Is Financial Globalization Polluting? – Symmetric and Asymmetric Effects of External Liabilities on CO2 Emissions in the MENA Region


 This paper aims to study the symmetric and asymmetric effects of financial globalization on CO2 emissions in the MENA region. Using a panel dataset of seven non-OPEC MENA countries over the period 1980-2014, we perform a comprehensive econometric analysis based on the panel ARDL and non-linear panel ARDL (NARDL) models and a battery of tests, including cross-sectional dependence tests, second-generation unit root tests and cointegration tests. The findings reveal a significant long-term impact of financial globalization on CO2 emissions that can be symmetric or asymmetric depending on the nature of financial globalization. While external debt liabilities appear to be polluting because they increase CO2 emissions significantly and linearly, the long-term impact of FDI and portfolio investment liabilities on CO2 emissions is asymmetric, with only negative shocks of FDI and portfolio investment decreasing CO2 emissions. This suggests that financial globalization through foreign investment (FDI and portfolio investment) is more environmentally friendly than financial globalization through debt, which provides interesting insights for policy makers.


Introduction
Since the Industrial Revolution, CO2 levels have increased by more than 40% 1 worldwide. About 7% of the 32 gigatons of carbon dioxide (CO2) emitted each year comes from the MENA region (World Bank, 2016). However, these emissions are not homogenous across MENA countries. The major oil producers and members of OPEC (Organization of the Petroleum Exporting Countries) emit the most CO2, and they have access to this fossil fuel at a low cost and no or very low taxes on petroleum products. On the contrary, despite the fact that the non-OPEC MENA countries are relatively less CO2 emitters, they have made very strong commitments at COP21 (the 2015 UN climate change conference) to reduce their emissions, as is the case of Tunisia 2 . This leads the governments of these countries to review their growth model, taking into account their internal constraintsincluding high unemployment ratesand especially external constraintsincluding the negative trade balance and the low quality of technological transferin an increasingly globalized world. Thus, rethinking financial openness is essential to reassess the growth model of non-OPEC MENA countries in light of the link between financial globalization and pollution. Since the 1990s, these countries have opted for greater financial integration allowing foreign capital flows, particularly foreign direct investment (FDI) and external debt, to increase investment, consumption, technology transfer, and income, as is the case for many other developing countries (Hakimi and Hamdi, 2017;Gaies and Nabi, 2019). However, in terms of its effects on climate, financial globalization appears to be a complex and multifaceted phenomenon, especially in the MENA region. For the "pessimists", FDI positively impacts CO2 emissions and thus increases climate risk (Pethigm, 1976;Walter and Ugelow, 1979;Radermacher, 1994;Omri et al., 2014;Abdouli andHammami, 2016, Raggad, 2020). This effect is even stronger in countries with few environmental regulations and many polluting industries, such as the cement and chemical industries. 3 For the "optimists", the effect of FDI on CO2 emissions is negative (Birdsall and Wheeler, 1993;Al-Mulali and Tang, 2013;Ong and Sek, 2013). From this perspective, the technology transfer achieved by FDI provides access to new, less polluting technologies and leads to a reduction in CO2 emissions (Liu et al., 2017). The opposition between these two theses has generated a real debate in the literature, known as the "pollution haven hypothesis" opposing the "pollution halo hypothesis". Recently, this led Xie et al. (2020) to study the direct and spillover effects of FDI inflows on CO2 emissions in emerging countries using a non-linear analysis based on the panel smooth transition regression (PSTR) model. The authors found that both the pollution haven hypothesis and the population halo hypothesis