A Study On The Realized GARCH-HAR Model Based On High-frequency Data

As volatility is unobserved, measuring volatility has been a central topic in fi-nancial market since volatility plays a very crucial role in pricing of financial instru-ments.asset allocations and risk management. As high-frequency financial data are now widely available, the GARCH family can no longer satisfy the need of study-ing high-frequency data, while realized volatility, which is obtained by aggregating high-frequency return series, provides a good measurement of market volatility. So it is meaningful to combine the advantages of the two in order to predict volatility.This paper computes realized volatility from the5-minute high frequency data of the Shanghai Composite Index, and decomposes the realized volatility into the time series of continuous volatility and jump volatility as well as positive semivari-ance and negative semivariance. It introduces a new framework, Re-GARCH-HAR, for the joint modeling of returns and realized volatility. As we can see in this pa-per, the empirical research suggests that our model has greater prediction ability than the standard GARCH model, and in the medium and long term prediction, our model also outperforms the realized GARCH model. When looking for the right form of our model, we find it interesting to see that modeling negative semivariance leads to significantly better in-sample and out-of-sample forecast performance.