Energy Consumption, CO2 Emissions and Economic Growth: A
Comparative Analysis in Guangdong, China
Hao Guo
School of Urban Planning and Design, Peking University Shenzhen Graduate School, Shenzhen, Guangdong, China
Keywords:
Energy Consumption, Carbon Dioxide Emissions, Environmental Kuznets Curve.
Abstract: Carbon dioxide (CO2) is the chief greenhouse gas causing global warming, and the relationship between
this gas and economic development is a major subject of research. This study uses panel data and models for
coastal cities in Guangdong province, China, from 2005 to 2017, as well as the robustness results obtained
using Driscoll-Kraay standard errors. The study finds that an inverted U-shaped relationship exists between
income and emissions in the Pearl River Delta region; the expected income per capita at the inflection point
on the U curve is between US$24,920 and 27,860, and the expected per capita CO2 emissions at the
inflection point are projected to be approximately 13 tons annually. However, the EKC hypothesis is not
verified in the non-Pearl River Delta area. Population agglomeration is found to benefit the reduction in per
capita CO2 emissions, and a population scale effect exists; however, the emissions reduction effect of
population agglomeration in the non-Pearl River Delta area is greater than that in the Pearl River Delta
region. This study also provides relevant policy recommendations and suggestions for future research.
1 INTRODUCTION
Climate change has become a major global issue,
and excessive greenhouse gas emissions pose a
threat and disaster to the survival of life on Earth.
Carbon dioxide (CO2) is the chief greenhouse gas
causing global warming and currently accounts for
approximately 75% of the earth's greenhouse effect
(Sirag, 2018). While CO2 emissions are tightly
linked with fossil fuel consumption, fossil fuels are
an important factor driving the rapid economic
development of numerous countries and areas. The
relationship between environmental quality and
economic development can be represented by what
is known as the environmental Kuznets curve
(EKC), which has become a major topic of research.
The Kuznets curve was first proposed by the
economist Simon Kuznets in 1955 while researching
per capita income and the fairness of income
distribution (Kuznets, 1955). Afterwards, during the
early 1990s, many scholars innovatively applied the
Kuznets curve in the field of environmental quality
and proposed that the relationship between per
capita income and environmental quality took the
form of an inverted U-shaped curve (Grossman,
1995). While numerous empirical studies have
supported the EKC hypothesis (Diao, 2009), other
scholars have reached different conclusions and
believe that there is no empirical support for EKC
(Sanchez, 2016). Some studies have reached
different conclusions concerning the shape of the
EKC curve due to differences in regional
characteristics (Churchill, 2018) or differences in the
pollutants studied (Kaika, 2013). As a result, doubts
still exist concerning the relationship between CO2
emissions and income that form the basis of the
EKC hypothesis, and further efforts must be made to
verify the hypothesis employing representative
geographical areas and methods.
This study takes CO2 emissions per capita as an
indicator of environmental degradation to analyze
and compare the EKC of the Pearl River Delta
region with other areas in Guangdong province,
during the period from 2005 to 2017.
2 METHODS
2.1 Data Sources and Variables
This study employs balanced panel data from 21
cities in Guangdong province from 2005 to 2017 as
Guo, H.
Energy Consumption, CO2 Emissions and Economic Growth: A Comparative Analysis in Guangdong, China.
DOI: 10.5220/0011731000003607
In Proceedings of the 1st International Conference on Public Management, Digital Economy and Internet Technology (ICPDI 2022), pages 99-104
ISBN: 978-989-758-620-0
Copyright
c
2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
99
a sample and obtains GDP per capita and population
density data from the Guangdong Statistical
Yearbook for various years. This paper takes 2005
as the base year for GDP per capita, corrected for
inflation, and converts all amounts into US dollars
from the average annual exchange rates. The cities’
energy consumptions are calculated from each city's
energy intensity and GDP, and CO2 emissions are
estimated by the energy and economic data. The
definitions of the variables involved are shown in
Table 1.
