Section 4 introduces the forecast error variance decomposition model and its result. Section 5 summarizes and concludes the paper.
2. Data description
The data consist of daily closing prices for the New York SP 500 index and the following 11 major Asia – Pacific equity market indices: Tokyo Nikkei 225, Hong
Kong Hang-Seng, Singapore Straits Times, Sydney All Ordinaries, Seoul Com- posite Index, Taiwan Composite Index, Kuala Lumpur Composite Index, Manila
Composite Index, Bangkok Composite Index, Jakarta Composite Index and Shang- hai B-shares index. The prices are collected for the period from July 1, 1996 to June
30, 1998.
To investigate the possible differences of cointegrational relationship among the 12 stock market indices before and during the Asian crisis, the data are divided into
two groups: 1. July 1, 1996 – June 30, 1997: the period before the crisis.
2. July 1, 1997 – June 30, 1998: the period during the crisis. All daily index prices are taken a natural logarithm ln in front. Table 1 shows
statistical characteristics in the 12 country indices. We also plot the 12 indices time series in Fig. 1 for the period before the crisis and for the period during the crisis.
In order to ensure the same number of observations, we omitted all the observa- tions from Saturday’s trading mainly in Bangkok, Taiwan and Seoul. For a
specific not common holiday in a country, we use the preceding day’s observation as a proxy for that day.
3. Unit roots test
Following the discussion by Dickey and Fuller 1979, 1981, consider the time series of x
t
has the pth-order autoregressive process: x
t
= a
+ a
1
x
t-1
+ a
2
x
t-2
+ ... + a
p
x
t − p
+ o
t
1 The representation can be transformed and added by a time-trend term.
Dx
t
= a
+ rx
t − 1
+ a
2
t +
p i = 2
b
i
Dx
t − r + 1
+ o
t
2 r =
1 −
p i = 1
a
i
n
b =
p j = i
a
j
In the above Augmented Dickey – Fuller test, the null hypothesis is H : r = 0. If
this is true, x
t
has a unit root. Dickey and Fuller 1979 found that the critical values for r = 0 depend on the form of the regression and sample size. With 100
observations, there are three different critical values for the t-statistic r = 0. For a regression without the intercept and trend term a
= a
2
= 0, use the section labeled
H .-
C .
Sheng ,
A .H
. Tu
J .
of Multi
. Fin
. Manag
.
10 2000
345 –
365
349 Table 1
Statistical characteristics of the 12 countries’ indices Number of
Standard Sample period
Variance Skewness
Kurtosis Minimum
Maximum Average
observation deviation
0.1907 SP 500
− 0.7343
Before 626.64
898.70 254
747.80 68.95
6.36 0.4177
− 1.4247
876.99 1138.49
5.84 250
During 76.52
1003.32 Before
85.40 −
0.4229 −
1.0957 17303.65
22455.49 247
19969.19 Nikkei225
1305.90 1548.46
141.44 0.7275
− 0.3895
14664.44 20575.26
246 16952.13
During 2.68
− 0.1712
− 0.9962
1974.37 2136.59
2291.53 Singapore Straits
245 Before
75.67 Times
During 43.63
− 0.0408
− 0.7486
1048.96 2007.23
248 1617.68
265.68 Before
133.33 7.46
0.3362 0.0507
2096.10 2725.90
249 Sydney All
2383.02 Ordinaries
− 0.3164
0.1107 2299.20
2881.40 99.76
3.74 2661.98
251 During
60.67 4.98
0.1653 −
1.0387 611.05
858.79 246
Seoul Composite 738.91
Before 0.3115
− 0.9322
280.00 781.70
37.45 518.36
267 During
139.33 886.94
Before 108.44
0.3535 −
1.3545 5955.50
9030.28 289
7254.29 Taiwan
Composite 0.1788
− 0.9336
7089.56 10116.84
During 277
8584.19 762.68
67.76 105.17
− 0.0563
− 0.7379
10585.86 12713.86
15196.79 Hong Kong
1156.31 251
Before Hang-Seng
During 518.05
0.5613 −
0.9126 7462.50
16673.27 246
11671.74 2458.97
3.15 0.1536
− 1.2461
1041.27 1271.57
60.41 Before
1158.53 249
Kuala Lumpur Composite
36.95 During
0.5549 −
0.4143 435.84
1084.88 246
704.77 161.38
202.14 Before
13.22 −
0.8301 0.0797
2499.00 3448.00
245 3090.23
Manila Composite
0.6193 −
0.1284 1518.00
2816.00 280.51
During 37.55
2095.45 249
H .-
C .
Sheng ,
A .H
. Tu
J .
of Multi
. Fin
. Manag
.
