اثرات عدم تقارن شوکها دربازارهای سهام نوسانات چین : یک روش غیر پارامتری افزودنی عمومی

The unique characteristics of the Chinese stock markets make it difficult to assume a particular distribution for innovations in returns and the specification form of the volatility process when modelling return volatility with the parametric GARCH family models. This paper therefore applies a generalized additive nonparametric smoothing technique to examine the volatility of the Chinese stock markets. The empirical results indicate that an asymmetric effect of negative news exists in the Chinese stock markets. Furthermore, compared with other parametric models, the generalized additive nonparametric model demonstrates a better performance for return volatility forecasts, particularly for the out-of-sample forecast. The results from this paper have important implications in risk management, portfolio selection, and hedging strategy.

The Chinese stock markets have grown rapidly since the establishment of the Shanghai Stock Exchange (SSE) in December 1989 and the Shenzhen Stock Exchange (SZSE) in April 1991. Specially, with the recent boom in China's economy, China's stock markets have been attracting an enormous amount of attention from policy makers, investors, and academics. Chinese stock markets are interesting and deserve attention also because they exemplify many unique characteristics that differ from well-developed Western financial markets. One of the unique characteristics is that the Chinese stock markets are the only equity markets covered by the International Finance Corporation that have completely segmented trading between domestic and foreign investors (Chui and Kwok, 1998 and Yang, 2003). The A-share market is only open to Chinese domestic investors while the B-share market was only open to foreign investors before February 2001.1 Many studies (Chui and Kwok, 1998 and Yang, 2003) also address the fact that the Chinese stock markets are tightly controlled by the government: The markets are at most partially privatized, and the state maintains state shares in varying amounts. The presence of market segmentation and heavy government regulations give rise to mispricing and information asymmetry, making the market clearly imperfect and incomplete (Chan et al., 2007). Further, stock trading is still new to most domestic participants. The A shares are dominated by domestic individual investors who typically lack sufficient knowledge and experience in investments (China Securities and Futures Statistical Yearbook, 2004). Given the unique characteristics of the markets and given that the typical Chinese investor is more prone to speculation and less sophisticated than those from more mature markets (Tan et al., 2008), Chinese stock volatility behaves very differently from that of other markets. This paper seeks to demonstrate that conventional volatility models, such as the GARCH-family approaches, employed for modelling stock returns in unique markets such as the Chinese stock markets are inferior to the nonparametric approach. GARCH models of stock returns rely heavily on volatility specification and known distributions of returns. In Chinese stock markets where it has been documented that information structures of stock returns are prone to changes subject to changing regulatory structures (Brooks and Ragunathan, 2003), the use of GARCH models of stock returns may insufficiently characterize the volatility of the Chinese stock market returns. Bülman and McNeil (2002) propose a nonparametric GARCH model (hereafter NP model), in which the hidden volatility process is a function of the lagged volatility and lagged value of the innovations from returns and is estimated by an iterative nonparametric algorithm. This model is more attractive than the parametric GARCH-family models because it requires neither a specification of the functional form of the hidden volatility process nor that of the distribution of the innovations. This paper contributes to the stock return modelling literature by focusing on modelling the Chinese stock return volatility and the asymmetric effect of shocks on return volatility2 using the NP model. Specifically, we restrict our attention to the univariate case of modelling stock returns and does not explore the possibility of spillover effects and interactions between the A and B-share markets.3 The reason being nonparametric modelling method in a multivariate GARCH framework is at its infancy stage, and for the purpose of comparison with the standard univariate GARCH modelling approach it is important that we restrict our attention to the univariate GARCH NP model. On the methodical front, we contribute by developing a generalized additive model and applying the nonparametric approach (hereafter GAM NP model) involving the iterative estimation algorithm to the generalized additive model of Hastie and Tibshirani (1990). The motivation for estimating the GAM NP model is that it can avoid the curse of dimensionality, which is a common problem for the nonparametric estimation of a multidimensional regression.4 Further, as will be shown in the Monte Carlo simulation and the empirical investigation, this newly proposed GAM NP model can deliver a more accurate volatility estimate than the parametric GARCH-family models and becomes computationally more efficient than the NP model. Also novel in our approach is that we extend the news impact curve from Engle and Ng (1993) to the nonparametric context and use it to measure and examine the asymmetric effect of shocks on volatility. It is without doubt that GARCH-family models are the most commonly employed models in the investigation of Chinese stock-return volatility and its asymmetric effects. For example, Yeh and Lee (2000) use the GJR model proposed by Glosten et al. (1993) to examine Chinese stock market volatility from May 22, 1992, to August 27, 1996. They find that investors in China chase after good news indicating that the impact of good news (positive unexpected returns) on future volatility is greater than that of bad news (negative unexpected returns). By estimating both the GJR and the EGARCH model, Friedmann and Sanddorf-Köhle (2002) report that bad news increases volatility more than good news in A-share and composite indices, whereas good news increases volatility more than bad news in B-share indices based on a sample beginning on May 22, 1992, and ending on September 16, 1999. The good-news-chasing-investor phenomenon in China makes the Shanghai and Shenzhen stock markets relatively unique and different from many other stock markets in the world. Lee et al. (2001) provide the same result as Friedmann and Sanddorf-Köhle (2002) with the EGARCH model and daily return data from December 12, 1990, to December 31, 1992. Zhang and Li (2008) investigates the asymmetry effect of bad news on Chinese stock volatility with a partial adjustment process. They find that the asymmetry effect begins to appear in May 1996. Dividing the total sample into two periods, Huang and Zhu (2004) produce results from the EGARCH and GJR models showing that the asymmetry effect only exists in the period between February 2001 and September 2003. In view of the different findings from past research regarding the leverage effect of Chinese stock-return volatility, we examine Chinese stock market volatility and the asymmetric effect of market news on the volatility using data from January 2, 1997, to August 31, 2007. Several questions will be addressed in the investigations: Do Chinese stock market volatilities react asymmetrically to shocks as in most mature stock markets in the world? Are investors in the Chinese stock markets still chasing after good news? Do volatilities in the Shanghai and in the Shenzhen stock markets react similarly to the market news? The answers to these questions have important implications for market practitioners forecasting stock returns and volatility, and for risk managers formulating optimal strategies for portfolio selection, risk management and hedging. The results from this paper suggest that the leverage effect exists in the Chinese stock markets: Bad news does affect return volatility more than good news. However, as implied by the news impact curve (NIC) from the GAM NP model, a small amount of good news is needed to keep the market calm. Further, compared with the superior performance of the GAM NP in the in-sample estimation and the out-of-sample forecast, the GJR and EGARCH models tend to overestimate the volatility process in turbulent periods and yield larger estimation errors. Our results suggest that the nonparametric smoothing approach is a more appropriate tool for estimating Chinese stock-return volatility than the parametric GARCH models. The rest of the paper is organized as follows. In Section 2, we present the nonparametric models and the model estimation algorithm. Section 3 performs the Monte Carlo simulation to evaluate the performance of the parametric and nonparametric models. Section 4 compares the performance of the nonparametric models with various GARCH-family models and examines the asymmetric effects of shocks on the Chinese stock market volatility. Section 5 concludes.

