Economic Modelling 40 (2014) 33–41

Contents lists available at ScienceDirect

Economic Modelling

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

The impact offinancial development, income, energy and trade on carbonemissions: Evidence from the Indian economy

Mohamed Amine Boutabba ⁎EPEE, University of Evry-Val d'Essonne, 4 bd. François Mitterrand, 91025 Evry Cedex, France

⁎ Tel.: +33 1 69 47 79 22.E-mail address: mohamedamine.boutabba@univ-evry

http://dx.doi.org/10.1016/j.econmod.2014.03.0050264-9993/© 2014 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Accepted 7 March 2014Available online xxxx

JEL classification:C32O53Q43Q53Q56

Keywords:Carbon emissionsFinancial developmentGrowthEnergy consumptionTrade

This paper examines the long-run equilibrium and the existence and direction of a causal relationship betweencarbon emissions, financial development, economic growth, energy consumption and trade openness for India.Our main contribution to the literature on Indian studies lies in the investigation of the causes of carbon emis-sions by taking into account the role of financial development and using single country data. The results suggestthat there is evidence on the long-run and causal relationships between carbon emissions,financial development,income, energy use and trade openness. Financial development has a long-run positive impact on carbon emis-sions, implying that financial development improves environmental degradation. Moreover, Granger causalitytest indicates a long-run unidirectional causality running from financial development to carbon emissions andenergy use. The evidence suggests that financial system should take into account the environment aspect intheir current operations. The results of this study may be of great importance for policy and decision-makersin order to develop energy policies for India that contribute to the curbing of carbon emissions while preservingeconomic growth.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

Climate change and global warming are the greatest and most con-troversial environmental issues of our times. There is broad consensusamong scientists that accumulated carbon dioxide emitted from theburning of fossil fuels, along with contributions from other human-induced greenhouse gas emissions, are warming the atmosphere andoceans of the earth (IPCC, 2007). The global effects of climate changeare already apparent in increasing the frequency of extreme weatherevents, altering precipitation patterns, heightening storm intensity,reversing ocean currents and a rising sea level. These changes, in turn,can have significant impacts on the functioning of ecosystems, theviability of wildlife, and the well-being of humans.

With the world's second largest population and over 1.1 billionpeople, India is one of the lowest greenhouse gas emitters in theworld on a per-capita basis. Its emission of 1.18 tonnes of carbon equiv-alent per capita in 2008 was nearly one-fourth of the correspondingglobal average of 4.38 tonnes. However, India is highly vulnerable toclimate change, as a large population are dependent on agriculture

.fr.

and forestry for livelihood. The Indian economy is also dependent onnatural resources and any adverse impact on these and related sectorswill negate government's efforts to eradicate poverty and ensure sus-tainable livelihood for the population.

India accords high priority to its development. The economy hasbeen growing, on average, at 7.7% per year between 2000 and2007, and fossil-fuel carbon emissions have increased by 125% be-tween 1950 and 2008, becoming the world's third largest fossil-fuel CO2-emitting country. As outlined in India's 12th Five YearPlan (2012–2017), the government of India has provisionally set a9% GDP growth target, which will require energy supply to growat 6.5% per year. Being aware of achieving its growth trajectory inan environmentally sustainable manner, India has announced inDecember 2009 that it would aim to reduce the emissions intensityof its GDP by 20–25% from 2005 levels by 2020. Therefore, India isfaced with the challenge of identifying the common ground betweenclimate change policy and economic growth and pursuing measuresthat achieve both.

However, to control the greenhouse gas emissions and to ensurethe sustainability of the economic development, it is importantto better understand the inter-temporal links in the environment–energy–income nexus. In the literature, there have been few researchesto explore the relationship between these variables in the case of India.

34 M.A. Boutabba / Economic Modelling 40 (2014) 33–41

Ghosh (2010) investigated the causal relationship between carbonemissions and economic growth using ARDL bounds testing approachcomplemented by Johansen–Juselius maximum likelihood proce-dure in a multivariate framework by incorporating energy supply,investment and employment. The result revealed the absence of along-run causality between carbon emissions and economic growth;however a bi-directional short-run causality between the two isfound. Alam et al. (2011) applied the Toda and Yamamoto causalitytest to examine the dynamic relationship between carbon emissions,economic growth, energy consumption, labour forces and gross fixedcapital formation. They found a bi-directional Granger causalitybetween energy consumption and carbon emissions in the longrun but neither carbon emissions nor energy consumption causesmovements in economic growth. Jayanthakumaran et al. (2012)analyzed the long-run relationship between carbon emissions andother variables such as growth, energy, trade and endogenouslydetermined structural breaks. They found evidence for the exis-tence of an EKC hypothesis for India. However, they failed to derivea clear picture regarding the association of structural change andcarbon emissions. Kanjilal and Ghosh (2013) examine the EKC hy-pothesis using threshold cointegration with endogenously deter-mined structural breaks. The study advocates that the existenceof structural breaks in the period of the study can produce mislead-ing results if they are not incorporated in the cointegration testingmodels.

This paper extends the above-mentioned multivariate frame-work further by including the impacts of financial developmentinto the nexus. To the best of our knowledge, there has never beenan attempt to investigate the causes of carbon emissions for Indiaby taking into account the role of financial development and usingsingle country data. This study tries to fulfil this gap. In this respect,we argue that the analysis of the relationship between carbon emis-sions and financial development may reduce the problems of omit-ted variable bias in econometric estimation. This attempt may alsobe of great importance for policy and decision-makers to better ap-prehend the determinants of carbon emissions in order to developeffective energy policies that will palliate the impacts of human ac-tivities, and thereby contribute to the curbing of carbon emissionswhile preserving economic growth.

The remainder of the paper is organized as follows. Section 2presents a brief literature review related to financial development andcarbon emissions. Section 3 describes the data and methodology.Empirical results are given in Section 4 while the summary and theconcluding remarks are outlined in Section 5.

2. A brief literature review

The impact of financial development on environmental conditionshas gained increasing attention in the recent literature. Yuxiang andChen (2011) used provincial data of Chinese economy to examinethe impact of financial development on industrial pollutants andfound improvements in environment due to financial development.They claimed that financial development improves environmentalquality by increasing incomeand capitalization, exploitingnew technol-ogy and implementing regulations regarding environment. Jalil andFeridun (2011) investigated the impact of financial development, eco-nomic growth and energy consumption on CO2 emissions in the caseof China from 1953 to 2006. The results of the analysis revealed a nega-tive sign for the coefficient of financial development, suggesting thatfinancial development in China has not taken place at the expense ofenvironmental pollution. On the contrary, it is found that financialdevelopment saves the environment from degradation. Moreover, theresults confirm the existence of a long-run relationship between carbonemissions, income, energy consumption and trade openness whilesupporting the presence of EKC hypothesis. Similarly, Zhang (2011)explored the effect of financial development on carbon emissions.

