The Role of Natural Gas in Mitigating Greenhouse Gas Emissions: The Environmental Kuznets Curve Hypothesis for Major Gas-Producing Countries ^†

Abstract

Since global warming has become a serious threat and GHG emissions are one of the main causes of it, analyzing the interactions between variables related to climate change has gained importance. This study investigates the nexus of per capita CO₂ emissions, per capita real GDP or income, per capita natural gas consumption, urban population, and trade openness by examining the validity of the environmental Kuznets curve (EKC) hypothesis for a panel of selected gas-producing countries over the period 1990–2020. To these data, slope homogeneity test, Granger causality in panels, stationarity tests, and cointegration tests are applied. A particular focus is on procedures that enable cross-sectional dependence. Admitting slope heterogeneity, the estimators provide mixed results. The findings, however, do provide evidence in favor of the EKC hypothesis in at least some of our sample countries. Furthermore, there are important policy implications that must be taken into consideration. This includes investing in clean technologies to reduce emissions and accelerating reform of fossil fuel subsidies.

1. Introduction

Climate change and global warming are among the most discussed issues in the world, and they have important implications for the economy. Global energy-related carbon dioxide (CO₂) emissions have been growing substantially over recent decades. Between 1990 and 2020, CO₂ emissions increased by 51% from 2.1 gigatons (Gt) of CO₂ to over 3.2 Gt [1]. The largest sources of CO₂ emissions are power generation, transport, and the industrial sector. In 2020, these three sectors together comprised about 72% of the total CO₂ emissions in the world (see Figure 1).

The key climate-related objective agreed upon at the 2015 Paris Conference on Climate Change (COP21) was to restrict global warming to less than 2 °C, ideally 1.5 °C, by 2100 compared to pre-industrial levels. The Paris Agreement has served as the backdrop for most of today’s greenhouse gas (GHG) emissions debate.

The main task for the Glasgow meeting (COP26)—held in November 2021—was to finalize the procedures for implementation of the COP21. The Glasgow Climate Pact asks countries to accelerate efforts towards phase-out of unabated coal power and inefficient fossil fuel subsidies. Although the COP26 appears to reinforce momentum toward deep decarbonization and clean energy transition, uncertainties loom large, affecting the ability to deliver on the announced pledges and achieve 1.5 °C compatible emissions pathways by mid-century.

That being said, abandoning use of all fossil fuels may seem like an ideal path toward 1.5 °C. Although cost of renewable energies is declining, renewables are still facing major obstacles, including but not limited to high capital costs, intermittency and variability, transmission infrastructure, affordability in many countries and regions, as well as lack of storage capacity [2].

In this context, natural gas as an affordable, reliable, and abundant source of energy represents an excellent alternative to reduce CO₂ emissions and help to combat global warming. Combustion of natural gas emits about half as much CO₂ as coal and 30% less than oil, as well as far fewer pollutants per unit of energy delivered. In addition, proven technology exists to reduce natural gas emissions, which ultimately could make natural gas even cleaner.

Moreover, natural gas exhibits complementary characteristics to address the drawbacks of renewables, namely a reliable, foreseeable production profile as gas is already part of the “baseload” production profile (displacing coal) in several countries. Furthermore, natural gas is becoming an important pillar of decarbonization and will increase its share in the global energy mix from 23% in 2020 to 27% in 2050 (see Figure 2) [1].

Taking into consideration the above-mentioned aspects, adoption of suitable policies regarding energy and gas consumption, economic development, and environmental considerations is crucial for governments and policymakers. Therefore, it is vital to investigate the relationship between natural gas consumption and CO₂ emissions, which is important both for battling against CO₂ emissions and promoting natural gas demand.

In this context, the environmental Kuznets curve (EKC) hypothesis became popular among scholars as it represents an important tool for environmental policy. The EKC claims that, in the early stages of economic growth (determined by per capita), environmental degradation increases, but, beyond some level of economic growth, the trend will reverse, so, at high-income levels, economic growth leads to environmental improvement. This implies an inverted U-shaped relationship between economic growth and environmental degradation [3,4,5].

In the context of sustainability, the EKC suggests that, as countries become more economically developed, they will be better equipped to address environmental problems, leading to reduction in environmental damage. However, the relationship between economic growth and environmental damage is complex, and addressing sustainability challenges will require a multifaceted approach, including use of renewable energy sources, investment in clean technologies, and other mitigation strategies (e.g., changing customer behavior and increasing and enhancing energy efficiency in buildings).

This study addresses two main research questions: first, what is the role of natural gas consumption in mitigating CO₂ emissions in the sample of 12 major gas-producing and exporting countries? Second, does an inverted U-shaped reaction curve represent the relationship between economic growth (income) and CO₂ emissions?

In particular, the study aims to fill a gap in the EKC literature by focusing on two main objectives:

• To the best of our knowledge, very few empirical studies have examined the validity of the EKC hypothesis with the impact of natural gas consumption on CO₂ emissions in the case of our sample countries.
• Standard panel data techniques assume cross-sectional independence. However, this assumption is hard to satisfy due to the high degree of socioeconomic integration among countries, and this may create biased and inconsistent estimates in misleading conclusions [6,7,8]. Therefore, this analysis utilizes second-generation econometric approaches, such as the common correlated effects mean group (CCEMG) of Pesaran [9] and the augmented mean group (AMG) estimator of Eberhardt and Teal [10], assuming cross-sectional dependence to provide more robust analysis. Although there are many studies using either first- or second-generation panel data methods, there are few studies that use both methods.

Based on the above-mentioned theorem, this study aims to examine the causal relationship between CO₂ emissions, economic growth, natural gas consumption, population, and trade openness for a panel of 12 major natural-gas-producing countries, spanning 1990 to 2020 in the context of EKC hypothesis modelling.

The remainder of this paper is structured as follows. Section 2 reviews the related literature. Section 3 details the methodology and data that have been used for analysis. Section 4 presents the empirical results. The final section concludes this study and proposes some energy policy implications.

2. Literature Review

The EKC was named for Kuznets [11], who argued that income inequality first increases with economic development and then falls as the economy develops to a certain extent. Grossman and Krueger [12] first proposed the hypothesis in the environmental context when they used three air pollutant measures for forty-two countries to study the relationship between air quality and economic growth. They found that sulfur dioxide and smoke first increase with per capita income but then decrease as per capita income reaches a certain level.

Following the seminal research, the EKC hypothesis was tested in plenty of studies. Different control variables were included in the EKC models to investigate the role of various kinds of energy in affecting carbon emissions. The results vary depending on the method and sample used. A summary of some of those studies is provided in Table 1.

To the best of our knowledge, however, very few studies examine the EKC hypothesis by analyzing effectiveness of natural gas consumption, especially in the case of natural-gas-producing countries. Among them, Zambrano-Monserrate et al. [13] examine the impact on emissions from electricity consumption from renewable sources as well as from petroleum and natural gas in Peru. The findings do not support the EKC hypothesis, but they show that gas and petroleum consumption have positive effects on CO₂ emissions, while electricity consumption from renewable sources has a negative impact.

Dong et al., [7] investigate effectiveness of natural gas consumption in a group of 14 Asia-Pacific nations from 1970 to 2016. They argue that the EKC will arise in any country regardless of whether it is a low-income or high-income country. Moreover, CO₂ emissions are negatively related to natural gas consumption over the long run. However, positive causality from natural gas consumption to CO₂ emissions might appear in some Asia-Pacific countries.

