Regularization parameter selection for penalized empirical likelihood estimator in misspecified models


3

Day & Time
4th March 2016, 15:00-16:00
Venue
Room C816, Building of the Faculty of Science Graduate school of Science
Lecturer
Prof. Naoya SUEISHI (Kobe University)
Presentation Title
Regularization parameter selection for penalized empirical likelihood estimator in misspecified models
Abstract
This paper considers the issue of regularization parameter selection for penalized empirical likelihood estimator in possibly misspecified moment restriction models. As previous literature has focused on asymptotic properties of penalized estimators, there is no selection method of a regularization parameter that has a sound theoretical background. The main contribution of this paper is to propose a novel information criterion for selecting the regularization parameter of the penalized empirical likelihood estimator in a date-driven way. Our information criterion is derived as a bias-corrected estimator of the expected value of the Kullback-Leibler information criterion from the estimated model to the true data generating process.Furthermore, we investigate the asymptotic properties of the penalized empirical likelihood estimator under misspecification. Herein, the consistency, asymptotic normality, and oracle property are established. These results are also new in the literature.