Modeling And Empirical Study Of Realized HAR GARCH Based On Fat Tail Distribution And Long Memory

In recent years,China's futures market has developed rapidly.In the first half of 2018,China's commodity futures volume reached 1.4 billion,accounting for 48%of the world.Due to the continuous and steady growth of market size,China's futures derivatives market is gradually developing.Become the most important risk management market in the world.Studying the volatility in the futures market and the relevant laws of the market,modeling the volatility,plays an important role in the field of financial risk management,investment portfolio and asset derivatives pricing.In the financial market,as high-frequency data becomes more and more available,a series of high-frequency volatility models have been generated.These models have achieved good results in comparison with traditional low-frequency data-based models..The Realized GARCH model using high frequency volatility has been relatively simple,the parameters are easy to estimate,and the empirical results are good enough.Studies have shown that both in the stock market and in the futures market,the data shows a long memory phenomenon,which means that the influence of past information cannot be ignored,but the phenomenon cannot be explained in the setting of the Realized GARCH model mentioned above.In the long memory model,the HAR model proposed by Corsi(2009)is also modeled by the realized volatility.The model is constructed using OLS to estimate the parameters.The empirical results are comparable to the much more complex ARFIMA models.The idea of constructing long memory in HAR model is introduced into the modeling of Realized GARCH model to obtain Realized HAR GARCH model.Based on the actual long memory futures data,the simulation compares the fitting effects of different models.At the same time,for the thick tail phenomenon existing in the data,the normal assumption of the residual distribution cannot explain the phenomenon,and the residual distribution of the model is extended to The Skewed-t distribution can explain the thick tail phenomenon.In the selection of data objects,the Shanghai copper and white sugar futures data were selected as research objects for empirical analysis.The results show that in different volatility forms,the logarithmically realized volatility is the most suitable model for this paper.In the long memory test of the data,both sets of futures data have long memory phenomena,and the length of the average cycle period indicates that the long memory can be explained by using the ideas in the HAR model.The empirical analysis based on the normal residual distribution model shows that the residual distribution does not conform to the normal distribution.The normal distribution can not fit the thick tail phenomenon in the data.The extension of the residual distribution to Skewed-t is more suitable for thick tail data..Comparing the results of different models using MCS test shows that the Realized HAR GARCH model is the best model,and extending the residual of the model to the Skewed-t distribution can further improve the fitting ability of the model.