In another study, by employing the ARDL approach and a data span from 1995 to 2014, Dong et al., [8] explore the causal relationship between natural gas consumption and CO₂ emissions in the short- and long run in Beijing, China. They find that the move away from coal and petroleum to natural gas can reduce CO₂ emissions. However, the mitigation effect of natural gas consumption would be weakened over time.

Layachi [14] investigates the association between crude oil, natural gas, and heating oil prices and environmental pollution in Algeria from 1980 to 2017. The study finds that economic growth has a positive and significant impact on CO₂ emissions. Furthermore, all energy prices confirm a negative and significant impact on CO₂; however, no causal relationship between natural gas prices and CO₂ emissions is found.

Lotfalipour et al. [15] study the link between fossil fuel, natural gas, and petroleum products with CO₂ emissions in Iran. The study used data from the period of 1967 to 2007. Applying the method of Toda–Yamamoto Granger causality, the study shows unidirectional Granger causality running from GDP, petroleum products, and natural gas consumption to carbon emissions.

Li et al. [16] assess the impact of natural gas consumption on CO₂ emissions based on the panel data of 30 provinces in China from 1998 to 2016. The study realizes that per capita GDP is effective in controlling emissions. Moreover, in regions with low natural gas usage, unidirectional causality is established from natural gas consumption to CO₂ emissions. The different causalities support that use of natural gas has a negative impact on emissions.

Murshed et al. [17] notice that consumption of natural gas, liquefied petroleum gas (LPG), and hydropower tend to decrease CO₂ emissions. They suggest that natural gas and LPG could both function as transitionary fuels to mitigate environmental pollution in Bangladesh.

Solarin and Lean [18] examine the impact of natural gas consumption on CO₂ emissions in China and India for the period of 1965 to 2013. They find that natural gas has a long-run positive impact on emissions in both countries.

In a series of publications (see, for example, Müller-Fürstenberger and Wagner [19] and Wagner [20]), Martin Wagner and co-authors discuss several major econometric problems that have been ignored in the empirical EKC literature, including but not limited to use of nonlinear transformations of integrated regressors of potentially nonstationary regressors and cross-sectional dependence, as well as develop a FMOLS estimator for cointegrating polynomial regressions (CPRs). These are critical of cointegration methods that use a variable and its square, something characteristic of EKC. Thus, the problems are not just of academic interest but can lead to contrary conclusions concerning the prevalence of an EKC hypothesis.

For instance, Fürstenberger and Wagner [19] argue the implications of the nonlinear transformations of nonstationary regressors and of neglecting cross-sectional dependence in a nonstationary panel context in the econometric analysis of EKC. They conclude that, in estimating the EKCs, the literature should pay attention to the fact that Kuznets curve regressions involve nonlinear transformation of integrated regressors (i.e., GDP) and that almost all panels of economic time series are cross-sectionally dependent.

Employing empirical analysis for CO₂ and SO₂ emissions data of 19 countries, Wagner [20] clarifies some conceptual shortcomings of the empirical EKC literature that arise because of the hitherto inadequate application of unit root and cointegration techniques. He presents a methodology that is suited to analysis of CPRs and argues that the powers of integrated processes are themselves not integrated processes. Using appropriate methods leads to strongly reduced evidence for a cointegrating EKC compared to typical but conceptually not sound findings.

Based on annual data per capita of CO₂ emissions, energy consumption, and income of eight developed and developing countries, Moosa and Burns [21], suggest that an EKC can be estimated separately from the conventional environmental Kuznets curve as a quadratic function of income. By using a variable addition test, as well as the unobserved components model, they found that the quadratic function fits better than the linear function and demonstrated strong support for the Kuznets curves in developed but not developing countries.

Table 1. Summary of literature review (alphabetical order).

Authors	Period	Study Area	Variables	Method	Results
Alam and Adil [22]	1971–2016	India	CO2, GDP, PES, FD, TO	ARDL	No credence to EKC
Apergis and Ozturk [23]	1990–2011	Asian countries	CO2, GDP, LAND, POP, Industry Shares in GDP	GMM	Validity of EKC
Atasoy [24]	1960–2010	The US	CO2, GDP, EC, POP	AMG, CCEMG	AMG validates the EKC. CCEMG provides only weak evidence on the EKC
Baek [25]	1980–2009	12 major nuclear-generating countries	CO2, GDP, EC, NE	DOLS, FMOLS	CO2 emissions tend to decrease monotonically with income growth, providing no evidence in support of EKC for CO2 emissions
Jammazi and Aloui [26]	1980–2013	GCC countries	CO2, EC, GDP	Wavelet Approaches	Bidirectional causality between EC and GDP, and unidirectional causality from EC to CO2
Kohler [27]	1960–2009	South Africa	CO2, GDP, EC, TO	ARDL, Johansen Cointegration, VECM	Positive bidirectional causality exists
Lin et al., [28]	1980–2011	Nigeria	CO2, EI, POP, CI, IVD, GDP	VECM	An inverse significant relationship between IND and CO2
El-Aasar and A. Hanafy [4]	1971–2012	Egypt	GHG, GDP, RE, TO	ARDL	EKC hypothesis does not exist for GHG emissions in Egypt for both short- and long term
Ozcan, B. [29]	1990–2008	Middle East countries	CO2, GDP, EC	FMOLS, VECM	Unidirectional causality from economic growth to energy consumption in the short run
Paramati et al., [30]	1991–2012	G20 countries	CO2, RE, NRE, POP, GDP, FDI, SMC	Panel Granger causality	Confirm a significant long run equilibrium relationship among the variables
Rafindadi [31]	1961–2012	Japan	CO2, GDP, EX, IM	ARDL	Presence of EKC despite the deteriorating income of the country
Saboori et al., [32]	1977–2008	OPEC countries	CO2, GDP, Labor, Capital, OC, OP	ARDL, TYDL	EKC approves. Oil prices reduce environmental damage by their negative effect on the ecological footprint
Saidi and Mbarek [33]	1990–2013	19 emerging economies	CO2, NE, RE, GDP	ARDL	Invalidity of EKC
Ulucak et al., [34]	1992–2016	BRICS countries	GDP, RE, Urbanization, Natural Resource Rent, Ecological Footprint	FMOLS, DOLS	Approve of EKC in case of BRICS Countries
Wang et al., [35]	1990–2012	China	CO2, EC, GDP	VECM	Shocks in CO2 have a small effect on EC and GDP
Zoundi [36]	1980–2012	25 African countries	CO2, GDP, RE	FMOLS, DOLS	No evidence of a total validation of EKC. However, CO2 emissions increase with GDP

Source: authors’ elaboration. Note: Carbon Dioxide (CO₂), Carbon Intensity (CI), Energy Consumption (EC), Energy Efficiency (EE), Energy Intensity (EI), Financial Development (FD), Foreign Direct Investment (FDI), Greenhouse Gas (GHG), Gross Domestic Product (GDP), Industrial Value-Added (IVD), Nuclear Energy (NE), Oil Consumption (OC), Oil Price (OP), Population (POP), Primary Energy Supply (PES), Real Exports (EX), Real Imports (IM), Renewable Energy (RE), Stock Market Capitalization (SMC), Trade Openness (TO), Augmented Mean Group (AMG), Autoregressive Distributed Lag (ARDL), Common Correlated Effects Mean Group (CCEMG), Dynamic Ordinary Least Squares Estimator (DOLS), Fully Modified OLS (FMOLS), Generalized Method Of Moments (GMM), Toda–Yamamoto–Dolado–Lutkepohl (TYDL), Vector Error Correction Model (VECM).

