Forecasting commodity prices: empirical evidence using deep learning tools.

Area of Science:

• * Financial Markets and Operational Research
• * Artificial Intelligence (AI) and Data Science

Background:

• * Financial markets have transformed due to crises, increasing interest in alternative assets like commodities.
• * Advancements in AI, including machine learning and deep learning, offer new tools for market analysis.
• * Algorithm selection in AI is critical; deep learning is applied when machine learning is insufficient.

Purpose of the Study:

• * To investigate the effectiveness of deep learning algorithms for forecasting commodity prices.
• * To analyze the Bloomberg Commodity Index and its five subindices (Agriculture, Precious Metals, Livestock, Industrial Metals, Energy).
• * To evaluate forecasting performance using daily data from January 2002 to December 2020.

Main Methods:

• * Utilized deep learning algorithms for time-series forecasting.
• * Employed the Bloomberg Commodity Index and its component subindices.
• * Analyzed daily data spanning nearly two decades (2002-2020).

Main Results:

• * The Long Short-Term Memory (LSTM) method demonstrated significant effectiveness in commodity price forecasting.
• * The Bloomberg Livestock Subindex and Bloomberg Industrial Metals Subindex proved superior for assessing other commodity indices.
• * Findings highlight the practical utility of specific deep learning models and indices.

Conclusions:

• * Deep learning, particularly LSTM, is a valuable tool for commodity price forecasting.
• * Specific subindices offer superior insights for broader commodity market assessment.
• * Results provide crucial information for investor risk management and public policy adjustments, especially during geopolitical events.