feat: v_mr 实盘扫描器——回测因子原样落地,腾讯qfq数据源与stock_daily零偏差

- mr_scanner.py: v_mr 6条件筛选(bias60<-10%/RSI<42/ret60<-15%/mom20<5%/额<1500万/rsi_delta>=2) - 股票池=stock_daily distinct code,与回测run_mr_backtest完全同口径(含ST,不含退市/301新创业板) - 数据源腾讯qfq日K: 因子与回测零偏差(实测bias60/ret60/mom20 Δ=0.000000) - regime门控: trend_down/choppy才扫; 幂等保护: 当天已有候选则跳过 - market_watch.py: 接入mr_scanner调用(每30分钟调度,幂等防重复) - 首轮实盘验证: 2026-07-31 trend_down市命中3只(*ST四通/*ST冀凯/*ST椰岛),全为回测验证过的ST股
2026-08-02 20:27:31 +08:00
parent 21401f91cf
commit 2af2e93408
2 changed files with 386 additions and 0 deletions
@@ -229,6 +229,27 @@ def main():
    except Exception as _e:
        print(f"[regime] 计算失败: {_e}", flush=True)

+    # ── v_mr 均值回复扫描（每天最多一次，regime 门控内置）──
+    # trend_down/choppy 时扫描深超跌小盘票，写 candidates(sector='v_mr')
+    # mr_scanner 自带幂等保护：当天已有 v_mr 候选则跳过，避免重复全扫描
+    try:
+        import subprocess as _sp
+        import sys as _sys
+        _script = Path(__file__).parent / "mr_scanner.py"
+        if _script.exists():
+            _r = _sp.run(
+                [_sys.executable, str(_script), "--top", "15"],
+                capture_output=True, text=True, timeout=1500)
+            if _r.stdout:
+                for _line in _r.stdout.strip().splitlines():
+                    print(f"[mr_scanner] {_line}", flush=True)
+            if _r.returncode != 0 and _r.stderr:
+                print(f"[mr_scanner] stderr: {_r.stderr[:500]}", flush=True)
+        else:
+            print("[mr_scanner] 脚本不存在，跳过", flush=True)
+    except Exception as _e:
+        print(f"[mr_scanner] 调用失败: {_e}", flush=True)
+
    # 靜默：只寫文件，不輸出到stdout，避免cron推送


@@ -0,0 +1,365 @@
+#!/usr/bin/env python3
+"""mr_scanner.py — v_mr 均值回复策略实盘扫描器
+
+回测侧 v_mr（strategy_lab.run_mr_backtest / mr_engine_v2.py）已 10y 全市场验证
+（21690笔 / WR 53.6% / avg +3.34%），但实盘此前没有该策略的扫描机制 ——
+本脚本把 v_mr 的入场筛选原样搬到实盘，按日扫描全市场深超跌小盘票。
+
+与回测完全一致的入场条件（mr_cfg v2 定稿）：
+  1. bias60  < -10%    : 收盘价在 MA60 下方超 10%（单边，无下限）
+  2. RSI14   < 42      : 超卖
+  3. ret60   < -15%    : 60 日跌幅超 15%（单边，无下限）
+  4. mom20   < 5%      : 20 日低动量（还没启动反弹）
+  5. amount20 < 1500万 : 小盘（20 日均成交额，百万元）
+  6. rsi_delta >= 2    : RSI 5 日回升 ≥2（止跌回升确认）
+出场建议（exit_cfg v2）：tp=+18% / sl=-8% / max_hold=25 交易日
+
+市场门控（2026-08-02 新增）：
+  - 只读 market_regime 表，regime 为 trend_down 或 choppy 时启用扫描
+    （v_mr 主战场：下跌趋势 + 震荡市；趋势市让位 v_next4）
+  - trend_up 时跳过（v_next4 追涨主战场，不扫超跌）
+
+数据源：Sina 240 分钟线（日K），datalen=120（覆盖 MA60 + 60日回看 + RSI 收敛）
+指标算法：与 backtest_framework.py 完全一致（内联，零偏差）
+输出：candidates 表（sector='v_mr'），与 accumulation_scanner 同 UPSERT 模式
+
+用法：
+  python3 mr_scanner.py            # 完整扫描（regime 门控）
+  python3 mr_scanner.py --force    # 忽略 regime 门控强制扫描
+  python3 mr_scanner.py --top N    # 输出前 N 只（默认 10）
+"""
+import sys, json, urllib.request, re, time, sqlite3
+from pathlib import Path
+from datetime import datetime
+
+DB_PATH = Path("/home/hmo/MoFin/data/mofin.db")
+UA = "Mozilla/5.0"
+
+# ── v_mr 参数（与 mr_engine_v2.py 注册的 mr_cfg/exit_cfg 完全一致）──
+MR_CFG = {
+    "bias_max": -10,      # MA60 下方超 10%（单边无下限）
+    "rsi_max": 42,        # RSI 超卖
+    "ret_max": -15,       # 60 日跌超 15%（单边无下限）
+    "mom20_max": 5,       # 20 日低动量
+    "amount_max": 15,     # 百万元 = 1500 万日成交额
+    "rsi_delta_min": 2,   # RSI 5 日回升 ≥2
+    "mkt_mode": "any",    # 实盘由 regime 门控替代（trend_down/choppy 才扫）
+}
+EXIT_CFG = {"tp_pct": 0.18, "sl_pct": 0.08, "max_hold_days": 25}
+
+TOP_N = 10
+
+
+# ── 技术指标（与 backtest_framework.py 完全同算法）──
+
+def calc_ma(series, n):
+    result = []
+    for i in range(len(series)):
+        if i < n - 1:
+            result.append(None)
+        else:
+            result.append(sum(series[i - n + 1:i + 1]) / n)
+    return result
+
+
+def calc_rsi(series, n=14):
+    deltas = [series[i] - series[i - 1] for i in range(1, len(series))]
+    gains = [d if d > 0 else 0 for d in deltas]
+    losses = [-d if d < 0 else 0 for d in deltas]
+    result = [None] * (n + 1)
+    avg_gain = sum(gains[:n]) / n
+    avg_loss = sum(losses[:n]) / n
+    if avg_loss == 0:
+        result.append(100)
+    else:
+        rs = avg_gain / avg_loss
+        result.append(100 - 100 / (1 + rs))
+    for i in range(n, len(gains)):
+        avg_gain = (avg_gain * (n - 1) + gains[i]) / n
+        avg_loss = (avg_loss * (n - 1) + losses[i]) / n
+        if avg_loss == 0:
+            result.append(100)
+        else:
+            rs = avg_gain / avg_loss
+            result.append(100 - 100 / (1 + rs))
+    while len(result) < len(series):
+        result.insert(0, None)
+    return result[:len(series)]
+
+
+# ── 数据获取 ──
+
+def fetch_tx_klines(code, datalen=120):
+    """腾讯前复权日K（qfq），与 stock_daily 数据零偏差，返回 [{date,open,close,high,low,volume}]"""
+    raw = str(code).strip()
+    if raw.startswith(("6", "9")):
+        prefix = "sh"
+    elif raw.startswith(("0", "3")):
+        prefix = "sz"
+    else:
+        return None
+    url = f"http://ifzq.gtimg.cn/appstock/app/fqkline/get?param={prefix}{raw},day,,,{datalen},qfq"
+    try:
+        req = urllib.request.Request(url, headers={"User-Agent": UA})
+        opener = urllib.request.build_opener(urllib.request.ProxyHandler({}))
+        with opener.open(req, timeout=8) as r:
+            text = r.read().decode("utf-8", errors="replace").strip()
+        data = json.loads(text)
+        node = data.get("data", {}).get(f"{prefix}{raw}", {})
+        bars = node.get("qfqday") or node.get("day") or []
+        if not bars or len(bars) < 70:
+            return None
+        result = []
+        for b in bars:
+            if len(b) < 6:
+                continue
+            result.append({
+                "date": b[0][:10],
+                "open": float(b[1]),
+                "close": float(b[2]),
+                "high": float(b[3]),
+                "low": float(b[4]),
+                "volume": float(b[5]),  # 手
+            })
+        return result
+    except Exception:
+        return None
+
+
+# 兼容别名（供外部引用）
+fetch_sina_klines = fetch_tx_klines
+
+
+def get_stock_pool():
+    """待扫描股票池：stock_daily 的 distinct code（与回测 run_mr_backtest 完全同口径）
+
+    回测股票池 = SELECT DISTINCT sd.code FROM stock_daily（4266只，含300/688，
+    不含301新创业板——数据源未收录）。实盘扫描用同一口径，保证 v_mr
+    信号覆盖的股票都是回测验证过的。
+    """
+    conn = sqlite3.connect(str(DB_PATH), timeout=5)
+    try:
+        existing = set()
+        for r in conn.execute("SELECT code FROM holding_strategies WHERE status='active'"):
+            existing.add(str(r[0]))
+        for r in conn.execute("SELECT code FROM holdings WHERE is_active=1"):
+            existing.add(str(r[0]))
+        # 与回测完全一致：stock_daily 有K线的股票（回测 universe='a' 排除5位港股）
+        all_stocks = [str(r[0]) for r in
+                      conn.execute("SELECT DISTINCT code FROM stock_daily").fetchall()]
+    finally:
+        conn.close()
+    # 只留 A 股（6位数字），排除港股（5位0开头）—— 与回测 is_hk_code 逻辑一致
+    a_stocks = [c for c in all_stocks if len(c) == 6 and c.isdigit()]
+    return a_stocks, existing
+
+
+def load_regime():
+    """读取 market_regime 最新状态"""
+    try:
+        conn = sqlite3.connect(str(DB_PATH), timeout=5)
+        row = conn.execute(
+            "SELECT date, above_ma20, adx, regime FROM market_regime "
+            "ORDER BY date DESC LIMIT 1").fetchone()
+        conn.close()
+        if row:
+            return {"date": row[0], "above_ma20": bool(row[1]), "adx": row[2], "regime": row[3]}
+    except Exception:
+        pass
+    return None
+
+
+# ── v_mr 筛选（与回测 run_mr_backtest 的 6 条件一致）──
+
+def check_vmr(klines):
+    """对单只股票做 v_mr 入场筛选。命中返回信号 dict，否则 None。
+    klines 为升序日K（Sina 返回顺序）。"""
+    if not klines or len(klines) < 70:
+        return None
+    closes = [k["close"] for k in klines]
+    highs = [k["high"] for k in klines]
+    lows = [k["low"] for k in klines]
+    i = len(klines) - 1  # 最新一日
+    close = closes[i]
+    if close <= 0:
+        return None
+
+    ma60 = calc_ma(closes, 60)
+    rsi_all = calc_rsi(closes)
+    rsi = rsi_all[i] if i < len(rsi_all) else None
+    m60 = ma60[i]
+    if not m60 or m60 <= 0 or rsi is None:
+        return None
+
+    # 1. MA60 下方超跌（单边）
+    bias60 = (close - m60) / m60 * 100
+    if bias60 > MR_CFG["bias_max"]:  # 只拦上沿（-10%），无下限
+        return None
+
+    # 2. RSI 超卖
+    if rsi > MR_CFG["rsi_max"]:
+        return None
+
+    # 3. 60 日跌幅（单边）
+    prev60 = closes[i - 60] if i >= 60 else 0
+    prev_ret60 = (close - prev60) / prev60 * 100 if prev60 > 0 else 0
+    if prev_ret60 > MR_CFG["ret_max"]:  # 只拦上沿（-15%），无下限
+        return None
+
+    # 4. 20 日低动量
+    prev20 = closes[i - 20] if i >= 20 else 0
+    mom20 = (close - prev20) / prev20 * 100 if prev20 > 0 else 0
+    if mom20 > MR_CFG["mom20_max"]:
+        return None
+
+    # 5. 小盘（20 日均成交额，百万元）
+    # 腾讯日K volume 单位=手（×100股），成交额=手×100×均价
+    amt20 = []
+    for k in klines[max(0, i - 19):i + 1]:
+        avg_px = (k["high"] + k["low"] + k["close"]) / 3
+        amt20.append(k["volume"] * 100 * avg_px)  # 元
+    amt_valid = [a for a in amt20 if a and a > 0]
+    amount_ma20 = sum(amt_valid) / len(amt_valid) if amt_valid else 0
+    amount_ma20_m = amount_ma20 / 1e6  # 元 → 百万元
+    if MR_CFG["amount_max"] is not None and amount_ma20_m > MR_CFG["amount_max"]:
+        return None
+
+    # 6. RSI 5 日回升（止跌确认）
+    rsi0 = rsi_all[i - 5] if i >= 5 else None
+    rsi_delta = (rsi - rsi0) if rsi0 is not None else 0
+    if rsi_delta < MR_CFG["rsi_delta_min"]:
+        return None
+
+    # 命中 → 出场建议（与回测 exit_cfg 一致）
+    tp_pct = EXIT_CFG["tp_pct"]
+    sl_pct = EXIT_CFG["sl_pct"]
+    target = round(close * (1 + tp_pct), 2)
+    stop = round(close * (1 - sl_pct), 2)
+
+    return {
+        "price": close,
+        "bias60": round(bias60, 2),
+        "rsi": round(rsi, 2),
+        "prev_ret60": round(prev_ret60, 2),
+        "mom20": round(mom20, 2),
+        "amount_ma20": round(amount_ma20_m, 2),
+        "rsi_delta": round(rsi_delta, 2),
+        "target": target,
+        "stop_loss": stop,
+        "date": klines[i]["date"],
+    }
+
+
+def main():
+    force = "--force" in sys.argv
+    top_n = TOP_N
+    if "--top" in sys.argv:
+        try:
+            top_n = int(sys.argv[sys.argv.index("--top") + 1])
+        except (ValueError, IndexError):
+            pass
+
+    print(f"[MR] {datetime.now().strftime('%H:%M')} v_mr 实盘扫描开始", flush=True)
+
+    # ── regime 门控 ──
+    regime = load_regime()
+    if regime:
+        rg = regime["regime"]
+        print(f"  市场阶段: {regime['date']} → {rg} (adx={regime['adx']})", flush=True)
+        if rg not in ("trend_down", "choppy") and not force:
+            print(f"  ⏭ {rg} 非 v_mr 主战场（trend_down/choppy 才扫），跳过", flush=True)
+            return
+        if rg not in ("trend_down", "choppy"):
+            print(f"  ⚠ --force 强制扫描（当前 {rg}）", flush=True)
+    else:
+        print("  ⚠ market_regime 不可用，默认执行扫描（v_mr 全市场可用）", flush=True)
+
+    # ── 幂等检查：当天已扫过 v_mr 则跳过（避免 30 分钟调度重复全扫描）──
+    import sqlite3 as _sq
+    _conn = _sq.connect(str(DB_PATH), timeout=5)
+    try:
+        _today = datetime.now().strftime("%Y-%m-%d")
+        _n = _conn.execute(
+            "SELECT COUNT(*) FROM candidates WHERE sector='v_mr' AND substr(created_at,1,10)=?",
+            (_today,)).fetchone()[0]
+    except Exception:
+        _n = 0
+    _conn.close()
+    if _n > 0 and not force:
+        print(f"  已有 {_n} 条今日 v_mr 候选，跳过重复扫描（--force 可强制）", flush=True)
+        return
+
+    # ── 股票池 ──
+    all_stocks, existing = get_stock_pool()
+    print(f"  股票池: {len(all_stocks)}只A股, 已有策略: {len(existing)}只", flush=True)
+    if not all_stocks:
+        print("  ⚠ stocks 表为空", flush=True)
+        return
+
+    # 并发拉日K（ThreadPool 8 并发，Sina 单只逐个拉）
+    from concurrent.futures import ThreadPoolExecutor, as_completed
+    pool = [c for c in all_stocks if c not in existing]
+    found = []
+    done = 0
+    with ThreadPoolExecutor(max_workers=8) as ex:
+        fut_map = {ex.submit(fetch_sina_klines, c): c for c in pool}
+        for fut in as_completed(fut_map):
+            code = fut_map[fut]
+            done += 1
+            klines = fut.result()
+            if klines:
+                sig = check_vmr(klines)
+                if sig:
+                    found.append((code, sig))
+            if done % 400 == 0:
+                print(f"  已扫描 {done}/{len(pool)}", flush=True)
+
+    print(f"  命中 v_mr 条件: {len(found)} 只", flush=True)
+
+    # 排序：超跌越深越优先（bias60 越小越靠前）
+    found.sort(key=lambda x: x[1]["bias60"])
+
+    # ── 写 candidates 表（UPSERT，保留计算列）──
+    conn = sqlite3.connect(str(DB_PATH), timeout=5)
+    inserted = 0
+    for code, sig in found[:top_n]:
+        name = code
+        try:
+            r = conn.execute("SELECT name FROM stocks WHERE code=?", (code,)).fetchone()
+            if r and r[0]:
+                name = r[0]
+        except Exception:
+            pass
+        price = sig["price"]
+        entry_low = round(price * 0.97, 2)
+        entry_high = round(price * 1.02, 2)
+        sl = sig["stop_loss"]
+        tp = sig["target"]
+        reasons = (f"v_mr超跌(bias60={sig['bias60']}% rsi={sig['rsi']} "
+                   f"ret60={sig['prev_ret60']}% mom20={sig['mom20']}% "
+                   f"额{sig['amount_ma20']}M rsi_delta={sig['rsi_delta']})")
+        # 检查是否已在 candidates 且未 promoted
+        exists = conn.execute(
+            "SELECT code FROM candidates WHERE code=? AND (promoted IS NULL OR promoted=0)",
+            (code,)).fetchone()
+        if exists:
+            continue
+        conn.execute(
+            "INSERT INTO candidates (code, name, sector, reason, "
+            "entry_range, stop_loss, target, created_at) "
+            "VALUES (?,?,?,?,?,?,?,datetime('now','localtime')) "
+            "ON CONFLICT(code) DO UPDATE SET "
+            "name=excluded.name, sector=excluded.sector, reason=excluded.reason, "
+            "entry_range=excluded.entry_range, stop_loss=excluded.stop_loss, target=excluded.target",
+            (code, name, "v_mr", reasons,
+             f"{entry_low}~{entry_high}", sl, tp))
+        inserted += 1
+        print(f"  🟢 {code} {name} 价{price} bias60={sig['bias60']}% "
+              f"rsi={sig['rsi']} ret60={sig['prev_ret60']}% {reasons}", flush=True)
+    conn.commit()
+    conn.close()
+    print(f"  ✅ 新增 {inserted} 只 v_mr 候选（前 {top_n}）", flush=True)
+
+
+if __name__ == "__main__":
+    main()