Estimating forecast variance with exponential smoothing Some new results

Stuart Bretschneider

Research output: Contribution to journalArticle

6 Scopus citations

Abstract

This note compares MAD and MSE smoothing approaches to estimating the forecast variance for the simple exponential smoothing forecast model. Using simulation techniques, the MSE approach is found to be more efficient than the MAD approach. These results hold for a wide assortment of cases in which both the mean and variance of the underlying demand series are potentially non-stationary. The results are found to be robust even in the presence of outliers. Several heuristic rules are developed for selection of a smoothing coefficient for the variance term, the most significant one being based on an inverse relationship between the optimal smoothing coefficient for the mean and the optimal coefficient for the variance.

Original languageEnglish (US)
Pages (from-to)349-355
Number of pages7
JournalInternational Journal of Forecasting
Volume2
Issue number3
DOIs
StatePublished - 1986

    Fingerprint

Keywords

  • Confidence interval - predictions
  • Data - simulation
  • Data errors - outliers
  • Exponential smoothing - monitoring
  • Forecast variance
  • time series

ASJC Scopus subject areas

  • Business and International Management

Cite this