Intra-Week Risk Controls (4h replay) — Backtest Report ⛔ REJECTED (both hurt)
Date: 2026-06-12
Analyst: Claude (for Jacques)
Test: /root/xsmom_intraweek_120626.py · Engine: /root/backtest_xsmom_PY-00027_100626.py
Builds on: 2026-06-12_volatility-scaled-sizing.md (the config these stack on).
1. Question
Weekly rebalancing holds positions through whatever happens mid-week. Does adding intra-week protection help against a mid-week crash or pump? Tested at 4h resolution (the weekly backtest can't see intra-week path) on the vol-scaled config: - Fast regime exit: flatten ALL positions to cash if BTC closes below its 50d MA at any 4h step. - Stop-loss X%: close any single position whose loss from entry breaches X% intra-week.
Honest bar: must help on tune AND OOS, robust across thresholds. 4h data covers 2024-06-04 → 2026-06-12 (full tune+OOS). Baseline 4h-replay reproduces the daily backtest exactly (+21%, Sharpe 1.17, April −13%) — simulator validated.
2. Results — every variant is worse than holding
| Config | Tune Sharpe | OOS return | OOS Sharpe | April | OOS stops / regexits |
|---|---|---|---|---|---|
| Baseline (hold to week-end) | 1.50 | +21% | 1.17 | −13% | 0 / 0 |
| Fast regime exit | 1.13 | −6% | −0.13 | −14% | 0 / 4 of 11 wk |
| Stop 10% | 0.24 | −36% | −2.74 | −15% | 40 stops |
| Stop 15% | 0.40 | −29% | −2.30 | −13% | 35 |
| Stop 20% | 1.12 | −20% | −1.48 | −14% | 28 |
| Stop 25% | 1.01 | −11% | −0.93 | −17% | 24 |
| Stop 30% | 1.05 | −18% | −1.07 | −26% | 19 |
| Regime exit + stop 20% | 0.77 | −35% | −2.76 | −20% | 26 / 4 |
No threshold improves tune or OOS. All hurt.
3. Why protecting against mid-week moves backfires
- Stop-losses get whipsawed. These coins swing 20–30% intra-week as normal noise. A stop sells the local low; the coin then recovers but the loss is locked and you're in cash for the bounce. In 11 OOS weeks a 10% stop fired 40 times, a 20% stop 28 times — systematically selling bottoms, converting recoverable dips into permanent losses. OOS fell from +21% to −20% or worse.
- Fast regime exit whipsaws too. BTC dips below its 50d line for hours (noise) → dump to cash → it pops back above → loss locked, rebound missed. A 50-day MA is a slow signal; sampling it every 4h makes it jittery. +21% → −6%.
- They don't even fix April (all −13% to −26%). April's coins gapped down and stayed down, so a stop realizes the same loss slightly earlier while bleeding on every other week where a position dipped and recovered.
4. Conclusion
Rejected — do not add stop-losses or a fast regime exit. The strategy is already protected the right way: structurally (market-neutral → broad moves hedged; vol-scaled → wild coins tiny; flat in bear), not reactively. Reactive intra-week rules fight the mean-reverting nature of these assets and lose. The correct response to a mid-week crash/pump is to hold and let the weekly rebalance handle it.
The one caveat: OOS is thin (11 bull weeks), but the failure is consistent on the tune set too and matches well-established theory (stops hurt on noisy mean-reverting assets), so the rejection is robust.
Status: FAIL (both controls). Recommended strategy unchanged: vol-scaled weekly, hold to rebalance.