J. Asafu-Adjaye r Energy Economics 22 2000 615]625 619
the significance of the A s conditional on the optimum lags.
1
Through the ECT,
i j
Ž an error correction model offers an alternative test of causality or weak exogeneity
. of the dependent variable . If, for example, l
is zero, then it can be implied that
en
the change in en does not respond to deviation in long-run equilibrium in period
t
t y 1. Also, if l
is zero and both A and A
are zero, it can be implied that
en 11
13
income and prices do not Granger-cause energy consumption. The non-significance of both the t and Wald F-statistics in the ECM will imply that the dependent
variable is weakly exogenous.
2
If the variables, y , en and p are cointegrated then it is expected that at least
t t
t
one or all of the ECT s should be significantly non-zero. Granger causality of the Ž .
Ž . dependent variables is tested as follows: 1 by a simple t-test of the l s; 2 by a
i
joint Wald F-test of the significance of the sum of the lags of each of the Ž .
explanatory variables in turn; and 3 by a joint Wald F-test of the following Ž .
Ž . Ž
. Ž .
Ž .
interactive terms: Eq. 1 } l and A , l and A
; Eq. 2 } l and A
,
y 22
y 23
en 11
Ž .
Ž . Ž
. Ž .
l and A
; and Eq. 3 } l and A , l and A
.
en 13
p 31
p 32
Annual time series data were utilised in this study. The series for India and Indonesia cover the period 1973]1995, while those for Thailand and the Philip-
pines cover the period 1971]1995. The data were obtained from World De
¨
elop -
ment Indicators WDI 1998, published by the World Bank. The choice of the
starting period was constrained by the availability of data on energy consumption. The precise definitions of the variables are as follows:
en : commercial energy use in kilograms of oil equivalent per capita.
y :
real income, defined as GDP in constant 1987 prices in local currency units. p
: prices. Since energy prices were not available, this variable was proxied by
Ž .
the consumer price index CPI , 1987 s 100.
4. Empirical results and discussion
Table 2 reports the results for both the ADF and PP test results. It can be seen that, with exception of Indonesian prices, the null hypothesis of nonstationarity
cannot be rejected at the 10 level for the levels of the variables. However, when first differences are taken, the null hypothesis of nonstationarity is rejected for
most of the variables. We have mixed results for the differenced Thailand energy and income variables. The null hypothesis of non-stationarity cannot be rejected by
the ADF test but is rejected by the PP test. It can therefore be concluded that in
Ž . most cases, income, energy and prices are integrated of order one, that is, I 1 ,
except Thailand energy and income which could be integrated of order two.
1
The lag lengths were assigned on the basis of minimising Akaike’s AIC criterion.
2
Ž .
Weak exoegeneity is a precondition for super-exogeneity see Engle et al., 1983 . The latter issue is not pursued in this study.
J. Asafu-Adjaye r Energy Economics 22 2000 615]625 620
Table 2
a
Results of unit root tests Ž
. Ž
. Countryr
Augmented Dickey]Fuller ADF Phillips]Perron PP
variable Levels
First Levels
First differences
differences India
y y
1.26 y
3.71 y
2.51 y
6.29
t
en y
1.92 y
2.81 y
1.39 y
5.83
t
p y
0.35 y
4.53 y
0.07 y
3.79
t
Indonesia y
y 0.03
y 3.65
y 0.24
y 4.74
t
en y
0.96 y
3.77 y
1.35 y
3.73
t
p y
4.49 y
3.11 y
5.35 y
2.64
t
Thailand y
y 0.61
y 2.27
y 1.23
y 2.87
t
en y
1.73 y
2.55 y
2.39 y
2.80
t
p y
2.73 y
3.15 y
3.04 y
2.61
t
Philippines y
y 1.58
y 3.15
y 1.90
y 2.32
t
en y
1.95 y
3.35 y
1.39 y
5.78
t
p y
1.16 y
4.33 y
1.30 y
3.97
t
Critical values 1
y 3.75
y 3.73
5 y
3.00 y
2.99 10
y 2.63
y 2.63
a
Note : The optimal lags for the ADF tests were selected based on optimising Akaike’s Information
Ž .
