Topic an analysis of the driving factors of carbon dioxide equivalent emissions using linear regression model

Country classification: 5 DATAAND STATISTICS DESCRIPTION OF VARIABLES 6 QUANTITATIVE ANALYSIS 10 Initial model 10 Error Testing 11 The Specification Error 11 Multicollinearity 12 Heteros

FOREIGN TRADE UNIVERSITY

Topic AN ANALYSIS OF THE DRIVING FACTORS OF

CARBON DIOXIDE EQUIVALENT EMISSIONS USING

LINEAR REGRESSION MODEL

Ord Fullname Student ID % Contribution 1 Nguyễn Thùy Dương

2012140012

2 Lé Thi Thanh Ha 2013140007

3 Nguyén Tran Ngoc Son 2012140047

4 Bach Thanh Tra 2012140050

Country classification: 5

DATAAND STATISTICS DESCRIPTION OF VARIABLES 6

QUANTITATIVE ANALYSIS 10 Initial model 10 Error Testing 11

The Specification Error 11 Multicollinearity 12 Heteroskedasticity 13 Normality of

residuals u 16 Final results 17

REFERENCES 22 APPENDIX 27

Page 1

AN ANALYSIS OF THE DRIVING FACTORS OF CARBON

DIOXIDE EQUIVALENT EMISSIONS USING LINEAR

REGRESSION MODEL Nguyen Thuy Duong, Le Thanh Ha, Bach Thanh Tra,

Nguyen Tran Ngoc Son Foreign Trade University

Abstract

This paper aims to identify the factors that affect carbon dioxide and carbon dioxide

equivalent methane emissions from energy usage and cement production worldwide in 2021

The model uses cross-country data for 72 nations and utilizes the Ordinary Least Square

regressions for estimating the parameters FDI inflow, population and urban rate were found

to have a positive influence on the emissions of carbon dioxide equivalents, while the effect

was the opposite regarding GDP growth A transitional economy is likely to exhaust more

greenhouse gas than a developed or developing country

I INTRODUCTION

During the 21st century, climate change and global warming have emerged as some of

the most serious problems facing the world Notable factors causing these problems,

especially carbon emissions levels, are now the highest in history Therefore, across the

world, a considerable amount of attention has been paid to controlling carbon and other

greenhouse gas emission levels, as well as studying how different aspects of the economy

correlate with it However, the majority of studies only link their analyses on carbon

dioxide equivalent emissions with energy consumption and economic growth but not other

factors It is not to mention that very few researchers care to examine a large number of

countries, but only resort to one or a few groups of countries Therefore, this research was

conducted to look at other variables of the economy that significantly impacts the rise of

carbon and methane emissions from fuel production in multiple countries, using the latest

data collectibles

The research applied linear regression model, particularly Ordinary Least Squares

regression (OLS) since the data is cross-sectional data By using OLS, the parametric

form makes it relatively more efficient to interpret data The process of result

interpretation and policy implication is thus also easier to carry out

Page 2

If LITERATURE REVIEW

The problem of rising carbon emissions caused by energy consumption has been

recognized in most countries around the world Being the main cause of global warming

and climate change, carbon emissions are continuously setting all-time high records,

making environmental deterioration a global challenge As a result, numerous studies

have been carried out to investigate factors affecting carbon emissions, in the hope of

finding an effective policy to control the situation The findings generally conclude that

the remarkable economic growth of countries over the past few decades tops the list of

factors affecting carbon emissions, followed by urban rate and population Other factors

examined to be closely related to the carbon emissions of a country include FDI (Foreign

Direct Investment) inflow, and (although pointed out by very few researchers) country

classification

1 Economic growth

The first thought-of reason explaining high carbon emission levels is global economic

growth As seen in a research examining the correlation between economic growth,

energy consumption, financial development, trade openness, and carbon emissions from

1975 to 2011 in Indonesia; economic growth increase does raise carbon dioxide

equivalent emissions (Shahbaz, Hye, Tiwari and Leitao, 2013) Another study also finds a

causality between GDP and energy consumption of the five main ASEAN countries

(Munir, Lean and Smyth, 2020) (Kahouli, 2018) also supported the linkage between

emission levels and economic growth with a research into electricity consumption, carbon

emissions, R&D stocks, and economic growth for Mediterranean countries from 1990 to

