A Study of Modeling of Nonstationary Time Series based on Poisson Distribution

摘要

Gaussian and linear state space model is a model for nonstationary time series. When we deal with non-Gaussian data based on the poisson distribution, however, this model is not adequate for the analysis. So some approaches to extend Gaussian and linear state space model have been proposed. For example, West et al proposed the Dynamic Generalized Linear Model. One of problems on his model is that exact posterior distribution of the parameter can not be obtained. In this paper, we propose model that enable us to calculate exact posterior distribution of the parameter analytically. Additionally, we show some simulataion results to make sure properties of this model.