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

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
第二层分析：在排除死亡区后，找出真正把2023-2024和2025-2026分开的指标
"""
import sys, io
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
import warnings
warnings.filterwarnings('ignore')

df = pd.read_csv(
    'D:/work/project/cyb50-quant/cat-fly/t1/t1_trades_with_environment_20260327_141655.csv',
    encoding='utf-8-sig'
)
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','持仓周期数','持仓小时数',
    'T1调整','原平仓时间','原平仓价格','原盈亏','盈亏变化',
    '入场信号','开仓市值','平仓时资金','市场状态',
    '趋势短期','趋势中期','趋势强度','波动率分位','波动率水平',
    '成交量分位','布林带位置','布林带区域','RSI分位','RSI区域',
    '1日动量','入场价格'
]
df.columns = cols
df['开仓时间'] = pd.to_datetime(df['开仓时间'])
df['年份'] = df['开仓时间'].dt.year
df['盈利'] = df['盈亏金额'] > 0
df['好年份'] = df['年份'].isin([2025, 2026])

# 排除死亡区
gz = df[df['市场状态'] != '下跌趋势低波'].copy()

SEP = '=' * 70

def show(t):
    print(f'\n{SEP}\n  {t}\n{SEP}')

# ─── 在非死亡区，比较好年份和差年份各指标分布 ───────────────────
show('非死亡区内：好年份(2025-2026) vs 差年份(2023-2024) 指标对比')

good_gz = gz[gz['好年份']]
bad_gz  = gz[~gz['好年份']]

print(f'\n  差年份(非死亡区): {len(bad_gz)}笔, 胜率{bad_gz["盈利"].mean():.1%}')
print(f'  好年份(非死亡区): {len(good_gz)}笔, 胜率{good_gz["盈利"].mean():.1%}')

print()
print('  各市场状态分布:')
for ms in gz['市场状态'].value_counts().index:
    b = (bad_gz['市场状态'] == ms).mean()
    g = (good_gz['市场状态'] == ms).mean()
    b_wr = bad_gz[bad_gz['市场状态'] == ms]['盈利'].mean() if (bad_gz['市场状态'] == ms).any() else float('nan')
    g_wr = good_gz[good_gz['市场状态'] == ms]['盈利'].mean() if (good_gz['市场状态'] == ms).any() else float('nan')
    marker = ' ←' if abs(g - b) > 0.08 else ''
    print(f'  {ms:<15} | 差:{b:.1%}(胜率{b_wr:.1%}) → 好:{g:.1%}(胜率{g_wr:.1%}){marker}')

print()
print('  各波动率水平分布:')
for vl in ['极低', '低', '中等', '高', '极高']:
    b = (bad_gz['波动率水平'] == vl).mean()
    g = (good_gz['波动率水平'] == vl).mean()
    b_wr = bad_gz[bad_gz['波动率水平'] == vl]['盈利'].mean() if (bad_gz['波动率水平'] == vl).any() else float('nan')
    g_wr = good_gz[good_gz['波动率水平'] == vl]['盈利'].mean() if (good_gz['波动率水平'] == vl).any() else float('nan')
    marker = ' ←' if abs(g - b) > 0.05 else ''
    print(f'  {vl:<8} | 差:{b:.1%}(胜率{b_wr:.1%}) → 好:{g:.1%}(胜率{g_wr:.1%}){marker}')

print()
print('  各RSI区域分布:')
for rsi in gz['RSI区域'].value_counts().index:
    b = (bad_gz['RSI区域'] == rsi).mean()
    g = (good_gz['RSI区域'] == rsi).mean()
    b_wr = bad_gz[bad_gz['RSI区域'] == rsi]['盈利'].mean() if (bad_gz['RSI区域'] == rsi).any() else float('nan')
    g_wr = good_gz[good_gz['RSI区域'] == rsi]['盈利'].mean() if (good_gz['RSI区域'] == rsi).any() else float('nan')
    marker = ' ←' if abs(g - b) > 0.05 else ''
    print(f'  {rsi:<10} | 差:{b:.1%}(胜率{b_wr:.1%}) → 好:{g:.1%}(胜率{g_wr:.1%}){marker}')