2016 Strong feedback effects were found between the examined factors (Li et al.,2021)

discussed the effect of economic growth, economic structure, and other factors on per

capita carbon emissions in 147 countries from 1990 to 2015 The results showed that at

the global level, economic growth and economic structure were respectively the most

significant positive and negative factors affecting carbon emissions

Page 3

2 Urban rate and population size

In addition to economic growth, there are other factors contributing to the increase in

carbon emission levels, namely urban rate and population size More than half of the

world’s population lives in cities (JNDESA, 2014) Thus, it can be said that urban areas

make up the large part of carbon emissions, given that material and energy consumption

processes primarily take place here In the case of China, the most populous country in the

world, (Dhakal, 2009) further finds that the 35 largest cities in China contributed up to

40% of the national carbon dioxide equivalent emissions

The impact of population size on environmental quality is also very obvious

(Engelman, 1994) plotted long-term trends in global carbon dioxide equivalent emissions

and population, and found that since 1970 both emissions and population have grown at

similar rates, leading to a hypothesis that population growth has been a major force in

driving up global emissions over recent decades (Meyerson, 1998) also found that the

global increase in carbon emissions has been quite closely correlated with population

growth over the last 25 years Birdsall (1992) specified two mechanisms through which

population growth could contribute to carbon emissions First, a larger population could

result in increased demand for energy for power, industry, and transportation, hence

increasing fossil fuel emissions Second, rapid population growth can cause deforestation,

other changes in land use, and combustion of wood for fuel These might contribute to

greenhouse gas emissions significantly Moreover, based on data from 93 countries from

1975 to 1996, (Shi A, 2003) found that global population growth is more than

proportionally associated with an increase in carbon emissions and that the impact of

population growth on carbon emissions is much more pronounced in developing countries

than in developed countries

3 FDI

Although FDI has become increasingly important, few details have been discussed

about it and its effects on greenhouse emissions Indeed, the rising FDI flow in developing

countries raises an important question regarding whether it has any environmental

consequences (Zeng and Eastin, 2012) Previous studies may examine the correlation

between FDI inflow and carbon dioxide equivalent emissions, but lack

Page 4

an analysis of the complexity and causality between the two variables, potentially leading

to poorer discernment in the overall result However, the conventional view may suggest

that to attract foreign investment, developing countries have a tendency to ignore

environmental concerns through relaxed or non-enforced regulation Such relaxed

environmental standards in developing countries may cause FDI to promote carbon

dioxide equivalent emissions at large (Pao and Tsai, 2011) This hypothesis is supported

by (Merican et al., 2007), whose research assesses the relationship between FDI and

pollution in the ASEAN-S By using autoregressive distributive lag estimation, it is clear

that FDI increases emissions in Malaysia, Thailand, and the Philippines; however, there

appears to be an inverse relationship between FDI and pollution in Indonesia Such

inconclusive and mixed results, therefore, call for further research into the effects of FDI

on carbon emissions

4, Country classification:

Although no specific research has been conducted on the relation between country

classification and carbon emission levels, existing analysis of each country type’s

environmental-related nature may reveal the answer The World Economic Situation and

Prospects (WESP) categorizes countries into three groups of developed, transitional and

developing countries; all with different economic conditions The consensus is that

developed countries are the largest contributors to global carbon dioxide equivalent

emissions, however, there have been recent calls for the developing countries to play an

active role in global emissions reduction (Winkler et al., 2002) The level of carbon

dioxide equivalent emissions from developing countries has been rapidly exceeding that

of the developed countries and in 2003 accounted for almost 50% of the world's carbon

dioxide equivalent emissions (World Development Indicators, 2007)

Meanwhile, countries with transitional economies are by their nature unstable Thus,

while environmental problems are now recognised widely in developed countries,

environmental benefits and willingness to pay environmental services in transitional

economies are generally low In a risky economic environment, it is cheaper to grow GDP

based on old technologies The result is that environmental

Page 5

degradation becomes inevitable in the absence of a robust environmental management

system Unfortunately, this is the case in most transition economies (Golub et al., 2003)

In summary, a consensus has not been reached regarding the factors that affect the

world's carbon emissions significantly Additionally, litthe comprehensive research has

focused on the relationships between FDI, country classification and carbon emissions

Moreover, considering the urgent needs for emission reduction and the current aging

crisis, more up-to-date data on these variables needs to be collected from a wider range of

countries Hence the research purpose of this article

Il DATAAND STATISTICS DESCRIPTION OF VARIABLES The data for this

research was gathered from 2021 mainly Missing data in 2021 was replaced by data from

