Palm oil is the most traded edible oil globally, and accurate forecasting of its price is critical for producers, traders, and policymakers across Southeast Asia. Conventional time-series models assume stationarity and regime homogeneity, conditions rarely satisfied in commodity markets subject to geopolitical shocks, weather anomalies, and evolving policy environments. This study proposes a Wavelet-Neural Network (WNN) framework that (i) decomposes the Crude Palm Oil (CPO) price series into multi-resolution components using the Daubechies wavelet basis, (ii) quantifies market regimes via the rolling Hurst Exponent and Spectral Entropy, and (iii) trains a separate feedforward neural network on each regime-specific component before recombining predictions. Using weekly CPO price data (USD/MT) from January 2015 to December 2024 (n = 520 observations), together with four macroeconomic covariates — Brent crude oil price, USD/MYR exchange rate, soybean oil price, and a global demand index — the WNN model outperforms Support Vector Regression (SVR), XGBoost, Naïve Baseline, and Exponential Smoothing across all three identified market regimes. The overall out-of-sample MAPE of 2.31 % (Regime I), 3.95 % (Regime II), and 1.88 % (Regime III), together with statistically significant Diebold–Mariano test results (p < 0.05), confirm the superiority of the proposed approach. The regime-switching mechanism enables the model to adapt dynamically to structural breaks without requiring explicit intervention, offering a practical tool for commodity price risk management.
Copyrights © 2026