Results indicated that, first, China's financial development constitutesan important driver for carbon emissions increase, which should betaken into account when carbon emissions demand is projected. Sec-ond, the influence of financial intermediation scale on carbon emissionsoutweighs that of other financial development indicators but itsefficiency's influence appears by far weaker although it may cause thechange of carbon emissions statistically. Third, China's stock marketscale has a relatively larger influence on carbon emissions but the influ-ence of its efficiency is very limited. Finally, among financial devel-opment indicators, China's FDI exerts the least influence on thechange of carbon emissions, due to its relatively smaller volumecompared with income. Ozturk and Acaravci (2013) examined thecausal relationship between financial development, openness, eco-nomic growth, energy consumption and carbon emissions in Turkeyfor the period 1960–2007. Empirical results yielded evidence of along-run relationship between carbon emissions, energy consumption,income, openness ratio and financial development. The results alsosupported the validity of EKC hypothesis in Turkish economy. However,financial development has no significant effect on carbon emissions inthe long- run.

For cross-country case studies, Talukdar and Meisner (2001) ex-amined the impact of private sector involvement on carbon emis-sions using data from 44 developing countries over nine years(1987–95). They found that both foreign direct investments and do-mestic financial capital markets in an economy are likely to havepositive impacts on the environment. Claessens and Feijen (2007)analyzed the role of governance in reducing CO2 emissions and re-ported that with the help of more advanced governance firms canlower the growth of carbon emissions. They suggested that financialdevelopment might stimulate the performance of firms due to theadoption of energy efficient technologies, which reduce carbonemissions. Tamazian et al. (2009) investigated the linkage betweenfinancial development, economic development and environmentalquality for BRIC countries using panel data over the period 1992–2004. Their results revealed that higher degree of economic andfinancial development decreases the environmental degradation.Tamazian and Bhaskara Rao (2010) tested the role of economic,financial and institutional developments on environmental degrada-tion with a sample of 24 transition countries for the period from1993 to 2004. Their findings showed that financial liberalizationmay be harmful for environmental quality if it is not accomplishedin a strong institutional framework. In addition, the findings confirmthe existence of an EKC.

India is included in some of the above-mentioned panel data studies.However, it is widely recognized that any potential inference drawnfrom these cross-country studies provides only a general understandingof the linkage between the variables, and thus are unable to offer muchguidance on policy implications for each country (Ang, 2008; Lindmark,2002; Stern et al., 1996). Hence, the aim of this research is to investigatethe impact of financial development on carbon emissions in the case ofIndia.

3. Methodology and data

Following the empirical literature in energy economics, it is plausibleto form the long-run relationship between carbon emissions, financialdevelopment, economic growth, energy consumption, and foreigntrade in linear logarithmic quadratic form, with a view of testing thelong-run and causal relationships between these variables in India, asfollows:

CO2t ¼ β0 þ α F Ft þ αYYt þ αY2Y2t þ αEEt þ αTTt þ εt ð1Þ

where t and ε denote time and error, respectively. CO2 is carbonemissions (measured in metric tonnes per capita), F stands for

35M.A. Boutabba / Economic Modelling 40 (2014) 33–41

financial development that is the total value of domestic credit toprivate sector1,2 as a share of GDP, Y indicates per capita real GDP(measured in local constant currency), Y2 is the square of per capitareal GDP, E means the energy consumption (measured as kg of oilequivalent per capita), which is used as a proxy for economic growth,and T represents trade openness, which is the total value of exportsand imports as a share of GDP.

The parameters αF, αY, αY2, αE, and αT are the long-term elasticityestimators of CO2 emissions with respect to financial development,per capita real GDP, the square of per capita real GDP, the per capitaenergy consumption and the trade openness, respectively.

Financial development may be harmful for environmental qualityαF N 0 (Talukdar and Meisner, 2001; Zhang, 2011), otherwise αF b 0 ifthe focus of financial sector is to improve environmental quality byenabling firms in adopting advanced cleaner and environment friendlytechniques (Claessens and Feijen, 2007; Jalil and Feridun, 2011;Tamazian et al., 2009).

The EKC hypothesis suggests that αY N 0 and αY2 b 0. αY being posi-tive reveals the phenomenon wherein as income increases, the CO2

emissions increase as well; αY2 being negative reflects the inverted-Ucurve-shaped pattern of the EKC, where once income passes the thresh-old, the CO2 emissions will decrease.

The expected sign of αE is positive because a higher level of energyconsumption should result in greater economic activity and stimulateCO2 emissions.

αY is expected to be negative or positive, depending on the level ofeconomic development stage of a country. In general, developing coun-tries, which are abundant in labour and natural resources, attempt topromote heavy industries, which usually are pollution-intensive, byaccepting foreign direct investment of developed countries. In contrast,developed countries change from energy-intensive industries to ser-vices and knowledge-based technology-intensive industries, whichare environmentally cleaner (Grossman and Krueger, 1995).

The sample period runs from 1971 to 2008 based on the annual timeseries data availability. The data originate from the world developmentindicator data base, the World Bank. All variables are employed withtheir natural logarithms form to reduce heteroskedasticity and to obtainthe growth rate of the relevant variables by their differenced logarithms.

3.1. Estimation strategy

This study employs the Autoregressive Distributed Lag (ARDL)bounds testing procedure recently developed by Pesaran et al. (2001).The ARDL has several advantages over other techniques of cointegrationsuch as Engle and Granger (1987) and Johansen and Juselius (1990).First, it can be applied irrespective of whether the underlying variablesare I(0), I(1) or a combination of both (Pesaran and Pesaran, 1997). Sec-ond, the ARDL procedure is statistically a more significant approach todetermine the cointegration relation in small samples than that of theJohensen and Juselius cointegration technique (Pesaran and Shin,1999). Third, evenwhere someof themodel regressors are endogenous,

1 Domestic credit to private sector refers to financial resources provided to the privatesector, such as through loans, purchases of non-equity securities, and trade credits andother accounts receivable, that establish a claim for repayment.

