Volatility Spillovers and Contagion during the Asian Crisis: Evidence from Six Southeast Asian Stock Markets

Author(s): Kanokwan Chancharoenchai and Sel Dibooglu Reviewed work(s): Source: Emerging Markets Finance & Trade, Vol. 42, No. 2 (Mar. - Apr., 2006), pp. 4-17 Published by: M.E. Sharpe, Inc. Stable URL: http://www.jstor.org/stable/27750488 . Accessed: 15/01/2013 06:16
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

M.E. Sharpe, Inc. is collaborating with JSTOR to digitize, preserve and extend access to Emerging Markets Finance &Trade.

http://www.jstor.org

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

March-April 2006, pp. 4-17. ? 2006 M.E. Sharpe, Inc.All rights reserved. ISSN 1540^96X/2006 $9.50 + 0.00.

Emerging

Markets

Finance

and Trade,

vol. 42, no. 2,

Kanokwan

Chancharoenchai

and

Sel Dibooglu

Volatility Spillovers and Contagion During theAsian Crisis

Evidence fromSix SoutheastAsian StockMarkets

Abstract:

conditional generalized heteroskedasticity autoregressive in six Southeast Asian stock mar investigate volatility spillovers crisis. We focus on interactions with the U.S. market kets around the time of the 1997Asian as a world financial market. market, and with the Japanese market as a regional financial Using (GARCH-M) model, we use bivariate GARCH-M models to examine the behavior their interactions with other markets u the Asian contagion" which started in the region. All models in Thailand and rapidly stock markets, of individual markets and lend support to the idea of spread to other markets.

a multivariate

We also

Key

words:

Asian financial

crisis,

contagion,

time series models.

The Southeast Asian economies were theenvy ofmany countries before thefinan cial and currency crisis of July 1997, which began inThailand and spread rapidly to Malaysia, thePhilippines, Indonesia, Korea, Taiwan, and Hong Kong. The cri sis had a surprising and dramatic effecton both thefinancial and real sectors of the afflicted countries. At its core were the large-scale foreign capital inflows into Southeast Asian financial systems,which became vulnerable to panic and sudden reversals of market confidence (Charumilind et al. 2006; Jeon and Seo 2003;

at Kasetsart Uni is a lecturer in the Faculty of Economics Kanokwan Chancharoenchai Sei Dibooglu is an associate Thailand. (dibooglus@yahoo.com) versity, Bangkok, profes sor at the University of Missouri-St. Louis. Sei Dibooglu research support acknowledges ofMissouri-St. Louis. The opin from the Center for International Studies at the University ions expressed 4 herein are solely those of the authors.

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

MARCH-APRIL

2006

Nagayasu 2001). Most of the economic activity that the capital inflows supported in the affected countries was highly productive, so the loss of economic activity from the sudden inflow reversal was enormous. This forced the economies into sharp downturns. The crisis prompted the largest financial bailouts in history,and itwas themost severe financial crisis to hit the developing world since the 1982 debt crisis. Indonesian gross domestic product (GDP) contracted bymore than 15 percent in 1998, and theKorean and Thai economies contracted by approximately 7 and 10 percent, respectively.The crisis also threatened thegrowth of other emerg ing and transitioneconomies, and as such, it is importantto understand thepattern of volatility spillovers, the extent towhich such spillovers might have influenced

otherwise sound economies, and how these effects could be mitigated.1 This paper explores the contagion effects of the Asian crisis on Asian regional stockmarkets, including Japan, and global markets, as proxied by theU.S. stock market. To capture the interactions among the larger markets and emerging mar kets,we use amultivariate generalized autoregressive conditional heteroskedasticity As thecrisis started fromThailand and rapidly spread to (GARCH-M) framework. other neighboring countries,we examine thedynamics of contagion from theThai stockmarket to fiveAsian emergingmarkets using a bivariate GARCH-M model. We use data from six emergingAsian stockmarkets: Thailand (TH), thePhilip

ketswith twomajor stockmarkets, those of Japan (JP) and the United States (US). We then examine the contagion and spillover effects of the Asian crisis between each possible pair of countries in the sample: Thailand, thePhilippines, Indonesia, Asian Malaysia, Korea, and Taiwan. We consider two samples: data prior to the crisis (January 3, 1994, toDecember 31, 1996), and an extended sample (January 3, 1994, toDecember 31, 1999). and Methodology

pines (PH), Indonesia (IN), Malaysia (MA), Korea (KO), and Taiwan (TW). As a Asian mar preliminary step,we explore the dynamic interactions of each of the

Data

To investigate thebehavior of excess returnvolatility and volatility spillovers, we consider the daily closing price of six emerging Asian markets. All indices are denominated in local currency and expressed indaily percentages. The daily stock price indices are all drawn fromDatastream. To proxy the risk-freerates of return, we use the three-month T-bill rate for the Philippines, the six-monthmiddle de posit rate for Indonesia, the interbank overnight rate forThailand, the interbank two-month offered rate for Malaysia, the negotiable certificate of deposit (NCD) ninety-one-day yield forKorea, themoney market 180-day middle rate for Tai middle rateT-bill rate for Japan, and the three-month T-bill wan, the three-month second market middle rate for the United States. The portfolio weights reflect the relative size of the markets and are calculated from monthly market capitalization data inU.S. dollars. Daily data for both the interestrate and market capitalization
are also from Datastream.

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

6 EMERGING MARKETS FINANCEAND TRADE

The excess returns, rt, are the stock returns, t,net of the risk less rate, Rf t,

Rs

rt=Rsj-Rf,n

(1)

where Rs tis thefirstdifference of stock prices in logarithms, \nSPIt InSPI^, and all data used comprise daily closing values of stockmarket indices. Let the conditional variance of the excess returnsand covariance between the re domestic market portfolio and foreignmarket portfolio be ht and cov(rd,rfit), The relationbetween theexpected excess return and both htand spectively. cov(rd,rft) can be estimated by regressing both theexcess returnsrton thepredictable compo
nent of stock market

With daily data from differentgeographical regions, trading hours generally overlap in a calendar day. The Asian markets open before theU.S. market does. Therefore, theU.S. market returns may predict emerging market and Japanese
market returns. Because the Asian markets close before the U.S. market does?

volatility,

and

the covariance,

cov(rd,rfit).

more specifically, on theprevious calendar day?the Asian market returnsdo not As inChan et al. (1992), to account for help to explain previous-day U.S. returns. the lack of synchronization in tradinghours, one lagged disturbance of theU.S. returns is incorporated in the Asian returns. In addition, returns for each country depend on one lagged disturbance to capture the effects of infrequent trading on the dynamics of index returns.2 For our three-dimensional model, with a major regional market (Japan), a global market (theUnited States), and each of the six Pacific Asia emerging stockmarkets (/?), the typical conditional mean equation can be expressed as =

rpj
+

+aprPJ-l +?Pl?PJ-l +dp2?usJ-l+^pla)pJhPJ

?p,t

C0V + + ' > {rP,t rUSJ) ?py Vjpj {rpj rjP,t ) ?USl ?W = + rus,t +dUs[?us,t-\ ^US^USlKlSJ ^us0 +ausrus,t-\

+
= KJP.r

+ + ?PlMpj C0V(rPj>rus.t ) ?jPl^jpj cov(rw,rjN) eust

X.iPo +(X.ipriP.t-\ +^Pl?jpa-l+djP2?us^+^jp^jpahJPa

?P2a)pj ^(^t^jpj^?us^usj

+ c?v(rusnrjp t) eJpJ

[et]~N(0,[Ht]),
where [Ht] is the variance-covariance matrix, and [et] is the vector of error terms on a typicalAsian emergingmarket portfoliomay be influenced by nondomestic
factors. from estimating r , rus, and rjp. Formulated this way, the monthly excess returns rp t