Table 1. Summary of literature review (alphabetical order).

Authors	Period	Study Area	Variables	Method	Results
Alam and Adil [22]	1971–2016	India	CO2, GDP, PES, FD, TO	ARDL	No credence to EKC
Apergis and Ozturk [23]	1990–2011	Asian countries	CO2, GDP, LAND, POP, Industry Shares in GDP	GMM	Validity of EKC
Atasoy [24]	1960–2010	The US	CO2, GDP, EC, POP	AMG, CCEMG	AMG validates the EKC. CCEMG provides only weak evidence on the EKC
Baek [25]	1980–2009	12 major nuclear-generating countries	CO2, GDP, EC, NE	DOLS, FMOLS	CO2 emissions tend to decrease monotonically with income growth, providing no evidence in support of EKC for CO2 emissions
Jammazi and Aloui [26]	1980–2013	GCC countries	CO2, EC, GDP	Wavelet Approaches	Bidirectional causality between EC and GDP, and unidirectional causality from EC to CO2
Kohler [27]	1960–2009	South Africa	CO2, GDP, EC, TO	ARDL, Johansen Cointegration, VECM	Positive bidirectional causality exists
Lin et al., [28]	1980–2011	Nigeria	CO2, EI, POP, CI, IVD, GDP	VECM	An inverse significant relationship between IND and CO2
El-Aasar and A. Hanafy [4]	1971–2012	Egypt	GHG, GDP, RE, TO	ARDL	EKC hypothesis does not exist for GHG emissions in Egypt for both short- and long term
Ozcan, B. [29]	1990–2008	Middle East countries	CO2, GDP, EC	FMOLS, VECM	Unidirectional causality from economic growth to energy consumption in the short run
Paramati et al., [30]	1991–2012	G20 countries	CO2, RE, NRE, POP, GDP, FDI, SMC	Panel Granger causality	Confirm a significant long run equilibrium relationship among the variables
Rafindadi [31]	1961–2012	Japan	CO2, GDP, EX, IM	ARDL	Presence of EKC despite the deteriorating income of the country
Saboori et al., [32]	1977–2008	OPEC countries	CO2, GDP, Labor, Capital, OC, OP	ARDL, TYDL	EKC approves. Oil prices reduce environmental damage by their negative effect on the ecological footprint
Saidi and Mbarek [33]	1990–2013	19 emerging economies	CO2, NE, RE, GDP	ARDL	Invalidity of EKC
Ulucak et al., [34]	1992–2016	BRICS countries	GDP, RE, Urbanization, Natural Resource Rent, Ecological Footprint	FMOLS, DOLS	Approve of EKC in case of BRICS Countries
Wang et al., [35]	1990–2012	China	CO2, EC, GDP	VECM	Shocks in CO2 have a small effect on EC and GDP
Zoundi [36]	1980–2012	25 African countries	CO2, GDP, RE	FMOLS, DOLS	No evidence of a total validation of EKC. However, CO2 emissions increase with GDP

Source: authors’ elaboration. Note: Carbon Dioxide (CO₂), Carbon Intensity (CI), Energy Consumption (EC), Energy Efficiency (EE), Energy Intensity (EI), Financial Development (FD), Foreign Direct Investment (FDI), Greenhouse Gas (GHG), Gross Domestic Product (GDP), Industrial Value-Added (IVD), Nuclear Energy (NE), Oil Consumption (OC), Oil Price (OP), Population (POP), Primary Energy Supply (PES), Real Exports (EX), Real Imports (IM), Renewable Energy (RE), Stock Market Capitalization (SMC), Trade Openness (TO), Augmented Mean Group (AMG), Autoregressive Distributed Lag (ARDL), Common Correlated Effects Mean Group (CCEMG), Dynamic Ordinary Least Squares Estimator (DOLS), Fully Modified OLS (FMOLS), Generalized Method Of Moments (GMM), Toda–Yamamoto–Dolado–Lutkepohl (TYDL), Vector Error Correction Model (VECM).

3. Methodology and Data

3.1. Empirical Model

We examine the relationship between CO₂ emissions per capita as a dependent variable, GDP per capita (income), natural gas consumption per capita, urban population (urbanization), and trade openness as the explanatory variables.

In this study, the following linear/logarithmic model is employed to investigate the validity of the EKC hypothesis and determinants of carbon emissions for a sample of 12 major gas-producing countries:

$$CO{2}_{it}= {\propto }_0+{\beta }_{1}GD{P}_{it}+{\beta }_{2}GD{P}_{it}^2+{\beta }_{3}G{C}_{it}+{\beta }_4{U}_{it}+{\beta }_{5}T{O}_{it}+{\epsilon }_{it}$$

(1)

Here, ${\propto }_0$ and ${\epsilon }_{it}$ are the intercept and error term representing other factors affecting CO₂ emissions, respectively. In the sample, $i$ indexes the country $\left(i=1,2,\dots 12\right)$ and $t$ indicates the time index in years (t = 1990, …, 2020).

A quadratic function of $GDP$ is used to capture decreasing marginal income effect on CO₂ emissions. If ${\beta }_1={\beta }_2=0$, this indicates no relationship between economic growth and pollution. If ${\beta }_1>0$ and ${\beta }_2<0$, the relationship between economic growth and pollution is inverted U-shaped, supporting the EKC hypothesis. This means the CO₂ emissions tend to rise with an increase in income up to a certain level, beyond which CO₂ emissions decline with higher income levels. Once the EKC hypothesis is confirmed, interest focuses on the position of the turning point in GDP per capita, beyond which level of pollution declines [4,13,25,37,38].

The coefficient ${\beta }_3$, the partial effect of natural gas demand on CO₂ emissions, is of interest. If an increase in natural gas demand leads to a decrease in CO₂ emissions, then ${\beta }_3$ will be negative. Finally, the signs of ${\beta }_4$ and ${\beta }_5$ are expected to be mixed according to the level of economic development of the country. Nonetheless, in general, a higher level of urbanization means a higher level of economic development, and, consequently, a higher level of CO₂ emissions. Therefore, a bidirectional causal relationship is expected between economic growth, urbanization, and CO₂ emissions. Furthermore, the relationship between economic activities and environmental degradation can be through trade. According to Sobrinho [39], some economists believe that trade brings economic growth, which helps to protect the environment through raised incomes. Conversely, others think that trade increases environmental damage by raising unsustainable consumption and production patterns (e.g., increasing water and air pollution, land and forest degradation, and waste generation). Thus, there could be a bidirectional causal link between economic growth and environmental pollution [40].

Similar to Li et al. [16], Murshed et al. [17], Dong et al. [7,8], Zambrano-Monserrate et al. [13], and Atasoy [24], the estimation methodology in this study contains the following steps:

• Cross-sectional dependence tests to check interdependencies between cross-sectional units (countries) (Section 4.1)
• Slope heterogeneity tests for slope homogeneity in large panels (Section 4.2)
• Panel unit root tests to check the stationarity of the variables (Section 4.3)
• Cointegration tests to check the validity of a cointegration relationship between the variables (Section 4.4)
• Estimate the long-run parameters of explanatory variables (Section 4.5)
• The Dumitrescu and Hurlin (D–H) method to reveal the causal relationships between the variables (Section 4.6)

3.2. Data and Descriptive Statistics

We use annual data from 1990 to 2020 for a sample of 12 gas-producing countries as follows: Algeria, Azerbaijan, Egypt, Iran, Malaysia, Nigeria, Oman, Qatar, Russia, Saudi Arabia, Turkmenistan, and the United Arab Emirates (UAE). According to the GECF [1], as of 2021, these countries together approximately hold 71% (134 tcm) of the world proven gas reserves, 41% (1,610 bcm) of the world dry gas production, and 30% (1150 bcm) of world gas consumption. Furthermore, these countries were responsible for 12% (4030 mt CO₂) of the global CO₂ emissions in 2020 (see Figure 3).

