报告题目：Heterogeneous Autoregressive Model for Symmetric Matrix-valued Time Series

报告时间：2026年8月18日下午16:30

报告地点：南湖校区办会楼二楼会议室1256室

主办单位：萝莉社

报告人：刘广应

报告人简介：刘广应，南京审计大学统计与数据科学学院教授，博士生导师，复旦大学博士，浙江大学博士后，香港科技大学访问学者。现为南京审计大学统计与金融联合实验室副主任，江苏省“青蓝工程”学术带头人，中国现场统计研究会旅游大数据学会理事、中国管理科学与工程学会理事、中国管理科学与工程学会金融计量与风险管理分会理事、江苏省概率统计协会理事。研究领域与兴趣：金融高频数据、应用统计、深度学习、金融数学等。在《中国科学》《Journal of the American Statistical Association》《Journal of Econometrics》《Journal of Business & Economic Statistics》《Statistica Sinica》等国内外杂志发表或录用论文40多篇。主持2项国家自然科学基金项目、1项国家社会科学基金项目、10余项省部级课题。

摘要：Matrix-valued time series data are common in the fields of economics and finance, and some of these are symmetric matrices, such as realized covariance matrices obtained from financial high-frequency data. This paper constructs a matrix heterogeneous autoregressive (MHAR) model for symmetric matrix-valued time series data to describe their dynamics. The coefficient matrices are estimated using the alternating direction method of multipliers, and the algorithmic convergence results and asymptotic properties of these estimators are established. Simulation results confirm these asymptotic properties. We then use our MHAR model to predict future realized covariance matrices and apply these predicted matrix values in portfolio management. Compared to other models for predicting realized covariance, our simulation studies and real-data analysis both demonstrate that our proposed model has superior predictive accuracy and investment performance.