The Symmetric and Asymmetric Impacts of Energy Consumption and Economic Growth on Environmental Sustainability

Abstract

Climate change has emerged as a global challenge because of its threat to sustainable development goals. Fossil fuels and economic growth are responsible for pollution and, thus, for climate change. In this context, this study explored the environmental Kuznets curve hypothesis for the case of 17 MENA countries over the period of 1990–2020. It investigated the symmetric and asymmetric impact of energy consumption and economic growth on CO₂ emissions by controlling for population density, trade openness, and FDI inflows using panel linear and nonlinear ARDL models. The robustness of the results was checked using the fully modified OLS and dynamic OLS methods. Moreover, the Dumitrescu–Hurlin panel causality test was employed to determine the directions of causality between the variables. Overall, the empirical results of both panel linear and nonlinear ARDL models validate the environmental Kuznets curve hypothesis for the selected sample of MENA countries. Economic growth leads to environmental degradation only in the long run, whereas a rise in energy consumption leads to an increase in pollution in both the short and long run. These results are confirmed by the fully modified OLS and dynamic OLS methods. The findings of the Dumitrescu–Hurlin panel causality test also indicate the existence of bidirectional causality between energy consumption and CO₂ emissions and between economic growth and CO₂ emissions. Therefore, policy makers in the MENA region should invest in clean technologies and accelerate the transition to renewable energies such solar energy, wind power, and hydropower to align with sustainable development goals.

1. Introduction

A report on the limits of the growth, together with the 1973 and 1979 oil shocks, sounded the alarm about the effects of energy consumption on economic growth and the environment. Indeed, apart from economic aspects, environmental aspects have been neglected in the industrialization process of many countries. Therefore, the Rio Earth Summit (1992) cast opprobrium and came to the conclusion that the various economies wishing to sustain their development should no longer solely focus on economic aspects but must, more importantly, closely consider environmental protection. From traditional, classical, or neoclassical theories to contemporary theories, energy has had a prominent role in the productive process of the economy. Although it is well recognized that energy has a significant impact on economic growth, the fact remains that its impact on the environment should not be neglected. Forster [1] and Luptacik and Shubert [2] argued that we cannot explore the relationship between energy consumption and economic growth without integrating the environmental component. Meadows report in 1972 and the studies of Georgescu-Roegen [3], Hall et al. [4], and Kaufmann [5] are the first documents that mentioned environment in an economic analysis. The introduction of environmental issues into growth analysis gave rise to the concept of sustainable development.

The relationship between economic growth and the environment remains a subject of discussion among various researchers and policymakers. Economic theory suggests a positive relationship between economic growth and environmental degradation. As in early stages, increased economic activity increases the use of fossil fuels that are harmful to the environment in the absence of clean technologies. However, after this occurs to some extent, economic development then contributes to environmental sustainability via a change in economic structures. This famous relationship between economic growth and the environment is known as the environmental Kuznets curve (EKC) hypothesis. EKC was named after Kuznets [6], who suggested that income inequality first increases and then decreases as economic development progresses. It is famously known for its inverted U-shaped relationship. Later, Grossman and Krueger [7] advanced a similar type of association between economic growth and environment. Many controversial empirical works explored the relationship between energy consumption, economic growth, and the environment. However, if the empirical literature recognizes the impact of fossil fuel consumption on economic growth and the environment in the case of industrialized countries, this relationship is a paradox in the case of less developed and developing countries. The Middle East and North Africa (MENA) region is the part of the world that holds the greatest potential in terms of energy resources. In 2020, Middle Eastern countries held 48.3% of the world-proven oil reserves and 40.3% of the world-proven natural gas reserves altogether; of these, Saudi Arabia held 17.2% of the world-proven oil reserves, Iran held 9.1% of the world-proven oil reserves and 17.1% of the world-proven natural gas reserves, Iraq held 8.4% of the world-proven oil reserves, and Qatar held 13.1% of the world-proven natural gas reserves (Energy Institute Statistical Review of World Energy, 2023) (https://www.energyinst.org/statistical-review, accessed on 23 October 2023).

The issue of growth based on energy consumption and its potential effects on the environment is no longer solely a matter for countries with high levels of pollution but must also be the responsibility of countries that aspire to development. Nevertheless, MENA is a minor polluter; it has enormous energy resources whose use could allow it to develop. However, respecting the agreements and summits regarding the climate—in particular, the reduction of carbon emissions with the aim of protecting the environment—is of importance. The objective of this article is to analyze the symmetric and asymmetric influence of energy consumption and economic growth on CO₂ emissions in a panel of 17 MENA countries using a panel linear autoregressive distributed lag (ARDL) model for annual data from 1990 to 2020. Our results could help policymakers take the necessary precautions to limit the impact of energy consumption on environmental degradation in MENA countries. The novelty of this study lies firstly in its contribution and distinction from previous studies that did not address the asymmetric impact of energy use on CO₂ emissions. Moreover, to our knowledge, this is the first study to simultaneously analyze the symmetric and asymmetric impacts of changes in energy consumption on environmental sustainability in MENA countries. The impacts of positive and negative changes in energy consumption on CO₂ emissions were estimated using the panel nonlinear ARDL (NARDL) model developed by Shin et al. [8]. This model permitted us to distinguish between the impact of increases in energy consumption and decreases in energy consumption on CO₂ emissions. This caused our study to be one of the few works to use a more innovative econometric approach for the case of MENA countries. Secondly, our research is among the few works that considered a relatively long and very recent period (1990–2020) for a large group of MENA countries. Thirdly, the results of panel ARDL models were checked for their robustness.

This article is organized as follows. In Section 2, we present a review of the literature. Section 3 shows the data and methodology. In Section 4, we report the empirical results and present discussions. In Section 5, we check for the robustness of the results. Finally, we present the conclusions and policy implications in Section 6.

