Dawit Woubishet - Ethiopian Economic Association

The driving forces of CO2 emissions and its responsiveness in Ethiopia: An
integrated Analysis
Dawit Woubishet Mulatu1 and Zerayehu Sime Eshete2
Abstract
Sound environmental policy is fully dependent upon sound science. However, we have little scientific
knowledge of the driving forces behind environmental changes. Therefore, research concerning which
driving factors has an impact on environment and CO2 emissions and the extent of their impact have
been of great importance, since these impact factors will directly influence the constitution of
improving environmental impacts and CO2 abatement measures, policies and strategies in SubSaharan Africa (SSA) region. Ethiopia is one of the fastest growing economies in the SSA region
andpromotes green-growth-strategy and sustainable development. However, the linkages between the
economy and environmental impacts, and the responsiveness of CO2 emissions due to changes in
economy still require further investigation. This study tries to fill this knowledge gap and explore the
human-environment interactions usingan integrated approach, by taking Ethiopia as an embodiment.
This study applied the so-called environmental impactsare the multiplicative product of Population,
Affluence, and Technology (IPAT) identity as a framework to assess the driving forces of CO2
emissions using vector autoregressive (VAR) model specification with co-integration and Vector Error
Correction Model (VECM).
The results indicate the long-run responsiveness of CO2 emissions for GDP per-capita and energy
intensity is positive and significant,and appears reasonably strong. The findings imply that higher GDP
per-capita can induce more energy consumption and more CO2 emissions in Ethiopia context.On the
other hand, the effect of population on CO2 emissions is negative, which has some similarity with
Boserupian view that the impact of population on the state of the environment and contrary to the
Malthusian perspective, which holds that population growth increases environmental impacts. The
findings contribute for better understanding of thedriving forces behind environmental impacts and
indicate possibilities to anticipate the impact of trends in population growth and economic
development on CO2 emissions. Therefore, considering the driving forces and the responsiveness of
CO2 emissions would enable us to inform decision-makers to make emission reduction measures and
to maintain economic development, population, and energy use, especially in lower-income level
countries.
Key words:Environmental impacts,CO2 emissions, Integrated approach (IA), Human-environment
interactions, IPAT and Ethiopia
1
Corresponding author, Environment and Climate Research Center (ECRC), Ethiopian Development
Research Institute (EDRI), Addis Ababa,Ethiopia, Tel: +251911603699, [email protected]
2
Yom
Institute
of
Economic
472215,[email protected]
Development,
YIED,
1
Addis
Ababa,
Ethiopia,
Tel: +251-934-
1. Introduction
The concern on environmental impacts attributed by various development activities continues to be an
important agenda that have captured the attention of researchersand policy-makers.For the past few
decades developing nations including the Sub-Saharan Africa (SSA), have experienced a radical
change
in
policy
and
strategy
following
the
actions
previously
undertaken
in
developed
1
nations .Environmental resources are the foundation for social and economic development in SSA
region. In this region, about 60% of the population depends on agriculture for livelihood and the most
vulnerable region in the world for climate variability2.Ethiopia is one of the developing countryand the
fastest growing economy in SSA region. The country economy is heavily dependent on rain-fed
agriculture and its productivity is associated with the use of environmental assets such as expansion
into uncultivated lands and supported by agricultural inputs. Environmental degradation in Ethiopia
threatens the physical and economic survival and reduces the environment's ability to provide
ecosystem services. Mismanagement of natural resources and their underutilization has undermined
their contribution to the country’s overall development. Moreover, the growth of population is one of
the influencing factors for the degradations of environmental assets3. To overcome the challenges and
to ensure sustainable development, the government of Ethiopia has initiated different strategies,
policies, and institutional arrangements.However, sound environmental policy is fully dependent upon
sound science and still we have little scientific knowledge of the driving forces behind environmental
change4.Therefore, research concerning which driving factors have an impact on environment and CO2
emissions and the extent of their impact have been of great importance, since these driving factors
will directly influence the constitution of improving environmental impacts and CO 2 abatement
measures, policies and strategies5.
A considerable number of research has been contributed to improve the understanding of driving
forces of environmental impacts
6,7,8,9
and CO2 emissions10,11,12.Every country shares the responsibility
to reduce the rapid growth ofgreenhouse gases emissions in order to alleviate global climate change
worldwide5.A multidisciplinary approach is required to understand the dynamics of environmental
impacts and human-environment interaction to address theory, concepts, models,and applications
relevant to environmental and socioeconomic problems13. The crucial challenge of a sustainability
oriented environmental management is to find the proper balance between humans and the impacts
their activities have on ecosystems14. However, limited research has been conducted in evaluating the
driving forces of environmental impacts and CO2 emissions in developing countries, particularly in SSA
region6,5,15. Thus, this study evaluates the driving forces of CO2 emissions and explores the humanenvironment interactions.On top, an integrated approach employed to better understand the linkages
between the economy and environmental impacts in SSA region,by taking Ethiopia as an embodiment.
The human-environmental impacts interaction, has been conceptualized mathematically by the socalled environmental impacts are the multiplicative product of Population, Affluence, and Technology
(IPAT) equation16. IPAT has been utilized as an analytical framework for assessing human impacts on
2
the natural environment
11
.The IPAT model reformulated into a stochastic form as the Stochastic
estimation of Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model in
order to analyse the effects of driving forces and impact factors on a variety of environmental
impacts17.The IPAT framework is a worthwhile analytic tool for girding the environmental policy
because of its combination of simplicity andparsimonious specification with its robustness of
application. IPAT is also widely criticized, primarily for its simplistic formulation while it applied widely
and captures the important driving forces behind environmentalimpacts4.
Consequently, this study usesIPAT identity as a framework to assess the driving forces of CO2
emissions. Besides, because of the current economic activities inevitably induce more energy
consumption and CO2 emissions, and heavily dependent on natural resources, considering the
responsiveness of CO2 emissions is vital for the governments in SSA region to influence the behavior
of people. The findings may also contribute to identify which environmental impacts have to be
considered to create coherent indicators to be used to inform decision-making for a wide range of
purposes.This paper is organized as follows: Section 2 discusses literature review, Section 3describes
the methodology (i.e. data and methods, and estimation technique/econometrics approach), and
Section 4discloses the results and discussion. Finally, conclusions are provided in Section 5. The main
objective of this research is to evaluate the driving forces of CO2 emissionsin Ethiopia and to
contribute to the country sustainable development strategy by improving environmental changes and
mitigatingthe level of CO2 emissions.
2. Literature Review
At micro level, environmental valuation studies reveal information on both the structure and
functioning of ecosystems and the varied and complex roles of ecosystems in supporting human
welfare
18
. At macro level, the environmental impacts of economic activity can be looked at in terms of
extractions from or insertion into environment and by the responsiveness of the environmental
impacts due to changes in economic development16.Understanding the relationship between economic
development and the factors causing the environmental pressures is the basic premise of formulating
and adjusting the environmental policy. A sound environmental policy should be effective to reduce or
mitigate
the
environment
pressures
and
simultaneously
maintain
economic
development 19.
