Model leaderboard
forecast skill vs. no-change baseline| # | Model | n | Skill | RMSE | MAE | MAPE | Bias | Dir. |
|---|---|---|---|---|---|---|---|---|
| No graded predictions yet | ||||||||
Skill score = 1 − RMSE ÷ baseline RMSE, where the baseline is the naive no-change forecast (price stays at its reference value). A positive skill score means the model genuinely beats persistence; 0 ties it; negative is worse. Click any column to re-rank.
Forecast quality
interval calibrationTrade profitability
| Model | Pairs | Cover | Dir | Sharpe | t | Hit | PF | Gated@τ |
|---|---|---|---|---|---|---|---|---|
| No graded min/max bands yet | ||||||||
Each min–max band is read as a 1-hour trade: take a position in the direction the band midpoint points (sized by conviction z = predicted move ÷ band half-width, clipped to ±1), hold to the target time, and close at the graded actual price. Sharpe = mean ÷ stdev of those per-trade returns — how consistently, not just how much, the band would have paid. A positive, significant (t ≥ 2) Sharpe means you can rely on the signal; negative means trading it loses (and fading it would profit). Raise τ to keep only high-conviction bands — it re-scores live, no resubmission needed.
Hold-the-trend (carry)
| Model | In mkt | Trades | Hold | Gross | Net | Net ret | Churn ret |
|---|---|---|---|---|---|---|---|
| No graded min/max bands yet | |||||||
The carry strategy holds a directional position across windows instead of re-trading every hour: it opens when conviction clears enter, keeps holding while |z| stays above exit, and closes (or flips) only when it crosses back. Gross returns are identical to churning the same positions hourly — the only difference is transaction cost, charged per unit of position change. So Net vs Churn ret shows exactly how much the hold-the-trend rule saves at your assumed cost. All three knobs re-score live.