提交

backend/scripts/active_mv_probe.py

@@ -0,0 +1,116 @@
+"""复刻指南针 0AMV（活跃市值）——可行性验证原型。
+
+模型：
+    活跃筹码比例(t) = 1 - Π_{近N个交易日} (1 - 换手率)     （换手衰减）
+    0AMV(t) = Σ_个股 流通市值(t) × 活跃筹码比例(t)
+流通市值不额外拉接口：成交量(股) / (换手%/100) 反推流通股本，× close 即得，
+且随解禁/增发每日自适应。N 为指南针未公开的窗口参数，输出多组候选供对照 app 校准。
+
+用法（backend 目录）：
+    env -u SSLKEYLOGFILE uv run python scripts/active_mv_probe.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
+
+WINDOWS = (20, 60, 120)      # 活跃窗口候选（交易日）
+LOOKBACK_DAYS = 500          # 日历日回看，足够 120 交易日窗口
+SHOW_DAYS = 20               # 打印最近 N 个交易日
+
+
+async def _load() -> tuple[pd.DataFrame, pd.DataFrame]:
+    async with async_session() as session:
+        rows = (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()
+        snap = (await session.execute(text("""
+            SELECT ts_code, circ_mv, trade_date
+            FROM daily_snapshot
+            WHERE trade_date >= (SELECT max(trade_date) - interval '10 days' FROM daily_snapshot)
+              AND circ_mv IS NOT NULL
+        """))).fetchall()
+    df = pd.DataFrame(rows, columns=["ts", "symbol", "close", "volume", "amount", "turnover"])
+    snap = pd.DataFrame(snap, columns=["ts_code", "circ_mv_wan", "snap_date"])
+    return df, snap
+
+
+def _validate(df: pd.DataFrame, snap: pd.DataFrame) -> None:
+    """反推流通市值 vs daily_snapshot（万元→元）交叉验证。"""
+    if snap.empty:
+        print("校验：daily_snapshot 为空")
+        return
+    snap["symbol"] = snap["ts_code"].str.split(".").str[0]
+    snap["d"] = snap["snap_date"].dt.date
+    common = set(snap["d"]) & set(df["date"])
+    if not common:
+        print("校验：daily_snapshot 与 candles 无可对比日期")
+        return
+    snap_date = max(common)
+    day = df[df["date"] == snap_date]
+    m = day[["symbol", "circ_mv"]].merge(
+        snap.loc[snap["d"] == snap_date, ["symbol", "circ_mv_wan"]],
+        on="symbol", how="inner",
+    )
+    if m.empty:
+        print("校验：无可对比个股")
+        return
+    ratio = m["circ_mv"] / (m["circ_mv_wan"] * 1e4)
+    print(f"校验（{snap_date}，{len(m)} 只）：反推/官方流通市值中位数 = {ratio.median():.4f}，"
+          f"P10={ratio.quantile(0.1):.3f}，P90={ratio.quantile(0.9):.3f}")
+    big = m.nlargest(5, "circ_mv")
+    for _, r in big.iterrows():
+        print(f"  {r['symbol']}: 反推 {r['circ_mv']/1e8:,.0f}亿 vs 官方 {r['circ_mv_wan']/1e4:,.0f}亿"
+              f"（比值 {r['circ_mv']/(r['circ_mv_wan']*1e4):.3f}）")
+
+
+def main() -> None:
+    df, snap = asyncio.run(_load())
+    df["date"] = df["ts"].dt.date
+    # 关键：分组滚动前必须按 (symbol, date) 排序，SQL 返回顺序不保证
+    df = df.sort_values(["symbol", "date"], kind="stable").reset_index(drop=True)
+    print(f"candles 载入 {len(df):,} 行，{df['symbol'].nunique():,} 只，"
+          f"{df['date'].min()} ~ {df['date'].max()}")
+
+    t = (df["turnover"] / 100.0).clip(upper=0.9999)
+    df["float_shares"] = df["volume"] / t             # 流通股本（股）；volume 单位=股
+    df["circ_mv"] = df["float_shares"] * df["close"]  # 流通市值（元）
+
+    _validate(df, snap)
+
+    df["log_inactive"] = np.log1p(-t)
+    agg_cols = {"circ_mv": "sum", "amount": "sum"}
+    for n in WINDOWS:
+        decay = df.groupby("symbol")["log_inactive"].transform(
+            lambda s: s.rolling(n, min_periods=1).sum()
+        )
+        df[f"amv_{n}"] = df["circ_mv"] * (1.0 - np.exp(decay.to_numpy()))
+        agg_cols[f"amv_{n}"] = "sum"
+
+    daily = df.groupby("date").agg(agg_cols).sort_index().tail(SHOW_DAYS)
+    print("\n日期        全市场流通市值(万亿)  日成交额(万亿)  " +
+          "  ".join(f"0AMV_{n}(万亿) 活跃占比_{n}" for n in WINDOWS))
+    for d, r in daily.iterrows():
+        print(f"{d}  {r['circ_mv']/1e12:8.3f}            {r['amount']/1e12:6.3f}        " +
+              "        ".join(
+                  f"{r[f'amv_{n}']/1e12:7.3f}    {r[f'amv_{n}']/r['circ_mv']*100:5.1f}%"
+                  for n in WINDOWS
+              ))
+
+
+if __name__ == "__main__":
+    main()