This project develops an end-to-end risk modeling framework that combines traditional financial risk metrics, statistical backtesting, and machine learning–driven risk prediction. I built an equal-weighted portfolio using four ETFs (GLD, QQQ, SPY, TLT) and analyzed its behavior across market cycles. The project evaluates risk through volatility, drawdowns, Value-at-Risk (VaR), and Conditional VaR (CVaR). It then validates VaR estimates using the Kupiec Unconditional Coverage Test. To extend traditional risk modeling, I implemented a machine learning classifier that predicts high-risk days those with extreme downside returns. The model uses lagged returns, rolling volatility, momentum, and drawdown features. I evaluated all models using time-series cross-validation to avoid look-ahead bias and overfitting. The result is a production-ready risk research pipeline that bridges quantitative finance theory with applied risk management and machine learning.

Key Components:

• Portfolio construction and equity curve analysis
• Rolling volatility and drawdown diagnostics
• Historical VaR and CVaR estimation
• VaR backtesting using the Kupiec test
• ML-based downside risk probability forecasting
• Time-series–aware model validation

Tools & Technologies:

• Python, Pandas, NumPy
• Matplotlib for risk and performance visualization
• SciPy for statistical testing
• Scikit-learn for ML pipeline and time-series cross-validation
• Market data from Stooq (ETFs)

Links:

• Link