2. Literature Review

The EKC model has been used to study the relationship between environment, energy, and economic growth. Most studies using this model have found that pollution increases with economic growth until a turning point, where the relationship between pollution and growth reverses. Grossman and Krueger [7] originally developed the link between growth and the environment. According to the EKC hypothesis, economies, especially those in development, have a trajectory of CO₂ emissions and GDP in the form of an inverted “U”. At some point, economies experiencing GDP growth also experience an increase in environmental degradation until they reach a turning point where society seeks a healthy environment that reduces CO₂ emissions. Stern [9] criticized the EKC hypothesis and proposed alternative approaches such as decomposition of emissions. Additionally, Stern criticized the econometric methods used in empirical studies validating the EKC.

Since the 1990s, many studies have been conducted on the relationship between growth and the environment using cross-section, time series, and panel data, as well as different methodologies and environmental indicators. The findings of these studies are controversial, even for the same country or group of countries. Some studies support the EKC hypothesis, while others do not. For example, studies by [10,11,12], support the EKC hypothesis, while studies by [13,14] do not. Many other studies have found mixed results [15,16,17].

Among the first group of studies supporting the EKC, Haggar [10] investigated the existence of a long-term equilibrium relationship between greenhouse gas (GHG) emissions, energy consumption, and economic growth for the Canadian industrial sector using cointegration techniques during the period of 1990 to 2007. The results show that energy consumption has a significant impact on GHG emissions. Moreover, the EKC hypothesis was validated. Alshehry and Belloumi [18] analyzed the relationship between economic growth, aggregate and disaggregate fossil fuels consumption, and carbon dioxide emissions for Saudi Arabia using the Johansen cointegration technique and annual data from 1971 to 2012. They found the presence of a long-run relationship between the different variables. In addition, there is evidence of the presence of bidirectional causality between natural gas consumption and CO₂ emissions and unidirectional causality from economic growth to CO₂ emissions in both the short and long run. Alvarado et al. [11] investigated the energy-growth-environment nexus using panel data techniques in 151 countries during the period of 1980 to 2016. Their results validated the EKC hypothesis and showed that energy consumption has a positive and significant impact on environmental indicators. Alharthi et al. [12] studied the influence of real income, renewable and non-renewable energy consumption, and urbanization on CO₂ emissions in MENA countries using quantile techniques over the period 1990–2015. They found that increases in renewable energy consumption led to a reduction in CO₂ emissions, whereas high levels of non-renewable energy consumption led to high levels in CO₂ emissions. In addition, their results validated the EKC hypothesis for the group of MENA countries.

Among the second group of studies, Nasir et al. [13] analyzed the EKC hypothesis for the case of Australia by studying the influence of economic growth and energy consumption on CO₂ emissions during the period of 1980 to 2014. They included different control variables such as industrialization, trade openness, and financial development. Their main finding reported that the EKC hypothesis is not validated. They explained this result via the positive influence of financial development, energy consumption, and trade openness on CO₂ emissions in the long term. Similarly, Kongkuah et al. [14] investigated the dynamic causal relationship between energy consumption, economic growth, and CO₂ emissions in China by including urbanization and international trade as control variables. Their main results indicate that the EKC hypothesis is not validated, whereas both economic growth and energy consumption have positive and significant effects on China’s CO₂ emissions in the long run.

In the case of the third group of studies, Ahmad et al. [15] explored the validity of the EKC in Croatia for the period 1992:1 to 2011:1 by estimating an ARDL model. The results demonstrated the existence of the EKC in the long term but its non-existence in the short term. In the same line, Hove and Tursoy [16] studied the EKC for a panel of 24 emerging countries using the generalized method of moments (GMM) for a panel data model over the period 2000–2017. They found that real GDP has a negative and significant effect on CO₂ emissions but a positive impact on nitrous oxide emissions. In addition, the square of real GDP increases CO₂ emissions but decreases nitrous oxide emissions. Not later, Majeed and Mazhar [17] explored the EKC hypothesis for the case of 76 high-, middle-, and low-income countries during the period of 1961 to 2018. Their findings are mixed and they depend on the data and methods used and countries investigated.

Recent studies have also investigated the asymmetric impact of energy consumption on environmental indicators. For example, Azam et al. [19] explored the non-linear effects of disaggregate energy (natural gas, nuclear energy, and renewable energy) on economic growth and CO₂ emissions in a selected sample of leading polluter countries by estimating panel fixed and random effects models during the period of 2000 to 2016. Their main finding indicates that renewable energy consumption reduces CO₂ emissions in the selected sample of countries. Not later, Majeed et al. [20] studied the asymmetric impacts of disaggregate energy consumption and economic growth on environmental quality in Pakistan using a NARDL model over the period 1971–2014. They found that positive changes in oil and gas consumption have positive effects on the environment, and negative changes have negative effects. Liu et al. [21] studied the asymmetric impacts of economic development and energy consumption on CO₂ emissions in China using different time series techniques over the period 1990–2020. The results of the Granger causality test indicate that energy consumption causes economic development and dioxide carbon emissions. In addition, the findings of impulse response functions show that energy consumption and agroforestry development affect CO₂ emissions in China.

In the case of MENA countries, some studies have investigated the EKC hypothesis. Kahia et al. [22] investigated the influence of renewable energy consumption and economic growth on CO₂ emissions in a selected group of MENA countries using the panel vector autoregressive model during the period of 1980 to 2012. Their results report that economic growth leads to pollution, whereas an increase in renewable energy consumption leads to decreases in CO₂ emissions. Dkhili [23] analyzed the relationship between CO₂ emissions, economic growth, and renewable energy by controlling for international trade and FDI for a selected group of MENA countries using panel data techniques over the period 1990–2018. The findings show a long-term decline between renewable energy and CO₂ emissions. More recently, Alkasasbeh et al. [24] studied the effects of energy consumption and economic growth on CO₂ emissions in five MENA countries using panel cointegration techniques and the Dumitrescu and Hurlin panel causality test over the period 1980 to 2020. The results show the presence of a long-run relationship between the three variables in the presence of structural breaks. Moreover, economic growth leads to environmental degradation in the investigated countries.

The above literature summary provides a good overview of information from ongoing discussions on how energy consumption and economic growth affect environmental indicators. However, it is clear from the literature review that these studies have varying results. Overall, the results of empirical studies on the EKC hypothesis are controversial and vary depending on the country, region, study period, type of data, and econometric techniques used. This study aims to fill the gap in the literature by analyzing the symmetric and asymmetric effect of energy consumption on environmental sustainability in MENA countries and by including a large sample of MENA countries in the study.

3. Data and Methods

3.1. Presentation of Data

This analysis focuses on the impact of energy consumption per capita and economic growth (real GDP per capita) on environmental quality (measured by CO₂ emissions per capita) for 17 MENA countries (Algeria, Egypt, Lebanon, Morocco, Saudi Arabia, United Arab of Emirates, Bahrain, Iran, Jordan, Libya, Oman, Syria, Iraq, Kuwait, Qatar, Tunisia, and Yemen) from 1990 to 2020. CO₂ emissions serve as a proxy for environmental sustainability. It is important to note that while there are 21 countries in the MENA region, we only have complete data for 17 countries for the period under investigation. In our analysis of the impact of energy consumption and economic growth on carbon dioxide emissions, we will consider the variables of CO₂ emissions per capita (CO2C) in metric tons per capita, energy consumption per capita (ECPC) in kg of oil equivalent per capita, and real GDP per capita in constant 2015 USD (GDPPC). The control variables include population density (PD) in number of people per square km, FDI inflows as percentage of GDP (FDI), and trade openness (TO) as the sum of exports and imports to GDP. The data for variables are sources from the World Bank’s world development indicators [25]. All variables are transformed to their natural logarithms.

Table 1. Descriptive statistics.

	LCO2C	LECPC	LGDPPC	LPD	LFDI	LTO
Mean	1.703	7.616	8.874	4.002	4.625	4.260
Median	1.384	7.205	8.487	4.112	4.615	4.337
Maximum	3.864	9.972	11.204	7.553	4.894	5.347
Minimum	−1.175	5.236	6.588	0.878	4.550	−3.863
Std. dev	1.150	1.224	1.172	1.401	0.033	0.754
Observations	527	527	527	527	527	527

provides the descriptive statistics for the log variables. Figure 1 illustrates the trends in CO₂ emissions per capita, energy use per capita, and real GDP per capita for the entire sample from 1990 to 2020, showing that the fluctuations in the three time series are nearly identical.

3.2. Methods

Our methodology is based on estimating environmental Kuznets curve panel ARDL and NARDL models using annual data from 1990 to 2020 for CO₂ emissions, energy consumption per capita, real GDP per capita, population density, FDI inflows, and trade openness in 17 MENA countries. We followed several steps in our methodology. First, we examined the data for cross-sectional dependence to identify any heterogeneity [26]. Second, we tested the series for stationarity using a second-generation panel unit root test. Third, we investigated the existence of cointegration between the variables. Finally, we estimate the EKC panel ARDL and NARDL models. Technical terms are provided in Appendix A 

Table A1. Technical terms and their abbreviations.

Terms	Abbreviations
Environmental Kuznets curve	EKC
Autoregressive distributed lag	ARDL
Nonlinear autoregressive distributed lag	NARDL
Cross-sectional dependence	CSD
Lagrange multiplier	LM
Dickey–Fuller	DF
Augmented Dickey–Fuller	ADF
Cross-sectional augmented Dickey–Fuller	CADF
Cross-sectional augmented IPS	CIPS
Generalized method of moments	GMM
Pooled mean group	PMG
Mean group	MG
Dynamic fixed effect	DFE
Fully modified ordinary least squares	FMOLS
Dynamic ordinary least squares	DOLS

.

3.2.1. Tests of Cross-Sectional Dependence

When examining relationships in a panel data model, it is important to consider the issue of cross-sectional dependence (CSD). This means that a shock affecting one country can also impact other countries in the sample due to direct and indirect economic relations between countries. Testing for CSD is a crucial step in panel data analysis. To address this issue, the first two tests used were the Lagrange multiplier (LM) cross-sectional dependence test developed by Breusch and Pagan [27], followed by the Pesaran [28] scaled LM cross-section dependence test. However, these tests may produce biased results when the group average is zero and the individual average is different from zero. Pesaran et al. [29] addressed this bias by incorporating variance and mean into the test statistics at the cross-sectional level. As a result, ref. [29] developed the bias-corrected scaled LM test for cross-section dependence. The null hypothesis indicates that there is cross-section independence between the series, while the alternative hypothesis suggests cross-section dependence.

3.2.2. Panel Unit Root Tests under Cross-Section Dependence

Pesaran [26] developed a panel unit root test that considers the cross-sectional dependence in the data. This test expands on the standard Dickey–Fuller (DF), or augmented Dickey–Fuller (ADF) regressions by incorporating cross-sectional averages. As a result, new asymptotic results were derived for both the ADF statistics in individual cross-sections (CADF) and for their simple averages. Pesaran [26] also computed the cross-sectional augmented IPS (CIPS) statistic by averaging the individual CADF test statistics for the entire panel. The null hypothesis of the test indicates that each cross-section of the panel is nonstationary. The CIPS test follows an asymptotically standard distribution.

