**1. Introduction**

Energy is of widespread concern because of its effects on life, development, and the existence of current as well as future generations. Early in human history, fire was the primary energy source. Since then, we have exploited energy from various sources, such as coal, oil, hydropower, wind, solar, geothermal, and nuclear. Each source of energy has different advantages and disadvantages. In the past, fossil fuels were cheaper than renewables and had stable production. However, they caused pollution, whereas renewables were clean but limited in production. However, the selection of a source of energy depends on governmental direction, without deep and overall analysis of the economy and environment simultaneously.

Following the general trend of sustainable economic growth and development, which is generally known as green growth, the sustainable aspect of economic growth focuses on policies that can achieve economic growth not only for this generation, but also for many generations to come. The OECD countries have been formulating and implementing energy policies that are based on limiting CO2 emissions by cutting and moving towards zero oil and coal use. The energy use of the United Kingdom has been transferred dramatically from fossil to clean energies, which accounted for 52 percent of the total energy consumption in 2017 [1]. In the US, the current governmen<sup>t</sup> is still interested in fossil fuels. However, the governmen<sup>t</sup> is planning to switch to solar and wind because of its low cost and environmentally friendly attributes. However, the US economy is still heavily dependent on fossil energy, particularly coal and oil. Limiting the use of fossil fuels will reduce the amount of CO2 released into the environment, but at the same time, slow down economic growth [2].

Some countries take advantage of their natural advantages to develop and exploit renewable energy. In Sweden, renewable energy (mainly wind, nuclear and hydro) accounts for a share of more than half of the domestic demand. This current level is expected to increase further by 100 percent by 2040 [1]. The most significant problem for energy policies with a focus on renewable energy is the guarantee of energy security for the nation.

On the other hand, the problem with oil-use countries is the fluctuations in oil prices. We note that oil prices are unpredictable and uncontrollable, especially in the event of unexpected events like the COVID-19 pandemic. The impact of oil prices on the consumer price index (CPI) in these countries is very significant, requiring quick actions from the governmen<sup>t</sup> in seeking alternative energy sources or supporting the economy with stimulus packages. The need to balance sustainable economic growth and development and environmental protection should be a top priority in energy policy.

In general, energy policies are based on various factors, including internal and external factors, such as price stabilization, reducing CO2 emissions and ensuring energy security, affordability, and suitability to the economy. However, decisions are mainly based on specific information for each factor, without considering the balance of economics and the environment simultaneously. As such, we consider that this paper will provide an additional piece of empirical evidence for governments to consider when they formulate and implement energy mix policies in their countries.

The empirical papers on energy economics to date appear to focus on the investigation of a relationship among variables of interest. For example, empirical studies on the environmental Kuznets curve hypothesis (EKC), a highly cited concept in energy economics studies, generally focus on the three main streams of analyses. The first stream empirically examines the change in the traditional EKC theory. The second stream investigates the nexus between environmental quality and total energy use. The last stream of research investigates the inter-relationship between trade openness, proxied by foreign investment flows, and environmental quality.

This paper is unique and different from other empirical papers in the area of energy economics and policy implications. In this paper, we seek to find a balanced structure of energy sources that can simultaneously achieve important goals in both domains: (i) the environment and (ii) the economy. Advanced countries such as Japan, the United Kingdom and France (and many others) have been advancing towards the use of cleaner energy sources to minimize the negative impacts of energy consumption on environmental degradation. The governments of developing and emerging countries appear to prioritize economic growth and development. The debate about striking the right balance between what we call the environmental goal and the economic goal appears to have been ignored in the current literature.

The structure of this paper is as follows: Section 2 discusses selected empirical studies on energy economics to date, with a focus on the three strands of research in response to the environmental Kuznets curve hypothesis. Section 3 presents the research methodology and data. The empirical findings of this paper, including the sensitivity analyses, are included in Section 4, followed by the conclusions in Section 5 of the paper.

### **2. Literature Review**

This paper is based on two traditional theories/hypotheses, including the Kuznets environment curve from environmental studies and economic growth theory. For the first hypothesis, in the 1950s, Simon Kuznets examined the relationship between economic growth and initial inequality. The Kuznets curve hypothesis states that when a nation follows industrialization, especially in agricultural mechanization, the economic center of a nation will move gradually towards the urban zones. The consequence of this development is that farmers and unskilled laborers from rural areas have to change their workplace by moving to large cities in order to earn more. This movement causes a substantial gap in earnings between people living in rural areas and downtown areas. Business owners earn profits. Workers in these industries receive an increase in income at a slower rate. However, the incomes of farmers fall because the population in rural areas declines, while the urban population increases. Nevertheless, inequality then declines as economic growth reaches the highest level of average income. An increase in Gross Domestic Product (GDP) per capita follows after the country reaches the optimal level of industrialization. Kuznets states that this inequality tends to resemble a U-shaped curve, where it increases first and then decreases with the increase in GDP per capita.

After decades of hypothesizing, Kruger and Grossman [3] apply the concept to their research in the field of the environment. Kuznets' curve is used to illustrate the relationship between economic growth and environmental degradation. An inverted U-shaped correlation is found. These results indicate that the nation's early stage of economic growth can be associated with the sacrifice of environmental quality. This view supports observations from developing and emerging countries. However, when a relatively high level of economic growth and development is achieved, the concerns for environmental quality emerge and increase. As a result, the inverted U-shape of the EKC curve has been supported by various empirical studies, including Shafik [4] and Omotor and Orubu [5]. This relationship has been considered a standard feature in engineering for formulating and implementing environmental policies. Onafowora and Owoye [6] indicate the long-term relationship between economic development and CO2 emissions. Al-Mulali and Oxturk [7] also confirm the U-shaped relationship between GDP and CO2 emissions.

Given the importance of the concept, may empirical studies have been conducted to examine the validity of the EKC hypothesis. In the beginning, time series analyses with many di fferent techniques are employed such as Auto Regressive Distributed Lag (ARDL), Vector Auto Regression (VAR), the Vector Error Correction Model (VECM), and Granger causality to investigate the nexus of economic growth and environmental quality for specific countries or groups of countries. Empirical papers to date appear to focus on the investigation of a relationship among variables of interest using panel data. Mixed results on this relationship are reported, including studies from Ang [8], Hossain [9], Lean and Smyth [10], Magazzino [11], and Magazzino [12]. In particular, Rahman and Velayutham [13] confirm no causal relationship or unidirectional causality in the short and long run. In contrast, Zhang [14], Shahbaz et al. [15], Salahuddin et al. [16] report on the bidirectional relationship between economic growth and environmental quality. Niu et al. [17] indicate the unidirectional causality findings in the short run, and the directional relationship is observed in the long run.

In addition to testing the validity of traditional EKC theory (the inverted U-shape curve), a new stream of research examines whether a so-called N-shaped relationship between income and CO2 emissions does exist. This stream of research has raised many empirical research questions, which have been explored by various scholars such as Rahman and Velayutha [13], To et al. [18], Sarkodie and Strezov [19], Churchill et al. [20], Zhou et al. [21], Magazzino [22], and Magazzino [14]. Churchill et al. [20] test the N-shape relationship for the OECD countries in the period 1870–2014 using mean group estimators (Mean Group (MG), Pooled Mean Group (PMG), Augmented Mean Group (AMG), and Common Correlated E ffects Mean Group (CCEMG)). They find two turning points in terms of GDP per capita, i.e., the relationship exhibits an N-shape in some countries such as Australia, Canada, and Japan, but not in others, such as Spain and the UK. Moreover, Sarkodie and Strezov [19] also test this N-shape in the top five developing countries that emit a significant level of greenhouse gases, including China, Iran, Indonesia, India, and South Africa, using panel quantile regression with data from 1982 to 2016. The findings of this study confirm the N-shaped relationship between per capita income and CO2 emissions in selected countries, leading to support for the validity of the EKC hypothesis.

In addition, unlike other papers, the second background theory utilized in this paper is based on economic growth theory. Economic growth has been considered an important and interesting topic for many economists. Barro [23], who supplements the classical and neoclassical growth models, studies the growth model, which considers additional variables of energy and other macro variables, including the impact of governmen<sup>t</sup> on economic growth in the long run. This model simultaneously tests the validity of Keynes' theory and provides evidence on an unclear relationship between economic growth and environmental quality. By assuming governmen<sup>t</sup> spending is complementary to private-sector production, Barro's model points to the non-monotonous relationship between governmen<sup>t</sup> spending and economic growth. Hence, the neoclassical growth model with the participation of governmen<sup>t</sup> such as the model of Barro [23] is often used in the research to test the factors a ffecting economic growth [24].

