Predicting Global Stock Returns Using Commodities: A Gradient Boosting Decision Tree Approach

Abstract

We examine the predictability of stock returns using commodity futures prices across 39 countries from 1999 to 2024 using the XGBoost implementation of the Gradient Boosting Decision Tree approach. There is evidence of increased integration between commodities and stock markets. Despite this, research examining the predictability of commodities on stock returns is limited, especially on a global scale. The aim is to build on previous studies and explore heterogenous effects of commodity price changes on countries. We find evidence of predictability for four individual commodities and two commodity indices after sampling twelve commodities and four indices. Copper and crude oil show the strongest predictability among individual commodities, while industrial metals and energy demonstrate the strongest predictability among commodity indices. Our results also indicate strong heterogeneous effects, with some countries exhibiting significantly greater exposure to commodity prices. In particular, the Australian stock market is more exposed to price changes in copper and industrial metals, while the Norwegian market shows large sensitivity to oil and energy. Based on our findings, a competitive long-short trading strategy is proposed.