Criteria AIC , using a range of lags. Truncation lags for the PP test were determined using the highest significant lag from either the autocorrelation or partial autocorrelation function of the first differenced
series.
Given that most of the variables are integrated of the same order, the next step was to test for cointegration using Johansen’s multivariate maximum likelihood
procedure.
3
The test results are reported in Table 3, where r represents the number of cointegrating vectors. It can be seen that, for India, the null hypothesis
of no cointegration relationships is rejected against the alternative of one cointe- grating relationship at the 1 level. In the case of Indonesia, the test results
suggest the presence of two cointegrating relationships. Finally, the results for Thailand and the Philippines suggest that, in both cases, the null hypothesis of no
cointegrating relationship can be rejected in favour of the alternative of a single cointegrating relationship.
4
The existence of cointegrating relationships among income, energy and prices suggests that there must be Granger causality in at least
3
Ž .
The Johansen approach Johansen, 1988; Johansen and Juselius, 1990 has been shown to be superior to Engle and Granger’s residual-based approach. Among other things, the Johansen approach
is capable of detecting multiple cointegrating relationships.
4
A similar pattern of results was obtained using trace tests.
J. Asafu-Adjaye r Energy Economics 22 2000 615]625 621
Table 3 Ž
Results of Johansen’s maximum likelihood tests for multiple cointegrating relationships intercept, no .
trend Countryrnull
Characteristic Test
5 Critical 1 Critical
a
hypothesis roots
statistics value
value India
UUU
r s 0 r s 1
0.77 38.78
29.68 35.65
r s 1 r s 2
0.27 8.29
15.41 20.04
r s 2 r s 3
0.08 1.64
3.76 6.65
Indonesia
UUU
r s 0 r s 1
0.80 54.83
29.68 35.65
UU
r s 1 r s 2
0.62 21.29
15.41 20.04
r s 2 r s 3
0.02 0.37
3.76 6.65
Thailand
UU
r s 0 r s 1
0.75 41.78
29.68 35.65
r s 1 r s 2
0.33 9.53
15.41 20.04
r s 2 r s 3
0.01 0.18
3.76 6.65
Philippines
U
r s 0 r s 1
0.61 34.00
29.68 35.65
r s 1 r s 2
0.30 12.61
15.41 20.04
r s 2 r s 3
0.17 3.31
3.76 6.65
a
The test statistic is the l value. Significant at the
UUU
1 level, the
UU
5 level and the
U
10
ma x
level.
one direction. However, it does not indicate the direction of temporal causality between the variables. To determine the direction of causation, we must examine
the ECM results. In addition to providing an indication of the direction of causality, the ECM
enables us to distinguish between ‘short-run’ and ‘long-run’ Granger causality. In Table 4, we provide joint Wald F-statistics of the lagged explanatory variables of
the ECM. These tests give an indication of the significance of short-run causal effects. We also provide t-statistics for the coefficients of the ECT s which give an
indication of long-run causal effects. Finally, we provide joint Wald F-statistics for
Ž .
the interactive terms i.e. the ECT s and the explanatory variables which give an indication of which variables bear the burden of short-run adjustment to re-estab-
lish long-run equilibrium, given a shock to the system. Ž
. Turning first to the short-run results for India Table 4 , it can be seen that the
Ž .
F -statistic for energy in the income equation is significant at the 5 level.
