“Energy consumption, CO² emissions and economic growth in MENA countries”

This study investigates the relationship between economic growth, final consumption, investment, energy use and CO² emissions in two groups of Middle East and North Africa (MENA) countries: Oil Poor Countries (OPC) and Oil Rich Countries (ORC). It is assumed and verified that the structural relationship between GDP growth, energy use and CO² emissions is different in these two groups of countries. FGLS panel esti- mations were carried out over the period 1974–2014. In ORC, no significant relationships are observed between energy use and GDP, whereas CO² emissions and GDP are positively linked. In OPC, there are opposite connections: a positive link between GDP and energy use, whereas the impact of CO² emissions on GDP tends to be nega- tive. In both groups of countries, a positive and bi-directional link is observed between energy use and CO² emissions. The strength of this link is twice bigger in OPC than in ORC. This indicates that CO2 reduction policies conducted through energy use con-trol (quantitative and qualitative) will have higher effect in OPC than in ORC. This also shows that the relationships between economic growth, energy use and CO² emissions differ noticeably and structurally between OPC and ORC. These results provide new insights into the opportunities and threats faced by CO2 reduction policies in OPCs and ORCs. initiated the idea that the relationship between economic growth and the environment may be divided into two stages. development stage uses intensively inexpensive environmental resources. Through time, their depletion induces a social demand for stronger environmental policies and a rise in their costs, which stimulates the development of eco-friendly solutions. Consequently, leads to a second stage of cleaner economic growth and possibly to an environmental recovery from the degradations initiated in the first stage of development. The relationship between the environmental pollution and economic growth, as well as energy consumption, been an area of intense research. Many subsequent studies test these relationships


INTRODUCTION
Increasing CO² emissions pose a serious threat to the environment, which is a common anxiety for both developing and developed countries (Zhang et al., 2015;Muhammad & Khan, 2019). Air pollution was ranked 5 th highest risk factor of mortality in 2017 at the global level (HEI, 2019). In 1992, the World Development Report discussed in detail the relationship between the environment and economic activity. It was argued that their connection is not linear, but has an inverted U shape (World Bank, 1992, pp. [36][37][38][39][40][41][42][43]. Grossman and Krueger (1993) developed similar analysis. These researchers initiated the idea that the relationship between economic growth and the environment may be divided into two stages. The first development stage uses intensively "free" inexpensive environmental resources. Through time, their depletion induces a social demand for stronger environmental policies and a rise in their costs, which stimulates the development of eco-friendly solutions. Consequently, this leads to a second stage of cleaner economic growth and possibly to an environmental recovery from the degradations initiated in the first stage of development. The relationship between the environmental pollution and economic growth, as well as energy consumption, has been an area of intense research. Many subsequent studies test these relationships and are now referred to as the Environmental Kuznets Curve (EKC 1 ) approaches. They have investigated various types and causes of environmental degradations in many countries (Kaika & Zervas, 2013; Antonakakis et al., 2017;Acheampong, 2018;Mikayilov et al., 2018;Muhammad, 2019). However, the empirical confirmation of the relationships hypothesized by the EKC has remained ambiguous and controversial until now. The results differ according to the periods of time, the countries under study and the econometric methodologies. For example, Acheampong (2018) and Muhammad (2019) compare the relationship between economic growth, CO² emissions and energy use in different regions. They notice that the parameter estimates differ strongly according to the countries and the estimation methods. Furthermore, as stated by Davis and Caldeira (2010, p. 5690), accounting based on consumption reveals that a substantial part of the CO² emissions is traded internationally. Therefore, these emissions are not reported in the national emission inventories when statistics are based on production data. Assigning responsibility of these emissions for consumers may be a solution of compromise between the need of economic growth in the countries where the energy is produced (currently those countries that are rich in oil) and those who use it (oil poor countries).
This study aims to investigate the relationship between economic growth, final consumption, investment, energy use and CO² emissions. Several empirical analyses have already been conducted on a large scale of heterogeneous countries. Their findings are often very general and difficult to convert into relevant recommendations for the national public policies. Initially national and natural resource endowments are scarcely considered, whereas they may be seen as a critical incentive to engage into environmentally friendly policies. Regarding oil, which is a non-renewable natural resource with many potential substitutes, the main long-term problem faced by oil rich countries (ORC) may be an "apocalyptical run" for the appropriation of the benefits of their gradually ending oil reserves before the emergence of less expensive green substitutes. At the international market scale, these strategies may result in keeping oil prices relatively low until the total depletion of their natural reserves. Low oil prices reduce the incentives to invest in initially more expensive green energies and delay the reduction of CO² emissions by the oil poor countries (OPC). Poor countries often try to promote their international competitiveness through productivity gains obtained by cost reduction strategies rather than by innovation. They often specialize in low-tech markets driven by price competition. Consequently, in a poor country with poor oil endowment, investment decisions in green vs fossil energy production may have important consequences for their international competitiveness and long-term growth prospects. They may face the traditional tradeoff between early stage economic growth and environmental preservation hypothesized by the EKC. In all cases, the goal of OPCs and ORCs is to promote their "self-interest" through GDP growth.
The EKC raises the question of a link between growth and environmental degradation without explicitly considering the possible impact of the natural resource endowments. This work tries to address this question by focusing on the impact of countries oil endowments on the relationship between economic activities, energy use and environmental damages caused by CO2 emissions. The first section briefly reviews the literature and explains the research process. The second section explains the methodology, describes the data and presents the results of the estimations. In conclusion, research limitations and implications are discussed. 1 The EKC is named for Kuznets (1955) who hypothesized that income inequality first rises and then falls as economic development proceeds.

