Impacts of Urbanization on Energy Consumption in the South Asian Association for Regional Cooperation Zone

Abstract

Energy resources play a vital role in the process of urbanization, and the high level of energy consumption has significantly created an alarming situation for environmental degradation. Increased demand for energy consumption in the South Asian Association for Regional Cooperation (SAARC) zone is a core concern for decreasing the existing reserves of energy, especially nonrenewable energy, when the growth of urbanization is increasing also. This study investigates the impacts of urbanization on energy consumption in this region by identifying factors that influence energy use. We employed globally used econometric techniques to examine the relationship between energy use and urbanization. The results of the study indicate that all the independent variables used in the model (except urban population growth) were statistically significant with a 99% level of confidence. In addition, the findings of this study recognized three long-run causalities running from the GDP (gross domestic product) to energy consumption, energy consumption to GDP, and energy consumption to the industry’s share of the countries’ GDP. We recommend (i) taking the initiative to invest in renewable energy, (ii) implementing green energy-efficient technologies in the industrial sector, and (iii) raising public awareness of the negative effects of energy use on the environment through education.

1. Introduction

Development is a continuous process where economic growth, urbanization, migration, structural transformation, technological change, education, the environment, and ethics are different indicators of this process [1]. Urbanization is defined as the process in which the population migrates from rural to urban areas; in fact, the labor force transfers from the agricultural sector to the industrial and service sectors [2]. Urbanization is considered one of the important indicators for sustainable development in any economy linked with the UN Sustainable Development Goal 11- sustainable cities and communities [3].

Agricultural and non-agricultural (industry and service) sectors are two of the significant driving sectors of economic growth as well as economic development of a country, and urbanization is a result of the structural change of these sectors. The structural transformation of the economy causes fundamental changes in natural resource use and influences energy demand in several ways [4]. For instance, urban life is expected to require more energy because of traveling to work by driving fuel-using vehicles and also due to more efforts constructing, operating, and maintaining municipal infrastructure and services including housing, water supply, roads, and bridges as compared to rural living [5]. However, economic growth has also affected energy consumption indirectly through rises in the urbanization rate, and the pattern of energy use which gradually changed in the urban areas [6]. Behera and Dash [7] believed the industrial–urban inter-linkages have been the main way to realize the growth and economic development of a society caused by the industrial revolution.

Precedent studies have shown a positive correlation between urbanization and energy consumption [8,9]. The main reason for this correlation is that the process of urbanization causes an increase in urban population yielding high urban density [10,11], which, in turn, increases the demand for new buildings and transportation, as well as the rising quality of energy-intensive lifestyles. National energy consumption is expected to increase because urban households use 50% more energy per capita than rural households [11]. The growth of urbanization in developing countries is higher compared to developed countries [12]. Moreover, increasing demand for energy consumption in developing countries like India and China is a core concern for decreasing the existing reserves of energy, especially nonrenewable energy [1]. It is predicted that 68% of the world’s population will be urban citizens by 2050, much of which will occur in Africa and Asia, notably in the SAARC countries, which will add 20% more city dwellers by this period [13]. This region has represented 3% of the world’s area, 24% of the world’s population, and 4% of the global economy, as of 2017 [14,15]. The tremendous rise of these nations requires new urban infrastructure, which increases resource consumption and puts further strain on their already vulnerable ecosystems [10]. As the above arguments indicate, more empirical analyses from different contexts are required to be able to generalize existing knowledge of the effects of urbanization on energy use.

Most of the countries in the SAARC zone (Afghanistan, Bangladesh, Bhutan, India, Maldives, Nepal, Pakistan, and Sri Lanka) are developing, especially Bangladesh, India, and Sri Lanka, and urbanization is an important indicator of their socio-economic development [16]. The aim of this study is to examine how urbanization affects energy consumption in the SAARC zone, given the hypothesis that there is a positive correlation between urbanization and energy consumption, which is graphically shown in Figure 1.

2. Literature Review

2.1. Evidence from Outside the SAARC Zone

“Urbanization can be defined broadly as an interrelated process of economic, demographic, political, cultural, technological, environmental and social changes, which involves the concentration of population and economic activities in urban areas” [18] (p. 302). The impacts of urbanization on energy consumption are explained by this concept of urbanization because energy use depends on the relationships between urban growth and human activities such as increases in production and consumption, the rising use of motor vehicles and other household energy usage, etc. [19]. Sodri and Garniwa [20] explained that economic growth increases transport demand in a country because of urbanization. Poumanyvong et al. [21] found that about 80% of the global gross domestic product in 2007 was produced in cities. An increased rate of urbanization in Africa and Asia is a challenging issue in the 21st century [22]. Madlener and Sunak [23] explained that urban growth has significantly arisen in less developed countries compared to more developed countries between 1970 and 2010, and it is projected that the urban population will almost double with the highest average urban growth rate of 3.3% per annum between 2010 and 2050 in these countries. Urban areas cover only 2% of the world surface, but they represent about 75% of the world’s consumption of resources and produce 70% of the world’s CO₂ emissions [23]. Moreover, the worldwide urban energy demand is dominated by fossil fuels, and individual transport is the major factor of urban energy demand in the world.