Althoughthe scientific consensus that humans have dramatically altered the global environment, we
have a limited knowledge of the specific forces that are driving those impacts. One key limitation to a
precise understanding of anthropogenic impacts is the absence of a set of refined analytic tools 10.
A
relationship
between
environmental
impacts
and
population,
has
been
conceptualized
mathematically by the so-called theenvironmental impacts are the multiplicative product of Population,
Affluence, and Technology (IPAT) model16.The IPAT model is an important theoretical and empirical
framework for identifying the drivers of environmental impacts and for estimating potential changes in
impacts due to changes in any of the driving forces/factors, including CO2 emissions,5 energy
consumption, ‘water footprint’, air pollution11, and ‘ecological footprint’20.The land use land cover
3
change (LULC) dynamics and other environmental impacts appear to track well with the IPAT model
6,13
.The rapid increase of CO2 emissions has caused many concerns among policy makers, and
population growth has been taken as one of the major driving forces behind increasing CO2 emissions
worldwide over the last two decades21.Despite its limitations, the IPAT model provides a useful
starting point for developing a better framework and for structuring empirical tests of competing
arguments17.Considerable progress has been made in developing a better understanding on how
human-actions lead to environmental change and in particular on what is most often called
anthropogenic drivers, which means the range of human-actions that cause environmental change and
the factors that shape those actions
22
. The nature and extent of the relationships between driving
forces of environmental impacts and CO2 emissions means that, with current knowledge, it is not
possible to develop probabilistic future for environmental impacts and emission scenarios. Though it
were possible to derive (subjective) probability distributions of the future evolution of individual
scenario driving force variables (for example: population, economic growth, or technological changes),
the nature of their relationships is known only qualitatively at best or remains uncertain (and
controversial) in many instances23.
There are two perspectives on the impact of demographic growth on environmental impacts: the
Malthusian tradition and the Boserupian approach5. IPAT-linked work invokes a neo-Malthusian or ecocentric vision-closed system with inflexible limits-and an exogenous role for technology in determining
these limits. This view claims that the environmental impacts takes placedue to population pressure
and solutions have to be found in limiting population, rather than in changing consumption and
behavioral patterns. In contrast, Boserupian or anthropocentric view-open systems with flexible limitsand an endogenous role for technology. This view, with the support of a significant number of case
studies, interprets the role of population growth in the context of broader conditions, with potentially
positive outcomes for welfare and the environment. Besides, at case study and regional level, the IPAT
formulation is insufficiently sensitive to capture the diversity, variability and complexity of real-world
situations24. IPAT bridge the approach to describe how our growing population contributes to our
environment, both positively and negatively25. Indeed, proper application of IPAT as an analytical tool
requires attention to certain statistical issues. Variability is a necessary, but not sufficient, condition
for assessing causality, and is, certainly, a requirement for statistical analyses4.
The concept of ecological elasticity has been also introduced to analyze environmental questions.
Ecological elasticity (EE) refers to the responsiveness or sensitivity of environmental impacts to a
change in any of the driving factors/forces. The ecological elasticity of population, affluence and other
factors for cross-national emissions of CO2 from fossil fuel combustion and for the energy footprint, a
composite measure comprising impacts from fossil fuel combustion, fuel wood, hydropower and
nuclear power are computed using STIRPAT model10. It can easily be calculate the EE of any of the
driving factors of environmental impacts. However,the operational measure of technology is not free
of controversy and is usually included in the error term since appropriate direct measures of
technology are lacking and any specific indicator is highly disputed5.Therefore, this study may
4
contribute to the knowledge gap on the linkages between the economy and the environment using
IPAT identity as an integratedapproach to assess the driving forces and the response ofCO2 emissions
in Ethiopia Context using vector autoregressive (VAR) model specification with co-integration and
Vector Error Correction Model (VECM)to estimate the long-and the short-run relationships between
CO2 emissions and the driving forces/factors.
3. Methodology
4.1. Data
A time-series data of Ethiopia on socioeconomic variables (population and GDP per-capita) for the
period 1971-2011 have been collected from Ethiopia Ministry of Finance and Economic Development
(MoFED) and from the database of the World Bank (WB) for Ethiopia economic indicators.Data on CO2
emissions, energy intensity and carbon intensity for the period 1971-2011 have been collected from
International Energy Agency (IEA). The definitions of variables used and their unit of measurement
presented in Table 1.
Table 1: Definitions of variables used in this paper
No
variable
Definition
Unit of measurement
1
CO2 emissions
Mt
2
Population
Carbon dioxide emissions from the burning of fossil
fuels and the manufacture of cement. It includes carbon
dioxide produced during consumption of solid, liquid,
and gas fuels and gas flaring
Mid-year population
3
GDP per-capita
Gross domestic product divided by midyear population
constant local price year 2000
Constant local price year
2000
4
Energy intensity
It is calculated using the ratio of total primary energy
supply (TPES) per PJ (including biofuels and other nonfossil forms of energy) per GDP
TPES per GDP
Number
Energy intensity of the economic output.
5
Carbon Intensity
CO2 emissions per unit of total primary energy supply
(TPES).Carbon intensity of the energy mix.
CO2 emissions per unit
TPES
Note: The logarithm of each variable presented as log in the regression models result.
4.2. IPAT equation
The IPAT equation and related formulas were renowned, along with the modern environmental
movement, since 1970’s as part of the ongoing debate on the driving forces of environmental
change11. Although first used to quantify contributions to unsustainability, the formulation has been
reinterpreted to assess the most promising path to sustainability 26. The first simple original
5
formulation of the theoretical framework to analyse environmental impact (I) was presented by 25.This
simple formulation of the IPAT model is:
I  P*F
......................................................................... (1)
Where I is the total impact, Fmeasures the per capita impact and P is population size. Impact
increases as either P or F increases or if one increases faster than the other declines 26. This model
presents some limitations that the factors are not independent among each other, due to
multiplicative and non-linear relationships. To indicate that the equation is non-linear and the
variables are interdependent, the model can be expanded as:
I  P( I , F ) * F .................................................................................... (2)
This variant shows that P depends on I and F. Technology (T) is not introduced explicitly at this stage,
but its impact is hidden in F, as a per capita impact. More general, the IPAT model specifies that
environmental impacts (I) are the multiplicative product of population size, affluence (per capita
production or consumption), and technology (impact per unit of production/consumption).
I  P * A *T
27
....................................................................... (3)
revised the IPAT model by disaggregating T into consumption per unit of GDP (C) and impact per
unit of consumption (T) and renamed the model as ImPACT.
28
proposed and modified the IPAT model
as I=PBATmodel, where B are behavioral choices. However,
29
argued that in I=PBAT model the
aspects of behavior are implicitly involved in each factor in the right-hand side of the equation I=PAT.
