Available online at www.sciencedirect.

com

ScienceDirect
Available online at www.sciencedirect.com
Procedia Computer Science00 (2017) 000–000
www.elsevier.com/locate/procedia
ScienceDirect
Procedia Computer Science 120 (2017) 671–675

9thInternational
9th InternationalConference
Conferenceon
onTheory
Theoryand
andApplication
ApplicationofofSoft
SoftComputing,
Computing,Computing
Computingwith
with
Words and Perception, 22-23 August 2017, Budapest, Hungary
Words and Perception, ICSCCW 2017, 24-25 August 2017, Budapest, Hungary

The relationship between energy consumption and economic

growth: Evidence from non-Granger causality test
Faisal Faisal*a, Turgut Tursoys, Ozlem Ercantanb
a
Department of Banking and Finance, Near East University, Nicosia 99138, North Cyprus, Mersin 10 Turkey.
b
Department
Department of Medical Biology and Genetics, Near East University, Nicosia 99138, North Cyprus, Mersin 10 Turkey

Abstract

This study examines the association between GDP and EC for Belgium using the sample period from 1960-2012. The study
applied ARDL followed by a T-Y approach to identify the causality. The study validated the existence of long-run relationship
between EC and GDP. Furthermore, GDP positively affects EC in long-run and short-run as well. The energy consumption
congregates to the long-run by 17% convergence rate that confirms the system stability.
The T-Y approach is used to investigate the direction of causality between EC and GDP. The findings confirmed a uni-
directional relationship from GDP to EC proving the validity of conservation hypothesis in Belgium. The policies which are
based on energy conservation, if implemented properly by the Government of Belgium, will have no adverse impact on economic
growth.
© 2018 The Authors. Published by Elsevier B.V.
© 2017 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of
Peer-review underComputing
Soft Computing, the responsibility of the
with Words Organizing
and PerceptionCommittee of ICSCCW 2017.
Keywords: Energy; GDP; bounds test; Toda-Yamamoto approach.

1. Introduction

The connection amongst GDP and EC has been widely investigated in the past three decades and one of the hotly
debated issues that have never been made a unanimous consensus by the researchers. The results may vary for
different data sets even for the same countries with the use of different econometric methodology. The literature

* Corresponding author. Tel.: +9-039-267-51000.

E-mail address:faisal.faisal@neu.edu.tr

1877-0509© 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the Organizing Committee of ICSCCW 2017.

1877-0509 © 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of Soft
Computing, Computing with Words and Perception.
10.1016/j.procs.2017.11.294
 672 Faisal Faisal et al. / Procedia Computer Science 120 (2017) 671–675
2 Faisal et al./ Procedia Computer Science00 (2017) 000–000

suggested that even for a particular country the same data sets, the causality results may vary for GDP and energy
consumption. For example, Thoma (2004) Kraft and Kraft (1978) both utilise the data set for USA 1973-2000 and
1947-1974. The estimated results of both the studies indicate contradictory results to each other. Both have an
opposing view of the causal relationship between energy consumption and GDP. Likewise, the same GDP and
energy consumption estimations confirmed by empirical study for turkey. For instance, Soytas et al. (2003), used
data set from 1970-2006. The empirical results of Jobert and Karanfil (2007) study revealed unidirectional causality
from GDP to EC, however the opposing results were confirmed by Soytas et al. (2003) and bidirectional causality
was observed by Faisal et al. (2016) and unidirectional causality from GDP to EC was confirmed by Faisal et
al.(2017) for Pakistan. This means that the differences and findings in the causality vary with the utilisation of
different econometric methods, variations among the estimated variables, the time span, characteristics of the
particular country and the sample selected.
The causality evidence among energy consumption and income for any particular country or countries in any
direction have significant policy have a significant impact upon the policy which is implemented in the best interest
of the country/countries. For instance, if the economy is characterised by the unidirectional relationship that draws
from EC to GDP, that indicates that the economy is more dependent on energy and without energy, the growth of
country will not be possible. That indicates that if the amount of energy is reduced below then the level of demand
employing that this reduction of energy will negatively affect the level of GDP. In this situation, energy conservation
is not a good and positive choice for a healthy economy and may harm the GDP (Masih and Masih, 1998). In
contrast, if the unidirectional relationship has been found from GDP to EC, this suggests that the country is having
less reliance on EC and would have less adverse effect on income by properly implementing the energy conservation
policies. Finally, if the estimations confirm no causality among GDP and energy, in either direction then this
confirms the neutrality hypothesis. This implies that both EC and real income doesn’t affect each other. The study of
(Faisal et al.2016) confirms that energy conservation policies, in this case, may be perused without adversely have
an impact on income.
The above mention causality hypothesis has motivated the researchers to investigate the direction and its impact
on each other that motivated many investigators to predict the causal association among GDP and EC. From the
above discussion, the conclusion can be derived from the mentioned studies for causality between EC and GDP in
such a way that the past studies in the literature focus on co-integration and causality relationship between GDP and
EC can be useful in providing policies recommendation. The purpose of present study is to analyse and evaluate the
relationship between energy consumption and economic growth in Belgium from 1960-2012 via using ARDL
bounds testing approach of cointegration as suggested by Pesaran et al.(2001). This study further focuses on the
robustness and usefulness of a T-Y approach that can be applied to any series regardless of their integration order.
The rest of the paper is organised as follows. Section 2 explains the model specification and the data. Section 3
describes the empirical findings from the study. Section 4 concludes the paper .