The effect of urbanization on energy demand has been documented by precedent studies, which expressed that energy demand responds positively to the changes in urbanization level [24,25]. The urbanization process is associated with an increase in demand for housing and energy-intensive lifestyles. Another important consideration is that per capita household expenditure depends on the per capita household income that is higher in urban areas compared to rural regions [26]. This high-income level may ensure higher quality lifestyles in the end. The higher living standards in urban areas increase directly or indirectly energy consumption and, as a result, intensify global warming [27]. The demand for energy is increasing sharply due to economic growth, rapid urbanization, and industrial development [9]. It is also argued that the standard of living is greatly affected by the level of energy consumption in any country [28]. There are two sources of energy, namely non-renewable and renewable, which are used to meet the increasing demand for energy worldwide. Natural gas, oil, and coal are the main sources of non-renewable energy, which support approximately three-fourths of the world’s energy consumption [29]. It is worth mentioning that Thomas et al. [30] assessed that the world has 50 more years of oil and gas reserves and Kahia et al. [31] argued that the production of gas and oil will drop to roughly 40–60% by 2030 compared to the 1970s.

Energy security is a problem facing both energy importing and energy producing countries. This problem stems from the environmental consequences of producing and using fossil fuels, fluctuations in energy prices, and the geopolitical climate surrounding fossil fuel production in the world [31]. Energy security and environmental challenges can lead to an increase in energy efficiency and the search to find alternative sustainable energy sources to replace nonrenewable energies [29]. Renewable energy sources are considered the most effective sources that provide some solutions to the problems of energy security, sustainable development, and environmental degradation [32]. Renewable energy (wind, solar, hydrogen) is projected to be the fastest-growing world energy source that in 2021 contributed approximately 12.8% of the total electricity production in the world [33,34].

There are many strategies in the process of urbanization that can help to reduce energy consumption. The use of renewable energy can play a significant role in the process of urbanization. For example, the combining of energy-efficient building design with renewable energy is a sustainable way to use energy for increasing urbanization [35]. The stochastic and distributed optimal energy management approach is another interesting strategy that can be used for active distribution networks (ADNs) within office buildings [36]. Li et al. [36] defined this approach as scheduling office buildings fitted with heating ventilation and air conditioning (HVAC) systems, and electric vehicles (EV) charging piles to participate in the ADN optimization. The stochastic bi-level optimum allocation strategy of intelligent buildings (IBs), including energy storage sharing (ESS) services, is being developed in order to reduce energy storage costs for urbanization and to consider the uncertainties of electricity prices [37]. Extreme environments (strong winds, extreme cold, and strong ultraviolet radiation etc.) cause a significant loss in the inertial response capacity of power systems and increase the frequency instability of transient problems [38]. The polar power system is another method that must provide dependable and safe electricity in various extreme environments. Zhang et al. [38] mentioned that the doubly fed wind generator (DFWG) connected to a two-region interconnected polar microgrid can enhance DFWG frequency stability and ensure the safe and reliable operation of polar microgrids.

2.2. Evidence from the SAARC Zone

There are three sectors in the economy of any country: agriculture, manufacturing and services. Sen [39] explained that structural transformation is not only an important factor of economic growth but also a core condition of economic development. Structural transformation is the movement of workers from low productivity sectors such as agriculture to high productivity sectors such as manufacturing and services in any country, accompanying an increase in energy consumption. For example, Tang and Shahbaz [40] assessed the causal relationship between electricity consumption and real output at the aggregate and sectoral levels using annual data from 1972 to 2010 in Pakistan. The results revealed that there was a unidirectional causality running from electricity consumption to real output at the aggregate level in the country. However, at the sectoral levels, this relationship is also consistent for the manufacturing and services sectors, while not for the agricultural sector. Imran and Siddiqui [41] found that there was a positive relationship between energy consumption and economic growth in the long run for Bangladesh, India, and Pakistan, using data from 1971 to 2008. Azam and Khan [42] studied the relationship between urbanization and environmental degradation in Bangladesh, India, Pakistan, and Sri Lanka. The researchers found that there was a nearly 24% increase in the global population living in urban regions between 1950 and 2014, which will increase to 66% by 2050. More than 20% of the world’s total population resides in the SAARC zone, with a 34% urban population [43]. It is obvious that demand for energy from conventional energy sources has been increasing because of the increase in the urban population. South Asia is one of the regions in the world with the lowest ranked per capita energy consumption and has produced electricity with less than 50% of their available potential [44]. Due to the lack of investment to introduce renewable energy technologies, these countries continue to rely on nonrenewable sources. Zeb et al. [45] found that GDP growth and attempts to power poverty have a positive impact on energy production, while carbon dioxide emissions have a negative impact. Akhmat et al. [46] found that environmental indicators have shown significant long-term equilibrium with electric power consumption in this region.

