TeX 53.6%
Python 46.2%
1# EMPIRICAL_DESIGN.md23Working design document — expands as stages complete. Authoritative parameter values4live in `config/`. Protocol reference: project CLAUDE.md §§7–16.56## Objects of interest71. **Shocks:** idiosyncratic residual-return extremes (lagged conditional quantiles /8 robust SD multiples), separated by sign; volume, range, and illiquidity-proxy shocks.92. **Propagation outcomes:** breadth, speed (censoring-aware), intensity, amplification10 (ISAR), persistence, asymmetry, at horizons h ∈ {1,2,5,10,15,30,60,120} min within session.113. **Networks:** estimated strictly pre-shock (rolling/frozen windows): correlation,12 partial correlation, lead–lag, volatility spillover, sparse VAR/glasso; centrality roles.134. **Evolution:** flexible year effects first; era bins only as secondary presentation.1415## Identification strategy (threat: common information)16- Exclude scheduled macro windows from the idiosyncratic baseline.17- Isolation filters: low contemporaneous market/sector return, no simultaneous18 extreme in >2% of universe, leave-one-out factors.19- Placebo timing matched on stock, minute-of-day, volatility state.20- Permuted networks preserving degree/sector/size/liquidity.21- Negative controls: future network weights (falsification only), pre-shock outcomes.2223## Inference24Two-way clustering (source × date) as baseline; day-block bootstrap for25distribution-level statements; randomization inference for event studies;26multiple-testing corrections within outcome families.2728## Status29- [x] Stage A pilot audit (2026-07-13)30- [x] Stage B full download (2026-07-13)31- [x] Stage C full raw audit32- [x] Stage D clean panels33- [x] Stage E descriptive evidence34- [x] Stage F shocks & propagation (baseline, F2 network/cumulative/co-jumps, robustness)35- [x] Stage G paper (v1.0 2026-07-13; v1.1 revised 2026-10-17)36- [x] Revision diagnostics from stored events (scripts/17, 2026-10-17)3738Implemented vs planned (as of 2026-10-17): implemented — residual shocks |z|≥6 with isolation,39placebo timing, ΔΔ, day/month/ticker bootstrap, lagged top-20 correlation neighbours, cumulative40breadth, co-jump minutes, pre-event levels, co-arrival classes, stratified estimator. Not implemented41— two-way clustering regressions, local projections with fixed effects, permuted networks, future42network weights, multiple-testing corrections, sector proxies (no security master), macro-window43propagation as a separate object, ML extension. See paper §8 and REVIEW-REPORT §8.44