Hypothesis
39% of YELLOW days upgrade to GREEN by 10:00 AM. If we detect the upgrade and enter late, we could capture additional P/L on days we'd otherwise sit out — without increasing risk, since the regime has confirmed GREEN conditions.
Baseline
115 days · 364 trades · +$1,360 P/L · 53% day WR · 48% trade WR · Sharpe 0.24
Dataset
250 trading days (March 2025 – March 2026). 4,848,591 five-minute bars across 215 symbols. All candidates are tickers that appeared in the morning scan on YELLOW-at-open days (60 qualifying days with tradeable candidates).
Procedure
Sweep 7 entry times from 09:45 to 10:35. For each checkpoint:
- Reconstruct regime at
check_timeusing SPY gap, VWAP slope, intraday return, and breadth (4 of 6 regime signals) - If YELLOW→GREEN upgrade detected, enter all scan candidates at the
entry_timebar close - Run V3 exits: ATR disaster floor (2×ATR₄, 3% minimum, 09:45 grace, 2-bar confirm) + 15:20 time exit
- Apply 15bps round-trip slippage to all trades
Results
| Check | Entry | Upgrades | Trades | P/L | Day WR | Trade WR | Sharpe | Max DD |
|---|---|---|---|---|---|---|---|---|
| 09:40 | 09:45 | 19 | 60 | -$1,734 | 37% | 42% | -1.57 | $6,196 |
| 09:45 | 09:50 | 23 | 82 | -$55 | 43% | 46% | -0.04 | $6,221 |
| 09:50 | 09:55 | 28 | 95 | -$4,525 | 43% | 43% | -3.21 | $6,842 |
| 09:55 | 10:00 | 25 | 85 | -$6,829 | 40% | 35% | -5.36 | $6,829 |
| 10:00 | 10:05 | 27 | 87 | -$7,132 | 41% | 33% | -5.85 | $7,666 |
| 10:15 | 10:20 | 22 | 68 | -$5,890 | 32% | 31% | -5.50 | $6,790 |
| 10:30 | 10:35 | 22 | 75 | -$11,484 | 18% | 25% | -14.03 | $11,845 |
Total P/L by Entry Time
Every entry time produces a net loss. The later the entry, the worse the outcome.
Win Rate Decay
Trade win rate degrades from 48% (baseline at open) to 25% (10:35 entry). The edge is at the open.
Conclusion
Hypothesis rejected. Late entries on YELLOW→GREEN upgrade days lose money at every tested time. The more upgrades detected (later checks find more), the more money lost — proving these are survivors of the first fade, not future winners.
Win rate degrades monotonically from 48% to 25%. Sharpe ratio goes from 0.24 (baseline) to -14.03 (worst case). Max drawdown exceeds $11k at the latest entry.
Implications
- Keep 9:30:01 entry. The gap-and-go edge is concentrated at the open. Delayed entry misses the initial momentum burst.
--late-entrystays notification-only. The flag triggers a watching message and 10:00 AM re-check, but does not auto-execute. Manual confirmation preserved as a safety gate.- YELLOW sit-out is correct. Even when YELLOW days upgrade to GREEN, the late entry destroys value. The regime filter works — trust it.
- Shadow tracker validates. Continue logging would-be P/L on sit-out days via
core/shadow_tracker.pyfor ongoing confirmation.
Next Experiments
- Sector-specific late entry: Do certain sectors (biotech, crypto) hold momentum longer than others after late entry?
- Partial position late entry: Would entering with 50% size on upgrades reduce drawdown enough to be profitable?
- ORB filter on late entries: If we wait for an opening range breakout before entering on upgrade days, does the signal quality improve?
- Exit optimization by entry time: V3 uses a fixed 15:20 time exit — do later entries need earlier exits?
References
- LeBeau, C. — Chandelier Exit: ATR-based trailing stop methodology, basis for V3 disaster floor
- Clenow, A. (2019) — Trading Evolved: Systematic momentum strategy design and regime filtering
- Internal:
sim_late_entry_sweep.py— full simulation source (250-day backtest, 7 time slots, V3 exits) - Internal:
core/regime.py— 6-signal traffic light gate (GREEN/YELLOW/RED)