2 In the literature, there are many proxies used for representing financial development.For example, the monetary aggregate M2 as a ratio of nominal GDP is used in measuringfinancial deepening. However, the availability of foreign funds in the financial systemmakes the monetary aggregate an inappropriate measure of financial development. An-other commonly used variable is the ratio of deposit liabilities to nominal GDP, which cap-tures the broad money stock excluding currency in circulation. But, this measure doesn'ttake into account the allocation of capital. Several studies have also employed the ratioof commercial bank assets divided by commercial bank plus central bank assets whichmeasures the importance of the commercial banks in the financial system. In this study,we use the domestic credit to private sector as a percentage of GDP, which constitutesthe most common variable used in the literature to represent financial development. Infact, this measure represents more accurately the role of financial intermediaries inchannelling funds to private markets participant.

the bounds testing approach generally provides unbiased long-run esti-mates and valid t-statistics (Narayan, 2005). Fourth, the model takes asufficient number of lags to capture the data generating process in gen-eral to specific modeling frameworks (Laurenceson and Chai, 2003).Fifth, the error correction model (ECM) can be derived from ARDLthrough a simple linear transformation, which integrates short run ad-justments with long run equilibrium without losing long run informa-tion (Pesaran and Shin, 1999).

Basically, the ARDL approach to cointegration involves two steps forestimating long-run relationship. The first step is to investigate theexistence of long-run relationship among all variables in the equationunder estimation. If there is an evidence of cointegration between vari-ables, the second step is to estimate the long-run and short-runmodels.

3.2. Stationarity

As discussed earlier, the ARDL bounds testing procedure can beapplied irrespective of whether the variables are I(0), I(1) (Pesaranand Pesaran, 1997). However, according to Ouattara (2004), in the pres-ence of I(2) variables the computed F-statistics provided by Pesaranet al. (2001) become invalid. This is because the bounds test is basedon the assumption that the variables should be I(0) or I(1). Therefore,the implementation of unit root tests in the ARDL procedure is neces-sary to ensure that none of the variables is integrated at an order ofI(2) or beyond.

It is well known that the presence of structural breaks in the seriesmay bias the results toward non rejection of the null hypothesis of aunit root when there is none. This consideration is of particular impor-tance since the economic system in India has been subject to somedras-tic changes in policy and regulations. An alternative to the unit root testagainst a single-break stationarity was proposed by Zivot and Andrews(1992). It was extended to a two-break stationarity alternative byLumsdaine and Papell (1997) and up to five-break stationarity alterna-tive, with a priori unknown number of breaks, by Kapetanios (2005).However, these tests maintain the linearity assumption under the unitroot null hypothesis. If a break exists under the null of a unit root, itwill exhibit size distortions that not only “over-reject” the null hypoth-esis of a unit root, but also will tend to estimate the break point incor-rectly. To overcome this problem, Lee and Strazicich (2003, 2004)have developed an alternative (at most two) endogenous break unitroot test that uses the LagrangeMultiplier (LM) test statistic, and allowsfor breaks both under null and alternative hypotheses. Thus, rejection ofthe unit root null based on LM test provides a quite strong evidence ofstationarity.

Lee and Strazicich (2003, 2004)'s unit-root test considers the datagenerating process as follows:

Δyt ¼ δ0ΔZt þ ϕeSt−1 þ ut

where eSt ¼ yt−eψx−Zteδ t ¼ 2;…Tð Þ and Zt is a vector of exogenous

variables defined by the DGP;eδ is the vector of coefficients in the regres-sion ofΔyt onΔZt respectivelywithΔ as the difference operator; andψx ¼y1−Z1

eδ, with y1 and Z1 the first observations of yt and Zt respectively.The unit-root null hypothesis is described by ϕ = 0. The augmentedterms ΔeSt− j; j ¼ 1;…k; were included to correct for serial correlation.The value of k is determined by the general-to-specific search proce-dure. To endogenously determine the location of the break (TB), theLM unit-root searches for all possible break points for the minimum(the most negative) unit-root t-test statistic, as follows:

Inf eτ eλ� �¼ Inf λ eτ λð Þ;λ ¼ TB

T:

The critical values of the endogenous two-break LM unit-root testare reported in Lee and Strazicich (2003) and the critical values of theone-break LM unit-root test are tabulated in Lee and Strazicich (2004).

36 M.A. Boutabba / Economic Modelling 40 (2014) 33–41

In the present study, when the two-break LM test results showedthat only one structural break is significant to at least the 10 per centlevel for some series, we perform the one-break LM test of Lee andStrazicich (2004). This was done not only because the one-break LMtest appears more appropriate in this case, but also because we wantedto determine if including two breaks instead of one can adversely affectthe power to reject the unit root hypothesis for these countries. For thesame reason, when the one-break or two-break LM test results showedthat no break is significant, we employ the Augmented Dickey–Fuller(ADF), Phillips–Perron, Augmented Dickey–Fuller GLS (ADF-GLS), andKwiatkowski–Phillips–Schmidt–Shin (KPSS) unit-root techniques. Thefirst three techniques test the null hypothesis of a unit root against thealternative of stationarity. The KPSS method tests the hypothesis thatthe series is stationary against the alternative of non-stationarity.

3.3. Cointegration analysis

The ARDL procedure involves the estimation of Eq. (1) as follows:

ΔCO2t ¼ a0 þXpi¼1

a1iΔCO2t−i þXpi¼0

a2iΔFt−i þXpi¼0

a3iΔYt−i þXpi¼0

a4iΔY2t−i

þXpi¼0

a5iΔEt−i þXpi¼0

a6iΔTt−1 þ λ1CO2t−1 þ λ2 Ft−1 þ λ3Yt−1

þλ4Y2t−1 þ λ5Et−1 þ λ6Tt−1 þ μ t

ð2Þ

whereΔdenotes thefirst difference operator, a0 is the drift component, etμt is the usual white noise residuals, and the variables CO2, F, Y, Y2, E, and Tare as defined earlier. The terms with summation signs represent theerror correction dynamics, while the second part of the equation with λcorresponds to the long run relationship. This equation incorporates thetime trend variable to capture the autonomous time-related changes.

The ARDLmethod estimates (p+1)k number of regressions in orderto obtain the optimal lag length for each variable, where p is the maxi-mum number of lags to be used and k is the number of variables inthe equation. This is an appropriate lag selection based on criteriasuch as Akaike Information Criterion (AIC) and Schwarz Bayesian Crite-rion (SBC). The bounds testing procedure is based on the joint F-statisticor Wald statistic that is tested the null hypothesis of no cointegration.

Pesaran et al. (2001) and Narayan (2005) individually report twosets of critical values for a given significance level. One set of criticalvalues assumes that all variables included in the ARDL model are I(0),while the other is calculated on the assumption that the variables areI(1). If the computed test statistic exceeds the upper critical boundsvalue, then the H0 hypothesis is rejected. If the F-statistic falls into thebounds then the cointegration test becomes inconclusive. In this case,following Kremers et al. (1992) and Banerjee et al. (1998), the errorcorrection term will be a useful way for establishing cointegration. Ifthe F-statistic is lower than the lower bounds value, then the null hy-pothesis of no cointegration cannot be rejected. Two sets of criticalvalues are reported in Narayan (2005) for sample sizes ranging from30 observations to 80 observations. Given the relatively small samplesize in the present study (38 observations), we extract appropriatecritical values from Narayan (2005).

