The ML chip keeps the OU mean-reversion backbone but replaces its constant drift with a Random Forest prediction — a standard machine-learning method that averages many decision trees. Here it is trained on the Fear & Greed index's own sub-indicators: VIX, S&P 500 momentum, market breadth, the put/call ratio, junk-bond demand and safe-haven flows.
The headline Fear & Greed number is an average, and averages lag. Often one or two of the underlying components — a VIX spike, breadth rolling over — move before the composite reflects it. The ML drift term tries to read those early movers and nudge the forecast accordingly, while the OU backbone still pulls everything back toward a sensible long-run level. Which components are carrying the current forecast is surfaced in the calibration line under the chart.
A learned drift can find signal, but it can also learn noise from a limited history — more flexibility, more ways to be fooled. So don't take the ML label as a badge of quality: open the "Model performance" table under the Fear & Greed forecast, which shows the ML model's coverage, MAPE, RMSE, Brier and hit-rate beside the plain OU baseline so you can see whether the learned drift actually helps on current data.
A standard machine-learning method that averages the predictions of many decision trees. Here it learns a drift for the Fear & Greed forecast from the index's own sub-indicators instead of assuming a constant one.
The headline index is an average, and averages lag. Individual components — a VIX spike, breadth rolling over — often move before the composite does, so the ML drift tries to read those early movers.
No. A learned drift can find signal but can also learn noise from limited history. Check the 'Model performance' table under the chart to see whether the ML model actually beats the plain OU baseline on current data.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.