lable at ScienceDirect

Energy 64 (2014) 524e532

Contents lists avai

Energy

journal homepage: www.elsevier .com/locate/energy

Differential electricity pricing and energy efficiency in South Africa

Marcel Kohler*

School of Accounting, Economics and Finance, University of KwaZulu-Natal, Private Bag X 54001, Durban 4000, South Africa

a r t i c l e i n f o

Article history:Received 8 May 2013Received in revised form14 October 2013Accepted 17 November 2013Available online 15 December 2013

Keywords:Electricity consumptionIndustrialSouth Africa

* Office J367, Westville Campus, University of KwaZDurban, South Africa. Tel.: þ27 31 2602574; fax: þ27

E-mail address: kohler@ukzn.ac.za.

0360-5442/$ e see front matter � 2013 Elsevier Ltd.http://dx.doi.org/10.1016/j.energy.2013.11.047

a b s t r a c t

By international standards the economy of South Africa is extremely energy intensive with only a fewcountries having higher intensities. SA’s primary energy use per unit of GDP is amongst the highest in theworld. The high energy and electricity intensity of the economy partly reflects SA’s resource endowments(in particular the abundance of coal) but is also a function of the historical under-pricing of coal andelectricity by the authorities. South African mining & industrial electricity efficiency is particularlyconcerning and considerably lower than the global average. This paper sets out to fill a significant gap inthe South African energy literature by highlighting the importance of incorporating electricity demandfactors as part of the country’s energy policy and electricity planning horizon. The paper focuses itsattention on modelling the electricity consumption of SA’s industrial and mining sectors given theseaccount for the lion’s share of electricity demand. A differential electricity pricing policy which targetselectricity intensive industrial and mining activities (as practised in China since 2004) is viewed by theauthor to be a superior policy to blanket electricity price increases administered by authorities in aneffort to encourage electricity savings and improve energy efficiency in South Africa.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

By international standards the economy of South African isextremely energy intensive with just a handful of countries notablyIceland, Russia and China having higher intensities. South Africa’sprimary energy use per unit of GDP is amongst the highest in theworld standing at 0.13 toe (tonnes of oil equivalent) per thousand2005 US dollars of GDP in 2010 calculated using purchasing powerparities. This compares with values of other energy intensiveeconomies like Iceland (0.25), Russia (0.22) and China (0.16) andaverages of 0.09 and 0.15 respectively for OECD and non-OECDcountries. According to energy statistics published by the [23]there has been a reduction in South Africa’s energy use per unitof GDP in recent years but this compares unfavourably with largeraverage reductions for both OECD and non-OECD countries.

The high energy intensity (and in the case of this paper specif-ically the electricity intensity) of the economy partly reflects SouthAfrica’s natural resource endowments in particular the localabundance of coal and othermineral resources but is also a functionof the domestic under-pricing of coal and electricity by the au-thorities for a long period of time. Historically, the country hasfollowed a heavily capital and electricity-intensive development

ulu-Natal, 1 University Road,31 2607871.

All rights reserved.

trajectory largely based on the use of coal. In 1991, Eskom (thenational electricity provider) proposed a price agreement withgovernment to reduce the real price of electricity to benefitelectricity-intensive activities within South Africa and place themin a stronger position to compete on international markets.

Given the country’s history of low and stable electricity prices,South African electricity efficiency is substantially lower on averagethan in other countries and improvements to date have been smallby international standards. Although under-emphasised in the IRP(Integrated Resource Plan) which sets out South Africa’s plan forelectricity generation over the next 20 years, one of the maintriggers (identified by market commentators) to encourage im-provements in South Africa’s electricity efficiency is to allow energyprices to rise to fully cover operating and capital costs and toproperly value electricity production, transmission and distributionexternalities. Related research by Ref. [26] in the case of China hasindicated that artificially low electricity tariff’s need to be replacedby a system that better reflects the capital costs of power genera-tion and transmission in order to encourage local & foreign in-vestment and efficiency improvements in power generatingcapacity. This paper sets out to fill a significant gap in the SouthAfrican energy literature by highlighting, as in research conductedin the case of China and reported in Energy by Wang et al. (2010),the importance of incorporating electricity demand factors as partof South Africa’s energy policy and electricity planning horizon. Thepaper focuses its attention on modelling electricity consumptionfor South Africa’s industrial and mining sectors given these two

Table 1Electricity intensity: South Africa and Rest of World (GWh/PPP adj. $ million).

1971 1980 1990 2000 2010 Change

OECD 0.249 0.269 0.267 0.264 0.253 1%EU-27 n/a n/a 0.223 0.211 0.204 �8%Non-OECD 0.196 0.220 0.272 0.264 0.277 42%China 0.360 0.443 0.345 0.301 0.371 3%Russia n/a n/a 0.442 0.483 0.362 �18%South Africa 0.258 0.383 0.493 0.553 0.451 75%

Source: Own calculations based on [23].

M. Kohler / Energy 64 (2014) 524e532 525

sectors account for the lion’s share of the country’s electricity de-mand. Our research sets out to support claims in Energy by Ref. [22]that differentiated electricity price policies are required if SouthAfrica is to create an effective energy efficiency policy. Finally, ourstudy estimates long-run output and price elasticities of electricitydemand for the various South African industrial sub-sectors similarto research by Ref. [21]. It does so however by employing differenteconometric techniques and by analysing a longer and more recenttime period: 1989e2009 in an attempt to establish which sectorswould be the best target candidates of a proposed differentialelectricity-pricing scheme. A differential electricity pricing policy(as that practised in China since 2004) and critically reviewed inEnergy by Ref. [28] is viewed by the author to be a superior policy toblanket electricity price increases administered by authorities in aneffort to encourage electricity savings and improve energy effi-ciency in South Africa.

The remainder of the paper is set out as follows. Section 2provides a review of the relevant energy efficiency and energydemand literature whilst section 3 sets out the empirical meth-odology and data employed in the current study. Section 4 reportsthe econometric results of South African industrial electricityconsumption. Section 5 briefly presents international experienceswith industrial energy efficiency policies, Section 6 sets out ourconclusions and policy recommendations.

