*Article* **The Nexus between Economic Growth, Energy Consumption, Agricultural Output, and CO<sup>2</sup> in Africa: Evidence from Frequency Domain Estimates**

**Adedoyin Isola Lawal**

Department of Economics, Bowen University, Iwo 232102, Nigeria; adedoyin.lawal@bowen.edu.ng or l.adedoyin@yahoo.com; Tel.: +234-8035233567

**Abstract:** This study examined the nexus between economic growth, energy consumption, and the environment with the moderating role of agricultural value addition and forest in Africa based on data sourced from 1980 to 2019. We employed both the time domain and frequency domain panel Granger causality estimation techniques to compare results across the different horizons. Extant literature suggests the inability of time domain estimation techniques to account for causality at different frequencies. The study also accounts for the nexus among our variables both at the singlecountry and multi-country levels. The results at the single-country level are at best mixed. The results of the panel Granger causality at the frequencies domain suggest that a bi-directional relationship exists between energy consumption and economic growth, and that energy consumption Granger causes carbon emissions in Africa. The results align with the feedback hypothesis on the one hand but contradict the conservation hypothesis on the other hand. The study has some policy implications.

**Keywords:** energy consumption; carbon emissions; agricultural output; economic growth; Africa

#### **1. Introduction**

In attaining sustainable development, energy, economics, and the environment play significant roles [1–5]. For instance, energy is crucial to the human economic and social development of any nation. It is estimated that global energy consumption will increase by about 56% from its current state in 2010 by the year 2040, as global aggregate demand is expected to double, given the expected increase in population [6–12]. However, the projected increase in total energy consumption is expected to be accompanied by an increase in carbon dioxide (CO2) emissions, which is a core factor in total greenhouse emission (GHG). The energy sector is responsible for about 61.4% of the total global GHG [13–16]. Ref. [7] noted that the contributions of agriculture sector to the GHG are estimated to be between 14–30%, though evidence abounds to show that the agricultural sector possesses the ability to reduce GHG by 80–88%. It is opined that forests possess the capacity to accumulate atmospheric carbon after converting CO<sup>2</sup> into carbon and oxygen, and that about 430 tons of carbon per hectare is absorbed in the wet forest, hence, halting the effects of carbon emissions [17–22].

In the same vein, environmental degradation plays a crucial role in the continuous occurrence of natural disasters with unprecedented impacts on the economy. Disasters related to oil spillage, water pollution, solid waste management, deforestation, soil erosion, salinity and water, logging, and desertification, among others, affects the socio-economic wellbeing of a nation and increases climate change. Environmental degradation worsens with the exploitation of fossil fuels [23–27]. In order to mitigate this without losing a significant part of the energy output, economies over the years have opted for renewable energy sources [28–30]. Renewable energy offers clean and safer energy and can be derived from solar, tidal, wind, geothermal, hydro and biofuel power. Besides its alternative energy potential, it is useful in supporting employment, output, income, and job creation. Extant

**Citation:** Lawal, A.I. The Nexus between Economic Growth, Energy Consumption, Agricultural Output, and CO<sup>2</sup> in Africa: Evidence from Frequency Domain Estimates. *Energies* **2023**, *16*, 1239. https:// doi.org/10.3390/en16031239

Academic Editor: George Halkos

Received: 23 December 2022 Revised: 10 January 2023 Accepted: 16 January 2023 Published: 23 January 2023

**Copyright:** © 2023 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

literature shows that the increase in economic growth and agricultural outputs have a positive impact on renewable energy [31,32]. Furthermore, given a global temperature increase of between 2–2.4 ◦C, renewable energy can help reduce carbon emissions by 50% by the year 2050. Besides its positive impact on the environment, renewable energy can reduce overdependency on foreign energy, given the fact that it is sourced domestically [33,34].

The United Nations Sustainable Development Goals (SDGs) emphasized the need to eradicate hunger (SDG 2), achieve clean energy utilization (SDG 7), achieve sustainable economic growth (SDG 8), adopt sustainable production and consumption (SDG 12), mitigate climate change through a sustainable clean environment (SDG 13), and adopt a global partnership model to achieve these goals (SDG 17). The nexus between these laudable metrics for sustainable development is key to exploring the linear and circular economic growth in any economy, be it regional or single country (Sarkodie 2020). Sub-Saharan Africa needs more energy than most continents of the world, given its everincreasing, teaming population and quest for sustainable growth [35]. Even though the continent is endowed with an abundance of non-renewable energy like petroleum and other fossil fuels, the negative impacts of fossil fuel on the environment, such as the increase in GHG and other pollutants, calls for concerns. Although the contribution of Africa to global warming at present may be negligible compared with other continents, it is obvious that the continent will be disproportionately affected by its impact if nothing is done. To mitigate the impact of GHG on the continent, the African Development Bank (AfDB) adopted a ten-year green growth strategy (2013–2022) with an emphasis on developing the renewable energy potential capable of promoting resource efficiency and sustainable development.

Several theoretical models exist that explain the links between energy, the economy and the environment. For instance, Environmental Kuznets Curve (EKC) models suggest that at the initial stage of development, a direct positive relationship exists between economic growth proxy by real gross domestic product (RGDP) and environmental pollution, but the relationship becomes indirect after a threshold level of income is achieved. The pollution haven model suggests that in developing economies characterized by weak pollution protection laws, trade and investment liberalization laws often induce environmental degradation as pollution-intensive firms will find it easier to produce in such economies than in developed economies with stringent environmental protection policies. The causality model employs unit roots, cointegration and causality measures to examine the nexus between energy consumption and economic growth. This model offers four possibilities, firstly (i) the growth-led hypothesis, which suggests the existence of unidirectional causality from economic growth to energy consumption. This suggests that conservation policies will have no impact on economic growth. This is common in energy-sufficient economies. Secondly, (ii) the energy-led hypothesis, which suggests that energy consumption stimulates growth, therefore, energy conservation policies will impact negatively on economic growth, thus, energy expansion policies are required. This is common in economies that are energy-dependent like most developing economies. Third is (iii) the feedback model, which suggests the existence of a bi-directional causality between energy consumption and economic growth. The model suggest that both constructs are jointly determined and affected simultaneously. Lastly is (iv) the neutrality model, stating that no causality exists between energy consumption and economic growth. It also suggests that environmentally-friendly policies can be achieved without obstructing economic growth.

Extant literature has attempted to examine the link between the environment, energy, and the economy with mixed results. For instance, Refs. [36–38] were of the view that causality runs from economic growth to energy consumption while Refs. [39–43] opined that causality is from energy consumption to economic growth. Furthermore, Refs. [41,42,44,45] noted that causality runs from economic growth to CO<sup>2</sup> emissions. The bulk of these studies focused on developed economies with little attention on African economies. Africa is faced with plurality of issues, key among them being the need to stimulate growth, ensure a sustainable environment and reduce energy poverty. The World Bank global monitoring report (2008) highlights the need for the continent to be on a sustainable development path

that embraces clean energy, a sustainable environment, and accelerated growth, noting the continuous increase in CO<sup>2</sup> emission and fall in per capita water resources. Given the low state of renewable energy development and the potential environmental hazards emanating from existing conventional fossil fuel amidst the desire to stimulate growth, it is imperative to examine the nature of the relationship between energy consumption (renewable and non-renewable), economic growth, and CO<sup>2</sup> emissions with the moderating impact of agriculture and agro-allied resources in Africa. Our study presents a short, intermediate, and long run analysis for 34 African economies. Unlike existing studies that employed time domain estimates like the traditional Granger causality estimates, VAR and other time domain estimates [16,30,46–49], the current study employed both the single and multi-country frequency domain Granger casualty estimates based on datasets sourced from 1980–2019. Even though frequency domain techniques offer better estimation models, because they allow for examination of the direction and level (strength) of the nexus at heterogeneous scales for frequency [2,3,9,50–52], they are yet to be explored especially in studies in Africa.

Our choice of Africa was induced by the fact that Africa is endowed with an abundance of potential energy resources (both renewable and non-renewable). It is estimated that in Africa, the potential energy generation capacity is up to 1.2 terawatts, excluding solar, and more than 10 terawatts including solar, with a high potential of achieving more than a 25% increase in clean energy by 2040 [8,53,54]. The continent is the world's youngest and fastest urbanizing continent, but it is the least energy-supplied, with annual consumption being 518 kwh in sub-Saharan Africa, equivalent to what a single member country of the OECD will use. Economic indices show that recently, African economies largely outperformed the global average (IMF, WB 2019) with the continent's overall GDP increasing 3.8% against the global average of 3.4%. Data availability large influences the choice of sample economies.

Against this background, this research attempts to know whether various energy policies in the continent offer the ability to end Africa's energy poverty, stimulate growth, and promote environmental sustainability. We intend to answer the following questions: (i) What drives the African economic, energy and environmental nexus—an environmental Kuznets curve, causality, or the pollution haven model? (ii) What is the nature of the causality between energy consumption (renewable and non-renewable) and economic growth, carbon emissions, and agricultural output in Africa? (iii) If causality is established, to what extent will the increase in energy consumption support economic growth, agricultural output, and reduce carbon emissions in Africa economies? Answering our questions will provide insights into at least five SDGs: SDG 2—zero hunger; SDG 7—achieve clean energy utilization; SDG 8—achieve sustainable economic growth; SDG 12—adopt sustainable production and consumption; and SDG 13—mitigate climate change through a sustainable clean environment.

This study will make essentially four contributions to the literature. First, in terms of methodology, we will provide a frequency-based panel Granger causality analysis that offers short, intermediate and long run casual estimates of the nexus between economic growth energy and the environment with a focus on African economies. Our method provides individual estimates for each of the economies studied, unlike the conventional methods that offer lump-sum causality estimates. Second, the study will calibrate the moderating impact of agriculture and agro-allied resources to the discourse on energy, economics and the environment in Africa. Africa is largely agrarian and to the best of the author's knowledge, no literature of the African extraction has considered the moderating role of agriculture in absorbing carbon emissions in the economic-energy-environmental nexus. Thirdly, in term of coverage and scope, our study will cover more African economies than most of the existing studies and use more recent data when compared with others. Fourthly, our study will also calibrate both the energy conservation and expansion policies into the energy, environment, and economic growth discourse. Our finding offers some policy implications for policy makers at both the national and regional levels, as well as for

international organizations and researchers on the link between energy, economic and the environment. The rest of the study is as follows: Section 2 presents the literature review; Section 3 offers the data and methodology; Section 4 deals with the presentation of results, while Section 5 concludes the study and offers some policy implications.

#### **2. Literature Review**

A critical assessment of extent literature clearly suggests that frequency domain estimates are yet to be sufficiently employed in examining the nature of the relationship between energy, economics and the environment with the moderating role of agriculture, especially based on evidence from Africa, despite its attractiveness and potential strength in providing measures in shaping the African policy space. Africa economies are in dire need of energy, with the need to advance economic growth at the front of the policy framework amidst the global quest to reduce CO2. It is pertinent, especially when faced with few publications on the subject matter, to examine the moderating role of agriculture in mitigating CO<sup>2</sup> emissions, stimulating economic growth and ending energy poverty. Such effort would not only offer a valuable platform to examine the nature of cointegration and the direction of causation, among the variables (energy, economics, environment and agriculture), it will equally initiate and stimulate further research and model specifications.

Table 1 presents the result of extent literature on the nexus between energy, economic growth, agriculture, and carbon emissions for a number of economies across the globe. The results as presented can be categorized into four main streams—methodological, results (findings), hypothesis or policy trust and variables employed. In methodological strands, a number of studies employed cointegration and/or Granger causality methods to investigate the link between energy, economic growth, and the environment [6,16,19,20,22,23,28,49,55–63] with mixed results. For instance, while [19] noted that a bi-directional relationship exists between non-renewable energy and climate change and that climate change Granger causes renewable energy for 16 African countries, ref. [16] observed that causation is from RGDP to renewable energy in the long run for China, with a negative impact on renewable energy in the short run. Similarly, ref. [13] documented the existence of a bi-directional relationship between renewable energy and non-renewable energy for India and South Africa, suggesting validity of the feedback hypothesis. The study further noted that causality runs from non-renewable energy to economic growth for Brazil and USA, an indication that the growth hypothesis is valid in these economies but noted no causal relationship exists between non-renewable energy and economic growth for Russia, India and South Africa, implying the validity of the neutrality hypothesis. For South Africa, ref. [6] noted that growth hypothesis is valid as the direction of causation is from energy use to RGDP. Ref. [19] offers multifaceted results, for instance, the authors documented that bi-directional relationships exist between fossil fuel and RGDP, between fossil fuel and CO2, and between CO<sup>2</sup> and RGDP for the oil-exporting economies. These results support the feedback hypothesis from oil prices to each of RGDP and CO<sup>2</sup> for the oilconsuming economies, suggesting the validity of the growth hypothesis. Ref. [57] results are at variance with those of [22–24,28,29,58] who noted causality is from RGDP to CO2, and that no causality exist between energy consumption and economic growth, thereby supporting the validity of the neutrality hypothesis in the studied economies.


**Table 1.** Summary of Literature review.







method, vector error correction model, error correction model, fully modified ordinary least square, dynamic ordinary least square, autoregressive distributed lag

model based on pooled mean group estimation, respectively.

The second strand of literature employs nonlinear models like quantile regression, system frequency domain estimate PMG, threshold regression, bootstrap estimates, NARDC, and recursive to examine the nature of relationship between energy, economic growth and CO<sup>2</sup> emissions with mixed results. For instance, [8,13,18,36,49,63,71,72,77,79,80,83] employed different versions of nonlinear models to examine the nexus between energy, economic growth, and CO<sup>2</sup> emissions with different results. Ref. [13] noted that fossil energy causes GHG, and that economic growth does not cause CO<sup>2</sup> emissions for 41 sub-Sahara African economies. Ref. [18] results from N-ARAL observed mixed findings; for example, the study noted that renewable energy reduces CO<sup>2</sup> emission for Nigeria, but no causality was documented between renewable energy and CO<sup>2</sup> for Angola and Egypt. The study further noted that renewable energy causes economic growth for Gabon, suggesting the validity growth hypothesis. Ref. [84] employed panel threshold for some selected OECD economies and reported the existence of positive and non-linear relationships between renewable energy and economic growth, an indication that the growth hypothesis holds. Ref. [49] employed the N-ARAL model and noted that environmental quality causes economic growth and that the neutrality hypothesis is valid, based on the results from environmental quality and capital stock. In a related development, [8] employed panel quantile regression to examine the nature of the relationship between energy, economic growth, and CO<sup>2</sup> for some selected 66 developing economies and noted that renewable energy reduces CO<sup>2</sup> with substantial effect at the 10th quantile, and that GDP increases CO2. Ref. [63] results, based on quantity ARDL, suggest the validity of the feedback hypothesis among economic complexity, energy consumption and the ecological footprint. For emerging economies [36] employed a bootstrap panel causality test and noted that the neutrality hypothesis is valid for all the economies except Poland, whose results suggest that causality is from renewable energy to economic growth. The single country (Turkey) estimates from [80] analysis shows that renewable energy reduces the ecological footprint in the long run; surprisingly, the results documented that non-renewable energy and economic growth positively impact on the ecological footprint.

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

This study examined the nature of the relationship between CO<sup>2</sup> emissions, energy consumption, agriculture and economic growth for some selected [34] Africa economies. Though Africa is made up of 54 independent countries, the selection of countries is largely influenced by data availability. The collected data cover the period 1980–2019. This period and the countries covered allow for examination of convergence issues inherent in the literature with adequate geographical covering of the African continent. The variables employed are annual data of GDP per capita (constants are 2010 and USD); CO<sup>2</sup> emissions per capita (metric tons); EC representing energy consumption; agriculture proxy by agricultural value added (AVA) per capita contribution of agriculture to GDP; and forest area (forest area as percentage of total land mass). The variables are expressed in natural forms such that *InCO*2, *Inγ*, *InEC*, *InAVA*; *InFoR* represent carbon emissions, economic growth, energy consumption, agricultural value chain and forest area, respectively. The data for the study are sourced as follows: CO<sup>2</sup> and RGDP from World Development Indicators (various issues), agriculture value addition and forest areas from Food and Agricultural Organization (various issues), and energy consumption data were from the OECD.

#### *Methodology*

As stated earlier, the study employed a frequency domain analysis to examine the relationship among energy, economic growth, and carbon emissions with the moderating impact of agriculture. Our preference of frequency domain estimates over time domain techniques is largely influenced by the weakness noticed in time domain estimates. For instance, time domain estimates cannot examine causality at different frequencies as they can only calculate a single test statistic over time [85–87]. Further, if the nexus among the variables is connected to more than one frequency, the ability of time dimension estimate to explore the information from the original data set becomes ineffective [88,89]. To overcome this, Geweke (1982) developed the Wald test procedure that employed linear constraints on coefficient parameters to test Granger causality in a certain frequency range. This procedure was extended by [90,91] as single country frequency domain causality test [85]. The [91] single country frequency domain causality test was further extended to a multicountry model by [92]. This extended frequency domain (panel Granger causality test) allows us to determine if the predictive power is concentrated at quick or slow fluctuating components. The current study aims at examine the nexus between the variables using both single-country and multi-country causality tests by following [85,93–95]. The tests are thus presented.

#### **Single-Country Causality Test:**

We begin our single country causality test by following [2] Gorus and Aydin 2019 specification of the [90] single test procedure stated as follows:

$$X\_t = \sum\_{j=1}^p \theta\_{11.j^{X\_{t-j} + \sum\_{j=1}^p \theta\_{12.j^{Y\_{t-j} + x\_{1t}}}}} \tag{1}$$

Here, *θ*<sup>11</sup> and *θ*12, are the coefficients of the polynomials, *ε*1*<sup>t</sup>* represents the error term, *p* represent the lag length, the constraint is on the first VAR, we express the constraints on the null hypothesis of "no Granger causality from *Y<sup>t</sup>* to *X<sup>t</sup>* at the frequency *w*" as stated below:

$$\sum\_{j=1}^{p} \theta\_{12,j} \cos(jw) = 0,$$

$$\sum\_{j=1}^{p} \theta\_{12,j} \sin(jw) = 0. \tag{2}$$

To test these constraints, we employed the incremental *R* <sup>2</sup> measurement test, calculated as follows:

$$R\_I^2 = R^2 - R\_\*^2 \tag{3}$$

Here, *R* <sup>2</sup> and *R* 2 <sup>∗</sup> are derived from the unrestricted and restricted models, respectively. (\*\*) The null hypothesis is rejected if this condition is observed:

$$R\_I^2 > F\_{(2T-2p, 1-\alpha)} \frac{2}{T-2p} \left(1 - R^2\right) \tag{4}$$

#### **Multi-Country Causality Test:**

Following [92], the study employed the seemingly unrelated regression (SVR) model stated as follows:

$$X\_{i,t} = \sum\_{j=1}^{p} \beta\_{i,j} X\_{i,t-j} + \sum\_{j=1}^{p} \gamma\_{i,j} Y\_{i,t-j} + \varepsilon\_{i,t,\ i=1,2,3,\dots,N} \tag{5}$$

Here, *Xi*,*<sup>t</sup>* and *Yi*,*<sup>t</sup>* are the variables of country *i* at time *t*, *p* is the lag length, *N* represent the number of countries and *εi*,*<sup>t</sup>* represents the error term at time *t* of country *i*. The null hypothesis constraints are expressed as follows:

$$\sum\_{j=1}^{p} \gamma\_{i,j} \cos(jw) = 0, \; i = 1, \; 2, \; 3, \; \dots, \; N$$

$$\sum\_{j=1}^{p} \gamma\_{i,j} \sin(jw) = 0, \; 1, \; 2, \; 3, \; \dots, \; N. \tag{6}$$

We tested these constraints using the incremented *R* <sup>2</sup> measured test, expressed as follows:

$$R\_I^2 = R^2 - R\_\*^2 \tag{7}$$

Here, *R* 2 represent the unrestricted and *R* 2 ∗ represents the restricted McElroy *R* <sup>2</sup> value expressed as follows:

$$R\_I^2 > F\_{\text{(2N, }N(T-2P), 1-\infty)} \frac{2N}{N(T-2p)} \left(1 - R^2\right) \tag{8}$$

We rejected the null hypothesis of no Granger causality from *Y<sup>t</sup>* to *X<sup>t</sup>* at the frequency '*w* 0 in the studied countries if Equation (8) was observed.

#### **4. Results**

The descriptive statistics and normality results of the variables employed in this study are presented in Table 2. The results suggested that the value of the Jarque-Bera statistics was greater than 5% for the variables, suggesting validity of normality in each of the variables studied.

**Variables Descriptive Analysis Normality Analysis (Natural Log-Form) Mean Max. Min. SD Skewness Kurtosis Jarque-Bera Probability** *Inγ* 175.98 298.77 142.67 39.09 −0.78 2.44 4.97 0.07 *InEC* 63.18 28.07 32.62 32.12 −0.48 2.14 4.22 0.06 *InAVA* 158.78 197.09 102.11 28.09 −0.55 3.09 498 0.08 *InCO*<sup>2</sup> 1.97 2.41 1.66 0.31 0.17 1.55 3.21 0.22

*InFOR* 2.99 4.01 1.98 0.55 0.05 1.61 2.76 0.22

**Table 2.** Descriptive statistics of the variables.

Source: Authors' computations 2022.

The results of both the cross-section dependency (CD) tests and the panel unit root tests are presented in Table 3. We began our analysis by investigating the cross-section dependency (CD) of the series, followed by conducting a check on the stationary properties of the series using the panel unit root test. The result in Table 3 suggest that cross-sectional dependency exists among the variables. This implies that shocks in any of the economies study can affect any of the rest. Having established cross-sectional dependency, we employed the cross-sectional augmented Dickey-Fuller test developed by [96], which is effective in detecting stationary properties of panel data as used in the current study [85,94,95]. The results suggests that *Inγ* and *InAVA* are stationary at the first different I(1), and that *InEC*, *InCO*2, and *InFOR* are stationary at their level value I(0).

**Table 3.** Cross-section dependence and panel unit root tests for the series.


Note: \*\*\* and \*\* suggest the rejection of the null hypothesis at 1% and 5% significance level, respectively. CIPS Statistics provides the simple average of the individual CADF statistics (*CADF<sup>i</sup>* ).

#### **5. Discussion**

#### *Frequency Domain Results*

As earlier stated, the study intends to examine the nature of relationship among energy, economic growth, carbon emissions, forests, and agricultural added value at three (3) clear

frequencies: short, intermediate and long run denoted as 2.5, 1.5 and 0.5, respectively. Results in the long run (0.5) implies that a permanent causality exists while the results in the short run (2.5) suggest temporary causality exists. In Tables 4–10, we present the results of the frequency domain causality based on single-country estimates. Table 4 presents the results of the link between economic growth and CO<sup>2</sup> emission for each of the 34 African economies. The results as presented suggest that a unidirectional (at the three spectra) causality runs from economic growth to CO<sup>2</sup> emission for Algeria, Angola, Benin, Burkina Faso, Ghana, Kenya, Morocco, Nigeria, Senegal, South Africa, and Zambia. The findings are in line with [13,29,61], but contradict [17,18,67] The results further reveals that a oneway causality both at the intermediate and long run is noted to exist from emission to economic growth for Congo, Madagascar, Mali, Rwanda and Zimbabwe. The results from the rest of the economies studied suggest that no link can be established between CO<sup>2</sup> and economic growth. This finding supports the validity of the neutralization hypothesis in these economies; thus, emission curbing policies can be applied in these economies. The results from Algeria, Angola, Benin, Burkina Faso, Ghana, Kenya, Morocco, Nigeria, Senegal, South Africa, and Zambia suggest that environmental protection laws could be harmful to the economy.

