Munich Personal RePEc Archive The Environmental cost of Skiing in the Desert? Evidence from Cointegration with unknown Structural breaks in UAE Shahbaz, Muhammad and Sbia, Rashid and Hamdi, Helmi COMSATS Institute of Information Technology, Lahore, Pakistan, Free University of Brussels, Central Bank of Bahrain 22 June 2013 Online at https://mpra.ub.uni-muenchen.de/48007/ MPRA Paper No. 48007, posted 05 Jul 2013 04:19 UTC
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Munich Personal RePEc Archive
The Environmental cost of Skiing in the
Desert? Evidence from Cointegration
with unknown Structural breaks in UAE
Shahbaz, Muhammad and Sbia, Rashid and Hamdi, Helmi
COMSATS Institute of Information Technology, Lahore, Pakistan,
Free University of Brussels, Central Bank of Bahrain
22 June 2013
Online at https://mpra.ub.uni-muenchen.de/48007/
MPRA Paper No. 48007, posted 05 Jul 2013 04:19 UTC
1
The Environmental cost of Skiing in the Desert?
Evidence from Cointegration with unknown Structural breaks in UAE
Granger causality, dynamic ordinary least squares, non-linear cointegration and error correction model. CO2 (SO2) emissions per capita, energy consumption
per capita, electricity production, energy prices, real GDP per capita, trade openness, urbanization, gross fixed capital formation, labor force and coal
consumption is indicated by CO2 (SO2), E, EP, ENP, RGDP, TR, URB, GRFC, LB and CO.
The Table-1 presents the summary of all time series studies of single countries. We found that there is no even single country study
while investigating the relationship between economic growth and CO2 emissions using the framework of environmental Kuznets
curve in case of UAE. This is a humble effort to fill the gap in exiting literature in the case of United Arab Emirates. The present study
opens up new insights for policy making authorities to design comprehensive economic, energy and environmental policy to sustain
long run economic growth while improving environmental quality.
16
III.I The Data and Empirical Modelling
The data on CO2 emissions (metric tons), real GDP, energy consumption (kt of oil equivalent)
and urban population has collected from world development indicators (CD-ROM, 2012). We
have combed international financial statistics to obtain data on exports and converted it into real
terms by dividing exports series on GDP deflator. The population series is used to convert all
series into per capita. The study covers time period of 1975-20113. Following Soytas et al.
(2007); Jalil and Mamud, (2009); Halicioglu, (2009); Iwata et al. (2010); Esteve and Tamarit,
(2012a, b); Shahbaz et al. (2013), the general form of our empirical equation is modeled as
following:
),,,,( 2
tttttt XUEYYfC (1)
We have converted all the series into logarithm to obtain reliable and efficient empirical
evidence. Shahbaz et al. (2012) pointed out that log-linear specification reduces sharpness in the
time series data and provides better results controllable variance as compared to simple
specification. The log-linear specification of our empirical equation is modeled as following:
itXtUtEtYtYt XUEYYC lnlnlnlnlnln 2
1 2 (2)
where ttttt UEYYC ln,ln,ln,ln,ln 2and tXln is natural log of CO2 emissions (metric
tons) per capita, natural log of real income per capita, natural log of squared of real income per
3We have used Eviews software to convert annual frequency data into quarter frequency using quadratic-match-sum
method.
17
capita, energy consumption (kt of oil equivalent) per capita, urbanization per capita and real
exports per capita. i is error term with constant variance and zero mean and having normal
distribution. We expect inverted U-shaped i.e. EKC hypothesis between economic growth and
CO2 emissions if 0Y and 02 Y
otherwise U-shaped relationship exist. The 0E
implies that efficient use of energy lowers CO2 emissions otherwise energy consumption
degrades environmental quality if 0E . We can expect positive or negative impact of
urbanization on CO2 emissions. If urbanization is planned and urban population has easy access
to energy efficient technology such as consumer durables for consumers and advanced
technology for producers then 0U otherwise 0E . The impact of exports on CO2
emissions depends upon the technology to be implemented in an economy if an economy uses
energy efficient technology then 0X otherwise an increase in exports will raise CO2
emissions i.e. 0x .
III.II Zivot-Andrews Unit Root Test
The usual first step in empirical analysis is to test the stationarity properties of the variables.
Traditional unit root tests are ADF by Dickey and Fuller (1979), P-P by Philips and Perron
(1988), KPSS by Kwiatkowski et al. (1992), DF-GLS by Elliott et al. (1996) and Ng-Perron by
Ng-Perron (2001). However, as pointed by Baum, (2004), empirical evidence on order of
integration of the variable by ADF, P-P and DF-GLS unit root tests are not reliable in the
presence of structural break in the series. In fact, unit root tests may be biased and inappropriate
in absence of information about structural break occurred in series.
18
To overcome this problem, Zivot-Andrews (1992) suggested three models to test the stationarity
properties of the variables in the presence of structural break point in the series. (i) First model
permits a one-time change in variables at level form, (ii) second model allows a one-time change
in the slope of the trend component i.e. function and (iii) last model has one-time change both in
intercept and trend function of the variables to be used in the analysis. Zivot-Andrews, (1992)
adopted three models to check the hypothesis of one-time structural break in the series as
follows:
k
j
tjtjtttxdcDUbtaxax
1
1 (3)
k
j
tjtjtttxdbDTctbxbx
1
1 (4)
k
j
tjtjttttxddDTdDUctcxcx
1
1 (5)
where tDU represents the dummy variables displaying mean shift occurred at each point with
time break while trend shift variables is presented by tDT 4. So,
TBtif
TBtifDU t
...0
...1and
TBtif
TBtifTBtDU t
...0
...
The null hypothesis of unit root break date is 0c which indicates that series is not stationary
with a drift not having information about structural break point while 0c hypothesis implies
that the variable is found to be trend-stationary with one unknown time break. Zivot-Andrews
unit root test fixes all points as potential for possible time break and does estimation through
4We used model-4 for empirical estimations following Sen, (2003)
19
regression for all possible break points successively. After that, this unit root test selects that
time break which decreases one-sided t-statistic to test 1)1(ˆ cc . Zivot-Andrews indicate
that in the presence of end-points, asymptotic distribution of the statistics is diverged to infinity
point. It is compulsory to choose a region where end-points of sample period are excluded. To do
so, we followed Zivot-Andrews suggestions by choosing the trimming regions i.e. (0.15T,
0.85T).
III.II The ARDL Bounds Testing
We employ the autoregressive distributed lag (ARDL) bounds testing approach to cointegration
developed by Pesaran et al. (2001) to explore the existence of long run relationship between
economic growth, electricity consumption, urbanization, exports and CO2 emissions in the
presence of structural break. This approach has multiple econometric advantages. The bounds
testing approach is applicable irrespective of whether variables are I(0) or I(1). Moreover, a
dynamic unrestricted error correction model (UECM) can be derived from the ARDL bounds
testing through a simple linear transformation. The UECM integrates the short run dynamics
with the long run equilibrium without losing any long run information. The UECM is expressed
as follows:
tD
t
m
mtm
s
l
ltl
r
k
ktk
q
j
jtj
p
i
ititUtXtEtYtCTt
DUXEY
CUXEYCTC
1
0000
1
111111
lnlnlnln
ln lnlnlnlnlnln
(6)
20
tD
t
m
mtm
s
l
ltl
r
k
ktk
q
j
jtj
p
i
ititUtXtEtYtCTt
DUXEC
YUXEYCTY
2
0000
1
111111
lnlnlnln
lnlnlnlnlnlnln
(7)
tD
t
m
mtm
s
l
ltl
r
k
ktk
q
j
jtj
p
i
ititUtXtEtYtCTt
DUXCY
EUXEYCTE
3
0000
1
111111
lnlnlnln
lnlnlnlnlnlnln
(8)
tD
t
m
mtm
s
l
ltl
r
k
ktk
q
j
jtj
p
i
ititUtXtEtYtCTt
DUEYC
XUXEYCTX
4
0000
1
111111
lnlnlnln
lnlnlnlnlnlnln
(9)
tD
t
m
mtm
s
l
ltl
r
k
ktk
q
j
jtj
p
i
ititUtXtEtYtCTt
DXEYC
UUXEYCTU
5
0000
1
111111
lnlnlnln
lnlnlnlnlnlnln
(10)
where Δ is the first difference operator, D is dummy for structural break point based on Z-A test
and t is error term assumed to be independently and identically distributed. The optimal lag
structure of the first differenced regression is selected by the Akaike information criteria (AIC).
