Research On Volatility Forecast Of China's Stock Market Based On HAR Model With Realized Volatility

In the current financial market,volatility plays an important role in derivatives trading,risk control and pricing.In this paper,Shanghai Component Index,Shenzhen Component Index and Shanghai and Shenzhen 300 Index are empirical research samples.The 5-minute high-frequency data is used to calculate the realized volatility of China's stock market.Based on the realized volatility,8 HAR models and continuous fluctuations model And the jump fluctuation model are constructed.The mean structure mutation test method is used to analyze the structural mutation characteristics of China's stock market.The average combination model and median combination model were constructed using 8 HAR class models.The traditional loss function method,DM test and SPA test are used to analyze the prediction effect of the volatility prediction model.The research finds:(1)China's Shanghai Composite Index,Shenzhen Composite Index and Shanghai and Shenzhen 300 Index realized volatility,continuous fluctuations and jump fluctuations,all of which have peak and fat tail distribution characteristics.The results of the HAR model,which did not decompose the realized volatility,indicate that the most influential stock market volatility in the market is the trading behavior of mid-term investors,while short-term investors have the least impact on market volatility.The results obtained by decomposing the continuous and jumping parts of the fluctuations are more in line with the actual situation.The long-term investors' influence on the stock market fluctuations is weakened,and the short-term investors' contribution to the fluctuations increases.The volatility prediction model with asymmetric effect shows that the positive and negative yield volatility coefficients are significant,and the negative yield volatility coefficient is greater than the positive yield volatility coefficient,indicating that the contribution of the negative semivariance is greater in the realized volatility,indicating The existence of leverage.On the basis of further decomposition of the jump fluctuations,the results show that positive("good")jumps have no significant effect on predicting future fluctuations,and negative("bad")jumps have an effect on predicting future fluctuations,which will make future fluctuations larger.,indicating that the jump will significantly affect the prediction of future volatility.(2)This paper adopts the BP structure mutation test method,and empirically finds that when the structural change of the volatility mean has been realized,the structural mutations of continuous fluctuation and jump fluctuation are close,and the point is almost the same as when the volatility matrix structure has been changed.The mutation characteristics of the three are basically the same.When there are many investors who are extremely optimistic in the market or the irrational degree of investors is high,there is a certain time lag in the jump fluctuation.(3)Modeling the continuous part and the jumping part separately,the daily average continuous volatility contributed the most to the prediction of continuous fluctuation,followed by the weekly effect and the monthly effect.In terms of scale effect,in China's stock market,the absolute daily yield and weekly yield of lagging first-order will not have a significant impact on investor behavior.In addition,in the model,the coefficient of negative return is significant,indicating that the negative gain of the first order of lag helps to predict the continuous wave sequence,and the effect is positive.In the jump wave model,the daily effect contributes the most to predicting future jump fluctuations.The contribution of the monthly effect initially reflects the continuous fluctuation as the main component in the fluctuation prediction.(4)The prediction accuracy analysis of the HAR model shows that both the traditional loss function method and the SPA test show that the HAR-RV-SJ-II model has the best intra-sample prediction performance in the wave prediction model with the leverage effect.It shows that the interaction term between the volatility and the negative rate of return can explain the asymmetry effect well,and the decomposition of the active part and the negative part of the jump can improve the prediction accuracy of the model.In addition,although both the loss function method and the DM test show that the overall prediction error of the HAR-RV-J model is small,the analysis of continuous fluctuations and jump fluctuations in the model can improve the prediction accuracy of the model,but the SPA test results show that the jump is performed with the rest.Compared with the decomposed model,only the jump volatility is roughly decomposed,and the prediction effect is not ideal.