About this program
Risk modeling is where most automated investment systems either earn their keep or quietly fail.
Why traditional risk metrics fall short
Standard deviation and beta are useful but assume relationships that markets regularly break. Spend the opening weeks documenting exactly where these assumptions fail — during credit events, liquidity crises, and correlation breakdowns — and why that matters for automated systems running without human oversight.
ML approaches to factor risk
Cover dimensionality reduction techniques like PCA and sparse autoencoders to identify latent risk factors in a portfolio. Compare these against Fama-French factor models and measure which approach gives earlier warning signals during drawdown periods in historical data.
Tail risk and anomaly detection
Isolation forests, one-class SVMs, and LSTM-based anomaly detectors each behave differently on financial time series. Work through the tradeoffs: false positive rates, detection lag, and computational cost. Building something that fires alerts too often is almost as bad as one that fires too rarely.
Connecting risk models to automated responses
- Position reduction triggers based on predicted volatility spikes
- Correlation monitoring between holdings
- Automated hedging logic using options proxies
- Circuit breakers for model confidence thresholds
On realistic expectations
ML risk models are tools for earlier detection and better-informed decisions, not guarantees against losses. Markets produce genuinely novel conditions that no training set fully anticipates.