Table 2 provides a statistical description of the
panel sample data. The sample consists of balanced
panel data with a period of 13 years. The Pearl River
Delta region includes Guangzhou, Shenzhen,
Huizhou, Dongguan, Foshan, Zhaoqing, Zhuhai,
Zhongshan, and Jiangmen, and the overall sample
size is 117. This study classifies the remaining areas
of Guangdong province, including the eastern
region, western region, and mountain areas, as the
non-Pearl River Delta area. This area includes 12
cities and has an overall sample size of 156. From
this table, the CO2 emissions per capita, GDP per
capita, and population density of the Pearl River
Delta region are 2, 3, and 2.77 times greater than the
corresponding figures for the non-Pearl River Delta
area, respectively. This outcome indicates that
significant differences exist between these two areas.
The standard deviations of the research variables are
relatively small, which indicates that the fluctuations
in values of the variables around the mean values are
relatively small; this also shows that the data are
stable and suitable for further study.
Table 1: Summary of variable definitions.
Name of
variable
Abbreviation Description Units Source
CO2
emissions
per capita
CO2pc
Average per
capita CO2
emissions
Tons/person
Estimated from energy consumption,
GDP, and population data from the
Guangdong Statistical Yearbook
Income
level
GDPpc GDP per capita
US$10,000/pers
on
Guangdong Statistical Yearbook for
various years
Population
density
Popuden
Permanent
population/area
1,000
persons/km2
Guangdong Statistical Yearbook for
various years
Table 2: Descriptive statistics of variables.
Region Pearl River Delta Non-Pearl River Delta
Variable CO2pc GDPpc Popuden CO2pc GDPpc Popuden
Mean 9.336 0.975 1.729 4.694 0.32 0.624
St.d 2.912 0.509 1.53 2.216 0.146 0.618
Min 2.762 0.145 0.247 1.029 0.091 0.154
Max 13.972 2.122 6.272 10.899 0.675 2.55
Size 117 117 117 156 156 156
Figure 1: Scatter diagram of the GDP and CO2 emissions per capita.
ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology
100
Figure1 shows the results of the quadratic curve
fitting for GDP per capita and CO2 emissions per
capita in the Pearl River Delta and the non-Pearl
River Delta area. From this figure, the Pearl River
Delta region has a significantly higher income level
than that in the non-Pearl River Delta area, and CO2
emissions per capita display a slowing trend as the
income level increases. This very likely indicates
that the EKC inflection point has been reached,
while emissions per capita in the non-Pearl River
Delta area continue to increase rapidly.
2.2 Model Design
Equations used to test EKC models include
quadratic term equations and cubic term equations,
and linear equations are usually used for model
testing in the case of countries or areas that are not
fully industrialized. A quadratic polynomial model is
a very versatile means of assessing the form of an
EKC curve. To observe whether an inverted U-
shaped curve is present, and whether environmental
degradation continues as income increases, this
study uses per capita income as a quadratic term in
model testing. The equation used to test the EKC
model in this study is as follows:
Y
= 𝛼
+ 𝛽
𝑋
+ 𝛽
𝑋
+ 𝛽
𝑍
+ 𝜀
,
(1)
Here, Y is the variable expressing the degree of
environmental degradation; X is the income level
(GDP per capita); Z consists of other explanatory
variables; ε
,
is an error term; i indicates different
entities; and t expresses time. The coefficients in
front of X and its quadratic term determine the form
of the curve. Accordingly, the following situations
can be used to judge whether the EKC hypothesis is
correct:
If β_1>0 and β_2=0, then X and Y have a
monotonically increasing relationship.
If β_1<0 and β_2=0, then X and Y have a
monotonically decreasing relationship.
If β_1>0 and β_2>0, then X and Y have a U-
shaped relationship
If β_1>0 and β_2<0, then X and Y have an
inverted U-shaped relationship.
All of these different curve shapes have different
implications. A monotonically decreasing curve
indicates that environmental quality improves as
income increases, while a monotonically increasing
curve indicates that environmental quality
deteriorates as income increases. A U-shaped curve
indicates that while environmental quality improves
as income increases, it begins to deteriorate after
reaching a certain point. When an inverted U-shaped
curve is present, this outcome indicates that
environmental quality first deteriorates steadily with
rising income but begins to improve with rising
income after reaching an inflection point.
