HybridStack: Explainable hybrid ensemble for accurate inflation forecasting with high-frequency macro-financial data

Abstract & Highlights

Macro-financial time series forecasting faces severe non-linearities, regime changes, and structural shifts. HybridStack introduces a multi-tier hybrid ensemble framework combining gradient-boosted decision trees (XGBoost, LightGBM, CatBoost) coupled with regularized meta-regressors to deliver robust predictive power alongside transparent model attribution.

• Held-Out Test RMSE: 0.0006, significantly outperforming standard statistical models and single-architecture deep neural baselines.
• Hyperparameter Optimization: Automated multi-dimensional Bayesian optimization spanning 10 high-frequency macro-financial indicator sources.
• Policy-Grade Explainability: Full integration of TreeSHAP, Kernel SHAP, Local Interpretable Model-agnostic Explanations (LIME), and Partial Dependence Plots (PDP) to quantify exact marginal feature attribution and macroeconomic predictor dynamics for central banks and policy institutions.

Authors

• S. M. Mohiuddin Khan Shiam (BRAC University)
• Meherunnesa Neela
• Amrijit Biswas
• Mahdy Rahman Chowdhury (North South University)

Citation & DOI

@article{shiam2026hybridstack,
  title={HybridStack: Explainable hybrid ensemble for accurate inflation forecasting with high-frequency macro-financial data},
  author={Shiam, S. M. Mohiuddin Khan and Neela, Meherunnesa and Biswas, Amrijit and Chowdhury, Mahdy Rahman},
  journal={Array},
  volume={31},
  pages={101003},
  year={2026},
  publisher={Elsevier},
  issn={2590-0056},
  doi={10.1016/j.array.2026.101003}
}

Read on ScienceDirect (Elsevier)

Recommended citation: S. M. Mohiuddin Khan Shiam, Meherunnesa Neela, Amrijit Biswas, and Mahdy Rahman Chowdhury. (2026). "HybridStack: Explainable hybrid ensemble for accurate inflation forecasting with high-frequency macro-financial data." Array, Elsevier, Vol. 31, Art. 101003, ISSN: 2590-0056. DOI: 10.1016/j.array.2026.101003.
Download Paper