Empirical Economics Social Sciences (miscellaneous)

Non-ferrous metal price prediction based on a three-stage prediction model and optimized double decomposition

Wang, Rui, Bai, Yulong, Song, Qi, Liu, Wenhui
Springer Nature · 2026-10-07 · DOI 10.1007/s00181-026-02984-5

Abstract

Accurate prediction of non-ferrous metal prices is crucial for decision-makers, yet it remains highly challenging due to the high volatility and nonlinearity of the markets. This study presents a three-stage hybrid model that integrates marine predators algorithm-optimized variational mode decomposition (VMD), bidirectional long short-term memory (BiLSTM) network, and ARIMA-based error correction. The method first applies MPA-optimized VMD to decompose the price series into meaningful components. Each subseries is then predicted using BiLSTM, while the component with the highest prediction error is further decomposed to capture underlying patterns. Finally, residual errors are corrected using ARIMA to improve prediction accuracy. Experimental evaluation based on aluminium, copper, and gold futures data from the London Metal Exchange demonstrates the improved performance of the proposed model compared with 12 benchmark methods. The results show that: (1) MPA-optimized decomposition leads to more effective component extraction than standard VMD; and (2) the hybrid model achieves competitive forecasting accuracy across all tested metals, offering practical insights for market participants.

Read at publisher →