9

Click here to load reader

Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

Embed Size (px)

Citation preview

Page 1: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

Energy Economics 32 (2010) 1374–1382

Contents lists available at ScienceDirect

Energy Economics

j ourna l homepage: www.e lsev ie r.com/ locate /eneco

Energy consumption, pollutant emissions and economic growth in South Africa

Kojo Menyah a,⁎, Yemane Wolde-Rufael b

a London Metropolitan Business School, London Metropolitan University, 84 Moorgate, London EC2M 6SQ, United Kingdomb Independent Researcher, United Kingdom

⁎ Corresponding author.E-mail addresses: [email protected] (K. M

(Y. Wolde-Rufael).

0140-9883/$ – see front matter © 2010 Elsevier B.V. Adoi:10.1016/j.eneco.2010.08.002

a b s t r a c t

a r t i c l e i n f o

Article history:Received 11 November 2009Received in revised form 2 August 2010Accepted 14 August 2010Available online 21 August 2010

JEL classification:C32O55Q20Q43

Keywords:Carbon dioxide emissionEconomic growthEnergy consumptionBounds testCausality

This paper examines the long-run and the causal relationship between economic growth, pollutantemissions and energy consumption for South Africa for the period 1965–2006 in a multivariate frameworkwhich includes labour and capital as additional variables. Using the bound test approach to cointegration, wefound a short-run as well as a long-run relationship among the variables with a positive and a statisticallysignificant relationship between pollutant emissions and economic growth. Further, applying a modifiedversion of the Granger causality test we also found a unidirectional causality running from pollutantemissions to economic growth; from energy consumption to economic growth and from energyconsumption to CO2 emissions all without a feedback. The econometric evidence suggests that South Africahas to sacrifice economic growth or reduce its energy consumption per unit of output or both in order toreduce pollutant emissions. In the long-run however, it is possible to meet the energy needs of the countryand at the same time reduce CO2 emissions by developing energy alternatives to coal, the main source of CO2

emissions. However, the econometric results upon which the policy suggestions are made should beinterpreted with care, as they may not be sufficiently robust enough to categorically warrant the choice of anunpalatable policy option by South Africa.

enyah), [email protected]

ll rights reserved.

© 2010 Elsevier B.V. All rights reserved.

1. Introduction

Like many coal-abundant countries, South Africa is facing a crucialpolicy dilemma relating to the use of coal vis-à-vis the development ofother sources of energy (see Büscher, 2009; Winkler, 2007). It isconfronted with the crucial issue of producing more coal to meet itsenergy requirements,while at the same time grapplingwith the issue ofreducing greenhouse gas (GHG) emissions. The issue is furthercomplicated by concerns over increases in the price of coal relative toother energy sources and coal reserves exhaustion. These issues areforcing South Africa to define an energy strategy that departs fromover-reliance on coal-generated electricity (Department of Energy andMinerals, South Africa, 2008). Even nuclear energy, despite itscontroversies and costs (see Apergis et al., 2010; Menyah and Wolde-Rufael, 2010; Wolde-Rufael and Menyah, 2010), is being promoted asone of the important energy sources of the future. However, critics ofthe South African energy strategy point out that despite the hugeenvironmental challenges it faces, the country still has a weakenvironmental policy that is not addressing the issue of greenhouseemissions adequately. According to the critics, policymakers see thecutting of GHG emissions as a “benevolent gesture towards mankind”

and not as a serious means for solving the environmental challengesfacing South Africa (Sebitosi and Pillay, 2008a).

What makes South Africa an interesting case study is that itseconomy is heavily dependent on the energy sectorwhich accounts for15% of the country's GDP with coal being the dominant one(Department of Energy and Minerals, South Africa, 2008). About 70%of South Africa's total primary energy supply is derived from coal, andcoal-fired power stations provide more than 93% of electricityproduction (World Bank, 2008). This over dependence on coal isleading to high levels of CO2 emissions relative to the size of the SouthAfrican economy and population (Winkler, 2007). In terms ofpollutants, the coal sector accounts for 87% of CO2 emissions, 96% ofsulphur dioxide (SO2) emissions, and 94% of nitrous oxide emissions.As Table 1 indicates, South Africa is one of the highest emitters of GHGemissions when compared to many developed and developingcountries, whether this is measured in emissions per person or perunit of GDP (Winkler, 2007;World Bank, 2008). South Africa ranks the7th largest emitter of GHG emissions per capita in the world (Sebitosiand Pillay, 2008a) and has experienced almost a 7-fold increase infossil-fuel CO2 emissions since 1950,with 80–90% of emissions comingfrom coal (Winkler, 2007).

Between 1980 and 2006, South Africa's per capita carbon dioxideemissions from the consumption and flaring of fossil fuels increasedcontinuously while they were falling for other major regions of theworld (see Fig. 1). To aggravate the problem further, the growth in

Page 2: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

Table 1Energy efficiency indicators, 2005.Source: World Bank, World Development Indicators, 2008.

CO2 emissions(kg per 2000 US $ of GDP)

CO2 emissions(kg per 2000 PPP $ of GDP)

CO2 emissions(metric tons per capita)

GDP per unit of energy use(PPP $ per kg of oil equivalent)

South Africa 2.54 1.03 8.72 3.12East Asia and Pacific 2.63 0.92 3.59 3.33Europe and Central Asia 2.65 0.74 7.01 3.32Euro area 0.38 0.27 8.07 7.45High income: OECD 0.43 0.37 12.45 6.13Latin America and Caribbean 0.59 0.28 2.49 7.19South Asia 1.92 0.52 1.08 4.64Sub-Saharan Africa 1.53 0.49 0.85 2.91World 0.80 0.52 4.53 5.04

1375K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

coal use is expected to continue or even accelerate over the next fewyears as South Africa is building more coal-powered stations to meetits growing energy needs and also to redress the past inequitabledistribution of electricity use among its races (Winkler, 2007; Sebitosiand Pillay, 2008a).

South Africa's energy consumption per capita and CO2 emissionsper capita have been growing faster than real GDP per capita andthere is no doubt that the current emission profile poses a significantchallenge to the country's energy development strategy (Fig. 2).

Studies that link economic growth, energy consumption andpollutant emissions in the same framework tend not to focus onSouth Africa. Most of the studies that relate to South Africa onlyinvestigate the link between energy consumption and economicgrowth (see, Ziramba, 2009; Odhiambo, 2009; Wolde-Rufael, 2006,2009). To our knowledge, there are no specific studies for South Africathat have employed modern advances in time series econometrics ofcointegration and causality to test the relationship between energyconsumption, pollutant emissions and economic growth in a coherentframework. The aim of this paper is to fill this gap by investigating therelationship between economic growth, energy consumption andpollutant emissions in a multivariate framework by including labourand capital as additional variables. The purpose is to show howenvironmental degradation and other crucial variables such as energy,combine with capital and labour to affect the growth process. From aneconometric point of view, we include these additional variablesbecause the exclusion of relevant variables makes the estimates notonly biased but also inconsistent, but the absence of causality in abivariate system can arise from omitted variables (Lütkepohl, 1982).However, since a five-variable case incorporates more informationthan a bivariate case, the causal inference drawn may be relativelymore reliable (Loizides and Vamvoukas, 2005).

In this paper, tests for establishing the long-run relationshipbetween the variables are carried out by using the cointegrationprocedure developed by Pesaran et al. (2001), hereafter PSS, whiletesting for causality is conducted using a modified version of the

40%

60%

80%

100%

120%

140%

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

Euroasia

South Africa

Europe

North America

Fig. 1. Trends of variables (1980=100). Per capita carbon dioxide emissions from theconsumption and flaring of fossil fuels, 1980–2006.

