Optimizing smoothing parameters for the triple exponential forecasting model

Exponential smoothing has always been a popular topic of research in forecasting. The triple exponential smoothing in particular involves modeling a function that is a combination of level, trend and seasonal factors. While simulating the model, each of the factors is associated with a parameter whose value has a significant impact on the accuracy of the forecast, yet optimizing these parameters for a time series has received relatively little attention in literature. In this thesis we will explore the results of multi-step forecasting by using parameters optimized through an algorithm centered around h-step ahead errors. An empirical study conducted on forecasting the monthly time series from the M3-Competition across a range of horizons gave us promising results. We show that this method proves to be better than the standard Holt-Winters procedure for the entire forecasting horizon in five out the six categories of data considered . We also show that this method significantly improves the accuracy over the short term forecasting horizon when compared to the automated Holt-Winters procedure used by experts in the M3 competition. Encouraged by these results, we recommend replicating this methodology to other models of the triple exponential smoothing in the future.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Industrial Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2016

Author dc:creator

• Narasingaraj, Harish Balaji

Contributors dc:contributor

• Nagi, Rakesh

• Holt Winters
• Triple Exponential Smoothing parameters
• M3 Competition

Statement dc:rights

• Copyright 2016 Harish B Narasingaraj

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/90834

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/90834