Thus, behavioral choices could only include aspects of behavior that are not already included in P, A
and T, and as such, B is also very difficult to define precisely.4introduced the concept of Ecological
Elasticity (EE) that refers to the responsiveness or sensitivity of environmental impacts to a change in
any of the driving factors and the elasticity of IPAT model. 17 reformulated the IPAT model into a
stochastic form as the Stochastic estimation of Impacts by Regression on Population, Affluence, and
Technology (STIRPAT) model in order to analyse the effects of driving forces on a variety of
environmental impacts. The basic STIRPAT model formulation is:
I i  aPi Ai Ti  i .............................................................................. (4)
b
c
d
Where a is the constant scale of the model, b, c and d are the exponents of P, A, and T, respectively
and ε is the error term. The subscript ivaries across observations over units, over time or both. The
original IPAT model assumes proportionality and sets that a=b=c=d=εi=1, while STIRPAT treats these
as parameters/coefficients to be estimated17,11. The STRIPAT model is the most standard formulation
of IPAT model for quantitative social research and can be estimated using regression techniques 4.
Statistical analysis of ordinary least square (OLS) and partial least square (PLS) regression will be
employed to analyse the responsiveness of CO2 emissions using SRIRPAT model6. The STRIPAT model
can also be transformed in to logarithmic function, which measures the responsiveness or sensitivity
of environmental impacts to a change in any of the driving forces17. The model then takes the
following form:
ln I i  a  b(ln Pi )  c(ln Ai )  d (ln Ti )   i ................................................... (5)
6
Where a andεi is the natural logarithmic of a and εi in Equation 5, respectively.In the STIRPAT model,
it is possible to substitute a vector of cultural, political and social structural variables for technology
(T) and examine the net effect of each on environmental impacts20,6. We employ STIRPAT as an
integratedapproach for better understanding of the responsiveness of CO2 emissions as an
environmental impact due to changes in driving forcesin Ethiopia economy context using the following
estimation techniques or econometrics approach.
4.3. Estimation techniques (or Econometrics approach)
Once the model is specified, the study will use vector autoregressive (VAR) model to examine the
responsiveness of CO2 emissions using STIRPAT modelas a base and the interactions of all variables in
the specification.This estimation technique put forth no need of restriction on the variables as
exogenous and endogenous and blends.30(1988) procedure for vector autoregressive (VAR) model
specification with co-integration and error correction techniques to estimate long-and short-run
coefficientsthat indicate the relationship of the variables, if the variables are stationary after
differencing.
A VAR system of equation may be specified as:
yt  A1 yt 1  A2 yt  2  ...  Ap yt  p  ut …………………………………………………………………………………….. (6)
Where the Ai’s are (nxn) coefficient matrices, Ytare all variables mentioned in the model specification,
and
ut  (u1t , u2t ,...,unt )' is n dimensional vectors of multivariate random errors with zero mean and
covariance matrix, that is innovation term. The optimal lag length (p) will be determined by Akaike
Information Criteria, AIC=T.log(SSR)+2n, Schwarz and Risanen Criteria SBC=T.log(SSR)+n.log(T)
also called Schwartz andHannan-Quinn Criteria HQC=T. log(SSR)+2n.log(log(T)).
The study also applies endogenous structural breaks unit root test of both31 (1992) test and32(1998)
tests in order to address problems that are associated with single break and two-breaks, respectively.
If there is no structural breaks in a time series data, the result generated from these tests and
Augmented Dickey-Fuller (ADF) test should be the same. However, the Zivot and Andrews (ZA) model
considers one structural break and uses many dummy variables for each structural break year31.As
the exact endogenous break is unknown, the ZA model then assumes every point as a potential break.
It, therefore, sequentially conducts a regression for every structural break point, in which the
minimum t-statistic indicates where the endogenous structural break pointis found.
The following equation gives the ZA model.
k
yt     t   DU1t   DT1t   yt 1   yt i   t ………………………………………… (7)
i 1
7
1 if t > TB 
1, if t > TB 
DU1t = 
 and , DT1t = 

0, otherwise 
0, otherwise 
Where
yt
is a time series variable,
t
is the time trend,
indicating the mean shift (change in the level),
DU1t is
the intercept dummy variable
DT1t and stands for the slope dummy representing
change in the slope of the trend function. Besides,
TB
represents a potential break point, k denotes
lag length.The null hypothesis states that the time series that excludes any structural break is nonstationary whereas the alternative hypothesis indicates that the series that includes one structural
break is stationary.
32
(1998), alternatively in the presence of two breaks,the time
Clemete Montanes
and Reyes (CMR) test of stationarity proposed two models:- Additive outlier (AO) model and Innovative
outlier (IO) model in order to address instantaneous structural breaks and gradual change,
respectively.
The following equation gives the IO model as below.
k
yt     yt 1  1DT1t   2 DT2t  1t DU1t  2t DU 2t    i  yt i   t ……………………….(8)
i 1
The AO modelhas two stages in order to test for stationarity. The first step removes the deterministic
part of the variable by modeling:
yt    1t DU1t  2t DU 2t  yt
The study uses and tests the following model after removing the deterministic party of the variable as
explained in equation 9:
k
yt   yt 1  1i DT1t i  2i DT2t i   yt i   t ………..………..………………….… (9)
i 1
Where
1, if t > TB 
1, if t > TB 
DU1t = 
 , DU 2t = 
 for representing intercept dummy
0, otherwise 
0, otherwise 
1, if t > TB 
1, if t > TB 
DT1t = 
 , DT2t = 
 for representing the slop dummy
0, otherwise 
0, otherwise 
After presenting the VAR results, the Granger Causality Test (GCT) can be run for testing the Granger
causality using Wald test that involves the effect of past values of all other variables on the current
value of the remaining variable. It is different from exogeneity tests, which look at whether the
8
current values of z explains current and future values of y. Moreover, it checks the stability condition
of the VARX (X stand for exogenous variables) model and analysis of one-time shock to apply
forecasting techniques for some specified target year. The requirement of satisfying the stability
condition of the VARX model indicates that the unit roots or the solutions of the VARX system are
below one, or all the Eigen values lie inside the unit circle, which is the necessary and sufficient
condition for stability. Otherwise, the impact of the impulse (shock) in some variables might not
decrease/increase with time. A crucial condition for the VAR model to be valid and consistent requires
the covariance to be stationary in order to avoid the formation of explosive roots. This confirms that
using the VARX model satisfies the stability condition and can be used for forecasting.
As the VAR model is stable, the next points will be to discuss the impulse response functions and the
variance decomposition in response to a one-time shock in the system. The impulse response function
presents the dynamic interactions among endogenous variables and traces the effect of a one-time
shock on current and future values of the endogenous variables. It sheds light for empirical causal
analysis and policy effectiveness. The variance decomposition also presents the separation of the
variation in an endogenous variable into the component shocks during the forecast period.
