北航经管学院“经济统计论坛”系列讲座
（2026年第1期，总第38期）
闫晗博士后讲座通知
讲座题目：Statistical Inference for General Moment Restrictions under Covariate Shift Adaptation
讲座时间：2026年1月14日（周三），10:30-11:30
会议地址：新主楼 A618
讲座嘉宾：闫晗 伦敦政治经济学院博士后
讲座嘉宾
Han Yan is a Research Officer in the Department of Statistics at the London School of Economics and Political Science (LSE), working with Professor Qiwei Yao. His current research interests include time series analysis, distribution shift problems and deep learning. Prior to joining LSE, he obtained his PhD in statistics at Peking University in 2025, under the supervision of Professor Song Xi Chen.
邀请人：康雁飞 教授
讲座概要
This study considers statistical inference with the presence of covariate shift for parameters defined by general moment restrictions. Different from the commonly used density ratio weighting approach, we construct the Neyman orthogonal moment function, which mitigates the first-order effect of nuisance function estimation. The Neyman orthogonal moment function involves the density ratio and the conditional mean of the moment function as nuisance functions. The deep multiple imputation method bypasses the need for repeated regression at many possible parameter values for the conditional mean function estimation. A semiparametric efficient estimator is formulated via the empirical likelihood based on the orthogonal moment functions with the DNN-based nuisance function estimators. We show that the empirical likelihood formulation permits Wilks's theorem, which facilitates simple inference despite the presence of nuisance functions. The proposed method is evaluated by simulations and an empirical study on an air quality study.