Granger causality test proposed by Toda and Yamamoto (1995),hereafter, TY which is valid regardless whether a series is I(0), I(1)or I(2), non-cointegrated or cointegrated of any arbitrary order. Toreinforce our empirical findings, we also quantify how much feedbackexists from one series to the other using the generalized forecast errorvariance decomposition technique proposed by Pesaran and Shin(1998), which is invariant to the ordering of the variables.

The rest of the paper is structured as follows. In Section 2we presenta brief review of the empirical literature followed by a discussion of themethodology used in Section 3. The empirical evidence is presented inSection 4 while the summary and the concluding remarks are outlinedin Section 5.

2. An overview of the literature

While coal is an important source of energy that raises relativelyfewer security concerns than do oil and natural gas, many believe thatcoal consumption is the major source of global warming as powerplants that burn coal are the major contributors to rising atmosphericconcentration of greenhouse gas emissions (see Wolde-Rufael, 2010).As a result of this, global warming has become an important globalenvironmental challenge facing the world, including South Africa.Primarily motivated by this concern, there have been several studiesthat have attempted to investigate the causal relationship betweenpollutant emissions and economic growth (see, Aslanidis, 2009;Galeotti et al., 2009). However, as Ang (2008) rightly argues, eventhough the relationship between output and pollution has beenextensively studied, most of these studies mainly focus on testing thevalidity of the so-called Environmental Kuznet's Curve (see, Aslanidis,2009; Galeotti et al., 2009) and do not consider investigating the causalrelationship between energy consumption, pollutant emissions andeconomic growth in the same framework. Since fossil-fuel energy useis the main source of global warming, incorporating energy consump-tion andother growthdetermining factors such as labour and capital inthe same growth accounting framework can enhance our understand-ing of the issues that can affect global warming.

60%70%80%90%

100%110%120%130%140%150%

1971

1974

1977

1980

1983

1986

1989

1992

1995

1998

2001

2004

Inde

x (1

971=

100)

Real GDP per capita

per capita energy consumption

per capita CO2 emissions

Fig. 2. Trends of variables (1971=100).

Page 3: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

1376 K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

Recently, few studies have attempted to investigate the causalrelationship between pollutant emissions, energy consumption andeconomic growth but there seems to be no consensus regarding thedirection of causality. For instance, in the case of Malaysia, Ang (2008)found that pollution and energy use were positively related to outputin the long-run with strong support for causality running fromeconomic growth to energy consumption, both in the short-run andlong-run. In the case of China, Ang (2009) found that more energy use,higher income and greater trade openness tend to cause more CO2

emissions. In a multivariate causality study for China, Zhang andCheng (2009) found a unidirectional Granger causality running fromGDP to energy consumption, and a unidirectional Granger causalityrunning from energy consumption to carbon emissions in the long-run but neither carbon emissions nor energy consumption leadseconomic growth. In contrast, Jalil and Mahmud (2009) found thatcarbon dioxide emissions were mainly determined by income andenergy consumption in the long-run. For Turkey, Soytas and Sari(2009) found that carbon emissions seem to Granger cause energyconsumption, but the reverse was not true. In Nigeria and Venezuela,Sari and Soytas (2009) found a unidirectional causality running fromeconomic growth to CO2. In contrast to the findings of Soytas and Sari(2009); Halicioglu (2009) found that there was a bi-directionalGranger causality (both in short- and long-run) running betweencarbon emissions and income in Turkey. In the case of the USA, Soytaset al. (2007) found that income does not Granger cause carbonemissions in the short-run but they found that there was a long-runcausal relationship between energy use and carbon emissions. For agroup of South American countries Apergis and Payne (2009) usingpanel cointegration and panel causality tests found that energy usehad a positive and a statistically significant impact on emissionswhereenergy consumption and economic growth cause emissions in theshort-run. In the long-run, Apergis and Payne (2009) also found thatthere was evidence of a feedback between energy consumption andemissions but no feedbackbetween real output andpollutant emissions.Further, Apergis and Payne (2010) for a group of Commonwealth ofIndependent States found that both energy consumption and economicgrowth cause carbon dioxide emissions in the short-run. In the long-runthere appears to be bi-directional causality running between energyconsumption and carbon dioxide emissions.

Concerning South Africa, there is no empirical evidence thatinvestigates the relationship between output growth and CO2

emissions using modern advances in time series econometrics ofcointegration and causality. However there are few studies that relateeconomic growth to energy consumption. For instance, Odhiambo(2009) found a bi-directional causality running between electricityconsumption and economic growth in South Africa. Ziramba (2009)also found a bi-directional causality between oil consumption andindustrial production but for other forms of energy consumption,there was evidence in support of the energy neutrality hypothesis.Wolde-Rufael (2009) also found a unidirectional causality runningfrom energy consumption to economic growth.

The above conflicting evidence has major implications for abatingCO2 emissions and economic growth. If there is a unidirectionalGranger causality running from CO2 emissions to economic growth,where increases in CO2 emissions lead to increases in economicgrowth, then an energy policy that decreases CO2 emissions could leadto a fall in economic growth. This may imply that economic growthcould be sacrificed in order to reduce CO2 emissions. On the otherhand, if causality runs from economic growth to CO2 emissions, whereincreases in economic growth cause increases in CO2 emissions, anenergy strategy that reduces CO2 emissions may not have a negativeimpact on economic growth. It may be possible to reduce CO2

emissions without harming economic growth. On the other hand, ifthere is no causality running in any direction, the neutrality hypothesisis accepted, and reducing CO2 emissions may not affect income andCO2 emission reduction policies may not affect economic growth. In

contrast, if there is a bi-directional causality running between the two,and if economic growth leads to more CO2 emissions, then this mayincrease the degradation of the environment.

3. Methodology and data

Despite the fact that it has been recognised that the interrelation-ships between environmental pollution, capital accumulation andother growth variables are of central importance in growth theory(Xepapadeas, 2005, p. 1221), there are not many studies that haveinvestigated the causal relationship between economic growth,energy consumption and pollutant emissions which also includelabour and capital in the analysis. This lack of empirical investigationis particularly true for South Africa. Recently however, a few studieshave attempted to highlight the importance of both energy andpollutant emissions as additional variables to capital and labour in thegrowth process for some countries (see, Ang, 2008, 2009; Sari andSoytas, 2007, 2009; Soytas and Sari, 2009; Apergis and Payne, 2009;and Zhang and Cheng, 2009). In this paper, following some of theseauthors, we employ the autoregressive distributive lag (ARDL)approach to cointegration test developed by PSS and the Grangercausality test proposed by TY to investigate the long-run and thecausal relationship between energy consumption, CO2 emissions,capital and labour for South Africa for the period 1965–2006.

3.1. Bounds test approach to cointegration

The bounds test approach to cointegration has certain econometricadvantages in comparison to other single equation cointegrationprocedures. As pointed out by Emran et al. (2007), the bounds testapproach to cointegration is preferred to other conventional coin-tegration tests becauseMonte Carlo evidence shows that it has severalimportant advantages over other conventional tests. The approacheffectively corrects for a possible endogeneity of explanatory variablesand the estimates derived from the approach exhibit desirable smallsample properties. Another important advantage of the ARDLapproach is that one can avoid the uncertainties created by unitroot pre-testing as the test can be applied regardless of whether theseries are I(0) or I(1). An added bonus of this approach is that unlikeother conventional tests for cointegration, it can be applied to studiesthat have a small sample size (Narayan, 2005). In addition, both theshort- and the long-run relationship can be simultaneously estimated.