In Table 5, we present the results of the link between energy consumption and economic growth for the selected African economies. The results suggest that a bi-directional relationship exist between the two for the economies of Algeria, Ghana, Kenya, Morocco, and Nigeria (at the three periods), South Africa (at intermediate and long run), Egypt (at the short run and intermediate), and at least one for each of Cameroon, Guinea, and Madagascar. These results support the validity of the feedback hypothesis in these economies. The results further reveal that an un-directional causality runs from economic growth to energy consumption for the economies of Mozambique, Namibia, Tanzania and Uganda in the short run, this suggests that the conservation hypothesis is rational in these economies. The growth hypothesis is validated based on the existence of causality from economic growth to energy consumption for the economies of Algeria, Ghana, Kenya, Morocco and Nigeria. The results are in line with the findings of [30,33,74].

Table 6 presents the results of the nexus between energy consumption and CO<sup>2</sup> emissions in the studied economies. The results reveal that energy consumption Granger causes carbon emissions in Nigeria, Algeria, Egypt, Tunisia and Ghana, suggesting that the pollution haven hypothesis is valid for these economies at short, intermediate and long runs. The results support the findings of [65] but disagree with [55].

The results of the causality between economic growth and agricultural value addition, as presented in Table 7, suggest that bi-directional causality is noted for almost all the studied economies at the short, intermediate, and long runs. The result is not surprising because agriculture constitutes the bulk of African GDP.

Table 8 shows that for most the studied economies, a unidirectional relationship runs from forestry to economic growth; this suggests that wood sourced from the forest support economic growth in the studied economies.

In Table 9, we present the results of the relationship between energy consumption and agricultural value addition across the three spectra of our analysis. The results reveal that there is a unidirectional causality from energy consumption to agricultural value addition in Egypt, Ghana, Tunisia and Uganda, whereas a bi-directional causality is documented for the economies of Nigeria, South Africa, Angola. This suggests that the feedback hypothesis is validated based on the relationship between energy consumption and agriculture in these economies. The results of the relationship between forestry and energy consumption are almost the same with those of agriculture and energy consumption, except that a one-way causality is noted to exist between forestry and energy consumption, suggesting the validity of the conservative hypothesis in these economies.

Table 10 we present the results of causality between CO<sup>2</sup> emission and agricultural value addition for the selected Africa economies. Our results reveal that no causality exists between these variables for the economies studies.


**Table 4.** Granger causality tests in the frequency domain estimates (*Inγ*, *InCO*<sup>2</sup> ).

\*\*\*, \*\*, \* represent 1%, 5%, 10% significant levels, respectively.


**Table 5.** Granger causality tests in the frequency domain estimates *Inγ*, *InEC*.

\*\*\*, \*\*, \* represent 1%, 5%, 10% significant levels, respectively.


**Table 6.** Granger causality tests in the frequency domain estimates *InEC*, *InCO*2.

\*\*\*, \*\*, represent 1%, 5% significant levels, respectively.


**Table 7.** Granger causality tests in the frequency domain estimates *Inγ*, *InAVA*.

\*\*\*, \*\*, \* represent 1%, 5%, 10% significant levels, respectively.


**Table 8.** Granger causality tests in the frequency domain estimates *Inγ*, *InFOR*.

\*\*\*, \*\*, \* represent 1%, 5%, 10% significant levels, respectively.


**Table 9.** Granger causality tests in the frequency domain estimates *InEC*, *InAVA*.

\*\*\* represent 10% significant level.


**Table 10.** Granger causality tests in the frequency domain estimates *InCO*2, *InAVA*.

The results of the panel Granger causality in the frequency domain for all the examined African economies suggest the existence of bi-directional relationships across the three spectra between economic growth and energy consumption. The results further reveal that a one-way Granger causality runs from energy consumption to CO<sup>2</sup> emission in the studied economies. A further examination of the results also suggests that there is a causal nexus between carbon emissions and economic growth for the entire spectra studied, and that no evidence suggests that causality runs from economic growth to carbon emissions. In term of theoretical underpinning, one can deduce that the feedback hypothesis is valid for the relationship between energy consumption and economic growth in the studied African economies. This suggests that African economies could grow their economies by increasing energy consumption, and that energy consumption could also be enhanced by growing the economy, suggesting that demand for energy consumption is a booster of economic growth. For the nexus between energy consumption and CO<sup>2</sup> emission, the results suggest the validity of the pollution haven hypothesis, as energy consumption has a bi-directional relationship with growth driving carbon emissions in African economies, thus, Africa economies, while pursuing growth, should start looking at clean energy consumption. Though the results of the study suggest that no causality runs from economic growth to carbon emissions, ruling out the possibility of the pollution haven hypothesis, the existence of causality from energy consumption to carbon emissions points to the existence or potential of the pollution haven hypothesis, which could be from an indirect perspective. On the meditating role of agricultural value addition and forests, the results noted that the impact of both forests and agricultural value addition is only significant on economic growth across all the spectra, and on energy consumption in the short run. No causality is established between either of forests and agricultural value addition, and CO<sup>2</sup> emission for the studied economies.

For comparison, we conducted time domain estimates for the entire region by employing the Dumitrescu–Hurlin panel causality estimate. From the results, it could be deduced that a bi-directional relationship exists between economic growth and energy consumption, and that a one-way causality runs from energy consumption to carbon emissions. The results suggest the feedback hypothesis is valid on the nexus between energy and economic growth in Africa. The results of the one-way nexus, however, suggest that the conservation hypothesis is not valid in Africa. Unlike the frequency domain estimate, the moderating variables failed exhibit any form of causality in the time domain model.

The study has made some significant contribution to knowledge by being among the first set of studies that has examined the nexus among energy, environment and economic growth in Africa within the context of frequency domain estimate, and that calibrated the moderating roles of forest and agricultural value addition to this nexus.

#### **6. Conclusions**

The essence of this study was to examined the causal relationships between energy consumption, economic growth and CO<sup>2</sup> emission with the moderating roles of forestry and agricultural value addition in Africa, by employing both time domain and frequency domain estimates to analyzed data sourced from 1980 to 2019. The study provides both single-country and multi-country estimates of this nexus. The results of the single country estimate are at best mixed across the various frequencies. The study recommends that policymakers in the studied economies should take into consideration these empirical findings when designing policy tools to achieving the correct mix of energy that will stimulate economic growth without causing havoc to the environment.

The results of the panel Granger causality estimates in the frequency domain suggest that a bi-directional relationship exists between energy consumption and economic growth in Africa economies. This implies that to achieve economic growth, the energy sector should be enhanced, and that enhanced energy space will further drive or stimulate growth. The results further suggest the existence of a one-way causality from energy consumption to carbon emissions, ruling out the validity of the conservation hypothesis in these economies. This could be a result of heavy dependency/consumption of non-renewable energy in the region. It is therefore recommended that policymakers in this region should start looking at movement toward clean energy consumption. Our results are in line with the findings

of Aydin (2019 for OECD economies, Gorus and Aydin 2018 for MENA economies, but contradicts [33,97].

The study is not an all-inclusive one, as there are limitations, which could be areas to be considered by other studies. For instance, alternative estimation techniques could be employed, other variables like ecological footprints, macroeconomic variables like foreign direct investment, and socio-political variables, among others. Other studies could examine the cost-benefit analysis of different energy options as they relate to the environment, economic growth, among others. Future research can employ multi-criteria analyses useful for quantifying the nexus between the different components.

The global economy is moving towards adopting renewable energy with the intension of mitigating climate change and reducing CO<sup>2</sup> emissions; hence, the economies of Africa should make concerted efforts to develop their renewable energy potential to support economic growth. This is in line with the UN resolution of the 2015 Paris Agreement that by the 21st Conference of Parties (COP21) of the United Nations Framework Convention on Climate Change (UNFCCC), countries should focus on investing in sustainable energy and de-emphasizing the consumption of fossil fuel, among others. African economies are encouraged to formulate and implement policies that will encourage consumption of renewable energy technologies such as laws protecting the production and usage of domestic solar panels, wind turbine production, granting tax incentives to renewable energy investments, stimulate green bonds and investment, among others.

**Funding:** This research received no external funding.

**Data Availability Statement:** The data for the study are sourced as follows: CO<sup>2</sup> and RGDP from World Development Indicators (various issues), agriculture value addition and forest areas from Food and Agricultural Organization (various issues), and energy consumption data was from OECD.

**Acknowledgments:** We acknowledge the support of Bowen University Management for proving the APC for this article.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

## *Article* **Forecasting the CO<sup>2</sup> Emissions at the Global Level: A Multilayer Artificial Neural Network Modelling**

**Pradyot Ranjan Jena 1,\*, Shunsuke Managi <sup>2</sup> and Babita Majhi <sup>3</sup>**


**Abstract:** Better accuracy in short-term forecasting is required for intermediate planning for the national target to reduce CO<sup>2</sup> emissions. High stake climate change conventions need accurate predictions of the future emission growth path of the participating countries to make informed decisions. The current study forecasts the CO<sup>2</sup> emissions of the 17 key emitting countries. Unlike previous studies where linear statistical modeling is used to forecast the emissions, we develop a multilayer artificial neural network model to forecast the emissions. This model is a dynamic nonlinear model that helps to obtain optimal weights for the predictors with a high level of prediction accuracy. The model uses the gross domestic product (GDP), urban population ratio, and trade openness, as predictors for CO<sup>2</sup> emissions. We observe an average of 96% prediction accuracy among the 17 countries which is much higher than the accuracy of the previous models. Using the optimal weights and available input data the forecasting of CO<sup>2</sup> emissions is undertaken. The results show that high emitting countries, such as China, India, Iran, Indonesia, and Saudi Arabia are expected to increase their emissions in the near future. Currently, low emitting countries, such as Brazil, South Africa, Turkey, and South Korea will also tread on a high emission growth path. On the other hand, the USA, Japan, UK, France, Italy, Australia, and Canada will continuously reduce their emissions. These findings will help the countries to engage in climate mitigation and adaptation negotiations.

**Keywords:** CO<sup>2</sup> emission; artificial neural network model; forecasting; simulation

#### **1. Introduction**

There is wide consensus among scientists and policymakers that global warming as defined by the Intergovernmental Panel on Climate Change (IPCC) should be pegged at 1.5◦ Celsius above the pre-industrial level of warming in order to maintain environmental sustainability [1]. The threats and risks of climate change have been evident in the form of various extreme climate events, such as tsunamis, glacier melting, rising sea levels, and heating up of the atmospheric temperature. Emissions of greenhouse gases, such as carbon dioxide (CO2) are the main cause of global warming. The Kyoto protocol and the subsequent Paris climate summit have urged the global North and South to cooperate and bear the responsibility of reducing the CO<sup>2</sup> emissions together on a partnership basis. However, climate politics is often not in sync with all the agreements of the Paris climate deals. Especially, since the United States (US) is not a signatory to the Paris climate accords, the international cooperation sought between the industrialized and industrializing countries is slow. Given this broad context of looming climate change threats and the slow pace of actions on reducing CO<sup>2</sup> emissions by the countries, more scientific research must be undertaken to understand the exact nature of the threats. Knowing the level of CO<sup>2</sup> emissions by the high emitting countries in near future will provide actionable insights on climate policy. Such information will aid in fostering the cooperation talks in the

**Citation:** Jena, P.R.; Managi, S.; Majhi, B. Forecasting the CO<sup>2</sup> Emissions at the Global Level: A Multilayer Artificial Neural Network Modelling. *Energies* **2021**, *14*, 6336. https://doi.org/10.3390/en14196336

Academic Editor: Fernando Morgado-Dias

Received: 7 August 2021 Accepted: 26 September 2021 Published: 4 October 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

upcoming United Nations (UN) COP26 climate conference from 31 October–12 November 2021 in Glasgow, United Kingdom (UK).

Estimating CO<sup>2</sup> has often been done in the context of a school of thought in research, popularly known as the environmental Kuznets curve (EKC) hypothesis. This hypothesis states that environmental degradation, such as air pollution (CO2, SO2, NO2, and SPM emissions), water pollution, and solid waste generation follow an inverted-U relationship with economic growth [2,3]. During the initial level of a country's economic growth, the environmental pollution increases due to rapid expansion in economic activities, however after a threshold level of income per capita in the country is reached, the environmental quality improves because of a higher share of public funds being devoted to improving the environmental quality [4–6]. Despite the last three decades of empirical research in an attempt to estimate the turning point of this EKC, there has still not been consensus about a global turning point. However, there has been tremendous growth in terms of methodological sophistication to estimate both time-series and panel data available for various environmental pollutants and countries [7–12].

A detailed literature review has been undertaken covering the most recent published papers to present the state-of-the-art advancements in EKC studies. Most of these studies have highlighted the role of renewable energy in reducing CO<sup>2</sup> emissions. Dong et al. [13] examined the dynamic causal links among per capita carbon dioxide (CO2) emissions, per capita GDP, per capita fossil fuels consumption, per capita nuclear energy consumption, and per capita renewable energy consumption for China. They found that both nuclear energy and renewable energy play important roles in mitigating CO<sup>2</sup> emissions in both the short and long run, while fossil fuels consumption is indeed the dominant reason for promoting CO<sup>2</sup> emissions. They observed that renewable energy has a higher CO<sup>2</sup> mitigating effect than nuclear power. Kim and Park [14] from a study of 30 countries for a period of 2000–2013, suggested that a developed financial market in a country helps deploy more renewable energy and, in turn, can reduce CO<sup>2</sup> emissions. Paramati et al. [15] from panel data of G20 countries show that foreign direct investment (FDI) inflows significantly reduces CO<sup>2</sup> emissions in both developed and developing economies while stock market growth reduces in developed economies. They also found that renewable energy consumption substantially reduces CO<sup>2</sup> emissions and increases economic output across the countries in their panels.

In a study, Li et al. [16] used the data from China and Nigeria from 1991–2014 to derive the energy efficiency measures in the mining and extractive related sectors. Using several econometric time series methods, they concluded that energy efficiency in the mining and extractive-related sector and the circular economy have not translated into CO<sup>2</sup> emission reduction in both countries. However, economic growth, energy use (nonrenewable energy), and clean energy substitution (renewable energy) are essential factors in mitigating CO<sup>2</sup> emissions. Lorente et al. [17] employed a carbon emission function to investigate the relationship between economic growth and CO<sup>2</sup> emissions in five European Union countries, namely, Germany, France, Italy, Spain, and the United Kingdom, for the 1985–2016 period. They found an N-shaped relationship between economic growth and CO<sup>2</sup> emissions in the EU-5 countries. Further, they observed that renewable electricity consumption, natural resources, and energy innovation improve environmental quality. Using a panel of 20 organisations for economic co-operation and development (OECD) nations for the period, 1870 to 2014, Churchill et al. [18] found support for the EKC hypothesis for the panel as a whole with turning points in income per capita that lie between \$18,955 and \$89,540 (in 1990 US\$).

A study by Chen et al. [19] used the Chinese data for the period 1980–2014 and explored the relationships among per capita CO<sup>2</sup> emissions, GDP, renewable and nonrenewable energy production, and foreign trade. They found that there is a long-run relationship among those variables. They also found that China does not follow the EKC for CO<sup>2</sup> emissions under the influence of economic growth, non-renewable energy

production, and foreign trade. However, the addition of renewable energy production variables supported the U-shaped EKC hypothesis in the long run.

Using data for 1995–2018, pooled mean group-autoregressive distributed Lag (PMG-ARDL) estimator, and heterogeneous causality tests, Gyamfi et al. [20] failed to confirm an N-shaped EKC in the emerging seven, rather they confirm the existence of an inverted U-shaped EKC in the study countries. They suggested the increased use of renewable energy to mitigate pollutant emissions in these countries. Using the data from a study of BRICS economies for the period of 1980 to 2016, Khattak et al. [21] investigated the complex interaction between innovation, renewable energy consumption, and CO<sup>2</sup> emissions, under the EKC framework. They found that innovation activities have failed to disrupt CO<sup>2</sup> emission in China, India, Russia, and South Africa, except for Brazil. They also showed that renewable energy consumption has mitigated CO<sup>2</sup> emission in the BRICS panel, Russia, India, and China but not in South Africa. Further, except for India and South Africa, they observed the EKC hypothesis in all the BRICS economies. Employing a stochastic impacts by regression on population, affluence, and technology (STIRPAT) framework to the data for the period of 1990–2017 from West Asia and Middle East nations, Kihombo et al. [22] probed the effects of technological innovation, financial development (FD), and economic growth (GDP) on the ecological footprint (EF) controlling for urbanization. They observed that a 1% upsurge in technological innovation decreases EF by 0.01%. However, a 1% rise in FD boosts the level of EF by 0.0016%, inferring that FD stimulates ecological degradation. They also showed the EKC hypothesis in the selected countries.

In India's case, using data for a period of 1990–2015 and several time series econometric models, Kirikkaleli and Adebayo [23] found a long-run cointegration relationship between consumption-based carbon dioxide emissions and its possible determinants. They also found that public-private partnership investment in energy makes a positive contribution to consumption-based carbon dioxide emissions in the long run. Further, public-private partnership investment in energy and renewable energy consumption also significantly causes consumption-based carbon dioxide emissions at different frequency levels in the country. Using annual data from six South Asian economies for a period of 1980–2016 and autoregressive distributed lag (ARDL) regression, Murshed [24] examined the validity of the greenhouse emissions-induced EKC hypothesis, controlling for liquefied petroleum gas (LPG) consumption, FDI inflows, and trade openness. The analysis confirms the authenticity of the EKC hypothesis for Bangladesh, India, Sri Lanka, and Bhutan. They suggested fuel-diversification policies for the government's of these countries. Using the data for a period of 1995–2017 from 34 high-income countries from three continents (Asia, Europe, and America), Khan et al. [25] explained the nexus of GHG emission with tourism, financial development index, energy use, renewable energy, and trade. They observed a country-level reciprocal connection of GHG with financial development in 11 countries, renewable energy in 22 countries, trade openness in five countries, and tourism in 12 countries. Using two-panel data sets of 17 major developing and developed countries as well as six geo-economic regions of the world during 1990–2014, Yao et al. [26] examined the dynamic relationship between renewable energy consumption rate (RER) and the EKC hypothesis. Using several econometric methods, they verified both the EKC and renewable energy Kuznets Curve (RKC) hypotheses, indicating that a 10% rise in RER would lead to a 1.6% carbon emission reduction. Saleem et al. [27] used the data for a period of 1980–2015 from selected Asian countries and employing several econometrics models, found the presence of an EKC hypothesis, where the impact of GDP growth and the square of GDP growth on CO<sup>2</sup> emissions are positive and negative, respectively. They also found that lower-income economies do not support the EKC hypothesis.

Employing the second-generation panel cointegration methodologies and data for 1984–2016, Ahmad et al. [28] analyzed the linkages between natural resources, technological innovations, economic growth, and the resulting ecological footprint in emerging economies. They observed the existence of slope heterogeneity across countries and correlation amongst cross-sectional units. They also found a stable, long-run relationship

between the ecological footprint, natural resources, technological innovations, and economic growth. Another study in India by Usman et al. [29] studied the role of energy consumption and democratic regimes in the environmental degradation function for a period of 1971–2014. Using different time series econometric models, they confirmed the EKC hypothesis and divulged that energy consumption increases environmental degradation both in the long and short run. They suggested prioritizing energy conservation policy to mitigate environmental degradation and spur economic growth. Using data from 25 manufacturing subsectors in 38 countries from 2000 to 2014 and using an endogenous finite mixture model, Yang et al. [30] probed the effect of renewable energy in the EKC relationship. They found that with the growing impact of renewable energy consumption, nearly half of the sample countries and two-thirds of the subsectors have experienced the transformation of the nexus between manufacturing growth and emissions. Bilgili et al. [31] employed the panel quantile regression technique on a dataset from thirteen developed countries over the period 2003–2018 to find an inverted U-shaped nexus between economic growth and carbon emissions only in higher carbon-emitting countries, thus, confirming the EKC hypothesis. However, the U-shaped nexus is more predominant in lower carbon-emitting countries. They also found that energy efficiency research and development is more effective in curbing carbon emissions than fossil fuels and renewable energy research and development.

The literature review shows that significant advancement has taken place in the study of EKC in terms of the methods used. In particular, the dynamic time-series and panel cointegration models with the use of structural breaks have produced credible evidence. However, these dynamic time-series models used mostly the lag length to make the model dynamic and estimate the long-run relationship. Moreover, the time series or panel data estimations produce a single estimated parameter for the relationship within the whole sample period. The long-run relationship between CO<sup>2</sup> emissions and its predictors, such as GDP per capita, renewable energy consumption, and trade openness may not have been linear as the previous studies with statistical methods had tried to estimate. A few of the studies used the structural breaks to account for the major shifts in the environmental regulations and policies that may have affected the long-run relationship, but they finally showed constant estimates in observing the effect of GDP on environmental degradation for the whole time period. If the apparent nonlinearities existing in this relationship over a period of time are considered explicitly, more accurate predictions can be made, which has been done in the current study.

The current study aims to forecast the level of CO<sup>2</sup> emissions for 2017–2019 at the global level. CO<sup>2</sup> emission is the key contributor to climate change and there is a global consensus that the mean global surface temperature must be contained at 1.5 degrees C above the pre-industrial level. Consequently, several countries have signed the Paris agreement to reduce emissions within their national boundaries. Against this backdrop, it is essential to forecast the CO<sup>2</sup> emissions levels in the countries that emit a higher share. Such forecasting will help the national governments to adjust their climate policies.

Forecasting of CO<sup>2</sup> emissions at business as usual (BAU) scenario is a necessary tool for major greenhouse gas emitting countries for two main reasons. First, the global circulation models that are used to assess the physical impacts from climate change needs emissions as inputs. Since the countries included in this study are responsible for 79% of global emissions, forecasts of their emission level in the short run will be essential to gauge the impacts of climate change at the global level. Second, the responsibility to reduce CO<sup>2</sup> emissions as agreed at the Paris climate convention is proportional to the BAU levels of emissions. Hence, accurate prediction of emissions will put the right value of resources that these countries need to commit for the reduction of emissions. Since there is a trade-off between emission reduction and economic growth, these countries will be anxious that their emission levels are not underpredicted. Some of the countries may withdraw from a multilateral climate treaty if they find that they are at an economic disadvantage due to their pledge to reduce emissions. Accurate prediction of the BAU emission levels holds significance for a feasible action plan by the countries to reduce the global CO<sup>2</sup> emissions.

Considering that there might be a nonlinear relationship between the indicators of economic growth and the CO<sup>2</sup> emissions, we develop a multilayer artificial neural network (MLANN) model. A multilayer artificial neural network model is more efficient in capturing the nonlinearity present in the time series data and provides higher accuracy in forecasting the CO<sup>2</sup> emissions based on the past values of the emissions and the economic indicators, such as GDP, population density, and urbanization. Such forecasts for the near future will provide insights into regulations on pollution control.

The contributions of the paper are:


The rest of the paper is organized as follows. Materials and methods are discussed in Section 2. Section 3 deals with the development of a CO<sup>2</sup> prediction model using MLANN. Details of the simulation study are given in Section 4. It also contains data collection and preprocessing, training and testing of the model. Section 5 presents results and discussion. Finally, conclusion of the paper is presented in Section 6.

#### **2. Materials and Methods**

We have considered two types of countries—first, countries that emit 2% or more share of global CO<sup>2</sup> emissions and countries that emit less than 2% share. The selection of countries in this study is based on the data compiled by the International Energy Agency (IEA), which estimates carbon dioxide (CO2) emissions from the combustion of coal, natural gas, oil, and other fuels, including industrial waste and non-renewable municipal waste. The specific data used are reproduced from the website Each Country's Share of CO<sup>2</sup> Emissions | Union of Concerned Scientists (ucsusa.org) and are given below in Figure 1.

Table 1 describes the countries considered in this study under two groups—high emission and low emission countries.


**Table 1.** High and Low emission countries.

**Figure 1.** Share of CO<sup>2</sup> emissions in high and low emitting countries.

The data on the output parameter, i.e., CO<sup>2</sup> emissions and the input parameters, such as GDP in constant US\$ measured in 2011, trade as a percentage of GDP, and urban population for all the countries are drawn from the World Bank database. The period of the study is 1960 to 2016. The forecasting period is 2017, 2018, and 2019.