Having found that there exists a long-run relationship between thevariables, the next step is to estimate the error-correction model:

ΔCO2t ¼ a0 þXpi¼1

a1iΔCO22t−i þXpi¼0

a2iΔFt−i þXpi¼0

a3iΔYt−i þXpi¼0

a4iΔY2t−i

þXpi¼0

a5iΔEt−i þXpi¼0

a6iΔTt−i þ ηECTt−1 þ μ1t

ð3Þ

where ηmeasures the speed of adjustment to obtain equilibrium in theevent of shock(s) to the system and ECTt − 1 is the residuals that areobtained from the estimated cointegration model of Eq. (1).

To gauge the adequacy of the specification of the model, diagnosticand stability tests are conducted. Diagnostic tests examine the modelfor serial correlation, functional form, non-normality and hetero-scedasticity. As suggested by Pesaran and Pesaran (1997), the stabilityof the short-run and long run coefficients are checked through thecumulative sum (CUSUM) and cumulative sum of squares (CUSUMSQ)tests proposed byBrown et al. (1975). The CUSUMandCUSUMSQ statis-tics are updated recursively and plotted against the breaks points. If theplots of the CUSUM and CUSUMSQ statistics stay within the criticalbounds of a 5% level of significance, the null hypothesis of all coefficientsin the given regression is stable and cannot be rejected.

3.4. Granger causality

The ARDL method tests the existence or absence of cointegrationrelationship between variables, but not the direction of causality. Ifwe do not find any evidence for cointegration among the variablesthen the specification of the Granger causality test will be a vectorautoregression (VAR) in first difference form. However, if we findevidence for cointegration then we need to augment the Granger-typecausality test model with a one period lagged error correction term(ECTt − 1). This is an important step because Engle and Granger(1987) caution that if the series are integrated of order one, in the pres-ence of cointegration VAR estimation infirst differenceswill bemislead-ing. The augmented form of Granger causality test with ECM isformulated in multivariate qth order of VECM model as follows:

1−Bð Þ

CO2t

FtYt

Y2t

EtTt

26666666664

37777777775¼

b1b2b3b4b5b6

2666666664

3777777775þXqi¼1

1−Bð Þ

c11;i c12;i c13;i c14;i c15;i c16;ic21;i c22;i c23;i c24;i c25;i c26;ic31;i c32;i c33;i c34;i c35;i c36;ic41;i c42;i c43;i c44;i c45;i c46;ic51;i c52;i c53;i c54;i c55;i c56;ic61;i c62;i c63;i c64;i c65;i c66;i

2666666664

3777777775

CO2t−i

Ft−i

Yt−i

Y2t−i

Et−i

Tt−i

26666666664

37777777775

þ

δ1δ2δ3δ4δ5δ6

2666666664

3777777775ECTt−1½ � þ

γ1t

γ2t

γ3t

γ4t

γ5t

γ6t

2666666664

3777777775

ð4Þ

where (1 − B) is the lag operator, ECT is the lagged error-correctionterm and yts serially independent random errors with mean zero andfinite covariance matrix.

The VECM allows us to capture both the short-run and long-runGranger causality. The short-run causal effects can be obtained by theF-test of the lagged explanatory variables, while the t-statistics on thecoefficient of the lagged error correction term indicates the significanceof the long-run causal effect.

4. Empirical results and discussion

4.1. Unit root tests

Table 1 reports the unit root results from the two-and one-break LMtests.We tested each variable for a unit root using the two-break LM testat the 1-, 5- and 10 percent levels of significance. As noted above, whenthis test showed that only one structural break is significant weemployed the one-break LM test at the same levels of significance. Inorder to determine the number of lags, we used a “general to specific”procedure at each combination of break points for the two-break test,and at each single break point for the one-break test.

As shown in Table 1, the unit root hypothesis with two structuralbreaks cannot be rejected for CO2 in level. Similar results were foundfor Y and T, all of which have experienced one break in their term struc-tures. However, if we take the first differences, the unit root null for all

Table 1Two/one-break minimum LM unit-root tests.

Level First Difference

t TB1 TB2 t TB1 TB2

CO2 −5.182 1986 2000 −7.310⁎⁎⁎ 1975 2000Y −3.055 1989 – −7.266⁎⁎⁎ 1990 –

T −4.665 1980 – −5.369⁎⁎⁎ 1981 –

Note: Results are based on the model C, which allows for two changes in the level andtrend of the series.⁎⁎⁎ Rejection of the null hypothesis at the 1% significance level.

Table 3The results of f-test for cointegration.

Model F-statistics Conclusion

F(CO2/F Y Y2 E T DU1986 DU2000) 2.731 Inconclusive

Note: The critical value ranges of F-statistics are 2.152–3.296, 2.523–3.829 and3.402–5.031 at 10%, 5% and 1% levels of significance, respectively, which are takenfrom the Appendix in Narayan (2005).

37M.A. Boutabba / Economic Modelling 40 (2014) 33–41

the series can be rejected at the 1% level, suggesting thereby that theyare integrated of order 1, i.e. I(1).

Table 2 presents the ADF, PP, KPSS and ADF-GRS test results for F, Y2

and E and, for which the results from the two- and one-break LM testsshowed no significant breaks. The results reveal that all the four testsalmost unanimously indicate that all variables are non-stationary intheir level data. However, the stationarity property is found in the firstdifference of the variables in 5% or 1% critical level.

While the first structural break for CO2 in 1986 is inexplicable interms of energy consumption, the second in 2000 reflects the steadyincrease of energy use since the late 1990s which contributes to theincrease of carbon emissions.

The break date of 1989 for Y shows an upward trend, which may berelated to the economic reforms undertaken by Rajiv Ghandi soon aftertaking over as Prime Minister in 1985. Reforms include the abolition oflicences for some industries, sale of shares in selected public enterprises,removal of price controls and establishment of the Stock ExchangeBoard of India.

The break date of 1983 for T coincides with the second oil shock,which deteriorated India's terms of trade as well as its balance-of-payment.

As none of the variables is integrated of order two, the ARDL boundsprocedure can be used to examine the existence of a long-run relation-ship in the following step.

4.2. Cointegration test results

Given the establishment of structural breaks in the CO2 series, Eq. (2)incorporates two dummy variables (DU1986 and DU2000). Thecointegration test under the bounds testing approach involves compar-ing the F-statistics against critical values. Given that the value of theF-statistic is sensitive to the number of lags imposed each time on thedifferenced variables (Bahmani-Oskooee and Nasir, 2004), we selectthe optimal order of lags of the model based on the Akaike Informa-tion (AIC) and the Schwarz–Bayesian(SBC) information criteria assuggested by Pesaran et al.(2001). The results of the lag selectioncriteria indicate that the optimal number of lags is one.