are verified under a threshold effect of FDI on CO2. As far as we know, this is the first study that examines the non-linear relationship between financial globalization and carbon emissions that integrates the two hypotheseshaven and halointo the methodological design. In addition, previous studies on the financial globalization-CO2 emissions nexus appear to have shortcomings in terms of measuring financial globalization in developing countries. Many authors demonstrate the importance of the three pillars of financial globalization, which are FDI, debt, and portfolio investment (Estrada et al., 2015;Trabelsi and Cherif, 2017) on the economic and financial development. Although financial globalization is largely dominated by both FDI and debt in developing countries with increased growth in portfolio investment (Levy-Yeyati and Williams, 2011;Gaies and Nabi, 2019), it is questionable why almost all empirical studies on the relationship between financial globalization and CO2 emissions use FDI alone as an indicator to capture the phenomenon of financial globalization in these countries (Koçak, and Şarkgüneşi, 2018). In order to overcome these two limitations and better understand the relationship between financial globalization and carbon emissions in non-OPEC MENA countries, this study develops a comprehensive econometric analysis based on the panel ARDL and non-linear panel ARDL (NARDL) models and a battery of tests including cross-sectional dependence tests, second-generation unit root tests, and cointegration tests. This allows to examine the symmetric and asymmetric effects of financial globalization on CO2 emissions and to incorporate both the pollution haven and the pollution halo hypotheses in the research design. To the best of our knowledge, our study is the first attempt to study the financial globalization-pollution nexus in the MENA region using the panel ARDL and non-linear panel ARDL (NARDL). Recently, the studies by Raggad (2020) and Boufateh and Saadaoui (2020) used the NARDL technique to examine the link between macro-finance and carbon emissions, but they focused on financial development, not financial globalization, and did not consider the case of MENA countries. This study is also one of the few analyses on the relationship between financial globalization and CO2 emissions that uses three different indicators of financial globalization, namely FDI liabilities, portfolio investment liabilities, and debt liabilities. As mentioned above, previous studies primarily used FDI while ignoring other types of external liabilities.
Overall, the contribution of this paper is twofold. First, using the ARDL and NARDL models, it demonstrates that financial globalization can have various and complex impacts on carbon emissions, thus reconciling the views of "optimists" and "pessimists". It highlights the asymmetric relationship between financial globalization and CO2 emissions. Second, this paper shows the relevance of analyzing financial globalization in MENA countries through its three pillars: FDI, external debt, and portfolio investment. It thus provides a deeper understanding of the relationship between financial globalization and carbon emission. This is particularly important for non-OPEC MENA countries, as a better grasp of the complexity of this link would allow them to review their growth model in light of the global environmental issue in a context of financial globalization. Moreover, while non-OPEC MENA countries may not be the most polluting in the MENA region, they already suffer from high temperatures and water shortages and have little room to adapt given their economic and financial underdevelopment (Gilmont, 2015).
The remainder of this paper is organized as follows. Section 2 presents a selection of empirical studies on the relationship between financial globalization and CO2 emissions in developing countries, particularly in the MENA region. The data, empirical framework, and results are discussed in section 3. Section 4 explains the autoregressive distributed lag (ARDL) and the non-linear autoregressive distributed lag (NARDL) models and discusses their results. Section 5 concludes.

