Extreme BTC funding does not fade the next window

Verdict: NO_EDGE · Asset: BTCUSDT perpetual funding vs spot 1h · Sample: chronological 60/20/20 by funding event

Hypothesis: extreme positive (negative) funding predicts underperformance (outperformance) over the next funding window — classic crowding fade.

What happened

Funding from Binance USDT-M data.binance.vision monthly archives (2020-01 → 2026-07, n=7212), aligned to spot BTCUSDT 1h closes. Frozen discovery thresholds: P10 = −1.79e−5, P90 = +3.76e−4.

Q1 (primary 8h, validation): fade signs look right (HIGH mean < MID < LOW) but the gate fails — t(H−M)=−0.57, t(L−M)=+0.80; P(neg|HIGH)=0.47, P(pos|LOW)=0.50. Continue also fails. Discovery LOW side looked real (t≈2.08) and did not hold on validation. OOS: zero HIGH events under the frozen P90 (regime shift — extremes left the sample).

Q2: validation OLS β on funding is dead (t≈−0.11).

Trading

Q3 not opened (prereg).

Final Verdict

NO_EDGE

Do not build “fade extreme funding” bots from this Binance USDT-M / spot-1h alignment. Crowding folklore is directionally familiar in discovery and evaporates out of sample — including the extreme bucket itself.