Empirical research based on growth theory and the growth model of Barro [23] conducts two main empirical tests. The first empirical test examines the impact of the government's role on economic development. The second test examines the link between economic growth and energy consumption, a stream that normally runs alongside EKC research.

The mechanism and level of the economic impact of public spending remain controversial and are explained by di fferent theories. In Keynesian economics, economic output is determined by aggregate demand. Meanwhile, along with factors such as consumption, income, and net exports, public spending is seen as an important derivative of aggregate demand [25]. Consequently, therefore, Keynes argued that governmen<sup>t</sup> involvement in the economy is necessary. When the economy is in recession, the governmen<sup>t</sup> needs to maintain demand for investment to stimulate private investment with large public investment programs, also known as the "crowding-in e ffects" hypothesis of public spending with private investment [26,27].

In contrast, neoclassical growth models argue for the "crowding-out e ffects" of public spending on private investment [28–32]. Government spending can directly substitute private investment, thereby slowing future growth [31]. Furthermore, governmen<sup>t</sup> demand for goods and services may cause interest rates to rise. As a result, capital becomes more expensive, negatively a ffecting access to private sector capital. By raising taxes or borrowing to finance public spending, public spending also makes it di fficult for the private sector to access scarce financial resources [30,31].

Many economists, such as Devarajan et al. [33], Chen [34], and Ghosh and Gregoriou [35], extended Barro's model to examine the impact of di fferent components of governmen<sup>t</sup> spending on economic growth. By assigning di fferent elasticity coe fficients to di fferent sectors of governmen<sup>t</sup> expenditure, their models can determine the optimal scale and structure of the public sector for economic growth.

The empirical findings of other papers confirm the negative e ffect of public spending on economic growth [36,37]. In contrast, a positive contribution of public spending to economic growth has also been found in other studies [38]. Meanwhile, a few studies have found public spending to have non-linear e ffects on economic growth [39]. Interpreting the results of a mixed test, Gemmell et al. [24] point out the role of budget constraints in the relationship between public spending and economic growth. Nevertheless, empirical studies examining the role of budget constraints in the relationship between public spending and economic growth are quite limited [24,40].

In other words, empirical studies that test the relationship between energy consumption and economic growth face problems with inconsistency and conflicting results among researchers. In the beginning, researchers found a one-directional e ffect of this nexus; however, the direction between them is actually the opposite. For instance, Soytas and Sari [41] find no bidirectional nexus between the two, while Lee [42] confirms a causal relationship in which energy consumption a ffects economic growth and vice versa.

Huang et al. [43]; To et al. [18]; Vo et al. [44] state the reasons for their inconsistent results: (1) the di fference in the period of the time series; (2) the use of time series techniques without analyzing or controlling for structural change (change in the short run) and the business cycle; (3) the sample period not being long enough to analyze the long-run e ffects. Thus, to address these issues, especially the disadvantages of time series data, To et al. [18] used macro panel data (panel data with a large time dimension) on 25 emerging and developing countries to determine the causality nexus between energy consumption, foreign direct investment (FDI), CO2 emissions, and GDP. They found an inverted N-shaped relationship between GDP and environmental degradation. An inverted U-shaped nexus between FDI and CO2 emissions is also found in these emerging and developing countries, which implies a trade-o ff between economic growth and the quality of the environment, in which environmental standards are relaxed to attract more foreign investment. Moreover, they also stated the

positive impact of energy consumption on CO2 emissions. This finding is consistent with the results of Chandran and Tang [45] and Acaravci and Ozturk [46].

In this paper, the authors simultaneously use the EKC hypothesis on the environment and Barro's economic growth model. Unlike previous papers, our study does not utilize energy consumption as the total amount of energy. In contrast, our study breaks down energy consumption into various energy sources. We then carefully analyze the impact of each energy source on the environment and economic growth. This analysis is done together with the use of macroeconomic panel data to estimate the long-run e ffect for each energy source. Based on the estimated coe fficients, all energy sources are ranked in order of those that are the least harmful to the environment and that provide the most significant contribution to economic growth. We believe that the approach which was taken for solving our research objective is new. Multi-criteria decision making (MCDM) analyses, including five energy sources and two criteria, are considered to combine these two rankings from two criteria (environmental degradation and economic growth). This study uses the weighted scoring method (WSM), the most popular method for MCDM [47–49], to score all rankings from five sources and two criteria. This method chooses a set of several alternatives (energy sources), which depend on the score for each alternative and the weighting for each criterion. The final optimal structure of energy sources is a set of five sources that satisfy the two most important criteria, including: (i) the most positive and significant e ffect on economic growth, and (ii) the least harmful e ffect on the environment by reducing CO2 emissions.

