The Relationship between Electricity Prices and Household Welfare in South Africa

Abstract

The study examines the relationship between electricity prices and household welfare in South Africa. The study employs a demand system framework on annual time-series data from 2000 to 2018 and the analysis involves the calculation of price elasticities and measurement of welfare changes. The price elasticities in this study are drawn from the linear expenditure demand model. To analyse welfare change, we consider the impact of electricity pricing policies on cost of living (proxied by the consumer price index and households’ expenditure patterns). The study achieves this: (i) by comparing electricity price movements to changes in the rate of inflation between 2000 and 2018; (ii) by regressing total household energy expenditure against household expenditure on electricity, to examine how electricity costs affect a household’s overall energy bills; and (iii) thirdly, by regressing household food expenditure against households’ electricity expenditure to determine how the latter affects a household’s ability to spend on other basic goods and services. The results of the study show: (i) South African household electricity demand is inelastic to changes in price of electricity; (ii) electricity prices in the country increased at a higher rate than the rate of inflation for most of the time during the study period, suggesting that households incurred increased expenditures to achieve their desired utility or satisfy their energy needs during this period; (iii) household total electricity expenditure is positively related to household total energy expenditure, implying that high household expenditure on electricity exerts upward pressure on the overall household energy budgets; and (iv) household total food expenditure is negatively related to household total energy expenditure. This shows that while policy makers achieved significant success with providing physical access to electricity, affordable access to this basic service is still a concern and affects the overall welfare of households in the country. The study recommends a review of the country’s electricity tariff structure to make affordability a key objective. Moreover, the study calls for coordinated efforts in addressing Eskom challenges which have also played a contributing role to the current energy crisis, characterized by an unreliable electricity supply and constantly increasing electricity prices.

1. Introduction

Energy is a basic need and an important component in the household’s monthly consumption basket. The inability of a household to afford adequate access to energy affects the quality of life and welfare of individuals [1]. In theory, individual welfare is defined as a person’s own assessment of his/her satisfaction given prices and income [2]. In practice, a person’s utility or level of satisfaction is difficult to measure or observe; hence, economists usually measure welfare by using some money-metric indicator which can be observed. Consumption expenditure is the most common and preferred household welfare indicator [3]. Higher energy prices can reduce household welfare by inducing households to spend more money to satisfy their energy needs. A household that spends “more than 10% of its disposable income on energy is considered to be facing energy poverty/insecurity” [4]. Moreover, the higher the household’s expenditure on energy, the more their welfare will be disrupted because they now have less income (resources) to acquire other basic goods and services after paying their energy bills [5].

Despite the use of multiple energy sources, electricity remains the primary source of household energy in South Africa. This is partly due to the adoption of the integrated national electrification program in 1994, which aims to achieve universal access (electrification rate of 97%) by 2025. By 2016, almost 90% of households in the country were electrified [6]. South African households are responsible for about 20% of the economy’s total electricity consumption [7]. Households use electricity for cooking, heating, lighting, cooling, and operating domestic appliances. The results of Stats SA’s income and expenditure surveys of 1999/2000, 2005/2006, and 2010/2011, and living conditions surveys of 2008/2009 and 2014/2015 show that more than 80% of the household energy budget is allocated towards electricity, whereas their expenditure on other energy sources is relatively smaller. Over the years, Eskom has undergone a series of challenges which include rising debt, corruption, and a declining volume of sales, amongst other things [8]. Consequently, all this has resulted in the country experiencing more than a decade of unreliable electricity supply and constantly increasing electricity tariffs. In the decade between 2008 and 2018, South African consumers experienced over inflation increases in the prices of electricity, with electricity prices during this period increased by approximately 300% [9,10]. The residential sector has experienced one of the most significant increases during this period, with the prices of electricity increasing from ZAR 0.5343 per kilowatt-hour in 2008 to approximately ZAR 1.257 per kilowatt-hour in 2018 [11]. Furthermore, in its national energy efficiency strategy, the government identified historically low energy prices as a potential barrier to investment in energy efficiency in the country. To correct this, the government took bold steps to increase the retail prices of electricity with the main goal of establishing cost reflective tariffs by 2015.

Empirical studies produced in other countries such as the United States and Iran have concluded a negative relationship between high electricity prices and household welfare, e.g., [12,13], whereas in South Africa not much work has been carried out specifically around this topic area. In addition, existing econometric studies for South Africa on welfare analysis employ the almost ideal demand system, and most focus on the demand for food or meat, e.g., [14,15,16,17]. To date, no study from South Africa has employed a demand analysis to research specifically about the relationship between electricity prices and household welfare. This study aims to fill this gap in the literature by studying the relationship between the two variables using the linear expenditure demand system framework. Moreover, a study of this nature is also important since electricity is a basic need and this relationship has implications for broader development objectives of the country such as inclusive growth, poverty eradication, managing climate change, and ensuring affordable access to basic services. Against this backdrop, the study evaluates the relationship between electricity prices and household welfare in South Africa.

