Journal of International Money and Finance 25 (2006) 528e550

www.elsevier.com/locate/econbase

Linkages between extreme stock market and currency returns

Phornchanok Cumperayot a,*, Tjeert Keijzer b, Roy Kouwenberg

a Chulalongkorn University, Faculty of Economics, Bangkok 10330, Thailand

b AEGON Asset Management NL, AEGONplein 20, 2591 TV, The Hague, Netherlands

c Asian Institute of Technology, School of Management, P.O. Box 4, Pathumthani 12120, Thailand

Abstract

We investigate the link between extreme events on the currency and stock markets for 26 countries by estimating a simultaneous equations probit model, using a sample of 2500 daily returns in the period from 1996 to 2005. In a number of emerging markets that went through a period of crisis an extreme stock mar­ket decline increased the probability of extreme currency depreciation on the same day. For currency mar­kets we find evidence of spillover of extreme events within regions, but limited influence outside the region. Extreme events on stock markets are much more interrelated globally, particularly when they orig­inate from the US. ? 2006 Elsevier Ltd. All rights reserved.

JEL classification: F31; F37; G15; C35

Keywords: Currency market; Stock market; Extreme events; Spillover

1. Introduction

The Asian crisis of 1997 sent the region into a prolonged period of currency depreciations and stock market declines, with grave consequences for the real economy during the subsequent years. As the crisis unfolded initially in Thailand with the devaluation of the Baht in July,

* Corresponding author. Tel.: +66 2 218 6241; fax: +66 2 218 6201. E-mail address: (P. Cumperayot).

0261-5606/$ - see front matter ? 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.jimonfin.2006.01.007


P. Cumperayot et al. / Journal of International Money and Finance 25 (2006) 528e550 529

policy makers in countries that were still unaffected by the crisis e such as Malaysia and Korea e still had some time to react and could have taken measures to protect their currencies. However, at that time a full-blown regional crisis was not anticipated. A relevant question is whether policy makers in Asia could have used the rapidly increasing number of extreme events in the region, and the increased volatility in their local stock markets, as early warning signals of further extreme currency depreciations in the region. In turn, could investors have anticipated the effects of what began as a currency crisis on global equity markets in general and Asian equity markets in particular?

This study investigates the empirical link between extreme returns in local stock and cur­rency markets and the pattern of spillover within and across regions.1 By focusing on extreme events, we aim to isolate the effect of true market shocks from the ‘‘normal’’ pattern of returns, in order to gain insight into the mechanics of possible cross-market spillover effects in extraor­dinary market environments. The first question that we would like to answer is whether extreme stock market returns and extreme currency market returns tend to occur simultaneously and whether one market leads another. If the latter is the case, can the occurrence of an extreme return in the stock (currency) market then be used to predict future extreme returns in the cur­rency (stock) market? And apart from these effects in local markets, do we find evidence of spillover of extreme returns from one region to another?

To address these questions we estimate a country-specific bivariate simultaneous equations probit model, with the probabilities of extreme currency and stock market events as dependent variables. We focus on extreme events across a large selection of countries (26) and a long sam­ple of daily data (2500 days). A number of authors have applied bivariate or multivariate ex­treme value theory (EVT) to study spillover behavior of extreme stock market returns, such as Straetmans (2000), Longin and Solnik (2001) and Poon et al. (2004). Hartmann et al. (2003a,b) apply EVT to extreme returns and spillover in currency markets, while Hartmann et al. (2004) address the link between extreme returns in stock markets and bond markets. We are not aware of any papers in the multivariate EVT literature that focus on the link between stock market extremes and currency market extremes.

The approach to modelling extreme linkages applied in this paper is, however, not based on EVT, but on limited dependent variable models, similar to Bae et al. (2003). Whereas Bae et al. (2003) use the total number of extreme events in a region as the dependent variable in order to study contagion effects in global stock markets, we try to predict the occurrence of extreme events for 26 individual countries. Moreover, we investigate the contemporaneous link between extreme events in stock markets and currency markets. Other papers related to our research include Kaminsky and Reinhart (1999) and Glick and Hutchison (2001), who study the problem of linkages between currency and banking crises. These papers focus on low-frequency data (monthly and annually, respectively) and use macroeconomic variables to explain and/or predict currency and banking crises. Glick and Hutchison (2001) model the linkage between currency and banking crises with a simultaneous equations probit model. Phylaktis and Ravazzolo (2005) study short- and long-run dynamic links between stock prices and exchange rates in the Pacific Basin region in a cointegration relationship framework using monthly data. Our paper complements the work of Phylaktis and Ravazzolo

