Empirical Study Of Volatility Based On Financial High-Frequency Data

Abstract:Since high-frequency data contained plenty of market information, Researching on high-frequency data became more and more important. In order to further research the market structure, the study of financial volatility has been the focus of econometric scholars both at home and abroad. This paper analysis the volatility and model of financial high-frequency date.First, the paper did statistical characteristic analysis on one minute date from Chinese stock market, we find that the high-frequency return series of Chinese stock markets were not normal distribution, but peak and heavy tail, and had significant "calendar effect".Second, the paper introduced realized volatility and realized bipower variation, Through analysis it is concluded that realized bipower not only has the advantages of realized volatility, but also has robustness and more efficient than realized volatility. Under the condition of considering micro structure and structure mean square error, we had that the sampling frequency of realized volatility and bipower variation is5minutes.Third, it was proved that bipower variation had long memory property by ADF-KPSS joint inspection、R/S analysis method and modified R/S analysis method, but the daily return rate of Shanghai composite Index hadn’t.Last, we decomposed the series of continuous volatility and jump volatility from the high-frequency date by significant testing method, we found that continuous volatility and jump volatility had autocorrelation, then based on HAR-RV model,the paper prosed LHAR-CV model and LHAR-LJ model, our results found that LHAR-CV could capture the leverage effect of the continuous volatility, it could fit and forecast continuous volatility efficiently, the previous week jump volatility on the impact of the current jump volatility is the most significant.