基于独立成分分析的高频高维金融资产组合风险估计与动态预测

近年来全球金融风险不断扩大，因此，组建高维资产组合，准确估计和预测资产组合价格波动，对金融风险防范与规避具有重大的理论与实际意义。已实现协方差（Realized Covariance Matrix, RCM）使用高频金融数据，可刻画高频高维资产组合的风险水平。为估计及预测RCM，目前主流的做法使用主成分分析提取RCM特征，然而此方法忽略了高频数据的非高斯性特征，且估计精度有待提高。鉴于此，本研究使用对非高斯性数据更适用的独立成分分析，提取RCM的有效独立成分和残差矩阵，并证明其渐近统计性质。然后，考虑高频波动率特征，分别对独立成分和残差矩阵进行动态的、正定性的预测。最后，实证检验独立成分分解及动态预测给高维投资组合带来的风险规避作用与经济效益。本研究发展了独立成分分析在高频高维金融资产组合风险估计与预测中的理论和方法，为我国学者和业界将此方法运用于我国金融市场实证分析提供参考。

In recent years, global financial risks have continued to expand. Therefore, the establishment of high-dimensional asset portfolios and accurate estimation and prediction of asset portfolio price fluctuations have great theoretical and practical significance for financial risk prevention and avoidance. Realized Covariance Matrix (RCM) uses high-frequency financial data to characterize the risk level of high-frequency and high-dimensional asset portfolios. For estimation and prediction, the current mainstream approach uses principal component analysis to extract RCM features, but this method ignores the non-Gaussian features of high-frequency data, and the estimation accuracy needs to be improved. In view of this, this study uses independent component analysis, which is more suitable for non-Gaussian data, to extract the effective independent component and residual matrix of RCM, and prove its asymptotic statistical properties. Then, considering the characteristics of high-frequency volatility, the independent component and residual matrix are respectively dynamically and positively predicted. Finally, empirically test the risk aversion and economic benefits of independent component decomposition and dynamic prediction to high-dimensional investment portfolios. This research has developed the theories and methods of independent component analysis in high-frequency and high-dimensional financial asset portfolio risk estimation and forecasting, which provides a reference for Chinese scholars and the industry to apply this method to the empirical analysis of my country's financial market.