Research On The Forecasting Performance Of The HAR-RV Model And Its Expand Models

Description and forecasting for financial asset income rate is the modern financial research of hot and difficult problem in academic field and practice field. Early research about volatility focuses on GARCH, SV and their extended mode by low frequency model, but these low frequency models’behave,both in describing and forecasting volatility, is not ideal. In recent years, Realized Volatility (shorthand for the RV) and dynamics models based on the Realized Volatility (ARFIMA-RV or ARFIMA-LnRV model) put forward by the scholars represented by Andersen.These models has better prediction ability which is superior to the low frequency models,such as GARCH and SV. But as a result of their economic interpretation is not very clear and in the process of building a difference operator may cause the loss of a large amount of trading information,Corsi is proposed Heterogeneous Autoregressive model of Realized Volatility(shorthand for HAR-RV) based on Heterogeneous market hypothesis.In recent years,the model of volatility of high frequency,especially the HAR-RV model coursed wide attention of scholars both at home and abroad.Under this background,we combined with the C_TZ jump test put forward by Corsi and the Signed Jump Variation (SJV) estimator put forward by Barndorff-Nielsen recently, constructed the HAR-RV-SJV, HAR-RV-J-SJV, HAR-RV-CJ-SJV and HAR-RV-TCJ-SJV models. And then, we decomposed the Signed Jump Variation (SJV) into positive and negative Signed Jump Variation, to build models: HAR-RV-SJV-D, HAR-RV-J-SJV-D, HAR-RV-CJ-SJV-D and HAR-RV-TCJ-SJV-D. Compared to SPA method,We took a more rigorous method called MCS.And we used5minutes high-frequency data of Shanghai and Shenzhen300Index from April8,2005to June30,2013as sample data to empirical compare the forecasting ability of the12kinds of volatility high frequency model in China’s stock market.The empirical results show that:1, C_TZ statistics identify the times of jump is far more than the Z statistics;2, the Signed Jump Variation can not only improve the fitting precision of the models, but also helps to enhance the predictive power;3, the model fitting and forecast precision was obviously improved when the positive and negative Signed Jump Variation,which had been decomposed from Signed Jump Variation (SJV),as the explained variable to join to the base model (HAR-RV, HAR-RV-J, HAR-RV-CJ and HAR-RV-TCJ).Furthermore,the negative jump variation has a more far-reaching impact on future than the positive jump variation (|βSJV-|>|βSJV|>|βSJV+|);4, Based on MCS inspection’s result of this paper, we found that HAR-RV-TCJ-SJV’s prediction accuracy is the highest of12kinds of model we mentioned. In terms of structure, this article take the general structure of master thesis framework: the first part is introduction, briefly expounds the development of the situation of the volatility forecasting currently, and the research approach which we will take in this paper; The second part is the literature review in domestic and foreign about volatility forecast up to now; The third part is about Jump and Signed Jump Variation’s measure and the definition of HAR class volatility forecasting model; The fourth part is to describe the volatility forecast method, and method of testing their prediction accuracy; The fifth part obtained the best prediction model by MCS test; the last part is the article’s conclusion,prospect,thanks, and literature review.