3.2.3. Panel Cointegration Tests

Unit root tests can be used to check the stationarity of a series. However, when dealing with empirical questions involving multivariate relationships, it is important to determine if a specific set of variables is cointegrated. In the context of time series, cointegration means that if a set of variables is individually integrated of order one, it is possible that certain linear combinations of these variables are stationary. In this case, the cointegration vector refers to the vector of the slope coefficients. The presence of panel cointegration can be confirmed using the Kao [30] panel cointegration test.

3.2.4. Panel ARDL Models

Once the cointegration relationship is determined, it should be estimated using the panel ARDL method. The panel ARDL specification has two major advantages: it allows for the joint estimation of short-term and long-term parameters, and it includes variables that can be integrated of different orders (0 or 1). Additionally, when endogeneity is a problem for some regressors, the panel ARDL model provides unbiased long-run approximations and test statistics. Thus, the panel ARDL model incorporates short-run adjustment to long-run equilibrium without losing long-run information. Furthermore, the problem of multicollinearity can be solved by selecting the appropriate lag order [20].

Our environmental sustainability model is based on the EKC hypothesis, which is a hypothetical relationship between various environmental indicators and GDP per capita. In the early stages of economic growth, pollution emissions increase and environmental quality decreases, but beyond a certain level of GDP per capita, the trend is reversed, so that at high income levels, economic growth leads to an improved environment. This implies that per capita CO₂ emissions are a U-shaped inverted function of per capita GDP. Subsequent empirical studies have considered the EKC model as a relationship between energy consumption, economic growth, and environmental indicators. Additionally, several control variables were introduced in the model [31]. In this study, our model linking real GDP per capita, energy consumption per capita, population density, FDI inflows, and trade openness to CO₂ emissions per capita has the following form:

$$CO2C=f \left(ECPC, GDPPC, GDPPC2, PD,FDI, TO\right)$$

(1)

When Equation (1) is transformed in its logarithmic form with a time series specification on panel data, we obtain the following equation:

$$\begin{gathered} {LCO2C}_{it}={\beta }_0+{\beta }_{1 }{LECPC}_{it}+{{{\beta }_2}_{ }{LGDPPC}_{it}+{\beta }_{3 }{LGDPPC2}_{it}+ \\ {\beta }_{4 }{LPD}_{it}{+{\beta }_{5 }}_{ }{LFDI}_{it}{+{\beta }_{6 }}_{ }{LTO}_{it}+{\mu }_i+\epsilon }_{it } \end{gathered}$$

(2)

where the index i (i = 1, …, 17) indicates the country i of the selected sample, and t (t = 1990, …, 2020) indicates the period. Environmental sustainability is proxied by CO₂ emissions per capita (LCO2C). The literature has shown that both real GDP per capita (LGDPPC) and energy use (LECPC) are the main factors influencing CO₂ emissions, with both variables expected to have a positive effect. If the EKC hypothesis is validated, the square of real GDP (LGDPPC2) is expected to have a negative sign. The control variables (PD, TO, and FD) are expected to have mixed signs [32]. The population growth rate is relatively high in major MENA countries, and it is anticipated that population density could negatively impact the environment in these countries. Rapid population growth rate leads to increased pressure on arable land, water, and energy, resulting in environmental degradation. Additionally, FDI inflows can have both negative and positive impacts on the environment. While most studies found a negative impact due to the transfer of polluting industries, some have concluded that FDI can have a positive impact if it brings green technologies and spillover effects for domestic industries [33]. Similarly, international trade openness can have both negative and positive effects on the environment. Trade openness can increase production and revenues, affecting carbon emissions through the scale and technique effects. It also has a positive impact through the transfer of clean technologies to developing countries [34].

If all the variables are stationary and not cointegrated, a dynamic panel data model can be estimated using first difference or system GMM estimators. However, if some variables are stationary at their levels and others at their first differences, a panel ARDL model based on EKC hypothesis is estimated.

The EKC panel ARDL model (p, q) to be estimated has the following form:

$$\begin{gathered} {LCO2C}_{it}={\alpha }_0+{\sum }_{j=1}^p{\alpha }_j{LCO2C}_{i, t-j} +{\sum }_{j=0}^q{\delta }_{1j}{LECPC}_{i, t-j} + \\ {{\sum }_{j=0}^q{\delta }_{2j}{LGDPPC}_{i, t-j}+{\sum }_{j=0}^q{\delta }_{3j}{LPD}_{i, t-j}+{\sum }_{j=0}^q{\delta }_{4j}{LFDI}_{i, t-j}+ \\ {\sum }_{j=0}^q{\delta }_{5j}{LTO}_{i, t-j}+{\sum }_{j=0}^q{\delta }_{6j}{LGDPPC2}_{i, t-j}+\mu }_i+{\epsilon }_{it } \end{gathered}$$

(3)

Equation (3) yields the EKC panel ARDL(p, q) error correction model:

$$\begin{gathered} {∆LCO2C}_{it}={\alpha }_0+{\eta }_i \left({LCO2C}_{i, t-1}-{\upsilon }_{1i}{LECPC}_{i,t}-{\upsilon }_{2i}{LGDPPC}_{i,t}- \\ {\upsilon }_{3i}{LPD}_{i,t}-{\upsilon }_{4i}{LFDI}_{i,t}-{\upsilon }_{5i}{LTO}_{i,t}{-\upsilon }_{2i}{LGDPPC2}_{i,t}\right)+ \\ {\sum }_{j=1}^{p-1}{\alpha }_{ij}{∆LCO2C}_{i, t-j} +{\sum }_{j=0}^{q-1}{\delta }_{1ij}{∆LECPC}_{i, t-j} + \\ {\sum }_{j=0}^{q-1}{\delta }_{2ij}{∆LGDPPC}_{i, t-j}+{\sum }_{j=0}^{q-1}{\delta }_{3ij}{∆LPD}_{i, t-j}+ \\ {\sum }_{j=0}^{q-1}{\delta }_{4ij}{∆LFDI}_{i, t-j}+{\sum }_{j=0}^{q-1}{\delta }_{5ij}{∆LTO}_{i, t-j}+ \\ {\sum }_{j=0}^{q-1}{\delta }_{6ij}{∆LGDPPC2}_{i, t-j}+{\mu }_i+{\epsilon }_{it } \end{gathered}$$

