Bayesain Estimation of ACP(1,1) with a Change point
| Autor: | Yao-Ting, Wang, 王耀霆 |
|---|---|
| Rok vydání: | 2014 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 102 Structural change can happen in a time series of counts. The goal of this paper has two fields. First, we review the development of integer-valued time series models. Second, we propose autoregressive conditional Poisson (ACP) models with a change point via a Bayesian approach. This change point is unknown in the ACP model, which allows asymmetric unconditional mean in the two time spans. Bayesian MCMC methods are used to estimate parameters simultaneously in the mean process and the change point. The change point is estimated by a posterior mode since it is an integer value. We also propose a diagnostic checking based on important sampling method which is within a Bayesian approach. We conduct a simulation study to investigate the estimation performance of ACP model with a change point. In order to illustrate the proposed method, three real data sets are considered, in which the ACP model and ACP model with a change point are applied to the crime data sets in Australia. We apply the Ljung-Box test and normality tests to check the generalized residuals whether the proposed model is adequacy. The diagnostic checking result show that there is no sign of inadequacy in all three data series. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|