The calculated F-statistics, together with the critical values, arereported in Table 3. The F-test has a non-standard distribution thatdepends on four factors, namely (i) the order of variables included in

Table 2Conventional unit root tests.

ADF PP

Level First difference Level First difference

Test statistics Test statistics Test statistics Test statistics

F 2.016 −2.142⁎⁎ 2.920 −4.11⁎⁎⁎Y2 3.187 −2.980⁎⁎⁎ 5.779 −3.026⁎⁎⁎E 3.357 −2.523⁎⁎ 7.450 −2.354⁎⁎

Note: ADF: Augmented Dickey–Fuller test. PP: Phillips–Perron test. KPSS: Kwiatkowski–Phillipsand DF-GLS critical values are taken fromMacKinnon (1991). KPSS critical values are sourced fronull is stationarity.⁎⁎⁎ Rejection of the null hypothesis at the 1% significance level.⁎⁎ Rejection of the null hypothesis at the 5% significance level.

the ARDL model, (ii) the number of explanatory variables,(iii) whetherthe ARDL model includes an intercept and/or time trends, and (iv) thesample size.

The calculated F-statistic F(CO2/F Y Y2 E T DU1986 DU2000) = 2.731lies between the lower and upper bounds of critical values, indicatingthat it is inconclusive whether or not the null hypothesis of nocointegration relationship should be rejected. In this case, as mentionedearlier, the error-correction term is a useful way of establishingcointegration (Banerjee et al., 1998; Kremers et al., 1992).

4.3. Long- and short-run elasticities

TheARDL cointegration procedurewas implemented to estimate theparameters of the Eq. (2). The AIC criterion has been utilized to find thecoefficients of the level variables. Because, AIC is known as parsimoni-ousmodel, as selecting the smallest possible lag length and it minimizesthe loss of degree of freedom as well.

The long-run results are reported in Table 4. Except for the coeffi-cient of T in the model, all estimated coefficients are statistically signif-icant and have correct signs as expected.

The results indicate that financial development has a long-run posi-tive impact onper capita CO2 emissions. A 1% increase in domestic creditto private sector will lead to about 0.182% increase in per capita CO2

emissions, which is significant at the 1% level. This suggests that finan-cial development improves environmental degradation. This finding isdifferent from Tamazian et al. (2009), whose study uses panel dataover period 1992–2004 and reveal that higher degree of financial devel-opment decreases the environmental degradation in the BRIC countries.However, our result lends support to Zhang (2011), who note that thebank loans provide solid support for Chinese companies to access exter-nal finance and enhance investment scale. This boosts economic growthand carbon emissions which depend on the bank asset scale expansion.

Both linear and non-linear terms of real GDP provide evidence insupporting the inverted-U relationship between economic growth andCO2 emissions. The result indicates that a 1% rise in real GDP will raiseCO2 emissions by 11.984% at the 1% significance level while a negativesign of squared term seems to corroborate the delinking of CO2 emissionsand real GDP at the higher level of income. These evidences support theEKC hypothesis, revealing that CO2 emissions increase in the initialstage of economic growth and decline after a threshold point, i.e.19,380 Indian rupees. This finding is consistent with Jayanthakumaranet al. (2012) and Kanjilal and Ghosh (2013), whose studies do not in-clude financial development, and with those various studies in which

KPSS ADF-GLS

Level First difference Level First difference

Test statistics Test statistics Test statistics Test statistics

0.659⁎⁎ 0.167 1.569 −3.123⁎⁎⁎0.729⁎⁎⁎ 0.352 1.272 −1.933⁎⁎⁎0.744⁎⁎⁎ 0.331 0.520 −4.635⁎⁎⁎

–Schmidt–Shin. ADF-GLS: Elliot–Rothenberg–Stock Dickey–Fuller GLS detrended. ADF, PPmKwiatkowski et al. (1992). All null hypotheses except KPSS are unit root; while, in KPSS

Table 4Long-run estimation results.

Regressor Coefficient

F 0.182⁎⁎⁎(0.036)

Y 11.984⁎⁎⁎(1.513)

Y2 −0.607⁎⁎⁎(0.071)

E 1.704⁎⁎⁎(0.294)

T 0.047(0.042)

DU1986 0.001(0.032)

DU2000 −0.005(0.030)

Constant −70.140⁎⁎⁎(6.631)

Note: The numbers in parentheses are standard errors.⁎⁎⁎ Significant at the 1% level.

38 M.A. Boutabba / Economic Modelling 40 (2014) 33–41

the relationship between GDP growth and CO2 emissions is examinedsuch as Song et al. (2008), Halicioglu (2009), Fodha and Zaghdoud(2010), Lean and Smyth (2010) and Ozturk and Acaravci (2013) andShahbaz et al. (2012).

The results show that energy consumption has as expected a long-run positive impact on per capita CO2 emissions. A 1% increase in energyconsumption will lead to a 1.704% increase in the CO2 emissions, whichis significant at the 1% level. However, the sign of trade openness is pos-itive but not significant in the long run. These results are consistentwithJayanthakumaran et al. (2012) and Kanjilal and Ghosh (2013) findingsfor India.

The short run dynamics results are reported in Table 5. The coeffi-cient on the lagged error correction term is significant with the correctsign at the 1% level, supporting the evidence of a stable long-run rela-tionship among the variables. The coefficient of −0.664 suggests thata deviation from the long run equilibrium level of CO2 emissions inone year is corrected by 64% over the following year.

The elasticity of CO2 emissions with respect to openness ratio or tofinancial development in the short run is negative but not statisticallysignificant. The finding on the insignificance of the openness trade ratiois consistent with that of Jalil and Mahmud (2009) for China. The signsof coefficients of Y and Y2 support again the EKC hypothesis in the shortrun and are significant at 1% level respectively. The short-run elasticityof CO2 emissions,with respect to energy consumption, is 1.003 and statis-tically significant at 1% level. It implies that a 1% increase in energyconsumption will raise CO2 emissions by 1.003% over the short run.

The diagnostic tests for the model are as presented at the bottom ofTable 5. Only the normality is violated and there is no neglected auto-correlation or heteroscedasticity present in the residuals of the modelacross the sample period. Hence, the outcome of the diagnostic tests in-dicates that the model have the desired econometric properties.

The graphs representing the CUSUM and CUSUMof squares tests arepresented in Figs. 1 and 2. As can be seen from the following figures, theplots of CUSUM and CUSUMSQ statistics are well within the criticalbounds, implying that all coefficients in the error-correction model arerelatively stable.

