A0485
Title: Nonparametric maximum likelihood estimation for GINAR(p) models
Authors: Taito Kihara - Keio University (Japan) [presenting]
Abstract: The nonparametric maximum likelihood estimator (NPMLE for short) for GINAR(p) models, which is based on generalized thinning operator, is defined and its asymptotic properties are discussed. An estimation method of semiparametric INAR(p) models has been previously proposed, however, estimation methods of semiparametric GINAR(p) models have not been addressed before. Previous work is extended to GINAR(p). Furthermore, semiparametric GINAR(p) models are more flexible than parametric GINAR(p) models when one analyzes actual integer-valued data. Particularly, it would be a useful tool to analyze data when underlying innovation distribution is multimodal or hard to detect the shape of the distribution function. We will show a numerical experiment with a GINAR(p) model which has multimodal innovation distribution as the data generating model to show the consistency of NPMLE which we can prove theoretically.