For estimation, all variables are expressed in logarithmic form. The dataset contains a strongly balanced panel composed of 2232 observations (N = 12 and T = 31). After eliminating periods with no data, the sample was reduced to 2220 observations.

Table 2. Data: definition, sources, and descriptive statistics, 1990–2020.

Variables.	Definition	Source	Unit	Obs.	Mean	Std. Dev.	Min	Max
lnCO2	Energy-related CO2 emissions	GECF	Million ton CO2 (Mt CO2)/person	372	1.9021	1.1877	−0.9790	3.9279
lnGDP	Real GDP	GECF	Real 2020 US$/person	372	8.8399	1.1701	6.8237	11.0257
lnGDP2	Quadratic GDP	GECF	Real 2020 US$/person	372	79.5100	21.2761	46.5637	121.567
lnGC	Natural gas consumption	GECF	Million Cubic Meters (MCM)/person	372	7.5359	1.3254	4.1875	10.0420
lnU	Urbanization	United Nations	Share of total population	372	−0.4577	0.2797	−1.2146	−0.0076
lnTO	Trade openness	World Bank	Trade (% of GDP)	360	4.2714	0.4830	3.0312	5.3954

shows definition of variables, data sources, and descriptive statistics.

4. Estimation Techniques and Results

4.1. Cross-Sectional Dependence Test

Cross-sectional dependence is an important issue that should be investigated before estimating any panel data models. Due to the high degree of socioeconomic and political integration among countries, cross-sectional dependence is a priori likely to exist. Ignoring this cross-sectional dependence may create biased estimates and mislead conclusions.

There are a variety of tests for cross-sectional dependence in the literature. In this study, we use the Breusch and Pagan [41] LM test, Pesaran scaled LM test, and the Pesaran CD test [42]. However, according to the literature [7,8,24,42], the Breusch–Pagan LM test is not appropriate for large cross-section dimension (N) or time dimension (T). Since N and T in this study are both rather moderate (12 and 31, respectively), the Pesaran CD test is applied to examine the potential common correlation effects between the variables that use the following statistic that is asymptotically normal under the null hypothesis of cross-section independence:

$$CD=\sqrt{\frac{2T}{N\left(N-1\right)}}\left({\displaystyle \sum }_{i=1}^{N-1}{\displaystyle \sum }_{j=i+1}^N{\widehat{\rho }}_{ij}\right)=>N\left(0,1\right)$$

(2)

Here, ${\widehat{\rho }}_{ij}$ indicates the sample estimate of the pairwise correlation of the residuals obtained by OLS. The correlation coefficients are obtained from the residuals of the model, and the CD statistic is asymptotically standard normal for $T\to \infty$ and $N\to \infty$ in any order. As shown in

Table 3. Cross-sectional dependence test results.

Test	Value
Breusch–Pagan LM	565.4625 ***
Pesaran scaled LM	43.4726 ***
Pesaran CD	5.2719 ***

Note: *** indicates that statistics are significant at a 1% level of significance. H₀: no cross-section dependence (correlation).

, the null hypothesis of no cross-sectional dependence is rejected at a 1% level, and thus a possible shock in any of the sample countries may affect others.

4.2. The Slope Homogeneity Test

Even though many researchers assumed homogeneous slope coefficients in panel data models across individual units, the slope homogeneity assumption often fails to hold in panels with large N and T observations. Therefore, in this section, the slope homogeneity tests proposed by Pesaran and Yamagata [43], and Blomquist and Westerlund [44], which are based on the early work of Swamy [45], are employed. The Blomquist and Westerlund test is valid in the presence of cross-sectional dependence. Moreover, the assumptions of homoscedasticity and serial independence of the Pesaran and Yamagata test can be dropped. The proposed version is consistent with heteroscedasticity and autocorrelation consistent (HAC) counterpart and formed two delta test statistics as follows:

$${\Delta }_{HAC}=\sqrt{N}\left(\frac{N^{-1}{S}_{HAC}-k}{\sqrt{2k}}\right)$$

(3)

$${({\Delta }_{HAC})}_{adj}=\sqrt{N}\left(\frac{N^{-1}{S}_{HAC}-k}{v\left(T,k\right)}\right)$$

(4)

Here, N denotes the number of cross-section units, S denotes the Swamy test statistic, and k denotes the number of explanatory variables. As listed in

Table 4. Slope heterogeneity test results.

Test	Value
Pesaran and Yamagata
$\mathsf{\Delta }$	12.696 ***
${\mathsf{\Delta }}_{adj.}$	14.500 ***
Blomquist and Westerlund
$\mathsf{\Delta }$	11.081 ***
${\mathsf{\Delta }}_{adj.}$	12.656 ***

Note: *** indicates that statistics are significant at a 1% level of significance. H₀: slope coefficients are homogenous.

, the null hypothesis of slope homogeneity is rejected.

4.3. Testing for Stationarity

Since there is cross-sectional dependence in the data, imposing homogeneity when coefficient heterogeneity is present can result in misleading conclusions [46]. In order to investigate stationarity, determine integration level of variables, and overcome spurious regression, in this study, the Levin–Lin–Chu test (LLC test) proposed by Levin et al. [47], the Fisher–ADF panel unit root test developed by Maddala and Wu [48], and the Im, Pesaran, and Shin (IPS test) proposed by Im et al. [49] are employed.

According to the literature [7,8,21,30,50], however, the first-generation panel unit root tests are not valid in the presence of cross-sectional dependence. Therefore, we also employed the second-generation panel unit root test proposed by Pesaran [50], namely the CIPS test, to identify the order of integration as follows:

$$CIPS\left(N,T\right)=N^{-1}{\displaystyle \sum }_{i=1}^N{t}_i\left(N,T\right)$$

(5)

Here, $t_i\left(N,T\right)$ is the cross-sectionally augmented Dickey–Fuller statistic for the ith cross-section unit given by the t-ratio of the coefficient of $y_{i,t-1}$ in the CADF regression defined as follows:

$$\Delta y_{it}={\alpha }_i+b_i{y}_{i,t-1}+c_i{\overline{y}}_{t-1}+d_i\Delta {\overline{y}}_t+{\epsilon }_{it}$$

(6)

Here, ${\overline{y}}_t$ is the average at time T of all N countries, and $\Delta$ is the first difference operator.

As indicated in

Table 5. Results of panel unit root tests.

Variable	Level		1st Difference
LLC Test	Intercept	Intercept and Trend	Intercept	Intercept and Trend
lnCO2	−4.5124 ***	−0.4769	−4.1597 ***	−3.0191 ***
lnGDP	−1.0392	−1.5491 *	−3.3349 ***	−2.0718 *
lnGDP2	−0.9070	−1.5123 *	−3.3014 ***	−2.0293 *
lnGC	−3.8289 ***	−1.3094 *	−11.275 ***	−10.6326 ***
lnU	−8.7882 ***	−6.8910 ***	−8.6010 ***	−52.3669 ***
Fisher–ADF test	Intercept	Intercept and trend	Intercept	Intercept and trend
lnCO2	2.4906 **	1.2074	37.1258 ***	35.7304 ***
lnGDP	−1.8509	−0.1505	12.7983 ***	10.9182 ***
lnGDP2	−1.9454	−0.1126	12.7340 ***	10.8641***
lnGC	1.2259	0.4684	43.5791 ***	39.4963 ***
lnU	41.9397	14.5531	2.4923 **	10.1618 ***
lnTO	4.2778 ***	2.1929 **	35.9704 ***	30.6290 ***
IPS test	Intercept	Intercept and trend	Intercept	Intercept and trend
lnCO2	−1.1238	−0.9496	−9.9337 ***	−10.4638 ***
lnGDP	2.8790	0.3618	−5.2711 ***	−5.8034 ***
lnGDP2	3.0644	0.3481	−5.2458 ***	−5.7710 ***
lnGC	−1.4087 *	−1.3933 *	−10.1395 ***	−10.6683 ***
lnU	1.9344	1.1152	1.5460	−2.5687 **
lnTO	−1.9039 **	−2.1075 **	−9.4524 ***	−9.9113 ***
CIPS test	Intercept	Intercept and trend	Intercept	Intercept and trend
lnCO2	−2.861 ***	−2.698	−5.475 ***	−5.989 ***
lnGDP	−2.444 ***	−2.536	−3.745 ***	−4.198 ***
lnGDP2	−2.492 ***	−2.512	−3.690 ***	−4.114 ***
lnGC	−2.845 ***	−2.694 *	−4.955 ***	−5.383 ***
lnU	−2.034	−2.372	−1.852	−2.686 **
lnTO	−2.110	−2.395	−4.618 ***	−4.797 ***

Note: *, **, *** indicate that statistics are significant at 10%, 5%, and 1% level of significance, respectively. H₀: panels contain unit roots.

, the LLC, the Fisher–ADF, the IPS, and the CIPS tests show that the series are integrated of order one I(1) when a time trend is included. Therefore, variables become stationary after applying first differences. These results provide evidence for a possible cointegration relationship between the variables.

4.4. Cointegration Tests

After concluding that the series are first-order integrated (in symbols, I(1)), in this section, three panel cointegration tests, namely the Kao [51], the Pedroni [52], and the Westerlund [53] tests, are applied to control for the validity of the long-run equilibrium relationship among the variables.

By applying two variance ratio statistics, namely panel variance ratio statistic and group mean variance ratio statistic, Westerlund takes cross-sectional dependence into account (see also [7,8,21,30]) as follows:

$$V{R}_G\equiv {\displaystyle \sum }_{i=1}^N{\displaystyle \sum }_{t=1}^T{\widehat{E}}_{it}^2{\widehat{R}}_i^{-1}$$

(7)

$$V{R}_P\equiv {\displaystyle \sum }_{i=1}^N{\displaystyle \sum }_{t=1}^T{\widehat{E}}_{it}^2{\left({\displaystyle \sum }_{i=1}^N{\widehat{R}}_i\right)}^{-1}$$

(8)

Here, ${\widehat{E}}_{it}\equiv {\displaystyle \sum }_{j=1}^t{\widehat{e}}_{ij}$ and ${\widehat{R}}_i\equiv {\displaystyle \sum }_{t=1}^T{\widehat{e}}_{it}^2$ and ${\widehat{e}}_{it}^2$ are the model residuals.

As specified in

Table 6. Panel cointegration test results.

Test	Value
Kao
Modified Dickey–Fuller	−4.3886 ***
Dickey–Fuller	−5.6569 ***
Augmented Dickey–Fuller	−3.9340 ***
Unadjusted modified Dickey–Fuller	−6.2985 ***
Unadjusted Dickey–Fuller	−6.2220 ***
Pedroni
Modified Phillips–Perron	1.3345 *
Phillips–Perron	−2.8852 ***
Augmented Dickey–Fuller	−2.4163 **
Westerlund
1 Variance ratio	−1.8754 **
2 Variance ratio	−3.2259 ***

Note: *, **, *** indicate that statistics are significant at 10%, 5%, and 1% level of significance, respectively. H₀: no cointegration. ¹ Include one lag on lCO₂. ² ΔCO₂ is considered.

, the results reject the null hypothesis of no cointegration, which reveals that the variables in the panel of 12 gas-producing countries have a long-run association over the 1990 to 2020 period. In other words, even though the variables may contain stochastic trends (i.e., be nonstationary), they nonetheless tend to move closely together over time and the difference between them is stationary [25].

4.5. Estimate the Long-Run Parameters

In order to test for validity of an EKC and calculate Equation (1), this study employs panel cointegration techniques, such as panel fully modified ordinary least squares (FMOLS), dynamic ordinary least squares (DOLS), common correlated effects mean group (CCEMG), and augmented mean group (AMG), for the full panel and individual countries.

The FMOLS estimator is a non-parametric approach and does not consider cross-sectional dependence, while the DOLS estimator is a parametric approach where lagged first-differenced terms are explicitly estimated. However, both approaches incorporate corrections for endogeneity and serial correlation and have been widely used in the EKC literature to estimate long-run coefficients [7,8,17,25,54].

The CCEMG estimator holds under slope homogeneity, particularly even in the presence of cross-sectional dependence, as well as endogeneity and serial correlation. The AMG estimator is also robust to parameter heterogeneity and cross-sectional dependence. It employs a two-stage method to estimate the unobserved common dynamic effect and allows for cross-sectional dependence by including the common dynamic effect parameter [7,8,10,25] as follows:

$$\mathrm{Stage\; I} \Delta y_{it}=\alpha +{\beta }_i\Delta x_{it}+{\phi }_i{f}_t+{{\displaystyle \sum }}_{t=2}^T{d}_t\Delta D_t+{\epsilon }_{it}$$

(9)

$$\mathrm{Stage\; II} AMG=N^{-1}{{\displaystyle \sum }}_{i=1}^N{\hat{\beta }}_i$$

(10)

Here, $\Delta$ is the first difference operator, $y_{it}$ and $x_{it}$ represent observables, $a_i$ represents the intercept, ${\hat{\beta }}_i$ is country-specific coefficient, $f_t$ is unobserved common factor with heterogeneous factor loadings, ${\phi }_i$ and $d_t$ are coefficients of time dummies, and $\Delta D_t$ and ${\epsilon }_{it}$ are error terms. In Stage I, Equation (9) is estimated by OLS. In Stage 2, the estimates are averaged.

Table 7. Panel cointegration coefficients for 12 natural-gas-producing countries, 1990–2020.

FMOLS							DOLS
Country/Panel	lnGDP	lnGDP2	lnGC	lnU	lnTO	EKC	lnGDP	lnGDP2	lnGC	lnU	lnTO	EKC
Algeria	−1.2288 *** (−5.9797)	0.1704 *** (8.7982)	0.2211 ** (2.3972)	−0.3165 (−1.3744)	−0.4560 *** (−7.7259)	No	−15.393 ** (−3.5182)	1.4098 ** (3.5430)	4.9374 ** (3.3886)	−24.264 ** (−4.1054)	−2.0256 (−2.7508)	No
Azerbaijan	−1.2615 *** (−3.4049)	0.0671 ** (2.5410)	1.0003 *** (9.5615)	−0.4274 (−0.3257)	−0.0415 (−0.4075)	No	41.465 *** (10.294)	−2.6830 *** (−10.343)	1.0319 *** (10.822)	−51.410 *** (−7.7455)	−2.0599 *** (−10.393)	Yes
Egypt	−0.3622 (−0.6790)	0.0599 (1.6943)	0.1611 (2.7494)	1.5249 (0.6564)	0.0423 (0.8654)	No	21.293 *** (6.3334)	−1.2946 *** (−6.0513)	−0.3288 *** (−4.8434)	6.6936 ** (4.0557)	0.2155 ** (4.1834)	Yes
Iran	−0.6704 ** (−2.1706)	0.0816 *** (4.3929)	0.3051 * (1.9981)	0.5521 (0.8515)	−0.0211 (−0.4547)	No	−0.6020 (−1.6941)	0.0707 *** (3.2700)	0.3192 * (1.8277)	0.6528 (0.8695)	−0.0017 (−0.0315)	No
Malaysia	−0.3162 (−1.6903)	0.0249 (1.4895)	0.3802 *** (5.9629)	0.9330 *** (3.7757)	0.0723 (1.0705)	No	115.75 ** (3.0992)	−6.6645 ** (−3.0717)	1.4602 ** (2.9340)	−3.5419 (−0.4345)	−0.9107 (−1.0625)	Yes
Nigeria	−1.0181 *** (−6.6289)	0.0996 *** (5.2751)	0.2085 ** (2.1291)	−0.0523 (−0.3171)	0.0952 (2.3188)	No	−2.4275 *** −4.8910	0.2815 *** 4.9065	−0.0791 −0.4680	−1.1132 ** −2.5698	0.4273 ** 3.1881	No
Oman	−0.3397** (−2.6795)	−0.0040 (−0.2317)	0.7260 *** (10.941)	−1.1016 ** (−2.6799)	−0.0246 (−0.1265)	No	−0.6881 *** (−4.9179)	0.0475 ** (2.5027)	0.6106 *** (9.6642)	−0.7520 ** (−2.1586)	−0.1182 (−0.8884)	No
Qatar	−0.9110*** (−8.3222)	0.0518 *** (7.6562)	0.8254 *** (12.661)	2.8573*** (3.6693)	−0.1169 (−1.0005)	No	−48.725 ** (−2.8378)	2.2822 ** (2.8486)	1.2324 *** (9.1319)	5.2539 ** (2.9818)	−0.7066 ** (−2.8119)	No
Russia	−2.2301 *** (3.0560)	0.1112 *** (3.0361)	1.4610 *** (3.6776)	−7.2650 ** (−2.1191)	−0.0947 * (−1.7235)	No	7.8218 * (2.1907)	−0.4318 * (−2.1465)	0.3443 ** (3.1545)	−32.424 ** (−4.7310)	−0.0903 * (−2.3112)	Yes
Saudi Arabia	−0.4266 (−1.0652)	0.0273 (1.1866)	0.5075 (2.2187)	2.4701 ** (1.2033)	0.1844 (2.3497)	No	−134.88 *** (−3.4987)	6.8013 *** (3.4984)	−0.7370 * (−1.8678)	12.574 ** (3.6900)	0.6735 *** (5.1091)	No
Turkmenistan	−1.8434 *** (−3.2464)	0.1341 *** (3.2073)	0.7543 ** (7.2875)	−4.0372 ** (−2.2264)	−0.1797 *** (−3.4626)	No	5.2139 *** (3.8249)	−0.3124 *** (−3.5168)	0.5137 *** (8.7780)	3.6680 (1.0454)	−0.0598 * (−1.9807)	Yes
UAE	−1.0595 ** (−2.7761)	0.0709 *** (5.0749)	0.5177 * (1.7813)	−2.4160 ** (−2.4237)	0.2863 ** (2.4562)	No	7.2298 * (2.0479)	−0.3216 * (−1.9233)	0.1097 (0.8257)	−896.24 *** (−13.362)	0.2226 ** (2.7583)	Yes
Panel	−0.8244 *** (−8.0283)	0.0553 *** (7.9496)	0.5500 *** (11.919)	0.6150 *** (2.7656)	0.2314 *** (3.0264)	No	1.2793 ** (1.9678)	−0.0758 ** (−1.9115)	0.5720*** (10.476)	−2.0604 (−1.6648)	0.0655 * (1.7201)	Yes
CCEMG							AMG
Country/Panel	lnGDP	lnGDP2	lnGC	lnU	lnTO	EKC	lnGDP	lnGDP2	lnGC	lnU	lnTO	EKC
Algeria	−11.867 (7.4538)	0.77533 * (0.46404)	0.27069 ** (0.09984)	−3.2029 ** (1.3386)	−0.05258 (0.08540)	No	−4.7723 (6.9398)	0.354167 (0.43681)	0.32699 *** (0.09197)	−6.1451 *** (1.1084)	−0.03408 (0.08720)	No
Azerbaijan	1.5779 * (1.1611)	−0.08666 * (0.07217)	0.58224 *** (0.14506)	1.179 (2.2351)	0.09036 (0.08493)	Yes	1.8295 * (1.3394)	−0.11959 * (0.08684)	0.1879 * (0.11131)	6.084 *** (1.0645)	−0.04665 (0.06744)	Yes
Egypt	17.825 *** (4.6707)	−1.142 *** (0.29807)	−0.30615 * (0.07743)	−5.4118 * (3.0150)	0.13315 *** (0.04511)	Yes	0.01985 * (4.8554)	−0.02829 * (0.31241)	−0.15847 * (0.09565)	0.00456 (3.3778)	0.07530 (0.05247)	Yes
Iran	−15.218 (10.512)	0.95094 (0.66042)	0.21741 (0.22734)	7.888 *** (2.822)	0.04094 (0.06781)	No	−14.278 * (7.9859)	0.92122 (0.50195)	0.11887 (0.17451)	5.8451 *** (2.0571)	0.05257 (0.05308)	No
Malaysia	−1.4256 (6.1761)	0.11849 (0.34923)	0.26896 *** (0.08147)	1.5536 (1.4143)	0.10239 (0.19701)	No	3.0977 * (4.4858)	−0.14973 * (0.25632)	0.33804 *** (0.06912)	1.3683 (1.0732)	−0.03291 (0.11823)	Yes
Nigeria	−0.11140 (6.466)	0.06602 (0.43450)	0.15508 (0.1161)	−7.0383 ** (3.577)	0.08922 ** (0.04479)	No	−3.5440 (4.0513)	0.29778 (0.28314)	0.14090 (0.08919)	−8.7369 ** (3.6017)	0.08727 ** (0.04079)	No
Oman	−28.704 * (14.8379)	1.5138 ** (0.7698)	0.43629 *** (0.08034)	0.82887 (0.51818)	−0.05435 (0.12376)	No	−35.393 (26.550)	1.8563 (1.3832)	0.60413 *** (0.09085)	1.4387 * (0.82057)	0.15523 (0.21913)	No
Qatar	7.2400 (12.262)	−0.33891 (0.57281)	0.81448 *** (0.10300)	4.4507 (3.2361)	−0.2076 * (0.11161)	No	−12.776 (10.2767)	0.59990 (0.48071)	0.88759 *** (0.06580)	6.0296 ** (2.4841)	−0.2349 * (0.12212)	No
Russia	−3.084 (2.4501)	0.20159 (0.13777)	0.76187 *** (0.13946)	0.86321 (3.7644)	0.03424 (0.0241)	No	−2.6678 (1.8526)	0.15814 (0.10502)	0.49916 *** (0.12431)	14.0736 *** (1.4684)	0.01631 (0.01710)	No
Saudi Arabia	−119.938 ** (60.691)	6.047 ** (3.0649)	0.21178 (0.34282)	21.576 ** (9.2040)	0.36412 ** (0.15958)	No	−120.67 ** (60.606)	6.0912 ** (3.0593)	0.44535 (0.27792)	1.3027 (4.5649)	0.23492 ** (0.11221)	No
Turkmenistan	2.5294 * (2.553)	−0.18008 * (0.15502)	0.49317 *** (0.1140)	10.210 *** (3.1237)	0.05224 (0.03959)	Yes	5.7003 *** (1.0965)	−0.3650 *** (0.0719)	0.52460 *** (0.04696)	10.504 *** (3.0939)	0.03902 (0.03055)	Yes
UAE	−60.562 *** (14.971)	2.8956 *** (0.71250)	0.8187 *** (0.25212)	−117.6 *** (36.680)	0.33227 ** (0.15561)	No	−6.6305 (7.7767)	0.33175 (0.36864)	0.28101 (0.31964)	−25.039 * (13.674)	0.32465 (0.19975)	No
Panel	−3.1452 (4.6954)	0.27606 (0.30010)	0.4314 *** (0.08371)	2.2542 (2.3309)	0.04972 90.03755)	No	−4.3443 (2.7246)	0.26488 * (0.15916)	0.36269 *** (0.0670)	2.7435 (2.3925)	0.05191 (0.04219)	No