Therefore, this estimated model can be used for policy decision-making purposes such that the impact of policy changes consideringfinancial development, income, energy use, and openness trade willnot cause a major distortion in the level of CO2 emissions, since the pa-rameters in this model seem to follow a stable pattern during the esti-mation period.3

3 In order to check themodel robustness, the energy variable is dropped from the equa-tion since it may explainmost of the CO2 emissions. The exclusion of energy variable doesnot affect the findings. These results are available upon request.

4.4. Granger causality results

The existence of a cointegrating relationship among CO2 emissions,financial development, income, energy consumption and trade suggeststhat theremust be Granger causality in at least one direction, but it doesnot indicate the direction of temporal causality between the variables.We examine both short-run and long-run Granger causality in this sec-tion. As the lag order of Eq. (4) is 1, significance of the differenced vari-ables can be measured directly through the corresponding t-statistics.Table 6 presents results of Granger causality in the short-and long-run.

Beginning with the short-run effects, financial development, percapita real GDP, the square of per capita real GDP, per capita energy con-sumption and trade are non statistically significant in the carbon emis-sions equation. This implies that there is no short-run causalrelationship between per capita carbon emissions, financial develop-ment, per capita real GDP, the square of per capita real GDP, per capitaenergy consumption and trade. In the energy use equation, per capitacarbon emissions, financial development, per capita real GDP and thesquare of per capita real GDP are statistically significant, implying thatper capita carbon emissions, financial development, per capita realGDP and the square of per capita real GDP Granger cause per capita en-ergy consumption in the short run. In trade equation, per capita realGDP and per capita energy consumption are significant at 5% level,while per capita carbon emissions and financial development appearto be non statistically significant. In sum, in the short run there is unidi-rectional causality from per capita carbon emissions, financial develop-ment, per capita real income and the square of per capita real income toper capita energy use and per capita real income and energy use totrade.

Turning to the long-run causality result, empirical evidence indicat-ed that financial development Granger causes carbon emissions and en-ergy use. The unidirectional causality is found running from financialdevelopment to carbon emissions without a feedback. This confirmsthat financial development contributes in enhancing carbon emissionsby facilitating access to credit for companies4 whose investment projectare not necessarily environmentally friendly and for consumers pur-chasing high-value and carbon intensive items such as houses, cars,heating and cooling systems. Energy use is Granger caused by financialdevelopment. This supports the argument that offering affordable creditincreases economic and investment activities for companies and thepurchase of electrical appliances for individuals, which raises signifi-cantly the demand for energy.

The unidirectional causality running from per capita real income toCO2 without feedback implies that emission reduction policies will notrestrain economic growth and might be a viable policy tool for Indiato achieve its sustainable development in the long-run. This result dif-fers from Ghosh (2010) for India but it is comparable to that found bythe previous studies such as Jalil and Mahmud (2009) for China,Fodha and Zaghdoud (2010) for Tunisia, Lotfalipour et al.(2010) forIran, Nasir and Rehman (2011) for Pakistan, Saboori et al. (2012) forMalaysia and Ozturk and Acaravci (2013) for Turkey.

We find bidirectional causality between per capita energy consump-tion and per capita carbon emissions in the long run. This implies thatIndia should overhaul the energy structure by encouraging energyefficient technologies in order to reduce both losses and carbon emis-sions. This is particularly important because India's primary energymix is predominantly fossil fuel based and coal is the mainstay of theenergy sector — accounting for 42% of primary energy demand andover 80% of electricity generation in 2008. According to the Internation-al Energy Agency, India will double its coal consumption by 2035. That,in turn, means that carbon emissions will keep growing substantially.However, it is assumed that any attempt at dealing with atmospheric

4 According to the World Bank's Doing Business 2011 Report, with an overall rank of134, India was among the top 50 countries in terms of obtaining credit and protectinginvestors.

Table 5Short-run estimation results.

Regressor Coefficient

ΔF −0.023(0.067)

ΔY 7.340***(2.432)

ΔY2 −0.368***(0.124)

ΔE 1.003***(0.332)

ΔT −0.061(0.058)

ΔDU1986 −0.002(0.016)

ΔDU2000 0.013(0.016)

Constant 0.026**(0.010)

ECM(−1) −0.640***(0.155)

R2

0.614SE of regression 0.021

Diagnostic tests (p-value)Serial correlation 0.199Functional form 0.149Normality 0.095Heteroscedasticity 0.382

Note: The asterisks *** and ** are 1% and 5% significant levels,respectively. The numbers in parentheses are standard errors. Δis the first difference operator. Both the SE of regression andadjusted R2 are “goodness of fit measures”, where SE of regres-sion should be as small as possible and adjusted R2 be as closeto unity as possible. The serial correlation is tested by theLagrange Multiplier test of residual serial correlation (the null isno serial correlation). The functional form is based on theRamsey's Reset test using the square of the fitted values (thenull is no specification errors and is conducted for one fittedterm using LR). The normality test is based on a test of skewnessand kurtosis of residuals. The heteroscedasticity is tested by theWhite test with cross terms and null is no heteroscedasticity.

-10

-5

0

5

10

2001 2002 2003 2004 2005 2006 2007 2008

CUSUM 5% Significance

Fig. 1. Plot of CUSUM.

1.2

1.6

39M.A. Boutabba / Economic Modelling 40 (2014) 33–41

pollution requires increasing the use of alternative sources of energythat are relatively free from pollutant emissions. In this way, the signingof the Indo–U.S. nuclear deal5 in October 2008 has opened up opportu-nities for the growth of nuclear power in India. Indeed, the country aimsto increase its installed capacity from 4000MW to 63000MW by 2032.Moreover, within the National Action Plan for Climate Change (NAPCC)adopted in 2008, the national solar mission suggested an annual 1% in-crease in renewables' share of total electricity consumption in India forthe next 10 years — implying a 15% total share by 2020. Great impor-tance was particularly given to solar power, due to the fact that Indiais a tropical country, where sunshine is available for longer hours perday and in great intensity. Furthermore, the National Mission forEnhanced Energy Efficiency proposed several targets for 2014–2015:annual fuel savings of at least 23 Mtoe, a cumulative avoided electricitycapacity addition of 19,000MW and a CO2 emissionmitigation of 98Mt.

0.0

0.4

0.8

5. Conclusion

Climate change is a major challenge for developing countries likeIndia, which are exposed to greater risk from this phenomenon. Theclimate change concerns of India led to the formulation of the NationalAction Plan on Climate Change, which outlines eight missions that are

5 The nuclear industry's development has been hamstrung by India's refusal to sign theNuclear Non-Proliferation Treaty, cutting the country off from cooperation and assistancein civil nuclear technology. In 2008, India and the Nuclear suppliers' Group agreed on awaiver to the embargo on trade in nuclear technology.

adaptive as well as mitigative in nature. As part of international mitiga-tion efforts, India registered with the UNFCCC its voluntary endeavourto reduce the emissions intensity of its GDP by 20–25% by 2020 in com-parison to the 2005 level even as it pursues the path of inclusive growth.Hence, it is important to better understand the causes of the greenhousegas emissions for India in order to tackle these pollutant emissions andto ensure the sustainability of the economic development.

