Fear & Greed Index forecast — composite ML ensemble
Eight selectable forecasting models — mean-reversion baselines (Ornstein-Uhlenbeck, AR(2), regime-switching, macro-conditioned OU), a Schrödinger path-integral model, and three machine-learning drift models (ML, quantum+ML, macro-GBM) — each fit on the index's own sub-indicators (VIX, breadth, momentum, put/call ratio) and compared on a single walk-forward leaderboard rather than asserting one model wins on every metric.
Current Fear & Greed score: 34 (fear). Top-ranked model on the walk-forward leaderboard: macro_gbm (RMSE 18.33, 82.9% CI90 coverage, 63.6% directional hit-rate).
Walk-forward model leaderboard
Every selectable model, ranked by closeness to 90% CI coverage plus RMSE on the same rolling backtest — the committed artifact behind the interactive leaderboard table.
- macro_gbm: RMSE 18.33, 82.9% CI90 coverage, 63.6% directional hit-rate (990 walk-forward predictions).
- ml: RMSE 17.42, 78.6% CI90 coverage, 72.7% directional hit-rate (990 walk-forward predictions).
- macro_ou: RMSE 18.95, 74.0% CI90 coverage, 75.8% directional hit-rate (990 walk-forward predictions).
- ou: RMSE 19.15, 73.9% CI90 coverage, 75.8% directional hit-rate (990 walk-forward predictions).
- ar2: RMSE 19.39, 73.6% CI90 coverage, 75.8% directional hit-rate (990 walk-forward predictions).
- quantum_ml: RMSE 18.23, 67.7% CI90 coverage, 63.6% directional hit-rate (990 walk-forward predictions).
- quantum: RMSE 20.33, 66.1% CI90 coverage, 66.7% directional hit-rate (990 walk-forward predictions).
- regime: RMSE 18.91, 61.4% CI90 coverage, 69.7% directional hit-rate (990 walk-forward predictions).
Frequently asked questions
- How is the Fear & Greed Index forecast built?
- A composite ensemble: an Ornstein-Uhlenbeck mean-reversion baseline combined with a Random Forest drift model trained on the index's own sub-indicators (VIX, breadth, momentum, put/call ratio, etc.). Backtested walk-forward against simpler single models on a committed leaderboard artifact. Educational research; not investment advice.
- Is the composite model always the most accurate?
- Not universally — the committed walk-forward leaderboard shows different models trade off RMSE, 90% coverage, and directional hit-rate differently; we publish the comparison rather than asserting one model is best on every metric.
- How many forecast models can I compare on this page?
- Eight: mean-reversion baselines (Ornstein-Uhlenbeck, AR(2), regime-switching, macro-conditioned OU), a Schrödinger path-integral quantum model, and three machine-learning drift models (ML, quantum+ML, macro-GBM) — every one backtested walk-forward and ranked on the same leaderboard.
Educational research only — not investment advice.