Skip to main content
verified results 15 arms synthetic FOMC shocks seed 7 · 200 bars

Does the event gate earn its cost?

15 cross-sectional momentum variants run against the same synthetic universe, same FOMC vol shocks, same 200-bar window. Every arm consumes the identical price stream — the only difference is the rule the strategy applies. Bars below sort Sharpe-descending; deltas anchor to the XsMomentumLiveStrategy baseline.

Synthetic data — no claim of an edge that transfers to real markets. The point is to show every arm under the same conditions so an operator can read the *relative* shape, not the absolute number.

All arms · Sharpe on one axis

sorted high → low

Sharpe domain -1.00 → 2.00. Pnl values in synthetic dollars (the universe starts at $100 base). Max drawdown is peak-to-trough over the 200-bar window.

Across 5 seeds · mean ± stdev

seeds 7..11

The single-seed leaderboard above can be a fluke. This table re-runs the same 15 arms across 5 consecutive seeds and sorts by *mean* Sharpe — what's consistent across the sweep, not what happened to lead on one seed. The bar shows the mean; the numbers beside it show the 5-seed stdev and min/max range.

  • mean leader Time-series momentum · absolute-threshold
    stdev 0.78 · range 0.63 → 2.14
  • Vol-weighted momentum
    stdev 0.74 · range 0.26 → 2.15
  • FOMC blackout · flatten 24h pre-event
    stdev 0.48 · range 0.36 → 1.61
  • Spread filter · tuned threshold
    stdev 1.00 · range -0.26 → 2.10
  • Event damping · continuous risk-off near FOMC
    stdev 0.49 · range 0.34 → 1.62
  • baseline Baseline · long-short cross-sectional momentum
    stdev 0.58 · range 0.17 → 1.66
  • Spread filter · gate on cross-sectional dispersion
    stdev 0.58 · range 0.17 → 1.66
  • Long-only · top-quantile momentum
    stdev 1.49 · range -0.93 → 2.45
  • Equal-risk long-only
    stdev 1.49 · range -0.93 → 2.45
  • Post-FOMC reversal · sign-flipped drift
    stdev 0.77 · range -0.64 → 1.14
  • Four-factor · adds skew
    stdev 1.03 · range -1.29 → 1.51
  • Post-FOMC drift · capture 4h after release
    stdev 0.77 · range -1.14 → 0.64
  • Three-factor · momentum + z-score + vol
    stdev 0.57 · range -0.88 → 0.39
  • Two-factor · momentum + z-score
    stdev 0.68 · range -1.14 → 0.37
  • Four-factor · tuned weights
    stdev 0.39 · range -0.59 → 0.35

What this reading changes: an arm that led on seed 7 but has stdev >1.0 across the sweep is high-variance, not actually better — the wide min/max range tells you which seed got which luck. An arm near the top with stdev under 0.58 (baseline's noise floor on this sweep) is the consistent leader.

By intervention family

  • Baseline

    1 arm
    • Baseline · long-short cross-sectional momentum sharpe 1.24

  • Factor composition

    4 arms
    • Two-factor · momentum + z-score sharpe -0.14
    • Four-factor · adds skew sharpe -0.23
    • Three-factor · momentum + z-score + vol sharpe -0.27
    • Four-factor · tuned weights sharpe -0.53

  • Weighted variants

    1 arm
    • Vol-weighted momentum sharpe 0.58

  • Constrained variants

    2 arms
    • Long-only · top-quantile momentum sharpe 1.88
    • Equal-risk long-only sharpe 1.88

  • Time-series momentum

    1 arm
    • Time-series momentum · absolute-threshold sharpe 0.63

  • Event gates

    2 arms
    • FOMC blackout · flatten 24h pre-event sharpe 1.13
    • Event damping · continuous risk-off near FOMC sharpe 1.11

  • Event capture

    2 arms
    • Post-FOMC drift · capture 4h after release sharpe 0.64
    • Post-FOMC reversal · sign-flipped drift sharpe -0.64

  • Regime gates

    2 arms
    • Spread filter · gate on cross-sectional dispersion sharpe 1.24
    • Spread filter · tuned threshold sharpe 0.83

reproduce this comparison

$ uv run python scripts/build_compare_fixture.py
wrote examples/fomc_blackout_compare/results.json (15 arms)

The script holds the pinned config (seed = 7, days = 200, fomc_vol_multiplier = 3.0000, fomc_drift_bps = 50.0000) and rewrites results.json. Each arm's Sharpe / PnL / drawdown is a real backtest number — no fixture fudge-factor. Bumping a config value in the script is a deliberate rotation of these on-site numbers.

Keyboard shortcuts

deskgd findgf strategiesgs featuresge promotionsgp apiga botgb writinggw

Shortcuts are no-ops while typing in an input or textarea.