LITERATURE REVIEW
Many studies examine the environmental impact of economic activities around the world through the lens of the EKC model (see, for example, Panayotou, 1993Panayotou, , 2003 and System Generalized Method of Moments (Sys GMM). He observes that economic growth influences energy consumption positively and significantly by SUR method and negatively and significantly by Sys GMM method in developed countries, positively and significantly in emerging countries (by SUR and GMM method), but negatively and significantly in MENA countries only by GMM method. He also shows that the effect of economic growth on the CO² emission is positive and significant in developed countries by SUR and GMM method and MENA countries by SUR and Sys GMM method, but is negative and significant in emerging countries by SUR and Sys GMM method. Additionally, the author shows that the impact of the CO² emission on the economic growth is positive and significant in developed countries, and is positive and significant in MENA countries by only dynamic models, but is negative and significant in emerging countries. He indicates that CO² emissions have a direct and significant impact on energy consumption in all developed, emerging and MENA countries by using the three methods of estimation. He notes that the impact of energy consumption on economic growth is positive and significant in developed countries by GMM method and in emerging countries by SUR and GMM method but is negative and roughly significant in MENA countries only by Sys GMM method. He emphasizes that the impact of energy consumption on the CO² emission is positive and significant in emerging and MENA countries by the three methods of estimation, but positive and significant in developed countries only by SUR and GMM method (Muhammad, 2019).
One possible reason explaining the diversity of results and their sensitivity to the econometric procedure may be that most studies implicitly consider that all the countries enjoy similar natural resource endowment and are engaged in similar growth regimes. The reality is much different: many key natural resources are unevenly distributed across countries. Countries with larger endowment usually exploit and export these resources to countries where the natural resource is scarcer. That international specialization may result in very different patterns of interaction between economic growth, investment, final consumption and pollution. From that point of view, the case of oil may be an interesting topic. That resource is unevenly distributed and its use has many harmful environmental consequences. Its production, transformation and consumption spread all over the world. One part of the harmful consequences of that business is internalized by each country 2 but some of them fall in the common pool; it is particularly the case of CO² emissions induced by oil. As a result, the direct relationship between GDP growth and polluting activities that is hypothesized by the EKC may not be easily applied to that kind of pollution. 2 Oil producing countries lose a non-renewable resource and often cause soil and water pollutions; countries that perform transformation generate localized air pollution and face industrial hazards; finally, the countries in which consumption occurs generate localized air pollution (in the form of fine particles).
This study aims to contribute to this debate by measuring the impact of countries' oil degree of abundance on the connection between GDP growth, final consumption, investment, energy use and CO² emissions. Two samples of MENA countries are compared, namely Oil Rich MENA's Countries (ORC) and Oil Poor MENA's Countries (OPC). The hypothesis is that the nature and intensity of the relationship between economic activity and CO 2 emissions differs from OPC to ORC. If that hypothesis proves right, it will be an opportunity to discuss its implications from the point of view of environmental policymaking.
According to the above-mentioned literature and discussion it is expected that the intensity of the relationship between the economic vs energetic variables differs between OPC and ORC.
Concerning the nature of the influential variables on GDP, it is known that a large part of the OPC national income results from energy consuming activities. On the contrary, in ORC, a large part of the national income comes from CO² emitting extractive activities. Consequently, it is expected that energy use should have a stronger impact on GDP growth in OPC than in ORC (H1 : "The impact of energy use on GDP is larger in OPC than in ORC"). In ORC, petrol rents result in a trade surplus that is not directly consumed but saved and invested abroad. Therefore, in ORC, GDP growth may not be associated with higher energy use. On the contrary, in oil importing OPC, it is assumed that GDP growth has a direct impact on the internal consumption that should result in higher energy use. Consequently, it is expected that the impact of GDP growth on energy use should be stronger in OPC than in ORC (H2 : "GDP growth has a stronger impact on energy use in OPC than in ORC). Concerning CO² emissions and their strong link with extractive activities, it is expected that there is a positive impact of CO² emissions on GDP growth in ORC, but no impact in OPC (H3: "In ORC, CO² emissions have a positive impact on GDP growth"; H3': "In OPC, CO² emissions have no impact on GDP growth"). Finally, in both groups of countries, a positive relationship between energy use and CO² emissions is expected (H4: "In both groups of countries, CO² emissions stimulate energy use"; H5: "In both groups of countries, energy use induces higher CO² emissions").

