research: 低滞后信号口径定稿与实盘前偏差审计
fast_bsp3 改用 tol=-1 + require_touch=False，信号滞后从 5.8 根降到 2.2 根。 滞后与收益严格单调（年化 370% -> 906%，同一份数据同一套成本）， 这是本轮提升的主因，也意味着实盘延迟会直接侵蚀收益。 新增 step31~39 验证策略能否落地： - 跨品种样本外——8 个未参与调参的币，PF 2.73 / t 28.5，无一为负 - 时点重建——只喂到信号那一根重算，同根命中 100%，确认无未来函数； 1m 在 2000 根窗口即饱和，计算耗时 0.20s - 偏差审计——多空对称、中枢生效时刻零回退、滑点稳健至 30bp、持仓几乎不重叠 - 消融——alpha 来自缠论中枢的上下文定位，而非「收盘转强」这个触发动作 补 research/HANDOFF.md：记录确切口径与参数、已排除的偏差、 已验证无效因而不必重做的方向，以及下一步用影子交易器实测执行滑点的方案。 清理 step1~20 的输出：早期方法论已被推翻（存在未来函数偏差）， 其结论不再被引用；脚本保留，需要时可重跑。 Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
jackyu66git
@@ -0,0 +1,192 @@
|
||||
"""Step 33：去掉「回抽必须触及中枢边界」这一条，在最优配置下完整验证。
|
||||
|
||||
step32 的消融实验里，不要求回抽触及边界的档位（C）信号多 43%、PF 持平、
|
||||
中位数几乎翻倍。原因是它把「突破后一去不回头」的强势段也收了进来。
|
||||
本步在三个最优级别对上、叠加大级别同向与中枢阶梯过滤后完整对比，
|
||||
并检查稳健性（分年、分品种、尾部），确认不是样本波动。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import os
|
||||
import sys
|
||||
import warnings
|
||||
from concurrent.futures import ProcessPoolExecutor, as_completed
|
||||
from pathlib import Path
|
||||
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
warnings.filterwarnings("ignore")
|
||||
for v in ("OMP_NUM_THREADS", "OPENBLAS_NUM_THREADS", "MKL_NUM_THREADS"):
|
||||
os.environ.setdefault(v, "1")
|
||||
|
||||
HERE = Path(__file__).resolve().parent
|
||||
sys.path.insert(0, str(HERE))
|
||||
sys.path.insert(0, str(HERE.parent))
|
||||
pd.set_option("display.width", 320)
|
||||
|
||||
SL, TP, MAXB = 1.5, 3.0, 48
|
||||
FEE, SLIP = 0.0004, 0.0001
|
||||
BEST = {"5m": "30m", "15m": "1h", "30m": "2h"}
|
||||
|
||||
|
||||
def run_one(task: tuple) -> dict | None:
|
||||
import warnings as _w
|
||||
_w.filterwarnings("ignore")
|
||||
sys.path.insert(0, str(HERE))
|
||||
sys.path.insert(0, str(HERE.parent))
|
||||
|
||||
from chanlun import TF_DF
|
||||
from lib.breakout import run_trades
|
||||
from lib.data import fetch_ohlcv
|
||||
from lib.fast_bsp3 import find_fast_bsp3
|
||||
from lib.fx_signal import extract_fx_signals, signals_to_frame
|
||||
from lib.nested_bsp import attach_htf_context, htf_fx_timeline
|
||||
from lib.nested_level import build_htf_zones
|
||||
|
||||
sym, ltf = task
|
||||
htf = BEST[ltf]
|
||||
pair = f"{sym}/USDT:USDT"
|
||||
try:
|
||||
df_l = fetch_ohlcv(pair, ltf, 10**9)
|
||||
if df_l is None or len(df_l) < 3000:
|
||||
return None
|
||||
chan_l = TF_DF(df_l, 1, ltf)
|
||||
cdf = chan_l.dataframe
|
||||
zones = build_htf_zones(cdf, ltf, chan=chan_l).reset_index(drop=True)
|
||||
if zones.empty:
|
||||
return None
|
||||
z = zones.copy()
|
||||
pg, pdn = z["zg"].shift(), z["zd"].shift()
|
||||
z["z_above"], z["z_below"] = z["zd"] > pg, z["zg"] < pdn
|
||||
z["zone_i"] = np.arange(len(z))
|
||||
|
||||
df_h = fetch_ohlcv(pair, htf, 10**9)
|
||||
if df_h is None or len(df_h) < 300:
|
||||
return None
|
||||
chan_h = TF_DF(df_h, 1, htf)
|
||||
s = signals_to_frame(extract_fx_signals(chan_h, chan_h.dataframe))
|
||||
tl = htf_fx_timeline(s, chan_h.dataframe)
|
||||
|
||||
out = []
|
||||
for name, req in (("要求回抽触边界(现用)", True), ("不要求回抽触边界", False)):
|
||||
sig = find_fast_bsp3(cdf, zones, require_touch=req)
|
||||
if sig.empty or len(sig) < 20:
|
||||
continue
|
||||
sig = sig.merge(z[["zone_i", "z_above", "z_below"]], on="zone_i", how="left")
|
||||
sig = attach_htf_context(sig, cdf, tl, "h1")
|
||||
sig = sig.drop_duplicates("entry_idx")
|
||||
entries = list(zip(sig["entry_idx"].astype(int),
|
||||
sig["direction"].astype(int)))
|
||||
tr = run_trades(cdf, entries, SL, TP, MAXB, fee=0.0, entry_delay=1)
|
||||
if tr.empty:
|
||||
continue
|
||||
m = sig.set_index("entry_idx")
|
||||
tr["mode"], tr["symbol"], tr["ltf"] = name, sym, ltf
|
||||
tr["date"] = cdf["date"].to_numpy()[tr["entry_idx"].to_numpy()]
|
||||
for c in ("h1_agree", "z_above", "z_below", "direction", "lag", "depth"):
|
||||
if c in m.columns:
|
||||
tr[c if c != "direction" else "dir_sig"] = tr["entry_idx"].map(m[c])
|
||||
out.append(tr)
|
||||
if not out:
|
||||
return None
|
||||
return {"task": f"{sym} {ltf}", "trades": pd.concat(out, ignore_index=True)}
|
||||
except Exception as e:
|
||||
return {"task": f"{sym} {ltf}", "error": repr(e)[:250]}
|
||||
|
||||
|
||||
def stat(g: pd.DataFrame, label: str, minn: int = 25) -> dict:
|
||||
r = g["gross"].to_numpy() - FEE - SLIP
|
||||
if len(r) < minn:
|
||||
return {}
|
||||
w, o = r[r > 0], r[r <= 0]
|
||||
sd = r.std(ddof=1)
|
||||
t10 = r[r <= np.quantile(r, 0.90)]
|
||||
return {"分组": label, "笔数": len(r),
|
||||
"滞后": f"{g['lag'].mean():.1f}" if "lag" in g else "—",
|
||||
"胜率": f"{(r > 0).mean() * 100:.1f}%",
|
||||
"均收益": f"{r.mean() * 100:+.3f}%",
|
||||
"中位": f"{np.median(r) * 100:+.3f}%",
|
||||
"PF": f"{w.sum() / abs(o.sum()):.2f}" if len(o) else "inf",
|
||||
"偏度": f"{pd.Series(r).skew():.2f}",
|
||||
"t值": f"{r.mean() / (sd / np.sqrt(len(r))):+.2f}",
|
||||
"剔10%PF": f"{t10[t10 > 0].sum() / abs(t10[t10 <= 0].sum()):.2f}"}
|
||||
|
||||
|
||||
def main() -> None:
|
||||
ap = argparse.ArgumentParser()
|
||||
ap.add_argument("--reuse", action="store_true")
|
||||
args = ap.parse_args()
|
||||
cache = HERE / "out" / "step33_notouch.csv"
|
||||
|
||||
if args.reuse and cache.exists():
|
||||
allt = pd.read_csv(cache, parse_dates=["date"])
|
||||
print(f"[复用] {len(allt)} 笔\n")
|
||||
else:
|
||||
tasks = [(s, l) for l in BEST for s in ("BTC", "ETH", "SOL")]
|
||||
print(f"[验证] {len(tasks)} 个任务\n", flush=True)