# ─── 在非死亡区，找出在差年份里最集中的"毒药"组合 ──────────────
show('非死亡区内：差年份的主要亏损来自哪些组合')

combos = []
for ms in gz['市场状态'].unique():
    for vl in gz['波动率水平'].unique():
        sub = bad_gz[(bad_gz['市场状态'] == ms) & (bad_gz['波动率水平'] == vl)]
        if len(sub) < 4:
            continue
        wr  = sub['盈利'].mean()
        avg = sub['盈亏金额'].mean()
        total = sub['盈亏金额'].sum()
        combos.append((f'{ms} × {vl}', len(sub), wr, avg, total))

combos.sort(key=lambda x: x[4])  # 按总盈亏升序（最亏排前）
print(f'\n  差年份非死亡区内各组合（按总盈亏升序）:')
print(f'  {"组合":<28} 笔数  胜率     均盈亏      总盈亏')
for cond, n, wr, avg, total in combos[:12]:
    print(f'  {cond:<28} {n:>4}  {wr:.1%}  {avg:>+8,.0f}元  {total:>+10,.0f}元')

# ─── 逐步叠加过滤条件，看能否让差年份也变好 ───────────────────
show('逐步叠加过滤：能否让差年份胜率提升至可接受水平')

# 基础: 非死亡区
print(f'\n  [基础] 非死亡区: {len(gz)}笔, 胜率{gz["盈利"].mean():.1%}')

# 过滤1: 再排除震荡低波
f1 = gz[gz['市场状态'] != '震荡低波']
print(f'  [+过滤震荡低波] {len(f1)}笔, 胜率{f1["盈利"].mean():.1%}')
for y in [2023, 2024, 2025, 2026]:
    sy = f1[f1['年份'] == y]
    print(f'    {y}年: {len(sy)}笔, 胜率{sy["盈利"].mean():.1%}, {sy["盈亏金额"].sum():+,.0f}元')

# 过滤2: 再排除 下跌趋势高波×极高波动率
f2 = f1[~((f1['市场状态'] == '下跌趋势高波') & (f1['波动率水平'] == '极高'))]
print(f'\n  [+过滤 下跌趋势高波×极高] {len(f2)}笔, 胜率{f2["盈利"].mean():.1%}')
for y in [2023, 2024, 2025, 2026]:
    sy = f2[f2['年份'] == y]
    if len(sy) == 0:
        continue
    print(f'    {y}年: {len(sy)}笔, 胜率{sy["盈利"].mean():.1%}, {sy["盈亏金额"].sum():+,.0f}元')

# 过滤3: 再排除T+1
f3 = f2[~f2['T1调整'].str.contains('T0', na=False)]
print(f'\n  [+过滤T+1调整] {len(f3)}笔, 胜率{f3["盈利"].mean():.1%}')
for y in [2023, 2024, 2025, 2026]:
    sy = f3[f3['年份'] == y]
    if len(sy) == 0:
        continue
    print(f'    {y}年: {len(sy)}笔, 胜率{sy["盈利"].mean():.1%}, {sy["盈亏金额"].sum():+,.0f}元')

# 过滤4: 再排除 波动率分位>0.7（非死亡区内高波动仍然不好）
f4 = f3[f3['波动率分位'].fillna(0) <= 0.70]
print(f'\n  [+过滤 波动率分位>0.70] {len(f4)}笔, 胜率{f4["盈利"].mean():.1%}')
for y in [2023, 2024, 2025, 2026]:
    sy = f4[f4['年份'] == y]
    if len(sy) == 0:
        continue
    print(f'    {y}年: {len(sy)}笔, 胜率{sy["盈利"].mean():.1%}, {sy["盈亏金额"].sum():+,.0f}元')

# ─── 检查各层过滤对2026年的影响 ─────────────────────────────
show('关键检查：各过滤条件对2026年17笔的影响')
y26 = df[df['年份'] == 2026]
print()

c1 = y26['市场状态'] == '下跌趋势低波'
c2 = y26['市场状态'] == '震荡低波'
c3 = (y26['市场状态'] == '下跌趋势高波') & (y26['波动率水平'] == '极高')
c4 = y26['T1调整'].str.contains('T0', na=False)
c5 = y26['波动率分位'].fillna(0) > 0.70

print(f'  过滤条件                      2026年命中  胜率     盈亏')
for name, cond in [
    ('下跌趋势低波', c1),
    ('震荡低波', c2),
    ('下跌趋势高波×极高', c3),
    ('T+1调整', c4),
    ('波动率分位>0.70', c5),
]:
    hits = y26[cond]
    missed = y26[~cond]
    if len(hits) == 0:
        print(f'  {name:<30} 0笔 (不影响)')
    else:
        wr = hits['盈利'].mean()
        total = hits['盈亏金额'].sum()
        print(f'  {name:<30} {len(hits)}笔 | 胜率{wr:.1%} | 盈亏{total:+,.0f}元  ← 误杀!')

# 所有条件组合对2026的影响
all_filter = c1 | c2 | c3 | c4 | c5
kept_26 = y26[~all_filter]
lost_26 = y26[all_filter]
print(f'\n  全部条件叠加: 保留{len(kept_26)}笔(胜率{kept_26["盈利"].mean():.1%}, {kept_26["盈亏金额"].sum():+,.0f}元) | 过滤{len(lost_26)}笔(胜率{lost_26["盈利"].mean():.1%}, {lost_26["盈亏金额"].sum():+,.0f}元)')

# ─── 找到不误杀2026且能改善差年份的最优组合 ──────────────────
show('目标：找出 不误杀2026，且差年份胜率最高 的过滤策略')

# 策略A: 只排除死亡区
sA = df[df['市场状态'] != '下跌趋势低波']
# 策略B: 排除死亡区 + 震荡低波
sB = df[~df['市场状态'].isin(['下跌趋势低波', '震荡低波'])]
# 策略C: 排除死亡区 + 下跌趋势高波×极高
sC = df[~(df['市场状态'] == '下跌趋势低波') &
        ~((df['市场状态'] == '下跌趋势高波') & (df['波动率水平'] == '极高'))]
# 策略D: 排除死亡区 + T+1
sD = df[(df['市场状态'] != '下跌趋势低波') & (~df['T1调整'].str.contains('T0', na=False))]
# 策略E: 排除死亡区 + 震荡低波 + T+1
sE = df[~df['市场状态'].isin(['下跌趋势低波', '震荡低波']) &
        (~df['T1调整'].str.contains('T0', na=False))]

print()
for name, s in [('A:仅排死亡区', sA), ('B:+排震荡低波', sB), ('C:+排下跌高波×极高', sC),
                ('D:+排T+1', sD), ('E:+排震荡低波&T+1', sE)]:
    y26s = s[s['年份'] == 2026]
    y25s = s[s['年份'] == 2025]
    y24s = s[s['年份'] == 2024]
    y23s = s[s['年份'] == 2023]
    total_pnl = s['盈亏金额'].sum()
    print(f'\n  策略{name}: {len(s)}笔, 总盈亏{total_pnl:+,.0f}元')
    for yr, sy in [('2023', y23s), ('2024', y24s), ('2025', y25s), ('2026', y26s)]:
        if len(sy) == 0:
            continue
        print(f'    {yr}年: {len(sy)}笔, 胜率{sy["盈利"].mean():.1%}, {sy["盈亏金额"].sum():+,.0f}元')