5. Results and Discussion
5.1.
Results
5.1.1. Descriptive Analysis
This part of the result gives the general features of the data in terms of both central tendency and
variation measures of the distribution as well as shape of the distribution over the period 1971-2011.
The statistical summary of the variables are presented in Table 2. Except Energy intensity, the
measures of skewness indicate that all variables have tails on the right side longer than the left ones
and bulk of values falls to the left of the mean. On top of this, Kurtosis also explains the peakedness
of the probability distribution of each variable so that energy intensity behaves in the way of having
mesokurtic distribution while per capita income is being characterized with leptokurtic shape. All other
variables are indicatinga platykurtic shape of the distribution,as their values are below three.
Table 2: Descriptive Analysis
Variable
Obs.
Mean
CO2emissions 41
120.9512
Energy intensity
Carbon intensity
population
Per capita income
41
Std. Dev.
Min
Max Skewness
Kurtosis
70.49786
54.0
265 0.862443
2.276553
41
89.756114.49617
55.0
117
3.000909
41
106.4634
28.16744
67.0
-0.593627
163
0.7600003
2.147187
54.06098
16.32654
31.7
84.7 0.3395142
1.838586
41 1063.171
219.7626
855
1860 2.210609
7.409546
9
Looking at Table 3, the partial and semi partial correlation of CO 2 emission with energy intensity is
negative and statistically insignificant while all others are positive and statistically significant.
Table 3: Partial and Semi Partial Correlations of CO2 emission
Variable
Partial
corr.
Semipartial
corr.
Partial corr.
^2
Semipartial
corr. ^2
Significant
value
Population
0.9246
0.1252
0.8550
0.0157
0.0000
Per Capita Income
0.4051
0.0229
0.1641
0.0005
0.0116
Energy Intensity
-0.1243
-0.0065
0.0154
0.0000
0.4573
Carbon Intensity
0.9308
0.1314
0.8664
0.0173
0.0000
5.1.2.
Stationarity Test
All time-series data must be stationary, meaning constant mean and variance over time, in the
regression model. Otherwise, the regression result becomes spurious. This paper uses Zivot-Andrews
unit root test that assumes one structural break and Clemente-Montanes-Reyes unit-root test that
accounts for two structural breaks in the time-series. These tests have comparative advantage over
the DF test and ADF test that presume there is no structural beak in the time-series. Table 4 provides
the Zivot-Andrews unit root test and all variables are non-stationary at level form (before making the
data at first difference form). However, except energy intensity, all of them are stationary at first
difference form.
Table 4: Zivot-Andrews unit root test for allowing for one break in intercept
At Level Form
At Difference
Variables (in log)
CO2emissions
Population
Per Capita Income
Energy Intensity
Carbon Intensity
Break year
Minimum t-statistics
Break year
Minimum t-statistics
2000
1986
2004
2004
2001
-4.068
-2.617
-0.476
-0.982
-4.189
1994
1982
2004
2004
2003
-7.139*
-6.507*
-6.846*
-4.141
-7.221*
Critical Values: 1% level of significance (-5.43) and 5% level of significance (-4.80)
One of the interesting points in this test is that the year chosen for structural break for each variable
is not uniform. The ZA test points out that energy intensity is non-stationary in the existence of one
structural break. This claims the CMR unit-root test that enables to examine the stationarity condition
in the existence of two structural breaks in the time series for both additive outlier (AO) and
innovation outlier (IO). The result presented in Table 5 confirms that energy intensity is stationary at
10
the first difference form, as t calculated is greater than t tabulated. Note that both t-values are
considered in absolute value in order to compare for decision.
Table 5: Clemente-Montanes-Reyes unit-root test for two breaks with AO and IO models
Variable
At Level form
(log at level form)
Min t
At Difference form
Optimal Breakpoints
Min t
Optimal Breakpoints
(year)
(year)
Energy Intensity (in AO model)
-2.827
1982 and 2005
-5.631*
1993 and 2004
Energy Intensity (in IO model)
-2.679
1982 and 2003
-6.555*
1994and 2004
N.B:- Min.‘t’ is the minimum t-statistics calculated. 5% critical value for the two breaks; -5.490
Finally, all variables that are expressed in first difference form are stationary when the study considers
structural break using by ZA unit root test and its complement, the CMR unit root test. Once it is
confirmed that all variables are stationary at first difference, the next step is to conduct co-integration
and Error correction model with the optimal lag length.
5.1.3. Selection and Determination of Lags order criteria
One of the most common practices in the VAR model is to select the optimal lagged term using some
criteria such as FPE: Final prediction error AIC: Akaike information criterion
SBIC: Schwarz
information criterion or Bayesian information criterion (BIC), and HQIC: Hannan-Quinn information
criterion. Before selecting the lag length, two situations should be identified. First, too short lag length
in the VAR may not capture the dynamic behavior of the variables
33
,thus,the optimal lag length would
be selected by the smallest lag shown under the criteria. Second,
34
(1992) point out that too long lag
length will distort the data and lead to a decrease in
in power of explaining the dynamic behavior of the
variable.
Table 6:-Selection of the Optimal Lag Length
Sample: 5 - 41
lag
LL
LR
Number of obs
df.p
FPE
AIC
=
37
HQIC
6.0e-13
SBIC
0
263.033
1
460.933
395.8
2
482.696
43.526
25
0.012
6.9e-17
-23.1187
3
506.507
47.622
25
0.004
9.1e-17
-23.0544
-21.8265
-19.5714
4
547.756
82.497* 25
0.000
6.0e-17
-23.9327*
-22.3211
-19.3612
25
0.000
5.3e-17*
-13.9477
-23.2937
-13.871
-22.8332*
-22.2745
-13.73
-21.9875*
-20.7241
According to Table 6, the optimal lag length is one based on FPE, HQIC and SBIC criteria, and the LR
test and AIC criteria indicate a lag length of four. Therefore, one lag length is considered in the VAR
11
model for having well defined co-integration and Vector Error Correction Model (VECM) as presented
below.
5.1.4. Johansen tests for Co-integration
In order to check whether there is co-integration or long-run relationship among variables, it is
common to apply the Johansen tests based on the co-integration rank that shows the number of cointegrating vectors
30
. The co-integration analysis also provides a framework for estimation, inference,
and interpretation when each variable is not stationary individually while has stationary linear
relationship together. If there is a stationary linear combination of non-stationary variables, the
variables combined are said to be co-integrated. The best way of testing co-integration is by using the
system Maximum Likelihood (ML) estimator of Johansen test which presented in Table 7.
Table 7: Johansen tests for co-integration
Johansen tests for cointegration
Trend: constant
Sample: 2 - 41
maximum
rank
0
1
2
3
4
5
parms
5
14
21
26
29
30
Number of obs =
Lags =
LL
459.30824
475.71236
488.99006
499.54296
502.46824
502.83786
eigenvalue
.
