You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
I Luk Kim 261f67dfde Promote PeerSympathy v2 reaction_close sweep10 — corr 0.80 + top_n 1
Threshold sweep over PeerSympathy v2 reaction_close found a viable config.
After 10 ablation runs (correlation_min, leader_reaction_min, top_n_peers,
event_types, VIX gate), the dominant lever is correlation_min:

  baseline 0.55:  -2.35% / MDD 24.5% / SQS 30.2 / PF 0.977
  sweep01 0.65: +11.10% / MDD 16.1% / SQS 50.2 / PF 1.204
  sweep02 0.75: +15.76% / MDD  9.1% / SQS 62.5 / PF 1.835
  sweep07 0.80: +14.11% / MDD  3.8% / SQS 80.2 / PF 2.198
  sweep10 0.80 + top_n 1: +15.35% / MDD 3.83% / SQS 81.5 / PF 2.46
  sweep09 0.85:  +0.98% / MDD  3.8% / SQS 21.4 / PF 1.156 (cliff)

Mechanical insight: pushing correlation gate from 0.55 → 0.80 trims
coincidental ETF-shared peers and concentrates on structurally-clean
pairs where the leader's print is new information for the peer's
fundamentals — NOT the noisy NVDA/AVGO/AMD/MU semis cluster (their
mutual corr usually 0.55-0.70 with high cross-noise from competing
product cycles).

Sympathy pairs that surfaced in sweep10:
  Utilities      VST → CEG (independent power generators)
  Ag equipment   AGCO → DE
  Industrials    DAL → CAT, HWM → GE
  Software       DT → CRM, IT → INTU
  Datacenter     VRT → PWR (electrification)
  Energy         OVV/OXY/EOG/COP cluster

Hits all PROMOTE thresholds: SQS 81.5 > 50, return +15.35% > +5%,
MDD 3.83% < 25%. Sub-scores: profitability 100, risk 62.9, consistency
84.3, robustness 59.3 (lowest — sample-size discount). Sharpe 0.721,
Calmar 0.906, expectancy 0.072 R, avg holding 3.56 days, win rate 55.6%.

CAVEATS:
- N=27 trades is thin. Bootstrap PF 95% CI is [0.68, 10.82]. Treat as
  directional candidate, not production. Robustness 59.3 reflects this.
- corr_min cliff at 0.85 (universe runs out). 0.80 is at the inside edge
  of viable; re-baseline on snapshot shifts.
- macro_vix_max=30 was a no-op in sweep08. Either no candidates on
  VIX>30 days, or gate not wired to peer_sympathy in selector.py.
  Falsified hypothesis; not promoted as a lever.

Run: bt_return_max_long_v1_midlarge-liq_20260509072637244341_e1c62252.
Parent: peer_sympathy_poc_v2_reaction_close (commit 68102f1).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
3 months ago
..
backtest Optimize v7 strategy: v7.356 achieves CW 2159% + SQS 90.7 (Pareto improvement over v7.314) 4 months ago
experiments Promote PeerSympathy v2 reaction_close sweep10 — corr 0.80 + top_n 1 3 months ago
intraday Clean up superseded configs and commit accumulated R&D infrastructure 4 months ago
ranking_models Add overlay engine, ranking models, snapshot pipelines, and research tools 5 months ago
snapshots Fix snapshot refresh: enable auto-rebuild, fix date-range edge cases, log incremental failures 4 months ago
app.yaml feat: PEAD mid-cap strategy + pipeline hardening + README cleanup 5 months ago
fred_series.yaml feat: implement ACE-F v1 Phase 1 -- Stock Oracle 기반 이벤트 파이프라인 5 months ago
symbols.yaml feat: PEAD mid-cap strategy + pipeline hardening + README cleanup 5 months ago
symbols_backup_largecap.yaml feat: PEAD mid-cap strategy + pipeline hardening + README cleanup 5 months ago
symbols_broad_snapshot_3408.yaml Investigate compound mode: V23 is absolute champion in all modes 4 months ago
symbols_largecap.yaml feat: PEAD mid-cap strategy + pipeline hardening + README cleanup 5 months ago
symbols_midcap.yaml feat: PEAD mid-cap strategy + pipeline hardening + README cleanup 5 months ago
symbols_midcap_smallmid.yaml Investigate compound mode: V23 is absolute champion in all modes 4 months ago
symbols_midlarge_snapshot_exact.yaml Add expanded ORB simulator features and metrics 4 months ago
symbols_orb_smoke.yaml Investigate compound mode: V23 is absolute champion in all modes 4 months ago
symbols_smallmid.yaml feat: PEAD mid-cap strategy + pipeline hardening + README cleanup 5 months ago