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

MARCH-APRIL

2006

model empirically, it is importantto specify thedynamics of To implement the conditional variance and covariance. Extending the standard (univariate)GARCH-M model, Bollerslev et al. (1988) propose a multivariate GARCH-M specification thatallows the covariance terms to influence the domestic returnprocess. For the three-dimensional case (trivariate GARCH-M process), the conditional variance covariance specification can be expressed as

( <PP )
Vec = [//J h
c?v(rP,t>rus,t) c?v(rp,t>ruP,t) <t>us

<t>JP
<t>p,us

<t>pjp \0usjp )

0 ^22 0

0 0 0

0 0 0 0

0 0 0 0 0 0 0

0 0

a55

o ^ 0 0 0 0
a66/ h

-pj-i ?usJ-\

epJ-\eusJ-\ e p,t-\? jpJ-\

\?usJ-\e jpJ-\ ) PJ-\

(3)
PJ

7n
0
+

0
y22

0 0
^33 0 0 0

0 0 0 744 0 0

0 0 0 0 755 0

o ^ o o o 0

Ksj-\ jp,t-l jp-t pus.t PJPJ

0 0 0 ^0

0 0 0 0

+ covfr^.!,^)

^{rpJ.{J(jpJ.{) 766/ [cov(w-i'r^-i)4

J \VusjpJ

where Vec(

matrix [//J, and ) is thevector operator that stacks the columns of the matrices. Additionally, with Equations (2) [0], [a], and [y] are diagonal coefficient and (3), theanalysis of dynamic patterns of variances is modeled byGARCH( 1,1), as commonly used in the literature. We ignore higher-order termsof lagged condi

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

8 EMERGING MARKETSFINANCEAND TRADE tional variances or prediction errors, following the empirical findings of French et tomodel conditional variance of al. (1987), who show thatusing a GARCH(2,1) excess returns /zrdoesnot appear todiffer significantly fromusing a GARCH(1,1). The trivariate conditional variance-covariance specification of Equation (3) allows the conditional variances to depend only on past squared residuals, and
covariances to depend on past products of error terms. The important cross-market

effects, highlighted by Hamao et al. (1990) for the national stockmarkets of the United States, United Kingdom, and Japan, are not sufficiently included in the GARCH-M model. Even though a more general Bollerslev et al. (1988) trivariate to be could process capture cross-market spillover effects,positive semi specified definiteness of the conditional covariance matrix in thatprocess is not assured. To more general dynamics in model (3), the [a] and [y] matrices would have to reflect
include nonzero,

fication is, therefore,respecified intoEquation (4), as originally proposed by Baba et al. (1989), denoted as BEKK below:

off-diagonal

elements.

The

conditional

variance-covariance

speci

= + [G]'[>_,][et_{]' + [Fl [Ht_x][F] [G], [Ht] [Pl[[P]

(4)

where [Ht] denotes the 3x3 variance-covariance matrix conditional on informa denotes the vector of disturbances fromEquation (2). The tion at time t,and term [P] is an upper triangular matrix of three coefficients, whereas [F] and [G] are free (square) matrices of coefficients with nine parameters for each. Unlike fullparameterization, thisapproach economizes thenumber of parameters inEqua

variance of returnswith othermarkets. This covariance is a weighted average of the variance of themarket portfolio of domestic assets and the covariance of the returns on themarket portfolio of domestic assets with themarket portfolio of foreign assets (foreign influences),where theweights are the proportions of do mestic and other stocks in theworld market portfolio. Empirical Results Ideally, ourmodel should estimate an eight-variablemultivariate GARCH-M model of the full set of stock excess returns,to account for contagion or spillover effects among the eightmarkets. Unfortunately, thiswould require estimating 80 param moment and 162 parameters in the second moment, which is impos eters in thefirst siblewith prevailing computing technology and numerical methods. Therefore, this and two-variable GARCH-M models studyfocuses on two subsystems: three-variable on thedaily stock excess returns. The three-variable (trivariate GARCH-M) model

tion (3) (twenty-four, including the interceptparameter for the trivariate system used here), and guarantees that the covariance matrices are positive definite. model, with the threeequations above, the conditional mean Consequently, the (2), and the conditional variance (3) and (4), allow for considerable dynamics in market portfolio of own assets and the co the risk-premium relation between the

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

MARCH-APRIL

2006

markets. The multivariateGARCH spilloversbetween any given pair of theemerging IV!approach shows the relation between the variance of the respectivemarket and the variance of othermarkets and describes the effect that the covariance between market as a volatility spillover effect. the markets has on the excess returnson that co on are The portfolioweights based U.S. dollars, rela daily capitalization data in sum to tive thevalue of the of therelevant markets, and reflect the relative size of the
markets

Asia and twodeveloped markets, includes one of the six emerging stockmarkets in as a United States the Stock Exchange regionalmarket) and the Japan (using Tokyo as to the New Stock the York capture volatility Exchange global market), (using transmission from regional and world markets to the sixAsian emerging stockmar kets.We use the two-variable (bivariate GARCH-M) model to explore volatility

Even thoughwe alluded to excess returnshaving a multivariate ^-distribution, Chan et al. (1992) showed that restoring normality in the sample by removing outliers did not significantly change the results. Given their results and conver gence problems, we use themultivariate normal distribution in the optimization
routine.

investigated.

Evidence from Trivariate GARCH-M Models Table 1presents estimation results for theThailand-Japan-United States trivariate GARCH-M model using theBEKK parameterization,3 relating each stockmarket of the sixAsian emergingmarket indices to the regional (Japan) and global (United States) stockmarkets. In themean equations in each estimated model, the inter surprising,because reduced capital gains taxes on long-termassets provide incen tives to hold those assets, despite otherwise unfavorable rates of returns(Bollerslev et al. 1988). They also reflect thatequity holders did consistently worse over the most country sample period. Furthermore, the time-series analysis indicates that stock index returns exhibit first-order serial correlation, which can be explained by institutional factors, such as bid-ask spreads and nonsynchronous trading in individual stocks.4The significant coefficients of one lagged disturbance for the
cept parameters are mostly negative. These large negative-intercept terms are not

emerging market (d{) and theUnited States (d2) also show that the asynchronism in trading times is successfully captured in themodel, as Chan et al. (1992) sug gest. This finding thus strongly supports the effect of differentcalendar days and cannot be ignored in the analysis. In addition, themean equations show that the effects of conditional variance in each individual market?the values of the Xx parameter?are, with few exceptions, all positive, but have weak explanatory power
for market excess returns in almost all cases,

This lack of significance of the coefficient on thevariance is somewhat surprising and implies that time variation in the conditional variance of the entiremarket is
not an

according

to asymptotic

^-statistics.

premium effectdoes not necessarily indicate the absence of a premium associated

important

source

of variation

in excess

returns. However,

this weak

market

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

0.136 0.351 (12.31)** 0.008 -60.087 (-3.29)** 0.310 (1.56) (1.58) (-1.29) -0.114 (1.44) Japan 0.112 (0.04)

0.019 0.071 States United (January 21, 1994, December 1999) 31, to Extended sample 0.024 (0.56) (0.30) (0.10)

(1.80)* -10.079 0.096 (-0.63)

1.134 (2.27)**

Thailand

(8.21)** -0.085 0.208 0.421 (2.24)* (-0.06) (-0.87) (-0.07) -0.008 -0.398

-0.145 0.078 (-0.74) (0.14)

rus,t= XusQ + ausrus,t-: + ?us/us,t^ + xus^us,thus,t

+ ?jp2a,jp,tcow(rus,t>rjp,t) + i:us,t

?p^}p,tco^rp,t'rusf + + cov(rP,f>rP.t) cov(w-0p.r) + ?jpAEjp,t^ djp2cus,t^ ljp^wjp,thip,t rjpt ajprjp,t-l ?p2a>p,t + ?us2?>us,t + = ejp,t ?w)+ ?jp^jpjc?v(rP,t>rjp,t) rp,t=XP0+ aprpX-\ + ?Pi?p,t-1 + ?p2Fus,t-Jl + ?us^>us,t Xjp0 cov(rp,f + ?p,t Estimates coefficients of conditional returns excessxp^>p,thp,t 0.174 (1.26) Japan -0.083 (-0.76) 0.320 0.253 5.485 1.419 (0.35) (0.34)

