Professional Documents
Culture Documents
Grant Colthup Thesis
Grant Colthup Thesis
Grant Colthup Thesis
1
0
(1)
where
t
x is a 1 n vector of nonstationary I (1) dependent variables,
0
A is a 1 n vector of
intercepts,
i
A is a n n coefficient matrix and
t
a 1 n vector of disturbance terms.
Equation (3) can be written in its Vector Error Correction (VEC herein) representation:
t t
p
i
i t t
x x A x + + + =
=
1
1
1
1 0
(2)
where
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
40
=
=
p
i
i
I A
1
,
+ =
=
p
i j
j i
A
1
(3)
Grangers original theory proposed that if the coefficient matrix has a reduced rank, such
that r < n then there is n r matrices and each with rank r. Furthermore, = and
t
x is I (0) where each column of the matrix is an error correction term and the speed
of adjustment. Johansens technique allows the matrix to be estimated from an unrestricted
VAR. If the rank equals 0 then a VAR in the differences is the correct specification. When the
rank equals n (number of equations contained in the VAR) then the VAR is stationary and no
correction is required. A rank equal to 1 indicates a single cointegrating vector, when the 1 <
rank < n there are multiple cointegrating vectors. Assuming 0 < rank < n, a test for the
number of cointegrating vectors is required. To conduct the test two test statistics are
computed:
+ =
=
n
r i
i trace
T r
1
)
1 ln( ) ( (4)
)
1 ln( ) 1 , (
1 max +
= +
r
T r r (5)
where
i
\
|
=
T
np T
unadjusted Adjusted
trace trace
(6)
where T is the total number of observations, n is the number of equations in the VAR and p is
the lag length. As discussed above the first step in conducting any test for cointegration is to
select an appropriate lag length using selection criteria. The lags must also be long enough to
remove any traces of serial correlation in the residuals of the VAR. All the cointegration tests
have been completed with differing lag structures in order to maintain the robustness of the
results. The first series of tests completed are bivariate tests between the sixty-six unique pairs
of markets
34
based on the full sample period (5/01/1988 to 7/09/2004).
The second set of cointegration tests are weekly recursive bivariate tests based on a 416-week
moving window. For week 0, all sixty-six tests of cointegration are completed based on data
from week -415 to week 0. The window then moves one week and the tests are re-computed
based on the data from week -414 to week 1. The procedure repeats until it reaches week 454.
3.3.2 Persistent Profile Analysis
Persistent profile analysis is a methodology developed by Pesaran and Shin (1996) that can be
used to examine the speed of convergence in response to a system-wide shock in the price
cointegrating relation. Darrat and Zhong (2004) note that the estimated profiles are unique
and do not require the prior orthogonalisation of shocks. As the time horizon increases, the
profile will approach zero as the effect of a shock on the cointegrating vectors is transitory by
nature and dissipates as the system returns back to equilibrium. The procedure used by Darrat
and Zhong and designed by Pesaran and Shin is outlined below. The first step is to estimate
the VECM as seen in equation (2):
t t
p
i
i t t
x x A x + + + =
=
1
1
1
1 0
(7)
where
t
x is a 1 12 vector of stock price indices,
t
x is a 1 12 vector of continuously
compounded returns,
0
A is a 1 12 vector of intercepts, and are 12 12 coefficient
34
There is only sixty-six unique pairs, because the test for cointegration between the US and Korea is the same
as the test between Korea and the US.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
42
matrices and is the r 12 long run coefficient matrix. Equation (2) is estimated using the JJ
method to derive an approximation of the coefficient matrix and the covariance matrix of
the residuals, .
The Beveridge and Nelson (1981) procedure is used to decompose the price vector
t
x , such
that:
t t t
x + = (8)
where
=
0 l
l t l t
C , and
=
+ =
0 l
l l l
B A C , and
=
=
l
j
j l
A B
0
(9)
1 1 0 1
+ + + =
t t t t
A A (10)
The price cointegrating relation can be written as:
+ + = =
0
1 0
) (
l
t l t t
B t x (11)
With r cointegrating vectors, the unscaled ( j i, ) elements of the persistence profile matrix
t
are given by:
j n n i ij
B B H = , where j i , and for r j i ,....., 2 , 1 , = ,.... 2 , 1 , 0 = n (12)
where
i
,
j
and are obtained from (8), and
n
B is the cumulative effect matrix defined
in (10) and is computed from the following recursive relation:
p n p n n n
B B B B
+ + + = ...
2 2 1 1
, n = 1, 2 and p is the order of the VAR (13)
where 0 ,
12 12 0
= =
n
B I B for 0 n and the s
i
' come from the matrices and such that:
p
I =
...
2 1 12 12
(14)
Therefore, the scaled measure of the persistent profile matrix of
t
is obtained as the elements
of:
)}, ( { ) ( ) ( n h G n GH n h
ij Z Z
= = where } ) 0 ( , ) 0 ( , ) 0 ( {
2 / 1 2 / 1
22
2 / 1
11
=
rr
H H H diag G (15)
Following a unit composite shock to the cointegrating vectors in a stationary system,
convergence to zero will occur.
.
Using a variety of measures, the short run linkage hypothesis is studied. The most obvious
choice would be to measure the level of correlation between the returns in the twelve markets.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
43
However, as discussed in Chapter 2 the presence of heteroscedasticity causes correlation
coefficients to become upwardly biased. The present solution to this problem is not
acceptable, requiring the imposition of a number of assumptions by the user. This thesis
proposes three alternative measures of short run linkage in place of using correlation analysis.
3.3.3 Generalised Variance Decomposition
Generalised forecast error variance decomposition (GVD henceforth) developed by Pesaran
and Shin (1998) is used to further examine the behaviour of market linkage. When markets
are not linked, a shock in one market will not be transmitted to others. Conversely, as the
level of market linkage intensifies, return innovations in one market explained by their
domestic variation should decline, while shocks from other markets assume greater
importance. The traditional method to conduct forecast error variance decomposition is the
Choleski procedure. The major weakness of the Choleski methodology is that the results are
sensitive to the ordering of the VAR. Furthermore, there is no firm theoretical or empirical
evidence to suggest how the variables should be ordered. The generalised form of variance
decomposition is unaffected by the order of the variables and produces unique results that
fully take account of the historical patterns of correlations observed amongst the different
shocks (Pesaran and Shin, 1998, p.4).
Another estimation issue needs to be addressed at this stage. The issue is whether or not to
conduct the GVD using a Vector Error Correction model (if needed) or an unrestricted VAR
in differences. Dekker et al (2001) provide an excellent discussion on the topic. They cite a
paper by Naka and Tufte (1997) who find that imposing the cointegrating vector does not
necessarily improve performance over short time horizons. In fact, Clements and Hendry
(1995), Engle and Yoo (1987) and Hoffman and Rasche (1996) found that an unrestricted
VAR is superior (in terms of forecast variance) to a restricted VECM over short time
horizons. Since the GVD is conducted over a 20-day horizon this thesis will use employ the
unrestricted differenced VAR. This also applies to the Generalised Impulse Response
functions outlined below.
Following the procedure from Darrat and Zhong (2004), the first step is to obtain estimates of
expected returns. The estimates are generated through using the VAR to forecast observed (or
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
44
actual) returns. The forecast errors are taken to represent the unexpected returns (shocks), the
VAR model of returns is as follows:
=
+ +
+ + =
k
l
t l t l k t
r r
1
1
(16)
where
t
r is a vector of continuously compounded stock returns of the twelve equity markets in
period t. is a vector of intercepts,
l
is a matrix of autoregressive parameters and
t
is a
white noise vector of disturbance terms. The n-period forecast error of returns is represented
by the following equation:
=
+ + +
=
1
0
n
l
l n t l n t t n t
D r E r (17)
The GVD of the n-step forecast error of return i explained by innovation, j is derived from the
following computation:
(
=
=
n
l
i l l i ij
n
l
j l i
ij
e D D e
e D e
n GVD
0
0
2
) (
) (
) (
(18)
where
i
e is an 1 12 selection vector with unity as its ith element and zeros elsewhere, is the
sum of squared residuals, ) (
t t
v v E and
ij
is the residual variance in the ith equation in the
VAR.
3.3.4 Generalised Impulse Response Functions
An alternative measure of market linkage is to examine the speed at which a market adjusts to
an innovation in another market. The argument put forward is that the greater the level of
market linkage then the impact of an innovation in a foreign market will quickly decay in the
domestic market. Bivariate generalised impulse response analysis will be considered. For the
bivariate analysis, the effects of innovations in the Japanese and US markets upon each other
and the PB countries will be examined. Like variance decomposition, Choleski decomposition
has traditionally been used to orthogonalise the innovations. As discussed above there are
several inherent deficiencies related to Choleski decomposition. Building on the work by
Koop, Pesaran and Potter (1996), Pesaran and Shin (1998) provide the procedure required to
conduct the generalised form of impulse response (GIR herein) analysis in the linear setting
35
.
35
Koop et al (1996) originally developed GIR analysis for non-linear multivariate models.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
45
Consider the following VAR of order p:
T t for x A x
t
p
i
i t i t
,..., 2 , 1
1
= + + =
=
(19)
where
t
x is a 1 n vector of dependent variables, is a 1 n vector of intercepts and
i
A is a
n n coefficient matrix. Pesaran and Shin make the following assumptions:
1.
t t all for E( matrix definite positive n n
an is n j i where t, all for E E
t t
ij t t t
= =
= = = =
0 )
} ,..., 2 , 1 , , { ) ( , 0 ) (
(20a)
2. circle unit the outside fall z A I of roots the All
p
i
i
i n
0
1
=
=
(20b)
3. collinear perfectly not are T 1,2,..., t for x x x
p t t t
=
, ,..., ,
2 1
(20c)
Assumption 2 ensures that
t
x is a covariance stationary process. Equation (19) can be re-
written as an infinite moving average (VMA) series:
= =
0
,
i
i t i t
T 1,2,..., t x (21)
where
i
is a n n coefficient matrix estimated from the following recursive estimation
process:
1,2,..., i A A A
p i p i i i
= + + + =
, ...
2 2 1 1
(22)
with
n
I =
0
and 0 =
i
for 0 < i , and
i i
G = .
The GIR function of
t
x at horizon k as stated in Koop et al (1996) is defined as;
) | ( ) , | ( ) , , (
1 1 1 + +
= =
t k t t t k t t X
x E x E k GI (23)
where is a 1 n vector of shocks and
1
t
is a non-decreasing information set that captures
the known history of the series up to time 1 t . Assume a shock is then applied to the jth
element of the error vector
t
. The effects are integrated out using an assumed or historically
observed distribution of the errors. A more specific form of equation (25) is the result:
) | ( ) , | ( ) , , (
1 1 1 + +
= =
t k t t j jt k t t j X
x E x E k GI (24)
If
t
is assumed multivariate normal, then Koop et al show that:
j jj j j jj nj j j j jt t
e E
1 1
2 1
) ,..., , ( ) | (
= = = (25)
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
46
where
j
e is an 1 n selection vector with unity as its jth element and zeros elsewhere, is the
sum of squared residuals, ) (
t t
v v E , as seen in equation (19). The 1 n vector of the unscaled
GIR of the effect of a shock in the jth equation at time t on
k t
x
+
is as follows:
0,1,2,... k
e A
jj
j
jj
j k
=
|
|
.
|
\
|
|
|
.
|
\
|
,
(26)
The scaled GIR function is derived by setting
j jj
= :
0,1,2,..., k e A k
j k jj
g
j
= =
, ) (
2 / 1
(27)
Due to the stationarity of the system, there is no concern over the responses converging to 0.
3.3.5 Block Exogeneity
The concept of block exogeneity is based upon Granger causality. X is said to cause Y if Y can
be more accurately forecast based on past values of X and Y compared to Y alone. If
substantial market linkage is discovered, then causality is expected. Furthermore, endogenous
markets like Japan and the US are likely to cause other markets, but no other markets are
likely to exert causality over them. In certain cases of strong market linkage causality could
possibly exist in both directions (i.e. market A causes market B and market B causes market
A). Consider the following VAR:
(
(
(
+
(
(
(
(
(
(
+
(
(
(
(
(
(
+
(
(
=
(
(
(
US
t
AU
t
US
t
AU
t
US
t
AU
t
US
AU
US
t
AU
t
r
r
B B
B B
r
r
A A
A A
a
a
r
r
M
M
L
M O M
K
M
L
M O M
K
M
M
2
2
12 , 12 1 , 12
12 , 1 1 , 1
1
1
12 , 12 1 , 12
12 , 1 1 , 1
(28)
To test the null hypothesis that returns in the US does not causes Australian returns, a
standard Wald test imposing the restriction 0
12 , 1 12 , 1
= = B A is no longer valid
36
. The
extension of the test for Granger causality to a multivariate setting is referred to as a block
exogeneity test.
36
Testing the restriction is only valid in a bivariate VAR model. This is the traditional Granger test.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
47
For Australia, the block exogeneity test essentially imposes a zero restriction on the lags of
AU
t
r in the eleven remaining equations. The test is formally conducted using a log likelihood
test as seen below
37
:
|) | log | | )(log (
u r
c T (29)
where | |
u
is the determinant from the covariance matrix from the unrestricted model, | |
r
is the determinant from the covariance matrix from the restricted model, T is the number of
observations c is the number of parameters estimated in each equation of the unrestricted
system
38
. If the null hypothesis is rejected it implies that the variable adds significant
information to the model.
The use of three measures to examine the level of short-term market linkage in the PB, will
present a broad and complete picture of the true relationship. All three measures have been
used in the previous literature and are seen as legitimate methods to observe the dynamics of
short run associations.
3.4 The time variation of market linkage
Based on the discussed in section 3.2 it would be likely that the level of market linkage varies
over time for such reasons as changes in trade and investment restrictions, the level of
investor confidence, legal and prudential regulation, composition of the economy and trading
partners and political and social attitudes. Furthermore, the EA crisis and events of September
11 are likely to have a significant enough affect to cause either a temporary or a permanent
change to the relationship between markets. Based upon this line of reasoning the following
hypotheses are proposed:
time over altered has Basin - Pacific the of markets stock the between ip relationsh The H
A
:
3
isturbance a market d following
nsifies Basin inte e Pacific- ets of th stock mark tween the ionship be :The relat H
3B
37
The testing procedure is provided by Enders (1995)
38
The unrestricted model is estimated by regressing
HK
t
r ,
ID
t
r ,
JP
t
r ,
KO
t
r ,
MY
t
r ,
NZ
t
r ,
PP
t
r ,
SG
t
r ,
TA
t
r ,
TH
t
r ,
US
t
r
against p lagged values of
AU
t
r ,
HK
t
r ,
ID
t
r ,
JP
t
r ,
KO
t
r ,
MY
t
r ,
NZ
t
r ,
PP
t
r ,
SG
t
r ,
TA
t
r ,
TH
t
r ,
US
t
r . The restricted
model is the same as the unrestricted model with the lagged values of
AU
t
r excluded from all eleven equations.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
48
This hypothesis requires the use of several methods to get the full picture. Weekly recursive
cointegration tests establish the fact that the cointegrating relationship has changed through
time. To enable easy interpretation of the volumes of data provided by the sixty-six pairs of
weekly tests all the trace statistics are graphed. An example can be seen below in Figure 3.4.
Figure 3.4 Recursive cointegration test between Australia and Hong
Kong
0
0.5
1
1.5
2
/
0
1
/
9
6
2
/
0
5
/
9
6
2
/
0
9
/
9
6
2
/
0
1
/
9
7
2
/
0
5
/
9
7
2
/
0
9
/
9
7
2
/
0
1
/
9
8
2
/
0
5
/
9
8
2
/
0
9
/
9
8
2
/
0
1
/
9
9
2
/
0
5
/
9
9
2
/
0
9
/
9
9
2
/
0
1
/
0
0
2
/
0
5
/
0
0
2
/
0
9
/
0
0
2
/
0
1
/
0
1
2
/
0
5
/
0
1
2
/
0
9
/
0
1
2
/
0
1
/
0
2
2
/
0
5
/
0
2
2
/
0
9
/
0
2
2
/
0
1
/
0
3
2
/
0
5
/
0
3
2
/
0
9
/
0
3
2
/
0
1
/
0
4
2
/
0
5
/
0
4
2
/
0
9
/
0
4
Date
N
o
r
m
a
l
i
s
e
d
T
r
a
c
e
S
t
a
t
i
s
t
i
c
Normalised Trace Statistic Normalised Critical Value Asian Crisis Sep-11
The design of the graph is such that when the line is above one, significant cointegration
exists between the two markets. Again, all trace statistics reported in this thesis have been
adjusted for small sample bias.