2. Materials and Methods: Conceptual Framework

This study employs a demand system methodological framework to examine the relationship between electricity prices and household welfare in South Africa over the period from 2000 to 2018. This section will describe the methods followed by the researcher to carry out the objectives of the research. It describes the conceptual framework followed, the model, empirical specifications of the model, the definition of variables, the data sources, and lastly, the steps followed in the estimation of the results. The results of the study are obtained from a demand system framework which involved the calculation of elasticities and measurement of welfare.

2.1. Conceptual Framework

The study is based on the demand system framework. According to [18], “the tendency of consumers to substitute between products in response to a price change has implications for the measurement of welfare”. A proper measurement of welfare must, therefore, account for this substitution behaviour, otherwise it may be at risk of overestimating welfare loss from a price increase, or else, underestimating welfare gain from a price decline. This can be achieved by estimating a demand function which reports the own-price, cross-price, and income elasticity for each good in question [18]. Several demand systems have been developed and applied in the empirical analysis of consumer behaviour. Amongst these is the linear expenditure system (LES) developed by [19], the Rotterdam (ROTT) demand system of [20]), the almost ideal demand system (AIDS) developed by [21]), and the quadratic almost ideal demand system (QUAIDS) developed by [22]. For a household buying n goods, the demand function can take the following general form:

$$q_i=q_i\left(P_1, P_2\dots \dots .pn,I\right)$$

(1)

where i = 1, 2…n denotes commodity, q_i is the quantity of commodity i demanded, P is the price, and I denotes household income.

Although different models have different strengths and weaknesses, the almost ideal demand system remains the most popularly used estimation model. This is mainly because it satisfies almost all the properties of a theoretical demand system and is highly flexible [23]. Furthermore, its functional form is consistent with household budget data and its estimation is simple [23]. The suitability of demand systems for this form of price analysis may also depend on the extent of price variations during the period of study.. As noted by [24], African countries (including South Africa), suffer from a dearth of energy data. This may potentially reduce the performance of the almost ideal demand system in calculating the elasticities required for the analysis. When there is limited price and expenditure data, the linear expenditure demand system is more suitable for use in calculating the elasticities required for the analysis [18]. The linear expenditure demand system satisfies all the theoretical restrictions imposed by consumer theory, and hence, also provides an internally consistent system to measure consumption behaviour [25]. Secondly, the method expresses the quantity of a good consumed as a function of total expenditure and relative prices, which simplifies the estimation and interpretation of model estimates [25]. Against this backdrop, this study employs a linear expenditure demand system instead of an almost ideal demand system to examine the impact of electricity prices on household welfare in South Africa.

2.2. Linear Expenditure Demand System

Stone developed the model in 1954 to study the consumption patterns of families in Britain. The methodology that is employed in determining the parameters of this system is derived from the neoclassical theory of choice [26]. The methodology postulates that a consumer’s utility function is an additively separable function of the form: U (x1, x2, x3, …, xn), and can be represented after a monotonic transformation as a set of individual partial utility functions which can take the form:

$$U={\displaystyle \sum }_{i=1}^n{\beta }_i \ln \left(x_i-{\gamma }_i\right)$$

(2)

where x_i = quantity of the ith good consumed, and β_i and γ_i are the parameters of the utility function, with β_i ˃ 0 x_i > γ_i and the requirement that Σβ_i = 1 so that the preference structure in Equation (4) is well behaved [26]. The behavioural relationships are found by maximising the utility subject to a budget constraint. The budget constraint takes the following form:

$${{\displaystyle \sum }}_{i=1}^n{P}_i{x}_i=M$$

(3)

where i is an energy source/product (e.g., electricity or liquid fuels), n is the number of products, P_i is the price of product i, and M is the total expenditure/income. Solving the constrained maximisation problem gives us an LES model of the form:

$$P_i{x}_i=P_i{\gamma }_i+{\beta }_i\left(M-{\displaystyle \sum }_{j=1}^n{P}_j {\gamma }_j\right)+{\epsilon }_i$$

(4)

where P_ix_i is the total expenditure by households on energy source i, and it is assumed that the household first buys a subsistence quantity of product I = γ_i at cost = P_iγ_i. This is called committed expenditure. ε_i is a disturbance term that is assumed to “follow a normal distribution, with a mean of zero and a variance -covariance matrix for all I” [18]. The total cost of the subsistence/committed expenditure is ΣP_jγ_j. M − ΣP_jγ_j is supernumerary/extra spending. The β_i denotes how households allocate their supernumerary spending over different commodities. The parameter γ_i can be positive or negative. A positive γ_i implies inelastic demand, whereas a negative γ_i implies elastic demand [26].

