spb/anomaly-atlas Public License

1---2project: anomaly-atlas3document: expD_leadlag_scan/analysis4author: Simon-Pierre Boucher5contact: contact@spboucher.ai6data_source: hfmarketdata.io7created: 2026-08-128modified: 2026-08-129status: reviewed10---1112# Analysis — expD_leadlag_scan1314Run: `results/expD_leadlag_scan/20260812T070742Z/results.json` (protocol15pre-specified; alignment primitives §8.1-gated — the gate caught an16interval-freshness bug before real data). All conclusions Level 0.1718## 1. The decay measurement (H07/H08 scan half)1920Both-fresh leads at +1min, 2006-2007 → 2014-2015:2122| Pair family | 2006-2007 | 2014-2015 |23|---|---|---|24| SPY → sector ETFs | +0.07 .. **+0.148** (all FDR) | −0.02 .. +0.028 |25| SPY → mega-caps | +0.01 .. +0.03 (mostly FDR) | −0.015 .. +0.046 |26| SPY → sparse names | +0.09 .. +0.12 | +0.06 .. +0.075 (still FDR) |2728Minute-scale market→component diffusion was REAL and large in 2006-2007 on29synchronized samples, and collapsed by 2014-2015 — direct, open-data30confirmation of the Chordia–Roll–Subrahmanyam convergence story at the 1min31floor. The residual 2014-2015 structure is statistically detectable (huge n)32but an order of magnitude smaller; several cells survive FDR, so the33pre-registered expected-negative clause for H08 is formally falsified —34recorded as such and handed to expF (costs will likely be decisive).3536## 2. ES ↔ SPY (H09 preview)3738corr(SPY_{t−1}, ES_t) = **−0.032** both-fresh, FDR-significant, and39identical to 4 decimals across all three continuous-futures splice variants40— the built-in robustness check held. Sign and size are consistent with41minute-scale basis reversion rather than one-way price discovery (the +142ES-leads direction is ≈0). A genuine structural candidate for expF/candidate43work; economically tiny.4445## 3. SPX → SPY: content staleness ≠ print staleness (H11 confirmed at scan level)4647SPY leads SPX by **+0.132** on the both-fresh subsample, with artifact share48≈ 0 — SPX "prints" every minute (fresh_frac 0.996), so synchronizing on49print times removes nothing: the staleness lives INSIDE the index50computation (Fisher 1966, measured live). Confirms candidate 03's design:51the correction must model constituent staleness, not print gaps.5253## 4. A T3 refinement for the taxonomy5455expB showed LOCF joins *inflate* leads for mid-stale names (+0.047). expD56adds the other arm: for ultra-stale names the LOCF zero-return mass *dilutes*57the measured lead (HTD: raw +0.048 vs fresh +0.073). The non-synchronicity58artifact is **non-monotone in staleness** — naive joins can hide as well as59manufacture lead-lag. Added to T3.6061## 5. Verdicts6263Clause (ii) falsified (tiny FDR residuals remain in 2014-2015); clauses64(iii), (iv), (v) held. Hand-off to expF: the 2014-2015 FDR survivors, the65ES→SPY basis effect, and the decay curve. expE (calendar) is next; its66budget is pre-counted in research_gaps.md.67