In this paper the ARDL approach to cointegration is estimatedusing the following unrestricted error correction (UREC) regressions:

Δ lnYt = α1 + ∑ρ

i=1β1iΔ lnYt−i + ∑

ρ

i=0κ1iΔ lnCt−i + ∑

ρ

i=0ϖ1iΔ ln Et−i

+ ∑ρ

i=0σ1i lnΔKt−i + ∑

ρ

i=0ξ1i lnΔLt−i + η1Y lnYt−1

+ η2Y lnCt−1 + η3Y lnEt−1 + η4Y lnKt−1 + η5Y lnLt−1

+ μ1t

ð1Þ

Δ lnCt = α2 + ∑ρ

i=0β2iΔ lnYt−i + ∑

ρ

i=1κ2iΔ lnCt−i + ∑

ρ

i=0ϖ2iΔ ln Et−i

+ ∑ρ

i=0σ2i lnΔKt−i + ∑

ρ

i=0ξ2i lnΔLt−i + η1C lnYt−1

+ η2C lnCt−1 + η3C lnEt−1 + η4C lnKt−1 + η5C lnLt−1

+ μ2t

ð2Þ

Page 4: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

1377K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

Δ ln Et = α3 + ∑ρ

i=0β3iΔ lnYt−i + ∑

ρ

i=0κ3iΔ lnCt−i + ∑

ρ

i=1ϖ3iΔ ln Et−i

+ ∑ρ

i=0σ3i lnΔKt−i + ∑

ρ

i=0ξ3i lnΔLt−i + η1E lnYt−1

+ η2E lnCt−1 + η3E lnEt−1 + η4E lnKt−1 + η5E lnLt−1

+ μ3t

ð3Þ

Δ lnKt = α4 + ∑ρ

i=0β4iΔ lnYt−i + ∑

ρ

i=0κ4iΔ lnCt−i + ∑

ρ

i=0ϖ4iΔ ln Et−i

+ ∑ρ

i=1σ4i lnΔKt−i + ∑

ρ

i=0ξ4i lnΔLt−i + η1K lnYt−1

+ η2K lnCt−1 + η3K lnEt−1 + η4K lnKt−1 + η5K lnLt−1

+ μ4t

ð4Þ

Δ ln Lt = α5 + ∑ρ

i=0β5iΔ lnYt−i + ∑

ρ

i=0κ5iΔ lnCt−i + ∑

ρ

i=0ϖ5iΔ ln Et−i

+ ∑ρ

i=0σ5i lnΔKt−i + ∑

ρ

i=1ξ5i lnΔLt−i + η1L lnYt−1

+ η2L lnCt−1 + η3L lnEt−1 + η4L lnKt−1 + η5L lnLt−1

+ μ5t

ð5Þ

where lnYt is the log of real GDP (proxy for economic development),lnCt is the log of CO2 emissions (measured in million tons of oilequivalent), lnEt is the log of energy consumption (measured inmillion tons of oil equivalent), lnKt is the log of real gross fixed capitalformation and lnLt is the log of employment. All the data are annualfor 1965–2006. Real GDP and gross fixed capital formation are at 2000constant prices. With the exception of the employment variable,which is from the Conference Board and Groningen Growth andDevelopment Centre (2008), the rest of the data was obtained fromthe World Bank (2008). In line with many researchers, in the absenceof physical capital stock, gross fixed capital formation has been used asproxy for the stock of physical capital (Sari and Soytas, 2007). Usinggross fixed capital formation instead of the stock of gross physicalcapital has its own limitations. Gross capital formation as a flowvariable, does not measure the stock of capital accumulated over theyears and it does not take into account any adjustments for thedepreciation of assets. Aggregate investment is merely a change inaggregate physical capital stock, less depreciation. Gross capitalformation also includes stocks. Nevertheless, Sari and Soytas (2007)point out that since the perpetual inventory method (PIM) ofestimating the physical capital stock assumes a constant depreciationrate, any variance in capital is mostly related to a change ininvestment. Thus, it is possible to obtain a fairly reliable measure ofthe trend in physical capital stock from new fixed investment data(Sari and Soytas, 2007; Lee et al., 2008).

In Eq. (1) tests for cointegration are carried out by testing the jointsignificance of the lagged levels of the variables using the F-test wherethe null of no cointegration is defined by HO: η1Y=η2Y=η3Y=η4Y=η5Y=0 against the alternative that H1: η1Y≠0, η2Y≠0, η3Y≠0,η4Y≠0, η5Y≠0. Similarly, in Eq. (2) we test for the joint significanceof the lagged levels of the variables using the F-test where the null ofno cointegration is defined by HO: η1C=η2C=η3C=η4C=η5C=0against the alternative that H1: η1C≠0, η2C≠0, η3C≠0, η4C≠0,η5C≠0. Other tests for cointegration using energy, capital and labour

in Eqs. (3), (4) and (5) can also be carried out using similarprocedures.

The asymptotic distribution of the F-statistic is non-standardunder the null and it was originally derived and tabulated by PSS butlatermodified by Narayan (2005) to accommodate small sample sizes.Two sets of critical values are provided: one, which is appropriatewhen all the series are I(0) and the other for all the series that are I(1).According to PSS, in testing for cointegration, if the computed F-statistic falls above the upper critical bounds, a conclusive inferencecan be made regarding cointegration without the need to knowwhether the series were I(0) or I(1). In this case, the null of nocointegration is rejected regardless of whether the series are I(0) or I(1). Alternatively, when the test statistic falls below the lowercritical value, the null hypothesis is not rejected regardless ofwhether the series are I(0) or I(1). In contrast, if the computed teststatistic falls inside the lower and the upper bounds, a conclusiveinference cannot be made unless we know whether the series wereI(0) or I(1).

3.2. Granger non-causality test

To complement the above discussion, we have also carried outGranger non-causality test using the TY procedure which is validregardless of whether a series is I(0), I(1) or I(2), not-cointegrated orcointegrated of any arbitrary order. The novelty of the TY procedure isthat it does not require pre-testing for the cointegrating properties ofthe system and thus avoids the potential bias associatedwith unit rootand cointegration tests as it can be applied regardless of whether aseries is I(0), I(1) or I(2), not-cointegrated or cointegrated of anarbitrary order (see Rambaldi and Doran, 1996). As has been pointedout by Clarke and Mirza (2006) pre-tests for unit root andcointegration might suffer from size distortions, which often implythe use of an inaccurate model for the non-causality test. To obviatesome of these problems, TY, based on augmented VAR modelling,introduced a Wald test statistic that asymptotically has a chi square(χ2) distribution irrespective of the order of integration or cointegra-tion properties of the variables. The TY approach fits a standard vectorauto-regression model on levels of the variables (not on their firstdifferences) and therefore makes allowance for the long-run informa-tion often ignored in systems that require first differencing and pre-whitening (Clarke and Mirza, 2006). The approach employs a modifiedWald test (MWALD) for restrictions on the parameters of the VAR (k)where k is the lag length of the system. The basic idea of the TY approachis to artificially augment the correct order, k, by the maximal order ofintegration, say dmax. Once this is done, a (k+dmax)th order of VAR isestimated and the coefficients of the last lagged dmax vectors are ignored(Caporale and Pittis, 1999).

To undertake the TY version of the Granger non-causality test, forVAR with 4 lags, (k=3 and dmax=1), we estimate the followingsystem of equations:

ln Ytln Ct

ln Etln Kt

ln Lt

26666664

37777775= A0 + A1

ln Yt−1

ln Ct−1

ln Et−1

ln Kt−1

ln Lt−1

26666664

37777775+ A2

ln Yt−2

ln Ct−2

ln Et−2

ln Kt−2

ln Lt−2

26666664

37777775+ A3

ln Yt−3

ln Ct−3

ln Et−3

ln Kt−3

ln Lt−3

26666664

37777775

+ A4

ln Yt−4

ln Ct−4

ln Et−4

ln Kt−4

ln Lt−4

266666664

377777775+

εln Yt

εln Ct

εln Et

εln Kt

εln Lt

26666664

37777775:

ð6Þ

Page 5: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

Table 2Misspecification test for the optimal lag selected.

Test statistics LM version F version

A: Serial correlation CHSQ(4)=7.794 [0.099]⁎ F(4,7)=0.467 [0.759]B: Functional form CHSQ(1)=0.168 [0.280] F(1,10)=0.326 [0.581]C: Normality CHSQ(2)=0.754 [0.686] Not applicableD: Heteroscedasticity CHSQ(1)=0.929 [0.335] F(1,35)=0.902 [0.349]E: Autoregressiveconditionalheteroscedasticity

CHSQ(4)=1.027 [0.906] F(4,7)=0.0500 [0.994]

A: Lagrange multiplier test of residual serial correlation.B: Ramsey's RESET test using the square of the fitted values.C: Based on a test of skewness and kurtosis of residuals.D: Based on the regression of squared residuals on squared fitted values.⁎Significant at 10%.

Table 4Estimated long-run coefficients, dependent variable, lnY.

Regressors Coefficient Standard error t-Ratio [probability]

CO2 (lnC) 1.19 0.30 3.97 [0.00]⁎⁎⁎

Energy (lnE) −0.70 0.27 −2.64 [0.01]⁎⁎⁎

Capital (lnK) 0.31 0.03 10.69 [0.00]⁎⁎⁎

Labour (lnL) 0.11 0.09 1.23 [0.23]Constant 6.64 2.82 2.44 [0.02]⁎⁎

Notes: ⁎⁎⁎ and ⁎⁎ denote significant levels at 1% and 5% respectively.

1378 K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

In Eq. (6), A1…A4 are five 5×5 matrices of coefficients with A0

being the 5×1 identity matrix, εs are the disturbance terms with zeromean and constant variance. From Eq. (6) we can test the hypothesisthat CO2 (lnCt) does not Granger cause economic growth (lnYt), withthe following hypothesis: Ho=a12

1 =a122 =a12

3 =0, where a12'si are the

coefficients of the CO2 emission variable in the first equation of thesystem presented in Eq. (6). Additionally, we can test the oppositenon-causality from economic growth (lnYt) to CO2 emissions (lnCt) inthe following hypothesis: Ho=a21

1 =a212 =a21

3 =0, where a21'si are the

coefficients of the economic growth variable in the second equation ofthe system presented in Eq. (6). Similar tests for testing causalitybetween output, CO2 emissions, energy consumption, capital andlabour can also be carried out using similar procedures.

4. Empirical results

4.1. Cointegration test

Tests for cointegration require that the variables under consider-ation are not integrated at an order higher than one. In the presence of I(2) or higher variables, the computed statistics provided by PSS andNarayan (2005) are not valid (Ang, 2007). Recognising that there aremany controversies surrounding unit root testing, we conduct severalunit root tests where our strategy was to compare results obtainedfrom several of these tests and examinewhether the preponderance ofthe evidence makes a convincing case for stationarity or non-stationarity of the variables used in this paper. Apart from theconventional unit root tests, since our time series may experiencestructural breaks thatmay affect cointegration and non-causality tests,we carried out three unit tests with structural breaks: the Perron(1997); Harvey et al. (2003) and the Zivot and Andrews (1992) tests.Results of these tests (which are not reported here to conserve space),show that the CO2 and energy consumption series were I(0) while theremaining series were I(1). Since our series were not integrated of thesame order, PSS is the most appropriate approach for testing for

Table 3F-statistic results of the joint null hypothesis that the coefficients of the levels of thelagged independent variables are zero.

F-statistic Lags

1 2 3 4

FlnY(FY|lnC,lnE,LnK,lnL) 2.198 2.090 3.958⁎ 4.026⁎,+

FlnC(FE|lnE,lnK,LnL,lnY) 1.566+ 1.532 1.427 1.962FlnE(FK|lnC,lnL,LnK,lnY) 3.718+ 3.609 2.552 3.469FlnK(FY|lnC,lnE,LnL,lnY) 3.145 3.157+ 4.922⁎ 1.707FlnL(FY|lnC,lnE,LnK,lnY) 1.660+ 1.699 3.032 3.345

Notes: + denotes optimum lag; ⁎ denotes rejection of the null hypothesis of nocointegration at 10% significant level. The upper bound critical value for 40 observationsis 3.838 (Narayan, 2005, p. 1988).

cointegration while the TY approach is themost appropriate approachfor testing for causality.

In addition to ensuring the order of integration of our series, it isalso important to ascertain that the optimal lag order ρ, of theunderlying Eqs. (1)–(5) is chosen appropriately so that the distur-bance terms in each of Eqs. (1)–(5) are not serially correlated.Consequently, the lag order should be high enough to reduce theresidual serial correlation problems but at the same time it should belowenough so that the conditional ECM(error correctionmodel) is notsubject to over-parameterisation problems (Ang, 2007; Narayan,2005; and PSS). We selected our optimum lag by the AIC (AkaikeInformation Criteria) and we tested the reliability of our models byapplying a number of diagnostic tests, including tests of autocorrela-tion, autoregressive conditional heteroscedasticity (ARCH), normalityand heteroscedasticity. In general, as can be seen from Table 2, wefound no evidence of serious violation of all the above tests.

The results of our cointegration tests are presented in Table 3. As isevident from the table, there is a long-run cointegrating relationshipamong the series under consideration. The calculated F-statistic of4.026 denoted by FlnY(FY|lnC,lnE,LnK,lnL), is higher than the upperbound critical value of 3.838 at 10% as tabulated in Narayan (2005) for40 observations. This shows that the null hypothesis of no cointegra-tion among output, CO2 emissions, energy consumption, employmentand capital is rejected. In contrast, no cointegrating relationship wasfound when the other series (CO2 emissions, energy, capital andlabour) were used as independent variables. For results using theseindependent variables, see Table 3.

Since our results support the existence of cointegration among theseries, we estimated the long-run coefficients. Table 4 shows that thesign of the CO2 coefficient is positive and statistically significantindicating that higher CO2 emissions promote economic growth.Specifically, a 1% increase in CO2 leads to 1.2% increase in economicgrowth, while a 1% increase in capital increases output by only 0.3%.Labour is not significantly related to output. In contrast, the sign of thecoefficient of the energy variable is negative.

Whilst this result is counter-intuitive from an economic point ofview, it is by no means unique. A negative long-run relationshipbetween energy consumption and GDP has been reported for Nigeria(Wolde-Rufael, 2005, 2006), Saudi Arabia and UAE (Squalli, 2007),Gabon (Wolde-Rufael, 2006) and for the US between industrialprimary energy consumption and GDP (Bowden and Payne, 2009).

Table 5Error correction representation of the selected ARDL model.

Regressors Coefficient Standard error t-Ratio [probability]

ΔlnCt−1 0.23 0.09 2.63 [0.01]⁎⁎⁎

ΔlnCt−2 −0.25 0.07 −3.57 [0.00]⁎⁎⁎

ΔlnEt−1 −0.17 0.09 −1.78 [0.09]⁎

ΔlnKt−1 0.22 0.04 5.41 [0.00]⁎⁎⁎

ΔlnKt−2 −0.04 0.03 −1.22 [0.23]ΔlnLt−1 −2.08 0.45 −4.06 [0.00]⁎⁎⁎

ΔlnLt−2 −1.15 0.56 −2.07 [0.05]⁎⁎

Constant 3.06 1.69 1.81 [0.08]⁎

ECM(−1) −0.46 0.10 −4.49 [0.00]⁎⁎⁎

Notes: R2=0.87; adjusted R2=0.78; F-stat. (8,31) 18.71 [0.00]⁎⁎⁎; DW-statistic 2.44.Notes: ⁎⁎⁎, ⁎⁎ and ⁎ denote significant levels at 1%, 5% and 10% respectively.

Page 6: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

Table 6VAR diagnostic tests.

Variable Lagrange test of serial correlation CHSQ(4):A ARCH CHSQ(4) Ramsey's RESET CHSQ(4): B Normality CHSQ(2): C Heteroscedasticity CHSQ(1): D

YY 4.687 [0.321] 1.308 [0.860] None 0.647 [0.968] 1.337 [0.248]CC 2.871 [0.580] 2.871 [0.580] 1.600 [0.206] 0.952 [0.621] 0.818 [0.366]EE 8.087 [0.088]⁎ 8.076 [0.089]⁎ 0.607 [0.436] 1.094 [0.579] 0.116 [0.734]KK 9.473 [0.050]⁎⁎ 2.605 [0.626] NONE 0.697 [0.706] 0.170 [0.680]LL 19.945 [0.001]⁎⁎⁎ 11.500 [0.022] 3.207 [0.073]⁎ 1.332 [0.514] 1.497 [0.221]

A: Lagrange multiplier test of residual serial correlation.B: Ramsey's RESET test using the square of the fitted values.C: Based on a test of skewness and kurtosis of residuals.D: Based on the regression of squared residuals on squared fitted values.ARCH: Autoregressive conditional heteroscedasticity test of residuals.⁎⁎⁎Significant at 1%; ⁎⁎significant at 5%; ⁎significant at 10%.

Table 7Granger causality test.

Null hypothesis χ2 ρ-value Σ of lagged coefficients

CO2 does not cause GDP 15.99 0.00⁎⁎⁎ CO2=0.59GDP does not cause CO2 1.06 0.79 GDP=−0.38Energy does not cause GDP 23.46 0.00⁎⁎⁎ Energy=−0.44GDP does not cause energy 1.31 0.73 GDP=0.19CO2 does not cause energy 5.70 0.13 CO2=0.78Energy does not cause CO2 10.33 0.02⁎⁎ GDP=0.20Capital does not cause GDP 7.49 0.06⁎ Capital=0.50

1379K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

The question is what might explain this unusual finding for SouthAfrica. We suggest a number of factors that could have contributed tothis. As pointed out in previous studies, the negative relationship mayreflect the possibility of inefficient and excessive energy consumptionin South Africa. It is not uncommon for some developing countriesthat are endowed with abundant energy resources to experienceinefficiency in their energy use (Bowden and Payne, 2009; Payne andTaylor, 2010; Squalli, 2007; Wolde-Rufael, 2009). In particular, SouthAfrica's past energy strategy based on vast, inexpensive supplies ofcoal and large public investments coupled with the apartheid era'ssingle-minded emphasis on energy security could have also contrib-uted to inefficient use of energy during this period (Spalding-Fecheret al., 2000). The apartheid era historically favoured supply-sideactions, rather than encouraging more efficient use of energy(Spalding-Fecher et al., 2000). Indeed, there is a real concern aboutenergy inefficiency in South Africa and there are plans to reduceenergy inefficiency by 12% by 2014 (Winkler, 2007). According toWinkler (2007), over the period 2006–2024, energy efficiency isprojected to make the greatest impact when seen against indicators ofsustainable development in South Africa and that energy efficiencywill be critical in making South Africa's energy development moresustainable. It has also been suggested that growing economies tendto move to the production of less energy intensive sectors (Squalli,2007). In the case of South Africa, there is an indication that energydemand is shifting towards the service sector where it requires lessenergy per unit of output (Winkler, 2007). It should, however, bepointed out that the possible causes of the negative coefficient for theenergy variable outlined above is not exhaustive. However, a detailedconsideration of this issue is beyond the scope of this paper.1

The short-run tests are presented in Table 5. As the table shows theerror correction term (ECT) is statistically significant and has thecorrect sign indicating that there was a long-run cointegratingrelationship among output, energy consumption, CO2 emissions,capital and labour. The speed of adjustment towards the long-runequilibrium is reasonably high with almost 46% of the disequilibriumcorrected in the first year. Therefore, both in the short- and in thelong-run, CO2 emissions have a statistically significant positive impacton output.

4.2. Granger causality test

In order to avoid spurious causality or spurious absence ofcausality, apart from determining the order of integration of theseries (dmax), it is also important to determine the optimal lag lengthk, in Eq. 6. Granger causality test is very sensitive to the selection ofthe lag length. If the chosen lag length is less than the true lag length,the omission of relevant lags can cause bias. On the other hand, if the

1 As also pointed out by one of the referees to this Journal, the negative coefficienton the energy variable is incompatible with production theory or what one mightexpect intuitively. An exhaustive investigation of this issue could therefore be thesubject of further research.

chosen lag length is more than required, the irrelevant lags in theequation cause the estimates to be inefficient (Clarke and Mirza,2006). It is therefore important to determine the optimum lag. Inselecting the optimal lag length, we followed Lütkepohl's (1993)procedure where he suggests linking the lag length (mlag) andnumber of endogenous variables in the system (m) to the sample size(T) according to the formula,m*mlag=T1/3 (Kónya, 2004). FollowingHatemi-J and Irandoust (2000) a combination of AIC, Schwarz'sBayesian Criterion (SBC), likelihood ratio (LR) test, and severaldiagnostic tests was used to select the number of lags required. Iftwo different orders of lags are obtained by the AIC and SBC criteria,then we apply the LR test to choose one of these two orders of lags(see Hatemi-J and Irandoust, 2000).We then check to see whether thechosen orders of lags pass several diagnostic tests; if not, we increasethe order of the lags successively until the diagnostic tests show betterresults when we test for the reliability of our models by applying anumber of diagnostic tests, including tests of autocorrelation,autoregressive conditional heteroscedasticity (ARCH), normality andheteroscedasticity. As can be seen from Table 6, we found no evidenceof serious violation of all the above tests.

Having established the order of integration of the series (dmax) andthe optimum lag length (k), the next step is to conduct Granger non-causality test by augmenting the VAR (k) by the maximum order ofintegration of the series, dmax. Table 7 presents results of the Grangernon-causality tests. As we are relatively more interested in therelationship between GDP, CO2 and energy consumption, we shallconcentrate on results pertaining to these variables. As Table 7 showsthe most interesting outcome of our causality test is that there is aunidirectional causality running from carbon dioxide emissions (CO2)to economic growth without feedback. Consistent with the resultsfrom the log-run test presented in Table 4, the sum of the laggedcoefficients of the pollutant emissions (CO2) was positive indicatingthat higher CO2 emissions promote economic growth. The implicationof our finding is that it is not possible to reduce emissions withoutsacrificing economic growth as reduction in CO2 emissions can causeoutput to decline. Our result is consistent with the empirical evidence

GDP does not cause capital 9.00 0.29⁎⁎ GDP=0.36Labour does not cause GDP 15.04 0.00⁎⁎⁎ Labour=−0.11GDP does not cause labour 3.42 0.33 GDP=0.06

Notes: ⁎⁎⁎, ⁎⁎ and ⁎ denote rejection of the null hypothesis of no causality at significantlevels of 1%, 5% and 10% respectively.

Page 7: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

1380 K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

found by Ang (2008) for Malaysia and Soytas and Sari (2009) forTurkey but not consistent with Soytas et al. (2007) for the UnitedStates and Zhang and Cheng (2009) for China.

On the relationship between energy consumption and CO2

emissions, Table 7 shows that there was a unidirectional causalityrunning from energy consumption to CO2 without feedback implyingthat an increase in energy consumption leads to an increase in CO2

emissions. This result is consistent with the view that energyconsumption is the main cause of CO2 emissions (degradation of theenvironment). This implies that reducing energy consumption,especially the consumption of fossil fuels, is a viable option that canhelp to reduce carbon dioxide emissions. This is particularlyimportant because about 70% of South Africa's total primary energysupply is derived from coal, and coal-fired power stations providemore than 93% of electricity production (World Bank, 2008). SouthAfrica's ratio of fossil-fuel CO2 emissions to total GHG emissions isamongst the highest in the world (81.2%).

The energy strategy of South Africa indicates that coal consump-tion is going to increase in the near future. This will inevitablyexacerbate further environmental degradation unless measures aretaken to reduce emissions and to find alternative environmentalfriendly sources of energy. The expansion of nuclear energy alonemaynot be adequate to meet the growing demand for energy and act as apanacea for growing CO2 emissions. There is also a plan to promoterenewable and clean energy, where the Government has set a 10-yeartarget to increase the share of renewable energy in total energyconsumption from 9–14% by 2012. There is great potential for usingsolar, wind, biomass, ocean waves, and hydropower. Reductions incarbon emissions of 3 million and 19 million tons are conceivable in2010 and 2025, respectively, based on baseline emission projectionsfor bulk electricity (Department of Energy and Minerals, South Africa,2002).

While there are several programs envisaged to address energyefficiency and energy sustainability, critics of the South African energystrategy still point out the there are several factors that are hinderingthese programs, such as the slow restructuring of the energy sector.South Africa is endowed with a variety of sustainable renewableenergy sources such as abundant sunshine and good wind regimes.These resources could offer a valuable complement to the currentalmost exclusive dependence on coal-based electricity generation(Sebitosi and Pillay, 2008a).

Some critics such as Surridge and Cloete (2009) argue thatsustainable coal technologies to mitigate CO2 emissions should beimposed on existing and new coal-powered plants. Additionally, asthe literature on carbon emission policies suggests, a possible tax onpolluters and carbon emission trading scheme are other alternativemeasures to combat CO2 emissions. Currently, carbon-capture andstorage are being investigated as a mitigation measure for carbondioxide emissions until renewable and nuclear energies are viableenough to reduce the use of fossil-fuel energy (Surridge and Cloete,2009). However it is beyond the scope of the paper to suggestwhether a pollution tax or emission trading scheme is more relevantto the South African case.

Despite the growing environmental challenge, the South Africanenvironmental policy does not seem to be serious about implement-ing a coherent strategy for sustainable environmental development aswell as a means of combating GHG emissions (Sebitosi and Pillay,2008b). The energy challenge facing South Africa is mostly being metby building more coal-powered stations to generate more electricity(Sebitosi and Pillay, 2008b). This will further contribute to more CO2

emissions. Moreover, in the past, the lack of a binding internationalagreement has encouraged countries like South Africa to pay lipservice to environmental protection (Sebitosi and Pillay, 2008a).

On the causal relationship between energy consumption andeconomic growth, we find a unidirectional causality running fromenergy consumption to economic growth without a feedback which is

consistent with Wolde-Rufael (2009). As can been seen from Table 7,the sum of the lagged coefficients of the energy variable is negative.This finding seems counter-intuitive from an economic point of viewas the lagged coefficient of the energy variable is negative implyingthat more energy consumption retards economic growth. However, aspointed out in Section 4.1 of this paper, there are several factors atwork that could have made this to happen including the inefficientuse of energy which South Africa is trying to solve. Therefore energyefficiency should be used not only as ameans of saving energy but alsoas a means of reducing pollutant emission. A recent study (Winkleret al., 2002) showed that a 5% increase in electricity efficiency in 2010would lead to a net increase of some 39,000 jobs and income of about$80 million. A national drive towards energy efficiency on this scalewould reduce emissions of carbon dioxide by roughly 1.5 million tonsof carbon in 2010. These demand-side management measures couldlead to reductions of annual carbon emissions of 2.2 million tons and5.2 million tons per year in 2010 and 2025, respectively (Winkleret al., 2002).

4.3. Variance decomposition analysis

The causality test presented above indicates only Granger causalitywithin the sample period and does not allow us to gauge the relativestrength of the Granger causality tests among the series beyond thesample period (Payne, 2002). Thus, to complement the abovediscussion, we apply the generalized impulse response approachproposed by Pesaran and Shin (1998) that does not requireorthogonalization of shocks and is invariant to the ordering of thevariables in the VAR. The difference between the orthogonalized andgeneralized forecast error variance decomposition is that in anorthogonalized forecast error variance decomposition, the percentageof the forecast error variance of a variable which is accounted for bythe innovation of another variable in the VAR will sum to one acrossall variables whereas the generalized forecast error variance decom-position allows one to make robust comparisons of the strength, size,and persistence of shocks from one equation to another (Payne,2002). This method does not evaluate the percentages of the forecasterror variance explained by each variable in absolute terms but only inrelative terms. Unlike the orthogonalised case, the row values for thegeneralized decompositions do not sum up to 100 (Payne, 2002 andSari and Soytas, 2007). The generalized version gives an optimalmeasure of the amount of forecast error variance decomposition foreach series (see Sari and Soytas, 2007; Payne, 2002).

Thus, to complement the above cointegration and causality tests,we decomposed the forecast error variance of GDP into proportionsattributed to shocks in all variables in the system including itself. Bydoing so, we can provide an indication of the strength of the Grangercausality test beyond the sample period. As we are more interested inthe contribution of CO2 emissions and energy consumption to outputgrowth as compared to the other two inputs of capital and labour, weonly decompose the forecast error variance of the output variable(lnYt) in response to innovations in CO2 emissions (lnCt) and energyconsumption (lnEt), capital (lnKt) and (lnLt). As can be seen fromTable 8, in the short-run (1 to 5 years), the forecast error variance ofCO2 emission explains about 8% of the forecast error variance of GDPwhile in the long-run it explains about 13% of the forecast errorvariance of GDP. In contrast, the forecast error variance of GDPexplains only 3% of the forecast error variance of CO2 emissions bothin the short and the long-run period. This seems to mildly support theunidirectional causality running from CO2 emissions to economicgrowth found earlier by the Granger causality test. Similarly, in theshort-run, the forecast error variance of energy consumption explainsmore than 8% of the forecast error variance of GDP while in the long-run it explains about 15% of the forecast error variance of GDP. Table 8further shows that CO2 emissions and energy consumption werecomplements, as a large part of their forecast error variance is

Page 8: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

Table 8Results of generalized forecast error variance decomposition analysis.

Variable Horizon ΔY ΔC ΔE ΔK ΔL

ΔY (GDP) 1 85.5 4.7 9.1 37.6 36.05 41.2 7.9 8.1 33.0 39.5

10 37.1 12.5 13.6 29.8 35.815 36.2 12.8 14.8 29.0 35.820 35.9 12.9 15.0 28.8 35.8

ΔC (CO2) 1 2.4 79.5 60.0 10.4 0.55 3.0 71.3 56.5 10.6 3.3

10 3.1 70.3 56.1 10.3 3.815 3.1 69.9 56.1 10.3 3.920 3.1 69.8 56.1 10.3 3.9

ΔE (Energy) 1 12.7 58.3 92.6 18.5 6.65 11.5 49.7 72.3 17.6 13.8

10 11.3 47.0 68.9 16.1 15.815 11.0 46.2 68.2 15.6 16.420 11.0 45.9 68.0 15.6 16.6

ΔK (Capital) 1 41.4 0.8 8.1 93.1 4.55 23.7 4.5 9.7 63.8 21.7

10 22.6 9.3 13.1 56.4 21.715 22.0 9.9 14.6 54.7 22.020 21.9 10.0 14.7 54.4 22.1

ΔE (Labour) 1 42.8 2.1 5.7 4.5 96.15 26.5 10.1 12.9 11.9 71.2

10 26.1 11.0 13.1 11.8 69.515 25.9 11.2 13.4 11.7 69.220 25.9 11.2 13.4 11.7 69.2

Notes: Unlike the orthogonalised case, the row values for the generalizeddecompositions do not sum up to100. The generalized version gives an ‘optimal’measure of the amount of forecast error variance decomposition for each series (seeSari and Soytas, 2007). The cost is that one cannot evaluate the percentages of theforecast error variance explained by each variable in absolute terms, only in relativeterms.

1381K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

explained by each other. For instance, in the short-run, CO2 emissionsexplain around 50% of the forecast error variance of energy while theforecast error variance of energy explains around 57% of the forecasterror variance of CO2 emissions.

In summary, the econometric evidence presented above presentsSouth Africa with two unpalatable policy options. On the one handsince causality runs from CO2 emissions to economic growth, SouthAfrica has to sacrifice economic growth if the country wants to reduceCO2 emissions. On the other hand, since there is a unidirectionalcausality running from energy use to CO2 emissions, energy use has tobe reduced to mitigate the adverse effects of CO2 emissions. In thelong-run however, it is possible to meet the energy needs of thecountry and at the same time reduce CO2 emissions by developingalternative energy sources to coal, the main source of CO2 emissions.Given that nuclear energy currently accounts for only 6% of totalenergy, the proposal to reduce emissions by developing nuclearenergy may not be adequate to combat the huge environmentalchallenge facing South Africa. A large proportion of energy supplywould still be obtained from fossil fuel. This creates a dilemma for acountry like South Africa which needs rapid economic growth to dealwith the legacy of apartheid, within the context of a global concern forthe reduction of greenhouse emissions. In the long-run, however,South Africa cannot shy away from adopting clean fossil-fueltechnologies in the generation of energy from coal.

5. Concluding remarks

Like many coal-abundant and major coal consuming countries,South Africa is facing a crucial challenge on the extent to which fossilfuel can be used to generate electricity. The country is confrontedwiththe crucial issue of balancing its sectoral energy strategies to producemore coal while at the same time having to reduce greenhouse gasemissions (GHG) which is largely generated by coal-fired powerstations.

This paper examined the cointegration and the causal relationshipbetween economic growth, pollutant emissions, energy consumption,labour and capital in South Africa for the period 1965–2006 byapplying the cointegration approach developed by Pesaran et al.(2001) and using the modified version of the Granger causality testproposed by Toda and Yamamoto (1995). Our empirical resultssuggest that there was a short-run as well as a long-run relationshipbetween the variables with a positive and a significant relationshipbetweenpollutant emissions and economic growth. Our causality testsalso indicate that there was a unidirectional Granger causality runningfrom pollutant emissions to economic growth; from energy consump-tion to economic growth and from energy consumption to CO2, allwithout feedback. The evidence seems to suggest that economicgrowth is not a solution to reducing the levels of CO2 emissions. IfSouth Africa is to reduce its emission levels, it has to reduce not onlyenergy consumption per unit of output but it may sacrifice itseconomic growth. An alternative and viable option for South Africais to increase the use of alternative sources of energy that are relativelyfree from pollutant emissions. South Africa is endowed with adequatesources of renewable energy that can simultaneously address both theenergy needs as well as the environmental concerns due to CO2

emissions. So far, the evidence indicates that South Africa has not fullyexploited these alternative energy resources. To combat energyshortages and address the environmental degradation facing thecountry, there is a need for South Africa to develop strategies to exploitits clean energy potential and increase their utilization. Research andinvestment in clean energy should be an integral part of the process ofcontrolling GHG emissions and finding alternative sources of energy tocoal. Improving energy efficiency and the exploitation of the country'srenewable energy are viable options formeeting the significant energyand environmental challenges facing the country. However, theeconometric results upon which the policy suggestions are madeshould be interpreted with care, as they may not be robust enough tocategorically warrant the choice of an unpalatable policy option bySouth Africa.

Acknowledgements

We are very grateful to two anonymous referees whose construc-tive comments have helped to improve upon the quality of the paper.We are also grateful to the Editor of the Journal, Richard S J Tol for hisencouragement. The usual caveats apply.

References

Ang, J.B., 2007. Are saving and investment cointegrated? The case of Malaysia (1965–2003). Applied Economics 39, 2167–2174.

Ang, J.B., 2008. Economic development, pollutant emissions and energy consumption inMalaysia. Journal of Policy Modeling 30, 271–278.

Ang, J.B., 2009. CO2 emissions, research and technology transfer in China. EcologicalEconomics 68, 2658–2665.

Apergis, N., Payne, J.E., 2009. CO2 emissions, energy usage, and output in CentralAmerica. Energy Policy 37, 3282–3286.

Apergis, N., Payne, J.E., 2010. The emissions, energy consumption and growth nexus:evidence from the commonwealth of independent states. Energy Policy 38,650–655.

Apergis, N., Payne, J.E., Menyah, K., Wolde-Rufael, Y., 2010. On the causal dynamicsbetween emissions, nuclear energy, renewable energy, and economic growth.Ecological Economics 69, 2255–2260.

Aslanidis, N., 2009. Environmental Kuznets curves for carbon emissions: a criticalsurvey, Department of Economics, University Rovira Virgili FCEE, AvingudaUniversitat, Department of Communication, Working Paper No 51.

Bowden, N., Payne, J.E., 2009. The causal relationship between U.S. energy consumptionand real output: a disaggregated analysis. Journal of Policy Modeling 31, 180–188.

Büscher, B., 2009. Connecting political economies of energy in South Africa. EnergyPolicy 37, 3951–3958.

Caporale, G.M., Pittis, N., 1999. Efficient estimation of cointegrating vectors and testingfor causality in vector autoregressions. Journal of Economic Surveys 13, 3–35.

