Empirical Study Of The Ability Of Prediction Of Two Models Based On Realized Volatility

Volatility is the most important character of financial assets. The accurate measurement of it is absolutely meaningful. Since capital markets abroad are more developed, correlative research has been more mature. There are kinds of restrictions in domestic capital market, so research in this area is relatively backwards. However, as the process of opening up of China's capital market speeds up, the research of volatility is around the corner. This is also important to measure the price of financial assets precisely and to construct risk management system of our country's financial institutions.First, this paper introduces "integrated volatility" and "realized volatility" and shows the definitions of them in foreign literature. Then it shows the character of the two volatilities. At last, we choose the optimal sampling frequency of high-frequency data with the minimization of the variation of the integrated volatility.About the part of model and empirical research the paper introduces two models: SV model and GARCH model. Based on the character of high-frequency data we build long-memory stochastic model (LMSV model) and (seasonality adjusted) GARCH model. According to the literature of Deo et. al (2006), based on the high-frequency data of Shang Zheng index, firstly, this paper makes seasonal adjustment to the original log yield series. Secondly, estimates the parameters of the model in accordance with enhanced frequency-domain quasi-maximum likelihood estimation. Finally, predicts the volatility. Adjusting the seasonality in the mean equation of GARCH model and then we estimate the parameters in this model with maximum likelihood estimation and predict the volatility in succession. The regression of volatility predicted to model-free Realized Range-based Volatility (RRV) is the target assessment of ability of prediction. RRV is the consistent estimator of integrated volatility, so it is the proxy of real volatility. We can make judgments of the ability of the prediction through comparison of regression coefficients and Goodness-of-fit.As we can see in this paper, through empirical research, the prediction's precision of LMSV model is much better no matter how long of the length of prediction. Both of the two models have better prediction ability during medium term. About the aspect of Goodness-of-fit, the ability of prediction of LMSV model is better than GARCH during short term, however, the latter is better than the former during medium and long terms.