ON THE CUMULATED MULTI-STEP-AHEAD PREDICTIONS OF VECTOR AUTOREGRESSIVE MOVING AVERAGE PROCESSES

被引：7

作者：

DEGOOIJER, JG

KLEIN, A

机构：

[1] Department of Economic Statistics, University of Amsterdam

来源：

INTERNATIONAL JOURNAL OF FORECASTING | 1992年 / 7卷 / 04期

关键词：

CUMULATED PREDICTIONS; KULLBACK-LEIBLER INFORMATION MEASURE; MULTI-STEP-AHEAD PREDICTIONS; RENYIS INFORMATION MEASURE; VECTOR AUTOREGRESSIVE MOVING AVERAGE PROCESSES;

D O I：

10.1016/0169-2070(92)90034-7

中图分类号：

F [经济];

学科分类号：

02 ;

摘要：

When a time series model is used for making predictions, then it is often meaningful to evaluate its performance on the basis of cumulated multi-step-ahead prediction errors. In this paper some theoretical properties of cumulated multi-step-ahead predictors and cumulated multi-step-ahead prediction errors for vector autoregressive moving average processes are considered. A general expression for the optimal cumulated multi-step-ahead predictor is derived. The predictors are based on the Kalman filter algorithm. To determine the maximum prediction horizon of cumulated multi-step-ahead predictions, two information measures are introduced. For univariate ARMA(p, q) processes with p < 3 and q < 3, these measures are evaluated analytically as well as numerically. It is shown that the information content of cumulated multi-step-ahead predictions depends on the prediction horizon and the location of the roots of the AR and MA polynomials.

引用

页码：501 / 513

页数：13

共 15 条

[1]

Anderson B., 1979, OPTIMAL FILTERING

[2]

ASHLEY R, 1983, J FORECASTING, V3, P211

[3]

Astrom K.J., 1970, INTRO STOCHASTIC CON

[4]

BAILLIE RT, 1979, BIOMETRIKA, V66, P675