High Frequency Financial Time Series Forecasting via Particle Filtering

Of the strong non-Gauss characteristic, the high frequency financial time series could not be analyzed and forecasted by traditional statistics method any more. For inaccurately estimating the realized volatility using the limited high frequency data created by the market operation, a novel forecasting method is proposed: after modeling the realized volatility, the particle filtering technology for non-Gauss non-liner process is adopted to analyze and predict the volatility, hence the intra-day transaction data could be treated. The method is applied in the MSFT intra-day quote forecasting and a perfect result is obtained.