How the Bitcoin forecast works
Where the data comes from, what the model does, how the range is built and how it is checked. Including what it does not do.
Current Bitcoin forecast
As of 29 Oct 2026, 11:00 UTC · Binance BTC/USDT · closed candles
| Horizon | Model estimate | Change | Lower | Upper | Reference close | Target (UTC) |
|---|---|---|---|---|---|---|
| 24 hours | 80,116 | −0.21% | 78,654 | 81,333 | 80,285 | 30 Oct 2026, 11:00 UTC |
| 7 days | 79,638 | −0.28% | 77,283 | 87,527 | 79,862 | 5 Nov 2026, 08:00 UTC |
| 30 days | 82,415 | +1.32% | 73,387 | 99,070 | 81,344 | 28 Nov 2026, 00:00 UTC |
Prices in USDT. Experimental estimates, not guaranteed outcomes; the actual price may fall outside the range.
Data
Forecasts use Binance spot BTC/USDT candles from Binance's public market-data endpoint (documentation). No API key or private data is involved. Each request reads 1,000 candles per horizon:
| Horizon | Candle | Steps ahead | Look-back windows | Updates |
|---|---|---|---|---|
| 24 hours | 1 hour | 24 | 12, 48, 144 candles | after each closed hour |
| 7 days | 4 hours | 42 | 12, 42, 126 candles | after each closed 4-hour candle |
| 30 days | 1 day | 30 | 14, 50, 150 candles | after each closed UTC day |
Only closed candles are used. Data with gaps, duplicates, out-of-order timestamps, too little history or a stale last candle is rejected, and that horizon is shown as unavailable. There is no synthetic or random fallback. Prices are in USDT, which is not the same as USD or a Chainlink USD settlement price.
Model
The model is a fixed, deterministic heuristic, not a machine-learning model trained on outcomes. For each horizon it combines:
- linear-regression trends of log prices across the three look-back windows;
- R² of each fit as a measure of trend consistency;
- recent momentum;
- realised volatility, which scales the projected move.
Volume is displayed for context only and is not a fitted factor. The model does not use news, order books, funding rates, Polymarket odds or social sentiment. It has not established a predictive edge and is not calibrated to probabilities.
Scenario range
The lower and upper scenarios use the 10th and 90th percentiles of the model's historical log-return errors for that horizon, expanded if needed to contain the point estimate. This is a descriptive range of past misses, not an empirically proven 80% interval. It applies to the end price only, not the path, high or low within the period. Because each horizon uses different data, a longer horizon can occasionally show a narrower range.
Walk-forward checks
To see how the model would have done, it is re-run at up to 24 past origins using only data available at each origin (a walk-forward evaluation). Target windows do not overlap. The page reports the number of checks and the mean absolute percentage error next to an unchanged-price baseline. Current figures:
| Horizon | Checks | Model error | Unchanged-price error |
|---|---|---|---|
| 24 hours | 24 | 1.02% | 0.96% |
| 7 days | 19 | 3.21% | 3.27% |
| 30 days | 24 | 10.27% | 10.19% |
These are retrospective checks. Forecasts are not yet archived as an immutable live record, so they should not be read as a track record or a promise of future accuracy.
What the forecast is not
- Not financial advice, and not an instruction to trade.
- Not a prediction of the intraperiod high or low.
- Not the same thing as PolySignal's short-term Polymarket signals, which use a separate engine. See Polymarket Bitcoin Up or Down.