Revisiting the Environmental Kuznets Curve Hypothesis in the MENA Region: The Roles of International Tourist Arrivals, Energy Consumption and Trade Openness

Abstract

This paper employed advanced panel methods to examine the effects of international tourist arrivals on CO₂ emissions in the Middle East and North Africa (MENA) region over the period of 1995–2018. To analyze the predictive power of tourist arrivals for CO₂ emissions, the panel Granger non-causality test was employed. Employing the common correlated effects mean group estimator showed that tourist arrival reduces CO₂ emissions, while energy consumption and trade openness are the main contributors of CO₂ emissions. Results also showed that although first-generation estimators confirmed the Environmental Kuznets Curve (EKC) hypothesis, according to the CCE-MG estimator, an inverted U-shaped association between economic progress and CO₂ emissions does not exist. Furthermore, the null hypothesis of non-Granger causality from tourist arrivals to CO₂ emissions was rejected. The findings suggest taking a more sustainable approach to tourism development and energy conservation in the long run.

1. Introduction

With rapid international tourism development, concerns over growing CO₂ emissions and environmental pollutions are globally increasing, specifically in the developing world [1,2,3]. This situation will be worrying if demands for consuming primary fossil energies are rising for the purpose of tourism development [4]. Previous studies have investigated the relationship between tourism growth, energy consumption and CO₂ emissions in different parts of the world, confirming either positive or negative associations between these constructs [1,3,5,6,7]. For example, in a most recent study, Katircioglu et al. found that there is a positive and statistically significant association between tourism and the levels of CO₂ emissions in Northern Cyprus, revealing that growth in the tourism sector causes degradation in the environment [7]. Applying the Environmental Kuznets Curve (EKC) hypothesis for the northeast Asian countries, Zhang and Liu found that the development of tourism has led to environmental degradation and significant energy consumption [8].

These empirical studies, however, reveal inconsistent findings about the effects of international tourist arrivals on environmental degradation and CO₂ emissions, leaving room for further empirical studies and robust investigations [1,9]. Therefore, it is not yet clear whether the development of international tourism will lead to an increase in greenhouse gases and environmental degradation, especially in the developing world, where due to high operation costs, the possibility of using clean energy is limited.

While previous studies have applied different variables, such as CO₂ emissions, primary energy consumption, GDP per capita, real GDP and tourism developments [1,10,11,12], in this study, we considered trade openness, international tourist arrivals, energy consumption and CO₂ emissions, which to the knowledge of the authors, no study has tested the relationships between all these variables altogether in the MENA region. There are also inconsistent findings regarding the association between trade openness and CO₂ emissions. For example, several studies have shown a positive relationship between trade openness and CO₂ emissions [13,14], whereas other studies found negative or no associations between them [15,16]. Thus, the current study will shed some light on this issue by empirically testing this relationship using the second-generation econometrics approaches in the MENA region countries. Hence, the purpose of this study is to empirically test the relationship between international tourist arrivals and CO₂ emissions in the MENA region. Likewise, we test the relationship between trade openness, as an influential factor in developing international tourism and CO₂ emissions as the result of trade openness. We considered eight countries in the Middle East and North African (MENA) region (Qatar, Kuwait, Oman, Iran, Saudi Arabia, Algeria, Morocco and Egypt), for the period of 1995–2018. We selected the MENA region as countries in this region are characterized by an unstainable increase in energy demand and consumption, driven by economic development, industrialization, population growth [17], as well as an increasing demand for tourism [18]. In addition, Koçak and Şarkgüneşi found a significant increase in pollution-damaging industries in the MENA region which demand immediate actions in CO₂ emissions [19]. Past studies found a positive relationship between international tourist arrivals and environmental CO₂ emissions as international tourists use various transportation modes, including air transport, road, rail and water [20,21]. It is estimated that tourism contributes 5% of the total CO₂ emissions and up to 14% of all emissions when other greenhouse gases are considered [22]. However, in this study, we employ advanced panel methods that will address cross-sectional dependence to examine the effects of international tourist arrivals on CO₂ emissions in the MENA region.

2. Literature Review

2.1. Theoretical Setting

Various theories, such as the Pollution Haven Hypothesis (PHH) and the EKC Hypothesis, have been used in the literature to explain the link between CO₂ emissions and other variables, such as energy consumption, trade openness, GDP and tourism [23]. Similarly, much research has investigated the impact of environmental pollution on economic growth that mostly used the theoretical support of EKC [24,25,26,27,28,29]. The EKC is defined as a “hypothesized relationship between various indicators of environmental degradation and income per capita” [30] (p. 1419). It is argued that as economies expand, more energy will be sought and produced; thus, changes in energy demand and, ultimately, consumption, will lead to changes in the environmental quality levels [7,31,32]. In the 2000s, an increasing volume of environmental protection regulations appeared in the tourism policy agenda [33] and the link between tourism expansion and the EKC acquired prominence. While much research on tourism-induced EKC has used this theory as a proxy for CO₂ emissions from environmental deterioration [11,34], few studies have integrated two or three theories to investigate the link between international tourist arrivals and CO₂ emissions [1,7]. Since international tourism depends on many internal and external factors, such as safety and security [35], accessibility, policy stability, freedom of mobility, pull and push factors, etc. [36], due to its wide application and its relevancy, this study has applied EKC as a theoretical lens that associates economic growth with pollutant emissions.

Another critical hypothesis associated with environmental deterioration which has been well-studied is PHH. This hypothesis indicates that developed nations shift their polluting sectors to developing states and increase their pollution intensity [37]. A considerable number of studies have investigated PHH in various countries and regions [38]. For example, PHH claims that due to the less stringent environmental restrictions in developing countries, developed states have established their polluting industries in less developed nations. However, this development comes at the expense of environmental quality as a result of emissions from development activities [39,40,41]. Supporting the EKC and PHH, Koçak and Şarkgüneşi found that in the MENA region, pollution-damaging industries are on the rise, and urgent action is required to reduce CO₂ emissions [19]. Although proponents of PHH argue that EKC does not incorporate the influence of changes in trade patterns on a country’s environment, they claim that growing international commerce has turned developing countries into a haven for polluting enterprises. This is, however, against the notion of sustainable tourism, which promotes environmentally friendly consumption behavior of both tourists (demand) and stakeholders (suppliers) [42]. Differing from other business sectors (e.g., industrial, agriculture, mining, etc.), the tourism industry relies on both supply and demand sustainable behavior to conserve the environment [43]. Such sustainable behaviors include using clean energies [44], consuming environmentally friendly products and services and contributing to carbon reduction.

2.2. Energy Consumption, CO₂ Emission and International Tourist Arrivals

Over the past decades, there has been a growing interest in examining the relationship between the constructs of tourism development, energy consumption, CO₂ emissions and other economic variables in the literature [3,5,8,11,45]. This interest stems from the rapid growth of tourism from one side, and environmental effects of tourism development from the other side [46,47,48]. While governments and policy makers attempt to develop their tourism industry through increasing the number of international tourist arrivals and the revenue they generate, environmentalists are concerned about the adverse impacts of tourism on the local environment [9,42].

Critics believe that tourism development requires consuming primary energy resources such as oil, natural gas and coal, and as such, tourist destinations use large amounts of energy to supply services [9]. Likewise, tourist attractions, ski resorts, accommodations and transportations are consuming high amounts of energy due to the application of mechanized activities [49,50]. Moreover, the land-use policy change to cater to the large tourism demand, particularly in developing and less developed countries, has resulted in deforestation and environmental degradation, and consequently in increased CO₂ emissions [51,52,53,54].

Studies have revealed both positive and negative associations between tourism development and CO₂ emissions. These contradictory findings leave doubt whether tourism contributes to the decrease or increase in CO₂ emissions [9]. Rectifying this doubt could be mediated through undertaking more empirical studies focusing on individual or multi-country data and analyzing the existing research. In this study, the related works are categorized into two groups: the first group includes studies that investigated the relationship between tourism development and CO₂ emissions, focusing on single-country data, while the second group includes studies that attempted to analyze this relationship by considering a group of countries.

Amongst the studies in the first group, Katircioglu applied the ARDL approach and demonstrated that tourism development has a positive effect on both energy consumption and CO₂ emissions in Turkey, one of the most visited destinations in the world [55]. Similarly, Eyuboglu and Uzar examined the connections between CO₂ emissions, visitor arrivals, energy consumption and economic development by employing three co-integration tests in Turkey [56]. Their findings suggest that tourism, economic development and energy consumption positively affect CO₂ emissions in the short and long run. Furthermore, Katircioglu et al. tested the impact of tourism growth on environmental degradation in Cyprus and they found a positive and statistically significant impact of tourism on the levels of carbon emissions in the long term, showing that growth in the tourism sector leads to environmental degradation in this country [7]. Applying the wavelet transform method, Raza et al. investigated the effect of tourist arrivals on CO₂ emissions in the USA. Their findings indicate that international tourism flow has an increasing impact on CO₂ emissions [57].

In an endeavor to examine the nexus between business, financial development, tourist expenses and CO₂ emissions in Greece, Işik et al. found that the tourism sector increases CO₂ emissions [58]. Similarly, Sharif et al. [21] found that tourist arrival has a strong effect on CO₂ emissions in Pakistan. Unlike the above-mentioned findings, Zhang and Gao noted that tourism has a decreasing impact on CO₂ emissions in the East and West of China, while it has no significant impact in the central region [59]. Likewise, Naradda et al. discovered a reducing impact of tourism revenues on CO₂ emissions in Sri Lanka [60]. Recently, Peng et al. examined the effect of tourism market diversification on CO₂ emissions in Australia. They found that tourism market diversification in Australia has a positive impact on CO₂ emissions in the long run [61].

In the second group, other studies considered multi-country data across the world. Kocak et al. examined the impact of tourism developments on CO₂ emissions in top-ten tourist destinations, and their findings demonstrate that tourist arrival has an increasing impact, while tourism receipt has a reducing effect on CO₂ emissions [9]. Zhang and Liu claim that Northeast and Southeast Asia is one of the fastest growing regions in terms of CO₂ emissions and international tourism [8]. They investigated the linkage between tourism, CO₂ emissions, real GDP, non-renewable and renewable energy, for several Asian countries covering the period of 1995–2014. They discovered that a significant amount of CO₂ emission comes from non-renewable energy sources, while renewable energy can actually reduce CO₂ emissions when compared to non-renewable sources.

Gao et al. adapted a panel data method, and explored the relationship between CO₂ emissions, energy usage, economic and tourism development in eighteen Mediterranean countries for the period of 1995–2010 [6]. The results revealed that the unidirectional causality is running from tourism to CO₂ emissions across the regions. In another study for the top 50 tourist destination countries, the authors of [1] analyzed the role of tourism development on CO₂ emissions. They applied multiple tests on panel data over the period of 1995 to 2016 and concluded that CO₂ emissions levels significantly changed over the years through tourism development. Their results also indicate that tourism development had positive effects on the level of CO₂ emissions for 18 out of the 50 countries. Sghaier et al. inspected the effect of tourist arrivals on CO₂ emissions in Tunisia, Egypt and Morocco, and their results showed that there was no statistically significant association between tourist arrivals and CO₂ emissions in Morocco, whereas tourist arrival had a decreasing effect on CO₂ emissions in Egypt but an increasing effect in Tunisia [62]. Tang and Abosedra applied the panel data for two-dozen countries in the MENA region to investigate the impacts of tourism, energy usage and political instability on economic growth. Their findings indicate that energy usage and tourism significantly contribute to the economic growth [63]. Lee and Brahmasrene studied the effect of tourism on economic growth and CO₂ emissions in European Union countries [3]. Their findings indicate that a long-term equilibrium connection exists between these variables.

Dogan and Aslan employed the second-generation approaches to study the possible cross-sectional dependence among the EU countries [5]. Findings demonstrate that energy consumption increases the level of emissions, while real income and tourism mitigate CO₂ emissions. Similarly, Balli et al. employed advanced methodologies, such as the common correlated effects mean group approach and augmented mean group (AMG), to examine the relationship between tourism, CO₂ emissions and economic development in the Mediterranean countries [64]. They found that tourism increases the level of CO₂ emissions and has a statistically significant effect on economic development in the Mediterranean region.

Reviewing the studies in this group shows that most of them have used first-generation econometrics tools to identify relationships between variables, and there are very limited numbers of studies (e.g., [5,64,65]) that have employed the second-generation econometrics approaches. Since the first-generation methodologies have shortcomings of assuming cross-sectional independence, they may produce inaccurate results. To address this drawback, the employment of second-generation econometrics approaches which allow for cross-sectional dependence is suggested. International tourism is dependent on several factors, including long-haul air travel, use of hotels and resorts that consume significant amounts of energy and activities that damage the environment and require a lot of energy consumption [43]. Further investigation is required to understand the thorough effects of various variables on the relationship between international tourism flow and CO₂ emissions. The current study adds to the recent knowledge in the literature.

2.3. Trade Openness, CO₂ Emission and Tourist Arrivals

Trade openness or trade liberalization is defined as “the openness or the orientation of the country in perspective of international trade” [66] (p. 3). It helps the world economy to develop faster by enhancing the trade volume and income, both in developed and developing nations. Nevertheless, this development trend is followed and influenced by environmental impacts as it is believed that trade openness increases pollution and CO₂ emissions [67].

