Abstract：This study develops a rolling forecasting model for commodity futures prices based on a decomposition-reconstruction mechanism to address challenges associated with nonlinearity, strong noise, and endpoint effects. The methodology comprises four steps: performing multi-scale decomposition on the price series using the ICEEMDAN algorithm, selecting features from each sub-sequence via Adaptive Lasso, merging the sub-sequences via an optimization algorithm for reconstruction, and building a forecasting model using an extreme learning machine (ELM). Empirical evaluations on multiple futures varieties show favorable forecasting performance in both level and directional dimensions. Additional back-testing of trading strategies demonstrates that the proposed model outperforms several benchmarks in terms of cumulative returns, win rate, and risk metrics. The results indicate that controlling endpoint effects can substantially improve the predictive performance of decomposition-prediction models and exhibit strong applicability in commodity futures markets.
姜旭初，李明，魏子琛，李斌. 基于分解-重构混合模型的期货价格预测及末端效应控制研究[J]. 管理学报, 2025, 22(11): 2158-.
JIANG Xuchu，LI Ming，WEI Zichen，LI Bin. Futures Price Prediction and End Effect Control Based on a Decomposition-Reconstruction Hybrid Model. Chinese Journal of Management, 2025, 22(11): 2158-.
