Prediction Of Volatility Of CSI 300 Index Based On Wavelet And EEMD Denoising

The prediction of volatility of stock indexes is always an important research objective in econometrics.But the high frequency data,which is used in recent researches of index volatility and measured in units of minutes and seconds,can be easily influenced by microstructural effect,people who use high frequency data concentrate more on removing noise from the volatility model than those who use daily data.Therefore,this paper researches the accuracy of denoising algorithm when predicting the volatility of stock index.The objective of this paper is CSI 300 index,which can reflect the whole operation of Shanghai and Shenzhen stock exchange simultaneously.We collect high frequency daily data of these two stock indexes from 18 May,2015 to 3 Jan,2019 every 5 minutes to predict the volatility.Firstly,process daily,weekly and monthly realized volatility by using EEMD and gain the break even point of high frequency and low frequency in the IMF through significance test.Secondly,using single reconstruction and multiresolution analysis,which are two methods in wavelet analysis,process the high frequency data in IMF,then obtain the realized volatility after denoising.At last,based on a heterogeneous market hypothesis,verify the accuracy of volatility by HAR-RV model,predict future volatility of stock market and compare it to the result of using original raw data.There are two main conclusions:a)Comment in respect of accuracy,comparing to using the original data,the method combining EEMD and wavelet analysis improves the accuracy of predicting future index volatility in the model,decreasing error by more than 75% and even MSE by more than 95.65%.b)Comment in respect of predicting the trend of volatility,the original one is apparently different from the actual result,while the one processed by denoising is very closed to the actual result.In mostly intervals of time,the basic size of volatility is small and any slight prediction error affects the subsequent research,so it is meaningful to improve the accuracy when we denoising.This paper introduces a method combining EEMD and wavelet analysis to denoising the volatility of stock index,and provides an effective denoising method of processing stock data,which has significant theoretical and applied value.