[Statlist] Reminder *Research seminar in statistics*

[Sylvie Huber]

RESEARCH SEMINAR IN STATISTICS



Organisers :                                                                                   .

E. Cantoni - E. Ronchetti - S. Sperlich - M-P. Victoria-Feser



Friday February 21st, 2014

at 11h15

Room M 5220, Uni Mail (40, bd du Pont-d'Arve)



Prof. Taisuke OTSU

London School of Economics



«Relative Error Refinement in Nonparametric Likelihood»

Abstract:

This article proposes a class of tests for moment condition models that admit higher order refinements in relative errors. In particular, we focus on tail or large deviation behaviors of p-values of tests under null hypotheses, and show that chi-square calibrations of tilted analogs of the generalized empirical likelihood statistics admit the relative errors of order O(1/n), which is typically attained by saddlepoint approximations. Compared to higher order analysis for absolute errors, commonly applied to the bootstrap, based on Edgeworth expansions, our higher order analysis for relative errors based on saddlepoint approximations is more relevant to investigate tail properties. In addition to simple hypothesis testing for just-identified moment conditions, composite hypothesis and testing for over-identified moment conditions is also considered.
http://www.stat-center.unige.ch/ResSem.html

Sylvie Huber
Université de Genève
GSEM Geneva School of Economics and Management
Secrétaire

Bureau 5263, 5ème étage
UNI MAIL, Bd du Pont d'Arve 40
Suisse - 1211 Genève 4
Tél. : + 41 22 379 82 63
Fax : + 41 22 379 82 93
E-mail : Sylvie.Huber at unige.ch<mailto:Sylvie.Huber at unige.ch>


	[[alternative HTML version deleted]]