Other studies explored the linkage between urbanization and environmental degradation in connection with various explanatory variables while considering different regions and countries and novel econometric techniques, but the results were mixed [47]. In the Gasimli et al. [27] study, carbon emissions were used as a proxy for environmental degradation that was directly related with economic growth. Abbasi et al. [47] explained that human activities are mainly (more than 95%) responsible for the rise in global temperature, which is due globally to the growth in both urbanization and globalization in the last two decades. The authors found a positive and significant impact of urbanization and energy consumption on CO₂ emissions, indicating that urban development and high energy consumption are barriers to improving environmental quality in the long run [48]. The efficiency of energy savings depends on the relationship between urbanization management and city growth in urban areas [49]. The findings of another research study showed that energy consumption increased by 52% between 1993 and 2003 in the SAARC countries, which raised the carbon emissions levels in the region [50]. The authors argued that countries in the SAARC zone could mitigate carbon emissions by reducing their energy consumption through technological improvements in the energy sector.

3. Methodology

3.1. Study Area

The South Asian Association for Regional Cooperation (SAARC) was founded on 8 December 1985 in Dhaka, Bangladesh to provide a platform for the peoples of South Asia to improve their quality of life through socioeconomic and cultural development. There are eight countries in the SAARC zone, including Afghanistan, Bangladesh, Bhutan, India, Maldives, Nepal, Pakistan, and Sri Lanka [51]. Recently, the SAARC countries have contemplated an environmentally friendly urbanization in their national development planning [52] which is consistent with the proposed major targets set in the 2015 UN Sustainable Development Goals (SDG) framework [3]. This study investigates the impacts of urbanization on energy consumption by comparing the relationship between urbanization and energy consumption in five nations (i.e., Bangladesh, India, Nepal, Pakistan, and Sri Lanka) from the SAARC zone (Appendix A), and using panel data for the period 1975–2014. These SAARC countries were selected because of the availability of data and the countries’ crucial importance to a developing region. This study distinguishes between “economic growth” understood as “a quantitative increase in the scale of the physical dimensions of the economy” and “development” understood as “the qualitative improvement in the structure, design and composition of the physical stocks of wealth that results from greater knowledge, both of technique and of purpose” [53]. While crude economic growth can lead to destruction of the environment, the process of development can lead to a sustainable stream of improvement that does not compromise the ability of future generations to meet their own needs given their preferences [54].

3.2. Model Specification

This study used a framework based upon the theory of energy–urbanization nexus employed in a multivariate context to examine the relationship between energy consumption and urbanization for selected countries in the SAARC zone. We defined a standard energy usage function in which final energy consumption ‘EC’ is specified as a function of urbanization ‘U’:

$${EC}_t=F\left[U_t\right]$$

(1)

where “F” is a linear homogenous function and ‘t’ is the time index. The process of urbanization mainly consists of two aspects: population urbanization (PU) and economic urbanization (EU) [55]. Therefore, Equation (1) takes the following form:

$${EC}_t=F\left[{EU}_t,{PU}_t\right]$$

(2)

Following Wang et al.’s [55] study, we used the following explanatory variables to measure the economic urbanization: per capita GDP (GDP), the industrial share of the GDP (SIG), the service share of the GDP (SSG), urban population (UP), and urban population growth rate (UPG). Therefore, the final energy consumption function, after this extension, can be expressed as:

$${EC}_t=f\left({{GDP}_t,{SIG}_t,{SSG}_t,UP}_t,{UPG}_t\right)$$

(3)

which can be re-written as Equation (4):

$${EC}_{it}=B_0{GDP}_{it}^{{\beta }_{1i}}{SIG}_{it}^{{\beta }_{2i}}{SSG}_{it}^{{\beta }_{3i}}{UP}_{it}^{{\beta }_{4i}}{UPG}_{it}^{{\beta }_{5i}}$$

(4)

All dependent and explanatory variables used in the model were converted to their logarithmic format to (i) reduce the impact of outliers, (ii) transform skewed data to approximate normality, (iii) linearize relationships between variables, (iv) stabilize variance in heteroskedastic data (heteroskedastic refers to a condition in which the variance of the error term varies widely), and (v) attain direct elasticities (Appendix B). Thus, Equation (5) presents the model that was used in the empirical analysis of this study:

$${lnEC}_{it}={\beta }_0+{\beta }_{1i}ln{GDP}_{it}+{\beta }_{2i}ln{SIG}_{it}+{\beta }_{3i}ln{SSG}_{it}+{\beta }_{4i}ln{UP}_{it}+{\beta }_{5i}ln{UPG}_{it}+{\mathrm{є}}_{it}$$

(5)

where ${\beta }_0$ and ${\mathrm{є}}_{it}$ represent ln($B_0$) and the error term, respectively, of the ith country at time t. Most importantly, ${\beta }_1, \dots \dots \dots \dots \dots \dots ,{\beta }_5$, represent the long-run elasticities of the dependent variable with respect to the explanatory variables.

