ROBUST ESTIMATION OF LEVEL AND TREND

被引：3

作者：

FERELLI, M

WILSON, GT

机构：

[1] UNIV LANCASTER,DEPT MATH,LANCASTER LA1 4YL,ENGLAND

[2] UNILEVER RES LABS,WIRRAL L62 4XN,CHESHIRE,ENGLAND

来源：

JOURNAL OF FORECASTING | 1990年 / 9卷 / 02期

关键词：

Bayes; Non‐gaussian; Non‐linear filtering; Numerical; Outliers; Robust time series; State space models;

D O I：

10.1002/for.3980090207

中图分类号：

F [经济];

学科分类号：

02 ;

摘要：

Numerical state space models are efficiently implemented for the estimation of the underlying level and trend of a time series. The model specification is chosen so that the estimation is insensitive to outliers yet adapts rapidly to step changes in level. An example illustrates, by means of projection plots, how at times of uncertainty in the evolution of the series the inferred distribution of level and trend may be multi‐modal. Copyright © 1990 John Wiley & Sons, Ltd.

引用

页码：151 / 172

页数：22

共 7 条

[1]

Gersch W, 1988, BAYESIAN ANAL TIME S, P431

[2]

HARRISON PJ, 1976, J R STAT SOC B, V38, P205

[3] NON-GAUSSIAN STATE-SPACE MODELING OF NONSTATIONARY TIME-SERIES [J].