METHODS
This study adopts the Cobb-Douglas production function to study five-way linkages between economic growth, final consumption, investment, energy use and CO² emissions. In this respect, the general form of the Cobb-Douglas production used by many authors (Acheampong, 2018; Muhammed, 2019; Muhammad & Khan, 2019) is as follows: where Y is the output, A is a global constant, and π is the residual term assumed to be identically, independently and normally distributed. The returns to scale are associated with investment, final consumption, energy use and CO² emissions, which are drawn by α 1 , α 2 , α 3 and α 4 , respectively.
The logarithmic transformation of the Cobb-Douglas production function (1) is given by: The five-way linkages are examined using the following system of equations: Eq4:  (Baum, 2001;Podestà, 2002). In the two groups of countries, it was deduced that the phenomenon of heteroscedasticity was present in five equations (see Appendices C and D). The Breusch-Pagan LM test is used to identify possible cross-sectional dependence in the error term between the studied entities (Baum, 2001 An individual fixed effect model correlates with the panel data. Estimations are performed using the feasible generalized least square (FGLS) method (Blackwell, 2005, p. 204). According to Podestà (2002, p. 14), "It is feasible to use FGLS method because it uses an estimate of variance-covariance matrix, avoiding the generalized least squares (GLS) assumption that Ω is known." Reed and Ye (2011, p. 986) find that "the FGLS is the best overall estimator with respect to efficiency, but the worst when it comes to estimating confidence intervals. This means that researchers may have to use one estimator if they want the 'best' coefficient estimates, and another if they desire reliable hypothesis testing." The estimations are examined individually on each of the five equations in the two groups of countries.

