Study On Jump Volatility Of Financial High-frequency Data

The obvious long memory,the “leptokurtosis and fat-tail distribution” as well as the noise of market microstructure are the important characteristics of high frequency financial data.Because of these characteristics,the GARCH class models as well as SV class models,which have wonderful statistical properties and modelling effects in the research of low frequency data,could not get good results in the modelling of high frequency data,which bring great challenges for us to further the researches of high frequency financial data.Based on the 5 minutes high frequency data of CSI 300,this paper fully takes the superposition effects of short-term,medium-term as well as long-term volatility,the impact of jump on volatility into consideration,and weights the realized volatility and the bi-powered volatility.The HAR-RWV class models,the root HAR-RWV class models and the logarithmic HAR-RWV class models are selected to predict the short-term volatility.Then the best model is combined with the support vector machine under different kernel functions,therefore,the best combination of models for short-term volatility prediction of high-frequency data is obtained.