| Model | Predictions | Accuracy | Avg Error % | Total PnL | Win Rate | Sharpe | Max drawdown | Always-up | Persistence | Buy & hold PnL | vs best baseline |
|---|---|---|---|---|---|---|---|---|---|---|---|
| linear 61 versions, trained on data up to 2026-07-31 to 2026-09-29 Replay 🏆 Best Return | 61 | 42.6% | 1.4% | $16,372.85 | 26.2% | 3.93 | -5.23% | 54.1% n=61 | 54.1% n=61 | $20,735.72 | ✗ Does not beat baseline -11.5 pts vs always-up, p=0.973, n=61 |
Baselines are scored on the same days as each model. Always-up predicts a rise every day; persistence repeats yesterday's direction. A model beats the baseline only when its accuracy is higher and the gap is statistically significant (one-sided binomial test, p < 0.05). Baselines are available for the daily timeframe only.