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
4
.
1
. Tests for cointegration and error-correction Malliaris and Urrutia 1992 used the two-step procedure of Engle and Granger
1987 EG hereafter to test the cointegrational relationships among price move- ments on six different markets during the market crash of October 1987. EG’s
procedure has been shown to be most appropriate for systems of only two variables with one possible cointegrating vector. Recent advances in multivariate cointegra-
tion and error correction modeling provide a useful framework for analyzing equilibrium price adjustments in information-linked markets. Johansen 1988 and
Stock and Watson 1988 maximum likelihood estimators can circumvent the use of EG’s two-step estimators and estimate and test for the presence of multiple
cointegrating vectors. Moreover, these tests allow the researchers to test restricted versions of the cointegrating vectors and the speed of adjustment parameters.
Following Johansen 1988 procedure, we focus on the model
3
X
t
= A
+ A
1
X
t − 1
+ o
t
3 This can be rewritten as
DX
t
= A
+ P X
t − 1
+ o
t
P = A
1
− I
4 Both procedures by Johansen 1988 and Stock and Watson 1988 rely heavily
on the relationship between the rank of P and its characteristic roots. The key feature to note in Eq. 4 is the rank of the matrix P; the rank of P equals the
number of cointegrating vectors. Clearly, if rank P = 0, the matrix is null. Since there is no linear combination of the x
it
processes that is stationary, the variables are not cointegrated. Instead, if P is of rank n, the vector process is stationary. In
intermediate cases, if rank P = 1, there is a single cointegrating vector and
2
For details concerning the difficulties of the unit root tests refer to Campbell and Perron 1991 or Enders 1995.
3
We follow Sims 1980 procedure and find that the optimal lag length is 1. Alternatively, you may select lag length p using the multivariate generalizations of the AIC or SBC. The result is the same.
expression PX
t − 1
is the error correction factor. For other cases in which 1 B rank P B n, which are multiple cointegrating vectors.
We first examine the cointegrational relationship among nine Asian countries excluding Thailand due to its stationary properties: Japan, Singapore, South
Korea, Taiwan, Hong Kong, Malaysia, the Philippines, Indonesia and Mainland China. The results are presented in Table 4. At 95 confidence interval, l
trace
and l
max
tests both show there is one cointegrating vector during the period of the crisis, but none before the period of the crisis. Including Thailand, we then re-examine the
cointegrational relationship among ten Asian countries. Table 4 also summaries the results. It shows that there are two cointegrating vectors during the period of the
crisis, representing the key role of Thailand in the Asian financial crisis. Since Thailand’s index is I0, we ignore only examination of the cointegrational relation-
ship before the period of the crisis.
In the aftermatch of the crisis, we found that the cointegrational relationship had improved in the Asian countries as a whole. It is interesting to further
examine the cointegrational relationship in each economic or political region in Asia. In the following empirical tests, we divide the 12 country indices into two
groups:
North-East Asia group: Japan, South Korea, Taiwan, Hong Kong and China. South-East Asia group: Thailand, Malaysia, the Philippines, Singapore and
Indonesia
4
.
4
.
2
. Granger
’
s causality As found in Eun and Shim 1989, the US stock market is the most influential in
the world. Shocks in the US stock market are rapidly transmitted to other markets in clearly recognizable movements. In this section, we employ Granger’s 1969
causality methodology to see whether the lead – lag relationship between the U.S. and Asian stock markets differs before and during the period of the financial crisis.
Due to the fact that national equity markets are generally operating time zones with different opening and closing times, testing for lead – lag or causality relation-
ships between US and Asian markets presents a problem of data synchronization due to time-zone shift differences. The adjustment of the causality test is of primary
importance in the execution and interpretation of the empirical tests conducted in this study. Table 5 shows the trading hours of the national stock exchanges in
Greenwich mean time, local time, and New York time. The middle column local time of Table 5 shows that all of the 11 Asia – Australia stock exchanges in our
sample trade within or around the same time interval. Further, all of the Asia – Aus- tralia exchanges are closed when the New York Stock Exchange NYSE opens for
the day. The last column New York time of Table 5 indicates that in any given trading day the closing prices for all of the Asia – Australia exchanges in our sample
are already known by the time the NYSE closes for the day.
4
The five countries are all members of ASEAN Association of South-East Asian Nations.
H .-
C .
Sheng ,
A .H
. Tu
J .
of Multi
. Fin
. Manag
.
10 2000
345 –
365
357 Table 4
Test of cointegration among Asian countries
a
Critical value 90 Null hpothesis
Alternative hypothesis Test statistic
Critical value 95 After the crisis
Before the crisis Nine Asian countries
excluding Thailand r\
186.39 r = 0
0172.03 192.89
204.49 150.53
r\1 131.35
150.03 156.00
r50 124.24
118.53 r\2
r52 97.24
107.27 94.15
89.48 r53
68.88 r\3
73.19 r54
64.84 r\4
49.74 46.18
68.52 r55
r\5 29.62
29.75 47.21
43.95 l
max
tests r = 1
53.98 40.67
54.46 r = 0
57.12 51.42
48.33 42.76
r = 1 34.12
r = 2 45.28
42.32 r = 2
r = 3 28.35
34.08 r = 3
36.76 39.37
r = 4 23.45
22.71 30.90
r = 5 16.56
20.00 33.46
r = 4 27.07
24.73 13.35
r = 5 15.23
r = 6 Ten Asian countries
including Thailand l
trace
tests r\0
225.85 –
272.62 r = 0
233.13 192.89
186.39 202.98
r51 –
r\1 r\2
150.53 –
147.71 r52
156.00 r\3
118.53 r53
– 124.24
104.78 89.48
r\4 –
70.30 94.15
r54 68.52
64.84 r\5
r55 –
46.03 l
max
tests r = 0
59.62 62.81
r = 1 69.64
– 53.98
r = 2 –
55.27 57.12
r = 1 48.33
r = 3 –
42.94 51.42
r = 2 45.28
42.32 34.47
r = 3 –
r = 4 39.37
36.76 r = 4
r = 5 –
24.27 r = 5
33.46 20.27
30.90 r = 6
–
a
r denotes the number of cointegrating vectors. Critical values are summarized from Johansen and Juselius 1990. and
denote, respectively, the significance at 95 and 90 confidence intervals.