2. Model specification and data

According to the empirical literature on energy, we use energy consumption and GDP to specify a long-run
relationship in the above mention variables. The long run relationship can be expressed in the form of the
econometric model as follows.
ECt  f ( 0  GDPt   t ) (1)
Where Ect represents the energy consumption (in Kg of oil equivalent per capita), and GDPt represents real GDP
per capita (constant 2000 US$) and represents the error correction term. The data has been collected from the
World Bank 2015. The estimated variables are converted into their natural logarithm to decrease the effect of
heteroscedasticity (Tursoy and Faisal, 2016).
LECt  f (  0  LGDPt   t ) (2)
The expected sign for β is positive. The article will follow the ARDL methodology (Faisal et al., 2016) for more
details to investigate the short and long-run relationship.
 Faisal Faisal et al. / Procedia Computer Science 120 (2017) 671–675 673
Faisal et al./ Procedia Computer Science00 (2017) 000–000 3

2.1. The Toda-Yamamoto approach to Granger causality test

This study utilised Toda and Yamamoto (1995) approach to analyse the causality that shows flexibility and
choosing the variables different order of integration as normally followed and Granger causality test by ignoring
those series which are non-stationary or having co-integration when testing for causality. The T-Y approach for
causality is applied on the estimated variables at levels. The approach can be utilised by augmenting the extra lag
order under the VAR system, by the dmax that represents the higher order of integration that has been determined
using the unit root test.

3. Results and Discussions

3.1. Empirical Evidence

The first step in empirical estimation is to determine the order of integration (Tursoy and Faisal, 2017). Table 1
and Table 2 give the result of unit root test. ADF and PP tests were applied that are most common in the
Table 1. Augmented Dickey-Fuller test
Model ADF ADF
Belgium (1960-2012) Level First Difference
CO CO & TR CO CO & TR
LEC -2.5427 (3) -1.3235 (3) -5.4594***(0) -6.2798***(0)

LGDP -2.4582 (4) -1.1201 (4) -4.9951***(0) -6.7064***(0)

Note: (i) *** represents significance at 1%.

literature to determine the integration order. The results of the ADF suggests the series is integrated of order one.
The PP also supports the results of ADF. The maximum of integration of series is 1. The ARDL model, in this case,
can be suitable since it can be applied regardless of the order of cointegration and small sample.

Table 2. Philips-Perron (PP) Unit Root test results for Stationarity of variables
Model PP PP
Belgium (1960-2012) Level First Difference
CO CO & TR CO CO & TR

LEC -3.4867 (0) -1.7572 (0) -5.4594*** (0) -6.2798***(0)

LGDP -4.9754***(0) -1.2129 (0) -4.9951*** (0) -6.7064*** (0)

Note: (i) *** represents significance at 1%.