However, none of the lagged explanatory variables in the other two equations Ž
. energy and price are statistically significant. These results imply that, in the short
run, there is unidirectional Granger causality running from energy consumption to income, while price has a neutral effect on both energy and income. Looking at the
t -statistics, it can be seen that the coefficient of ECT is significant in the income
equation but is not significant in either the energy or price equation. This result can be interpreted as follows. Given a deviation of income from the long-run
equilibrium relationship as defined by ECT s y y p y en , all three variables
t t
t
J. Asafu-Adjaye r Energy Economics 22 2000 615]625 622
Table 4 Temporal Granger-causality results for India, Indonesia, Thailand and the Philippines
Countryr Short-run effects
Source of causation: dependent
a
Dy Den
Dp ECT
Den ,
Dp ,
Dy ,
t t
t t
t t
variable only
ECT ECT
ECT Wald F-statistics
t -Ratio
Wald F-statistics India
UU UU
UU UU
Dy ]
8.14 1.01
y 2.85
4.09 4.63
]
t
Den 0.18
] 0.08
y 1.01
] 0.77
0.83
t
Dp 0.39
0.07 ]
y 0.41
0.11 ]
0.20
t
Indonesia
UU
Dy ]
0.81 1.68
y 2.08
2.19 2.40
]
t
Den 0.87
] 0.63
1.83 ]
1.68 1.70
t UU
UU UU
Dp 1.06
0.81 ]
y 2.83
4.00 ]
4.67
t
Thailand
UUU UUU
Dy ]
11.3 0.26
y 0.54
9.98 0.20
]
t UUU
U UUU
UU UUU
Den 9.66
] 5.22
y 2.84
] 6.55
16.9
t UUU
UUU UUU
UUU
Dp 0.40
9.02 ]
y 3.48
7.48 ]
6.60
t
Philippines
U UUU
UUU
Dy ]
2.66 21.6
y 1.16
y 1.43
14.4 ]
t U
U UUU
Den 2.98
] 0.00
y 2.16
] 2.38
6.14
t UUU
UUU
Dp 10.6
0.21 ]
y 1.43
1.05 ]
8.51
t a
ECT-error correction term in the error-correction model. Significance at the
UUU
1 level, the
UU
5 level and the
U
10 level.
interact in a dynamic fashion to restore long-run equilibrium. However, the non-significance of the F-statistics for price indicates it is exogenous in the system,
implying that energy consumption bears the burden of the short-term adjustment to long-term equilibrium. The Wald F-test results in the last three columns of
Table 4 suggest that, in the long run, both energy and price Granger-cause income.
The results for Indonesia are not much different from those of India. The standard Granger tests would have concluded that there are no causal relationships
among y , en and p . However, the coefficient of ECT in the energy equation is
t t
t
significant at the 5 level. This implies that, as in the case of India, energy and prices interact in the short-term to restore long-run equilibrium after a change in
income. None of the interaction terms are statistically significant, implying that the long-run energy and price effects are weak. The results for the other equations
suggest that, in the long term, income and prices have no effect on energy consumption. However, both energy consumption and income cause price changes.
In the case of Thailand, it can be seen from the income equation that energy Granger-causes income in both the short- and long-run. However, the energy
equation results indicate that income also Granger-causes energy and therefore there is bidirectional Granger causality. The results for the price equation suggest
that prices also Granger-cause energy consumption. For the Philippines, we again
J. Asafu-Adjaye r Energy Economics 22 2000 615]625 623
observe Granger causality running from energy and prices to income in both the short and long runs, with reverse causality running from energy to income.
Ž .
Our results are consistent with the findings of Masih and Masih 1997 , Hwang Ž
. Ž
. and Gum 1992 and Glasure and Lee 1997 who found evidence of bidirectional
causality between income and energy for South Korea and Taiwan. However, they refute the neutrality hypothesis advanced in respect of the United States for the
Ž .
energy]income relationship Erol and Yu, 1989; Yu and Jin, 1992 in three out of four cases. It is only in the case of Indonesia and India where neutrality between
energy and income is observed in the short run. However, this can be expected in the case of Indonesia since it is a net energy exporter and therefore can be
shielded from energy shocks.
The ECMs displayed reasonable goodness-of-fit based on the R
2
and F statistics Ž
. not reported here and passed most of the diagnostic tests including the Godfrey
LM test for serial correlation, the Engle test for first-order autoregressive het- w
Ž .x eroscedasticity ARCH 1 , the Bera]Jacque test for normality and the Ramsey
Ž .
RESET test for model misspecification.
5. Conclusion