TP/SL odds: track record
On BTC, ETH and SOL perps, CLOBr gives the odds that price hits your take profit first, your stop first, or neither within 1h, 4h or 24h. This page shows how those odds scored, month by month, against a random walk.
How our odds work · Odds board
Every month is predicted by a model fit only on the 12 months before it. Decisions every 15 minutes, levels at 1 and 2 standard deviations, July 2023 to now. The random walk is what the odds would be if price moved randomly with the same volatility.
| Horizon | All three outcomes vs random walk | Months better than random walk | Which level first vs random walk | Decisions scored |
|---|---|---|---|---|
| 1h | +11.7% better | 40 of 40 | +1.9% better | 1,396,164 |
| 4h | +9.0% better | 40 of 40 | +2.2% better | 1,396,028 |
| 24h | +8.0% better | 39 of 40 | +1.3% better | 1,395,084 |
Better = lower Brier score than the random walk, in percent. Pooled over every month since July 2023.
All three outcomes: TP first, SL first, neither
Monthly Brier score, lower is better.
Which level first, when one is hit
Monthly Brier score, lower is better. The gain here is small.
What the odds are good at, and what they are not
- Most of the gain is knowing how often neither level gets hit. A plain random walk overestimates how far price travels.
- The gain on which level comes first is small, and at 24h it is not reliable month to month.
- Inputs are volatility, momentum, funding, time of day and how active the coming hours usually are (weekends and late Asian hours are quieter). The model also allows for error in its own volatility estimate, so big moves and tight ranges both get realistic odds. Liquidation heatmaps and order book depth did not add accuracy in our 3-year backtest, so they are not inputs.
- Odds are probabilities. A 55% chance still fails 45% of the time.
Model 1.1.0+fit20261020, trained through 2026-10-20 on Binance USD-M 1-minute candles. Refit monthly.