 
Summary: UNIVERSITY OF REGINA
Department of Mathematics and Statistics
Graduate Student Seminar
Speaker: Xiaojing Wu
Date: 26 April 2006
Time: 11.00 o'clock
Location: College West 307.20 (Math & Stats Lounge)
Title: GMM estimation in integervalued AR(1) time series models
Abstract: A couple of discrete time series AR (1) models with Poisson and Negative
Binomial marginals have been proposed by several authors. Due to having quite complex
conditional densities, those models seem to be more feasible to be estimated by using the
Generalized Method of Moment (GMM) than conditional MLE methods. Even though the
GMM estimators are well known to be asymptotically consistent with the true parameter
values, there is still considerable smallsample bias with those GMM estimators. In this
seminar, the GMM procedures as well as the corresponding estimating functions for those
discrete AR (1) models will be presented first. Furthermore, an analytical form of the
smallsample bias will also be given in the seminar. The methods to derive the analytical
form will be presented as well. The simulation results to be reported in the end of seminar
show that the estimation bias can be largely reduced by introducing this analytical form
into the estimating function and thus improve the precision of the GMM estimators for
