A network visualization approach and global stock market integration

• DOI: http://doi.org/10.1002/ijfe.1617
• Date: 01 July 2018
• Author: Jing Chen,Mike J. Buckle,Chen Tong
• Published date: 01 July 2018

RESEARCH ARTICLE

A network visualization approach and global stock market

integration

Chen Tong

1

| Jing Chen

2

| Mike J. Buckle

3

1

School of Management, Swansea

University, Singleton Park, Swansea SA2

8PP, UK

2

School of Mathematics, Cardiff

University, Cardiff, UK

3

Management School, Liverpool

University, London, UK

Correspondence

Chen Tong, School of Management,

Swansea University, Singleton Park,

Swansea SA2 8PP, UK.

Email: chen.tong1ct@outlook.com

JEL Classification: G12; G15; G18; C32;

F02; C45

Abstract

This paper applies a visualized network approach to analyse the degree of inte-

gration or comovement among global equity markets. We utilize daily prices of

stock market indices of 57 countries from January 1997 to August 2012 to estab-

lish both a minimum spanning tree network and graph network. We examine

network features including connectivity and centrality through a variety of

indicators. Our results clearly show that there has been a tendency over time

for markets to become more integrated globally even during periods of market

stress. The centrality results suggest that the U.S. and Hong Kong markets have

been the most dominant markets in their geographic region. For Europe, we

find 3 dominant centres, the United Kingdom, France, and Germany in con-

trast to previous literature suggesting that the United Kingdom was the domi-

nant country. We further identify that Japan and Australia, instead of acting

as dominant countries in their region, serve as bridging countries between

the region and the rest of the world. Finally, we find that Africa does not form

a cluster, and individual African countries tend to connect to other developed

markets.

KEYWORDS

integration, network, stock market

1|INTRODUCTION

Global financial market integration is an important topic

in finance, and the degree to which financial markets

are interconnected has an important bearing on informa-

tion shock transmission and financial stability. The 2008

global financial crisis revealed both the benefits and costs

of greater financial market integration. It could be argued

in a financially integrated global market that risks can

spread and spill over to other segments of the financial

market, increasing the likelihood of contagion of financial

fragilities. Lane and Milesi‐Ferretti (2010), Crotty (2009),

Huyghebaert and Wang (2010), and Dabrowski (2010)

all examined the relationship between the degree of finan-

cial market integration in the world financial system and

contagion. The general finding is that the shock from a

crisis spreads more quickly in a highly integrated finan-

cial system than a less connected system. In contrast, an

individual financial market or segment inside a highly

integrated system appears to benefit from the high level

of integration when the financial system is recovering

from the turbulence.

Previous studies of financial market integration nor-

mally utilize econometric techniques (see Campbell &

Hamao, 1992; Forbes & Rigobon, 2002; Voronkova,

2004; etc.). Such approaches are restricted to examining

the dynamics of a system, and the modelling techniques

are commonly correlation‐based methods. One prominent

issue with such methods is that even perfectly

cointegrated variables do not necessarily guarantee high

Received: 29 June 2015 Revised: 6 September 2016 Accepted: 19 December 2017

DOI: 10.1002/ijfe.1617

296 Copyright © 2018 John Wiley & Sons, Ltd. Int J Fin Econ. 2018;23:296–314.wileyonlinelibrary.com/journal/ijfe

correlations. Additionally, in multifactor modelling, there

is the problem of choosing relevant factors. Moreover,

when a few countries are significantly driven by the same

economic factor, the R‐squares given by these models are

usually large, which leads to inflated cointegration results

(Pukthuanthong & Roll, 2009). Even where a positive

relationship between correlation and integration is found

(see Edison, Levine, Ricci, & Sløk, 2002; Forbes &

Rigobon, 2002; Bekaert, Harvey, & Ng, 2005; Bekaert,

Hodrick, & Zhang, 2009), we still cannot conclude

whether such results are unbiased.

