Research On Realized Covariance Matrix Model Based On Measurement Errors

In recent years,modeling covariance matrix with high frequency data has been becoming more and more popular.However,the statistics estimated by high frequency data is affected by measurement errors easily,there will be a problem of measurement errors in variables in the process of modeling.Because the relative research on this field is less,this paper tries to build a covariance matrix prediction model considering measurement errors to reduce the influence of measurement errors.To consider the problem of measurement errors easily in the process of the covariance matrix modeling,this paper uses DCC method to decompose the realized covariance matrix into the realized variance matrix and the realized correlation coefficient matrix.By the log-HARQ model proposed in this paper,the paper realizes the purpose of considering the measurement errors in the process of modeling the covariance matrix.At the same time,this paper further considers the risk assets'interactive effect of variance,asymmetry and mean reversion characteristics,and also proposes the model that considers all of the characteristics and uses the Lasso method to filter out non-forecasting factor and so on.Those models get a further higher capacity of prediction.This paper investigates the empirical application value of the above corrected model in the fields of investment portfolio and risk management.First applies the realized covariance matrix corrected model to investment portfolio,using the out-of-sample predictive covariance matrix of the benchmark model and the corrected model to construct portfolios,this paper finds that the realized covariance matrix model corrected by measurement errors performs better than the benchmark model under consideration of various indicators,the corrected models further considering some common features of the stock almost improve portfolio performance.This paper further applies the above realized covariance matrix models in risk management,this paper finds that realized covariance matrix model corrected by measurement errors has a better performance than the benchmark model,most of realized covariance matrix model further considering the characteristics of stock have a higher predictive accuracy.