提交

2026-09-16 09:09:10 +08:00
parent 71a0f6e404
commit a490fdc110
8 changed files with 487 additions and 196 deletions
--- a/backend/scripts/active_mv_calib.py
+++ b/backend/scripts/active_mv_calib.py
@@ -0,0 +1,152 @@
+"""0AMV 校验 v2：对照指南针 app EOD 收盘读数，寻找能否精确一致。
+
+指南针目标（亿元，app 收盘读数）：
+    2026-09-11: 172,364.3
+    2026-09-14: 169,908.5（开 171,979.5 高 175,413.6 低 169,908.5）
+    2026-09-15: 166,791.5（开 169,782.7 高 173,060.4 低 166,658.0）
+
+模型：0AMV = Σ 自由流通市值 × active20，active20 = 1-Π(1-流通换手) 滚动20日。
+自由流通市值 = volume/(turnover_rate_f%) × close（daily_snapshot 近期才有 tff）。
+检验变体：全市场 / 剔北交所 / 再剔科创板 / 剔次新（上市<90自然日），
+若均差一个常数因子 → 指南针自由流通股本口径私有，无法精确复刻。
+
+用法（backend 目录）：
+    env -u SSLKEYLOGFILE PYTHONIOENCODING=utf-8 uv run python scripts/active_mv_calib.py
+"""
+from __future__ import annotations
+
+import asyncio
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+import numpy as np
+import pandas as pd
+from sqlalchemy import text
+
+from app.db import async_session
+
+N_WIN = 20
+TARGETS = {"2026-09-11": 172364.3e8, "2026-09-14": 169908.5e8, "2026-09-15": 166791.5e8}
+
+
+async def _load() -> tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame, pd.DataFrame]:
+    async with async_session() as session:
+        hist = (await session.execute(text("""
+            SELECT ts, symbol, close, volume, amount, turnover
+            FROM candles
+            WHERE timeframe = '1d'
+              AND turnover IS NOT NULL AND turnover > 0
+              AND volume > 0
+              AND ts >= now() - interval '500 days'
+        """))).fetchall()
+        pend = (await session.execute(text("""
+            SELECT ts, symbol, close, volume, amount
+            FROM candles
+            WHERE timeframe = '1d' AND volume > 0
+              AND ts::date = (SELECT max(ts)::date FROM candles
+                              WHERE timeframe='1d' AND volume > 0)
+        """))).fetchall()
+        snap = (await session.execute(text("""
+            SELECT ts_code, trade_date::date AS d, turnover_rate, turnover_rate_f
+            FROM daily_snapshot
+            WHERE trade_date >= now() - interval '20 days'
+              AND turnover_rate IS NOT NULL AND turnover_rate_f > 0
+        """))).fetchall()
+        basic = (await session.execute(text("""
+            SELECT symbol, market, exchange, list_date FROM stock_basic
+        """))).fetchall()
+    return (
+        pd.DataFrame(hist, columns=["ts", "symbol", "close", "volume", "amount", "turnover"]),
+        pd.DataFrame(pend, columns=["ts", "symbol", "close", "volume", "amount"]),
+        pd.DataFrame(snap, columns=["ts_code", "d", "turnover", "turnover_rate_f"]),
+        pd.DataFrame(basic, columns=["symbol", "market", "exchange", "list_date"]),
+    )
+
+
+def main() -> None:
+    hist, pend, snap, basic = asyncio.run(_load())
+    snap["symbol"] = snap["ts_code"].str.split(".").str[0]
+
+    # 最新 candles 交易日若缺换手率（夜间 3.5 步未跑），用 snapshot 补
+    if not pend.empty and not snap.empty:
+        d_pend = pend["ts"].dt.date.max()
+        sp = snap[snap["d"] == d_pend][["symbol", "turnover"]]
+        pend = pend[pend["ts"].dt.date == d_pend].merge(sp, on="symbol", how="inner")
+        pend = pend[pend["turnover"] > 0]
