Time-Varying Portfolio Selection Based On DCC-MIDAS

DCC-MIDAS model combines the characteristics of DCC-GARCH model and GARCH-MIDAS model,which can make full use of their advantages.Firstly,it considers the influence of past market information on the correlation relationship and reflects the nature of time-varying.Secondly,it skillfully distinguishes the dynamic correlation into short-run component and long-run component,which can fully reveal the time-varying correlation among financial time series.Thirdly,it directly models raw mixed frequency data without any frequency conversion,which can effectively exploit rich information contained in mixed frequency data.Therefore,the DCC-MIDAS model can improve the estimation accuracy of time-varying correlation relationship and realize the robust and effective estimation of time-varying covariance matrix.In this regard,the dissertation applies the DCC-MIDAS model to the mean-variance portfolio selection research,and conducts the empirical analysis in the cases of medium-and large-scale financial assets.In the case of medium-scale financial assets,in order to improve the traditional mean-variance model in two aspects including the covariance matrix estimation and static weight processing,the dissertation proposes a new time-varying portfolio selection model based on the DCC-MIDAS model and a parametric scheme.Specifically,the DCC-MIDAS model is introduced to improve the prediction accuracy of dynamic association relationship among financial assets by exploiting high frequency information.In addition,considering the influence of the time-varying characteristics of financial assets on portfolio weights,the dissertation incorporates them to design a parametric weight function,which helps to improve portfolio performance.The dissertation then applies the new model to conduct empirical analysis on several stocks and industry groups in China's stock markets.The research finds that book-to-market ratio(BTM)and price earnings ratio(PE)are positively correlated with portfolio weights,while market equity(ME)presents negatively.The empirical results show that the proposed model outperforms several competing models in terms of standard deviation risk,Sharpe Ratio,and efficient frontier.In the case of large-scale financial assets,the dissertation proposes a novel norm constrained time-varying minimum variance model with DCC-MIDAS,labeled as NC-MVP-DCC-MIDAS.It applies the DCC-MIDAS model to improve the estimation of dynamic correlations among financial assets by exploiting rich information contained in mixed frequency data.Additionally,it imposes norm constraints on the minimum variance optimization problem with the elastic-net penalty to pick a reasonable number of financial assets and prevent extreme positions in the resulting portfolio.The efficacy of new approach is illustrated via portfolio studies on the constituent stocks in the Shanghai Stock Exchange(SSE)50 Index of China.The empirical results show that the proposed approach is efficient to solve a large portfolio selection problem and outperforms several competing models in terms of the mean,standard deviation,Sharpe ratio,and portfolio backtesting.To sum up,the DCC-MIDAS model has competitive advantages in describing the time-varying correlation relationship of financial assets.Introducing the DCC-MIDAS model into the Markowitz portfolio selection model can significantly improve portfolio performance.The dissertation empirically demonstrates the feasibility and superiority of DCC-MIDAS model in the field of portfolio selection,which provides a new perspective for the analysis of portfolio selection.At the same time,the dissertation also expands the application scope of the DCC-MIDAS model.In addition,the conclusions of this dissertation can provide decision support for market investors and financial regulators in risk measure and risk management.