10 2000
345 –
365
350
Table 1 Continued Standard
Number of Sample period
Variance Skewness
Kurtosis Minimum
Maximum Average
observation deviation
853.08 203.06
48.34 0.0523
− 0.9347
464.77 1268.20
Bangkok 246
Before Composite
− 0.0160
− 0.6428
257.44 682.16
103.23 22.69
469.58 244
During Before
54.59 4.76
− 0.0988
− 1.3150
530.89 725.00
257 Jakarta Compo-
626.52 site
0.9264 0.2125
340.00 741.00
99.66 During
504.11 248
19.70 3.42
0.5438 −
0.9786 44.87
Shanghai B- 98.21
Before 244
63.90 14.79
Shares 0.4320
− 0.9584
39.44 90.36
13.15 During
2.84 60.96
244
t . Including an intercept term but not a trend term a
2
= 0 only, use the section
labeled t
m
. Finally, with both intercept and trend, use the section labeled t
t
., Table 2 lists critical values for 95 and 99 confidence intervals for t-statistic t, t
m
and t
t
. Dickey and Fuller 1981 provided three additional F-statistics f
1
, f
2
and f
3
to test joint hypothesis on the coefficients. The null hypothesis r = a
= 0 is tested
using the f
1
statistic. Including a time trend in the regression, the joint hypothesis a
= r = a
2
= 0 is tested using the f
2
, statistic and the joint hypothesis r = a
2
= 0 is
tested using the f
3
statistic. Table 3 also provides critical values for 95 and 99 confidence intervals for t-statistic f
1
, f
2
and f
3
.
Fig. 1. Time series plots for 12 country indices.
Fig. 1. Continued
Finally, it is possible to test hypotheses concerning the significance of the drift term a
and time trend a
2
. Under the null hypothesis r = 0, the test for the presence of the time trend is given by the t
bt
statistic. Thus, the statistic is the test a
2
= given that r = 0. Similarly, to test the hypothesis a
= 0 given that r = 0 use the t
at
and to test the hypothesis a =
0 given that r = 0 and a
2
= 0, use the t
am
. The critical
H .-
C .
Sheng ,
A .H
. Tu
J .
of Multi
. Fin
. Manag
.
10 2000
345 –
365
353
Table 2 Summary results of unit roots tests
a
South Korea US
Taiwan Hong Kong Malaysia
The Philippines Thailand
Indonesia China
Japan Singapore
Australia Country
I1 I1
I1 I1
I1 I1
I0 I1
I1 I1
I1 I1
Before the crisis I1
I1 I1
I1 I1
I1 I1
I1 During the crisis
I1 I1
I1 I1
a
In denotes integration of order n; I1 denotes unit root and I0 denotes stationarity. The test statistics can be provided by the corresponding author on request.
values of 95 and 99 confidence intervals for t
bt
, t
at
and t
am
are also listed in Table 3.
Unless the researcher knows the actual data-generating process, it is questionable as to whether it is appropriate to estimate. Hence, we consider the following three
alternatives of Eq. 2. Dx
t
= rx
t − 1
+
p i = 2
b
i
Dx
t − r + 1
+ o
t
2a Dx
t
= a
+ rx
t − 1
+
p i = 2
b
i
Dx
t − r + 1
+ o
t
2b Dx
t
= a
+ rx
t − 1
+ a
2
t +
p i = 2
b
i
Dx
t − r + 1
+ o
t
2c It might seem reasonable to test the hypothesis r = 0 using the most general of
the models of Eq. 2. For example, if the true process is a random walk, this regression should find that a0 = r = a
2
= 0. One problem with this line of reasoning
is that the presence of the redundant estimated parameters reduces degrees of freedom and the power of the test. In other words, there is the possibility that the
research will conclude that the process contains a unit root, where, in fact, none is present.
Campbell and Perron 1991 found four important difficulties concerning the unit root tests. These difficulties imply that the research may fail to reject the null
hypothesis of a unit root because of a mis-specification concerning the deterministic part of the regression. Too few or too many regressors may cause a failure of the
Table 3 Summary of the Dickey–Fuller Tests
Test statistic Hypothesis
Critical values for 95 and 99 confidence Model
intervals for sample size of 100
a
t
t
Dx
1
= a
+ rx
t−1
r = 0 −
3.45 and −4.04 +
a
2
t+o
t
t
at
3.11 and 3.78 a
= 0 given
r = 0 t
bt
2.79 and 3.53 a
2
= 0 given
r = 0 6.49 and 8.73
f
3
r = a
2
= a
= r = a
2
= f
2
4.88 and 6.50 r = 0
Dx
t
= a
+ rx
t−1
t
m
− 2.89 and −3.51
+ o
t
a =
0 given 2.54 and 3.22
t
am
r = 0 4.71 and 6.70
f
1
a =
r = 0 r = 0
Dx
t
= rx
t−1
+o
t
t −
1.95 and −2.60
a
Source, Enders 1995, or Dickey and Fuller 1979, 1981.
test to reject the null of a unit root
2
. To avoid the inappropriate use of a unit root test, we follow the procedure suggested by Doldado et al. 1990.
The empirical results of unit root tests are summarized in Table 2. Before the crisis, the calculated values of test statistics, with the exception of Thailand, are all
smaller than their corresponding critical values at 5 significance level. We do not reject the null hypothesis that the 11 indices not Thailand contain a unit root.
During the period of the crisis, the calculated values of test statistics of all 12 indices are smaller than their corresponding critical values at 5 significance level.
We do not reject the null hypothesis that the 12 indices all contain a unit root.
4. Cointegration and causality