Graveyard / Listings and unlocks / Binance perps
Shorting a coin when Binance launches its perpetual
Does a coin fall after Binance announces a futures contract for it?
The idea
Before a perpetual future exists, bearish traders cannot easily bet against a coin. When Binance launches one, those traders can short, and the coin may fall. Two ways to play it: short the new perpetual right after launch, or sell the spot coin from the announcement.
How we tested it
| Data | Binance announcements of new USDT-margined perpetuals for coins already on spot: 127 announcement clusters (138 coin events). Binance one-minute prices and funding. Arm B: short the perp one minute after first trade, hold 1 or 5 days. Arm A: sell or avoid the spot coin from the notice to the launch. |
|---|---|
| Period | November 2021 to November 2026. Development: first 60% (82 events, to 27 November 2023). Holdout: last 40% (56 events). |
| Costs | 10 bp round trip plus actual funding for arm B. 20 bp round trip for arm A. |
| Holdout | Dev and holdout shown separately, with correction for several variants. |
| Control | Three random coins per event, same clocks and costs. |
Result
| Launches (clusters) | 127 |
|---|---|
| Arm B, 5-day net | −1.40% |
| Arm B, 5-day excess vs control | −0.33% (t −0.13) |
| Arm B, development / holdout excess | −0.23% / −0.47% |
| Arm A, 5-day excess | +1.41% (t 0.59) |
| Arm A, 5-day net | +0.05% |
| Arm A, development / holdout excess | +1.24% / +1.67% |
| Arm A win rate | 59% to 62% |
Why it failed
Shorting after launch has no edge and loses in both halves. Selling from the announcement shows a +1.41% excess, but it is not distinguishable from chance, nets only +0.05%, the year-by-year sign flips and the win rate is high but the mean is about zero because of fat tails. The notice also comes only hours before the listing on average, so the short-sale relief story does not show up.
What would make it worth retesting
For arm A: margin-short history showing it was executable and at least 300 events. For arm B: nothing. Only about 25 new events arrive each year.
Past results of a backtest say nothing certain about the future.