报告题目： Effective Model Averaging for Generalized Linear Models in Diverging Model Spaces
报告摘要：Model averaging has attracted a lot of attentions in the past decades as a powerful forecasting tool in statistics and econometrics. While abundant frequentist model averaging methods have been proposed for many models, existing weight choice criteria for generalized linear models are based on a model size penalized Kullback-Leibler (KL) loss or just cross-validation. In this paper, we propose a novel model averaging approach for generalized linear models motivated by an asymptotically unbiased estimator of the KL loss which is penalized by an ``effective model size". When all the candidate models are misspecified, the proposed method has asymptotic optimality as usual, but it allows number of candidate models and dimension of covariates both to be diverging. Furthermore, when correct models are included, we prove that the weight vector of the wrong candidate models converges to zero and hence the weighted regression coefficient estimator is consistent. Simulation studies and real data examples show strong merits of our new method over the existing frequentist model averaging methods for generalized linear models.
报告人简介：方方，华东师范大学bwin必赢线路教授。本科和博士先后毕业于北京大学数学系和美国威斯康星大学统计系。在2013年加入华东师大之前，曾在通用电气金融集团和上海浦东发展银行任职多年。主要研究方向为缺失数据、模型平均、多源碎片化数据分析。在包括 AOS、JOE、Biometrika 在内的国际一流统计期刊上发表论文30余篇。主持国家自然科学基金重点项目子课题、面上项目、青年项目各1项。参与国家重点研发计划重点专项课题2项和上海市重点项目1项。曾获上海市自然科学二等奖。SCI期刊 Journal of Nonparametric Statistics 副主编。出版统计科普小说《统计王国奇遇记》。