源文件:research/quant_digests/2026-03-20_1132_strict-bms-impulse-not-shared-gate.md
这轮主看一个很新的仓库 Madrycrypto/fibo71-bot(2026)。我没有复刻它“Fib 71-79”主叙事,而是只抽其中更适合我们 desk 的旁支:src/indicators/bms_detector.py 里把 BMS 冲击质量写成可计算规则——3 根同向动量K、每根 body≥60% 区间、且突破距离>0.5ATR(14)。
BTC/ETH/SOL 近 120 天 15m 做事件快检,对比 raw breakout 与 strict impulse breakout 的 4-bar signed return 与 re-entry 率。raw=2385,strict impulse 只有 10(0.42%);short 侧 raw=2560,strict impulse 33(1.29%)。它太稀疏,直接 shared 化会明显压缩交易频次。4-bar re-entry 显著更低(long 10.0% vs raw 44.1%;short 9.1% vs raw 39.3%),说明它确实更“冲得直”,但 short 侧仍不保证收益方向。high-conviction subset 或 size-down 反向否决 参考。strict BMS impulse 作为 shared gate 会过度降频;把它降级为“打分项/分层项”后,可能在不显著砍交易数的前提下改善 follow-up 质量。
strict_impulse=1 当且仅当:1) 最近 3 根K 与 breakout 同向; 2) 每根 |close-open|/(high-low) >= 0.60; 3) break_distance / ATR14 > 0.5。
2/3 满足 + median body>=0.50 + distance>0.25ATR。post-cost expectancy(按 long/short 分开)trade retention(相对 baseline 的留存率,防止“只靠极端降频换好看”)MadrycryptoN/Ahttps://github.com/Madrycrypto/fibo71-bothttps://github.com/Madrycrypto/fibo71-bothttps://github.com/Madrycrypto/fibo71-bot/blob/main/src/indicators/bms_detector.pyhttps://github.com/Madrycrypto/fibo71-bot/blob/main/README_BMS_STRATEGY.mdN/Ahttps://developers.binance.com/docs/derivatives/usds-margined-futures/market-data/rest-api/Kline-Candlestick-Datahttps://fapi.binance.com/fapi/v1/klines?symbol=BTCUSDT&interval=15m&limit=1500--- 快检文件:
reports/artifacts/literature/strict_bms_impulse_pool_summary_2026-03-20.csvreports/artifacts/literature/strict_bms_impulse_asset_summary_2026-03-20.csvreports/artifacts/literature/strict_bms_impulse_event_examples_2026-03-20.csvreports/artifacts/literature/strict_bms_impulse_metadata_2026-03-20.jsonreports/artifacts/literature/strict_bms_impulse_all_events_2026-03-20.csv