The Impact of Renewable Energy Consumption and Economic Growth on CO 2 Emissions: New Evidence using Panel ARDL Study of Selected Countries

Most countries consume more non-renewable energy to generate economic activities. Hence, economic growth plays a vital role in contributing to higher CO 2 emissions. Therefore, this type of energy has reduced and replaced by renewable energy. Renewable energy is said not to be detrimental to the environment. Consequently, it is imperative to examine the effects of renewable energy consumption and economic growth on CO 2 emissions in selected countries by per capita income. Using a sample of high-income, upper-middle-income, and lower middle -income, and low-income countries for the period of 1990-2017, and the estimation method of the panel ARDL, the main results show that in the long run, overall renewable energy consumption can reduce CO 2 emissions. However, economic growth and population growth can result in higher CO 2 emissions in the long term. In the short run, the results show that higher overall economic growth can contribute to higher CO 2 emissions. Contrarily, higher population growth, and renewable energy consumption can help reduce CO 2 emissions in the short run.


INTRODUCTION
Energy consumption has increased to boost economic growth. Bildieci (2014), in his study, found that economic growth has intimately connected with energy consumption. This finding suggests that depletion in energy sources can serve a stumbling block for economic development. The importance of energy in generating economic activities is on par with the stress of labor and capital. Hence, the demand for all energy sources such as coal, gas, and oil in all economics sectors exhibits a steady rise every year. According to the International Energy Agency (2018), the transportation and industrial sectors in Malaysia consume the largest share of total energy. Silva (2018) stated that higher economic activities require more energy. In the absence of energy, economic development has disrupted, and high unemployment ensues.
Despite the importance of energy in economic growth, it can trigger harmful effects on the environment. Dogan and Seker (2016) gave credence to the fact that non-renewable energy, such as oil and coal can increase CO 2 emissions. Environmental degradation stems from CO 2 emission released in the aftermath of inexorable energy consumption. Harmful gas emissions are released as a result of combustion. The release of CO 2 emissions is inevitable in power generation. As a result, humans have to suffer several environmental problems such as haze, acid rain, and higher global temperature. This Journal is licensed under a Creative Commons Attribution 4.0 International License This will be complex when a country consumes non-renewable energy in every sector. This suggests that greater economic activities can culminate in environmental degradation. Hence, Saidi and Mbarek (2016) found that an increase in economic growth can lead to higher CO 2 emissions. The economic sectors, such as industries and transportation that are dependent on energy, can harm the environment. The transportation and industrial sectors contribute most to CO 2 emissions as these sectors consume the largest share of total energy in Malaysia. Therefore, non-renewable energy must be replaced by renewable energy to mitigate environmental effects. For example, biofuel is one of the alternative energy sources which are not detrimental to the environment.  supported that the use of renewable energy can reduce carbon emissions. Renewable energy is the largest contributor to the reduction of CO 2 emissions . A vast array of literature proposed that renewable energy as the best alternative energy to ensure that economic growth can be boosted and the environment can be preserved. This is because their findings showed that renewable energy consumption could reduce CO 2 emissions. For example, Bhattacharya et al. (2017) and  stated that renewable energy should be used to promote economic growth and reduce CO 2 emissions. This paper contributes to policymakers and the existing literature in the following ways. First, this study employs the panel ARDL method to examine the impacts of renewable energy consumption and economic growth on CO 2 emissions in various countries. This method has not been used by previous studies on renewable energy consumption. Furthermore, none of the previous studies delved into the effects of renewable energy consumption and economic growth on CO 2 emissions in comparison among different income groups. Hence, this current study splits into four groups according to per capita GNI: high-income, upper-middle-income, lower-middle-income, and low-income groups. This study can determine whether economic growth in countries with high income contributes more CO 2 emissions and whether the use of renewable energy in those countries can reduce CO 2 emissions. We can also compare these findings with the findings in upper-middle-income, lower-middle-income, and low-income countries. The amount of CO 2 emissions in high-income countries is larger than that in lowincome countries. Figure 1 shows the total CO 2 emitted by selected countries by per capita GNI in 2015. The United States was the largest emitter of CO 2 with a total of 5,225,394 kilotons, followed by Iran (623,255 kilotons), Saudi Arabia (608,804 kilotons), and Canada (589,780 kilotons). Comoros was the smallest producer of CO 2 with a total of 189 kilotons, followed by Uganda (4,693 kilotons), and Tajikistan (5,383). The marked difference between CO 2 in the high-income country (the United States) and lowincome country (Comoros) stood at 5,225,205 kilotons. Based on the figure, it shows that high-income countries produced more CO 2 than low-income countries created. These findings suggest that as income increases, pollution goes up simultaneously and vice versa.
Given this backdrop, this article attempts to examine the impact of renewable energy consumption and economic growth on CO 2 emissions in selected countries by per capita GNI. The writing of this article has segregated into five sections. The second section reviews past studies, followed by methodology and model specification in the third section. The fourth section deals with study results; and finally, the conclusion in the fifth section.

LITERATURE REVIEW
Numerous studies investigated the effects of energy consumption and economic growth on CO 2 emissions without specifying whether non-renewable or renewable energy such as Alam et al. (2015), Aiyetan and Olomola (2017) as well as Stamatiou and Dritsakis (2019). However, their findings are mixed. For example, Islam and Ghani et al. (2017) tested their hypotheses by using panel co-integration and panel granger causality, and the results showed that there are positive relationships among energy consumption, economic growth, and CO 2 emissions. Aiyetan and Olomola (2017), on the other hand, found that economic growth plays an essential role in determining CO 2 emissions in the long run but not in the short run. Apart from that, energy consumption is a positive determinant of CO 2 emissions in the long run and short run in Nigeria. In Italy, the findings are slightly different, as Stamatiou and Dritsakis (2019) discovered that economic growth could increase CO 2 emissions. Other than that, a decrease in energy consumption does reduce not only CO 2 emissions but also economic growth. Several previous studies are focusing on the effect of renewable energy and economic growth on CO 2 emissions in various countries (Jebli and Youssef 2017;Dogan and Seker 2016;Bhattacharya et al. 2017;Sinha and Shahbaz 2018;Cherni and Jouini, 2017;Paramati et al. 2017;Bekhet et al. 2018;Chen and Geng 2017;Dong et al. 2017;Waheed et al. 2018;Zaidi 2017;Chen et al. 2018;Ozcan and Ozturk, 2019). However, their results are also not consistent with each other.
According to Jebli and Youssef (2017), in the long run, an increase in renewable energy consumption and economic growth can result in higher CO 2 emissions. The study employed a panel data analysis, namely FMOLS, DOLS, and OLS, focusing on the North African country to examine the effects of renewable energy consumption and economic growth on CO 2 emissions from 1980 to 2011. The FMOLS approach has employed by Paramati et al. (2017) (2017), Chen et al. (2018), and Sinha and Shahbaz (2018) employed the ARDL approach to examine the effects of renewable energy consumption and economic growth on CO 2 emissions, but their findings were not consistent. Cherni and Jouini (2017) discovered that economic growth contributes to CO 2 emissions, but renewable energy consumption does not contribute to CO 2 emissions in Tunisia and Turkey, respectively. Contrarily, Sinha and Shahbaz (2018) found that renewable energy consumption can reduce CO 2 emissions in India from 1971 to 2015. The results also showed that economic growth increases CO 2 emissions in the early stages, and then it can reduce CO 2 emissions in the final stages. Thus, the findings supported the Environmental Kuznets Curve (EKC).  also provided consistent results that the ECK exists, and renewable energy consumption plays a vital role in reducing CO 2 emissions. The study also employed the same method but was conducted in China to analyze data from 1965 to 2016. Chen et al. (2018) supported that renewable energy consumption can influence CO 2 emissions negatively from 1980 to 2014 in China.
However, Dogan and Seker (2016) argued that renewable energy consumption did not reduce CO 2 emissions in the European Union using the OLS approach. The findings gave credence to the results found by Paramati et al. (2017) that renewable energy consumption could reduce CO 2 emissions in Turkey. Besides, FDI and stock market growth play an essential role in reducing CO 2 emissions.
Most previous studies focused on renewable energy consumption on CO 2 emissions in a single country by using time series data analysis. Their findings are still mixed and still baffling why renewable energy consumption can lead or not lead to CO 2 emissions. A limited number of previous studies such as Zoundi (2016) investigated using a panel data analysis. However, the research just focused on developing countries in Africa. The study did not compare among high-income, upper-middle-income, lower-middle-income, and low-income countries. Therefore, this study provides a better understanding by comparing the effects of renewable energy consumption and economic growth in different countries by per capita GNI.

METHODOLOGY
This study employs the panel ARDL approach to investigate the effects of renewable energy consumption and economic growth on CO 2 emissions in selected countries by per capita GNI.
Where CO 2 represents CO 2 emissions (kilotons), GDP represents a real gross domestic product (GDP) (LCU), POP is population, and RE is renewable energy consumption (% of total energy consumption). All of the variables have transformed into their logarithms.

Panel Unit Root Test
Panel data analysis needs to test for stationarity issues, and hence unit root tests are conducted. The tests are essential to examine the presence of stationarity for panel data, as suggested by Levin-Lin-Chu (2002) (LLC) and Im-Pesaran-Shin (2003) (IPS). All the data set are tested to determine the integration order of I (0) or I (1).
The tests have conducted to check whether our variables are not stationary at level, but they must be stationary at first difference. Furthermore, Levin et al. (2002) found that the use of panel unit root tests is more efficient than time series unit root. There are two-panel unit root test methods used in this study, namely Levin et al. (2002), Breitung (2000, and Im et al. (1997). However, the IPS unit root test is more critical than the LLC unit root test due to its appropriateness for regression of heterogeneity unit root.

Panel Estimation
Pooled Mean Group (PMG) estimation or panel ARDL model has the advantage of determining dynamic long-run and shortrun relationships. The PMG estimator can estimate relationships in the short run, including the coefficients and the adjustment for long-run equilibrium (speed of adjustment) and error variance to be heterogeneous. The long-run coefficients are restricted to be homogenous across countries. The use of this method is appropriate as it is more efficient and consistent with the existence of long-run relationships. The second method of estimation is Mean Group (MG). According to Pesaran and Smith (1995), it has less restrictive procedures that can estimate the diversity of parameters. It can also estimate different coefficients for each country. Both of the MG and PMG estimators require the selection of appropriate lag lengths using the Schwarz Bayesian Criterion (SBC) or Akaike Information Criterion (AIC). The MG estimator provides consistent long-run mean estimation, although it is inefficient with homogeneity. In the presence of long-run homogeneity, pooled estimators are consistent and efficient. The third estimator is the Dynamic Fixed Effect (DFE). This estimator is the same as the PMG estimator. It can limit the co-integration vector coefficient to have consistency for all long-run panels. Apart from that, it also limits the time adjustment coefficient and produces consistent short-run estimation. DFE limits the coefficients of integration vectors for all panels. All the estimators (PMG, MG and DFE) can show the long-run and short-run effects of each variable. According to Pesaran and Shin (1999), these approaches are more consistent in generating long-run coefficients regardless of whether the order of integration is I (0) or I (1). This method uses the combination of time series and cross-section data with T larger than N. According to Shin (1999, 2001), the most appropriate number of N is about 20-30 countries. Next, the Hausman tests have used to determine which one is better in this study: PMG, MG or DFE.
Where, i represents the number of countries (1, 2, 3...,20), t is the number of years (1990-2017), (p,q,r,s) is the optimum time lag, α i is the countries specific effect, and ε it refer to the remainder error terms. The short-run relationship with an error correction model is as follows: , Where λ i are long-run parameters, and ϕ i is the parameter for the error-correction term that measures the speed of adjustment to the long-term equilibrium of lnCO 2 due to changes in lnGDP, lnPOP, and lnRE. ϕ i indicates the existence of a long-run relationship. Thus, a negative and significant value of ϕ i shows the existence of a co-integrating relationship among lnCO2, lnGDP, lnPOP, and lnRE. All ECM dynamics and terms can freely change. Besides, the parameter estimation for this model is consistent and asymptotically normal to estimate long-run coefficients for both stationary and non-stationary regressors I (1

Hausman Test
Hausman test has used to choose the preferred estimator between the PMG or MG estimator and either the PMG or DFE estimator. According to Pirotte (1999), the MG estimator allows parameters to be independent across groups and do not take into account the heterogeneity between groups. However, Pesaran and Shin (1999) argued that the PMG is better because it gives coefficients of different short-run variance by country. In contrast, for longterm coefficients, it is assumed that all countries are homogeneous (similar). In contrast, for the MG estimator, it allows only shortand long-term coefficients heterogeneous (different) length of time between countries. The choice between PMG or MG estimators depends on the null hypothesis testing. If the null hypothesis is accepted, then the PMG estimator is selected because it is more efficient than the MG estimator. If the null hypothesis has rejected, then the MG estimator is chosen over the PMG estimator. Next, to choose either the PMG or DFE estimators, if the null hypothesis is accepted, the PMG estimator is better than the DFE estimator.

