# The weekend futures gap reversal does not show up in Bitcoin or Ether

> Trading against Bitcoin and Ether weekend moves of 2% or more lost 15.9 bp per trade over 24 hours; the move did not reverse, and the slope has the wrong sign and is not significant (t +0.59).

E420 · Verdict: KILL · Cause: No signal · Market: Crypto & DeFi · Mechanism: Calendar · Horizon: 1d · Tested: October 2026

Source: https://alphanotfound.eu/experiments/E420-weekend-gap-fill-in-crypto

Crypto traders say the price "fills the futures gap": a weekend move should partly reverse when the regulated futures market reopens on Sunday evening. We measured the move from the Friday close to the Sunday reopening in Bitcoin and Ether perpetuals, 708 weekend observations in all, and traded against moves of 2% or more for 24 hours. The 253 trades lost 15.9 bp on average (1 bp = 0.01%). The move did not reverse (slope +0.045, wrong sign, t +0.59). After the spot-ETF launch the trades earned +17.3 bp, but placebo windows shifted by a day earned +12.5 and +6.3 bp.

## Key numbers

- **Net per trade:** -15.9 bp (253 trades in 176 weekends with a move of 2% or more)
- **Interval of the weekend mean:** -83 to +59 bp (95% interval; includes zero)
- **Slope of next-day return on weekend move:** +0.045 (t +0.59; reversal needs a negative slope with t below -2)
- **Worst single trade:** -1,923 bp (a continuation of a large weekend move)

## The idea being tested

**Belief:** Weekend price moves in Bitcoin and Ether partly reverse when the regulated futures market reopens: the so-called futures gap gets filled.

**Why people believe it:** Regulated crypto futures and spot ETFs are shut from Friday evening to Sunday evening, so weekend moves happen without the institutions that trade through them. When the futures market reopens, and the ETFs trade on Monday, those participants re-hedge, and the weekend move should partly revert toward the Friday close.

**Why it would persist:** The closure is fixed by the calendar and institutional mandates, not by a pricing error, and the large players who would have to trade against the move are limited by risk limits and by funding costs of holding a perpetual position.

## How it was tested

For every weekend since 2020 we took the perpetual-futures price at the Friday close of the regulated futures market and at its Sunday reopening, and computed the weekend move W. One hour after the reopening we bet against that move (short after a rise, long after a fall) and held for 24 hours, paying 10 bp round-trip costs (1 bp is 0.01%) and the actual funding. The rule, thresholds and criteria were written down before the data was looked at. Two placebos used the same construction with all times shifted by one day, and the same rule was run on 22 other coins.

**Pre-registered:** yes

**Data:** Hourly perpetual-futures open prices for Bitcoin and Ether, plus 22 other coins as a secondary set, and historical funding rates (largest USD-margined crypto perpetual venue, public market-data archive; 1 hour; 2020-01..2026-10); Opening and closing times of the weekend closure of regulated crypto futures (exchange trading calendar, Friday 17:00 to Sunday 18:00 New York time; weekly; 2020-01..2026-10)

**Universe:** Bitcoin and Ether perpetuals, 354 weekends each, 708 observations (420 before the spot-ETF launch in 2024-01, 288 after). Secondary: 22 other perpetuals, same rule.

**Signal:** W = P(Sunday 18:00 New York) / P(Friday 17:00 New York) - 1, using the hourly open of the perpetual. Primary set: weekends with |W| of at least 2%. Slope: next-day return R (hour +1 to hour +25 after the reopening) regressed on W over all weekends, with errors clustered by weekend.

**Execution:** One hour after the Sunday reopening (the first full hour, no first-trade pricing) take the opposite side of the sign of W; hold 24 hours. Simple returns.

**Costs:** 10 bp round trip, plus funding paid or received from the historical funding rates over the holding window.

**Controls:** Placebos: the identical construction with the anchors and window shifted 24 hours earlier and 24 hours later. Sample split at the spot-ETF launch (2024-01). Secondary set of 22 other perpetuals.

**Sample split:** Before the spot-ETF launch (2020-01 to 2024-01) versus after (2024-01 to 2026-10); rule fixed before any data was examined.

**Pass criterion:** The inverse of the pre-registered rejection criteria: net mean above 0 in the primary set, slope t of -2 or below over all weekends, correct slope sign after the ETF launch, and placebos weaker than the main result.

**Kill criterion:** Primary net mean at or below 0, or slope t above -2 over all weekends, or wrong slope sign after the ETF launch, or a placebo as strong as the main result.

**Power:** Minimum detectable effect about 33 bp per trade at the 24-hour horizon.

## Result

**Why it was rejected:** Net return is negative, the slope has the wrong sign and is nowhere near significant, and placebo windows earn as much as the best subsample.

**Observations:** 253

**Net mean:** -15.9 bp

**Median:** +8.8 bp

**Worst:** -1,923 bp

- Across all 708 weekend observations the slope of next-day return on weekend move is +0.045 (t +0.59): the sign is positive (wrong for a reversal) and not significant, so the weekend move did not reverse.
- With |W| of at least 2%, 253 trades (176 weekends) lost 15.9 bp on average; the median trade made +8.8 bp and the worst lost 1,923 bp. The weekend-level mean is -12.8 bp with a 95% interval of -83 to +59 bp.
- Before the spot-ETF launch the 174 trades lost 31.0 bp (weekend mean -39.3 bp, interval -140 to +60 bp); after it the 79 trades earned +17.3 bp (weekend mean +38.5 bp, interval -54 to +131 bp, slope -0.10, t -0.65).
- The two placebo windows, with all times shifted by one day, earned +12.5 and +6.3 bp, as strong as the post-ETF result; the 22 other perpetuals earned +7.9 bp (slope t -0.73).
- By size of weekend move (cut points chosen after the test, descriptive only, not part of the pre-registered test): -2.7 bp for moves of 2% to 3%, +32.4 bp for 3% to 5% and -71.2 bp at 5% and above. These bins were picked after the test, so the +32.4 bp is not evidence.

### No gap fill in any period (bp)

Mean net return per weekend of trading against the weekend move, with 95% intervals clustered by weekend. All three intervals include zero; the pre-ETF period is negative, the post-ETF period positive but with a wide interval.

|   | Mean | 95% low | 95% high |
| --- | --- | --- | --- |
| All weekends (n 176) | -12.8 | -83 | 59 |
| Before spot ETFs (n 116) | -39.3 | -140 | 60 |
| After spot ETFs (n 60) | 38.5 | -54 | 131 |

### Placebos earn as much as the rule (bp)

Mean net return per trade for the main rule, its subsamples, the one-day-shifted placebos and the other 22 perpetuals. The post-ETF cell, the only positive primary result, is no stronger than the placebos.

|   | Mean net per trade |
| --- | --- |
| Main rule, all (n 253) | -15.9 |
| Before spot ETFs (n 174) | -31.0 |
| After spot ETFs (n 79) | 17.3 |
| Placebo, shifted -24h | 12.5 |
| Placebo, shifted +24h | 6.3 |
| 22 other perpetuals | 7.9 |

### Move size does not line up with reversal (bp)

Mean net return per trade by size of the weekend move. Cut points chosen after the test, descriptive only, not part of the pre-registered test. The middle bucket is positive, the other two negative; no trend was tested.

|   | Mean net per trade |
| --- | --- |
| 2% to 3% (n 77) | -2.7 |
| 3% to 5% (n 84) | 32.4 |
| 5% and above (n 92) | -71.2 |

## Interpretation

**Why it failed (interpretation, not established):** Crypto trades around the clock on its own venues, so the weekend price is formed by the same market that sets it on Monday; the regulated futures price is a follower, not an anchor the perpetual has to snap back to. A gap on a futures chart then measures only that the futures market was closed, and the test found no detectable relation between the weekend move and the next day's return (slope t +0.59; minimum detectable effect about 33 bp). That is an explanation consistent with the numbers, not something the test isolated.

**Lesson:** A named chart pattern ("gap fill") is a claim about a causal link between a market's closing hours and the next move. The test has to show that link in the data: here a subsample (+17.3 bp after the ETF launch) looked like support until placebo windows one day away earned +12.5 and +6.3 bp. A subsample result means little until its placebos have been run on the same construction.

**Scope:** Covers Bitcoin and Ether perpetuals, hourly anchors, moves of 2% or more, a 24-hour hold and weekends from 2020-01 to 2026-10, plus 22 other perpetuals under the same rule. Regulated crypto futures have traded around the clock since 2026-05, so the weekend closure that the idea rests on existed only before that month.

## Reproducibility

**Steps:** 1) Take hourly perpetual open prices for Bitcoin and Ether and compute, for each weekend, W from the Friday 17:00 to the Sunday 18:00 New York open. 2) At the Sunday 18:00 open plus one hour, trade against sign(W) and exit 24 hours later; subtract 10 bp round trip and the summed funding in the window. 3) Keep weekends with |W| of at least 2%; report the mean, the median and a weekend-clustered bootstrap interval. 4) Regress the return of the same window on W over all weekends with errors clustered by weekend. 5) Repeat with times shifted by -24 and +24 hours, before and after the ETF launch in 2024-01, and on the other perpetuals.

**Grade:** COMPLETE

---
AI-generated research, checked but not peer-reviewed. Historical, experimental research, not investment advice. Dataset Terms of Use: https://alphanotfound.eu/terms