cyb50-quant/quant/cyb50_strategy.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
创业板50指数点位量化交易策略回测框架
训练集：2017-2023 | 验证集：2024-2025
"""

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from datetime import datetime, timedelta
import warnings
warnings.filterwarnings('ignore')

# 设置中文显示
plt.rcParams['font.sans-serif'] = ['DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

# ==================== 1. 数据加载 ====================

def load_real_data():
    """加载真实数据 - cyb50_baostock.csv"""
    df = pd.read_csv('cyb50_baostock.csv')
    df['date'] = pd.to_datetime(df['date'])
    df = df.set_index('date').sort_index()
    
    # 转换数据类型
    for col in ['open', 'high', 'low', 'close', 'volume']:
        df[col] = pd.to_numeric(df[col], errors='coerce')
    
    print(f"真实数据加载成功: {df.index[0].date()} ~ {df.index[-1].date()}")
    return df

# ==================== 2. 技术指标计算 ====================

def calculate_atr(high, low, close, period=20):
    """计算ATR（平均真实波幅）"""
    tr1 = high - low
    tr2 = abs(high - close.shift(1))
    tr3 = abs(low - close.shift(1))
    tr = pd.concat([tr1, tr2, tr3], axis=1).max(axis=1)
    atr = tr.rolling(window=period).mean()
    return atr

def calculate_rsrs(high, low, n=20, m=250):
    """
    计算RSRS指标（阻力支撑相对强度）- 优化版
    返回: (rsrs_score, r_squared)
    """
    # 使用rolling计算斜率和R²
    def rolling_beta(x):
        if len(x) < n or np.std(x[:n//2]) == 0:
            return np.nan
        low_vals = x[:n//2]
        high_vals = x[n//2:]
        if np.std(low_vals) == 0:
            return 0
        beta = np.corrcoef(low_vals, high_vals)[0,1] * np.std(high_vals) / np.std(low_vals)
        return beta
    
    # 简化计算：使用 rolling.apply
    slopes = pd.Series(index=high.index, dtype=float)
    r2s = pd.Series(index=high.index, dtype=float)
    
    for i in range(n-1, len(high)):
        low_window = low.iloc[i-n+1:i+1].values
        high_window = high.iloc[i-n+1:i+1].values
        
        if np.std(low_window) > 0:
            beta = np.corrcoef(low_window, high_window)[0,1] * np.std(high_window) / np.std(low_window)
            # R²
            y_pred = np.mean(high_window) + beta * (low_window - np.mean(low_window))
            ss_res = np.sum((high_window - y_pred) ** 2)
            ss_tot = np.sum((high_window - np.mean(high_window)) ** 2)
            r2 = 1 - ss_res / ss_tot if ss_tot > 0 else 0
            
            slopes.iloc[i] = beta
            r2s.iloc[i] = r2
    
    # 计算标准分（滚动M日）
    rsrs = pd.Series(index=high.index, dtype=float)
    for i in range(m+n-2, len(slopes)):
        slope_window = slopes.iloc[i-m+1:i+1]
        if slope_window.std() > 0:
            zscore = (slopes.iloc[i] - slope_window.mean()) / slope_window.std()
            rsrs.iloc[i] = zscore * r2s.iloc[i]
    
    return rsrs, r2s

def calculate_rsi(close, period=14):
    """计算RSI指标"""
    delta = close.diff()
    gain = (delta.where(delta > 0, 0)).rolling(window=period).mean()
    loss = (-delta.where(delta < 0, 0)).rolling(window=period).mean()
    rs = gain / loss
    rsi = 100 - (100 / (1 + rs))
    return rsi

# ==================== 3. 市场状态判断 ====================

def classify_state(close, ma20, ma60, ma120, atr_percent, rsrs, close_series):
    """
    判断市场状态：BULL（牛市）/ BEAR（熊市）/ OSCILLATE（震荡）
    close_series: 用于计算涨跌幅的收盘价序列
    """
    # 趋势得分
    trend_score = 0
    if close > ma20: trend_score += 1
    if ma20 > ma60: trend_score += 1
    if ma60 > ma120: trend_score += 1
    
    # 波动率
    high_volatility = atr_percent > 5
    
    # 熔断检测（极端情况）
    daily_return = close_series.pct_change().iloc[-1] if len(close_series) > 1 else 0
    crash = daily_return < -0.07
    
    if crash or (trend_score <= 1 and high_volatility and close < ma60):
        return "BEAR"
    elif trend_score >= 2 and not high_volatility:
        return "BULL"
    else:
        return "OSCILLATE"

# ==================== 4. 策略核心 ====================

class CYB50Strategy:
    """创业板50指数交易策略"""
    
    def __init__(self, params=None):
        # 默认参数
        self.params = params or {
            'rsrs_n': 20,
            'rsrs_m': 250,
            'bull_buy': 0.5,
            'bull_sell': -0.7,
            'bear_buy': 1.5,
            'bull_max': 1.0,
            'osc_max': 0.6,
            'stop_loss': 0.10,
            'min_change': 0.20
        }
        self.current_position = 0
        self.state_history = []
        self.entry_price = None
    
    def generate_signal(self, data):
        """生成交易信号"""
        close = data['close']
        high = data['high']
        low = data['low']
        
