.


International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017356

International Journal of Energy Economics and 
Policy

ISSN: 2146-4553

available at http: www.econjournals.com

International Journal of Energy Economics and Policy, 2017, 7(3), 356-363.

Revisiting the Income, Energy Consumption and Carbon 
Emissions Nexus: New Evidence from Quantile Regression for 
Different Country Groups

Yaya Keho*

Ecole Nationale Supérieure de Statistique et d’Economie Appliquée Abidjan, Côte d’Ivoire. *Email: yayakeho@yahoo.fr

ABSTRACT

The environment Kuznets curve (EKC) hypothesis has been widely tested in the energy economics literature. However, previous studies focused 
only on the mean effects and have not yet examined the role of pollution levels in the income-pollution nexus. This paper uses quantile regression 
to reexamine the effect of economic growth and energy consumption on dioxide carbon (CO2) emissions for five panels of 59 countries. The results 
reveal that energy consumption increases CO2 emissions in all panels, the effect being larger in low pollution countries. They also provide evidence 
supporting the EKC hypothesis for Sub-Saharan, American and European countries at all quantiles, and for Asian and MENA countries at lower 
levels of CO2 emissions. These findings suggest that economic growth is not everywhere and always the cause and the cure of pollution. Therefore, 
environmental control policies should be tailored differently across low and high pollution countries.

Keywords: Carbon Dioxide Emissions, Economic Growth, Energy Consumption, Quantile Regression 
JEL Classifications: C23, F18, O58, Q53

1. INTRODUCTION

Since the pioneering works by Grossman and Krueger (1991), 
Holtz-Eakin and Selden (1992), Selden and Song (1994) 
and Shafik (1994), the environmental Kuznets curve (EKC) 
hypothesis has been an active field of empirical research in 
energy economics. This was a result of the emergence of the 
literature on unit root and cointegration techniques. The EKC 
hypothesis describes an inverted U-shaped relationship between 
income and environmental degradation. That is, environmental 
degradation increases as income increases and then declines 
after income exceeds a threshold level. The main motivation for 
testing this relationship is that it allows policy makers to judge the 
response of the environment to economic growth. Furthermore, 
the EKC hypothesis suggests that growth is the cause and the 
cure of air pollution. To curb the pollution level countries have 
to accelerate economic growth to surpass the income threshold 
level. However, the empirical evidence regarding the validity of the 
EKC hypothesis has yielded mixed results (Dinda, 2004; Ozturk, 
2010; Smith and Narayan, 2015, for a review). For example, while 
some studies found evidence supporting the EKC hypothesis 

(Dinda and Coondoo, 2006; Apergis and Payne, 2009; 2010; Jalil 
and Mahmud, 2009; Lamla, 2009; Iwata et al., 2010; Lean and 
Smyth, 2010; Pao and Tsai, 2010; Shahbaz et al., 2013; Shahbaz 
et al., 2014; Al-Mulali et al., 2015a; Apergis and Ozturk, 2015; 
Jebli et al., 2016, Zambrano-Monserrate et al., 2016; Rosado, 
2017), others found no support for this hypothesis or mixed results 
(Dinda and Coondoo et al., 2000; Harbaugh et al., 2002; Perman 
and Stern, 2003; Martinez-Zarzoso and Bengochea-Morancho, 
2004; Galeotti, et al., 2006; Aslanidis and Iranzo, 2009; Al-Mulali 
et al., 2015b; Baek, 2015; Yang et al., 2015; Al-Mulali et al., 2016; 
Ozturk and Al-Mulali, 2015; Shahbaz et al., 2015; Gill et al., 2017).

One problem with the existing studies is that they have relied on 
mean-based regression approaches such as ordinary least squares 
(OLS) or instrumental variables, which implicitly assumes that 
the impact of income along the distribution of carbon dioxide 
(CO2) emissions is the same. However, theory does not provide 
any guide as to why we should focus the analysis on the mean 
effects only. Further, the assumption of a mean shift is a testable 
hypothesis in a statistical framework. While estimating how “on 
average” income affects CO2 emissions yields straightforward 



Keho: Revisiting the Income, Energy Consumption and Carbon Emissions Nexus: New Evidence from Quantile Regression for Different Country Groups

International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017 357

interpretations, the standard methodology may miss what is 
crucial for policy purposes, namely how economic growth affects 
emissions at different points of the distribution of CO2 emissions. 
For example, while increases in income may not matter for average 
CO2 emissions levels, it would be useful to know if it matters for 
countries at the tails of the pollution distribution. It is likely that 
the effect of income on CO2 emissions is different in high and 
low pollution countries. In this paper, we are interested to know 
where economic growth matter not just whether or not it matters 
on average.

