Dynamic correlation analysis of financial contagion: Evidence from Asian markets

https://doi.org/10.1016/j.jimonfin.2007.06.005Get rights and content

Abstract

We apply a dynamic conditional-correlation model to nine Asian daily stock-return data series from 1990 to 2003. The empirical evidence confirms a contagion effect. By analyzing the correlation-coefficient series, we identify two phases of the Asian crisis. The first shows an increase in correlation (contagion); the second shows a continued high correlation (herding). Statistical analysis of the correlation coefficients also finds a shift in variance during the crisis period, casting doubt on the benefit of international portfolio diversification. Evidence shows that international sovereign credit-rating agencies play a significant role in shaping the structure of dynamic correlations in the Asian markets.

Introduction

During the period from July 1997 through early 1998, Asian financial markets experienced a series of financial distresses, which spread rapidly and sequentially from one country to another in a short interval of intense crises. Later on, it spread further to Russia and Latin America. The short-term damage of the crisis not only caused asset prices to plunge across these markets but also created speculative runs and capital flight, leading to considerable financial instability for the entire region. A longer-run consequence triggered by the crisis and its spillover effect was that it brought about a dramatic loss of confidence for investors who had intended to invest in Asian markets, jeopardizing the economic growth of the region. Such a shift in the attitudes of investors may produce prolonged damage to portfolio investments because their concerns may not subside until another successful story of economic growth in the region develops, and that may take a long time. As such, academic researchers and policy makers alike have paid close attention to identify the channels of shock transmission across countries and to measure the damaging impact of crises on the environment for investments in Asian markets.

Since the financial shocks and the contagion process in the Asian-crisis episode were attributable to a variety of factors beyond economic linkages, many researchers have focused on financial contagion by providing evidence of significant increases in cross-country correlations of stock returns and/or volatility in the region (Sachs et al., 1996). Yet, the existence of contagion in relation to the crisis remains a debatable issue. Some studies show a significant increase in correlation coefficients during the Asian crisis and conclude that there was a contagion effect (Baig and Goldfajn, 1999). Other researchers find that after accounting for heteroskedasticity, there is no significant increase in correlation between asset returns in pairs of crisis-hit countries, reaching the conclusion that there was “no contagion, only interdependence” (Forbes and Rigobon, 2002, Bordo and Murshid, 2001, Basu, 2002).1 However, in their tests for financial contagion based on a single-factor model, Corsetti et al. (2005) find “some contagion, some interdependence.” Further, focusing on different transmission channels, Froot et al. (2001) and Basu (2002) confirm the existence of the contagion effect.2 Thus, the evidence on the financial contagion is not conclusive.

The existing literature on the empirical research of financial contagion has several limitations and drawbacks. First, there is a heteroskedasticity problem when measuring correlations, caused by volatility increases during the crisis. Second, in addition to a lagged dependent variable, an omitted variable problem arises in the estimation of cross-country correlation coefficients due to the lack of availability of consistent and compatible financial data in Asian markets. Third, since contagion is defined as significant increases in cross-market co-movements, while any continued market correlation at high levels is considered to be interdependence (Forbes and Rigobon, 2002), the existence of contagion must involve evidence of a dynamic increment in correlations. Thus, the dynamic nature of the correlation needs to be sorted out. Fourth, a common problem encountered by these studies is the fact that virtually all of the tests are affected by identifying the source of crisis and the choice of window length (Billio and Pelizzon, 2003). Moreover, the choice of sub-samples conditioning on high and low volatility is both arbitrary and subject to a selection bias (Boyer et al., 1999).3 Fifth, it is generally recognized that indicators of sovereign creditworthiness represented by sovereign credit ratings announced by international credit-rating agencies and publications are based on economic fundamentals; the changes in ratings are perceived to reflect an external assessment of the risk associated with changes in economic fundamentals or political risk, which should have an impact on stock returns and, in turn, the correlation coefficients (Beers et al., 2002, Kaminsky and Schmukler, 2002).4 Sovereign rating downgrades in one country may create an international contagion effect through the wake-up call to neighboring countries that have similar macroeconomic environments and the cross-market hedging channels. Baig and Goldfajn (1999) find an increase in the correlations in the sovereign spread during the crisis periods as compared to tranquil periods. Their analysis, however, lacks dynamic elements and fails to provide a systematic framework to capture the intervention from credit-rating changes.

To overcome the limitations found in the existing literature, this paper employs a cross-country, multivariate GARCH model, which is appropriate for measuring time-varying conditional correlations. This methodology will enable us to address the heteroskedasticity problem raised by Forbes and Rigobon (2002) without arbitrarily dividing the sample into two sub-periods.5 In the meantime, using lagged U.S. stock returns as an exogenous factor and estimating the system simultaneously help us to resolve the omitted variable problem and, at the same time, to account for the global common factor.6 More important, the model provides a mechanism to trace the time-varying correlation coefficients for a group of Asian stock markets. Analyzing the derived time series of correlation coefficients allows us to detect dynamic investor behavior in response to news and innovations. Particularly, our empirical analysis provides new evidence of the significant impact of sovereign credit-rating changes around the announcement dates, domestic and foreign, on cross-country correlation coefficients of stock returns in the Asian countries. This new insight will be informative for global investors, helping them to make better decisions with regard to asset and risk management, including asset allocation, portfolio diversification, and hedging strategy (Fong, 2003).