(-1.01) (6.32)**(0.82) -0.146

GARCH-M Trivariate from Estimates Excess Thailand, Daily United Returns the of and States Japan,

Before Asian the crisis United States 1994, 5, to (January December 31, 1996)

0.214 (0.66) (-1.16) -0.243 (-0.36) -0.119

11.636 (1-32) 1.362 (0.21)

0.039 (0.02)

Thailand

(3.73)**(-2.55)**

0.409 -0.134 (-0.56)

12.181 (1.19) 0.509 (5.99)** -0.312

-4.491 1.105 (1.37) (-1.02)

Table 1

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

G21 0.084

(-2.57)** (0.86) -0.093

Estimates coefficients of matrix variance-covariance Notes: Portfolio weights

]' [HM][F] ][G] [>M [Ht ] [F] [G] [P] + =

reflect

relative

sizes

of the three

markets.

in parentheses Numbers

are

^-statistic

P13

1.136

(-7.77)** (7.95)** G33 0.344 -0.096 (-0.01) -0.039 0.457 G31 0.216 0.258 (15.99)** 0.218 (8.09)** F13 -0.241 (-21.34)** -0.491 0.162 -0.073 (0.16) 0.051 0.061 (3.15)** 0.576 P22 (-0.01) 0.053 (5.48)** 0.007 0.086 (4.55)** (0.19) 0.032 0.004 G12 G23 0.053 0.070 F22 Fn 0.132 F12 -0.047 -0.073 (2.80)** (-5.90)** -0.167 (-1.50) 0.903 0.220-0.332 (36.89)** F23 -0.130 (-4.75)** F33 0.871 0.242 (4.73)** (0.01 (0.02e-5) 0.002 0.002e-1 P33 e~3) (0.02e~1) (2.24)** (-3.62)** G22 -0.019 0.242 (14.71)** 0.002 P23 -0.045 -0.150 (-0.06) G32 0.066 (0.92) F31 (0.14) -0.240 1.788 (21.56)** F21 -0.426 0.380 (2.578)** F32 -0.006 -0.098 (-1.88)* (0.68) (-0.53) (-0.12) (-0.20) (-0.47) (-0.20) (-1.54) (-0.08) (2.67)** (12.21)** (1.25) (0.33) (0.19) (2.17)* (16.65)** * (1.32) ** Statistically significant at the 1 percent Statistically level.

sig

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

12 EMERGING MARKETSFINANCEAND TRADE movements of the total with nondiversifiable risk in the market. It may be a sign of or weak evidence of time variation in market risk. multicollinearity, While the market's variance (X{) has a positive effect ingeneral, the coefficients on the covariance termshave negative and positive signs. Specifically, in the com plete sample, Indonesian expected excess returnsdepend negatively on the covari ance of Indonesian excess returns with Japan, or cov(r/nt,rjp and positively on the t),
covariance with U.S. excess returns, or

ceive volatility from Malaysia and transmitvolatility toKorea in theprecrisis pe riod. Similarly,U.S. volatility transmitsto the Taiwanese market in the same period. When theentire sample is considered,Malaysia transmitsvolatility to Japan,which transmits it to the United States. Both Japan and the United States transmitvolatil ity to Indonesia. Japan receives a considerable volatility effect fromMalaysia throughcov(rma t,rjp t) in both sample periods. This finding generally contrastswith the literature, inwhich developed markets have a major effect on emerging mar kets, but not theotherway around. These volatility transmissionpaths are summa rized inFigure 1. was some interdepen Despite the literature,ourmodel demonstrates that there dence in volatility between emerging markets and developed markets before and after the Asian crisis.The covariance termsimply a Shockwave thattraveled through the differentfinancialmarkets.

cov(rjnT,rust).