Note: *, **, *** indicate that statistics are significant at 10%, 5%, and 1% level of significance, respectively. The values in parentheses represent t-statistics. DOLS: Algeria: including fixed leads (2) and lag specification (1). Azerbaijan, Malaysia, Russia, and the panel: the quadratic model has been considered, including fixed lead (1) and lag specification (1). Egypt: the linear model has been considered, including fixed lead (1) and lag specification (1). Nigeria: including fixed lead (1) and lag specification (1). Oman, Qatar, Saudi Arabia: including lag specification (1). Turkmenistan: the linear model has been considered. UAE: the quadratic model has been considered.

presents the estimation results using FMOLS, DOLS, CCEMG, and AMG. Accordingly, the four mentioned estimation methodologies provide mixed results.

Among four full panel estimates of the long-run relationship, the only models that have statistically significant variables are the FMOLS and DOLS. With FMOLS, the estimated effect of income on CO₂ emissions is negative and statistically significant at a 1% level. Nonetheless, the estimated coefficient on the quadratic income term is positive, providing presence of the U-shaped EKC hypothesis. On the other hand, the DOLS model shows that the coefficient on the income and squared income term supports the evidence of the EKC hypothesis. However, in this model, no significant relationship is found between urbanization and CO₂ emissions.

The CCEMG and AMG models neither have significant variables, nor do they support the EKC hypothesis. Inevitably, we accept the fact that the evidence does not support the EKC hypothesis for the full panel of 12 major gas-producing countries. This finding of no evidence in the full panel is consistent with the results of Destek et al. [55], Liddle [56], and partially Erdogan et al. [3], Destek and Sarkodie [57], and Acaravci and Akalin [58].

As for the relationship between natural gas consumption and CO₂ emissions, in all the outcomes of individual estimators (except Egypt), natural gas consumption has a significantly positive effect on CO₂ emissions. Similarly, according to the full panel regression results, natural gas consumption is the singular variable among four models, which is statistically significant at the 1% level. That is, a 1% increase in natural gas consumption causes CO₂ emissions to increase by 0.36% to 0.57% in different models. Therefore, natural gas consumption will not decrease CO₂ emissions, but we can argue that, by increasing consumption of gas in the energy mix, the speed and slope of CO₂ emissions are slowing.

Regarding the relationship between income and CO₂ emissions, only in the FMOLS estimator is the coefficient of GDP significant and negative, while the coefficient of squared GDP is positive and significant, meaning that GDP does not contribute to reduction in CO₂ emissions in the full panel. Moreover, population growth has a significant positive coefficient only under FMOLS, implying that a 1% increase in population growth increases CO₂ emissions by around 0.61%. Finally, the effect of trade openness on CO₂ emissions in the FMOLS and DOLS estimators is significant and positive, while, in the CCEMG and AMG, it is insignificant. Therefore, in the full panel, trade openness plays a role in increasing CO₂ emissions.

As mentioned, one thing that can be instantly noticed is that the results depend dramatically on the method used. The country-specific results of the individual FMOLS estimator indicate that the EKC does not hold in the sample countries, while the individual DOLS estimator shows that the EKC holds in six of twelve countries, namely Azerbaijan, Egypt, Malaysia, Russia, Turkmenistan, and the UAE. Moreover, the CCEMG results specify that the EKC hypothesis is maintained in Azerbaijan, Egypt, and Turkmenistan, while the individual AMG estimator results propose that the EKC holds in Azerbaijan, Egypt, Malaysia, and Turkmenistan. Accordingly, we can conclude that the EKC hypothesis is valid in Azerbaijan, Egypt, Turkmenistan, and to some extent in Malaysia. This means that income levels have increased environmental degradation at the initial stages of economic development, but it declined after attaining a specific turning point in income level in these countries.

Concerning the relationship between natural gas consumption and CO₂ emissions, Egypt’s energy policy prioritizes natural gas usage as a substitute for oil products. The country is implementing reforms aimed at encouraging competition and the opening of its gas sector, and it has accelerated the bidding rounds for gas exploration and development, specifically in deep waters and western areas. In 2020, natural gas constituted more than 50% of the country’s energy mix. Moreover, 79% of the power generation mix is supplied by natural gas. Natural gas is also being encouraged in the transport sector by incentivizing vehicle conversions and building gas-refueling stations. Last but not least, Egypt has paid a great deal of attention to countering climate change in the past few years. The country is to host the UN climate change conference COP27 in 2022. The mentioned energy policies in Egypt support our modelling results. Accordingly, a 1% increase in natural gas consumption causes a decline in CO₂ emissions by 0.32%, 0.30%, and 0.15% in the DOLS, CCEMG, and AMG models, respectively.

4.6. Panel Causality Test

Additional to the reported cointegration tests, the direction of causality among CO₂ emissions, income, natural gas consumption, urbanization, and trade openness variables helps to develop specific energy policies to tackle CO₂ emissions. Therefore, the D–H panel causality test proposed by Dumitrescu and Hurlin [59] is employed to determine the direction of causality based on an average Wald statistic of Granger [60]. The D–H test with the linear regression model can be described as follows:

$$y_{i,t}={\alpha }_i+{\displaystyle \sum }_{k=1}^p{\gamma }_i^{\left(k\right)}{y}_{i,t-k}+{\displaystyle \sum }_{k=1}^p{\beta }_i^{\left(k\right)}{x}_{i,t-k}+{\epsilon }_{i,t}$$

(11)

Here, ${\alpha }_i$, ${\gamma }_i^{\left(k\right)}$, and ${\beta }_i^{\left(k\right)}$ represent the constant term, the autoregressive parameters, and the regression coefficient slopes, respectively, and they differ across groups. The null hypothesis of D–H states that no causal relationship exists for all individuals, whereas the alternative hypothesis claims that the causal relationship occurs at least once for the panel.

As indicated in

Table 8. Results of D–H panel causality test.

Null Hypothesis	Wald Statistics
LGDP = LCO2	6.7533 ***
LCO2 = LGDP	7.3271 ***
LGC ≠ LCO2	3.0876
LCO2 = LGC	3.5557 **
LUR ≠ LCO2	5.6659 ***
LCO2 = LUR	4.7675 ***
LTO = LCO2	5.2695 ***
LCO2 = LTO	3.6270 **
LGC ≠ LGDP	2.9962
LGDP = LGC	6.5912 ***
LUR = LGDP	7.3501 ***
LGDP = LUR	4.4033 ***
LTO ≠ LGDP	2.3198
LGDP = LTO	4.7253 ***
LUR = LGC	9.1723 ***
LGC ≠ LUR	3.1961
LTO ≠ LGC	2.4157
LGC ≠ LTO	3.1662
LTO = LUR	3.4838 *
LUR = LTO	7.0099 ***

Note: *, **, *** indicate that statistics are significant at 10%, 5%, and 1% level of significance, respectively. H₀ (≠): A does not homogeneously cause B.

and Figure 4, the panel causality test reveals a strong bidirectional causal link between CO₂ emissions and trade openness, CO₂ emissions and income, income and urbanization, as well as urbanization and trade openness for the panel of 12 major gas-producing countries. The results also indicate unidirectional panel causality running from CO₂ emissions toward urbanization and natural gas consumption, from income to natural gas consumption and trade openness, and from urbanization to natural gas consumption. These relationships support the argument that natural gas consumption may not be a limiting driver of CO₂ emissions in the panel of sample countries.

Furthermore, bidirectional causality between urbanization and income and urbanization and trade openness suggests that urbanization is an instrument of economic development and is associated with industrialization. Moreover, the bidirectional causal relationship between CO₂ emissions and income and CO₂ emissions and trade openness approves the hypothesis that not only does level of economic growth have an impact on CO₂ emissions but CO₂ emissions can also have an impact on economic growth and level of development. This implies that more economic activities are likely to be associated with CO₂ emissions.

5. Conclusions

Using robust methods, this study aims to provide consistent and unbiased answers to its main research questions regarding the validity of the EKC hypothesis for the sample of 12 major gas-producing countries during 1990–2020 and the role of natural gas consumption in mitigating CO₂ emissions in these countries.

Considering cross-sectional dependence, the slope homogeneity test, as well as the Granger causality framework, including stationarity tests, cointegration tests, and causality tests, are employed, enabling cross-sectional dependence. Moreover, the D–H panel causality test is used for causality analysis. The main findings of this study are as follows:

First, among four full panel estimates (i.e., panel FMOLS, DOLS, CCEMG, and AMG estimators), the only procedures that yield statistically significant variables are FMOLS and DOLS. However, in FMOLS, the estimated coefficient on the quadratic income term is positive, failing to support the U-shaped EKC hypothesis. The results of DOLS support the EKC hypothesis. No significant relationship is found between urbanization and CO₂ emissions. The CCEMG and AMG models neither have significant variables nor support the EKC hypothesis. Therefore, we conclude that the evidence does not favor the EKC hypothesis for the full panel.

Second, regarding country-specific results, four estimators provided mixed results. Accordingly, the FMOLS estimator shows that the EKC does not hold in sample countries, while the individual DOLS estimator shows that the EKC holds in six of twelve countries. The CCEMG and AMG estimators specify that the EKC hypothesis holds in three and four countries, respectively. Therefore, we conclude that the EKC hypothesis is valid in four countries, namely Azerbaijan, Egypt, Turkmenistan, and Malaysia. Accordingly, environmental degradation increases as income levels increase with economic development. Beyond a certain level of income per capita, however, the trend reverses. Therefore, economic growth leads to environmental improvement at high-income levels in the above-mentioned countries.

In this connection, Stern [61] proposes that the ambiguity surrounding the EKC is because the EKC theory ignores thermodynamic laws. According to the first law, energy cannot be created or destroyed, but it can be transferred. Economic activity requires utilization of resources, and use of resources implies production of waste. Therefore, transforming pollution from one form to another is not the same as mitigating it.

Furthermore, Panyotou [62] provides a few explanations for inversion of pollution patterns that may not have been observed yet in our sample countries. He argues that the turning point for pollution is the result of more affluent, prosperous, and progressive societies placing greater emphasis on environmental issues and thus putting into place institutional and non-institutional measures to effect change. Furthermore, in the early phase of a country’s industrialization, due to rudimentary technology, pollution increases. When industrialization achieves more advanced levels, pollution will stop increasing and start to take a U-turn. In addition, service industries will gain prominence, resulting in even more pollution reduction.

Our third conclusion is that natural gas consumption—neither in the full panel nor in the individual countries’ estimators (except in the case of Egypt)—has not decreased level of CO₂ emissions. However, as natural gas is the lowest-carbon hydrocarbon compared to other fossil fuels, substituting gas with oil and coal (e.g., in the power generation sector) would reduce speed and slope of CO₂ emissions, and, most probably, if we include other fossil fuel variables, such as oil and coal, we could see the positive effect of natural gas on reducing CO₂ emissions. Furthermore, the feature of gas as “easy-to-store energy” stands out as a type of the best option to support intermittent renewable generation output when the sun does not shine and the wind does not blow. Moreover, the existing natural gas infrastructure and abundant and affordable supply of gas resources make gas a driver of energy transition in scaling up of blue hydrogen.

6. Policy Implications

Adoption of suitable policies regarding energy consumption, economic development, and environmental considerations is crucial for governments and policymakers. Governments should implement energy policies to reduce energy intensity and increase energy efficiency in order to keep CO₂ emissions under control. The findings of this study also highlight important policy implications:

(i)

To invest in technologies to prevent emissions from the entire gas value-chain: considering the impact of natural gas production and exports and economic development on CO₂ emissions, policymakers should consider measures to not only develop their natural resources but also employ technologies to prevent release of CO₂ and methane emissions generated through conventional production facilities. Today, proven technology exists to reduce gas flaring, methane venting, and even to capture, use, or store CO₂ emissions (i.e., CCS/CCUS). Accordingly, the natural gas industry is capital-intensive, with long lead times and payback periods that require policy and regulatory stability. Sufficient investments through the entire gas value-chain, as well as in clean technologies, including CCS/CCUS capacity, are required to attain the UN Sustainable Development Goals and for the very battle of mitigating and adapting to climate change.

(ii)

To accelerate the reform of fossil fuel subsidies: the implicit subsidies typically occur in countries with relatively rich oil and gas reserves, where state-owned oil and gas companies can be mandated to sell refined products for the domestic market at lower prices than production costs (e.g., in the mentioned 12 sample countries). Fossil fuel subsidies that are offered for a long period are not only inefficient but also dangerous due to gradual increases in energy demand and consequent increases in CO₂ emissions. Furthermore, as energy cost is artificially lower due to the subsidies, this will drive continued dependence on fossil-based fuels. As a result, subsidies may allow governments to build or develop new fossil fuel infrastructure, extending use of fossil fuels and, as a result, delaying economic transition to cleaner energies. Further, low domestic fossil fuel prices have led to immense oil and gas demand growth that cannot continue if oil- and gas-exporting countries wish to continue exporting. Only very high price jumps can stop this development, but, politically, these price jumps are very costly for many governments [63]. Thus, it is recommended that countries gradually remove fossil fuel subsidies, introduce true energy pricing, and promote mechanisms and policies, such as tax changes, feed-in tariffs, carbon pricing, and renewable portfolio standards.