BTC open-interest swings mark vol, not the next direction
BTC open-interest swings mark vol, not the next direction
Verdict: WATCH · Asset: BTCUSDT USDT-M OI vs spot 1h · Sample: chronological 60/20/20
Hypothesis: extreme open-interest changes predict the next 8h return (long-crowding fade when +ΔOI meets prior up-move), or at least mark higher forward |r|.
What happened
Feature A frozen as prior 8h relative ΔOI. Vision daily metrics (2020-09 → 2026-08, native 5m sum_open_interest → last-in-hour 1h), n=52 299.
| Gate | Validation |
|---|---|
| Q1 long crowding fade (+ΔOI & prior r>0 vs MID) | FAIL (t=−0.74, ΔP=+0.003) |
| Q1b extreme |ΔOI| → higher |fwd r| | CLEAR (t=+4.41, lift +31%) |
| Q3 | Not run (directional Q1 failed) |
Q2: OI is not collinear with funding/basis, but the |ΔOI| vol lift is largely shared with prior |r| — another vol-regime proxy, same family as volume and |r| clustering.
Trading
No directional unlock. Useful as a risk texture marker, not as a fade/continue signal.
Final Verdict
WATCH
Same story as session, weekday, and volume: microstructure and positioning mark volatility, not a tradeable mean-return edge on this sample. Revisit only with a cleaner residual vs prior |r|, or a different OI definition.
Comments (0)