Unlike the previous studies, this paper contributes to the 
EKC literature by using the quantile regression methodology 
developed by Koenker and Bassett (1978). This approach allows 
us to investigate possible differences in the effect of income 
on CO2 emissions at different points of the distribution of CO2 
emissions. Using this approach, we also address the issue of 
parameter heterogeneity in the pollution-income relationship. 
CO2 is considered to be the major cause of global warming and 
has been the most widely used pollutant indicator in the EKC 
literature. We focus on this pollutant to reassess the validity of the 
EKC for 59 countries divided into five panels. To our knowledge, 
the quantile regression approach has not been applied to test the 
EKC hypothesis.

The remainder of the article is organized as follows. Section 
2 outlines the estimation methodology and describes the data. 
Section 3 discusses the empirical results, while Section 4 
concludes.

2. MODEL, DATA AND METHODOLOGY

2.1. Empirical Model
Following the empirical literature (Apergis and Payne, 2009; Lean 
and Smyth, 2010; Pao and Tsai, 2010), we specify the econometric 
model as follows:

2
üüüüüüüCO = + E + I + I +θ θ θ θ µ  (1)

where i is for country i in the panel, t refers to the time period, 
CO2 stands for per capita carbon dioxide emissions, E is energy 
use per capita, I refers to income measured by real gross domestic 
product (GDP) per capita. It is expected that higher energy use 
results in more CO2 emissions. Therefore the expected sign of θ1 
is positive. Under the EKC hypothesis, the sign of θ2 is expected 
to be positive whereas a negative sign is expected for θ3. If θ3 is 
statistically insignificant, it indicates a monotonic increase in the 
relationship between CO2 emissions and income.

2.2. The Quantile Regression Approach
Previous studies use standard methods (OLS, instrumental 
variables, generalized method of moments and generalized least 
squares) that deliver the average effects of explanatory variables 
over the whole distribution of the dependent variable. However, 
in this study we are interested in studying the impact of income on 
the entire distribution of CO2 emissions. Therefore we rely on the 
quantile regression method which was first introduced by Koenker 
and Bassett (1978) and discussed in further works (Koenker and 

Machado, 1999; Koenker and Hallock, 2001). This method has 
two main advantages. First, compared to OLS regression, it is 
more robust to outliers and to non-normal distribution. Second, it 
allows for the estimation of the effect of income at different points 
of the distribution of CO2 emissions.

The quantile regression model can be formulated as follows:

2
2it t 0 1 it 2 it 3 it itq(CO / ) = + E + I + I +τ τ τ τΩ θ θ θ θ µ  (2)

Where q(CO2it/Ωit) is the conditional quantile of CO2 and Ωt 
contains the available information known at time t. Equation 2 
can be written as follows:

yit = xitθτ+εit (3)

Whereis ( )2it it it, itx = 1,E ,I ,I the vector of explanatory variables; 
θτ are the k×1 regression coefficients at the τ

th quantile of the 
dependent variable y.

Contrary to OLS which is based on minimizing the sum of squared 
residuals, the τth quantile regression estimator of θ minimizes an 
asymmetrically weighted sum of absolute errors:

Min y x 1 y x

=mi

it it

y ³x

it it

y xit it it it
θ

τ
θτ

τ
ξ θ

τ − θ + −τ − θ
τ

∑ ∑ ( )










nn y x
t=1

T

it it
θ

τ τϕ − θ∑ ( )
 (4)

Where ϕτ (z) is a loss function defined as ϕτ(z) = |z|+(2τ−1)z, 0<τ<1.

The quantile regression method allows the marginal effects 
of covariates to change at different points in the conditional 
distribution of CO2 emissions by estimating θτ using different 
values of τ. It is in this way that quantile regression allows for 
parameter heterogeneity in the pollution-income nexus.

2.3. Data and Descriptive Statistics
The empirical analysis uses data for 59 countries divided into 
five groups: 15 Sub-Saharan African countries, 13 American 
countries, 10 Asian countries, 13 European countries, and 8 
MENA member countries. The list of countries is presented in 
Table 1. The countries were chosen based on data availability. 
We use annual time series for real GDP per capita expressed 
in constant 2000 US dollar, per capita energy consumption 
in kg oil equivalent and per capita CO2 emissions measured 
in metric tons. All the data are obtained from the 2015 World 
Development Indicators by the World Bank. The sample period 
varies across regions and has been dictated by availability of 
the data for all the series. All the data were converted into 
natural logarithms.