+        print(f"最新交易日 {d_pend}：{len(pend):,} 行用 snapshot 换手率补齐")
+
+    df = pd.concat([hist, pend[hist.columns]], ignore_index=True)
+    df["date"] = df["ts"].dt.date
+    df = df.sort_values(["symbol", "date"], kind="stable").reset_index(drop=True)
+    df = df.merge(basic, on="symbol", how="left")
+    print(f"合计 {len(df):,} 行，{df['symbol'].nunique():,} 只，{df['date'].min()} ~ {df['date'].max()}")
+    print("板块分布：", df.drop_duplicates("symbol")["market"].value_counts().to_dict())
+
+    t = (df["turnover"] / 100.0).clip(upper=0.9999)
+    df["log_inactive"] = np.log1p(-t)
+    decay = df.groupby("symbol")["log_inactive"].transform(
+        lambda s: s.rolling(N_WIN, min_periods=1).sum()
+    )
+    df["active20"] = 1.0 - np.exp(decay.to_numpy())
+
+    # ---- 每个有 tff 的日期：个股级 FF20 明细 ----
+    ms: dict = {}
+    for d in sorted(snap["d"].unique())[-12:]:
+        day = df[df["date"] == d]
+        if day.empty:
+            continue
+        m = day.merge(snap.loc[snap["d"] == d, ["symbol", "turnover_rate_f"]],
+                      on="symbol", how="inner")
+        m = m[m["turnover_rate_f"] > 0]
+        tff = (m["turnover_rate_f"] / 100.0).clip(upper=0.9999)
+        m["free_mv"] = m["volume"] / tff * m["close"]
+        m["ff"] = m["free_mv"] * m["active20"]
+        listed = pd.to_datetime(m["list_date"], format="%Y%m%d", errors="coerce")
+        m["age_days"] = (pd.Timestamp(d) - listed).dt.days.fillna(10**6)
+        ms[d] = m
+
+    # ---- 参数扫描：剔除次新窗口 × 是否剔北交所 ----
+    print("\n== 参数扫描（对指南针三日的比值；理想=1.00000 稳定）==")
+    print(f"{'剔北交所':<6}{'剔次新(自然日)':>12}{'09-11':>10}{'09-14':>10}{'09-15':>10}{'均值':>10}{'极差':>9}")
+    results = []
+    for bse_ex in (False, True):
+        for w in (0, 60, 90, 120, 150, 180, 270, 365, 550):
+            ratios = []
+            for tgt_d, tgt in TARGETS.items():
+                d = pd.Timestamp(tgt_d).date()
+                if d not in ms:
+                    break
+                m = ms[d]
+                mask = m["age_days"] >= w
+                if bse_ex:
+                    mask &= m["market"] != "北交所"
+                ratios.append(m.loc[mask, "ff"].sum() / tgt)
+            if len(ratios) < 3:
+                continue
+            mean_r = float(np.mean(ratios))
+            spread = max(ratios) - min(ratios)
+            results.append((abs(mean_r - 1) + spread, bse_ex, w, ratios, mean_r, spread))
+            print(f"{'是' if bse_ex else '否':<6}{w:>12}{ratios[0]:>10.5f}{ratios[1]:>10.5f}"
+                  f"{ratios[2]:>10.5f}{mean_r:>10.5f}{spread:>9.5f}")
+
+    results.sort()
+    _, best_bse, best_w, best_ratios, best_mean, best_spread = results[0]
+    print(f"\n最优配置：剔北交所={'是' if best_bse else '否'}，剔上市<{best_w}自然日")
+    print(f"  三日比值 {['%.5f' % r for r in best_ratios]}，极差 {best_spread:.5f}")
+
+    # 最优配置下的每日序列
+    print("\n== 最优配置近 12 日 FF20（亿）==")
+    for d in sorted(ms):
+        m = ms[d]
+        mask = m["age_days"] >= best_w
+        if best_bse:
+            mask &= m["market"] != "北交所"
+        v = m.loc[mask, "ff"].sum() / 1e8
+        mark = f"  ←指南针 {TARGETS[str(d)]/1e8:,.1f}" if str(d) in TARGETS else ""
+        print(f"{d}  {v:>10,.0f}  n={int(mask.sum())}{mark}")
+
+
+if __name__ == "__main__":
+    main()