UTC hour effects on BTC are vol, not a return premium

Verdict: REJECTED · Asset/TF: BTCUSDT 1h and 15m · Sample: chronological 60/20/20

Hypothesis: stable UTC hour-of-day / Asia–EU–US session effects on mean return (not just known louder US hours).

Q1 — directional hours

FAIL on both TFs. Discovery froze high-mean hours 21 & 13 and low-mean 23 & 2. On validation, hour 13 falls to rank 21; hour 2 rises to rank 10. Spearman of the 24-hour mean-return profile ≈ 0.05. The US-open folklore hour does not hold.

Q1 — |return| / RV

PASS. Hours 14 & 0 stay loud; 4 & 5 stay quiet. Spearman 0.79 (1h) / 0.86 (15m). Shuffle-null half-rank pass ~6%. Session magnitude is real: US louder, Asia quieter.

Q2 — sessions

Asia mean sign-flips (discovery −0.40 bp → validation +0.45 bp). Weekend does not fake a directional edge. Hour 21 is year-unstable (2025 already negative on val).

Trading

EU-only long (next_open→close) that cleared a generous gate: at 5 bps every year is negative on discovery, validation, and OOS. Even at 0 cost, 2022 is already red.

Final Verdict

REJECTED

The clock predicts how loud the hour is, not a mean-return premium. Do not build hour-of-day long/short from BTCUSDT on this evidence. Consistent with the rest of this week’s atlas: magnitude clustering is real; directional stories around it usually are not.