2. Background

2.1. Electricity efficiency and intensity

Energy efficiency according to the IEA (International EnergyAgency) and the WEC (World Energy Council) involves a reductionin the energy input of a given service (such as heating/cooling, etc.)or level of economic activity. The resulting reduction in energyconsumption whilst usually associated with technological changescan also come about as a result of better organisation and man-agement or improved economic conditions in the sector underinvestigation. Electricity efficiency which is the focus of thisresearch paper is measured as the change recorded in electricityintensity in order to account for its quantitative nature. A commondefinition of electricity intensity adopted in studies by Refs. [30,44]and [20]measures this intensity in terms of electricity consumptionper national production unit such as the J (joule) per US$ of GDP. Inthis paper we follow this approach and measure the electricityintensity of industrial sectors as the electricity consumption tooutput contribution of that sector. Improving the electricity effi-ciency of production processes is generally regarded as a low costand effective way of curbing energy demand in an economy.

2.2. Electricity intensity: the South African case

Energy statistics published by the [23] indicate that SouthAfrica’s electricity intensity has been rising at an alarming rate andby 2010 stood at 0.451 GWh per 2005 US million dollars compa-rable to values for China (0.371), Russia (0.362) and far in excess ofthe OECD and non-OECD average ranges of 0.249e0.253 and0.196e0.277 over the review period 1971 to 2010 respectively (seeTable 1 for details).

The high overall electricity intensity of the South Africaneconomy when compared internationally is the result of a heavilycapital and electricity-intensive development path that has beendriven by the extraction of resources and a set of inter-connectedeconomic activities termed the ‘MineralseEnergy Complex’ [13].This Complex is primarily based on mining, and limited mineralbeneficiation that is underpinned by the provision of cheap elec-tricity. Eskom (the national electricity provider) has been of

fundamental significance to the MineralseEnergy Complexthrough its electricity price fixing agreement with government.This agreement has reduced the real price of electricity since theearly 1990s to benefit electricity-intensive activities within theeconomy. South Africa has thus enjoyed electricity prices amongstthe lowest in the world ande although prices started rising sharplyin 2008 after a series of power outages e by 2011 South Africa stillhad extremely low electricity tariffs compared to other countries(see Fig.1). Whilst statistics for China are not included in this figure,it is noted that the fare charged by the State Grid for 2010 stood at0.16 yuan (US$26) per GWh. Eskom estimates that current SouthAfrican electricity prices are still only about two thirds of the levelneeded to cover total costs, even though average prices have morethan doubled in real terms since 2007 (see Fig. 2).

The alarming rate of increase in South Africa’s electricity-intensityfor the period 1971e2010 implies that South African economy wideelectricity efficiency compares poorly internationally (for details onthis refer back to the percentage changes indicated in Table 1). Ref.[48] suggests that South African industrial electricity efficiency isparticularly concerning and considerably lower than global averages.In particular, industrial activities linked to the MineralseEnergyComplex account for most of the country’s electricity consumptionwhilst contributing far less to South Africa’s GDP. According to the SADepartment of Energy, industry and mining consumed 54% of theelectricity produced in the country in 2010 which has only slightlychanged from the 66% consumed in 1989 (see Table 2).

Our estimates of South Africa’s industrial electricity intensity forthe period 1989e2010 are presented in Table 3. Related research byRef. [20] found South Africa’s primary minerals extraction andprocessing industries linked to the ‘Minerals Energy Complex’ to beextremely electricity-intensive by OECD standards.

Whilst the reported estimates in themselves do not prove thatSouth African industry is inefficient it does suggest that largequantities of electricity are used (per unit value produced) in thecountry’s industrial processes. Information on electricity intensity/efficiency is essential to energy policy makers in understandinghow a country’s demand for electricity changes when the economyundergoes changes in its economic structure. This takes on addedsignificance in a country facing critical energy supply constraints ashas been the case in South Africa since the major electricityblackouts of 2008.

Although now relatively dated, the 1998 White Paper [5] formsthe back-bone for all energy related policy in South Africa. In termsof the energy efficiency of South Africa’s industrial and commercialsectors, the White Paper, commits government to the following:

� Promotion of energy-efficiency awareness;� Encouragement of the use of energy-efficiency practices;� Establishment of energy-efficiency standards for commercialbuildings; and

� Monitoring the progress

In 2005 the Department of Minerals and Energy released SouthAfrica’s first Energy Efficiency Strategy [6]. A national target of 12%

Fig. 1. Industrial Electricity price, international comparison. 2011 or latest year available, USD per MWh. Source: Ref. [23] Energy Prices and Taxes, OECD Estimates and ESKOM.

M. Kohler / Energy 64 (2014) 524e532526

for electricity efficiency improvement by 2015 was set by thestrategy. The aim of the Strategy was to set a policy frameworkallowing for affordable energy to all whilst at the same timediminishing the negative environmental consequences of theextensive energy use in the country. As reported in Energy by Ref.[40] whilst South Africa’s electricity efficiency target was set inlight of the fact that the country was the seventh biggest emitter ofgreenhouse gases on a per capita basis and the national electricityintensity was almost twice the average of the OECD countries, thecountry’s Energy Efficiency Strategy has had limited impact to date.Follow up research by Ref. [22] which focused on factors affectingtrends in energy efficiency in South Africa from 1993 to 2006indicated that structural changes of the economy have played asignificant role in the increasing economy-wide energy inefficiency.This according to the authors contrasts with the utilisation effi-ciency of South Africa’s energy intensity which has contributed topositive improvements in the country’s energy efficiency.

The research by Ref. [22] highlights the urgent need for elec-tricity efficiency improvements in South African industry in light ofthe sector’s large percentage consumption of total electricity pro-duced. According to [12] electricity efficiency improvements pre-sent an opportunity for South Africa firms to: increase profit;improve environmental compliance; mitigate to some extentcompetition from rival producers; and help overcome capital in-vestment constraints. Electricity efficiency improvements should assuch be embraced by the country’s industrial producers.

