EcoSta 2019: Start Registration
View Submission - EcoSta2019
Title: Penalized methods for quasi likelihood analysis with locally asymptotic quadratic properties Authors:  Yoshiki Kinoshita - The University of Tokyo (Japan) [presenting]
Nakahiro Yoshida - University of Tokyo (Japan)
Abstract: Penalized methods are applied to the quasi likelihood analysis in order to perform a variable selection in stochastic models. We consider locally asymptotic quadratic properties to describe a local structure of quasi likelihood functions. The polynomial type large deviation inequality is introduced to bound the tail of quasi likelihood functions. We estimate not only the limiting distribution of the estimator but also the probability that correct model is selected.