The explained variable in this model is CO2
emissions per capita, and the explanatory variables
are GDP per capita and its quadratic term. Model 1.1
is used to determine whether the EKC hypothesis is
applicable to the study area. Degree of population
agglomeration is subsequently added to model 1.1 as
an explanatory variable, yielding model 1.2. This
study also establishes models 2.1 and 2.2 for the
non-Pearl River Delta area during the same period
and adopts a fixed effects model and the random
effects model to compare the results of other models.
These models' equations are as follows:
CO2pc
= 𝛼
+ 𝛽
𝐺𝐷𝑃𝑝𝑐
+ 𝛽
𝐺𝐷𝑃𝑝𝑐
+ 𝜀
,
(2)
CO2pc
= 𝛼
+ 𝛽
𝐺𝐷𝑃𝑝𝑐
+ 𝛽
𝐺𝐷𝑃𝑝𝑐
+
𝛽
𝑃𝑜𝑝𝑢𝑑𝑒𝑛
+ 𝜀
,
(3)
Here, the subscript i=1,⋯ ,n indicates cross-
sectional units, and the subscript t=1,⋯ ,T
indicates time.
While many factors influence CO2 emissions per
capita, this study adopts per capita income and
degree of urban population agglomeration as the two
chief research variables. One of the chief difficulties
affecting analysis is model selection because the
choice of model has a large effect on the analysis.
Apart from the ordinary least squares method (OLS),
the fixed effects and random effects models are used
to analyze the panel data. An F-test is used to
determine the importance of individual effects and
compares a mixed OLS model with the fixed effects
model. In addition, the Lagrange multiplier (LM)
test is used to compare the OLS regression model
with the random effects model. Finally, the
Hausman test is used to confirm whether to use a
random effects model or a fixed effects model.
Cameron and Trivedi suggested that the sigmamore
option is the best in Stata for the Hausman test
because this option indicates that the two covariance
matrices are based on an estimation variance by the
same effective estimator (Cameron, 2010). These
tests indicate that model 1.1, model 1.2, and model
2.2 are fixed effects models, while model 2.1 is a
random effects model. Diagnostic testing is
performed on all models. The Pesaran cross-
sectional dependence test (Pesaran CD) is employed
to determine whether the residual is relevant across
different cities (Pesaran, 2004). The revised Wald
test is used to determine the models'
heteroscedasticity. The Wooldridge test is used to
Energy Consumption, CO2 Emissions and Economic Growth: A Comparative Analysis in Guangdong, China
101
test the serial correlation of the panel data.
According to Hoechle, if models are heterogeneous,
autocorrelated, and cross-regionally dependent, the
Driscoll-Kraay standard errors method should be
used (Hoechle, 2007).
3 RESULTS
From the F statistic of model 1.1 in Table 3, the
results of the F-test are significant for a fixed effects
model (F (2, 106) = 441.819, p-value = 0.000).
Model fitting reveals that a regression estimate using
the CO2 emissions per capita function can explain
up to 89% of the variation in CO2 emissions per
capita (R-squared = 0.848). The fixed effects
modeling results also reveal that the coefficients of
all variables are significant to a level of 1%.
Furthermore, model 1.1 verifies the existence of an
inverted U-shaped curve. After the population
density variable is added, this study finds that the
fixed effects model is also applicable to model 1.2.
Based on the results of testing for heteroscedasticity,
autocorrelation, and panel dependence, this study
employs the Driscoll-Kraay standard errors method
to overcome and minimize these problems. The
results of the fixed effects estimation and Driscoll-
Kraay standard errors estimation are shown in Table
3. From the F statistic of model 1.2, the results of the
F-test are significant for a fixed effects model (F (3,
105) = 313.032, p-value = 0.000). Similarly, the
model estimation results indicate that the data and
model have a good fit, and the model can explain
62.7% of the variation in per capita CO2 emissions
(R-squared = 0.627). The coefficients of all variables
are significant in a fixed effects model. For
population density, each increase in population
density by 1,000 persons per square kilometer could
reduce per capita CO2 emissions by 0.734 tons.
