Adaptive Forecasting with an AR(1) Model
Abstract
Updating formulas for the forecasts of a one-parameter autoregressive model are obtained when the parameter is assumed random. It is shown that the updated forecasts are similar to those derived from exponentially weighted moving average forecasts with the important difference that forecasts can lie outside the interval containing the old forecast and the new observation. Based on the growth of the new observations the updated confidence intervals may become larger or smaller than the old ones. Similarities to and differences between a Box-Jenkins model, a Kalman Filter and a model proposed by Makridakis and Wheelwright are illustrated.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1977
- Accession Number
- ADA060337
Entities
People
- Robert M. Oliver
Organizations
- University of California, Berkeley