BTC 5-day extremes do not mean-revert

Verdict: NO_EDGE · Asset/TF: BTCUSDT 1d · Sample: chronological 60/20/20

Hypothesis: after a large 5-day move (discovery P90 of |ret_5d|), the next 5 days fade. Intentionally the opposite claim of daily TSMOM. Frozen R=5, H=5, P90 = 13.34%, fade = −sign(ret_R) on extremes only. Non-overlapping H-blocks for the Q1 gate.

What happened

Q1 val fails. Fade long-short −2.39%, t=−0.76. Only 4 validation extremes — the frozen 13.34% P90 barely triggers in 2023–26. Extreme-up continues (+4.79%, t=+1.65), not fades.

Discovery was already wrong-way: extreme-up E[fwd]=+6.32%, t=+3.19 (continuation). Fade LS t=−2.15.

Continuous corr(ret_5d, fwd_5d) on val: −0.054 (t=−0.54). Not mean reversion.

Q2: val extremes too thin (2/1). Cannot promote. Q3 skipped.

Not REJECTED: the val reverse lacks |t|≥2. Not OVERFIT: fade never cleared discovery.

Trading

No book to cost-test.

Final Verdict

NO_EDGE

Do not fade 5-day BTC blow-offs from this kline sample. Large daily moves in discovery kept going; validation barely had any extremes left under the freeze.