Pesaran et al. (2001) suggest F-test for joint significance of the coefficients of the lagged level of
variables. For example, the null hypothesis of no long run relationship between the variables is
0:0 UXEYCH against the alternative hypothesis of cointegration
0: UXEYCaH . Pesaran et al. (2001) computed two set of critical value (lower
and upper critical bounds) for a given significance level. Lower critical bound is applied if the
regressors are I(0) and the upper critical bound is used for I(1). If the F-statistic exceeds the
upper critical value, we conclude in favor of a long run relationship. If the F-statistic falls below
21
the lower critical bound, we cannot reject the null hypothesis of no cointegration. However, if the
F-statistic lies between the lower and upper critical bounds, inference would be inconclusive.
When the order of integration of all the series is known to be I(1) then decision is made based on
the upper critical bound. Similarly, if all the series are I(0), then the decision is made based on
the lower critical bound. To check the robustness of the ARDL model, we apply diagnostic tests.
The diagnostics tests are checking for normality of error term, serial correlation, autoregressive
conditional heteroskedasticity, white heteroskedasticity and the functional form of empirical
model.
III.III The VECM Granger Causality
After examining the long run relationship between the variables, we use the Granger causality
test to determine the causality between the variables. If there is cointegration between the series
then the vector error correction method (VECM) can be developed as follows:
t
t
t
t
t
t
t
t
t
t
t
mmmmm
mmmmm
mmmmm
mmmmm
mmmmm
t
t
t
t
t
t
t
t
t
t
ECM
U
X
E
Y
C
BBBBB
BBBBB
BBBBB
BBBBB
BBBBB
U
X
E
Y
C
BBBBB
BBBBB
BBBBB
BBBBB
BBBBB
b
b
b
b
U
X
E
Y
C
5
4
3
2
1
1
5
4
3
3
1
1
1
1
1
1
,55,54,53,52,51
,45,44,43,42,41
,35,34,33,32,31
,25,24,23,22,21
,15,14,13,12,11
1
1
1
1
1
1,451,441,431,421,41
1,451,441,431,421,41
1,351,341,331,321,31
1,251,241,231,221,21
1,151,141,131,121,11
4
3
2
1
)(
ln
ln
ln
ln
ln
...
ln
ln
ln
ln
ln
ln
ln
ln
ln
ln
(11)
22
where difference operator is and 1tECM is the lagged error correction term, generated from
the long run association. The long run causality is found by significance of coefficient of lagged
error correction term using t-test statistic. The existence of a significant relationship in first
differences of the variables provides evidence on the direction of short run causality. The joint
2 statistic for the first differenced lagged independent variables is used to test the direction of
short-run causality between the variables. For example, iiB 0,12 shows that economic growth
Granger causes CO2 emissions and economic growth is Granger of cause of CO2 emissions if
iiB 0,11 .
IV. Results Interpretations
We find descriptive statistics and pair-wise correlations between the variables (see Table-2). The
results reveal that CO2 emissions, economic growth, electricity consumption, urbanization and
exports are found to be normally distributed. The statistics by Jarque-Bera indicate that all the
series are having zero mean and finite covariance. The findings of pair-wise correlation show
that economic growth and urbanization are positively correlated with CO2 emissions while
negative correlation is found from electricity consumption and exports to CO2 emissions. The
positive correlation exists from electricity consumption and exports to economic growth.
Urbanization is inversely correlated with economic growth. Exports and urbanization are
positively linked with electricity consumption and same inference is drawn between urbanization
and exports.