4 DISCUSSION
This study consequently discovers that a significant
inverse correlation exists between population density
and CO2 emissions per capita. Furthermore, the
inverted U-shaped curve in the Pearl River Delta
region is also confirmed by this model. From the
results of the fixed effects model estimation
employing models 1.1 and 1.2 and Driscoll-Kraay
standard errors estimation in the table, the monomial
coefficient of GDP per capita is significantly
positive, while the second-order coefficient is
significantly negative. Model 1.1 indicates that per
capita income at the inflection point is US$24,920
while CO2 emissions per capita at the inflection
point are 12.922 tons, with a 95% confidence
interval of 12.178 to 13.667 tons.
Table 3: Panel regression results for the pearl river delta region.
Model 1.1 FE Model 1.2 FE
Square FE
Driscoll- Kraay
Standard Errors
Square FE
Driscoll- Kraay
Standard Errors
GDPpc 6.983*** 6.983*** 7.020*** 7.020***
(-14.65) (-15.59) (-15.12) (-18.84)
GDPpc
2
-1.401*** -1.401*** -1.260*** -1.260***
(-7.09) (-7.35) (-6.31) (-8.10)
Popuden -0.734** -0.734***
(-2.61) (-6.60)
_cons 4.220*** 4.220*** 5.282*** 5.282***
(-17) (-19.33) (-11.17) (-16.78)
Observations 117 117 117 117
Cities 9 9 9 9
F-test or Wald chi2 441.819 415.713 313.032 606.414
R-square/ R-square within 0.848 0.893 0.627 0.899
EKC Holds Yes Yes Yes Yes
Turning points (real 2005 US$) 2.492 2.492 2.786 2.786
Note: ***, **, and * indicate 1%, 5%, and 10% levels of significance.
ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology
102
Table 4: Panel regression results for the non-pearl river delta region.
Model 2.1 RE Model 2.2 FE
Square RE
Driscoll- Kraay
Standard Errors
Square FE
Driscoll- Kraay
Standard Errors
GDPpc 9.654*** 9.654*** 10.808*** 10.808***
(-8.44) (-5.6) (-9.68) (-7.87)
GDPpc
2
-2.721 -2.721 -3.538** -3.538
(-1.64) (-1.04) (-2.24) (-1.69)
Popuden -5.811*** -5.811***
(-4.20) (-3.39)
_cons 1.939*** 1.939** 5.295*** 5.295***
(-3.58) (-2.38) (-6.49) (-5.48)
Observations 156 156 156 156
Cities 12 12 12 12
F-test or Wald chi2 921.321 394.165 347.441 165.244
R-squared 0.396 0.396 0.881 0.881
EKC Holds No No Yes No
Turning points (real 2005 US$) - - 1.527 -
Note: ***, **, and * indicate 1%, 5%, and 10% levels of significance.
The results of the LM test, the F test, and the
Hausman test reveal that while model 2.1 for the
non-Pearl River Delta area should be treated as a
random effects model, model 2.2 for the Pearl River
Delta region should be treated as a fixed effects
model. Similarly, after confirming the
heteroscedasticity, autocorrelation, and panel
dependence of these models, this study employs the
Driscoll-Kraay standard errors method to overcome
these problems. Table 4 displays the random effects
results for model 2.1 and the fixed effects results for
model 2.2. However, although the quadratic term for
GDP per capita in the fixed effects results for model
2.2 is significant (as indicated by two asterisks),
after employing the Driscoll-Kraay standard errors
to overcome the problems of heteroscedasticity,
autocorrelation, and panel dependence, this term is
not significant, which is consistent with
expectations. Based on the F test and the Wald chi-
squared values of the two models, this study finds
that these two models are significant, and there are
significant inverse correlations between increasing
population density and CO2 emissions per capita. In
the non-Pearl River Delta area, each increase in
population density by 1,000 persons per square
kilometer could reduce per capita CO2 emissions by
5.8 tons. However, an inverted U-shaped curve is
not verified for the non-Pearl River Delta area.
