Measuring Stock Market Risk Based On Realized HAR-GARCHSK Model

The stock market ofen appear severe fluctuation.The stock market crash of 2015 is a very exciting process,and the crash of the Shanghai Stock Exchange Composite Index shattered peopele’s confidence,in addition,stock market volatility severe fluctuation is not conducive to the healthy development of the financial market in china.At this time,the risk management becomes more and more important.The first step of risk management is risk measurement,because VaR is additive and can be estimated easily,so people often use VaR to calculate risk of a portfolio of financial assets in a certain confidence level value of the maximum loss,including loss of stock index of stock market.VaR measurement firstly requires accurate prediction of volatility,i.e.,choose effective volatility model.With the availability of high-frequency data,many scholars now use intraday data volatility modeling,such as the realized GARCH model,the realized EGARCH model and so on,which compared to the traditional model has better volatility forecasting and VaR effect estimations,but these models did not consider long memory and time-varying property of volatility series.In fact,volatility of stock market has obvious persistent characteristics and conditional skewness(three moments)and conditional kurtosis(four moments)will chang with the change of time.Considering these characteristics of returns and volatility,this paper constructs the realized HAR-GARCHSK model in volatility modeling,the model will be set to the distribution of GCE,which higher moments of this distribution is not fixed,the distribution of GCE can fully use information in conditional skewness and kurtosis of the return series and a more accurate estimation of VaR.After the construction of the realized HAR-GARCHSK model,the realized GARCH model,the realized HAR-GARCH model and the realized HAR-GARCHSK model were compared,using the real data of closing price of the Shanghai Composite Index in empirical analysis,the results of three models show that: Returns exhibit the characteristics of higher peak and fat tail and not obey the normal distribution,so in this paper we use GCE distribution to build the realized HAR-GARCHSK model compared with the realized GARCH model and the realized HAR-GARCH model.Based on parameter estimations of the realized HAR-GARCHSK model,the historical data of returns have time-varying characteristics,which change severe with time.The realized GARCH model,the realized HAR-GARCH model and the realized HAR-GARCHSK model have severe conditional variance persistence,which the later conditional variance obviously affected by the front conditional variance,then we use the results of model estimation to predict volatility and estimate VaR.Volatility forecasting of the realized HAR-GARCHSK model is better than that of the realized HAR-GARCH model which can well describe the return properties of the peak and thick tail.The realized HAR-GARCHSK model which based on time-varying higher moments has better effect in VaR estimation than the realized GARCH model and the realized HAR-GARCH model.Finally,this paper introduces the extended direction and main application fields of the realized HAR-GARCHSK model,which builds foundation of further development of the model.