It is crucial for an economy to be able to generate and distributea sufficient supply of electricity if sustainable economic growth is tobe achieved in that the availability of energy resources and thereliability of these inputs are important determinants of industrialproductivity. At present, there is no feasible method to store elec-trical power on a country-wide scale. The installed capacity must,therefore, be able to generate enough electricity to meet peak de-mand [10]. Growth in capacity to generate power must keep up

Fig. 2. SA Industrial Electricity price 1989e2012, nominal versus real. (SA cents perMWh). Source: Eskom & Own Calculations using Ref. [42].

with growth in demand from consumers in order to avoideconomically damaging blackouts or brownouts. Power shortageshinder growth not only by decreasing productivity, but by forcingfirms to re-optimise among factors by using more material, andfewer energy inputs [14]. Firms will tend to produce fewer (andpossibly import more) of the inputs required in the production oftheir final output. If blackouts become too frequent, firmsmay evenresort to generating their own energy inputs. In the case of China’selectricity supply shortages of the early 2000s [14], found that theoverall effect of the power shortages was to increase companiesproduction costs, finding no evidence of an increase in self-generation by firms.

2.3. Balancing South Africa’s electricity supply and demand

The price industrial consumers pay for electricity in South Africais determined by regulators, and not demand-supply forces in themarket. The lack of an equilibrating price mechanism can lead totemporary demandesupply imbalances, especially if government isslow to react to market signals. In a situation where the price ofelectricity is driven by supply and demand, a high price wouldsignal excess demand and would soon drive more investment intothe industry. High electricity prices also encourage innovation inalternative methods of power generation, as well as greater effi-ciency in consumption and in production processes in whichelectricity is a key input [10]. Without a market-determined elec-tricity price it is up to regulators to forecast the future energy needsof the economy and make the appropriate capacity investments.

The electricity supply problem faced by South Africa can bethought of in terms of two related dimensions. Total installedgeneration capacity was insufficient during 2007/2008, and iscurrently still struggling to meet peak demand. The subsequentscheduled blackouts, termed ‘load shedding’, had highly adverseeffects on economic growth, job creation, and foreign and localinvestor confidence. At present, the safety margin between supplycapacity and demand is too low to allow the required routinemaintenance of capacity, leaving the state electricity utilityvulnerable to unexpected demand increases and down time (HSRC,2009). Both facets of the problem are caused by inadequate gen-eration capacity and a lack of quality coal inputs. Unable to financethe required capacity expansion through profits, Eskomwas forcedto source a ‘controversial’World Bank loan to fund the constructionof the Medupi coal-fired power station and other smaller projects[16]. Through its Eskom Power Investment Support project, theWorld Bank will contribute a total of US$3.75 billion to theexpansion, of which US$3.05 billion will go towards the 4.8 GWMedupi power station, US$260 million towards a 100 MW windand 100 MW concentrated solar power project, and US$485 to-wards efficiency improvements including the conversion of coaltransportation from road to rail [46].

Table 2South African electricity consumption by economic sector (MWh).

1989 % 2010 %

Industry excl mining 57,480,112 41.72 88,864,830 41.54Mining 34,667,867 25.16 28,772,620 13.45Transport sector 4,229,831 3.07 3,640,190 1.70Agriculture 3,438,991 2.50 6,163,900 2.88Non-specified (other) 12,000,000 8.71 14,211,860 6.64Commerce and public services 14,445,983 10.48 30,412,450 14.22Residential 21,518,989 15.62 41,844,740 19.56Total economy 137,781,773 100.00 213,910,590 100.00

Source: Ref. [8] Energy Balances & Ref. [23] Energy Balances for Non OECD countriesvarious issues.

M. Kohler / Energy 64 (2014) 524e532 527

The scope of supply-side solutions to the problem is limited inthe short-run, primarily due to the long lead times associated withpower station construction. Demand side solutions where users areinduced to decrease electricity demand are, therefore, more usefulin emergency situations where a rapid reduction in consumption isneeded. Incentives put in place to curb demand could includegovernment subsidies for installing less electricity-intensive pro-duction equipment, or a tariff structure that encourages users toconserve electricity where possible [19].

Cheap electricity has placed electricity-intensive South Africanindustries in a strong position in international markets, which hasencouraged investment in these industries. Regulators are nowreluctant to raise tariffs for fear of eroding the competitiveness ofthese firms and precipitating widespread job cuts. Eskom has alsoentered into a 25 year pricing contract with Alusaf, a subsidiary ofBHP Billiton and South Africa’s primary aluminium producer, whichguarantees a constant supply of electricity at a reduced tariff that islinked to the LondonMetal Exchange aluminium price. Eskom’s SPA(Special Pricing Agreements) have required subsidisation fromresidential users in the form of both higher residential tariffs andless reliable access to electricity. The present pricing regime is,therefore, transferring wealth from residential users to energy-intensive big business, in contrast to government policy thatspecifies that large-scale industrial users are to cross-subsidisepoor domestic consumers [16].

Ref. [21] note that the steady decline in real electricity pricesbetween the 1980s and 2007 led to a lower electricity consumptionresponsiveness to changes in the electricity price. On the otherhand, the early 1980s saw a sharp rise in the real price of electricityand a subsequent increase in magnitude of the price elasticity ofelectricity consumption. Electricity tariffs have been rising rapidlysince 2007, providing an opportunity for the sensitivity of demandto the changing electricity price to be observed. Real GDPwas 10.4%higher in the first half of 2012 than it was in 2007, while the real

Table 3SA industrial electricity intensity: 1989e2010.

Industrial sector 1989 1999 2010 Ave

Non-ferrous metals 0.6609 0.5836 0.6523 0.5964Iron & steel 0.2541 0.4211 0.2173 0.3108Non-specified industry 0.2424 0.3248 0.2548 0.2618Mining 0.2596 0.1608 0.1798 0.1914Non metals 0.0539 0.0552 0.0827 0.0724Chemical & petro 0.0535 0.0141 0.0380 0.0338Wood & products 0.0475 0.0389 0.0123 0.0222Paper & print 0.0216 0.0217 0.0230 0.0244Tex, leather & foot 0.0120 0.0114 0.0167 0.0120Food, bev & tob 0.0037 0.0044 0.0040 0.0042Machinery & equipment 0.0024 0.0007 0.0005 0.0005Transp equipment 0.0001 0.0002 0.0003 0.0005Construction 0.0002 0.0003 0.0004 0.0004Total industry 0.1078 0.0921 0.0715 0.0842

Source: Ref. [8] Energy Balances & Ref. [23] Energy Balances for Non OECD countriesvarious issues.

electricity price had more than doubled over the period. Electricityoutput had fallen by 2.6%, partly as a result of the rapid tariff in-crease and partly due to sectoral shifts that are unrelated tochanges in the electricity price [32]. The fixed nature of manystructures and pieces of equipment used in electricity-intensiveproduction processes generally leads to the belief that demandelasticities are low in the short-run. In the long-run, the variabilityof these factors allows a greater degree of optimisation in pro-duction processes and substitution of inputs, leading to higherdemand elasticities. A sharp rise in the real price of electricity overthe next decade is likely to amplify the role that electricity tariffsplay as a demand determinant in the South African economy whichis why accurate estimates of industrial electricity demand respon-siveness to output and price changes are crucial in the South Afri-can context.