Figure 2 shows the GDP (constant in 2010 US\$) for the countries considered in this study, in 1990 and 2016. Although the period of study is from 1960 to 2016, we chose the more recent years to compare the growth of the GDP. The countries shown in the *X*-axis are ordered from the highest emission status to the lowest among the 17 countries. The *Y*-axis shows the cumulative annual growth rate (CAGR) between 1990 and 2016. The countries showing a high growth rate in GDP are expectedly China with a CAGR of 9.5%, India with 6.2%, Indonesia with 4.8%, and Turkey with 4.4%. the countries that experienced low growth rates are Italy with 0.7%, Japan with 0.96%, and France with 1.56%.

**Figure 2.** The GDP figures and its growth for the selected countries.

Figure 3 shows the CO<sup>2</sup> emissions of the 17 countries and their CAGR for the period 1990 and 2016. The countries that accounted for the highest growth in CO<sup>2</sup> emissions between 1990 and 2016 are China with 6.17%, India with 5.5%, Saudi Arabia with 5%, Iran with 4.8%, Brazil with 4%, and Turkey with 3.7%. The countries that have managed to rein in their emissions growth are UK with −1.16%, Italy with −1.1%, France with −0.88%, the USA with 0.33%, Japan with 0.41%, and Canada with 0.9%. The growth trends in Figures 2 and 3 suggest that the highly developing countries tend to emit more CO<sup>2</sup> while the already developed countries have slowed down their emissions. This evidence for the period 1990–2016 is close to the assertions of the EKC.

However, future forecasts are needed to convince the developed countries to commit more financial support for the developing countries to motivate the latter to sacrifice some of their economic ambitions. The trade-off that the highly emitting developing countries, such as China, India, and Brazil have to accept to reduce their CO<sup>2</sup> emissions in order to comply with their commitments at the Paris climate summit agreement, is substantial. Unless they receive financial support from the industrialized countries as agreed upon by the Paris climate summit, these countries are unlikely to reduce their emission levels. We attempt to forecast the CO<sup>2</sup> emission levels of 17 countries that account for nearly 79% of the global emissions. By using the highly complex and non-linear artificial neural network (ANN) models that can accurately forecast the future emission values, we provide actionable insights to the policymakers to engage in more active dialogues to achieve the Paris agreement. Using the multilayer ANN model, we forecast the CO<sup>2</sup> emissions for Group 1 and 2 types of countries (Table 1) for 2017–2019.

**Figure 3.** The CO<sup>2</sup> emission (kt) and its CAGR for the selected countries.

#### **3. Development of the Multilayer Artificial Neural Network (MLANN) Based CO<sup>2</sup> Forecasting Model**

Statistical models are not able to estimate the relationships accurately when the data are uncorrelated, non-stationary, nonlinear and chaotic [32]. To overcome this problem various intelligent models are proposed by the researchers. The MLANN is a nonlinear, multi-layered, fully connected feedforward network that can model the nonlinearity of the data appropriately [33]. The MLANN model is trained using past data and optimizes the weights that will be used to forecast the CO<sup>2</sup> emissions based on the inputs given. The flowchart shown in Figure 4 is used for the development of a MLANN based prediction model.

The complete structure of the MLANN based prediction model is given in Figure 5. Let *I*, *JandK* represent the indices for the input, hidden and output layers respectively. Where *I* = the number of inputs, *J* = the number of neurons in the hidden layer, and *K* = the number of neurons at the output layer. In this CO<sup>2</sup> prediction model the output is one value, so for this study the value of *K* = 1. Let *P* be the number of input patterns and let any *i*th input pattern is given as *p<sup>i</sup>* . Each input pattern is supplied to the MLANN model sequentially, multiplied with the weights, sum together, and finally passed through the nonlinear activation function (*tanh*) to produce the output at the first hidden layer. This process is repeated for the next hidden layers and output layer. Let the estimated output of the network is *est<sup>k</sup>* . The error value is obtained by comparing the estimated value with the desired value or target value, *t<sup>k</sup>* . The backpropagation learning rule [33] given in Equations (6)–(11) is used to update the weights and bias values of each layer. This process continues until the squared error is minimum. The detailed equations of feed-forward computation and rules to update the weights and bias are discussed below.

**Figure 4.** Methodology of MLANN based CO<sup>2</sup> prediction model.

**Figure 5.** A MLANN based CO<sup>2</sup> emission prediction model.

Refereeing to the above figure, the output of the *k*th output neuron *est<sup>k</sup>* is given [33] as:

$$est\_k = \tan h(h\_k) \tag{1}$$

where

$$h\_k = \sum\_{j=1}^{I} est1\_j w\_{kj} + w\_{bk} \tag{2}$$

*est*1*<sup>j</sup>* = the output obtained at *j*th hidden neuron.

*wkj* = weights connecting *j*th hidden neuron and *k*th output neuron.

*wbk* = bias at *k*th output neuron.

In the same way, the output at neuron of *j*th hidden layer, *est*1*<sup>j</sup>* is computed [33] as—

$$est1\_j = \tan h(h\_j) \tag{3}$$

where

$$h\_{\dot{j}} = \sum\_{i=1}^{I} p\_i w\_{\dot{j}i} + w\_{b\dot{j}} \tag{4}$$

*p<sup>i</sup>* = *i*th *input pattern wji* = *weights between i*th *input and j*th *hidden neuron wbj* = *bias at j*th *hidden neuron*

The error value is obtained by comparing the output of the prediction model, *est<sup>k</sup>* with the corresponding target value, *t<sup>k</sup>* . So,

$$e\_k = t\_k - \varepsilon t\_k \tag{5}$$

The weights connecting the neurons of hidden and output layers, *wkj* are updated [33] by:

$$w\_{k\dot{j}} = w\_{k\dot{j}} + \mu \times \delta\_k \times \text{est1}\_{\dot{j}} \tag{6}$$

where

$$\delta\_k = e\_k \times \frac{\left(1 - est^2\_k\right)}{2} \tag{7}$$

*µ* = *learning parameter*,(0 < *µ* < 1)

The bias weight is updated as:

$$w\_{bk} = w\_{bk} + \mu \times \delta\_k \tag{8}$$

Similarly, the weights connecting the input and the hidden layer neurons, *wji* are updated [33] as:

$$w\_{\vec{\mu}} = w\_{\vec{\mu}} + \mu \times \delta\_{\vec{\jmath}} \times p\_{\vec{\imath}} \tag{9}$$

where

$$
\delta\_{\dot{j}} = \delta\_k \times w\_{k\dot{j}} \times \frac{\left(1 - est1\right)^2}{2} \tag{10}
$$

The updating of bias weight of *jth* neuron in the hidden layer is done [33] as:

$$w\_{b\circ} = w\_{b\circ} + \mu \times \delta\_{\circ} \tag{11}$$

The Equations (1)–(11) are the key equations used in the development of the MLANN based CO<sup>2</sup> forecasting model.

#### **4. Simulation study**

The CO<sup>2</sup> emissions prediction is formulated as an optimization problem. The model is having three inputs and one output. The inputs are fed to the model and the obtained output is compared with the available target value until the squared error is minimum. Matlab 2016 package is used for the simulation of the problem.

#### (a) Data collection and preprocessing:

The data for 9 countries under Group 1 and 8 countries under Group 2 are collected from 1960 to 2016 till which the comprehensive data are available in the World Bank database. CO<sup>2</sup> emissions are used as the output parameter for which the out-of-sample forecasting has been done. The variables, such as GDP (in constant 2010 US\$), trade ratio, and urban population are used as the inputs in the MLANN model. The data has been preprocessed as the first step in the modeling wherein the data for all the variables have been normalized. During normalization, each value of the four variables is divided by the corresponding maximum value so that all values can lie between 0 and 1. Normalized data generally makes the learning process and the convergence speed faster. If all features do not have similar ranges of values, then the gradients can move to and fro and take a long time before they can attain the global minimum. To circumvent this problem in the learning process, normalization of the data is necessary. Normalization of the data is followed by preparation of training set and testing set. Randomly selected 80% of data are used for training of the model and the rest 20% of data is used for testing of the developed model. Simulation is carried out by varying the ratio of data division (70:30, 80:20, and 90:10) and an 80:20 ratio is selected finally as it gives the best result. Further, the three missing values of CO<sup>2</sup> emission for the year 2017–2019 are calculated using the optimized weights of MLANN based model.

(b) Training of the model:

Out of the total of 57 patterns (1960 to 2016), the training set consists of 46 patterns that are randomly chosen, and the remaining 11 patterns are used for testing of the calibrated model. An input pattern of data consists of the values of trade ratio, urban population, and GDP. The corresponding CO<sup>2</sup> emission value is the target value for the training of the model. A 9:3:1 structure is used for the simulation. It consists of two hidden layers with nine and three neurons respectively. The connecting weights between the layers and the bias weights are randomly initialized to lie between −0.5 to 0.5. The 9:3:1 structure is fixed after doing experiments by varying different structures of MLANN as it gives minimum error value. In each iteration, one input pattern is given to the model, and feedforward processing is done to get the estimated output from the model. Feedforward processing involves summing the weighted inputs, adding the bias weights, and then passing it through the activation function or nonlinear function (tanh). The estimated output is compared with the corresponding target value to obtain the error. The backpropagation (BP) training algorithm is used to update the connecting weights and bias weights. The value of the learning parameter is taken as 0.1. The same process is repeated until all training patterns are exhausted. This completes one experiment. The experiment is repeated until the mean squared error (MSE) is minimized. The MSE value for each experiment is stored and plotted to observe the convergence characteristics. The final value of connecting weights and bias weights are frozen for testing of the developed model.

#### (c) Testing of the model:

Once the training process is complete, the developed MLANN based model is ready to be used for evaluation. The 20% of the testing patterns are applied to the model sequentially and the estimated output is noted. The estimated output is compared with the corresponding desired value and the mean absolute percentage error (MAPE), mean absolute error (MAE) and root mean square error (RMSE) are tabulated in Tables 2 and 3 which indicate the performance of the model. The MAPE, MAE and RMSE are calculated using Equations (12)–(14). Also, the comparison of the actual and estimated CO<sup>2</sup> values during testing are plotted and exhibited in Figure 6a–i for Group-1 countries. For Group-2 countries, it is given in Figure 7a–h.

$$MAPE = \frac{1}{n} \sum\_{i=1}^{n} \frac{| (t(n) - \text{est}(n)) |}{t(n)} \times 100\tag{12}$$

$$MAE = \frac{1}{n} \sum\_{i=1}^{n} |(t(n) - est(n))| \tag{13}$$

$$RMSE = sqrt(\frac{1}{n}\sum\_{i=1}^{n} \left(t(n) - est(n)\right)^2) \tag{14}$$

$$MAPE = \frac{1}{n} \sum\_{i=1}^{n} \frac{|(Out\_{obs} - Out\_{est})|}{Out\_{obs}} \text{ where } t(n) = \text{target value}$$

$$est(n) = \text{estimated value}$$

$$n = \text{Number of testing patterns}$$

**Table 2.** MAPE, MAE and RMSE values obtained during testing for Group-1 countries.


**Table 3.** MAPE, MAE and RMSE values obtained during testing for Group-2 countries.


**Figure 6.** Comparison of the actual and estimated value of CO<sup>2</sup> emissions using the MLANN for Group-1 countries during the testing of the model.

**Figure 7.** *Cont*.

**Figure 7.** Comparison of the actual and estimated value of CO<sup>2</sup> emission using MLANN for Group-2 countries during testing.

#### **5. Results and Discussion**

From Table 2 it is observed that the MAPE values for Group-1 countries lie between 1.78 to 3.52% except for Indonesia and Saudi Arabia. The MAPE value is 5.91 for Saudi Arabia and 9.68 for Indonesia. The MAPE is an indicator of how close the predicted values are to the actual values. The RMSE values lie between 0.01 to 0.05 for Group I countries except for Indonesia. The MAE values lie between 0.01 to 0.07 for Group I countries. The MAPE values for the Group-2 countries are given in Table 3 which shows that the values lie between 1.92 and 3.8 except for Brazil and Italy. The value is 5.33 for Brazil and 8.08 for Italy. The RMSE values lie between 0.02 to 0.07 and the MAE values lie between 0.01 to 0.06 for Group II countries. As the MAPE values are less than 4% for most of the countries considered in this study, the MLANN model is able to predict the values reasonably accurately with less percentage of error except for a few cases. The comparison of actual and estimated CO<sup>2</sup> values obtained during testing is shown in Figures 6 and 7 for Group-1 and Group-2 countries respectively. In most cases, the actual and estimated values are close to each other.

However, the gap between the actual and predicted values of CO<sup>2</sup> emissions found during the testing phase of the model for Indonesia, Saudi Arabia, Brazil, and Italy is due to the wide fluctuations observed in their emissions data during the period of the study. Although the MLANN model developed in this study is robust to the nonlinearities in the

data, wide fluctuations may still increase the percentage of error as is the case for these four countries.

The simulation study is carried out by varying the ANN structure. Different combinations of hidden layer and neurons are used to simulate the model and the results in terms of the training and testing times, as well as the performance achieved, are obtained and displayed in Table S1 in Supplementary Materials. For each country data, initially, combinations of one hidden layer where five, six, seven and eight neurons are used, and thereafter two hidden layers with the same variations of neurons are used for simulation. From Table S1 in Supplementary Materials, it is exhibited that comparing the training time, testing time, MSE in training, and MAPE in testing, the proposed structure of the MLANN model is better in comparison to other combinations of hidden layer and neurons. Further, the simulation is also carried out with different data division ratios and it is observed from Tables S2 and S3 in Supplementary Materials, that the 80–20% ratio is suitable for the proposed study as it gives the minimum MAPE value in all cases.

As suggested in Wu et al. [34], other machine learning methods, such as the support vector machine (SVM) model is simulated and the resultant MAPE values are provided in Table S4 in Supplementary Materials. It is observed that the MAPE values of all countries of Group-I and Group-II are higher in comparison to the proposed MLANN model. We have not added the methods of SVM and a detailed comparison between MLANN and SVM in the main text since it will require substantial expansion of the manuscript.

#### *Forecasting of CO<sup>2</sup> Emissions*

In this section, we present the forecasted values of CO<sup>2</sup> emissions for the Group-1 and Group-2 countries for the years 2017, 2018, and 2019 given in Tables 4 and 5 respectively. These are out-of-sample forecasts of CO<sup>2</sup> emissions based on the optimized weights from the calibrated MLANN model and the values for inputs, such as GDP (in 2010 constant US\$), urban population, and trade ratio for 2017, 2018, and 2019. The data of CO<sup>2</sup> emissions for these years are not available, however, the data for inputs for these three years are available for most of the countries considered in this study except for Iran, the USA, and Japan. For Iran, input data is available only for 2017, and for the USA and Japan, it is available for 2017 and 2018. Accordingly, the forecasts are done for these countries for the years the input data are available. The EKC hypothesis stands on the empirical evidence that the elasticity of income effect is larger than the combined elasticities of scale and composition effects [35,36]. The literature review in this study has discussed many recent articles that have either established the EKC relationship in the long run or failed to find evidence for it. A few other studies have used a similar framework as EKC to forecast the out-of-sample values of CO<sup>2</sup> emissions [37]. Aufhammer and Carson forecasted the CO<sup>2</sup> emissions for the Chinese provinces for the single year of 2010 by using the estimated coefficient values of different predictors of their 'best' model and the projected values of the predictors, such as GDP per capita and population figures whose values were unknown when they published this study. Two other noteworthy studies by [38] and [39] have used a similar approach and forecasted the time path of CO<sup>2</sup> emissions for the year 2100 and 2050 respectively. We improve upon these studies in two ways. First, we develop a sophisticated neural network nonlinear model to calibrate the EKC relationship and obtain the optimized input weights that are used to predict the CO<sup>2</sup> emissions based on the predictors, such as GDP, urban population, and trade ratios. These optimized weights provide a more realistic time-series relationship between the emissions and the predictors. Secondly, we forecast the CO<sup>2</sup> emissions for high emitting and low emitting countries based on the known values of the predictors, not their projected values.


**Table 4.** Forecasted CO<sup>2</sup> emission values for the year 2017–2019 for Group-1 countries.

**Table 5.** Predicted CO<sup>2</sup> emission values for the year 2017–2019 for Group-2 countries.



Figure 8a,b depict the CO<sup>2</sup> emission values for the Group-1 countries from 2010 to 2019. The emissions from 2010 to 2016 are the actual data obtained from the World Bank database, whereas the values from 2017 to 2019 are the forecasted values. Figure 8a shows that the forecasted emissions for both China and India have increased. China surpassed the USA in 2005 and since then the rate of emission growth is substantially higher for China. During the same period of 2005–2019, the USA's emissions levels have dipped and during the short period of 2017–2019, it shows a declining trend. This is a noteworthy observation in the context of international climate negotiations. Although the USA is not a signatory to Paris climate agreements, it has its internal pollution regulation mandates that have yielded a reduction in CO<sup>2</sup> emissions. On the other hand, China has taken great strides in transforming its economic structure following a circular economy model [40]. Despite these reforms, the emission levels are expected to rise in the short-run horizon. China's past high emission levels and the high growth rate in emission will render it a high emitting country in the near future despite its significant improvement in restructuring the economic models. India is the third highest CO<sup>2</sup> emitter in the world and the rate of emission growth shows a rising trend for the country. The forecasted values for 2017–2019 signify the uphill challenge India is facing to comply with its commitments towards Paris agreements as the emission levels are expected to rise during this period. Japan's emission levels are predicted to reduce further following its declining trend that started around 2007.

Figure 8b shows that the trajectory of CO<sup>2</sup> emissions in Indonesia is quite volatile which is the reason for a higher percentage error in our forecasts for Indonesia. The forecasted values for the period 2017–2019 show arising rend for the country. The other countries in the Group-1 category that shows a rising expected level of CO<sup>2</sup> emissions are Iran, South Korea, and Saudi Arabia. Whereas Canada's emission levels have been stabilized and it embarked on a declining phase of CO<sup>2</sup> emissions since 2008. Figure 9a shows the CO<sup>2</sup> emission trajectory and the forecasted levels for the Group-2 countries. Although the global share of CO<sup>2</sup> emission in countries, such as Brazil, South Africa, Mexico, and Turkey are either 1% or less than 1%, their expected emission level will rise in the near future. Brazil, in particular, shows a high emission growth path which weakens the country's position in the future global climate summits, such as COP26. The reported burning of large tracts of Amazonian forest in Brazil has been heavily criticized by the rest of the globe. The country needs to be more proactive and engaged in complying with its Paris agreement commitments. The expected trajectory of the CO<sup>2</sup> emission growth path for the industrialized countries, such as France, the UK, Australia, and Italy are shown in Figure 9b. The emission levels in France and UK are continuously declining and are expected to decline further. Italy and Australia have reached their peak levels of CO<sup>2</sup> emission in 2006 and 2011 respectively. Since then, their emission levels have stabilized at lower levels and are expected to decline further.

**Figure 8.** The actual and forecasted CO<sup>2</sup> emissions values for countries: (**a**) China, USA, India, and Japan; (**b**) Canada, Indonesia, Iran, Saudi Arabia, and South Africa.

**Figure 9.** The actual and forecasted CO<sup>2</sup> emissions values for countries: (**a**) Brazil, South Africa, Mexico, and Turkey; (**b**) Australia, UK, Italy and France.

#### **6. Conclusions**

The IPCC report [41] warns that the current level of national pledges on mitigation of greenhouse gas emissions and adaptation to climate change are not enough to constrain

global warming to the level agreed upon by the countries in the Paris Agreement. The report urges the signatory countries to upscale and accelerates the implementation of multilevel and cross-sectoral climate mitigation actions. To be able to do so, accurate prediction of future CO<sup>2</sup> emission path in business-as-usual conditions holds importance. Such predictions would lead the countries to accelerate their mitigation and adaptation measures. This study forecasts the CO<sup>2</sup> emissions for the high and low emitting countries by their global shares of emission, for the years 2017, 2018, and 2019. Among the high emitting countries, China and India have been treading a high emission growth path, whereas the US and Japan are on the declining trend. Following the EKC hypothesis literature, we model the CO<sup>2</sup> emissions as the output of the model and GDP in constant 2010 US\$, urban population, and trade ratios as the predictors. Several past studies have used the same variables to predict the EKC relationship, however, their methods had been static and mostly linear. Considering that the relationship between CO<sup>2</sup> emissions and its predictors may be nonlinear in the long run, we develop a multilayer artificial neural network model to estimate this relationship.

Based on the World Bank database of 17 countries, of which nine are placed in high emitting (Group-1) and the remaining eight in the low emitting (Group-2) countries spanning from 1960 to 2016, a MLANN model is developed. After the model simulation, it is observed that the prediction accuracy of the in-the-sample data has been 96% leaving 4% to the prediction error. With this high level of prediction accuracy, the model is well calibrated to forecast the out-of-the-sample emission growth path. The data for the input predictors have been available for the years 2017, 2018, and 2019 but not for the CO<sup>2</sup> emissions of the selected countries. Hence, we forecast the CO<sup>2</sup> emissions of these years based on the optimal weights and the input data. From the results, it is observed that China despite its aggressive transformation of economic activities to a circular economy model, is still on the path of increasing emissions in near future. Similarly, India will continue to emit higher levels of CO<sup>2</sup> in the short run that has been studied. Other high emitting countries, such as Iran, Indonesia, Saudi Arabia, and South Korea are expected to continue with their high CO<sup>2</sup> emission growth path if they remain on the BAU economic production-consumption trajectory. These countries need to restructure their economic activities in more sustainable ways to reduce greenhouse gas (GHG) emissions. However, the US and Japan are expected to further reduce their carbon footprint by emitting less CO<sup>2</sup> into the atmosphere. France, UK, Italy, Australia, and Canada are poised to stabilize their emission levels at a low emission growth path and are on course to comply with the Paris agreement. Finally, although low emitting countries, Brazil, South Africa, Turkey, and have been on the rising path of GHG emissions. These countries prioritize their economic growth over the reduction of CO<sup>2</sup> emissions. Hence, they are not expected to comply with the Paris agreement's emission reduction goals.

Based on these results, it is incumbent upon the national policymakers and multilateral policy supporting bodies, such as the UN, OECD, World Bank, and IMF to commit more financial resources for the reduction of CO<sup>2</sup> emissions. Most of the countries that we studied that are on a high emission growth path are currently industrializing. Their goal is to achieve higher economic growth, create more employment, and increase income per capita. Hence, these countries are less likely to change their economic structure suitable for a low carbon economy. The already industrialized countries who have achieved a reduction in their national CO<sup>2</sup> emission goals must come forward to support the countries who are not close to achieving the pledges they made at the Paris climate conference. The next multilateral climate summit which is scheduled to take place in the UK in October-November 2021, known as COP26 will have to focus on issues of greater climate cooperation and finance.

The MLANN model used in the study though has forecasted the CO<sup>2</sup> emission quite accurately in most cases, there are a few cases where the prediction error was high. This is a limitation of the study. Future studies can use other ANN-based models like radial basis function neural network (RBFNN), recurrent neural network (RNN), extreme learning machine (ELM), etc., to reduce the percentage of error. Further, the scope of this study can be expanded by using the mean impact value (MIV) based method to select features and by using the optimal lag order of input data as suggested by Lee and Ou [42] and Wu et al. [43].

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/en14196336/s1, The Tables S1–S4 are available as Supplementary Materials.

**Author Contributions:** Conceptualization, P.R.J. and S.M.; methodology, B.M., P.R.J. and S.M.; software, B.M.; validation, B.M., P.R.J. and S.M.; formal analysis, P.R.J. and B.M.; investigation, P.R.J. and S.M.; data curation, B.M.; writing—original draft preparation, P.R.J., B.M. and S.M.; writing—review and editing, P.R.J. and B.M.; supervision, P.R.J.; project administration, P.R.J.; funding acquisition, S.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available in Supplementary Materials.

