学术沙龙主题:On robustness of Prasad-Rao linearization-type MSPE estimators under model misspecification in small area estimation

报告人:刘小惠 江西财经大学教授

报告时间:2022年6月30日(周四)；上午9:00—10:30

报告地点：腾讯会议ID：302359056

邀请人：李本崇

报告人简介：刘小惠，江西财经大学统计学院，教授，博士生导师，副院长，主要研究领域为稳健统计、统计计算、时间序列分析、混合效应模型等，先后在《中国科学数学》，《数学学报》，Journal of Econometrics、Journal of Business & Economic Statistics、Journal of Computational and Graphical Statistics、Journal of Statistical Software、Statistica Sinica、Oxford Bulletin of Economics and Statistics及Annals of Tourism Research等国内外期刊上发表录用相关学术论文50余篇。先后主持江西省自然基金重点项目，杰出基金项目，国家自然科学基金地区项目、青年项目及面上项目等10余项省部级项目。

报告摘要:This paper is regarding the robustness of Prasad-Rao linearization-type MSPE estimators under potential model misspecification. It was thought that these methods could only work when the underlying model is correctly specified. However, the empirical results show that with respect to the unconditional MSPE estimation, the performance of Prasad-Rao linearization-type methods appear surprisingly satisfactory when the underlying model is misspecified in its mean function. In view of this, this paper attempts to prove the empirical robustness from a theoretical perspective. Both theorems and simulation results confirm that although the existence of model misspecification destroys the first/second unbiasedness, the distance between real MSPE and empirical MSPE will get smaller as the degree of model bias increases.