Time series models for multivariate series of count data
Time series models for multivariate series of count data
复制标题
多元计数数据序列的时间序列模型
DOI：
10.1007/978-1-4899-4515-0_21
复制
发表时间：
1993
期刊：
影响因子：
--
通讯作者：
A. Harvey
中科院分区：
文献类型：
--
作者：
K. Ord;C. Fernandes;A. Harvey
Harvey和Fernandes（1989）的早期论文（后来称为HF）提出了计数数据（即由非负整数组成的观测值）的各种时间序列模型。这些模型导致预测的基础上指数加权移动平均（EWMA）的参数确定贴现率计算的最大似然（ML）。本文考虑了一种方法来扩展这样的模型，以科普多变量时间序列的计数观测。在贝叶斯的背景下，计数数据的单变量处理已经由West，Harrison和Migon（1985）发展。HF提出的模型可以被认为属于结构时间序列模型（Harvey，1989）。这些模型是根据感兴趣的成分直接建立的。最简单的结构模型（局部水平加噪声）的形式为
An earlier paper, Harvey and Fernandes (1989), denoted subsequently as HF, proposed various time series models for count data, that is, observations consisting of non-negative integers. These models led to forecasts based on the exponentially weighted moving average (EWMA) with the parameter determining the rate of discounting being computed by maximum likelihood (ML). This paper considers a method for extending such models to cope with multivariate time series of count observations. In a Bayesian context, an univariate treatment of count data has been developed by West, Harrison and Migon (1985).The models proposed by HF can be regarded as falling within the class of structural time series models (Harvey, 1989). These are models which are set up directly in terms of components of interest. The simplest structural model, the local level plus noise, takes the form