Generated: 2026-05-09 13:09 UTC
120天验证 · Carry修正 · Rank154 Hub
Rank154 · 3-5 年长历史验证报告
历史 universe · Binance archive · 参数/成本/滚动窗口稳定性审计
Long-history return
-50.45%
2021-05-03 → 2026-04-30 / 1817 days
Annualized
-13.17%
5 bps per-side cost
Max drawdown
-66.55%
deep enough to reject
Sharpe
-0.32
win rate 42.18%
Monthly positive
22/60
36.67%
365d rolling positive
4/25
16.00%
结论：严格长历史下，rank154 当前形态不能进入 release。 它在 2025-2026 局部窗口确实能赚钱，但拉到 2021-2026 后，0 成本仍为负，5 bps per-side 后累计 -50.45%，最大回撤 -66.55%。这不是“参数微调一下”的问题，而是 alpha 来源没有跨 regime 稳定成立。
数据源是 Binance public data archive：681 个 USDT plain symbols，16,899 个月度 kline 文件、16,325 个 funding 文件，共 503,589 行日频 panel。Universe 每天用当时的 30d trailing quote_volume 排 TopN，没有用今天的 24h ticker 排名去筛过去。月度归档当前覆盖到 2026-04-30，所以本轮长历史主报告截至 2026-04-30。
因果性与数据口径审计
| 检查项 | 结论 | 说明 |
|---|
| Universe | PASS | daily TopN selected by that day's 30d trailing quote_volume across archive symbols; no current 24h ticker ranking |
| Listing age | PASS_WITH_LIMITATION | uses first observed archive kline as listing proxy; avoids current exchangeInfo but may differ from official onboardDate by a few days |
| Momentum | PASS | 10d pct_change through signal close |
| Breakout | PASS | 20d high distance through signal close |
| Carry | PASS | last settled funding record by signal date; PnL uses next day's realized funding sum |
| Execution timing | PASS | rebalance at close D, earn close-to-close return and realized funding on D+1 |
| Survivorship | IMPROVED_NOT_PERFECT | uses Binance archive symbol directories including delisted monthly data when present; still limited to data.binance.vision coverage |
年度表现：regime 稳定性不够
| 年份 | 累计 | 年化 | 最大回撤 | Sharpe | 手续费$ | 日均换手 |
|---|
| 2021 | 9.65% | 15.33% | -12.23% | 0.95 | 173.2532046627456 | 0.14 |
| 2022 | -36.53% | -36.63% | -40.78% | -1.54 | 1038.2013558304134 | 0.68 |
| 2023 | -3.77% | -3.78% | -30.30% | -0.07 | 1204.505425705984 | 0.68 |
| 2024 | -32.25% | -32.27% | -34.44% | -1.53 | 1052.5643982861743 | 0.74 |
| 2025 | 2.45% | 2.46% | -34.56% | 0.28 | 1551.3475657923157 | 0.80 |
| 2026 | 7.53% | 24.96% | -18.34% | 0.75 | 494.8453978007588 | 0.80 |
2022 和 2024 是核心反证：两个完整年份分别约 -36.5% 和 -32.3%。2025/2026 的正收益不足以抵消历史大回撤。
成本敏感性：0 成本也不够强
| Cost bps/side | 累计 | 年化 | 最大回撤 | Sharpe | 日均换手 |
|---|
| 0.0 | -10.40% | -2.18% | -49.32% | 0.07 | 0.65 |
| 5.0 | -50.45% | -13.17% | -66.55% | -0.32 | 0.65 |
| 10.0 | -72.61% | -22.92% | -79.91% | -0.70 | 0.65 |