A secondary technique used to determine the effect of the EA crisis and September 11 on the
level of market linkage is to include an intercept dummy variable for the East Asian crisis and
September 11, in the test for cointegration. The use of this technique represents a substantial
improvement in accurately detecting changes in the long run relationship and is only
employed for the bivariate and multivariate tests on the full period (5/01/1988 to 7/09/2004).
To determine if the level of short run market linkage fluctuates over time, the sample is
broken into three sub-samples and all the tests re-computed. The first sub-sample starts from
the beginning of the sample (1/05/1988) and runs to the 31 October 1997. The second period
runs from the 1 November 1997 to September 11 2001 and the last period extends from
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
49
September 12 2001 to 7 September 2004. This provides significant evidence as to how the
degree of short-run market linkage varies over time.
3.5 Factors responsible for market linkages
This thesis, postulates that several variables influence the level of market linkage. Table 3.5
summarizes the expected affect of the variables on market linkage. The following discussion
highlights the theoretical underpinnings as to how these variables influence the level of
market linkage.
Table 3.5 Expected relationship between
market linkage and the factors proposed
Variable
Proposed
Effect
Goods Market Integration +
Trade Dummy +
Market Development
Turnover ratio -
Size Difference -
Size Dummy +
Exchange rate volatility -
Returns volatility +
Money Market
Money Market integration +
Interest rate differential +
Location -
A. Goods Market Integration and Trade
Goods market integration is likely to be a dominant factor in explaining market linkage.
Referring back to economic theory, the goods market is defined as the IS curve. The IS and
LM curves combine to form the aggregate demand function (AD), AD is equivalent to the
growth of the economy. The inclusion of this variable also helps to control for any shocks in
aggregate demand. It also is a good proxy for the level of economic growth in that market.
The effect of economic growth on equities markets is obvious, that is significant appreciations
will occur during times of good growth and depreciations during contractions.
Consider two economies that are characterised by large natural resource and primary
production sectors, that is, they have a similar industry composition. If commodity prices rise,
then those companies in both economies will have higher earnings without incurring any extra
costs and therefore increase profit and cash flows to the firm. The market would interpret this
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
50
news favourably and subsequently the stock markets in both economies will exhibit an
appreciable increase.
Now let us extend the model further by including two other economies that produce durable
consumer and capital goods. Neither economy has any significant natural resource or primary
production. As a result, they import all their necessary production inputs from the two
economies mentioned above. The effect of an increase in commodity prices will make
production more expensive. The companies of these two economies then have two options
available to them. They can keep prices fixed, remain competitive in the global market,
absorb the cost themselves, and earn lower profits and cash flows. Alternatively, they could
increase price to retain their profit margin but lose competitiveness and again suffer a
reduction in profit and cash flows. Either option will result in a decline in profit and cash
flows. Investors in these economies will view this news with displeasure and the price of
these firms will depreciate along with the respective national stock markets. Obviously if the
price of capital goods increased, the scenario would occur the other way around but the result
would still be the same. That is, the two small open natural resource and primary producers
would move together and the industrial production economies move together.
Finally, consider the effect of a positive (negative) shock to the industrial production
economies. Production in the economy rises (declines); therefore, more (less) inputs are
required from the other two economies. In this case, the stock markets in all four economies
will experience an appreciation (depreciation).
The scenarios described above all rely upon the possibility of trade between markets. If trade
is not possible then good market integration is unlikely. The lack of international trade is an
indication of market segmentation. In segmented markets, domestic factors are the only
providers of information. As discussed in the previous section the Pacific-Basin region has a
high level of trade between each other and the world. Therefore, trade barriers do not limit the
probability of goods market integration. Most nations in the Pacific-Basin are net exporters
and therefore rely heavily upon trade to promote growth and innovation. Thus, trade is of vital
importance to these economies.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
51
In this thesis, the level of goods market integration is captured by running bivariate
cointegration tests on the industrial production statistics of each economy. Since industrial
production statistics are used to measure the level of goods market production, the trace
statistic from the cointegration test must provide an insight into the long run relationship
between the goods markets of two economies. Sixty-six tests in total are conducted based on
monthly data from 15/01/1989 to 15/06/2004. Darrat and Zhong (2004) use a similar metric to
indicate the level of goods market integration between the two markets.
Two proxies are presented in this study to describe the effect of trade upon market linkage. As
per the discussion above it is theorised that trade will be a vital role in determining market
connections. The two measures employed in this thesis are dummy variables. The first
dummy considers the effect of trade organisations upon market linkage. If a pair of markets
both belongs to ASEAN then the dummy takes a value of one and zero otherwise.
B. Location
The geographical location of markets is likely to affect the level of linkage for two reasons.
The first is that, markets that are close to each other are more likely to experience economic
integration. This is because it has been shown through out the world that markets that are
close to each other display a higher than average level of trade. As discussed above, trade
provides a transmission channel between economies to transmit stock market comovements.
Secondly, markets close in geographical proximity are likely to share similar factor
endowments. That is, markets that are close to one another have a similar environment and
climate. Therefore, it is probable that they have similar deposits of natural resources and
ability to cultivate certain primary products. As argued above, markets of similar industry
composition and factor endowments are going to comove in the same direction.
The second theory as to why markets of a close geographical location are likely to comove
with each other is the home bias puzzle. Expanding this theory to an international level,
investors are more likely to invest in countries that are close because they believe they 'know'
them better
39
. Therefore, if market A performs well and investors from market B have an
investment in market A, they will have higher wealth. Investors from market B may then
39
For example investors from Singapore are likely to invest in Malaysia instead of New Zealand
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
52
consume this additional wealth in country B, consequently firms in market B perform better
and the markets display comovement.
The manner in which this variable is operationalised in this thesis is just to measure the
distance between two markets in kilometres. As seen in table 3.5 a negative relationship is
expected, however the effect of the US could significantly influence the results and a positive
relation discovered because the US is so open, but also so far away.
C. Market Development
The effect of market development upon market linkage would appear to be obvious. The
logical conclusion would be to assume developed markets are display higher levels of market
linkage. However, as markets become further developed they also become more exogenous.
Therefore, the determining factor becomes not the absolute level of development, but the
relative level of development. In the literature to date there have been numerous suggested to
describe this variable but no solid findings. Bekaert and Harvey (1997) argue that market
capitalisation is a good proxy for the level of development in a market. Expanding on this
proposition, this thesis presents two variables as proxies for size. The first is the large market
group dummy variable that takes a value of one when Australia, Hong Kong, Japan, Korea,
Taiwan or the US are in the pair and a zero otherwise. The small market group dummy
variable takes a value of one when the pair contains Indonesia, Malaysia, New Zealand, the
Philippines, Singapore or Thailand
40
. Another variable computed is also based on the size of
the relevant markets. Section 4.2 provides an in depth description of this size difference
variable. Lastly, a concept new to this thesis is the use of a turnover ratio to proxy for market
liquidity and ultimately market development. The effect of deep and liquid markets and their
relation to market development has been well established in the previous literature. Again,
section 4.2 provides a comprehensive discussion surrounding the properties of this variable.
D. Money Market Linkage
This thesis proposes that the level of money market linkage affects the level of market
linkage. As seen in section 3.2 the benefits and costs from capital flows are considerable and
can be long lasting. While much of the growth in the East Asian economies can be linked to
40
This thesis realizes that these markets are not necessarily small or unimportant. The large dummy variable was
created by finding the markets with the six highest market capitalizations over the 16-year period.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
53
massive capital flows from Japan and the US during the 1980s, the EA crisis just as readily
highlights the pitfalls associated with unrestricted capital movements. The probability of
market linkage increases significantly in the presence of capital market integration for several
reasons.
In economics, the LM curve represents capital markets. The stance of the monetary policy
adopted by the respective monetary authority dictates the position of the LM curve.
Consequently, capital market integration has significant ramifications for the breadth and
depth of the conduct of monetary policy. Assume a country adopts a contractionary monetary
policy stance. The increased interest rate raises the cost of capital and subsequently
investment in the economy falls along with economic growth. According to the theory of
interest rate parity, the rise in interest rates will also cause the domestic currency to
appreciate. In turn, exports become more expensive and imports cheaper, leading to a
worsening of the balance of trade and ultimately a decline in GDP. The decline in aggregate
output can potentially have significant consequences for other economies.
Another way for the money market to influence the level of the linkage between markets is
through the Present Value Model (PVM herein). Essentially under PVM the value of a firm is
its future dividends discounted back to today, where the discount rate is a function of the risk
free rate. Therefore, if the risk free rate rises, so will the appropriate discount rate for that
firm. Ceteris paribus, if the discount rate increases then the price of the firm will fall. If two
markets have display money market linkage then it is highly likely that their interest rates will
move in a similar direction, thus both stock indexes will appreciate.
To examine the effect of money market linkage, two variables have been put forward. The
first is the absolute weekly interest rate differential between the two markets averaged over
the period 5 January 1993 to 7 September 2004. Under the concept of interest rate parity, it
would be assumed that any deviations away would cause substantial flows in capital. The
discussion above supposes that capital flows can cause significant linkage between markets.
However there has been some literature that has found that uncovered interest rate parity (UIP
henceforth) does not hold. If, a negative relation is found between the proxy for monry market
linkage and equity market linkage then it is proof toward the PVM. If a positive relationship
is uncovered then it is proof in the favour UIP holding.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
54
The second measure used to proxy for money market linkage is the trace statistic from a test
for cointegration conducted on the interest rates of the respective markets. The use of this
measure largely extends from the work of Phylaktis (1999) that uses cointegration tests to
proxy for capital market integration.
E. Market Volatility
Extensive research has established the importance of volatility in stock markets. Therefore,
the volatility of markets will influence the level of market linkage. The market contagion
hypothesis provides the theoretical underpinning for this expectation. Market contagion
dictates that markets tend to exhibit higher linkage when they display higher volatility.
Therefore, it would be expected that volatility is positively related to market linkage if the
contagion theory is credible. However, emerging literature led by Forbes and Rigobon (2002)
have found that the EA crisis did not result in contagion between the markets of the PB. The
result of this is that either situation is legitimate considering that both are based on sound
economic theory and robust empirical evidence.
F. Exchange Rate Volatility
Exchange rates play an important part of any economy. This study supposes two ways for
exchange rate volatility to influence the level of market linkage. The first is that when
investors are making their decisions, they compute returns denominated in their own domestic
currency. Therefore, volatility in exchange rates can raises the systematic risk of that nation.
Therefore risk averse investors, are less likely to invest into the market, thus driving the two
markets apart.
Furthermore, fluctuations in exchange rates affect the level of trade in an economy through
deviations away from purchasing power parity (PPP). PPP dictates that a countries level of
exports and imports will vary according to the prevailing real exchange rate. As discussed
above trade plays a crucial role in promoting market linkage and goods market integration.
Therefore the greater the volatility in the exchange rates, the higher the probability of
deviations away from PPP and subsequently a higher level of trade However, it is important
to note that PPP has found not to hold in empirical testing. Therefore, if a positive relation is
found then it is likely that PPP is holding, as opposed to a negative relationship which
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
55
provides evidence toward the first argument presented above.. The two theories put forward
to explain the effect of exchange rate volatility on market linkage actually seem to oppose one
another. However, both are economically sound and valid arguments, thus making it difficult
to predict the direction of the relationship. This actually provides for an interesting
comparison to see if it is investors decision making or wider economic variables that
influence market linkage.
Based upon the above theoretical underpinnings, a cross-sectional regression model is
proposed to determine which factors significantly affect market linkage.
i SMALL LARGE OECD ASEAN j i
rate X
j i
returns
j i j i j i, j i, j i j i, j i
D D D D TO
r Trace IR Sdiff Location Trace IP Trace
+ + + + + + +
+ + + + + + + =
12 11 10 9 , 8 , 7
, 6 , 5 4 3 , 2 1 ,
where,
j i
Trace
,
is the trace test statistic extracted from a bivariate cointegration test between the
stock markets indices of country i and j. Sixty-six tests in total are conducted based on data
from the full sample (5/01/1988 to 7/09/2004). Since over 760 observations are included for
the tests of cointegration, the results obtained are highly robust and persuasive.
j i
Trace
,
is an
indication of the strength of the relationship between the two stock markets. A value in excess
of 19.96 indicates the presence of cointegration at the 95% confidence level. In general, high
trace statistics are an indication of stronger market linkage. To ensure the robustness of the
result obtained, the maximum test statistic and eigenvalues from the cointegration test will be
taken as further proxies for market linkage. Masih and Masih (2001) note that eigenvalues are
an indication of the intensity of the cointegrating relationship.
j i,
Trace IP is the trace statistic extracted from a bivariate cointegration between the industrial
production indexes of markets i and j. Sixty-six tests in total are conducted based on monthly
data from 15/01/1989 to 15/06/2004. Darrat and Zhong (2004) use a similar metric to indicate
the level of goods market integration between the two markets. A value in excess of 19.96
indicates the presence of cointegration at the 95% confidence level. Similar to
j i
Trace
,
, a high
statistic represents greater goods market integration.
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
56
j i
Location
,
is the distance measured in kilometres between the cities in which the stock
exchanges are situated.
j i,
Sdiff is the average weekly logged market capitalisation of market i deflated by the average
weekly logged market capitalisation of market j. For further clarification, refer to section 4.2.
j i
TO
,
is the bilateral turnover ratio between markets i and j
41
. The manner in which the
statistics were computed a low ratio, indicates high liquidity.
j i,
Trace IR is the trace statistic extracted from a bivariate cointegration test between the
interest rates of markets i and j. Sixty-six tests in total are conducted based on weekly data for
the period 5
th
January 1993 to 7
th
September 2004.
j i,
Trace IR is an indication of the level of
capital market integration between the two markets. A value in excess of 19.96 indicates the
presence of cointegration at the 95% confidence level.
returns
j i,
is the equally weighted standard deviations of market i and j averaged over the full
period. The individual standard deviations are extracted from a bivariate GARCH (1, 1)
model. All sixty-six standard deviations are derived from returns starting the 12
th
January
1988 and ending the 7
th
September 2004.
rate
,
X
j i
is the volatility between the exchange rates of markets i and j extracted from a
univariate GARCH (1, 1) model. Sixty-six tests in total are conducted based on weekly
exchange rate returns starting on the 2
nd
of January 1988 and finishing the 7
th
of September
2004.
j i
r
,
is the equally weighted mean return of market i and j for the period 12
th
January 1988 to
the 7
th
September 2004.
41
The method used to compute the weekly turnover ratios is discussed in Section 4.2
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
57
ASEAN
D is a dummy variable that takes a value of 1 when markets i and j are ASEAN member
nations, and 0 otherwise
OECD
D is a dummy variable that takes a value of 1 when markets i and j are OECD member
nations, and 0 otherwise
LARGE
D is a dummy variable that takes a value of 1 when markets i and j are a pair containing
Australia, Hong Kong, Japan, Singapore, Taiwan or the US and 0 otherwise.
SMALL
D is a dummy variable that takes a value of 1 when markets i and j are a pair containing
Indonesia, Korea, Malaysia, New Zealand, the Philippines or Thailand and 0 otherwise.
Not all twelve variables will be regressed against the dependent variable in a single equation.