This demand function satisfies three conditions: firstly, homogeneity in prices and total expenditure; secondly, the adding-up condition requires that the estimated/predicted expenditure for the different commodities equals the total expenditure in any period; and lastly, the regularity condition, which implies quasi-concavity of the utility function [26]. The study adopts the seemingly unrelated regression (SUR) approach proposed by Arnold Zellner in 1962.

Zellner SUR

The concept of seemingly unrelated regression models’ equations is used to describe systems whose equations at first examination appear unrelated, but are related through the correlation in the errors [27]. The correlation may arise from various sources, including correlated shocks in household income ]. The seemingly unrelated regression gained popularity over the years and has been largely applied in estimating demand functions for different households for a given commodity, as well as in estimating the demand of a given household for different commodities. There are alternative estimation techniques such as the generalized least squares estimator (GLS); however, the seemingly unrelated regression was selected for use in this study for gaining efficiency in estimation by combining information from different equations, as well as its relative ease of use.

The analysis was caried out using Stata 13′s nlsur command, which uses the iterative feasible nonlinear least squares estimator. The procedure adjusts for cross-equation contemporaneous correlation, and consequently, considers the optimization process underlying the demand system. The iterative process ensures that the obtained estimates approach asymptotically those of the maximum likelihood method.

After estimation of the linear expenditure demand model parameters (i.e., γ_i and β_i) in Equation (4), the own-price and income elasticities are calculated as follows:

$$E_i=-1+\frac{Pi\gamma i}{Pixi} \left(1-{\beta }_i\right)$$

(5)

$$E_m=\frac{\beta iM}{Pixi}$$

(6)

2.3. Welfare Analysis

Although the estimation of the model in Equation (4) and calculation of the elasticities show how household electricity demand responds to changes in electricity prices, it is not a complete assessment of the true impact on the welfare of households. It is also important in the evaluation of welfare to consider the effect of electricity pricing policies on a household’s expenditure patterns, cost of living, and overall budgets. The study achieves this by comparing changes in electricity price movements with the rate of inflation during the study period to determine the compensating variation. Compensating variation can be used to find the effect of a price change on an agent’s overall welfare. It measures welfare as the difference between expenditure at the initial price and expenditure at the new price (and initial utility). In monetary terms, it can be described as the amount of additional money an electricity consumer would need to reach their initial utility (keep initial utility unchanged) after a change in prices.

Secondly, the study regressed total household energy expenditure against total electricity expenditure to assess how the latter has affected the former during the period under study. The study further regressed food expenditure against electricity expenditure and household expenditure on liquid fuels to assess how energy cost may affect spending on other household goods and services. For this purpose, the study followed [28], which developed the following model to study the effect of energy pricing policies on food prices:

$$lnFPit=\beta 0_i+\beta 1_{i}LE{P}_{it}+\beta 2_{i}lnE{X}_{it}+{\epsilon }_{it}$$

(7)

In this equation, FP represents food prices, EP represents energy prices, i = 1, 2 is the subscript for 10 food items, and T is the subscript for time (March 1995 to February 2018). However, for the purpose of the current study, we are interested in how households’ total expenditure on electricity affects total household energy expenditure, as well as how it affects household expenditure on other basic necessities (food). The above equation is adopted and modified to fit this purpose. The models estimated are specified as follows:

$$lnEnergy Exp=\beta 0_t+{\beta }_{1}lnElectricty Exp{P}_t+{\beta }_{2}lnLiquidfuelEx{p}_t+{\epsilon }_t$$

(8)

where lnEnergyExp Exp is the log of household overall energy expenditure (expenditure on electricity, liquid fuels, and solid fuels), lnElectricity Exp is the log of total electricity expenditure, and lnLiquidfuelExp is the log of household total expenditure on liquid fuels. t = 18 represents the time period (2000–2018).

$$lnFoodExpenditure=\alpha 0_t+{\alpha }_{1}lnElectricityEx{p}_t+{\alpha }_{2}lnLiquidFuelEx{p}_t+{\epsilon }_t$$

(9)

where lnFoodExpenditure is the log of household total expenditure on food and nonalcoholic beverages, lnElectricity Exp is the log of household electricity expenditure, and lnLiquidfuelExp is the log of household total expenditure on liquid fuels. t = 18 represents the time period (2000–2018). The two models were estimated using the same seemingly unrelated regression due to the relative ease of use.

2.4. Estimation Strategy

The analysis of this study involves the calculation of price elasticities and conducting the measurement of welfare change.

