_______________________________________________________________
Centre for International Capital Markets
Discussion Papers ISSN 1749-3412
_______________________________________________________________
Global and Regional Spillovers in Emerging Stock Markets:
a Multivariate GARCH-in-Mean Analysis
John Beirne, Guglielmo Maria Caporale, Marianne Schulze-Ghattas, Nicola Spagnolo
No 2009-17
2
Global and regional spillovers in emerging stock markets: a multivariate
GARCH-in-mean analysis
John Beirne
a1
Guglielmo Maria Caporale
b*
Marianne Schulze-Ghattas
c1
, Nicola Spagnolo
b
a
European Central Bank
b
Centre for Empirical Finance, Brunel University, London, UK
c
Financial Markets Group, London School of Economics, London UK
___________________________________________________________________________
Abstract
This paper examines global (mature market) and regional (emerging market) spillovers in
local emerging stock markets. Tri-variate VAR GARCH(1,1)-in-mean models are estimated
for 41 emerging market economies (EMEs) in Asia, Europe, Latin America, and the Middle
East. The models capture a range of possible transmission channels: spillovers in mean
returns, volatility, and cross-market GARCH-in-mean effects. Hypotheses about the
importance of different channels are tested. The results suggest that spillovers from regional
and global markets are present in the vast majority of EMEs. However, the nature of cross-
market linkages varies across countries and regions. While spillovers in mean returns
dominate in emerging Asia and Latin America, spillovers in variance appear to play a key
role in emerging Europe. There is also some evidence of cross-market GARCH-in-mean
effects. The relative importance of regional and global spillovers varies too, with global
spillovers dominating in Asia, and regional spillovers in Latin America and the Middle East.
JEL classifications: F30; G15
Keywords: Volatility spillovers; contagion; stock markets; emerging markets
___________________________________________________________________________
1. Introduction
The empirical finance literature abounds with studies of cross-border links in stock market
returns. This is not surprising. Empirical modelling of such links is relevant for trading and
hedging strategies and provides insights into the transmission of shocks (news) across
______________________
1
Marianne Schulze-Ghattas was on sabbatical from the International Monetary Fund and visiting fellow at the
Financial Markets Group, London School of Economics when the research was done. The views expressed in
this paper are those of the authors and do not necessarily reflect those of the European Central Bank or the
International Monetary Fund.
* Corresponding author. Centre for Empirical Finance, Brunel University, West London, UB8 3PH, UK. Tel.:
+44 1895 266 713; fax: +44 1895 269 770. E-mail address: Guglielmo-Maria.Caporale@brunel.ac.uk.
3
markets. Informed by standard asset pricing models and supported by advances in the
econometric modeling of volatility, research in the past two decades has focused on
interdependencies in terms of both first and second moments of return distributions.
Early studies of spillovers across national stock markets primarily covered advanced
countries. Prompted by the October 1987 stock market crash in the US, Hamao, Masulis and
Ng (1990), King and Wadhwani (1990) and Schwert (1990) examined spillovers across
major markets before and after the crash. Subsequent research refined and expanded the
analysis of advanced market links by examining spillovers in high frequency (e.g., hourly)
data (Susmel and Engle, 1994); asymmetry in the transmission of positive and negative
shocks (Bae and Karolyi, 1994; Koutmos and Booth, 1995); differences in the transmission
of global and local shocks (Lin, Engle and Ito, 1994), and interactions among larger sets of
advanced markets (Theodossiou and Lee, 1993; Fratzscher, 2002).
Research into cross-border links in emerging stock markets was boosted by the growth and
increasing openness of these markets, as well as the speed and virulence with which past
financial crises in emerging market economies (EMEs) spread to other countries. Bekaert and
Harvey (1995, 1997, 2000) and Bekaert, Harvey and Ng (2005) analyse the implications of
growing integration with global markets for local returns, volatility, and cross-country
correlations, covering a diverse set of EMEs in Africa, Asia, Latin America, and the
Mediterranean. Most other studies of EME stock markets focus on specific regions.
Scheicher (2001), Chelley-Steeley (2005), and Yang, Hsiao and Wang (2006) examine extent
and effects of stock market integration in Central and Eastern Europe, both within the region
and with advanced markets, while Chen, Firth and Rui (2002) look at evidence of regional
linkages among Latin American stock markets. Floros (2008) focuses on the Middle East,
while Ng (2000), Tay and Zhu (2000), Worthington and Higgs (2004), Caporale, Pittis and
Spagnolo (2006), Engle, Gallo and Velucchi (2008), and Li and Rose (2008) examine stock
markets in emerging Asia.
These studies generally point to increasing links among emerging stock markets, and
between these markets and mature markets. However, results are difficult to compare across
countries because they are based on different methodologies, time periods, and data
frequencies. This paper seeks to remedy this problem by applying a uniform specification to
a large set of EMEs - 41 in all - spanning four regions: Asia, emerging Europe, the Middle
East and North Africa, and Latin America. A downside of this approach is that, given the
large number of countries in each region, we cannot model simultaneously the links among
all local markets, and between these markets and major mature markets. We focus on links
between local emerging markets and aggregate global and regional markets as we are
interested in the impact of the latter on the former.
The paper relies on a broad model framework that encompasses several channels through
which news in global and regional markets may influence local emerging markets. More
specifically, we apply a tri-variate VAR-GARCH-in-mean framework with the BEKK
representation proposed by Engle and Kroner (1995) to model and test for cross-market
spillovers in means and variances of stock returns as well as own and cross-market spillovers
from second to first moments (GARCH-in-mean effects). This approach builds and expands
on the methodologies adopted in earlier studies such as Hamao, Masulis and Ng (1990), Ng
4
(2000), and Bekaert, Harvey and Ng (2005). The global market in each tri-variate model is a
GDP-weighted average of the US, Japan, and Europe (Germany, France, Italy, and the UK),
1
and the regional market is a weighted average of all emerging markets in the region included
in our country sample, except for the model’s local market.
2
Our analysis is based on weekly
stock returns in local currency. Time series end in mid-March 2008 and start in 1993 for
emerging Asia, and in 1996 for Latin America, most markets in emerging Europe, South
Africa, the Middle East and North Africa.
We use Wald tests to examine several hypotheses about spillovers in means and variances, as
well as GARCH-in-mean effects, from global and regional markets to local markets. The
results suggest that spillovers from regional and global markets are present in the vast
majority of EMEs. However, the nature of cross-market linkages varies across countries and
regions. While spillovers in mean returns dominate in emerging Asia and Latin America,
spillovers in variance appear to play a key role in emerging Europe. There is also some
evidence of cross-market GARCH-in-mean effects. The relative importance of regional and
global spillovers varies too, with global spillovers dominating in Asia, and regional spillovers
in Latin America and the Middle East.
The paper is organised as follows. Section 2 describes the econometric model. Section 3
provides details on the data set and outlines the hypotheses tested. Section 4 discusses the
results; and section 5 offers some concluding remarks.
2. Methodology
We represent the first and second moments of returns in local, regional and global stock
markets by a tri-variate VAR-GARCH(1,1)-in-mean process.
3
In its general specification the
model has the following form:
x
t
= α + Β'x
t-1
+ Γ' h*
t
+ u
t
(1)
with x
t
a 3x1 vector of returns in local emerging markets, regional emerging markets, and
mature markets; x
t-1
a corresponding vector of lagged returns; h*
t
= (√h
11,t
,√h
22,t
,√h
33,t
) a
vector of the conditional standard deviations in local, regional, and global markets; and u
t
=
(e
1,t
, e
2,t
, e
3,t
) a residual vector. The parameters of the mean return equations (1) comprise the
constant terms α = (α
1
, α
2
, α
3
); the parameters of the autoregressive terms Β = (β
11
, 0, 0 | β
21
,
β
22
, 0 | β
31
, β
32
, β
33
), which allow for mean return spillovers from mature markets to regional
and local emerging markets, and from regional markets to local markets; and Γ = (γ
11
, 0, 0 |
γ
21
, 0, 0 | γ
31
, 0, 0) the parameters of the GARCH-in-mean terms.
1
We used GDP weights because time series on market capitalisation were not available for all emerging
markets in our sample.
2
Bekaert, Harvey, and Ng (2005) adopt a similar approach.
3
The model is based on the multivariate GARCH(1,1)-BEKK representation proposed by Engle and Kroner
(1995).
5
The residual vector u
t
is tri-variate and normally distributed u
t
| I
t-1
~ (0, H
t
) with its
corresponding conditional variance-covariance matrix given by:
h
11,t
h
12,t
h
13,t
H
t
= h
21,t
h
22,t
h
23,t
(2)
h
31,t
h
32,t
h
33,t
In the multivariate GARCH(1,1)-BEKK representation proposed by Engle and Kroner
(1995), which guarantees by construction that the variance-covariance matrices in the system
are positive definite, H
t
takes the following form:
a
11
0 0
'
e
1,t-1
2
e
1,t-1
e
2,t-1
e
1,t-1
e
3,t-1
a
11
0 0
H
t
= C'
0
C
0
+ a
21
a
22
0 e
2,t-1
e
1,t-1
e
2,t-1
2
e
2,t-1
e
3,t-1
a
21
a
22
0
a
31
a
32
a
33
e
3,t-1
e
1,t-1
e
3,t-1
e
2,t-1
e
3,t-1
2
a
31
a
32
a
33
g
11
0 0
'
g
11
0 0
g
21
g
22
0 H
t-1
g
21
g
22
0 (3)
g
31
g
32
g
33
g
31
g
32
g
33
Equation (3) models the dynamic process of H
t
as a linear function of its own past values H
t-1
as well as own and cross products of past innovations e
1,t-1
, e
2,t-1
, e
3,t-1
, allowing for own-
market and cross-series influences in the conditional variances. The parameters of (3) are
given by C
0
, which is restricted to be upper triangular, and two matrices A
11
and G
11.
Each of
these two matrices has three zero restrictions as we are focusing on volatility spillovers
(causality-in-variance) running from mature stock markets to regional and local emerging
stock markets, and from regional to local emerging markets.
Given a sample of T observations, a vector of unknown parameters θ
4
and a 3 x 1 vector of
variables x
t
, the conditional density function for the model (1)-(3) is:
4
Standard errors are calculated using the quasi-maximum likelihood method of Bollerslev and Wooldridge
(1992), which is robust to the distribution of the underlying residuals – these are not reported for reasons of
space. A residual vector u
t
following a t-student distribution has also been considered, but the results were
qualitatively similar and therefore are not reported. The full set of results is available from the authors upon
request.