Several variables proxy for the same effect, therefore multiple models will be estimated with
variables certain factors omitted. Equations including slope dummy variables (interaction
terms) are also employed. To ensure the robustness of the regression several diagnostic tests
are completed. All regressions are estimated with robust standard errors using the Newey-
West procedure. Furthermore, the correlation matrix of independent ensures multicollinearity
is not an issue. A Probit model, where the dependent variable takes a value of 1 if significant
bivariate cointegration is found
42
and 0 otherwise. The Probit model will have the same set of
independent variables as the OLS equation. For both models, a series of univariate and
multivariate equations will be estimated.
i SMALL LARGE OECD ASEAN j i
rate X
j i
returns
j i j i j i, j i, j i j i, j i
D D D D TO
r Trace IR Sdiff Location Trace IP
+ + + + + + +
+ + + + + + + =
12 11 10 9 , 8 , 7
, 6 , 5 4 3 , 2 1 ,
Pr
The reasons behind using two different estimation techniques for are:
1. The OLS equation allows linkage to be measured in a continuous rather than discrete
setting. This is important distinction because there are likely to be situations where
strong linkage is discovered but is just not strong enough to qualify as significant.
2. An interesting feature of Probit models is the following relation:
i i
probabilty * = (30)
42
At the 5% significance level
CHAPTER 3HYPOTHESIS AND THEROY DEVELOPMENT
58
where
i
is the coefficient for variable i and
i
is the standard deviation of variable i. The
probabilty measures the increase (decrease) in the probability of markets being linked due
to a one standard change in i with all other coefficients held constant. This allows the
economic significance of variable i upon market linkage to be examined.
CHAPTER 4SAMPLE DATA
59
Chapter 4: Sample Data
4.1 Introduction
Considering the importance of statistical testing procedures in this study, it is imperative that
the data is of the highest quality. Section 2 of this chapter outlines the data collected, any
adjustments required to ensure the integrity of the data and the methods used to compute
certain variables. Lastly, section 3 contains the descriptive statistics for the returns series of
each market.
4.2 Data and its stochastic properties
Many prior studies have shown that a long sample period is necessary to be able to discover a
cointegrating relationship. Weekly price levels were collected for the following stock
markets; the S&P500 (the U.S.), Nikkei 225 (Japan), Hang Seng (Hong Kong), Jakarta
Composite (Indonesia), KLSE Composite (Malaysia), PSE Composite (Philippines), Straits
Times (Singapore), Seoul Composite (South Korea), SET (Thailand) and the Taiwan
Weighted (Taiwan). For Australia and New Zealand indexes constructed by DataStream were
employed because of missing data, the methodology used to construct the indexes is outlined
in Appendix 1. For all markets, the sample starts on the 5
th
of January 1988 and finishes on
the 7
th
of September 2004, giving 871 observations per country. The week runs from Tuesday
to Tuesday in an attempt to avoid the bias often times associated with data from Monday or
Friday. . The returns series were generated using continuous compounding in the following
calculation:
) ln( ) ln(
1
=
it it it
P P r (1)
where
it
r is the continuously compounded return on index i at time t and
it
P is the price level
of index i at time t. Table 4.1 panel A and B show that in all four periods that stock prices are
found to be non-stationary, while returns are stationary at the 1% significance level. To proxy
for the market contagion effect, the equally weighted standard deviation of the pair i and j is
computed:
CHAPTER 4SAMPLE DATA
60
2
,
j i
j i
+
= (2)
Table 4.1. Augmented Dickey-Fuller and Phillips-Perron Unit Root Tests (p-values)
Panel A: Log Stock Price Levels
Country Full Period
Pre- East Asian
Period
Post - East Asian
Period
Post - September
11 Period
Australia
ADF 0.564 0.789 0.519 0.700
PP 0.617 0.783 0.491 0.739
Hong Kong
ADF 0.405 0.816 0.488 0.758
PP 0.407 0.819 0.488 0.756
Indonesia
ADF 0.941 0.249
a
0.259 0.942
PP 0.930 0.248
a
0.146 0.924
Japan
ADF 0.773 0.708 0.909 0.482
PP 0.765 0.641 0.930 0.532
South Korea
ADF 0.194
a
0.194 0.577 0.282
PP 0.124
a
0.195 0.569 0.265
Malaysia
ADF 0.377
a
0.173 0.426 0.766
PP 0.298
a
0.173 0.365 0.741
New Zealand
ADF 0.503 0.785 0.470 0.922
PP 0.541 0.801 0.454 0.923
Philippines
ADF 0.398 0.661 0.534 0.870
PP 0.315 0.629 0.417 0.797
Singapore
ADF 0.191
a
0.575
a
0.633 0.869
PP 0.124
a
0.572
a
0.616 0.885
Taiwan
ADF 0.060
a
0.248
a
0.932 0.361
PP 0.060
a
0.158
a
0.922 0.361
Thailand
ADF 0.455 0.190 0.179 0.868
PP 0.371 0.185 0.108 0.860
US
ADF 0.946 0.994 0.318 0.449
PP 0.927 0.999 0.369 0.543
Panel B: Log Returns
CHAPTER 4SAMPLE DATA
61
Country Full Period
Pre- East Asian
Period
Post - East Asian
Period
Post - September
11 Period
Australia
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Hong Kong
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Indonesia
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Japan
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
South Korea
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Malaysia
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
New Zealand
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Philippines
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Singapore
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Taiwan
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Thailand
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
US
ADF 0.000** 0.000** 0.000** 0.000**
PP 0.000** 0.000** 0.000** 0.000**
Notes: This table reports the unit root tests for the log price levels and log returns of the
twelve national stock indexes. The full period is January 5, 1988 to September 7, 2004, the
pre- East Asian period is January 5, 1988 to October 21, 1997, the post East Asian period is
October, 22, 1998 to September, 11, 2001 and the post September 11 period is September
12, 2001 to September 7, 2004. The numbers reported are the p-values of the ADF and PP
tests statistics for unit roots. If not otherwise specified a constant was included in the unit root
test.
a
indicates a trend term was included in the test regression. An ** indicates the rejection
of the null hypothesis that the series contains a unit root at the 5% significance level.
CHAPTER 4SAMPLE DATA
62
where
i
and
j
is the standard deviation of markets i and j respectively. A GARCH (1, 1)
model provided the estimates for the weekly conditional variance. The sixteen-year average
for each of the two elements was determined and the square root taken to obtain the standard
deviation.
The city distance tool hosted on http://www.geobytes.com/CityDistanceTool.htm gave the
distance (in kilometres) between the markets i and j. To determine capital market integration,
suitable interest data was required. The most appropriate rate to use would be the cash or
overnight rate set by the relevant monetary authorities. However, due to the limitations of
available data over the testing period other short-term rates had to be employed. Following
the example of Bracker et al (1999), various three-month interest rates were employed. For all
interest rates, the data was available from the 5
th
January 1993 to the 7
th
of September 2004.
Three-month T-bill rates were used for the U.S., Singapore and the Philippines. For
Indonesia, Malaysia and Thailand 3-month deposit rates were obtained, while for Australia,
Hong Kong and New Zealand three-month interbank rates were collected.
Table 4.2 Augmented Dickey-Fuller and Phillips-Perron Unit Root Tests (p-
values) for Interest Rates
Log Levels Log Differences
ADF PP ADF PP
Country
Australia 0.645 0.449 0.000*** 0.000***
Hong Kong 0.488 0.224 0.000*** 0.000***
Indonesia 0.592 0.322 0.000*** 0.000***
Japan 0.247
a
0.166
a
0.000*** 0.000***
South Korea 0.461 0.538 0.000*** 0.000***
Malaysia 0.553 0.563 0.000*** 0.000***
New Zealand 0.372 0.315 0.000*** 0.000***
Philippines 0.115 0.147 0.000*** 0.000***
Singapore 0.108 0.113 0.000*** 0.000***
Taiwan 0.663 0.659 0.000*** 0.000***
Thailand 0.959 0.933 0.000*** 0.000***
US 0.946 0.927 0.000*** 0.000***
Notes: This table reports the unit root tests for the log interest rates and the
differenced log interest rates. The full period is 5
th
January 1993 to the 7
th
of
September 2004. The numbers reported are the p-values of the ADF and
PP tests statistics for unit roots. An intercept was included in the test
regression.
a
indicates a trend term was included in the test regression. An
*** indicates the rejection of the null hypothesis that the series contains a
unit root at the 1% significance level, while ** and * indicate rejection at the
5% and 1% levels respectively.
CHAPTER 4SAMPLE DATA
63
The Japanese three-month certificate of deposit rate, South Korean 91-day certificate of
deposit and the Taiwanese 90 day money market rate were also collected. These rates still
have a short enough time horizon to accurately reflect the monetary policy goals of the
respective authorities. Table 4.2 shows that interest rates are non-stationary in their levels, but
stationary in the first differences. Therefore, it can be concluded that interest rates are I (1)
and therefore a cointegration test can be conducted in order to extract the trace statistics
required to proxy for money market linkage. The use of the trace statistics generated from the
cointegration tests on interest rates to explain money market linkage is unique to this study.
The exchange rates for the different countries were collected from DataStream. Due to the
lack of data for the period being examined, the rates were computed using cross rates. For all
thirteen markets, the bilateral exchange rate with the US was available for the entire period.
The cross rates were computed as follows:
HKD
USD
USD
AUD
HKD
AUD
* = (2)
The use of cross rates is not ideal, but the level of error is extremely small and still allows for
strong comparisons. Furthermore, in practice virtually all currency traders convert their
domestic currency into USD in order to purchase other foreign currency and implicitly use
cross rates anyway.
The bilateral exchange rates were required to compute the volatility of the bilateral exchange
rate. The first step was to generate the weekly continuously compounded returns using the
procedure outlined in equation (1) for the period 2
nd
January 1990 to 7
th
September 2004. A
GARCH (1, 1) model was employed to extract the weekly conditional variance. The variance
was averaged over the sixteen years and the square root taken to derive the standard deviation.
Industrial production statistics were collected from DataStream on a monthly basis for the
twelve markets being examined, for the period 15/01/1989 to 15/06/2004. Industrial
production statistics are a good proxy for both industry composition and output of the
economy. Similar to capital market integration, the proxy for goods market integration will be
the trace statistic from a bivariate test for cointegration. The motivation for this choice of
CHAPTER 4SAMPLE DATA
64
statistic stems from Darrat and Zhong (2004) whom employ a similar methodology. As per
the discussion pertaining to capital market integration, cointegration tests require the system
of variables to be integrated of the same order. Table 4.3 and 4.4contains the results for the
unit root test for the industrial production statistics.
Table 4.3 Unit Root tests for Industrial Production Statistics in log levels
Critical Values Critical Values
Country
Augmented
Dickey-Fuller
test statistic 90% 95% 99%
Phillips-Perron
test Statistic
90% 95% 99%
AU -1.4247 -2.575 -2.877 -3.466 -1.406 -2.575 -2.877 -3.466
HK -1.2303 -2.575 -2.877 -3.466 -1.585 -2.575 -2.877 -3.466
ID -0.2092 -2.575 -2.877 -3.466 0.047 -2.575 -2.877 -3.466
JP -1.9762 -2.575 -2.878 -3.467 -11.383*** -2.575 -2.877 -3.466
KO -0.3180 -2.575 -2.877 -3.466 -0.293 -2.575 -2.877 -3.466
MY -0.5969 -2.575 -2.877 -3.466 -0.529 -2.575 -2.877 -3.466
NZ 0.3147 -2.575 -2.877 -3.466 0.308 -2.575 -2.877 -3.466
PH -1.1725 -2.575 -2.877 -3.466 -1.580 -2.575 -2.877 -3.466
SG -2.0599 -2.575 -2.877 -3.466 -2.272 -2.575 -2.877 -3.466
TA -2.0521 -2.575 -2.877 -3.466 -2.373 -2.575 -2.877 -3.466
TH -1.1342 -2.575 -2.877 -3.466 -1.116 -2.575 -2.877 -3.466
US -0.4265 -2.575 -2.877 -3.466 -0.534 -2.575 -2.877 -3.466
Table 4.4 Unit Root tests for the log differenced Industrial Production Statistics
Critical Values Critical Values
Country
Augmented
Dickey-Fuller
test statistic 90% 95% 99%
Phillips-Perron
test Statistic
90% 95% 99%
AU -9.49609*** -2.575 -2.877 -3.466 -13.30209*** -2.575 -2.877 -3.466
HK -16.8483*** -2.575 -2.877 -3.466 -19.25698*** -2.575 -2.877 -3.466
ID -7.20959*** -2.575 -2.877 -3.466 -7.52741*** -2.575 -2.877 -3.466
JP -25.39081*** -2.575 -2.877 -3.466 -20.60083*** -2.575 -2.877 -3.466
KO -9.46027*** -2.575 -2.877 -3.466 -15.66235*** -2.575 -2.877 -3.466
MY -12.94404*** -2.575 -2.877 -3.466 -21.19589*** -2.575 -2.877 -3.466
NZ -9.76598*** -2.575 -2.877 -3.466 -13.82546*** -2.575 -2.877 -3.466
PH -15.60689*** -2.575 -2.877 -3.466 -20.9483*** -2.575 -2.877 -3.466
SG -22.29304*** -2.575 -2.877 -3.466 -28.08833*** -2.575 -2.877 -3.466
TA -15.63103*** -2.575 -2.877 -3.466 -24.01909*** -2.575 -2.877 -3.466
TH -11.66317*** -2.575 -2.877 -3.466 -21.33933*** -2.575 -2.877 -3.466
US -14.73194*** -2.575 -2.877 -3.466 -22.61475*** -2.575 -2.877 -3.466
Notes: This table reports the unit root tests for the log price levels and log returns for the industrial
production statistics for the twelve countries. Monthly data is used. The full period covers 15/01/1989 to
15/06/2004. AU ! Australia, HK ! Hong Kong, ID ! Indonesia, JP ! Japan, KO ! Korea, MY !
Malaysia, NZ ! New Zealand, PP ! The Philippines, SG ! Singapore, TA ! Taiwan, TH ! Thailand
and US ! The United States. An intercept is included in all test regressions. An *** indicates the
rejection of the null hypothesis that the series contains a unit root at the 1% significance level, while **
and * indicate rejection at the 5% and 1% levels respectively.
CHAPTER 4SAMPLE DATA
65
In order to calculate the size differentials between markets, the weekly market capitalisations
(denominated in US dollars) of the twelve markets was collected for the period between
3/04/1990 to 7/09/2004. This data was also required to compute turnover ratios. In this thesis,
size differentials were computed in the following fashion:
j
i
j i
SI ln Avg
SI ln Avg
sdiff =
,
(3)
where
i
SI ln Avg is the natural log of stock market capitalisation i computed at weekly
intervals. The average of the logged stock index i is then calculated for the full period. The
same procedure is repeated for market j. The impetus behind this proxy derives from Bracker
et al (1999) who utilize a similar metric.
Before, continuing some preliminary notes regarding this statistic are required. The first is
that the US never appears in the numerator because of the ordering. The consequence of this
is that, for the eleven cases when the US is in the denominator an average difference of
around 0.6 is returned (excluding Japan). This thesis proposes that most markets display
strong linkage the US, therefore when is smaller
j i
sdiff
,
higher market linkage is expected.
Size differences of over one likely represent cases when Australia, Japan, South Korea and
Taiwan appear in the numerator and the less developed economies of SE Asian in the
denominator. Based on past empirical evidence, economic links and information presented in
chapter 3, the proposition that Australia, South Korea, Taiwan and especially Japan do not
display significant market linkage with the less developed markets in SE Asia is reasonable.
The combination of facts leads to the conclusion that, when
j i
sdiff
,
,is high market linkage is
expected to decline.
When the two markets are of the same size it is expected that market linkage will higher..
Values of one are expected in two situations. The first is when two large markets are
included in the computation and the second when two small markets are compared. When
considering the large markets group (Australia, Hong Kong, Japan and Taiwan and the US),
this thesis predicts that the level of linkage between these markets to exceed the average.
CHAPTER 4SAMPLE DATA
66
Turnover ratios serve, as a proxy for market development is a proposition presented in this
thesis. Market capitalisation and trading volume figures were obtained from DataStream.