EMPIRICAL RESULTS
This study employs the panel ARDL technique to examine the effects of renewable energy consumption and economic growth on CO 2 emissions in selected countries by per capita GNI. The countries have divided into four groups: high-income, upper-middle-income, lower-middle-income, and low-income groups. The LLC and IPS unit root tests have conducted, and the results are summarized in Table 1. The LLC results show that all the variables (lnGDP, lnPOP & lnRE) except for lnCO 2 are not significant at level, suggesting that they are not stationary. Therefore, the null hypothesis is accepted. However, all the variables, including lnCO2 are significant at first difference. This findings mplies that they are all stationary, and thus the alternative hypothesis is accepted. In addition to the LLC tests, the IPS tests are also performed, and the results reveal that all the variables (lnGDP, lnPOP, lnCO2 and lnRE) are not significant at level but significant at first difference. This indicates that the variables are not stationary at level but stationary at first difference. Next, the panel ARDL technique is employed.  Table 3 shows the results of short-run estimation using PMG, MG and DFE. The values of error correct term (ECT) are negative and significant for all of the three estimators, and thus they confirm the existence of long-run relationships. The results of PMG and DFE indicate that economic growth and renewable energy consumption can influence CO 2 emissions in the short run. Higher economic growth can increase CO 2 emissions but higher renewable energy consumption can reduce CO 2 emissions. These findings are not consistent with the findings obtained from the MG estimator.
The results of the MG estimator reveal that all the variables do not have any effect on CO 2 emissions in the short run. Only the results of PMG show that population growth can lead to lower CO 2 emissions in the short run. The results of the Hausman tests suggest that PM is better than MG and DFE. Table 4 shows the results of short-run estimation in specific countries. The results are divided into four categories: highincome, upper-middle-income, lower-middle-income, and lowincome countries. In the high-income countries (Canada, the United States, Poland, Belgium and Saudi Arabia), renewable energy consumption does not affect CO 2 emissions in the short run. Population growth does not also contribute to CO 2 emissions in the high-income countries except for Saudi Arabia. In Canada and the United States, economic growth can increase CO 2 emissions in the short run. However, in Poland, Belgium, and Saudi Arabia, economic growth does not affect CO 2 emissions.
In the upper-middle-income countries (Algeria, Gabon, Iran, Malaysia and Turkey), renewable energy consumption does not

SUMMARY AND CONCLUSIONS
This study aims to examine the effects of renewable energy consumption and economic growth on CO 2 emissions in selected countries (Canada, the United States, Poland, Belgium, Saudi Arabia, Algeria, Gabon, Iran, Malaysia, Turkey, Bangladesh, Egypt, Indonesia, Nigeria, Pakistan, Benin, Comoros, Senegal, Tajikistan & Uganda) by per capita GNI. The countries have divided into four categories: high-income, upper-middle-income, lower-middle-income, and low-income countries. The panel ARDL method was employed, and the results show that in the long run, overall renewable energy consumption can reduce CO 2 emissions. However, economic growth and population growth can result in higher CO 2 emissions in the long term. In the short run, the results show that higher economic growth can contribute to higher CO 2 emissions. Contrarily, higher population growth, and renewable energy consumption can help reduce CO 2 emissions in the short run.
In all of the high-income and upper-income countries, renewable energy consumption does not play any role in reducing CO 2 emissions. This is because renewable energy in those countries accounts for a tiny percentage of total energy consumption. This means that the countries are highly dependent on non-renewable energy instead of renewable energy. However, in all of the lowermiddle-income countries and most of the low-income countries, renewable energy consumption can reduce CO 2 emissions. This is because those countries are highly dependent on renewable energy instead of non-renewable energy. For example, renewable energy contributed to 73% of total energy consumption in Bangladesh in 1991, and that was the highest percentage over the period 1990(World Bank, 2019.
Economic growth in the low-income countries does not affect environment degradation. In some high-income, upper-middleincome, and lower-middle-income countries, economic growth can contribute to environmental degradation. Population growth does not result in higher CO 2 emissions in most countries.
These findings are essential for policymakers to formulate the right policies. A shift from non-renewable energy such as oil and coal to renewable energy such as solar and biofuel is a good move to mitigate environmental degradation. Other than that, the governments can give fiscal incentives such as tax reductions to firms that use clean energy in their production. This incentive is essential to reduce CO 2 emissions.