The econometric estimation of energy demand elasticities canbe traced back in time to a period substantially earlier than theseminal work of [25] which sparked renewed interest in energy-growth studies. Despite the importance of reliable elasticity esti-mates in energy modelling to inform economic policy, there is asurprising scarcity of literature on industrial energy demand elas-ticities particularly so in the case of electricity. Table 4 provides asummary of industrial electricity demand studies to date.

The studies differ largely with respect to the econometric meth-odology used, the time span covered, and the country analysed. Interms of the elasticity of industrial electricity demand to economicactivity this is indicated to vary between 0.15 and 1.22 whilst theelasticity with respect to price is indicated to vary between �0.04and �0.45 in the short-run and �0.31 and �1.94 in the long-run.

3. Methodology and data

In order to estimate a long-run relationship for the South Afri-can economy’s industrial sectors’ electricity demand we employthe following general function specification:

Et ¼ f ðQt ; Pt ; Zt ;XtÞ (1)

where Et (electricity consumption) is contemporaneously depen-dent on the level of Qt (real economic activity), Pt (real electricityprice), other Zt (endogenous variables) such as the real price of asubstitute for electricity, and Xt (exogenous variables), such as asector-specific coefficient for autonomous technical change,energy-saving technological progress or changes in the structure ofindustrial production. Such structural changes may be due to thesubstitution of labour by electricity-using capital and/or the off-shoring of labour intensive production processes to other countries.Changes such as these tend to increase the electricity intensity ofindustrial sub-sectors in contrast to energy-saving technologicalprogress. Since these factors affect the relationship between theother variables we can account for these indirectly through theinclusion of a deterministic term.

Numerous studies (for an up to date survey of these see Ref.[43]) find that electricity in industrial processes is not easilysubstituted by other energy inputs. Taking cognisance of this in ouranalysis we do not control for inter-fuel substitution and thusexclude the prices of other energy carriers. We therefore, adopt thefollowing standard constant elasticity (Cobb-Douglas type) repre-sentation in our empirical analysis:

Et ¼ C0 expðdummyÞQbqt Pbpt (2)

where Xt ¼ C0 exp(dummy) is the deterministic term, C0 is a con-stant, exp(dummy) is a time dependent dummy and bq and bp arethe demand elasticities in respect of economic activity and

Table 4Industrial electricity demand studies and elasticity estimates.

Study Country Method Elasticity estimates

Output Price

Fisher & Kaysen (1962)a US OLS �1.25Baxter & Rees (1968)a UK OLS �1.50Anderson (1971)a US OLS �1.94Mount et al. (1973)a US Pooled OLS LR: �1.82 SR: �0.22[17] Kenya OLS �0.09 to �0.78[2] Israel Cointegration LR: 0.99 to 1.22 LR: �0.31 to �0.44[3] India Pooled regression 0.49 to 1.06 SR: �0.04 to �0.45[24] USA Simul. equations �0.34 to �0.55[37] Greece Cointegration LR: 0.85 SR: 0.61 LR: �0.85 SR: �0.35[18] China CGE model LR: �0.017 to �0.019[9] Turkey Cointegration 0.15 �0.16[29] Germany Cointegration LR: 0.7 to 1.9

SR: 0.17 to 1.02LR: 0.00 to �0.52SR: �0.31 to �0.57

SR e short-run; LR e long-run.a These studies are reviewed in Ref. [45].

M. Kohler / Energy 64 (2014) 524e532528

electricity price, respectively. The advantage of this standard log-linear specification is its simplicity and limited data requirementsand according to [36] performs better than more complex models.

Econometric studies on the estimation of energy demand elas-ticities are oftenbased on time series data. Since the seminalwork of[11]; cointegration analysis has increasingly become the favouredmethodological approach for analysing time series data to overcomethe spurious regressionproblemwhen the time series are integratedof order one, I(1) or higher. Instead of taking first differences of thedata, the common methodological approach adopted previously, itis possible to deal with the problem by identifying existing sta-tionary linear combinations of two or more non-stationary timeseries. The presence of stationary linear combinations indicatescommon stochastic trends (i.e. cointegration), these are interpretedas long runequilibrium relationships between the variables and, cantherefore, according to [11]; be characterised by being generatedthrough an error correction mechanism.

Unfortunately, unit root and cointegration testing undertaken ina pure time series context suffers from the problem of low pre-dictive power and small sample size. The inclusion of a cross-sectional dimension in the analysis is often employed to helpovercome this problem. An alternative approach is the ARDL(autoregressive distributed lag) bounds testing approach to coin-tegration. This method, introduced by Refs. [34] and [35]; hasenjoyed considerable support over recent years. The advantage ofthe ARDL approach is that information regarding the order ofintegration of the variables is not required. The pretesting for unitroots, which is needed in other cointegration methodologies can beomitted. The significance of a long-run relationship is tested usingcritical value bounds, which are determined by the two extremecases that all variables are I(0) (the lower bound) and that all var-iables are I(1) (the upper bound).

Taking natural logarithms of Eq. (2) and adding an error termyields the econometric specification of our long-run industrialelectricity demand function:

et ¼ b0 þ b1dummyþ b2qt þ b3pt þ εt (3)

where et ¼ ln(Et); qt ¼ ln(Qt) and pt ¼ ln(Pt). The bs are the long-runcoefficients and εt is a white noise error term.