(4)

where ${\eta }_i$ is the coefficient of the speed of adjustment (${\eta }_i<0); {\upsilon }_{1i}, {\upsilon }_{2i}, {\upsilon }_{3i}{, \upsilon }_{4i}{ and \upsilon }_{5i}$ represent the long-run coefficients of different independent variables (LGDPPC, LECPC, LPD, LFDI, LTO, and LGDPPC2) on CO₂ emissions. The variable LGDPPC2 represents the square of LGDPPC. The variable, ECT = $\left(\left({LCO2C}_{i, t-1}-{\upsilon }_{1i}{LECPC}_{i,t}-{\upsilon }_{2i}{LGDPPC}_{i,t}-{\upsilon }_{3i}{LPD}_{i,t}-{\upsilon }_{4i}{LFDI}_{i,t}-{\upsilon }_{5i}{LTO}_{i,t}-{\upsilon }_{6i}{LGDPPC2}_{i,t}\right)\right),$ is the error correction term; ${\delta }_{1ij},$ ${\delta }_{2ij}{, \delta }_{3ij}, {\delta }_{4ij},$ and ${\delta }_{5ij}$ are the coefficients of short-run dynamics; p is the optimal lag for the dependent variable; q are the optimal lag orders for the regressors; ${\mu }_i$ is the country specific effects; and ${\epsilon }_{it }$ are the error terms. In the case of asymmetric impact of energy consumption, the variable ECPC is replaced by its positive (@CUMDP(LECPC)) and negative changes (@CUMDN(LECPC)).

The panel ARDL models can be estimated using the pooled mean group (PMG), mean group (MG), and dynamic fixed effect (DFE) estimators, which allow for heterogeneity in the dynamics of adjustment of variables to the long-term relationship. The PMG estimator is advantageous for dynamic panels with more temporal observations than individuals [35]. It assumes that the model constant, short-term coefficients, and error variances may differ between individuals, while the long-term coefficients are forced to be identical for all countries. This makes the PMG estimator an intermediate procedure between the MG and DFE estimators.

If the assumption of similarity of long-term coefficients is accepted, the PMG estimator increases the accuracy of the estimates relative to the MG estimator. However, the hypothesis of homogeneity of long-term coefficients cannot be accepted a priori. The Hausman statistical test was performed to determine which of these estimators is most efficient in estimating the data. If the long-term coefficients are identical across countries, the PMG estimates will be consistent and effective while the MG estimates will be consistent but not effective. However, if long-term restrictions are misplaced, PMG estimates are not consistent while MG estimates will provide consistent estimates of the average of long-term coefficients across countries.

3.2.5. Robustness Test: Use of the FMOLS and DOLS Methods

To test the robustness of PMG estimates, the fully modified ordinary least squares (FMOLS) estimator and/or the dynamic ordinary least squares (DOLS) estimator can be used to estimate the long-term or equilibrium relationship if there is a cointegration relationship between the variables. According to Kao and Chiang [36], these two techniques lead to asymptotically distributed estimators towards a normal, zero mean and constant variance distribution.

3.2.6. Dumitrescu–Hurlin Panel Causality Test

Dumitrescu and Hurlin [37] developed a heterogeneous fixed-effect non-causality test based on Granger’s definition for panel data. They acknowledged that in many economic areas, it is highly probable that if a causal link exists for one country, it could also exist for other countries. In this context, causality can be more effectively tested in a panel context with (NT) observations. However, this test differs in some characteristics from other panel causality tests. It yields more effective results than other tests as it considers the issue of cross-sectional dependence. Additionally, the test can be utilized if the time dimension (T) is greater or less than the individual dimension N [37]. This test is generally based on the null hypothesis Ho (no causal relationship between variables) against an alternative hypothesis H1 (there is a causal relationship between variables for at least one of the individuals in the panel).

4. Results and Discussion

4.1. Results of the Cross-Sectional Dependence Tests

The findings from the four cross-sectional dependence tests are presented in

Table 2. Results of cross-section dependence tests.

Tests	Breusch-Pagan LM		Pesaran Scaled LM		Bias-Corrected Scaled LM		Pesaran CD
	Stat.	Prob.	Stat.	Prob.	Stat.	Prob.	Stat.	Prob.
LCO2C	1285.25	0.00	69.68	0.00	69.40	0.00	11.56	0.00
LGDPPC	1608.47	0.00	89.28	0.00	88.99	0.00	21.36	0.00
LPD	3900.28	0.00	228.24	0.00	227.95	0.00	62.37	0.00
LECPC	1387.13	0.00	75.86	0.00	75.57	0.00	17.67	0.00
LFDI	566.99	0.00	26.13	0.00	25.84	0.00	15.58	0.00
LTO	766.39	0.00	38.22	0.00	37.94	0.00	15.36	0.00
LGDPPC2	1611.21	0.00	89.44	0.00	89.16	0.00	21.37	0.00

Note: null hypothesis (H₀): absence of cross-section dependence. LGDPPC2 is the square of the variable LGDPPC.

, showing that we reject the null hypothesis of no cross-section dependence for all variables at a 1% significance level. This suggests that there is cross-section dependence within the panel. As a result, a shock in one MENA country has the potential to impact other countries in the region.

4.2. Results of Panel Unit Root Tests

Table 2 shows that the unit root tests for cross-sectional dependence indicate that each series has cross-section dependence. As a result, the Pesaran [26] CIPS (Z(t-bar)) test for unit roots was used. This test considers cross-sectional dependence and the results are presented in

Table 3. Results of Pesaran [26] CIPS panel unit root test.

