Electricity consumption and population growth in South Africa: A panel approach

: This study investigates the relationship between population growth and electricity consumption in South Africa for the period from 2002 to 2021 collected from StatsSA. The study utilises Seemingly Unrelated Regression model and Dumitrescu and Hurlin (2012) causality tests to analyse the relationship between the variables. Empirical results revealed that there is a negative statistically significant relationship between population growth and electricity consumption in South Africa. The results further reveal one-way causality running from population growth to electricity consumption. The study recommends that the government and policy makers must implement policies aimed at increasing renewable electricity generation to match the gap between electricity demand and growing population thereby reducing constant loadshedding in South Africa.


Introduction
Electricity is the backbone of an economy to grow in most societies today. South Africa has been marred with continuous loadshedding that has resulted from poor electricity generation that fails to meet the overgrowing electricity demand. The impact of loadshedding has resulted in some of the households having to find alternative sources of energy such as wood fuel in the rural areas, gas, and solar systems.
This has also resulted in electricity consumption declining as people switch to alternative sources of energy. Lenoke (2017), Khobai and Le Roux (2017), , Hlongwane and Daw (2021), Stungwa, Hlongwane et al. (2022) has conducted a study on electricity consumption but their studies was much focusing on the relationship between electricity consumption and economic growth.
Electricity consumption is predicted to rise over next few decades. Increasing demand for electricity poses a problem to South Africa's administration because it relates to a significant reliance on expensive fuel imports. The study of primary factors of electricity consumption in South Africa will contribute to a better understanding and description of the nature of aggregate electricity consumption, as well as an endeavour to build a solid electricity policy.
Overview of the study: There are insufficient studies focusing on the relationship between population growth and electricity consumption in South Africa and the significance of this study is to investigate on that nexus. Electricity infrastructure is divided into three sub-sectors: generation, transmission, and distribution. In terms of generation, Eskom is the market leader in the production of power. Eskom generates, transmits, and distributes electricity in South Africa to industrial, mining, commercial, agricultural, mining, commercial, agricultural, and residential clients, as well as municipalities, which redistribute electricity to companies and homes within their jurisdictions (Ratshomo and Nembahe 2019). In addition, the utility acquires power from Independent Power Producers (IPPs) under different agreement schemes, as well as electricity generating facilities located outside of the country's boundaries. Electricity is the backbone of the South African economy, and it is a critical industry that generates employment and value by extracting, processing, and delivering energy commodities and services across the country. South Africa's consistent economic expansion, along with an increased emphasis on industrialisation and a mass electrification initiative to provide power to deep rural regions, has resulted in a sharp increase in electricity consumption in recent years. Ratshomo and Nembahe (2019), Modise and Mahotas (2020) and Gabrielle (2020) summarizes that residential was accountable for 8%, commerce and public services 14%, agriculture 6%, transport sector 19%, industry sector 52% and 1% was not specified in terms of electricity consumption in South Africa in 2016. Residential sector accounted for 72% of electricity consumption in South Africa according Ratshomo and Nembahe (2019). Population has been increasing but electricity consumption has not been increasing enough to match with the population. Figure 1 below shows the sum of population growth and electricity consumption in nine provinces in South Africa.

Figure 1: Sum of population and electricity consumption by province from 2002 to 2021
Source: Author's own compilation using data from StatsSA to 2021. According to Figure 1, Gauteng has the highest electricity consumption followed by KwaZulu-Natal, Mpumalanga, and Western Cape province in terms of total gigawatt hours consumed. This is because Gauteng is the main tertiary and industrial hub of South Africa and holds a lot of population as compared to other provinces. In terms of the sum of population, Gauteng has the highest number, followed by KwaZulu-Natal, Eastern Cape, Western Cape, and Limpopo province. In South Africa we have a problem of high electricity consumption and high population growth, hence the significance of this study is to investigate the relationship between population growth and electricity consumption in South Africa. This will enable the policy makers, government, and Eskom to know how much electricity they should generate to meet the increasing consumption from the growing population.

