Skip to content

Add execution-sensitivity robustness suite (anchor, grid, real-market)#144

Merged
weich97 merged 1 commit into
mainfrom
exec-sensitivity-robustness-suite
Jun 15, 2026
Merged

Add execution-sensitivity robustness suite (anchor, grid, real-market)#144
weich97 merged 1 commit into
mainfrom
exec-sensitivity-robustness-suite

Conversation

@weich97

@weich97 weich97 commented Jun 15, 2026

Copy link
Copy Markdown
Owner

Reviewer-driven robustness suite for paper 01 (also re-adds two scripts a bad git pathspec dropped from the prior PR):

  • Anchor initialization (warm-start harness + run_anchor_robustness.py): init-cost gap 0 at ideal, +0.025 (up to +0.046) at harsh corner -> part of the cold anchor's penalty is construction, reported honestly.
  • Parameter grid (run_execution_param_grid.py): liquid tau=1.0 (no reordering) -> illiquid tau=0.43; reordering appears already at typical-equity params (0.81).
  • Real-market (run_execution_sensitivity_real.py): reordering persists on real Yahoo OHLCV (tau 0.81-0.91), milder than synthetic.
  • Turnover control (analyze_turnover_control.py): within-tercile tau 0.63 vs full 0.24.

Test plan

  • factory/runner/sweep tests pass; ruff clean
  • all four analyses run end-to-end deterministically (no API)

Reviewer-driven robustness checks for paper 01, plus warm-start harness
support and two scripts the previous PR dropped (a bad git pathspec
aborted that add):

- Warm-start holdings: ExperimentConfig.initial_holdings_weights seeds the
  portfolio at first-bar prices free of construction cost; runner applies
  it; factory plumbs it through.
- run_anchor_robustness.py: cold vs warm-start buy-and-hold across the
  ladder. The init-cost gap is 0 under ideal fills but +0.025 (up to +0.046
  in jump/tail) at the harsh corner, so part of the cold anchor's penalty is
  one-shot construction; the warm-start anchor is reported alongside.
- run_execution_param_grid.py: impact x participation x latency grid from
  liquid (tau 1.0, no reordering) to illiquid (tau 0.43); reordering already
  appears at typical-equity parameters (tau 0.81).
- run_execution_sensitivity_real.py and analyze_turnover_control.py
  (re-added): real Yahoo OHLCV check (reordering persists, tau 0.81-0.91,
  milder than synthetic) and turnover-tercile control.
- Sweep records per-run turnover_events.
@weich97 weich97 merged commit 3b7e30b into main Jun 15, 2026
10 checks passed
@weich97 weich97 deleted the exec-sensitivity-robustness-suite branch June 15, 2026 07:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant