Archive / Crypto & DeFi / E390
Shorting large altcoins around token unlocks shows no detectable drift
Across 197 scheduled token unlocks, shorting the day before earned +52.0 bp net and the day after lost 29.1 bp; both intervals include zero, none of 4 cells survived correction, and unlock size did not matter.
Many traders expect a token to fall when a large batch of locked coins becomes sellable, and the dates are published in advance. We compiled 197 scheduled token unlocks of 10 large altcoins and shorted each coin's perpetual future for 24 hours, either the day before the unlock or the day after, paying 14 bp in fees plus actual funding. The day before earned +52.0 bp on average and the day after lost 29.1 bp. Both 95% intervals include zero, none of four tested cells survived multiple-testing correction, and the test found no detectable dose response.
When a large batch of locked tokens vests, the recipients sell, so the price falls around the unlock date and a short earns a premium.
Why people believe it
Team members, early investors and contributors receive their tokens on a fixed schedule, and the common belief is that many of them sell. The calendar is public, so anyone can see when the new supply arrives.
Why it would persist The argument, as stated before testing
Perpetual futures on large coins are liquid enough to short in size. Capacity is not the obstacle: at 2% of the coin's 24-hour perpetual volume, the median event allows about $1.6M.
We built a calendar of 290 release tranches for 10 large altcoins from official documents and on-chain contracts, of which 197 fell in March 2023 to October 2026 while the coin had a perpetual future. For each release we simulated two short trades: one held for the 24 hours before the release time, one opened 15 minutes after it and held 24 hours. Costs were 14 bp per round trip plus the funding actually paid. The same trade shifted by one week served as a control.
A cell survives Benjamini-Hochberg correction at q = 0.10 over the 4 cells, passes the power gate, and shows a dose response: monotone increase from low to high dose with a positive Spearman p below 0.10.
Both (a) and (b) have a confidence interval that includes zero; or no dose response in the surviving cell; or funding paid by the short exceeds the gross drift in the surviving cell.
7 bp per side, 14 bp per round trip (28 bp for the BTC-adjusted two-leg version), plus actual funding. Funding received averaged -0.7 bp per window in cell (a) and -0.1 bp in cell (b).
The same coin and clock window shifted by +7 days, reported as event minus control. Dose: release value (tokens times perp price at T-24h) divided by the 30-day average daily perp quote volume, tested by Spearman correlation with the trade result and by terciles. Timing sensitivity: unknown release times set to 12:00 UTC instead of 00:00 UTC.
Both pre-registered cells have intervals that include zero and no cell survives correction; there is also no dose response.
- Cell (a), short from 24 hours before the unlock to the unlock: +52.0 bp net per event (n 197, 95% interval -17.3 to +119.1, median +52.3, worst -2,590). Gross drift was +66.7 bp before the 14 bp cost and funding.
- Cell (b), short from T+15 minutes to T+24h15m: -29.1 bp net (95% interval -110.6 to +57.1, median +7.2, worst -3,018). Gross drift was -15.0 bp.
- Against the same windows shifted by a week, cell (a) is +75.1 bp better than control (interval -28.3 to +180.6); the control itself averaged -28.9 bp. For cell (b) the difference is +1.3 bp (interval -126.3 to +129.5).
- BTC-adjusted cells at 28 bp cost: +41.7 bp for (a) and -40.8 bp for (b); no cell passes Benjamini-Hochberg at 10%.
- No detectable dose response. Spearman correlation between dose and result is -0.002 (p 0.98) in cell (a) and -0.045 (p 0.53) in cell (b). By dose tercile, cell (a) earned +118.7, -64.5 and +100.1 bp from low to high, which is not monotone.
- The minimum detectable effect was about 64 bp, against a 14 bp round-trip cost, so a true drift smaller than roughly 64 bp could not be ruled out; the upper ends of the intervals reach +119.1 bp (cell a) and +180.6 bp (a minus control).
- Results flip sign between time windows: cell (a) earned +71.4, -81.9 and +125.4 bp in the train, validation and test windows; cell (b) earned -44.8, +113.8 and -122.0 bp.
