On Multi-Horizon Forecasting of Copper Price Returns Using Deep Learning Techniques

Abstract

Forecasting copper prices is vital for stakeholders in industries reliant on this commodity. The challenge arises from the market's dynamism and the multitude of factors affecting prices. This study introduces neural network models for predicting short-term copper price returns. Utilizing historical pricing data and macroeconomic indicators from 2007 to 2021, we discover that models dedicated to specific forecasting horizons outshine those designed for multiple horizons. Notably, Long Short- Term Memory (LSTM) models consistently delivered the most accurate predictions for both one-week and one-month future returns, confirming their robustness in capturing the complex patterns inherent in the copper market.