|
||||
res = []
|
||||
with ProcessPoolExecutor(max_workers=5) as ex:
|
||||
futs = {ex.submit(run_one, t): t for t in tasks}
|
||||
for i, f in enumerate(as_completed(futs), 1):
|
||||
r = f.result()
|
||||
if r is None or "error" in (r or {}):
|
||||
print(f" [{i}] 跳过 {(r or {}).get('error', '')}", flush=True)
|
||||
continue
|
||||
res.append(r)
|
||||
print(f" [{i}/{len(tasks)}] {r['task']}", flush=True)
|
||||
if not res:
|
||||
return
|
||||
allt = pd.concat([r["trades"] for r in res], ignore_index=True)
|
||||
allt.to_csv(cache, index=False)
|
||||
|
||||
allt["date"] = pd.to_datetime(allt["date"])
|
||||
allt["push"] = np.where(allt["dir_sig"] == 1, allt["z_above"], allt["z_below"])
|
||||
allt["push"] = allt["push"].fillna(False).astype(bool)
|
||||
fin = allt[(allt["h1_agree"] == 1) & allt["push"]]
|
||||
|
||||
print("=" * 118)
|
||||
print("########## 1. 最终配置对比（同向+阶梯，全级别合并）##########")
|
||||
print(pd.DataFrame([r for r in [stat(g, m) for m, g in fin.groupby("mode")] if r]
|
||||
).to_string(index=False))
|
||||
|
||||
print("\n########## 2. 分级别 ##########")
|
||||
for ltf, g in fin.groupby("ltf"):
|
||||
rows = [stat(x, f"{ltf} {m}") for m, x in g.groupby("mode")]
|
||||
print(pd.DataFrame([r for r in rows if r]).to_string(index=False))
|
||||
|
||||
print("\n########## 3. 分年（看是否某一年独大）##########")
|
||||
for m, g in fin.groupby("mode"):
|
||||
g = g.copy()
|
||||
g["y"] = g["date"].dt.year
|
||||
rows = [stat(x, str(y), minn=20) for y, x in g.groupby("y")]
|
||||
rows = [r for r in rows if r]
|
||||
if rows:
|
||||
print(f"-- {m} --")
|
||||
print(pd.DataFrame(rows).to_string(index=False))
|
||||
|
||||
print("\n########## 4. 分品种 ##########")
|
||||
for m, g in fin.groupby("mode"):
|
||||
rows = [stat(x, s, minn=20) for s, x in g.groupby("symbol")]
|
||||
rows = [r for r in rows if r]
|
||||
if rows:
|
||||
print(f"-- {m} --")
|
||||
print(pd.DataFrame(rows).to_string(index=False))
|
||||
|
||||
print("\n########## 5. 资金曲线 ##########")
|
||||
for m, g in fin.groupby("mode"):
|
||||
g = g.sort_values("date")
|
||||
r = g["gross"].to_numpy() - FEE - SLIP
|
||||
lev = np.clip(0.01 / np.clip(g["risk_pct"].to_numpy(), 0.002, None), 0, 20)
|
||||
pnl = r * lev
|
||||
eq = np.cumprod(1 + pnl)
|
||||
yrs = (g["date"].max() - g["date"].min()).days / 365.25
|
||||
dd = (1 - eq / np.maximum.accumulate(eq)).max()
|
||||
print(f" {m:>18}: n={len(r):>4} 年{len(r) / yrs:>3.0f}笔 "
|
||||
f"年化 {(eq[-1] ** (1 / yrs) - 1) * 100:+7.1f}% 回撤 {dd * 100:5.1f}% "
|
||||
f"Sharpe {pnl.mean() / pnl.std(ddof=1) * np.sqrt(len(pnl) / yrs):5.2f}")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
Reference in New Issue
Block a user