Literature
Theoretical literature: The theoretical that underpins the investigation is presented in this section of the study. This study focuses on the growth theories in greater depth. Malthus (1798) believed that population increase will outpace the earth's ability to produce food, resulting in humanity's destitution. Solow (1956) underlined in a research that contributed to economic growth theory that a country with faster population growth rates will have lower levels of capital and income per worker in the long run.
According to Kremer (1993), population expansion leads to economic growth. More people in the country equals more geniuses, scientists, and engineers, which lead to quicker technical advancement (Stungwa and Daw 2021). This study, however, tries modifying the growth theories by making electricity consumption depend on population growth and economic growth in South Africa to investigate the nature of relationship that exists among the variables.
A review of developing countries: Mohanty and Chaturvedi (2015) examine the weather electricity energy consumption on economic growth in Indian using the annual data spanning 1970-1971 to 2011-2012. Using granger causality test and Engle-Granger technique, their study suggested that electricity energy consumption has a positive relationship on economic growth in the short run and long run.  Athukorala and Wilson (2010) investigate it the short run dynamics and long run equilibrium relationship between the residential electricity demand and the factors influencing demand, like capita income, price of electricity, price of kerosene oil and price of liquefied petroleum gas using an onion data for Sri Lanka for the period of 1960 to 2007. The main findings of the paper were that increasing the price of electricity is not the most effective tool to reduce electricity consumption. Niu, Jia et al. (2013)  Granger causality results revealed unidirectional causal relationship from economic growth to electricity consumption that indicates economic growth stimulates electricity consumption in the long run. The researchers recommend that it is essential for Pakistan policymakers to plan and increase infrastructure development to meet increasing electricity demand. The researchers also recommend that government should adopt policies to sustain electricity supply. Hussain, Rahman et al. (2016) forecasted electricity consumption in Pakistan. The study borrowed available annual time series data spanning for the period from 1980 to 2011. The study employed Holt-Winter and Autoregressive Integrated Moving Average (ARIMA) models to forecast electricity consumption in Pakistan. The empirical results revealed that electricity demand is higher in the household sector than in other sectors and that electricity generation would be lesser than the increase in electricity generation. The researchers recommend that policymakers should focus on short-and long-term projects such as renewable sources of electricity to balance the supply-demand gap in Pakistan. A review of developed countries: Kahouli (2018) (2018) Whereby, LELC represents the logged electricity consumption, LPOP is the logarithm of population growth, LGDP is the logarithm of gross domestic product, 0 is the error term and constant.

Data Sources:
The study utilises the annual time series data spanning for the period from 2002 to 2021 for electricity consumption, population and gross domestic product collected from Quantec and Statistics South Africa (StatsSA).