Table 5 Trading hours of stock exchanges
New York time h Local time h
Greenwich mean time h Stock exchanges
Sydney 19:00–21:15
0:00–02:15 10:00–12:15
23:00–0:15 14:00–15:15
04:00–05:15 19:00–21:00
Tokyo 0:00–02: 00
9:00–11:00 23:00–01:00
13:00–15:00 04:00–06:00
19:40–21:40 09:40–11:40
0:40–02:40 Seoul
13:20–15:20 04:20–06:20
23:20–01:20 21:15–22:45
10:15–11:45 Shanghai
02:15–03:45 0:00–14:30
05:00–07:30 13:00–15:30
20:00–23:00 01:00–04:00
Taiwan 09:00–12:00
Manila 09:30–12:15
20:30–23:15 01:30–04:15
Jakarta 22:00–12:00
03:00–5:00 10:00–12:00
Hong Kong 21:00–11:30
02: 00–04:30 10:00–12:30
01:30–02:30 06:30–07:30
14:30–15:30 21:00–23:30
02:00–04:30 10:00–12:30
Singapore 14:30–16:00
01:30–03:00 6:30–08:00
21:00–22:00 02:00–03:00
Kuala Lumpur 10:00–11:00
11:15–12:30 22:15–23:30
03:15–04:30 01:30–03:00
06:30–08:00 14:30–16:00
20:30–05:00 01:30–10:00
Bangkok 08:30–17:00
New York 09:30–16:00
09:30–16:00 1430–21:00
Following Malliaris and Urrutia 1992, we make the following adjustments for the Granger causality. Suppose that a major world event occurs in Hong Kong or
other exchange in Asia and is announced at a certain point in time on a given trading day. The closing price of the same trading day on the SP 500 index in
New York will reflect this information. This illustrates that closing prices on day t in Hong Kong or other exchange in Asia affect closing prices in New York on the
same calendar day t. Thus, a Granger regression investigating whether Hong Kong or other exchange in Asia leads New York looks as follows:
X
t US
= C
+
P i = 1
a
i
X
t − i HK
+
P j = 1
b
j
X
t − j US
+ o
t
On the other hand, if important economic news occurs in the US, and is released at a certain point in time on a given trading day, the closing price of the next
trading day, rather than the same trading day, on the Hong Kong or other exchange in Asia will reflect this information. In other words, closing prices on day
t in New York may affect closing prices in Hong Kong or other exchange in Asia on day t + l and not on day t. Thus, a Granger regression postulating that New
York leads Hong Kong, after adjusting for time-zone differences, becomes:
X
t + 1 HK
= g +
P i = 1
a
i
X
t − i US
+
P j = 1
b
j
X
t − j HK
+ o
t
4
.
3
. Results The results of cointegration tests for each group are given in Table 6, which
shows that there is no evidence of any cointegrational relationship for the five North-East Asian country indices during and before the period
of the crisis. However, the evidence shows that at least one cointegrational relationship exists for the five South-East Asian country indices during the period
of the crisis.
The results of the Granger’s causality tests are presented in Table 7. Each entry in the table denotes the P-value of the market on the left-hand side caused by the
market at the top. The dominant role of the US stock market remains unchanged during the whole period of the financial crisis, and the results in Table 7 indicate
that the US market still causes some Asian markets during the period of the financial crisis. Behind South Korea, the US market plays the second dominant role
in the Asian financial crisis. On the other hand, only three Asian markets Hong Kong, South Korea and China cause feedback the US market. According to the
above results, the Asian financial crisis seems to have been not an intra-regional crisis affecting only the equity markets in East Asia.
Table 7 Granger’s casuality tests among US and Asian countries during the period of the crisis
H .-
C .
Sheng ,
A .H
. Tu
J .
of Multi
. Fin
. Manag
.
10 2000
345 –
365
360
Table 6 Test of cointegration among five North-East and five South-East Asian countries
a
Five South-East Critical value 95
Critical value 90 Alternative
Null hypothesis Five North-East
hypothesis Test statistic
Test statistic Before the
During the During the
Before the crisis
crisis crisis
crisis l
trace tests r = 0
68.52 r\0
64.84 45.09
61.29 –
90.76 47.21
43.95 46.43
– r51
38.11 21.41
r\1 24.53
r\2 29.68
26.79 11.77
20.34 –
r52 15.41
13.33 6.23
r53 –
r\3 5.16
7.93 r54
3.76 r\4
2.69 0.09
0.73 –
1.62 l
max tests –
44.33 33.46
30.90 23.68
r = 0 23.18
r = 1 27.07
24.73 21.90
– r = 1
17.77 9.64
r = 2 –
18.31 20.97
18.60 r = 3
r = 2 6.61
12.41 14.07
12.07 4.61
r = 3 –
r = 4 5.07
7.20 3.76
r = 4 2.69
r = 5 0.09
0.73 –
1.62
a
r denotes the number of cointegrating vectors. Critical values are summarized from Johansen and Juselius 1990. and
denote, respectively, the significance at 95 and 90 confidence intervals.
5. Variance decomposition