Table 2 gives some descriptive statistics of the data. As can be 
seen, there are some variations in the data within and across 
groups. European countries have greater per capita income, 



Keho: Revisiting the Income, Energy Consumption and Carbon Emissions Nexus: New Evidence from Quantile Regression for Different Country Groups

International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017358

are greater consumers of energy and pollute more, followed by 
American and MENA countries. The Kurtosis exceeds 3 in most 
cases suggesting that series have heavy tails. This shows that 
data is not normal which is also proved with the Jarque-Bera test 
statistic. Therefore, estimation technique based on linear Gaussian 
models will be biased, hence it more appropriate to use quantile 
regression technique.

3. RESULTS AND DISCUSSION

Tables 3-7 display both pooled OLS and quantile regression 
estimates. OLS estimates provide a baseline of mean effects, 
and we compare these to estimates for separate quantiles in the 
conditional distribution of CO2 emissions. We report results 
for the 10th, 25th, 50th, 75th and 90th quantiles. In order to obtain 

Table 1: List of countries and sample period
Regions Countries Sample period
Sub-Saharan Africa Benin, Cameroon, Congo democratic, Congo Republic, Cote d’Ivoire, Gabon, Ghana, Kenya, 

Mauritius, Nigeria, Senegal, South Africa, Togo, Zambia, Zimbabwe
1976-2011

America Argentina, Bolivia, Brazil, Canada, Chile, Colombia, Ecuador, Guatemala, Honduras, Mexico, Peru, 
Uruguay, Venezuela

1971-2011

Asia Bangladesh, China, India, Indonesia, Malaysia, Pakistan, Philippines, Singapore, Sri Lanka, Thailand 1971-2011
Europe Austria, Finland, Greece, Turkey, Belgium, Denmark, France, Italy, Luxembourg, Portugal, Spain, 

Sweden, UK
1971-2011

MENA Algeria, Egypt, Iran, Jordan, Morocco, Oman, Saudi Arabia, Tunisia 1976-2011

Table 2: Descriptive statistics
Variables Observe Mean±SD Minimum Maximum Kurtosis Skewness JB
Panel A: Sub-Saharan African countries

CO2 emissions 540 −0.652±1.221 −3.391 2.387 3.536 0.807 65.21
Energy use 540 6.280±0.627 5.336 8.000 3.655 1.153 129.44
GDP 540 6.899±0.946 5.279 9.514 2.857 0.945 80.87
GDP squared 540 48.502±13.931 27.872 90.527 3.133 1.125 114.43

Panel B: American countries
CO2 emissions 533 0.596±0.937 −0.885 2.901 3.167 0.792 56.37
Energy use 533 6.830±0.784 5.402 9.033 4.471 1.395 221.13
GDP 533 8.062±0.867 6.660 10.523 3.840 0.878 84.22
GDP squared 533 65.758±14.690 44.355 110.742 4.591 1.199 184.13

Panel C: Asian countries
CO2 emissions 410 0.082±1.255 −3.282 2.950 2.745 0.212 4.18
Energy use 410 6.364±0.910 4.449 8.905 3.304 0.529 20.77
GDP 410 7.003±1.191 5.013 10.495 3.722 1.069 87.04
GDP squared 410 50.463±18.297 25.135 110.156 4.673 1.437 189.00

Panel D: European countries
CO2 emissions 533 2.034±0.567 0.293 3.703 4.126 −0.135 29.817
Energy use 533 8.064±0.612 6.307 9.474 3.118 −0.532 25.522
GDP 533 10.052±0.602 8.081 11.382 4.628 −1.075 161.640
GDP squared 533 101.411±11.730 65.309 129.561 4.226 −0.845 96.968

Panel E: MENA countries
CO2 emissions 288 1.202±0.863 −0.458 3.005 2.267 0.359 12.628
Energy use 288 6.951±0.845 5.457 8.981 2.488 0.541 17.219
GDP 288 8.028±0.902 6.271 9.998 2.324 0.567 20.939
GDP squared 288 65.265±14.968 39.321 99.977 2.391 0.724 29.657

JB refers to the χ2 statistic from the Jarque-Bera test of normality, with P values in parentheses, GDP: Gross domestic product, SD: Standard deviation

Table 3: Determinants of CO2 emissions for Sub-Saharan African countries
Variables OLS Quantile regression Test of symmetry1 Test of equality2

q10 q25 q50 q75 q90 q10=q90 q25=q75
Energy use 1.160*  

(21.85)
1.207*  
(11.73)

1.305*  
(18.53)

1.239*  
(24.32)

1.205*  
(23.63)