2.4.1. Unit Root Test

A necessary condition to meet when conducting research using time-series data is stationarity. If a time series is nonstationary or has a unit root, it will have a time varying mean, variance, and covariance, increasing the likelihood of a spurious regression problem. David Dickey and Wayne Fuller developed the augmented Dickey–Fuller (ADF) test for stationarity in 1979. The test was developed as an improvement of the previous Dickey–Fuller (DF) test, which assumed that there is no correlation in error terms. The ADF removes the possibility of autocorrelation amongst error terms by assuming the following model:

$$\mathsf{\Delta }{Y}_t={\beta }_1+{\beta }_{2t}+\delta y_{t-1}+{{\displaystyle \sum }}_{i=1}^m\alpha i \mathsf{\Delta }{y}_{t-1}+{\epsilon }_t$$

(10)

where:

ΔY_t−1 = (Y_t−1-Y_t−2) ΔY_t−2 = (Y_t−2-Y_t−3)

(11)

The ADF test is conducted by adding lag values of the dependent variable ΔY. The number of lagged terms to include is normally determined empirically and the idea is to include enough terms so that the error terms in Equation (10) are serially uncorrelated with δ, the coefficient of Y_t−1 [29]. The null hypothesis of the test a particular series has a unit root against the alternative hypothesis which states the series is stationary/no unit root. In order to reach a conclusion, the t-statistic is compared to critical values. If the t-statistic is greater than the critical values, we reject the null of nonstationarity, or otherwise, we fail to reject the null hypothesis.

2.4.2. Demand Analysis

In the first step of the analysis, the parameters of the linear expenditure model described in Equation (4) were estimated using seemingly unrelated regression. The study estimates the linear expenditure model for two different energy commodities, namely, electricity and liquid fuels (petroleum products). The energy price and demand data were obtained from the Department of Energy’s price statistics and energy balance reports from 2000 to 2018. Equation (4) indicates that the demand for product 𝑖 (electricity) depends directly on its own subsistence quantity 𝛾𝑖 and indirectly on the subsistence levels of all other energy commodities 𝛾j in the model. Incorporating these cross-equation restrictions requires estimating the model parameters via a system estimator. The study adopts the seemingly unrelated regression (SUR) approach proposed by [27] After estimating Equation (4) and obtaining the parameters (β_i and γ_i), the elasticities were calculated in the form presented in Equations (5) and (6).

2.4.3. Welfare Analysis

The welfare analysis of the study was performed in two steps. Firstly, the study constructs a value of compensating variation by comparing changes in electricity prices to changes in inflation between 2000 and 2018. If benchmark electricity prices increase at a lower rate than inflation, the compensating variation will be negative, indicating that households will have incurred less expenditure to maintain their level of wellbeing/utility after a price rise. Similarly, if electricity prices increased at a higher rate than inflation, the compensating variation will be positive, indicating that households incurred a higher expenditure to maintain their existing level of wellbeing, and this will be interpreted as welfare loss.

In the second step of the welfare analysis, the study regressed total household energy expenditure against household expenditure on electricity to find how it affects households’ overall energy bills during the study period. The study further regressed household expenditure on food against household electricity expenditure to assess how the electricity cost affects spending on other household goods and services. The two models are estimated using the seemingly unrelated regression system estimator.

2.5. Definition of Variables

Household electricity expenditure (lnElectricity): The log of household expenditure on electricity (measured in ZAR per year) is the main dependent variable and is used in the study as a measure of welfare change. To obtain the actual household expenditure on electricity, the quantity of electricity consumed (demand measured in kilowatt-hours) in a year was multiplied by the price of electricity measured in ZAR cents per kilowatt-hours. Economic theory suggests that people consume to obtain utility or satisfaction, and that different quantities of different goods and services provide different levels of utility. Moreover, people tend to purchase a combination of goods and services that will maximise their utility. A change in the price of electricity affects the amount spent by households on utility maximisation, and hence, affects the welfare of households. Other empirical studies have also used household energy expenditure as a proxy of welfare, e.g., [13,30,31].

Household expenditure on liquid fuels (lnLiquidFuels): The log of household expenditure on liquid fuels. Examples of liquid fuels include, diesel, kerosene (illuminating paraffin), and liquefied petroleum gas. To obtain the actual household expenditure on liquid fuels, the quantity of liquid fuels consumed by the residential sector (demand measured in litres) in a year was multiplied by the price of liquid fuel measured in ZAR cents per litre. The transition from traditional fuels (such as wood, dung, and crop residue) to modern fuels has been associated with development and economic improvements. As the prosperity of households increases, they tend to start using nonsolid fuels such as kerosene for cooking, heating, and conducting other household activities. Furthermore, the household income and expenditure surveys from South Africa show that households’ total energy expenditure includes spending on liquid fuels [32,33,34,35]. Household expenditure on liquid fuels (measured in ZAR per year) was, therefore, included to analyse household expenditure behaviours for alternative energy products.

