Risk-Based Quantitative Modeling & ML Risk Forecasting

This project develops an end-to-end risk modeling framework that integrates traditional portfolio risk metrics, statistical backtesting, and machine learning–based downside risk forecasting.

Objectives

• Measure portfolio risk using volatility, drawdowns, VaR, and CVaR
• Backtest Value-at-Risk using the Kupiec Unconditional Coverage Test
• Forecast high-risk market days using machine learning
• Apply time-series–aware validation to avoid look-ahead bias

Data

• Daily ETF data (GLD, QQQ, SPY, TLT)
• Source: Stooq (free market data)

Methodology

1. Portfolio construction (equal-weight)
2. Risk metrics: volatility, drawdown, VaR, CVaR
3. VaR backtesting using Kupiec test
4. Feature engineering for risk prediction
5. ML classification (Logistic Regression)
6. TimeSeries cross-validation

Tools & Technologies

• Python, Pandas, NumPy
• Matplotlib
• SciPy
• Scikit-learn

Outputs

• Equity curve and drawdown plots
• Rolling volatility & VaR diagnostics
• ML-based probability forecasts for high-risk days

How to Run

pip install -r requirements.txt