0.55966
0.48515
0.41001
0.13607
0.01831
40
1
5%
trace
critical
statistic
value
87.0592
68.52
54.2510
47.21
27.6956*
29.68
6.5898
15.41
0.7392
3.76
As the rule of thumb, as the log-likelihood of the unconstrained model with the co-integrating
equations is significantly different from the log-likelihood of the constrained model that excluded the
co-integrating equations, we reject the null hypothesis of no co-integration. Besides, the above result
shows that the trace statistic value (27.6956) is less than the critical value (29.68) as moving from
the rank zero in ascending order. This leads to accept the null hypothesis that states the maximum
rank is two and thereby two co-integration equations is used in the VAR model. Therefore, the
existence of co-integration approves to conduct the Vector Error Correction Model (VECM) in order to
evaluate the both long-run and short-run relationships and theuse of vector error correction model is
also justified.
5.1.5. Vector Error Correction Model: The Long Run dynamics equilibrium
The Vector Error Correction Model (VECM) is a special case of the VAR for variables that are stationary
in their first differences after taking into account any co-integrating relationships.The focus of this
paper is the long-run response of CO2 emissions due to changes in the driving forces, the result
presented in Table 8.The results show the long-run relationshipamong CO2 emission, and per capita
12
income, population, energy intensity and carbon intensity for the first co-integrating vector over the
period of 1971-2011.
Table 8: Vector Error Correction Model Co-integration Equation
Cointegrating equations
Equation
Parms
_ce1
4
Identification:
chi2
P>chi2
144225
0.0000
beta is exactly identified
Johansen normalization restriction imposed
beta
Coef.
_ce1
logco2emis~s
logpop
logpicg
logenergyi~y
logcarboni~y
_cons
1
-1.114158
.198008
.2397298
-.9363568
1.638943
Std. Err.
.
.0138183
.0439834
.0433167
.015536
.
z
.
-80.63
4.50
5.53
-60.27
.
P>|z|
.
0.000
0.000
0.000
0.000
.
[95% Conf. Interval]
.
-1.141241
.111802
.1548306
-.9668069
.
.
-1.087074
.284214
.3246289
-.9059067
.
In Table 8, all of the variables coefficients are statistically significant at onepercent. As the variables
are expressed in terms of logarithms, the coefficients are being interpreted as long-run elasticities.
The coefficients measure the responsiveness of the CO2 emissions due to changes in the driving forces
of CO2 emissions. Accordingly, both a percentage increase in per capita income and energy intensity
intensifies the level of CO2 emission by 0.198 percent and 0.2398 percent, respectively.However, a
one percentage increase in population likely leads to a decreasing CO2 emissions by 1.115 percent in
the long run. Similarly, there is 0.9364 percent negative response of CO2 emissions when there is an
increase by a unit percentage change in the level carbon intensity.
5.1.6. Vector Error Correction Model: The Short-Run Dynamics equilibrium
Having a result on the long-run equilibrium dynamic relationships among variables, the short-run
dynamic relationship should be the subsequent point of presentation. As presented in Table 9, the
estimates of the short-run parameters that explain the speed of adjustment towards the long run
equilibrium position per annum, implying rapid adjustment toward equilibrium. For all variables except
energy intensity the ECM or the error term at lag one are negative and statistically significant,
showing the fact that any short-run variation and fluctuation between variables will move towards a
stable long-run equilibrium relationship. This indicates that these variables quickly fall back toward the
long equilibrium position as the values increase over time. However, it is positive and statistically
insignificant for the energy intensity. This means that energy intensity moves up toward the long run
equilibrium position as its value increases.
13
Table 9: Vector Error Correction Model3
Vector error-correction model
Sample:
2 - 41
Log likelihood =
Det(Sigma_ml)
=
Equation
475.7124
3.22e-17
Parms
D_logco2emissi~s
D_logpop
D_logpicg
D_logenergyint~y
D_logcarbonint~y
No. of obs
AIC
HQIC
SBIC
2
2
2
2
2
Coef.
RMSE
.082466
.009919
.051508
.057139
.076652
Std. Err.
R-sq
chi2
P>chi2
0.4791
0.8679
0.1986
0.0822
0.4053
34.95622
249.5602
9.41943
3.404987
25.9006
0.0000
0.0000
0.0090
0.1822
0.0000
z
P>|z|
=
40
= -23.08562
= -22.87189
= -22.49451
[95% Conf. Interval]
D_logco2em~s
_ce1
L1.
-6.985845
1.348294
-5.18
0.000
-9.628453
-4.343238
_cons
.0090453
.0141211
0.64
0.522
-.0186316
.0367223
_ce1
L1.
-.3290528
.1621694
-2.03
0.042
-.6468991
-.0112066
_cons
.0232469
.0016985
13.69
0.000
.019918
.0265758
D_logpicg
_ce1
L1.
-2.120246
.8421441
-2.52
0.012
-3.770818
-.4696742
_cons
.0057694
.0088201
0.65
0.513
-.0115176
.0230564
D_logenerg~y
_ce1
L1.
1.468319
.9342086
1.57
0.116
-.3626961
3.299334
_cons
-.0028304
.0097843
-0.29
0.772
-.0220073
.0163464
D_logcarbo~y
_ce1
L1.
-6.270079
1.25323
-5.00
0.000
-8.726365
-3.813792
_cons
-.0139117
.0131255
-1.06
0.289
-.0396372
.0118139
D_logpop
Having results of co-integration and VECM, the subsequent presentations are post/estimation tests
which are important to confirm and improves the consistence of the model results.
5.1.7. Granger causality Test
Noting the VAR model considers all variables as endogenous variables, the Granger causality test is so
an important statistical hypothesis test for setting whether one time series is useful in forecasting
another or not. The Granger causality test also points out the causality relationship between variables
in the sense of testing the predication capacity of the past values independent variables about
dependent variable, as presented Table 10.
In the Granger causality test, the null hypothesis states that the excluded variable from equation does
not Granger-causes. Against this hypothesis, both per capita income and energy intensity do Granger3
Note that the presence of autocorrelation is being tested by Lagrange-multiplier test and it gives that the Chi square
value becomes 32.8250 with the probability of getting Chi2 is 0.13550. Thus, we fail to reject the null hypothesis
that states there is not autocorrelation at lad order.
14
cause CO2 emission. In terms of joint statistical test, all variables do jointly Granger-cause one
another. Specifically, all explanatory variables (population, per capita income, energy intensity and
carbon intensity) do jointly Granger cause CO2 emission.