0.930*  
(20.81)

6.58*  
(0.010)

1.96  
(0.162)

12.11*  
(0.000)

GDP 2.911*  
(7.07)

3.248*  
(11.57)

2.982*  
(7.17)

1.654*  
(2.29)

2.380*  
(4.98)

1.911*  
(3.74)

5.70*  
(0.017)

1.36  
(0.243)

1.99**  
(0.094)

GDP squared −0.162*  
(−5.63)

−0.191*  
(−9.00)

−0.160*  
(−5.10)

−0.077**  
(−1.57)

−0.133*  
(−4.05)

−0.094*  
(−2.75)

6.14*  
(0.013)

0.52  
(0.469)

2.14**  
(0.074)

Constant −20.136*  
(−12.64)

−22.05*  
(−18.57)

−21.976*  
(−15.84)

−16.014*  
(−5.80)

−17.827*  
(−10.15)

−14.455*  
(−7.36)

11.70*  
(0.000)

5.34*  
(0.021)

3.81*  
(0.004)

The numbers in parentheses are t-statistics computed from heteroskedasticity-robust standard errors. Quantile regression results are based upon 1000 bootstrapping repetitions. The 
asterisks ** and * denote significance at the 10% and 5% levels, respectively. 1F-statistic and associated P values for symmetry test. 2F-statistic and associated P values are reported for the 
test of equality of the coefficients across quantiles (i.e., q10=q25=q50=q75=q90), OLS: Ordinary least squares, GDP: Gross domestic product



Keho: Revisiting the Income, Energy Consumption and Carbon Emissions Nexus: New Evidence from Quantile Regression for Different Country Groups

International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017 359

heteroskedasticity-robust estimates, we report robust t-statistics 
for OLS estimates and quantile regression results from 1000 

bootstrapping repetitions. Figures 1-5 represent the effects of 
income and energy use across the CO2 emissions distribution.

Table 4: Determinants of CO2 emissions for American countries
Variables OLS Quantile regression Test of symmetry1 Test of equality2

q10 q25 q50 q75 q90 q10=q90 q25=q75
Energy use 1.065*  

(22.83)
1.496*  
(24.29)

1.298*  
(27.44)

1.110*  
(28.18)

1.051*  
(28.13)

1.017*  
(20.38)

42.07*  
(0.000)

20.71*  
(0.000)

13.29*  
(0.000)

GDP 1.705*  
(8.45)

2.407*  
(11.75

2.778*  
(17.42)

1.637*  
(8.26)

1.346*  
(5.06)

0.773*  
(3.41)

32.38*  
(0.000)

27.43*  
(0.000)

18.80*  
(0.000)

GDP squared −0.095*  
(−7.73)

−0.151*  
(−11.36)

−0.164*  
(−16.04)

−0.095*  
(−8.70)

−0.079*  
(−6.02)

−0.048*  
(−3.60)

33.80*  
(0.000)

34.59*  
(0.000)

17.72*  
(0.000)

Constant −14.128*  
(−14.71)

−19.457*  
(−19.92)

−20.09*  
(−26.55)

−13.848*  
(−15.43)

−12.057*  
(−10.85)

−9.103*  
(−8.85)

61.02*  
(0.000)

47.18*  
(0.000)

21.16*  
(0.000)

The numbers in parentheses are t-statistics computed from heteroskedasticity-robust standard errors. Quantile regression results are based upon 1000 bootstrapping repetitions. 
The asterisks * denotes significance at the 5% level. 1F-statistic and associated P values for symmetry test. 2F-statistic and associated P values are reported for the test of equality of the 
coefficients across quantiles (i.e., q10=q25=q50=q75=q90), OLS: Ordinary least squares, GDP: Gross domestic product

Table 6: Determinants of CO2 emissions for European countries
Variables OLS Quantile regression Test of symmetry1 Test of equality2

q10 q25 q50 q75 q90 q10=q90 q25=q75
Energy use 0.812*  

(19.28)
0.916*  
(9.64)

0.696*  
(11.94)

0.730*  
(10.26)

1.017*  
(29.65)

0.994*  
(56.92)

0.65*  
(0.420)

29.58*  
(0.000)

11.59* 
(0.000)

GDP 1.023*  
(2.88)

5.146*  
(6.65)

2.359*  
(5.09)

1.536**  
(1.76)

0.934*  
(5.60)

1.668*  
(10.67)

18.75*  
(0.000)

10.13*  
(0.001)

12.88*  
(0.000)

GDP squared −0.051*  
(−2.75)

−0.285*  
(−6.74)