Price of electricity (lnElectricityPrice): This is the log of the price of electricity measured in KwH and is the main explanatory variable of interest in this study. With factors such as income and price of substitute goods remaining unchanged, a rise in the price of electricity can lead to a rise in electricity affecting the quantity of electricity bought, as well as the amount spent on electricity by households.

Price of liquid fuels (lnParrafinPrice): The price of illuminating paraffin (kerosene) is used in the study to proxy liquid fuel prices. Illuminating paraffin is the primary cooking fuel for approximately two million South Africans [36]. The fuel continues to be used as it is easily decanted, widely available in neighbourhood outlets, perceived as affordable, and often the only available option for low-income urban settlements [36].

Average electricity price changes: The log of annual percentage electricity prices increases in South Africa (2000–2018). It is also possible to assess welfare by comparing the annual aeverage electricity price increase to changes in inflation. The variable will help us measure changes in welfare by comparing the difference between electricity and inflation, which is also a measure of the cost of living.

Total household energy expenditure (lnEnergyExp): This is the log of household total expenditure on energy. This includes household expenditure on electricity, liquid fuels, and solid fuels. An increase in electricity expenditure can reduce welfare by inducing households to spend more money for satisfying their energy needs. Hence, a positive relationship is expected between the two variables. Other empirical studies that examined the effect of electricity pricing policies on household welfare have also considered the impact on total household energy cost, e.g., [13,30,31].

Household food expenditure (lnFoodExpenditure): This is the log of household food and nonalcoholic beverage expenditure measured in ZAR. Alongside energy, food is a basic need and a key component in the household total monthly consumption basket. As stated in the earlier parts of the study, high and increasing household energy bills can reduce welfare by reducing the amount of money resources available to cover other basic necessities, including food [5,37].

2.6. Data Sources

The analysis for this study focuses on the residential/household sector. The data used in the study was sourced from the Department of Energy’s annual energy statistics publication from 2000 to 2018; Eskom annual reports; Stats SA’s income and expenditure surveys of 1999/2000, 2005/2006, and 2010/2011, and living conditions surveys of 2008/2009 and 2014/2015; and the South African Reserve Bank online historical statistics database.

3. Results

This section provides a detailed discussion of the summary of findings obtained from the econometric analysis. This includes the descriptive analysis, stationarity analysis, estimation of elasticities, and analysis of welfare change.

3.1. Pre-Estimation Test

This subsection provides the results of the pre-estimation tests, which include a summary of the descriptive statistics and unit root test. These tests were conducted to check the individual characteristics of the variables of interest. The results are shown in Appendix B.

Table A2. Summary of descriptive statistics.

Variable	Mean	Std. Dev	Min	Max
EnergyExp	111.741.50	759,762	95,071	119,850
Electricity	54.33%	24.67	17.03%	94%
LiquidFuels	12.69%	10.71	0.001	52%
FoodExp	254,459.74	131,544.31	84,783.00	488,951.00
EectricityPrice	43.23	29.07	13.23	93.79
ParrafinPrice	550.25	192.33	246.13	924.83
Inflation	5.7%	2.03%	1.4%	10.9%
Avprice incr (electricity)	12.35%	9.41%	−0.003	31.30%

Authors’ own computation using Stat 14.0 and data from Eskom, the Department of Energy statistical price reports and annual energy balances from 2000 to 2018, and South African Reserve Bank statistics database.

in Appendix B shows the results from the summary of the descriptive statistics. These results show that household annual total energy expense averaged ZAR 111,741.50 between 2000 and 2018, whereas expenditure on food and nonalcoholic beverages had a mean of ZAR 254,459.70. As a share of total household energy demand, electricity demand averaged 54.33% between 2000 and 2018, whereas liquid fuels averaged 12.69. Electricity tariffs averaged ZAR 0.4323 per kilowatt-hour between 2000 and 2018, whereas prices of illuminating paraffin averaged ZAR 5.5025 per litre during the same period. Inflation, on the other hand, averaged 5.7% between 2000 and 2018 with a minimum of 1.4% and a maximum of 10.9%. The electricity price increase averaged 12.35 % between 2000 and 2018 with a minimum of −0.003% and a maximum of 31.30%. Household food expenditure had a relatively higher standard deviation (131,544.3), followed by their total energy expenditure (7597.62), paraffin prices (192.33), electricity tariffs (24.67), average electricity price increases (9.41), liquid fuels (10.71), and inflation (2.03). In general, higher standard deviations indicate that the data points are spread out, whereas lower values of standard deviation indicate that the data points are closer to the mean.

The variables were also examined for the unit root using the augmented Dickey–Fuller test, and the results are presented in

Table A3. ADF unit-root test results.

Variable	Level Series	First Difference	Level of Integration
EnergyExp	−1.392	−3.750 ***	I (1)
Electricity	−0.109	−3.598 **	I (1)
LiquidFuels	−2.216	−4.380 ***	I (1)
FoodExp	−2.870	−3.524 ***	I (1)
lnEectricityPrice	2.079	3.240 **	I (1)
lnParrafinPrice	−1.743	4.981 ***	I (1)
lnFoodExp	−4.750 ***	−3.383 ***	I (0)
lnCPI	−2.830 *	−3.873 ***	I (0)
Elect Price increases	−1.608	−3.006 ***	I (1)

Note: *** p-value < 1%; ** p-value < 5%; * p-value < 10%. Source: authors’ own computation using Stat 14.0 and data from Eskom, the Department of Energy statistical price reports and annual energy balances from 2000 to 2018.

in Appendix B. Results of the unit root analysis show that inflation and household food expenditure are stationary at level, whereas all the other variables of the study become stationary after first differencing.

Table 1. Own-price and expenditure elasticities.

Variable	Own-Price Elasticity (Ei)	Income/Expenditure Elasticity (EM)
Electricity	0.701 (0.04) **	0.076 (0.012) ***
Liquid fuel	−2.527 (1.722)	−0.023 (0.12)

Note: *** p-value < 1%; ** p-value < 5% indicate siginificance at 1% and 5% level of significance respectively: authors’ own calculations using data from Eskom and the Department of Energy’s statistical price reports and annual energy balances, 2000–2018.

below provides a summary of findings from the demand estimation.

3.2. Demand Analysis

Table 1 show the own-price and income elasticity of the two different energy sources (electricity and liquid fuels) over the period from 2000 to 2018. The own-price elasticities measure the change in the demand for each of the energy commodities in question in response to price movements. The results show a positive and statistically significant association between the price of electricity and household electricity demand. Furthermore, the own-price elasticity of electricity demand is = 0.70 < 1, indicating that electricity demand is less responsive to changes in the prices. This result is in line with the previous literature, which found electricity demand to be inelastic, e.g., [38,39,40,41,42]. The own-price elasticity of liquid fuels shows that price movements have a negative but insignificant impact on the demand for liquid fuels.

The income elasticity coefficient of electricity demand is positive and statistically significant. This shows that electricity is a normal good and as income rises, household electricity consumers will demand more electricity. The income elasticity of liquid fuel is negative and statistically significant, indicating that liquid fuels are considered by consumers as relatively inferior energy products and that a higher income may lead to less demand for these products.

Although the price elasticity of demand is important to understand how sensitive South African households are to changes in the price of electricity, the study further attempts to estimate welfare by considering the impact of pricing policies on households’ high prices against the cost of living (inflation) and the need to satisfy other basic necessities (e.g., food). This study achieves this in three ways: Firstly, by comparing electricity price movements to changes in the rate of inflation between 2000 and 2018 to determine the compensating variation, which shows whether an electricity consumer would need less or more money to reach their initial utility after a price rise. Secondly, the study regressed total household energy expenditure against household expenditure on electricity to assess how expenditure on electricity has affected households’ overall energy bills during the study period. The study further regressed household expenditure on food against households’ electricity expenditure to determine how electricity cost may affect spending on other household goods and services. The next section presents the results of the welfare analysis.

3.3. Welfare Analysis

To understand the welfare effects of electricity pricing policies, firstly, the study compares the prices of electricity to changes in the inflation rate between 2000 and 2018 to construct a value of compensation variation. The results are shown in

Table A1. Results of computed compensating variation.

Energy Inflation Set Equal to CPI until 2018	Electricity Price %	CPI%	Change in Welfare% (CV)
2000	6.35%	5.39%	0.96%
2001	4.06%	5.64%	−1.58%
2002	9.01%	9.15%	−0.14
2003	7.27%	5.87%	1.4%
2004	2.5	1.4%	1.1
2005	4.1	3.4	0.7
2006	5.1	4.6	0.5
2007	5.9	5.2	0.7
2008	27.5	6.6	20.9
2009	31.3	6.2	25.1
2010	24.8	5.4	19.4
2011	25.8	4.5	21.3
2012	16.0	5.2	10,8
2013	8.0	5.7	2.3
2014	8.0	6.1	1.9
2015	12.7	6.1	6,6
2016	9.4	5.7	3.7
2017	2.2	6.6.	−4.4
2018	2.8	6.6	−3.8

Source: authors’ own calculations based on data from DOE energy price reports from 2000 to 2018.

in Appendix A. Secondly, the study estimates two regression models to understand: (i) how energy expenditure responds to changes in household expenditure on the two different energy sources (electricity and liquid fuels), and (ii) how household expenditure on other necessities is affected. The same seemingly unrelated regression estimator was employed in the estimation of these models due to its relative ease of use.