Table 10: Granger causality Wald tests
Granger causality Wald tests
Equation
Excluded
logco2emissions
logco2emissions
logco2emissions
logco2emissions
logco2emissions
logpop
logpicg
logenergyintens~y
logcarbonintens~y
ALL
1.6767
19.939
23.534
.561
38.247
chi2
df Prob > chi2
1
1
1
1
4
0.195
0.000
0.000
0.454
0.000
logpop
logpop
logpop
logpop
logpop
logco2emissions
logpicg
logenergyintens~y
logcarbonintens~y
ALL
.04577
6.3777
5.6999
.10108
8.1375
1
1
1
1
4
0.831
0.012
0.017
0.751
0.087
logpicg
logpicg
logpicg
logpicg
logpicg
logco2emissions
logpop
logenergyintens~y
logcarbonintens~y
ALL
2.6973
3.1009
8.9
2.6002
22.278
1
1
1
1
4
0.101
0.078
0.003
0.107
0.000
logenergyintens~y
logenergyintens~y
logenergyintens~y
logenergyintens~y
logenergyintens~y
logco2emissions
logpop
logpicg
logcarbonintens~y
ALL
2.2341
2.4207
.09873
2.2204
13.194
1
1
1
1
4
0.135
0.120
0.753
0.136
0.010
logcarbonintens~y
logcarbonintens~y
logcarbonintens~y
logcarbonintens~y
logcarbonintens~y
logco2emissions
logpop
logpicg
logenergyintens~y
ALL
.46864
.92772
20.404
23.575
38.049
1
1
1
1
4
0.494
0.335
0.000
0.000
0.000
As per the result presented in Table 10, the lagged values of per capita income and energy intensity
help to predict CO2 emission. The converse does not hold true, its reveals that the relationship
between CO2 emission and the driving factors confirms the existence of unidirectional causality.
5.1.8. The stability of the model
Checking the stability of the model is an important test in time-series econometrics. The stability of
the VAR model requires the moduli of the Eigen values to lie within the unit circle. Otherwise, the
system is not stationary. Rather it is explosive or non-convergent. Table 11 and the associated
graphical presentation confirm that all roots are less than oneand no root lies outside the unit circle.
This result signifies and satisfies the stability condition of the model. As the VAR model is stable, it is
possible to present the impulse response functions and variance decomposition in response to a onetime shock in the system.
15
Table 11: Eigen vales Stability Condition
Figure 1: Roots of the co-integration matrix
Roots of the companion matrix
All the eigenvalues lie inside the unit circle.
VAR satisfies stability condition.
N.B: - For a characteristic equation of the type,
roots:
0
.995829
.995829
.74725
.263654
.049973
-.5
.9952561 + .03376271i
.9952561 - .03376271i
.7472496
.2636535
.04997289
-1
Modulus
Imaginary
Eigenvalue
.5
1
Eigenvalue stability condition
a 2  b2
35
-.5
0
Real
b
a2  b  c  0 1 , 2  2a 
,
1 , 2     Imaginary roots: 1 , 2    i where i 
a complex number, a + bi is
-1
i
.5
1
b 2  4ac
   i
2a
; Real
 1 . In general case, the modulus of
.
5.1.8. Impulse Response Function and Variance Decomposition
The Impulse Response Function (IRF) refers to the dynamic interactions among endogenous variables
and traces the effect of a one-time shock on current and future values of the endogenous variables.
The overriding objective of IRF is to trace the time path of structural shocks in the VAR system and
additionally check of co-integration test’s findings. IRF show the current and lagged effects overtime
of changes in errors (shocks) on the variables in the VAR. If the system of equations is stable, any
shocks should decline to zero, and if it is an unstable system, it would produce an explosive time path.
It is important to specify the orthogonal impulse response function because it gives impulse responses
assuming (counterfactually) that the VAR residuals are uncorrelated. The resulting IRFs are shown in
Figure 2.
16
Figure 2: Impulse Response Function for preferred ordering
L1, logcarbonintensity, logcarbonintensity
L1, logcarbonintensity, logco2emissions
L1, logcarbonintensity, logenergyintensity
L1, logcarbonintensity, logpicg
L1, logcarbonintensity, logpop
L1, logco2emissions, logcarbonintensity
L1, logco2emissions, logco2emissions
L1, logco2emissions, logenergyintensity
L1, logco2emissions, logpicg
L1, logco2emissions, logpop
L1, logenergyintensity, logcarbonintensity
L1, logenergyintensity, logco2emissions
L1, logenergyintensity, logenergyintensity
L1, logenergyintensity, logpicg
L1, logenergyintensity, logpop
L1, logpicg, logcarbonintensity
L1, logpicg, logco2emissions
L1, logpicg, logenergyintensity
L1, logpicg, logpicg
L1, logpicg, logpop
L1, logpop, logcarbonintensity
L1, logpop, logco2emissions
L1, logpop, logenergyintensity
L1, logpop, logpicg
L1, logpop, logpop
.1
0
-.1
-.2
.1
0
-.1
-.2
.1
0
-.1
-.2
.1
0
-.1
-.2
.1
0
-.1
-.2
0
5
10
15
20
0
5
10
15
20
0
5
10
15
20
0
5
10
15
20
0
5
10
15
20
step
95% CI
orthogonalized irf
Graphs by irfname, impulse variable, and response variable
The diagonal panels in above show the estimated effects of shocks to each variable on future values of
its own value in terms of percentage change. In all cases, the shock dies out over time, reflecting the
stationarity of the variables. A one-standard deviation shock to each variable falls within zero and one
percent. The off-diagonal panels show the effects of a shock in one variable on the dynamic time path
of other variables in terms of percentage. The effect decays to zero over time. Overall, the short-run
equilibrium adjustment process is quite fast.
5.1.9. Variance Decomposition with Optimal Lag Length
Variance Decomposition or forecast error indicates the amount of information each variable
contributes to the other variables in a VAR model. It measures the extent to which each shock
contributes to unexplained movements and forecast errors in each variable. In other word, it
determines how much of the forecast error variance of each of the variable can be explained by
exogenous shocks to other variables over a series of time horizons. Note that all variance
decompositions start at zero because there is no forecast error at a zero lag.Accordingly, the Figure 3
indicates that the variation in CO2 emission is majorly explained by its own lagged values, followed by
energy intensity and per capita income.
17
Figure 3: Variance decompositions with preferred ordering
L1, logcarbonintensity, logcarbonintensity
L1, logcarbonintensity, logco2emissions
L1, logcarbonintensity, logenergyintensity
L1, logcarbonintensity, logpicg
L1, logcarbonintensity, logpop
L1, logco2emissions, logcarbonintensity
L1, logco2emissions, logco2emissions
L1, logco2emissions, logenergyintensity
L1, logco2emissions, logpicg
L1, logco2emissions, logpop
L1, logenergyintensity, logcarbonintensity
L1, logenergyintensity, logco2emissions
L1, logenergyintensity, logenergyintensity
L1, logenergyintensity, logpicg
L1, logenergyintensity, logpop
L1, logpicg, logcarbonintensity
L1, logpicg, logco2emissions
L1, logpicg, logenergyintensity
L1, logpicg, logpicg
L1, logpicg, logpop
L1, logpop, logcarbonintensity
L1, logpop, logco2emissions
L1, logpop, logenergyintensity
L1, logpop, logpicg
L1, logpop, logpop
1.5
1
.5
0
-.5
1.5
1
.5
0
-.5
1.5
1
.5
0
-.5
1.5
1
.5
0
-.5
1.5
1
.5
0
-.5
0
5
10
15
20
0
5
10
15
20
0
5
10
15
20
0
5
10
15
20
0
5
10
15
20
step
95% CI
fraction of mse due to impulse
Graphs by irfname, impulse variable, and response variable
5.2.
