统计与管理学院2016年学术报告第16期
【主 题】Functional linear models with fixed effects
【报告人】 朱仲义
复旦大学
【时 间】 2016年5月13日（星期五）15:00－16:00
【地 点】 上海财经大学统计与管理学院大楼1208室
【摘 要】In this paper, we introduce a functional linear model with fixed effects for functional data where the predictor and response are random processes. The proposed model can be viewed as a generalization of the classical functional linear model, and characterizes individual specific source of variability. We implement the regularity procedure through a projection on the eigenfunction basis of the response process, leading to a special version of linear mixed effects model for panel data. In order to deal with the difficulty caused by a large number of individual effects, we use the penalty method to shrink individual effects, and propose a class of penalized least squares estimators. In a theoretical investigation, we establish asymptotic normality for the deviation between estimated and true regression function coefficients, and derive some asymptotic consistent properties for the predictions obtained from the fitted functional linear models with fixed effects. Some simulation studies and an application of intra-day volatility patterns of the $S\&P$ 500 index are conducted to illustrate the finite sample performance of the proposed modeling framework and estimation methods.
【邀请人】 周勇