3.3.1. Compensating Variation

Table A1 in Appendix A displays the results of the compensation welfare analysis based on data from DOE energy price reports published between 2000 and 2018. The results show that prior to 2003, electricity price increases in the country were below the rate of inflation. From 2004 to 2016, electricity price increases in South Africa were higher than increases in the rate of inflation. This indicates that households incurred higher expenditure to achieve their desired welfare or maintain their existing welfare, and hence, the value of compensating variation during this period (2004–2016) was positive. Table A1 also shows that in 2017 and 2018, electricity prices increased at a lower rate than the rate of inflation, and hence, the value of compensating variation for these years was negative, indicating that South African households incurred lower expenditure to attain their desired welfare or to keep their utility unchanged after the price rise. These results are in line with a priori expectations as they imply a negative association may exist between electricity prices and household welfare. In a related study for South Africa, ref [43] found that electricity prices have an upward pressure on inflation, and hence, the cost of living. Another research study conducted by [23] on the impact of energy pricing policies on consumer welfare found that between 1984 and 2012, the rise in energy prices in Pakistan was greater than the rise in the general consumer price index. The study concludes that households in that country experienced welfare loss as they incurred above-inflation expenditures to satisfy their energy needs during the above-mentioned period.

For Indonesia, ref [30] conducted research on how fuel and electricity subsidies impact a household’s welfare. The study employed a linear expenditure framework. The model was estimated empirically using National Social Economic Survey data. Compensating variation and equivalent variation were computed to measure how households in that country respond to a reduction in the amount of subsidy they receive. Amongst other things, results showed that a reduction in the electricity subsidy of IDR 50, IDR 100, IDR 150, and IDR 200 can reduce a poor household’s welfare by IDR 12,946, IDR 25,893, IDR 38,839, and IDR 51,785 per month per household, respectively.

In another study, ref [31] estimated a linear version of an almost ideal demand system on price and energy expenditure data from 1987 to 2012 and found that electricity, coal, and high-speed diesel (HSD) can be considered as some of the leading household energy sources in Pakistan, and the demand for these energy products was found to be relatively less elastic or unit elastic [31]. The computed values of compensating variation showed that the higher the price of energy, the more income compensation is required by the household to remain at the same level of utility as before the price change [31].

3.3.2. Household Energy Expenditure and Household Electricity Expenditure Nexus

The authors performed their own computation using the Stata 14.0 statistical software package and data from Eskom, the Department of Energy, Stats SA publications, and the South African Reserve Bank from the period of 2000–2018.

The regression results displayed in

Table 2. Regression results: household energy expenditure, electricity expenditure, and liquid fuel expenditure nexus.

Variable	Coefficient	Std. Error	p-Values
Electricity	0.104	0.017	0.000 ***
Liquid fuel	0.006	0.017	0.725

Note: *** p-value < 1%; indicates significane at 1% level of significance. Authors’ own computation with Stata 14.0 statistical software package and data from Eskom, Department of Energy, Stats SA publications, and South African Reserve Bank; 2000–2018.

show how the household expenditure on two different energy sources (electricity and liquid fuels) affects the household’s total energy bill. To obtain the actual expenditures, the quantities of electricity and liquid fuel demanded by households was multiplied by the respective prices. As per the results displayed above, there is a positive and statistically significant association between total household energy expenditure and electricity expenditure. This shows that an increase in the budget allocated by households to electricity will result in an increase in the overall household energy bill. These results are consistent with a previous study for South Africa by [37], who found that 73% of the sample respond to higher electricity prices by increasing their electricity budgets to keep their welfare unchanged. In 2013, the Department of Energy’s survey of energy-related behaviours and perceptions in South Africa found that households in the country spend approximately 14% of their income on energy [44], a figure which is significantly higher than the 10% threshold for energy poverty. In another relevant study for South Africa, ref [45] estimated that around more than 15% of households in the country lived in energy poverty in 2014/2015, whereas [46] found that on-grid households in the country are relatively more energy poor than off-grid households. This shows that while connectivity is one part of the problem, affordability of electricity is another genuine issue in the country which is yet to be addressed. The next section presents the results for the food and household energy expenditure nexus.

3.3.3. Household Food Expenditure, Electricity Expenditure, and Liquid Fuel (Petroleum Products) Expenditure Nexus

Table 3. Regression results: household energy expenditure, electricity expenditure, and liquid fuel expenditure nexus.

