Machine Learning Inveils Hidden Price Responsiveness: A SHAP-based Analysis of Indonesia's Aggregate Coffee Production Under Global Price Volatility

This study employs a machine learning framework, utilizing Random Forest regression and SHAP value analysis, to investigate the nonlinear price responsiveness of Indonesian coffee production to global market fluctuations. The findings reveal distinct patterns between Arabica and Robusta varieties. While Arabica exhibits greater variability in its price elasticity (coefficient of variation: 157.2%; range: -1.3236 to 19.4309), Robusta demonstrates a stronger absolute responsiveness (range: -0.4260 to 25.5859) with lower variability (139.4%). A critical finding is the inverse relationship between Robusta prices and its production, which contrasts with the positive correlation observed for Arabica. These results underscore the necessity for variety-specific policy, suggesting a focus on Robusta price monitoring due to its heightened sensitivity. The study also highlights a key methodological limitation stemming from the use of aggregate production data, underscoring the urgent need for Indonesian statistical agencies to provide disaggregated statistics by variety. This would enable more precise policy analysis and affirms the value of advanced analytical techniques in addressing complex agricultural supply responses in commodity markets.