| 20.0 | -91.63% | -39.27% | -92.88% | -1.48 | 0.65 |
| 30.0 | -97.45% | -52.16% | -97.70% | -2.25 | 0.65 |
最关键的是 0 bps 仍然 -10.4%。如果裸 alpha 都不是明显正的，那么真实成本只会把问题放大。
参数范围：不是一个窄参数最优的问题
| Universe | 累计 | 年化 | 最大回撤 | Sharpe | 中位持仓数 |
|---|
| 15.0 | 6.41% | 1.26% | -54.10% | 0.19 | 14.0 |
| 20.0 | -43.06% | -10.70% | -73.32% | -0.17 | 24.0 |
| 30.0 | -50.45% | -13.17% | -66.55% | -0.32 | 32.0 |
| 40.0 | -61.37% | -17.40% | -69.84% | -0.61 | 42.0 |
| 50.0 | -67.26% | -20.11% | -71.70% | -0.89 | 52.0 |
| 75.0 | -60.03% | -16.83% | -69.01% | -0.92 | 70.0 |
| Carry | Breakout | 累计 | 年化 | 最大回撤 | Sharpe | 日均换手 |
|---|
| 0.0 | 0.8 | -2.44% | -0.50% | -49.72% | 0.11 | 0.32 |
| 0.1 | 0.7000000000000001 | -10.20% | -2.14% | -52.16% | 0.05 | 0.33 |
| 0.2 | 0.6000000000000001 | -6.12% | -1.26% | -52.98% | 0.09 | 0.40 |
| 0.35 | 0.45 | -33.06% | -7.75% | -60.68% | -0.14 | 0.53 |
| 0.5 | 0.3 | -50.56% | -13.20% | -66.62% | -0.32 | 0.65 |
| 0.65 | 0.15 | -64.90% | -18.98% | -73.04% | -0.56 | 0.69 |
| 0.8 | 0.0 | -62.90% | -18.07% | -68.89% | -0.54 | 0.70 |
Universe=15 在长历史里勉强为正，但回撤仍 -54.1%，且这属于另一个更集中、更高风险的策略形态；carry 权重降低能改善，但没有把策略变成稳健 alpha。
滚动窗口稳定性
| 窗口 | 样本数 | 正收益比例 | 中位收益 | 最差 | 最好 |
|---|
| 180d | 55 | 29.09% | -7.69% | -31.00% | 25.52% |
| 365d | 25 | 16.00% | -21.09% | -39.90% | 23.96% |
最差 365d 窗口
| 开始 | 结束 | 收益 | 回撤 | Sharpe |
|---|
| 2022-01-05 | 2023-01-04 | -39.90% | -42.11% | -1.75 |
| 2023-12-26 | 2024-12-24 | -30.92% | -34.44% | -1.41 |
| 2023-04-30 | 2024-04-28 | -30.63% | -30.63% | -1.60 |
| 2024-02-24 | 2025-02-22 | -29.66% | -36.76% | -1.60 |
| 2022-07-04 | 2023-07-03 | -26.55% | -31.00% | -1.23 |
| 2023-06-29 | 2024-06-27 | -26.04% | -29.76% | -1.21 |
| 2022-03-06 | 2023-03-05 | -25.49% | -41.83% | -0.95 |
| 2022-05-05 | 2023-05-04 | -25.02% | -40.71% | -1.00 |
最好 365d 窗口
| 开始 | 结束 | 收益 | 回撤 | Sharpe |
|---|
| 2025-04-19 | 2026-04-18 | 23.96% | -34.54% | 0.62 |
| 2024-06-23 | 2025-06-22 | 19.75% | -23.84% | 0.58 |
| 2024-08-22 | 2025-08-21 | 7.73% | -23.84% | 0.37 |
| 2021-05-03 | 2022-05-09 | 1.45% | -15.53% | 0.17 |
| 2025-02-18 | 2026-02-17 | -3.65% | -34.54% | 0.11 |
| 2024-10-21 | 2025-10-20 | -4.94% | -25.14% | 0.06 |
| 2022-12-31 | 2023-12-30 | -5.35% | -30.30% | -0.17 |
| 2021-07-04 | 2022-07-08 | -6.21% | -21.88% | -0.16 |
我的判断
这个验证基本推翻了“rank154 是强候选”的短样本结论。它更像一个在近期市场阶段有效的 trend/carry 混合暴露，而不是跨 3-5 年稳定存在的结构性套利。下一步如果继续做，不建议直接优化原参数；更合理的是拆开做因子归因：单独测 carry、momo、breakout 的 long/short leg、按牛熊/高低 funding regime 分桶，再决定是否保留某个 sleeve。