**2. Literature Review**

The causality between energy consumption and economic growth has been approached by several authors. For example, the study of Caraiani et al. [15] confirms the causality between primary energy consumption from different sources and gross domestic product for Bulgaria, Poland, Romania, Hungary and Turkey between 1980 and 2012/2013. Paul and Bhattacharya [22] studied the relationship between economic growth and energy consumption in India for the 1950–1996 period and they found that there is both a bidirectional causality and same direction causality between the two variables. Asafu-Adjaye [9] proves in his study on Indonesia, India, Thailand and the Philippines that the relationship between energy and income is not a neutral one, hence, energy consumption is influenced by the income variation. A similar observation is made by Aqeel and Butt [23] who prove that the economic growth of a country directly influences the growth of petroleum consumption in that country.

Other experts [10,24,25] demonstrate, with different approaches, that carbon emissions and energy consumption do not lead to economic growth, so their suggestion to the authorities is to pursue energy conservative policies and carbon reduction policies since they don't interfere with the economic development of a country. Their results are very important for supporting the global strategies of energy consumption reduction.

Another study [26] aims at determining the existing relations between the energy consumption from different energy sources and the economic growth in the case of Romania, Spain and the EU27 average, on short and long term. The authors [26] prove the existence of unidirectional relations from energy consumption towards gross domestic product (GDP), notably the causal influence of renewable energy consumption on economic growth in Romania and Spain. On the other hand, Aspergis and Payne [27,28] find that there is a bidirectional causality between economic growth and renewable energy consumption, based on their studies on different sets of countries. Further studies of Aspergis and Payne [29] on a panel of 80 countries showed that there is a positive impact on real GDP from both renewable and non-renewable energy consumption, since there is only a small difference in the elasticity estimates of the renewable and non-renewable energy consumption.

Çoban and Topcu [16] investigate the effects of economic growth, energy prices and financial development indices on energy consumption at the EU27 level. At the same time, comparing the 15 older EU states with the 12 newer ones, they note that an increase in energy consumption due to higher financial development in the EU15 countries, as the rest of the member states have a less developed financial system, especially with regard to stock exchanges, which limits its impact on energy consumption. However, a poorly developed financial system produces negative effects on investments in energy efficient technologies, these effects being suggested by energy intensity above the EU average of the EU12 countries [16].

More recent studies [14] confirm the differences between developed and developing economies, showing that the use of energy from renewable sources has a higher influence on economic growth of the countries with higher GDP than for those with lower GDP.

Other approaches [30] show the influence of emissions, GDP, financial development, capital stock and population on energy consumption. In addition, Stadelmann and Castro [31] propose unusual indicators to be analysed in the estimation of energy consumption in relation to public policies in 106 states during 1998–2009.

The causality between carbon dioxide emissions, energy consumption, gross domestic product, and foreign direct investments is investigated by Kim [32] who claims that there is no direct short-term causality between foreign direct investments and CO2 emissions, based on a study on 57 developing

countries in the 1980–2013 time frame. His results do not support the idea that foreign investments positively influence the CO2 emissions. However, Nasir et al. [33] claim the contrary. By using a panel data analysis on data from 1982 to 2014 in five East-Asian economies, they [33] find that financial development, economic development and foreign direct investments have a statistically significant long-term relationship with environmental degradation represented by CO2 emissions. In fact, financial development and foreign investments leads to an increase in environmental degradation.

By studying the relationship between energy consumption and economic growth in the countries known as BRICS (Brazil, Russia, India, China and South Africa), Baloch et al. [34] suggest that there might be a correlation between the abundance of natural energy resources and CO2 emissions, based on a panel data study from 1990 to 2015. Their [34] results show that the abundance of resources mitigates the emissions in Russia, while in South Africa it increases them. Aydin [35] finds that in the same BRICS countries the increase of biomass energy consumption use would have positive results towards sustainable environment, economic growth and energy dependency reduction.

Salim et al. [36] uses a linear and non-linear econometric approach to see the effects of urbanization on pollutant emissions and energy intensity in developing economies of Asia. Their results show that population, prosperity, and non-renewable energy consumption are major influencers of pollutant emissions. A more important result is that for the countries which achieved a certain level of development, their emissions tend to decline.

While investigating the literature review on the methodology used for studies in this field, it was observed that the top-down methodological approaches on the evolution of environmental conditions and the energy sector take into account energy market components, but do not include technological development in the analysis, while bottom-up approaches use the model of overall balance to capture the determinants of change in the energy system and the natural environment, such as emissions, energy efficiency and technology, without considering the feedback from the economy [37]. These limits have led to the emergence of a mixed approach that suggests feedback, but due to the nature of the equilibrium models, it still fails to surprise those [37].

Considering the proposed policies suggested by different authors, we mention the Colombian case, where a low-carbon policy would preserve low emissions in electricity generation [37]. Other studies [38] use the non-causality in heterogeneous panel test to see if the exploitation of renewable energy sources in the EU-28 countries is an achievable solution for environmental pollution mitigation. Their results suggest that it is possible to reach the sustainable development goals until 2030 through renewable energy consumption and carbon emission mitigation, so they support the policies regarding renewable energy promotion. Some authors [39] support the idea that one policy could not work for each case, so a mixed policies approach should be considered based on the specificities of every country. Another study suggests a policy of rewarding the most efficient countries by granting them potential increases in emission and energy consumption while the least efficient countries must bring decreases to achieve full efficiency by applying the modelled reallocation [40].

According to Belke et al. [41], most of the current models for analysing energy relations and economic growth are based on the model of production functions such as Saidi and Hammami's [30] study, which, however, does not include the price of energy, as most studies in this area. Even so, the data panel is preferred over time series and cross-sectional analysis due to its higher accuracy by including binary variables that capture different time series and different cross-sectional units with the fixed or impact effects model [42]. On the other hand, the Wang et al. [43] study includes influence factors such as the following: Energy prices, urbanization and GDP on energy consumption through a panel data analysis on 186 countries between 1980–2015, and it finds that energy prices negatively affect the energy consumption in low-and medium-income countries. Also, the study finds that urbanization is a very important factor which affects energy consumption per capita. Also, Lv et al. [3] support the idea that urbanization influences carbon emissions from energy consumption, but due to the new ecological or green trends followed by the urban population, the urbanization has an alleviation effect on the emissions level.

In opposition to urbanization, there is another high energy consumption factor, namely agriculture. Harchaoui and Chatzimpiros [44] discuss the possibility of reducing the energy consumption in agriculture. Their results show that the current agricultural model is structurally energy deficient. Basically, its functional energy requirements are almost equal with the final production. The energy potential from manure and crop residues (as biomass) could only equal the external energy needs of agriculture [44]. For agriculture to become an energy source it is supposed to stop feeding from cropland and to reach the maximum amount possible from the agricultural residues [44]. Tian et al. [45] propose a more thoughtful choice of production ways to improve the sustainability of agriculture, by reducing the energy consumption. Their observations prove that the amount of energy consumed for growing the agricultural product is very high and unadjusted to the geographical conditions; however, it could be easily reduced by adjusting to the area of growing [45].

Other authors [46] go further to propose renewable energy sources along with a pros and cons evaluation of the source. The main reason against the alternative energy source (geothermal energy) is the high investment needed to turn it into a viable system, which makes it an option only for developed countries. Whereas, in countries like Turkey, where there is an abundance of geothermal sources, the rapidly growing population and economic growth do not allow for a stagnation in the use of pollutant energy sources (such as coals) and the investment in harnessing the renewable energy source [47]. After Temiz Dinç and Akdo ˘gan [48] demonstrate a bidirectional causal relationship between renewable energy and economic growth based on 1980–2016 data, they claim that increasing renewable energy production and decreasing energy consumption are a must for ensuring Turkey's sustainable development. Some authors [49] come with solutions appropriate for reducing the final energy consumption, based on a multi criteria analysis on the case of Italy, which proves the efficiency of using solar thermal panels combined with the heat pumps instead of the current system used for providing hot water and heat.

Mostly based on panel data analysis, on longer or shorter periods, most of the studies demonstrate the negative effects of energy consumption over the environment, through the polluting effect it has, but also the fact that it does not affect the economic growth of a country in a significant way. In this case, most of the recommendations of the experts incline to designing new policies, which should integrate investments in renewable energy production and replacing non-renewable sources, mostly used in the current situation.

#### **3. Materials and Methods**

This study uses both bibliometric analysis and panel data techniques. The first method represents a quantitative analysis of the literature review in the field in order to emphasize the importance trend of a topic, as well as its main areas of interest. The bibliometric analysis is conducted on the 671 articles found on the Web of Science database, in the 1975–May 2019 time frame, by using terms such as "energy", "emission\*" and "economy" in the query and refining the results after the "relationship" filter word, in order to keep only the results which have a model included. After finding these articles on the Web of Science database, quantitative analysis was performed by investigating the trend of the scientific production in the field, the most prolific authors, the areas of interest, as well as the affiliated countries of the publications found on Web of Science. The method is useful in overviewing the previous results on the researched area, and it is constantly used in other studies [50,51]. Also, by using the Vosviewer software, the concepts mostly used in these articles and the relationships between them will be exposed [52]. This software creates word networks by analysing the title, the abstract and the keywords of research data from Web of Science. The limit of this technique is given by the fact that the information provided by the title, the abstract and the keywords of research data has a marketing purpose and, sometimes, it might reach subjects which are not thoroughly debated in the full corpus of the scientific publications. Nonetheless, the word networks give an overview of the areas of interests in the field.

Further, panel data is the econometric technique chosen to observe energy consumption determinants, as it is more comprehensive than time series or cross-sectional analysis [42]. In comparison with other studies in the field [17,30,53–56], this piece of research offers a more comprehensive analysis because of the numerous variables considered in the energy consumption influence analysis.

The estimation of the panels was conducted by testing the three methods explained by Wooldridge [57], namely: The common or non-effect constant method, implying assigning a common constant to all countries considered in the panel, not making the difference between states and periods; the fixed effect method, which allows the assignment of a fixed constant for each state/period, that is, the constant varies cross-sectional; and the random effect method, which allows the treatment of constants, which are not fixed for each state/period, as random parameters, which involve the inclusion of a new error due to differences in the fixed effect in the error term.

The multiple regression models include data on the categories of data presented in Tables 1–3.

Also, in the analysis was considered the square of GDP (GDP-2) in order to test the potential existence of a hyperbola shape.



\* Part of the energy mix. Source: Our own abstracting of the considered factors.

**Table 2.** World Bank variables used in the models [20].


Source: Our own abstracting of the considered factors.


**Table 3.** Energy Information Administration (EIA) variables used in the models [21].

All these variables were chosen by investigating several scientific studies [16,17,30,53,58–63].

Our first hypothesis is that GDP, GHG (greenhouse gas emissions), the energy mix based on nuclear energy and fossil fuels, capital accumulation, financial development, trade opening, military development, internet access, agricultural area, population density, labour force, and the degree of urbanization has positive relationships with energy consumption, so it causes negative effects on the holistic system, which considers all three pillars of sustainable development: Economic, social and environmental.

At the same time, the use of renewable energy, the price of oil, net energy imports, research and development funding, environmental taxes, exhaustion of natural resources, female decision makers, health system financing and high level of education might have negative relationships with energy consumption, thus generating positive effects on the holistic system by reducing the use of energy. The more educated people, whom should be more aware of the current climate change and energy challenges and the higher access to internet might contribute to reducing energy consumption. In addition, the increased financing in the health system could contribute at energy savings and energy efficiency improvements.

For the study of the influencing factors of energy consumption, many models described by Equations (1) and (2) were tested and obtained.

$$PEC\_{\rm it} = \alpha\_{\rm it} + \sum \beta\_{\rm it} \times Ecomonic\_{\rm \\_it} + \sum \gamma\_{\rm it} \times Energy\_{\rm \\_it} Var\_{\rm it} + \sum \lambda\_{\rm it} \times Sccoio - Eco\\_Var\_{\rm it} + \varepsilon\_{\rm 1it} \tag{1}$$

$$FEC\_{\text{it}} = \phi\_{\text{it}} + \sum \eta\_{\text{it}} \times Ecownic\_{\text{-}}Var\_{\text{it}} + \sum \kappa\_{\text{it}} \times Enrgr\_{\text{-}}Var\_{\text{it}} + \sum \eta\_{\text{it}} \times Soco - Eco\\_Var\_{\text{it}} + \varepsilon\_{\text{2it}} \tag{2}$$

 *Economic*\_*Varit*—Variables related to GDP, capital stock, internet users, gross fixed capital formation as a share of GDP, financial development, the external balance of goods and services, military expenditure, research and development expenditure in GDP. Each variable from this category tested in the panel model is attributed a different coefficient β*it* or η*it*, which offers information on the effect of the relationship with the endogenous variable and its impact. This means that the size and the sign of the impact of each variable in this category is different on primary energy consumption (PEC) and final energy consumption (FEC).