2020 In total, cross-sectional information was gathered from 72 countries in the world

ranging from developed, in transition and developing countries

The data on carbon dioxide equivalent emission amount was taken from bP (2022) and

Knoema (2022) Specifically, it was taken from the table of Carbon Carbon dioxide

equivalent emissions from energy, process emissions, methane, and flaring of bP and the

World Atlas The data on GDP growth, population and urban rate were taken from the

World Bank (2021) The data on 2021 FDI inflow was taken from the UNCTAD (2022)

The classification of countries (developed, developing, in transition) was according to the

UNCTAD (2021) Table 1 below presents each of the variables, their units, and their

predicted effects on carbon dioxide equivalent emission amount

emissions The sum of carbon Million tonnes Dependent

dioxide emissions from of carbon variable

flaring, industrial equivalent processes and methane

emissions in carbon dioxide equivalent

people living in urban population areas as

defined by national statistical offices

based on basic economic

TABLE I - Variable names, description, unit of variables and predicted signs Page 7

The Ordinary Least Square regressions were used to best estimate the parameters of the model The relationship between carbon dioxide equivalent emission with other listed

variables was determined in the year 2021

The greenhouse gas emission is scrutinized by the dependent variable of carbon dioxide equivalent emission In the model, the indicator for this variable is emissions For each country, it accounts for carbon dioxide and methane emissions from various processes of human use, where the methane is converted to carbon dioxide equivalent for

better measurement of the carbon footprint of that country

The first independent variable is of foreign direct investment inflow, whose indicator

in the model is fd FDI inflow in 2021 was recorded with some values being negative The negative signs of FDI inflow showcase that the value of disinvestment by foreign investors was more than that of capital newly invested The higher the FDI inflow, it is expected that there would be more economic growth, especially for developing countries

Since economic growth is linked to carbon footprint, the expected sign of the coefficient 1S positive

The second independent variable is annual gross domestic product growth, denoted as gdpgrowth in the model It is the annual percentage growth rate of GDP at market prices based on constant 2015 prices, expressed in U.S dollars The available data for this

variable in 2021 is not adequate as there is such a small time difference between the

examined year and the year this research was conducted The lacking data (of 5 out of 72

countries) was replaced by the data in 2019 and 2020 from the same source (World Bank) The expected coefficient for gdpgrowth is positive

The third independent variable is the population of the country, denoted by population

in the model The unit of measurement for population is million people The data is based

on the de facto definition of population, which counts all residents regardless of legal

status or citizenship The values gathered are mid-year estimates, as stated by the World Bank More people are expected to pollute more than less, so the expected sign of this variable’s coefficient is positive

Page 8

The fourth independent variable is urban population, which is wrbanrate in the model The

variable depicts the percentage of the population that lives in urban areas instead of rural

areas The higher this rate is, the lower the environmental quality because of the high density

of cities Hence, wrhanrate is expected to have a positive effect on the dependent variable

The fifth independent variable is country classification, denoted by 3 dummy variables

developed, developing and transitional with transitional bemg the base dummy variable It 1s based on the WESP classification of countries From the findings cited in the literature

review, we speculate that transitional economies are more likely to have the biggest carbon footprint due to the stage of growth they are in As a consequence, the expected signs of

developed and developing are negative

Descriptive statistics were taken after gathering data on all variables, except the dummy

variables FDI inflow is the lowest in the Netherlands, at -81056 GDP growth ranged from - 18% to 13.5% Population in China remains to be the highest, with the lowest in the data set being Estonia The highest urban rate was Singapore with 100% population residing in urban areas, a special case; while the lowest was China Hong Kong SAR with 10% Most

economies are former countries of the Soviet Union

Figures are rounded to 3 decimal places

TABLE 2 - Descriptive statistics The Ordinary Least Squares model used in this study was as follows:

5599990999 999999999008 +

Bs 9999999999 9990990909000

IV QUANTITATIVE ANALYSIS

1 Initial model

The initial regression results before error testing were presented in Figure 1 Because

of missing data, the sample size was limited to 72

The R-square for this test was about 0.8004, signifying that the model explained about 80% of the variation in carbon dioxide equivalent emission levels in 2021 The F- statistics, at 43.44 which was higher than F-critical (at about 2.25), and the p-value of F- statistics, at almost 0, indicate that the model was overall significant

Out of the 6 independent variables tested (2 of which are dummy variables), there were

