41% annual return. Sharpe 2.51. Predicting crashes before they happen.

A new paper from Oxford introduces a hybrid ML ensemble that forecasts short-horizon market risk — and trades it.

• Combines neural nets + tree ensembles to predict 5-day SPY drawdowns (> 1%)

• Uses cross-asset features from equities, bonds, FX, commodities, and volatility

• Finds that oil, FX, and Treasury signals lead equities — warning of crashes before they hit

• A simple long/short SPY strategy earns Sharpe 2.51 with beta 0.51 over 2005–2025

The insight: systematic alpha emerges from modeling risk itself — not returns — through interpretable, causal ensembles.