Forecasting Research On Realized Volatility Of Sse Industry Index Based On Markov Regime Switching Mixing Data Sampling Model

Volatility is closely related to risk management,derivatives pricing,trading strategy and portfolio optimization,which makes the research on volatility more preferred by financial scholars.With the continuous development of data technology,the realized volatility measurement based on high-frequency data has become the mainstream of research regarding volatility.Among them,the Heterogeneous Autoregressive Realized Volatility(HAR-RV)model has become a hot model in recent years because of its simple estimation and strong economic explanation.However,with the continuous understanding of the model,scholars have found that there is a time-varying lag structure in the realized volatility,and it is more helpful to improve the model prediction accuracy considering the structural uncertainty in its research.In addition,the financial time series have different fluctuations in different periods,which also attracted the attention of scholars.At the same time,realized volatility not only has the characteristic of"leptokurtosis and fat-tail",but also has"structural breaks".Therefore,this paper combines the mixed data sampling model(MIDAS)with the Markov Regime Switching,and establishes a new volatility model(MS-MIDAS)to describe and predict stock market volatility in China.The purpose of this paper is to establish an appropriate volatility model to describe and predict the volatility.Two problems need to be solved in the research.First,the mixed data sampling model smoothes the lagged structure of realized volatility through different weight equations,and there is no limit on the maximum lagged order.Although the model structure is more flexible,can it accurately describe the long memory characteristics of asset volatility?Second,existing literature find structural breaks exist in the financial market.In the modeling analysis,we need to consider not only the characteristics of volatility itself,but also the impact of structural breaks in the asset market.Therefore,this paper introduces Markov Regime Switching based on MIDAS model.But can the changes of the model beat the benchmark model and have better out of sample prediction ability?In this paper,literature and empirical analysis are combined to solve these two problems.First,through literature review,this paper summarizes the shortcomings and defects of the existing volatility models,which also provides theoretical support for this paper to select MIDAS model as the benchmark model.Second,the structural breaks of the selected research object is determined by minimizing the penalty comparison function,which also brings the direct evidence for the modeling analysis.Last,the empirical research on MS-MIDAS model and benchmark Midas model is carried out by using MATLAB,and conclusions are obtained as follows.First,this paper finds that there are structural breaks in the industry index of Shanghai stock market,and the low volatility period is more persistent and stable.In the period of high volatility,lagged realized volatility has a greater impact on the future volatility.Second,through the out of sample R~2test,it is found that the prediction accuracy of MS-MIDAS model is significantly better than that of the benchmark MIDAS model.In the robustness tests,we changed the lag order,selected ten minute sampling frequency data,introduced the jump component in MIDAS and MS-MIDAS models,and changed the out of sample prediction accuracy test method.These four empirical results support the conclusion and prove that the Markov mechanism transformation structure has a positive effect on improving the prediction ability of MIDAS model.Finally,this paper further discusses whether the prediction ability of the new model is still superior in three special periods.