Turnover ratios are traditionally estimated as the value of the volume traded deflated by the
market capitalisation. Data restrictions made this not possible for this thesis. In an attempt to
mitigate the effects of exchange rates and market composition, all market capitalisations
figures are denominated in USD terms. The trading volume (measured in the number of
shares not value) was also collected for the twelve markets. The major problem that arises
from using the physical number of shares traded is that in some markets many relatively
cheap stocks might be traded quite often, while in other markets more expensive stocks are
traded at a lower frequency. Thus, it would appear that the first market is liquid when in fact it
may not be. Therefore, by collecting all market capitalisation figures in USD and using
domestic trading volumes allows for greater comparability. For example, the turnover ratio
for Korea would be completed as follows:
USD) (in Korea of tion Capitalisa Market
Korea) (in traded shares of Number
TO
KO
= (3)
All turnover ratios are computed on a weekly basis. These ratios are then averaged over the
period 23/04/1991 to 7/09/2004, such that one ratio is obtained for each market.
Table 4.5 - Turnover ratios
for all 12 markets
Country Turnover
Australia 2.992405
Hong Kong 3.012569
Indonesia 15.26897
Japan 0.8147
Korea 2.922903
Malaysia 3.253496
New Zealand 4.123634
Philippines 14.34273
Singapore 4.195414
Taiwan 3.985235
Thailand 13.52558
US 0.575817
The bilateral turnover ratio is the equally weighted turnover ratio between market i and j. The
lower the ratio the more advanced the market; Japan and the US have the two lowest, while
CHAPTER 4SAMPLE DATA
67
Indonesia, the Philippines and Thailand have the three highest. As can be seen the results
obtained appear to be fairy accurate proxies of market development.
4.3 Descriptive Statistics
Table 4.6 displays the descriptive statistics for the returns series of the twelve stock markets.
The table highlights several issues. The first is the outstanding performance of Australia and
the US over the sample period. Australia had a mean weekly return of 0.151% (4
th
highest)
and the lowest variance, while the US had a weekly mean return of 0.169% (3
rd
highest) and
the second lowest variance. By contrast, the poor performance of the Japanese market is
evident, as it is the only market to have a negative return. As expected, all markets strongly
rejected the null hypothesis of normality and kurtosis.
Table 4.6 - Descriptive Statistics: Stock Index Returns for the full period
Country Mean Variance Skewness Kurtosis Jarque-Bera
Australia 0.00151** 0.00041 -0.21166** 3.38445** 421.72246**
Hong Kong 0.00195 0.00152 -0.96946** 8.45726** 2729.07022**
Indonesia 0.00259 0.00278 5.61822** 99.47763** 363299.578**
Japan -0.00074 0.00090 0.10695 1.35944** 68.65168**
South Korea 0.00050 0.00193 -0.00426 1.92855** 134.82717**
Malaysia 0.00129 0.00140 0.17016** 8.18368** 2431.95431**
New Zealand 0.00094 0.00065 0.27354** 7.4037** 1997.88645**
Philippines 0.00083 0.00173 0.11595 1.69941** 106.63966**
Singapore 0.00094 0.00079 -0.16393** 2.53159** 236.22112**
Taiwan 0.00105 0.00231 -0.5785** 1.92253** 182.51053**
Thailand 0.00085 0.00198 0.20676** 3.5089** 452.52362**
US 0.00169** 0.00050 -0.18182** 2.99853** 330.72471**
Notes: The full period begins 1
st
January 1988 and finishes 7
th
September 2004.
The returns are weekly continuously compounded returns calculated as per (1). **
denotes significance at the 5% level.
Significant skewness was found in all cases except Japan, South Korea and the Philippines.
The uncharacteristic performance of the Japanese market during the sample period compared
to the other countries examined is likely to reduce the possibility of it being linked with other
markets
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
68
CHAPTER 5: EMPIRICAL RESUTS AND FINDINGS
5.1 Introduction
This chapter outlines the major empirical results of this thesis. Conclusions are made from the
results in reference to the hypotheses and propositions outlined in Chapter 3. The chapter is
divided into several major sections. Section 5.2 address the results from the cointegration
analysis, while Section 5.3 presents the findings obtained from the persistent profile analysis.
5.4 and 5.5 cover the findings regarding the generalized impulse response functions, and
block exogeneity tests. Lastly, section 5.7 contains the results from the regression analysis
concerning the factors driving market linkage.
5.2 Cointegration Analysis
This section is broken into; Section 5.2.1 examines the bivariate and multivariate tests for
cointegration on the full sample period. 5.2.2 contains, the results pertaining to the recursive
bivariate cointegration analysis.
5.2.1 Bivariate and Multivariate Tests for Cointegration for the full sample period
The tests for stationarity are contained in panel A and B of Table 4.1. For the full sample
period, all twelve markets are found to be I (1), thus satisfying the first condition required for
cointegration. Following the procedure outlined in section 3.3.1, the first step is to choose an
appropriate lag length (p). The Sims likelihood ratio test, AIC or SBC are be used to
determine the optimal lag length. Based on a mixture of evidence from the selection criteria
and previous empirical findings, the bivariate cointegration tests are completed with 10, 15,
20, 30 and 52 lags in the test VAR to ensure robustness. The results from all five-lag
structures are displayed in Table 5.1. All trace statistics were deflated by the small sample
bias correction factor as per the discussion in section 3.3.1. Due to the test VAR containing
twelve equations for the multivariate tests, the degrees of freedom rapidly decline. Therefore,
lag lengths of 5, 8, 10, 15 and 20 are employed. The use of longer lag lengths in cointegration
tests is preferred for two reasons. The first reason as mentioned above and in Chapter 5, serial
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
69
correlation between variables can significantly affect the performance of the JJ procedure.
Therefore, the order of the VAR should be increased until serial correlation is removed from
the residual series. Secondly, because cointegration tests are used to identify the possibility of
long run relationships, an excessively short lag length is inappropriate. Furthermore, uses of
relatively short lag lengths decrease the probability of significant cointegration being
discovered. As discussed in Chapter 2, the drawback of over-parameterization is the loss of
discriminatory power in the JJ procedure. This issue is not relevant to this thesis, due to the
full sample containing 871 observations for each series. Therefore, even with a lag length of
fifty-two weeks, some 819 observations per series are still available for testing.
For Australia, some interesting results were obtained. The first was that under all five-lag
structures no strong linkage was found to exist with Hong Kong, Malaysia, New Zealand, the
Philippines, Thailand or the US
43
. Significant cointegration with Indonesia was found in all
cases except when 52 lags were included. The conclusion that could be drawn from this result
is that Australia affects the short-term return in the Indonesian market over a long run time
horizon. When p = 30 and 52, Australia, exhibits significant linkage with Japan. It would be
thought that movements in the Japanese market affect Australia in the long run. At a lag,
length of 10 and 52 significant cointegration is discovered between Taiwan and Australia.
While at the 52-week lag, Korea and Singapore were found to be connected to Australia. This
thesis proposes the role of Australia in the Pacific-Basin is one of a market clearer, that is it
returns the system to equilibrium. Movements in Japan, Hong Kong and the US are
transmitted by Korea, Singapore and Taiwan into Australia.
Cointegration is found at p = 10, 15 and 30 between Hong Kong and Indonesia. It can be
safely assumed that Indonesia is influenced by Hong Kong in the long run. South Korea and
Singapore display marginal linkage
44
with Hong Kong. Indonesia is found to take its lead
from multiple markets. Indonesia shows strong linkage with Japan, Korea, New Zealand,
Singapore, Taiwan and the US at lengths, 10, 15, 20 and 30. Fro all situations besides the
relationship with New Zealand it is supposed that the other five markets exert influence of
43
With 10 lags, a cointegrating relationship between Australia and the US was found to exist at the 10%
significance level. While a cointegrating relationship was found between New Zealand and Australia at the 10%
significance level, when p = 52.
44
Marginal linkage refers to significance at the 10% level.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
70
Indonesia. There also appears to be weak linkage with Malaysia. Besides the results discussed
above, Japan also demonstrates weak significance with New Zealand and Singapore.
From the results obtained, South Korea appears to be one the most exogenous markets in the
Pacific-Basin. There is substantial support for the notion that Australia, Indonesia, Hong
Kong, Malaysia, New Zealand, Singapore, Taiwan, and the US are all linked with Korea in
the long run. A believable story is that shocks from Hong Kong, Singapore and the US flow
into Korea and it then transmits them to Australia, Indonesia and New Zealand. The relation
between Taiwan and Korea is not so clear. Even looking at the evidence from the GVD, both
markets seem to be important to one another.
The Malaysian market was found to have a robust relationship with Korea and Taiwan and a
less well developed relations with the US and Indonesia. Disturbances in Hong Kong, Japan
and Singapore are likely to flow through Korea and Taiwan into Malaysia then to Indonesia.
Malaysias dependence on investment from the US is likely to influence the relationship. The
behaviour of New Zealand in the Pacific-Basin is not unexpected. As discussed above New
Zealand is affected by the Indonesian and Japanese market, while the New Zealand does not
influence any other market in the Pacific-Basin.
The Philippines is likely to be endogenous because of the relative lack of development. The
bivariate cointegration tests support this proposition. Only in one case with Taiwan do the
Philippines display market linkage with any other markets in the PB. Like South Korea,
Singapore proves to be an exogenous market. As previously, discussed Singapore is proposed
to affect the Australian and Indonesian markets, while it is influenced by Hong Kong and
Japan. This thesis suggests that Singapore as a conduit to link Hong Kong and Japan with
Korea. Furthermore, Taiwan and the US appear to be linked in a less significant fashion.
Taiwan assumes a similar position to that of South Korea as a transmitter in the Pacific-
Basin. As discussed in the section above, Taiwan is found to be linked to Australia, Indonesia,
Korea, Malaysia, the Philippines and Singapore. However, the other interesting characteristic
of Taiwan is its strong relationship with the US. From the evidence presented above, Taiwan
stands as the only market in the PB that exhibits robust ties with the US. This is likely to have
considerable consequences on the importance of Taiwan in the PB.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
71
Following the procedure developed by Husted (1992) and outlined in Darrat and Zhong
(2004), the tests of cointegration are recomputed with the inclusion of two dummy variables.
The EA crisis dummy takes a value of 1 for the period October 17, 1997 to 31 December
1998 and 0 otherwise
45
. The choice of this period is based on the work of Forbes and Rigobon
(2002) whom use a similar time horizon. The second dummy takes a value of 1 for the four
weeks following September 11 2001 and 0 otherwise
46
. The choice of twenty-six weeks
derives from the findings of Hon et al (2004) whom find that correlation between the US and
Hong Kong, Japan, Korea, Australia, Singapore, Malaysia and Taiwan significantly increased
post-September 11 for the next six months. Three different specifications are estimated. The
first includes both the dummies, while the other two excludes one dummy and re-tests, for all
tests p = 52. Panel B of Table 5.2 contains the results. Caution must be employed when
interpreting the result of these tests. The inclusion of a stationary variable (i.e. the dummy
variables) is likely to lead to the true cointegrating relationship being overstated. Furthermore,
the Osterwald-Lenum critical values may not be applicable. This thesis takes a conservative
stance and only concludes that the event has caused an increase in linkage if the increase is
considerable.
The results show that the inclusion of the EA crisis dummy appreciably intensifies the
cointegrating relationship
47
. Figure 5.2a plots the trace statistics from the bivariate tests with
and without the dummy variable. The trace statistics have been normalized by the 95%
Osterwald-Lenum critical values, such that a value greater than 1 indicates significance at the
95% level. The increase in the strength of the cointegrating relationship produced after
including the dummy for the EA crisis period is so intense that even from a conservative
standpoint it can be concluded that the EA crisis caused a once off structural change in the
equity markets of the PB.
45
Dummy variables for four and twelve weeks post October 17 were also tested. The results did not display any
material difference and therefore are not reported. However, they are available upon request.
46
Dummy variables for twenty-six and twelve weeks post September 11 were also tested. The results did not
display any material difference and therefore are not reported. However, they are available upon request.
47
All trace statistics were adjusted using the small sample bias correction factor.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
72
Figure 6.2a The impact of the East Asian crisis on market linkage
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
T
r
a
c
e
s
t
a
t
i
s
t
i
c
Trace with dummy Trace without dummy Normalised Critical Value
Out of the sixty-six pairs, the cointegration relationship increased for fifty-two. An interesting
result is that in seven of the cases when the relationship did not intensify Australia was one of
the markets included. This indicates the relatively small impact the EA crisis had upon the
Australian market and its standing in the PB. These findings provide further support for
hypothesis 3
A
and 3
B
. The effect of September 11 is presented in Figure 5.2b.
Figure 6.2b The impact of September 11 on market linkage
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
T
r
a
c
e
S
t
a
t
i
s
t
i
c
Trace with dummy Trace without dummy Normalised Critical Value
Considering the caution with which the results should be interpreted, the effect of September
11 on the PB is still note worthy. This graph though highlights the relatively mild impact of
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
73
September 11 on the PB compared to the EA. Finally, both dummy variables were
simultaneously included and the tests recomputed. Figure 5.3c can be seen below.
Figure 6.3c The impact of both the EA crisis and September 11 on
market linkage
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
Pair
T
r
a
c
e
S
t
a
t
i
s
t
i
c
Both dummies No Dummy Asian dummy Normalised Critical Value
The most obvious result from the graph above is that the increase in the intensity of the
cointegrating relationship is mainly attributable to the EA crisis and not September 11. The
results of the multivariate results can be viewed in table 5.3.1.
Table 5.3.1 Cointegration Tests for the presence of long run market linkage between the nations of
the Pacific-Basin
lag =5 lag = 8 lag =10
Null
Unadjusted
Trace
Adjusted
Trace
Unadjusted
Trace
Adjusted
Trace
Unadjusted
Trace
Adjusted
Trace
r = 0 399.58 372.05 403.81 375.99 421.97 392.90
r 1 324.83 302.45 324.23 301.89 327.20 304.66
r 2 257.35 239.62 252.38 234.99 263.95 245.77
r 3 194.22 180.84 198.79 185.10 212.08 197.47
r 4 145.04 135.05 151.83 141.37 166.87 155.37
r 5 112.58 104.82 114.84 106.93 121.96 113.56
r 6 81.31 75.71 80.00 74.49 87.93 81.87
r 7 58.06 54.06 57.21 53.27 61.10 56.89
r 8 39.49 36.77 41.74 38.86 42.32 39.40
r 9 25.05 23.32 26.65 24.81 26.63 24.80
r 10 12.98 12.09 13.91 12.95 14.84 13.82
r 11 6.16 5.74 6.57 6.12 6.76 6.29
lag =15 lag = 20 Critical Values
Null
Unadjusted
Trace
Adjusted
Trace
Unadjusted
Trace
Adjusted
Trace
95% 90%
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
74
r = 0 410.17 381.91 439.14 408.89 340.39 333.26
r 1 319.27 297.28 349.47 325.40 291.40 281.63
r 2 249.57 232.38 272.91 254.11 244.15 237.35
r 3 199.99 186.21 213.97 199.23 202.92 196.66
r 4 155.08 144.40 171.05 159.27 165.58 159.74
r 5 117.28 109.20 130.02 121.06 131.70 126.71
r 6 87.70 81.66 100.60 93.67 102.14 97.17
r 7 63.67 59.28 74.87 69.71 76.07 71.66
r 8 43.15 40.18 52.15 48.56 53.12 49.91
r 9 29.39 27.37 35.88 33.41 34.91 31.88
r 10 17.36 16.16 21.85 20.34 19.96 17.79
r 11 8.41 7.83 9.54 8.88 9.24 7.50
Notes: This table reports the Johansen-Juselius (1992) trace statistics for the multivariate
cointegration tests conducted on the log of stock market indexes of all twelve markets included in the
system. Weekly data is used. The full period covers 5/01/1988 to 7/09/2004. Intercepts are included in
the estimated cointegration vectors. The small sample correction factor suggested by Cheung and Lai
(1993) is applied to all the trace statistics. A lag length of ten was used for all tests. Other lag lengths
were tested. The differences between the results were immaterial. The results are available upon
request. The 95% and 90% critical values are obtained from Osterwald-Lenum (1992). An * indicates
rejection of the null of no cointegration at the 10% level, while an ** indicates rejection at the 5%
significance level.
