Demand forecasting with four-parameter exponential smoothing

doi:10.1016/j.ijpe.2016.08.004

Return

Demand forecasting with four-parameter exponential smoothing

2016-11-01

PRE

Liljana Ferbar Tratar

Blaž Mojškerc

Aleš Toman *

DOI：10.1016/j.ijpe.2016.08.004

Original

Original request for help

Save

Abstract

En 中文

Exponential smoothing methods are powerful tools for denoising time series, predicting future demand and decreasing inventory costs. In this paper we develop a smoothing and forecasting method that is intuitive, easy to implement, computationally stable, and can satisfactorily handle both, additive and multiplicative seasonality, even when time series contain several zero entries and large noise component. We start with the classical additive Holt-Winters method and introduce an additional smoothing parameter in the level recurrence equation. All parameters are required to lie within [0,1 and estimated by minimizing the one-step-ahead forecasting errors in the sample. Doing so, the errors decrease sub-stantially, especially for the time series with strong trends. The newly developed method produces more accurate short-term out-of-sample forecasts than the classical Holt-Winters methods and the Holt-Winters methods with damped trend. The performance of the method is evaluated using a battery of real quarterly and monthly time series from the M3-Competition. A simulation study is conducted for further in-depth analysis of the method under different demand patterns. We developed and justified the use of a symmetric relative efficiency measure that allows researchers ad practitioners to evaluate the performance of different smoothing and forecasting methods. (C) 2016 Elsevier B.V. All rights reserved.

Keywords:

Demand forecasting

Exponential smoothing methods

Seasonal data

Holt-Winters methods

Damped trend methods

M3-Competition

Individual products

Symmetric relative efficiency measure

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.