Day & Time
21st November 2014, 16:40~18:20
Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall), The University of Tokyo
Name and Affiliation
(Emeritus Professor of Hiroshima University, Advisor of the Hiroshima University Statistical Science Research Core)
High-Dimensional Consistency Properties of Estimation Methods of the Rank in Multivariate Linear Model
This talk is concerned with the problem of estimating the rank in multivariate linear model. As one of our main results we show consistency properties of AIC, BIC and Cp under a high-dimensional asymptotic framework. The results are extended for the criteria with tuning parameters. We consider ridge-type criteria whose performances are studied through numerical experiments. We also discuss methods for selecting the relevant variables in reduced-rank regression model and for selecting the endogenous variables in a single structural equation in econometrics.