Tests	Levels		First Difference		Order of Integration
	Stat.	Prob.	Stat.	Prob.
LCO2C	−1.176	>=0.10	−2.929	<0.01	I(1)
LGDPPC	−1.597	>=0.10	−3.808	<0.01	I(1)
LPD	−1.248	>=0.10	−2.566	<0.01	I(1)
LECPC	−1.367	>=0.10	−3.954	<0.01	I(1)
LFDI	−2.391	>=0.10	−5.184	<0.01	I(1)
LTO	−2.064	>=0.10	−3.54207	<0.01	I(1)
LGDPPC2	−1.582	>=0.10	−3.095	<0.01	I(1)

Notes: Null hypothesis (H₀): presence of unit root with cross-section dependence.

. The findings reveal that all variables are non-stationary at their levels but become stationary at their first differences.

4.3. Results of Panel Cointegration Test

The results of the panel Kao cointegration test are shown in

Table 4. Results of Kao cointegration test.

	t-Statistic	Prob.
ADF	−4.847 ***	0.00

Notes: Null hypothesis is no cointegration; level of significance: *** p-value < 0.01.

. These results indicate that the null hypothesis of no cointegration is not supported, suggesting that there is a cointegration relationship between the series and that they should move together in the long term.

4.4. Results of EKC Panel ARDL Model

4.4.1. Results of EKC Panel Symmetric ADRL Model

Prior to estimating the panel ARDL model, we conducted the Hausman test to select among the three estimators pooled mean group, mean group, and difference fixed effect. The results of the Hausman test are displayed in

Table 5. Results of PMG Hausman specification test.

Null Hypothesis: Estimator Is Statistically Similar to the PMG Estimator
Estimator	Stat.	DOF	p-Value
Mean group	8.550	7	0.286

, indicating that the assumption of long-term coefficient homogeneity cannot be rejected. Consequently, the pooled mean group estimates were deemed more consistent and efficient than the mean group estimates. Therefore, we proceeded to estimate our panel ARDL model using the pooled mean group method. The results of the panel ARDL model using the pooled mean group estimator are presented in

Table 6. Results of EKC panel ARDL model using PMG estimator.

Dependent Variable: D(LCO2C)
Selected Model: PMG(1,3,3,2,3,3,3)
Variable	Coefficient	Std. Error	t-Statistic	Prob.
	Long-Run (Pooled) Coefficients
LPD	−0.061 *	0.035	−1.723	0.085
LECPC	0.568 ***	0.033	17.106	0.000
LFDI	−0.936 ***	0.276	−3.387	0.000
LGDPPC	3.822 ***	0.259	14.728	0.000
LTO	0.081 ***	0.015	5.248	0.000
LGDPPC2	−0.182 ***	0.014	−12.372	0.000
@TREND	−0.006 ***	0.001	−6.175	0.000
	Short-run (Mean Group) Coefficients
COINTEQ	−0.377 ***	0.116	−3.249	0.001
D(LPD)	−0.296	2.771	−0.107	0.914
D(LPD(-1))	1.411	3.643	0.387	0.698
D(LPD(-2))	5.126	4.293	1.193	0.233
D(LECPC)	0.491 ***	0.098	4.978	0.000
D(LECPC(-1))	0.136	0.092	1.469	0.142
D(LECPC(-2))	0.025	0.080	0.319	0.749
D(LFDI)	0.709 **	0.323	2.189	0.029
D(LFDI(-1))	0.486	0.412	1.179	0.238
D(LGDPPC)	20.555	15.830	1.298	0.194
D(LGDPPC(-1))	−16.905	13.214	−1.279	0.201
D(LGDPPC(-2))	−11.178	12.473	−0.896	0.370
D(LTO)	0.006	0.034	0.173	0.862
D(LTO(-1))	0.039	0.040	0.975	0.329
D(LTO(-2))	0.007	0.044	0.176	0.860
D(LGDPPC2)	−1.178	0.897	−1.312	0.189
D(LGDPPC2(-1))	0.861	0.692	1.243	0.214
D(LGDPPC2(-2))	0.608	0.628	0.968	0.333
C	−6.774 ***	2.072	−3.269	0.001

Notes: The maximum lag order used the dependent variable and the regressors is equal to 3; the model selection method used is the Akaike information criterion (AIC); level of significance: *** p-value < 0.01, ** p-value < 0.05, * p-value < 0.10.

, revealing that all the independent variables significantly explain CO₂ emissions in the long run. As anticipated, energy consumption and economic growth have a positive and significant impact on CO₂ emissions at a 1% level. Specifically, a 1% increase in energy consumption leads to a 0.56% increase in CO₂ emissions in the long run, while a 1% rise in economic growth results in a 3.82% increase in CO₂ emissions. Additionally, the variable square of real GDP per capita exhibits a significant and negative coefficient in the long run, validating the EKC for the selected sample of MENA countries. Trade openness also has a positive and significant effect on CO₂ emissions at a 1% level of significance, with a 10% increase in trade openness leading to a 0.81% increase in CO₂ emissions in the long run. This suggests that international trade contributes to environmental degradation by increasing energy consumption in polluting industries in MENA countries. However, FDI inflows have a negative and significant influence on CO₂ emissions at a 1% level of significance, with a 10% increase in FDI inflows potentially resulting in a 9.3% decrease in CO₂ emissions in the long run. This indicates that foreign investments help reduce environmental degradation through the import of clean technologies and the transfer of less polluting industries to MENA countries. Surprisingly, population density has an unexpected effect on the environment, with a high population density leading to lower CO₂ emissions. Specifically, a 10% increase in population density results in a 0.61% decrease in CO₂ emissions in the long run. This can be explained by the fact that the least densely populated countries, such as Saudi Arabia and Algeria, are the most energy-consuming and polluting countries due to their large areas.

