← 返回首页

Scout · Crypto Pairs Stat-Arb 15m · Rank 4b Reframe

生成时间:2026-03-16 18:50 UTC | 这是原 Rank 4 之后的窄重开:不翻案原 park,只回答 rolling-beta / stricter entry 是否足够把它推进到下一刀轻量稳定性。

一句话结论

Rank 4b rolling-beta clean replication 已把部分 pair 从原先显著负值收回到轻微正 pocket,但最新时间切片把 surviving pairs 一起压回负值,因此当前更诚实 verdict 是 park。

这里仍不是 paper candidate verdict。它只是在判断:Rank 4b 这条窄重开,是否已经从“直接 park”抬升到“值得补一刀 Light Stability Pack”。

Frozen clean-room spec v2

itemvalue
sourceRank 4b reframe from Tadi et al. (2021/2023) + repo-inspired stat-arb template
scope15m crypto pairs on cached BTC/ETH/SOL history; same 3 pairs as Rank 4 baseline
what_changed_vs_rank4Freeze narrow reframe: rolling-beta z-score spread instead of frozen-beta; stricter entry_z = 2.5; no extra regime gate in the formal v2 spec
trade_onAfter 60% train split, use prior-bar rolling z-score; if >= +2.5 short spread, if <= -2.5 long spread; enter on next bar open
trade_offExit on next bar open when prior-bar z-score mean-reverts back through 0.0, or max hold 32 bars
beta_modelRolling beta via 192-bar covariance / variance; weights frozen per trade at entry beta
positioningDollar-neutral proxy using 1/(1+|beta|) and |beta|/(1+|beta|) leg weights
cost_model6.0 bps per side per leg; roundtrip charged as 4 legs = 0.0024
lookahead_guardSignals use prior-bar rolling z-score only; entries/exits execute on next bar open
repaint_guardNo future extrema / no revised labels; only cached OHLCV closes and next open execution
why_no_extra_gate_yetThe reframe first asks whether model calibration alone can rescue the pocket; regime/vol gates stay second-order until this clean replication v2 passes
next_light_check_if_survivesTime stability + cost/trade-count stability + cross-pair stability before any paper-candidate discussion

Rank 4b pair summary

pairtrain_rowstest_rowstrain_log_price_corrrolling_beta_windowtrain_beta_medianentry_zexit_zmax_hold_barstrade_countwin_ratemean_net_returncumulative_net_returnavg_hold_barspair_verdict
ETH-USD/SOL-USD6918.004613.0071.98%1920.7552.500.0032.0020.0050.00%0.11%2.28%32.0one_more_light_check
BTC-USD/SOL-USD6918.004613.0081.24%1920.5362.500.0032.0015.0040.00%0.05%0.74%29.5one_more_light_check
BTC-USD/ETH-USD6918.004613.0091.15%1920.6332.500.0032.0021.0028.57%-0.34%-6.99%30.7clean_replication_complete_but_negative

当前读法应优先看:是不是至少有不止一组 pair 从明显负 pocket 被抬回到轻微正 pocket,以及交易数有没有薄到失真。

Rank 4 vs Rank 4b compare

pairrank4_trade_countrank4_cumulative_net_returnrank4_verdictrank4b_trade_countrank4b_cumulative_net_returnrank4b_verdictcum_net_deltatrade_count_delta
ETH-USD/SOL-USD127-27.77%clean_replication_complete_but_negative202.28%one_more_light_check30.06%-107
BTC-USD/SOL-USD117-22.91%clean_replication_complete_but_negative150.74%one_more_light_check23.64%-102
BTC-USD/ETH-USD83-12.42%clean_replication_complete_but_negative21-6.99%clean_replication_complete_but_negative5.43%-62

这张对照不是为了证明 stat-arb 已经成立,而是为了回答:窄 reframe 后,结果有没有从“全线负值”抬回到“值得补下一刀”的状态。

Light Stability Pack · 时间稳定性(决策刀)