**Acknowledgments:** We duly acknowledge the excellent technical support provided by Purna Chandra Tanti, Sunil Khosla, and Loshini to complete this manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **The Dynamics of US Gasoline Demand and Its Prediction: An Extended Dynamic Model Averaging Approach**

**Sakar Hasan Hamza \* and Qingna Li**

School of Mathematics and Statistics, Beijing Institute of Technology, Beijing 100081, China

**\*** Correspondence: hassan.sakar.math@gmail.com

**Abstract:** This study contributes to the body of literature on modeling and predicting gasoline demand by using nonlinear econometric techniques. For this purpose, dynamic model averaging (DMA) and Bayesian model averaging (BMA) combined with Artificial Bee Colony (ABC) are used to forecast gasoline consumption in the United States. The article's independent variables include demographic characteristics, economic activity, income, driving expenditures, automobile price, and road availability for annual data from 1960 to 2020. In the proposed model, not only may the coefficients and elasticity of a predictor of gasoline demand change over time, but other sets of predictors can also emerge at different periods. Moreover, this study aims to automate the process of picking two forgotten variables of the DMA model using the ABC model. Our findings indicate that dynamic model averaging significantly improves forecasting performance when compared to basic benchmark techniques and advanced approaches. Additionally, integrating it with an Artificial Bee Colony (ABC) may result in improved outcomes when time-varying forgetting variables are present. The findings of this research provide policymakers in the fields of energy economics and the environment with helpful tools and information.

**Keywords:** gasoline demand; dynamic model averaging (DMA); artificial bee colony (ABC); time-varying parameter; dynamic model

#### **1. Introduction**

Gasoline demand in the United States has been steadily increasing since the 1990s. In 2019, Approximately 143 billion gallons of gasoline was used in the United States, with the transportation sector accounting for over 70% of the total consumption [1]. The demand increase can be attributed to factors, such as population growth, urbanization, and increased consumer spending on vehicles. Furthermore, the EIA [1] reported that gasoline demand is highly sensitive to changes in economic activity, fuel prices, and weather patterns. For example, during the COVID-19 pandemic in 2020, gasoline demand in the US fell significantly due to reduced economic activity and stay-at-home orders. However, as the economy recovers and restrictions are lifted, demand is expected to increase once again. In addition, the EIA [1] predicts that gasoline demand will continue to rise in the coming years, reaching approximately 151 billion gallons by 2050.

Numerous research has been conducted on the effectiveness of gasoline demand factors and their capacity to forecast. In this context, some previous studies have adopted a direct approach to estimation by examining the demand for car sales [2–5]. Apart from forecasting vehicle sales, research in the area of travel demand has also looked at gasoline use as a response variable when evaluating fuel price elasticities [6–9]. Huo and Wang [5] discovered that pricing and income elasticities in China are based on consumer vehicle stock and projected vehicle sales in China up to 2050 using the FEEI model. Bento et al. [10] conducted similar research for the United States, using a simultaneous equations model for US households and taking into account the new discarded vehicle markets, among other factors. Graham and Glaister [9] conducted a thorough literature review of 113 studies

**Citation:** Hamza, S.H.; Li, Q. The Dynamics of US Gasoline Demand and Its Prediction: An Extended Dynamic Model Averaging Approach. *Energies* **2023**, *16*, 4795. https://doi.org/10.3390/en16124795

Academic Editor: George Halkos

Received: 10 March 2023 Revised: 28 April 2023 Accepted: 4 May 2023 Published: 19 June 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

conducted in the United Kingdom. Goetzke and Vance [11] and Bento et al. [10] found comparable results in terms of fuel consumption's reactivity to fuel prices and in contrast to vehicle mile travel's response to fuel prices. Meanwhile, Oladosu [6] described individual family fuel consumption choices using a vehicle–fuel expenditure allocation model (or AIDS model) for multi-vehicle families in the United States.

In addition, there is a lack of consensus on the estimated coefficients in the studies that have been carried out in the process of modeling and predicting gasoline demand. According to Goetzke and Vance's [11] review of the literature, the average gasoline price elasticity is roughly −0.18, with estimates ranging from −1.01 to 0.01. Thus, the majority of studies interpret these fuel price elasticities as evidence for the existence of a rebound effect, in which the cost savings associated with a reduction in the cost of driving or a gain in fuel efficiency eventually result in an unforeseen rise in fuel consumption. The rebound effect describes a situation in which drivers are presented with lower travel costs (such as falling gas prices) and/or increased fuel efficiency, which unintentionally results in increased fuel consumption and/or vehicle travel. Consequently, in terms of policy, such a phenomenon might result in erroneous calculations and, thus, incorrect interpretations for decision makers. Dimitropoulos et al. [12] conducted a meta-analysis of 74 studies that included 1120 estimates of reported rebound effects and discovered an average rebound effect of 10% to 12%. This unintended impact on driving and fuel consumption habits has significant consequences for the efficacy of policy planning and interventions aimed at reducing emissions and fuel consumption. Overall, studies in the transportation literature employ a variety of methods in terms of model selection, with the majority of scholars being aware of obvious causes for observed discrepancies in the findings. As a result, the effective modeling and forecasting of gasoline demand may offer a critical foundation for policymakers to consider the policy implications of their energy market activities, which is the goal of this paper.

At a minimum, typical forecasting models have two shortcomings: first, numerous studies have shown that predictors change over time, and factors, such as market cycles and macroeconomic policy changes, may result in structural breakdowns in the relationship between fundamental principles and dynamics. Additionally, the effect of each input on the dependent variable changes according to the period and market conditions [13,14]. A model with a static list of predictors may also lose accuracy and consistency over time. Extensive and precise analysis may be performed at any time to pick a model. In other words, if we have *N* predictors, we must evaluate and compare, 2 *<sup>N</sup>* models at each time point (the number of subsets of *N* variables that accurately represent all possible combinations and inclusions of *N* variables in the model) with *T* × 2 *<sup>N</sup>* as the total number of models should be tested throughout T. Therefore, while N and T are large, their analysis is impossible or, at least, difficult.

The accuracy of forecasts has been improved by using model averaging approaches, such as "forecasting combination", in recent research. Both "Bayesian Model Averaging" (BMA) and " forecasting combination" models are characterized by fixed weight values given to models throughout time; however, they do not offer sufficient flexibility to manage the time gap between the contributions of the modeling [15,16]. Therefore, dynamic model selection (DMS) and dynamic model averaging (DMA) were suggested by Raftery et al. [17] to overcome the limitations of the other models. Findings show that macroeconomic forecasting may benefit from this method [18,19]. The appropriateness of each model throughout time is shown in several studies on this subject. The time-varying parameter (TVP) model may employ DMA to compute the average likelihood of each variable being present in the best prediction model. As a more exact definition, one may argue that the average forecast across models is based on an average likelihood of the existence of a variable at time *t* based on prior knowledge [19–21]. Selecting the optimal prediction model is based on determining which variables have the greatest likelihood of being present in this model, and the model's prediction will be based on this calculation [19]. Although DMS picks a model that comprises variables most likely to be included in forecast models

among those estimated in each period, it does so in a more efficient manner. Inspired by the works of Koop and Korobilis [18] and Bork and Mller [22], Raftery et al. [17] found that the DMA model's forecasting accuracy was 30 percent higher than that of other timeseries approaches, such as AR and OLS regression. A DMA technique is presented by Wei and Cao [23] to predict a housing price increase in Chinese cities. Research shows that DMA is a better forecasting model than BMA, equal-weighted averaging (EW), and information-theoretic modeling. Dong and Yoon [24] employed a DMA approach to explore the global economic drivers that have a large impact on developing Asian stock market returns, notably during the financial crisis. Moreover, other applications for predicting are noteworthy: aggregate equity returns [25], commodity prices [26,27], exchange rates [28,29], Government bond yields' term structure [27], and commodity price volatility and equity return [30].

Therefore, the following is the study's primary contributions: (1) This study aims to estimate and forecast the gasoline demand in the USA using TVP techniques, particularly the DMA approach, which is much more accurate than prior methods. (2) In most investigations, Bayesian TVP is used to estimate the model's parameters [31,32]. Although this approach approximates the generation of model parameters and switching probabilities using two forgetting elements, the inclusion of forgotten factors might be helpful since full Bayesian models may be quite large and time consuming in terms of computational volume. It also assumes that the two factors are constant over time, which is not the case for the single mechanism addressed in the study by Koop and Korobilis [18]. In addition, removing this constraint to reduce the computing cost of the model may lead to an improvement in model prediction accuracy. In this study, we attempt to execute a random process of forgetting factor selection using an algorithm called the ABC. Therefore, another key contribution in this work is to integrate ABC with DMA to improve the forecast accuracy.

The remaining parts of the article are organized as described below. In the second section, a research approach is presented. In Section 3, we provide a summary of both our data and the empirical findings of the forecasting. The conclusion is presented in Section 5.

#### **2. Research Methodology**

The DMA technique employed in the study at hand was introduced by Raftery et al. [17]. The following is the standard models for State-Space approaches, namely the Kalman filter:

$$\mathbf{y}\_t = \mathbf{z}\_t \boldsymbol{\theta}\_t + \boldsymbol{\varepsilon}\_t \tag{1}$$

$$
\boldsymbol{\theta}\_{\text{t}} = \boldsymbol{\theta}\_{\text{t}-1} + \boldsymbol{\mu}\_{\text{t}} \tag{2}
$$

where θ<sup>t</sup> = h <sup>ϕ</sup>t−1, <sup>β</sup>t−<sup>1</sup> , <sup>γ</sup>t−<sup>1</sup> , · · · , <sup>γ</sup>t−<sup>p</sup> i denotes a vector of m × 1 coefficients, and µ<sup>t</sup> ∼ N(0, Q<sup>t</sup> ) and ε<sup>t</sup> ∼ N(0, Ht) with a mean of zero and variances of Q<sup>t</sup> and H<sup>t</sup> are normally distributed. y<sup>t</sup> denotes a dependent variable, and z<sup>t</sup> = h 1, xt−1, yt−<sup>1</sup> , · · · , yt−<sup>p</sup> i denotes a 1 × m vector of variable interruption and intercept estimators depending on the model. As a consequence, the State-Space method is defined as follows, given a subset of K models at a given time:

$$\mathbf{y}\_{\mathbf{t}} = \mathbf{z}\_{\mathbf{t}}^{(\mathbf{k})} \boldsymbol{\theta}\_{\mathbf{t}}^{(\mathbf{k})} + \boldsymbol{\varepsilon}\_{\mathbf{t}}^{(\mathbf{k})} \tag{3}$$

$$
\boldsymbol{\theta}\_{\mathbf{t}+1}^{(\mathbf{k})} = \boldsymbol{\theta}\_{\mathbf{t}}^{(\mathbf{k})} + \boldsymbol{\mu}\_{\mathbf{t}}^{(\mathbf{k})} \tag{4}
$$

In this equation, ε (k) <sup>t</sup> ∼ N(0, H (k) t ) and µ (k) <sup>t</sup> ∼ N(0, Q (k) t ) with ϑ<sup>t</sup> = (θ (1) t , · · · , θ (k) t ) reveal which model of K subsets performs best during whatever period. Dynamic model averaging is a technique that permits a distinct model to be estimated at every given moment [19]. Raftery et al. [17] proposed a DMA approach that involves two parameters of α and λ, dubbed the forgetting factors. A recurrence estimate or forecast is feasible based on

the information of conventional filtering when the constants H<sup>t</sup> and Q<sup>t</sup> are being considered. The following formula serves as the foundation for the Kalman filtering (KF) process:

$$\Theta\_{\mathbf{t}-1} \Big| \mathbf{y}^{\mathbf{t}-1} \sim \mathbf{N}(\hat{\theta}\_{\mathbf{t}-1}, \sum\_{t=1 \mid \mathbf{t}-1} \mathbf{y}) \tag{5}$$

In Equation (5), the calculation of <sup>∑</sup>t−1|t−<sup>1</sup> and θˆ <sup>t</sup>−<sup>1</sup> is performed using a conventional approach that is a function of H<sup>t</sup> and Q<sup>t</sup> , and then the KF process is performed using the following equation:

$$
\Theta\_t \Big| \mathbf{y}^{t-1} \sim \mathbf{N}(\boldsymbol{\Theta}\_{t-1}, \sum\_{\mathbf{t} \mid \mathbf{t}-1}) \tag{6}
$$

Since <sup>∑</sup>t|t−<sup>1</sup> =∑t−1|t−<sup>1</sup> +Q<sup>t</sup> , to simplify, Raftery et al. [17] substituted <sup>∑</sup>t|t−<sup>1</sup> <sup>=</sup> <sup>1</sup> <sup>λ</sup>t|t−<sup>1</sup> ∑t−1|t−<sup>1</sup> with <sup>∑</sup>t|t−<sup>1</sup> =∑t−1|t−<sup>1</sup> +Q<sup>t</sup> , accordingly with <sup>0</sup> < <sup>λ</sup> ≤ 1, <sup>Q</sup><sup>t</sup> = (<sup>1</sup> − <sup>λ</sup>t|t−<sup>1</sup> −1 ) <sup>∑</sup>t−1|t−<sup>1</sup> . The value of λ<sup>t</sup> that is near to one suggests that the coefficients change more gradually. Raftery et al. [17] awarded it a value of 0.99 for the last five years' quarterly statistical data; the preceding figure shows that the observations from the previous five years account for 80 percent of the most current observation. If it is 95%, it indicates that the most recent five years of data accounted for 35% of the weight of the earlier observation. As a result, it is critical to choose the forgetting factors, which are often believed to be between 95 and 99 percent. The estimate in the model will be completed by using updated estimators using the following functions:

$$
\lambda\_{\mathbf{t}|\mathbf{t}} = \lambda\_{\mathbf{t}-1|\mathbf{t}-1}
$$

$$
\Theta\_{\mathbf{t}}|\mathbf{y}^{\mathbf{t}} \sim \mathcal{N}(\hat{\Theta}\_{\mathbf{t}} \sum\_{\mathbf{t}|\mathbf{t}}) \tag{7}
$$

In which

$$\boldsymbol{\hat{\Theta}}\_{\rm t} = \boldsymbol{\hat{\Theta}}\_{\rm t-1} + \sum\_{\mathbf{t}|\mathbf{t}-\mathbf{1}} \mathbf{z}\_{\mathbf{t}} \left(\mathbf{H}\_{\rm t} + \mathbf{z}\_{\mathbf{t}} \sum\_{\mathbf{t}|\mathbf{t}-\mathbf{1}} \mathbf{z}\_{\mathbf{t}}'\right)^{-1} (\mathbf{y}\_{\mathbf{t}} - \mathbf{z}\_{\mathbf{t}} \boldsymbol{\hat{\Theta}}\_{\rm t-1}) \tag{8}$$

$$\sum\_{\mathbf{t}|\mathbf{t}} = \sum\_{\mathbf{t}|\mathbf{t}-\mathbf{1}} - \sum\_{\mathbf{t}|\mathbf{t}-\mathbf{1}} \mathbf{z}\_{\mathbf{t}} \left(\mathbf{H}\_{\mathbf{t}} + \mathbf{z}\_{\mathbf{t}} \sum\_{\mathbf{t}|\mathbf{t}-\mathbf{1}} \mathbf{z}\_{\mathbf{t}}'\right)^{-1} \mathbf{z}\_{\mathbf{t}} \sum\_{\mathbf{t}|\mathbf{t}-\mathbf{1}} \tag{9}$$

Recursive prediction operates based on the predictive distribution in the following manner:

$$\mathbf{y}\_t | \mathbf{y}^{\mathbf{t}-1} \sim \mathbf{N} \left( \mathbf{z}\_t \hat{\mathbf{e}}\_{\mathbf{t}-1}, \mathbf{H}\_{\mathbf{t}} + \mathbf{z}\_t \sum\_{\mathbf{t} \mid \mathbf{t}-1} \mathbf{z}\_{\mathbf{t}}' \right) \tag{10}$$

Depending on the model, the above-mentioned functions for k may be expressed as follows, whereas the KF in the fixed estimators' model can be represented as (5)–(7), using ϑ<sup>t</sup> as a vector of all parameters (3) and (4).

$$\left| \, \vartheta\_{\mathbf{t}-1} \right| \mathbf{L}\_{\mathbf{t}-1} = \mathbf{k} \left| \mathbf{y}^{\mathbf{t}-1} \sim \mathrm{N}(\hat{\theta}\_{\mathbf{t}-1}^{(\mathbf{k})} \boldsymbol{\sum}\_{\mathbf{t}-1 \mid \mathbf{t}-1}^{(\mathbf{k})}) \tag{11}$$

$$\left. \vartheta\_{\mathbf{t}} \right| \mathbf{L}\_{\mathbf{t}} = \mathbf{k} , \mathbf{y}^{\mathbf{t}-1} \sim \mathbf{N}(\hat{\theta}\_{\mathbf{t}-1'}^{(\mathbf{k})} \sum\_{\mathbf{t} \mid \mathbf{t}-1}^{(\mathbf{k})}) \tag{12}$$

$$\left. \vartheta\_{\mathbf{t}} \right| \mathbf{L}\_{\mathbf{t}} = \mathbf{k}\_{\mathbf{t}} \mathbf{y}^{\mathbf{t}} \sim \mathbf{N}(\hat{\theta}\_{\mathbf{t}}^{(\mathbf{k})}, \sum\_{\mathbf{t} \mid \mathbf{t}}^{(\mathbf{k})}) \tag{13}$$

The value of θˆ (k) <sup>t</sup> and (∑ (k) t|t ) and (∑ (k) t|t−1 ) was acquired with the use of KF and Equations (8) and (9) and <sup>∑</sup>t|t−<sup>1</sup> <sup>=</sup> <sup>1</sup> <sup>λ</sup>t|t−<sup>1</sup> ∑t−1|t−<sup>1</sup> . We employed the Raftery et al. [17] technique, which incorporates a forgetting factor termed α for state equations in various estimating models, and so the aforementioned components are analogous to the forgetting

factor. Equation (4) is the starting point for the Kalman filter's application. When DMA is utilized, similar effects are obtained:

$$\mathbf{P}(\boldsymbol{\theta}\_{t-1}\Big|\mathbf{y}^{t-1}) = \sum\_{\mathbf{k}=1}^{K} \mathbf{p}(\boldsymbol{\theta}\_{t-1}^{(\mathbf{k})} \Big|\mathcal{L}\_{t-1} = \mathbf{k}, \mathbf{y}^{t-1}) \text{Pr}\left(\mathcal{L}\_{t-1} = \mathbf{k} \Big|\mathbf{y}^{t-1}\right) \tag{14}$$

The model's prediction function was replaced by the following equation introduced by Raftery et al. [17].

$$\pi\_{\mathbf{t}\mid\mathbf{t}-1,\mathbf{k}} = \frac{\pi\_{\mathbf{t}-1\mid\mathbf{t}-1,\mathbf{k}}^{\alpha\_{\mathbf{t}\mid\mathbf{t}-1}}}{\sum\_{\mathbf{l}=1}^{\mathbf{K}} \pi\_{\mathbf{t}-1\mid\mathbf{t}-1,\mathbf{l}}^{\alpha\_{\mathbf{t}\mid\mathbf{t}-1}}} \tag{15}$$

If 0 ≤ α < 1, the interpretation will be identical to that of λ, resulting in the following updated function:

$$\pi\_{\mathbf{t}|\mathbf{t},\mathbf{k}} = \frac{\pi\_{\mathbf{t}|\mathbf{t}-\mathbf{1},\mathbf{k}}^{\alpha\_{\mathbf{t}|\mathbf{t}-\mathbf{1}}} \mathbf{p}\_{\mathbf{k}}(\mathbf{y}\_{\mathbf{t}}|\mathbf{y}^{\mathbf{t}-1})}{\sum\_{\mathbf{l}=1}^{\mathbf{K}} \pi\_{\mathbf{t}|\mathbf{t}-\mathbf{1},\mathbf{l}}^{\alpha\_{\mathbf{t}|\mathbf{t}-\mathbf{1}}} \mathbf{p}\_{\mathbf{l}}(\mathbf{y}\_{\mathbf{t}}|\mathbf{y}^{\mathbf{t}-1})} \tag{16}$$

$$\mathfrak{a}\_{\mathfrak{t}|\mathfrak{t}} = \mathfrak{a}\_{\mathfrak{t}-1|\mathfrak{t}-1}$$

where p<sup>l</sup> (yt y t−1 ) indicates the predictive density in terms of y. The weighted mean may be applied to the predictive outputs of each model by using <sup>π</sup>tbt−1,k to perform recursive prediction on those outputs. As a result, the DMA point prediction is as follows:

$$\mathbb{E}(\mathbf{y}\_t \Big| \mathbf{y}^{t-1}) = \sum\_{\mathbf{k}=1}^{\mathbf{K}} \pi\_{\mathbf{t}|\mathbf{t}-1,\mathbf{k}} \mathbf{z}\_{\mathbf{t}}^{(\mathbf{k})} \hat{\boldsymbol{\Theta}}\_{\mathbf{t}-1}^{(\mathbf{k})} \tag{17}$$

DMS operates in such a manner that it picks the model with the greatest quantity of <sup>π</sup>tbt−1,k at any point in time. When <sup>α</sup> equals 0.99, the effectiveness of the previous 5 periods will account for 80% of the weighting for the current time. When α equals 0.99, 80 percent of the weighting for the current period will be determined by the performance of the preceding five periods. When <sup>α</sup> equals one, <sup>π</sup>tbt−1,k is precisely determined using the BMA model. Moreover, when λ equals one, BMA uses a traditional linear prediction model with constant coefficients.

Additionally, the suggested model's recursive estimation will begin with past values for <sup>π</sup>0b0,k and <sup>θ</sup> (k) 0 :

$$\mathbb{E}(\mathbf{y}\_t|\mathbf{y}^t) = \sum\_{\mathbf{k}=1}^{\mathbf{K}} \pi\_{\mathbf{t}|\mathbf{t},\mathbf{k}} \mathbf{z}\_{\mathbf{t}}^{(\mathbf{k})} \hat{\boldsymbol{\Theta}}\_{\mathbf{t}-1}^{(\mathbf{k})} \tag{18}$$

After calculating the equations, period t information is used to update the values. As previously stated, the purpose of including forgotten components is to minimize computational volume, as employing comprehensive Bayesian models may significantly increase computational volume. On the other hand, the sole process provided by Koop and Korobilis [18] is the manual selection of random values, which cannot result in plain values and also presupposes that the two parameters remain constant throughout time. In this work, we attempted to randomize the process for the selection of forgetting factors, α, and λ, using the ABC method. This approach is designed to decrease the sum of squared errors, which indicates the difference between computed and observed data. The mathematical expression is as follows:

$$\text{Minimize } \mathbf{e\_t} = \left( \mathbf{y\_t} - \mathrm{E}(\mathbf{y\_t}|\mathbf{y^t}) \right)^2$$

The following is the pseudocode of the algorithm's implementation procedure:

• Step 1: Choose a curve fitting function. Equations (16) and (18) may be combined to create the following function:

$$\mathbf{E}(\mathbf{y}\_{\mathbf{t}}|\mathbf{y}^{\mathbf{t}}) = \sum\_{\mathbf{k}=1}^{\mathbf{K}} \frac{\pi\_{\mathbf{t}|\mathbf{t}-\mathbf{1},\mathbf{k}}^{\alpha\_{\mathbf{t}|\mathbf{t}}} \mathbf{p}\_{\mathbf{k}}(\mathbf{y}\_{\mathbf{t}}|\mathbf{y}^{\mathbf{t}-1})}{\sum\_{\mathbf{l}=1}^{\mathbf{K}} \pi\_{\mathbf{t}|\mathbf{t}-\mathbf{1},\mathbf{l}}^{\alpha\_{\mathbf{t}|\mathbf{t}}} \mathbf{p}\_{\mathbf{l}}(\mathbf{y}\_{\mathbf{t}}|\mathbf{y}^{\mathbf{t}-1})} \mathbf{z}\_{\mathbf{t}}^{\mathbf{(k})} \boldsymbol{\hat{\theta}}\_{\mathbf{t}-1}^{(\mathbf{k})} \tag{19}$$

whereas recursive prediction operates using predictive distributions in the following manner:

$$\mathbf{y}\_{\mathbf{t}}|\mathbf{y}^{\mathbf{t}-1} \sim \mathcal{N}\left(\mathbf{z}\_{\mathbf{t}}\boldsymbol{\hat{\theta}}\_{\mathbf{t}-1}, \mathbf{H}\_{\mathbf{t}} + \mathbf{z}\_{\mathbf{t}}\frac{1}{\lambda\_{\mathbf{t}|\mathbf{t}}}\sum\_{\mathbf{t}=\mathbf{t}-1|\mathbf{t}-1}\mathbf{z}\_{\mathbf{t}}^{\mathbf{t}}\right)$$


$$\mathbf{w}\_{\rm s} \mathbf{w}\_{\rm j}^{\rm new} \mathbf{s}\_{\rm s} = \mathbf{w}\_{\rm j}^{\rm low} + \chi (\mathbf{w}\_{\rm j}^{\rm up} - \mathbf{w}\_{\rm j}^{\rm low})\_{\prime} \begin{array}{l} \mathbf{w}^{\rm low} \le \mathbf{w}\_{\rm j} \le \mathbf{w}^{\rm up} \\ \mathbf{s} = 1, \ldots, \mathbf{SN} \end{array} \tag{20}$$

where SN represents the food supply in total. w up j and wlow j represent the top and lower limits of the j − th design variable, while γ is a random real value between zero and one.