Generally, in terms of the effect of trade openness, two perspectives are considered in the literature. The first approach assumes that the effect of trade openness on environmental pollutions such as carbon emissions is unclear, and it can be divided into scale effect, technology effect and composition effect (see, for example, [67,68]). The second perspective relates to the PHH theory, which argues that due to the relatively poor environmental standards and regulations applied in developing and less developed countries, contaminating sectors shift their production to countries which thus become a “pollution haven” [37,38]. However, differing from other trade and business sectors, tourism deals with the freedom of human mobility dominated by country-specific or regional policies and regulations [69]. Within the context of our study, trade openness refers to the extent to which a country is involved in the international trade, because an increase in the degree of participation in the global trading system is likely to boost international tourism flows [70]. In addition, Chaisumpunsakul and Pholphirul note that trade openness improves international tourism flows [71]. This is because the countries in business relationships are inclined to travel across international borders for business deals, the creation and development of networks or business ventures and business transactions. Existing research suggests that trade openness is a key determinant of international tourism arrivals [71,72]. Nevertheless, other investigations confirm that trade liberalization with trading partners may not necessarily boost tourist arrivals from those countries but can be seen as a main catalyst for the growth and development of the tourism sector [73]. Considering the contradictory findings on the nexus between trade openness and international tourist arrivals and CO₂ emissions, this study sheds light on this issue, by examining if trade openness increases international tourism flow, as well as if this tourism growth increases CO₂ emissions in the MENA region. As the EKC and PHH theories inform, countries in the MENA region are prone to negative environmental impacts due to their trade openness policies, regional agreements on travel liberalization and the unsustainable development path they have taken towards “fast development” [70,74].

3. Model, Data and Methodology

3.1. Model Specification

To analyze the effect of tourist arrivals and other selected variables on CO₂ emissions, the following model proposed by several empirical studies in the literature [4,6,75] was estimated:

$$\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)+{\mathsf{\beta }}_2{(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2+{\mathsf{\beta }}_3\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)+{\mathsf{\beta }}_4\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)+{\mathsf{\beta }}_5\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(1)

where ${\mathrm{CO}}_{2\mathrm{it}}$, ${\mathrm{GDPP}}_{\mathrm{it}}$, ${\mathrm{ENU}}_{\mathrm{it}}$, ${\mathrm{OPEN}}_{\mathrm{it}}$ and ${\mathrm{TARIV}}_{\mathrm{it}}$ are carbon dioxide emissions (in metric tons) per capita, real GDP per capita (in constant 2010 US$), primary energy consumption (kg of oil equivalent) per capita, trade openness (export plus import as share of GDP) and the number of international tourist arrivals, respectively. All variables reported here are in logarithmic form. According to the EKC hypothesis, we expect that ${\mathsf{\beta }}_1>0$ and ${\mathsf{\beta }}_2<0$. We expect that ${\mathsf{\beta }}_3>0$, which indicates that the consumption of fossil fuel primary energies such as oil, coal and natural gas increases the pollution. In the literature, there are controversial findings about the pollution effects of trade openness. According to the trade theory of Helpman and Krugman, trade openness via promotion of physical output may worsen the CO₂ emissions, and in contrast, according to Yanikkaya (2003) and Ang (2009), trade openness via importing new technologies and increasing the productivity of factors, including energy, help to use the energy in a more sustainable way, and reduce its negative externalities. Additionally, empirical studies found contradictory findings about the effects of environmental degradation of international tourist arrivals; thus, in the model, i and t are the i^th country and the t^th year, and ${\mathsf{\epsilon }}_{\mathrm{it}}$ is a random error term that has a mathematical expectation of zero [76,77,78].

3.2. Data

The data were collected for eight countries in the MENA region, including Algeria, Egypt, the Islamic Republic of Iran, Kuwait, Morocco, Oman, Qatar and Saudi Arabia, over the period of 1995–2018, the longest time period for which the data were available for the maximum number of countries. Real GDP per capita and trade openness were collected from the World Development Indicators (2020). The data of primary energy consumption per capita were compiled from the BP Statistical Review of World Energy (2019) and data of international tourist arrivals were collected from The United Nations World Tourism Organization (2020) [79,80].

3.3. Methodology

The panel time-series models were more appropriate for the selected dataset since T (number of years) was greater than N (number of selected countries) [81,82]. Therefore, we selected the panel time-series models rather than the conventional methods including fixed and random effects. Considering the common factors across group members, this panel time-series estimator also considers heterogeneous slope coefficients and cross-sectional correlation across members (cross-section dependence). To this end, first, Im et al.’s panel unit root test was implemented to test the stationary properties of dependent and explanatory variables [83]. In addition, to control the cross-sectional dependence of the data, the null hypothesis of cross-sectional independence was tested by using the cross-sectional dependence test (CD) [84] and then the null hypothesis of the unit root was tested by using the panel unit root test [85].

Next, the long-run relationship between fossil CO₂ emissions ($\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$) and explanatory variables was tested by using a second-generation test, such as Westerlund’s panel co-integration, which considers the cross-sectional dependence [86].

Westerlund advanced an error correction-based panel co-integration test and developed four statistical tests. G_t, G_α, P_t and P_α, which are distributed normally with good performance in a small sample size [87]. The mean group test statistics, i.e., G_t and G_α, were calculated under the assumption of individual error correction parameters (between dimensions) and P_t and P_α were calculated based on the assumption of common error correction parameters (within dimensions) across selected countries. To test the co-integration null hypothesis, Westerlund [87] developed the following error correction model:

$$\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)={\mathsf{\gamma }}_i^{\prime }{\mathrm{d}}_{\mathrm{t}}+{\rho }_i(\ln \left({\mathrm{CO}}_{2\mathrm{it}-1}\right)-{\displaystyle \sum }_{j=1}^k{\mathsf{\tau }}_{\mathrm{ij}}{X}_{ijt-1})+{\displaystyle \sum }_{l=1}^{p_i}{\vartheta }_{il}\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}-\mathrm{l}}\right)+{\displaystyle \sum }_{r=-q_i}^{p_i}{\displaystyle \sum }_{j=1}^k{\theta }_{ir}\Delta X_{ijt-r}+{\mathsf{\omega }}_{\mathrm{it}}$$

(2)

where $d_t$, i, j, t and X are defined as Equation (2), $p_i$ is optimum lags of dependent and explanatory variables and $q_i$ is optimal leads of explanatory variables. $(\ln ({\mathrm{CO}}_{2\mathrm{it}-1})-\sum _{j=1}^k{\mathsf{\tau }}_{\mathrm{ij}}{X}_{ijt-1})$ is an error correction term and ${\rho }_i$ is its coefficient and indicates the speed of adjustment, and a negative sign indicates the convergence from the short to the long run. Westerlund [87] defined the two mean group test statistics G_t and G_α as follows:

$$G_t=N^{-1}{\sum }_{i=1}^N\frac{{\widehat{\rho }}_i}{{\sigma }_{{\widehat{\rho }}_i}}$$

(3)

and

$$G_{\alpha }=N^{-1}{\sum }_{i=1}^N\frac{T{\widehat{\rho }}_i}{{\widehat{\rho }}_i\left(1\right)}$$

(4)

where ${\widehat{\rho }}_i$ is the estimated OLS of ${\rho }_i$ in Equation (4), ${\sigma }_{{\widehat{\rho }}_i}$ is the standard error of ${\widehat{\rho }}_i$ and ${\widehat{\rho }}_i\left(1\right)$ is defined as the ratio of the long-run variance of ${\widehat{\mathsf{\omega }}}_{\mathrm{it}}$ to the long-run variance of $\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$. To compute P_t and P_α panel statistic tests, Westerlund [87] recommended a three-step procedure, as follows: In the first step, the $\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$ and $\ln \left({\mathrm{CO}}_{2\mathrm{it}-1}\right)$ are regressed into ${\mathrm{d}}_{\mathrm{t}}$, lags of $\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$, as well as the lags and leads of $\Delta X_{ijt}$, separately, and the projection errors $\Delta \widetilde{\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)}$ and $\widetilde{\ln \left({\mathrm{CO}}_{2\mathrm{it}-1}\right)}$ are computed as follows:

$$\Delta \widetilde{\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)}=\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)-{\widehat{\mathsf{\gamma }}}_i^{\prime }{\mathrm{d}}_{\mathrm{t}}-{\widehat{\rho }}_i{\displaystyle \sum }_{j=1}^k{\mathsf{\tau }}_{\mathrm{ij}}{X}_{ijt-1}+{\displaystyle \sum }_{l=1}^{p_i}{\widehat{\vartheta }}_{il}\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}-\mathrm{l}}\right)+{\displaystyle \sum }_{r=-q_i}^{p_i}{\displaystyle \sum }_{j=1}^k{\theta }_{ir}\Delta X_{ijt-r}+{\mathsf{\omega }}_{\mathrm{it}}$$

(5)

$$\widetilde{\ln \left({\mathrm{CO}}_{2\mathrm{it}-1}\right)}=\ln \left({\mathrm{CO}}_{2\mathrm{it}-1}\right)-{\widehat{\mathsf{\gamma }}}_i^{\prime }{\mathrm{d}}_{\mathrm{t}}-{\widehat{\rho }}_i{\displaystyle \sum }_{j=1}^k{\mathsf{\tau }}_{\mathrm{ij}}{X}_{ijt-1}+{\displaystyle \sum }_{l=1}^{p_i}{\widehat{\vartheta }}_{il}\Delta \ln \left({\mathrm{CO}}_{2\mathrm{it}-\mathrm{l}}\right)+{\displaystyle \sum }_{r=-q_i}^{p_i}{\displaystyle \sum }_{j=1}^k{\theta }_{ir}\Delta X_{ijt-r}+{\mathsf{\omega }}_{\mathrm{it}}$$

(6)

In the second step, $\rho$ (common error correction parameter) and its standard error (SE) ($\widehat{\rho }$) are computed as follows:

$$\widehat{\rho }={({\displaystyle \sum }_{i=1}^N{\displaystyle \sum }_{t=2}^T\tilde{{\left(\ln ({\mathrm{CO}}_{2\mathrm{it}-1})\right)}^2})}^{-1}{\displaystyle \sum }_{i=1}^N{\displaystyle \sum }_{t=2}^T\frac{1}{{\widehat{\rho }}_i\left(1\right)}\left(\widetilde{\ln \left({\mathrm{CO}}_{2\mathrm{it}-1}\right)}\right)\left(\Delta \widetilde{\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)}\right)$$

(7)

$$SE\left(\widehat{\rho }\right)={({(N^{-1}{\displaystyle \sum }_{i=1}^N{\widehat{S}}_i^2)}^{-1}{\displaystyle \sum }_{i=1}^N{\displaystyle \sum }_{t=2}^T{\left(\ln ({\mathrm{CO}}_{2\mathrm{it}-1})\right)}^2)}^{-\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}$$

(8)

where ${\widehat{S}}_i=\widehat{\pi }/{\widehat{\rho }}_i\left(1\right)$ and $\widehat{\pi }$ is the OLS estimated standard error of Equation (4).

In the third step, two of the panel test statistics, P_t and P_α, are computed as $P_t=\frac{\widehat{\rho }}{SE\left(\widehat{\rho }\right)}$ and $P_{\alpha }=T\widehat{\rho }$. Upon rejection of the no co-integration null hypothesis, using the above-mentioned panel co-integration tests, the long-run relationship between fossil CO₂ emissions ($\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$) and explanatory variables is estimated by using Equation (1). To estimate the long-run relationship between variables, the two group estimators were developed in the econometrics of panel data models. The first one disregards the cross-sectional dependence of the panel members, including pooled mean group (PMG), fully modified OLS (FMOLS) and dynamic OLS (DOLS) estimators (first-generation tests), and the second one considers the cross-sectional dependence, e.g., the common correlated effects mean group (CCE-MG) estimator (second-generation test). A few empirical studies have shifted from first-generation panel econometric techniques to second-generation tests as a result of globalization, to account for cross-sectional dependence and parameter heterogeneity [5,64,65].

The PMG estimator, developed by Pesaran et al., restricts the homogeneity of long-run slope coefficients across cross-sections (by taking the simple average of individual country coefficients), but allows for heterogeneity of coefficients in the short run [88]. The error correction form of Equation (1) is defined as follows:

$$\begin{array}{cc}\hfill \Delta ln\left({\mathrm{CO}}_{2\mathrm{i}.t}\right)={\rho }_i& (ln\left({\mathit{CO}}_{2i.t-1}\right)-{\mu }_{1.i}ln\left({\mathrm{GDPP}}_{\mathrm{it}}\right)-{\mu }_{2.i}{\left(ln\left({\mathrm{GDPP}}_{\mathrm{it}}\right)\right)}^2-{\mu }_{3.i}ln\left({\mathrm{ENU}}_{\mathrm{it}}\right)-{\mu }_{4.i}ln\left({\mathrm{OPEN}}_{\mathrm{it}}\right)\hfill \\ \hfill & \left.-{\mu }_{5.i}ln\left({\mathrm{TARIV}}_{\mathrm{it}}\right)\right)+{\pi }_{1.ij}\sum _{j=1}^{q_0-1}\Delta ln\left({\mathit{CO}}_{2i.t-1}\right)+{\tau }_{1.ij}\sum _{j=0}^{q_2-1}\Delta ln\left({\mathrm{GDPP}}_{\mathrm{it}}\right)\hfill \\ \hfill & +{\tau }_{2.ij}\sum _{j=0}^{q_1-1}\Delta {\left(ln\left({\mathrm{GDPP}}_{\mathrm{it}}\right)\right)}^2+{\tau }_{3.ij}\sum _{j=0}^{q_2}\Delta ln\left({\mathrm{ENU}}_{\mathrm{it}}\right)+{\tau }_{4.ij}\sum _{j=0}^{q_3-1}\Delta ln\left({\mathrm{OPEN}}_{\mathrm{it}}\right)\hfill \\ \hfill & +{\tau }_{5.ij}\sum _{j=0}^{q_4-1}\Delta ln\left({\mathrm{TARIV}}_{\mathrm{it}}\right)+{ϵ}_{it}\hfill \end{array}$$

(9)

where p, $q_0$, $q_1$, $q_2$, $q_3$ and $q_4$ are the optimal lag(s) of $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$ and explanatory variables. The optimal values of p, $q_0$, $q_1$, $q_2$, $q_3$ and $q_4$ are selected by employing BIC criteria. $\mu$ and $\tau$ are the vectors of coefficients, ${\rho }_i$ is the coefficient of error correction terms and ${\epsilon }_{it}$ are error terms.

According to the PMG methodology, the term: $(\ln (C{O}_{2i.t-1})-{\mu }_{1.i}\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)-{\mu }_{2.i}{(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2-{\mu }_{3.i}\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)-{\mu }_{4.i}\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)-{\mu }_{5.i}\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right))$, represents the long-run relationship between CO₂ emissions ($\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$) and other independent variables.

Pedroni developed the FMOLS estimator to test the long-run relationship between panel time-series variables [89,90]. Thus, in this study, to estimate the long-run relationship between variables by employing FMOLS, we specify the following regression model:

$$\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)={\mu }_i+{\acute{H}}_{i.t}\psi +{\nu }_{it}$$

$$H_{i.t}=H_{i.t}+{\zeta }_{i.t}$$

(10)

where $H$ is 5 × 1 vector of explanatory variables, ${\mu }_i$ is the country-specific intercept and the error terms ${\zeta }_{i.t}$ and ${\nu }_{i.t}$ are assumed to be I (0). The FMOLS estimation of $\psi$ is stated as follows:

$${\widehat{\psi }}_{FMOLS}={\left[{\displaystyle \sum }_{i=1}^N{\displaystyle \sum }_{t=1}^T(H_{i.t}-{\overline{H}}_{i.t})*{(H_{i.t}-{\overline{H}}_{i.t})}^{\prime }\right]}^{-1}*\left[{\displaystyle \sum }_{i=1}^N\left({\displaystyle \sum }_{t=1}^T(H_{i.t}-{\overline{H}}_{i.t})*\widehat{\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)}-T{\widehat{\mathsf{\delta }}}_i\right)\right]$$

(11)

However, the long-run relationship by employing Kao and Chiang’s [91] approach is estimated under the following framework:

$$\begin{array}{c}ln\left({\mathrm{CO}}_{2\mathrm{t}}\right)={\mu }_i+{\dot{x}}_{i.t}{\psi }_i+{\displaystyle \sum _{j=-p}^p}{\pi }_{j}ln\left({\mathrm{CO}}_{2\mathrm{it}-\mathrm{j}}\right)+{\displaystyle \sum _{j=-q_0}^{q_0}}{\rho }_{1.j}\mathit{Ln}{\left(GDPP\right)}_{i.t-j}+{\rho }_{2.j}{\displaystyle \sum _{j=-q_1}^{q_1}}{\left(\mathit{Ln}{\left(GDPP\right)}_{i.t-j}\right)}^2\hfill \\ +{\rho }_{3.j}{\displaystyle \sum _{j=-q_2}^{q_2}}\mathit{Ln}{\left(ENU\right)}_{i.t-j}+{\rho }_{4.j}{\displaystyle \sum _{j=-q_3}^{q_3}}\mathit{Ln}{\left(\mathit{OPEN}\right)}_{i.t-j}+{\rho }_{5.j}{\displaystyle \sum _{j=-q_4}^{q_4}}\mathit{Ln}{\left(TARIV\right)}_{i.t-j}+{\epsilon }_{it}\hfill \end{array}$$

(12)

where q and p are numbers of leads/lags of $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$ and explanatory variables, which are selected by the BIC criteria.

The main shortcomings of the first-generation tests, such as PMG, FMOLS and DOLS estimators, are that they do not allow for cross-sectional dependence. To address this flaw, the CCE mean group estimator was employed, which was developed in [85]. To provide more explanations about the CCE mean group estimator, the following regression model is specified:

$$\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)={\mu }_i^{\prime }{d}_t+{\psi }_i^{\prime } H_{i.t}+{\nu }_{it}$$

$${\nu }_{i.t}={\pi }_i^{\prime }{f}_t+{\zeta }_{i.t}$$

(13)

where H is a 5 × 1 vector of explanatory variables, $d_t$ includes a (n × 1) vector of observed common effects and $f_t$ includes a (m × 1) vector of unobserved common factors, which could be correlated with $H_{i.t}$ and $d_t$. The CCE-MG estimator of ${\psi }_i^{}$ (${\widehat{\psi }}_{CCE-MG})$ is expressed as follows:

$${\widehat{\psi }}_{CCE-MG}={\left(N\right)}^{-1}*\left[{\displaystyle \sum }_{i=1}^N\left({\widehat{\psi }}_i\right)\right]$$

(14)

$${\widehat{\psi }}_i={\left[(H_i^{\prime }\overline{R}{H}_i\right]}^{-1}{H}_i^{\prime }\overline{R}\left(\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)\right)$$

(15)

where $H_i$ and $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$ are 5 × 1 vectors of explanatory variables and the dependent variable of the i^th member of the panel, respectively. ${\widehat{\psi }}_i$ is the estimated value of ${\psi }_i^{\prime }$ of the i^th member of the panel.

In the next step, the predictive power of tourist arrivals for the CO₂ emissions per capita were tested. To this end, the Dumitrescu and Hurlin panel Granger non-causality (G-NC) test was employed [92]. The test extends a method of determining causal relationships between time series originally proposed in [93]. A number of advantages of the test have been identified as follows: In addition to being simple to implement, it has an asymptotic distribution that is normal, it increases the strength of Granger’s non-causality tests even for small samples with small “T” and “N” dimensions, it does not need special panel estimation techniques, and finally, it can be easily applied to panels that are unbalanced or have a different lag order “K” for each individual.

The panel G-NC test [92] is based on the following bivariate Granger causality regression model:

$$\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)={\alpha }_i+{\displaystyle \sum }_{k=1}^{p_i}{\rho }_{i,k}\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)+{\displaystyle \sum }_{j=1}^{p_i}{\gamma }_{i,j}\ln \left(TARI{V}_{i,t-j}\right)+{ϵ}_{1t-j}$$

(16)

where it is assumed that the optimal values of lags of dependent and explanatory variables (i.e., $p_i$) are identical but vary across members of the panel, and they are selected using AIC information criteria. The null hypothesis is ${\sum }_{j=1}^q{\widehat{\gamma }}_{i,j}=0$ for all members of the panel, and the alternative hypothesis is that the null hypothesis is rejected, at least for one member of the panel. To test the null hypothesis for each member of the panel, conventional Wald test statistics (namely $W_i$) were used, and to test the null hypothesis for the panel level, three test statistics, namely $\overline{W}$, $\overline{Z}$ and $\tilde{Z}$, were applied as follows:

$$\overline{W}=\frac{1}{N}{W}_i$$

(17)

$$\overline{Z}=\sqrt{\frac{N}{2K}}\left(\overline{W}-K\right)\begin{array}{c}d\\ \underset{T,N\to \infty }{\to }\end{array}N\left(0,1\right)$$

(18)

$$\tilde{Z}=\sqrt{\frac{N}{2K}\frac{T-3K-5}{T-2K-3}}\left(\frac{T-3K-3}{T-3K-1}\left(\overline{W}-K\right)\right)\begin{array}{c}d\\ \underset{T,N\to \infty }{\to }\end{array}N\left(0,1\right)$$

(19)

Testing the null hypothesis of Granger causality is based on the $\overline{Z}$ and $\tilde{Z}$ test statistics, and to decide about the rejection of the null hypothesis, we computed the critical values of the $\overline{Z}$ and $\tilde{Z}$ test statistics using a block bootstrapping procedure, which was developed in [91] and is a conventional way to solve the cross-sectional dependence problem.

4. Empirical Results

To estimate Equation (1), we performed triple pre-tests, including panel unit root, cross-section dependence and panel co-integration tests. First, we examined the null hypothesis of no cross-section dependence for $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$, $\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$, ${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$, $\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$, $\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$ and $\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$, using four tests: the LM test [94], the bias-corrected scaled LM test [95], the scaled LM test [82] and the cross-dependence (CD) test [84]. The results are shown in

Table 1. The results of cross-section dependence tests and first- and second-generation panel unit root tests.

Panel A: The Results of Cross-Section Dependence Tests
Variables	Breusch-Pagan LM	Pesaran Scaled LM	Bias-Corrected Scaled LM	Pesaran CD
$\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$	453.856 (0.000)	49.245 (0.000)	49.049 (0.000)	6.729 (0.000)
$\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$	389.679 (0.000)	41.681 (0.000)	41.486 (0.000)	13.278 (0.000)
${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$	389.844 (0.000)	41.701 (0.000)	41.505 (0.000)	13.251 (0.000)
$\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$	418.398 (0.000)	45.066 (0.000)	44.870 (0.000)	7.607 (0.000)
$\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$	212.320 (0.000)	20.779 (0.000)	20.584 (0.000)	11.226 (0.000)
$\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$	622.554 (0.000)	69.126 (0.000)	68.930 (0.000)	24.457 (0.000)
Panel B: The Results of the Panel Unit Root Test of [83]
Variables	Model with Constant		Model with Constant and Trend
	Level	First Difference	Level	First Difference
$\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$	−0.375 (0.354)	−6.865 (0.000)	0.895 (0.815)	−5.593 (0.000)
$\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$	0.106 (0.542)	−7.556 (0.000)	1.000 (0.841)	−6.636 (0.000)
${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$	0.240 (0.595)	−7.587 (0.000)	0.881 (0.811)	−6.565 (0.000)
$\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$	−1.845 (0.033)	−8.757 (0.000)	1.290 (0.902)	−8.895 (0.000)
$\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$	−2.368 (0.009)	−7.479 (0.000)	−0.945 (0.172)	−7.777 (0.000)
$\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$	−0.453 (0.325)	−7.174 (0.000)	1.051 (0.853)	−5.758 (0.000)
Panel C: The Results of the Panel Unit Root Test of [85]
Variables	Model with Constant		Model with Constant and Trend
	Level	First Difference	Level	First Difference
$\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$	−1.637 (0.051)	−3.702 (0.000)	0.388 (0.651)	−2.272 (0.012)
$\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$	0.886 (0.812)	−6.034 (0.000)	1.216 (0.888)	−5.015 (0.000)
${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$	1.084 (0.861)	−6.045 (0.000)	1.501 (0.933)	−5.046 (0.000)
$\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$	−1.769 (0.038)	−2.989 (0.001)	0.272 (0.607)	−2.141 (0.016)
$\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$	1.358 (0.913)	−2.431 (0.008)	0.413 (0.660)	−1.916 (0.028)
$\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$	−1.315 (0.094)	−3.932 (0.000)	0.399 (0.655)	−2.124 (0.017)

Note: $\ln ({\mathrm{CO}}_{2\mathrm{it}})$, $\ln ({\mathrm{GDPP}}_{\mathrm{it}})$, ${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$, $\ln ({\mathrm{ENU}}_{\mathrm{it}})$, $\ln ({\mathrm{OPEN}}_{\mathrm{it}})$ and $\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$ are natural logs of fossil CO₂ emissions (in metric tons) per capita, real GDP per capita (in constant 2010 US$), squared real GDP per capita, primary energy consumption (kg of oil equivalent) per capita, trade openness (export plus import as share of GDP) and the number of international tourist arrivals, respectively. The figures in the parenthesis are p-values of the tests.

(Panel A). As it can be seen, the null hypothesis of no cross-section dependence was rejected by all four tests for all variables at the 1% significance level. This may justify the employment of the second generation of panel data econometric techniques. To this end, in the panels B and C of Table 1, the results of the panel unit root test in [83] and the panel unit root test in [85] are depicted. We obtained the results of panel unit root tests for the two cases: the model with a constant trend and the model with a constant and linear trend, and for each of the models, we tested the null hypothesis of a unit root in the variables. The results of the panel unit root test of [81] in Table 1, panel B indicate that $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$, $\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$, ${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$ and $\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$ are integrated in order one I(1), with only one intercept and with a constant and linear trend. The null hypothesis of the unit root was rejected for two variables, $(\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$ and $\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$, in the model with an intercept at the 5% significance level, but the null hypothesis was not rejected in the model with the intercept and linear trend, and thus both were integrated in order one I(1).

The results of the panel unit root test of [85] are depicted in panel C of Table 1. The results indicate that $\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$, ${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$ and $\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$ are integrated in order one with only one intercept and an intercept and linear trend. According to the results of the unit root test for the model with only one intercept, the variables $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$, $\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$ and $\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$ were I(0) at 10%, 5% and 10%, respectively. In contrast, the results of the unit root test for the model with an intercept and linear trend indicate that all variables were I(1) at least at a 5% level of significance (Table 1).

According to the first and second generations of panel unit root tests, the selected variables were I(1) at the 5% significance level. Hence, in the next step, the presence of the long-run relationship between $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$ (as the dependent variable) and explanatory variables was tested. To this end, we tested the null hypothesis of no co-integration between the dependent variable ($\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right))$ and the independent variables (lnGDPPit, lnENUit, lnOPENit and lnTARIVit) using panel co-integration tests of [87] (as the second-generation panel co-integration tests), which allowed for testing the cross-section dependence. The results of panel co-integration in [84], which allowed for cross-section dependence (second generation), are presented in panel C of

Table 2. The results of first and second generations of panel co-integration tests.

Panel A: Residual Co-Integration Test [90]
Tests	Tests Based on the within Dimension of Panel		Tests	Tests Based on the between Dimension of Panel
	Statistic	p-Value		Statistic	p-Value
Panel v-Statistic	−0.401	0.656	Group $\rho$-Statistic	2.041	0.979
Panel $\rho$-Statistic	0.959	0.831	Group PP-Statistic	−2.206	0.014
Panel PP-Statistic	−2.121	0.017	Group ADF-Statistic	−1.650	0.049
Panel ADF-Statistic	−1.671	0.047
Panel B: Panel Co-Integration Test [91]
Test	Statistic	p-value
ADF	−8.490	0.000
Panel C: Panel Co-Integration Test [87]
Tests	Statistic		p-Value
Gt	−3.126		0.020
Ga	−9.586		0.020
Pt	−10.854		0.010
Pa	−9.526		0.060

Note: The null hypothesis of [87,90,91] is no co-integration. Model with constant only was selected for all three co-integration tests. The critical values were prepared by [87,90,91].