The tests seem to indicate the presence of two to three cointegrating vectors depending on
which lag length is believed. To determine whether Japan or the US is responsible for driving
cointegration in the PB, the tests were re-computed with Japan excluded, then the US
excluded. As can be seen from table 5.3.2 panel A, the trace statistic from when Japan is
excluded is considerably higher than when the US is excluded.
Table 5.3.2 Cointegration Test Results for the presence of Long-Run Equity Market
Linkage in the Pacific-Basin on the full period (January 5, 1998 - September 7, 2004)
Panel A. The effect of Japan and the US on market linkage
Null
Trace
statistic for
all markets
Trace statistic for
when Japan is
excluded
Trace statistic for when
the US is excluded
Critical Values
95% 90%
r = 0 392.90** NA NA 340.39 333.26
r 1 304.66** 310.6807** 278.2973 291.40 281.63
r 2 245.77** 230.937 218.098 244.15 237.35
r 3 197.47* 176.3436 167.0039 202.92 196.66
r 4 155.37 137.2844 129.8755 165.58 159.74
r 5 113.56 102.1726 95.92551 131.70 126.71
r 6 81.87 72.35816 69.53845 102.14 97.17
r 7 56.89 52.9183 49.07055 76.07 71.66
r 8 39.40 35.3941 33.44207 53.12 49.91
r 9 24.80 21.32065 20.73436 34.91 31.88
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
75
r 10 13.82 10.2989 9.658377 19.96 17.79
r 11 6.29 4.455487 3.871029 9.24 7.50
Notes: This table reports the Johansen-Juselius (1992) trace statistics for the
multivariate cointegration tests conducted on the log of stock market indexes of all
twelve markets included in the system. Weekly data is used. The full period covers
5/01/1988 to 7/09/2004. Intercepts are included in the estimated cointegration vectors.
The small sample correction factor suggested by Cheung and Lai (1993) is applied to
all the trace statistics. A lag length of ten was used for all tests. Other lag lengths were
tested. The differences between the results were immaterial. The results are available
upon request. The 95% and 90% critical values are obtained from Osterwald-Lenum
(1992). An * indicates rejection of the null of no cointegration at the 10% level, while an
** indicates rejection at the 5% significance level.
It must be noted that this does not provide concrete evidence to support the proposition that
the US influences the cointegrating relation more than Japan; however, it does provide
persuasive proof to suggest it does. The effect of the EA crisis and September 11 are also
considered in the context of the multivariate setting. The effect of the EA crisis is found to
have a far greater impact upon the level of cointegration than did September 11. It can be seen
that the inclusion of the Asian dummy leads to the identification of two extra cointegrating
vectors compared to the result from the September 11 dummy. This evidence provides
enough proof is from a conservative standpoint to conclude that the EA crisis affected the
long run market linkage between the countries of the PB. Such compelling findings are not
seen when only the September 11 dummy is included.
Panel B. The impact of the East Asian crisis and September 11 on market linkage
Null
Trace statistic
for no dummy
variables
Trace statistic for
when the EA crisis
dummy is included
Trace statistic for
when the
September 11
dummy is included
Trace statistic for
when both dummy
variables are
included
Critical Values
95% 90%
r = 0 392.90** 386.23** 371.42** 387.54** 340.39 333.26
r 1 304.66** 307.92** 289.44* 308.20** 291.4 281.63
r 2 245.77** 253.04** 233.67 253.14** 244.15 237.35
r 3 197.47* 199.28* 183.62 199.44* 202.92 196.66
r 4 155.37 153.86 139.82 154.01 165.58 159.74
r 5 113.56 116.00 106.82 116.04 131.7 126.71
r 6 81.87 83.77 76.81 84.88 102.14 97.17
r 7 56.89 57.51 54.95 57.79 76.07 71.66
r 8 39.4 39.68 36.99 39.82 53.12 49.91
r 9 24.8 22.41 22.65 22.54 34.91 31.88
r 10 13.82 12.08 11.77 12.18 19.96 17.79
r 11 6.29 5.69 5.43 5.88 9.24 7.5
Notes: This table reports the Johansen-Juselius (1992) trace statistics for the multivariate cointegration tests
conducted on the log of stock market indexes of all twelve markets included in the system. Weekly data is
used. The full period covers 5/01/1988 to 7/09/2004. Intercepts are included in the estimated cointegration
vectors. The small sample correction factor suggested by Cheung and Lai (1993) is applied to all the trace
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
76
statistics .A lag length of ten was used for all tests. Other lag lengths were tested. The differences between the
results were immaterial. The results are available upon request. The 95% and 90% critical values are obtained
from Osterwald-Lenum (1992). An * indicates rejection of the null of no cointegration at the 10% level, while an
** indicates rejection at the 5% significance level.
5.2.2 Bivariate Recursive Cointegration Analysis
The methodology for the recursive tests is outlined in section 3.3.1. Because there are 416
observations available per series, to ensure the credibility of the results, lags of 10, 30 and 52
were tested. Even when p = 52, there are still some 260 observations on which the test can be
completed, thus the test still maintains significant discriminatory power. The results from the
30 lag tests will be focused upon in the following discussion and all sixty-six graphs can be
seen in Figure 5.2.2 of the Appendix. Some quick explanatory notes about the graphs
presented, allow for easier interpretation. The first observation corresponds to the 2/01/1996
and finishes on the 7/09/2004, the EA crisis occurs around the 90 observation, while
September 11 is observation 298. The trace statistics have been adjusted for small sample bias
and have been standardized by the 95% critical value provided by Osterwald-Lenum (1992),
such that a valued in excess of 1 indicates significance at the 5% level. The abbreviations
used for the different markets are as follows: AU ! Australia, HK ! Hong Kong, ID !
Indonesia, JP ! Japan, KO ! Korea, MY ! Malaysia, NZ ! New Zealand, PP ! The
Philippines, SG ! Singapore, TA ! Taiwan, TH ! Thailand and US ! The United States.
Therefore, the graph TATH is the weekly recursive cointegration test between Taiwan and
Thailand.
Due to the intricate explanation required to properly describe the findings, only those
containing Australia will be discussed in detail, while other important results also deserve
attention. The overall result that can be taken away from all sixty-six graphs is that, the level
of market linkage is extremely volatile over time and can intensify or decline quickly.
Furthermore, without any doubt it can be concluded that the EA crisis and September 11 to a
lesser extent led to an increase in the level of market linkage. Additionally for some markets,
the fear concerning the Y2K bug caused an increase in market segmentation leading up to the
year 2000; however, the evidence seems to suggest that post 2000 markets quickly reverted to
displaying higher levels of market linkage. Lastly, some evidence emerges supporting the
notion that in the build up to the EA crisis, September 11 and the Y2K bug that the level of
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
77
linkage was frequently below its average levels. These conclusions opens up a completely
new question to be addressed by the literature and that is, are the effects of market
disturbances more long lived and heightened due to the transmission channels between
markets not operating at full efficiency. This question is far to complex and broad to be
covered in this study, but provides and interesting platform for further study to be completed
upon.
The results pertaining to the tests including Australia and the other eleven markets illustrate
some interesting findings. Australias relationship with Hong Kong is only significant at very
few times over the eight-year period. In the build up to the EA crisis, the level of linkage
increased substantially to its highest before rapidly declining prior to the EA crisis. Following
the crisis, the relationship increased dramatically and attained significance several times in the
following months. In the lead up to the year 2000, the markets displayed a visible lack of
linkage. However, almost immediately into the new year the relationship experienced a
considerable upsurge. This evidence tends to suggest that the fear surrounding the Y2K bug
had a noticeable impact upon markets behaviour. The events of September 11 had an almost
identical effect upon the level of market linkage as the EA crisis did. Pre- September 11,
linkage was relatively low; however almost immediately following the attacks the strength of
the relationship nearly doubled and was significant for an extended period of time (2-3
months). Post- September 11 the intensity declined dramatically, but in the last year has once
again experienced a sizeable increase and a return to significance.
The results for Australia and Indonesia and Australia and Japan, both demonstrate an upward
trend. Prior to the year 2000, Indonesia displayed no significant long run market linkage with
Australia. In the post 2000 period, there has been considerably strong linkage between the
two markets, with a significant relationship being evident constantly for almost the last two
and a half years. For Japan, a similar relationship is found, instead though it appears that
September 11 acted as a catalyst behind increasing the level of association. This result is
mirrored by the GVD results that finds the amount of variance explained by Japan
considerable increase post- September 11. Significant linkage can be seen to have been
present for around the last 130 weeks between Australia and Japan.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
78
The importance of the Korean market in regard to Australia can be seen to have increased
gradually over the last eight years. The three events (EA crisis, y2k and September 11) all
seem to result in considerable spikes in the cointegrating relationship. The relationship
displays brief significance post y2k, but quickly deteriorated, before the events of September
11 sparked a marked increase in the level of linkage. The connection continued to be
significant to the end of the sample period, with only small periods of insignificance. The
GVD results exhibit a comparable pattern, showing an increase of 7.5% from the pre- EA
period to the post- September 11 sample.
Malaysia and Australia do not demonstrate any long periods of significant linkage over the
past eight years besides the last couple of months. Again, the three events discussed above all
produce noticeable spikes in the cointegrating relationship, at times to such an extent to cause
brief significance. The connection between Australia and New Zealand exhibits an
unexpected amount of volatility. For the first 250 observations, there is an obvious upward
trend in the relationship and shows sporadic significance over a 60-week period. However,
the decline was swift and by week 300, the intensity had more than halved. Since the decline,
September 11 caused a sharp rise but no significance, while in the last year there has been a
considerable reduction in the strength of the relationship. The GVD produces a result that
describes this relationship almost exactly. In the post- EA/pre September 11 period
Australia accounts for 13.19% of the forecast error variance in New Zealand, while New
Zealand explains 12.71% of the Australian variance. In the post September period Australia
accounts for only 7% in new Zealand and New Zealand a considerably smaller 4.43%.
Immediately preceding the EA crisis the level of association between Australia and the
Philippines is almost negligible. Like other results, the EA crisis caused the relationship to
undergo a considerable change. Even though the spike was only temporary, the underlying
level of linkage remained quite high. Almost simultaneously, following the new year in 2000,
the level of linkage tripled and fir the first time displayed significance. From the time since,
the relationship has continued to expand and a considerably high concentration of linkage is
apparent. On the back of extensive and robust growth, it is not surprising that Australia and
Singapore have become linked to one another. The relationship can best be described as
spasmodic expansion followed by an immediate decline then a period of static followed by
another jump in the intensity and such the cycle continues. The driving forces behind the
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
79
spasmodic increases are the three major events as described above. However, only in the past
year has a significant relationship been realized. The GVD results provide further
corroborating evidence to support the gradual increase in linkage between the two markets.
Australia and Taiwan have been significantly linked to one another briefly at several times
over the sample examined. The first period followed the EA crisis, while the other two
periods of significance can be directly attributed to the passing of the y2k bug and September
11. Thailand and Australia displays an almost identical relationship as to Australia and
Taiwan, that is, the three periods over significance are driven by the three events already
mentioned. In the last year, the relationship for both markets and Australia has remained
robust, but still insignificant.
Lastly, the result between Australia and the US as per expectations has been remarkably
strong during the sample period. The strong association with the US began in early 2000 as
the fears surrounding the Y2K bug subsided. Almost immediately, the relationship
experienced a dramatic appreciation. For the next three and a half years the US and Australia
were significantly linked to one another. It has only been since September 2003 that
significance has been lost. An interesting observation is the relatively small affect of
September 11 compared to the EA crisis and Y2K bug.
5.3 Persistent Profile Analysis
Section 5.4 outlines the methodology required to estimate persistent profiles. The first
condition is to ensure the variables are integrated of the same order. The result of ADF and
PP tests for stationarity can be seen in Table 4.1. For all twelve markets, the null hypothesis
of non-stationarity for stock prices could not be rejected for the full sample period as well as
the three sub periods. The null hypothesis of non-stationarity is rejected in all cases for all
periods for the differenced data. Thus, the condition of being integrated of the same order is
satisfied.
As outlined in section 3.3.1, the JJ test for cointegration is performed. Table 5.3 displays the
results from the multivariate JJ cointegration tests conducted on the full sample. The presence
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
80
Table 5.3 Cointegration Test Results for the presence of Long-
Run Equity Market Linakge in the Pacific-Basin
Null Adjusted Trace Statistic Critical Value
full
period pre EA post EA 95% 90%
r = 0 392.90** 360.71** 388.22** 340.39 333.26
r 1 304.66** 292.6** 304.63** 291.4 281.63
r 2 245.77** 229.7932 236.956 244.15 237.35
r 3 197.47* 182.4899 184.5456 202.92 196.66
r 4 155.37 144.3046 133.9664 165.58 159.74
r 5 113.56 108.9757 100.4413 131.7 126.71
r 6 81.87 79.86194 69.35261 102.14 97.17
r 7 56.89 56.06952 43.93166 76.07 71.66
r 8 39.4 35.89299 29.65278 53.12 49.91
r 9 24.8 20.8248 18.73039 34.91 31.88
r 10 13.82 12.37812 19.50358 19.96 17.79
r 11 6.29 4.494546 3.973298 9.24 7.5
of three cointegrating vectors
48
can be identified. Similar results are found for the pre and post
EA crisis period. Unlike the short run measures, the multivariate tests for cointegration
consume a considerable number of degrees of freedom. As a result, it makes the sample too
small to accurately conduct the tests on the three sub samples.
The results from the PP for the full period are contained in the Appendix in Figure 5.3a, 5.3d
and 5.3e. The cointegration tests identify three cointegrating vectors. An alternative
interpretation is that there are nine different stochastic processes in the Pacific-Basin. The
results show that near convergence is achieved after thirty weeks for vectors 1, 2 and 3. The
rate of convergence is very stable, implying a low level of volatility between markets.
Looking at Figure 5.3b, 5.3d and 5.3e the PP for the pre-East Asian crisis period can be seen.
The results show that cointegrating vectors obtains near convergence after 20 weeks.
Cointegrating vector two is faster to revert to equilibrium with over 95% of convergence
completed after thirty fifteen weeks. Like the results for the full period, convergence for the
pre-EA crisis period occurs steadily, thus indicating low volatility is the system.
Figures 5.3c, 5.3d and 5.3e show the PP for the post-EA period. The notable difference is the
in the speed of the convergence. Both vector 1 and 2 achieve over 95% of convergence within
48
Assuming a lag length of 10 is being employed
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
81
ten weeks. The increased speed in convergence indicates that the markets of the PB became
more efficient and linked to one another following the EA crisis.
5.4 Generalized Impulse Response Functions
Figures 5.4a 5.4g display the results from the GIR functions. As mentioned in earlier
sections generalized impulse functions show how quickly a market reacts and reverts to
equilibrium following a one standard deviation shock in another market. Fast convergence is
an indicator of strong market linkage. A two standard error bound can be observed on all
graphs of the GIR functions. This bound should be interpreted in the following manner. When
the both the bounds are in the positive or negative section of the graph, then it can be
concluded that the response by market A to a shock in market B is significant. The situation
when both bounds are not in the same region indicates an insignificant response. This can
indicate two vastly different situations. The first is that market A is not linked to market B
and therefore does not respond to the disturbance. The alternate interpretation is that the
response is so fast that it cannot be discerned using weekly returns. For the full period and
the three sub-samples, Japan and the US were shocked and the responses observed.
Figures 5.4a and 5.4b show the GIR functions for the full sample period. Looking at the effect
of a shock in the US, most markets exhibit significant reversion after the two-week mark.
Australia, Hong Kong, Japan, Korea, Malaysia, New Zealand and Taiwan display strong
linkage with the US. The conclusion arises from the speed at which the markets converge to
equilibrium. The markets are highly efficient with most of the convergence completed by the
second week and only an immaterial level of noise remaining. These results provide
supporting evidence for the bivariate cointegration tests. From table 5.2, significant
cointegration with the US is evident between Australia, Japan, Korea, Malaysia and Taiwan at
shorter lag lengths.
Singapore and Thailand seem to respond slower with significant adjustments still present after
two weeks. However, this result still reinforces the existence of a noticeable relationship. As
postulated in this thesis, both the Philippines and Indonesia take the longest to return to
equilibrium. The Philippines takes over three weeks, while Indonesia displays significant
responses in the fourth week following the initial shock.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
82
Markets in the PB appear to react differently to a shock originating from the Japanese market.
For all markets, the response to a shock originating form Japan is more quickly incorporated
than a shock from the US. The US displays the fastest convergence and the lowest volatility
with full convergence achieved in less than two weeks. . The results pertaining to Australia,
Hong Kong, Korea, Malaysia and New Zealand are not materially different from the
responses to a shock in the US. This finding provides evidence to support the proposition that
more developed markets are linked with one another. It might can argued that the Malaysian
economy is not as advanced as the other four markets discussed above, therefore how can this
finding be explained. The linkage between Malaysia and the US arises from the enormous
amount of capital investment US companies have contributed over the last thirty years. The
relationship with Japan extends from the dominance of the Japanese market in the Pacific-
Basin.
Singapore and Thailand exhibit faster reversion to a Japanese shock, considering all
significant adjustment occurs in less than two weeks. Both the Philippines and Indonesia,
display substantially faster convergence considering no significant corrections take place after
two weeks have passed. This evidence reveals that the markets of the Pacific-Basin
demonstrate a greater level of linkage with Japan than with the US.
The results for the pre-EA crisis period displayed in Figures 5.4c and 5.4d, uncovers some
intriguing results. Figure 5.4d shows the effect of a one standard deviation shock to the US
market. The major conclusions that can be reached from the impulse response functions are
that Hong Kong and Japan completely adjust in less than a week. Australia, Malaysia, New
Zealand, Singapore and Thailand also rank highly considering no significant reaction takes
place after week two. This result is an improvement over the full period result for Singapore
and Thailand. The Philippines displays markedly faster reversion, with only residual noise
present after week 3. No significant convergence takes place by Indonesia, Korea or Taiwan.
This finding is somewhat unexpected considering the strong cointegration results obtained for
Korea and Taiwan, but not for Indonesia that appears to be segmented from the US.
The impulse response functions generated by a shock to the Japanese market are presented in
Figure 5.4c. The results further provide evidence toward the suggestion that the Pacific-Basin
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
83
has a different relationship with Japan than it does with the US. Hong Kong, Malaysia, New
Zealand, Singapore, Thailand and the US all demonstrate an extremely high level of market
linkage with Japan considering no significant adjustment occurs past week one. Australia,
Korea and Taiwan follow suit and complete any necessary reversion by the end of week two.
In almost all cases, this marks a considerable improvement in the level of linkage compared to
the full period result. Indonesia and the Philippines do not display any significant adjustment
over all time horizons.
In the post-ea/pre-September 11 period the level of the speed of convergence for all markets
improves dramatically. The impulse response functions in relation to a shock in the US are
exhibited in Figure 5.4f. All markets besides Indonesia achieve full convergence in or less
than two weeks. Korea, Malaysia and Taiwan require special mention considering they show
no significant adjustments past week one. This is a noticeable improvement for Korea and
Taiwan whom in the pre- EA period did not display any significant response. Thus, it can be
argued that following the EA crisis, either US investors shifted their attention toward Korea
and Taiwan or vice versa. This result provides convincing evidence towards hypothesis 3
A
and 3
B
, showing that in the period following the EA crisis the level and nature of market
linkage underwent an appreciable change. In opposition to this Indonesia does not
significantly respond to the shock in the US. That is, it appears that following the EA crisis
that Indonesia became even further segmented from the US.
The result of a shock to the Japanese market during this time is similar to that of the US
except for two differences. The first is that the speed of convergence to the shock in Japan is
slightly faster than that originating from the US. Secondly, unlike in the US case, Indonesia
responds quickly and significantly to a deviation in the Japanese market. Phylaktis and
Ravazzolo (2002), who found that the EA crisis reduced the level of global integration but
increased regional (i.e. the Pacific-Basin) integration, support this result.
Finally, the post-September 11 findings are outlined in Figures 5.4g and 5.4h. The obvious
result is the distinct increase in the level of market linkage concerning both markets. When
the US is shocked most markets experience significant reversion in week 1 and only some
noise after. The only other result worthy of mention is the lack of meaningful response by
Indonesia following a disturbance in the US. Hon et al (2004) find that post September 11
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
84
Indonesia has one of the lowest level of correlation (below 0.13) with the US, while on
average correlation increased between the US and all other Pacific-Basin markets.
The GIR functions for a shock to the Japanese market in the post-ea/pre-September 11period
can be viewed in Figure 5.4g. New Zealand and Taiwan demonstrate the highest level of
market linkage with Japan. The continuing speed at which New Zealand reverts to
equilibrium following a disturbance provides considerable evident to support the proposition
that New Zealand takes its lead from Japan and not necessarily Australia. Australia, Hong
Kong, Korea, Malaysia, Singapore and the US also quickly converge back to equilibrium
following the shock to the Japanese market. The interesting fact to note is that both Indonesia
and the Philippines display significant adjustment for up to two weeks. This finding is in stark
contrast from the result obtained from when the US is shocked and indicates that Indonesia
shifted away from being linked with world markets and became more reliant upon nations in
the Pacific-Basin. Therefore, it can be concluded that following September 11 the level of
market linkage experienced a substantial improvement.
These results provide overwhelming evidence to support hypothesis 2 that is the markets of
the Pacific-Basin are linked to one another in a short run fashion. In addition, the results
obtained form the generalized impulse response functions help to validate the findings of the
bivariate cointegration tests, often finding a significant relationship in both the short and long-
run. Furthermore, it provides evidence as to the structure of market linkage both pre and post
market disturbances. The evidence clearly supports hypothesis 3
A
and 3
B
. That is, not only
has the relationship between markets in the PB altered over time, but also following market
turmoil the level and structure of market linkage undergoes an evolution. Building on the
point further it is possible that the increase in market linkage in the post-ea/pre-September 11
period led to the events of September 11 not having such a significant impact because markets
had become more efficient and therefore the shock passed through faster. Possibly a more
credible argument is that, as suggested by the evidence and past empirical work that the PB is
not as strongly linked with the US as it is with itself.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
85
5.5 Generalized Variance Decomposition
the generalized variance decomposition was completed using TSP. As discussed in section
3.3.3 the computations were completed for the full period and three sub-samples. The results
of the analysis can be observed in Panel A D of Table 5.5.
The results from all four periods provide some interesting findings. The results from the full
period show that Hong Kong, New Zealand, Singapore and the US strongly influence
Australia. Furthermore, at the 20-week horizon foreign markets explain 63% of the variance
in Australia. This evidence clearly displays the high level of market linkage between Australia
and the Pacific-Basin in the short run. Hong Kong is found to be even more exogenous than
Australia, with over 68% of its variance explained by other Pacific-Basin economies. As per
expectations, Australia is found to account for about 9% of the variance in Hong Kong, while
Singapore contributes around 12.75%. Malaysia, the Philippines and Thailand are also worthy
of mention, explaining 7.52%, 6.84% and 7.23% (at the 20-week horizon) respectively.
Indonesia is found to be highly endogenous accounting for 58% of its own variance. This
result complements the findings from the impulse response functions. Additional evidence to
support this conclusion comes from the fact that the Philippines contribute 40.5% to its own
variance.
On the other hand, Japan and the US are not as endogenous as would have been thought. At
the 20-week horizon, all other markets explain 54% of the variance for Japan and 55% for the
US. Several markets make material contributions to the variance of Japan; they include
Australia (6.63%), Hong Kong (6.89%), Korea (5.89%), Singapore (8.16%) and the US
(6.89%). For the US, Australia (10.84%), Hong Kong (7.47%), Japan (6.6%) and Singapore
(6.7%) are all important markets.
Malaysia and Singapore, like Australia and Hong Kong are highly exogenous, only
accounting for 37.75% and 31.19% of their own variances. Furthermore, the well-documented
relationship between Singapore and Malaysia is highlighted by the results. Singapore explains
10.94% of the variance in Malaysia, while Malaysia accounts for 8.67% of variance in
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
86
Singapore. In both markets, Hong Kong plays a critical role, while the contribution from
Thailand should not be underestimated
49
.
Korea and Taiwan display considerable endogeneity over the full period. Korea is responsible
for 45.59% of its variance, while the contribution of Taiwan to itself is an incredible 61.25%
after week 1 and 55.84% at week 20. Again, Hong Kong and Singapore are vital to both
markets, explaining some 9.34% and 8.43% in Korea and 5.89% and 6.04% in Taiwan
respectively.
The market of New Zealand is somewhat endogenous, while Thailand for a developing
market is quite exogenous. The only note worthy result pertaining to New Zealand is that
Australia is responsible for over 14% of the forecasted error variance. Hong Kong (9.21%),
Malaysia (8.92%), Singapore (8.13%) and Taiwan (11.19%) all considerably influence
Thailand.
The pre- EA results are similar to the generalized impulse response functions, that is, markets
are far more endogenous and not linked during this time. In terms of the GVD, it essentially
means that markets explain more of their own variance and less of each others. Australia
explains material levels of variance in Hong Kong (6.61%), Japan (5%), New Zealand
(13.14%) and the US (9.41%), while Hong Kong, Malaysia, new Zealand, Singapore and the
us are discussion all these levels are down compared to the full period results. Not
surprisingly, Australia accounts for more than 13% of the error variance for New Zealand.
Besides Australia, the only other country to explain any other material amount in New
Zealand is the US. The conclusion that can be drawn from these results is the effect of close
economic relationships on the level of market linkage. The four countries mentioned above
represent the major trading partners of one another in the pre-ea period.
The endogeneity of Indonesia is even further exacerbated in the pre EA sample. Even after
20 weeks, Indonesia still accounts for 69.26% of its own variance, while no other market
accounts for more than 5% and it explains no more than 3% in any other market. Similar to
Indonesia, Japan also demonstrates considerable independence. However, it is essential to
49
For Malaysia, Hong Kong explains 9.49%, while Thailand accounts for 9.22%. In the case of Singapore, Hong
Kong contributes a staggering 12.93% and Thailand a considerable 8.86%.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
87
note that during the period being examined that Japan at times was the largest equity market
in the world and therefore is expected to display a certain level of segmentation. This same
argument is also applicable to the US. However, unlike Japan significant contributions are
made by individual countries including Australia (9.41%), New Zealand (5.61%) and
Singapore (6.25%).
Both Korea and Taiwan display a distinct lack of integration, with each market contributing
over 59% of their own variance. The only country worthy of mention is Singapore, which has
the greatest level of explanatory power in both markets. The level of fragmentation restricts
the contributions made by Korea and Taiwan to other markets in the Pacific-Basin to below
5%.
Hong Kong is found to exert considerable influence over Malaysia, the Philippines, Singapore
and Thailand, while all four markets explain more than 5% of the forecasted error variance in
Hong Kong. In addition, it must be noted that the other four other markets are the largest
contributors to the variance decomposition of one another beside that explained by the
domestic market. The evidence supports the notion that these five markets form a bloc within
the PB.
As has been noted in the previous sections, the EA crisis caused a significant transformation
of relationships in the PB. This conclusion is further supported by the results outlined in table
5.5 panel C. The overriding finding is that markets explain considerably less of their own
variance and are explained noticeably more by foreign markets.
Some of the most compelling results from the post ea crisis include, the incredible change in
the Korean and Taiwan markets from being endogenous to being more exogenous than
several other markets. Korea established strong links with Australia, Hong Kong, Japan, the
Philippines, Singapore and most importantly the US. This result almost exactly mirrors the
relationship found to exist between Korea and the US from the impulse response functions
presented in section 5.4. Taiwan on the other hand demonstrates considerable influence over
Korea and Japan, while Hong Kong, Japan, Korea, the Philippines and Singapore all make
sizeable contributions towards the forecasted error variance of Taiwan.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
88
The proposed bloc between Hong Kong, Malaysia, the Philippines, Singapore and Thailand
seems to be weakened in the post EA period. The main change seems to be related to
Malaysia whose orientation moves away from the four markets towards Australia, Korea, new
Zealand and the US. However, the other four markets still display above average levels of
market linkage.
The level of independence shown by Japan and the US continued to decline, to such an extent
that they became more exogenous than other smaller markets in the PB. It is also interesting
to see that in the case of Japan that the US is the most important market, while for the US,
Japan ranks second behind Australia. Meanwhile, the strong relationship between Australia
and New Zealand continued to be evident.
The level of segmentation of Indonesia and the Philippines between the markets in the PB
decreased substantially. The level of integration of the Philippines increased to such an extent
that it is responsible for explaining more than 7% of the variation in 10 out of 11 other
markets. Indonesia does not display the same increase in market integration, but still
demonstrates significant improvement.
The other standout discovery is the reaction of Malaysia to the EA crisis. The amount of
variance explained by itself for the 1, 5 and 10-week horizon after a shock on average is
greater than that in the pre EA sample. Furthermore, as discussed above the Malaysian
market appeared to shift away from the group of markets including Hong Kong, the
Philippines, Singapore and Thailand and toward Korea and New Zealand. This could have
been the result of Malaysia introducing capital restrictions in the wake of the EA crisis, or a
structural shift in the economy or trading partners.
The post September 11 period highlights how the markets of the PB have become
increasingly dependent upon each other. For all 12 markets at the 20-week horizon the level
of forecast error variance in the domestic market, explained by the domestic market does not
exceed 24%. The credibility and significance of this result should not be underestimated. It
shows that the level of market linkage in the PB is not the result of a spurious correlation.
Instead, the increase in intensity can be traced back to a systematic reduction in barriers that
has led to immeasurable changes in the region. The results from this sample are extremely
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
89
interesting considering that this thesis uses data for up to September 7, 2004, and therefore is
likely to highly relevant to the current market environment. All the figures referred to in the
analysis can be observed in table 5.5 Panel D.
The composition of markets affecting Australia post September 11 is markedly different
from that from previous periods. The importance of Hong Kong, New Zealand, and the US
have declined,
50
while Japan, Korea, Singapore and Taiwan have assumed greater meaning
51
.
Hong Kong experienced a comparable shift in composition. Japan, Korea, Singapore and
Taiwan consolidated their positions as being markets of considerable importance to Hong
Kong, while Australia and the Philippines suffered a reduction in their level of influence
52
.
Following the events of September 11, a valid argument supposing that Indonesia is one of
the most exogenous markets in the PB. This conclusion appears to be in direct opposition
with the results from the GIR. However, a more careful inspection finds that both of the
results actually complement one another. The critical point to remember is that in the impulse
response functions only the Japanese and US markets were shocked. After examining the
figures from table 5.5 panel D, it is clear to see that the involvement of the US in Indonesia is
considerably lesser than that of Japan. This result supports the findings from the GIR
function, which found Indonesia to be linked with Japan, but not the US. Nearly identical
proof is found for the Philippines, that is, the variance decomposition shows a stronger
relationship with the US than with Japan, as per the GIR functions.
In the post-September 11 world, the effect of the US over markets in the PB is not as
influential as it once was. For all 11 cases, the level of forecast error variance explained by
the US does not surpass 10%, and only averages 5.94%. By contrast, the Japanese market
becomes highly influential, it describes over 10% of the forecast error variance in Australia,
Hong Kong, Korea, Malaysia, Singapore and Taiwan and averages over 7.5% for all markets.
50
The decrease in the amount of forecasted error variance explained by New Zealand in the post September 11
sample is around 10%, while for the US and Hong Kong around 1% depreciation is recorded.
51
The increase in the contribution attributable to Japan is around 4.5%, Korea 1.5%, Singapore 5.2% and
Taiwan a substantial 6%.
52
The Philippines declined by more than 7%, while the decrease in Australias explanatory power was only
marginal
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
90
The markets of Korea and Taiwan continued to expand their influence over the PB region.
