源文件:research/quant_digests/2026-03-23_0803_pfe-extreme-veto-not-followup-greenlight.md
这次主看 fmzquant/strategies 里的 ChaoZhang (2024)《Polarized Fractal Efficiency (PFE) Trading Strategy》。它最值得我们 desk 拿来用的旁支,不是“再加一个主触发器”,而是把 路径效率极值变成 follow-up 的拒单/降仓条件。
|PFE_EMA| 高阈值不该被当成 continuation 绿灯,更像“末端冲刺/追单风险”提示;更诚实的角色是 veto / size-down overlay。Length=9, LengthEMA=5)在 BTC/ETH/SOL 近 180 天 15m 数据做轻量分桶,比较 |PFE_EMA| 不同区间的后续方向延续率。本地最小快检(pooled,n=52,401):
|PFE_EMA|<=20:1-bar continuation=49.08%,4-bar continuation=48.84%;20<|PFE_EMA|<=50:1-bar continuation=48.50%,4-bar continuation=47.91%;|PFE_EMA|>50:1-bar continuation=47.13%,4-bar continuation=47.77%。可执行读法:high_abs 相比 low_abs 在 1-bar continuation 低约 1.95pct,不支持“PFE 越极端越该追”。
补充:按资产看(BTC/ETH/SOL)high_abs 的 cont_hit_1 都低于各自低分桶,方向一致。
|PFE_EMA| 极值可做 no-chase veto,减少追在最后一脚。|PFE_EMA| 极值区作为 anti-chase 风险过滤。size-down,次选 temporary veto。high-|PFE_EMA| veto/降仓能降低 post-break 假延续与回吐损失,而不靠“砍光交易数”取胜。PFE = sqrt((close-close[9])^2 + 100)C2C = sum(sqrt((close-close[1])^2 + 1), 9)xFracEff = sign(close-close[9]) * round((PFE/C2C)*100)PFE_EMA = EMA(xFracEff, 5)|PFE_EMA| > rolling_q90 时 veto 新 follow-upsize × 0.5BTC/ETH/SOL perpetual,120~180 天,成本 6/10/15 bps。post_cost_expectancy、false_follow_ratio@4bars、trade_count_retention、max_drawdown。fmzquant/strategies)reports/artifacts/quant_digests/pfe_path_efficiency_proxy_20260323/btcusdt_pfe_frame.csvreports/artifacts/quant_digests/pfe_path_efficiency_proxy_20260323/ethusdt_pfe_frame.csvreports/artifacts/quant_digests/pfe_path_efficiency_proxy_20260323/solusdt_pfe_frame.csvreports/artifacts/quant_digests/pfe_path_efficiency_proxy_20260323/bucket_summary_by_asset.csvreports/artifacts/quant_digests/pfe_path_efficiency_proxy_20260323/bucket_summary_cross_asset.csvreports/artifacts/quant_digests/pfe_path_efficiency_proxy_20260323/bucket_summary_pooled.csv