According to the standard errors estimation results
for models 2.1 and 2.2, the first-order coefficient of
GDP per capita is significantly positive, but the
second-order coefficient is not significant.
5 CONCLUSIONS
An inverted U-shaped relationship exists between
income and environmental degradation in the Pearl
River Delta region. This study finds that an income
per capita between US$24,920 and 27,860 is
expected at the EKC inflection point for the Pearl
River Delta region. Although economic growth
could cause pollution in the short run, it could
reduce emissions and pollution in the long term.
However, reductions in environmental degradation
may not spontaneously appear as per capita income
increases, and attention should be paid to the
specific mechanisms by which this positive effect
occurs. An inverted U-shaped relationship does not
exist in the non-Pearl River Delta area, but a
significant positive correlation exists between
income and environmental degradation. The reason
for this difference between the Pearl River Delta
region and the non-Pearl River Delta area lies in the
relatively high level of urbanization and
industrialization of the Pearl River Delta region;
even some cities in the Pearl River Delta region have
already reached the middle or late stage of
Energy Consumption, CO2 Emissions and Economic Growth: A Comparative Analysis in Guangdong, China
103
urbanization and industrialization. However, most
areas in the non-Pearl River Delta area are still in the
initial stage of rapid urbanization and
industrialization, and these processes will continue
in the future. Finally, we find that population
agglomeration can facilitate a reduction in the per
capita CO2 emissions in the Pearl River Delta and
the non-Pearl River Delta area. The population
agglomeration brings a population scale effect and
can significantly reduce the level of carbon dioxide
emissions.
Although the article has made some significant
findings, it still has certain limitations. Besides CO2
emissions, other pollutants such as sulfur dioxide,
methane and so on, could be also used to measure
environmental degradation. Furthermore, CO2
emissions data cannot be obtained directly, so the
carbon emissions data estimated by the energy and
economic data in this paper are not completely
accurate, so more precise data should be obtained
through more advanced tools and methods in the
future.
ACKNOWLEDGEMENTS
The authors would like to thank Shenzhen Low
Carbon City Big Data Engineering Laboratory for
providing with all the necessary support.
REFERENCES
Churchill S A, Inekwe J, Ivanovski K, et al. The
Environmental Kuznets Curve in the OECD: 1870–
2014 [J]. Energy Economics, 2018, 75: 389-399.
Cameron A C, Trivedi P K. Microeconometrics using stata
[M]. College Station, TX: Stata press, 2010.
Diao X D, Zeng S X, Tam C M, et al. EKC analysis for
studying economic growth and environmental quality:
a case study in China [J]. Journal of Cleaner
Production, 2009, 17(5): 541-548.
Grossman G M, Krueger A B. Economic growth and the
environment [J]. The quarterly journal of economics,
1995, 110(2): 353-377.
Hoechle D. Robust standard errors for panel regressions
with cross-sectional dependence [J]. The Stata Journal,
2007, 7(3): 281-312.
Kuznets S. Economic growth and income inequality[J].
The American economic review, 1955, 45(1): 1-28.
Kaika D, Zervas E. The Environmental Kuznets Curve
(EKC) theory—Part A: Concept, causes and the CO2
emissions case [J]. Energy Policy, 2013, 62: 1392-
1402.
Pesaran H. General diagnostic tests for cross section
dependence in panels[J], Cambridge Working Papers
in Economics, 0435, University of Cambridge, 2004.
Sirag A, Matemilola B T, Law S H, et al. Does
environmental Kuznets curve hypothesis exist?
Evidence from dynamic panel threshold [J]. Journal of
environmental economics and policy, 2018, 7(2): 145-
165.
Sanchez L F, Stern D I. Drivers of industrial and non-
industrial greenhouse gas emissions [J]. Ecological
Economics, 2016, 124: 17-24.
ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology
104