This paper examines the long-run equilibrium and the existence anddirection of a causal relationship between carbons emissions, financialdevelopment, economic growth, energy consumption and trade open-ness for India during the period 1971–2008. Our main contribution tothe literature on Indian studies is the analysis of the impact of the finan-cial development on carbon emissions. To examine this relationship, weuse a two-step procedure: In the first step, we explore the cointegrationbetween the variables by using ARDL bounds testing approach. Second-ly, we employ a dynamic VEC model to test causal relationshipsbetween these variables as well as stability tests.

The results suggest that there is evidence on the long-run and causalrelationships between per capita CO2 emissions, financial development,per capita real GDP, the square of per capita real GDP, per capita energyuse and trade openness. But probably the most interesting finding isthat financial development has a long-run positive impact on per capita

-0.42001 2002 2003 2004 2005 2006 2007 2008

CUSUM of Squares 5%Significance

Fig. 2. Plot of CUSUMSQ.

Table 6Granger causality results.

DependentVariables

Sources of causation

Short-run Long-run

ΔCO2 ΔF ΔY ΔY2 ΔE ΔT δi

ΔCO2 – 0.078 −6.649 0.347 0.616 −0.111 −0.229***(0.080) (4.304) (1.563) (0.465) (0.071) (0.079)

ΔF 0.282 – 7.195 −0.343 −0.194 0.194 0.227(0.401) (0.755) (0.492) (1.030) (1.228) (0.175)

ΔY 0.171 −0.054 – 0.302 0.492 0.098 0.006(0.194) (0.086) (1.267) (0.499) (0.077) (0.085)

ΔY2 3.183 −0.974 −127.561 – 8.804 1.994 0.079(3.787) (1.667) (90.081) (9.733) (1.493) (1.656)

ΔE −0.216*** 0.055* −4.101** 0.225** – −0.015 −0.084***(0.070) (0.031) (1.666) (0.086) (0.028) (0.031)

ΔT 0.012 −0.293 – 1.315 −2.799** – −0.215(0.463) (0.204) 25.130** (0.568) (1.191) (0.203)

(11.021)

Notes: ***, ** and * indicate that the null hypothesis of no causation is rejected at the 1%, 5% and 10% significance levels, respectively. The numbers in parentheses are standard errors. Δ isthe first difference operator. The number of appropriate lag is one according to Akaike information criterion, Schwarz information criterion and Hannan–Quinn information criterion.

40 M.A. Boutabba / Economic Modelling 40 (2014) 33–41

CO2 emissions, suggesting that financial development improves environ-mental degradation. Furthermore, the Granger causality test indicateslong-run unidirectional causality running from financial developmentto carbon emissions and energy use. The policy advice is therefore thatfinancial system should take into account the environment aspect intheir current operations. For example, banking systemmay encourage in-vestments in energy efficient technology by offering interest discountsand including carbon related conditions in their financial products suchas business vehicle and investment real estate term loans. Hence, a setof practical policies and incentives that promote more low-carbon fi-nance is an important part of building up India's resource-conservingsociety.

Causality tests also indicate that there was a long-run unidirectionalGranger causality running from per capita real income to per capitacarbon emissions in the long-run, without feedback. Moreover, bidirec-tional causality between per capita energy consumption and per capitacarbon emissions is found in the long run. The evidence seems to sug-gest that emission reduction policies will not restrain economic growthand might be a viable policy tool for India to achieve its sustainable de-velopment in the long-run. This would require that India adopt alterna-tives sources of supply and increase energy efficiency across the energyvalue chain. In this respect, India has opened up gas reserves for explo-ration and production by private and foreign firms under the OpenAcreage Licensing Policy and has fixed a target of 15% renewable contri-bution to the electricity generationmix by 2020. Moreover, increases inenergy efficiency are being targeted by the National Mission on En-hanced Energy Efficiency. However, the success of installed renewablecapacity differs markedly on a state by state basis (World EconomicForum, 2012). States with large growth rates tend to have burgeoningrenewable sectors – e.g. Gujarat and Maharashtra – while those withlow growth rates, particularly those in the north-east, do not have anestablished renewable sector. This suggests that India should alsoconsider creating a unified energy regulator to help align incentives be-tween the states and the central government. Furthermore, Graus et al.(2007) and Chikkatur (2008) outlined that the thermal efficiency ofcoal power plants in India is about 29–30%, while in developed coun-tries a much higher level is achieved. This implies that India shouldadopt new technologies to improve the energy efficiency of powergeneration.

While the above analysis has provided interesting insights, it shouldbe noted that the development of efficient energy policies, that willcontribute to curb carbon emissionswhile preserving economic growth,needs to consider other variables than the underlying factors in our re-search. A promising extension of this work would be to consider the se-curity of energy supply, rural development concerns, urbanization andother environmental variables in the case of India.

Acknowledgements

I would like to thank anonymous referees as well as participantsto the 62nd Congress of the French Economics Association (AFSE)(Aix-Marseille University, June 24–26, 2013) for their helpful commentsand suggestions. The usual disclaimers apply.

References

Alam,M.J., Begun, I.A., Buysse, J., Rahman, S., Huylenbroeck, G.V., 2011. Dynamicmodelingof causal relationship between energy consumption, CO2 emissions and economicgrowth in India. Renew. Sust. Energ. Rev. 15, 3243–3251.

Ang, J., 2008. Economic development, pollutant emissions and energy consumption inMalaysia. J. Policy Model 30, 271–278.

Bahmani-Oskooee, M., Nasir, A., 2004. ARDL approach to test the productivity biashypothesis. Rev. Dev. Econ. 8 (3), 483–488.

Banerjee, A., Dolado, J.J., Mestre, R., 1998. Error-correction mechanism tests forcointegration in a single-equation framework. J. Time Ser. Anal. 19 (3), 267–283.

Brown, R.L., Durbin, J., Evans, J.M., 1975. Techniques for testing the constancy of regres-sion relationships over time. J. R. Stat. Soc. 37, 149–163.

Chikkatur, A., 2008. A Resource and Technology Assessment of Coal Utilisation in India, PewCentre of Global Climate Change, Arlington, VA, USA. see http://www.pewclimate.org/docUploads/india-coal-technology.pdf.

Claessens, S., Feijen, E., 2007. Financial sector development and the millennium develop-ment goals. World Bank Working Paper No. 89.

Engle, R.F., Granger, C.W.J., 1987. Cointegration and error correction representation: esti-mation and testing. Econometrica 55, 251–276.