−0.118*  
(−4.46)

−0.074** 
(−1.68)

−0.049*  
(−5.91)

−0.085*  
(−11.18)

21.10*  
(0.000)

7.27*  
(0.007)

13.35*  
(0.000)

Constant −9.560*  
(−5.72)

−28.484*  
(−7.63)

−15.450*  
(−7.05)

−11.724*  
(−3.01)

−10.408*  
(−13.43)

−13.828*  
(−17.77)

14.22*  
(0.000)

5.58*  
(0.018)

10.79*  
(0.000)

The numbers in parentheses are t-statistics computed from heteroskedasticity-robust standard errors. Quantile regression results are based upon 1000 bootstrapping repetitions. The 
asterisks ** and * denote significance at the 10% and 5% levels, respectively. 1F-statistic and associated P values for symmetry test. 2F-statistic and associated P values are reported for the 
test of equality of the coefficients across quantiles (i.e., q10=q25=q50=q75=q90), OLS: Ordinary least squares, GDP: Gross domestic product

Table 7: Determinants of CO2 emissions for MENA countries
Variables OLS Quantile regression Test of symmetry1 Test of equality2

q10 q25 q50 q75 q90 q10=q90 q25=q75
Energy use 0.648*  

(12.35)
0.962*  
(29.99)

0.921*  
(39.60)

0.852*  
(39.34)

0.629*  
(10.49)

0.494*  
(7.79)

50.29*  
(0.000)

28.05*  
(0.000)

13.73* (0.000)

GDP 0.485**  
(1.93)

0.831*  
(3.79)

0.484*  
(1.98)

0.488*  
(2.24)

0.563*  
(1.98)

0.966*  
(3.14)

0.14  
(0.713)

0.07  
(0.791)

1.08 (0.365)

GDP squared −0.007  
(−0.52)

−0.051*  
(−3.61)

−0.026**  
(−1.74)

−0.020  
(−1.43)

−0.011  
(−0.68)

−0.028  
(−1.58)

1.12  
(0.291)

0.67  
(0.413)

1.35 (0.252)

Constant −6.690*  
(−6.68)

−9.028*  
(−9.78)

−7.507*  
(−7.85)

−7.376*  
(−8.37)

−6.809*  
(−6.76)

−7.899*  
(−7.95)

0.74  
(0.391)

0.42  
(0.518)

1.14 (0.336)

The numbers in parentheses are t-statistics computed from heteroskedasticity-robust standard errors. Quantile regression results are based upon 1000 bootstrapping repetitions. The 
asterisks ** and * denote significance at the 10% and 5% levels, respectively. 1F-statistic and associated P values for symmetry test. 2F-statistic and associated P values are reported for the 
test of equality of the coefficients across quantiles (i.e., q10=q25=q50=q75=q90), OLS: Ordinary least squares, GDP: Gross domestic product

Table 5: Determinants of CO2 emissions for Asian countries
Variables OLS Quantile regression Test of symmetry1 Test of equality2

q10 q25 q50 q75 q90 q10=q90 q25=q75
Energy use 1.522*  

(40.72)
2.043*  
(18.24)

1.841*  
(26.16)

1.567*  
(31.42)

1.422*  
(34.95)

1.391*  
(71.83)

33.71*  
(0.000)

37.22*  
(0.000)

14.19*  
(0.000)

GDP 0.807*  
(4.41)

1.066*  
(2.45)

0.866*  
(3.05)

0.356*  
(2.12)

0.057  
(0.16)

−0.021  
(−0.10)

5.02*  
(0.025)

3.96*  
(0.047)

2.15**  
(0.073)

GDP squared −0.064*  
(−4.88)

−0.107*  
(−3.95)

−0.085*  
(−3.94)

−0.038*  
(−3.05)

−0.008  
(−0.30)

0.006  
(0.40)

13.28*  
(0.000)

6.28*  
(0.012)

4.34*  
(0.001)

Constant −12.026*  
(−17.82)

−15.344*  
(−9.48)

−13.599*  
(−11.42)

−10.447*  
(−14.72)

−8.746*  
(−6.85)

−8.508*  
(−11.50)

14.85*  
(0.000)

10.54*  
(0.001)

5.26*  
(0.000)

The numbers in parentheses are t-statistics computed from heteroskedasticity-robust standard errors. Quantile regression results are based upon 1000 bootstrapping repetitions. The 
asterisks ** and * denote significance at the 10% and 5% levels, respectively. 1F-statistic and associated P values for symmetry test. 2F-statistic and associated P values are reported for the 
test of equality of the coefficients across quantiles (i.e., q10=q25=q50=q75=q90), OLS: Ordinary least squares, GDP: Gross domestic product