1 Financial spillover occurs when extreme market returns in one specific market or country trigger a similar extreme event in a different, yet adjacent, market or country. Due to the ambiguity surrounding the definition and measurement of the related concept ‘‘contagion’’ (see, e.g. Dornbusch et al., 2000; Karolyi, 2003), we focus exclusively on spillover effects of extreme returns.


530 P. Cumperayot et al. / Journal of International Money and Finance 25 (2006) 528e550


(2005) by focusing on extreme events, using daily data for 26 developed and emerging markets.

The empirical results presented in this paper show that an extreme stock market decline increased the probability of an extreme currency depreciation on the same day in a number of emerging markets that went through a period of crisis. The opposite effect, from cur­rency depreciation to stock market decline, is not supported strongly by the data. In a num­ber of countries an extreme depreciation actually decreased the probability of a stock market decline, probably reflecting the positive effect of currency depreciation on exports and growth. For currency markets we find evidence of spillover of extreme events within regions, but limited effects outside the region. Extreme events on stock markets are much more interrelated globally, with events in the region America having the strongest im­pact worldwide.

This paper proceeds as follows: Section 2 describes the research methodology and modelling approach adapted in this paper, as well as our definition of extreme events. Section 2 also re­views the theoretical literature motivating our empirical work. Section 3 provides a description of the data, the procedure for generating out-of-sample conditional volatility predictions and an analysis of conditional extreme event probabilities, providing a first impression of the charac­teristics of the data. In Section 4, we present the estimation results of the simultaneous equa­tions probit models. Section 5 summarizes results and concludes the paper.

2. Methodology

Since we aim to study market linkages in times of stress, we first need to specify our def­inition of extreme returns. Similar to Bae et al. (2003), we select the a%-percentile — with a located in the tail area (e.g. a = 1%, 2.5%, or 5%) — of the return distribution as the threshold for the definition of extremes. We focus on the left tail of the stock market return distribution, i.e. the a% most negative stock market returns, and study the right tail of the currency returns, i.e. the a% largest increases of the foreign currency price of 1 US dollar, representing depre­ciation (or devaluation) of the local currency. Typically, periods of economic crisis involve a weakening of the local currency and for this reason we ignore the left tail of the currency return distribution, i.e. extreme appreciation or revaluation.

The binary variable yt ,c i indicates the occurrence of an extreme currency return in country i on day t:

y c ti = 1, if the currency return (measured in local currency per US dollar) of country i on day t is above the (1 — a)%-percentile threshold,

y

c; i ?

0; otherwise.

Similarly, the binary variable yt ,si indicates the occurrence of an extreme stock market return in country i on day t:

yst;i ? 1; if the stock market return in local currency on day t in country i is below the a%-percentile threshold, yst;i ? 0; otherwise.

We model the probability of the occurrence of extreme stock and currency market events with a simultaneous equations probit model for each country i. This model allows us to estimate


P. Cumperayot et al. / Journal of International Money and Finance 25 (2006) 528e550 531

to what extent extreme returns in the currency market and the stock market of a country are linked to each other and to a number of additional explanatory variables.2 The simultaneous equations probit model is defined as follows:

yc = g1 iyTi + b1'i j-^im + ￡1 ti ! for each country i and t= 1,2, ...,T (1)

yt*i = g2 iy*c i + b2iX 2 tM + e2ti! for each country i and t = 1,2,..., T (2)

with y c ti = 1, ifyt*i>0,

y c ti = 0, otherwise,

and y s ti = 1, if y*s > 0,

y s ti = 0, otherwise.

The matrix X1 ti (X 2t,i ) contains observations of k1 (k 2 ) exogeneous explanatory variables at time t for predicting currency (stock) market crises in country i. The variables e1 t , i and e2t , i are the disturbance terms of Eqs. (1) and (2), which are assumed to be identically and indepen­dently distributed (IID).