This paper examines the return volatility and the asymmetric effect of market news on the volatility in the Chinese stock markets using a nonparametric approach. In order to avoid the curse of dimensionality, the back-fitting algorithm from the generalized additive model of Hastie and Tibshirani (1990) is applied to the nonparametric smoothing technique from Bülman and McNeil (2002). When compared with the parametric asymmetric GARCH models commonly used for capturing asymmetric volatility, the nonparametric models perform much better in capturing the asymmetry effect and in describing the dynamic of Chinese stock-return volatility. With respect to the predicted return volatility's asymmetric reaction to good news and bad news, we find that the return volatility responds more strongly to bad news in the Chinese stock markets in our sample period. We extend the news impact curve to the nonparametric setting to further examine the asymmetry effect implied by the GAM NP model. Interestingly, the evidence based on the news impact curve of the GAM NP model suggests that the good-news-chasing behavior of the Chinese domestic investor continued. Additionally, the markets behave such that they require a certain amount of good news in order to remain as calm as possible. When all the models are employed to obtain the overnight out-of-sample forecast, the nonparametric models yield the lowest forecast errors and outperform the parametric models by capturing the observed spikes in the volatility of returns. In contrast, the EGARCH and the GJR models tend to overestimate the volatility and returns in the high-volatility periods. The forecasted returns are therefore more accurate from the nonparametric model especially when the market is very volatile. Finally, we show that the results from this paper have important implications in the risk management such as the VaR forecasts.