*3.4. Data*

Data are used for both model 1 (on the environmental goal) as presented in Equation (1) and model 2 (on the environmental goal) as presented in Equation (2). Data are collected for the period from 1980 to 2017 for 28 developed countries in the Organization for Economic Co-operation and Development (OECD), which include the following countries: Australia, Austria, Belgium, Canada, Chile, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Japan, Korea, Rep., Latvia, Lithuania, Luxembourg, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, Turkey, United Kingdom, and the United States.

The description of all variables in both model 1 and model 2 are presented in Table 1.



**Table 1.** *Cont.*

\* CO2 emissions are affected by burning data from fossil fuels, soil, and cement equipment, collected by the Carbon Dioxide Information Analysis Center (CD CDIAC). The center collected global carbon dioxide emissions between 1950 and 1982, estimated by Marland and Rotty [56] from fuel production data from the UN's Energy Statistics Yearbook [57]. We consider that the main reason for the use of fuel production data is due to a higher level of reliability in comparison with fuel consumption data at the global level. This choice of using fuel production data is widely utilized in empirical analyses. Moreover, doing so will also avoid creating an accounting identity in Equation (1). We consider that when energy consumption data is used, the total of estimated coefficients for GDP, GDP<sup>2</sup> in Equation (1) is equal to zero. \*\* Collected from governmen<sup>t</sup> sources and published data, including data from the Energy Research of the Institute of Geosciences and Natural Resources which is available in BP Statistical Review [58].

Table 2 presents the descriptive statistics of the variables utilized in our two models.


### **Table 2.** Descriptive statistics.

\* The unit production data is a million tonnes of oil equivalent (Mtoe). \*\* Renewable energy includes biomass, geothermal, solar, wind, and other renewable sources.

### **4. Empirical Results**

### *4.1. Test of Presence of Cross-Sectional Dependence*

The first step in the regression technique is to test for the presence of cross-sectional dependence, and the results from this test affect all the techniques used in the subsequent steps. As a result, to obtain strong test results, we use three simultaneous tests: Pesaran [59], Friedman [60], and Frees [61]. Although the three tests have their advantages and disadvantages, they also provide an overview of the robustness of the results. Moreover, two of the specifications (fixed effect and random effect) are employed in the three tests in both models to reveal the change in the test results.

If the null hypothesis is accepted, there is no cross-sectional dependence, and the appropriate unit root test for all data is the Pesaran test [62]—the second generation of panel unit root tests, and the long-run estimator methods are pooled using FMOLS and DOLS [18]. In contrast, all the results in Table 3, including six tests for each model, strongly reject the null hypothesis at the one percent and five percent significance levels. That means that all data samples have cross-sectional dependence or are sample country specific. This leads to a change in the test used in the following steps. In this case, the unit root test used should be the one by Im, Pesaran, and Shin (IPS test; [63]), which is expanded in the Choi test for cross-sectional dependence. Furthermore, mean group regressions, such as main mean group analysis, including fully modified ordinary least squares (FMOLS), dynamic ordinary least squares (DOLS) and the other mean group analysis methods, including Mean Group (MG), Common Correlated Effects Mean Group (CCEMG) and Augmented Mean Group (AMG), should be used to determine the long-run effects [64].


**Table 3.** Sectional independence tests.

Notes: Fixed effects (FE) and random effects (RE) models. \*\*\* and \*\* indicate statistical significance at the one and five percent level, respectively.

### *4.2. Panel Unit Root Tests*

A unit root test is conducted to determine the stationarity and the integration of the same order for variables used in the paper [18]. This test is required before the cointegration tests are conducted to examine the long-run nexus between CO2 emissions and each source of energy (model 1) and economic growth and each source of energy (model 2). The results of the unit root tests and robustness checks are presented in Table 4 below. The robustness checks from all four tests are presented for all variables, with the constant and the trend and constant shown in both the level and first difference forms.


**Table 4.** Unit root test using Pesaran test.

Standard errors in parentheses.

### *4.3. Panel Cointegration Test Results*

Cointegration tests, including those by Kao [65], Pedroni [66], and Westerlund [67], are employed after confirming the stationarity at the same order I(1) in Table 4. This step helps to avoid spurious results [18], and we conduct these three tests at the same time to obtain robust results. All the results are in Tables 5 and 6.


Notes: \*\*\*, \*\*, and \* show the rejection of the null hypothesis of no cointegration is statistically significant at the 1, 5, and 10 percent levels, respectively.


**Table 6.** Cointegration tests for model 2 (Equation (2)).

Notes: \*\*\* and \*\* show the rejection of the null hypothesis of no cointegration is statistically significant at the 1, 5, percent levels, respectively.

In Table 5, the result is highly statistically significant at one percent in both the Kao and Pedroni tests and at five percent in the Westerlund test using model 1, a model of the environment. We conclude that some or all panels show cointegration between variables. In other words, there may be at least one long-run nexus between the variables in model 1. Therefore, the use of Panel Vector Auto Regression (P-VAR), which is used for evaluating the short-run effect, is not considered in this research.

Similarly, the growth model for the environment has the same results in the cointegration test. All the results, which are in Table 6, reject the null hypothesis at a highly significant level (one percent). This critical step ensures that at least one variable in model 2 has a long-run relationship with the dependent variable (economic growth rate). Overall methods, both models 1 and 2 show robustness in their test results, which predict high reliability in conclusion to this study.

### *4.4. Regression and Ranking Results*

We consider that ordinary least squares (OLS) regression is inappropriate, leading to a biasedness in estimating the long-run equilibrium relationship. In this paper, we apply fully modified ordinary least squares (FMOLS) in order to take the endogeneity problems, as well as the serial correlation issues, into account [68,69]. In addition, dynamic ordinary least squares (DOLS) is also employed, as this DOLS technique can also eliminate endogeneity problems and serial correlation issues using contemporaneous values, leads, and lags in the first difference. Due to the greater use of assumptions and the reduction in the degrees of freedom by using leads and lags [70,71], FMOLS is the preferred model in this study. However, we use the result of the DOLS model to confirm the direction of the estimated coefficients.

### 4.4.1. Model 1: Environmental Goal

We employ the traditional model of the environment (Kuznet [50]) with the control variables (income and square of income) and the proxy variable for energy consumption, which is divided into five main sources of energy (coal, gas, oil, hydropower, and renewable energy) that have enough data for analysis. Furthermore, in this model, the multicollinearity problem between GDP and sources of energy is very clear in the variables in model 2 (the impact of energy use on the growth rate). To analyze the effect of the multicollinearity problem on the coefficient of the variables of concern, we use both models (FMOLS and DOLS) and other mean group models with and without control variables (GDP and GDP2). The regression results are shown in Tables 7 and 8.


**Table 7.** Regression results for model 1 (dependent variable: lnCO2).

Standard errors in parentheses, \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.1.



Standard errors in parentheses, \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.1. All variables are transferred into logarithmic form. Note: A rank of one denotes the least environmental harm, and a rank of five denotes the most environmental harm.

This step aims to determine the impact of energy sources on environmental degradation before we do any further rankings. The main concern is how the coefficient or final rankings of these sources change across multicollinearity problems and multiple specifications. If there is no difference or the deviation in regressors between specifications is small, or the statistical significance is still high, the final rank based on these coefficients is the most reliable.

With these concerns in mind, the first considerations in the results in both Tables 7 and 8 are the sign and magnitude of the estimated coefficients for all five energies (coal, gas, oil, hydropower, and renewable energy). In the main models, FMOLS and DOLS with and without a multicollinearity check, the coefficient of coal consumption is from +0.253 to +0.273. This deviation is quite low and highly significant (one percent). This coefficient means that when coal consumption increases/decreases by one percent, on average, CO2 emissions increase/decrease by 0.273 percent (FMOLS model), ceteris paribus. The coefficient for gas consumption is from +0.152 to +0.19, and all regressors are also highly statistically significant (at one percent). The economic meaning is similar to that for coal

consumption, in that a change in the use of gas of one percent leads to a 0.17 percent change in CO2 emissions (in the same direction). Oil and hydropower consumption have high statistical significance as well, but oil use has the biggest effect on CO2 emissions (a one percent increase/reduce in oil use, on average, leads to a 0.53 percent increase/decrease in CO2 emissions). In contrast, the effect of renewable energy is unclear and has a weak significance. These results indicate no evidence that the use of renewables leads to environmental degradation.

Based on the magnitude and signs of the regressors, the following steps are used to rank the five energy sources: (1) a negative impact on CO2 emissions or a negative coefficient (statistically significant); (2) no impact (zero coefficient with statistically significant) or the coefficient is statistically insignificant; (3) positive coefficient (statistically significant). The final rank is as follows: hydropower, renewable energy, gas, coal and oil in the order of the least environmental harm to the most environmental harm, as indicated in Table 8.

### 4.4.2. Model 2: Economic Goal

This step considers the impact of each energy source on economic growth in the long run by applying Barro's growth model. In accordance with the approach adopted for the environmental model (Equation (1)), this analysis for the economic growth model (Equation (2)) also uses the five mean group methods, FMOLS, DOLS, MG, CCEMG and AMG, to regress the effects. The results are presented in Table 9.



Standard errors in parentheses, \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.1. Note: A rank of one denotes the most contribution to economic growth, and a rank of five denotes the least contribution to economic growth.

#### *4.5. A Combination of These Two Criteria Using the Weighted Scoring Method*

We then use the weighted scoring method (WSM) to obtain a score for the five sources, in which the order of these scores shows their final contribution to the balanced structure of energy sources (including coal, gas, oil, hydropower, and renewable energy). The most preferred source of energy is the energy source with the highest score. In this method, the weight of each criterion plays an important role in the score and directly affects the final ranking. Thus, to obtain an overview of the final rank and its role in achieving both environmental and economic goals, we analyze changes in the final rank across the two scenarios step by step with sensitivity analyses. Using the weights of each criterion range from 70% to 80%, Table 10 shows how the WSM method works and the influence of this weight on the final structure.


**Table 10.** Final ranking for each of the energy sources.

\* Hydro stands for hydropower, and Renew stands for other types of renewable energy, including wind, geothermal, solar, biomass, and waste. \*\* With the environmental scenario, the weighting of the environmental goal ranges from 70% to 80% and the remaining proportion of 20–30% applies to the economic goal. The same approach applies to the economic scenario, with a weight of 70–80%, leaving 20–30% for the environmental goal. \*\*\* The final ranking takes into account the rankings from estimated coefficients from models 1 and 2, together with the assumed weighting range. Robustness analyses are presented in the Appendix A of this paper. Note: A rank of one denotes the most contribution to economic growth, and a rank of five denotes the least contribution to economic growth. A rank of one denotes the least environmental harm, and a rank of five denotes the most environmental harm.

In the environmental goal scenario, we assume that a governmen<sup>t</sup> prioritizes the environmental goal rather than the economic goal. We assign scores for each energy source using the WSM, which increases the weight of the environmental goal from 70 percent to 80 percent, and the remaining proportion for the economic goal is from 30 percent to 20 percent. This selected range demonstrates the overwhelming priority of one goal, being the environmental goal, which in this case might create a "crowding-out effect" on the other goal, being the economic goal. We also conduct sensitivity analyses, which are included in Appendix A, Table A1.

Table 10 shows that, with the priority of the environmental goal, the top ranking belongs to clean energy such as hydropower (ranked 1) and renewable (ranked 2) and fossil sources including gas, oil and coal. These findings have important implications for countries who make the environmental goal a policy priority. These countries should focus on policies that can encourage the use of clean energies. This scenario may be relevant for developed countries who have achieved a certain level of economic development, as these countries are not completely reliant on fossil fuels. For instance, many countries such as Germany, France, and Britain have set targets to ban the sale of petrol and diesel vehicles in the future [72,73]. Similarly, many cities around the world have started to convert public transportation to electric vehicles, and have banned or put taxes on diesel vehicles coming into their cities, such as Paris, Athens, Mexico City, Madrid and London [74].

The economic scenario uses the same method to analyze the role of the economic goal in the final ranking. We assign a weight from 70 percent to 80 percent to prioritize the economic goal. We also conduct sensitivity analyses using various weights for the economic scenario, which are included in Appendix A, Table A2.

When the priority is on the economic goal, a major change in ranking is observed from the first scenario (with a focus on the environmental goal) to the second scenario (with a focus on the economic goal). Table 10 shows that fossil fuels are ranked first (oil is ranked first, and gas is ranked second), and then clean energy follows (hydro and renewable energy). Making the economic goal a priority may be relevant in practice for underdeveloped countries and some developing countries. Currently, at a low economic growth rate, these countries are willing to trade o ff environmental degradation to attract more foreign investment in order to boost the economy [18]. It is argued that the widespread use of fossil fuel-based energy in these countries, such as oil and gas, in the process of industrialization and modernization, will lead to significant economic growth.

### **5. Conclusions and Policy Implications**

This paper aimed to determine a balanced energy structure, in the long run, using data from the OECD countries for the period from 1980 to 2017 [75]. In this paper, five energy sources were considered including coal, gas, oil, hydropower, and renewable energy. The proposed optimal energy mix was developed with the view of achieving two fundamental goals at the same time: (i) to minimize environmental degradation; and (ii) to support economic growth.