	- (a) Create new options for an employed bee using the following equation, where a new candidate food source (swnew j ) is identified using two prior food source locations remembered by an employed bee (swold j ) and a randomly chosen neighborhood of a food source (swold k ):

$$\mathbf{s}\_{\rm s} \mathbf{w}\_{\rm j}^{\rm new} = \mathbf{s}\_{\rm j} \mathbf{w}\_{\rm j}^{\rm old} + \boldsymbol{\varrho} (\mathbf{s}\_{\rm s} \mathbf{w}\_{\rm j}^{\rm old} - \boldsymbol{s}\_{\rm s} \mathbf{w}\_{\rm k}^{\rm old}) \tag{21}$$


$$\mathbf{p\_i} = \frac{\oslash\_i}{\sum\_{i=1}^{\text{SN}} \oslash\_i}$$

where ∅<sup>i</sup> represents a measure of the solution's fitness i, as determined by the employed bee. This corresponds to the nectar content in the food supply at location i. • Step 9. Traverse each source of food (i = 1, . . . , SN).


position; if there is no change, the candidate food source that the observer bee visited will not be selected.


Another objective of this study aimed to compare the effectiveness of various prediction methods. The Mean Absolute Forecast Error (MAFE) and the Mean Squared Forecast Error (MSFE) are employed as standard indices in this research.

$$\text{MSFE} = \frac{\sum\_{\tau=\tau\_0}^{\text{T}} \left[ \mathbf{y}\_\tau - \text{E}(\mathbf{y}\_\tau | \text{Data}\_{\tau-\text{h}}) \right]^2}{\text{T} - \tau\_0 + 1} \tag{22}$$

$$\text{MAFE} = \frac{\sum\_{\mathbf{\tau}=\mathbf{\tau}\_0+1}^{\mathbf{T}} |\mathbf{y}\_\mathbf{\tau} - \mathbf{E}(\mathbf{y}\_\mathbf{\tau} | \mathbf{Data}\_{\mathbf{\tau}-\mathbf{h}})|}{\mathbf{T} - \mathbf{\tau}\_0 + 1} \tag{23}$$

where Dataτ−<sup>h</sup> is the data that were obtained from the time <sup>τ</sup> − h, <sup>h</sup> is the horizon for time prediction, and E(y<sup>τ</sup> |Dataτ−h) is the forecast point of y<sup>τ</sup> . This study begins with the results of DMA and DMS, followed by the events that determine which variables are most suited for predicting the gasoline demand function. Then, the performance of DMS and DMA is contrasted. In addition, it assesses the sensitivity of models and prediction results concerning the selection of forgetting factors.

#### **3. The Estimated Model and Data**

Annual observations for the United States from 1960 to 2020 were utilized in this analysis. Exogenous variables include measurements of demographic traits, economic activity, income, driving expenses, car pricing, and road availability. These variables in Table 1 are chosen based on an extensive review of the available literature.


**Table 1.** Research literature for estimating gasoline demand function to determine model variables.

Table 2, below, provides a brief description of the variables included in our analysis, as well as a definition and reference to the source. Moreover, summary statistics for the variables that are used in the empirical analysis are presented in Table 3.


#### **Table 2.** Variables and definitions.

FRED: Federal Reserve Economic Data; https://fred.stlouisfed.org, accessed on 20 June 2022; EIA: Energy Information Administration; https://www.eia.gov/, accessed on 20 June 2022; FHWA: Federal Highway Administration; https://highways.dot.gov/, accessed on 18 June 2022.

**Table 3.** Descriptive statistics.


#### **4. Results**

By comparing DMA predictions, we examine forecast performance. We attempted to empirically test several configurations of the ABC model to increase the accuracy of the forecast while achieving the quickest feasible computation speed. In conclusion, the number of bees was fixed at five, the greatest quantity of repetitions at five, and the lower and upper bound between 0.9% and 1%. Finally, we demonstrate the sensitivity of our findings to the choice of forgetting factors, α and λ. We provide findings for prediction horizons of one year (h = 1) and four years (h = 4). A prediction horizon of 4 means that we used the values of the independent variables in the previous 4 periods to predict the dependent variable in the current period. Obviously, with an increase in the prediction horizon, the prediction accuracy of the independent variables decreases. Our models all incorporate an intercept and a single lag between the dependent and independent variables. Experiments with lag lengths up to two revealed that a single lag produces the highest prediction results. Using the ABC approach, we sought to randomize the forgetting components in this study. Thus, our methodology not only provides for the automated determination of the two forgetting elements but also for their evolution over time to minimize the prediction model's inaccuracy. These computations are carried out at a low computational cost. Thus, rather than selecting manually, we use a more precise selection mechanism. Figure 1 illustrates the outcome of estimating the components across time and the prediction horizons one and four. After estimating the model using the combined DMA-ABC model, the chance of each of the model's independent variables being present is supplied. The posterior inclusion probability is shown in Figures 2 and 3. That is, they

quantify the likelihood that a predictor will help predict at time t. They are equivalent to the weights applied using DMA to models that incorporate a predictor. These graphs illustrate which predictors are significant at any given moment in time. These graphs demonstrate that DMS nearly always selects sparse models. These results are compelling evidence of model evolution. DMA has a significant theoretical advantage over other forecasting methodologies in that it enables the forecasting model to evolve. Of course, this gain may be negligible in a given empirical application if the forecasting model does not vary much over time. While the same trend remains true to a lesser degree, it is apparent that there is a significant change over time. That is, the forecasting model's collection of predictors evolves with time. After 1980, practically all surface variables enter the model with varying probability. Intermittent values, of course, provide various outcomes. Between 2000 and 2015, the likelihood of existence, or the initial lag, of the majority of model variables is questioned. In comparison to other variables, vehicle registration has the lowest likelihood of being present, while public road mileage at the level and first log values indicate a high possibility of being included in gasoline demand forecasting. equivalent to the weights applied using DMA to models that incorporate a predictor. These graphs illustrate which predictors are significant at any given moment in time. These graphs demonstrate that DMS nearly always selects sparse models. These results are compelling evidence of model evolution. DMA has a significant theoretical advantage over other forecasting methodologies in that it enables the forecasting model to evolve. Of course, this gain may be negligible in a given empirical application if the forecasting model does not vary much over time. While the same trend remains true to a lesser degree, it is apparent that there is a significant change over time. That is, the forecasting model's collection of predictors evolves with time. After 1980, practically all surface variables enter the model with varying probability. Intermittent values, of course, provide various outcomes. Between 2000 and 2015, the likelihood of existence, or the initial lag, of the majority of model variables is questioned. In comparison to other variables, vehicle registration has the lowest likelihood of being present, while public road mileage at the level and first log values indicate a high possibility of being included in gasoline demand forecasting.

*Energies* **2023**, *16*, x FOR PEER REVIEW 9 of 14

horizons of one year (h = 1) and four years (h = 4). A prediction horizon of 4 means that we used the values of the independent variables in the previous 4 periods to predict the dependent variable in the current period. Obviously, with an increase in the prediction horizon, the prediction accuracy of the independent variables decreases. Our models all incorporate an intercept and a single lag between the dependent and independent variables. Experiments with lag lengths up to two revealed that a single lag produces the highest prediction results. Using the ABC approach, we sought to randomize the forgetting components in this study. Thus, our methodology not only provides for the automated determination of the two forgetting elements but also for their evolution over time to minimize the prediction model's inaccuracy. These computations are carried out at a low computational cost. Thus, rather than selecting manually, we use a more precise selection mechanism. Figure 1 illustrates the outcome of estimating the components across time and the prediction horizons one and four. After estimating the model using the combined DMA-ABC model, the chance of each of the model's independent variables being present is supplied. The posterior inclusion probability is shown in Figures 2 and 3. That is, they quantify the likelihood that a predictor will help predict at time t. They are

**Figure 1.** α and λ over time (h = 1, h = 4). **Figure 1.** α and λ over time (h = 1, h = 4).

**Figure 2.** Posterior probability (h = 1). **Figure 2.** Posterior probability (h = 1). **Figure 2.** Posterior probability (h = 1).

**Figure 3.** Posterior probability (h = 4). **Figure 3.** Posterior probability (h = 4). prediction error in model BMA with <sup>a</sup> prediction horizon of 1. In addition, this value is **Figure 3.** Posterior probability (h = 4).

Figure 4 illustrates the actual and predicted value of gasoline along with the forecast horizons h = 1 and h = 4. The accuracy of the model in estimating gasoline demand is seen

Figure 4 illustrates the actual and predicted value of gasoline along with the forecast horizons h = 1 and h = 4. The accuracy of the model in estimating gasoline demand is seen

estimated model's accuracy. Our earlier DMA and DMS findings were for our benchmark example, in which we used the ABC technique to determine a random forgetting factor that changes over time. As previously stated, researchers in this area use predetermined values for α and λ. As a consequence, Raftery et al. [17] used λ ൌ α ൌ 0.99 and suggest that the findings will remain resilient to accepting modifications in these variables. To test these claims of resilience, the results of our forecasting experiment utilizing different combinations of forgetting components are shown in Table 4. MSFE and MAFE values for various models of DMA-ABC, DMS-ABC, DMS, DMA, BMA, TVP-BMA, and TVP are provided in Table 4 for prediction horizons 1 and 4. The results of the comparison of several models in Table 4 indicate that the combined model of DMS and ABC, with the option of automatically acquiring forgetting factors over time, obtains the greatest results in forecasting gasoline demand. According to the DMA-ABC model, the mean values of the forgetting components are equal to α = 0.9449 and λ = 0.9662. Even taking the constant mean values of computational forgetting factors into account produced satisfactory results. It is noteworthy that the value α = 0.9449 enables relatively fast model evolution over time. This is similar to a previous tale we mentioned: it seems that allowing models to evolve is more significant than allowing parameters to vary with λ = 0.9662 for increasing forecast performance. The BME model (with λ = α = 1) does not have any dynamic approach, which means that while the estimated coefficients are constant over time, the input variables to the model are also constant over time. To investigate the effects of adding dynamics to the model in increasing the forecasting accuracy, we added two more columns to Table 4. In these two columns, the ratio of MAFE and MSFE of different models is calculated with the MAFE and MSFE values of the BMA model (with B index). Based on the results, the prediction error values in model DMS-ABC are about 0.82 of the prediction error in model BMA with a prediction horizon of 1. In addition, this value is

example, in which we used the ABC technique to determine a random forgetting factor that changes over time. As previously stated, researchers in this area use predetermined values for α and λ. As a consequence, Raftery et al. [17] used λ ൌ α ൌ 0.99 and suggest that the findings will remain resilient to accepting modifications in these variables. To test these claims of resilience, the results of our forecasting experiment utilizing different combinations of forgetting components are shown in Table 4. MSFE and MAFE values for various models of DMA-ABC, DMS-ABC, DMS, DMA, BMA, TVP-BMA, and TVP are provided in Table 4 for prediction horizons 1 and 4. The results of the comparison of several models in Table 4 indicate that the combined model of DMS and ABC, with the option of automatically acquiring forgetting factors over time, obtains the greatest results in forecasting gasoline demand. According to the DMA-ABC model, the mean values of the forgetting components are equal to α = 0.9449 and λ = 0.9662. Even taking the constant mean values of computational forgetting factors into account produced satisfactory results. It is noteworthy that the value α = 0.9449 enables relatively fast model evolution over time. This is similar to a previous tale we mentioned: it seems that allowing models to evolve is more significant than allowing parameters to vary with λ = 0.9662 for increasing forecast performance. The BME model (with λ = α = 1) does not have any dynamic approach, which means that while the estimated coefficients are constant over time, the input variables to the model are also constant over time. To investigate the effects of adding dynamics to the model in increasing the forecasting accuracy, we added two more columns to Table 4. In these two columns, the ratio of MAFE and MSFE of different models is calculated with the MAFE and MSFE values of the BMA model (with B index). Based on the results, the prediction error values in model DMS-ABC are about 0.82 of the

Figure 4 illustrates the actual and predicted value of gasoline along with the forecast horizons h = 1 and h = 4. The accuracy of the model in estimating gasoline demand is seen in Figure 3. Additionally, expanding the prediction horizon resulted in a decline in the estimated model's accuracy. Our earlier DMA and DMS findings were for our benchmark example, in which we used the ABC technique to determine a random forgetting factor that changes over time. As previously stated, researchers in this area use predetermined values for α and λ. As a consequence, Raftery et al. [17] used λ = α = 0.99 and suggest that the findings will remain resilient to accepting modifications in these variables. To test these claims of resilience, the results of our forecasting experiment utilizing different combinations of forgetting components are shown in Table 4. MSFE and MAFE values for various models of DMA-ABC, DMS-ABC, DMS, DMA, BMA, TVP-BMA, and TVP are provided in Table 4 for prediction horizons 1 and 4. The results of the comparison of several models in Table 4 indicate that the combined model of DMS and ABC, with the option of automatically acquiring forgetting factors over time, obtains the greatest results in forecasting gasoline demand. According to the DMA-ABC model, the mean values of the forgetting components are equal to α = 0.9449 and λ = 0.9662. Even taking the constant mean values of computational forgetting factors into account produced satisfactory results. It is noteworthy that the value α = 0.9449 enables relatively fast model evolution over time. This is similar to a previous tale we mentioned: it seems that allowing models to evolve is more significant than allowing parameters to vary with λ = 0.9662 for increasing forecast performance. The BME model (with λ = α = 1) does not have any dynamic approach, which means that while the estimated coefficients are constant over time, the input variables to the model are also constant over time. To investigate the effects of adding dynamics to the model in increasing the forecasting accuracy, we added two more columns to Table 4. In these two columns, the ratio of MAFE and MSFE of different models is calculated with the MAFE and MSFE values of the BMA model (with B index). Based on the results, the prediction error values in model DMS-ABC are about 0.82 of the prediction error in model BMA with a prediction horizon of 1. In addition, this value is equal to 0.76 in the forecast horizon of 4. Therefore, by increasing the prediction horizon, moving towards dynamic models leads to a further increase in prediction accuracy. *Energies* **2023**, *16*, x FOR PEER REVIEW 11 of 14 equal to 0.76 in the forecast horizon of 4. Therefore, by increasing the prediction horizon, moving towards dynamic models leads to a further increase in prediction accuracy.

**Figure 4.** The actual and predicted value of gasoline demand in the forecast horizon h = 1 and h = 4 α = 0.9449; λ. **Figure 4.** The actual and predicted value of gasoline demand in the forecast horizon h = 1 and h = 4 α = 0.9449; λ.

h = 1 h = 4

DMA λ = 0.9698; α = 0.9556

DMS λ = 0.9698; α = 0.9556

DMA λ = α = 0.99

DMA λ = 0.95; α = 0.99

DMS λ = 0.95; α = 0.99

 

 

**MAFE MSFE**

0.71 4.86 0.96 0.96 DMA-

0.58 3.87 0.78 0.76 DMS-

0.60 3.98 0.80 0.79 DMS <sup>λ</sup> <sup>=</sup>

0.72 4.98 0.97 0.98 DMA <sup>λ</sup> <sup>=</sup>

0.60 3.98 0.80 0.79 DMS <sup>λ</sup> <sup>=</sup>

0.71 4.85 0.95 0.96

0.58 3.87 0.78 0.76

0.73 5.03 0.98 0.99

0.72 4.95 0.97 0.98

0.60 3.93 0.80 0.78

 

 

**Table 4.** Comparison of models.

ABC DMA-ABC 0.55 3.60 1.01 0.98

**Prediction Method MAFE MSFE**

DMA λ = 0.9662; α = 0.9449 0.53 3.60 0.98 0.98

DMS λ = 0.9662; α = 0.9449 0.44 3.00 0.81 0.81

DMA λ = α = 0.99 0.54 3.65 0.99 0.99

DMA λ = 0.95; α = 0.99 0.55 3.56 1.01 0.97

DMS λ = 0.95; α = 0.99 0.46 2.99 0.85 0.81

<sup>α</sup> <sup>=</sup> 0.99 DMS <sup>λ</sup> <sup>=</sup> <sup>α</sup> <sup>=</sup> 0.99 0.45 3.04 0.82 0.83

<sup>α</sup> <sup>=</sup> 0.95 DMA <sup>λ</sup> <sup>=</sup> <sup>α</sup> <sup>=</sup> 0.95 0.55 3.57 1.01 0.97

<sup>α</sup> <sup>=</sup> 0.95 DMS <sup>λ</sup> <sup>=</sup> <sup>α</sup> <sup>=</sup> 0.95 0.46 2.99 0.85 0.81


**Table 4.** Comparison of models.

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

Accurate modeling and forecasting of gasoline demand may provide a valuable framework for policymakers to consider the policy implications of their energy market activities, which is the goal of this study. The primary shortcoming in prior forecasting models was their inability to accurately predict over time. Policymakers, on the other hand, should disregard short-term and temporary variations in gasoline demand in favor of economic stability. Therefore, the objective of this study was to develop a nonlinear dynamic model DMA-ABS to forecast gasoline consumption in the United States using annual data from 1960 to 2020. These models may be used to determine changes in both the input variables and the parameters of variables through time. The inclusion of two forgetting variables in the DMA model may be used to control the speed of such dynamics in the model, which has been previously determined manually in earlier research. In this work, we sought to implement a random process of forgetting factor selection using the ABC method. Therefore, one of the primary objectives of this study is to merge ABC with DMA to increase prediction accuracy. The findings of the DMS estimation model indicated that the input variables fluctuate with time, emphasizing the need of employing dynamic models rather than constant input variables for estimating gasoline demand.

Gasoline demand prediction helps policymakers to make informed decisions on issues related to energy security, environmental regulations, and transportation infrastructure. For instance, it can assist in determining the number of gas stations required to meet demand in a particular area, the type of fuel to be used in different transportation modes, and the amount of investment needed to maintain or upgrade the transportation infrastructure. Moreover, gasoline demand prediction can aid in managing the price of gasoline. It helps in determining the price level that will meet the demand and supply equilibrium. Therefore, it is recommended that in future research, the DMA model will be integrated with other evolutionary algorithms, such as particle swarm optimization (PSO), genetic algorithm (GA), etc., to compare the results and provide a more accurate prediction of the gasoline market through the expansion of the model presented in this research.

**Author Contributions:** Conceptualization, S.H.H.; Methodology, S.H.H.; Software, S.H.H.; Formal analysis, S.H.H.; Resources, Q.L.; Writing—original draft, S.H.H. and Q.L.; Writing—review & editing, Q.L.; Visualization, S.H.H.; Supervision, Q.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** No funding has been received for this research.

**Data Availability Statement:** Data is available upon request.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

## *Article* **Assessing Fossil Fuels and Renewables' Impact on Energy Poverty Conditions in Europe**

**George Halkos \* and Eleni-Christina Gkampoura**

**\*** Correspondence: halkos@uth.gr

**Abstract:** The disadvantages of fossil fuels and their impact on the environment have made the transition to renewable energy sources essential to cover our energy needs. However, different energy resources have a different impact on energy poverty conditions in the world, an issue that is important to examine and properly address. This study examines the impact that fossil fuels final energy consumption in households per capita and renewables and biofuels final energy consumption in households per capita have on energy poverty conditions in Europe, using panel data from 28 European countries for the time period 2004–2019 and static and dynamic regression models, while also performing various econometric tests. The findings indicate that GDP per capita and fossil fuels are linked to an inverse relationship to energy poverty conditions. Renewables and biofuels are also linked to an inverse relationship to the inability to keep homes adequately warm and the presence of leaks, damp, or rot in the dwelling, but they could be considered a driver of arrears on utility bills. In addition, a comparative analysis between Sweden, Germany, and Greece and their conditions on energy poverty and energy transition was conducted, highlighting the differences existing between the three European countries. The findings of the research can be useful for governments and policy makers to develop strategies that promote energy transition while protecting energy consumers.

**Keywords:** energy poverty; fossil fuels; renewables; Europe

#### **1. Introduction**

The discovery and use of fossil fuels can be viewed as the main foundation of humankind's prosperity, growth, and well-being [1]. Coal, oil, and gas have been in the center of the industrialized world since the early 1800s and they constitute the main driving force of economic and social growth in the world [2]. Fossil fuels are still used to cover most of the world's energy needs and this usage is expected to increase more in the future, due to the expected increase in the global population and the new, energy-intense way of life [3].

The disadvantages that emerge from the usage of fossil fuels are many and significant. Fossil fuels are primarily responsible for enormous greenhouse gas emissions into the atmosphere, and they contribute on a great level to global warming, something that could be proven catastrophic for the environment as well as for human health, life, and civilization as we know it [4]. In addition, fossil fuels are finite, and some scientists believe that they might reach their peak soon [1]. Their depletion means that the world should not rely on them anymore and, instead, turn to alternative sources.

All the disadvantages that come as a result from the use and combustion of fossil fuels make it obvious that the world should focus on alternative energy sources, or other energy saving methods and measures for the mitigation of GHG emissions [5]. A transition to renewable energy sources for the satisfaction of our energy needs is considered to be essential to address climate change and achieve the target of limiting the global average temperature increase under 2 ◦C. Despite the fact that the usage of renewables has increased over past years, fossil fuels still cover around 80% of global energy demands, leading to an urgent need of change in future energy policies [6].

**Citation:** Halkos, G.; Gkampoura, E.-C. Assessing Fossil Fuels and Renewables' Impact on Energy Poverty Conditions in Europe. *Energies* **2023**, *16*, 560. https:// doi.org/10.3390/en16010560

Academic Editor: Idiano D'Adamo

Received: 8 December 2022 Revised: 29 December 2022 Accepted: 29 December 2022 Published: 3 January 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Department of Economics, University of Thessaly, 38333 Volos, Greece

To achieve a successful full transition to clean energy sources, it is important that energy R&D is promoted on a global level and that innovation, investment, and deployment policies and strategies regarding energy storage technologies are evolved and adopted by policy makers [7]. Cities, which are the biggest energy consumer, should be a part of that transition, promoting a cleaner urban energy system that will lead to cleaner cities [8]. In addition, the transition requires political support and efficient governance, as well as market designs and financial incentives [9]. This transition, unlike other energy transitions that have happened in the past around the world, will have to occur rapidly to successfully tackle climate change. Even though it is supported that this transition is unlikely to happen within the next few decades, governments should promote and support such policies, while international agreements concerning the issue should be implemented [10]. The COVID-19 pandemic that had an important impact on the global economy, has also widened the uncertainty of energy transition [11].

