An empirical study of LMSV model in China stock market based on realized volatility

This paper studies the characteristics of long memory in high-frequency financial time series. First, it introduces realized volatility (RV) and long memory stochastic volatility model (LMSV). Because there are too many parameters in LMSV model, usual methods to estimate parameters are not useful any more. Second, we adopt the semi-parametric estimation method-Local Whittle estimator. Third, combining RV and LMSV model together and using the intraday data of every five minutes in the Shanghai Stock Exchange´s Shanghai Composite Index from 2000 to 2008, we estimate the parameter of long memory and compare it with the ARFIMA model, which is a model being spread widely over the past few years. We find that the results of estimation, long memory parameter d, are in accord with the definition of long memory, and LMSV model is more effective in practice.