In this paper, the weighted scoring method (WSM), the most popular method of the multi-criteria decision-making (MCDM) techniques, was used to combine the rankings using five energy sources and two goals. Various tests, including the cross-sectional test, the stationarity test, and panel cointegration test, were conducted in this paper. Furthermore, this paper employed mean group regressions to consider the long-run e ffect of the estimates. These mean group techniques included two groups: (i) the main mean group analysis, including fully modified ordinary least squares (FMOLS) and dynamic ordinary least squares (DOLS); the other mean group analysis, including Mean Group (MG), Common Correlated E ffects Mean Group (CCEMG) and Augmented Mean Group (AMG). These techniques were used to determine the long-run e ffects between the variables utilized in the paper. Sensitivity analyses were also conducted to ensure the robustness of the findings.

Our empirical findings indicate that, in the long term, in achieving both the environmental goal and economic goals, the OECD countries may consider adopting a balanced energy mix in which the following structure, associated with preferences for each source of energy, is considered: (i) hydropower, (ii) renewables, and (iii) fossil fuels (oil, then gas, and then coal). However, we are aware that determining an optimal energy structure is not a solid scientific process because the decision on optimal energy mix heavily depends on various factors, including internal and external factors. Some of these factors may be well beyond the control of the governments of the OECD countries. For example, in designing an optimal energy structure, a ffordability is very important. A ffordability represents the financial capacity the general public can pay to use energy. An energy structure is not optimal if the general public is unable to pay for its energy consumption. In addition, security is also a very important aspect of any optimal energy mix because the economy and society cannot be without energy. Last but not least, sustainability in economic growth and development, together with sustainability in energy consumption, are equally important compared to any other aspects. Designing an optimal energy structure is not only for current generations, but also for the many generations to come. As a consequence, we are aware of and agree with the view that designing and implementing an optimal energy structure is an extremely complicated issue. In addition, there may not be a one-size-fits-all approach because each country will face di fferent challenges in the process of designing an optimal energy policy. The members of the OECD are mainly advanced countries, and they may share similarities in terms of their economic growth and development progress, social inclusion and culture. However, this does not mean that one policy for an optimal energy structure can be developed and applied to all members. We also consider that there may not be an optimal energy structure for any nation because energy policy has been moving and changing very quickly, particularly due to the current progress of technology. An optimal energy structure for a country today may no longer be optimal in the very near future as technology can change at the pace of days or months.

Based on the above observations, we consider that the findings of this paper should be considered as an additional piece of empirical evidence for the governments of the OECD countries to take into account, alongside all other pieces of evidence currently available within their constraints and contexts. As a result, based on the findings of this paper, the policy implications can be summarized as follows. When the environmental goal is prioritized, the optimal energy structure will start with clean energy sources, including hydropower and renewable energy. Fossil fuel energy will follow, including oil, gas and then coal. This scenario appears to be relatively consistent with the current environment for most of the developed countries in the OECD. On the other hand, in our economic scenario, in which the economic growth goal is prioritized, the important role of fossil fuel in boosting the economy is observed. This scenario confirms the view that it is difficult to replace fossil fuels with cleaner sources of energy when the first priority is to achieve economic goals. This scenario reflects the reality of the developing and emerging markets in the process of industrialization and modernization.

**Author Contributions:** Conceptualization, A.H.T. and D.H.V.; methodology, D.H.V.; software, A.H.T.; validation, A.H.T. and D.H.V.; formal analysis, A.H.T.; investigation, A.H.T.; resources, D.H.V.; data curation, A.H.T.; writing—original draft preparation, A.H.T. and D.H.V.; writing—review and editing, A.H.T. and D.H.V.; visualization, A.H.T.; supervision, D.H.V.; project administration, A.H.T. and D.H.V.; funding acquisition, D.H.V. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was funded by the Ministry of Education and Training of Vietnam under gran<sup>t</sup> B2020-MBS-03.

**Acknowledgments:** The authors acknowledge constructive comments from the Editor of the journal and three reviewers. With their expert views on the issue, three reviewers provided us with very helpful, insightful and practical comments on this important topic. We also greatly appreciate the comments and suggestions from participants at Vietnam's Business and Economics Research Conference (VBER2019), July 2019, Ho Chi Minh City, Vietnam.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.