Data Analysis:
The study employs a basic linear Seemingly Unrelated Regression (SUR) model developed by Moon and Perron (2006) to analyse the relationship between population growth and electricity consumption in South Africa. Suppose that is a dependent variable, = (1, ,1 , ,2 , … . , , −1 ) ′ is a vector of explanatory variables observational unit , and is unobservable error term, where the double index denotes the ℎ observation of the ℎ equation in the system, denotes the time dimension. The classical linear SUR model is a system of linear regression equations as given below: Whereby, = 1, … . , , and = 1, … , . Denote = 1 + ⋯ + . This study however modifies the simple SUR model to a multivariate regression with parameter restrictions since this study employs more variables as proposed by Moon and Perron (2006). In this modification, = [ ′ 1 , ′ 2 , … , ′ ] ′ and ( ) = ( 1 , … , ) to be (L×N) block diagonal matrix. The multivariate SUR model can therefore be rewritten as given below: Where the coefficient ( ) satisfies: For some (NL×L) full rank matric . In a special case where 1 = ⋯ = = , we have = ( 1 , … , ) ⊗ where denotes the ′th column of the × identity matrix . In the errors of are assumed to be overtime with mean zero and homoscedastic variance ∑ = ( , ′ | ). We assume that ∑ positive definite and denote by the ( , ) ℎ element of ∑, that is = ( , ′ | ).
The Dumitrescu-Hurlin Causality Test: Dumitrescu and Hurlin (2012) provide an extension of Granger (1969) causality test designed to detect causality in panel data. The underlying regression is: Whereby, , and , are the observations of two stationary variables for individual period .
Coefficients are allowed to differ across individuals but are assumed to be time invariant. The lag order of is assumed to be identical for all individuals, and the panel must be balanced. As given in Granger (1969), the procedure to determine the existence of causality is to test for significant effects of past values of on the present value of . The null hypothesis is therefore defined as: Which correspond to the absence of causality for all individuals to the panel. The DH test assumes there can be causality for some individuals but not necessarily for all. Thus, the alternate hypothesis is: Where 1 ∈ (0, − 1) is unknown. If 1 = 0, there is causality for all individuals in the panel. 1 must be strictly smaller than N, otherwise, there is no causality for all individuals, and 1 reduces to 0 . In opposition of the above notion, Dumitrescu and Hurlin (2012) proposes the following procedure: run the N individual regressions implicitly encloses in equation 5, perform F tests of the K linear hypothesis 1 = ⋯ = = 0 to retrieve the individual Wald statistic , and finally compute the average Wald statistic ̅ : 1 : Lopez and Weber (2017) emphasizes that the test is designed to detect causality at panel level and rejecting 0 does not exclude noncausality for some individuals. Following Monte Carlo simulations, Dumitrescu and Hurlin (2012) show that ̅ is asymptotically well behaved and can genuinely be used to investigate panel causality. Under the assumption that the Wald statistics are independent and identically distributed across individuals, it can be shown that the standardized statistic ̅ when T→ ∞ first and then N → ∞ (sometimes interpreted as T should be large relative to N) follows a standard normal distribution: The testing procedure of the null hypothesis finally based on ̅ and ̃. If these are larger than the standardized critical values, then Lopez and Weber (2017) highlight that the null hypothesis (Ho) must be rejected and conclude that Granger causality exists. For large N and T panel datasets, ̅ can be reasonably considered and for large N but relatively small T datasets, ̃ should be favoured. The study therefore continues to provide the results and interpretations as shown in Section 4 below. The study performed a panel unit root test as shown in Table 1 above by employing the Levin, Lin, and Chu test, Im, Pesaran, and Shin test and ADF-Fisher test. The results shows that LELC and LGDP are integrated of I(1) while LPOP is integrated of I(0). The study therefore continues to perform the panel cointegration test as shown in Table 2 below to determine long run relationships among the variables.  Table 2 above shows the cointegration results of Pedroni (1999) and Kao (1999). The results of Pedroni (1999) are separated into two sections: within the dimension and between dimensions. The null hypothesis of Pedroni (1999) stress that there is no cointegration between the variables. Within the dimension, Panel rho-statistic, PP-statistic and ADF-statistic are all significant at 1% level of significance. As a result, the null hypothesis cannot be accepted and the conclusion the conclusion that cointegration exists is reached. Because the four tests were a tie, the study cannot establish that there is cointegration without testing the between dimensions. The results of the between dimensions confirms presence of cointegration since the group rho-Statistic (0.0783) is statistically significant. The group PP-statistic and ADF-statistic are also significant at 1% level of significance. The results of Kao (1999) also confirms that there is cointegration within the variables in the model since the ADF probability value is significant at 1% implying the rejection of the null hypothesis of no cointegration. The study therefore continues to perform the Durbin-Watson test to detect autocorrelation as shown in Table 3 below. The study begins by using the DW statistics to check for autocorrelation of the residuals sequence of fixed effects model (FEM), random effects model (REM) and Cross-section Seemingly Unrelated Regression (SUR). The results are presented in Table 3 above and the DW statistics for FEM, REM and SUR are greater than 2, implying that there is no presence of autocorrelation in the residuals when FEM, REM and SUR are used to investigate the relationship between total population, economic growth, and electricity consumption in the nine South Africa provinces. After some diagnostic tests, the results of the FEM and REM models cannot be accepted due to presence of heteroskedasticity and nonnormal residuals, therefore, the study will employ the Cross-section Seemingly Unrelated Regression to analyse the relationship between electricity consumption, total population, and economic growth in South Africa. The study performed the SUR model to show long run relationship among the variables in the model as given in Table 4 above when LELC is a dependent variable explained by LPOP and LGDP. The results shows that there is a negative statistically significant relationship between population growth and electricity consumption in South Africa. A 1% increase in population growth in South Africa, will significantly result in electricity consumption declining by 0.09%, ceteris paribus. These results are inconsistent with the studies conducted by Al-Bajjali and Shamayleh (2018), Zaman, Khan et al. (2012) and Huang (2015) that found a positive relationship between population growth and electricity consumption. This means that an increase in population growth in South Africa has a detrimental effect on electricity consumption. This may be a result of people in South Africa using alternative sources of energy. The recent load shedding affecting South Africa since 2009 has resulted in the people switching to alternative sources of energy such as paraffin, firewood, gas, solar and biogas for their daily activities.