3.3. Methods for Estimation

From an econometric point of view, the testing procedure consists of the following steps: (i) panel unit root tests, (ii) panel cointegration tests and its estimates, and (iii) Granger causality analysis, which are briefly explained. Interested readers can find out more about these tests in [47,56].

3.3.1. Panel Unit Root Test

In the first step of the estimation process, we examined the stationary properties of the data series to provide valid empirical evidence on long-run relationships among variables [57]. We used a panel unit root test to determine the stationarity of the data used in the model, to examine the effects of urbanization on energy consumption in the region of the study. One of the advantages of using a panel unit root test versus an individual unit root test is that the former has higher significance than the latter for maintaining persistence of individual time series regression errors across its cross section [58]. In this study, we employed three panel unit root tests, namely the Levin-Lin-Chu (LLC) test [59], the Im-Pesaran-Shin (IPS) test [60], and the MW test [61], to enhance the robustness of the results. Both the LLC and the IPS tests are based on the Augmented Dickey–Fuller principle, whereas the MW test is based on the Fisher test [62].

3.3.2. Panel Cointegration Test

After the panel unit root tests have confirmed that the panel data are non-stationary, the conditions of the panel cointegration test are tested. A cointegration test in time series examines if there is any long-run relationship between variables when they are non-stationary [47]. To do this, various testing procedures can be used, including the Maddala and Wu [61], Kao [63], and Pedroni [64,65] tests. We have used the last two methods in this study, as they have been used worldwide. Pedroni’s test suggests seven different statistic tests to examine if any cointegration relationship exists in heterogeneous panel data to check for bias in the explanatory variables used in the model. Kao [63] developed a residual-based test to examine if any cointegration relationship is available in heterogeneous panels. These two tests are very similar in both structure and hypothesis testing [56,58].

If the cointegration tests are satisfied and all the variables are cointegrated, the next step is to estimate the long-run coefficients of all the selected variables [47]. There are a number of methods in the literature that have been used for estimating the parameters of the model. Some examples of these methods are the ordinary least squares (OLS), fixed effects (FE), random effects (RE), generalized method of moments (GMM), feasible generalized least squares (FGLS), fully modified least squares (FMOLS), linear regression with panel-corrected standard errors (PCSE), linear regression with Newey–West standard errors (N-W), and linear regression with Driscoll–Kraay standard errors (DK). Among these methods, the OLS, fixed effects, random effects, and GMM methods do not always provide efficient estimators, yielding biased and inconsistent coefficients in the presence of serial correlations in the panel data, and instead, we have used the FMOLS method of Pedroni [65], which is very popular amongst the researchers [57]. The FMOLS method resolves the problems of serial correlation, endogeneity, simultaneity bias, and heterogeneous dynamics [47]. The panel FMOLS estimators (Equation (6) can be specified as follows:

$${\stackrel{̑}{\mathrm{\beta }}}_{FMOLS}={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\left({\sum }_{t=1}^T{\left(X_{it}-{\overline{X}}_i\right)}^2\right)}^{-1}\left({\sum }_{t=1}^T\left(X_{it}-{\overline{\mathrm{X}}}_i\right)Y_{it}^{\ast }-T{\stackrel{̑}{\mathrm{ɣ}}}_i\right)$$

(6)

where $Y_{it}^{\ast }=Y_{it}-{\overline{Y}}_i-\left({\displaystyle \raisebox{1ex}{${\stackrel{̑}{\mathrm{\Omega }}}_{2,1,i}$}\!\left/ \!\raisebox{-1ex}{${\stackrel{̑}{\mathrm{\Omega }}}_{2,2,i}$}\right.}\right)\Delta X_{it}$, ${\stackrel{̑}{\mathrm{ɣ}}}_i={\stackrel{̑}{\mathrm{\Gamma }}}_{2,1,i}+{{\stackrel{̑}{\mathrm{\Omega }}}_{2,1,i}}^0-\left({\displaystyle \raisebox{1ex}{${\stackrel{̑}{\mathrm{\Omega }}}_{2,1,i}$}\!\left/ \!\raisebox{-1ex}{${\stackrel{̑}{\mathrm{\Omega }}}_{2,2,i}$}\right.}\right)\left({\displaystyle \raisebox{1ex}{${\stackrel{̑}{\mathrm{\Gamma }}}_{2,1,i}$}\!\left/ \!\raisebox{-1ex}{${\stackrel{̑}{\mathrm{\Omega }}}_{2,2,i}$}\right.}\right)$ and ${\mathrm{\Omega }}_{it}$ is the long-run covariance matrix which can be further decomposed as; ${\mathrm{\Omega }}_i={\mathrm{\Omega }}_i^0+{\mathrm{\Gamma }}_i+{\mathrm{Ѓ}}_i$.