The first step of the bounds testing approach is to estimate thefollowing unrestricted error correction model using OLS:

Det ¼ cþ dummyþ f1et�1 þ f2qt�1 þ f3pt�1

þXk

i¼141iDet�i þ

Xl

i¼142iDqt�i þ

Xm

i¼143iDpt�i þ vt

(4)

where the f are the long-run multipliers, c is a drift term, 4 are theshort-run coefficients and vt is a white noise error term. Due to thefact that it is not clear a priori whether q and p, are the long-runforcing variables for electricity consumption, current values of Dqand Dp are excluded from Eq. (4).

As a second step, an F-test on the joint hypothesis that the long-run multipliers of the lagged level variables are all equal to zeroagainst the alternative hypothesis that at least one long-runmultiplier is non-zero is conducted, i.e.:

H0 : f1 ¼ f2 ¼ f3 ¼ 0;

H1 : f1s0; or f2s0; or f3s0:

Critical values which depend on the number of regressors andthe deterministic terms included are provided by Ref. [33]. For eachconventional significance level, two sets of critical values are given,which constitute the lower and the upper bound. The lower boundrepresents the critical values for the case in which all includedvariables are assumed to be I(0), while the upper bound assumes allthe variables to be I(1). Hence, all possible combinations of ordersof integration for the single variables are covered. If the calculatedF-statistic lies above the upper bound, the null hypothesis of nocointegration can be rejected, irrespective of the number of unitroots in the single variables. On the other hand, if it lies below thelower bound, the null hypothesis is not rejected. Only if the F-sta-tistic lies between the bounds, are the results of the inferenceinconclusive, given that the order of integration of the single vari-ables is unknown.

If the existence of a significant cointegration relationship isidentified by the bounds F-test, the next step is to select the optimalARDL specification of Eq. (4). This process is guided by the AIC(Akaike Information Criterion) and the SBC (Schwarz BayesianCriterion). Furthermore, the properties of the residuals are checkedto ensure the absence of serial correlation. A representation of theARDL(k, l, m) model in the general case is:

Det ¼ b0 þXk

i¼1a1;iet�i þ

Xl

i¼0a2;iqt�i þ

Xm

i¼0a3;ipt�i þ ut

(5)

where ut is an error term and k, l and m are the lag lengths of thesingle variables.

The long-run coefficients are constructed as non-linear func-tions of the parameter estimates of Eq. (5) as follows:

b0 ¼ ac=�1�

Xki¼1

a1;i

�(6)

Table 5Bounds F-tests for a cointegration relationship.

Lag length two Fe (ejq, p)¼Total industry 4.837**Non-ferrous metals 3.864*Iron & steel 3.193*Mining 1.062Non metals e

Chemical & petro e

Wood & products 3.186Paper & print 3.813*Tex, leat & foot e

Food, bev & tob 86.318***Machinery & equipment e

Transp equipment 1.962Construction 3.126

Notes: ***, ** and * denote significance at the 1%, 5% and 10% level, respectively.

M. Kohler / Energy 64 (2014) 524e532 529

b1 ¼ ad=�1�

Xk

i¼1a1;i

�(7)

and

bj ¼XZ

1aj;i=

�1�

Xk

i¼1a1;i

�(8)

with j ¼ 2, 3 and z ¼ k, l, m. b0 and b1 are the constant and thedummy in the long-run model represented by Eq. (3), respectively.The bj are the long-run slope coefficients.

Finally, the (dynamic) short-run coefficients for the errorcorrection representation are estimated according to:

Det ¼ qc þ qd þ qectECTt�1 þXk

i¼1a1;iDet�i þ

Xl

i¼1a2;iDqt�i

þXm

i¼1a3;iDpt�i þ ut

(9)

where ECTt�1 is the error correction term resulting from the esti-mated long-run equilibrium relationship, Eq. (3), and qect is thecoefficient reflecting the speed of adjustment to long-run equilib-rium, i.e. the percentage annual correction of a deviation from thelong-run equilibrium the year before.

3.1. Data

South African electricity consumption data at the level of thedifferent economic sub-sectors is taken from the Department ofEnergy’s Energy Balances (DoE various issues) and is measured inMWh. The Energy Balances classify the economy into five sectors:the industrial sector, the commercial, agricultural, residential andtransport sectors which are further disaggregated into 22 in-dustries. This data is collected by the Trade and Industry division inStatsSA in collaboration with the DoE. The main source of the in-formation is Eskom and the NERSA (National Energy Regulator).

The data series on electricity prices for the various economicsub-sectors is taken from the Energy Price Report, 2011 [7]. Thispublication identifies tariffs for various energy types includingelectricity. The tables for the electricity charges are derived fromEskom’s Statistical Yearbooks and Annual reports. The electricityprices are averages for the economic sectors and are imputed fromEskom’s revenue per kWh by customer category: Bulk, Domesticand Street Lighting, Commercial, Industrial, Mining, Rural/Farming,Traction and International. The data applies only to Eskom chargesto its direct customers and excludes sales by local authorities.Whilst these prices are presented in nominal terms we convertthese into real prices by using the annual CPI (Consumer Price In-dex), with 2005 as the base year, provided by StatsSA (StatisticsSouth Africa) [42].

The data series on real total output is taken from Ref. [39] In-dustry trends database. This is measured in millions of Rands andconverted into 2005 real prices by using the CPI (Consumer PriceIndex) from StatsSA (Statistics South Africa) [42].

4. Econometric results

As a first step of the ARDL bounds testing procedurewe estimateEq. (4) for each industrial sector using OLS. As our analysis is basedon annual data, we consider lag lengths of one and two. A timespecific dummy which takes into account a structural break in ourdata series for the 1993/1994 political transition to democracy inSouth Africa is included whenever significant. Next we undertake aF-test on the joint significance of the lagged variables in levels. Theresults of the F-tests for all sectors are shown in Table 5. The F-

statistic indicates no joint significance for the sectors non-metallicminerals; chemicals & petrochemicals; textiles, leather & footwear;andmachinery & equipment. For all other industrial sectors the nullhypothesis of no long-run relationship is rejected at least at the 10%level.