Results and Interpretation
Therefore, policies that results in increase in electricity consumption might have a detrimental effect on the environment and people resulting call to switch to greener energy as Eskom is heavily reliant on non-renewable sources of energy.
The results further shows that there is a positive statistically significant relationship between economic growth and electricity consumption in South Africa. A 1% increase in economic growth in South Africa, will significantly result in electricity consumption increasing by 0.01, ceteris paribus. These results are consistent with the studies conducted by Huang (2015), Al-Bajjali and Shamayleh (2018) and Zaman, Khan et al. (2012) that found that economic growth has a positive relationship with electricity consumption. This entails that increase in economic growth result in an increase in electricity consumption since firms will be expanding the scale of their activities and industries requiring more electricity to cater for increasing demand in electricity. This calls for policy makers and the government to implement policies that results in an increase in renewable electricity generation to match with the growing electricity demand from economic growth. The Durbin-Watson statistic is greater than the Rsquared which means the regression is not spurious. The R-squared is 0.45%, meaning 45% of the variation in electricity consumption is explained by population growth and economic growth in South Africa. The Adjusted R-squared is 0.41, meaning that 41% is adjusted for the degrees of freedom. The study therefore continues to perform the Cross-section dependence as shown in Table 5 below. The study performed cross-sectional dependence test as shown in Table 5 above. When there is presence of cross-sectional dependence across the panels, this impact on the efficiency of estimators and leads to biased results. The null hypothesis is that there is no cross-sectional dependence among residuals of the variables in the model. The results of Breusch and Pagan (1980) are reliable since it is good and powerful when time period (T) is greater than the cross-sectional dimension (N), while the Pesaran (2015) is perfect when either N is big or small (Stungwa and Daw 2021). The probability values of the Breusch-Pagan LM (1.0000) and Pesaran CD (0.9948) are greater than 5% implying that we fail to reject the null hypothesis cross-sectional dependence and concluding that there is cross-sectional independence between the cross-sectional units. This entails that the South Africa provinces are independent of each other when it comes to the relationship between electricity consumption, population growth and economic growth. The study continues to perform the residual normality test as shown in Figure 2 below.

