Research On Extension And Prediction Of GARCH Model Based On Negative Realized Semivariance

As an important attribute of financial assets,volatility is often used to measure the risk level of assets.It also plays an important role in the fields of financial derivative pricing,risk management,and investment portfolio construction,making the estimation and prediction of volatility a research hot issues.Since the beginning of the 21 st century,computer access technology and data processing capabilities have been greatly enhanced,reducing the difficulty of obtaining high-frequency transaction data.As high-frequency data contains more information,volatility measures and its modeling and predicting based on high-frequency data become the focus of academia and practice.As the earliest proposed realized measure,realized volatility is widely used in academia due to its simplicity and accuracy.Later,some scholars further divided the realized volatility into two parts according to the positive or negative returns of the intra-day,namely the positive realized semivariance and the negative realized semi-variance.This article uses the 5 minutes frequency of the CSI 300 Index and the S & P 500 Index as examples.In the framework of the GARCH model,the measures of positive realized semi-variance and negative realized semi-variance have been introduced,respectively,to establish volatility models and compare their out-of-sample prediction performance with the traditional GARCH family models and the original realized GARCH model.In evaluating out-of-sample prediction performance,this paper uses out-of-sample rolling time window technology,and selects five kinds of robust loss functions,and uses a rigorous and robust Model Confidence Set(MCS).The main empirical conclusions are:(1)Regardless of whether the CSI 300 Index or the S & P 500 Index is used as the research object to predict one step forward,the MCS test results show that,under the selected five kinds of robust loss functions,in most cases the volatility models based on high-frequency trading data have better out-of-sample prediction performance than the traditional GARCH family models,and the results of the forward five-step predictions are also consistent,indicating that the above conclusions are robust.(2)Whether in the Chinese market or the US market,whether it is a one-step forward forecast or a five-step forward forecast,after the introduction of realized semi-variance,the in-sample fitting and out-of-sample prediction effects of the volatility model are somewhat improved.Compared with the positive realized semivariance,the effect of the negative realized semi-variance on future volatility might be more important.The volatility model based on the negative realized semivariance has the highest volatility prediction accuracy among all models.