Svogun 2022 · Cost/Regime Experiment v1

这是基于现有 8 币 `60m_365d / 60m_730d` 本地缓存数据做的第一版 clean-room survival experiment：先不用论文全部规则族，只测试两个最小 baseline——ma_crossover 与 rolling_breakout_20——看 gross / net / regime 会不会明显重排排序。

data 8 crypto / 60m / 365d+730d rules MA crossover + breakout baseline cost gross / net_low(10bps) / net_high(30bps) hold 12 bars

这页回答什么？

在我们自己的数据上，cost 会不会显著改变 breakout / trend 规则的排序？
regime（这里先用 bubble_proxy）会不会重排结果？
这是否足以支持把 gross/net + regime 变成后续 breakout 主线的默认约束？

核心结论：在当前 8 币 60m 样本上，cost 的确会让 breakout / trend baseline 全部进一步变差，而 bubble_proxy 分层也会显著改变均值与胜率，所以 `gross/net + regime` 应当成为后续 breakout 研究的默认报告项，而不是可选附录。

证据：这轮用同一批本地缓存 bars，对 `ma_crossover` 与 `rolling_breakout_20` 同时跑了 `gross / net_low / net_high` 与 `bubble_proxy` 分层；无论在 `60m_365d` 还是 `60m_730d`，成本都会把均值进一步压低，而不同 regime 下的均值与胜率也出现明显分化。

实验口径

signal 在 bar close 触发；next bar open 入场；持有 12 bars 后 exit close 离场。
ma_crossover：EMA(12) 上穿 EMA(48)
rolling_breakout_20：close 上破前 20 根 close 高点
net_low = gross - 10bps；net_high = gross - 30bps（round-trip 近似）
bubble_proxy：价格在慢均线上方且 trend_strength / vol24 同时高于本 symbol 样本中位数

anchor view（60m_730d）

sample_key	rule	cost_case	trade_count	win_ratio	mean_return	median_return	iqr_return	positive_symbol_ratio
60m_730d	ma_crossover	gross	1712	48.07%	-0.08%	-0.10%	2.81%	25.00%
60m_730d	rolling_breakout_20	gross	8205	48.60%	0.12%	-0.07%	2.83%	75.00%
60m_730d	ma_crossover	net_high	1712	42.17%	-0.38%	-0.40%	2.81%	12.50%
60m_730d	rolling_breakout_20	net_high	8205	42.80%	-0.18%	-0.37%	2.83%	12.50%

overall summary

sample_key	rule	cost_case	trade_count	win_ratio	mean_return	median_return	iqr_return	positive_symbol_ratio
60m_365d	ma_crossover	gross	867	48.10%	-0.09%	-0.08%	2.74%	25.00%
60m_365d	rolling_breakout_20	gross	4084	49.19%	0.10%	-0.04%	2.64%	62.50%
60m_730d	ma_crossover	gross	1712	48.07%	-0.08%	-0.10%	2.81%	25.00%
60m_730d	rolling_breakout_20	gross	8205	48.60%	0.12%	-0.07%	2.83%	75.00%
60m_365d	ma_crossover	net_low	867	45.67%	-0.19%	-0.18%	2.74%	12.50%
60m_365d	rolling_breakout_20	net_low	4084	47.04%	0.00%	-0.14%	2.64%	50.00%
60m_730d	ma_crossover	net_low	1712	45.97%	-0.18%	-0.20%	2.81%	12.50%
60m_730d	rolling_breakout_20	net_low	8205	46.50%	0.02%	-0.17%	2.83%	50.00%
60m_365d	ma_crossover	net_high	867	41.64%	-0.39%	-0.38%	2.74%	0.00%
60m_365d	rolling_breakout_20	net_high	4084	42.95%	-0.20%	-0.34%	2.64%	0.00%
60m_730d	ma_crossover	net_high	1712	42.17%	-0.38%	-0.40%	2.81%	12.50%
60m_730d	rolling_breakout_20	net_high	8205	42.80%	-0.18%	-0.37%	2.83%	12.50%

bubble proxy split

sample_key	rule	cost_case	bubble_proxy	trade_count	win_ratio	mean_return	median_return	iqr_return	positive_symbol_ratio
60m_365d	ma_crossover	gross	False	626	48.24%	-0.02%	-0.06%	2.47%	37.50%
60m_365d	ma_crossover	gross	True	241	47.72%	-0.27%	-0.19%	3.65%	25.00%
60m_365d	rolling_breakout_20	gross	False	2686	51.45%	0.21%	0.08%	2.46%	100.00%
60m_365d	rolling_breakout_20	gross	True	1398	44.85%	-0.12%	-0.31%	3.11%	25.00%
60m_730d	ma_crossover	gross	False	1221	48.98%	-0.02%	-0.05%	2.55%	12.50%
60m_730d	ma_crossover	gross	True	491	45.82%	-0.21%	-0.28%	3.70%	37.50%
60m_730d	rolling_breakout_20	gross	False	5413	49.29%	0.08%	-0.03%	2.56%	62.50%
60m_730d	rolling_breakout_20	gross	True	2792	47.28%	0.19%	-0.17%	3.53%	50.00%
60m_365d	ma_crossover	net_low	False	626	45.69%	-0.12%	-0.16%	2.47%	12.50%
60m_365d	ma_crossover	net_low	True	241	45.64%	-0.37%	-0.29%	3.65%	12.50%
60m_365d	rolling_breakout_20	net_low	False	2686	49.37%	0.11%	-0.02%	2.46%	75.00%
60m_365d	rolling_breakout_20	net_low	True	1398	42.56%	-0.22%	-0.41%	3.11%	12.50%
60m_730d	ma_crossover	net_low	False	1221	46.85%	-0.12%	-0.15%	2.55%	12.50%
60m_730d	ma_crossover	net_low	True	491	43.79%	-0.31%	-0.38%	3.70%	25.00%
60m_730d	rolling_breakout_20	net_low	False	5413	47.09%	-0.02%	-0.13%	2.56%	50.00%
60m_730d	rolling_breakout_20	net_low	True	2792	45.34%	0.09%	-0.27%	3.53%	50.00%
60m_365d	ma_crossover	net_high	False	626	41.69%	-0.32%	-0.36%	2.47%	12.50%
60m_365d	ma_crossover	net_high	True	241	41.49%	-0.57%	-0.49%	3.65%	0.00%
60m_365d	rolling_breakout_20	net_high	False	2686	44.86%	-0.09%	-0.22%	2.46%	25.00%
60m_365d	rolling_breakout_20	net_high	True	1398	39.27%	-0.42%	-0.61%	3.11%	12.50%
60m_730d	ma_crossover	net_high	False	1221	42.75%	-0.32%	-0.35%	2.55%	12.50%
60m_730d	ma_crossover	net_high	True	491	40.73%	-0.51%	-0.58%	3.70%	12.50%
60m_730d	rolling_breakout_20	net_high	False	5413	43.01%	-0.22%	-0.33%	2.56%	0.00%
60m_730d	rolling_breakout_20	net_high	True	2792	42.41%	-0.11%	-0.47%	3.53%	37.50%

当前怎么读这页？

先看 same rule 在 gross → net_low → net_high 的退化幅度，确认成本是否值得被升格为默认报告项。
再看 bubble_proxy = True / False 的差异，确认 regime 是否会重排均值与胜率。
如果这两点都成立，那么后续内部 breakout / confirmation 研究就不应再只给 gross 总表。

提醒：这仍是最小 survival experiment，不是完整论文复现，更不是最终策略页。