Source: Author's own compilation
The study performed the residual normality test as shown in Figure 2 above. The value of the Jarque-Berra is 1.272486 and its corresponding probability value is 0.529277 meaning that we fail to reject the null hypothesis that the residuals are normally distributed. This means that the results from the model are reliable since the residuals are normally distributed that is consistent with the prior expectations of a normal OLS regression model. The study continues to perform the Dumtrescu Hurlin causality test as shown in Table 6 below to check for causal relationships among the variables in the model.   show bidirectional causality between economic growth and population growth in South Africa since the probability values are 0.0000 and 0.0244 which are statistically significant at 1% and 5% level of significance, respectively. This means that policies that affect economic growth and population growth will have causal effect on each other. The results reveal absence of causality between economic growth and electricity consumption since the probability values (0.3760 and 0.9119) are insignificant at 1%, 5% and 10% level of significance. This means that policies that affect economic growth will not have causal effect on electricity consumption and vice versa. The study therefore continues to give the conclusion and recommendations of the study as shown in Section 5 below.

Conclusion and Recommendations
The study examined the relationship between population growth and electricity consumption using economic growth as an intermittent variable and discovered that population growth and economic growth are significantly related to electricity consumption in South African provinces in both negative and positive ways, respectively. The study employed a Seemingly Unrelated Regression Model using panel data spanning the years 2002 to 2021. The panel unit root test was used in the study to establish the order of cointegration and to assist prevent the problem of spurious regressions. The study used the cross-section dependence diagnostic test and discovered that the provinces are independent of one another, avoiding misleading findings and inefficient parameters. The residual normality test findings indicated that the residuals are normally distributed, which is compatible with the expectations of a normal OLS model. The policy recommendations of this study are therefore as follows: Firstly, the negative statistically significant relationship between population growth and electricity consumption calls for the policy makers, the government and Eskom to speed up policies that increase renewable electricity consumption. This will help reduce reliance on non-renewable electricity consumption and leading to households altering to environmentally friendly sources of energy such as wind and solar. The government must increase investment in the wind farms in the Eastern Cape province and solar in the Northern Cape and Limpopo provinces to take advantage of the abundant wind and higher temperature to generate electricity.
Secondly, the one-way causality from population growth to electricity consumption calls for the government, Eskom and policy makers must audit on electricity consumption to reduce people who are illegally connected to the grid municipalities who does not pay for their electricity bills to comply and pay their debts. This will help reduce the problem of heavy debts on Eskom, reducing financial problems and allowing Eskom the opportunity to be able to produce electricity that matches with a growing population.
Thirdly, the positive relationship between economic growth and electricity generation calls for the policy makers to implement policies that result in an increase in electricity generation to match the growing demand in electricity consumption because of economic growth. The Electricity is the backbone of an economy, electricity is needed to grow the economy. An increase in economic growth means the expansion of the activities of firms, households, and other sectors in the economy. Most of the sectors in the economy depends on electricity for caring their daily activities such as in the primary, secondary and tertiary sectors. This growing in electricity demand because of economic growth then needs to be accounted for by an increase in electricity generation to avoid problems of constant load shedding that has recently marred the South African economy.
Fourthly, bidirectional causality between population growth and economic growth calls for the policymakers to revise policies aimed at increasing population growth. Policies that will have an impact on population growth will also have a causal effect on economic growth. If the government implement policies that increase population growth, this will result in a causal effect on economic growth in South Africa. This is what Solow Growth Model allude when a countries are having the same population growth, saving rate and capital accumulation, then they have the same steady state, so they will converge. Solow (1956) then alludes to say along the convergence path, a poorer country then grows faster.
In conclusion, the study's main objective was to investigate the relationship between population growth and electricity consumption in South Africa by utilising economic growth as an intermittent variable.
The objective was accomplished by the discovery of a negative relationship between population growth and electricity consumption and a positive relationship between economic growth and electricity consumption. This study therefore recommends that in future, the research should consider investigating the relationship between population growth and electricity consumption by employing the panel models and more variables such as unemployment, electricity generation and sectorial analysis to discover new knowledge in the field.