报告地点：行健楼学术活动室526
邀请人：吴奕飞教授
Abstract: Factor-augmented regression is a well-established tool for time series forecasting, yet its application to high-dimensional, large-scale datasets stored across distributed nodes remains constrained by challenges of data privacy, computational efficiency, and integration of local versus global information. To address these limitations, we propose the hierarchical factor-augmented forward-validation (HFFV) model averaging method---a novel forecast combination framework tailored explicitly to distributed data structures. The HFFV method first deploys a hierarchical factor-augmented model that extracts and integrates two layers of information: node-specific (local) factors capturing unique dynamics of each distributed server and global factors reflecting common trends across the entire dataset. We establish the uniform convergence rate of its coefficient estimators to confirm the reliability of factor-augmented inference in distributed settings. To aggregate local forecasts into a robust global prediction, HFFV further uses a forward-validation procedure for estimating combination weights. This weight estimator is proven asymptotically optimal in the sense that its squared forecast risk attains the infeasible lower bound. Critically, HFFV automatically assigns all non-negligible weights to correctly specified candidate models, downweighting misspecified ones. Simulation studies confirm its strong finite-sample performance, while an empirical application to province-level Chinese macroeconomic data not only demonstrates its forecasting superiority over alternatives but also highlights how it leverages regional heterogeneity to improve macro forecasts.
报告人简介：涂云东，北京大学博雅特聘教授，联合受聘于光华管理*女王调教-女王调教视频-女王 调教小说商务统计与经济计量系和北京大学统计科学中心。入选“日出东方”北大光华青年人才，北京大学优秀博士学位论文指导教师(2017,2021,2024)，北京大学优秀研究生导师(2024)，教育部“长江学者奖励计划”青年学者，国家级人才项目获得者。先后获武汉大学理学学士学位(2004)和经济学硕士学位(2006)、加州大学河滨分校经济学博士学位(2012)。环亚太青年计量经济学者(YEAP)会议发起人和主要组织者。50余篇学术论文发表在多个国际国内知名专业杂志。著作教材《时间序列分析》由人民邮电出版社于2022年9月出版。研究领域涵盖时间序列分析、非参数计量方法、大数据分析、金融计量和预测等。