RESULTS AND DISCUSSION
In light of the above, Table 1 shows that the mean of all variables is greater in ORC than their counterparts in OPC. The level of CO² emissions is double in ORC. In addition, energy use is more volatile in ORC. However, it is less volatile in OPC in terms of the standard deviation (Std. dev). Table 2 and Figures 1-4 clearly show that the common patterns are observed in both OPC and ORC: the links between lnY and lnC, lnY and lnK and between lnEu and lnCO² are significantly positive and bidirectional. This means that there are strong positive interactions between GDP and its two main components such as the final consumption and investment. Except the link from lnEu to lnK, which is not significant in both groups, the connections between the economic variables (lnY, lnC and lnK) and the energetic ones (lnEu and lnCO²) differ strongly. Many relationships with opposite signs are observed in OPC and ORC, basically related to the links between economic and energetic variables. The relationship between lnC and lnK differ strongly between OPC and ORC, which certainly reflects very different underlying productive structures.
Eq1 indicates that the GDP's fluctuations (lnY) are positively explained by those in the final demand (lnC) and the gross capital formation (lnK) in the two groups of countries. This is in line with the national accounting perspective. In the MENA countries, the coefficient associated with lnC is more important than that of lnK. It is over triple in oil rich countries (0.613 vs 0.179). It indicates  that in the MENA countries, economic growth is more sensitive to the domestic final demand than to investment behaviors.
Interestingly, it is observed that the variables lnEu and lnCO² have opposing impacts in OPC and ORC. LnEu has a positive and significant effect on lnY in OPC but not in ORC. The positive impact of lnEu in OPC is in line with the fact that energy is an input into many production processes and consumption activities such as transportation or cooking. In the economies where the service sector 5 accounts for a low proportion of GDP, the energetic intensity of business activities in OPC is certainly higher than that in ORC that are specialized in oil exports rather than in energy intensive activities such as agriculture or industry. In ORC, the impact of lnCO² on lnY is positive, whereas it is usually considered as a waste resulting from the economic activities (production and consumption) rather than as an input of the GDP. This may be explained by the fact that oil extraction is an important source of CO² emissions through the process of flaring (Masnadi et al., 2018, p. 851). Consequently, the CO² emission may appear as a precursor of the sales (and GDP) in ORC, which results in a positive relationship between lnCO2 and lnY. Conversely, in OPC, the impact of lnCO2 tends to be negative. This is certainly explained by the fact that in OPC the CO² emission mainly results from the oil imports whose impact on the GDP is mechanically negative. The conclusion is that OPC and ORC may face very different incentives to engage in CO² reduction policies: CO² increase promotes GDP of ORC, whereas it may be associated with a decrease in GDP in OPC due to fossil energy imports. It is also clear that OPC need energy to sustain their development, whereas in ORC energy consumption is not a key incentive for their economic growth.
Eq2 signifies a positive and significant impact of GDP on final consumption in MENA countries. A slight crowding-out effect of the investment on the final consumption (lnK → lnC = -.124) is observed in OPC, which is not observed in, ORC. This is certainly due to the fact that ORCs usually enjoy a trade surplus and a benefit from large financing capacities. On the contrary, OPCs usually face tight constraints on their trade balance, which leads to reduced financ-ing capacity. Concerning the impact of energy use and CO² on consumption, it is noticed that there are large differences between ORC and OPC. In ORC, fluctuations in final consumption are preceded by others in energy use (lnEu → lnC = .133). This is certainly explained by the behavior of consumers who need more transportation (and energy) to sustain their purchasing process. This phenomenon is not significant in OPC. This is certainly because energy is mainly used in the production processes.