The next step is to identify the cointegration among the estimated model. The results of the F-statistics bounds
test has been shown in Table 3.
Table 3. Results of the Bound test of Co-integration.
Model (LnEC/LnGDP)
AIC lag length (1,1)
F-Statistics (Bound Test) 3.6654***
NOTES: The ARDL model is estimated by using restricted intercept and no trend;*** represents 10 % significance level.

The result showed that the computed F-statistic is above I(1) critical values that confirm the cointegration by
10% among the variables. All the estimated variables move in a long-run together in a stochastic process.
Once the cointegration is identified, then the long-run and short-run relationships are estimated under the ARDL
framework simultaneously. Table 4 shows the results of the ARDL long-run and short-run results. The long-run
674 Faisal Faisal et al. / Procedia Computer Science 120 (2017) 671–675
4 Faisal et al./ Procedia Computer Science00 (2017) 000–000

results showed a positive impact of energy consumption on economic growth. This means that if economic growth
increase by 1%, it may cause the energy consumption to rise by 0.52% which is in accordance with the expectation.
Table 4. ARDL Long run and short run results
LnECt ARDL(1,1)
Long-run coefficient
Variable Coefficient Standard Error t-Statistics
Constant 2.9916 1.0451 2.8623*
LnGDP 0.5288 0.1023 5.1681*
R2 0.973 Adjusted R2 0.972
F-Statistics 595.4583* DW 1.83
SSR 0.0573 S.E Regr. 0.0345
Short-run coefficients with ECM
Variable Coefficient Standard Error t-Statistics
Constant 0.5133 0.2246 4.1902*
D(lnGDP) 1.2258 0.1881 6.5139*
ECMt-1 -0.1715 0.0506 -3.3844*
R2 0.41 Adjusted R2 0.37
F-Statistics 11.1685* DW 1.83
Sum Squared resid 0.0573 S.E of regression. 0.0345
Note: * represents significance at 1% level.
The same results are also followed in the short-run as well. The ECT coefficient is negative with 10% statistical
significance. The coefficient of ECT is 0.1715. This suggests the convergence of the long-run dynamics by 17.15%.
The convergence rate, in this case, is low, so it would take the time to converge the dynamics back to the
equilibrium position. However, this convergence implies the stability of the system which is important as expected.

Table 5. Diagnostic Tests for the ARDL model

Diagnostic Test 2 2 2 2 Ramsey Reset Test
 serial cor  Hett  Nor  Auto reg

F-statistics 0.3019 3.0611 0.6700 1.1937 2.9370

(p-values) (0.8599) (0.3823) (0.7153) (0.2746) (0.0932)
NOTE: The value in the parenthesis represents the corresponding P-value for each diagnostic test.

The reliability and validity of our estimations are confirmed by the diagnostic test. As the diagnostic tests suggest
that the residuals are white noise and free from serial correlation problem. The stability of the estimated model is
confirmed by Ramsey rest test. The CUSUM and CUSUM squared residuals plots are stable. The plots of CUSUM
and CUSUM squared residuals can be obtained from the authors upon request.
Table.6 Long run non-causality test due to Toda-Yamamoto (1995) for dmax=1
Non-Granger causality test Significance of the -value of the MWALD statistic
Direction of Causality
From to Lag 1 Lag 2 Lag 3 Lag 4 Inferences

LEC LGDP 2.2107 1.3003 1.1729 1.6368 LGDP ≠› LEC

(0.1439) (0.2829) (0.3321) (0.1856)
LGDP LEC a
8.6565* 5.7417* 3.5994* 3.4224* LEC => LGDP
(0.0033) (0.0062) (0.0215) (0.0177)
Note: GDP, EC represents economic growth and Energy consumption. * denotes the significance at 5 and 10% level respectively. a represents the
optimum lag selected on the basis of lag criteria that includes the length of the VAR (k + dmax).
Finally, the results of causality in this article has been documented by using Toda-Yamamoto approach. As
earlier documented by many studies, the confirmation of long-run relationship further implies the existence of
causality among the variables. The study applies the TYA approach determine the causality among the estimated
variables. As earlier stated that Toda-Yamamoto approach can be applied regardless of the integration order.
However, provided that the maximum order of integration can be determined. As the maximum order of integration,
 Faisal Faisal et al. / Procedia Computer Science 120 (2017) 671–675 675
Faisal et al./ Procedia Computer Science00 (2017) 000–000 5

