Banking Operational Risk Intelligence Platform

An end-to-end operational risk analytics system built with real FDIC regulatory data, ML risk scoring, KRI/KPI monitoring, interest rate analysis, and an interactive Streamlit dashboard - replicating workflows used in financial institution compliance and risk management.

---

Business Problem

Financial institutions and their regulators need to continuously monitor operational risk across thousands of banks. Manual processes are slow, inconsistent, and fail to catch early warning signs before they become systemic failures. This platform automates:

• Control exception detection - automated KRI breach flagging across Basel III thresholds
• Risk-based sampling - statistical sampling of high-risk institutions for control testing
• Predictive risk scoring - ML models identify at-risk institutions before failure
• KRI/KPI reporting - standardized reporting aligned with CCAR and regulatory frameworks
• Macro risk overlay - Federal Reserve interest rate analysis and yield curve monitoring

---

Data Sources

Source	Description	Access
FDIC BankFind API	Financial data for 4,500+ FDIC-insured institutions	Free, no API key required
FDIC Failures Endpoint	US bank failure history since 1934 with insurance cost	Free, no API key required
FRED (St. Louis Fed)	Fed Funds Rate, 10Y/2Y Treasury yields	Free via pandas-datareader; embedded fallback included

Note: Data availability depends on the respective agency's API uptime and rate limits. The FRED pipeline includes embedded fallback data if the API is unavailable.

---

Architecture

banking-risk-platform/
├── src/
│   ├── pipeline/
│   │   ├── fdic_pipeline.py      # FDIC API ingestion, KRI engineering, SQLite load
│   │   └── fred_pipeline.py      # FRED interest rate fetch, yield curve metrics
│   ├── ml/
│   │   └── risk_engine.py        # Logistic Regression, Random Forest, Gradient Boosting, K-Means
│   └── dashboard/
│       └── app.py                # Streamlit executive dashboard
├── sql/
│   └── schema_and_queries.sql    # SQLite schema + 6 analytical KRI queries
├── data/
│   ├── banking_risk.db
│   ├── institutions.csv
│   ├── failures.csv
│   ├── risk_scores.csv
│   ├── interest_rates.csv
│   └── models/
│       ├── model_metrics.json
│       ├── feature_importance.csv
│       └── cluster_centers.csv
└── run.py                        # Master pipeline runner

Quick Start

1. Install dependencies

pip install -r requirements.txt

2. Run the full pipeline

python run.py

This will:

• Pull 4,500+ banks and all historical failures from FDIC API
• Fetch Federal Reserve interest rate data from FRED
• Train 3 ML models and K-Means clustering
• Launch the Streamlit dashboard at http://localhost:8501

3. Run step by step

# Phase 1: FDIC data pipeline
python src/pipeline/fdic_pipeline.py

# Phase 2: Interest rate pipeline
python src/pipeline/fred_pipeline.py

# Phase 3: ML risk engine
python src/ml/risk_engine.py

# Phase 4: Dashboard
streamlit run src/dashboard/app.py

Note: If you encounter a 429 rate limit error from the FDIC API, the pipeline retries automatically with increasing wait times. If errors persist, wait 2-3 minutes before re-running.

---

Machine Learning

Supervised Models - Binary Classification: At-Risk vs Stable

Model	ROC-AUC	Accuracy	Recall	F1
Random Forest	0.84	71.8%	92.3%	0.463
Gradient Boosting	~0.82	~70%	~89%	~0.44
Logistic Regression	~0.78	~68%	~85%	~0.41

Label Construction: Banks are labeled at-risk based on historical FDIC failure records (1934-present), supplemented by a 30% random sample of currently HIGH-risk institutions. KRI threshold values are intentionally excluded from label construction to avoid circular learning - the same metrics used as features cannot also define the target variable.

Features:

• Return on Assets (ROA), Return on Equity (ROE)
• Capital Adequacy Ratio (Basel III)
• Non-Performing Loan Ratio (NPL)
• Loan-to-Deposit Ratio, Net Interest Margin, Cost-to-Income Ratio
• Fed Funds Rate, 10Y/2Y Treasury yields, Yield Curve Spread (FRED)

Unsupervised Model - K-Means Clustering

4 clusters discovered: Low Risk - Elevated Risk - High Risk - Critical Risk

---

KRI Framework - Basel III / CCAR Aligned

KRI	Warning Threshold	Critical Threshold	Regulatory Basis
Capital Adequacy Ratio	< 10%	< 8%	Basel III
Non-Performing Loan Ratio	> 5%	> 10%	FDIC guidance
Return on Assets	< 0.5%	< 0%	Peer benchmarks
Loan-to-Deposit Ratio	> 85%	> 90%	Liquidity risk
Cost-to-Income Ratio	> 60%	> 70%	Efficiency ratio

---

Dashboard

Section	Description
Executive Summary	KRI cards, risk tier distribution, capital histogram, failure timeline
Risk Heatmap	US choropleth with traffic-light color coding per KRI metric
Macro & Rates	Fed Funds Rate history, yield curve spread, rate environment analysis
ML Risk Scoring	Model comparison, feature importance, high-risk watch list, K-Means clusters
Control Testing	Exception counts, breach rates, historical failure analysis, risk-based sampling
Institution Lookup	Search any institution by name for full KRI and ML risk profile

---

SQL Queries

sql/schema_and_queries.sql includes 6 production-style KRI queries:

1. Control Exception Report - Basel III capital breach detection
2. Multi-Flag Watch List - Banks with 3+ simultaneous KRI breaches
3. State-Level KRI Summary - Regulatory dashboard view
4. Historical Failure Cost Analysis - Decade-by-decade failure trends
5. Risk-Based Sampling - Statistical sampling replicating CCAR examination methodology
6. YoY Trend Analysis - Institution-level ROA deterioration detection

---

Limitations

• Data reflects the most recent FDIC Call Report period available via API (Q3 2025)
• SQLite is used for portability and is not intended for production-scale concurrent workloads
• ML model performance is based on historical failure patterns and may not generalize to novel failure modes
• FRED data falls back to embedded quarterly averages if the API is unavailable

---