Eq3 shows that in OPC and ORC, lnY has always a positive and significant effect on lnK. However, it is as twice as larger in OPC (2.268) than in ORC (.944). This suggests that there is probably an initial situation of under-equipment in OPC and that there is a major incentive for the investment in the aggregated demand. However, the reverse impact of K on Y is weak; undoubtedly, this is due to the equipment imports in OPC and ORC (LnK → LnY = .335 and LnK → LnY = .179, respectively). The OPC final consumption has a significantly negative impact on investment (LnC → LnK = -1.222). The crowding-out effect is absent in ORC. This is certainly due to the mentioned situation of the lack of financial capacities in OPC. As Wen (2006, pp. 378-379) suggests, "such a crowding-out effect is more likely in situations of the temporary demand shocks and the weak returns to scale in the production technology. However, if the demand shocks are sufficiently persistent, then even in the absence of increasing returns to scale investment can be pro-cyclical and highly volatile". The intuition is mainly associated with the anticipated higher future demands after the shock can only be met by higher savings financed by the lower level of consumption when the access to the international financial markets is limited. The results show that in OPC, investment is neither determined by the energy use nor by CO² emissions, whereas in ORC, CO² emissions would stimulate the investment. This is not related to the implementation of environmental policies to combat CO² emissions from oil extraction and refining activities, as the reverse link (from lnK to lnEU) is not significant in these countries. Consequently, the positive effect of CO² emissions on investment results from the impact of the physical volume of oil extraction and refining activity on the adjustment of the production capacities of the national suppliers of the oil sector.
In Eq4, the variable lnY has a positive and significant effect on lnEu only for OPCs. It is coherent with the fact that they must use energy for agriculture and industry, whereas the service sector is weakly developed. Nevertheless, for ORCs, most of their GDP comes from oil sales, whose prices and volumes are highly volatile and critical to global markets. Consequently, GDP fluctuations in ORC are not strongly related to the national level of energy consumption. It is noticed that the effect of final consumption is positive and significant on energy use in ORC. However, this effect becomes negative and significant in OPC. This is probably explained by the higher per capita income in ORC than that in OPC. According to Zhang et al. (2015, p. 881), "the higher income is associated with the consumption baskets that are more intensive in energy (transportation, housing, food, …)" (see also Lenglart et al. (2010). The consequence would be that in ORC, the link between consumption and energy use would be stronger and more positive than in OPC. The effect of the variable lnK remains negligible on lnEu in ORC, but it is negative and significant in OPC. In OPC, investment significantly decreases energy use, which is consistent with the idea of the energy saving investment and the embodied technological change in the countries characterized by a low level of technology 6 and high labor intensity (Henry et al., 1988). As shown by Muhammad (2019), a positive impact of CO² emissions on energy use is observed in the MENA countries 7 ; it is stronger in OPC than in ORC. This may be accounted for the fact that in ORC, the variability of CO² emissions is partly due to the domestic oil consumption and partly to the process of flaring undertaken during the oil extraction, which is driven by the foreign oil demand rather than the domestic energy use. Consequently, the relationship between CO² emissions and the domestic energy use is weaker in ORC than in OPC, where domestic energy use fluctuations are mainly determined by domestic quantities of the oil uses causing CO² emissions. 6 The level of technology is defined by the World Bank as a Research and Development Expenditure (% of GDP  . 1565), who show that "the EKC usually holds for most of the developed countries but usually not for developing economy". In OPC, however, the impact of GDP on CO² emissions is not significant; the main incentives of CO² emissions are the level of Eu (.849) (whose impact is as twice as larger than those in ORC), final consumption (.136) and then investment (.073). In OPC, a small but positive and significant impact of final consumption and investment on CO² emissions is also observed. It is possibly due to the growth of a demand inclined towards goods with higher CO² emissions or energy use ratios. In OPCs with initial poor infrastructures and rapid demographic growth, such as Tunisia and Morocco, a large part of the demand in goods investment is made out of the new infrastructures and residential, commercial, and industrial buildings. Building activities lead to higher cement demand and production, which is a major source of CO² emissions that are not only due to energy use but also to the chemical process involved in the production of cement (Andrew 2019, p. 1675). In addition to cement, other industrial activities generate non-fossil fuel CO² emissions: mineral products, chemical products, and metal products. In 2016, their contribution to CO² emissions in China was, for example, 5% (Cui et al., 2019, p.1). The non-energy use of oil is also a source of the CO² that is not directly related to the energy use such as the production of petrochemical feedstocks, lubricants, solvents, and bitumen (