23
Table-2: Descriptive Statistics and Correlation Matrix
Variable tCln tYln tEln tUln tXln
Mean 3.4406 12.3273 9.0609 4.3876 8.3744
Median 3.4293 12.2620 9.1399 4.3826 8.3115
Maximum 4.1511 12.8449 9.5342 4.4355 9.2260
Minimum 2.7692 11.5962 7.7773 4.3607 7.7646
Std. Dev. 0.2863 0.2919 0.4321 0.0201 0.3835
Skewness 0.3600 -0.0706 -1.3198 0.9095 0.5170
Kurtosis 4.1451 2.9880 4.3433 2.8811 2.4247
Jarque-Bera 2.8212 0.0309 3.5247 5.1229 2.1590
Probability 0.2439 0.9846 0.1100 0.0771 0.3397
tCln 1.0000
tYln 0.6726 1.0000
tEln -0.7574 0.7267 1.0000
tUln 0.2205 -0.4709 0.3299 1.0000
tXln -0.3975 0.4582 0.4584 0.0812 1.0000
The assumption of the ARDL bounds testing is that the series should be integrated at I(0) or I(1)
or I(0) / I(1). This implies that the none of variables is integrated at I(2). To resolve this issue, we
have applied traditional unit root tests such as ADF, PP and DF-GLS5. Our empirical exercise
finds that CO2 emissions ( tCln ), electricity consumption ( tEln ), economic growth ( tYln ),
5Results are available upon request from authors
24
exports ( tXln ) and urbanization ( tUln ) are not found to be stationary at level with constant and
trend. This shows that the variables are integrated at I(1). The main issue with these unit tests is
that these tests do not seem to consider information about unknown structural beaks in the series.
This implies that traditional unit root tests provide ambiguous results regarding integrating
properties of the variables. The appropriate information about structural break would help policy
makers in designing inclusive energy, economic, urban and trade policy to sustain environmental
quality in long run. We have applied Zivot-Andrews unit root test which accommodates single
unknown structural break in the variables.
Table-3: Zivot-Andrews Structural Break Unit Root Test
Variable At Level At 1st Difference
T-statistic Time Break T-statistic Time Break
tCln -3.285 (2) 1999Q2 -9.170 (4)* 1997Q3
tYln -2.551 (5) 2000Q2 -7.632 (2)* 1988Q2
tEln -4.503 (3) 1997Q4 -8.277 (0)* 1992Q2
tUln -4.962 (5) 1995Q3 -7.943 (0)* 1995Q2
tXln -4.384 (4) 1982Q1 -8.359 (0)* 1980Q3
Note: * represent significant at 1% level of significance. The critical
value at1% is -5.57 and at 5% is -5.08. Lag order is shown in
parenthesis.
25
The results are detailed in Table-3. We find, while applying Zivot-Andrews, (1992) test with
single unknown break, that economic growth, electricity consumption, exports, urbanization and
CO2 emissions have unit root at level with intercept and trend. The structural breaks are found in
economic growth, electricity consumption, exports, urbanization and CO2 emissions in 2000Q2,
1997Q4, 1982Q1, 1995Q3 and 1999Q2 respectively. The variables are found to be stationary at
1st difference. This implies that series have same level of integration. The robustness of results is
validated by applying Zivot-Andrews, (1992) with single unknown structural break. Our findings
indicate that variables are integrated at I(1). The unique integrating order of the variables lends a
support to test the existence of cointegration between the variables. In doing so, we apply the
ARDL bounds testing approach in the presence of structural break to examine cointegration
between the variables. The results are reported in Table-4. The lag order of the variable is chosen
following Akaike information criterion (AIC) due to its superiority over Schwartz Bayesian
criterion (SBC). AIC performs relatively well in small samples but is inconsistent and does not
improve performance in large samples whilst SBC in contrast appears to perform relatively
poorly in small samples but is consistent and improves in performance with sample size
(Acquah, 2010).
The appropriate lag section is required because F-statistic variables with lag order of the
variables. The lag order of the variables is given in second column of Table-4. The results
reported in Table-4 reveal that our computed F-statistics are greater than upper critical bounds
generated by Pesaran et al. (2001) which are suitable for small data set. We find three
cointegrating vectors once CO2 emissions, electricity consumption and exports are used as
dependent actors. This validates that there is long run relationship between economic growth,
26
electricity consumption, exports, urbanization and CO2 emissions in case of UAE. The
diagnostic tests such as autoregressive conditional heteroskedasticity (ARCH), white
heteroskedasticity and functional form of the model are also applied. The results of diagnostic
tests are reported in Table-4. We find that there is no evidence of autoregressive conditional
heteroskedasticity. The results indicate that white heteroskedasticity does not exist. The
functional form of long run model is well designed confirmed by Ramsey RESET test.
Table-4: The Results of ARDL Cointegration Test
Bounds Testing to Cointegration Diagnostic tests
Estimated Models Optimal lag length Break Year F-statistics 2