Task #332 ATR 3.0x Follow-up Readout
As of: 2026-07-03 12:00 CST
Scope: research-only follow-up after Kaite proposed ATR 3.0x + Trend<=5 to reduce whipsaw in choppy regimes. No production files or cron were changed.
1. Original Task #332 LOSO lens
Runner: scripts/run_task332_atr30_followup_is_oos.py
Output: output/streaming_hist_2012/task332_atr30_followup_is_oos/
This reuses the original Task #332 canonical validation path, so it is comparable with the 2.5x / 3.25x IS/OOS robustness audit. It is not the Wanbai SWS L2 cap30 headline path.
Main result:
- ATR 3.0x vs 2.5x, Trend fixed <=5: pre-registered Sharpe/MDD fold rule passes only 1/3 folds under normal friction and 1/3 under high friction. Therefore 3.0x is not supported as a Sharpe edge.
- ATR 3.25x vs 3.0x, Trend fixed <=5: passes 0/3 folds under normal and high friction. 3.25x does not add robust incremental evidence over 3.0x under this lens.
Key LOSO points:
- 2018-2023: 3.0x improves Sharpe vs 2.5x and reduces exits; this is the choppy-regime benefit Kaite is trying to preserve.
- 2015-2017: 3.0x worsens MDD vs 2.5x in the canonical path.
- 2024-2026: 3.0x is near flat under normal friction and negative under high friction.
Engineering read: if the claim is "3.0x produces robust Sharpe alpha", the answer is no. If the claim is "3.0x is a risk/whipsaw tradeoff selected by mandate, not by Sharpe edge", use the SWS cap30 headline below.
2. Current production / Wanbai SWS L2 cap30 headline lens
Runner: scripts/run_task332_atr30_sws_cap30_headline.py
Output: output/streaming_hist_2012/task332_atr30_sws_l2_cap30_headline/
This reuses the Wanbai v5.5.14 replay setup: SWS L2 cap30, fixed 80m entry liquidity gate, Trend exit <=5, only ATR multiplier changes. This is the correct lens for the LP-document headline and the MDD mandate discussion.
Full sample headline
| Friction | ATR | Annual | Sharpe | MDD | Final NAV | Sells | ATR sells | Trend sells |
|---|---|---|---|---|---|---|---|---|
| normal | 2.5 | 16.16% | 0.827 | -29.48% | 5.06x | 348 | 189 | 55 |
| normal | 3.0 | 18.22% | 0.924 | -26.86% | 6.12x | 330 | 147 | 77 |
| normal | 3.25 | 17.52% | 0.885 | -28.02% | 5.74x | 327 | 139 | 83 |
| high | 2.5 | 14.46% | 0.737 | -31.38% | 4.31x | 352 | 193 | 56 |
| high | 3.0 | 15.77% | 0.802 | -27.91% | 4.88x | 333 | 147 | 81 |
| high | 3.25 | 15.72% | 0.795 | -29.90% | 4.86x | 330 | 138 | 89 |
Segment MDD
| Friction | ATR | 2015-17 MDD | 2018-23 MDD | 2024-26 MDD |
|---|---|---|---|---|
| normal | 2.5 | -21.95% | -27.27% | -18.50% |
| normal | 3.0 | -21.03% | -26.50% | -18.50% |
| normal | 3.25 | -21.10% | -26.24% | -18.50% |
| high | 2.5 | -23.43% | -28.25% | -18.87% |
| high | 3.0 | -21.64% | -27.60% | -18.87% |
| high | 3.25 | -22.82% | -27.62% | -18.87% |
Engineering read under Ray's risk-priority lens:
- ATR 3.0x + Trend<=5 materially improves full-sample MDD vs 2.5x in the current SWS cap30 production lens: normal -29.48% -> -26.86% (+2.62pp), high -31.38% -> -27.91% (+3.47pp).
- It does not restore the old v1.5 -24.43% MDD, and does not get normal friction back inside the 25% standard mandate; it moves the profile from roughly 30% pressure disclosure toward the middle.
- 2018-2023 MDD improvement is modest but directionally positive: normal +0.77pp, high +0.65pp.
- Under this SWS cap30 headline lens, 3.0x is better balanced than 3.25x: 3.25x has slightly fewer ATR sells but worse full-sample MDD than 3.0x in both normal and high friction.
3. 2026-07-02 P1 static impact
Switching from ATR 2.5x to ATR 3.0x is a looser ATR stop, so it creates zero immediate forced sells. It only lowers the ATR stop where the ATR entry stop is active.
Names affected on 2026-07-02 static P1 watch:
| Code | Name | Close | Trend | Effective stop @2.5 | Effective stop @3.0 | Buffer @2.5 | Buffer @3.0 | Main watch |
|---|---|---|---|---|---|---|---|---|
| 688127.XSHG | 蓝特光学 | 85.37 | 7 | 78.89 | 75.87 | 8.2% | 12.5% | ATR watch loosened |
| 300476.XSHE | 胜宏科技 | 306.09 | 7 | 286.00 | 274.17 | 7.0% | 11.6% | ATR loosened, but nearest live watch remains MA200 |
| 300502.XSHE | 新易盛 | 509.00 | 6 | 437.21 | 437.21 | 16.4% | 16.4% | unchanged; one Trend point above <=5 |
4. Governance status
- Active production remains v1.5.1_sws_l2_cap30_atr25_trend5 until Ray CIO sign-off and a new Kaite-confirmed production change.
- If accepted, this is not a return to the "3.25x Sharpe edge" claim. It should be framed as a mandate/risk preference change: reduce whipsaw and pull MDD back from the 30% pressure zone, while still disclosing that the result remains above the 25% standard mandate.