Research On Realized Covariance Forecasting Of Futures Market Under Structural Change:Liquidity Perspective

There is a very close relationship between the stock index futures and the benchmark index market.Investors can not only choose investment arbitrage strategy to obtain profits through this correlation,but also hedge to avoid risks.This paper takes CSI 300 stock index futures and index market as an example to study the common volatility between the two markets,and uses realized covariance to measure the common volatility.Liquidity is the vitality of the financial market.It helps the market to better realize the function of price discovery and optimize the allocation of resources.In addition,it plays an important role in financial risk management and asset pricing.This paper studies the influence of liquidity on the realized covariance of stock index futures and stock market in China,and reveals the influence mechanism of liquidity on the correlation between the two markets from a new perspective.The empirical study finds that the stock market liquidity has a significant impact on the realized covariance,and this impact is negative.From the perspective of time-varying effect,the influence mechanism is also verified.During the period of stock disaster,the influence of index liquidity on realized covariance has changed symbolically.This paper calculates the realized covariance based on the high-frequency data of CSI300 index and stock index futures market.At the same time,this paper uses the HAR model of volatility forecasting for reference,and uses the nonlinear support vector regression（SVR）method to study the forecasting of realized covariance.Because the sample period includes the period of 2015 stock disaster,the model is uncertain and the parameters are unstable.In order to overcome the problem of the decrease of the accuracy of the realized covariance forecasting caused by the systematic structural change,this paper proposes a new window weighted average prediction method based on the window average prediction method(Zhang et al.,2020^[1])with fixed sampling in the time dimension.The empirical analysis from the perspective of forecasting shows that the window weighted average method is superior to the window average method and the traditional OLS and SVR methods.In particular,the window weighted average method based on SVR obtains the best forecasting effect.The structural mutation of the market will lead to the uncertainty of the model and the instability of the parameters,and the window weighted average method can solve this problem.Besides liquidity,the jump and co-jump of asset prices also have an important impact on the correlation between markets.By decomposing realized variance and realized covariance,this paper obtains the jump of single market and the co-jump of two markets,and then constructs the model to analyze the influence mechanism of liquidity and the forecasting of realized covariance.Empirical analysis shows that the combination of jump,co-jump and liquidity is helpful to obtain higher forecasting accuracy.Finally,the robustness test is carried out from three different perspectives:firstly,a new method based on high-frequency return detection of co-jump;secondly,different liquidity measures are used;finally,the dynamic conditional correlation model is used to construct different realized covariance.The empirical analysis shows that under different rolling windows,the models constructed by different co-jump,liquidity and realized covariance calculation methods all get consistent conclusions,so they are more robust.