Clarke, J., Mirza, S.A., 2006. Comparison of some commonmethods of detecting Grangernoncausality. Journal of Statistical Computation and Simulation 76, 207–231.

Conference Board and Groningen Growth and Development Centre, Total EconomyDatabase, September 2008: http://www.conference-board.org/economics.

Page 9: Energy consumption, pollutant emissions and …faculty.smu.edu/millimet/classes/eco6375/papers/menyah...Energy consumption, pollutant emissions and economic growth in South Africa

1382 K. Menyah, Y. Wolde-Rufael / Energy Economics 32 (2010) 1374–1382

Department of Energy and Minerals, South Africa, DME. 2002. Draft White Paper onRenewable Energy and Clean Energy. Available at: http://www.dme.gov.za.

Department of Energy and Minerals, South Africa, 2008; Annual Report. http://www.dme.gov.za/pdfs/about_dme/AR%20pdf%20for%20web.pdf.

Emran, M.H., Shilip, F., Alam, M.I., 2007. Economic liberalization and price response ofaggregate private investment, time series evidence from India. Canadian Journal ofEconomics 40, 914–934.

Galeotti, M., Manera, M., Lanza, A., 2009. On the robustness of robustness checks of theenvironmental Kuznets curve hypothesis. Environmental Resource Economics 42,551–574.

Halicioglu, F., 2009. An econometric study of CO2 emissions, energy consumption,income and foreign trade in Turkey. Energy Policy 37, 1156–1164.

Harvey, D.L., Leybourne, S.J., Newbold, P., 2003. How great are the great ratios? AppliedEconomics 35, 163–177.

Hatemi-J, A., Irandoust, M., 2000. Time-series evidence for Balassa's export-led growthhypothesis. Journal of International Trade and Economic Development 9, 355–365.

Jalil, A., Mahmud, S.F., 2009. Environment Kuznets curve for CO2 emissions: acointegration analysis for China. Energy Policy 37, 5167–5172.

Kónya, L., 2004. Export-led growth, growth-driven export, both or none? Grangercausality analysis on OECD countries. Applied Econometrics and InternationalDevelopment 4, 73–97.

Lee, C.C., Cheng, C., Chen, P., 2008. Energy-income causality in OECD countries revisited:the key role of capital stock. Energy Economics 30, 2359–2373.

Loizides, J., Vamvoukas, G., 2005. Government expenditure and economic growth:evidence from trivariate causality testing. Journal of Applied Economics. 8, 25–152.

Lütkepohl, H., 1982. Non-causality due to omitted variables. Journal of Econometrics 19,367–378.

Lütkepohl, H., 1993. Introduction to Multiple Time Series, 2nd ed. Springer Verlag,Berlin.

Menyah, K., Wolde-Rufael, Y., 2010. CO2 emissions, nuclear energy, renewable energyand economic growth in the US. Energy Policy 38, 2911–2915.

Narayan, P.K., 2005. The saving and investment nexus for China, evidence fromcointegration tests. Applied Economics 37, 1979–1990.

Odhiambo, N.M., 2009. Electricity consumption and economic growth in South Africa: atrivariate causality test. Energy Economics 31, 635–640.

Payne, J.E., 2002. Inflationary dynamics of a transitional economy: the Croationexperience. Journal of Policy Modeling 34, 219–230.

Payne, J.E., Taylor, J.P., 2010. Nuclear energy consumption and economic growth in theUS: an empirical note. Energy Sources, Part B: Economics, Planning, and Policy 5,301–307.

Perron, P., 1997. Further evidence on breaking trend functions in macroeconomicvariables. Journal of Econometrics 80, 355–385.

Pesaran, M.H., Shin, Y., 1998. Generalised impulse response analysis in linearmultivariate models. Economics Letters 58, 17–29.

Pesaran, M.H., Shin, Y., Smith, S., 2001. Bounds testing approach to the analysis of levelrelationships. Journal of Applied Econometrics 16, 289–326.

Rambaldi, A.N., Doran, T.E., 1996. Testing for Granger non-causality in cointegratedsystem made easy. Working Papers in Econometrics and Applied Statistics No. 88. :Department of Econometrics, University of New England.

Sari, R., Soytas, U., 2007. The growth of income and energy consumption in sixdeveloping countries. Energy Policy 35, 889–898.

Sari, R., Soytas, U., 2009. Are global warming and economic growth compatible?Evidence from five OPEC countries? Applied Energy 86, 1887–1893.

Sebitosi, A.B., Pillay, P., 2008a. Grappling with a half-hearted policy: the case of renewableenergy and the environment in South Africa. Energy Policy 36, 2513–2516.

Sebitosi, A.B., Pillay, P., 2008b. Renewable energy and the environment in South Africa:a way forward. Energy Policy 36, 3312–3316.

Soytas, U., Sari, R., Ewing, B.T., 2007. Energy consumption, income, and carbonemissions in the United States. Ecological Economics 62, 482–489.

Soytas, U., Sari, R., 2009. Energy consumption, economic growth, and carbon emissions:challenges faced by an EU candidatemember. Ecological Economics 68, 1667–1675.

Spalding-Fecher, R., Williams, A., van Horen, C., 2000. Energy and environment in SouthAfrica: charting a course to sustainability. Energy for Sustainable Development 4 (4),8–17.

Squalli, J., 2007. Electricity consumption and economic growth: bounds and causalityanalyses of OPEC countries. Energy Economics 29, 1192–1205.

Surridge, A.D., Cloete, M., 2009. Carbon capture and storage in South Africa. EnergyProcedia 1, 2741–2744.

Toda, H.Y., Yamamoto, T., 1995. Statistical inference in vector autoregressions withpossibly integrated processes. Journal of Econometrics 66, 225–250.

Winkler, H. Howells, M. and Baumer, K. 2002. Sustainable development policies andmeasures. Institutional issues and electrical efficiency in South Africa, http://www.eri.uct.ac.za/Research/publications/05Winkler-Baumert_SDPAMS.pdf.

Winkler, H., 2007. Energy policies for sustainable development in South Africa. Energyfor Sustainable Development 11, 26–34.

Wolde-Rufael, Y., 2005. Energy demand and economic growth. The African ExperienceJournal of Policy Modelling 27, 891–903.

Wolde-Rufael, Y., 2006. Electricity consumption and economic growth: a time seriesexperience for 17 African countries. Energy Policy 34, 1106–1114.

Wolde-Rufael, Y., 2009. Energy consumption and economic growth: the Africanexperience revisited. Energy Economics 31, 217–224.

Wolde-Rufael, Y., 2010. Coal consumption and economic growth revisited. AppliedEnergy 87, 160–167.

Wolde-Rufael, Y., Menyah, K., 2010. Nuclear energy consumption and economic growthin nine developed countries. Energy Economics 32, 550–556.

World Bank. 2008. World Development Indicators.Xepapadeas, A., 2005. Economic growth and the environment. In:Mäler, K.G., Vincent,

J.R. (Eds.), Handbook of Environmental Economics, edition 1, volume 3. Elsevier,pp. 1219–1271.

Zhang, Xing-Ping, Cheng, Xiao-Mei, 2009. Energy consumption, carbon emissions, andeconomic growth in China. Ecological Economics 68, 2706–2712.

Ziramba, E., 2009. The demand for residential electricity in South Africa. Energy Policy36, 3460–3466.

Zivot, E., Andrews, D., 1992. Further evidence of great crash, the oil price shock and unitroot hypothesis. Journal of Business and Economic Statistics 10, 251–270.