CONCLUSION
Internationally, energy availability is one of the major drivers of growth. However, in the last century, most of the energy used in the world came from non-renewable sources, which in turn resulted in huge CO² emissions. Consequently, several research papers have recently tried to study the relationships between economic growth, final consumption, investment, energy use and CO² emissions at the international, regional and local levels. However, few have explicitly made comparisons between ORCs and OPCs. The value of the comparison lies in its relevance from an environmental policy perspective. In these two groups of countries, very different patterns of interconnections between economic variables and energy and CO² variables are observed. Panel FGLS estimations indicate that in all the MENA countries under investigation, the link between energy use and CO² emissions is bi-directional and strongly positive. However, it is nearly as twice as higher in OPC than in ORC. Accordingly, environmental policies aimed at loosening the link between energy use and CO² emissions should have a stronger environmental impact on OPC than on ORC. This is taking the form of programs of substitution of fossil energies by renewable energies and the promotion of more efficient methods of the energetic conversion and CO² capture.
It is also observed that in ORC, GDP is not associated with the level of energy use. This is strongly and positively related to CO² emissions. Besides, CO² emissions in ORC would also have a positive correlation with GDP, which is explained by the huge amounts of CO² generated through oil extraction and refinery with no connection to the energy use. This results in a low level of incentive to fight CO² emissions and may potentially have adverse consequences for the economic growth if the development and the implementation of the greener processes result in higher production costs. Therefore, specific CO² policies will have to be developed for ORC with the aim to weaken the link between CO² emissions and GDP. It takes the form of a new technology promotion in the oil industry but with a macro-economic diversification (IMF, 2003). The second option may be the best one in the long perspective, since it increases macroeconomic resilience in case of any cropping-up of oil crisis and future depletion of the reserves.
Regarding the OPC, a bi-directional relationship between GDP and the level of energy use is observed, whereas CO² emissions show a tendency to reduce GDP (the reverse link is negative but not significant). The negative link between CO² emissions and economic growth creates a positive incentive in favor of the CO² emission reduction programs through the promotion of renewable energy and the switch to technologies with higher energetic return (Johnstone et al., 2008, p. 19). In these countries, investment and final consumption have a negative impact on energy use. This is clearly the result of a shift in consumer and investment behaviors in favor of new energy saving goods and technologies. Despite that negative impact on energy use, a slight positive impact of final consumption and investment on CO² emissions is also observed. This could have been the result of evolution in demand that favors activities with higher levels of non-energy use of oil and/or higher non-oil CO² emissions. Unfortunately, this study does not offer evidence for that hypothesis.