4.1. Long-run and short-run elasticities

Based on the bounds test results, we proceed to estimate thelong-run elasticities and the corresponding error correction modelsfor mining and seven industrial sub-sectors. Equation (3) is esti-mated for each of these South African industrial sectors, the modelselection is guided by the SBC (Schwarz Bayesian Criterion) whichsuggests a maximum of two lag lengths be incorporated in ourmodel estimation. The estimated residuals are tested to ensure theyare not serially correlated. The parameter estimates are then usedto construct the long-run elasticities according to Eqs. (6)e(8).Finally, to establish the short-run dynamics of industrial sectorelectricity consumption, the corresponding error correction modelsaccording to Eq. (9) are estimated using the lagged ECTs obtainedfrom the long-run relationships estimated.

Tables 6e8 provide a summary of the estimated long-run co-efficients, the error correction estimation results and the diag-nostic tests (for serial correlation, normality, andheteroscedasticity) of the underlying ARDL models for therespective industrial sectors. The order of the industrial sector-specific ARDLs along with the estimated long-run coefficientsare presented in Table 6. For total industrial electricity consump-tion the signs of the statistically significant income and priceelasticities are positive and negative as expected confirming theresults of previous work over the short period of analysis 1993e2006 by Ref. [21]. Our results suggest a price inelastic electricitydemand (elasticity ¼ �0.939) for South Africa’s industrial sectorfor the period 1989e2009. This compares favourably with theprice elasticity ¼ �0.869 estimated by Ref. [21]. Our results inrespect of industrial sector output suggest this likewise is a highlysignificant factor which influences SA’s industrial electricity con-sumption with an output elasticity ¼ 0.628. Ref. [21] estimate anelasticity ¼ 0.712 in this regard. In respect of the industrial sectorspecific results the long-run output elasticities range between0.293 and 2.578 (in the case of the short-run dynamics between0.473 and 1.027). The long-run demand elasticities with regard toprice range between �0.586 and �1.774 and in the short-runbetween �0.944 and �2.589. The results suggest that significantresponsiveness’s in industrial electricity consumption to pricechanges are found to exist in the construction, paper, pulp & printand iron & steel industries.

Table 6Long-run coefficients for sector-specific ARDLs.

Sector order of ARDL Constant Dummy Output Price

Total industry ARDL(1,0,0,2) 12.314*** (0.003) �0.904*** (0.000) 0.628*** (0.006) �0.939** (0.018)Non-ferrous M ARDL(1,1,0,0) �6.396 (0.423) 0.275 (0.296) 1.797*** (0.001) 1.629 (0.243)Iron & steel ARDL(2,0,0,2) 14.955*** (0.000) 0.196*** (0.003) 0.293*** (0.000) �0.586** (0.011)Mining ARDL(1,0,0,0) 10.370 (0.107) �0.149 (0.225) 0.553 (0.264) 0.126 (0.618)Wood & product ARDL(1,2,0,1) 22.664*** (0.001) 0.175 (0.435) �1.049*** (0.008) 0.102 (0.908)Paper & print ARDL(2,1,0,0) 13.365*** (0.002) �0.328*** (0.004) 0.538** (0.031) �1.774*** (0.001)Food, bev & tob ARDL(2,0,2,2) 16.025*** (0.004) 1.969* (0.022) �0.530 (0.321) 0.706 (0.342)Transp equipment ARDL(1,1,1,0) �18.589** (0.020) 0.159 (0.656) 2.578*** (0.000) �0.347 (0.804)Construction ARDL(1,1,0,0) 32.870** (0.023) �1.456** (0.023) 0.032 (0.961) �7.765*** (0.002)

Notes: ***, ** and * denote significance at the 1%, 5% and 10% level, respectively. p-values are reported in brackets.

M. Kohler / Energy 64 (2014) 524e532530

5. International experience with industrial energy efficiencypolicies

Previous research in Energy by Ref. [31] on energy intensitytrends in 31 industrial and developing countries over the period1950e1988 suggests that electricity intensities are likely to developsimilarly to how energy intensities have developed as economicstructure and end-use efficiency continue to change. According to[27] there is a wealth of experience among industrialised countrieswith technologies and policies to increase electricity end-use effi-ciency. The authors indicate that some developing countries arebeginning to adopt these technologies and policies many of whichfocus on the demand-side management of electricity consumption.In the case of the industrial sector of Slovenia Ref. [1], indicate thatimprovements to internal industrial conversion systems, notablycogeneration of electricity and heat, are amongst the technologiesthat produce a major part of the overall electricity efficiency gains.Research by Ref. [15] on energy efficiency in the German pulp andpaper industry identifies heat recovery in paper mills and the use ofinnovative paper drying technologies as the most influential tech-nologies in reducing the sector’s energy demand.

Aided by an extensive discussion in Energy of worldwide ex-periences with the demand-side management of electricity [47],report on the integral role of China’s demand response programs inalleviating and coping with electricity supply shortages at a na-tional level. In addressing South Africa’s electricity supply short-ages [20], identify the need for a nation-wide demand-sidemanagement programme to improve energy efficiency. The authorssuggest that electricity price reform, such as that recentlyannounced in South Africa, whereby the electricity price level isincreased significantly in conjunction with block-rate tariffs thatcharge a higher rate to those that consume more is vital if thecountry is to reduce its electricity intensity. These claims are sup-ported by follow up research by Ref. [22] based on a decompositionanalysis of South Africa’s energy efficiency that calls for a differ-entiated energy pricing regime similar to that practiced in China.These claims are in line with earlier research reported in Energy byRef. [41] that highlighted the important role played by energy

Table 7Error correction representations for the underlying ARDL models.

Sector ECTt�1 Det�1 Dqt

Total industry �0.668*** [0.003] 0.419* [0.078]Non-ferrous metals �0.571*** [0.001] 1.027*** [0.000]Mining �0.618** [0.019] 0.342 [0.239]Iron & steel �1.615*** [0.000] 0.472 [0.077] 0.473*** [0.005]Wood & products �0.496** [0.035] 0.907 [0.221]Paper, pulp & print �0.928*** [0.000] 0.329 [0.096] 0.499* [0.074]Food, bev & tobacco �0.207*** [0.009] �0.730*** [0.000] 0.188 [0.163]Construction �0.738*** [0.001] 0.024 [0.962]

Notes: ***, ** and * denote significance at the 1%, 5% and 10% level, respectively. p-value

pricing policy in influencing patterns of energy consumption andproduction in the U.S. and in the Asia-Pacific region.

