Realized covariance matrix is good at forecasting volatility

Abstract

The analysis and modeling of high-frequency financial data are new research fields in financial econometrics. The realized covariance matrix, gotten by expanding realized volatility based on univariate high-frequency data to multivariate high-frequency data, can describe volatility and correlation of multivariate time series. The paper gains the realized covariance matrix of the high-frequency data of Shanghai Composite Index and Shenzhen Component Index, and uses VAR model to forecast variance. Then the result is compared with the ones which are gotten by using ARMA model on realized volatility and GARCH model on two indexes. By comparing those three forecast variance by mean squared error, the paper shows that the realized covariance matrix is better than realized variance, and the realized variance is better than GARCH model on variance forecasting.