Rank213 monthly_volume_universe_rebuild

目标:把静态 frozen30 改成每月月初按上一完整自然月 Binance UM perp quote-volume(USDT 成交额)重建的 causal top30 热门 universe,其余参数(15m / 64 / 12 / veto / gate / 4bps)全部冻结不变。

formal_strategy_review · universe_selection_audit · runner

重要限制

本页按月初以前的上一完整自然月 Binance UM perpetual 1d quote_volume 求和,作为 causal liquidity / 热门度 proxy 来重建每月 top30 universe。它不再依赖 CoinGecko market-cap proxy,也避免了 1000*/1000000* 合约 multiplier 对市值近似的扭曲;但它仍是 volume-proxy 近似,而不是原始研究当时保存下来的官方选池快照。

与 frozen30 的同口径对比

versionstrategyrebalancesnet_mean_bpsnet_cum_pctmax_drawdown_pctwin_rate_pctavg_turnover_xgate_on_rate_pct
frozen30plain baseline18087-1.13 bps-95.12%-99.21%46.44%1.000x
frozen30baseline+veto18087-0.84 bps-89.98%-99.06%46.04%1.066x
frozen30baseline+veto+gate180870.45 bps100.72%-32.01%2.54%0.065x
monthly_volume_rebuildplain baseline18087-1.58 bps-98.09%-99.62%46.66%1.000x100.00%
monthly_volume_rebuildbaseline+veto18087-2.53 bps-99.60%-99.79%46.01%1.074x100.00%
monthly_volume_rebuildbaseline+veto+gate18087-0.23 bps-37.13%-48.63%1.61%0.042x3.54%

compare csv: reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_monthly_volume_universe_rebuild_compare.csv

每月 universe 变动摘要

monthselected_countentered_countexited_countoverlap_with_frozen_countselected_not_in_frozen_countfrozen_not_in_selected_count
2020-02101006424
2020-032313091421
2020-04230091421
2020-05230091421
2020-06241091521
2020-072950101920
2020-083010111919
2020-093055111919
2020-103055102020
2020-113044121818
2020-123044102020
2021-013066102020
2021-023055121818
2021-033044121818
2021-043077121818
2021-053077131717
2021-063055131717
2021-073044121818
2021-083077111919
2021-093055121818
2021-103077121818
2021-113088121818
2021-123088111919
2022-013077111919
2022-023044111919
2022-033033111919
2022-043099131717
2022-053044121818
2022-063066111919
2022-073055102020
2022-083077121818
2022-093044121818
2022-103055121818
2022-113055111919
2022-123044111919
2023-013055111919
2023-023055121818
2023-033044111919
2023-043077111919
2023-053055111919
2023-063066121818
2023-073077131717
2023-083066131717
2023-09301111121818
2023-103088121818
2023-113088131717
2023-12301010111919
2024-01301010121818
2024-02301111111919
2024-03301111141616
2024-04301212151515
2024-053099141616
2024-063066141616
2024-073099111919
2024-083099131717
2024-09301111131717
2024-10301010121818
2024-11301010111919
2024-12301212131717
2025-013099151515
2025-023099131717
2025-033077131717
2025-043088141616
2025-05301010151515
2025-06301111131717
2025-073099141616
2025-08301010141616
2025-093077151515
2025-10301010141616
2025-113099161414
2025-12301010171313
2026-01301212161414
2026-02301111181212
2026-033099191111
2026-0430882288

full monthly universe csv: reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_monthly_volume_universe_rebuild_monthly_universe.csv

最终结论

在按月 volume/top30 热门度重建的 causal universe 下,gate 线已失去正 net_mean,说明 frozen30 版本的强结果有较大概率受静态选池/幸存者偏差放大。