An Empirical Study of Intra-day Stock Return Volatility
It is well known that microstructure noise could have substantial impact on volatility estimation of high frequency asset returns. The Two Scale Realized Volatility (TSRV) estimator makes use of all the available data and at the same time corrects the effect of market microstructure noise. In this study, 30-minute TSRV series is constructed from tick-by-tick Dow Jones 30 stock prices. Our results show that the 30-minute volatility estimate series has the stylized characteristics, including volatility clustering, long memory and displaying U-shape within the day. Also, the volatility for stocks during earning announcement period is significantly higher than that in non-announcement period. This phenomenon is particularly striking at the opening hour of the announcement day. Time series model is built on the periodic and long memory features with rolling window size of one month. We forecast the out-of-sample 30-minute volatility one day ahead based on Semi-parametric Fractional Autoregressive model and modified HAR-RV linear regression model.