🧠 Explainable AI in Finance: Fundamental Factor Modeling

This repository explores how Explainable Artificial Intelligence (XAI) methods can be applied to machine learning models used for a case study in fundamental factor investing. It accompanies the CFA Institute's report on 'Explainable AI in Finance: Addressing the Needs of Diverse Stakeholders' (2025), illustrating the practical use of post-hoc explainability tools in financial modeling.

📊 Project Overview

We implement a supervised learning workflow to predict monthly excess returns of stocks based on fundamental factor exposures using the XGBRegressor model from XGBoost. Our dataset contains normalized exposures to six BARRA-style factors across 218 stocks over a 100-month period starting in February 2008. We thank Bloomberg LP for use of the price and factor loadings data in this case study.

🧪 Workflow

1. Data Preprocessing
   The data provided has already been re-scaled.

2. Model Training

   • Split data into training and test sets
   • Use cross-validation to optimize hyperparameters
   • Train final model on full training set

3. Model Evaluation
   Assess predictive performance using root mean squared error (RMSE).

4. Explainability Methods
   We apply multiple XAI methods to understand model behavior:

   • Feature Importance: Gain-based feature ranking from XGBoost
   • Partial Dependence Plots (PDP): Global sensitivity to features
   • Individual Conditional Expectation (ICE): Instance-level visual insights
   • SHAP (SHapley Additive Explanations): Local and global contribution of features
   • LIME (Local Interpretable Model-agnostic Explanations): Simplified local explanations

Each technique aligns with post-hoc explainability practices identified in the CFA Institute's XAI report and provides a unique lens into model transparency.

🔍 Key Themes

• Post-Hoc Explainability: Applied to an ensemble model (XGBoost), demonstrating outcome-based and instance-level interpretability.
• Financial Relevance: The model emulates a factor-investing strategy, a common use case in portfolio construction and performance attribution.
• Stakeholder Interpretability: Tools used here can support explainability needs for different users, such as quantitative analysts, model validators, and regulators.

🧰 Tech Stack

• Python 3.x
• Jupyter Notebook
• XGBoost
• SHAP
• LIME
• Scikit-learn
• Matplotlib

📁 Files

• Notebook.ipynb: Main notebook demonstrating modeling and explainability
• data/: Folder containing processed datasets
• Plots/: Folder containing SVG files of the Notebook plots
• README.md: Project overview and usage guide

📘 Reference

This notebook was developed to complement the CFA Institute’s research on human-centered AI systems in finance, particularly:

"Explainable AI in Finance: Addressing the Needs of Diverse Stakeholders (CFA Institute, 2025)"

🤝 Contributing

Feel free to open issues or contribute improvements to expand examples, integrate more models (e.g., linear regression, neural networks), or include additional XAI methods (e.g., counterfactuals, rule-based models).

📜 License

MIT License