Date: 2026-09-10 | n=10951 predicted events.
a. pre- vs post-announcement attribution (signed fwd ret)
mean signed 10d fwd: +0.0018 (t=+2.31) pre-ann portion: -0.0005 post-ann portion: +0.0022
b. ex-surprise (|z|<1, no real news)
n=6666, mean signed fwd: +0.0008 high-news (|z|>=1) n=4285, mean: +0.0033
c. symmetry (by entry-drift sign)
drift>0 n=6217, signed fwd: +0.0122 drift<0 n=4734, signed fwd: -0.0120
Verdict
MIXED/WEAK: pre-share -0.26, low-news +0.0008, high-news +0.0033, negative-drift leg -0.0120. No clean mechanism — H10 stays CANDIDATE on portfolio stats alone, gated on forward proof.
Follow-up: is H10 just 10d momentum in disguise?
Unconditional 10d cross-sectional momentum (top 30, 10d hold, long-only): OOS Sharpe 0.93 (severe 0.14, IS 0.52) vs H10 1.00. Correlation between the two OOS net curves: 0.52. Verdict: NOT a clone — the pre-earnings window conditioning carries ~half-independent variance and a slightly better severe-cost profile (0.35 vs 0.14). H10 = short-horizon drift continuation filtered to scheduled-news windows; out-earns its unconditional cousin but the reason remains unresolved (leakage and premium stories both rejected above). Forward paper data is the next judge, not more decomposition.
Follow-up 2: hold-through-news vs pure pre-event (diagnostic, ex-post dates)
Signed 10d forward by whether the hold contains the actual announcement: through-ann (n=7724): +0.0015 (t=+1.52); pure pre-event (n=3261): +0.0024 (t=+2.24). The edge does NOT come from holding through news — it accrues before scheduled events regardless of the news itself. Leakage (needs news-selectability) and premium-capture (needs pre-accrual — also absent: pre-share was -0.26) are both out; what remains is coherent: a pre-event positioning drift — drift continuation into scheduled news that pays whether or not news lands inside the hold. Mechanism status: resolved enough to be testable (predict: H10 variants that exit before announcements keep the edge; news-conditional variants add nothing). H10 stays CANDIDATE, gated on forward proof.