Discussion
Anthropogenic greenhouse gas (GHG)emissions are mainly driven by population size, economic
activity, lifestyle, energy use, land-use patterns, technology and climate policy36.The importance of
informed decision on environmental policies has highlighted the need for a deeper scientific
understanding of the driving forces/factors impacting environment 20. However, the scientific
knowledge of the driving forces behind environmental impacts is meager in Ethiopian economy
context.Under the current development practices of Ethiopia, greenhouse gas (GHG) emissions would
more than double from 150 Mt CO2 emissions in 2010 to 400 Mt CO2 emissions in 2030. To address
CO2 emissions challenges, the country initiated the Climate-Resilient Green Economy (CRGE) using
sectoral approach and prioritized more than 60 initiatives, which could help to achieve the
country’sdevelopment goals while limiting greenhouse gas emissions to around today’s 150 Mt CO2
emissions around 250 Mt CO2 emissions in 203037. Thus, improved understanding of CO2 emissions
elasticitiesdue to changes in GDP per capita, population and technology is vital both for climate change
policy interventions and negotiations; and for generating projections of CO2 emissions. Indeed, the socalled Kaya Identity—essentially IPAT with energy intensity (energy consumption over GDP) and
carbon intensity of energy (carbon emissions over energy consumption) in place of technology (T)—
plays a core role in the Intergovernmental Panel on Climate Change (IPCC) estimates of future CO2
emissions38.
18
The STIRPAT formulation is amenable to examining the effects of driving factors on impacts along with
alternative
conceptualizations
of
the
driving
forces
of
environmental
impacts17.In
this
study,theSTIRPAT model estimatedby considering the time dimension of the data,stationarity test,
causality, stability and the dynamic interactions of endogenous variables. Besides, energy intensity
and carbon intensity considered as technology (T) in our formulation to comprehend the
responsiveness of CO2 emissions in Ethiopian economy.We treat CO2 emissions as the dependent
variable and establish the STIRPAT model to analyse the driving forces of CO 2 emissions.The main
focus of the discussion is the regression coefficientsthat show the long-run responsiveness of CO2
emissions due to changes in the driving forces ofCO2 emissions.
The key finding is that the long-run responsiveness of CO2 emissions for GDP per-capita and energy
intensity is positive and significant. The effect of energy intensity on CO2 emissions responsiveness is
highest. Thus, whenever the coefficients are positive an increase in any driving forces (GDP per capita
and energy intensity) increases impacts (CO2 emissions) through the multiplicative effect on the other
driving forces4. The result suggests that higher GDP per-capita can induce more energy consumption
(use) and leads to more CO2 emissions in Ethiopia.This result is consistent to the findings by 5(2006)
that indicated the effect of energy intensity on CO2 emissions is higher for lower income countries as
compared to high and upper income countries.These might be due to high investments that demand
energy, higher level of urbanization and the slow improvement of energy efficiency. Besides, the
coefficients are exponents in multiplicative function.
The result also suggests the impact of the growth of GDP per-capita on CO2 emissions presents an
approximately increasing trend with the increase of economic development level. The effect of energy
intensity on CO2 emissions is positive because energy intensity is affected by the level of economic
development, thus measures should be taken to improve energy-use and energy-consuming
structures to reduce the impact of energy intensity on CO 2 emissions. On the other hand, the effect of
population on CO2 emissions is negative, which has some similarity with Boserupian view that the
impact of population on the state of the environment is likely either a non-relationship between two
variables, or even a negative elasticity.This finding is contrary to the Malthusian perspective, which
holds that population growth increases environmental impacts.This result supports the findings of
11
(2004) that indicated the high level of consumption even in slow population growth in developed
nationsis at least as great threat to the environment as rapid population growth elsewhere. It is also
consistent to the argument that global climate change clearly belongs to the developed (affluent)
world and its moderately sized population, not to the less developed world and its large population 4.
The results from Granger causality test also reveal that all explanatory variables do jointly Granger
cause CO2 emission. The findings imply that the relationship between CO2 emissions and the driving
factors confirms the existence of unidirectional causality. As a result, measures should be taken to
sustainthe driving forces of environmental impacts in order to maintain the level of CO2 emissions.In
addition,evidence from this empirical analysis suggests that a detailed understanding of the underlying
driving forces of CO2 emissions is required prior to any policy interventionsin Ethiopia in order to
19
achieve the initiativesof CRGE. Our efforts to study the driving forces of CO2 emissions in a more
analytical and systematic manner are also greatly improve the acceptance of integrated approach
studies.
6. Conclusion
Ethiopia initiated the Climate-Resilient Green Economy (CRGE) using sectoral approach and prioritized
more than 60 initiatives, which could help to achieve the country’s development goals while limiting
greenhouse gas (GHG) emissions to around today’s 150 Mt CO 2 emissions around 250 Mt CO2
emissions in 2030. However, the linkages between the economy and environmental impacts, and the
responsiveness of CO2 emissions due to changes in economy still require further investigation. In
addition, the scientific knowledge of the driving forces behind environmental impacts is meager in
Ethiopia economy context. Thus, improved understanding of CO2 emissions elasticities due to changes
in GDP per-capita, population and technology is vital both for climate change policy interventions and
negotiations, and for generating projections of CO2 emissions. This study explores the humanenvironment interactions using an integrated approach to analyse the driving forces of CO2 emissions
and its responsiveness in Ethiopia.
This study applied the so-called environmental impacts are the multiplicative product of Population,
Affluence, and Technology (IPAT) identity as a framework to assess the driving forces of CO2
emissions using vector autoregressive (VAR) model specification with co-integration and Vector Error
Correction Model (VECM). The approach illustrated in this paper serves as a demonstration of the
integrated research, combining the environmental approach and time-series data analysis in a
coherent manner that is interactive and comprehensive for seeking better understanding of a unique
environmental concern.Consequently, we anticipate that our analytical and integrated approach
increases the relevance of studies to better understanding of human-environment interactions.
The finding indicates that the long-run responsiveness of CO2 emissions for GDP per-capita and energy
intensity is positive and significant. The effect of energy intensity on CO 2 emissions responsiveness is
highest.These might be due to high investments that demand energy, higher level of urbanization and
the slow improvement of energy efficiency. This result implies that higher GDP per-capita can induce
more energy consumption and more CO2 emissions in Ethiopia context. The impact of population on
CO2 emissions is negative, which has some similarity with Boserupian view and contrary to the
Malthusian
perspective,
which
holds
that
population
growth
increases
environmental
impacts.Therefore, considering the driving forces and the responsiveness of CO2 emissions would
enable us to inform decision-makers to make emission reduction measures and to maintain economic
development, population, and energy use, especially in lower-income level countries.