Currently, the world highly depends on fossil fuels. This means that certain features are configured, while they provide stability regarding technological artefacts and scientific knowledge, market structures, practices, and regulatory frameworks [12]. Based on that, a few experts argue that a transition to renewable energy sources would disturb the balance that fossil fuels have provided, and it would challenge the status quo. It is often believed that renewables might not be able to meet the global energy demand, because of their lower energy density, or that society and politicians will not adopt the best scientifically proven alternative that is available. Some experts argue that the implementation of Carbon Capture and Storage (CCS) is the best solution for CO<sup>2</sup> emissions mitigation and sustenance of the fossil fuels use at the same time [13], but it is found to be a more expensive and riskier alternative to renewables, while not being carbon neutral [9]. In the literature, some studies have focused on the social and political impacts of renewables [14], on their impact on social welfare [15], as well as on their impact on local development [16]. However, after taking both positive and negative impacts into consideration, most scientists believe that the transition to renewable energy sources is beneficial and is viewed as the best solution, in order to promote sustainability and energy security.

Energy poverty is another energy-related topic that is well-discussed in the recent literature. It refers to a situation where households do not have access to the necessary energy services, which are vital for the satisfaction of basic human needs [17]. A household can be characterized as energy poor when it is lacking sufficient, affordable, and safe energy services [18], something that is not observed in developing countries only; an important number of European households were found to be "fuel poor", especially during the period of the global financial crisis [19].

The linkages between the different types of energy sources and energy poverty, as well as between energy transition and energy poverty, have not been examined extensively in the existing literature. It is, however, a topic that should be addressed and the studies that focus on this matter could be used as a tool by policy makers for the successful implementation of energy-related strategies and policies, which have been in the center of many global initiatives for sustainability. For instance, the 7th Sustainable Development Goal proposed by the United Nations, referring to providing access to affordable and sustainable energy services for all, includes both the topics of energy poverty and energy transition to more sustainable sources [20]. This study aims to contribute and expand the existing knowledge on the topic, examining the relationship between energy poverty conditions in 28 European countries for the years 2004–2019 and energy consumption coming from fossil fuels and renewables in households, using an advanced econometric methodology.

#### **2. Literature Review**

A significant number of studies in the current literature have been focusing on the topic of energy poverty. A review of the problem of energy poverty has been presented by Halkos and Gkampoura; the authors included in their review various definitions that have been given to energy poverty in the literature, as well as the impact that energy poverty has on human health, on the society, the economy, and the environment. In addition, various energy poverty drivers are identified, including household characteristics and other socioeconomic and environmental factors. The authors also analyzed the different approaches to measuring energy poverty and presented the situation currently occurring in different parts of the world regarding the problem. Various actions that could assist in tackling energy poverty are also presented in the review [21].

The main approaches of measuring energy poverty that have been suggested in the literature include the expenditure approach, where metrics are compared to certain thresholds, and the consensual approach, where various indicators that are based on surveys can be used, while composite measurements can also be created [22]. The Multidimensional Energy Poverty Index (MEPI), developed by Nussbaumer et al. [23] is one of these composite indexes, which takes into account the multidimensional nature that energy poverty has, and it was used to estimate energy poverty for certain African countries. Based on the same index, energy poverty was calculated for certain Latin American countries by Santillán et al. [24].

When it comes to European countries, the European Union Statistics on Income and Living Conditions (EU-SILC) provide measures and data that can be used to calculate energy poverty, including the indicators for measuring energy poverty that are suggested in the literature: (i) inability to keep home adequately warm, (ii) arrears on utility bills and (iii) presence of leak, damp, or rot in the dwelling. These data was used by Thomson and Snell [25], who estimated fuel poverty for 25 European countries, by proposing a methodology of four different scenarios, where a different weight was assigned to each indicator. Based on this methodology, Halkos and Gkampoura [26] in their research also created four different scenarios with different weights, in order to evaluate energy poverty conditions in 28 European countries for the period 2004–2019, while identifying the drivers of energy poverty conditions and the impact that the economic crisis had on them.

A stochastic frontier analysis approach was used by Rodriguez-Alvarez et al. [27] to identify the determinants of energy poverty in 30 European countries for the years 2005–2018. The three previously mentioned indicators were also used by Bollino and Botti [19], who combined them with two additional variables, developing the Energy Poverty Multidimensional Index (EMPI) to capture and evaluate energy poverty for 2012 and 2014 in Europe.

In the current literature, a few studies have examined the relationship between renewables and energy poverty, as well as between energy transition and energy poverty. Most of these studies support that a transition from fossil fuels to renewable energy sources will have a positive impact on energy poverty, helping towards its eradication. However, it is also argued in a few studies that energy transition could have negative results on energy poverty-related problems, unless certain measures are being implemented to minimize their damage.

More specifically, Mastropietro [28] analyzed the effect that renewable energy sources for electricity might have on energy poverty. The author argued that the costs required to support renewables' technologies are often transferred to the consumers as surcharges, something that could lead to more intense energy poverty issues. It is suggested that measures, such as state finance or finance through the auctions for emission allowances should be implemented, in order to minimize the social cost that will follow after energy transition.

Specific guidelines that should be followed in order to eliminate the problem of energy poverty by using renewable energy sources and, more specifically, solar energy, were presented by Pagliaro and Meneguzzo [29]. The authors argue that energy poverty will be reduced if renewables are adopted for energy generation and they make specific suggestions to policy makers, such as: view energy poverty within its local social and economic context, get advice from energy managers with knowledge regarding new technologies in the sector and their socioeconomic impacts, make the community engage and be interested in the matter, and establish public institutions concerning renewables, with the target of providing education on the topic and strategies for a successful energy transition. These guidelines, according to the authors, will facilitate the transition process and will lead to environmental and socioeconomic benefits.

The linkages between energy transition and socioeconomic inequalities in Europe have also been explored in Bouzarovski and Tirado Herrero's [30] research, who emphasize the spatial and temporal variations in energy poverty's incidence. The authors found that there are significant regional inequalities concerning the drivers of energy poverty and the exposure of the countries to them, meaning that energy poverty is a problem presented in a variety of social strata. They also pointed out that energy poverty has increased in the EU since 2007 in general, even though energy transition does not have an extreme impact on these inequalities, according to the authors' results. Further investigation concerning the risks that result from energy transition is strongly suggested.

The case of the UK was studied by Hiteva [31], who examined the potential impact that energy transition might have on fuel poverty. The author took into consideration the conditions of energy transition and the costs, risks, and financial liabilities in the industry of renewable energy, that could have a negative effect on energy poverty. The case of the UK was analyzed and then compared with Bulgaria, the country with the biggest percentages of people living in energy poverty conditions in Europe. The author suggested that the issue of fuel poverty should be addressed throughout the whole renewable electricity production chain and not just at the consumption end and highlighted the need of implementing policies for fuel poverty alleviation throughout this chain.

The effect that energy poverty has on several development outcomes was analyzed by Adom et al. [32], while taking into consideration the influence of green energy transition. The findings highlight that a transition to green energy could potentially reduce vulnerability and provide partial resilience regarding energy poverty shocks, while it could also facilitate the improvement of several development outcomes, including GDP per capita, poverty, and income inequality.

Based on these studies found in the current literature, it can be observed that the results regarding energy poverty and different energy sources linkages, as well as energy poverty and energy transition linkages vary, especially when focusing on different world regions. Even though the topic is discussed in the recent literature, it is significantly important to further examine this relationship. This study examines the linkages between energy consumption coming from fossil fuels and renewable sources and energy poverty indicators for 28 European countries, using an in-depth econometric methodology, that is presented in Section 3. To the best of our knowledge, this methodology has not been used in any similar studies, while examining European countries and using recent data at the same time. In addition, a comparative analysis of energy poverty and energy transition conditions in three European countries is conducted, in order to better understand the progress of three countries with different socioeconomic and environmental conditions on these topics.

#### **3. Methodology**

For the analysis, data were collected for the three indicators that are considered to be the key elements of energy poverty, according to the current literature (Table 1: Indicators 1–3). In addition, data were collected regarding GDP per capita, final energy consumption in households per capita, as well as final energy consumption in households by fuel (Table 1: Indicators 4–6). These last two databases were combined, in order to create two new indicators: FFpc, which stands for fossil fuels (oil and petroleum products, natural gas and solid fossil fuels) final energy consumption in households per capita, and RESpc, which stands for renewables and biofuels final energy consumption in households per capita. All data were retrieved from Eurostat's database, for the period 2004–2019 and for 28 European countries (Figure 1). Statistical packages EViews and Stata were used for the analysis.


6 Final energy consumption in households by fuel % [36]

For the analysis, data were collected for the three indicators that are considered to be the key elements of energy poverty, according to the current literature (Table 1: Indicators 1–3). In addition, data were collected regarding GDP per capita, final energy consumption in households per capita, as well as final energy consumption in households by fuel (Table 1: Indicators 4–6). These last two databases were combined, in order to create two new indicators: FFpc, which stands for fossil fuels (oil and petroleum products, natural gas and solid fossil fuels) final energy consumption in households per capita, and RESpc, which stands for renewables and biofuels final energy consumption in households per capita. All data were retrieved from Eurostat's database, for the period 2004–2019 and for 28 European countries (Figure 1). Statistical packages EViews and

**Table 1.** Indicators retrieved from Eurostat and used for the analysis. Stata were used for the analysis.

Energies 2022, 15, x FOR PEER REVIEW 5 of 17

The linkages that exist between energy poverty and fossil fuels and renewable energy consumption are examined, taking into consideration panel data for the years 2004–2019. After the use of Box–Cox specifications that compare linear and logarithmic forms, three different regression models are formulated, where Indicators 1, 2, and 3 (Table 1) are considered as dependent variables, and GDP per capita, fossil fuels final energy consumption per capita, and renewables and biofuels final energy consumption per capita are considered as independent variables. The proposed model is: The linkages that exist between energy poverty and fossil fuels and renewable energy consumption are examined, taking into consideration panel data for the years 2004–2019. After the use of Box–Cox specifications that compare linear and logarithmic forms, three different regression models are formulated, where Indicators 1, 2, and 3 (Table 1) are considered as dependent variables, and GDP per capita, fossil fuels final energy consumption per capita, and renewables and biofuels final energy consumption per capita are considered as independent variables. The proposed model is:

$$(\text{Indicf})\_{\text{i}\text{i}} = a\_{\text{i},\text{t}} + \beta\_{\text{1i},\text{t}}(\text{GDPpc})\_{\text{i},\text{t}} + \beta\_{\text{2i},\text{t}}(\text{FFpc})\_{\text{i},\text{t}} + \beta\_{\text{3i},\text{t}}(\text{RESpc})\_{\text{i},\text{t}} + \gamma\_{\text{i}} + \delta\_{\text{1}} + \varepsilon\_{\text{1},\text{t}} \tag{1}$$

In this regression model, GDPpc stands for GDP per capita, FFpc stands for fossil fuels final energy consumption in households per capita, and RESpc stands for renewables and biofuels final energy consumption in households per capita. Additionally, *J* equals to numbers 1–3, indicating the three different indicators that are used as dependent variables.

For the models' estimations, Fixed (FE) and Random (RE) Effects methods are used, depending on how ai is handled: either as fixed predefined numbers or as random expulsions from a particular distribution [26,37–39]. In the case of Fixed Effects, where the cross-section specific components are viewed as fixed parameters, then the model becomes:

$$y\_{it} = a + X\_{it}'\beta + \sum\_{i=1}^{N} \mu\_i D\_i + v\_{it} \tag{2}$$

In the case of Random Effects, which can be used in cases were N individuals are drawn from a large population randomly, the following apply:

$$\mu\_i \sim \text{IID}\left(0, \sigma\_\mu^2\right), v\_{\text{it}} \sim \text{IID}\left(0, \sigma\_v^2\right) \tag{3}$$

with the *µ<sup>i</sup> s* being independent of the vits, as are the *Xits* of the *µ<sup>i</sup> s* and *vits* for all *I* and *t* [40]. Inconsistency is checked in the RE estimate with Hausman tests, which determine whether the FE or RE model should be used.

In addition, Generalized Method of Moments (GMM) is used, in terms of orthogonal deviations, in order to capture the models' dynamic nature. As Arellano and Bond have stated [41], in orthogonal deviations, each observation is indicated as a deviation from the average of the sample's future observations. Each deviation is weighted to standardize the variance:

$$\mathbf{x}\_{\text{it}}^{\*} = \left[ \mathbf{x}\_{\text{it}} - \left( \mathbf{x}\_{i(t+1)} + \dots + \mathbf{x}\_{iT} \right) / (T - t) \right] \sqrt{(T - t)} / \sqrt{T - t + 1}, t = 1, \dots, T - 1 \tag{4}$$

The (*Ti* − *q*) equations for individual units *i* are:

$$\mathcal{Y}\_{\dot{l}} = \delta w\_{\dot{l}} + d\_{\dot{l}} \eta\_{\dot{l}} + v\_{\dot{l}} \tag{5}$$

where *δ* is a parameter vector that includes *αk*, *β*, and *λ*, *w<sup>i</sup>* is a data matrix that includes the endogenous variables' time series, the interpretive variables *x* and the time dummies, and *d<sup>i</sup>* is a (*T<sup>i</sup>* − *q*) × 1 vector of ones.

Various problems might occur in panel data analyses; this is why several econometric tests are performed before the regression analysis. One of these problems is the correlation of the variables in the dataset. Pesaran's cross-section dependence test allows us to check if the timeseries are cross-sectional independent. In cases of cross-sectional dependence, OLS Dummy estimator (FEM) allowing for individual fixed effects with Driscoll-Kraay standard errors (in Fixed Effects models) can correct the variance–covariance matrix while, in Random Effects models, Breusch-Pagan LM test for individual effects and robust standard errors are applied.

Unit root tests are also performed, in cases where cross-section dependence is confirmed. Dickey-Fuller and Augmented Dickey-Fuller tests can be performed in panel data, with the issue of homogeneity in the autoregressive parameter. In addition, Westerlund tests are performed for panel cointegration, based on the significance of the error correction term in the error correction model. Four tests that check for panel cointegration are proposed: the Gt and Ga statistics, testing for the null hypothesis of no cointegration of all cross-sectional units. The rejection of the hypothesis implies cointegration for at least one unit. The Pt and Pa statistics also test the null hypothesis of no cointegration and their rejection implies cointegration for the panel in total [26,37–39].

#### **4. Results**

The descriptive statistics of the indicators used in the analysis (Section 3), are presented in this section. The highest percentages of people that were unable to keep their home adequately warm were observed in Bulgaria for various years, while high percentages were also observed in Portugal as well. The lowest percentages were found in Luxembourg and Norway. Greece presented the highest percentages of people facing arrears on utility bills, while high percentages were also found in Bulgaria. In contrast, the lowest percentages were found in Luxembourg and the Netherlands. Poland presented the highest percentages of people living in dwellings with leak, damp, or rot, while high percentages were also observed in Latvia and Cyprus. The lowest percentages were found in Finland for most years.

GDP per capita was also included in the analysis, expressed in purchasing power standards. The highest levels of GDP per capita were found in Luxembourg for most years, while the lowest levels were observed in Bulgaria. In addition, the highest levels of fossil

fuels final energy consumption in households per capita were found in Luxembourg for most years, while the lowest levels were found in Iceland. Similarly, the highest levels of renewables and biofuels final energy consumption in households per capita were found in Latvia, while zero levels were observed in Malta for various years.

Table 2 presents the descriptive statistics of the variables used in our analysis.


**Table 2.** Descriptive Statistics of used indicators.

To check for cross-section dependence, a Pesaran test is performed. All results reject the null hypothesis, indicating the existence of cross-section dependence, suggesting thus, the use of Driscoll-Kraay standard errors for the static regression models, in order to correct the variance–covariance matrix (Table 3).


**Table 3.** Pesaran CD test for cross-section dependence.

Note: The null hypothesis assumes that there exists no cross-section dependence (correlation). Significance at \*\*\* 1%.

Fisher-ADF and Fisher-PP unit root tests are performed, which suggest that the examined variables are I(1), with stationarity evidence in first differences (Table 4).

Westerlund tests are performed to test for panel cointegration. The results suggest that the Gt and Ga statistics reject the null hypothesis in most cases, implying cointegration for at least one unit. In addition, the Pt and Pa statistics reject the null hypothesis in every case, implying cointegration for the whole panel (Table 5).

Six regression models were formulated, where each one of the three main variables that are considered to be core elements of energy poverty are used as dependent variables. In the static models, fixed effects model specifications are used, based on the results of the Hausman tests, with FE Driscoll-Kraay standard errors, based on the results of the Pesaran CD tests.

The results indicate that GDP per capita is negatively linked to each one of the three studied variables, in both static and dynamic models. This confirms the fact that economic growth can improve the conditions of energy poverty, since an increase in GDP per capita would lead to a decrease in the percentages of energy poverty factors, as well as that a financial crisis can significantly impact energy poverty conditions. These findings can be supported by other similar studies in the literature [26,27].


**Table 4.** Fisher-ADF and Fisher-PP panel unit root tests.

Note: The null hypothesis assumes that the variable contains unit root. *p*-values in brackets. Significance at \*\*\* 1%.

#### **Table 5.** Westerlund Panel Cointegration Test.


Note: The null hypothesis assumes no cointegration. Significance at \*\*\* 1%.

Fossil fuels final energy consumption in households per capita is inversely linked in both static and dynamic models to two out of the three indicators, indicating that an increase in the consumption of energy derived from fossil fuels can improve energy poverty conditions. Thus, it is proven that the increased use of fossil fuels per capita, which implies higher energy consumption, leads to better conditions regarding energy poverty in households. In the case of inability to keep the home adequately warm, the static model also indicates that the use of fossil fuels in energy consumption can improve these conditions; in contrast, the dynamic model indicates the opposite, implying a static rather than a dynamic influence of fossil fuels usage in such analyses.

Renewables and biofuels final energy consumption in households per capita is found to be a driver of arrears on utility bills, indicating that in the studied time period, higher levels of renewables' use led to difficulties in paying utility bills on time. These findings can also be supported by studies in the literature, where it has been argued that sometimes the costs to support renewables' technologies are transferred to consumers [28], leading therefore to higher electricity prices [26] and explaining, thus, the existence of arrears on utility bills.

In contrast, energy consumption produced from renewables is linked to an inverse relationship to the presence of leaks, damp, and rot in dwellings, according to both static and dynamic model. This indicates that an increase in renewable energy consumption per capita can improve these conditions in households. At the same time, a similar relationship is observed between renewable energy consumption and inability to keep the home adequately warm, according to the dynamic model, indicating that higher levels of renewable energy consumption per capita, would improve people's ability to keep their houses adequately warm. In the static model, renewables final energy consumption in households per capita is statistically insignificant (Table 6).

**Table 6.** Regression results with three different dependent variables.


Note: t-Statistics in parentheses and *p*-values in square brackets. Parentheses in Wald and Sargan tests indicate degrees of freedom. Critical values for the Wald test of overall significance of the explanatory variables: χ20.05,3 = 7.815. Critical values for the Sargan test for over-identifying restrictions: χ20.05,24 = 36.415, χ20.05,25 = 37.652. Significance at \*\*\* 1%, \*\* 5% and \* 10%.

The lag of the dependent variables in the dynamic models are autoregressive-distributed lag specifications, that end up as an AD (1,0) formulation, showing the adjustment to equilibrium values. Table 7 presents the adjustment coefficients of each dynamic model, the discrepancy that is eliminated in a year between the actual and desired values and the periods that are required for the adjustment.


**Table 7.** Adjustment to equilibrium values, based on the lag in the dynamic models.

Wald tests of joint significance, as well as Sargan tests of over-identifying restrictions, are asymptotically distributed as χ <sup>2</sup> variables. Sargan statistics imply evidence of serially uncorrelated errors, since the null hypothesis of over-identifying restrictions is not rejected. AR(1) and AR(2) tests for first and second order serial autocorrelation do not reject the hypothesis of no autocorrelation.

#### **5. Case Studies: Sweden, Germany, and Greece**

Three countries, with different socioeconomic and environmental conditions, were chosen and their policies and progress on energy transition as well as their energy povertyrelated conditions are analyzed, compared, and discussed in this section. The chosen countries were Sweden, Germany, and Greece. The three European countries were selected due to their different characteristics and the different socioeconomic, environmental, climatic and energy conditions existing in each one of them.

Sweden is a country in Northern Europe that is characterized by its proactivity on environmental issues and its climate consensus [42]. The country has been characterized as a global leader when it comes to low-carbon economy and has followed a successful path towards energy transition [43]. Sweden's energy needs are covered mainly by hydropower and biomass and the country's geography with moving waters and a big percentage of forest coverage is assisting that, despite the cold climate that requires a high amount of energy for heating [44].

The Swedish energy policies, which aim to promote sustainability, are based on energy policies set by the EU. The EU targets refer to reducing energy consumption by 32.5%, to provide at least 32% of energy consumption from renewable sources and provide at least 14% of energy consumption in the transport sector from renewable sources. Specifically for the Swedish targets, the country has aimed to achieve by 2030, 50% more efficient energy consumption, compared to 2005. In addition, the country's goal is to cover 100% of its electricity needs from renewable energy sources, by 2040 [45].

Germany is a country in Central Europe and has a highly industrialized economy that has been promoting an ambitious plan of energy transition over the past years [46]. The country aims to promote an economy that is low carbon, sustainable, and energy efficient and has achieved a significant growth when it comes to renewable power generation capacity, actively promoting the transition to renewable energy [47].

Since 2010, Germany has initiated and promoted a plan for a more efficient energy system that is based mainly on renewable energy sources, called *Energiewende*. More specifically, *Energiewende*'s targets include the provision of 50% of electricity supply by renewable energy sources and coal's phase-out by 2038 [48]. Data shows that two thirds of Germany's power generation could be covered from renewables by 2030, while solar energy and wind energy could cover half of that proportion [47]. However, and despite the progress that has been made, the evidence shows that the country is struggling to meet its targets, mainly due to the uneven progress that exists across sectors and challenges, especially in transportation and heating. At the moment, Germany uses fossil fuels at a high degree to cover its energy needs and coal is the largest source of power generation, although it is planned to be phased out by 2038 [49].

Greece is a country in South-eastern Europe that is also implementing reforms in the energy sector in order to foster decarbonization and promote a just energy transition. More specifically, the country aims to achieve a reduction in its greenhouse gas emissions by more than 56%, by 2030 (compared to 2005 levels), aiming to achieve by 2050 a climate neutral economy. However, at the moment, fossil fuels are the primary energy supplier in the country [50].

As stated in the National Energy and Climate Plan 2021–2030, Greece aims to increase the share of renewables to 31% by 2030, in order to contribute to the achievement of the EU target, that aims to increase the share of renewables to at least 32% by 2030. In addition, the country aims to reduce the use of lignite that is used for power generation, and to shut down by 2028 the lignite-fired plants, while ensuring energy security and promoting energy efficiency [51].

From this evidence, it is obvious that Sweden is in the lead when it comes to energy transition policies and promotion, while Greece is comparatively slower in the process of decarbonization, promoting less ambitious policies. This can also be observed by the provided data. As seen in Figure 2, among the three studied countries, Sweden has the biggest share of renewable energy in gross final energy consumption, reaching 56.39% in 2019. The country has achieved the target of reaching 49% by 2020, towards Europe 2020 target, which aimed to increase the share of renewable energy in gross final energy consumption in the EU to 20% by 2020. In Greece, the share of renewable energy reached 19.68% in 2019, and the country has also achieved the target of reaching 18% by 2020. In Germany, the same share was estimated at 17.35% in 2019 and the country was very close in achieving the target of 18% by 2020 [52]. Energies 2022, 15, x FOR PEER REVIEW 12 of 17

Figure 2. Share of renewable energy in gross final energy consumption in Sweden, Germany, and Greece (%). **Figure 2.** Share of renewable energy in gross final energy consumption in Sweden, Germany, and Greece (%).

The Energy Transition Index takes into consideration the performance of the current energy system as well as the enabling environment for energy transition and aims to reflect the relationships and dependencies that exist between the energy system transformation and various factors (economic, social, political, regulatory) that determine whether a country is ready for transition. According to the Energy Transition Index 2021, and as seen in Figure 3, Sweden is the global leader, ranking in the first place, while Germany is found in the 18th place and Greece in the 54th [53]. These rankings highlight once more the differences on energy transition potential and progress that exist among the selected countries and the necessity of efforts required to move towards sustainability in each one of them. The Energy Transition Index takes into consideration the performance of the current energy system as well as the enabling environment for energy transition and aims to reflect the relationships and dependencies that exist between the energy system transformation and various factors (economic, social, political, regulatory) that determine whether a country is ready for transition. According to the Energy Transition Index 2021, and as seen in Figure 3, Sweden is the global leader, ranking in the first place, while Germany is found in the 18th place and Greece in the 54th [53]. These rankings highlight once more the differences on energy transition potential and progress that exist among the selected countries and the necessity of efforts required to move towards sustainability in each one of them.

After the comparative analysis of the energy transition progress and policies promoted in each one of the studied countries, an overview of the energy poverty situation and the policies regarding energy poverty is also presented. As seen in Figure 4, Greece was the country with the highest percentages of people living under energy poverty conditions among the three examined countries, according to Eurostat data. In contrast, low percentages were observed in Germany and Sweden for all three indicators; these

Figure 3. Energy Transition Index 2021.

Greece (%).

sustainability in each one of them.

After the comparative analysis of the energy transition progress and policies promoted in each one of the studied countries, an overview of the energy poverty situation and the policies regarding energy poverty is also presented. As seen in Figure 4, Greece was the country with the highest percentages of people living under energy poverty conditions among the three examined countries, according to Eurostat data. In contrast, low percentages were observed in Germany and Sweden for all three indicators; these After the comparative analysis of the energy transition progress and policies promoted in each one of the studied countries, an overview of the energy poverty situation and the policies regarding energy poverty is also presented. As seen in Figure 4, Greece was the country with the highest percentages of people living under energy poverty conditions among the three examined countries, according to Eurostat data. In contrast, low percentages were observed in Germany and Sweden for all three indicators; these countries perform better compared to the EU average on the specific indicators, while Greece has a significantly lower performance compared to the EU average. Energies 2022, 15, x FOR PEER REVIEW 13 of 17 countries perform better compared to the EU average on the specific indicators, while Greece has a significantly lower performance compared to the EU average.

Figure 2. Share of renewable energy in gross final energy consumption in Sweden, Germany, and

The Energy Transition Index takes into consideration the performance of the current energy system as well as the enabling environment for energy transition and aims to reflect the relationships and dependencies that exist between the energy system transformation and various factors (economic, social, political, regulatory) that determine whether a country is ready for transition. According to the Energy Transition Index 2021, and as seen in Figure 3, Sweden is the global leader, ranking in the first place, while Germany is found in the 18th place and Greece in the 54th [53]. These rankings highlight once more the differences on energy transition potential and progress that exist among the selected countries and the necessity of efforts required to move towards

Figure 4. Data regarding inability to keep home adequately warm, arrears on utility bills, and **Figure 4.** Data regarding inability to keep home adequately warm, arrears on utility bills, and presence of leak, damp or rot in the dwelling for Germany, Greece, and Sweden (% of population).

presence of leak, damp or rot in the dwelling for Germany, Greece, and Sweden (% of population). The data presented here for the three selected countries can also be compared to the findings from the analysis presented in Section 4. More specifically, we have found that an increasing GDP per capita can improve energy poverty conditions and, if we look The data presented here for the three selected countries can also be compared to the findings from the analysis presented in Section 4. More specifically, we have found that an increasing GDP per capita can improve energy poverty conditions and, if we look closely at the data of the selected countries, we can see that in periods when GDP per

closely at the data of the selected countries, we can see that in periods when GDP per capita was increasing in the studied countries, energy poverty levels were lower. Simi-

bles final energy consumption in households per capita, higher percentages of households facing arrears on utility bills were also observed. However, it is important to further research each country's social, economic, and environmental conditions and active policies, in order to better understand the differences in energy poverty conditions and

According to the EU Energy Poverty Observatory, both Germany [54] and Greece [55] have an active research community, concerning the field of energy poverty, while the research in Sweden does not specifically focus on energy poverty, but on other en-

Based on this evidence, it can be observed that Greece and Germany have developed and promoted various policies to lower energy poverty levels, while Sweden has not been actively addressing the problem, due to the already low levels that are observed in the country. Instead, Sweden is focusing on energy transition and renewable energy sources and has set ambitious goals. Germany and Greece have also been promoting energy transition policies, but their energy needs are still mainly covered by fos-

promote targeted and more effective policies and strategies.

ergy-related fields, such as energy transition and efficiency [56].

sil fuels.

6. Conclusions

capita was increasing in the studied countries, energy poverty levels were lower. Similarly, the results regarding the effect of renewable energy consumption per capita can also be validated, since we can observe that in most periods of higher levels of renewables final energy consumption in households per capita, higher percentages of households facing arrears on utility bills were also observed. However, it is important to further research each country's social, economic, and environmental conditions and active policies, in order to better understand the differences in energy poverty conditions and promote targeted and more effective policies and strategies.

According to the EU Energy Poverty Observatory, both Germany [54] and Greece [55] have an active research community, concerning the field of energy poverty, while the research in Sweden does not specifically focus on energy poverty, but on other energyrelated fields, such as energy transition and efficiency [56].

Based on this evidence, it can be observed that Greece and Germany have developed and promoted various policies to lower energy poverty levels, while Sweden has not been actively addressing the problem, due to the already low levels that are observed in the country. Instead, Sweden is focusing on energy transition and renewable energy sources and has set ambitious goals. Germany and Greece have also been promoting energy transition policies, but their energy needs are still mainly covered by fossil fuels.

#### **6. Conclusions**

At the moment, the world depends highly on fossil fuels, despite their disadvantages and their impact on the environment. Energy transition and the use of renewable energy sources has been promoted a lot more in the past few years and a lot of countries have made significant progress to that end. However, the impact that energy transition and the use of renewable sources could have on the problem of energy poverty should be continuously studied and addressed.

This study contributes to the existing literature and expands the knowledge on the topic, focusing on assessing the impact that fossil fuels and renewables usage had on energy poverty conditions in 28 European countries during the time period 2004–2019. The necessary data were extracted from the Eurostat database and an in-depth econometric methodology was followed which, to the best of our knowledge, has not been used in similar studies. The findings suggest that GDP per capita and fossil fuels final energy consumption in households per capita are linked to an inverse relationship to energy poverty conditions. In addition, the results indicate that an increase in renewables and biofuels final energy consumption in households per capita led to an increase in the percentage of people facing arrears on utility bills, while it led to a decrease in the percentage of people that cannot keep their home adequately warm, and of the percentage of population living in a dwelling with a leaking roof, damp walls, floors or foundation, or rot in window frames or floor.

These results highlight the fact that higher fossil fuels usage per capita can improve energy poverty conditions, while also highlighting the assistance that a higher level of renewables usage per capita can provide in certain energy poverty conditions. Attention should be given, though, on the mitigation of the impact that renewables' usage can have on arrears on utility bills. Government and policy makers should be aware of this relationship and develop strategies that promote energy transition while protecting energy consumers. More specifically, there should be given extra attention in not transferring the costs of renewables to consumers [28] and in promoting policies that assist households pay their bills on time, while also ensuring that the mitigation of fossil fuels usage will not have an impact on system's stability [12], leading to other social or economic problems.

Additionally, three case studies were examined and the conditions in three European countries with different socioeconomic and environmental characteristics (Sweden, Germany, and Greece) were presented and compared. The evidence shows that Germany and Greece have focused on energy poverty mitigation while Sweden, which manages to keep its energy poverty levels significantly low, promotes more ambitious strategies

regarding energy transition. Between the examined countries, Greece is the one with the highest energy poverty levels while Sweden is the one with the highest share of renewable energy in gross final energy consumption.

This comparison can be useful for policy makers, since it highlights the differences that exist among European countries in these fields and the importance to promote and implement targeted policies in each country, based on their progress and needs. Outside Europe, and when it comes to developing countries, policy makers should take into consideration other studies in the literature to support effective renewable energy development, focusing on market guarantee, lowering the costs of licensing for renewable projects, raising public consciousness, and increasing R&D, among others [57]. In general, the role that cities and communities play in ecological and energy transition should be examined and taken into consideration by policy makers, when promoting relevant strategies [58], while it would also be useful to explore the impact that subsidies towards green resources can have in supporting energy transition [59] and, subsequently, how these could impact energy poverty conditions in certain countries and regions. Finally, the impact of the COVID-19 pandemic on environmental matters and on renewable energy should be taken into consideration when promoting strategies for specific countries or regions, assessing the impact that the pandemic had on energy markets [60].

While the results of this study highlight the linkages that exist between energy coming from different sources and energy poverty conditions and can be proven helpful for governments and policy makers, future research on this relationship is strongly suggested. More specifically, extensive research targeted to specific countries or regions is essential, examining not only the current situation and the current linkages, but also the tailored policies and strategies that should be promoted for achieving a successful energy transition while ensuring energy security, minimizing energy poverty levels, and progressing on the targets of the 7th Sustainable Development Goal at the same time.

**Author Contributions:** Both authors G.H. and E.-C.G. contributed equally to each section of this paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Data Availability Statement:** Publicly available datasets were analyzed in this study. These data can be found in the References section.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

### *Article* **Determinants of Renewable Energy Consumption in Africa: Evidence from System GMM**

**Adedoyin Isola Lawal**

Department of Economics, Bowen University, Iwo 232102, Nigeria; adedoyin.lawal@bowen.edu.ng or l.adedoyin@yahoo.com; Tel.: +23-480-3523-3567

**Abstract:** The adoption of renewable energy remains Sub-Saharan Africa's best option to achieve sustainable growth and mitigate climate change. The essence of this study is to examine the factors that determine the adoption of renewable energy adoption in Africa by employing the System Generalized Methods-Of-Moment (GMM) to analyze data sourced from 1990 to 2019 on some selected African economies. The study examined the tripartite role of the economic, environmental, and sociopolitical factors on renewable energy adoption in Africa and noted that a positive relationship exists between economic and renewable energy adoption, supporting the validity of the feedback hypothesis. Hence, a policy that supports simultaneous growth of the economy and renewable energy could be adopted. The results further show that environmental factors such as carbon emission and ecological footprint negatively impact renewable energy (RE) adoption in Sub-Saharan African economies. The impact of socio-political factors is, at best mixed; for instance, the result of urbanization is positive and significant, suggesting that urbanization helps in the quick adoption of renewable energy in the studied economies, while the results of corruption show otherwise. To account for single-country dynamics, the study employed the full PMG and noted that the pollution haven hypothesis holds for a number of African economies. The results offer some policy implications.

**Keywords:** renewable energy; climate change; carbon emission; economic growth; Africa

#### **1. Introduction**

Top on the agenda of global policymakers is defining and designing suitable energy, economic, and environmental policies that can mitigate increasing global carbon dioxide emissions (CO2) [1–5]. This is premised on the fact that increasing CO<sup>2</sup> emission negatively impacts human wellbeing and health and poses a threat to handing over a secure and sustainable environment to the future generation [6,7]. Achieving sustainable environmental policies capable of reducing CO<sup>2</sup> emissions requires a comprehensive and robust understanding of its causes [8–13]. Extant literature suggests that to keep humanity and prevent negative alteration of man's state; concerted efforts must be taken to reduce and mitigate the impact of greenhouse gas (GHG) emissions and keep the average global temperature at the pre-industrial state of less than 2◦ C (IPCC 2007, Kyoto Protocol 1997) [4,14–16].

Evidence such as continuous occurrences of super droughts, wildfires, and hurricanes, among others that suggest the intensification of extreme weather events and natural disasters occurring in higher numbers or frequencies as well as magnitude across the globe call for urgent attention from governmental and non-governmental organizations, bilateral and multilateral institutions, to mitigate climate change/CO<sup>2</sup> to avert global disaster [4,5,17,18]. Several actions and policies have been canvassed by various international institutions to curb the negative impact of CO<sup>2</sup> emissions over the years [19–21]. Some of these policies often center on improving energy efficiency, conserving energy, and designing energy strategies [22]. The main drivers of these policies are reducing the high levels of CO<sup>2</sup> emission from intense nonrenewable energy sources and reducing the high percentage of nonrenewable energy in the total energy component (nonrenewable accounts for more than

**Citation:** Lawal, A.I. Determinants of Renewable Energy Consumption in Africa: Evidence from System GMM. *Energies* **2023**, *16*, 2136. https:// doi.org/10.3390/en16052136

Academic Editor: George Halkos

Received: 30 January 2023 Revised: 14 February 2023 Accepted: 17 February 2023 Published: 22 February 2023

**Copyright:** © 2023 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

80% of the global total energy components). At the center of these two policies is the need to increase the world component of renewable energy in the global energy mix.

Over the past decades, advocacy has identified renewable energy (RE) sources as reliable alternative sources of energy to conventional fossil energy sources such as crude oil, coal, and natural gas, stressing that they have some added advantages of being environmental-friendly, readily available, among others [23]. As noted by [22], there is a rapid decline in the generation cost of renewable energy. There has been strong advocacy for its usage by international organizations such as the 1997 Kyoto Protocol, the 2016 Paris agreement (COP21), the International Energy Agency, and the United Nations, just to mention a few, as it is environmentally friendly and possess the ability to mitigate climate change, produces either no or minimal global warming emissions [24,25]. Essentially, RE promotes economic growth in a number of ways. (i.) RE technologies support the diversification of the energy mix and support energy security via the provision of a reliable, vast supply of renewable energy necessary to achieve sustainable economic growth. (ii.) RE advances both social and environmental benefits as it reduces the amount of CO<sup>2</sup> emission into the environment, hence reducing the cost of addressing environmental pollution. (iii.) Developing RE sources assist economies in becoming self-reliant for energy and avoiding energy shortages arising from external shocks. (iv.) RE creates job opportunities, among others. It is also worth noting that the continuous shocks or upsurge in oil prices and prices of other fossil fuels against the continuous fall in RE technologies are incentives to shifts towards RE sources adoption [26].

Despite the strength of RE as a source of energy, its universal adoption has been relatively slow. For instance, 80% of the world's energy mix is still comprised of nonrenewable energy. This will have a negative effect on the effort to switch toward a green and sustainable energy system. Hence there is a need to explore the drivers of the deployment of RE to know what factors maximize the achievement of sustainable energy. According to [27], factors that can influence the adaption of RE can be classified into nine strands: political, institutional, economic, social, environmental, regulatory, technical, technological, and logistics.

Extant literature on the determinants of RE adoption is multi-dimensional, focusing on energy indicators, environmental factors, explanatory variables, regions and countries, time periods, econometric models, and estimation techniques [27]; for instance, [8,28–31]. In terms of the methodology adopted, ref. [32] canvassed for strong modeling techniques, ref. [33] employed panel data estimation techniques, ref. [34] employed panel autoregressive distributed lag (P-ARDL), and ref. [35] employed bootstrap ARDL, among others. A closer look at most of the extant studies suggests that though Africa has a huge reserve of renewable energy, few studies have been conducted on the possibility of switching toward the adoption of renewable energy. There are few appreciable studies on the determinants of drivers of RE adoption on the continent. This, among others, is the essence of the current study.

A major factor in mitigating increased emission rates is adopting RE in the production and consumption life. RE is healthy for public health, the environment, and the economy; hence, focusing on adopting RE is key to achieving environmentally sustainable economic growth. RE, among others, helps in diversifying the energy mix, increases energy security as it provides a reliable, vast, and renewable supply of energy needed for sustainable growth, and reduces environmental costs owing to addressing issues related to CO<sup>2</sup> emissions. Specifically, RE can be influenced by three main constructs: economic, environmental, and socio-political factors [36–38]. The impact of economic growth on RE adoption could be explained by the influence of macroeconomic variables such as real gross domestic product (RGDP), foreign direct investment (FDI), financial development (FD), and trade openness (TRD), among others. Similar to every source of energy, four possibilities exist in explaining the linkages between economic growth and RE. They are RE-leading, economic growth following hypothesis; economic growth leading, RE following hypothesis; feedback hypothesis where a bilateral relationship exists between RE and economic growth; the

fourth possibility is the neutrality hypothesis, where no causality exists between economic growth and RE [20,39,40].

The impact of environmental factors on RE adoption is essentially influenced by two models: the environmental Kuznets hypothesis (EKC); and the pollutant haven models. The EKC noted that a U-shape relationship exists between economic growth and environmental pollution. The theory simply described a non-linear relationship between growth and environmental degradation. The pollutant haven model stressed that the existence of legislation to punish the deployment of environmentally harmful energy sources would motivate the adoption of RE [41–43].

The socio-political strands focus on the ability of governance structure, government policies, and urbanization, among others, to influence the adoption of RE [44–46]. Urbanization as a socioeconomic factor impacts energy consumption and environmental condition as it may induce the enlargement of energy-intensive industries such as steel and concrete, the power industry, and the transport sector, thereby provoking upward shocks to the environment [47]. Another dimension to the contributions of urbanization to energy consumption and the environment suggests that urbanization might improve the environmental quality, provided man is willing to be environmentally conscious and friendly. As important as these constructs are to the adoption of RE, few studies have accounted for them in RE adoption works. For instance, refs. [1,2] did not account for the impact of socio-political factors, and [48–51] only focused on environmental factors. Refs. [52–54] focused on both economic and environment but did not address socio-political factors. The crux of the current study is to calibrate these constructs to discuss RE adoption with a focus on Africa.

Several factors induced our motivation on Africa (as of 2017, African CO<sup>2</sup> emission was 4% of global CO<sup>2</sup> emissions. It grew at an average of 4.6% yr−<sup>1</sup> over the period 1990–2017 against the global rate of 12.2% yr−<sup>1</sup> ); for instance, extant studies and reports have noted the deteriorating nature of the environmental space in Africa over the last few decades [55,56]. Air pollution and CO<sup>2</sup> emissions account for environmental degradation in the region more than other types of pollution, such as water or land pollution. The region is reported to have one of the most prolonged CO<sup>2</sup> emission growth rates in the world, with more than a 123% growth rate between 1979 and 2017, surpassing the global average of 60% [57,58]. With the current trend in CO<sup>2</sup> emission growth rate, Africa will, by the year 2030, have a 30% CO<sup>2</sup> emission growth rate.

The essence of this study is to investigate the nexus between RE and economic growth, the environment, and socio-political factors by employing a System Generalized Methods-Of-Moment (GMM) model. Our choice of system GMM was influenced by its many advantages over alternative estimation techniques, such as the difference GMM. For instance, system GMM has three clear-cut advantages: (i.) It is useful in reducing endogeneity bias; (ii.) it reduces time-varying measurement error bias; (iii.) it reduces weak instrument error bias [20,59,60]. The system GMM helps address the issues related to endogeneity resulting from the inclusion of other potential endogenous explanatory variables, as well as other possibilities of measurement errors owing to the use of cross-country data displaying high persistence [61–65]. The study intends to ask the following research questions: (i) What drives renewable energy adoption in Africa? (ii) To what extent do macroeconomic variables impact renewable energy adoption in Africa? (iii) Do environmental factors impact renewable energy adoption in Africa? (iv) What is the role of socio-political factors in renewable energy adoption in Africa?

Our study's novel contribution to literature is four-fold. First, to the best of our knowledge, we are among the first studies to examine the drivers of RE consumption in Africa. Africa, as a growing economy, is in dire need of energy and is simultaneously faced with the need to have a safe environment given the alarming rate of CO<sup>2</sup> emission of 123%, surpassing the global average of 60%. Therefore, Africa needs to switch from traditional fossil fuel-dominated energy sources to clean and safer RE sources; hence the need to understand the drivers of RE adoption for appropriate policy adjustment. Secondly, we account for the role of macroeconomic variables, environmental constructs, and sociopolitical factors in the nexus between energy, economics, and the environment. Thirdly, we employed novel and appropriate estimation techniques, the system GMM which is useful in reducing endogeneity bias, time-varying measurement error bias, and weak instrument error bias, and reducing measurement errors owing to the adoption of cross-country data. Fourthly, we offer some policy implications.

Our study will provide insights into at least six Sustainable Development Goals (SDGs): SDG 7- affordable and clean energy; SDG 8- economic growth; SDG 11- sustainable cities and communities; SDG 12- responsible consumption and production; SDG13- climate action; and SDG 17- partnership for the goal with trade offering leadership.

The remainder of this study is as follows: Section 2 presents the literature review; Section 3 presents the materials and estimation techniques; Section 4 deals with the presentation and discussion of results, while Section 5 concludes the paper.

#### **2. Literature Review**

The theoretical note that governs this study is threefold: cointegration (economic growth-related), environmental, and impact. The cointegration (economic growth) strands are further divided into four hypotheses that explain the possibility of causality between RE and economic growth. These hypotheses are energy-leading growth following hypothesis, which states that it is the demand for energy that spurs economic growth; hence, conservative measures to conserve the environment will have negative consequences on economic growth. The second leg of this strand is the economic growth-leading following hypothesis that suggests that it is growth that drives energy demand. The third strand is the feedback hypothesis which states that a bilateral relationship exists between economic growth and energy consumption. The fourth hypothesis is the neutrality hypothesis which suggests that no causality exists between economic growth and energy consumption. Hence, any policy introduced to manipulate either of the two will have little or no effect on the other [20,66,67].

The discussion of the extant literature on the impact of macroeconomic variables on energy behavior remains inconclusive; for instance, ref. [2] examined the dynamic effect of nonrenewable energy, renewable energy, economic growth, and foreign direct investment on the environment based on data sourced from the year 2000 to 2015 for some selected African economies. The study employed panel ARDL that calibrates the pooled mean group, mean group, and dynamic fixed effect estimator to examine the validity of both the environmental Kuznets curve and/or pollution haven hypothesis. The result attained shows that while a negative and significant relationship exists between renewable energy and CO<sup>2</sup> emissions, the relationship between CO<sup>2</sup> and other explanatory variables is positive and significant, both in the short and long runs, except for FDI, which is positive only in the long run. The study noted that EKC does not hold for the studied economy; as a result, it tilts towards the pollution haven hypothesis. This suggests that African economies are less concerned about their environment but place a high premium on growth. A major difference between ref. [2] and the current study is the fact that whereas the former does not discuss socio-political factors, the latter calibrated it into their model; the current study accounts for single-country analysis.

For some selected 55 economies, ref. [68] employed a two-system GMM procedure to examine the nexus between financial development and renewable energy adoption based on data sourced from 2005 to 2014. The study noted that a positive and significant relationship exists between financial development and renewable energy for high-income economies though the relationship is insignificant for low-income economies. The study noted that sophisticated financing is key to achieving RE in the studied economies. The study also noted that the impact of trade openness and carbon emission are statistically insignificant for the economies studied, suggesting that trade has no impact on RE adoption. The results from the impact of carbon emission on RE adoption are intriguing, especially for high-income economies. The authors concluded that the EKC model is valid for the studied economies.

In a related development, ref. [69] noted that financial development is key to achieving the adoption of RE in China. The study emphasized the role of green financing and a green reputation in achieving the deployment of renewable energy that will support growth. The study employed several econometric techniques to analyze both micro and macro data on the Chinese economy from 2015 to 2020. The study identified oil price volatility and geopolitical risk as key obstacles to adopting RE in China. In a related development, ref. [70] noted that financial development is key to achieving RE consumption in Africa based on the study estimation of the generated method of moments (GMM) and quantitative regression (QR) in analyzing data sourced from 2004 to 2014. The study noted that financial inequality is a major setback to progress in RE consumption in Africa.

Ref. [71] noted that financial development, agriculture, and economic growth are key to the adoption of RE in Africa, while corruption and bad governance negatively affects Africa's adoption of RE. The study analyzed case studies, research articles, policy briefs, and project reports across and beyond Africa. It noted that for Africa to achieve the SGDs, the operations of Power Africa, Sustainable Energy for All (SE4ALL) initiative, concerted efforts must be put in place to address corruption on the continent.

Ref. [72] noted that FDI negatively impacts the environment on the one hand and RE consumption on the other hand for China based on the results obtained on the deployment of systems GMM, random effect, and fixed effect on the annual date from 2011 to 2016. The study noted that the pollution haven hypothesis is valid for the study economy.

Ref. [73] noted that RE and nonrenewable energy (N-RE) are key determinants of FDI inflows. Trade, tourism, and market size play positive but less significant roles in attracting FDI for the BRICS, stressing that a negative relationship exists between FDI and inflation rate. Ref. [74] noted that RE has a neutral effect on FDI. Instead, the institutional environment and land availability are the core factors that stimulate FDI. Ref. [75] noted that a long-run relationship exists between FDI, RE, and economic growth for some selected nine countries identified in the Climate Change Performance Index 2018 report

Ref. [76] estimation of data from G-C economies based on data sourced from 1978 to 2014 shows that capital market expansion and trade openness are the leading drivers of CO<sup>2</sup> emission. The results further noted that CO<sup>2</sup> is respectively related to RE adoption (see also ref. [77]. Their results tilt toward the pollution haven hypothesis

In agriculture, ref. [78] shows that a long-run relationship exists between agricultural land expansion and CO<sup>2</sup> emission in Peru though RE improves environmental quality by reducing CO<sup>2</sup> emission. Ref. [79] noted that a positive relationship exists between agriculture and RE, but no such relationship is found to exist between agriculture and CO<sup>2</sup> for the economies of the US, Canada, China, and Poland. Ref. [80] noted that a bidirectional relationship exists between energy and agriculture for the EU. Ref. [81] noted that agriculture, RE, trade, and globalization negatively impact CO<sup>2</sup> emissions in Turkey. The study tilts toward the pollution haven hypothesis for Turkey.

The theoretical note from the environmental strands can be classified into two main types: The Environmental Kuznets Curve and the pollution haven hypotheses. The EKC opined that the relationship between economic growth and environmental pollution is in the form of an inverted U-shaped, such that at the early stage of a nation's economic growth, environmental pollution deepens, and after reaching a certain threshold level, environmental pollution begins to decline. The proponents of this hypothesis are of the view that at the initial stage of development, economies are concerned with achieving economic growth with less concern for protecting the environment, but with time and advancement in economic growth comes a surge in environmental pollution, and attention begins to shift towards achieving clean energy [41,42,82,83].

A variety of these models has been canvassed in the literature focusing on CO<sup>2</sup> emissions as indicators of environmental pollution [42,84]. Some have calibrated the ecological footprint [49,85]. Recent studies have calibrated macroeconomic and financerelated variables to the studies on EKC [86]. The discussion on the relevance of EKC is continuous and yet to be concluded.

The pollution haven hypothesis (PHH) is the view that multinational companies that engage in rigorous pollution fields prefer to move to developing countries with fewer environmental/ecological protection laws. The reverse of the pollution haven hypothesis is the pollution halo hypothesis, which states that FDI could induce a downward trend in CO<sup>2</sup> emission, hence promoting energy-efficient technology usage that revolved around sustainability methods. Accordingly, it is believed that FDI can positively impact the ecosystem of an economy in three channels: scale effect (economic size), technical effect (improved technology), and structural effect (improvement in manufacturing design). The interaction of these effects will improve growth and reduce CO<sup>2</sup> emissions. The proponent of this hypothesis has identified FDI and yawning for development as the key drivers of CO<sup>2</sup> emissions in developing economies [7,42,47]. Closeness to colonial masters by former colonies and globalization, among others, are the reasons that account for the movement of multinational firms with toxic production outlets to less developing economies [72].

The studies on the impact of ecological footprints suggest that a functional relationship exists between the ecological footprint and several variables. For instance, ref. [86] noted that financial debt and renewable energy help reduce environmental degradation and that financial debt, RE, and NRE positively impact the growth of the 15 highest emitting economies. Ref. [86] noted that economic growth and national resources advance the ecological footprint and that human capital in the current state cannot mitigate environmental deterioration. Though RE does decrease ecological footprint, the study established the existence of feedback causality between human capital, urbanization, and ecological footprint. Ref. [31] noted that RE decreases ecological footprint in the long run in Turkey and that a bi-directional relationship exists between RE and economic growth and ecological footprint.

The theoretical note on impact assessment focuses on the role of governance and other socio-political factors in shaping the choice of energy usage to achieve carbon neutrality. The proponents of this thought believe that climate change is a global public issue and requires effective climate governance to address it [87–90]. As noted by ref. [91], energy governance is key to decoupling carbon emissions as it is vital to promoting RE adoption. For a sample of 36 emerging economies, ref. [92] observed that good governance especially economic and institutional governance is key to mitigating CO<sup>2</sup> emission and progressive adoption of RE. Ref. [93] designed a novel, holistic analytical approach to examine energy access governance for the Southern African economies of Uganda and Zambia by employing three data collection methods: qualitative document analysis, semistructured stakeholder interviews, and closed surveys. The study noted that the rule of law, transparency standards, accountability, and inclusiveness are key to accessing RE for the studied economies. The study also noted that competing regulatory frameworks distort access to RE. Ref. [90] cautioned on the danger of monopolized power in designing and implementing RE for the economies of Nepal and Indonesia. The authors noted that RE designed in the studied economies was bedeviled with the inability to carry the major stakeholders along in its design and running.

Ref. [94] calibrated the role of corruption perception and political governance in energy consumption-economic growth nexus for a team of 49 economies using a dynamic data environment analysis model based on data sourced from 2007 to 2016. The study noted that political governance proxied by political stability, bureaucratic quality, personal safety and security of private property, and legal and regulatory frameworks positively impact energy consumption.

Ref. [89] employed machine learning techniques to analyze the impact of green governance on renewable energy consumption in India and noted that governance structure influences the adoption of energy choices. The study further noted that the taxonomy of green governance proxy by global governance, adaptive governance, climate governance, ecological governance, self-governance, energy governance, and information technology governance are related and work on the same objectives by pursuing different activities.

For Switzerland, ref. [95] examined the role of public awareness and governance structure in the effective transition from nonrenewable energy consumption to renewable energy sources. The study noted that public awareness and good governance are crucial to the effective transition and adoption of RE (see also ref. [88]. Ref. [96] explored the role of both internal and external governance structures in the adoption of renewable energy for some selected 1027 firms spread across 47 economies/regions. The study noted that internal governance structure tends to have a negative influence on RE adoption as it often induces a declining influence on RE, whereas external governance has a negative impact.

In Brazil, ref. [97] employed quantitative measures to access the nexus between water, energy, food, and land as it affects the adoption of biofuels emanating from sugarcane. The study concluded that each of these factors is key to achieving sustainable/green energy adoption in the studied economy. A study by ref. [97] was further expanded by ref. [98], who calibrated the role of geopolitics in adopting RE in Mexico. The study employed an external multi-regional input-output model (EMRIO) that calibrates import dependence and governance quality into the RE adoption framework for the Mexican economy. The study noted that better governance is key to the successful adoption and implementation of RE in the studied economy.

Ref. [50] noted that for governance structure and effectiveness to influence the adoption of RE positively, there is a need to have a holistic view of the consequences of RE adoption by calibrating natural resources extortion into the equation. The study argued that evidence abounds to show that the transition from a fossil-dominated system towards RE will have negative consequences on metal by more than a fraction of 7 by 2050 when compared with the 2015 levels, especially in economies with weak, poor, and failing resource governance up to between 32 and 40%.

Ref. [99] noted that political interference in environmental management, poor or lack of effective implementation, and lack of political independence of environmental agencies, which increases the risk of consumption, are the main factors militating against the adoption of RE in Brazil (see also ref. [100]).

A critical look at the literature reviewed here suggests that little or no study has been conducted on the determinants of renewable energy consumption in Africa, and their findings are inconclusive. This is what the current study aims to do, and by extension, calibrate the role of environmental factors, economic growth, and socio-political factors to study RE adoption in Africa.

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

The section presents the data-generating set and sources. It also presents the methodology employed and the justification for employing it.

#### *3.1. Data*

The data for the current study were sourced from several reputable global data outlets. For instance, we obtained data on macroeconomic variables, including real gross domestic product (RGDP), foreign direct investment (FDI), financial development (FD), trade (TRD), and government spending (GOVT) from the World Development Indicators (various issues). The data on the inflation rate (INF) was sourced from the United Nations Statistics (UN Data). We sourced data on agricultural output (AGRIC) from the Economic Research Service of the US Department of Agriculture (USDA). Data on environmental factors proxy by CO<sup>2</sup> emission (thousand kt) and ecological footprint were sourced from the BP Statistical Review of the World Energy (various issues). To account for the impact of socio-political factors, we calibrated the impact of governance effectiveness (GOVE) and urbanization (URB) into our model. Data on these variables were sourced from the World Development Indicators (various issues). We also account for the impact of the life expectancy index, education index, and corruption perception index as part of our socio-political factors in shaping the adoption of RE. Data on the education index (EI) and life expectancy index (LEI) were sourced from the United Nations Development Program (UNDP) development reports (various issues). Meanwhile, data on the corruption perception index (COR) was sourced from the Transparency International database. To address issues relating to heteroscedasticity, we

standardized our variables by obtaining their natural log forms. Data on renewable energy were sourced from the International Energy Agency database (various issues).

#### A Priori Expectations

Theoretically, we expect a positive relationship between RE and each RGDP, financial development, trade openness, agricultural output, and FDI. A negative relationship is expected to exist between RE and inflation. On environmental variables, we expect an inverse relationship between RE and CO2. On socio-political constructs, we expect that a positive relationship should exist between government effectiveness and RE. The relationship between RE and urbanization could be either positive (if the people are en vironmentally conscious) or negative if smart cities and the environment are neglected. The relationship between RE and other variables could be either way.

#### *3.2. Methodology*

The essence of the current study is to examine the determinant of RE adoption in some selected African economies. The study explored the neoclassical production function employed by [20,26,101] to develop the model for the study as stated below:

$$RE\_{it} = \alpha\_1 RE\_{it-1} + \beta\_1 W\_{it} + \beta\_2 X\_{it} + \beta\_3 Z\_{it} + \delta\_l + \mathcal{Q}\_i + \varepsilon\_{it} \tag{1}$$

where *REit* is the renewable energy consumption per capita, *W* is the proxy of all macroeconomic variables that can influence renewable energy consumption, *X* is the proxy of all environmental factors that can influence renewable energy consumption, *Z* is the proxy of all socio-political factors that can influence renewable energy consumption, *α and β* are the coefficients of the model, ∅*<sup>i</sup>* is the time-invariant country effects, *δ<sup>t</sup>* is the unobservable time effects, *ε* is the residual term, *t* is the time period. The GMM estimation techniques proposed by Arellano and Bond for our model, as stated in Equation (2), are as follows:

$$E(y\_{it-s} - \Delta u\_{it}) = 0 \text{ for } t = 3, \dots, \dots \\ T \text{ and } 2 \le T \tag{2}$$

Here, *yit*−*<sup>s</sup>* is the suitable lags of the dependent variables. The implication is that the second and further lags of the dependent variables are employed as an instrument for the residual of Equation (1) in differences. As noted by refs. [20,59,102], the estimator of Equation (2) is prone to a huge small sample bias, given the fact that the number of periods is small, with the dependent variables presenting a high degree of persistence. To address this, our study employed the system GMM model as suggested by refs. [40,101,103]. The model is as follows

$$E(\Delta d\_{it-s} - (\delta\_i - \mu\_{it})) = 0 \text{ for } t = \mathbf{3}, \dots, \dots \text{.} T \tag{3}$$

It becomes unattractive and inappropriate to employ the ordinary least squares (OLS), fixed effects, or random effects because of the presence of lagged endogenous variable *y*, *t* − 1 in Equation (1), given that *yit* is correlated with *δ<sup>i</sup>* and it induces upward biases, which is inconsistent with the OLS assumption of independence of the error term from the regressors [101,104–107]. To address this problem, the extant literature on dynamic panel models employed the Arellano and Bond GMM estimation model that employs an internal mechanism to explore the correlation between *y*, *t* − 1. and *δ<sup>i</sup>* . The GMM techniques remove ∅*<sup>i</sup>* in short dynamic panels such as Equation (1) by differencing it first. To obtain a relatively consistent estimator, we employed lagged values of the levels of the independent variables as the predetermined variables [26,108]. In specific, when ∅*<sup>i</sup>* (I = 1,2, . . . ., n) are serially uncorrelated, then the second and higher-order lags of the independent variables are valid instruments. Extant literature has shown that a major problem of the [104] GMM model is that it produces poor instruments for the regressors when the regressors display persistence over time. To overcome this challenge, Arellano and Bond 1995, developed a system GMM that can estimate two sets of equations: (i) A set of levels that employ lags of the regressors in first differences as instruments; (ii) a set of equations in first differences that employs lags of the regressors in level as instruments. From the narrative, it can be

deduced that the system GMM is superior and appropriate for our model when compared with the difference GMM in at least three areas. (i.) It reduces endogeneity bias. (ii.) It reduces time-varying measurement error bias. (iii.) It reduces weak instrument error bias.

The current study employed the system GMM to address the endogeneity in the data generating set that could occur as a result of including *y*, *t* − 1, an indication that RE consumption and many of the other regressors may be jointly determined by the growth rate of the GDP, as well as the possibility of measurement errors that could occur because of employing cross country data that displays high persistence. To examine the validity of the orthogonality assumption of system GMM, we employed the Hansen test of overidentification and the Arellano and Bond tests for second-order and higher-order several correlations *AR*(2) test, given that system GMM techniques rely on internal instruments. The study adopted the [109] small sample correction of the standard errors for all the two-step system analyses, as suggested by ref. [110]. Some of the variables in some of the studied economies are heterogeneous; hence, we employed a full PMG method for the short-run nexus following [111–113].

#### **4. Results and Discussion**

We present the results of the current study in two parts. The first part focuses on the nexus between RE, economic growth, environmental factors, and socio-political factors in the selected African economies, based on system GMM estimation techniques. The second is our analysis focused on the country-specific output of these relationships in each studied economy.

#### *4.1. System GMM Estimates*

We present the descriptive analysis results of the relationship between the variables explored in Table 1. The results suggest that FDI has the lowest mean, while corruption has the highest mean value. The result, as presented by standard deviation, suggests that corruption has the highest standard deviation, while RGDP has the lowest standard deviation.


**Table 1.** Descriptive statistic.

Source: Author's computation 2023.

We present the results of the impact of our independent variables on the dependent variable (RE) in Table 2. From the results, as shown in columns (1–5), the results of the OLS, fixed effects, baseline system GMM, and alternative system estimates are presented respectively for robustness purposes. As earlier noted, OLS estimation of Equation (1) induces upward bias for the lagged per RE, while fixed effects induce a downward bias. Empirically, a valid estimate is expected to lie between the OLS and fixed effects [20,60,65,110]. Our results, as presented in column 3 of Table 2, suggests that the two-step system GMM coefficient on the lagged RE is −1.582, and it is between the upward-biased OLS estimates of

−1.143 and downward-biased fixed effect estimates of 5.446. The results also suggests that our estimation is negative and highly significant. This suggests the existence of conditional convergence across the selected African economies studied (See also [111]).



Note: standard errors are reported in []; \*\*\* represent 1%. Source: Author's computation 2023.

In column [3] of Table 2, we present the results of the impact of economic growth on RE for the selected economies based on the two-step system GMM. We obtained an estimate of 0.1015 with a 1% level of significance. This suggests that economic growth promotes RE adoption in the selected economies. We validated our results by testing for over-identification restrictions and second-order serial correlation based on *AR*(2) test and the Hansen test. The *ρ*-value result of the *AR*(2) at 0.104 rules out the possibility of second or higher-order serial correlation in the residuals. The results of the Hansen test for over-identification further validated the instruments employed.

A look at the results of other explanatory variates suggests that our results are in line with relevant economic theory and existing empirical findings. For instance, the coefficients of financial development are positive and significant, suggesting that financial development aids the consumption of renewable energy. This supports the finding of [26,69]. The results from each agriculture and trade openness are positive and significant, suggesting that each of them positively supports RE adoption [71]. The results from trade suggest that trade policies such as market liberalization that supports international trade advance the adoption of RE in Africa and, by extension, advance economic growth in the region (see ref. [112]). The result on agriculture suggests that agriculture significantly aids RE adoption in the studied economies and is in line with the findings of [79,81]. As expected, the results on the relationship between RE adoption and inflation are negative and significant. This suggests that with rising inflation, peoples' adoption of RE will be slow as the purchasing power ability of the people is eroded. Our result is in line with the findings of [113].

Our results are mixed for the other explanatory constructs (environmental and sociopolitical). For instance, while the results from CO<sup>2</sup> emission and ecological footprints are negative and significant, the result from urbanization is positive and significant. This suggests that as these economies get urbanized, the adoption of RE is embraced. This also connotes that, on average, men in these economies are environmentally conscious and friendly. Our results support the findings of [86] for the BRICS economies. The result of CO<sup>2</sup> emission suggests that CO<sup>2</sup> has an inverse relationship with RE in the studied economies (see also ref. [31]). On the socio-political factors, our results also show that governance structure has a positive and significant relationship with RE adoption, suggesting that with a good governance structure, more people will embrace the adoption of RE. The results agree with the findings of [89,92–94] but contradict [71] submission for sub-Saharan African economies.

#### *4.2. Single-Country Estimates Results*

Beyond panel estimation, we provide a single-country estimation in our model to account for the heterogeneous behavior of some variables in some of the selected African economies. Hence, our study followed [111–113] to employ a full PMG test for short-run nexus. Before we employed a full PMG model, we conducted unit root tests using In-Pesaran-Shin (IPS), Levin-Lin-Chu (LLC), and cross-sectional augmented Dickey-Fuller test. Our results suggest that none of the variables in the model is found to be *I*(2)). Results are available upon demand. We employed both the Pedroni test and the Westerlund test to conduct a cointegration estimate. The results show that the long-run estimates across all countries are stronger. Results are available upon demand. The result of the full PMG estimate showing country-specific estimate is presented in Table 3. From the results, it can be deduced that the impact of RGDP on RE is positive and significant for Nigeria, Ghana, Kenya, Ethiopia, Morocco, and South Africa, though a negative and significant relationship is noted for Algeria and DR Congo. The results of other macroeconomic variables are similar. For instance, a positive relationship exists between financial development and RE in Nigeria, South Africa, Ghana, and Kenya. The result also shows that foreign direct investment positively impacts renewable energy in Nigeria, Ghana, and South Africa. Trade openness supports RE adoption in Ghana, SA, Kenya, and Nigeria. The inflation rate negatively impacts RE for all the economies studied. On environmental factors, both CO<sup>2</sup> and ecological footprint has a negative and significant impact on RE for Angola, Tanzania, Ivory Coast, Nigeria, Ethiopia, Kenya, and DR Congo. Socio-political factors exhibit some level of positive impact on RE adoption. For instance, urbanization impacts RE adoption positively in South Africa, Nigeria, Ghana, Kenya, Ethiopia, and Angola. Life expectancy and education index impact RE positively and significantly for economies such as Nigeria, South Africa, Tanzania, Kenya, Angola, Ethiopia, and Egypt. The result for corruption index offers mixed results. For instance, it was reported that a negative relationship exists between corruption and RE adoption in Nigeria, Kenya, Ivory Coast, DR Congo, Egypt, and South Africa. Overall examination of our results suggests that most of the studied economies exhibit interacting trends in the adoption of RE.


**Table 3.** Country-specific full short-run PMG estimates.


Source: Author's computation 2023. Note: \*, \*\*, \*\*\* represent 10%, 5%, 1% significant level respectively.

#### **5. Conclusions**

Decarbonization of the energy sector is at the front burner of the 21st-century energy adoption policy among economies across the world. This is essential to achieving the global quest for sustainable growth via renewable energy that helps in mitigating climate change [10]. A good understanding of the drivers of RE adoption is key to achieving success in RE growth which is essential to attaining sustainable growth. As noted earlier, four macroeconomic-energy possibilities exist in the literature on the link between economic growth and energy. They are the energy-led growth-following hypothesis, growth-led energy-following hypothesis, feedback/bi-directional hypothesis, and neutrality/indifference hypothesis, with each of these possibilities offering unique

implications for policy modeling. Beyond macroeconomic variables, environmental factors (such as CO<sup>2</sup> emission and ecological footprint) and socio-political factors (education, life expectancy, and urbanization) often determine the adoption of RE. This paper contributes to the literature by employing appropriate estimation techniques—system GMM and full PMG—to examine the factors that influence the deployment of RE in the African sub-region based on data sourced from 1980 to 2019. Our choice of the system GMM was driven by the possibility that RE adoption and other control variables employed in our model could be jointly determined. The system GMM employed can deal with endogeneity-related issues. It can address the susceptibility of data to measurement error among other things. Validating the orthogonality assumptions in the system GMM, the study employed the Hansen test of over-identification, the Arellano and Bond (2000) test of second-order serial correlation, and the small sample correction of the standard errors.

The results of our estimation techniques and the results from a series of robust tests reveal that the relationship between RE and economic growth in the studied economies is positive and significant, suggesting that RE promotes economic growth on the one hand and economic growth promotes RE adoption on the studied economies, supporting the validity of feedback hypothesis in the studied economies. Hence, policies that support RE adoption should be advanced. A look at other explanatory variables suggests that a positive relationship exists between financial development, foreign direct investment, trade, governance, urbanization, and life expectancy that stimulates RE adoption.

This suggests that for these economies to achieve sustainable growth powered by RE, policymakers need to implement policies that will promote financial development, enhance trade, promote urbanization, and promote education and governance structure. The results of the nexus between inflation and each economic growth and RE adoption are negative, suggesting that policymakers should lower the inflation rate to promote economic growth influenced by RE adoption.

The second strand of our analysis focused on country-specific estimates. From the results, it can be deduced that the results obtained in country-specific estimation are not too far from the ones obtained at the aggregate level. For instance, the connection between RE and economic growth is positive and significant for the economies of SA, Nigeria, and Kenya, suggesting the possibility of a feedback hypothesis.

Though this study has advanced literature by examining the drivers of RE adoption from the point of economic, environmental, and socio-political views, there is a need for further empirical analysis on the subject to further enhance the knowledge of this nexus. Hence, we suggest that further study could examine the impact of RE on total energy adoption in Africa. Different estimation techniques can also be employed.

**Funding:** This research received no external funding.

**Data Availability Statement:** We obtained data on macroeconomic variables, including real gross domestic product (RGDP), foreign direct investment (FDI), financial development (FD), trade (TRD), and government spending (GOVT) from the World Development Indicators (various issues). The data on the inflation rate (INF) was sourced from the United Nations Statistics (UN Data). We sourced data on agricultural output (AGRIC) from the Economic Research Service of the US Department of Agriculture (USDA). Data on environmental factors proxy by CO2 emission (thousand kt) and ecological footprint were sourced from the BP Statistical Review of the World Energy (various issues). To account for the impact of socio-political factors, we calibrated the impact of governance effectiveness (GOVE) and urbanization (URB) into our model. Data on these variables were sourced from the World Development Indicators (various issues). We also account for the impact of the life expectancy index, education index, and corruption perception index as part of our socio-political factors in shaping the adoption of RE. Data on the education index (EI) and life expectancy index (LEI) were sourced from the United Nations Development Program (UNDP) development reports (various issues). Meanwhile, data on the corruption perception index (COR) was sourced from the Transparency International database. To address issues relating to heteroscedasticity, we standardized our variables by obtaining their natural log forms. Data on renewable energy were sourced from the International Energy Agency database (various issues).

**Conflicts of Interest:** The author declares no conflict of interest.

#### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