Literature review
The literature on the macro determinants of carbon emissions is clearly abundant. In order to study the impact of financial globalization on CO2 emissions in seven MENA countries, we focus our review on studies concerning developing countriesincluding MENA countriesconsidering one or more aspects of financial globalization as the explanatory variable(s) of interest. In doing so, it appears that almost all previous studies have used FDI inflows as the sole indicator of financial globalization and CO2 emissions as the indicator of CO2 emissions. It is no secret that FDI inflows have a considerable effect on GDP growth (e.g., Estrada et al., 2015;Trabelsi and Cherif, 2017;Gaies et al., 2020). Beyond economic growth, scholars have also investigated other impacts of FDI inflows, including on the environment. In this sense, two strands of literature have emerged, the first pointing towards a positive impact of FDI on CO2 emissions, and thus a harmful effect on the environment (Abdouli and Hammami, 2016).

Positive effect of FDI on CO2
This "pessimistic" current finds its origin in the Pollution Haven Hypothesis (PHH) developed by Pethigm (1976) and Walter and Ugelow (1979). In analyzing the PHH, some authors point to a link between FDI inflows and CO2 emissions, where the former increase the latter, particularly in industries with high levels of pollution in countries with few environmental regulations. In fact, with reference to Abdouli and Hammami (2016) who examine a sample of developed countries, Shahbaz et al. (2019) explore the PPH in MENA countries and the increase in industries with high pollution levels, confirming that FDI generates CO2 emissions. For the Middle Eastern countries Saudi Arabia, Oman and Qatar, Kari and Sadam (2012) find that the increase in FDI inflows also contributes to increasing per capita CO2 emissions. Contrary to the assumption that FDI inflows include the replacement of polluting technologies with less polluting ones (e.g., Gallagher 2004Gallagher , 2009Liu et al., 2017), the authors evidence that for the three countries cited, FDI inflows have not contributed to more sustainable growth, thus raising CO2 emissions. In the same vein, Neequaye and Oladi (2015) and Shahbaz et al. (2015) investigate the effect of FDI on CO2 emissions, supporting the PHH for the MENA region. Similar results have been found for Ghana (Solarin et al., 2017) and Turkey (Koçak and Sarkgünes, 2017). Other authors, including Lan (2012), Shahbaz et al. (2015), Ouyang and Lin (2015) and Nasir et al. (2019) also confirm the PPH in the case of developing countries. In addition, besides their direct effect, FDI inflows may also impact CO2 emissions indirectly via economic growth (Xie et al., 2019). In fact, on the one hand, FDI and GDP growth are indisputably linked, with FDI influencing growth, as has been shown in numerous studies (e.g., Estrada et al., 2015;Trabelsi and Cherif, 2017;Gaies et al., 2020). On the other hand, the existing literature provides evidence of a directequally undisputedrelationship between growth and CO2 emissions (e.g., Sinha et al., 2017;Nabavi-Pelesaraei et al., 2018;Kaab et al., 2019).

Negative effect of FDI on CO2
The second strand -an "optimistic" oneoften invokes the pollution halo hypothesis developed by Birdsall and Wheeler (1993). This current generally refers to the positive influence that technology transfer has on reducing CO2 emissions (Xie et al. 2019). Along these lines, and contrary to the previously mentioned investigations on the PHH, Al-Mulali and Tang (2013) refute the positive link between FDI and CO2 emissions in host countries. In fact, by testing the PHH for Middle Eastern countries of the Gulf Corporation Council, the authors find a negative impact of FDI on CO2 emissions. According to Gallagher (2004Gallagher ( , 2009, FDI inflows can negatively impact CO2 emissions if these inflows are related to the transfer of technology to developing countries, since they usually help to replace outdated and polluting technologies. In this sense, Liu et al. (2017) find that FDI inflows lower CO2 emissions, indicating that the former can serve as a means to roll out less-polluting technology. Ong and Sek (2013) obtained the same results using an ARDL approach for lowand middle-income countries between 1970 and 2008, showing that FDI inflows can reduce CO2 emissions. As a consequence, Bakhsh et al. (2017) argue that FDI should not be promoted at the expense of rising CO2 emissions, since studies examining the PHH have not reached a clear consensus. One of the first studies to address this issue is that of Xie et al. (2020). The authors incorporate both the pollution haven and the pollution halo hypotheses in their research design to study the effect of FDI inflows on CO2 emissions approximated by the increase or decrease in CO2 emissions. The analysis is based on the panel smooth transition regression (PSTR) model and a sample of emerging countries covering the period 2005-2014. Xie et al. (2020) show that both the pollution haven hypothesis and the population halo hypotheses are verified under a threshold effect of FDI on CO2.
In light of the foregoing, we contribute to the existing literature by two interesting ways.
First, in line with Xie et al. (2019), we try to reconcile the two strands of literaturethe "optimists" and the "pessimists"by assuming that financial globalization may have different and complex impacts on CO2 emissions. Thus, our study is the first to apply both the panel ARDL and panel NARDL models to investigate the possible symmetric and asymmetric effects of financial globalization on CO2 emissions in seven MENA countries. This research design allows us to capture the short-and long-term effects of financial globalization on CO2 emissions, as well as its positive short-and long-run shocks and its negative short-and longrun shocks. We can therefore conclude which of the two opposing hypotheseshaven or halo triumphs in our country sample.
Second, as mentioned earlier, previous studies have focused on FDI as an indicator of financial globalization. They therefore neglect external debt and portfolio investment, which are the other two main components financial globalization (e.g., Estrada et al., 2015;Trabelsi and Cherif, 2017;Gaies et al., 2020). Considering FDI, external debt and portfolio investment liabilities, we go one step further than the existing literature, providing a more comprehensive analysis.

Data and empirical framework
The dataset covers the period 1980 to 2014 for 237 country-year observations from seven MENA countries, including Bahrain, Egypt, Jordan, Lebanon, Morocco, Oman, and Tunisia. The sample was established under the constraint of data availability and the relative homogeneity of the countries. We thus exclude the largest oil-exporting countries in the MENA region (OPEC members) because they have heterogenous CO2 emissions, economic growth and energy consumption compared to the other countries of the region. We also exclude countries with too much non-available data for the entire period.
As in previous studies (see Section 2), we use CO2 emissions as a proxy for CO2 emissions extracted from the World Bank Open Data. We explain this dependent variable by three indicators of financial globalization, which are external debt, FDI and portfolio investment liabilities, from the External Wealth of Nations database developed by Lane and Milesi-Ferretti (2018). These variables are commonly used in the literature on the link between financial globalization and growth (e.g., Estrada et al., 2015;Trabelsi and Cherif, 2017;Gaies et al., 2020) and they allow us to go one step further than previous studies which only consider FDI as a variable of financial globalization. As control variables, we consider GDP per capita, energy consumption and urban population, which are also extracted from the World Bank Open Data. They are considered as the usual control variables by previous studies. Descriptive statistics for the variables are presented in Table 1 and Figure  1 below.

Figure 1. CO2 and external liabilities evolution over the period 1980-2014
Based on the data presented above, we study the impact of financial globalization on CO2 emissions using a comprehensive econometric analysis.
First, we test the cross-sectional dependence across our country sample to choose the adequate panel unit root test (first-or second-generation tests). Second, we check the order of integration of the variables to verify the use of the ARDL approach. This approach cannot be adopted if the order of integration exceeds one. Third, we examine the long-run association among the variables used in our model by applying a set of cointegration tests. Fourth, we estimate the panel ARDL model explaining the CO2 emissions by the external liabilities and control variables to investigate the symmetric impact of financial globalization on CO2 emissions in the short-and long-run. Finally, we apply the non-linear panel ARDL approach (NARDL) to capture the short-and long-term effects of financial globalization on CO2 emissions, as well as its positive short-and long-run shocks and its negative short-and longrun shocks.

Cross-sectional dependence and unit root tests
The unit root test is performed for variables to check the order of integration. It is important to test the cross-sectional dependence for panel datasets in order to choose the appropriate panel unit root test. 4 In homogenous panels, cross-sectional links may emerge from common economic and/or non-economic macro-dynamics, such as institutional, demographic, macroeconomic, geographic and/or political aspects. Economic and financial globalization has increased these common macro-dynamics. Thus, looked at in another way, a possible cross-sectional dependence in our sample corroborates its homogeneity and could indicate financial integration between the countries given their geographical position and their policy of financial openness in a context of financial globalization.
We perform the Breusch and Pagan (1980) test and the Pesaran (2004) scaled test to examine the existence of cross-sectional dependence among residuals.
The Lagrange multiplier test proposed by Breusch and Pagan (1980) ( ) is the appropriate choice when the number of cross section N is smaller than the number of periods T, which is the case for our framework. To calculate the statistic, we consider the following equation: Where , is the dependant variable, , is a × 1 vector of the explanatory variables, are the individual intercepts, and is a × 1 vector of the coefficients to be estimated. Under the null hypothesis of cross-sectional independence, , is supposed to be independent The statistic of Breusch and Pagan is given by: However, the is not suitable with large N. To resolve this problem, Pesaran (2004) develops the following scaled version of test: As explained above, before proceeding with the unit root test to study the integration order of our variables, we perform Breusch and Pagan (1980) test and Pesaran (2004)  scaled test of cross-sectional dependence (LM; CDLM). Table 2 shows a cross-sectional dependence in our three specifications -CO2-debt, CO2-portfolio investment and CO2-FDI with GDP per capita, energy consumption and urban population as control variables (Models 1, 2 and 3)according to the Breusch-Pagan (1980) and Pesaran (2004) statistics. This indicates a common dynamic among the countries that corroborates the relative homogeneity of our sample, and the important influence of the phenomenon of financial globalization in the MENA region. In addition, due to the cross-sectional dependence in our models, we apply the second-generation panel unit root tests developed by Pesaran (2007) Table 3 shows that the variables used in our Models 1, 2 and 3 are integrated in a different order [I (0) or I (1)]. This result indicates that these variables are stationary in level and/or first difference, which confirms the use of the panel ARDL and NARDL models since they do not require the variables to be integrated in the same order to capture the adjustment process of the variables to their long-run levels. However, we performed the unit root test to check whether the order of integration exceeds one [I (.) ≤ I (1)] because if it does, the estimation of the panel ARDL and NARDL becomes inconsistent (Pesaran et al., 1999;Shin et al., 2014).

Cointegration tests
In order to examine the possibility of a long-term relationship between our variables, we apply the two panel cointegration tests introduced by Pedroni (1999Pedroni ( , 2004 and Kao (1999), with the null hypothesis being the absence of cointegration. Pedroni (1999Pedroni ( , 2004 developed seven tests to test the null hypothesis. Four tests are based on the within-dimension methods and are computed by separately adding the numerator to the denominator over the N cross-sections. The other three tests are based on the between-dimension methods and are computed by splitting the numerator and the denominator prior to summing the N crosssections. The Kao (1999) panel cointegration test is based on the ADF test, following the standard approach adopted by the Engle-Granger step procedures.
We perform the Pedroni (1999Pedroni ( , 2004 and Kao (1999) tests to investigate a potential long-run relationship between the variables used in our Models 1, 2 and 3, namely CO2 emissions, external debt, FDI and portfolio investment liabilities, GDP per capita, energy consumption and urban population. The coefficients of these tests are reported in Table 4. They are mostly significant at the 1% and 5% levels, indicating that co-integration exists among the variables used in our three Models (1, 2 and 3). In other words, the null hypothesis of the Pedroni (1999; and Kao (1999) tests corresponding to non-co-integration is rejected at the 1% and 5% levels of significance, showing a significant long-term association between CO2 emissions, external debt, FDI and portfolio investment liabilities, GDP per capita, energy consumption, and urban population in the sample.

Panel ARDL model
Using panel data models to study the relationship between CO2 emissions and financial globalization has many benefits for empirical studies. It allows to take into account the cross-sectional attributes between countries, especially in a homogenous sample, and detects the time series interactions between the variables. Moreover, the large number of observations usually included in panel data allows for more efficient estimates. In this study, the sample includes seven countries over 35 years. Hence, the numbers of time series observations (T) is larger than the number of cross-sectional observations (N). As proposed by Pesaran and Smith (1995) and Pesaran et al. (1999), the panel ARDL model is the most suitable panel model for this study. In addition, the ARDL model can lead to efficient estimates irrespective of whether the variables are integrated in the same order I (0) or I (1) which is the case with our variables (see section 3.2).
The panel ARDL (p, q, q, …, q) model is written as follows: represents the specific effect of the group ; , is a × 1 vector of the regressors; are × 1 vectors of the coefficients; and are scalars. Based on equation (5), the error correction model is specified as follows: The term ( , −1 − ′ , ) is the error correction term. If the parameter ∅ is negative and significant, it indicates a convergence to the long-run equilibrium. In addition, the vector ′ is composed by the long-run links between the variables.
The estimations of the panel ARDL model are obtained using two different estimators: the mean group estimator (MG) and the pooled mean group estimator (PMG). The MG estimator, introduced by Pesaran and Smith (1995), allows to capture the heterogeneity of long-run and short-run parameters. It estimates equations for each individual country and then calculates the coefficients' means. Thus, the possible homogeneity across countries is not taken into account. The PMG estimator, developed by Pesaran et al. (1999), allows the shortrun coefficients to vary across countries, however, it assumes that the long-run coefficients are the same among countries (Dimitriadis and Katrakilidis, 2020). After running the MG and PMG models, it is necessary to choose the best estimator. For this purpose, the Hausman test is performed to compare the long-run coefficients of these estimators. If the null hypothesis is not rejected, the PMG estimator is preferred. If the alternative hypothesis is accepted, the MG estimator is more appropriate.
In order to study the impact of financial globalization on CO2 emissions , we develop the following model: 2 , = 0 + 1 , + 2 , + 3 , + 4 , + , Where 2 , , , , , , , and , represent respectively the natural logarithm of CO2 emissions per capita, the financial globalization indicators (debt, portfolio investment and FDI liabilities), the natural logarithm of GDP per capita, the natural logarithm of the per capita energy consumption, and the urban population. , is the error term.
According to Peseran et al. (1999), the ARDL model is written as follows:

=0
The Schwarz Bayesian Criterion (SBC) criterion and the Akaike Information Criterion (AIC) are employed to select the appropriate lag length.
As mentioned above, we apply the panel ARDL estimation to capture the symmetric impact of external liabilities (financial globalization indicators) on CO2 emissions (CO2 emissions indicator) for our sample of seven MENA countries for the years 1980 to 2014. As shown in Table 5, the Hausman statistics indicate that the maximum likelihood-based PMG estimator is more efficient than the maximum likelihood-based MG estimator at the significance level of 5%, which led us to apply the PMG model, as suggested by Pesaran et al. (1999). In addition, Table 5 reports the long-and short-run coefficients of the external liabilities indicators and the other explanatory variables. It displays a positive and significant long-run coefficient of debt liabilities at the 10% level, while the long-run coefficients of FDI and portfolio investment liabilities are statistically non-significant at conventional levels. The table also shows that the long-run coefficients of GDP per capita and energy consumption are positively and significantly linked to CO2 emissions. However, the model does not capture a significant long-run impact of urban population on CO2 emissions. With regard to the shortrun coefficients, Table 5 reveals that only the effect of urban population on CO2 emissions is positive and significant at the 5% level. Finally, the negative and significant coefficients of error correction term (ECT (-1)) at the 1% level, prove that a significant link exists between the short-and long-run effects captured by our model.

Non-linear panel ARDL (NARDL) model
In order to take into account the long-and short-run asymmetries in the impact of financial globalization on CO2 emissions , we perform the non-linear ARDL (NARDL) model proposed by Shin et al. (2014). Using this approach, the value of the financial globalization (GLOB) indicators (Debt, Portfolio and FDI) are divided into positive and negative shocks, noted respectively as By replacing + and − into in equation (9), we obtain the following non-linear panel ARDL model: Where 1 + and 1 − are the short-run coefficients associated with positive and negative shocks, respectively.
The error correction version of the equation (12) is expressed as follows: The mitigated effects of FDI inflows on CO2 (the pollution haven hypothesis vs the pollution halo hypothesis) that emerged from previous studies (see Section 2), as well as the non-significant symmetric impacts of FDI and portfolio investment liabilities on CO2, highlighted by our panel ARDL estimates, led us to reconduct our regressions by applying the panel NARDL model developed by Shin et al. (2014). The panel NARDL results are presented in Table 6 which includes the long-and short-run coefficients of the external liabilities variables, as well as GDP per capita, energy consumption and urban population. In addition, Table 6 reports the results of the Wald symmetry test which highlights the existence of long-term asymmetric effects at the 5% level for the FDI-CO2 nexus (Model 2) and the portfolio investment-CO2 nexus (Model 3), while the test indicates an absence of asymmetric effects for the external debt-CO2 nexus (Model 1) and for all models in the short term. The long-run coefficients of Models 2 and 3 reveal that a negative shock of FDI and portfolio investment liabilities decreases CO2 emissions. Thus, it appears that financial globalization through foreign investment is more environmentally friendly than financial globalization through debt (Table 5) for MENA countries. Moreover, the estimates of Models 2 and 3 show a positive impact of energy consumption, GDP per capita and urban population on CO2 emissions in the long run. The Hausman test statistics in Table 6 approve the use of the PMG model as is the case with the panel ARDL model (Table 5), and the negative and significant coefficients of the error correction term (ECT (-1)) at the 1% level confirm the convergence between the short-and the long-run shocks. In summary, our results confirm that the relationship between financial globalization and CO2 emissions is complex and multifaceted. They reveal that in the seven MENA countries we studied over the period 1980-2014, CO2 emissions are significantly impacted in the long term by external liabilities, while in the short term, this impact does not seem to be really significant. The long-run impact of financial globalization on CO2 emissions can be symmetric and asymmetric depending on the nature of this globalization. When financial globalization is based on debt liabilities, it seems risky for the climate because it increases CO2 emissions significantly and linearly. In contrast, the long-term impact of FDI and portfolio investment liabilities on CO2 emissions is asymmetric, with only negative shocks of FDI and portfolio investment decreasing CO2 emissions, thus demonstrating that a lower stock of FDI and portfolio investment equals higher CO2 emissions. It appears that financial globalization based on foreign investment (FDI and portfolio investment) is more environmentally friendly than financial globalization based on external debt. This can be explained by the fact that foreign investment, in particular FDI, unlike external debt, has a positive spillover effect on the climate by bringing green and ethical businesses and technologies that reduce CO2 emissions in recipient countries. It goes without saying that, in the meantime, FDI and portfolio investment can catalyze CO2 emissions in recipient countries, particularly developing countries, which aim to promote economic growth by relaxing environmental regulations to attract foreign capital. However, it appears from our panel NARDL results that only the environmentally beneficial effect of foreign investment (FDI and portfolio investment) is significant in the long term. This refutes the pollution haven hypothesis in favor of the pollution halo hypothesis. On the contrary, external debt does not ensure technology transfer. They could allow the financing of economic growth through the mobilization of foreign capital, but without a positive spillover effect on the environment, worse still, they increase CO2 emissions. This could be explained by the fact that in our seven MENA countries, external debt is oriented towards consumption and polluting production. The long-term positive impact of our control variables -GDP per capita, energy consumption and urban populationon CO2 emissions corroborates this pattern.