The principles, effects and problems associated with a DEPP(differential energy pricing policy) for energy intensive industriesin China, are discussed at length in Energy by Ref. [28]. In essence,in the case of China, the government instituted special energypricing policies in June 2004, in an attempt to improve energy ef-ficiency and abate pressure on installed generation capacity. Energyintensive industries that did not meet specific efficiency andenvironmental targets were taxed under the DEPP by being forcedto pay a higher electricity price [10]. Initially the ferroalloy,aluminium, caustic soda, cement, steel, and calcium carbide in-dustries were subject to the pricing policy, with phosphorus andzinc smelting being included later in September 2006. Firms inthese industries were divided into four categories, namelyencouraged, permitted, restricted and eliminated, with the lattertwo (low output, low efficiency firms) paying a surcharge on thebasic electricity price [10]. The former two categories received anadjustment to the provincial wholesale electricity price. Surchargesfor restricted and eliminated enterprises were 5 fen and 20 fen perkwh respectively, approximately 10%e20% of the basic price [38].The objective was to drive inefficient firms out of the market or toforce innovation or investment in less energy-intensive productionmethods.

In the context of the South African economy, the research pre-sented here supports claims made in Energy by Ref. [22]. Namely,based on experiences in China and the research findings of [28] and[10] the introduction of alternative demand-side electricity man-agement policies, such as time of use, which punishes inefficientusers and a more diversified pricing schedule (that places thehighest cost burden on the country’s industrial and commercialconsumers that are least efficient) should be supported in SouthAfrica.

Ultimately, such electricity pricing strategies if adopted inter-nationally in electricity-intensive economies should help incenti-vise energy efficiency improvements and encourage thedevelopment of renewable energy resources and smart-grid tech-nologies within these countries.

Dqt�1 Dpt Dpt�1 Dummy

�0.627** [0.021] �1.172*** [0.000]0.124 [0.803] 0.157 [0.205]0.078 [0.631] �0.092 [0.178]�0.944** [0.033] 0.011 [0.915]�0.976 [0.253] �2.589** [0.031] 0.379 [0.161]�0.432 [0.194] �0.304** [0.012]

0.718*** [0.004] 0.146 [0.198] 0.194*** [0.000]�0.695 [0.643] �1.075** [0.017]

s are reported in brackets.

Table 8Diagnostic tests for the underlying ARDL models.

Sector Lagrange multiplier statistics

Serial correlation:c2SC (1)

Normality:c2N (2)

Heteroscedasticity:c2H (1)

Total industry 0.310 [0.578] 0.143 [0.931] 1.577 [0.209]Non-ferrous M 0.313 [0.860] 1.134 [0.567] 0.286 [0.592]Iron & steel 0.333 [0.564] 0.937 [0.626] 0.745 [0.388]Mining 0.944 [0.331] 0.347 [0.841] 0.129 [0.719]Wood & products 1.418 [0.234] 0.191 [0.909] 0.008 [0.977]Paper, pulp & print 1.005 [0.316] 0.615 [0.735] 0.032 [0.859]Food, bev & tobco 0.795 [0.373] 23.353 [0.000] 1.278 [0.258]Construction 0.169 [0.681] 1.207 [0.547] 2.929 [0.087]

Notes: p-values are reported in brackets.

M. Kohler / Energy 64 (2014) 524e532 531

6. Conclusion

Government intervention in the form of taxes and surchargescan discourage investment in a particular target industry, reducingits share of GDP over time. The challenge to policy makers inter-nationally is to implement these surcharges in a manner that in-flicts the least damage on output and employment. The negativeeconomic impacts associated with an increase in the price ofelectricity in electricity-intensive economies could be minimised ifthe price increases are diversified amongst traditionally highelectricity consuming industries (such as non-ferrousmetals, iron &steel, mining, non-metallic minerals and paper & pulp). Byemploying a differential pricing policy, authorities can targetelectricity-intensive industries by charging them higher tariffs inorder to encourage greater production efficiency and reduceaggregate electricity demand. A differential tariff structure wouldraise the cost of energy inefficiency and induce a re-optimisation ofproduction processes so that more material inputs and fewer en-ergy inputs are used in energy-intensive industries. A differentialelectricity pricing regime in the case of South Africa would driveout the least electricity-efficient industries in the long term,changing the structure of the economy to one that is less energy-intensive, with a smaller carbon footprint. In so doing, theadverse impact of structural changes of the South African economyon economy-wide energy efficiency as highlighted in researchworkby Ref. [22] are brought into check. Our estimates of the long-runresponsiveness of South Africa’s industrial electricity consump-tion to price increases suggest that those industries which wouldmake ideal targets for such a differential electricity pricing schemeare the construction, paper, pulp & print and iron & steel industries.Our industrial electricity consumption elasticities suggest that theiron & steel industry would even respond to such price increases inthe short-run.

While maybe not as immediately obvious, the long-run costs ofa chronic electricity supply deficit on growth of industrial outputcertainly outweigh the short-run job and production lossesresulting from and an electricity price increase. A study by Ref. [4]found that scheduled load-shedding at 10% of total annual capacitywould shrink South African GDP by as much as 0.7%, and noted thatalthough difficult to quantify, the adverse impact of unscheduledblackouts would be far greater.

The effects of an insufficient and unreliable electricity supply inelectricity-intensive economies are not limited internationally toindustries in which electricity is a key input. As the output of firmsin electricity-intensive industries is constrained by an electricitysupply shortage, so their demand for other inputs is reduced. In thisway the impact of a supply deficit can be seen as truly economy-wide, as, theoretically, even industries which do not utilise elec-tricity in production may suffer. Ultimately, advocating the adop-tion of differential electricity pricing strategies internationally in

electricity-intensive economies should help incentivise energy ef-ficiency improvements and encourage the development ofrenewable energy resources and smart-grid technologies withinthese countries.

Acknowledgements

The author would like to thank Economic Research SouthernAfrica for financial support towards this research project.

References

[1] Al-Mansour F, Merse S, Tomsic M. Comparison of energy efficiency strategiesin the industrial sector of Slovenia. Energy 2003;28(5):421e40.

[2] Beenstock M, Goldin E, Nabot D. The demand for electricity in Israel. EnergyEcon 1999;21(2):168e83.

