Quant Model Validation & Numerical Risk Simulation Framework

An offline, end-to-end independent quant model validation framework for derivatives, interest-rate models, Monte Carlo risk simulation, VaR/Expected Shortfall backtesting, numerical reliability, and model-risk governance artifacts.

This repository was designed as a non-production portfolio project for market-risk/model-validation roles. It uses synthetic portfolios by default, optional public-data sample files shaped after Federal Reserve H.15 Treasury-rate and FRED ICE BofA OAS-style data, and conservative claim boundaries.

What this project demonstrates

• Model inventory and validation scope definition

• Black-Scholes, binomial tree, Monte Carlo option pricing, Heston simulation, Vasicek and CIR rate paths

• Independent benchmark comparison, no-arbitrage checks, Greeks validation, finite-difference tolerance tests

• Full-revaluation vs delta-gamma P&L, VaR, Expected Shortfall, stress loss

• 250-day rolling VaR backtesting with Kupiec and Christoffersen tests

• Public-data sample validation for Federal Reserve H.15-style Treasury rates and FRED ICE BofA OAS-style spread factors

• Subportfolio backtesting proxies for derivatives, rates, spread-risk, and aggregate portfolio P&L

• Regulatory-style mapping memo connecting VaR/ES/backtesting/assumption testing artifacts to market-risk model validation concepts

• Stressed-window VaR proxy using maximum rolling historical VaR window selection

• P&L attribution proxy comparing actual vs explained P&L and residual-risk budget

• APL/HPL backtesting proxy comparing actual P&L proxy against hypothetical/explained P&L proxy

• ES tail diagnostics comparing modeled ES to realized average tail losses

• FRTB-inspired liquidity-horizon ES proxy across derivatives, rates, spread-risk, and aggregate P&L proxies

• Stress scenario coverage and monotonicity checks for equity, volatility, and rate shocks

• Artifact manifest and run manifest for reproducibility and GitHub review

• Dockerfile for containerized local validation

• FRTB-inspired risk-factor modellability/data-sufficiency proxy for public-data samples

• Baseline-vs-final validation summary showing one-sided false-pass reduction from the specialty reliability layer

• Numerical reliability: Monte Carlo convergence, seed stability, tolerance gates, optional CPU/GPU consistency scaffold

• Specialty reliability layer: Validation Health Score, closed-form NumPy ridge residual error-correction challenger, asymmetric validation envelope

• Governance artifacts: validation memo, model card, known limitations, compensating controls, model-risk findings, release gate

Claim boundary

This is an offline/non-production validation framework using synthetic portfolios and public-data samples. It is not a bank trading-desk model, not a regulatory capital model, not FRTB-compliant, and does not use proprietary bank data.

Run

Recommended Python: 3.11. The GitHub Actions workflow uses Python 3.11 for reproducible CI validation.

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
python -m pip install -r requirements.txt
PYTHONDONTWRITEBYTECODE=1 python scripts/run_pipeline.py
PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 python -m pytest -q

Or:

make validate

Main outputs

After python scripts/run_pipeline.py, generated artifacts are saved under outputs/ and reports/.

Key files:

• outputs/model_inventory.csv
• outputs/benchmark_comparison.csv
• outputs/greeks_validation.csv
• outputs/convergence_summary.csv
• outputs/var_es_results.csv
• outputs/stress_loss_results.csv
• outputs/backtesting_exceptions.csv
• outputs/kupiec_christoffersen_results.csv
• outputs/assumption_sensitivity_matrix.csv
• outputs/validation_health_scores.csv
• outputs/model_risk_findings.csv
• outputs/baseline_vs_final_validation_summary.csv
• outputs/public_market_data_validation.csv
• outputs/subportfolio_backtesting_summary.csv
• outputs/regulatory_mapping_table.csv
• outputs/stressed_var_results.csv
• outputs/pnl_attribution_proxy_summary.csv
• outputs/apl_hpl_backtesting_comparison.csv
• outputs/es_tail_diagnostics.csv
• outputs/liquidity_horizon_es_proxy.csv
• outputs/stress_scenario_coverage.csv
• outputs/artifact_manifest.csv
• outputs/run_manifest.json
• outputs/risk_factor_modellability_proxy.csv
• reports/github_runnable_audit.md
• outputs/release_decision.json
• reports/public_market_data_validation_report.md
• reports/regulatory_mapping_memo.md
• reports/resume_metrics_report.md
• reports/model_risk_validation_memo.md
• reports/executive_summary.md

Optional dashboard

The dashboard dependencies are kept separate from the core validation dependencies so GitHub Actions and Docker validation stay lightweight.

pip install -r requirements-dashboard.txt
streamlit run src/dashboard/streamlit_app.py

Docker run

docker build -t qmv-framework .
docker run --rm qmv-framework

Resume-safe framing

Built an independent Python quant model validation framework for derivative pricing, interest-rate models, and market-risk simulation, covering Black-Scholes, binomial tree, Monte Carlo, Heston, Vasicek/CIR rates, no-arbitrage checks, Greeks validation, full-revaluation vs delta-gamma P&L, VaR/Expected Shortfall, stress diagnostics, and Kupiec/Christoffersen backtesting.

Added numerical-reliability and model-risk controls including public-data sample validation, subportfolio VaR backtesting, Monte Carlo convergence diagnostics, finite-difference tolerance gates, seed-stability tests, asymmetric validation envelopes, residual error-correction challengers, regulatory-style mapping, finding severity, validation memos, model cards, known limitations, and conservative synthetic/public-data claim boundaries.