The Construction And Empirical Research Of Dynamic Higher Moments Volatility Model Under High Frequency Data

The GEM market is an important branch of Chinese Stock market,which aims to enrich the trading varieties of the financial market,provide financing channels and growth space for middle and small-sized enterprises that cannot be listed on the main board market and have small capital scale but have development potential to further meet the financial demand of different entities and make up for the shortcomings of the simplification of the capital market structure.Since 2020,the Co VID-19 epidemic has caused a huge impact on global capital markets,and global stock markets have experienced violent shocks and circuit breaker and limited the influx.Although the GEM index fell to 1227.99 on January 31,2019,under the epidemic situation,the GEM index was able to rise against the trend.Compared with other indexes in the financial market,the GEM index can effectively help international investors to allocate systemic risks.Therefore,constructing appropriate models and applying appropriate methods to measure financial risks in the GEM market is crucial for investors to manage and prevent risks.Considering that the conditional higher moments of financial asset returns have dynamic characteristics and according to the Realized EGARCH-SK model proposed by Wu et al.(2020),the paper selects the realized volatility and the realized range-based volatility as the realized measures and expands the residual distribution to the skewed-t distribution,thereby constructing dynamic higher moments Realized GARCH model and dynamic higher moments Realized EGARCH model.In empirical research,firstly,we test the time-varying skewness and time-varying kurtosis effects of the data.Secondly,we estimate the parameters of each model,and compare the fitting ability of each model by using the likelihood function.Again,the recursive window method is used to make the first one-day-ahead forecasts,four sets of loss function and MCS test are used to measure the out-of-sample forecast performance.Finally,Kupiec test and Bootstrap test are used to compare the accuracy of the out-of-sample Va R and ES forecasts of each model.The empirical results based on 5 minutes returns of GEM index from January 4,2013 to December 31,2019 show that,the high frequency volatility model that takes into account the dynamic characteristics of skewness and kurtosis can help improve the model fitting ability,volatility prediction accuracy and risk measurement accuracy.The results do not depend on the model setting,distribution assumptions and measurement selection.The Gram-Charlier extended distribution is obviously inferior to the skewed-t distribution in the fitting ability of the model.In addition,the choice of measures and risk level will affect the performance of the model risk measurement.Under the extreme risk level of 1% of RV,the skewed-t distribution has the best performance of out-of-sample Va R and ES prediction,especially when RRV is selected as the realized measure,the prediction accuracy of the risk measurement is more obvious.From the perspective of the forecasting effect of volatility and the accuracy of risk measurement,the dynamic higher moments Realized EGARCH model is better than the dynamic higher moments Realized GARCH model.