In the short run, only energy consumption and FDI inflows have a positive and significant impact on CO₂ emissions in the selected sample of MENA countries. This suggests that an increase in energy consumption and greater foreign investments in the economy lead to increased CO₂ emissions in the short term. For example, a 10% increase in energy consumption results in a 4.91% increase in carbon dioxide emissions, while a 10% increase in foreign direct investment inflows leads to a 7.09% increase in carbon dioxide emissions in the short run in the sample of MENA countries. The estimated coefficient of the error correction term in the short term is negative and statistically significant at a 1% level, indicating that the system is dynamically stable and converging towards a long-term equilibrium in all countries.

4.4.2. Results of ECK Panel Asymmetric ADRL Model

The MENA region is abundant in energy resources, leading to extensive energy usage in some countries. Changes in energy consumption, whether positive or negative, can have disproportionate effects on CO₂ emissions. To determine if energy consumption has a symmetric or asymmetric impact on CO₂ emissions in the selected sample of MENA countries, we conducted a symmetry test. The results of this test are presented in

Table 7. Results of symmetry test.

Null Hypothesis: Coefficient Is Symmetric
Variable	Statistic	Value	Probability
Long-Run
LEUC	F-statistic	69.702 ***	0.000
	Chi-square	69.702 ***	0.000

Notes: the maximum lag order used the dependent variable and the regressors are equal to 3; the model selection method used is the Akaike information criterion (AIC); level of significance: *** p-value < 0.01.

, showing that we reject the null hypothesis of a symmetric long run impact of energy consumption on CO₂ emissions in the selected sample of MENA countries. This indicates that we utilized a panel nonlinear ARDL model. The results of the panel NARDL model estimation using the PMG estimator are reported in

Table 8. Results of EKC panel NARDL model estimates using PMG method.

Dependent Variable: D(LCO2C)
Selected Model: PMG(1,3,2,3,1,3,3)
Variable	Coefficient	Std. Error	t-Statistic	Prob.
	Long-run (Pooled) Coefficients
LPD	−0.110 *	0.065	−1.673	0.094
LFDI	−0.482	0.343	−1.404	0.161
LGDPPC	9.241 ***	1.007	9.171	0.000
LTO	0.062 **	0.030	2.039	0.042
LGDPPC2	−0.448 ***	0.052	−8.571	0.000
@CUMDP(LECPC)	0.952 ***	0.125	7.570	0.000
@CUMDN(LECPC)	−0.670 ***	0.192	−3.478	0.000
@TREND	−0.050 ***	0.006	−7.731	0.000
	Short-run (Mean Group) Coefficients
COINTEQ	−0.155 ***	0.046	−3.331	0.000
D(LPD)	−3.731 ***	1.322	−2.821	0.005
D(LPD(-1))	2.254	3.547	0.635	0.525
D(LPD(-2))	3.747	3.990	0.939	0.348
D(LFDI)	0.309	0.281	1.101	0.271
D(LFDI(-1))	0.550 *	0.300	1.828	0.068
D(LGDPPC)	25.571	17.772	1.438	0.150
D(LGDPPC(-1))	−20.087 *	11.006	−1.824	0.068
D(LGDPPC(-2))	−14.681	11.625	−1.262	0.207
D(LTO)	0.020	0.046	0.441	0.659
D(LGDPPC2)	−1.453	1.027	−1.414	0.158
D(LGDPPC2(-1))	1.033 *	0.588	1.757	0.079
D(LGDPPC2(-2))	0.813	0.608	1.337	0.181
D(LECPC)	0.585 ***	0.060	9.717	0.000
D(LECPC(-1))	0.161 **	0.070	2.290	0.022
D(LECPC(-2))	0.072	0.074	0.967	0.333
C	−6.597 ***	1.982	−3.327	0.000

Notes: The maximum lag order used the dependent variable and the regressors are equal to 3; the model selection method used is the Akaike information criterion (AIC); level of significance: *** p-value < 0.01, ** p-value < 0.05, * p-value < 0.10.

. In the long run, the effects of population density, FDI inflows, real GDP per capita, trade openness, and the square of real GDP per capita are similar to those of the panel ARDL model. Regarding energy use, positive changes in energy consumption led to increased CO₂ emissions, while negative changes led to decreased CO₂ emissions in the selected sample of MENA countries. However, the impact of positive changes in energy use on CO₂ emissions is more significant than that of negative changes. A 10% increase in energy use results in a 9.52% increase in CO₂ emissions, while a 10% decrease leads to a 6.7% decrease in CO₂ emissions in the long run. Additionally, energy consumption has a positive and significant influence on CO₂ emissions in the short run.

4.5. Results of Dumitrescu–Hurlin Panel Causality Test

The presence of the cointegration relationship between the variables shows that there could be a causal relationship between the dependent variable and the independent variables that needs to be confirmed by the Granger causality, as improved by Dumitrescu and Hurlin [36]. To identify the direction of causality between the independent variables and the dependent variable, we used the Dumitrescu–Hurlin panel causality test. The results of this test are presented in

Table 9. Results of Dumitrescu–Hurlin panel causality test.

Null Hypothesis:	W-Stat.	Zbar-Stat.	Prob.
DLDENS does not homogeneously cause DLCO2C	2.520 ***	3.627	0.000
DLCO2C does not homogeneously cause DLDENS	4.861 ***	9.538	0.000
DLEUC does not homogeneously cause DLCO2C	2.893 ***	4.569	0.000
DLCO2C does not homogeneously cause DLEUC	1.904 **	2.072	0.038
DLFDI does not homogeneously cause DLCO2C	0.634	−1.133	0.256
DLCO2C does not homogeneously cause DLFDI	0.825	−0.650	0.515
DLGDPC does not homogeneously cause DLCO2C	7.260 ***	15.59	0.000
DLCO2C does not homogeneously cause DLGDPC	16.061 ***	37.81	0.000
DLTO does not homogeneously cause DLCO2C	2.099 ***	2.565	0.010
DLCO2C does not homogeneously cause DLTO	1.688	1.528	0.126

Notes: Level of significance: *** p-value < 0.01, ** p-value < 0.05.