Keho: Revisiting the Income, Energy Consumption and Carbon Emissions Nexus: New Evidence from Quantile Regression for Different Country Groups

International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017360

First note the OLS results. As expected, energy consumption has 
positive impact on CO2 emissions in all panels, which is consistent 
with previous studies (Liu, 2005; Ang, 2007; Apergis and Payne, 
2009). Further, except for MENA countries, the estimated 
coefficients on income per capita and squared income per capita 

are both significant with a positive and negative sign, respectively. 
This result gives a support for the EKC hypothesis that the level 
of CO2 emissions initially increases with income, until it reaches 
a threshold level, then it declines. Hence, beyond a threshold level 
of real output, an increase in per capita GDP reduces emissions 

Figure 1: Ordinary least squares and quantile regression estimates over the conditional quantiles of CO2 emissions: Evidence for Sub-saharan 
African countries

Source: The x-axis represents the conditional quantile of CO2 emissions. The horizontal dashed line represents the OLS estimates. The two dotted 
lines depict the 95% confidence intervals for the OLS estimates. The solid line represents the quantile regression estimates; and the shaded grey 
area plots the 95% confidence band for the quantile regression estimates

Figure 2: Ordinary least squares and quantile regression estimates over the conditional quantiles of CO2 emissions: Evidence for American 
countries

Source: The X-axis represents the conditional quantile of CO2 emissions. The horizontal dashed line represents the ordinary least squares 
estimates (OLS). The two dotted lines depict the 95% confidence intervals for the OLS estimates. The solid line represents the quantile 
regression estimates; and the shaded grey area plots the 95% confidence band for the quantile regression estimates



Keho: Revisiting the Income, Energy Consumption and Carbon Emissions Nexus: New Evidence from Quantile Regression for Different Country Groups

International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017 361

as the demand for environmental quality increases and these 
economies grow.

The quantile regression results suggest some important differences 
across different points in the conditional distribution of CO2 
emissions. Energy consumption increases pollution in all quantile 

levels and its impact decreases with the pollution level in 
American, Asian and MENA countries, suggesting that the effect 
of energy consumption is larger in low-pollution countries. This 
result implies that more energy conservation policies will help to 
reduce CO2 emissions. It also implies that as incomes grow, in the 
absence of energy conservation policies, pollution will increase.

Figure 3: Ordinary least squares and quantile regression estimates over the conditional quantiles of CO2 emissions: Evidence for Asian countries

Source: The X-axis represents the conditional quantile of CO2 emissions. The horizontal dashed line represents the ordinary least squares 
(OLS) estimates. The two dotted lines depict the 95% confidence intervals for the OLS estimates. The solid line represents the quantile 
regression estimates; and the shaded grey area plots the 95% confidence band for the quantile regression estimates

Figure 4: Ordinary least squares and quantile regression estimates over the conditional quantiles of CO2 emissions: Evidence for 
European countries

Source: The X-axis represents the conditional quantile of CO2 emissions. The horizontal dashed line represents the ordinary least squares 
(OLS) estimates. The two dotted lines depict the 95% confidence intervals for the OLS estimates. The solid line represents the quantile 
regression estimates; and the shaded grey area plots the 95% confidence band for the quantile regression estimates



Keho: Revisiting the Income, Energy Consumption and Carbon Emissions Nexus: New Evidence from Quantile Regression for Different Country Groups

International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017362

With respect to the effect of per capita GDP, the results show that 
the EKC hypothesis holds for four panels, but not consistently 
throughout the conditional distribution of CO2 emissions. In 
particular, the EKC hypothesis holds for Sub-Saharan African, 
American and European countries at all quantile levels, whereas 
it holds for Asian and MENA countries with lower levels of CO2 
emissions. For Asian and MENA countries at the right tail of the 
distribution of CO2 emissions, the coefficients on income and its 
square are insignificant. This means that economic growth is good 
for the environment for Asian and MENA countries that have a 
low level of pollution. If Asian and MENA countries with high 
levels of pollution are to reduce their emission levels, they have 
to sacrifice their economic growth.

4. CONCLUSION

The EKC hypothesis has been a widely tested hypothesis in the 
energy economics literature over the past decades. However, 
previous studies focused only on mean effects and have not yet 
examined the role of pollution levels in the income-pollution 
nexus. This short paper uses quantile regression to reexamine 
whether CO2 emissions are linked to per capita GDP in a way 
that is described by the EKC. The empirical analysis has been 
conducted for five panels of 59 countries.

