Research On The Mixed Average GAS Factor Model And Its Application In China’s Stock Market

Risk assessment is a very important step in financial decision making,in a new outbreak of repeated impact and under the background of the grim situation of international trade,such as stock increases uncertainty in financial markets,the yield and the predictability of volatility index performance degradation,when it is necessary to use reasonable model to assess risk loss,in the research on all kinds of market risk measurement,The model based on GAS family shows good prediction performance,especially in the aspect of volatility and VaR prediction,which can well fit the characteristics of time-varying drive and thick tail of volatility.Andersen and Bollerslev et al.indicated that the established volatility fitted by intra-day high-frequency data showed superior prediction performance in predicting future variance,which indicated that the introduction of high-frequency data could better describe the linear dependence of asset volatility.In addition,refer to past studies found that financial product pricing is affected by its volatility,so adding a risk premium to the volatility in the equation,is introduced in accordance with the average yield of time-varying equations,can better capture the volatility and linear and non-stationary characteristics of yield,and accurate analysis of the underlying asset rate of return,namely the influence of rate of return risk premium to the price.In this thesis,the mean value equation and mixing data are introduced into the basic GAS model to improve the prediction performance of the model,expand be explained variable factor,the influence of the mixing in accordance with the average volatility research field to supplement,has certain practical significance and the innovation value,at the same time with the aid of the MCS inspection to find the optimal competition model,thus the precise value at risk VaR,It has strong application value to predict and avoid the risk of financial market fluctuation effectively and provide scientific and effective decision-making suggestions for market investors.Firstly,according to the MIDAS-GAS factor model proposed by Gorgi,Koopman and Li(2019),this thesis introduces the mean value equation into the this model,constructs a Mixed Data Sampling Generalized Autoregressive Score in Mean(MIDAS-GAS-M)factor model and its large-sample properties are proved.And establishes the relationship between return rate and risk,taking into account the influence of high-frequency variables on volatility.Furthermore,considering the impact of risk premium on yield rate,Based on the constructed reference model and the new model,Monte Carlo simulation is used for experiments and weighted maximum likelihood method is used for estimation.Four evaluation indexes,namely likelihood function value,BIC,RMSE and MAE,were set to compare the fitting prediction performance of volatility and return rate of the four models under different simulation times,different models and different residual difference arrangements.Secondly,the CSI 300 index is selected as the underlying asset,and the residual distribution of the return series is determined by marginal distribution test,and then the MIDAS-GAS-M model is constructed.The fitting effect is compared by referring to the loss function and the fitting chart,and the future return rate and volatility are predicted,and the MCS test is used to investigate the forecast situation.The prediction performance of different models was compared.Finally,the tail risk of CSI 300 index is measured,the VaR risk value of its out-of-sample predicted value is calculated,and the Kupiec backtracking test is used to measure the accuracy of VaR estimation,and the optimal model is selected and applied to the management of tail risk.Conclusions are as follows: firstly,the simulation experiments,the improved mixing in accordance with the average GAS within the sample fitting and sample prediction has the very big superiority,the volatility of timeliness,add a risk premium in accordance with the average part and high frequency variable assets,error t distribution and GED distribution helps to improve the fitting precision of the model.ARMA-GASM model and MIDAS-GAS-M factor model are superior to GAS-M model and MIDAS-GAS model in four precision indexes when fitting and forecasting the rate of return.Secondly,in CSI 300 index,the empirical study,the sequence of yield of the stock market is rush thick tail,the day has been realized volatility of CSI 300 index and hours of realized volatility showed significant autocorrelation characteristics of marginal distribution test to get yield sequence for the t distribution of residual distribution,the implemented measures introduced in accordance with the mean and days to the GAS model,It can improve the in-sample parameter estimation and out-of-sample prediction ability of the model.The MCS test results under the three evaluation indexes and the coverage level of 10% show that the improved model has a high pass rate and excellent performance.Thirdly,the VaR value of the underlying asset is calculated at different confidence levels,and the test results at the three significance levels are basically consistent,indicating that the measured value at risk is robust,and the conditional value at risk ES can better describe the tail change of the loss variable.In addition,the posterior test failure ratio of VaR series at risk is used to measure the prediction accuracy of VaR.Compared with the general model,the MIDAS-GAS-M factor model based on the mean mixing factor volatility has good fitting effect and more accurate estimation of VaR value.