### **3. Methodology and Data**

#### *3.1. Models Representing for the Environmental Goal and the Economic Goal*

To determine an energy structure that simultaneously achieves both the environmental degradation goal and the economic growth goal, we construct a model including two distinct parts for achieving these two goals simultaneously: (i) the environmental degradation goal and (ii) the economic growth goal.


To combine the rankings from these two parts, we develop a multi-criterion decision-making technique (MCDM) using five energy sources (including coal, gas, oil, hydropower, and renewable energy) and two criteria (environmental goal and economic goal). The weighted score method (WSM), the most popular method in MCDM [47–49], is used to score all the ranking results. The final structure for the energy mix demonstrates a source of energy (coal, gas, oil, hydropower, and renewable energy) in the order of preferences that satisfy the following two conditions: (i) a particular source of energy does the least harm to the environment or has the lowest CO2 emissions and (ii) a particular source of energy boosts economic growth the most.

### 3.1.1. The Environmental Goal

The rankings related to the environmental goal are employed following an examination of the validity of the traditional EKC hypothesis [50]. The non-linear relationships between environmental quality and income are reported. More specifically, the relationship has an inverted U-shape, which means that, after a threshold level of income, an increase in income will reduce the negative effect on environmental quality. On this basis, the model takes the following form [18]:

$$\text{EQ} = \text{f(GDP, GDP^2, EC)}$$

where EQ stands for the environment quality, which can be proxied by the level of emissions, such as CO2. In order to raise the reliability of the analysis and estimation, the proxy variable should have a long time period. As such, we consider that employing CO2 emissions as the proxy is appropriate. Income and square income are widely used in empirical analyses testing a non-linear relationship between economic growth and environmental degradation. EC stands for energy consumption, which comprises the consumption of five energy sources: oil, gas, coal, hydropower, and renewable energy. Employing various energy sources allows us to separate the contribution of each energy source to environmental quality. With these variables, we construct a regression model, model 1, as follows:

$$\text{CO}\_2\text{\text{\textdegree}} = \pi\_0 + \pi\_1 \text{GDP}\_{\text{\textdegree}} + \pi\_2 \text{GDP}\_{\text{\textdegree}}^2 + \pi\_3 \text{Co\text{\textdegree}}\_{\text{\textdegree}} + \pi\_4 \text{Gas}\_{\text{\textdegree}} + \pi\_5 \text{Oil}\_{\text{\textdegree}} + \pi\_6 \text{Hyd}\_{\text{\textdegree}} + \pi\_7 \text{Ren}\_{\text{\textdegree}} + \varepsilon\_{\text{\textdegree}} \tag{1}$$

where carbon emissions (CO2) are used as the dependent variable.

Various independent variables are used, including per capita real GDP, per capita real GDP squared and cubed, and per capita consumption of coal, gas, oil, hydropower, and renewable energy. All variables are transformed into their logarithmic form.

### 3.1.2. The Economic Growth Goal

According to growth theories, empirical studies on testing and estimating the e ffect of growth factors are commonly based on the production function, especially the Cobb-Douglas [51] function, divided into four main factors: technology, capital, human resources, and natural resources. These empirical studies normally transform the model into a logarithmic form to facilitate analysis. The output growth model of Barro [52] is basically presented as follows:

$$
\Delta \mathbf{Y} = \mathbf{F}(\mathbf{Y}, \mathbf{Y}^\*)
$$

where ΔY is the growth rate of income/output, Y is per capita income/output, and Y\* is the long-run level of income/output or potential income/output of an economy. The value of Y\* is based on governmen<sup>t</sup> policies such as investment in education, research activities, and increases in capital. ΔY is positively related to Y\* and negatively related to Y. The Barro model, which includes control variables, is as follows:

$$
\Delta \mathbf{Y}\_{\rm it} = \pi\_0 + \pi\_1 \mathbf{Y}\_{\rm o\,i} + \pi\_2 \Delta \mathbf{EC}\_{\rm it} + \pi\_3 \mathbf{X}\_{\rm it} + \varepsilon\_{\rm it}
$$