, showing bidirectional causality between energy use and CO₂ emissions, economic growth and CO₂ emissions, and population density and CO₂ emissions. Additionally, there is unidirectional causality from trade openness to CO₂ emissions, but no causality between FDI and CO₂ emissions. These results support the previous findings of the PMG. The bidirectional causality between energy use and CO₂ emissions suggests that increased energy consumption can lead to higher CO₂ emissions and vice versa. This indicates that Middle East and North Africa countries rely on carbon energy for their economic growth, resulting in significant CO₂ emissions. Furthermore, the two-way link between economic growth and CO₂ emissions suggests bilateral effects from economic growth to CO₂ emissions and from CO₂ emissions to growth in MENA countries. This implies that environmental policies aimed at reducing CO₂ emissions can have a significant impact on production, while accelerated growth policies can significantly increase CO₂ emissions in the selected sample of MENA countries.

5. Robustness of Results

To assess the strength of the panel ARDL results, the FMOLS and DOLS methods were used to estimate the long-term effects of energy use and economic growth on CO₂ emissions in the chosen sample of MENA countries. The findings of both methods are presented in

Table 10. Results of FMOLS and DOLS methods.

Dependent Variable: LCO2C
Variable	Panel FMOLS				Panel DOLS
	Coeff.	Std. Error	t-Statistic	Prob.	Coeff.	Std. Error	t-Statistic	Prob.
LDENS	−0.146 ***	0.001	−84.38	0.000	−0.053	0.026	−1.983	0.049
LECPC	0.755 ***	0.009	82.05	0.000	0.719	0.053	13.40	0.000
LFDI	0.393 ***	0.015	25.78	0.000	0.512	0.299	1.710	0.089
LGDPPC	1.686 ***	0.004	418.4	0.000	0.740	0.363	2.039	0.043
LGDPPC2	−0.089 ***	0.003	−25.46	0.000	−0.028	0.020	−1.400	0.163
LTO	−0.053 ***	0.008	−6.122	0.000	−0.024	0.015	−1.602	0.111
R-squared	0.992				0.999
Adjusted R-squared	0.991				0.997

Notes: level of significance: *** p-value < 0.01.

. It is evident that both methods yield the same results as the PMG estimator. The EKC was confirmed by both methods. Furthermore, high levels of energy consumption result in high levels of CO₂ emissions in the selected sample of MENA countries. In conclusion, it can be inferred that an increase in energy consumption leads to an increase in environmental degradation in MENA countries. In terms of the impact of energy consumption on the environment, an increase in the use of non-renewable energy could have negative effects on the environment. Therefore, our results are reliable and can be interpreted without reservation.

6. Conclusions

We analyzed the symmetric and asymmetric impacts of energy consumption and economic growth on environmental quality in a selected sample of 17 MENA countries from 1990 to 2020 using the EKC panel ARDL and NARDL models. The results of the symmetric and asymmetric panel ARDL models indicate that energy consumption has a positive and significant effect on the environment. An increase in energy use leads to environmental degradation while a decrease in energy use leads to an improvement in the environment in MENA countries. Additionally, the results of the PMG estimator support the EKC hypothesis, showing that an increase in economic activity leads to an increase in pollution, but with a high level of real GDP per capita, pollution decreases. Overall, the results demonstrate that increasing energy use leads to an increase in CO₂ emissions in both the short and long run. The long-run results of PMG estimator were confirmed by FMOLS and DOLS methods.

Our findings indicate that economic growth leads to environmental degradation only in the long run, while energy consumption has both short-run and long-run impacts on the environment in MENA countries. This can be explained by the direct, instantaneous, and continuous impact of energy consumption on the environment, as opposed to the potential effects (scale, composition, and technique effects) of economic growth on the environment. According to the EKC hypothesis, the scale effect of economic growth induces pollution, while the composition and technique effects lead to a reduction in pollution in the long run [38]. In the long run, changes in production composition and improved technology lead to reducing environmental degradation, with the scale effect outweighing the composition and technique effects of economic growth in the MENA region. Therefore, MENA countries should invest in less-polluting technologies and transition their economic activities to be based on technology-intensive service activities. Additionally, countries in the Middle East and North Africa, which are rich in oil, should focus on reducing energy use and adopting energy efficiency measures.

Furthermore, the results of the causality tests indicate a bidirectional causality between energy consumption and CO₂ emissions, as well as between economic growth and CO₂ emissions. This suggests that increased energy consumption can increase CO₂ emissions, and vice versa, indicating a dependence on fossil fuels for economic growth in the MENA region, resulting in significant CO₂ emissions. Additionally, the two-way link between economic growth and CO₂ emissions indicates bilateral effects, with environmental policies designed to reduce CO₂ emissions having a significant impact on production, while accelerated growth policies can significantly increase CO₂ emissions in the MENA region.

In conclusion, these results suggest that MENA countries need to implement policies to mitigate the negative impacts of rising energy use and economic growth on the environment. Cooperation and communication between countries for better controlled development is essential, as well as the development of energy-efficient and environmentally friendly infrastructure systems and the transition to renewable energy sources.

One limitation of this study is the exclusion of data from some countries in the MENA region, and future work could focus on comparing major oil exporters and importers separately. Additionally, including other control variables in the models could provide further insights. Despite these limitations, our results remain significant for countries in the Middle East and North Africa region.