Energy Shocks and Coffee Market Resilience under a Machine Learning Framework with the SDI+ Index

Abstract

This study analyzes the dynamic interdependence between Arabica coffee and Brent oil markets amid global energy shocks from 2010 to 2025. It employs a new Structural Decoupling Index (SDI+) combined with unsupervised machine learning algorithms, including K-Means, Gaussian Mixture Models, and Spectral Clustering. The analysis uncovers multiple dependence regimes and structural shifts. Findings indicate that 52% of the identified episodes involve strong decoupling, highlighting the fragile and episodic nature of the relationship between coffee and energy. The SDI+ predicts structural breaks up to 15 days in advance of traditional DCC-GARCH models, showcasing its potential as an early warning tool. Empirical results demonstrate that geopolitical and climatic shocks-particularly the Russia-Ukraine conflict and the 2023/24 ENSO event-heighten volatility and disrupt price transmission mechanisms. These insights advance the energy-agriculture literature by incorporating multiscale dependence metrics and machine learning for assessing commodity risk. The study provides practical recommendations for policymakers, cooperatives, and investors seeking to enhance the resilience of coffee-dependent economies against external energy shocks and global uncertainties.