[3] Bose RK, Shukla M. Elasticities of electricity demand in India. Energy Policy1999;27(3):137e46.

[4] Deloitte. Estimating the elasticity of electricity prices in South Africa. Johan-nesburg; 2009.

[5] Department of Minerals and Energy (DME). White paper on energy policy.Pretoria: Department of Minerals and Energy; 1998.

[6] Department of Minerals and Energy (DME). Energy efficiency strategy of theRepublic of South Africa. Pretoria: Department of Minerals and Energy; 2005.

[7] Department Of Energy (DoE). Energy price report 2011. Pretoria; 2012.[8] Department of Energy (DoE), (various issues). Energy balances. Pretoria:

Department of Energy; Available from: http://www.energy.gov.za/files/media/media_energy_balances.html.

[9] Dilaver Z, Hunt LC. Industrial electricity demand for Turkey: a structural timeseries analysis. Energy Econ 2011;33(3):426e36.

[10] Edwards T. China’s power sector restructuring and electricity price reforms.Asia Paper 2012;6(2).

[11] Engle RF, Granger CWJ. Co-integration and error correction: representation,estimation and testing. Econometrica 1987;55(2):251e76.

[12] Fawkes H. Energy efficiency in South African industry. J Energy South Af2005;16(4).

[13] Fine B, Rustomjee Z. The political economy of South Africa. From minerals-energy complex to industrialisation. London: Hurst; 1996.

[14] Fisher-Vanden K, Mansur E, Wang Q. Costly blackouts? Measuring produc-tivity and environmental effects of electricity shortages; 2013.

[15] Fleiter T, Fehrenbach D, Worelli E, Eichhammer W. Energy efficiency inGerman pulp and paper industry e a model-based assessment of savingspotential. Energy 2012;40(1):84e99.

[16] Greenpeace. The Eskom factor: power politics and the electricity sector inSouth Africa. The Electricity Governance Complex; 2012.

[17] Haji S, Haji H. A dynamic model of industrial energy demand in Kenya. EnergyJ 1994;15(4):203e24.

[18] He YX, Yang LF, He HY, Luo T, Wang YJ. Electricity demand price elasticity inChina based on computable general equilibrium model analysis. Energy2010;36:1115e23.

[19] Human Sciences Research Council. The impact of electricity price increasesand rationing on the South African economy; 2008.

[20] Inglesi-Lotz R, Blignaut JN. Electricity intensities of the OECD and South Africa:a comparison. Working Paper 204-University of Pretoria, NA; 2011.

[21] Inglesi-Lotz R, Blignaut J. Estimating the price elasticity of demand for elec-tricity by sector in South Africa. South Af J Econ Manag Sci 2011b;14:449e65.

[22] Inglesi-Lotz R, Pouris A. Energy efficiency in South Africa: a decompositionexercise. Energy 2012;42:113e20.

[23] International Energy Agency (IEA). Energy Balances for non-OECD countries;2012.

[24] Kamerschen DR, Porter DV. The demand for residential, industrial and totalelectricity, 1973e1998. Energy Econ 2004;26(1):87e100.

[25] Kraft J, Kraft A. On the relationship between energy and GNP. J Energy Dev1978;3:401e3.

[26] Lam P. Pricing of electricity in China. Energy 2001;29:287e300.[27] Levine MD, Koomey JG, Price L, Geller H, Nadel S. Electricity end-use effi-

ciency: experience with technologies, markets, and policies throughout theworld. Energy 1995;20(1):37e61.

[28] Lin B, Liu J. Principles, effects and problems of differential power pricing policyfor energy. Energy 2011;36(1):111e8.

[29] Madlener R, Bernstein R, Gonzalez M. Econometric estimation of energy de-mand elasticities. E.ON Energy Res Center Series 2011;3(8).

[30] Mukherjee K. Energy use efficiency in U.S. manufacturing: a nonparametricanalysis. Energy Econ 2008;30(1):76e96.

[31] Nilsson LJ. Energy intensity trends in 31 industrial and developing countries1950-1988. Energy 1993;18(4):309e22.

[32] OECD. OECD economic surveys: South Africa 2013. OECD Publishing; 2013.[33] Pesaran P, Pesaran MH. Time series econometrics using Microfit 5.0. Oxford:

Oxford University Press; 2009.

M. Kohler / Energy 64 (2014) 524e532532

[34] Pesaran MH, Shin Y. An autoregressive distributed-lag modelling approach tocointegration analysis. In: Strom S, editor. Econometrics and economic theoryin the 20th century. Cambridge: Cambridge University Press; 1999.

[35] Pesaran MH, Shin Y, Smith RJ. Bounds testing approaches to the analysis oflevel relationships. J Appl Econ 2001;16:289e326.

[36] Pesaran MH, Smith RP, Akiyama T. Energy demand in Asian developingeconomies. Oxford: Oxford University Press; 1998.

[37] Polemis ML. Modeling industrial energy demand in Greece using cointegra-tion techniques. Energy Policy 2007;35(8):4039e50.

[38] Price L, Wang X, Yun J. The challenge of reducing energy consumption of theTop-1000 largest industrial enterprises in China. Energy Policy 2010;38:6485e98.

[39] Quantec, (various issues). Quantec standardised industry database. Pretoria:Quantec.

[40] Sebitosi AB. Energy efficiency, security of supply and the environment inSouth Africa: moving beyond the strategy documents. Energy 2008;33(11):1591e6.

[41] Siddayao CM. Pricing policy and efficient energy use. Energy 1983;8(1e2):45e68.

[42] Statistics SA (various issues). Consumer price index. Pretoria: StatsSA.[43] Stern D. Interfuel substitution: a meta-analysis. J Econ Stud 2012;26(2):

307e31.[44] Sun JW, Ang BW. Some properties of an exact energy decomposition model.

Energy 2000;25(12):1177e88.[45] Taylor LD. The demand for electricity: a survey. Bell J Econ 1975;6(1):74e110.[46] World Bank. Eskom power investment support project fact Sheet; 2012.[47] Wang J, Bloyd CN, Hu Z, Tan Z. Demand response in China. Energy 2010;35:

1592e7.[48] Winkler H, Marquard A. Changing development paths: from an energy-

intensive to low-carbon economy in South Africa. Clim Dev 2009;1:47e65.