Volatility Contagion in the Asian Crisis: New Evidence of Volatility Tail Dependence

• Author: Rodrigo Herrera,Alexander Karmann
• DOI: http://doi.org/10.1111/rode.12089
• Published date: 01 May 2014
• Date: 01 May 2014

Volatility Contagion in the Asian Crisis:

New Evidence of Volatility Tail Dependence

Alexander Karmann and Rodrigo Herrera*

Abstract

We analyze empirically the existence and the extent of financial contagion by means of extreme value

theory in the Asian crisis. We consider two key markets, the stock exchange and the foreign exchange using

daily data in the period 1992–2001. We present several notions of financial contagion as a significant change

in volatility tail dependence (VTD) among different assets. To this end, we introduce a semiparametric

VTD estimator in the framework of regularly varying strictly stationary time series. Our analysis provides

mixed evidence with respect to the “interdependence vs contagion” dispute. Within-country contagion is

more likely to hold than across-country contagion. Because the latter is typically symmetric, contagion in

stocks and foreign exchange coincide, in line with “portfolio rebalancing” arguments. Across-market conta-

gion supports the “wake up call” argument of loss of confidence, as small countries’ currency markets affect

contagiously the stock markets of the larger economies.

1. Introduction

The Asian Crisis 1997/98 provoked many economists to analyze the regional eco-

nomic and especially financial markets to better understand the linkages across

markets and across countries. A common understanding is its regional character with

non-significant impact to large economies outside the region, such as the USA or the

EU (see Arestis and Glickman, 2002), and the prominent role of regional financial

linkages across markets as well as across countries, mainly attributed to portfolio

rebalancing effects of financial investors. There is still an ongoing debate of the

Forbes and Rigobon (2002) type “no contagion, only interdependence” and how to

identify extreme co-movements of financial markets. Finally, in terms of extreme

value theory (EVT), what can be said about the impact of economic or institutional

characteristics on financial linkages during the Asian crisis.

In the literature there are different approaches to define contagion and distinguish-

ing this concept from interdependence. The most accepted definition is the Forbes

and Rigobon (2002) approach. They defined contagion as a significant increase in

co-movement of markets after initial shock. Our definition of contagion is based on

the Forbes and Rigobon (2002) approach. Contagion in this paper is interpreted as a

significant increase in the tail dependence function of squared log returns, as a proxy

to the volatility tail dependence (VTD) that takes place during a turmoil period. In

contrast, if two markets show high tail dependence of squared log returns during the

period of stability and continue to be highly tail dependent after a shock to one

* Herrera: Facultad de Economía y Negocios, Universidad de Talca, Avenida Lircay s/n, Talca, Chile.

Tel: + 56-75-2200310; Fax: + 56-75-2200358; E-mail: rodriherrera@utalca.cl. Karmann: Chair of Monetary

Economics, Technische Universitaet Dresden, Dresden, Germany. Herrera thanks the Chilean Agency

CONICYT for their financial support. FONDECYT No.: 11110247.

Review of Development Economics, 18(2), 354–371, 2014

DOI:10.1111/rode.12089

© 2014 John Wiley & Sons Ltd

market, this constitutes interdependence. Thus, the definition of contagion risk is

based on the idea of high volatility change.

The concept of (volatility) contagion and in particular VTD is not new in finance.

For instance Edwards and Susmel (2001) using weekly stock market data for a group

of Latin American countries analyze the behavior of volatility through time. They

find strong evidence of volatility co-movements across countries, especially among

the Mercosur countries. Baur (2003) proposes a new test that is based on a regression

model that differentiates between mean contagion and volatility contagion in an

asymmetric way. Empirical results for 11 Asian stock markets show that there is

mean and volatility contagion in the Asian crisis. Kim et al. (2003) show that before

the Asian crisis, the Asian economies were more volatile than G-7 countries.

However, comovement and persistence properties of business cycles in the Asian

countries were very similar to those of the group of seven developed nations’ (G7)

economies. Brailsford et al. (2006) investigate risk and return in the banking sector in

Taiwan, China and Hong Kong. The study focuses on the risk–return relation in a

conditional factor generalized autoregressive conditional heteroskedasticity-in-mean

(GARCH-M) framework that controls for time-series effects. Finally, the study pro-

vides evidence on these relations before and after the Asian financial crisis. Dungey

et al. (2010) propose an identified structural GARCH model to disentangle the

dynamics of financial market crises. They apply this method to data of the 1997–98

Asian financial crises which consists of a complex set of interacting crises. They find

significant hypersensitivity and contagion between markets and show that links may

strengthen or weaken. Chiang and Wang (2011) propose a new approach to evaluate

volatility contagion in financial markets. They find empirical evidence of contagion

from the USA to other G7 stock markets owing to the subprime mortgage crisis.

Empirical evidence shows that volatility is contagious from the US market to several

markets examined. Engle et al. (2012) model the interrelations of equity market

volatility in eight East Asian countries before, during and after the Asian currency

crisis. Using a new class of asymmetric volatility multiplicative error models based on

the daily range, they find that dynamic propagation of volatility shocks occurs

through a network of interdependencies and shocks originating in Hong Kong may

be amplified in their transmission throughout the system, posing greater risks to the

region than shocks originating elsewhere.

To estimate volatility contagion two main classes of volatility models are used; the

GARCH and the stochastic autoregressive volatility (SARV) models. While a

GARCH model allows one to incorporate volatility spillovers in the model, it does

not allow one to incorporate volatility in all markets at once and it does not allow for

the endogeneity of all return or volatility measures as SARV models do. Using

SARV models, to study extreme-value dependence during periods of turmoil, may

lead to results deviating from a GARCH analysis when examining the volatility of a

time series. For instance, Davis and Mikosch (2009a) show there is no extremal clus-

tering for SARV processes in either the light- or heavy-tailed cases, in contrast to the

situation for GARCH processes (Davis and Mikosch, 2009b). That is, extreme events

do not tend to cluster in time, or alternatively, large sample behavior of maxima of a

non independent and identically distributed (iid) sequence is the same as that of the

maxima of the associated iid sequence.

In this paper we will consider that the asset markets returns analyzed are strictly

stationary sequences of random vectors whose finite-dimensional distributions are

jointly regularly varying with some positive tail index. Under the theory of regular

variation in EVT only the tail of financial returns is modeled, i.e. the estimation uses

VOLATILITY CONTAGION IN THE ASIAN CRISIS 355

© 2014 John Wiley & Sons Ltd