They both contain substantial explanatory for several markets and display a high level of
linkage between themselves. It is also interesting to note the connections concerning Taiwan
and Australia and the US. In the post-EA period, neither market exerted material influence
over Taiwan. However, post- September 11 and the relationship dramatically changed with
Australia accounting for 11.31% and the US 8.16% respectively. The combination of Korea
and Taiwan represents one of if not the most influential pairing in the PB following
September 11. this is extremely intriguing considering their beginnings as closed endogenous
markets. The findings that have been discovered for Taiwan supports the decision by the
Taiwan government to focus its attention toward the PB
53
.
Malaysias position in the PB following September 11, is one of a market that is considerably
influenced by the markets of Australia, Hong Kong, Japan, Singapore and Taiwan, while
displaying noteworthy explanatory power in the forecast error variance of Japan, Singapore,
Taiwan and the US. For New Zealand, September 11 greatly affected its relationships with
markets in the PB. The major role filled by Australia in the past was replaced by a
combination of Hong Kong, the Philippines, Singapore, Taiwan and the US.
Since the events of September 11, Singapore has continued to mature in its role as a market
leader in the PB. Singapore is found to be the second most exogenous market (only
marginally trailing Hong Kong) and one that accounts for substantial proportions of the
forecast error variance in nearly all markets, averaging of 9.5% of the variance for all 11 other
markets. Unexpectedly, the lowest level variance provided by Singapore is for Malaysia.
However, this result should not be that surprising for the following reason. Prior to the EA
crisis, the linkage between the markets was the strongest between any of the other 66 pairs.
Through the early-mid 1990s, a number of factors changed that drove a wedge between the
two. As the Malaysian economy became more sophisticated, it depended less upon Singapore,
subsequently their linking relationship diminished. The EA crisis then swept into Malaysia
and forced numerous reforms. The reforms attempted to restrict the level of integration of
Malaysia with the world. In addition, the level of market linkage has never reached the levels
prior to the EA crisis.
53
This is actually discussed in section 3.2
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
91
When the results from the GVD are examined in an overall context, a credible and convincing
set of conclusions can be drawn. The first refers to H
2
, which postulates the existence of
market linkage in the short run. After conducting the four sets of tests, it can clearly be
observed that the degree of market linkage in the PB over the short-run is exceedingly high
(especially in the post September 11 period). Further evidence from the findings allow the
conclusion that the linking relationship has altered through time and once more, has
experienced an appreciable increase following a period of market turmoil.
5.6 Block Exogeneity Tests
Four sets of block exogeneity tests were completed, the results of which can be seen in Panel
A D of table 5.6. As discussed in previous sections, block exogeneity tests are just the
extension of bivariate Granger causality tests to a multivariate system. The presence of
Granger causality can be interpreted two different ways. The first is that if market A causes
market B it can be said that market A leads market B. The alternative explanation is that if
market A causes market B, then it can be concluded that market B is efficient as it reacts to all
available market information, including that of market A. The first option is likely to apply
when if the US is found to cause Thailand, while the second explanation is more likely if
Thailand is found to cause the US.
The main results that come from the exogeneity tests are that:
1. In the full period the smaller less developed markets of Indonesia, Malaysia, New
Zealand, the Philippines and Thailand all have returns caused by several different
markets. Also for all the countries mentioned above when all markets are excluded
from the equation the evidence suggests that the Pacific-Basin as a whole holds
information for these markets. It is also interesting to note that Japan, Korea and
Taiwan are caused by some markets. However, considering the size of the Japanese
market it is likely that these countries arent causing returns in Japan, rather than
Japan responding to the news from those markets.
2. In the pre EA period the very few cases of causality does not warrant discussion,
except for one surprising result that finds the US is significantly affected by the
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
92
Pacific-Basin as a whole. This is likely to mean that in this period the US was
sensitive to information originating from the Pacific-Basin
3. Japan again is affected by several markets in the post EA similar to how it was in
the full period. Except this time Australia and the US are found to cause returns in
Japan. Even though Australia is a large developed market it would still is thought
that it does not cause returns in Japan. However, it would be argued that the US
does lead Japan in the post EA period. The other interesting point to make is that
seems that Indonesia is significantly influenced by Australia, Korea, Malaysia and
New Zealand.
4. The most interesting results from the block exogeneity results come from the post
September 11 period. Australia, Hong Kong, Japan and the US are all caused by
much smaller markets that just arent big enough to have a material effect over these
four markets. However, what it dies indicate is that Australia, Hong Kong, Japan
and the US have become more efficient post September 11 such that they react to
sources from information even from the smaller markets.
5.7 Regression Analysis
The cross sectional analysis from the OLS model is contained in panel A D of table 5.7.
Four different dependent variables are tested in the OLS model. Trace1 is the average trace
statistic from the weekly recursive tests with 10 lags, while Trace2 is from the test with 30
lags. Eigen1 and Eigen2 are the eigen values from the respective tests. For the probit model,
the values are restricted to take a value of either 1 or 0. From the tests for when p = 10,
Australia and Hong Kong was found to be cointegrated 0 times. The average number of
cointegrating vectors discovered over the 66 pairs was 58.3. Therefore, in this situation the
pair between Australia and Hong Kong would get a value of 0. That is, a value of 1 is given
when the number of cointegrating vectors for a pair exceeds the average number of
cointegrating vectors over the 66 pairs. This method is used to compute Pr1 and Pr2 (is based
on tests when there are 30 lags).
In three out of four cases, the ASEAN dummy is positive and significant, while in two cases
the interest rate trace is found to be negative and significant. A negative coefficient for the
interest rate trace statistic is unexpected. Location is found to be highly significant in all four
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
93
tests, and displays positive correlation with market linkage instead of the predicted negative
relation. This result can be explained quite easily, after examining the data the longest
distances between markets always is the distance between themselves and the US. Since the
US is the largest market in the world, and as discovered in the previous section is open to
other markets, then it would be expected that there is high market linkage between the
countries of the PB and the US, thus biasing this variable upward.
The variable sdiff is found to be strongly negative in all the regressions. The discussion
regarding the construction of this variable in chapter 4, predicted that a negative relationship
should be discovered. The turnover ratio is also found to be significant and negative as
predicted in an earlier section. The size dummies, display mixed results. From a theoretical
perspective the dummies both are as expected. That is, the large dummy has a positive
(although not significant) coefficient while the small group is negative. The results from these
variables demonstrate the importance of market development regarding market linkage. The
OECD dummy was found not to be significant in any model and appears to not be fitting with
the other variables. Therefore, in the multivariate tests it was not included.
Unexpectedly the industrial production trace is found to be negative and only significant in
one regression. This result is likely for the inability of the proxy to properly capture the
information it was supposed to. Keeping in mind that this section is highly experimental then
any results serve as an interesting platform for further work to launch from. In contrast to the
contagion hypothesis the volatility between markets is likely to cause a decline in the level of
market linkage volatility gets higher. This indicates that investors will try to stay away from
markets that are characterized by high volatility. Only the probit results for the multivariate
model is discussed, because of the immaterial difference with the OLS results
54
. Six
regressions were conducted for pr1 and pr2, an example of the output is provided below:
Extract of table 5.7
pr1 intercept aseand exvol iptrace irtrace location vol ldum sdum
coefficient -3.010 1.956 -4.423 -0.044 0.059 0.000 885.111 0.857 -0.198
t-stat -2.124 2.763** -0.009 -1.504 1.509 2.940** 1.135 1.716 -0.390
std err 1.417 0.708 516.633 0.029 0.039 0.000 779.679 0.500 0.507
std dev 0.361 0.001 9.023 7.790 4670.089 0.001 0.422 0.422
% 0.707 -0.002 -0.395 0.459 1.214 0.474 0.362 -0.084
R 0.432
54
The results are available upon request.
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
94
pr1 intercept aseand exvol iptrace irdiff Location vol ldum sdum
coefficient -1.230 2.011 691.121 -0.015 -0.075 0.000 106.723 0.680 0.184
t-stat -1.199 2.843** 0.897 -0.606 -0.756 2.674** 0.156 1.458 0.400
std err 1.026 0.707 770.853 0.025 0.100 0.000 685.361 0.466 0.460
std dev 0.361 0.001 9.023 7.790 4670.089 0.001 0.422 0.422
% 0.726 0.387 -0.136 -0.586 0.782 0.057 0.287 0.078
R 0.408
Notes: The results above are produced from a probit regression where pr1 is the dependent variable
and the independent variables are the same as those described in section 3.5. The t-stats have 55
degrees of freedom and at the 95% level the critical value is 2.0040 and 1.673 at the 90% level.
The full results are contained in table 5.7. In both the OLS and probit models the ASEAN
dummy was found to be postivie and significant. This result is credible as trade and economic
integration is likely to substantially affect the level of market linkage. Excahnge rate volatility
is found to be significant in eight of the OLS regressions, but not in the probit model. The
sign is almost always found to be positive. This indicates that the level of linkage increases in
the face of higher exchange rate volatility. As argued in chapter 3, greater exchange rate
volatility is likely to result in greater levels of trade and capital flowing between countries.
The trace statistic from the cointegration tests on the industrial production statistics is found
not to be significant in any of the cases. Again, this relationship can be attributed to the proxy
for goods market linkage, that is, the trace statistic appears not to be an appropriate proxy.
From the OLS regression the interest differential was significant in the majority of cases and
negative in all cases. This result provides evidence to support the PVM proposition, that is, a
higher differential will lead to a greater differential in prices and therefore reduce the level of
market linkage between markets. Location like the ASEAN dummy is one of the most
significant variables. Again, the slope is always positive. However, as discussed above this
result can be easily accounted for. The large market dummy was significant and positive in a
number of cases. This result is as per the predictions of this thesis. That is, the larger markets
are linked to one another, but display lower levels of linkage with the smaller markets. The
small market dummy again was found to be significant in the bulk of cases and the coefficient
was negative. This result was also not surprising as it would be thought that the small markets
might be linked with a medium to large country, but not another developing market.
While it appears that the regression results are not persuasive, the adjusted R squared provides
evidence to the contrary. In most regressions (both the OLS and probit) the R squared
CHAPTER 5EMPIRICAL RESULTS AND FINDINGS
95
exceeded 40%, indicating that the factors may not be significant individually, they still have
considerable explanatory power. Lastly, examining the % statistic gives an indication of the
economic significance of the variable. Location is found to be the most economically
significant variable. The ASEAN dummy also was economically significant, while the
interest rate trace and interest rate differential were alos found to be economically significant.
The remainder of the results is too erratic to be able to accurately interpret. However, it must
be noted that the testing procedures and proxies employed in this section were completely
original and experimental. The magnitude of the R squared statistics does still indicate that
the model is capturing a considerable amount of variation in market linkage.
CHAPTER 6CONCLUSION
96
CHAPTER 6: Conclusion
6.1 Contribution
As per the discussion in section 1.3, this thesis aimed to add to the current academic literature
in three aspects:
1. Are the equity markets of the Pacific-Basin region linked to one another in either the
short or long run;
2. If so, has this relationship varied over time; and
3. Which variables have been responsible for markets becoming more integrated with
one another
The actual contribution made regarding the first proposition is the undeniable level of
evidence produced that unequivocally establishes that the equity markets of the Pacific-Basin
demonstrate a certain level of interaction in both the short and long run. The use of recursive
bivariate cointegration provides the most accurate and complete picture of the structure of the
market linkage in the Pacific-Basin constructed to date. Furthermore, the results from this
thesis are made even more compelling due to all test statistics being adjusted by a small
sample bias correction. Therefore, the results generated by this thesis are extremely robust
and allows credible conclusions to be drawn from them. . The use of three methods to
establish the short run market linkage ensures a complete analysis of the subject.
Like the first proposition, this thesis provides overwhelming evidence that the level of market
linkage varies through time. The recursive bivariate tests provides some of the most
persuasive evidence in regard to the current literature. The use of the dummy variable analysis
in regard to the Pacific-Basin markets has not previously been employed. The inclusion of
this methodology was validated by the results produced from it. The persistent profile
analysis showed that the speed at which prices in the Pacific-Basin converge back to
equilibrium is significantly higher in the post- EA crisis period. All three short run measures
for substantial evidence to suggest the relationship has altered over time and in fact grown
stronger.
The evidence presented in relation to the final proposition was not as strong as that provided
for propositions 1 or 2. However, it still did make several contributions that could be used to
CHAPTER 6CONCLUSION
97
guide future research. First off it must be noted that the testing procedure implemented in this
thesis is completely original and experimental. That is, no previous works have used trace
staitistics or eigenvalues to proxy for the level of market linkage. Secondly, the applicaiton of
a probit model has not previously been used or propsed in the context of marklet linkage.
While some of the independent variables in this study were influenced by other pieces fo
literature, ultimately they too are unique to this study. Considering, the level of
experimentation surrounding this final proposition the fact that outstanding results were not
obtained should not come as a surprise.
6.2 Limitations
While the section above outlined the strengths of the results obtained in this thesis, there still
exists ample room for improvement in this study. The first and probably most obvious
limitation of this study was using weekly prices over a sixteen-year period. After completing
this thesis, it has been made obvious that when considering cointegration weekly data is
acceptable, but daily data is preferred. The use of daily data would have enabled an even more
robust examination of market linkage in the Pacific-Basin. The effect of having a small
sample severely limited the use of multivariate cointegration tests.
While this thesis presented a number of propositions regarding the direction of the
cointegration between markets it is not formally tested by imposing restrictions upon the
cointegrating vector. A detailed examination on the cointegrating vectors would be quite
interesting.
This thesis did not explicitly consider the impact of other major market disturbances, such as
the Russian default, both Gulf Wars, or effect of the Y2K bug. It is possible that it was one
of these events driving the results rather than those presented in this thesis. Furthermore, the
results pertaining to September 11 could be underestimated because there is only three years
of data available.
There are a number of other variables that could be employed in the final analysis that could
be more important than the variables proposed in this study. Other variables that could be
CHAPTER 6CONCLUSION
98
included are trade and capital flow statistics, sovereign debt ratings, level of unemployment or
inflation. These are also many others could be included.
6.3 Further Research
The area of market linkage has been extensively studied in the current literature; however
there still exist a number of areas that require further examination. The first is the inability of
this thesis and the current literature to explain the level of market linkage between national
equity indices. The question as to what causes market linkage is an important question and
deserves further attention.
Some of the limitations discussed above also open up new avenues for future research. The
identification of the cointegrating vectors would constitute a significant development in the
field. While the use of OLS and probit is adequate, there are other more robust methods such
as GMM that could be employed.
There exists a unique chance at the moment to be able to study the integration of the Chinese
market into the world economy. Considering, the size and importance of China, it would be
interesting to see how it merges into the Pacific-Basin.
99
REFERENCES
Ammer,J., Mei,J. (1996) Measuring International Economic Linkages with Stock Market
Data, The Journal of Finance, 51, 1743-1763
Bekaert, G., Harvey, C.R., 1995. Time-varying world market integration. Journal of Finance
50, 403444.
Bekaert, G., Harvey, C.R., 1997. Emerging equity market volatility. Journal of Financial
Economics 43, 29-77.
Bekaert, G., Harvey, C.R., 1998.Capital .flows and the behavior of emerging market equity
returns. Unpublished Working Paper 6669.National Bureau of Economic Research,
Cambridge, MA.
Bessler, D.A., Yang, J., 2003. The structure of interdependence in international stock markets.
Journal of International Money and Finance 22, 261287.
Blackman, S.C., Holden, K., Thomas, W.A., 1994. Long-term relationships between
international share prices. Applied Financial Economics 4, 297304.
Bracker, K., Docking, D.S., Koch, P.D., 1999. Economic determinants of evolution in
international stock market integration. Journal of Empirical Finance 6, 127.
Chan, K.C., Gup, B.E., Pan, M-S., 1992. An empirical analysis of stock prices in major Asian
markets and the United States. The Financial Review 27, 289307.
Cheung, Y.-W., Lai, K.S., 1993. Finite-sample sizes of Johansens likelihood ratio tests for
cointegration. Oxford Bulletin of Economics and Statistics 55, 313328.
Chung, P.J., Liu, D.J., 1994. Common stochastic trends in Pacific Rim stock markets.
Quarterly Review of Economics and Finance 34, 241259.
Connolly, R., Wang, A. (2003) International Equity Market Comovements: Economic
Fundamentals or contagion?, Pacific-Basin Finance Journal, 11, 23-43
Darrat,A., Zhong,M. (2003) Equity Market Linkage and Multinational Trade Accords: The
Case of NAFTA, Working Paper, Forthcoming in The Journal of International Finance and
Money
De Fusco, R.A., Geppert, J.M., Tsetsekos, G.P., 1996. Long-run diversification potential in
emerging stock markets. The Financial Review 31, 343363.
Dekker, A., Sen, K.and Young, M. (2001) Equity market in the Asia Pacific region: A
comparison of the orthogonalized and generalized VAR approaches, Global Finance Journal,
12, 1-33.
100
Dickey, D.A., Fuller, W.A., 1981. Likelihood ratio statistics for autoregressive time series
with a unit root. Econometrica 49, 10571072.
Enders, W., 1995. Applied Econometric Time Series. Wiley, New York.
Engle, R.F., Granger, C.W.J., 1987. Cointegration and error correction: representation
estimation, and testing. Econometrica 55, 251276.
Eun, C., Shim, S., 1989. International transmission of stock market movements. Journal of
Financial and Quantitative Analysis 24, 241256.
Forbes, K.J., Rigobon, R., 2002. No contagion, only interdependence: measuring stock
market comovements. Journal of Finance 57, 2223-2261.
Ghosh, A., Saidi, R., Johnson, K.H., 1999. Who moves the Asia-Pacific stock marketsUS
or Japan. Financial Review 34, 159169.
Gonzalo, J., 1994. Five alternative methods of estimating long run equilibrium relationships.
Journal of Econometrics 60, 203233.
Granger, C.W.J., 1986. Developments in the study of cointegrated economic variables.
Oxford Bulletin of Economics and Statistics 48, 213228.
Hansen, H., Johansen, S., 1993. Recursive Estimation in Cointegrated Models. Institute of
Mathematical Statistics Preprint No. 1, January, University of Copenhagen.
Hendrik, H., Juselius, K., 1995. Cats in Rats, Cointegration analysis of time series. Estima,
Evanston, IL.
Hon, M., Strauss, J., Yong, S. (2004) Contagion in Financial Markets after September 11:
Myth or Reality?, The Journal of Financial Research, XXVII, 95-114
Hung, B., Cheung, Y., 1995. Interdependence of Asian emerging equity markets. Journal of
Business, Finance and Accounting 22, 281-288.
Husted, S. 1992. The emerging U.S. current account deficit in the 1980s: A cointegration
analysis. Review of Economics and Statistics 74, 159-166.
Johansen, J., Juselius, K., 1993. Testing structural hypotheses in a multivariate cointegration
analysis of the PPP and the UIP for UK. Journal of Econometrics 53, 211244.
Johansen, S., 1991. Estimation and hypothesis testing of cointegration vectors in Gaussian
vector autoregression models. Econometrica 59, 15511580.
Johnson, R., Soenen, L. (2002). Asian economic integration and market comovement. The
Journal of Financial Research, XVV, 141-157
Karolyi, G.A., Stulz, R. M., 1996. Why do markets move together? An investigation of US-
Japan stock return comovements using ADRs. Journal of Finance 51, 951-986.
101
Kasa, K., 1992. Common stochastic trends in international stock markets. Journal of
Monetary Economics 29, 95124.
Kolari, J., Sutanto, P., Yang, J. (2004) On the stability of long-run relationships between
emerging and US stock markets. Journal of Multinational Financial Management 14, 233-248
Kwan, A.C.C., Sim, A.-B., Cotsomitis, J.A., 1995. The causal relationships between equity
indices on world exchanges. Applied Economics 27, 3337.
Leong, S., Felmingham, B. (2003) The interdependence of share markets in the developed
economies of East Asia. Pacific-Basin Finance Journal, 11, 219-237
Lessard, D.R., 1973. International portfolio diversification multivariate analysis for a group of
Latin American countries. Journal of Finance 28, 619633.
Levy, H., Sarnat, M., 1970. International diversification of investment portfolios. American
Economic Review 60, 668675.
Longin, F., Solnik, B., 1995. Is the correlation in international equity returns constant:
19601990? Journal of International Money and Finance 14, 326.
Lutkepohl, H., Reimers, H. E., 1992. Impulse response analysis of cointegrated systems.
Journal of Economic Dynamics and Control 16, 53-78.
Manning,N. (2002) Common trends and convergence? South East Asian equity markets,
1989-1999, Journal of International Money and Finance, 21, 183-202
Masih,R.and Masih,A.M.M.(2001)Long and short term dynamic causal transmission amongst
international stock markets, Journal of International Money and Finance,20 ,563-87.
Ng, A., 2000. Volatility spillover effects from Japan and the US to the Pacific-Basin. Journal
of International Money and Finance 19, 207-233.
Osterwald-Lenum, M., 1992. A note with quantiles of the asymptotic distribution of the
maximum likelihood of cointegration rank statistics. Oxford Bulletin of Economics and
Statistics 54, 461471.
Pesaran, M.H., Pierse, R.G., Lee, K.C., 1993. Persistence, cointegration, and aggregation: a
disaggregate analysis of output fluctuations in the US economy. Journal of Econometrics 56,
5788.
Pesaran, M.H., Shin, Y., 1996. Cointegration and speed of convergence to equilibrium.
Journal of Econometrics 71, 117-143.
Phillips, P.C.B., Perron, P., 1988. Testing for a unit root in time series regression. Biometrika
75, 335346.
Phylaktis, K., 1999. Capital market integration in the Pacific Basin region: An impulse
response analysis. Journal of International Money and Finance 18, 267-287.
102
Phylaktis, K., Ravazzolo, F., 2002. Measuring the financial and economic integration with
equity prices in emerging markets. Journal of International Money and Finance 21, 879-903.
Richards, A., 1995. Co-movements in national stock returns: Evidence of predictability not
cointegration. Journal of Monetary Economics 36 (3), 631 654.
Roll, R. (1992). Industrial structure and the comparative behavior of international stock
market indices. The Journal of Finance, XLVII, 3 41.
Siklos, P.L., Ng, P., 2001. Integration among Asia-Pacific and international stock markets:
Common stochastic trends and regime shifts. Pacific Economic Review 6 (1), 89 110.
Solnik, B., 1974. Why not diversify internationally rather than domestically? Financial
Analysts Journal 30, 4854.
103
Figure 5.2.2 All sixty six graphs of the weekly recursive bivariate cointegration tests
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
50 100 150 200 250 300 350 400 450
AUHK
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
AUID
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
AUJP
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
AUKO
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
50 100 150 200 250 300 350 400 450
AUMY
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
AUNZ
104
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
100 200 300 400
AUPP
0.2
0.4
0.6
0.8
1.0
1.2
1.4
100 200 300 400
AUSG
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
100 200 300 400
AUTA
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
100 200 300 400
AUTH
0.4
0.8
1.2
1.6
2.0
2.4
100 200 300 400
AUUS
0.4
0.6
0.8
1.0
1.2
1.4
1.6
100 200 300 400
HKID
105
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
50 100 150 200 250 300 350 400 450
HKJP
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
50 100 150 200 250 300 350 400 450
HKKO
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
50 100 150 200 250 300 350 400 450
HKMY
0.0
0.4
0.8
1.2
1.6
2.0
2.4
50 100 150 200 250 300 350 400 450
HKNZ
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
HKPP
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
50 100 150 200 250 300 350 400 450
HKSG
106
0.0
0.4
0.8
1.2
1.6
2.0
50 100 150 200 250 300 350 400 450
HKTA
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
50 100 150 200 250 300 350 400 450
HKTH
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
50 100 150 200 250 300 350 400 450
HKUS
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
50 100 150 200 250 300 350 400 450
IDJP
0.0
0.2
0.4
0.6
0.8
1.0
1.2
50 100 150 200 250 300 350 400 450
IDKO
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
50 100 150 200 250 300 350 400 450
IDMY
107
0.0
0.4
0.8
1.2
1.6
2.0
50 100 150 200 250 300 350 400 450
IDNZ
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
50 100 150 200 250 300 350 400 450
IDPP
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
50 100 150 200 250 300 350 400 450
IDTA
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
50 100 150 200 250 300 350 400 450
IDSG
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
IDTH
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
50 100 150 200 250 300 350 400 450
IDUS
108
.2
.3
.4
.5
.6
.7
.8
.9
50 100 150 200 250 300 350 400 450
JPKO
0.0
0.2
0.4
0.6
0.8
1.0
1.2
50 100 150 200 250 300 350 400 450
JPMY
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
JPNZ
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
50 100 150 200 250 300 350 400 450
JPPP
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
JPSG
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
JPTA
109
0.2
0.4
0.6
0.8
1.0
1.2
50 100 150 200 250 300 350 400 450
JPTH
0.4
0.6
0.8
1.0
1.2
1.4
1.6
50 100 150 200 250 300 350 400 450
JPUS
0.0
0.4
0.8
1.2
1.6
2.0
50 100 150 200 250 300 350 400 450
KOMY
0.0
0.2
0.4
0.6
0.8
1.0
1.2
50 100 150 200 250 300 350 400 450
KONZ
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
50 100 150 200 250 300 350 400 450
KOPP
.2
.3
.4
.5
.6
.7
50 100 150 200 250 300 350 400 450
KOSG
110
.2
.3
.4
.5
.6
.7
.8
50 100 150 200 250 300 350 400 450
KOTA
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
50 100 150 200 250 300 350 400 450
KOTH
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
50 100 150 200 250 300 350 400 450
KOUS
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
MYNZ
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
MYPP
0.4
0.6
0.8
1.0
1.2
1.4
1.6
50 100 150 200 250 300 350 400 450
MYSG
111
0.0
0.4
0.8
1.2
1.6
2.0
2.4
50 100 150 200 250 300 350 400 450
MYTA
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
50 100 150 200 250 300 350 400 450
MYTH
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
50 100 150 200 250 300 350 400 450
MYUS
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
50 100 150 200 250 300 350 400 450
NZPP
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
50 100 150 200 250 300 350 400 450
NZSG
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
50 100 150 200 250 300 350 400 450
NZTA
112
0.0
0.4
0.8
1.2
1.6
2.0
50 100 150 200 250 300 350 400 450
NZTH
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
50 100 150 200 250 300 350 400 450
NZUS
0.0
0.4
0.8
1.2
1.6
2.0
2.4
50 100 150 200 250 300 350 400 450
PPSG
0.0
0.4
0.8
1.2
1.6
2.0
50 100 150 200 250 300 350 400 450
PPTA
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
PPTH
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
50 100 150 200 250 300 350 400 450
PPUS
113
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
50 100 150 200 250 300 350 400 450
SGTA
0.2
0.4
0.6
0.8
1.0
1.2
50 100 150 200 250 300 350 400 450
SGTH
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
SGUS
0.2
0.4
0.6
0.8
1.0
1.2
50 100 150 200 250 300 350 400 450
TATH
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
50 100 150 200 250 300 350 400 450
TAUS
0.2
0.4
0.6
0.8
1.0
1.2
1.4
50 100 150 200 250 300 350 400 450
THUS
114
Figure 5.3a Persistence Profile for the Full period
0
0.2
0.4
0.6
0.8
1
1.2
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101
Weekly Horizons
P
e
r
s
i
s
t
e
n
c
e
P
r
o
f
i
l
e
Full Period CV1 Full Period CV2 Full Period CV3
Figure 5.3bPersistence Profile for the Pre-East Asian crisis preiod
0
0.2
0.4
0.6
0.8
1
1.2
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101
Weekly Horizons
P
e
r
s
i
s
t
e
n
c
e
P
r
o
f
i
l
e
Pre-Asian Crisis CV1 Pre-Asian Crisis CV2
115
Figure 5.3c - Persistence Profile for the Post East Asian crisis period
0
0.2
0.4
0.6
0.8
1
1.2
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101
Weekly Horizons
P
e
r
s
i
s
t
e
n
c
e
P
r
o
f
i
l
e
Post-Asian Crisis CV1 Post-Asian Crisis CV2
Figure 5.3d Persistence Profile for Cointegrating Vector 1 for all
periods
0
0.2
0.4
0.6
0.8
1
1.2
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101
Weekly Horizons
P
e
r
s
i
s
t
e
n
c
e
P
r
o
f
i
l
e
Full Period CV1 Pre-Asian Crisis CV1 Post-Asian Crisis CV1
116
Figure 5.3e Persistence Profile for Cointegrating Vector 2 for all
periods
0
0.2
0.4
0.6
0.8
1
1.2
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101
Weekly Horizons
P
e
r
s
i
s
t
e
n
c
e
P
r
o
f
i
l
e
Full Period CV2 Pre-Asian Crisis CV2 Post-Asian Crisis CV2
117
-.004
-.002
.000
.002
.004
.006
.008
.010
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.008
-.004
.000
.004
.008
.012
.016
.020
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.010
-.005
.000
.005
.010
.015
.020
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.004
-.002
.000
.002
.004
.006
.008
.010
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.004
-.002
.000
.002
.004
.006
.008
.010
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4a :Generalized impulse response functions to a one standard deviation shock in the Japanese market based on the full sample period
118
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.005
.000
.005
.010
.015
.020
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.005
.000
.005
.010
.015
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.004
-.002
.000
.002
.004
.006
.008
.010
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.005
.000
.005
.010
.015
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.010
-.005
.000
.005
.010
.015
.020
.025
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4b : Generalized impulse response functions to a one standard deviation shock in the US market based on the full sample period
119
-.004
-.002
.000
.002
.004
.006
.008
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.004
-.002
.000
.002
.004
.006
.008
.010
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.012
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.010
-.005
.000
.005
.010
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.004
-.002
.000
.002
.004
.006
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4c :Generalized impulse response functions to a one standard deviation shock in the Japanese market based on the pre-EA sample period
120
-.004
.000
.004
.008
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.012
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.004
.000
.004
.008
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.008
-.004
.000
.004
.008
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.004
.000
.004
.008
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.012
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.004
.000
.004
.008
.012
.016
.020
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4d :Generalized impulse response functions to a one standard deviation shock in the US market based on the pre-EA sample period
121
-.012
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.012
-.008
-.004
.000
.004
.008
.012
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.01
.00
.01
.02
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.012
-.008
-.004
.000
.004
.008
.012
.016
.020
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4e :Generalized impulse response functions to a one standard deviation shock in the Japanese market based on the post-EA sample period
122
-.010
-.005
.000
.005
.010
.015
.020
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.01
.00
.01
.02
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.015
-.010
-.005
.000
.005
.010
.015
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.02
-.01
.00
.01
.02
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4f :Generalized impulse response functions to a one standard deviation shock in the US market based on the post-EA sample period
123
-.016
-.012
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.06
-.04
-.02
.00
.02
.04
.06
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.016
-.012
-.008
-.004
.000
.004
.008
.012
.016
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.06
-.04
-.02
.00
.02
.04
.06
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.05
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.04
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4g :Generalized impulse response functions to a one standard deviation shock in the Japanese market based on the post-September 11 sample period
124
-.020
-.015
-.010
-.005
.000
.005
.010
.015
2 4 6 8 10 12 14 16 18 20
Response of Australia
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Hong Kong
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Indonesia
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Japan
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Korea
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Malaysia
-.015
-.010
-.005
.000
.005
.010
.015
2 4 6 8 10 12 14 16 18 20
Response of New Zealand
-.04
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of the Philippines
-.03
-.02
-.01
.00
.01
.02
.03
2 4 6 8 10 12 14 16 18 20
Response of Singapore
-.05
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Taiwan
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of Thailand
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
2 4 6 8 10 12 14 16 18 20
Response of the US
Figure 5.4h :Generalized impulse response functions to a one standard deviation shock in the US market based on the post-September 11 sample period