Movement odds — next 24 hours
How likely is ETH to move, and by how much — deliberately never which way. The engine is the options market’s own forecast with its measured calibration bias corrected, blended with a walk-forward statistical model. It was the only candidate that beat raw option-implied pricing out-of-sample, and its full scorecard is on this page.
ETH · spot $2,585
as of 2026-09-20 19:38 IST| Event | Engine | Options imply | Base rate |
|---|---|---|---|
| Move > 2% | 42% | 34% | 48% |
| Move > 4% | 11% | 6% | 20% |
| Expected |move| | ±1.7% | ±1.7% | ±1.9% |
The expected-move cone
amber = engine ±1.7% · dashed = options-implied ±1.7%What the engine sees
- ATM IV (0.9d tenor)
- 40.0% · 10th pctile
- Realized vol 24h / 5d
- 40.6% / 52.9%
- Moved in the last 24h
- ±2.1% vs ±1.7% expected
- Peak window
- not in effect on weekends
- Most likely window
- None — the weekday hour profile does not hold on weekends.
- Day-of-week
- Sunday daily — ETH sellers historically won only 34%; this window runs into the Monday reopen, when traditional-market flow returns.
No driver checklist here on purpose: ATR compression, funding, term-structure slope and skew were each tested and carry no measurable information about next-day movement. Listing them would be decoration.
Why the day of the week matters
Friday is the quietest day of the crypto week, and it is not close. Across 8.9 years and two venues, movement over the Friday-to-Saturday window ran 21% below average on BTC (lower in 9 of 10 years) and 24% below on ETH (10 of 10 years) — p < 0.00001 on both.
The cause is market hours, not crypto. We ran the same test on tokenized gold as a control, expecting it to show nothing. It showed the same pattern at the same size (−25% on Friday, −58% on Saturday) — and gold’s cash market is simply shut from Friday evening to Sunday evening. BTC and ETH trade 24/7, but the people who move them largely do not: when traditional markets close, crypto goes quiet too. That also explains why Sunday is the most active window of the week (+15% above average on both BTC and ETH) — it runs into the Monday Asia and Europe opens.
This matters for how much to trust it. An edge that comes from crowd positioning decays once enough people find it; an edge that comes from when the world’s exchanges are open is structural. Trading hours are not going to be arbitraged away.
We tested the same idea on market holidays — and published what it said
If trading hours drive this, a weekday when the NYSE is shut should behave like a mini-weekend. Across 97 closures since 2017, US-session movement fell 47% on BTC and 22% on ETH — and the size tracked how global the closure was: Christmas −76%, Good Friday −55%, but MLK Day (US-only, Europe open) +4%. The mechanism is real.
Over a full 24 hours, the effect vanished — BTC +1.4% (p = 0.86), ETH +0.4%. Crypto doesn’t lose the movement on a holiday, it displaces it into Asia and Europe hours. Gold, whose market is shut all day rather than for one session, stayed −38% at both horizons.
So there is no holiday premium-selling edge, and this page will not invent one: 1-day implied vol already marks holidays down 15–18%, which is about right given flat realized movement. The result is useful for timing — expect a dead US session — not for deciding what to sell.
Honest limit: this measures realized movement only. The seller win rates quoted above compare realized movement against what was priced, and the pricing half still rests on this venue’s options history alone — no public source publishes a long crypto-options tape. Saturday is the quietest day of all in absolute terms, yet Saturday selling lost money, because implied vol drops further than movement does. Quiet is not the same as cheap.
Does it tell the truth? (out-of-sample scorecard)
496 held-out days, walk-forward on ETH’s own tape. When the engine said “X%”, how often did ETH actually move more than 2%? A perfectly honest forecaster matches the observed column; the blend’s correction of implied pricing’s calibration bias is its entire edge (Brier 0.2389 vs raw implied 0.2399 (and 0.1546 vs 0.1569 on P(>4%))).
| Forecast bucket | Engine said | Actually happened | Days |
|---|---|---|---|
| 0.00-0.30 | 25% | 25% | 32 |
| 0.30-0.45 | 38% | 36% | 109 |
| 0.45-0.60 | 52% | 55% | 260 |
| 0.60-0.75 | 64% | 58% | 90 |
Educational analytics, not financial advice and not a prediction of direction. The engine forecasts movement magnitude only; its edge over option-implied pricing is real but thin, and it can be wrong on any given day. Models trained walk-forward per asset on Delta Exchange tapes (BTC through 2026-06-22, ETH through 2026-07-25). You place, size, and own every trade yourself.