age90_14d_skip1d_voladj Phase 3 严肃验证包
目标：判断它是否有资格进入 paper/live 候选，而不是继续扩展新策略。
生成时间：2026-05-06 03:34 UTC · 样本：2020-02-01 → 2026-04-13 · 15m symbol-month 覆盖：98.99%
结论
当前结论：继续研究，但仍不能进 live。 原 4bps close-to-close 为 455.65%，最大回撤 -60.08%；但执行成本和 2022-2023 弱段仍是硬门槛。
执行层：4h TWAP 在 12bps 下为 272.55% / DD -63.68%。如果真实执行接近 12bps，策略安全边际明显变薄。
风控层：prior_60d_return_positive 在 4bps 下为 382.29% / DD -33.68%，但 walk-forward 选择结果必须优先看下方逐年表，不能只看全样本。
结构层：收益主要来自 long 侧；full-cap short-only 为 -96.97% / DD -99.52%，所以“做空弱币”不能单独构成可交易理由。
1. 执行级验证：15m TWAP/VWAP + 成本
同一批 daily 选股，不重新打分；表里同时放 close_to_close_reference、signal_day_open_to_next_open、delayed_next_open_to_following_open 和 15m TWAP/VWAP。因为 score 明确跳过最近 1 天，只用 t-15d → t-1d，所以 UTC 00:00 的 signal-day 进场是 causal，不是拿当天 close 偷看。15m 窗口从 UTC 00:00 开始，30m/1h/4h 分别使用 2/4/16 根 15m bar。成本为 per-basket 扣减。
| scenario | cost_bps_per_basket | avg_coverage_pct | trading_baskets | net_mean_bps | net_cum_pct | max_drawdown_pct | win_rate_pct |
|---|---|---|---|---|---|---|---|
| close_to_close_reference | 4 | 100.00% | 2263 | 10.92 bps | 455.65% | -60.08% | 49.05% |
| close_to_close_reference | 8 | 100.00% | 2263 | 6.92 bps | 124.78% | -75.49% | 47.95% |
| close_to_close_reference | 12 | 100.00% | 2263 | 2.92 bps | -9.10% | -84.97% | 46.93% |
| close_to_close_reference | 16 | 100.00% | 2263 | -1.08 bps | -63.25% | -90.79% | 46.18% |
| close_to_close_reference | 20 | 100.00% | 2263 | -5.08 bps | -85.15% | -94.66% | 45.29% |
| signal_day_open_to_next_open | 4 | 100.00% | 2263 | 18.56 bps | 2983.62% | -43.73% | 50.60% |
| signal_day_open_to_next_open | 8 | 100.00% | 2263 | 14.56 bps | 1148.31% | -46.72% | 49.71% |
| signal_day_open_to_next_open | 12 | 100.00% | 2263 | 10.56 bps | 405.16% | -58.25% | 48.78% |
| signal_day_open_to_next_open | 16 | 100.00% | 2263 | 6.56 bps | 104.35% | -70.98% | 47.90% |
| signal_day_open_to_next_open | 20 | 100.00% | 2263 | 2.56 bps | -17.36% | -81.52% | 46.66% |
| delayed_next_open_to_following_open | 4 | 99.98% | 2263 | 10.99 bps | 464.11% | -59.34% | 48.96% |
| delayed_next_open_to_following_open | 8 | 99.98% | 2263 | 6.99 bps | 128.21% | -75.02% | 47.86% |
| delayed_next_open_to_following_open | 12 | 99.98% | 2263 | 2.99 bps | -7.71% | -84.69% | 46.84% |
| delayed_next_open_to_following_open | 16 | 99.98% | 2263 | -1.01 bps | -62.69% | -90.61% | 46.00% |
| delayed_next_open_to_following_open | 20 | 99.98% | 2263 | -5.01 bps | -84.92% | -94.59% | 45.21% |
| twap_240m | 4 | 99.44% | 2250 | 16.84 bps | 2151.05% | -48.02% | 50.71% |
| twap_240m | 8 | 99.44% | 2250 | 12.86 bps | 815.94% | -50.56% | 49.82% |
| twap_240m | 12 | 99.44% | 2250 | 8.88 bps | 272.55% | -63.68% | 49.11% |