. The statistics of the four tests, Gt, Ga, Pt and Pa, were statistically significant at 5%, 5%, 1% and 10% significance levels, respectively. Thus, according to the panel co-integration test in [86], the null hypothesis of no co-integration among variables in Equation (1) was rejected, at least at the 10% significance level, and thus there was a long-run relationship between $\ln \left({\mathrm{CO}}_{2\mathrm{it}}\right)$ (as the dependent variable) and the explanatory variables. Hence, in the final step, we estimated the long-run relationships between variables using PMG, FMOLS, DOLS (first generation) and CCE-MG (second generation) estimators.

We have presented the results of the mentioned estimators in

Table 3. The results of PMG, FMOLS, DOLS and CCE-MG estimators.

Panel A: The Results of PMG Estimator			Panel C: The Results of DOLS Estimator
Variable	Coefficient	p-Value	Variable	Coefficient	p-Value
$\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$	−0.041	0.000	$\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$	−0.070	0.010
$\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$	4.227	0.000	$\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$	2.066	0.078
${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$	−0.244	0.000	${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$	−0.114	0.077
$\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$	0.854	0.000	$\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$	0.875	0.000
$\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$	0.096	0.000	$\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$	0.165	0.014
Panel B: The results of FMOLS estimator			Panel D: The results of CCE-MG estimator
Variable	Coefficient	p-value	Variable	Coefficient	p-value
$\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$	−0.062	0.000	$\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$	−0.031	0.073
$\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$	0.528	0.002	$\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$	−7.705	0.474
${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$	−0.023	0.018	${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$	0.480	0.399
$\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$	0.904	0.000	$\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$	0.564	0.000
$\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$	0.094	0.000	$\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$	0.138	0.001

Notes: $\ln ({\mathrm{CO}}_{2\mathrm{it}})$, $\ln ({\mathrm{GDPP}}_{\mathrm{it}})$, ${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$, $\ln ({\mathrm{ENU}}_{\mathrm{it}})$, $\ln ({\mathrm{OPEN}}_{\mathrm{it}})$ and $\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$ are natural logs of fossil CO₂ emissions (in metric tons) per capita, real GDP per capita (in constant 2010 US$), squared real GDP per capita, primary energy consumption (kg of oil equivalent) per capita, trade openness (export plus import as share of GDP) and the number of international tourist arrivals, respectively. PMG, FMOLS, DOLS and CCE-MG are the pooled mean group estimator, the fully modified OLS estimator, the dynamic OLS estimator and the common correlated effects mean group estimator.

. In panel A of Table 3, we present the results of the PMG estimator. The coefficient of $\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$ was negative (${\mathsf{\beta }}_5=-0.041)$ and statistically significant at the 1% significance level. The results indicate a that 10% increase in tourist arrivals decreased ${\mathrm{CO}}_2$ emissions by 0.4%. The coefficient of $\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$ was positive (${\mathsf{\beta }}_1=4.227$) and the coefficient of ${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$ was negative (${\mathsf{\beta }}_2=-0.244$), and both were statistically significant at the 1% level. According to the results of the PMG estimator, the EKC hypothesis is not rejected in the MENA region. In addition, the coefficient of $\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$ was positive (${\mathsf{\beta }}_3=0.854)$ and statistically significant at the 1% significance level. It means that a 10% increase in primary energy consumption increased ${\mathrm{CO}}_2$ emissions by 8.5%. The coefficient of $\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$ was positive (${\mathsf{\beta }}_4=0.096)$ and statistically significant at the 1% significance level. It implies that a 10% increase in trade openness increased ${\mathrm{CO}}_2$ emissions by 0.9%.

The results of FMOLS and DOLS are presented in panels B and C of Table 3, respectively. As can be seen, the results of the two mentioned estimators were similar to the results obtained by the PMG estimator. According to the results of FMOLS and DOLS, a 10% increase in tourist arrivals decreased ${\mathrm{CO}}_2$ emissions by 0.7% and 0.6%, respectively. The results of both estimators indicate the existence of EKC in the MENA region and the environmental degradation effect of increasing energy consumption and trade openness.

In panel D of Table 3, we present the results of the CCE-MG estimator, which allowed for cross-sectional dependence across panel members. The results of the CCE-MG estimator indicate that the coefficient of $\ln \left({\mathrm{TARIV}}_{\mathrm{it}}\right)$ was negative (${\mathsf{\beta }}_5=-0.031)$ and statistically significant at the 10% significance level. The results indicate that a 10% increase in tourist arrivals decreased ${\mathrm{CO}}_2$ emissions by 0.3%. All four estimators indicate that development of the tourism industry in the MENA region accelerated the reduction of ${\mathrm{CO}}_2$ emissions.

In contrast with the results of the three estimators, PMG, FMOLS and DOLS, the results of the CCE-MG estimator indicate that the hypothesis of the existence of EKC in the MENA region was rejected. The coefficients of both variables, $\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right)$ and ${(\ln \left({\mathrm{GDPP}}_{\mathrm{it}}\right))}^2$, were statistically insignificant. The coefficient of $\ln \left({\mathrm{ENU}}_{\mathrm{it}}\right)$ was positive (${\mathsf{\beta }}_3=0.561)$ and statistically significant at the 1% significance level. This implies that a 10% increase in fossil fuel energy consumption increased ${\mathrm{CO}}_2$ emissions by 5.6%. The pollution effect of energy consumption according to the results of the CCE-MG estimator was less than the three other estimators, PMG, FMOLS and DOLS. The coefficient of $\ln \left({\mathrm{OPEN}}_{\mathrm{it}}\right)$ was positive (${\mathsf{\beta }}_4=0.138)$ and significant at the 1% significance level. This implies that a 10% increase in trade openness accelerated ${\mathrm{CO}}_2$ emissions by around 1.4%. The results demonstrate that the environmental degradation effect of trade openness according to the results of the CCE-MG estimator was greater than the three other estimators, namely, PMG, FMOLS and DOLS.

To test the Granger causality from tourist arrivals to CO₂ emissions per capita, the methodology of [92] was employed. At first, we selected the optimal value of lags using BIC criteria and then the bootstrap procedure was used to compute the critical values of the test statistics. The computed values of $\overline{W}$, $\overline{Z}$ and $\tilde{Z}$ test statistics were 21.390, 13.328 and 1.923. The conventional p-values of $\overline{Z}$ and $\tilde{Z}$ test statistics were 0.000 and 0.048, respectively. Using the conventional p-values, the null hypothesis of non-Granger causality between tourist arrivals and CO₂ emissions per capita was rejected, at least at the 5% significance level. The p-values of $\overline{Z}$ and $\tilde{Z}$ test statistics, which were computed by using 100 bootstrap replications, were 0.01 and 0.01, respectively, and indicate that the null hypothesis of non-Granger causality was rejected at the 1% significance level.