 *Energy*\_*Varit*—Variables related to different types of energy consumption, energy mix, oil price and the share of net energy imports. To each variables from this category tested in the panel model, it is attributed a different coefficient- γ*it* or κ*it*, which offers information on the effect of the relationship with the endogenous variable and its impact. This means that the size and the sign of the impact of each variable in this category is different on PEC and FEC.

 Socio − Eco \_Var*it*—Variables related to population, female population, degree of urbanization, female legislators, officials and managers, the proportion of women's mandates in national parliaments, the workforce, the share of the labour force that followed tertiary education in the total workforce, health expenditure in GDP, greenhouse gas emissions (GHG) emissions, share of agricultural area in total area, environmental taxes as a share of GDP and rental of natural resources as a share of GDP. To each variables from this category tested in the panel model, it is attributed a different coefficient; λit or ϕit, which offers information on the effect of the relationship with the endogenous variable and its impact. This means that the size and the sign of the impact of each variable in this category is different on PEC and FEC.

εit—Represents the error of each model [64];

*i*—Represents the geographic indication;

*t*—Represents the time considered in the analysis.

The energy mix has been tested by the energy consumption of fossil fuels (XFOS), renewable (XER) and nuclear energy (XEN) in total gross domestic energy consumption, according to several authors [54,59,65].

In addition, the independent variables were tested on both total and divided to the population and the results indicate insignificant differences. So, this research presents only the results which considered the total independent variables.

All the indicators used in the models present the state and the evolution of the society on a certain time frame. These have been introduced in the econometric models after fulfilling the hypotheses necessary for regressions validity. So, the data was tested for stationarity, the existence of a normal distribution, and multicollinearity between the variables used in the same model.

Data stationarity was tested for the variables of all models analysed by indicating the existence of the unit root at the panel and the individual series, i.e., the presence of autocorrelation between past data of the same variable [64], which should not appear in order to apply the regression model. According to the applied tests, most of the variables were not stationary at the first level, the stationarity being identified at their first difference, generally by considering a constant or a lack of trend and constant after the graph of the time series [18]. Also, natural logarithm was applied to the energy consumption, GHG, GDP accumulation of capital (K), population and workforce indicators for estimating the elasticity of coefficients of variables in regressions. Then the difference was applied in the case of non-static data [57], which leads to normalization of time series. This was the case of all variables considered in the models.

Further, the variables were considered for a model only if the collinearity coefficient was less than 0.5 and the causality between them was found present. The correlation matrix contributes to establishing regression models as it highlights the multicollinearity that does not have to be present between the regression variables [18].

Moreover, the Granger test provided the existence of the causality between two variables, two by two, but there are a multitude of determinants of the analysed endogenous variables that act together within the holistic system. Therefore, those variables that appear to have no influence on the dependent variable may in fact cause changes in the endogenous variables. This can be highlighted by presenting the results of multiple regression analysis, which also highlights the sign of the changes made. So, the results of the causality tests were also considered in the construction of the models at the EU28 level during 1995–2014 [18]. Thus, it was found that there is a causal relationship of GHG and economic growth on energy consumption at EU28 level in the short and medium term, as demonstrated in previous studies by Kasman and Duman [17] for the EU12 and Saidi and Hammami [30] for 58 countries, of which EU15 can be observed. In addition, the economic and social determinants common to all two types of energy consumption analysed as endogenous variables are as follows: Capital accumulation, environmental taxes, oil prices, population, representatives in female leadership positions, labour force, and number of Internet users. In addition, percentage changes in both primary and final energy consumption are caused by past financial development, as is also the case of Saidi and Hammami [30] article, however, it does not capture the influence of capital accumulation. At the same time, trade openness, along with previous levels of natural resource depletion and agricultural surface, causes changes in primary energy consumption, like the outcome of Kasman and Duman [17], but no such evidence was found for the final energy consumption. Another interesting result is the causal influence of the urbanization change on the final energy consumption, which has not been econometrically demonstrated yet at EU28 level. Another important explanatory variable, RD-GDP, causes changes in the final energy consumption, a fact that is politically relevant by highlighting the importance of research in developing the technologies needed for energy sustainability. Finally, the past percentages of tertiary education workforce are generating changes in final and primary energy consumption, which highlights the importance of educating and informing employees about sustainability requirements and new green energy policies.
