EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3537. A deep learning approach to forecasting commodity prices

Invited abstract in session WB-2: Optimal Portfolio Strategies, stream OR in Banking, Finance and Insurance: New Tools for Risk Management.

Wednesday, 10:30-12:00
Room: Glassalen (building: 101)

Authors (first author is the speaker)

1. Hayette Gatfaoui
Finance, IESEG School of Management

Abstract

Commodity prices are a strategic concern to countries and governments (e.g., energy or food security). Producers and end-users are sensitive to commodity price swings and volatility, which may imply lower profits and higher energy bills respectively. Commodity prices depend on supply and demand fundamentals but also specific drivers such as weather-linked seasonality and events, regional/country-specific infrastructures (e.g., bottlenecks) and regulatory/compliance constraints.
Commodity price forecasting is important to allow market participants to hedge their risk exposures (i.e., losses due to large price swings). We apply a deep learning approach to few commodities for forecasting prospects. In the area of big data and artificial intelligence, deep learning allows for detecting key patterns in commodity prices and exploit them for forecasting purposes. Such an approach is data driven and relies on neural network analysis without requiring explicit fundamental factors. Moreover, the dynamic complex behavior of commodity prices is captured. A study of several forecasting windows is also performed. Besides, we also perform an improved deep learning approach by including very few key fundamental factors to check if these ones improve the forecasting degree of the method and/or increase the forecast horizon. Hence, we test for the usefulness of an augmented information space on the forecasting power and accuracy of deep learning in the context of few commodity markets.

Keywords

Status: accepted

Back to the list of papers