Variable	Coefficient	Std. Error	p-Values
Electricity	−1.191	0.191	0.000 ***
Liquid fuels	0.188	0.178	0.291

Note: *** p-value < 1%, indicates significance at 1% level of significance. Source: Authors’ own computation with Stata 14.0 statistical software package and data from South African reserve bank and department of energy.

shows the results for the household food expenditure, electricity expenditure, and household liquid fuel expenditure nexus in South Africa between 2000 and 2018. The models were estimated using the same seemingly unrelated regression estimator. The results of the study show a negative and statistically significant association between household food expenditure and total electricity expenditure. This is in line with a priori expectations as it implies that higher energy expenditures can reduce the amount of money or resources available to households for spending on other competing household needs. There are few, if any, studies in the literature that examine the nexus between household food expenditure and energy expenditure. A more related study by [47] for the US concluded that, “an unexpected rise in the prices of gasoline, natural gas, and electricity increases the probability of food access problems, while an unexpected drop in the price of each energy source decreases the probability”. In addition, ref [48] also points out that lower energy costs can help households “afford higher outlays of food, education, and health care”. For South Africa, ref [49] found that the rapid rise in electricity prices in South Africa between 2008 and 2018 resulted in an increase in the energy cost burden for households, whereas [42] concluded that electricity and food are substitutes for all South African households. This implies that households may be willing to sacrifice certain units of one good to obtain additional units of the other good in consumption.

4. Summary of Findings

Based on the results presented in the above sections, the study makes the following findings:

• Household electricity demand is inelastic to changes in prices.
• Electricity prices in South Africa increased at a higher rate than inflation for most of the period from 2000 to 2018.
• High electricity prices can affect household welfare by: (i) increasing their energy budget or the amount of money households need to spend to satisfy their energy needs, and (ii) by reducing household food budgets.

The findings of this study are supported by the results of other studies. Studies by [38,39,40,41], for example, all found electricity demand to be inelastic to changes in price. Furthermore, [13,37,50,51,52] all identified a negative link between high energy prices (including electricity) and household welfare. Although policy makers have made significant progress with providing physical access to electricity, the result of this study shows that affordable access to this basic service is still a challenge for households in the country. The finding has important implications for energy policy and the broader economic objectives of the country, such as reducing poverty. The FBE policy, for example, was one of the measures put forward by the government to achieve affordable access. The quantity of FBE is inadequate to meaningfully improve the standards of living of households of the designated household groups in the country. Evidence suggests that low-income households spend a relatively larger portion of their budgets on energy as compared to other households. Results of this study show that the FBE program represents only a partial delivery of the Energy White Paper’s goal of affordable access, and much still needs to be done to attain it.

5. Conclusions, Recommendations, and Future Research Prospects

The study examined the relationship between electricity prices and household welfare in South Africa. The study followed a linear expenditure demand system framework involving both the estimation elasticities and the measurement of welfare. Based on the results, the study concludes that higher electricity prices reduce household welfare by inducing households to spend more on satisfying their energy needs, whereas higher electricity expenditures reduce available resources to cover other basic necessities such as food. Considering this, the following recommendations are put forward in this study:

5.1. Recommendations for the Government and Eskom

• A review of the country’s current electricity tariff structure shows to include affordability as one of the main objectives when designing tariffs. Eskom’s “cost recovery from users” model affects affordability and undermines other developmental objectives of the country, including those aimed at reducing poverty. The current Eskom tariff structure ensures that those that cannot pay cannot access, and this is inconsistent with the vision set out in the Energy White Paper of 1998.
• The study recommends coordinated efforts from the government, Eskom, and electricity consumers in finding the solution to the Eskom challenges which have perpetuated our current energy crisis. These challenges include ageing infrastructure, governance issues, and financial challenges. These all indicate the urgency with which we must act on the plans put in place to restructure and return the entity to efficiency (including the unbundling process).
• The study recommends a gradual shutting down of all old Eskom power stations and increasing investment towards renewable energy. This can be achieved by allowing more independent power producers in the industry and developing mechanisms to ensure that they charge a fair price.
• The government programs that encourage households to use alternative, safer sources of energy (e.g., rooftop solar power) must be made accessible to all households.
• Extend the reach of demand-side management initiatives and awareness campaigns to even the most remote areas in the country.
• The government can also extend the electricity self-generation threshold beyond 100 MW, as this will take pressure off the national grid and allow households access to reliable and cheap electricity.

5.2. Recommendations for Households

Households can practice efficient use of electricity. Energy is consumed to provide an energy service (e.g., lighting, heating, cooling), whereas capital equipment is used to convert the energy into an energy service (e.g., a light bulb is used to convert electricity into lighting). When the energy efficiency of capital equipment improves, households can enjoy welfare gains resulting from reduced consumption and lower energy bills. South African households must continue to make use of energy efficient technologies (e.g., energy-efficient light bulbs and refrigerators).

5.3. Future Prospects for Research

The results of this study are based on secondary data and quantitative analysis. To capture the true perceptions/experiences of households in the nation on the influence of energy prices on their welfare, future research on this topic should adopt a mixed methods approach (that combines quantitative and qualitative data).