The results suggest that energy use and income may have 
significant effects at points in the conditional distribution of CO2 
emissions other than the mean. In essence, our findings indicate 
where energy use and income may matter, not just whether or not 
they matter on average. We found that energy consumption tends 
to increase CO2 emissions in all panels, the effect being larger in 
low pollution countries. This suggests that energy conservation 

policies in low pollution economies may be more beneficial for 
the environment. The results also provide evidence supporting 
the EKC hypothesis for Sub-Saharan, American, and European 
countries at all quantiles. Also, the EKC hypothesis holds for 
Asian and MENA countries with lower levels of CO2 emissions. 
This means that economic growth is good for the environment for 
Sub-Saharan, American, and European countries and for Asian and 
MENA countries that have a low level of pollution. These findings 
give some hope that economic growth can help to reduce CO2 
emissions. They also suggest that economic growth is not always 
and everywhere the cause and the cure of pollution. A growing per 
capita income will reduce pollution in Sub-Saharan, American and 
European countries as well as in Asian and MENA countries with 
lower levels of pollution. Asian and MENA countries with high 
levels of pollution should do more to reduce their carbon emissions 
by adopting more strict environmental and energy policies, 
increasing the infrastructure for energy efficiency and improving 
the use of alternative sources of energy that are relatively free 
from pollutant emissions.

REFERENCES

Al-Mulali, U., Tang, C., Ozturk, I. (2015a), Estimating the environment 
Kuznets curve hypothesis: Evidence from Latin America and the 
Caribbean countries. Renewable and Sustainable Energy Reviews, 
50, 918-924.

Al-Mulali, U., Saboori, B., Ozturk, I. (2015b), Investigating the 
environmental Kuznets curve hypothesis in Vietnam. Energy Policy, 
76, 123-131.

Al-Mulali, U., Adebola, S.S., Ozturk, I. (2016), Investigating the presence 
of the environmental Kuznets curve (EKC) hypothesis in Kenya: An 
autoregressive distributed lag (ARDL) approach. Natural Hazards, 
80(3), 1729-1747.

Figure 5: Ordinary least squares and quantile regression estimates over the conditional quantiles of CO2 emissions: Evidence for MENA countries

Source: The X-axis represents the conditional quantile of CO2 emissions. The horizontal dashed line represents the ordinary least squares 
(OLS) estimates. The two dotted lines depict the 95% confidence intervals for the OLS estimates. The solid line represents the quantile 
regression estimates; and the shaded grey area plots the 95% confidence band for the quantile regression estimates



Keho: Revisiting the Income, Energy Consumption and Carbon Emissions Nexus: New Evidence from Quantile Regression for Different Country Groups

International Journal of Energy Economics and Policy | Vol 7 • Issue 3 • 2017 363

Apergis, N., Payne, J.E. (2009), CO2 emissions, energy usage, and output 
in Central America. Energy Policy, 37, 3282-3286.

Apergis, N., Ozturk, I. (2015), Testing environmental Kuznets curve 
hypothesis in Asian countries. Ecological Indicators, 52, 16-22.

Aslanidis, N., Iranzo, S. (2009), Environment and development: Is there 
a Kuznets curve for CO2 emissions? Applied Economics, 41(6), 
803-810.

Baek, J. (2015), Environmental Kuznets curve for CO2 emissions: The 
case of Artic countries. Energy Economics, 50, 13-17.

Dinda, S. (2004), Environmental Kuznets curve hypothesis: A survey. 
Ecological Economics, 49, 431-455.

Dinda, S., Coondoo, D., Pal, M. (2000), Air quality and economic growth: 
An empirical study. Ecological Economics, 34, 409-423.

Dinda, S., Coondoo, D. (2006), Income and emission: A panel data-based 
cointegration analysis. Ecological Economics, 57(2), 167-181.

Galeotti, M., Lanza, A., Pauli, F. (2006), Reassessing the environmental 
Kuznets curve for CO2 emissions: A robustness exercise. Ecological 
Economics, 57, 152-163.

Gill, A.R., Viswanathan, K.K., Hassan, S. (2017), Is environmental 
Kuznets curve still relevant? International Journal of Energy 
Economics and Policy, 7(1), 156-165.

Grossman, G., Krueger, A. (1991), Environmental Impacts of a North 
American Free Trade Agreement, National Bureau of Economics 
Research Working Paper, No. 3194. Cambridge: World Bank, NBER.

Harbaugh, W.T., Levinson, A., Wilson, D.W. (2002), Reexamining the 
empirical evidence for an environmental Kuznets curve. The Review 
of Economics and Statistics, 84(3), 541-551.

