“经济统计论坛”系列讲座
2025年第5期，总第36期
题目：Enhancing Forecast Reconciliation for Hierarchies of Time Series via Model Combination
主讲人：李茜茜，清华大学博士后
时间：2025年11月28日（星期五），10:00-11:00
地点：新主楼A618
报告摘要：
Forecast reconciliation for hierarchical time series and model combination—two closely related areas—have received significant attention in recent years. This paper unifies these two perspectives by reformulating forecast reconciliation as a model combination problem and formally establishing their equivalence. Building on this unified framework, we develop a quadratic programming formulation for estimating combination weights, which naturally encompasses the existing MinT forecast reconciliation as a special case. Several extensions illustrate the flexibility of the framework, allowing for covariance shrinkage or simplification and penalized optimization of combination weights within the same quadratic programming formulation. In addition, we show that the approach can be readily used as the basis for generating probabilistic forecasts. Empirical analyses of real-world datasets show strong potential for our proposed reconciliation approaches.
主讲人简介：
Xixi Li is a postdoctoral researcher in the Department of Management and Engineering at Tsinghua University, specializing in statistical time series forecasting and causal inference under the supervision of Prof. Xiaojie Mao. His research is supported by the Shuimu Tsinghua Scholar Program. His work focuses on developing novel forecasting methods by integrating interpretable statistical machine learning approaches with domain expertise in business. He earned his Ph.D. in Statistics from The University of Manchester, where he was advised by Dr. Jingsong Yuan. As the first or corresponding author, he has published in some journals, including the Journal of Applied Statistics, International Journal of Forecasting (four articles), International Journal of Production Research, and Expert Systems with Applications.
