The Study Of Realized Volatility Model Based On High-frequency Data

This paper mainly studies the modeling and forecasting of realized volatility based on high-frequency time series where our attention has been focused on the effect of jump process.As a new financial market,the jump phenomenon occurs frequently in China's security markets.As a common asset price behavior in the security markets,it has become a common understanding that jump process is important and should be introduced into the prediction of realized volatility.Based on the micro-structure theory of security markets,we have optimized the realized volatility forecasting model taking jump process into consideration,and the time-varying and multi-model realized volatility forecasting models are further considered in order to improve the accuracy of future volatility forecasting in China's securities market.In this paper,the modeling and forecasting of realized volatility based on high-frequency time series are studied and discussed from the following aspects:Firstly,we propose a novel truncated two-scale realized volatility estimator(TTSRV)and demonstrate its advantages in characterizing implied volatility theoretically.Our numerical simulation results also show that the new realized volatility estimator TTSRV can describe the implied volatility more effectively when considering the effect of jump process and market microstructure noise,so as to improve the prediction accuracy of future volatility.Secondly,we introduce TTSRV into the classical HAR-RV model to predict the volatility of China's CSI300 index.First,we compare the outof-sample prediction ability of classical HAR-RV model with that of HARTTSRV model,and then compare the out-of-sample prediction ability of two kinds of realized volatility prediction models considering overnight returns,leverage effect and measurement errors.It is concluded that TTSRV is a more accurate estimator to describe the jump process and can effectively improve the prediction accuracy of the model in various situations based on our empirical results,and a more effective realized volatility prediction model based on high-frequency data is constructed based on it.Thirdly,we introduce the constraints of margin buying and short selling into the realized volatility model,so that the jump process can be predicted more accurately and the overall prediction accuracy of the model can be improved.First,based on the in-sample results,we find that whether based on margin buying constraints,short selling constraints or both,the results of the prediction model clearly show that margin trading constraints have a significant impact on realized volatility.Furthermore,we find that the margin buying constraints and short selling constraints also have significant effects on the jump process,indicating that the introduction of margin trading constraints significantly improves the characterization accuracy of the jump process.In addition,we have also constructed modified RV-based model where the coefficients of the jump part of the classical HAR-RV model are changed from constant ones to time-varying ones which change with the power of the margin trading constraints.Based on the out-of-sample forecasting performances of the realized volatility of China's stock market,it is concluded that the forecasting accuracy of our newly constructed model is significantly improved.This further verifies the significance of margin trading constraints on jump process and volatility prediction from the perspective of liquidity shock.Fourthly,we consider constructing a time-varying and dynamic realized volatility forecasting model,and introduce Kalman Filter rule to make the parameters of a single model change with time.The empirical study proves that the effect of a single realized volatility forecasting model based on time-varying parameters is better than the ones based on constant parameters.On this basis,we further consider the multi-model dynamic time-varying realized volatility forecasting models where DMA and its improved method ADMA are introduced.By constructing a family of models and assigning weights to each of them,our empirical results also show that the forecasting accuracy of DMA and ADMA methods based on different parameters is higher than that of most single models,which reflects the robustness of the multi-model realized volatility forecasting model based on dynamic average method.Lastly,we measure the tail risk of China's stock market by using realized volatility based on high frequency data.We introduce extreme value theory to characterize tail risk,and the realized volatility based on high-frequency data is introduced into the tail risk measurement.Under the use of the extreme value theory and the high-frequency data,the RV-EVT model is constructed to measure the tail risk of CSI 300 stock index futures,and the prediction model based on realized volatility constructed in this paper is compared with the classical RV-based forecasting model.Our empirical results show that the introduction of extreme value theory can describe tail risk more accurately under extreme conditions,and the improvement of the accuracy of future volatility prediction can also effectively improve the accuracy of tail risk characterization of the China's market.