| twap_240m | 16 | 99.44% | 2250 | 4.91 bps | 51.48% | -74.77% | 48.18% |
| twap_240m | 20 | 99.44% | 2250 | 0.93 bps | -38.43% | -83.63% | 46.76% |
| twap_30m | 4 | 99.44% | 2250 | 17.50 bps | 2384.46% | -46.38% | 50.04% |
| twap_30m | 8 | 99.44% | 2250 | 13.52 bps | 910.93% | -49.07% | 49.29% |
| twap_30m | 12 | 99.44% | 2250 | 9.54 bps | 311.20% | -58.22% | 48.53% |
| twap_30m | 16 | 99.44% | 2250 | 5.57 bps | 67.20% | -70.95% | 47.51% |
| twap_30m | 20 | 99.44% | 2250 | 1.59 bps | -32.04% | -81.34% | 46.44% |
| twap_60m | 4 | 99.44% | 2250 | 17.21 bps | 2260.56% | -46.65% | 51.07% |
| twap_60m | 8 | 99.44% | 2250 | 13.23 bps | 860.50% | -49.32% | 50.27% |
| twap_60m | 12 | 99.44% | 2250 | 9.26 bps | 290.68% | -59.30% | 49.38% |
| twap_60m | 16 | 99.44% | 2250 | 5.28 bps | 58.85% | -71.70% | 48.36% |
| twap_60m | 20 | 99.44% | 2250 | 1.30 bps | -35.43% | -81.39% | 47.42% |
| vwap_240m | 4 | 99.44% | 2250 | 15.14 bps | 1414.11% | -50.06% | 50.98% |
| vwap_240m | 8 | 99.44% | 2250 | 11.16 bps | 515.98% | -57.05% | 49.82% |
| vwap_240m | 12 | 99.44% | 2250 | 7.19 bps | 150.51% | -70.16% | 48.58% |
| vwap_240m | 16 | 99.44% | 2250 | 3.21 bps | 1.84% | -79.27% | 47.56% |
| vwap_240m | 20 | 99.44% | 2250 | -0.77 bps | -58.61% | -87.08% | 46.80% |
| vwap_30m | 4 | 99.44% | 2250 | 16.85 bps | 2050.28% | -47.15% | 49.82% |
| vwap_30m | 8 | 99.44% | 2250 | 12.87 bps | 774.90% | -49.57% | 48.89% |
| vwap_30m | 12 | 99.44% | 2250 | 8.90 bps | 255.85% | -61.51% | 47.82% |
| vwap_30m | 16 | 99.44% | 2250 | 4.92 bps | 44.68% | -73.25% | 46.98% |
| vwap_30m | 20 | 99.44% | 2250 | 0.94 bps | -41.20% | -83.23% | 46.22% |
| vwap_60m | 4 | 99.44% | 2250 | 16.23 bps | 1790.06% | -47.33% | 50.80% |
| vwap_60m | 8 | 99.44% | 2250 | 12.25 bps | 668.98% | -49.97% | 50.00% |
| vwap_60m | 12 | 99.44% | 2250 | 8.27 bps | 212.75% | -63.42% | 48.98% |
| vwap_60m | 16 | 99.44% | 2250 | 4.29 bps | 27.15% | -74.57% | 47.96% |
| vwap_60m | 20 | 99.44% | 2250 | 0.32 bps | -48.32% | -83.91% | 46.98% |
2. 换仓日成交额占比限制
容量估算：每条腿权重约 1/6，用 entry/exit 窗口较小 quote_volume 约束，计算在 1%/5%/10% 参与率下可承载的 basket notional。它不是滑点模型，只是容量红线。
| scenario | participation_pct | days | p10_capacity_usdt | median_capacity_usdt | p90_capacity_usdt | pct_days_capacity_ge_100k | pct_days_capacity_ge_500k | pct_days_capacity_ge_1m |
|---|---|---|---|---|---|---|---|---|
| twap_240m | 1.00% | 2252 | $52.6K | $441.1K | $1.21M | 82.7709 | 45.4707 | 16.6519 |
| twap_240m | 5.00% | 2252 | $263.2K | $2.21M | $6.07M | 96.1368 | 82.7709 | 71.4476 |
| twap_240m | 10.00% | 2252 | $526.4K | $4.41M | $12.14M | 98.5346 | 90.7638 | 82.7709 |
| twap_30m | 1.00% | 2252 | $7.4K | $65.4K | $183.5K | 32.4600 | 0.4885 | 0.0888 |
| twap_30m | 5.00% | 2252 | $37.1K | $326.8K | $917.3K | 78.4192 | 32.4600 | 7.6821 |
| twap_30m | 10.00% | 2252 | $74.2K | $653.6K | $1.83M | 86.8561 | 59.1918 | 32.4600 |
| twap_60m | 1.00% | 2252 | $14.3K | $124.2K | $346.5K | 57.5933 | 3.3304 | 0.4440 |
| twap_60m | 5.00% | 2252 | $71.4K | $620.9K | $1.73M | 86.5453 | 57.5933 | 31.3499 |
| twap_60m | 10.00% | 2252 | $142.8K | $1.24M | $3.46M | 92.6732 | 73.9787 | 57.5933 |
| vwap_240m | 1.00% | 2252 | $52.6K | $441.1K | $1.21M | 82.7709 | 45.4707 | 16.6519 |
| vwap_240m | 5.00% | 2252 | $263.2K | $2.21M | $6.07M | 96.1368 | 82.7709 | 71.4476 |
| vwap_240m | 10.00% | 2252 | $526.4K | $4.41M | $12.14M | 98.5346 | 90.7638 | 82.7709 |
| vwap_30m | 1.00% | 2252 | $7.4K | $65.4K | $183.5K | 32.4600 | 0.4885 | 0.0888 |
| vwap_30m | 5.00% | 2252 | $37.1K | $326.8K | $917.3K | 78.4192 | 32.4600 | 7.6821 |
| vwap_30m | 10.00% | 2252 | $74.2K | $653.6K | $1.83M | 86.8561 | 59.1918 | 32.4600 |
| vwap_60m | 1.00% | 2252 | $14.3K | $124.2K | $346.5K | 57.5933 | 3.3304 | 0.4440 |
| vwap_60m | 5.00% | 2252 | $71.4K | $620.9K | $1.73M | 86.5453 | 57.5933 | 31.3499 |
| vwap_60m | 10.00% | 2252 | $142.8K | $1.24M | $3.46M | 92.6732 | 73.9787 | 57.5933 |
3. 预注册 Gate 验证
只测试事先定义的少数 gate：策略自身 30/60 日 trailing return、30 日 > -5%、BTCÐ 60 日趋势。gate off 时空仓，收益记 0，避免“只统计开仓日”美化。
| gate | cost_bps_per_basket | active_rate_pct | trading_baskets | net_mean_bps | net_cum_pct | max_drawdown_pct | win_rate_pct |
|---|---|---|---|---|---|---|---|
| trade_all_reference | 4 | 100.00% | 2263 | 10.92 bps | 455.65% | -60.08% | 49.05% |
| trade_all_reference | 8 | 100.00% | 2263 | 6.92 bps | 124.78% | -75.49% | 47.95% |
| trade_all_reference | 12 | 100.00% | 2263 | 2.92 bps | -9.10% | -84.97% | 46.93% |
| trade_all_reference | 16 | 100.00% | 2263 | -1.08 bps | -63.25% | -90.79% | 46.18% |
| trade_all_reference | 20 | 100.00% | 2263 | -5.08 bps | -85.15% | -94.66% | 45.29% |
| prior_30d_return_positive | 4 | 53.25% | 1205 | 10.09 bps | 479.86% | -52.30% | 50.87% |
| prior_30d_return_positive | 8 | 53.25% | 1205 | 7.96 bps | 258.23% | -58.28% | 49.71% |
| prior_30d_return_positive | 12 | 53.25% | 1205 | 5.83 bps | 121.27% | -64.90% | 48.63% |
| prior_30d_return_positive | 16 | 53.25% | 1205 | 3.70 bps | 36.65% | -70.47% | 47.88% |
| prior_30d_return_positive | 20 | 53.25% | 1205 | 1.57 bps | -15.63% | -75.16% | 46.80% |
| prior_60d_return_positive | 4 | 53.69% | 1215 | 9.28 bps | 382.29% | -33.68% | 49.88% |
| prior_60d_return_positive | 8 | 53.69% | 1215 | 7.13 bps | 196.74% | -42.14% | 48.72% |
| prior_60d_return_positive | 12 | 53.69% | 1215 | 4.98 bps | 82.54% | -52.80% | 47.98% |
