Regulatory Risk Intelligence

Running the pipeline

1. Place raw OCC data in 'data/raw/occ_enforcement.csv
2. Run: python -m src.preprocssing A cleaned dataset will be written to 'data/processed/'. A 200-row sample is included in 'data/sample/ for demonstration.

Regulatory Risk Intelligence

A forward-looking risk model that predicts whether a bank will receive a regulatory enforcement action within the next 6 months.

Problem Statement

Regulatory enforcement actions often occur in clusters. Once a bank receives an enforcement action, it is more likely to experience additional actions in subsequent months. Risk teams need a structured way to prioritize monitoring resources. This project answers: Given historical enforcement activity, can we estimate the probability of a bank receiving another enforcement ation in the next 6 months?

Data Used

• OCC enforcement action history data
• Monthly bank0level panel dataset
• Rolling 6-month event counts
• Peer enforcement activity metrics Each observation represents a: (bank,month) The target variable:
• y_next_6m = 1 if enforcement occurs in the rolling 6 months Time-based split was used to prevent lookahead bias.

Modeling Approach

events_last_6m -> bank-specific enforcement persistence peer_events_last_6m -> systemic regulatory activity Forward 6-month label construction Time-based train/test validation

Baseline model: Logistic Regression No leakage Imbalanced classification evaluation

Results

ROC-AUC:0.95 Top Decile Capture ~92% Interpretation: If monitoring resources are focused on the top 10% highest-risk bank-months, approximately 92% of future events are captured. This indicates a strong clustering of regulatory risks.

Limitations

The model captures short-term enforcement momentum rather than structural bank risk. Observations are monthly and may exhibit temporal dependence. Macro-economic and regulatory regime variables are not yet included. Model is intended for prioritization, not regulatory judgment.

Why this project matters

Demonstrates: Time-series feature engineering Forward-label construction Leakage Control Lift-based evaluation Modular Python analytics pipeline

Application to operational risk, compliance analytics, and regulatory intelligence functions.