where Δ Yit is the economic growth rate of country i at year t, Yo I stands for the logarithm of initial per capita GDP of country i, and ECit denotes the log of energy consumption of country i at year t. Barro [52] and Huang et al. [43] used control variables (Xit), including inflation, capital stock, governmen<sup>t</sup> spending, growth of labor, and degree of international openness. Model 2 is written as follows:

$$\begin{aligned} \Delta \text{lnY}\_{\text{il}} &= \pi\_0 + \pi\_1 \text{lnY}\_{\text{o}} + \pi\_2 \Delta \text{lnCos}\_{\text{il}} + \pi\_3 \Delta \text{lnCos}\_{\text{il}} + \pi\_4 \Delta \text{lnOil}\_{\text{il}} + \pi\_5 \Delta \text{lnHyd}\_{\text{il}} + \pi\_6 \Delta \text{lnRen}\_{\text{il}} + \\ &\pi\_7 \text{INF}\_{\text{il}} + \pi\_8 \text{CAP}\_{\text{il}} + \pi\_9 \text{GEX}\_{\text{il}} + \pi\_{10} \Delta \text{lnLF}\_{\text{il}} + \pi\_{11} \text{TRADE}\_{\text{il}} + \varepsilon\_{\text{it}} \end{aligned} \tag{2}$$

where:

> ΔlnYit: the first di fference in the logarithm of per capita income for the country i at year t; lnYo i: the log of initial per capita income of country i;

ΔlnCoa/Gas/Oil/Hyd/Renit: the first di fference in the logarithm of coal, natural gas, oil, hydropower, and renewable energy consumption for country i at year t;

INFit: inflation rate of country i at year t;

CAPit: gross fixed capital formation for country i at year t (%GDP);

GEXit: general governmen<sup>t</sup> final consumption expenditure for country i at year t (%GDP);

ΔlnLFit: the first di fference in the logarithm of the labor force for country i at year t;

TRADEit: total export and import as a share of GDP for country i at year t (%GDP).

### *3.2. Econometric Techniques*

Two models, model 1 representing the environmental goal and model 2 representing the economic goal, are analyzed using the same econometric techniques. Various econometric analyses, including the cross-sectional test, the stationary test, and panel cointegration test, are conducted. Nguyen and Vo [53] and To et al. [18] state that this procedure can be estimated by three steps. First, macro panel data have a long time dimension; thus, the first step for the macro panel data is the same as with the time series data. Second, the order of integration for each variable of the macro panel data needs to be tested and determined. Third, a prerequisite for the existence of a long-run relationship is the presence of cointegration between variables. Once cointegration is confirmed, the long-run relationship between the group of integrated variables can be investigated using the Vector Error Correction Model (VECM) (see [54,55]). These tests are discussed in detail by To et al. [18].

If these tests are verified, and the results show that there is at least one long-run nexus between explanatory variables and the dependent variable, then the long-run estimators are employed. The most popular models for estimating the mean group long-run e ffect, which can treat endogeneity problems and serial correlation in macro panel data, are fully modified ordinary least squares (FMOLS) and dynamic ordinary least squares (DOLS). These techniques, FMOLS and DOLS, are used in this paper. The purpose of this study is to find a balanced structure of energy sources (coal, gas, oil, hydropower, and renewable energy). As such, after conducting regressions, these five energy sources are ranked based on their regressors.


### *3.3. Weighted Scoring Method*

The weighted scoring method (WSM) is then used in the next step to combine the two sets of rankings (one set for the environmental goal and the other set for the economic goal) based on the score for each of the five energy sources (coal, gas, oil, hydropower, and renewable energy). The following equation is used: 

$$\mathbf{S(A\_i)} = \sum \mathbf{W\_j} \times \mathbf{S\_{ij}}\dots$$

The multi-criteria decision-making techniques (MCDM) in this paper now consist of two criteria {C1, C2} (being the environmental criterion and the economic criterion) and five energy sources {A1, A2, A3, A4, A5} (being coal, gas, oil, hydropower, and renewable energy) in a decision matrix of all choices {Sij}, where {Sij} is the score after calculating and evaluating the performance of choices using criterion {Cj}. The weights {W1, W2} indicate the importance or the role of a specific criterion. A sensitivity analyses using di fferent sets of weights are also conducted to ensure the robustness of the findings.