Japanese

excess

returns

re

Evidence from Bivariate GARCH-M Models This section uses the bivariate GARCH-in-mean model to examine volatility spillover effects fromone Asian emergingmarket to another: Thailand, thePhilip
pines, Indonesia,

sions, with eight parameters in the conditional mean equation and eleven in the
variance-covariance equation for each regression. Because the six

Malaysia,

Korea,

and Taiwan.

There

are fifteen

separate

regres

conditional

cients suggests thatall markets, except Taiwan, exhibit first-orderserial correla tion.Perhaps most interesting,time-varyingvariance in therespective stockmarket is not the only importantsource of variation in the stock excess returns. Overall, the value of theA1?the conditional variance coefficient, tends tohave theexpected positive sign. For the covariance, theThai excess returnsdo not depend on any otherAsian emerging stockmarkets' behavior, though at the same time, theThai market does apparently cause positive volatility throughcov(rthnrpU)in thePhilip Asia at the 10 percent level. pine market during thepost-financial crisis period in This is strong evidence for the belief of rapid spread of the fallout fromThailand to neighboring countries during the Asian crisis.

emerging markets are approximately in the same geographical region,we are not concerned with asynchronous tradingand do not include lagged disturbance terms. Results for thePhilippines-Thailand model are given5 inTable 2. The mean equations show that the interceptparameters overall are negative in sign but statistically insignificantfor all markets. The significance of thea coeffi

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

MARCH-APRIL

2006

13

Figure 1.The Pathways of VolatilityTransmissions

in the Trivariate Model

Asian crisis (January 3, 1994, toDecember 31, 1996) (a) Before the Korea

Malaysia

Japan

w+

United States

Taiwan

3, 1994, toDecember 31, 1999) (b) Extended sample (January Indonesia

Malaysia

* Japan m+ United States

In addition, over thepre-financial crisis period in Asia, thePhilippine and Tai wanese markets receive volatility from each other through The Phil t). cov(rph ,,rm. The ippinemarket transmitsvolatility to theKorean market through co\(rpll t,rko t). Taiwanese market, and transmits it Korean market also receives volatility from the back in turn, through cov(rw,rm.,), but the influence appears to be weak. Finally, These results are evidence of interdependence among the six emergingmarkets in Asia, even before the crisis. For theconditional mean equations over theextended sample period, the statis tics show a number of interestingresults regarding the covariance terms.The ex and Taiwanese market are strongly pected excess returns on theMalaysian
influenced the Malaysian market receives volatility through cov(rjnPrmat) and cov(r/mu,rm.,).

Our results indicate that the returncomovements among East Asian stockmarkets were strongprior to the Asian crisis and continued unabated after the Asian crisis. These results are broadly in line with those ofYang and Lim (2004). Moreover,

given inPanel C of Figure 2. There is strongevidence of contagion after the crisis. The Thai stockmarket's volatility spillover suggests that ithas played an increas ingly important role in theAsian stock markets after itsmid-1997 fallout. The emergingmarkets also seem to be highly connected to regional capital markets.

nrMt)to the respectively.The Taiwanese market transmitsvolatility throughcov(r/>; Indonesian market. The Malaysian market transmitsvolatility throughcov(rjn t,rfm t) to the Indonesian market; at the same time, itsexpected excess returnsdepend on The pathways of volatility transmis Indonesian volatility through cov(rin t,rmflJ). sion are given inFigure 2. We also estimate bivariate models fordaily stock excess returnsamong the six emerging markets from January 1, 1997, toDecember 31, 1999. The results are

by

theMalaysian

excess

returns

through

cov(rlna

t,rkot) and

cov(rm,

nr,a t),

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

14 EMERGING MARKETSFINANCEAND TRADE Table 2 Estimates Thailand from the Bivariate GARCH-M Model with the Philippines and of the conditionalexcess returns Estimates of the coefficients
+ rd,t=*d0 +adrd,t^ +^-(?f,t)hd,t +Xd2Uf,t c?v(rd,t>rf,t) ?d,t = + + + + ef,t rf,t \ afrf,t-\ ^f,thf,t V1-(0f,t)Q)djcov(rd,t>rf,t)

Before theAsian crisis (January5, 1994, to December 31, 1996) Philippines

0.013

Extended sample 5, 1994, to (January December 31, 1999) Philippines

-0.020 (-0.47) 0.170 (7.86)** -0.031

Thailand
-0.141 (-1.34) 0.071

Thailand
-0.105 (-1.19) 0.098 (4.81)** 0.054

(0.16)
0.168 (5.12)** -0.027 (-0.10)

(2.07)*
0.055

(0.57)
0.407

(2.12)*
0.131 (1.70)***

(1.04)
-0.008 (-0.09) matrix

k9

0.042

(0.32)
Estimates

(1.19)
of the coefficients

of the variance-covariance

= [Ht]H

+ [G]' + [F] [Ht_,][F] [P] [eM J[e?][G]

J{rd,t>rf,t)) d,t [<x>v{rd,t>rf,t) ht
0.006 -0.230 -0.134 0.230 0.363 0.152 -0.003 0.280 1.143 0.550 -0.346 (0.02e-2) (-0.22) (-0.16) (14.94)** (14.41)** (15.12)** (-0.18) (5.11)** (26.32)** (17.28)** (-6.32)** "f,t

with [Hf] =
0.067 0.665 0.704 -0.096 -0.148 0.308 0.441 -0.692 0.245 0.227 -0.787 Notes:

P11 ^12 P '22 ?n G12 G21 G22 F? Fl2

(0.34) (11.15)* (8.62)*

(-1.84)* (-2.97)*

(8.24)* (8.69)*
(-9.94)*

(3.31)* (4.98)*
;-15.05)*

excess returns are calculated in local currency:/= Daily Thai and Philippine reflect relative size of the two markets. d = Philippines. Portfolio weights Thailand, are r-statistics. *** Statistically in parentheses Numbers significant at the 10 percent * ** level. Statistically significant at the 5 Statistically significant at the 1 percent level. percent level.

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

MARCH-APRIL

2006

15

Figure 2. The Pathways of VolatilityTransmissions

in the Bivariate Model

Asian crisis (January 3, 1994, toDecember 31, 1996) (a) Before the

Philippines

Korea

Taiwan V Indonesia Malaysia

3, 1994, toDecember 31, 1999) (b) Extended sample (January

Thailand
Indonesia

wm^Philippines
Malaysia

t
Taiwan

t
Korea

(c) Asian crisisperiod (January1, 1997, toDecember 31, 1999)

Thailand ^

Philippines

Taiwan

4
Malaysia ^

*
Korea

I
Indonesia

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

16 EMERGING MARKETSFINANCEAND TRADE thereare contagion effects in the region, despite capital controls imposed by some
countries, such

Finally, the significance of the diagonal elements of the [F] and [G] matrices indicates thatGARCH effects are prevalent and strong.On the other hand, the assumption of cross-market effects can be confirmed by thehigh degree of signifi cance of the off-diagonal elements. This high degree of significance is consistent with the results of Hamao et al. (1990), who show that the cross-market effect cannot be neglected. This empirical finding confirms the estimation results from
the conditional mean equations.

as Malaysia.

Conclusions We examine volatility spillovers in Southeast Asian emerging stockmarkets in the context of the mid-1997 financial crisis, using a multivariate GARCH process with BEKK parameterization tomodel the volatility of excess returns.This procedure reflects thewell-known autoregressive behavior in volatility series, and accounts for spillover effects between various equity markets. The significant degree of these effects can reveal the relative size and openness of a particularmarket. The excess returnsexhibit first-orderserial correlation,which can be explained by institutional factors. The effect of nonsynchronous trading hours is found to have a significant effect in explaining index returns. In addition, results indicate significant foreign influences on the time-varying risk premiums in all specifica tions and models. Similarly, thebivariate GARCH-M model provides strong evi dence of reactions among the six neighboringmarkets inSoutheastAsia. The sudden fallout inThailand seems to have played an importantrole in the variation in ex cess returns inother Southeast Asian markets. This supports the idea of the "Asian contagion," suggesting that the crisis started inThailand and spread to other finan
cial markets.

Notes
use "contagion" of transmission in the broad sense, including cross-country or cross-country spillovers are indicated by signifi spillover effects. In our model, terms. For alternative definitions and a brief survey, see Yang cant cross-country covariance 1.We shocks

and Lim (2004).

2. For details 3. Results

Kim et al. (2000).

east Asian upon

on asynchronous States

trading,

see Chan

et al. (1992), Wei

et al. (1995),

and

for the United

countries

are detailed

and Japan in tables 3-7

together with each of the following South in an appendix, available from the authors and in Cohen in

request: the Philippines, 4. An extensive discussion

and Korea. Indonesia, Taiwan, Malaysia, can be found in Scholes and Williams (1977) combinations pairwise from the authors upon of countries request.

et al. (1986). 5. Results tables 8-21

of the appendix,

for all other possible available

are detailed

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions

MARCH-APRIL

2006

References

Baba,

Y.; R.F.

eralized Bollerslev,

1989. "Multivariate Simultaneous Engle; D.F. Kraft; and K.F. Kroner. ARCH." of California, San Diego. Paper, University Working and J.M. Wooldridge. 1988. "A Capital Asset Pricing Model T.; R.F. Engle; Covariance." Journal of Political Economy 96, no. 1 (February):

Gen with 116?

Time-Varying 131. Chan, K.C.; G.A. Premium Charumilind,

and R.M. Stulz. 1992. "Global Financial Markets and the Risk Karolyi; on U.S. Equity." Journal of Financial Economics 137-167. 32, no. 2 (October): and Y. Wiwattanakantang. 2006. "Connected Thailand C; R. Kali; Lending:

Before the Financial Crisis." Journal of Business 181-218. 79, no. 1 (January): S. Mai er; R. Schwartz; and D. Whitcomb. 1986. The Microstruc Cohen, K; G. Hawawini; ture of Security Markets. Cliffs, NJ: Prentice Hall. Englewood 1987. "Expected Returns and Volatility." Jour French, K; W. Schwert; and R. Stambaugh. nal of Financial Economics 3-30. 19, no. 1 (September): Hamao, Y; R.W. Masulis; International Across 281-307. and B. Seo. 2003. "The Impact of the Asian Financial Crisis on Foreign Ex Market of East Asian Countries." Finance change Efficiency: The Case Pacific-Basin Journal 509-525. 11, no. 4 (September): and I.Mathur. 2000. "Price Transmission Between ADRs Kim, M.; A. Szakmary; Dynamics Jeon, B.N., and Their and V. Ng. in Price Changes 1990. "Correlations and Volatility Stock Markets." Review of Financial Studies 3, no. 2 (Summer):

(August): 1359-1382.

Underlying

Foreign

Securities."

Journal

of Banking

and Finance

24, no. 8 Rates and

J. 2001. "Currency Crisis and Contagion: Evidence from Exchange Nagayasu, Sectoral Stock Indices of the Philippines and Thailand." Journal of Asian 12, no. 4 (Winter): 529-546. Scholes,

Economics

Williams. 1977. "Estimating Betas from Nonsynchronous Data." Jour M., and J. nal of Financial Economics 309-327. 5, no. 3 (December): and G.-S. Chaung. 1995. "Volatility and Price Change Wei, K.C.J.; Y.-J. Liu; C.-C. Yang; the Developed and Emerging Markets." Finance Spillover Effects Across Pacific-Basin Journal?*,no. Yang, T, Pacific 1 (May): 113-136. and J.J. Lim. 2004. "Crisis, Basin Financial Markets Contagion, and Policies and East Asian 7, no. Stock Markets." 119-151. Review of

1 (March):

To order reprints, call

1-800-352-2210;

outside

the United States,

call 717-632-3535.

This content downloaded on Tue, 15 Jan 2013 06:16:43 AM All use subject to JSTOR Terms and Conditions