 4.

Test for the number of regimes for OECD-34.

^***

statistical significance at 1%.

^**

statistical significance at 5% level.

Determination of the number of location parameters

PSTR 1	Optimal no. of transition functions	2
	Residual Sum of Squares	2.632
	Number of Parameters	13
	AIC Criterion	−5.441
	BIC Criterion	−5.351
PSTR 2	Optimal no. of transition functions	2
	Residual Sum of Squares	2.430
	Number of Parameters	19
	AIC Criterion	−5.492
	BIC Criterion	−5.361
Robustness Checks
PSTR 3	Optimal no. of transition functions	2
	Residual Sum of Squares	3.534
	Number of Parameters	13
	AIC Criterion	−5.146
	BIC Criterion	−5.056
PSTR 4	Optimal no. of transition functions	2
	Residual Sum of Squares	3.088
	Number of Parameters	19
	AIC Criterion	−5.253
	BIC Criterion	−5.122

Table 5.

Determination of the number of location parameters for OECD-34.

Notes: (i) For each model, the optimal number thresholds, denoted r (m), are determined according to a sequential procedure based on the LMF statistics of the hypothesis of non-remaining nonlinearity.

(ii) The RSS value is given for each couple (m; r). (iii) The AIC and BIC criterions are shown for each PSTR.

Models	Speed of transit. – $\gamma$1	Speed of transit. - $\gamma 2$	Threshold param- c1	Threshold param- c2
PSTR 1	22.6835	25.3073	10.1363	10.7031
PSTR 2	1273.8	28.2	10.1709	10.7091
PSTR 3	17.7708	10.1812	10.7002	9.2849
PSTR 4	10.9579	7.9411	10.7137	9.2925

Table 6.

Results of threshold values of GDP per capita for OECD-34.

Variable	PSTR 1		PSTR 2		PSTR 3		PSTR 4
	coeff.	t-Stat	coeff.	t-Stat	coeff.	t-Stat	coeff.	t-Stat
GDP per capita	0.3372^***	10.5552	0.4180^***	10.374	0.8670^***	8.5244	0.9405^***	7.8088
GDP per capita*g1 (.)	−0.3071^***	−4.2881	−0.3426^***	−4.833	−0.0508	−1.0829	−0.0503	−0.7263
GDP per capita*g2 (.)	−0.0148	−0.3491	−0.0412	−0.709	−0.5227^***	−6.6361	−0.4600^***	−4.1124
Transition parameters
Speed of transition – γ1	22.6835		1273.8		17.7708		10.9579
Speed of transition – γ2	10.7031		28.2		10.1812		7.9411
Threshold parameter – c1	10.1363; 25242.895		10.1709; 26131.585		10.7002; 44364.7272		10.7137; 44967.712
Threshold parameter – c2	10.7031	44493.572	10.7091	44761.336	9.2849	10774.095	9.2925	10856.2909
Control variables
Renewable energy consumption	−0.2784^***	−18.237	−0.2510^***	−15.724	−0.0456^***	−3.6457	−0.0433^***	−3.2781
Urbanization	−0.7158^***	−5.672	−0.6543^***	−4.864	−2.0202^***	−8.1128	−1.5299^***	−5.546
Openness	-	-	−0.1465^***	−3.408	-	-	−0.1722^**	−2.236
FDI	-	-	0.0055	1.236	-	-	−0.0199	−1.559
No. Obs	646 × 5		646 × 7		646 × 5		646 × 7
No. Countries	34		34		34		34

Table 7.

PSTR estimates with the real GDP per capita as the threshold variable: 1997–2015, OECD-34.

Note: i) ^{*** −} p-value < 1% ^{; **} - p-value < 5%; ^* - p-value < 10%; ii) the PSTR 3 and PSTR 4 include the renewable electricity output as a robustness checks exercise; iii) the coefficients of the control variables are those of the first linear regime; iv) in the second regime, RE coefficient is positive and statistically significant. The threshold parameter reflects the threshold of income around to which the income effect on pollution changes (from a positive one to a negative one or vice-versa).

JEL Classification

Q53, Q56, F64, C54