The major findings of this paper are summarized as follows. First, this study, which uses a longer data span, finds supportive evidence of contagion during the Asian-crisis period, resolving the puzzle of “no contagion, only interdependence” reported by Forbes and Rigobon (2002). Second, two different phases of the Asian crisis are identified. The first phase, from the start of the crisis to November 17, 1997, entails a process of increasing volatility in stock returns due to contagion spreading from the earlier crisis-hit countries to other countries. In this phase, investor trading activities are governed mainly by local (country) information. However, in the second phase, from the end of 1997 through 1998, as the crisis grew in public awareness, the correlations between stock returns and their volatility are consistently higher, as evidenced by herding behavior. Statistical analysis of correlation coefficients shows shifts in the level as well as in the variance of correlations, casting some doubt on the benefit of international portfolio diversification during the crises. Third, after controlling the variables involved in the crisis period, we find that the correlation coefficients respond sensitively to changes in sovereign credit ratings. This indicates that both market participants and financial credit-rating agents have their own dynamic roles in shaping correlation coefficients.

The remainder of the paper proceeds as follows. Section 2 describes the data and statistics of stock returns. Section 3 examines the correlation coefficients based on a simple-correlation analysis by adjusting the impact of volatility during different sample periods. Section 4 presents a multivariate GARCH model and discusses its application to our context. Section 5 reports the estimation results and tests the time-varying correlation coefficients in response to different shocks. Section 6 contains conclusions.

Section snippets

Data and descriptive statistics

The data used in this study are daily stock-price indices from January 1, 1990, through March 21, 2003, for eight Asian markets that were seriously affected by the 1997 Asian financial crisis. The data set consists of the stock indices of Thailand (Bangkok S.E.T. Index), Malaysia (Kuala Lumpur SE Index), Indonesia (Jakarta SE Composite Index), the Philippines (Philippines SE Composite Index), South Korea (Korea SE Composite), Taiwan (Taiwan SE Weighted Index), Hong Kong (Hang Seng Index), and

Correlation analyses

Since correlation analysis has been widely used to measure the degree of financial contagion, it is convenient to start our investigation by checking the simple pair-wise correlation between the stock returns for the markets under investigation. However, correlation coefficients across markets are likely to increase during a highly volatile period. That is, if a crisis hits country A with increasing volatility in its stock market, it will be transmitted to country B with a rise in volatility

The dynamic correlation-coefficient model

The multivariate GARCH model proposed by Engle (2002), which is used to estimate dynamic conditional correlations (DCC) in this paper, has three advantages over other estimation methods.

Estimates of the model

Table 3 reports the estimates of the return and conditional variance equations. The AR(1) term in the mean equation is significantly positive for Thailand, Indonesia, Malaysia, Philippines, and Singapore, while it is significantly negative for Hong Kong and Japan. This finding is in agreement with the evidence in the literature in that the AR(1) is positive in emerging markets due to price friction or partial adjustment and that AR(1) is negative as the presence of positive feedback trading in

Conclusions

This paper investigates the relationship between the stock returns of various crisis-hit markets during the 1997–1998 Asian financial crises. Heteroskedasticity-adjusted simple-correlation analysis with an extended length of window as well as dynamic correlation analysis concludes that there is evidence of contagion effects during the Asian financial crisis, a finding that does not agree with the “no contagion” conclusion reached by Forbes and Rigobon (2002).

While examining stock-market

Acknowledgments

We thank the co-editor, Michael Melvin, and an anonymous referee for valuable comments on a previous version, although we alone are responsible for any errors that may remain. The paper was presented at the 2005 ASSA meeting in Philadelphia. Huimin Li would like to acknowledge the summer research funding from the College of Business and Public Affairs, West Chester University. Thomas C. Chiang would like to thank the research support received from the Marshall M. Austin fund, LeBow Business

References (34)

  • K.-H. Bae et al.

    A new approach to measuring financial contagion

    Review of Financial Studies

    (2003)
  • T. Baig et al.

    Financial market contagion in the Asian crisis

    IMF Staff Papers

    (1999)
  • Basu, R., 2002. Financial contagion and investor ‘learning’: an empirical investigation. IMF working paper no....
  • D.T. Beers et al.

    Sovereign credit ratings: a primer. Standard & Poor's sovereign reports (April 3)

  • M.D. Bordo et al.

    Are financial crises becoming more contagious? What is the historical evidence on contagion?

  • Boyer, B.H., Gibson, M.S., Loretan, M., 1999. Pitfalls in tests for changes in correlations. Federal Reserve Board...
  • S. Claessens et al.

    Contagion: why crises spread and how this can be stopped

  • Cited by (645)

    • The impact of the Russia–Ukraine war on volatility spillovers

      2024, International Review of Financial Analysis
    • Dynamics of market states and risk assessment

      2024, Physica A: Statistical Mechanics and its Applications
    View all citing articles on Scopus
    View full text