Fodha, M., Zaghdoud, O., 2010. Economic growth and pollutant emissions in Tunisia: anempirical analysis of the Environmental Kuznets Curve. Energy Policy 38, 1150–1156.

Ghosh, S., 2010. Examining carbon emissions–economic growth nexus for India: a multi-variate cointegration approach. Energy Policy 38, 2613–3130.

Graus, W.H.J., Worrell, M., Voogt, E., 2007. International comparison of energy efficiencyof fossil power generation. Energy Policy 35, 3936–3951.

Grossman, G., Krueger, A., 1995. Economic environment and the economic growth. Q. J.Econ. 110, 353–377.

Halicioglu, F., 2009. An econometric study of CO2 emissions, energy consumption, incomeand foreign trade in Turkey. Energy Policy 37 (3), 1156–1164.

Intergovernmental Panel on Climate Change (IPCC), 2007. Climate Change SynthesisReport 2007. http://www.ipcc.ch/S.

Jalil, A., Feridun, M., 2011. The impact of growth, energy and financial development on theenvironment in China: a cointegration analysis. Energy Econ. 33, 284–291.

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

Jayanthakumaran, K., Verma, R., Liu, Y., 2012. CO2 emissions, energy consumption, tradeand income: a comparative analysis of China and India. Energy Policy 42, 450–460.

Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference oncointegration — with applications to the demand for money. Oxf. Bull. Econ. Stat.52, 169–210.

Kanjilal, K., Ghosh, S., 2013. Environmental Kuznet's curve for India: evidence from testsfor cointegration with unknown structural breaks. Energy Policy 56, 509–515.

Kapetanios, G., 2005. Unit root testing against the alternative hypothesis of m structuralbreaks. J. Time Ser. Anal. 26, 123–233.

Kremers, J.J., Erksson, N.R., Dolado, J.J., 1992. The power of cointegration tests. Bull. Econ.Stat. 54 (3), 325–347.

Laurenceson, J., Chai, J., 2003. Financial Reform and Economic Development in China. UK,Edward Elgar, Cheltenham.

Lean, H.H., Smyth, R., 2010. CO2 emissions, electricity consumption and output in ASEAN.Appl. Energy 87, 1858–1864.

41M.A. Boutabba / Economic Modelling 40 (2014) 33–41

Lee, J., Strazicich,M., 2003.MinimumLagrangemultiplier unit root testwith two structuralbreaks. Rev. Econ. Stat. 85 (4), 1082–1089.

Lee, J., Strazicich,M., 2004.MinimumLagrangemultiplier unit root testwith one structuralbreak. Working paper. Department of Economics, Appalachian State University, USA.

Lindmark, M., 2002. An EKC-pattern in historical perspective: carbon dioxide emis-sions, technology, fuel prices and growth in Sweden, 1870–1997. Ecol. Econ.42, 333–347.

Lotfalipour, M.R., Falahi, M.A., Ashena, M., 2010. Economic growth, CO2 emissions, andfossil fuels consumption in Iran. Energy 35, 5115–5120.

Lumsdaine, R., Papell, D., 1997. Multiple trends breaks and the unit root hypothesis. Rev.Econ. Stat. 79, 212–218.

MacKinnon, J.G., 1991. Critical Values for Co-Integration Tests. In: Engle, R.F., Granger, C.W.J.(Eds.), Long-Run Economic Relationships. Oxford University Press, pp. 267–276.

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

Nasir, M., Rehman, F.-U., 2011. Environmental Kuznets curve for carbon emissions inPakistan: an empirical investigation. Energy Policy 39, 1857–1864.

Ouattara, B., 2004. Foreign Aid and Fiscal Policy in Senegal. Mimeo University ofManchester.

Ozturk, I., Acaravci, A., 2013. The long-run and causal analysis of energy, growth, open-ness and financial development on carbon emissions in Turkey. Energy Econ. 36,262–267.

Pesaran, M., Pesaran, B., 1997. Working with Microfit 4.0: Interactive Economic Analysis.Oxford University Press, Oxford.

Pesaran, H.M., Shin, Y., 1999. Autoregressive distributed lag modelling approach tocointegration analysis. In: Storm, S. (Ed.), Econometrics and Economic Theory inthe 20th Century: The Ragnar Frisch Centennial Symposium. Cambridge UniversityPress, Cambridge.

Pesaran, M.H., Shin, Y., Smith, R.J., 2001. Bounds testing approaches to the analysis of levelrelationships. J. Appl. Econ. 16, 289–326.

Saboori, B., Suleiman, J., Mohd, S., 2012. Economic growth and CO2 emissions in Malaysia: acointegration analysis of the environmental Kuznets curve. Energy Policy 51, 184–191.

Shahbaz, M., Lean, H.H., Shabbir, M.S., 2012. Environmental Kuznets curve hypothesisin Pakistan: cointegration and Granger causality. Renew. Sust. Energ. Rev. 16 (5),2947–2953.

Song, T., Zheng, T., Tong, L., 2008. An empirical test of the environmental Kuznets curve inChina: a panel cointegration approach. China Econ. Rev. 19, 381–392.

Stern, D.I., Common, M.S., Barbier, E.B., 1996. Economic growth and environmental degra-dation: the environmental Kuznets curve and sustainable development. World Dev.24 (7), 1151–1160.

Talukdar, D., Meisner, C.M., 2001. Does the private sector help or hurt the environment?Evidence from carbon dioxide pollution in developing countries. World Dev. 29,827–840.

Tamazian, A., Bhaskara Rao, B., 2010. Do economic, financial and institutional develop-ments matter for environmental degradation? Evidence from transitional economies.Energy Econ. 32, 137–145.

Tamazian, A., Chousa, J.P., Vadlamannati, C., 2009. Does higher economic and financialdevelopment lead to environmental degradation: evidence from the BRIC countries.Energy Policy 37, 246–253.

World Bank, k. Doing Business 2011: Making a difference for Entrepreneurs. http://www.doingbusiness.org//media/fpdkm/doing%20business/documents/annual-reports/english/db11-fullreport.pdf.

World Economic Forum, 2012. New Energy Architecture: India. http://www3.weforum.org/docs/WEF_NewEnergyArchitecture_ExecutiveSummary_India.pdf.

Yuxiang, K., Chen, Z., 2011. Financial development and environmental performance: evi-dence from China. Environ. Dev. Econ. 16, 1–19.

Zhang, Y.J., 2011. The impact of financial development on carbon emissions: an empiricalanalysis in China. Energy Policy 39, 2197–2203.

Zivot, E., Andrews, D., 1992. Further evidence on the great crash, the oil price shock andthe unit root hypothesis. J. Bus. Econ. Stat. 10, 251–270.