2 significant ones, fdi and population Both population and FDI inflow had a positive correlation with carbon dioxide equivalent emissions levels The p-values for both variables were very low, at almost 0, suggesting statistical significance at 1%

From the positive signs of both coefficients, we can speculate that population and FDI

inflow both have a positive influence on the emissions of CO2 equivalent

Page 10

Specifically, holding other factors constant, a rise in FDI inflow by 1 million dollar in a country will raise greenhouse gas emissions by approximately 0.011 million tonnes and an increase of 1 million people will cause emissions to rise by about 4.738 million tonnes

emissions Coef Std Err E P>|t 95% Conf Interval)

gdpgrowth -27 88057 19.61687 -1.42 0.1606 -67 05816 11.29702

populetion 4.737745 4330771 10.94 0.000 3.87283 5.60266 urbanrate 12.11685 4.852874 2.50 0.015 2.424992 21.80871 developed -353.9495 299.7547 -1.18 0.242 -952, 601 244.702

2.1 The Specification Error

First, the specification error of the model was checked by the Ramsey RESET test to discover whether influential variables have been left out The results are presented below:

ovtest

Ramsey RESET test using powers of the fitted values of emissions

Ho: model has no omitted variables

Fi(3, 62) = 99.32

rob > F = 0.0000

Page 11

FIGURE 2 - Results of the Ramsey RESET test

As the p-value of the F-statistical was smaller than the significant level, 5%, we concluded that the model had misspecification of functional form or omission of

1mportant variables

2.2 Multicollinearity

Next, we tested for the multicollinearity of the model by checking the correlation matrix, then by using the variance inflation factor (VIF) test The consequence of perfect multicollinearity is that the population parameters cannot be estimated, while imperfect multicollinearity leads to inexact hypothesis tests The results of the matrix and the VIF

test were shown in Figure 3 and 4

emis3i-s fdi gdpgro~h popula-n urbanr-e develo-d develo~g emissions 1.0000

£ai 0.6506 1.0000 gdpgrcwth 0.0980 0.1093 1.0000

Page 12

developed 2.92 0.342958 developing 2.86 0.349060

fai 1.23 0.812559 urbanrate 1.18 0.847682 gđpgxrowt! 1.12 0.890631 Mean VIF | 1.78

FIGURE 4 - Results of the VIF test For the VIF test, the mean VIF was 1.78, smaller than 10, thus, we could conclude with confidence that the model did not have the problem of multicollinearity

2.3 Heteroskedasticity

Next, we detected whether there was a problem of heteroscedasticity If so, the OLS estimators and regression predictions based on heteroskedasticity will no longer be BLUE (Best Linear Unbiased Estimators), so the regression predictions will be inexact as well

We tested for this problem by graphing out the distribution of residuals, as well as using the White test and the Breusche-Pagan test The results were as follows:

Page 13

FIGURE 5 - Graph of the distribution of residuals

As the distribution of residuals did not converge into any certain direction and there existed many outliers, we could predict that the model might face heteroskedasticity

- imtest, white White's test for Ho: homoskedasticity against Ha: unrestricted heteroskedasticity

chi2 (24) - 71.90

Prob > chi2 = 0.0000 Cameron é& Trivedi's decomposition of IM-test

Skewness 6.54 6 0.3660 Kurtosis 2.40 1 0.1216 Total 80.83 31 0.0000

Page 14

FIGURE 6 - Results of the White test For the White Test, the p-value (Prob > chi2) was almost 0, smaller than 5%, therefore the model should have had the problem of heteroscedasticity

Variables: fitted values of emissions

ch12 (1) = 538.87

FIGURE 6 - Results of the Breusche-Pagan test For the Breusche-Pagan test, the p-value (Prob > chi2) was also almost 0, smaller than 5%, therefore, we could conclude with confidence that the problem of heteroscedasticity existed in the model

> Solution: We would fix the problem by using the Robust Standard Errors This is a technique to obtain unbiased standard errors of OLS coefficients under the problem of

Robust emissions Coef Std Err t P>ịtl| [95% Con Interval] tái 0111174 0028371 3.92 0.000 0054513 0167836

2.4, Normality of residuals u

Finally, we conducted a test to check whether u has a normal distribution If there is no normality of residuals, it means that the sample size is not large enough or not randomly selected, leading to an inexact hypothesis

We tested for the normality of u by both graphing the distribution of residuals as well

as using the Jarque-Bera test The results were as follows: