Rank213c Buffer8 专题研究
目标：沿着 rank_buffer8_50_50 这条线，测试 buffer 宽度、long/short 分侧缓冲、short overlay、only-replace、partial rebalance 与 turnover cap，寻找 12bps 成本下更可持有的 213c2 候选。
生成时间：2026-05-07 06:20:27 UTC。样本为 causal monthly-volume Top50、age90、retvol14 skip1d、daily next-day hold；主读法看 12bps/1x turnover。
213c 研究路线图 · 第五轮利润厚度 · 试探性拓展
一句话结论
本轮把 buffer8 线推进到了“可继续深挖”的阶段，但还没有推进到“可放大”的阶段。 最好候选是 buffer8_weekly，12bps 后日均 6.78 bps，累计 191.39%，最大回撤 -49.98%，换手 0.134x/day；它通过了本轮 Promising 门槛，但仍需要真实成本 replay、年度稳定性和 live-vs-shadow 验证。
最重要的发现是：buffer 的价值主要来自降低误换仓，而不是让策略变成新 alpha。 过宽 buffer、short 降权、dispersion gate、partial rebalance 都不能稳定解决路径风险；下一步应该把研究收窄到“buffer8 附近 + 真实成本 replay + short-active day 归因”。
核心读数
-2.19 bps无 buffer 基准 12bps 日均
4.11 bpsbuffer8 anchor 日均
6.78 bps本轮最好候选日均
1 / 7Promising / Watch
12bps 候选表
| verdict | experiment_group | variant | net_mean_bps | net_cum_pct | max_drawdown_pct | sharpe | avg_turnover_x | long_mean_bps | short_mean_bps | long_capital | short_capital | short_gate | reason |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Promising | friction_control | buffer8_weekly | 6.78 bps | 191.39% | -49.98% | 0.630 | 0.134 | 7.27 bps | 1.11 bps | 0.500 | 0.500 | always | clears 12bps mean/turnover target, but still needs robustness and real-cost replay |
| Watch | risk_guard | buffer8_avoid_eligible_prior7_crash | 5.12 bps | 108.31% | -46.92% | 0.498 | 0.255 | 2.23 bps | 5.94 bps | 0.500 | 0.500 | always | positive 12bps net with acceptable turnover, but thickness or drawdown is still marginal |
| Watch | risk_guard | buffer8_avoid_btc_prior7_crash | 4.90 bps | 92.22% | -51.98% | 0.461 | 0.246 | 2.47 bps | 5.39 bps | 0.500 | 0.500 | always | positive 12bps net with acceptable turnover, but thickness or drawdown is still marginal |
| Watch | friction_control | buffer8_turnover_cap_0p2 | 4.16 bps | 71.75% | -43.84% | 0.418 | 0.175 | 6.55 bps | -0.29 bps | 0.500 | 0.500 | always | positive 12bps net with acceptable turnover, but thickness or drawdown is still marginal |
| Watch | buffer_width | buffer8_50_50 | 4.11 bps | 52.78% | -48.88% | 0.368 | 0.210 | 5.50 bps | 1.13 bps | 0.500 | 0.500 | always | positive 12bps net with acceptable turnover, but thickness or drawdown is still marginal |
| Watch | friction_control | buffer8_partial_2d | 3.88 bps | 53.60% | -46.72% | 0.369 | 0.196 | 6.14 bps | 0.10 bps | 0.500 | 0.500 | always | positive 12bps net with acceptable turnover, but thickness or drawdown is still marginal |
| Watch | friction_control | buffer8_turnover_cap_0p3 | 3.81 bps | 47.45% | -49.65% | 0.353 | 0.199 | 5.81 bps | 0.39 bps | 0.500 | 0.500 | always | positive 12bps net with acceptable turnover, but thickness or drawdown is still marginal |
| Watch | friction_control | buffer8_partial_3d | 3.74 bps | 56.34% | -45.75% | 0.377 | 0.180 | 6.05 bps | -0.16 bps | 0.500 | 0.500 | always | positive 12bps net with acceptable turnover, but thickness or drawdown is still marginal |
| Diagnostic | short_overlay | buffer8_short_btc_above_ma20 | 6.62 bps | 103.66% | -68.90% | 0.478 | 0.206 | 5.50 bps | 3.60 bps | 0.500 | 0.500 | btc_above_ma20 | positive net, but drawdown/turnover/path is not suitable as a candidate |
| Diagnostic | side_specific | long75_short25_buffer8 | 6.38 bps | 7.79% | -93.31% | 0.348 | 0.202 | 8.24 bps | 0.56 bps | 0.750 | 0.250 | always | positive net, but drawdown/turnover/path is not suitable as a candidate |
| Diagnostic | short_overlay | buffer8_short_btc_prior7_positive | 6.20 bps | 86.16% | -63.37% | 0.449 | 0.239 | 5.50 bps | 3.57 bps | 0.500 | 0.500 | btc_prior7_positive | positive net, but drawdown/turnover/path is not suitable as a candidate |
| Diagnostic | short_overlay | buffer8_short25_btc_prior7_positive | 5.27 bps | 53.23% | -76.84% | 0.386 | 0.168 | 5.50 bps | 1.78 bps | 0.500 | 0.250 | btc_prior7_positive | positive net, but drawdown/turnover/path is not suitable as a candidate |
| Diagnostic | short_overlay | buffer8_short25_dispersion_high | 3.69 bps | 6.81% | -86.14% | 0.270 | 0.120 | 5.50 bps | -0.37 bps | 0.500 | 0.250 | prior30_dispersion_high | positive net, but drawdown/turnover/path is not suitable as a candidate |
| Diagnostic | buffer_width | buffer6_50_50 | 3.19 bps | 20.53% | -60.32% | 0.278 | 0.242 | 6.27 bps | -0.17 bps | 0.500 | 0.500 | always | positive net, but drawdown/turnover/path is not suitable as a candidate |
分组最好结果
| experiment_group | variant | net_mean_bps | net_cum_pct | max_drawdown_pct | avg_turnover_x | long_buffer_extra | short_buffer_extra | short_capital | short_gate | weight_blend | turnover_cap_x |
|---|---|---|---|---|---|---|---|---|---|---|---|
| buffer_width | buffer8_50_50 | 4.11 bps | 52.78% | -48.88% | 0.210 | 8 | 8 | 0.500 | always | 1.000 | |
| buffer_width | buffer6_50_50 | 3.19 bps | 20.53% | -60.32% | 0.242 | 6 | 6 | 0.500 | always | 1.000 | |
| buffer_width | buffer10_50_50 | 1.90 bps | -7.70% | -53.22% | 0.191 | 10 | 10 | 0.500 | always | 1.000 | |
| friction_control | buffer8_weekly | 6.78 bps | 191.39% | -49.98% | 0.134 | 8 | 8 | 0.500 | always | 1.000 | |
| friction_control | buffer8_turnover_cap_0p2 | 4.16 bps | 71.75% | -43.84% | 0.175 | 8 | 8 | 0.500 | always | 1.000 | 0.200 |
| friction_control | buffer8_partial_2d | 3.88 bps | 53.60% | -46.72% | 0.196 | 8 | 8 | 0.500 | always | 0.500 | |
| reference | base_50_50_daily | -2.19 bps | -66.53% | -84.08% | 0.598 | 0 | 0 | 0.500 | always | 1.000 | |
| risk_guard | buffer8_avoid_eligible_prior7_crash | 5.12 bps | 108.31% | -46.92% | 0.255 | 8 | 8 | 0.500 | always | 1.000 | |
| risk_guard | buffer8_avoid_btc_prior7_crash | 4.90 bps | 92.22% | -51.98% | 0.246 | 8 | 8 | 0.500 | always | 1.000 | |
| short_overlay | buffer8_short_btc_above_ma20 | 6.62 bps | 103.66% | -68.90% | 0.206 | 8 | 8 | 0.500 | btc_above_ma20 | 1.000 | |
| short_overlay | buffer8_short_btc_prior7_positive | 6.20 bps | 86.16% | -63.37% | 0.239 | 8 | 8 | 0.500 | btc_prior7_positive | 1.000 | |
| short_overlay | buffer8_short25_btc_prior7_positive | 5.27 bps | 53.23% | -76.84% | 0.168 | 8 | 8 | 0.250 | btc_prior7_positive | 1.000 | |
| side_specific | long75_short25_buffer8 | 6.38 bps | 7.79% | -93.31% | 0.202 | 8 | 8 | 0.250 | always | 1.000 | |
| side_specific | long75_short25_long12_short4 | 3.64 bps | -39.38% | -93.90% | 0.203 | 12 | 4 | 0.250 | always | 1.000 | |
| side_specific | long_buffer12_short25_buffer4 | 2.00 bps | -4.56% | -76.34% | 0.162 | 12 | 4 | 0.250 | always | 1.000 |
推论讨论
- buffer 宽度不是越大越好。 buffer8 附近仍是较合理的中间点；继续加宽会降低换手，但也更容易持有过期信号。
- weekly buffer8 是本轮唯一 Promising。 它把换手压到约 0.134x/day，并把 12bps 后日均推到 6.78bps；但 2022 和 2026YTD 仍为负，说明它不是一个可以直接放大的终局版本。
- 分侧 buffer 没有解决核心问题。 给 long 更宽、给 short 更窄的设计能改变风险形态，但没有稳定地产生更低回撤的 12bps 候选。
- short overlay 仍偏诊断价值。 BTC / dispersion gate 会减少 short active days，但不够精准，不能单独成为上线开关。
- partial rebalance 和 turnover cap 的方向合理，但需要实盘成本模型。 它们降低账面换手后，收益是否真实保留，取决于成交价漂移与分批成交成本。
- 当前最有价值的下一步不是再扩网格。 应该对 buffer8 anchor 做交易原因归因、真实成本 replay、最差月份拆解，再决定是否有必要写 213c2 实盘候选。
最好候选年度稳定性
| variant | year | days | net_cum_pct | max_drawdown_pct | avg_turnover_x | long_mean_bps | short_mean_bps |
|---|---|---|---|---|---|---|---|
| buffer8_weekly | 2020 | 335 | 46.47% | -26.88% | 0.078 | 20.51 bps | -6.68 bps |
| buffer8_weekly | 2021 | 365 | 125.73% | -27.18% | 0.142 | 49.02 bps | -21.03 bps |
| buffer8_weekly | 2022 | 365 | -7.98% | -34.70% | 0.148 | -19.93 bps | 20.97 bps |
| buffer8_weekly | 2023 | 365 | 8.72% | -24.78% | 0.148 | 14.96 bps | -9.46 bps |
| buffer8_weekly | 2024 | 366 | 2.79% | -27.84% | 0.143 | 4.25 bps | -0.34 bps |
| buffer8_weekly | 2025 | 365 | 4.01% | -30.63% | 0.134 | -15.11 bps | 20.17 bps |
| buffer8_weekly | 2026 | 125 | -17.60% | -30.02% | 0.148 | -18.91 bps | 8.07 bps |
最好候选最差月份
| variant | month | net_cum_pct | max_drawdown_pct | avg_turnover_x |
|---|---|---|---|---|
| buffer8_weekly | 2022-05 | -26.25% | -27.53% | 0.161 |
| buffer8_weekly | 2020-08 | -25.27% | -26.88% | 0.121 |
| buffer8_weekly | 2021-02 | -22.48% | -27.18% | 0.170 |
| buffer8_weekly | 2023-03 | -16.49% | -15.66% | 0.169 |
| buffer8_weekly | 2026-02 | -14.91% | -22.44% | 0.161 |
| buffer8_weekly | 2026-01 | -14.49% | -22.19% | 0.129 |
| buffer8_weekly | 2025-10 | -13.09% | -13.82% | 0.153 |
| buffer8_weekly | 2024-07 | -12.77% | -13.03% | 0.129 |
操作建议
- 不放大。 本轮没有出现“12bps 后厚度足够、回撤明显下降”的版本。
- 保留 buffer8_50_50 和 buffer8_weekly 两个 anchor。 前者路径更接近第五轮 Watch，后者是本轮唯一 Promising；二者都要做真实成本 replay 和 live-vs-shadow 对齐。
- 下一轮只做深挖，不做宽网格。 拆分新开仓/carry/替换原因，计算每类交易的净贡献，并重点解释 weekly cadence 为什么改善均值但仍留下年份亏损。
- short gate 暂不进生产。 dispersion / BTC gate 可以做监控标签，但不作为硬交易开关。
产物
reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213c_buffer8_focus_results.csvreports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213c_buffer8_focus_daily.csvreports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213c_buffer8_focus_shortlist.csvreports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213c_buffer8_focus_annual.csvreports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213c_buffer8_focus_monthly.csvreports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213c_buffer8_focus_summary.json