Generated: 2026-05-10 03:40 UTC
Rank154 Hub
Rank154b young funding 归档审计
180-365d young coin · funding crowding continuation · daily vs 5d staggered · cost sensitivity
Verdict
ARCHIVED
NO-GO · no paper lane
Core total return
-4.3%
20bps, 5d staggered
Annualized
-0.9%
2021-05-03 → 2026-04-30
Avg turnover
18.2%
日均换手，5d staggered 后
0 cost edge
+85.7%
有 gross edge，但不够厚
20bps cost drag
-66.3%
累计 turnover × cost
最终收口：154b 已归档为 research lead only，不进入 paper lane。它的经济故事（young coin 的 funding crowding/attention continuation）有一点 gross price edge，但在 funding 支付和 20bps turnover 成本下已经变成负净 alpha；2024/2025 OOS 连亏，2026 前四个月的暴拉把总曲线救回接近零。这不是可以继续调参推进的 release candidate。
1. 策略定义
- Universe：Binance USDT perp archive panel，按每个日期过滤上市年龄
180-365d，再取当日 30d trailing quote volume Top30。
- Alpha 因子：
carry_raw = funding_rate_last。154b 当前只关注这一个核心 alpha 因子；没有 momentum、breakout、volume quality 等第二因子参与打分。
- Signal：高 funding 做多、低 funding 做空。这个方向测试的是 young coin 的注意力/拥挤延续，而不是收 funding carry。
- Portfolio：高 funding 前 20% 做多，低 funding 后 20% 做空；核心版 5-day staggered，每天开 1/5 sleeve，目标 gross 约 65%。
- PnL：close-to-close price return + realized next-day funding；正 funding 下，long 付费、short 收费。
- Cost：核心版按 20bps × turnover 粗扣，另跑 0/10/20/30/50bps 敏感性。
1b. Funding 单因子 IC / ICIR
| horizon | target | days | ic_mean | ic_median | icir_daily | icir_ann_sqrt365 | t_stat | positive_rate |
|---|
| 1 | long_total | 1427 | -0.0025 | -0.0111 | -0.011 | -0.22 | -0.43 | 48.3% |
| 1 | price | 1427 | 0.0149 | 0.0083 | 0.069 | 1.31 | 2.60 | 51.7% |
| 5 | long_total | 1423 | -0.0089 | -0.0134 | -0.042 | -0.80 | -1.58 | 47.0% |
| 5 | price | 1423 | 0.0195 | 0.0170 | 0.091 | 1.75 | 3.45 | 53.8% |
| 10 | long_total | 1418 | -0.0125 | -0.0148 | -0.060 | -1.15 | -2.26 | 46.8% |
| 10 | price | 1418 | 0.0230 | 0.0178 | 0.111 | 2.11 | 4.16 | 53.5% |
解释：price target 只看未来价格；long_total target = 未来价格收益 - 未来 funding 成本，代表高 funding 做多后真正拿到的方向收益。154b 的 price IC 为正，但 long_total IC 转负，说明价格延续信号被 funding 成本抵消。
1c. Top 20% vs Bottom 20% spread
| horizon | target | spread_mean | spread_median | spread_t_stat | positive_rate | top_ret_mean | bottom_ret_mean |
|---|
| 1 | long_total | 0.00% | -0.06% | 0.03 | 48.8% | 0.04% | 0.03% |
| 1 | price | 0.29% | 0.09% | 2.19 | 51.3% | 0.08% | -0.20% |
| 5 | long_total | 0.26% | -0.06% | 0.76 | 49.8% | 0.26% | 0.00% |
| 5 | price | 1.25% | 0.73% | 3.69 | 53.7% | 0.42% | -0.83% |
| 10 | long_total | -0.81% | -0.22% | -1.13 | 48.9% | -0.12% | 0.69% |
| 10 | price | 0.77% | 0.59% | 1.10 | 52.9% | 0.11% | -0.66% |