In this paper, we propose to use a network visualiza-

tion approach to examine global stock market integration.

According to Allen and Babus (2008), mapping a financial

network is “the first and crucial step in understanding a

modern financial system.”In general, a network consists

of a large group of entities including both core and non-

core groups that are distinguished by the level of connec-

tivity of the entity to others. The connection between

entities can be direct or indirect through paths.

Our study is based on the rationale that a social type

network can capture connections within the global finan-

cial system and hence provide an alternative approach to

understanding market integration. We construct correla-

tion‐based distance measures for 57 stock market indices

over a 21‐year period and identify the connections

between them. Using this approach, we examine the rela-

tion between the dynamics of the network structure and

significant market events (such as financial crises). We

are also able to measure important network characteris-

tics such as the degree of connectivity, clustering, and

centrality.

1

In this study, we utilize two approaches from the net-

work literature: minimum spanning tree networks

(MSTNs) and graph networks (GNs; see Onnela, Kaski,

& Kertész, 2004). Each approach has its own distinct

advantages, but they also complement each other. Both

MSTNs and GNs separate the core from the noncore

countries in the network. The GN's most prominent

advantage is to visually reveal financial clusters, whereas

the MSTN is good at revealing the connectivity of

markets.

The first main finding of our study is that there is

increasing integration of stock markets throughout the

sample period occurring at the regional level. Since the

2008 crisis, the level of integration in the global market

has greatly increased. We also note that after the 2008 cri-

sis, the European region/cluster initially reduced its con-

nectivity to the United States before moving back to

precrisis levels. Another key finding is that the regional

financial centres appearing in the GN network provide

evidence to support the new economic geography with a

shift in dominance away from the United States and

West‐Europe towards Asia and some other fast develop-

ing countries and regions. Finally, at the individual coun-

try level, we find interesting evidence that although

countries like Japan and Australia are important in their

region in terms of size, connectivity, and importance coef-

ficient (IC), they are not centred in the Asia‐Pacific clus-

ter. Instead, they appear to act as bridging countries

between the region and the rest of the world. For Africa,

although many financial centres such as South Africa

are growing quickly, we do not find them to be highly sig-

nificant in the global financial network or to form a

regional cluster. Instead, African countries seem to con-

nect to developed markets in other continents

individually.

This paper contributes to the existing literature on

financial market integration in four ways. First, we estab-

lish a financial network without relying on the assistance

of a real goods trading network to provide a distance mea-

sure. Using financial market data rather than macroeco-

nomic data to construct the network is less likely to

suffer from measurement error. A common measure of

distance in a network is the Gower distance measure

(Gower & Ross, 1969), which associates correlation to

Euclidean distance. A complete network showing every

relation (correlation) between markets is too rich, so we

filter to identify the important links using the minimum

spanning tree (MST) procedure. This selects the most

important links between nodes based upon minimum dis-

tance (maximum correlation). By selecting the shortest

distance (maximum correlation), we avoid the problem

of selecting negative correlations. In this paper, we modify

the measure to make it simpler while still keeping the

advantages and features of the original measure (see

Equation 4). In this new measure, the highly connected

countries that carry large correlations will have shorter

distance, and this will provide greater clarity in showing

the closeness between such markets. Second, we develop

two different types of financial networks with separate

indicators of connectivity and centrality—the IC and clus-

tering coefficient (CC). These different measures provide

complementary results to explain the degree of centrality

and structure of financial clusters. Through comparing

the IC and CC, we are able to obtain new results that

the past literature on integration has not identified. For

example, we find that Europe has had several centres of

importance over most of the sample period rather than

just the United Kingdom, which has been seen histori-

cally as the sole leading country. Third, we use the con-

structed networks to examine the impact of various

financial crises on countries both globally and regionally.

Using the network measures, we are able to observe a

common pattern of countries to move away from the orig-

inal trouble countries/regions during crises but move

TONG ET AL.297