Wood-based panel futures price prediction incorporating supply chain features

This paper proposes a wood-based panel futures price prediction method incorporating supply chain features, aiming to improve prediction accuracy and explore price formation mechanisms. The model constructs a multi-dimensional feature system by integrating upstream material price indices including timber, chemical raw materials, and energy, as well as downstream indicators such as the construction industry prosperity index. To resolve the heterogeneity between daily price data and monthly supply chain features, we design an innovative dual-frequency fusion network (DM-FusionNet). This network processes time-series dependencies in daily data through a bidirectional LSTM branch, extracts long-term patterns from monthly features through a lightweight Transformer branch, and organically combines both types of information through a dynamic fusion mechanism. Experimental results demonstrate that compared to traditional methods, the proposed model achieves significant improvements across multiple evaluation metrics, with a 16.8 % reduction in MSE and an improved R2 value of 0.870. The model exhibits robust performance across different time scales and market conditions, particularly achieving a trend prediction accuracy of 96.1 % over a 60-day prediction period. Feature importance analysis reveals a “dual-dominant” influence mechanism constituted by downstream demand indicators (NHPI) and upstream chemical raw material prices. This research not only enriches the methodological framework for commodity futures price prediction but also provides a practical decision-support tool for participants in the wood-based panel futures market.