1d. 5d IC 年度稳定性
| year | target | days | ic_mean | icir_daily | positive_rate |
|---|
| 2021 | long_total | 62 | -0.0183 | -0.108 | 45.2% |
| 2021 | price | 62 | 0.0250 | 0.143 | 58.1% |
| 2022 | long_total | 361 | -0.0030 | -0.013 | 47.1% |
| 2022 | price | 361 | 0.0159 | 0.070 | 52.9% |
| 2023 | long_total | 154 | 0.0237 | 0.120 | 53.2% |
| 2023 | price | 154 | 0.0348 | 0.176 | 56.5% |
| 2024 | long_total | 366 | -0.0113 | -0.058 | 44.5% |
| 2024 | price | 366 | 0.0046 | 0.024 | 48.6% |
| 2025 | long_total | 365 | -0.0373 | -0.164 | 44.1% |
| 2025 | price | 365 | 0.0010 | 0.005 | 51.8% |
| 2026 | long_total | 115 | 0.0313 | 0.152 | 56.5% |
| 2026 | price | 115 | 0.1129 | 0.568 | 73.0% |
5d price IC 的主要强度来自 2026：2024/2025 的 price IC 分别只有约 0.0046 / 0.0010；long_total 在 2024/2025 为负。这进一步支持“regime burst，不是稳定可交易 alpha”的收口。
2. Core equity curve
3. 成本敏感性：edge 不够厚
| cost_bps | return | ann_return | max_dd | sharpe | avg_turnover | commission_return_sum |
|---|
| 0 bps | 85.7% | 13.2% | -51.1% | 0.55 | 18.2% | 0.0% |
| 10 bps | 33.3% | 5.9% | -56.1% | 0.34 | 18.2% | -33.1% |
| 20 bps | -4.3% | -0.9% | -63.1% | 0.14 | 18.2% | -66.3% |
| 30 bps | -31.3% | -7.3% | -71.7% | -0.07 | 18.2% | -99.4% |
| 50 bps | -64.6% | -18.8% | -84.0% | -0.49 | 18.2% | -165.7% |
0 cost 下总收益 +85.7%，10bps 下 +33.3%，20bps 下变成 -4.3%。这说明问题不是完全没有信号，而是 gross edge 太薄，无法覆盖真实执行成本。
4. 年度拆解：OOS 不稳，2026 单段救场
| period | return | max_dd | avg_turnover | price_return_sum | funding_return_sum | commission_return_sum | regime_note |
|---|
| 2021 | 9.7% | -6.9% | 5.1% | 16.1% | -3.9% | -2.4% | 早期正收益 |
| 2022 | -15.4% | -31.8% | 24.9% | 14.9% | -12.2% | -18.2% | 熊市/高换手，亏 |
| 2023 | 7.3% | -7.1% | 10.8% | 17.6% | -2.0% | -7.9% | 小正 |
| 2024 | -26.8% | -36.9% | 22.6% | -2.1% | -11.3% | -16.6% | OOS 亏 |
| 2025 | -27.2% | -47.6% | 22.2% | 41.9% | -41.8% | -16.2% | 价格腿强但 funding+cost 抹平 |
| 2026 | 84.7% | -14.3% | 21.2% | 107.0% | -36.3% | -5.1% | 前 4 个月暴拉，结果被它救回来 |
5. Train / OOS 粗拆
| split | return | positive_years | note |
|---|
| Train-ish 2021-2023 | -0.4% | 2/3 | 基本持平，不是强训练样本 |
| OOS-ish 2024-2026 | -1.6% | 1/3 | 2024/2025 连续亏，2026 单段救场 |
6. Rolling 365d 稳定性
| windows | positive_windows | positive_rate | median_return | min_return | max_return | median_sharpe |
|---|
| 25.0 | 10.0 | 40.0% | -11.0% | -49.7% | 23.1% | -0.39 |
Rolling 365d 只有 10/25 个窗口为正，中位窗口收益 -11.0%，最差窗口 -49.7%。这不是“稳定小 edge”，更像 regime-dependent burst。
7. 变体对照
| variant | return | ann_return | max_dd | sharpe | avg_turnover | price_return_sum | funding_return_sum | commission_return_sum |
