第 33 篇：择时策略市场状态识别

import numpy as np
import pandas as pd

# ---------- 1. 数据获取（akshare 优先 + 本地 fallback） ----------
def get_index():
    try:
        import akshare as ak
        df = ak.index_zh_a_hist(symbol="000300", period="daily",
                                start_date="20150101", end_date="20241231")
        df = df.rename(columns={"日期": "date", "收盘": "close"})
        df["date"] = pd.to_datetime(df["date"])
        return df.set_index("date")[["close"]]
    except Exception:
        rng = np.random.default_rng(33)
        n = 2437
        # 造三段行情：熊(2015-2018) → 震荡(2019-2021) → 牛(2022-2024)
        drift = np.concatenate([np.full(1000, -0.0006),
                                np.full(700, 0.0),
                                np.full(737, 0.0008)])
        vol = np.concatenate([np.full(1000, 0.018),
                              np.full(700, 0.010),
                              np.full(737, 0.014)])
        ret = rng.normal(drift, vol)
        idx = pd.bdate_range("2015-01-01", periods=n)
        return pd.DataFrame({"close": 3000 * np.exp(np.cumsum(ret))}, index=idx)

px = get_index()

# ---------- 2. 三态分类器 ----------
def regime_label(px, long_w=200, mid_w=60, vol_w=60, vol_hi=0.8):
    """返回 DataFrame: state(1=上,0=震荡,-1=下), vol_pct(波动率分位)"""
    out = pd.DataFrame(index=px.index)
    out["ma_long"] = px["close"].rolling(long_w).mean()
    out["ma_mid"] = px["close"].rolling(mid_w).mean()
    out["ret"] = px["close"].pct_change()
    out["vol"] = out["ret"].rolling(20).std() * np.sqrt(252)  # 20日年化波动
    out["vol_pct"] = out["vol"].rolling(vol_w).rank(pct=True) # 历史分位
    above = (px["close"] > out["ma_long"]) & (out["ma_mid"] > out["ma_long"])
    below = (px["close"] < out["ma_long"]) & (out["ma_mid"] < out["ma_long"])
    out["state"] = np.where(above, 1, np.where(below, -1, 0))
    return out.dropna(subset=["ma_long"])

reg = regime_label(px)

# ---------- 3. 状态统计画像 ----------
def regime_profile(reg):
    s = reg["state"]
    print("状态占比:", s.value_counts(normalize=True)
          .rename({1: "趋势上", 0: "震荡", -1: "趋势下"}).round(3).to_dict())
    # 平均持续时长：数连续同状态的段
    runs = (s != s.shift()).cumsum()
    dur = s.groupby(runs).agg(["first", "size"])
    for st, name in [(1, "趋势上"), (0, "震荡"), (-1, "趋势下")]:
        d = dur[dur["first"] == st]["size"]
        if len(d) > 0:
            print(f"{name}: 平均持续 {d.mean():.0f} 个交易日, "
                  f"最长 {d.max()} 天")
    # 状态内日收益特征
    for st, name in [(1, "趋势上"), (0, "震荡"), (-1, "趋势下")]:
        r = reg.loc[s == st, "ret"]
        print(f"{name}: 日均收益(示意) {r.mean():+.3%}, "
              f"年化波动 {r.std() * np.sqrt(252):.1%}")

regime_profile(reg)

# ---------- 4. 可视化：价格 + 状态着色 ----------
import matplotlib.pyplot as plt
plt.rcParams["font.sans-serif"] = ["SimHei"]
plt.rcParams["axes.unicode_minus"] = False

fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(11, 7), sharex=True,
                               height_ratios=[3, 1])
for st, color in [(1, "tab:red"), (0, "tab:gray"), (-1, "tab:green")]:
    mask = reg["state"] == st
    ax1.scatter(reg.index[mask], reg["close" if "close" in reg else "ma_long"][mask]
                if False else px.loc[reg.index[mask], "close"],
                c=color, s=3, label={1: "趋势上", 0: "震荡", -1: "趋势下"}[st])
ax1.plot(reg.index, reg["ma_long"], "b-", lw=0.8, alpha=0.6, label="200日线")
ax1.legend(); ax1.set_title("沪深300 市场状态识别（示意）")
ax2.plot(reg.index, reg["vol_pct"], "k-", lw=0.8)
ax2.axhline(0.8, color="r", ls="--", lw=0.8)
ax2.set_ylabel("波动率分位"); ax2.set_xlabel("日期")
plt.tight_layout(); plt.savefig("regime.png", dpi=120); plt.show()

# ---------- 5. 简单应用：状态驱动的仓位 ----------
def regime_position(reg, base=1.0):
    """趋势上满仓、震荡半仓、趋势下空仓；波动率高再打折"""
    pos = reg["state"].map({1: base, 0: base * 0.5, -1: 0.0})
    pos = pos.where(reg["vol_pct"] < 0.8, pos * 0.7)  # 高波动 → 打7折
    return pos.shift(1)  # 当日状态 → 次日仓位，无未来函数

pos = regime_position(reg)
strat = (pos * reg["ret"]).dropna()
bh = reg["ret"].dropna()
for name, r in [("状态择时", strat), ("买入持有", bh)]:
    nav = (1 + r).cumprod()
    dd = (nav / nav.cummax() - 1).min()
    print(f"{name}: 总收益(示意) {nav.iloc[-1] - 1:.1%}, "
          f"最大回撤 {dd:.1%}")

示例输出（示意，模拟行情结果）:
状态占比: {'趋势下': 0.31, '震荡': 0.27, '趋势上': 0.42}
趋势上: 平均持续 121 个交易日, 最长 402 天
震荡: 平均持续 34 个交易日, 最长 96 天
趋势下: 平均持续 98 个交易日, 最长 297 天
趋势上: 日均收益(示意) +0.052%, 年化波动 15.2%
震荡: 日均收益(示意) +0.004%, 年化波动 9.8%
趋势下: 日均收益(示意) -0.061%, 年化波动 21.5%
状态择时: 总收益(示意) 38.9%, 最大回撤 -19.7%
买入持有: 总收益(示意) 21.4%, 最大回撤 -38.6%