Examining Stock Volatility in the Segmented Chinese Stock Markets: A SWARCH Approach

Abstract

This study adopts the SWARCH model to examine the volatile behavior and volatility linkages among the four major segmented Chinese stock indices. We find strong evidence of a regime shift in the volatility of the four markets, and the SWARCH model appears to outperform standard generalized autoregressive conditional heteroskedasticity (GARCH) family models. The evidence suggests that, compared with the A-share markets, B-share markets stay in a high-volatility state longer and are more volatile and shift more frequently between high- and low-volatility states. In addition, the relative magnitude of the high-volatility compared with that of the low-volatility state in the B-share markets is much greater than the case in the two A-share markets. B-share markets are found to be more sensitive to international shocks, while A-share markets seem immune to international spillovers of volatility. Finally, analyses of the volatility spillover effect among the four stock markets indicate that the A-share markets play a dominant role in volatility in Chinese stock markets.

Keywords:

• Markov-switching ARCH
• volatility
• market segmentation
• Chinese stock markets
• volatility spillover

Notes

¹ It would be good to match the time span of the four markets. However, when we tried the empirical modeling, we found an extremely erratic movement in price and trading volume on the A-share markets before 1995. This is a widely observed scenario in a newly established emerging stock market, like domestic A-share stock markets in China. Incomplete regulations, frequent market reforms and extreme speculation by domestic investors cause a weird movement in the stock data, which makes modeling the data unreliable. Therefore, we deleted the earlier period of data for the SHA and the SZA in our analysis.

² We have tried the cases for three ARCH terms and above but only SZB got convergence. The third ARCH term we obtained is almost zero and insignificant for all cases. In addition, there are only a few times in the literature when ARCH terms are set to be higher than two. Thus, we set the maximum number of ARCH terms to be two.

³ For example, for the SHA, the MAE, LES, and ∣LE∣ all show that the best model in forecasting is a SWARCH type model, i.e. AR(1)-SWARCH(2,1)-L, while the MSE shows that AR(1)-SWARCH(2,2)-L is the best.

⁴ For the SHB, based on the LES and ∣LE∣, the best model in forecasting is an ARCH/GARCH type of model, i.e. AR(2)-GARCH(1,1), while for the SZB, the best model in forecasting is AR(2)-TGARCH(1,1).

⁵ The estimations of other GARCH specifications are available upon request.

⁶ We note that the fact that these probabilities are relatively large ensures that the decomposition of the time series is meaningful in terms of volatility regimes.

⁷ The China Securities Regulatory Commission (CSRC) adopted a new policy that allows new shares to be purchased by investors on the secondary market. Before this date, new securities were sold only to some “special” investors who could obtain much higher profit than investors on the secondary market. This policy innovation has a significant impact on fund flows, stimulating transactions on the two exchanges.

⁸ The 1997 Asian crisis hit Hong Kong in late October and the Hong Kong stock market crashed on October 20 (between October 20 and October 23 the Hang Seng Index dipped 23%).