The associated t-statistics are specified as Equation (7):

$$t_{{\stackrel{̑}{\mathrm{\beta }}}_{FMOLS}}={\displaystyle \frac{1}{\sqrt{N}}}{\sum }_{i=1}^N{t}_{{\stackrel{̑}{\mathrm{\beta }}}_{FMOLS}},i; \mathrm{where} t_{{\stackrel{̑}{\mathrm{\beta }}}_{FMOLS}},i=\left( {\stackrel{̑}{\mathrm{\beta }}}_i-{\mathrm{\beta }}_0\right){\left[{{\stackrel{̑}{\mathrm{\Omega }}}_{1,1,i}}^{-1}{\left(Y_{it}-\overline{\mathrm{Y}}\right)}^2\right]}^{{\displaystyle \raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}}$$

(7)

3.3.3. Panel Granger Causality Test

The cointegrating relationship is confirmed among the variables. That indicates not only the existence of a long-run relationship but also the presence of a causal relationship between these variables, at least in one direction; however, it does not give information on the direction of the causal relationship. If cointegration exists, then we employ the Granger causality statistical test using the panel vector error correction model (VECM) to investigate the direction of causality among the variables of the model. The VECM Granger causality test can capture the short-run causality based on the F-statistic, and the long-run causality based on the lagged error correction term [47]. This method essentially integrates the lagged of the residual from the specified long-run regression model as a right-hand side variable. Thus, to test the causal relationship between the variables of the model, a panel-based error correction model (VECM) is defined as Equation (8):

$$\begin{gathered} \left(\begin{array}{c}\Delta ln{EC}_{it}\\ \Delta ln{GDP}_{it}\\ \Delta ln{SIG}_{it}\\ \Delta ln{SSG}_{it}\\ \Delta ln{UP}_{it}\\ \Delta ln{UPG}_{it}\end{array}\right)=\left(\begin{array}{c}{\alpha }_1\\ {\alpha }_2\\ {\alpha }_3\\ {\alpha }_4\\ {\alpha }_5\\ {\alpha }_6\end{array}\right)+\sum _{k=1}^P\left(\begin{array}{c}{\mathrm{\beta }}_{11\mathrm{k}}{\mathrm{\beta }}_{12\mathrm{k}}{\mathrm{\beta }}_{13\mathrm{k}}{\mathrm{\beta }}_{14\mathrm{k}}{\mathrm{\beta }}_{15\mathrm{k}}{\mathrm{\beta }}_{16\mathrm{k}}\\ {\mathrm{\beta }}_{21\mathrm{k}}{\mathrm{\beta }}_{22\mathrm{k}}{\mathrm{\beta }}_{23\mathrm{k}}{\mathrm{\beta }}_{24\mathrm{k}}{\mathrm{\beta }}_{25\mathrm{k}}{\mathrm{\beta }}_{26\mathrm{k}}\\ {\mathrm{\beta }}_{31\mathrm{k}}{\mathrm{\beta }}_{32\mathrm{k}}{\mathrm{\beta }}_{33\mathrm{k}}{\mathrm{\beta }}_{34\mathrm{k}}{\mathrm{\beta }}_{35\mathrm{k}}{\mathrm{\beta }}_{36\mathrm{k}}\\ {\mathrm{\beta }}_{41\mathrm{k}}{\mathrm{\beta }}_{42\mathrm{k}}{\mathrm{\beta }}_{43\mathrm{k}}{\mathrm{\beta }}_{44\mathrm{k}}{\mathrm{\beta }}_{45\mathrm{k}}{\mathrm{\beta }}_{46\mathrm{k}}\\ {\mathrm{\beta }}_{51\mathrm{k}}{\mathrm{\beta }}_{52\mathrm{k}}{\mathrm{\beta }}_{53\mathrm{k}}{\mathrm{\beta }}_{54\mathrm{k}}{\mathrm{\beta }}_{55\mathrm{k}}{\mathrm{\beta }}_{56\mathrm{k}}\\ {\mathrm{\beta }}_{61\mathrm{k}}{\mathrm{\beta }}_{62\mathrm{k}}{\mathrm{\beta }}_{63\mathrm{k}}{\mathrm{\beta }}_{64\mathrm{k}}{\mathrm{\beta }}_{65\mathrm{k}}{\mathrm{\beta }}_{66\mathrm{k}}\end{array}\right)\left(\begin{array}{c}\Delta ln{EC}_{it-k}\\ \Delta ln{GDP}_{it-k}\\ \Delta ln{SIG}_{it-k}\\ \Delta ln{SSG}_{it-k}\\ \Delta ln{UP}_{it-k}\\ \Delta ln{UPG}_{it-k}\end{array}\right) \\ +\left(\begin{array}{c}{\delta }_1\\ {\delta }_2\\ {\delta }_3\\ {\delta }_4\\ {\delta }_5\\ {\delta }_6\end{array}\right){ECM}_{it-1}+\left(\begin{array}{c}{\epsilon }_{1it}\\ {\epsilon }_{2it}\\ {\epsilon }_{3it}\\ {\epsilon }_{4it}\\ {\epsilon }_{5it}\\ {\epsilon }_{6it}\end{array}\right) \end{gathered}$$

(8)

where i = 1, 2, … … …, n; t = P + 1, P + 2, P + 3, … … …, T; ∆ and ECM symbolize the first difference of the variable and the error-correction term, respectively; K denotes the optimal lag length, which is determined by the Schwarz Information Criterion (SIC); α’s and β’s are parameters of the model, and ε’s are adjustment coefficients.

4. Empirical Analysis

4.1. Panel Unit Root Tests Results

Table 1. Results of the LLC, IPS, and MW Panel Unit Root Tests.

Variables	LLC Test	Prob.	IPS Test	Prob.	MW Test	Prob.
Case 1a: Model with constant and trend terms [level form]
LNEC	−2.214 **	0.013	−2.194 **	0.014	17.970 ***	0.055
LNGDP	1.523	0.936	3.777	0.999	1.485	0.999
LNSIG	−1.769	0.038	−2.598 *	0.004	26.611 *	0.003
LNSSG	−3.066 *	0.001	−2.601 *	0.004	37.151 *	0.000
LNUP	−1.201	0.114	−2.949 *	0.001	63.350 *	0.000
LNUPG	−1.250	0.105	−1.942 **	0.026	21.080	0.020
Case 1b: Model with only constant term [level form]
LNEC	−2.162 **	0.015	0.629	0.735	12.389	0.259
LNGDP	6.734	1.000	9.317	1.00	2.731	0.987
LNSIG	−2.615 *	0.004	−2.714 *	0.003	36.530 *	0.000
LNSSG	−0.484	0.313	1.127	0.870	19.091	0.039
LNUP	0.275	0.608	1.400	0.919	60.699 *	0.000
LNUPG	−0.561	0.287	0.663	0.746	9.438	0.491
Case 2: Model with only constant term [first difference]
∆LNEC	−4.625 *	0.000	−7.111 *	0.000	126.312 *	0.000
∆LNGDP	−3.982 *	0.000	−4.670 *	0.000	89.114 *	0.000
∆LNSIG	−8.936 *	0.000	−10.973 *	0.000	134.396 *	0.000
∆LNSSG	−7.235 *	0.000	−10.257 *	0.000	144.238 *	0.000
∆LNUP	−3.314 *	0.000	−2.182 **	0.014	16.708 ***	0.081
∆LNUPG	−5.859 *	0.000	−7.316 *	0.000	77.509 *	0.000

Note: *, ** and *** indicate rejection of the null hypothesis of no unit root at 1%, 5%, and 10% level of significance, respectively. Source: Sample data.

shows the estimation results from conducting the panel unit root tests. As mentioned earlier, to avoid any spurious results and to investigate the possibility of panel cointegration, a panel unit root test is conducted considering all the regression variables to detect the existence of unit roots [66]. The three tests have the null hypothesis that all the panels contain a unit root. A cointegration test is applied to determine the long-term equilibrium relationship if the variables are stationary at the first difference. Table 1 shows that not all the variables are stationary with and without time trend specifications (Case 1) at level form using the LLC, IPS, and MW tests. Table 1 also shows Case 2, which presents the results of the tests at the first difference for the LLC, IPS, and MW tests with the intercept. Based on the tabulated results, the LLC, IPS and MW test statistics imply that energy consumption and urbanization variables reveal almost similar results at the first difference without time trends, indicating they are stationary at the first difference. This implies the null hypothesis of unit roots (i.e., non-stationary) is rejected with 99% confidence for all the explanatory variables except the UP independent variable, which is significant with 95% confidence for the IPS test and with 90% confidence for the MW test. Therefore, the explanatory variables are characterized as integrated of order one, I(1). The panel unit root tests results support that all the panel variables are stationary of order one using the LLC, IPS, and MW tests. If the I(1) variables are cointegrated with one another, they could be useful in further econometric analysis. Therefore, the panel cointegration method is applied to test the existence of cointegration relationships between energy consumption and the other variables.

4.2. Panel Cointegration Tests Results

Table 2. Results of Pedroni and Kao Panel Cointegration Tests.

Pedroni Residual Cointegration Test
	Statistics	p-Values
Panel v-stat	0.426	0.33
Panel rho-stat	0.764	0.00
Panel PP-stat	−0.735	0.00
Panel ADF-stat	0.033	0.005
Group rho-stat	1.747	0.00
Group PP-stat	−0.307	0.00
Group ADF-stat	0.923	0.82
Kao Residual Cointegration Test
ADF	−2.430	0.00

Source: Sample data.

presents the results of Pedroni [64,65] and Kao [63] panel cointegration tests. The results stem from testing the hypothesis whether any long-run relationship between the dependent and the independent variables exists. As it is shown in Table 2, most of the statistics, such as panel rho-stat, panel PP-stat, group rho-stat, and group PP-stat, are found statistically significant with 99% confidence for panels.