Holtz-Eakin, D., Selden, T.M. (1992), Stoking the Fires? CO2 Emissions 
and Economic Growth, NBER Working Papers No. 4248. NBER.

Iwata, H., Okada, K., Samreth, S. (2010), Empirical study on 
environmental Kuznets curve for CO2 in France: The role of nuclear 
energy. Energy Policy, 38, 4057-4063.

Jalil, A., Mahmud, S. (2009), Environment Kuznets curve for CO2 
emissions: A cointegration analysis for China. Energy Policy, 37(12), 
5167-5172.

Jebli, M.B., Youssef, S.B., Ozturk, I. (2016), Testing environmental 
Kuznets curve hypothesis: The role of renewable and non-renewable 
energy consumption and trade in OECD countries. Ecological 
Indicators, 60, 824-831.

Koenker, R., Bassett, G.J. (1978), Regression quantile. Econometrica, 
46, 33-50.

Koenker, R., Hallock, K. (2001), Quantile regression. Journal of Economic 
Perspectives, 15(4), 143-156.

Koenker, R., Machado, J.A.F. (1999), Goodness of fit and related inference 
processes for quantile regression. Journal of the American Statistical 
Association, 94, 1296-1310.

Lamla, M. (2009), Long-run determinants of pollution: A robustness 
analysis. Ecological Economics, 69, 135-144.

Lean, H.H., Smyth, R. (2010), CO2 emissions, electricity consumption 
and output in ASEAN. Applied Energy, 87, 1858-1864.

Liu, X. (2005), Explaining the relationship between CO2 emissions 
and national income - The role of energy consumption. Economics 
Letters, 87(3), 325-328.

Martinez-Zarzoso, I., Bengochea-Morancho, A. (2004), Pool mean group 
estimation for an environmental Kuznets curve for CO2. Economics 
Letters, 82(1), 121-126.

Ozturk, I. (2010), A literature survey on energy-growth nexus. Energy 
Policy, 38(1), 340-349.

Ozturk, I., Al-Mulali, U. (2015), Investigating the validity of the 
environmental Kuznets curve hypothesis in Cambodia. Ecological 
Indicators, 57, 324-330.

Pao, H.T., Tsai, C.M. (2010), CO2 emissions, energy consumption and 
economic growth in BRIC countries. Energy Policy, 38, 7850-7860.

Perman, R., Stern, D.I. (2003), Evidence from panel unit root and 
cointegration tests that the environmental Kuznets curve does 
not exist. The Australian Journal of Agricultural and Resource 
Economics, 47, 325-347.

Rosado, J.A. (2017), Economic growth, CO2 emissions and electric 
consumption: Is there an environmental Kuznets curve? An empirical 
study for North America countries. International Journal of Energy 
Economics and Policy, 7(2), 65-71.

S e l d e n ,  D . H . ,  S o n g ,  D .  ( 1 9 9 4 ) ,  E n v i r o n m e n t a l  q u a l i t y  a n d 
development: Is there a Kuznets curve for air pollution emissions? 
Journal of Environmental Economics and Management, 27(2), 
147-162.

Shafik, N. (1994), Economic development and environmental quality: 
An econometric analysis. Oxford Economic Papers, 46, 757-773.

Shahbaz, M., Ozturk, I., Afza, T., Ali, A. (2013), Revisiting the 
environmental Kuznets curve in a global economy. Renewable and 
Sustainable Energy Reviews, 25, 494-502.

Shahbaz, M., Khraief, N., Uddin, G.S., Ozturk, I. (2014), Environmental 
Kuznets curve in an open economy: A bounds testing and causality 
analysis for Tunisia. Renewable and Sustainable Energy Reviews, 
34, 325-336.

Shahbaz, M., Dube, S., Ozturk, I., Jalil, A. (2015), Testing the 
environmental Kuznets curve hypothesis in Portugal. International 
Journal of Energy Economics and Policy, 5(2), 475-481.

Smyth, R., Narayan, P.K. (2015), Applied econometrics and implications 
for energy economics research. Energy Economics, 50, 351-358.

Yang, H., He, J., Chen, S. (2015), The fragility of the environmental 
Kuznets curve: Revisiting the hypothesis with Chinese data via an 
extreme bound analysis. Ecological Economics, 109, 41-58.

Zambrano-Monserrate, M.A., García-Albán, F.F., Henk-Vera, K.A. 
(2016), Bounds testing approach to analyze the existence of an 
environmental Kuznets curve in Ecuador. International Journal of 
Energy Economics and Policy, 6(2), 159-166.