Bayesian-Optimized Multi-Task Gaussian Process Regression With Composite Kernels for Soybean Oil Futures Forecasting

Accurate forecasting of agricultural commodity prices is critical for economic stability and strategic decision-making, yet remains challenging owing to nonlinear dynamics and multifactorial volatility. This study addresses the forecasting of soybean oil prices in China’s wholesale markets by developing a Bayesian-optimized Multi-Task Gaussian Process Regression (MTGPR) framework. Traditional econometric models and machine learning approaches often struggle with non-stationary data and uncertainty quantification, while existing Gaussian process applications in agricultural markets remain underexplored. We propose a composite kernel architecture integrating Matérn 2.5 and Exponential Sine Squared (ESS) kernels to capture localized price trends, periodic cycles, and heteroskedastic volatility. The framework incorporates futures market indicators and employs spectral analysis with Reproducing Kernel Hilbert Space (RKHS) theory for feature alignment. Using a decade-long dataset (2015–2024) of daily soybean oil prices, the model undergoes rigorous validation via cross-validated Bayesian optimization, bootstrap error estimation, and posterior predictive intervals. Comparative evaluations against Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Support Vector Regression (SVR) and Extreme Gradient Boosting (XGBoost) demonstrate MTGPR’s superiority, achieving a mean absolute percentage error (MAPE) of 0.89% and R² of 0.9238. Key innovations include adaptive kernel selection, probabilistic uncertainty bands for risk-aware strategies, and cross-market transferability analysis. Results confirm the model’s capacity to disentangle complex market patterns while maintaining computational tractability. This work advances agricultural price forecasting by bridging theoretical Bayesian methods with practical trading applications, offering policymakers and market participants a robust tool for hedging and supply chain optimization.