Does Signal Decomposition Improve Stock Market Forecasting? Uncovering High- and Low-Frequency Information with a Hybrid EMD–Random Forest Model

Document Type : Research Paper

Authors

1 Department of Economics and Management, Science and Research Branch, Islamic Azad University, Tehran, Iran.

2 Faculty of Management and Economics, Tarbiat Modares University, Tehran, Iran.

3 Department of Economics and Accounting, Imam Ali Officers' University, Tehran, Iran.

Abstract

Accurate equity index forecasting requires models capable of capturing temporal dependencies across multiple scales. This study develops a hybrid empirical mode decomposition–Random Forest (EMD–RF) framework and evaluates its performance against standalone Random Forest (RF), support vector machine (SVM), and XGBoost models across the CAC 40, S&P 500, SSE Composite, and Tehran Stock Exchange (TSE). Forecasting performance is assessed using root mean squared error (RMSE), mean absolute error (MAE), and mean absolute scaled error (MASE), complemented by Diebold–Mariano (DM) tests and component-level ablation analysis. EMD–RF achieves the lowest observed values for all three error metrics across all four indices, demonstrating consistent improvements over the benchmark models. However, statistical significance varies across markets: improvements are significant against all benchmarks for the CAC 40 and TSE, significant only against XGBoost for the S&P 500, and statistically inconclusive for the SSE Composite. Ablation results further show that high- and low-frequency intrinsic mode functions (IMFs) provide complementary predictive information across all four markets, although their relative contributions vary by index. The residual component also contributes additional predictive information to differing degrees. These findings demonstrate that EMD-based decomposition enhances RF forecasting performance by preserving informative temporal structures, while highlighting the market-dependent contributions of frequency-specific components and the importance of distinguishing observed forecast improvements from statistically established gains.

Keywords

Main Subjects

Files

History

  • Receive Date: 05 September 2026
  • Revise Date: 23 October 2026
  • Accept Date: 30 October 2026

Share

How to cite

Statistics