TESTING FOR FIXED EFFECTS (F-TEST)
The hypotheses to be tested are: Here, the p-value is small enough (at ˂ 0.01 level) to reject the null hypothesis (H0) in five equations. So there is a significant fixed effect (FE), and the FE model is thus preferred than a Pooled OLS model.

TESTING FOR RANDOM EFFECTS (BREUSCH-PAGAN LM TEST)
The hypotheses to be tested are: Here, the p-value is big enough (at ˃ 0.05 level) to accept the null hypothesis (H0) in five equations. Thus, the random effect is not significant, and the Pooled OLS model is thus preferred to the random effect.
After testing for individual specific effects (fixed effect and random effect), it can be deduced that the regression model should be an individual FE model verified by the Hausman test.

TESTING BETWEEN FE AND RE (HAUSMAN TEST)
The hypotheses to be tested are: Here, the p-value is big enough (at ˂ 0.05 level) to reject the null hypothesis (H0) in five equations. Therefore, the effects are fixed and the regression model should be an individual FE.

TESTING FOR INTRA AND INTER INDIVIDUAL HETEROSCEDASTICITY (BREUSCH-PAGAN TEST AND MODIFIED WALD TEST)
• Intra-individual heteroscedasticity (Breusch-Pagan test) The hypotheses to be tested are: Here, the p-value is small enough (at ˂ 0.01 level). This leads to a strong rejection of the null hypothesis (H0) in five equations for any confidence level. Therefore, there is a phenomenon of intra-individual heteroscedasticity.
• Inter individual heteroscedasticity (modified Wald test) The hypotheses to be tested are: Here, the overall statistic χ 2 (N) has a p = 0.0000 ˃ 5% only in Eq3. This leads to the acceptance of the null hypothesis (H0) for any confidence level. Therefore, a phenomenon of inter individual heteroscedasticity is absent in Eq3 but is present in other four equations.

TESTING FOR CROSS-SECTIONAL CORRELATION (BREUSCH-PAGAN LM TEST)
The hypotheses to be tested are: Here, the overall statistic χ 2 ((N(N -1))/2) has a p = 0.0000 ≤ 5% in five equations. This leads to rejection of the null hypothesis for any confidence level. Consequently, the errors exhibit cross-sectional correlation in five equations.

TESTING FOR AUTOCORRELATION WITHIN UNITS (WALD TEST)
The hypotheses to be tested are: H0: No first-order autocorrelation. H1: There is first-order autocorrelation.

TESTING FOR FIXED EFFECTS (F-TEST)
The hypotheses to be tested are: H0: All individual intercepts = 0 (α i = 0 in the regression model Here, the p-value is small enough (at ˂ 0.01 level) to reject the null hypothesis (H0) in five equations. So there is a significant fixed effect (FE) and the FE model is thus preferred than a Pooled OLS model.

TESTING FOR RANDOM EFFECTS (BREUSCH-PAGAN LM TEST)
The hypotheses to be tested are: H0: All individual specific variance components = 0 (α i = 0 in the regression model ( ) ' . Here, the p-value as big enough (at ˃0.05 level) to accept the null hypothesis (H0) in five equations. Thus, the random effect is not significant and the Pooled OLS model is thus preferred to the random effect.
After testing for individual specific effects (fixed effect and random effect), it can be deduced that the regression model should be an individual FE model verified by the Hausman test.

TESTING BETWEEN FE AND RE (HAUSMAN TEST)
The hypotheses to be tested are: H0: Individual specifics are random: E(α i + ε it /x it = 0). H1: Individual specifics are fixed. Here, the p-value as big enough (at ˂ 0.05 level) to reject the null hypothesis (H0) in five equations. Therefore, the effects are fixed and the regression model should be an individual FE.

TESTING FOR INTRA AND INTER INDIVIDUAL HETEROSCEDASTICITY (BREUSCH-PAGAN TEST AND MODIFIED WALD TEST)
• Intra-individual heteroscedasticity (Breusch-Pagan test) The hypotheses to be tested are: H1: The variance of the error changes over time. Here, the p-value is small enough (at ˂ 0.01 level). This leads to the strong rejection of the null hypothesis (H0) in five equations for any confidence level. Therefore, a phenomenon of intra-individual heteroscedasticity is present.
• Inter individual heteroscedasticity (modified Wald test) The hypotheses to be tested are: Here, the overall statistic χ 2 (N) has a p = 0.0000 ≤ 5% in Eq1, Eq2, Eq3 and Eq4. This leads rejection of the null hypothesis (H0) for any confidence level in these equations. Therefore, a phenomenon of inter individual heteroscedasticity is present in the four first equations but absent in Eq5.

TESTING FOR CROSS-SECTIONAL CORRELATION (BREUSCH-PAGAN LM TEST)
The hypotheses to be tested are: H0: The error terms are not correlated across entities. H1: The error terms are correlated across entities. Here, the overall statistic χ 2 ((N(N -1))/2) has a p = 0.0000 ≤ 5% in five equations. This leads rejecting the null hypothesis for any confidence level. Consequently, the errors exhibit cross-sectional correlation in five equations.

TESTING FOR AUTOCORRELATION WITHIN UNITS (WALD TEST)
The hypotheses to be tested are: H0: No first-order autocorrelation. H1: There is first-order autocorrelation.