CAJ | 학술논문

本文在考察上证综指和深证成指日已实现波动率的长记忆性特征和结构突变的基础上构建了已实现波动率自回归结构突变模型,并用于进行预测。预测结果表明,在未来结构突变已知情形下,自回归结构突变模型的预测效果比其他预测模型要好,在未来结构突变未知情形下,该模型的预测效果比ABDL模型稍微差,而ABDL模型在两种情形下都是比较稳健的预测模型。
본문재고찰상증종지화심증성지일이실현파동솔적장기억성특정화결구돌변적기출상구건료이실현파동솔자회귀결구돌변모형,병용우진행예측。예측결과표명,재미래결구돌변이지정형하,자회귀결구돌변모형적예측효과비기타예측모형요호,재미래결구돌변미지정형하,해모형적예측효과비ABDL모형초미차,이ABDL모형재량충정형하도시비교은건적예측모형。
The realized volatility method based on the high frequency of financial data has obtained the widespread application in the financial domain and been developed into a brand-new research area of financial econometrics. In comparison with GARCH model and SV model, the computation of realized volatility method is convenient and does not need complex parameter estimation. In addition, it is an unbiased estimator of real volatility under certain conditions. However, the research on the high-frequency of volatility forecasts in domestic market is still in its infancy. Although the vast majority of studies are to apply the model, they rarely take into account the various features of high-frequency volatility in Chinese stock markets. In this paper, we focus on the possibility of structural breaks, trends and long memory in the daily realized volatility series in Chinese stock markets usinghigh-frequency data from Shanghai Composite Index and Shenzhen Component Index during 2000-2010. First, we use local whittle method, ARFIMA model and SEMIFAR model to test long memory and estimate long memory models for the realized volatility series in Shanghai Composite Index and Shenzhen Component Index. We find strong evidence for long memory in these realized volatility series. Second, we use Bai and Perron＇ s （1998, 2003） methods to test and estimate a multiple mean break model. We find that there are several common structural breaks in these realized volatility series n Shanghai Composite Index and Shenzhen Component Index, and there is no smooth flexible trend in these series by using SEMIFAR models. Third, we test long memory in the structural break adjusted data. We find that there is a partial reduction of long memory in these realized volatility series in Shanghai Composite Index and Shenzhen Component Index after adjusting structural break. This evidence suggests that the presence of structural breaks in the daily realized volatility series in Chinese stock markets can account for parts of the observed long memory. However, it is very difficult to separatelow-frequency dynamics from high-frequency fluctuations, in particular when short-run fluctuations have high variance. Therefore, this study proposes a short memory model or RV-AR-Break model to forecast the realized volatility in Chinese stock market based on long-memory behavior and the possibility of structural breaks. Forecasting results show that this short memory model provides a superior predictive ability when the timing of future break dates and their sizes are known. ABDL model is a robust forecasting model when the timing of future break dates and their sizes are both known and unknown. This research has great theoretical significance on enriching the volatility of prediction research methods and prediction theory. This study also has great practical significance on optimal portfolio, reasonable allocation of financial derivatives pricing, measuring, prediction and prevention of assets return＇ s risk for institutional and individual investors. Finally, it can further improve the stock market risk supervision level for the financial supervision department.