貝氏估計在時間序列上的應用

Abstract

我們在時間序列的模型下,提出估計的程序在貝氏的架構下,而主要的障礙牽涉到其初始狀態.我們利用在Wise(1951)所提到的exact likelyhood function用來做參數估計,我們也提出所選取的先驗分配不會與時間序列中的stationarity衝突,在Chib(1993)和Chib and Greenberg(1994)並沒有去做考慮,而經過模擬之後我們得到比較準確的推論。

We propose an estimation procedure for the time series regression models under the Bayesian inference framework. The major obstacle for estimating a time series involves its initial states. With the exact method of Wise (1951), an exact likelihood function can be obtained, which can be used to estimate the parameters. We also propose a prior that does not conflict with the stationarity of the time series, which was not token into consideration by Chib(1993) and Chib and Greenberg(1994). Simulation studies show that our method leads to more accurate inferences.