It is shown from the Pedroni cointegration test that there is a long-run stable relationship among variables in the panel data sets based on the p values, which is also verified by the Kao test. These findings confirm that there is a long-run relationship between the I(1) variables, which indicates that the intensity of the long-run relationship should be estimated properly.

4.3. Estimation Results—FMOLS Method

The Pedroni panel co-integration and the Kao estimation techniques confirm that the cointegration exists among the selected variables in the study; the next step is to examine the long-run relationship between variables. We estimated the parameters of Equation (5) using the fully modified ordinary least squares (FMOLS) method. Since all the data are converted into natural logarithmic form, the parameters of the equation express the long-run elasticities of the per capita energy consumption with respect to the five independent variables used in the model.

Table 3. Panel FMOLS Results.

Independent Variables	Coefficient	t-Statistic	Prob.
LNGDP	1.083	11.806	0.000
LNSIG	0.586	3.039	0.002
LNSSG	−0.566	−1.834	0.006
LNUP	1.359	7.405	0.000
LNUPG	0.296	4.450	0.096
R2 Adj. R2 D-W test			0.987 0.986 1.54

Notes: Panel method = Grouped estimation; Country dummy = Yes; Period dummy = Yes. Source: Sample data.

shows the panel FMOLS estimators specified in Equation (5). Table 3 shows that all the coefficients were statistically significant with 99% confidence, except the coefficient for urban population growth; that means that changes in the independent variables correlate with shifts in the dependent variable. Table 3 shows that there is a direct relationship between energy consumption and GDP, which explains that a 1% increase in GDP will increase energy consumption by approximately 1.083%. This result confirms our hypothesis that an increase in GDP requires more energy to produce goods and services, and the increased use of transportation shows a positive impact on energy consumption. These findings were statistically significant and consistent with precedent studies by Azam and Khan [42] and Anser et al. [26]. As for the industrial sector’s share of GDP, we found a direct relationship with energy consumption that was statistically significant. An increase in the industrial sector’s share of GDP of 1% consumes an additional per capita energy of 0.586% in the regions of the study (Table 3). Energy consumption increases because the portion of traditional technology-based industry is higher than the modern one in the industrial share of GDP in this region and, as a result, the energy use in the tertiary industry (traditional technology-based industry) is high. This finding was consistent with precedent studies by Ahmad and Majeed [57] and Wang et al. [55]. We also found an inverse relationship between energy consumption and the service sector’s share of GDP, implying that an increase in the service sector’s share of GDP of 1% would reduce per capita energy consumption by 0.56%. Similar to the findings of Ahmad and Majeed [57], this study concluded that urban infrastructure development, energy efficiency, and automobile technology efficiency, which are the main components of the service sector in GDP, had less energy consumption.

As for the urban population size, a significant positive relationship with energy consumption was found for the regions of the study, with a long-run elasticity of 1.359 (Table 3). This finding also conformed with the result of Wang et al.’s [67] study. In summary, we can conclude that energy consumption had a long-run direct relationship with GDP, the industrial sector’s share of GDP, and the size of urban population, and a long-run inverse relationship with the service sector’s share of GDP in the member states of the SAARC zone. This is due to the fact that most of the member states in the SAARC zone are developing economies, where energy consumption is higher due to the higher growth rate of the industrial sector and of the urban population.

4.4. Vector Error Correction Model (VECM) Results

The estimation of the long-run relationship between dependent and independent variables does not provide information about the causal relationships between them. This justifies the use of the panel vector error correction model (VECM) to examine the causality direction between variables.

Table 4. Panel Causality Test Results for the SAARC.

Dependent Variables	Source of Causation (Independent Variables) Short-Run						Long-Run
	∆LNEC	∆LNGDP	∆LNSIG	∆LNSSG	∆LNUP	∆LNUPG	ECT
∆LNEC	---	0.49 * (2.81)	0.13 (1.96)	0.16 ** (2.13)	2.79 ** (2.39)	0.016 (0.98)	−0.02 * (−3.06)
∆LNGDP	0.042 ** (1.09)	---	0.028 * (0.89)	0.026 ** (0.68)	0.548 (0.98)	0.013 (0.16)	−0.016 * (−4.47)
∆LNSIG	0.083 * (0.98)	0.317 ** (0.72)	---	0.076 (0.92)	0.28 ** (0.23)	0.017 (1.01)	−0.005 ** (−0.06)
∆LNSSG	−0.161 (2.22)	0.472 (3.00)	0.006 (0.11)	---	−0.158 (−0.511)	0.016 (1.13)	0.009 (1.43)
∆LNUP	0.004 (0.93)	0.023 (2.33)	0.006 ** (1.73)	0.027 * (5.98)	---	0.010 ** (0.11)	−8.98 (−0.198)
∆LNUPG	0.358 (0.84)	1.71 (1.85)	0.007 (0.02)	0.40 (0.98)	22.40 (0.66)	----	0.025 (0.630)

The p-values are presented in parentheses while t-statistics are in brackets. ECT = the estimated coefficient on the error correction term; * and ** denote statistical significance at 1% and 5% level, respectively. Source: Sample data.

shows the results of the panel causality test in both the short-run and long-run for the region of the study. From Table 4, we can detect three types of evidence of casual relationships in the long-run. The first relationship was from GDP, the industrial sector’s share of GDP, the service sector’s share of GDP, urban population size, and urban population growth rate to energy consumption. The coefficient of error correction term (ECT) was found to be equal to −0.02, which was statistically significant with 99% confidence. The second relationship was from energy consumption, the industrial sector’s share of GDP, the service sector’s share of GDP, urban population size, and urban population growth rate to GDP. The coefficient of error correction term (ECT) was found to be equal to −0.016, which was statistically significant at the 0.01 level. Finally, the third relationship was from energy consumption, GDP, the service sector’s share of GDP, urban population size, and urban population growth rate to the industrial sector’s share of GDP. As Table 4 shows, the coefficient of error correction term (ECT) was found to be equal to −0.005, which was statistically significant with 95% confidence.

From Figure 2, we can detect four types of evidence of unidirectional casual relationships in the short-run. These relationships are (i) from GDP, the service sector’s share of GDP, and urban population size, to energy consumption, (ii) from energy consumption, the industrial sector’s share of GDP, and the service sector’s share of GDP to gross domestic product, (iii) from energy consumption, GDP, urban population size to the industrial sector’s share of GDP, and (iv) from the industrial sector’s share of GDP, the service sector’s share of GDP, and urban population growth rate to urban population size. Figure 2 indicates that there are some short-run bidirectional causal relationships between GDP and energy consumption; GDP and the industrial sector’s share of GDP; and the industrial sector’s share of GDP and population size. The findings of this study explain how urban population’s size, GDP, and the industrial sector’s share of GDP (which are the main factors of urbanization) are influencing energy consumption. The findings also suggest that there is a positive relationship between the selected urbanization factors and energy consumption, that provides proof of the research hypothesis of the study.

5. Policy Recommendations

We investigated the impacts of urbanization on energy consumption in the SAARC zone. The parameters of the econometric model were estimated using the fully modified ordinary least squares (FMOLS) method, and the vector error correction model (VECM) method was employed to examine the direction of causality between dependent and independent variables. The findings show that all the variables are statistically significant at the 1% level of significance except urban population growth, although the effects vary among variables. The findings from the VECM, according to the value of error correction term (ECT), show that there are three casual relationships in the long-run and four unidirectional causal relationships in the short-run. Therefore, we can conclude that more (less) urbanization leads to more (less) energy consumption in the region of the study. In the following, we provide some policy recommendations for reducing energy consumption or for using energy more efficiently in the SAARC zone, a region challenged by rapid urbanization.

First, the level of GDP positively affects energy consumption. Most of the member states in the SAARC zone are developing economies, where energy consumption is higher, due to the higher growth rate of the levels of the economies. Higher GDP growth rates increase the demand for both unskilled and skilled labor force, physical capital, and raw materials, leading to an increase in energy consumption. In addition, most of the countries’ economies in the SAARC zone are experiencing structural transformation, as they transition from agriculture to industrial development in the economy. These countries’ governments should take new initiatives to support investment in renewable energy, by transferring resources towards a sustainable development of their countries.

Secondly, an increase in the size of the industrial sector’s share of the GDP is likely to increase energy consumption. As the industrial sector’s share of the GDP expands, production increases, and that leads to an increase in energy consumption. In addition, the SAARC countries are developing economies, which export different types of manufactured products to developed countries, due to the comparative advantage of producing these products at a relatively cheaper rate. This is another reason for the increase in the industrial sector’s share of the GDP, as well as the increase in energy consumption. The governments of these countries should modify their industrial policies by providing incentives to these industries to adopt efficient energy sources as well as green technologies.

Thirdly, this study has shown that an increase in urban population results in increased demand for economic output, which leads to an increase in energy consumption. This finding likely indicates that a rise in the existing economic growth in urban areas leads to an increase in urban population, which eventually leads to higher energy consumption. The governments of the countries in the SAARC zone should take immediate policy responses to the population growth by promoting sustainable development measures. For example, they should take initiatives to educate people to realize the consequences of a fast increase in the population rate. The governments can also develop policies encouraging the reduction of energy use at the household level, for instance, by rising gasoline prices and encouraging public transportation use.

Fourthly, another important policy that the governments in the SAARC zone could initiate is to support education concerning the environmental consequences of energy overuse, so that people are aware of energy overuse and its consequences for the environment, and would change their lifestyles.

6. Conclusions

The study has sought to provide a comprehensive and systematic analysis of urbanization’s impacts on energy consumption for selected member states of the SAARC zone. The findings of this study suggested important policy implications that could provide a useful framework for policymakers. We recommend further studies to cover other aspects of this research topic. For instance, land use and social urbanization variables could be contemplated as other factors affecting energy use, and inclusion of nonrenewable and renewable energies in the energy mix for the sustainable development of the countries in the SAARC region.