Acknowledgments: This research is financed by the Environmental Economics Unit (EEU), at the Department of
Economics, University of Gothenburg (UoG).We are very grateful for the Environment for Development (EfD)
initiative at the Environment and Climate Research Center (ECRC) and Ethiopian Development Research Institute
(EDRI) for their countless support in facilitating countless issues and favorable research environment.
20
Reference
1.
Mpakati-Gama, E. C., Wamuziri, S. & Sloan, B. Environmental monitoring and evaluation in SubSahara Africa – a state of the art review. Built, The Review, Human Environment4, 56–63 (2011).
2.
IAASTD. Agriculture at Crossroads: Sub-Saharan Africa. (2009).
3.
EPA. Ethiopia Environment Outlook: Environment for Development. (2007).
4.
York, R., Rosa, E. A. & Dietz, T. Bridging Environmental Science with Environmental Policy :
Plasticity of Population , Affluence , and Technology. 83, (2002).
5.
Fan, Y., Liu, L.-C., Wu, G. & Wei, Y.-M. Analyzing impact factors of CO2 emissions using the
STIRPAT model. Environmental Impact Assessment Review26, 377–395 (2006).
6.
Mulatu, D. W. Linking The economy to Ecosystems: Land use change and Ecosystem services
valuation at basin level, PhD Thesis, University of Twente. (2014).
7.
Long, H., Tang, G., Li, X. & Heilig, G. K. Socio-economic driving forces of land-use change in
Kunshan , the Yangtze River Delta economic area of China. 83, 351–364 (2007).
8.
Xie, Y., Mei, Y., Guangjin, T. & Xuerong, X. Socio-economic driving forces of arable land
conversion: A case study of Wuxian City, China. Global Environmental Change15, 238–252 (2005).
9.
Geist, H. J. & Lambin, E. F. Proximate Causes and Underlying Driving Forces of Tropical
Deforestation. BioScience52, 143 (2002).
10.
York, R., Rosa, E. A. & Dietz, T. STIRPAT, IPAT and ImPACT: analytic tools for unpacking the driving
forces of environmental impacts. Ecological Economics46, 351–365 (2003).
11.
Rosa, E. a, York, R. & Dietz, T. Tracking the anthropogenic drivers of ecological impacts. Ambio33,
509–12 (2004).
12.
Hooi Hooi, L. & Russell, S. CO2 Emissions, Electricity Consumption and Output in ASEAN,
Development Research Unit, Discussion Paper DEVDP 09‐13, Monash University, Business and
Economics. (2009).
13.
Turner, B. L., Lambin, E. F. & Reenberg., A. The emergence of land change science for global
environmental change and sustainability. Proceedings of the National Academy of Sciences.
Proceedings of the National Academy of Sciences105, 20690–20695 (2007).
14.
Falkenmark, M. Towards integrated catchment management: opening the paradigm locks
between hydrology, ecology and policy‐making. International Journal of Water Resources
Development20, 275–281 (2004).
21
15.
Campbell, D. J., Lusch, D. P., Smucker, T. A. & Wangui, E. E. Multiple Methods in the Study of
Driving Forces of Land Use and Land Cover Change : A Case Study of SE Kajiado District , Kenya.
33, 763–794 (2005).
16.
Perman, R., Ma, Y., Mcgilvray, J. & Common, M. Natural Resource and Environmental Economics.
(2005).
17.
Dietz, T., Va, F. & Rosa, E. A. Rethinking the Environmental Impacts of Population , Affluence and
Technology ’. (1994).
18.
Howarth, R. B. & Farber, S. Accounting for the value of ecosystem services. Ecological
Economics41, 421–429 (2002).
19.
Ru, X., Chen, S. & Dong, H. A Study on Relationship between CO2 Emissions and Economic
Development in China Based on Dematerialization Theory. Energy and Environment Research2,
37–44 (2012).
20.
York, R., Rosa, E. A. & Dietz, T. The Ecological Footprint Intensity of National Economies. Journal
of Industrial Ecology8, 139–154 (2004).
21.
Shi, A. Shi A. Population growth and global carbon dioxide
emissions.http://www.iussp.org/Brazil2001/s00/S09_04_Shi.pdf. (2001).
22.
Rosa, E. A. & Dietz, T. Human drivers of national greenhouse-gas emissions. Nature climate
change2, (2012).
23.
IPCC. Emissions Scenarios. (2000).
24.
Lambin, E. F. et al. The causes of land-use and land-cover change: moving beyond the myths.
Global Environmental Change11, 261–269 (2001).
25.
Ehrlich, P. R. & Holdren, J. P. Impact of Population Growth Science. Science171, 1212–1217
(1971).
26.
Chertow, M. R. The IPAT Equation and Its Variants. Journal of Industrial Ecology 13–29 (2000).
27.
Waggoner, P. E. & Ausubel, J. H. A framework for sustainability science: A renovated IPAT
identity. Proceedings of the National Academy of Sciences99, 7860–7865 (2002).
28.
Schulze, P. C. I=PBAT. Ecological Economics40, (2002).
29.
Diesendorf, M. I=PAT or I=PBAT? Ecological Economics42, 3 (2002).
30.
Johansen, S. Statistical Analysis of Cointegrating Vectors. Journal of Economic Dynamics and
Control12, 231–254 (1988).
22
31.
Zivot, E. & Andrews, W. further Evidence on the Great Crash, the Oil-Price Shock, and the Unit root
Hypothesis, , Vol. 10. Journal of Business and economics10, (1992).
32.
Clemete, J., Montanes, A. & Reyes, M. Testing for a unit root in Variables with a double change in
the Mean. Economic Letters59, (1998).
33.
Chen, H. C. & Patel, V. C. Near-wall turbulence models for complex flows including separation.
AIAA Journal26, 641–648 (1988).
34.
DeJong, D. N., Nankervis, J. C., Savin, N. E. & Whiteman, C. H. The power problems of unit root
tests in time series with autoregressive errors.53, . Journal of Econometrics53, 323–343 (1992).
35.
Hamilton. Time Series Analysis, Princeton University Press. (1994).
36.
IPCC. IPCC Fifth Assessment Synthesis Report. (2014).
37.
FDRE. Ethiopia’s Green Economy Strategy,Federal Democratic Republic of Ethiopia. (2014).
38.
Brant Liddle. What Are the Carbon Emissions Elasticities for Income and Population? New
Evidence From Heterogeneous Panel Estimates Robust to Stationarity and Cross-Sectional
Dependence, Asia Pacific Energy Research Centre October 15, 2012, USAEE Working Paper No.
12-1. (2012).
23