Note that Eqs. (1) and (2) are not estimated separately with two probit-regressions: this would lead to biased and inconsistent parameter estimates due to the presence of endogenous variables on the right-hand-side of the equations. An instrumental variable estimation technique can be applied to circumvent this problem: the simultaneous equations probit model above can be estimated with a two-stage estimation procedure, as described in Maddala (1983; pp. 246—247). The covariance matrix of the estimated coefficients is non-standard and is adjusted accordingly in the estimation procedure (see Maddala, 1983, p. 247).

The probit model allows us to estimate the contemporaneous relationship between extreme events in the currency market and the stock market. The literature provides diverse theories concerning the precise nature of this relationship (see, e.g. Phylaktis and Ravazzolo, 2005). We first concentrate on the link from exchange rate depreciation to the stock market. The cur­rent account is one channel through which a strong exchange rate depreciation (or devaluation) can affect the probability of a stock market decline. A currency depreciation increases the local currency value of exports and export competitiveness, leading to increased economic growth and corporate profits (see, e.g. Junz and Rhomberg, 1973). The stock market reflects future profit growth (see Fama, 1990; Schwert, 1990) and a strong depreciation can therefore lead to a lower probability of a simultaneous extreme stock market decline.3,4

A large unanticipated depreciation or devaluation can also have a negative impact on the economy, if domestic banks and firms have large amounts of unhedged foreign currency de­nominated debt. The increased debt burden leads to a decline in investment and economic ac­tivity (Mishkin, 1996). A currency collapse can also undermine an already ailing banking sector, leading to a combined currency and banking crisis (a twin crisis, see Kaminsky and Reinhart, 1999). Not every strong currency depreciation in emerging markets leads to a banking crisis or a reduction of future economic activity. Following an extreme currency market event,

2 See Glick and Hutchison (2001) for an application of the simultaneous equations probit model to twin banking and
currency crises.

3 As stock prices are the present value of all future dividends, we expect limited impact of the short-term J-Curve
effect (see, e.g. Magee, 1973).

4 An exception to this relationship might be trade-oriented countries where exports possess a significant import con­
tent, such as Singapore (Abeysinghe and Yeok, 1998).


532 P. Cumperayot et al. / Journal of International Money and Finance 25 (2006) 528e550

however, foreign investors might not yet know (or disagree on) whether the depreciation will be beneficial or detrimental to the economy in the longer term. The increased level of uncertainty may lead to a higher country risk premium and as a result, lower equity prices.5

We now turn our attention to the link from extreme stock market declines to the currency market. In countries with an open, relatively small and undeveloped domestic capital market, portfolio flows of foreign investors can exert a strong influence on the local stock market (Be-kaert et al., 2002). If foreign investors drastically lower their return expectations or increase their risk assessment of the local stock market, this could lead to large portfolio outflows and a stock market decline. In turn, portfolio outflows can lead to excess supply of the local currency and depreciation of the exchange rate, as illustrated in the theoretical model of Hau and Rey (2006). Froot and Ramadorai (2002) and Hau and Rey (2004) provide empirical evi­dence of the positive relationship between portfolio flows and currency returns. Hence, in emerging and small developed markets, we would expect extreme stock market declines to be linked to an increased probability of extreme currency depreciation through foreign investor portfolio flows. For large developed markets, however, where foreign investors are less domi­nant in the capital markets, we do not anticipate the link between extreme stock market declines and currency depreciation with the same conviction. This is due to the many economic factors that can influence the relationship, such as differences in inflation rates, domestic growth versus growth abroad, imports and exports.

3. Data, volatility and conditional probabilities of extreme events

3.1. Data

We use daily data on local stock index returns and exchange rates from the period 3 July 1995 through 29 July 2005, a sample of 2630 daily return observations.6 In total 26 countries are included in our study, consisting of 17 emerging markets and nine developed markets. We use S&P/IFCI stock indices in local currency for the emerging markets, which are available on a daily basis starting from 30 June 1995. FTSE World stock indices in local currency represent the developed stock markets. The FTSE World index series are value-weighted total return in­dices. Daily exchange rates for all countries versus the US dollar are from Datastream (original source: WM Company/Reuters). We measure exchange rates relative to the US dollar, as the US dollar is the main currency for invoicing international exports and imports.7 For the United States, we use the FED Nominal Broad Dollar Index.