Conclusion
In this paper, we studied the symmetric and asymmetric effects that financial globalization may have on CO2 emissions in the MENA region. Covering the period 1980-2014 for seven non-OPEC MENA countries, we applied the panel ARDL and NARDL models. As part of this comprehensive econometric investigation, we also conducted a battery of tests, including cross-sectional dependence tests, second-generation unit root tests and cointegration tests. Our results show a significant long-term impact of financial globalization on CO2 emissions. Depending on the nature of financial globalizationachieved through external debt or foreign investment (FDI and portfolio investment)the impact can be either symmetric or asymmetric. As for external debt, it increases CO2 emissions in a significant and linear way and therefore constitutes CO2 emissions. On the contrary, since the long-term impact of FDI and portfolio investment liabilities on CO2 emissions is asymmetric, only negative shocks of both FDI and portfolio investment liabilities decrease CO2 emissions, while positive shocks do not increase CO2 emissions. In other words, this suggests that financial globalization through foreign investment is more respectful of the environment than financial globalization through debt. These results can be explained by the fact that foreign investment in general and FDI in particular contributes to the establishment of green businesses and technologies in recipient countries that help reduce CO2 emissions, as opposed to external debt. In fact, not only does the latter fail to ensure the transfer of technology, but it also increases CO2 emissions, as the mobilization of foreign capital to finance economic growth does not have a positive spillover effect on the environment. This could be explained by the fact that in our seven MENA countries, external debt is oriented towards consumption and polluting production. In summary and drawing on our findings, we can refute the pollution haven hypothesis in favor of the pollution halo hypothesis. These results provide interesting insights for policy makers in MENA countries by highlighting the relevance of strict regulation of foreign capital flows, apart from portfolio investment and FDI, to reduce CO2 emissions and address the climate challenge. In future research, we suggest comparing the results of these non-OPEC MENA countries with other non-oil exporting emerging countries, as they are also subject to almost the same economic, social and environmental constraints.

Author Declarations
-Funding: Not applicable' -Conflicts of interest/Competing interests (include appropriate disclosures): Not applicable' -Availability of data and material/ Data availability (data transparency, if link please provide the link to access. For further information, go to https://www.springernature.com/gp/authors/research-data-policy/data-availabilitystatements/12330880): Not applicable' -Code availability (software application or custom code): Not applicable'