in this case, is I (1) as suggested by both ADF and PP. The results of the Toda-Yamamoto approach have been
shown in Table 6. The results showed that economic growth is causing energy consumption in this case for a period
1960-2012 at lag 1. However, traditional Granger causality test is sensitive to lag length. For this purpose, robust
checks were applied till lag 4. By increasing lag 1 to lag 4, it suggests that economic growth is causing energy
consumption and not vice versa. This indicates that results are robust from lag 1 to lag 4. This suggests the validity
of conservation hypothesis in Belgium in which economic growth stimulates the increase consumption of energy
that supported by uni-directional causality from GDP to energy consumption. This supports the fact that economic
growth stimulates the development of energy sector in an economy that is characterised by less dependence on
energy. The policies which are based on energy conservation, if implemented properly by the Government of
Belgium, will have no adverse impact on economic growth. However, the results of our studies are in contrast with
the studies conducted by Faisal et al. (2016) whom results supported the validity of feedback hypothesis.

4. Conclusion

This study examines the relationship between EC and GDP for Belgium covering the period from 1960-2012.
The study applied ARDL followed by TY approach to identify the causality. The study identified the existence of
cointegration among EC and GDP. Furthermore, the economic growth is having a positive impact on energy
consumption both in long-run and short-run. The energy consumption converges back to the long-run equilibrium
position by 17% convergence rate that identifies the system stability.
The findings of TYA to causality indicated a uni-directional relationship from GDP to EC. The robustness of the
results was estimated by extending it to different lags, and the results remain the same proving the validity of
conservation hypothesis in Belgium. The policies which are based on energy conservation, if implemented properly
by the Government of Belgium, will have no adverse impact on economic growth.

References

Brown, R. L., Durbin, J., & Evans, J. M. (1975). Techniques for testing the constancy of regression relationships over time. Journal of the Royal
Statistical Society. Series B (Methodological), 149-192.
D. Dickey and W. A. Fuller,(1981) “Likelihood ratio statistics for autoregressive time series with a unit root,” Econometrica, 49, 1057-72.
FAISAL, F., TÜRSOY, T., & REŞATOĞLU, N. G. (2017). Is there any Causality between Financial Development, Energy Consumption and
Economic Growth in Pakistan? Evidence from ARDL Bounds Testing Approach and Vector Error Correction Model. International Journal of
Ecological Economics and Statistics™, 38(2), 33-48.
Faisal, Tursoy, T., & Resatoglu, N. G. (2016). Energy Consumption, Electricity, and GDP Causality; The Case of Russia, 1990-2011. Procedia
Economics and Finance, 39, 653-659.
Jobert, T., Karanfil, F., (2007). Sectoral energy consumption by source and economic growth in Turkey. Energy Policy, 35, 5447-5456.
Kraft, J., Kraft, A., (1978). On the relationship between energy and GNP. Journal of Energy Development 3, 401–403.
Masih, A.M.M., Masih, R., (1998). A multivariate cointegrated modelling approach in testing temporal causality between energy consumption,
real income and prices with an application to two Asian LDCs. Applied Economics 30 (10), 1287–1298.
Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied
Econometrics, 16, 289–326.
Phillips, P.C.B., Perron, P., (1988). Testing for a unit root in time series regression. Biometrika 75, 335–346.
Soytas, U., Sari, R., (2003). Energy consumption and GDP: causality relationship in G-7 countries and emerging markets. Energy Econ. 25 (1),
33–37.
Thoma, M., (2004). Electrical energy usage over the business cycle. Energy Economics 26, 463–485.
Toda, H.Y., Yamamoto, Y., (1995). Statistical inference in vector autoregressions with possibly integrated process. J. Econ. 66, 225–250.
Tursoy, T., & Faisal, F. (2016). Causality between stock price and GDP in Turkey: An ARDL Bounds Testing Approach. Romanian Statistical
Review, 64(4), 3-19.
Tursoy, T., & Faisal, F. (2017). Re-testing for financial integration of the Turkish Stock Market and the US Stock Market: An Evidence from co-
integration and error correction models, 65(2), 43-55.