5. Discussion

The aim of this study was to investigate the relationship between CO₂ emissions and tourist arrivals in the MENA region over the period of 1995–2018, by using the first (PMG, FMOLS, DOLS) and second generation (CCE-MG) of panel data econometrics approaches. The results, however, showed that employing all four estimators, development of the tourism industry reduced the level of ${\mathrm{CO}}_2$ emissions in the MENA region. Although the coefficients of tourist arrivals were very small in all the estimated models (−0.041, −0.062, −0.070 and −0.031 by employing PMG, FMOLS, DOLS and CCE-MG, respectively), they were negative and statistically significant. This finding is consistent with the investigations of Sghaier et al., showing that tourist arrival has a decreasing effect on CO₂ emissions in Egypt [62]. Several reasons can help interpret this finding. Some countries in the MENA region, such as Qatar, Kuwait, Oman and Algeria, have fewer international tourists than other countries. For example, the number of international arrivals to the above countries in 2019 was: Qatar (1.1 million), Oman, (2.5 million), Algeria (3.4 million) and Kuwait (1.9 million) [80]. Second, in line with the sustainability policies, some countries in the region, such as Qatar, Kuwait, United Arab Emirates and Oman, have started to reduce greenhouse gas emissions through the development of low-carbon technologies, such as carbon capture and storage (CCS) technologies [96]. These policies have changed the consumption of non-renewable energies, which could be a major source of CO₂ emissions. The third reason could be that due to the trade openness and liberalization of travel within the region (GCC countries), most international arrivals to the MENA region originate within MENA itself, but the proportions can vary across destinations according to their source markets [79]. Hence, intra-regional and domestic travel consume less energy compared to international long-haul travel [97,98]. Moreover, countries such as Qatar, Kuwait and Oman in the region have shut down or are trying to shut down their polluting industries and turn to relatively “clean industries”, such as tourism, in recent decades [99]. This recommends that while consuming and applying clean energies, tourism could be a favorable economic development tool that can reduce CO₂ emissions and consequently prevent environmental degradation in the MENA countries. Favorable conditions of MENA countries for tourism development, including rich history and cultural heritage, ancient civilization and unique natural attractions, could be a good reason for boosting tourism [63]. In addition, hosting international mega events in the region such as the Qatar World Cup 2022, and relatively recovering from the pandemic, created the opportunity to boost international tourist flows.

Looking at the result of the EKC hypothesis, it is well-recognized that employing the first-generation econometrics approaches such as PMG, FMOLS and DOLS proved the presence of an inverted U-shaped association between economic development and CO₂ emissions. While previous research findings in the region show the absence of a consensus on the validity of the EKC [99], our results suggest that these conflicting findings may be attributed to country-specific policies, the use of different energy and income measures, as well as the econometric methodology. However, shifting towards the employment of second-generation approaches, the EKC hypothesis was rejected in the region. This implies the importance of econometrics approaches in testing the EKC hypothesis. The panel time-series approaches such as CCE-MG, rather than the conventional panel methods, may provide a better picture of the development of environmental degradation associated with economic growth.

The results also confirmed that trade openness increases CO₂ emissions in the MENA region. This result is in line with Al-Mulali, who reported the positive impact of trade openness on CO₂ emissions in the case of Middle Eastern countries [100]. Furthermore, the finding is consistent with the results of Le et al., indicating the negative relationships between trade openness and environmental quality in middle- and low-income countries [101]. This can be explained by the scale effect, which implies that trade openness increases CO₂ emissions through high levels of economic activities that can be harmful for the environment. Furthermore, based on the PHH, trade liberalization will transfer the pollution-intensive industries from countries with high environmental regulations to countries with lower regulations [102]. Thus, it is expected that developing countries gain comparative advantage in more polluted industries and become pollution havens. Since trade openness has a statistically significant impact on economic growth, therefore the environmental and trade goals should be mutually supported. Our finding implies that international trade liberalization allows some investment companies to invest in polluted industries (e.g., energy industry, manufacturing and construction industry, etc.), which could create more CO₂ emissions, and thus it is suggested that policy makers consider the environment while regularly revising trade policies. This is the case as the region has about 57% of the world’s oil reserves and 41% of natural gas resources [103].

6. Conclusions

According to the obtained results, it is concluded that energy consumption is one of the major determinants of CO₂ emissions in the MENA region. The coefficient of energy consumption variable was bigger than the coefficient of trade openness in all estimated models by all four estimators (PMG, FMOLS, DOLS and CCE-MG). This shows that fossil fuel energy consumption is playing a dominant role in increasing environmental impacts in the region [44]. The abundance of fossil fuel reserves in several MENA countries has made them more dependent on the consumption of conventional energy sources, thus causing a larger share of fossil fuel energy sources in the region’s energy portfolio. Thus, policies should lead to the reduction of fossil fuels’ consumption, and investment in energy-saving and energy-efficiency projects and renewable energy sources in the long run.

As the quality of the environment is important in sustainable tourism development [104] and increasing the levels of CO₂ can lead to global warming and climate change, it is therefore advised that investment and trade regulations should change in favor of environmental sustainability and renewable energy [73]. Policy makers need to apply strict laws and regulations for controlling polluting industries by investors so that trade freedom does not exacerbate greenhouse gases. This notion that developing countries could be a “pollution haven” should be put aside and promote sustainable business sectors. Trade agreements can also strengthen the capacity for governments to address environmental issues, particularly the reduction of trade barriers on environmental goods can lead to increased access to green technologies at a lower cost [101]. Trade openness is also a crucial factor in tourism development. On the one hand, the freedom of trade causes the movement of travelers, especially business travels, facilitates the transportation of goods and services and expands international communications and collaborations. On the other hand, it leads to an increase in CO₂, which reduces the quality of the environment. Therefore, trade openness policies should move towards clean industries and help maintain the quality of the environment through reductions in consuming polluting energies. In addition, governments can impose a carbon tax on polluting business sectors to reduce the greenhouse gas emissions that they emit. Policies that address emission reductions must focus on correcting the industrial structure and enhancing energy efficiency. This importance could be achieved through developing and adopting a sustainable tourism approach, especially in the post-pandemic era, where most economies are experiencing a recovery.

The current study contributes to the mainstream literature by empirically examining the effects of international tourist arrivals on CO₂ emissions in the MENA region over the period of 1995–2018. Previous studies found positive relationships between international tourist arrivals, energy consumption and CO₂ emissions, meaning that increasing tourist arrivals enhances CO₂ emissions and environmental degradation [7,8]. However, we conversely found that increasing international tourist arrivals in the MENA countries region will not lead to increases in CO₂ emissions, questioning this notion that international tourism development will lead to more CO₂ emissions.

Limitations and Directions for Future Research

This study has some limitations that need to be acknowledged. First, we only chose eight countries in the region due to the lack of comprehensive data for most countries in the specified period. Future studies might consider more samples or the whole region to investigate more robust findings. In addition, because it was beyond the objective of this study, we did not consider other influential variables such as political instability in the region, which might affect these findings. Hence, future research could examine the influence of political instability on international arrivals, trade openness and CO₂ emissions_. One particular limitation was the lack of distinction between low-income (e.g., Iran, Egypt, Algeria) and high-income countries (e.g., Oman, Qatar, Kuwait, UAE) in the region, and thus the income level could be an important factor to better understand the effects of tourist arrival flows on CO₂ emissions.