The Research Of SSE 50ETF Volatility Based On HAR Model

Volatility is one of the important indexes to reflect the risk of financial assets,and the measurement and prediction of the volatility is also an important research issue which concerned by scholars and investors.Early research about volatility focused on ARCH,GARCH and other models which based on low-frequency data,but they did not perform well in describing and forecasting volatility.In recent years,as high-frequency data become easier to obtain,realized volatility models based on high-frequency data begin to attract more attention.Based on heterogeneous market hypothesis,Corsi et al.proposed Heterogeneous Autoregressive model of Realized Volatility(HAR-RV).Because of its clear economic implication and accurate characterization of volatility characteristics,this model has become the mainstream model for studying high-frequency volatility.Based on the above background,this paper takes the five-minute high-frequency data of SSE 50 ETF as the research object to explore the characteristics of realized volatility in China's stock market.And then,according to the specific characteristics of volatility,we combined with the significant jump test put forward by Huang et al and the leverage term put forward by Corsi et al,constructed the HAR-RV,HAR-RV-J,HAR-RV-CJ,LHAR-RV,LHAR-RV-J,LHAR-RV-CJ model as the benchmark model.And on this basis,we introduced SSE 50 ETF Volatility Index(i VX)and CBOE China ETF Volatility Index(VXFXI)as implied volatility variable to construct HAR-RV-IV models.Finally,we maked out-of-sample rolling forecast for the 12 kinds of models,and compared the forecasting ability of these volatility models in China stock markets through the loss function method and MCS test.Then we can found the prediction models which suitable for the Chinese market and provide references for investors' investment and risk management.The empirical results show that:(1)volatility of China stock market has obvious characteristics of long memory,leverage effect and jump;(2)in both in-sample forecast and out-of-sample forecast,LHAR-RV-CJ model is the best-performing model of the HAR-RV model and its extended form,which shows that the combination of jump and leverage effect can effectively improve the fitting precision of the model.;(3)the model fitting and forecast precision was obviously improved after introducing the implied volatility as the explained variable,which shows that implied volatility does include additional information about the future volatility.However,the improvement of the extended form of HAR-RV model by introducing implied volatility is less than the improvement of HAR-RV model,and the improvement of LHAR-RV-CJ model is the least,which shows that a combination of volatility leverage and jump characteristics can make more efficient use of the information contained in realized volatility,thus reducing the room for implied volatility to improve the forecast ability;(4)compare the forecast ability of models in different periods,we found the forecast ability was better in the stable period than in the oscillatory period.The introduction of jump and leverage effect can improve the forecast ability of the model in the oscillatory period,but the result was reversed in the stable period.This may be caused by the inconsistency between the volatility characteristics captured by the jump & leverage term and the actual characteristics of the volatility in the stable period.Furthermore,introducing implied volatility can improve the forecast ability of models in the oscillatory period,but in the stable period,i VX didn't improve the forecast ability of models.This may be because i VX index is still imperfect compared with VXFXI index,which cannot provide effective information for the prediction of volatility in stable period.