Crypto forecast

Expected range and market regime — the part of crypto that is actually predictable. No price targets: we measured direction, and it is a coin flip.

Recomputed: 16.10.2026 00:00 MSK

BTC 82086.20 USDT
next 24 hours usual
±2.3%
80208 … 83964 USDT
80% of past moves stayed inside a band built this way
next 7 days usual
±5.3%
77770 … 86403 USDT
80% of past moves stayed inside a band built this way
ETH 2453.35 USDT
next 24 hours usual
±3.0%
2380 … 2527 USDT
80% of past moves stayed inside a band built this way
next 7 days usual
±7.8%
2261 … 2645 USDT
80% of past moves stayed inside a band built this way
SOL 105.59 USDT
next 24 hours usual
±4.1%
101 … 110 USDT
80% of past moves stayed inside a band built this way
next 7 days usual
±11.2%
94 … 117 USDT
80% of past moves stayed inside a band built this way

Honesty scoreboard

Direction: not predictable. Across every horizon we test, guessing which way the price goes lands on a coin flip, and our own model does not beat “tomorrow equals today”. That is why there is no price target anywhere on this page.

Coin / horizon Direction hit rate vs “no change” Volatility vs baseline Test points
BTC · 24h 0.485 1.024 +0.158 4494
BTC · 168h 0.445 1.118 +0.220 4278
ETH · 24h 0.529 1.013 +0.156 4494
ETH · 168h 0.464 1.140 +0.198 4278
SOL · 24h 0.473 1.034 +0.076 4494
SOL · 168h 0.474 1.135 +0.068 4278

These are backtest numbers: the method replayed over its own history, training and testing kept apart by a gap as long as the horizon. Hit rate 0.50 means a coin flip; “vs no change” above 1.000 means worse than not forecasting at all; volatility above zero means better than both baselines. How this is computed →

Live track record

Horizon Checked Landed in band Share
24h 513 409 0.80
168h 441 336 0.76

Every forecast on this page is stored when published and compared with what actually happened once its horizon closes. Below 30 checks the share is shown but means little — it is marked as accumulating.

Why there is no price target

Because we checked, and it does not work. Take a year of hourly prices, build the usual features — momentum, position in the range, realised volatility — train a model, and test it on data it has never seen, with a gap between training and testing so the two cannot overlap. The result is a hit rate of about 0.50 and an error slightly worse than assuming nothing changes at all.

Volatility is a different story. How much a coin moves is genuinely predictable — calm weeks follow calm weeks, storms cluster — and that survives the same test. So this section forecasts how far, not which way. The full method, baselines and caveats →