Research On The Risk Measurement Of High-frequency Data Portfolios Based On Mutual Information Vine Copula Model

Frequent financial crises have made the volatility of financial asset returns and the interdependence between assets more and more complicated.In order to effectively characterize the volatility of high-frequency data and accurately describe its interdependence,this paper constructs a Realized HAR GARCH model with weighted realized extreme multipower variation and a Vine Copula model based on mutual information to study the risk measurement of investment portfolios.The main contents are as follows:(1)Extend the weighted realized range multi-power variation to the Realized HAR GARCH modelHigh-frequency data contains more intraday fluctuation information than lowfrequency data.The range fluctuation estimator RRV developed for high-frequency data uses intraday price range to give the realized range fluctuation as an estimator of integral fluctuation.Significantly better than realized volatility RV.Therefore,this paper selects the optimal power based on the discussion of the realized range multipower variation family,and further constructs the weighted realized range multipower variation to eliminate the calendar effect,and finally extends it to the Realized HAR GARCH model to Improve the fitting effect of Realized HAR GARCH.(2)Propose a data-driven mutual information sequential selection algorithm to determine the Vine Copula structureWhen constructing joint distributions for investment portfolios,scholars mostly use the more general R Vine Copula model based on the sequential selection algorithm of Kendall’s rank correlation coefficient.Czado(2012)has pointed out that the traditional Kendall rank correlation coefficient will accumulate uncertainty in the model selection process,and the calculation difficulty increases with the number of variables.Therefore,based on the relationship between conditional mutual information and copula entropy,this paper proposes a sequential selection algorithm based on mutual information driven by raw data to determine the Vine copula structure.At the same time,the truncation procedure based on AIC criterion reduces the complexity of model estimation.The algorithm not only considers the uncertainty accumulated in the model building process is reduced,and the complexity of model estimation is reduced.(3)Empirical research on the risk measurement of high-frequency data portfolios of CSI five major industry indicesFor the marginal distribution,this paper constructs the Realized HAR GARCH-RRV,Realized HAR GARCH-RRBV,Realized HAR GARCH-RRQV and Realized HAR GARCH-WRRQV models based on the framework of the realized range multi-power variation family.The following conclusions are drawn: 1)The realized range quadruplepower variation(RRQV)is the most robust and effective in the realized range multipower variation family.2)The weighted realized range of the fourth power variation can eliminate the influence of the calendar effect.3)Realized HAR GARCH-WRRQV has the best fitting degree and prediction effect among the four models,and can be used as the marginal distribution of the asset returns of the five major industry indices.For joint distribution,this paper establishes D Vine,R Vine,C Vine and Vine Copula models based on mutual information,and then establishes a PCBN model to supplement the propagation path of portfolio risk.Through empirical analysis,it is found that: 1)The C Vine Copula model based on mutual information considers uncertainty in the process of model construction,and the structure selection is more reasonable.2)The C Vine Copula model based on mutual information has a large likelihood function value and small AIC and BIC,which can well model the joint distribution among assets.3)From the risk measurement effect,the prediction and test results show that the C Vine Copula model based on mutual information has high stability and prediction performance to measure risk.4)The risk propagation path given by the PCBN model illustrates the pivotal role of all-finger optional and all-finger energy in the systemic risk of the portfolio.5)According to the optimal weight of Monte Carlo simulation,investors should reduce the investment proportion of all-inclusive information and increase the investment proportion of all-indicated information.