- Setting unknown release times to 12:00 UTC cuts cell (a) from +52.0 to +10.9 bp (interval -56.9 to +78.1), so the pre-release drift depends on an assumption about the time of day.
- Per coin, cell (a) ranges from +207.8 bp (ENA, n 20) and +80.0 (SUI, n 41) to -79.5 (ARB, n 32, exact on-chain times) and -5.3 (APT, n 36). Events with official or on-chain schedules only (n 123): +56.6 bp, interval -46.3 to +156.6.
- On-chain check for one token: 31.7% of released tokens moved within 24 hours after a tranche boundary, against 3.3% if releases were spread evenly, with a median release about 101 hours after vesting.
Neither cell separates from zero
Net result per unlock for a 24-hour short, with 95% intervals clustered by coin and month. Both pre-registered cells include zero, the difference against a one-week-shifted control does too, and moving unknown release times to 12:00 UTC cuts the day-before result from +52.0 to +10.9 bp.
Chart data (bp)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| (a) Short T-24h to T | 52 | -17.3 | 119.1 |
| (a) minus control a week later | 75.1 | -28.3 | 180.6 |
| (a) with unknown times at 12:00 UTC | 10.9 | -56.9 | 78.1 |
| (b) Short T+15m to T+24h | -29.1 | -110.6 | 57.1 |
| (b) minus control a week later | 1.3 | -126.3 | 129.5 |
No detectable dose response
(a) Short T-24h to T(b) Short T+15m to T+24h
Mean net result by unlock size relative to the coin's 30-day average daily perp volume (dose terciles, low to high). A supply effect that grows with size would rise from left to right; both cells zigzag instead (Spearman -0.002 and -0.045).
Chart data (bp)
| (a) Short T-24h to T | (b) Short T+15m to T+24h | |
|---|---|---|
| Low dose | 118.7 | 36.5 |
| Middle dose | -64.5 | -74.3 |
| High dose | 100.1 | -50.2 |
No coherent pattern across coins
(a) Short T-24h to T(b) Short T+15m to T+24h
Mean net result per event by coin for the seven coins with at least 16 events, sorted by number of events. The sign of the day-before result changes between coins that have the same kind of monthly tranche. PYTH, TIA and W (5 events in total) are left out.
Chart data (bp)
| (a) Short T-24h to T | (b) Short T+15m to T+24h | |
|---|---|---|
| SUI (n 41) | 80 | -29.3 |
| APT (n 36) | -5.3 | 6.5 |
| ARB (n 32) | -79.5 | -51.7 |
| STRK (n 31) | 75.5 | 9.8 |
| ENA (n 20) | 207.8 | 102.6 |
| ZK (n 16) | -65 | 35.7 |
| ZRO (n 16) | 65.7 | -112.8 |
Why it failed Interpretation · not established
The likeliest explanation is smearing and pre-hedging: median monthly tranches were 0.16 to 0.9 days of perp volume per coin, and recipients of one token released over several days rather than at the boundary (median about 101 hours after vesting), so any pressure is spread out or hedged earlier. The test could not detect effects below about 64 bp against individual results as large as -2,590 and -3,018 bp. A small post hoc subset of the largest cliffs was also looked at; it is not evidence and is not used here.
Lesson
A known, dated supply event does not by itself give a trade. When the exact time of the event is uncertain, the result can depend on the assumed clock time, so that has to be tested before a pre-event drift is taken seriously. A dose-response check is cheap and hard to fake: an effect that does not grow with the size of the supply is weak evidence of a supply effect.
Scope of this result
Covers 197 scheduled vesting tranches of 10 large altcoins with perpetual futures, March 2023 to October 2026, 24-hour shorts at 7 bp per side plus funding. It does not cover small-cap tokens, windows other than 24 hours before or after, or recipient-level selling.
A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.
1. Test design
Pre-registeredDataUniverseSignalExecutionCostsControlsSample splitPass criterionKill criterionPower2. Result
ObservationsInstrumentsNet mean95% intervalMedianWorstFinding 1Finding 2Finding 3Finding 4Finding 5Finding 6Finding 7Finding 8Finding 9Finding 103. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade