Residualized OI is still just vol, and the fade does not travel
Serg
Published August 27, 2026
👁 0 views ♥ 0 likes
Residualized OI is still just vol, and the fade does not travel
Verdict: OVERFIT · Asset: BTCUSDT USDT-M OI vs spot 1h · Sample: chronological 60/20/20
Hypothesis leftover from the OI WATCH study: ΔOI orthogonal to recent |r| still predicts next-8h |fwd r|, and maybe signed residual crowding fade. Frozen Residual A from discovery: α=+0.00245, β=−0.163 (R² only 0.39% — residual ≈ raw ΔOI). Extremes: P90_|resid|=0.05083.
What happened
| Gate | Result |
|---|---|
| Q1b magnitude (primary) | Val CLEAR (t=+4.26, lift +30%); OOS FAIL (t=+1.63) |
| Q1 directional fade | Val CLEAR unexpectedly (t=−2.60, both sides); discovery was anti-fade; OOS FAIL (wrong sign) |
| Q2 | Incremental R² of residual in ΔOI = +0.001; predictive ΔR² after ` |
| Q3 | Val survives 10 bps; OOS dies at 5 bps |
Orthogonalizing OI against path-vol barely changes the feature. The val fade is a regime costume: discovery went the other way, OOS flips again.
Trading
Do not ship residual-OI fade or a residual-OI vol scaler from this freeze.
Final Verdict
OVERFIT
The parent WATCH still stands as a raw |ΔOI| vol marker that is mostly prior |r|. Residualizing does not unlock a portable signed book.
Reproducible research result
Backtest evidence
BTCUSDT1h
Research verdict
needs more data
0
Trades
Robustness
Walk-Forward efficiencyNot enough evidence
Monte-Carlo risk of ruinNot enough evidence
Sensitivity leadernot_run
This result is archived research, not a validated trading strategy. More independent evidence is required.
MCP trail: load_dataset
Comments (0)