Wavelet- and SVM-based forecasts: An analysis of the U.S. metal and materials manufacturing industry

Research output: Contribution to journalArticlepeer-review

35 Scopus citations

Abstract

This article compares four non-linear forecasting methods: multiplicative seasonal ARIMA, unobserved components (UC), wavelet-based and support vector machines (SVM). Whereas the first two methods are well known in the time series field, the other two rely on recently developed mathematical techniques. Based on forecasting accuracy and encompassing tests applied to shipments data of the U.S. metal and material manufacturing industry for 1958-2000, we conclude that that these two novel forecast techniques can either outperform the traditional ones or provide them with extra forecast information. In particular, based on the Granger-Newbold test, it appears that wavelets may be a promising new technique.

Original languageEnglish
Pages (from-to)80-89
Number of pages10
JournalResources Policy
Volume32
Issue number1-2
DOIs
StatePublished - Mar 2007

Keywords

  • Forecasting
  • Support vector machines
  • Wavelets

Fingerprint

Dive into the research topics of 'Wavelet- and SVM-based forecasts: An analysis of the U.S. metal and materials manufacturing industry'. Together they form a unique fingerprint.

Cite this