| prior_60d_return_positive | 16 | 53.69% | 1215 | 2.84 bps | 12.27% | -63.71% | 47.24% |
| prior_60d_return_positive | 20 | 53.69% | 1215 | 0.69 bps | -30.96% | -75.10% | 46.50% |
| prior_30d_return_above_minus5pct | 4 | 69.42% | 1571 | 10.83 bps | 528.55% | -46.01% | 50.10% |
| prior_30d_return_above_minus5pct | 8 | 69.42% | 1571 | 8.05 bps | 235.42% | -57.30% | 49.08% |
| prior_30d_return_above_minus5pct | 12 | 69.42% | 1571 | 5.28 bps | 78.95% | -67.51% | 48.12% |
| prior_30d_return_above_minus5pct | 16 | 69.42% | 1571 | 2.50 bps | -4.55% | -75.28% | 47.49% |
| prior_30d_return_above_minus5pct | 20 | 69.42% | 1571 | -0.28 bps | -49.10% | -82.32% | 46.53% |
| btc_eth_prior_60d_positive | 4 | 49.36% | 1117 | 6.90 bps | 229.40% | -45.21% | 50.40% |
| btc_eth_prior_60d_positive | 8 | 49.36% | 1117 | 4.93 bps | 110.76% | -47.50% | 49.33% |
| btc_eth_prior_60d_positive | 12 | 49.36% | 1117 | 2.95 bps | 34.83% | -50.14% | 48.08% |
| btc_eth_prior_60d_positive | 16 | 49.36% | 1117 | 0.98 bps | -13.77% | -59.35% | 47.18% |
| btc_eth_prior_60d_positive | 20 | 49.36% | 1117 | -1.00 bps | -44.85% | -70.12% | 46.02% |
4. Walk-forward Freeze
每个测试年只用前两年，从预注册 gate 中按 trailing 2Y Calmar 选一个，然后固定到下一年。这个表比全样本 gate 更接近“以后真的怎么选”。
| test_year | train_window | chosen_gate | trading_baskets | net_mean_bps | net_cum_pct | max_drawdown_pct | win_rate_pct |
|---|---|---|---|---|---|---|---|
| 2022 | 2020-2021 | btc_eth_prior_60d_positive | 57 | 4.75 bps | 17.87% | -9.83% | 52.63% |
| 2023 | 2021-2022 | btc_eth_prior_60d_positive | 237 | -2.74 bps | -13.35% | -38.25% | 45.15% |
| 2024 | 2022-2023 | prior_60d_return_positive | 169 | 8.32 bps | 30.78% | -31.52% | 52.66% |
| 2025 | 2023-2024 | prior_30d_return_positive | 237 | 18.01 bps | 78.76% | -16.60% | 50.63% |
| 2026 YTD | 2024-2025 | prior_30d_return_positive | 94 | 69.15 bps | 75.40% | -24.37% | 57.45% |
5. Long / Short 单边拆解
用原 close-to-close leg 重建收益，判断收益来自 long 强势币、short 弱势币，还是两边都有贡献。
| side_scenario | trading_baskets | net_mean_bps | net_cum_pct | max_drawdown_pct | win_rate_pct |
|---|---|---|---|---|---|
| dollar_neutral_original | 2263 | 10.92 bps | 455.65% | -60.08% | 49.05% |
| half_cap_long_contribution | 2263 | 13.53 bps | 700.12% | -67.79% | 51.44% |
| half_cap_short_contribution | 2263 | 1.40 bps | -38.41% | -84.36% | 49.14% |
| full_cap_long_only_minus_4bps | 2263 | 23.05 bps | 263.78% | -96.28% | 51.22% |
| full_cap_short_only_minus_4bps | 2263 | -1.21 bps | -96.97% | -99.52% | 48.74% |
保守读法
- 如果要进 paper，建议先只允许 very small notional，并强制使用 15m TWAP/VWAP 执行口径，不用日线 close 幻觉。
- 如果 12bps 后仍无法接受回撤，这条线应继续留在 research，不进入 live。
- 下一轮如果继续推进，应接真实成交模拟：订单簿/盘口深度、资金费率、逐笔换仓和残留仓位处理。