pairwindow_kindwindow_labeltrade_countwin_ratemean_net_returncumulative_net_returnwindow_verdict
BTC-USD/ETH-USDtime_terciletercile_1714.29%-0.58%-4.02%negative_or_flat
BTC-USD/ETH-USDtime_terciletercile_2728.57%-0.28%-1.92%negative_or_flat
BTC-USD/ETH-USDtime_terciletercile_3742.86%-0.17%-1.19%negative_or_flat
BTC-USD/ETH-USDcalendar_month2026-01333.33%-0.51%-1.52%negative_or_flat
BTC-USD/ETH-USDcalendar_month2026-02922.22%-0.35%-3.13%negative_or_flat
BTC-USD/ETH-USDcalendar_month2026-03933.33%-0.28%-2.50%negative_or_flat
BTC-USD/SOL-USDtime_terciletercile_1560.00%0.24%1.17%positive
BTC-USD/SOL-USDtime_terciletercile_2540.00%0.13%0.62%positive
BTC-USD/SOL-USDtime_terciletercile_3520.00%-0.21%-1.04%negative_or_flat
BTC-USD/SOL-USDcalendar_month2026-01366.67%0.16%0.49%positive
BTC-USD/SOL-USDcalendar_month2026-02650.00%0.25%1.46%positive
BTC-USD/SOL-USDcalendar_month2026-03616.67%-0.20%-1.20%negative_or_flat
ETH-USD/SOL-USDtime_terciletercile_1771.43%0.22%1.51%positive
ETH-USD/SOL-USDtime_terciletercile_2666.67%0.39%2.38%positive
ETH-USD/SOL-USDtime_terciletercile_3714.29%-0.23%-1.58%negative_or_flat
ETH-USD/SOL-USDcalendar_month2026-01366.67%0.34%1.01%positive
ETH-USD/SOL-USDcalendar_month2026-021258.33%0.17%1.99%positive
ETH-USD/SOL-USDcalendar_month2026-03520.00%-0.14%-0.72%negative_or_flat

这轮只补一刀时间稳定性,而且直接拿它做 promote / park 二选一:如果 surviving positive pairs 在最近 tercile 与最近月份都一起转负,就说明 pocket 主要来自前段样本,不够诚实地升 paper candidate。

Trade preview

pairsideentry_time_utcexit_time_utcentry_betaentry_zscoreexit_signal_zscorehold_barsleft_weightright_weightleft_leg_returnright_leg_returngross_returnnet_returnroundtrip_cost
ETH-USD/SOL-USDlong_spread2026-01-27 13:00 UTC2026-01-27 21:00 UTC0.795-2.59-2.973255.71%44.29%3.51%2.80%0.72%0.48%0.24%
ETH-USD/SOL-USDlong_spread2026-01-27 21:15 UTC2026-01-28 05:15 UTC0.952-3.08-2.263251.23%48.77%-0.79%-0.85%0.01%-0.23%0.24%
BTC-USD/SOL-USDlong_spread2026-01-28 07:45 UTC2026-01-28 15:45 UTC0.460-2.54-3.213268.50%31.50%0.28%-0.84%0.46%0.22%0.24%
BTC-USD/SOL-USDlong_spread2026-01-28 16:00 UTC2026-01-29 00:00 UTC0.601-3.11-1.453262.45%37.55%-0.20%-0.57%0.09%-0.15%0.24%
BTC-USD/ETH-USDlong_spread2026-01-29 10:15 UTC2026-01-29 18:15 UTC0.588-2.50-2.813262.97%37.03%-3.88%-3.96%-0.98%-1.22%0.24%
BTC-USD/ETH-USDlong_spread2026-01-29 18:30 UTC2026-01-30 02:30 UTC0.766-2.83-2.003256.61%43.39%-2.88%-3.13%-0.27%-0.51%0.24%
BTC-USD/ETH-USDshort_spread2026-01-31 10:30 UTC2026-01-31 18:30 UTC0.5202.531.963265.78%34.22%-5.56%-9.38%0.45%0.21%0.24%
ETH-USD/SOL-USDshort_spread2026-01-31 14:15 UTC2026-01-31 22:15 UTC0.6752.710.603259.70%40.30%-7.55%-8.70%1.00%0.76%0.24%
BTC-USD/SOL-USDshort_spread2026-01-31 15:30 UTC2026-01-31 23:30 UTC0.5472.531.923264.65%35.35%-2.88%-3.38%0.66%0.42%0.24%
ETH-USD/SOL-USDlong_spread2026-02-02 08:45 UTC2026-02-02 16:45 UTC1.099-2.51-2.903247.63%52.37%5.58%3.52%0.81%0.57%0.24%
BTC-USD/SOL-USDlong_spread2026-02-02 14:00 UTC2026-02-02 22:00 UTC0.657-2.64-1.663260.34%39.66%0.39%1.46%-0.34%-0.58%0.24%
ETH-USD/SOL-USDlong_spread2026-02-02 17:00 UTC2026-02-03 01:00 UTC1.422-2.76-1.313241.29%58.71%-0.20%-0.79%0.38%0.14%0.24%

这里只展示前 12 笔 trade preview;完整表见 reports/artifacts/scout_crypto_pairs_stat_arb_15m_rank4b/trades.csv

边界与下一步