Skip to main navigation Skip to search Skip to main content

Modeling commodity price co-movement: building on traditional time series models and exploring applications of machine learning algorithms

  • Luca L. Kozian
  • , Marcos R. Machado*
  • , Joerg R. Osterrieder
  • *Corresponding author for this work

Research output: Contribution to journal › Article › Academic › peer-review

134 Downloads (Pure)

Abstract

In this study, we explore the efficacy of various methodologies and co-movement measures for modeling the co-movement of cross-commodity prices, using macroeconomic variables. Applying Vector Autoregression (VAR), VAR with exogenous variables (VARX), multiple regressions, and Random Forest regressions, alongside Pearson correlations and Gerber statistics, we analyze the price co-movement of 20 key commodities over the period from mid-2003 to early 2023. Our results reveal that VAR and VARX models notably outperform Random Forests and multiple regressions, achieving R2 values of up to 89%. Although Random Forests marginally outperform multiple regressions, they are still inferior to VAR-based approaches. Furthermore, we observe that the inclusion of Gerber statistics (differently from the Pearson correlation metric) enhances the effectiveness of VAR and VARX models, while their influence on Random Forests and multiple regression models remains indeterminate. This research contextualizes and contributes to the existing literature on commodity price dynamics, providing insights into the relative strengths of various modeling approaches and setting the stage for future methodological advancements, especially in the applications of Machine Learning (ML) in the field.

Original languageEnglish
Article number109283
Pages (from-to)111-154
Number of pages44
JournalDecisions in economics and finance
Volume49
Issue number1
Early online date2 Feb 2025
DOIs
Publication statusPublished - Jun 2026

Keywords

  • UT-Hybrid-D
  • Commodities
  • Gerber statistic
  • Random Forest
  • VAR
  • VARX
  • Co-movement

Fingerprint

Dive into the research topics of 'Modeling commodity price co-movement: building on traditional time series models and exploring applications of machine learning algorithms'. Together they form a unique fingerprint.

Cite this