基于已实现协方差矩阵的高维金融资产投资组合应用

High-Dimensional Financial Assets Portfolio Selection Based on Realized Covariance Matrix

随着金融市场的发展,可配置金融资产种类不断增加,高维资产的投资组合应用引起了广泛的关注,因此高维协方差矩阵的建模及预测更加重要。基于已实现协方差矩阵,创新地将Elastic Net(弹性网)方法与向量自回归模型结合,对高维已实现协方差矩阵进行建模和预测。实证分析中模型取得了理想的预测精度,待估参数的数目显著下降;由于弹性网方法具备充分的变量选择功能和群组效应,得到的模型更加完善,因此资产之间动态相关结构也更加明晰;分析发现行业之间协方差变化比自身方差变化更加复杂,将VAR-LASSO、VAR-EN、DCC-MVGARCH、EWMA四种模型预测的协方差矩阵应用到投资组合中,结果表明VAR-EN优势明显。

With the development of the financial market, more and more types of financial assets are available, so that portfolio application of high dimensional assets has attracted wide attention, resulting that modeling and predicting high dimensional covariance matrix is more important. Based on realized covariance matrix this paper has combined vector autoregressive model with elastic net method and then proceed the application in the modeling and prediction of high dimensional covariance matrix. In the empirical analysis, the model has good prediction accuracy and the number of estimated parameters decreased significantly. With perfect ability of variable selection and group effect, dynamic correlated structure between assets is clarified Analysis shows that the covariances of industry are more complicated than the variances. Finally, we apply the covariance matrix predicted by VAR-LASSO, VAR-EN？ DCC-MVGARCH and EWMA into the investment portfolio respectively, VAR-EN outperforms the rest methods.