|---|
| long short staggered h5 cost0 | 85.7% | 13.2% | -51.1% | 0.55 | 18.2% | 195.4% | -107.6% | 0.0% |
| long short staggered h5 cost10 | 33.3% | 5.9% | -56.1% | 0.34 | 18.2% | 195.4% | -107.6% | -33.1% |
| long short daily h1 cost0 | 30.6% | 5.5% | -57.9% | 0.33 | 73.0% | 215.1% | -144.0% | 0.0% |
| young180 365 top50 long short staggered h5 cost20 | 23.6% | 4.3% | -44.6% | 0.30 | 18.2% | 191.2% | -89.1% | -66.2% |
| long only staggered h5 cost0 | -70.8% | -21.9% | -95.5% | 0.17 | 17.1% | 123.3% | -50.3% | 0.0% |
| long short staggered h5 cost20 | -4.3% | -0.9% | -63.1% | 0.14 | 18.2% | 195.4% | -107.6% | -66.3% |
| long only staggered h5 cost10 | -78.6% | -26.6% | -96.4% | 0.09 | 17.1% | 123.3% | -50.3% | -31.1% |
| long only daily h1 cost0 | -85.6% | -32.3% | -96.4% | 0.08 | 69.1% | 110.4% | -68.1% | 0.0% |
| long only staggered h5 cost20 | -84.3% | -31.1% | -97.2% | 0.02 | 17.1% | 123.3% | -50.3% | -62.3% |
| long only staggered h5 cost30 | -88.5% | -35.3% | -97.7% | -0.05 | 17.1% | 123.3% | -50.3% | -93.4% |
| long short staggered h5 cost30 | -31.3% | -7.3% | -71.7% | -0.07 | 18.2% | 195.4% | -107.6% | -99.4% |
| long only daily h1 cost10 | -95.9% | -47.4% | -98.6% | -0.17 | 69.1% | 110.4% | -68.1% | -125.5% |
Top50 版本在 20bps 下为正（+23.6%），但这更像降低波动/特定样本改善，不足以改变结论；需要单独做 IC、事件归因和近期 paper shadow 才能考虑作为下一条线索。
8. Causality / honesty audit
| item | status | detail |
|---|
| Universe | PASS | 每个交易日用 archive panel，listing_days 180-365，按当日 trailing 30d quote volume 取 Top30。 |
| Listing age | PASS_WITH_LIMITATION | 用首次 archive daily kline 近似上市日；因果但不是官方 onboardDate。 |
| Signal | PASS | signal date 已结算 funding_rate_last；高 funding 做多，低 funding 做空。 |
| Execution | PASS | D 日 close 后调仓，吃 D→D+1 close-to-close price return 和 D+1 realized funding。 |
| Cost | ROUGH | 按 turnover × bps 粗扣；尚未做订单簿/冲击成本模拟。 |
| Robustness | FAIL | 20bps 核心版净收益为负；rolling 365d 仅 10/25 个窗口为正。 |
9. Archive rule / 产物
Archive rule: 不要用 TopN / holding period / sleeve fraction 继续救 154b。任何 future funding-age work 必须新建 rank/name，并先声明 regime、after-funding target 与真实成本口径。
scripts/backtest_rank154b_young_funding.pyreports/artifacts/rank154b_young_funding_backtest/rank154b_backtest_stats.csvreports/artifacts/rank154b_young_funding_backtest/rank154b_backtest_yearly.csvreports/artifacts/rank154b_young_funding_backtest/rank154b_backtest_rolling365.csvreports/artifacts/rank154b_young_funding_backtest/154b_long_short_staggered_h5_cost20_equity.csv