Stock return dynamics and stock market interdependencies

Abstract

This article compares stock return behaviour in mature and emerging stock markets. The role of leading markets and the impact of the October 1997 East Asian financial crisis are examined in the context of stock market interdependencies. An extended AR(1)-GARCH-M (autoregressive generalized autoregressive conditional heteroskedasticity) model is used. Potential price and volatility transmission mechanisms stemming from leading markets and potential structural breaks in mean and variance caused by the above crisis are discussed. Daily data covering the 16 April 1991 to 29 November 2001 period are used. The results reveal significant differences in stock return behaviour between mature and emerging markets and confirm substantial interdependencies among stock markets, originating from both leading and emerging markets.

Acknowledgements

The author is indebted to Professors K.P. Prodromidis and N. Pittis for valuable help and guidance. Their comments and suggestions have improved the final version of this article. She would also like to thank Prof. E. Tsionas for helpful comments. The usual disclaimer applies.

Notes

¹ The classification of stock markets in mature and emerging and other relevant details are given in The Emerging Stock Market Factbook 2000 (2000), p. 2. The quantitative features basically refer to their trading activities and market participation. These can be evaluated by stock market indicators, such as market capitalization and value traded or their ratios to GDP (Levine and Zervos, 1998) and the number of listed companies. The qualitative characteristics include operational efficiency, quality of market regulation, supervision and enforcement, corporate governance practices, minority shareholder rights, transparency and level of accounting standards.

² The US and the Japanese stock markets are considered as the leading stock markets in terms of their performance concerning their quantitative and qualitative characteristics. Details on the histories of the New York Stock Exchange (NYSE) and the Tokyo Stock Exchange (TSE) are given in their Internet sites.

³ As a result of the escalation of capital withdrawals, losses to foreign investors and intense pressures on their national currencies (Radelet and Sachs, 1998), the stock markets of South East Asian countries (Indonesia, Korea, Malaysia, the Philippines and Thailand) experienced significant downward movements in their price indices. For example, the stock price indices for Malaysia (Kuala Lumpur Composite), Thailand (Bangkok S.E.T.) and Indonesia (Jakarta SE Composite) declined from 1012.840, 665.6200 and 721.2700, respectively, at the end of July 1997 to 664.6900, 447.2100 and 50.4180, respectively, at the end of October of the same year. The data is from Datastream.

⁴ Other financial variables (like interest rates, dividend yields, or various macroeconomic variables with a clear business cycle component) have also been found to account for return predictability (Pesaran and Timmermann, 1996).

⁵ Within the ARCH–GARCH framework, the (G)ARCH-M models, which specify the conditional mean as an explicit function of the conditional variance, focus on the risk-return tradeoff.

⁶ For MSMs, many of the relevant empirical studies have focused on modelling heteroskedasticity (Sokalska, 1997) on the assumption that the level of returns is difficult to predict in the ‘more efficient’ stock markets (relative efficiency; Campbell et al., 1997). They deal with the conditional variance and hence, with the ‘unpredictable’ part of stock returns. Still, in many cases it is assumed that the mean return exhibits a certain degree of predictability on the basis of past information. To that end, an adjustment procedure is employed for the removal of the predictable part of the return series (Akgiray et al., 1989; Pagan and Schwert, 1990a; Engle and Ng, 1993, Shields, 1997).

⁷ Other variables, like interest rates, are also found to occasionally influence stock return volatility (Glosten et al., 1993).

⁸ GARCH and ARCH models are sometimes classified as long and short memory models, respectively.

⁹ The interest on the issue of volatility persistence stems from an argument relating to stock market movements, i.e. to variation in share prices. According to this argument, these significant movements are attributed to changes in risk premia, which are induced by movements in stock return volatility. For this to hold, the influence of changing stock return volatility must be persistent, i.e. shocks to volatility have to persist for a long time. See Poterba and Summers (1987).

¹⁰ Within the GARCH framework, the sum of the ARCH and GARCH coefficients is used as a succint measure of the degree of persistence of shocks to volatility. In the case that this sum equals one (implying that current information remains important for forecasts of the conditional variance for all horizons), the unconditional variance does not exist (Fama 1963; Mandelbrot 1963; Pagan and Schwert 1990b).

¹¹ Not accounting for a shift in the unconditional variance caused, for example, by a regime change or other local and/or global effects, may give rise to the impression that a shock has a very long lasting or even permanent effect on the variance. It is shown (Tzavalis and Wickens 1995; Franses 1995; Aggarwal et al. 1999) that the estimates for the degree of volatility persistence are significantly reduced by including dummy variables to allow for changes in the unconditional variance.

¹² This class of work belongs to the wider category of studies examining the influence of factors other than local (world information variables, global factors) on stock return behaviour [for instance, Harvey (1995) and Bekaert and Harvey (1997).

¹³ The Australian and the Norwegian price index series consist of 2729 observations (16 April 1991 to 29 September 2001) and the Danish series of 2653 ones (16 April 1991 to 14 June 2001). It is worth noting that quite a few stock markets emerged in the 1990s thanks to tremendous developments with regard to liberalization processes. The question on how influential these developments have been for the stock return determination process, albeit an important one, is beyond the scope of this article.

¹⁴ The phenomenon of excess kurtosis in stock returns investigations is an empirical regularity. Details are given in the early work of Mandelbrot (1963) and more recently in Andreou et al. (2001).

¹⁵ For the basic features of the ARCH, GARCH, (G)ARCH-M and IGARCH processes and their applications to stock return data, see for instance, Engle (1982), Bollerslev (1986), Engle and Bollerslev (1986), Engle et al. (1987), Schwert (1989), Schwert and Seguin (1990), Bollerslev et al. (1992), Hamilton (1994) and Pagan (1996).

¹⁶ To determine the correct US and Japanese return series used in terms of trading time for each country, the respective local time as well as the exact trading hours for each stock market were taken into account in relation to the US and the Japanese time and trading hours. A detailed correspondence of the time subscripts for all stock markets under consideration is given in the Appendix, Table A4.

¹⁷ For references to non-synchronous trading see Sentana and Wadhwani (1992), Koutmos (1997) and Fortune (1998), to time-varying expected returns see Sentana and Wadhwani (1992), Koutmos (1997) and Patelis (1997), and to positive feedback trading see Sentana and Wadhwani (1992), Cuthbertson (1996) and Koutmos (1997).

¹⁸ In case of the presence of almost equally strong contemporaneous and temporal volatility spillover effects with alternating signs in stock market i, a possible adjustment and/or offsetting mechanism could take place between periods t and t − 1.

¹⁹ In the cases in which the contemporaneous and temporal volatility spillover effects are roughly equal in size but opposite in signs in stock market i, the underlying effect can be said to be adjusted or even offset between periods t and t − 1.

²⁰ According to the results of the additional investigation conducted on volatility persistence and IGARCH effects, the measures of volatility persistence and the existence of IGARCH effects would have been enhanced in the absence of the additional regressors and the dummy variable (accounting for a one-time change) from the variance equation. See Tables A7 and A8 in the Appendix.