cyb50-quant/cat-fly/t1/final_strategy.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
CYB50 T+1 最终策略回测（完整K线管道版）
──────────────────────────────────────────────────────────────────────
数据管道：
  原始K线CSV → 技术指标 → 多空信号 → 交易执行 → T+1转换 → 市场环境标注 → 过滤分析

策略逻辑：
  必要条件：市场状态 NOT IN ['下跌趋势低波', '震荡低波']（死亡区过滤）
  入场条件：加分模型评分 >= 5（Version D 阈值）
  仓位加成：评分≥5 且命中组合规则 → 1.5x 仓位（调整盈亏金额）

对比版本：
  B  — 仅排死亡区（基准）
  D  — 排死亡区 + 评分≥5（入场过滤）
  F  — 排死亡区 + 评分≥5 + 组合规则加仓（最终策略）

组合规则（走前向验证，训练2023-2024，验证2025，盲测2026）：
  规则1：RSI区域=中性偏弱 & rsi_bin=rsi偏低       (训练67% → 验证80%)
  规则2：vol_bin=vol中 & ts_bin=ts强              (训练57% → 验证60%)
  规则3：波动率水平=中等 & ts_bin=ts强             (训练57% → 验证60%)
  规则4：市场状态=震荡高波 & RSI区域=中性偏弱        (训练57% → 验证50%)
  规则5：RSI区域=中性偏弱 & ts_bin=ts弱            (训练57% → 验证50%)
"""
import sys, io, os, contextlib
if sys.platform == 'win32':
    sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8')
    sys.stderr = io.TextIOWrapper(sys.stderr.buffer, encoding='utf-8')

import pandas as pd
import numpy as np
from datetime import datetime
import warnings
warnings.filterwarnings('ignore')

from cyb50_30min_dual_direction import (
    IntradayDataFetcher, DualDirectionSignalGenerator, DualDirectionExecutor
)
from t1_converter import simulate_t1_trades
from comfort_zone_analyzer import MarketEnvironmentAnalyzer, ComfortZoneAnalyzer

# ──────────────────────────────────────────────────────────────────
# 配置项
# ──────────────────────────────────────────────────────────────────
DATA_CSV    = os.path.join(os.path.dirname(__file__), 'cyb50_30min_2023_to_20260325.csv')
INITIAL     = 1_000_000
DEATH_ZONES = {'下跌趋势低波', '震荡低波'}
BOOST_MULT  = 1.5     # 命中组合规则时的仓位乘数

USE_FEES    = True    # True=含手续费（t1_converter计算）; False=按价差×数量重算
EXPORT_CSV  = True    # True=每次运行后导出各版本明细到 CSV（含K线指标）
VERBOSE     = False   # True=显示策略执行过程详情（大量输出）

SEP = '=' * 72


# ──────────────────────────────────────────────────────────────────
# 工具：压制子模块 stdout
# ──────────────────────────────────────────────────────────────────
@contextlib.contextmanager
def suppress_stdout():
    old = sys.stdout
    sys.stdout = io.StringIO()
    try:
        yield
    finally:
        sys.stdout = old

def ctx():
    """每次返回新的上下文管理器（VERBOSE=False时压制输出）"""
    return contextlib.nullcontext() if VERBOSE else suppress_stdout()


# ──────────────────────────────────────────────────────────────────
# 指标分箱
# ──────────────────────────────────────────────────────────────────
def vol_bin(v):
    if pd.isna(v): return 'unknown'
    if v < 0.20: return 'vol极低'
    if v < 0.40: return 'vol低'
    if v < 0.60: return 'vol中'
    if v < 0.80: return 'vol高'
    return 'vol极高'

def rsi_bin(v):
    if pd.isna(v): return 'unknown'
    if v < 0.05: return 'rsi极底'
    if v < 0.10: return 'rsi底'
    if v < 0.20: return 'rsi低'
    if v < 0.40: return 'rsi偏低'
    if v < 0.60: return 'rsi中'
    if v < 0.80: return 'rsi偏高'
    return 'rsi高'

def ts_bin(v):
    if pd.isna(v): return 'unknown'
    if v < 1.0: return 'ts弱'
    if v < 1.5: return 'ts中弱'
    if v < 2.5: return 'ts中'
    if v < 4.0: return 'ts强'
    return 'ts极强'

def add_bins(df):
    df = df.copy()
    df['vol_bin'] = df['波动率分位'].apply(vol_bin)
    df['rsi_bin'] = df['RSI分位'].apply(rsi_bin)
    df['ts_bin']  = df['趋势强度'].apply(ts_bin)
    return df


# ──────────────────────────────────────────────────────────────────
# Version G: 日线指标 & 区域净分
# ──────────────────────────────────────────────────────────────────
def build_daily_indicators(data_df: pd.DataFrame) -> pd.DataFrame:
    """从30分钟K线聚合日线，计算日线技术指标"""
    d = data_df.resample('D').agg(
        Open=('Open', 'first'), High=('High', 'max'),
        Low=('Low', 'min'),    Close=('Close', 'last'),
        Volume=('Volume', 'sum')
    ).dropna(subset=['Close'])
    delta = d['Close'].diff()
    gain  = delta.where(delta > 0, 0).rolling(14).mean()
    loss  = (-delta.where(delta < 0, 0)).rolling(14).mean()
    d['RSI_D']      = 100 - 100 / (1 + gain / (loss + 1e-9))
    low9            = d['Low'].rolling(9).min()
    high9           = d['High'].rolling(9).max()
    rsv             = (d['Close'] - low9) / (high9 - low9 + 1e-9) * 100
    d['K_D']        = rsv.ewm(com=2, adjust=False).mean()
    d['D_D']        = d['K_D'].ewm(com=2, adjust=False).mean()
    d['J_D']        = 3 * d['K_D'] - 2 * d['D_D']
    d['MA5_D']      = d['Close'].rolling(5).mean()
    d['MA20_D']     = d['Close'].rolling(20).mean()
    d['MA5_slope']  = d['MA5_D'].diff() / d['MA5_D'].shift() * 100   # 日MA5斜率%
    d['pct_MA20']   = (d['Close'] - d['MA20_D']) / d['MA20_D'] * 100
    bm              = d['Close'].rolling(20).mean()
    bs              = d['Close'].rolling(20).std()
    d['BB_pos_D']   = (d['Close'] - (bm - 2*bs)) / (4*bs + 1e-9)
    d['Mom5_D']     = d['Close'].pct_change(5)  * 100
    d['Mom10_D']    = d['Close'].pct_change(10) * 100
    return d


def attach_zone_indicators(enriched: pd.DataFrame,
                            data_with_ind: pd.DataFrame,
                            daily: pd.DataFrame) -> pd.DataFrame:
    """附加入场时刻的30分钟K线指标 + 当日日线指标"""
    kline_cols = ['RSI', 'K', 'D', 'J', 'MACD_hist',
                  'BB_width', 'ATR_Pct', 'Momentum', 'Volume_Ratio']
    daily_cols = ['RSI_D', 'K_D', 'J_D', 'MA5_slope',
                  'pct_MA20', 'BB_pos_D', 'Mom5_D', 'Mom10_D']
    out = enriched.copy()
    # 30分钟指标 — 向量化
    for c in kline_cols:
        if c in data_with_ind.columns:
            out[c] = out['开仓时间'].map(
                lambda t, col=c: data_with_ind.at[t, col]
                if t in data_with_ind.index else float('nan'))
    # 日线指标 — 向量化
    entry_days = out['开仓时间'].dt.normalize()
    for c in daily_cols:
        if c in daily.columns:
            out[c] = entry_days.map(
                lambda t, col=c: daily.at[t, col]
                if t in daily.index else float('nan'))
    return out


# 舒适区规则（命中 +1 分）— 来自 kline_zone_analysis 分析结论
COMFORT_RULES = [
    # (col1, lo1, hi1,  col2,       lo2,  hi2,   描述)
    ('J',       20, 50,  'J_D',       20,  50,   'J跨周期共振低位'),       # WR 85.7%
    ('RSI',     45, 55,  'RSI_D',     60,  70,   'RSI 30min温和+日线强'),  # WR 80.0%
    ('K',       20, 40,  'K_D',       80, 999,   'K低位+日线K超买(反弹)'), # WR 71.4%
    ('K',       20, 40,  'K_D',       20,  40,   'K跨周期共振低位'),       # WR 62.5%
    ('J',        0, 20,  'J_D',       20,  50,   'J超卖+日线低位'),        # WR 66.7%
    ('RSI',     25, 35,  'MA5_slope', 0.1, 0.5,  'RSI弱+日MA5上升'),       # WR 66.7%
    ('RSI',     35, 45,  'RSI_D',     40,  50,   'RSI双偏弱共振'),         # WR 66.7%
    ('RSI',     35, 45,  'MA5_slope', 0.1, 0.5,  'RSI偏弱+MA5上升'),       # WR 66.7%
    ('RSI_D',   60, 70,  'Mom5_D',   -2,   0,   '日RSI强+小幅回调中'),    # WR 63.6%
    ('BB_pos_D',0.3,0.5, 'J_D',        0,  20,  'BB中下+日KDJ超卖'),      # WR 62.5%
    ('MACD_hist',0.8,999, None, None, None,       'MACD强多头'),           # WR 66.7%
    ('Momentum', 0, 0.01, None, None, None,       '30min小涨'),            # WR 62.5%
]

# 黑暗区规则（命中 -1 分）
DARK_RULES = [
    ('RSI',       45, 55,  'RSI_D',       50,  60,  'RSI双均衡无方向'),    # WR 25%
    ('Momentum',-999,-0.03,'Mom5_D',    -999,  -4,  '动量双大跌共振'),      # WR 20%
    ('Volume_Ratio',1.8,999,'ATR_Pct', -999,0.006,  '大放量+极低波动'),    # WR 28.6%
    ('MACD_hist', 0.2, 0.8, None, None, None,        'MACD弱多头'),        # WR 25%
    ('Volume_Ratio',-999,0.5,None,None, None,         '极度缩量'),          # WR 25%
    ('K',         20, 40,  'K_D',         60,  80,  'K低位+日线K高位'),    # WR 33.3%
    ('RSI',       35, 45,  'MA5_slope',  0.5, 999,  'RSI偏弱+急速上升'),   # WR 25%
    ('RSI',       35, 45,  'RSI_D',       70, 999,  'RSI偏弱+日线超买'),   # WR 25%
    ('Momentum',-999,-0.03, None, None, None,         '30min大跌'),        # WR 26.7%
    ('RSI',       25, 35,  'MA5_slope', -999,-0.5,  'RSI弱+日MA5急降'),    # WR 26.3%
    ('BB_pos_D',-999, 0.1, 'J_D',          0,  20,  'BB超下轨+日KDJ超卖'), # WR 28.6%
    ('Mom10_D',    6, 999,  None, None, None,         '日线10日大涨高位'),  # WR 32.1%
]


def _in_range(val, lo, hi) -> bool:
    v = pd.to_numeric(val, errors='coerce')
    return not pd.isna(v) and lo <= v < hi


def zone_net_score(row) -> int:
    """区域净分 = 舒适规则命中数 − 黑暗规则命中数"""
    def hit(c1, lo1, hi1, c2, lo2, hi2):
        if not _in_range(row.get(c1, float('nan')), lo1, hi1):
            return False
        if c2 is None:
            return True
        return _in_range(row.get(c2, float('nan')), lo2, hi2)
    comfort = sum(1 for r in COMFORT_RULES if hit(*r[:6]))
    dark    = sum(1 for r in DARK_RULES    if hit(*r[:6]))
    return comfort - dark


# ──────────────────────────────────────────────────────────────────
# 加分模型（Version D 使用，阈值≥5）
# ──────────────────────────────────────────────────────────────────
def comfort_score(row) -> int:
    s   = 0
    ms  = str(row.get('市场状态', ''))
    vl  = str(row.get('波动率水平', ''))
    rsi = str(row.get('RSI区域', ''))
    vq  = row.get('波动率分位', float('nan'))
    rq  = row.get('RSI分位', float('nan'))
    ts  = row.get('趋势强度', float('nan'))
    t1  = str(row.get('T1调整', ''))   # '否' 或 '是(T0→T1)'

    if ms == '下跌趋势高波' and vl == '极低':      s += 3  # 最优组合
    if pd.notna(rq) and 0.05 <= rq < 0.10:       s += 3  # 最优RSI区间
    if pd.notna(vq) and vq < 0.30:               s += 2  # 低波动率分位
    if pd.notna(ts) and 1.5 <= ts < 4.0:         s += 2  # 适中趋势强度
    if rsi == '中性偏弱':                          s += 2  # 最稳定RSI
    if 'T0' not in t1:                           s += 2  # 非T0延期单
    if vl in ('极低', '低', '中等'):               s += 1  # 低中波动率
    if pd.notna(rq) and rq >= 0.60:              s += 1  # RSI回升阶段
    return s


# ──────────────────────────────────────────────────────────────────
# 组合规则（走前向验证后锁定）
# ──────────────────────────────────────────────────────────────────
COMBO_RULES = [
    {'RSI区域': '中性偏弱', 'rsi_bin': 'rsi偏低'},     # 规则1: 训练67% → 验证80%
    {'vol_bin': 'vol中',   'ts_bin':  'ts强'},         # 规则2: 训练57% → 验证60%
    {'波动率水平': '中等',  'ts_bin':  'ts强'},          # 规则3: 训练57% → 验证60%
    {'市场状态': '震荡高波', 'RSI区域': '中性偏弱'},       # 规则4: 训练57% → 验证50%
    {'RSI区域': '中性偏弱', 'ts_bin':  'ts弱'},         # 规则5: 训练57% → 验证50%
]

def hits_combo(row) -> bool:
    for rule in COMBO_RULES:
        if all(str(row.get(col, '')) == str(val) for col, val in rule.items()):
            return True
    return False


# ──────────────────────────────────────────────────────────────────
# 权益曲线模拟（支持仓位乘数）
# ──────────────────────────────────────────────────────────────────
def simulate_equity(df, initial=INITIAL):
    df = df.copy().reset_index(drop=True)
    if 'pnl_mult' not in df.columns:
        df['pnl_mult'] = 1.0
    cap  = float(initial)
    caps = []
    for _, r in df.iterrows():
        cap += float(r['盈亏金额']) * float(r['pnl_mult'])
        caps.append(cap)
    df['资金余额'] = caps
    df['实际盈亏'] = df['盈亏金额'] * df['pnl_mult']
    df['盈利']     = df['实际盈亏'] > 0
    return df


# ──────────────────────────────────────────────────────────────────
# 绩效统计
# ──────────────────────────────────────────────────────────────────
def calc_stats(df, initial=INITIAL):
    if len(df) == 0:
        return None
    wr  = df['盈利'].mean()
    pnl = df['实际盈亏'].sum()
    cap = df['资金余额'].iloc[-1]
    ret = (cap - initial) / initial
    win = df[df['盈利']]['实际盈亏']
    los = df[~df['盈利']]['实际盈亏']
    plr = abs(win.mean() / los.mean()) if len(los) > 0 and los.mean() != 0 else float('inf')
    eq  = df['资金余额'].values
    pk  = np.maximum.accumulate(np.append([initial], eq))
    dd  = ((eq - pk[1:]) / pk[1:]).min() if len(eq) > 0 else 0
    return dict(n=len(df), wr=wr, pnl=pnl, cap=cap, ret=ret, plr=plr, dd=dd)


def print_yearly(df, initial=INITIAL):
    if len(df) == 0:
        return
    df = df.copy()
    df['年份'] = pd.to_datetime(df['开仓时间']).dt.year
    prev = initial
    for y in sorted(df['年份'].unique()):
        sy  = df[df['年份'] == y]
        pnl = sy['实际盈亏'].sum()
        wr  = sy['盈利'].mean()
        end = sy['资金余额'].iloc[-1]
        print(f"    {y}年: {len(sy):>3}笔 胜率{wr:.1%} | {pnl:>+12,.0f}元 ({pnl/prev:>+.2%}) → {end:,.0f}元")
        prev = end


def print_regime_breakdown(df):
    if len(df) == 0 or '市场状态' not in df.columns:
        return
    print(f"    {'市场状态':<15} {'笔数':>5} {'胜率':>7} {'均盈亏':>10} {'总盈亏':>12}")
    print(f"    {'-'*15} {'-'*5} {'-'*7} {'-'*10} {'-'*12}")
    for ms in sorted(df['市场状态'].unique()):
        sub = df[df['市场状态'] == ms]
        wr  = sub['盈利'].mean()
        avg = sub['实际盈亏'].mean()
        tot = sub['实际盈亏'].sum()
        print(f"    {ms:<15} {len(sub):>5} {wr:>7.1%} {avg:>+10,.0f} {tot:>+12,.0f}")


# ──────────────────────────────────────────────────────────────────
# 主流程
# ──────────────────────────────────────────────────────────────────
def main():
    print(SEP)
    print('  CYB50 T+1 最终策略回测（完整K线管道版）')
    print(f'  Version D (排死亡区+评分≥5) vs Version F (D + 组合规则加仓{BOOST_MULT}x)')
    print(SEP)

    # ── 步骤1: 加载原始K线 CSV ───────────────────────────────────
    print(f'\n📂 步骤1: 加载原始K线数据...')
    raw_csv = pd.read_csv(DATA_CSV, encoding='utf-8-sig')
    fetcher = IntradayDataFetcher()
    with ctx():
        data = fetcher._process_dataframe(
            raw_csv,
            pd.Timestamp('2000-01-01'),
            pd.Timestamp('2099-12-31')
        )
    print(f'   K线: {len(data)}条  {data.index[0].date()} ~ {data.index[-1].date()}')

    # ── 步骤2: 计算技术指标 ──────────────────────────────────────
    print(f'\n📈 步骤2: 计算技术指标...')
    with ctx():
        data_with_ind = fetcher.calculate_intraday_indicators(data)
    ind_cols = [c for c in ['RSI','K','D','J','MACD','MACD_hist','BB_width',
                             'ATR_Pct','Momentum','Volume_Ratio'] if c in data_with_ind.columns]
    print(f'   指标列: {ind_cols}')

    # ── 步骤3: 生成信号 & 执行多空双向交易 ──────────────────────
    print(f'\n🔄 步骤3: 生成多空双向信号并执行交易...')
    gen = DualDirectionSignalGenerator()
    with ctx():
        signals_df = gen.generate_dual_direction_signals(data_with_ind)
    executor = DualDirectionExecutor(initial_capital=INITIAL)
    with ctx():
        _, trades_df = executor.execute_dual_direction_trades(signals_df)
    long_trades  = trades_df[trades_df['交易方向'] == '做多'].copy()
    short_trades = trades_df[trades_df['交易方向'] == '做空'].copy()
    print(f'   总交易: {len(trades_df)}笔  做多: {len(long_trades)}笔  做空: {len(short_trades)}笔')

    # ── 步骤4: T+1规则转换（仅做多） ────────────────────────────
    print(f'\n📅 步骤4: T+1规则转换（做多交易）...')
    with ctx():
        t1_trades = simulate_t1_trades(data_with_ind, long_trades, INITIAL)
    if len(t1_trades) == 0:
        print('   ⚠️  T+1转换后无交易记录，退出。')
        return
    t1_adj = (t1_trades['T+1调整'] == '是(T0→T1)').sum()
    print(f'   T+1交易: {len(t1_trades)}笔  其中T0→T1调整: {t1_adj}笔')

    # ── 步骤5: 计算市场环境 & 标注每笔交易 ──────────────────────
    print(f'\n🔬 步骤5: 市场环境分析 & 交易标注...')
    market_analyzer = MarketEnvironmentAnalyzer(data_with_ind)
    cza = ComfortZoneAnalyzer(t1_trades, market_analyzer)
    with ctx():
        cza._enrich_trades_with_environment()
    enriched = cza.enriched_trades.copy()
    print(f'   成功标注: {len(enriched)}笔交易')

    # ── 规范化列名 & 数值类型 ────────────────────────────────────
    # T+1调整 → T1调整（供 comfort_score 使用）
    if 'T+1调整' in enriched.columns:
        enriched.rename(columns={'T+1调整': 'T1调整'}, inplace=True)

    enriched['开仓时间'] = pd.to_datetime(enriched['开仓时间'])
    enriched['平仓时间'] = pd.to_datetime(enriched['平仓时间'])
    for c in ['盈亏金额', '盈亏百分比', '波动率分位', 'RSI分位', '趋势强度',
              '布林带位置', '1日动量', '成交量分位']:
        if c in enriched.columns:
            enriched[c] = pd.to_numeric(enriched[c], errors='coerce')
    enriched = enriched.sort_values('开仓时间').reset_index(drop=True)
    enriched = add_bins(enriched)

    # USE_FEES 开关
    if not USE_FEES:
        enriched['盈亏金额'] = (enriched['平仓价格'] - enriched['开仓价格']) * enriched['仓位']
        fee_label = '不含手续费（按价差×数量重算）'
    else:
        fee_label = '含手续费（t1_converter计算）'

    print(f'   数据区间: {enriched["开仓时间"].min().date()} ~ {enriched["开仓时间"].max().date()}')
    print(f'   手续费模式: {fee_label}')

    # ── 步骤6: 评分 & 附加日线/K线指标 ──────────────────────────
    print(f'\n🎯 步骤6: 评分 & 附加日线指标 & 构建 B/D/F/G...')
    enriched['_score']     = enriched.apply(comfort_score, axis=1)
    enriched['_combo_hit'] = enriched.apply(hits_combo, axis=1)
    enriched['_year']      = enriched['开仓时间'].dt.year

    # 附加 K线+日线指标，计算区域净分
    daily = build_daily_indicators(data_with_ind)
    enriched = attach_zone_indicators(enriched, data_with_ind, daily)
    enriched['_zone_net'] = enriched.apply(zone_net_score, axis=1)
    enriched['_zone']     = enriched['_zone_net'].apply(
        lambda n: '舒适' if n >= 2 else ('黑暗' if n <= -2 else '中性'))

    not_dead   = ~enriched['市场状态'].isin(DEATH_ZONES)
    high_score = enriched['_score'] >= 5
    not_dark   = enriched['_zone'] != '黑暗'

    vB_df = enriched[not_dead].copy();                          vB_df['pnl_mult'] = 1.0
    vD_df = enriched[not_dead & high_score].copy();             vD_df['pnl_mult'] = 1.0
    vF_df = enriched[not_dead & high_score].copy()
    vF_df['pnl_mult'] = vF_df['_combo_hit'].apply(lambda x: BOOST_MULT if x else 1.0)
    # Version G：排死亡区 + 评分≥5 + 跳过黑暗区 + 舒适区×1.5仓位
    vG_df = enriched[not_dead & high_score & not_dark].copy()
    vG_df['pnl_mult'] = vG_df['_zone'].apply(lambda z: BOOST_MULT if z == '舒适' else 1.0)

    vB = simulate_equity(vB_df)
    vD = simulate_equity(vD_df)
    vF = simulate_equity(vF_df)
    vG = simulate_equity(vG_df)

    sB, sD, sF, sG = calc_stats(vB), calc_stats(vD), calc_stats(vF), calc_stats(vG)

    # ════════════════════════════════════════════════════════════
    # 输出：回测结果对比
    # ════════════════════════════════════════════════════════════
    print()
    print(SEP)
    print('  版本说明')
    print(SEP)
    print(f'  Version B = 排死亡区（基准）                               {sB["n"] if sB else 0}笔')
    print(f'  Version D = 排死亡区 + 评分≥5                             {sD["n"] if sD else 0}笔')
    print(f'  Version F = 排死亡区 + 评分≥5 + 组合规则加仓{BOOST_MULT}x         {sF["n"] if sF else 0}笔')
    print(f'  Version G = 排死亡区 + 评分≥5 + 区域过滤 + 舒适区{BOOST_MULT}x     {sG["n"] if sG else 0}笔')
    print()
    print(SEP)
    print('  回测结果对比')
    print(SEP)

    def fv(s, k, fmt):
        if s is None: return 'N/A'
        v = s[k]
        if fmt == 'n':    return str(int(v))
        if fmt == 'pct':  return f'{v:.1%}'
        if fmt == 'pct2': return f'{v:+.2%}'
        if fmt == 'f2':   return f'{v:.2f}'
        if fmt == 'money':return f'{v:+,.0f}'
        if fmt == 'cap':  return f'{v:,.0f}'
        return str(v)

    hdr = (f'  {"指标":<14} {"Version B(基准)":>14} {"Version D":>12}'
           f' {"Version F(+加仓)":>16} {"Version G(区域)":>16}')
    print(hdr)
    print('  ' + '-' * 74)
    for name, k, fmt in [
        ('交易笔数', 'n',   'n'),
        ('胜率',     'wr',  'pct'),
        ('盈亏比',   'plr', 'f2'),
        ('总收益率', 'ret', 'pct2'),
        ('最终资金', 'cap', 'cap'),
        ('总盈亏',   'pnl', 'money'),
        ('最大回撤', 'dd',  'pct'),
    ]:
        print(f'  {name:<14} {fv(sB,k,fmt):>14} {fv(sD,k,fmt):>12}'
              f' {fv(sF,k,fmt):>16} {fv(sG,k,fmt):>16}')

    # ── 年度明细 ─────────────────────────────────────────────────
    print(f'\n  Version B（排死亡区）年度明细:')
    print_yearly(vB)
    print(f'\n  Version D（评分≥5）年度明细:')
    print_yearly(vD)
    print(f'\n  Version F（评分≥5 + 加仓）年度明细:')
    print_yearly(vF)
    print(f'\n  Version G（区域过滤 + 舒适区加仓）年度明细:')
    print_yearly(vG)

    # ── 组合规则命中分析 ─────────────────────────────────────────
    print(f'\n{SEP}')
    print('  组合规则命中分析（在评分≥5的交易中）')
    print(SEP)

    combo_hit  = vF_df[vF_df['_combo_hit']].copy()
    combo_hit['实际盈亏'] = combo_hit['盈亏金额'] * BOOST_MULT
    combo_hit['盈利']     = combo_hit['实际盈亏'] > 0
    combo_miss = vF_df[~vF_df['_combo_hit']].copy()
    combo_miss['实际盈亏'] = combo_miss['盈亏金额'] * 1.0
    combo_miss['盈利']     = combo_miss['实际盈亏'] > 0

    print(f'\n  命中规则: {len(combo_hit)}笔 '
          f'| 胜率{combo_hit["盈利"].mean():.1%} '
          f'| 均盈亏{combo_hit["实际盈亏"].mean():+,.0f}元 '
          f'| 仓位乘数{BOOST_MULT}x')
    print(f'  未命中:   {len(combo_miss)}笔 '
          f'| 胜率{combo_miss["盈利"].mean():.1%} '
          f'| 均盈亏{combo_miss["实际盈亏"].mean():+,.0f}元 '
          f'| 仓位乘数1.0x')

    combo_hit['_year'] = pd.to_datetime(combo_hit['开仓时间']).dt.year
    print(f'\n  命中规则年份分布:')
    print(f'  {"年份":>5} {"命中笔数":>8} {"胜率":>7} {"总盈亏(1.5x)":>14}')
    print(f'  {"-----":>5} {"--------":>8} {"-------":>7} {"----------":>14}')
    for y in sorted(enriched['_year'].unique()):
        sy = combo_hit[combo_hit['_year'] == y]
        if len(sy) == 0:
            print(f'  {y:>5} {"0":>8} {"—":>7} {"0":>14}')
        else:
            print(f'  {y:>5} {len(sy):>8} {sy["盈利"].mean():>7.1%} {sy["实际盈亏"].sum():>+14,.0f}')

    # ── 市场状态分布（Version F）────────────────────────────────
    print(f'\n{SEP}')
    print('  Version F 各市场状态表现')
    print(SEP)
    print()
    print_regime_breakdown(vF)

    # ── Version G 区域分布说明 ───────────────────────────────────
    print(f'\n{SEP}')
    print('  Version G — 区域分布（在 Version D 85笔中）')
    print(SEP)
    vD_z = enriched[not_dead & high_score].copy()
    zone_grp = vD_z.groupby('_zone').agg(
        n=('盈亏金额', 'count'),
        wr=('盈亏金额', lambda x: (x > 0).mean()),
        avg_pnl=('盈亏金额', 'mean'),
        total_pnl=('盈亏金额', 'sum'),
    ).reset_index()
    print(f'\n  {"区域":<6} {"笔数":>5} {"胜率":>7} {"均盈亏":>10} {"总盈亏":>12}  {"Version G操作"}')
    print(f'  {"-"*6} {"-"*5} {"-"*7} {"-"*10} {"-"*12}  {"-"*16}')
    action_map = {'舒适': f'入场 ×{BOOST_MULT} 仓位', '中性': '入场 ×1.0 仓位', '黑暗': '⛔ 跳过'}
    for _, r in zone_grp.sort_values('wr', ascending=False).iterrows():
        act = action_map.get(r['_zone'], '')
        print(f'  {r["_zone"]:<6} {int(r["n"]):>5} {r["wr"]:>7.1%}'
              f' {r["avg_pnl"]:>+10,.0f} {r["total_pnl"]:>+12,.0f}  {act}')

    # 各年黑暗区过滤效果
    print(f'\n  黑暗区过滤效果（Version D → Version G 减少笔数）:')
    print(f'  {"年份":>5} {"D总笔":>6} {"G入场":>6} {"过滤":>5} {"D总盈亏":>12} {"G总盈亏":>12} {"改善":>12}')
    print(f'  {"-"*5} {"-"*6} {"-"*6} {"-"*5} {"-"*12} {"-"*12} {"-"*12}')
    for y in sorted(enriched['_year'].unique()):
        d_y  = vD[vD['开仓时间'].dt.year == y] if '开仓时间' in vD.columns else pd.DataFrame()
        g_y  = vG[vG['开仓时间'].dt.year == y] if '开仓时间' in vG.columns else pd.DataFrame()
        nd   = len(d_y);  ng = len(g_y)
        pd_  = d_y['实际盈亏'].sum() if nd else 0
        pg_  = g_y['实际盈亏'].sum() if ng else 0
        print(f'  {y:>5} {nd:>6} {ng:>6} {nd-ng:>5}'
              f' {pd_:>+12,.0f} {pg_:>+12,.0f} {pg_-pd_:>+12,.0f}')

    # ── K线指标详情（Version F 最近20笔）────────────────────────
    print(f'\n{SEP}')
    print('  Version F 最近20笔 — 入场时K线指标')
    print(SEP)
    print(f'\n  {"":1} {"开仓时间":<17} {"市场状态":<12} {"评分":>4} {"组合":>4}'
          f' {"RSI":>6} {"K":>6} {"D":>6} {"J":>6}'
          f' {"BB位置":>6} {"动量%":>7} {"量比":>6} {"实际盈亏":>10}')
    print(f'  {"-"*1} {"-"*17} {"-"*12} {"-"*4} {"-"*4}'
          f' {"-"*6} {"-"*6} {"-"*6} {"-"*6}'
          f' {"-"*6} {"-"*7} {"-"*6} {"-"*10}')

    for _, r in vF.tail(20).iterrows():
        t     = r['开仓时间']
        win   = '★' if r['盈利'] else '✗'
        boost = 'HIT' if r['pnl_mult'] > 1.0 else '   '
        sc    = int(enriched.loc[enriched['开仓时间'] == t, '_score'].iloc[0]) \
                if (enriched['开仓时间'] == t).any() else -1
        ms    = str(r.get('市场状态', ''))
        # K线指标
        if t in data_with_ind.index:
            bar = data_with_ind.loc[t]
            rsi_v = bar.get('RSI', float('nan'))
            k_v   = bar.get('K',   float('nan'))
            d_v   = bar.get('D',   float('nan'))
            j_v   = bar.get('J',   float('nan'))
            bb_u  = bar.get('BB_upper', float('nan'))
            bb_l  = bar.get('BB_lower', float('nan'))
            bb_p  = (bar.get('Close', float('nan')) - bb_l) / (bb_u - bb_l + 1e-9)
            mom   = bar.get('Momentum', float('nan')) * 100
            vr    = bar.get('Volume_Ratio', float('nan'))
        else:
            rsi_v = k_v = d_v = j_v = bb_p = mom = vr = float('nan')

        print(f'  {win} {str(t)[:16]} {ms:<12} {sc:>4} {boost:>4}'
              f' {rsi_v:>6.1f} {k_v:>6.1f} {d_v:>6.1f} {j_v:>6.1f}'
              f' {bb_p:>6.2f} {mom:>7.2f} {vr:>6.2f} {r["实际盈亏"]:>+10,.0f}')

    # ── 评分分布 ─────────────────────────────────────────────────
    print(f'\n{SEP}')
    print('  评分分布（非死亡区全量）')
    print(SEP)
    nondead = enriched[not_dead]
    print(f'\n  {"评分":>5} {"笔数":>5} {"胜率":>7} {"均盈亏":>10} {"累计盈亏":>12}  {"选入D/F?":>8}')
    print(f'  {"-"*5} {"-"*5} {"-"*7} {"-"*10} {"-"*12}  {"-"*8}')
    for sc in sorted(nondead['_score'].unique()):
        sub  = nondead[nondead['_score'] == sc]
        wr   = (sub['盈亏金额'] > 0).mean()
        avg  = sub['盈亏金额'].mean()
        tot  = sub['盈亏金额'].sum()
        flag = '✓ 入场' if sc >= 5 else '  跳过'
        print(f'  {sc:>5} {len(sub):>5} {wr:>7.1%} {avg:>+10,.0f} {tot:>+12,.0f}  {flag}')

    # ── CSV 导出 ─────────────────────────────────────────────────
    if EXPORT_CSV:
        ts      = datetime.now().strftime('%Y%m%d_%H%M%S')
        fee_tag = 'nofee' if not USE_FEES else 'fee'
        out_dir = os.path.dirname(__file__)

        # K线指标列（入场时刻）
        kline_cols = [c for c in ['RSI','K','D','J','MACD','MACD_hist',
                                   'BB_width','ATR_Pct','Momentum','Volume_Ratio']
                      if c in data_with_ind.columns]

        def attach_kline(df_in):
            """给 df_in 每行附加入场时刻的K线指标"""
            rows = []
            for t in df_in['开仓时间']:
                if t in data_with_ind.index:
                    rows.append(data_with_ind.loc[t, kline_cols].to_dict())
                else:
                    rows.append({c: float('nan') for c in kline_cols})
            return pd.DataFrame(rows, index=df_in.index)

        base_cols = [
            '交易方向', '开仓时间', '平仓时间', '开仓价格', '平仓价格', '仓位',
            '盈亏金额', '盈亏百分比', '退出原因', 'T1调整',
            '市场状态', '波动率水平', '波动率分位', 'RSI区域', 'RSI分位',
            '趋势强度', '布林带区域', '_score', 'vol_bin', 'rsi_bin', 'ts_bin',
            '成交量分位', '布林带位置', '1日动量',
            '_zone_net', '_zone',   # Version G 区域字段
        ]

        def build_export(df, label):
            out = df.copy().reset_index(drop=True)
            out.insert(0, '序号', range(1, len(out) + 1))
            out['版本']     = label
            out['手续费模式'] = fee_label
            out['仓位乘数'] = out.get('pnl_mult', pd.Series([1.0]*len(out)))
            out['实际盈亏'] = out['盈亏金额'] * out['仓位乘数']
            out['盈利标记'] = out['实际盈亏'].apply(lambda x: '盈' if x > 0 else '亏')
            out['验证盈亏%'] = ((out['平仓价格'] - out['开仓价格'])
                               / out['开仓价格'] * 100).round(4)
            # 附加K线指标
            kl = attach_kline(out)
            for c in kl.columns:
                out[c] = kl[c].values
            keep = (['序号', '版本', '手续费模式', '盈利标记', '仓位乘数',
                     '实际盈亏', '资金余额'] +
                    [c for c in base_cols if c in out.columns] +
                    kline_cols +
                    ['验证盈亏%'])
            keep = [c for c in keep if c in out.columns]
            return out[keep]

        print(f'\n📁 导出CSV...')
        exp_B = build_export(vB, 'B_排死亡区')
        exp_D = build_export(vD, 'D_评分≥5')
        exp_F = build_export(vF, 'F_评分≥5+加仓')
        exp_G = build_export(vG, 'G_区域过滤')

        for label, exp in [('B', exp_B), ('D', exp_D), ('F', exp_F), ('G', exp_G)]:
            fname = os.path.join(out_dir, f'backtest_v{label}_{fee_tag}_{ts}.csv')
            exp.to_csv(fname, index=False, encoding='utf-8-sig')
            print(f'   Version {label}: {os.path.basename(fname)}  ({len(exp)}笔)')

        combined = pd.concat([exp_B, exp_D, exp_F, exp_G], ignore_index=True)
        fname_all = os.path.join(out_dir, f'backtest_all_{fee_tag}_{ts}.csv')
        combined.to_csv(fname_all, index=False, encoding='utf-8-sig')
        print(f'   合并总表: {os.path.basename(fname_all)}  ({len(combined)}笔)')

    print()
    print(SEP)
    print('  回测完成')
    print(SEP)


if __name__ == '__main__':
    main()