        # 计算指标
        rsrs, r2 = calculate_rsrs(high, low, 
                                   self.params['rsrs_n'], 
                                   self.params['rsrs_m'])
        
        ma20 = close.rolling(20).mean()
        ma60 = close.rolling(60).mean()
        ma120 = close.rolling(120).mean()
        atr = calculate_atr(high, low, close, 20)
        atr_percent = atr / close * 100
        
        # 获取当前值
        curr_rsrs = rsrs.iloc[-1]
        curr_close = close.iloc[-1]
        curr_ma20 = ma20.iloc[-1]
        curr_ma60 = ma60.iloc[-1]
        curr_ma120 = ma120.iloc[-1]
        curr_atr_pct = atr_percent.iloc[-1]
        
        # 检查是否有足够数据
        if pd.isna(curr_rsrs):
            return 0, "INIT"
        
        # 判断市场状态
        state = classify_state(curr_close, curr_ma20, curr_ma60, 
                              curr_ma120, curr_atr_pct, curr_rsrs, close)
        
        # 状态防抖（连续3日确认，极端情况除外）
        self.state_history.append(state)
        if len(self.state_history) >= 3:
            # 熔断检测：单日大跌立即转熊
            daily_return = close.pct_change().iloc[-1]
            if daily_return < -0.07:
                state = "BEAR"
            elif len(self.state_history) >= 3:
                # 正常防抖
                recent_states = self.state_history[-3:]
                if len(set(recent_states)) > 1:
                    state = self.state_history[-2] if len(self.state_history) >= 2 else state
        
        # 根据状态确定仓位
        target_pos = self._calculate_position(state, curr_rsrs, curr_atr_pct)
        
        # 止损检查
        if self.entry_price is not None and self.current_position > 0:
            current_drawdown = (curr_close - self.entry_price) / self.entry_price
            if current_drawdown < -self.params['stop_loss']:
                target_pos = 0
                self.entry_price = None
        
        # 最小调仓幅度过滤
        if abs(target_pos - self.current_position) < self.params['min_change']:
            target_pos = self.current_position
        
        # 更新入场价
        if target_pos > 0 and self.current_position == 0:
            self.entry_price = curr_close
        elif target_pos == 0:
            self.entry_price = None
        
        self.current_position = target_pos
        return target_pos, state
    
    def _calculate_position(self, state, rsrs, atr_percent):
        """根据状态和指标计算目标仓位"""
        p = self.params
        
        if state == "BULL":
            if rsrs > p['bull_buy']:
                pos = p['bull_max']
            elif rsrs < p['bull_sell']:
                pos = 0
            else:
                pos = p['bull_max'] * 0.5
                
        elif state == "BEAR":
            # 熊市：空仓为主
            if rsrs < -p['bear_buy']:
                pos = 0.1  # 极端超卖，10%仓位博反弹
            else:
                pos = 0
                
        else:  # OSCILLATE
            if rsrs > 0.7:
                pos = p['osc_max']
            elif rsrs < -0.7:
                pos = 0
            else:
                pos = p['osc_max'] * 0.5
        
        # 波动率调整
        if atr_percent > 5:
            pos *= 0.6
        
        return np.clip(pos, 0, 1)

# ==================== 5. 回测引擎 ====================

def backtest(data, strategy, initial_capital=1000000, start_date=None, end_date=None):
    """
    回测引擎
    """
    # 数据切片
    if start_date:
        data = data[data.index >= start_date]
    if end_date:
        data = data[data.index <= end_date]
    
    dates = []
    positions = []
    navs = []
    states = []
    
    capital = initial_capital
    current_nav = 1.0
    
    # 跳过前250日（warm-up）
    start_idx = 250
    
    for i in range(start_idx, len(data)):
        curr_data = data.iloc[:i+1]
        curr_date = data.index[i]
        
        # 获取信号
        position, state = strategy.generate_signal(curr_data)
        
        # 计算当日收益（使用前一日仓位）
        if i > start_idx:
            daily_return = data['close'].iloc[i] / data['close'].iloc[i-1] - 1
            prev_position = positions[-1] if positions else 0
            strategy_return = daily_return * prev_position
            current_nav *= (1 + strategy_return)
        
        dates.append(curr_date)
        positions.append(position)
        navs.append(current_nav)
        states.append(state)
    
    # 构建结果DataFrame
    results = pd.DataFrame({
        'date': dates,
        'position': positions,
        'nav': navs,
        'state': states
    }).set_index('date')
    
    # 计算指数基准
    index_data = data.iloc[start_idx:].copy()
    results['index_close'] = index_data['close']
    results['index_nav'] = results['index_close'] / results['index_close'].iloc[0]
    
    # 计算指标
    metrics = calculate_metrics(results['nav'], results['index_nav'])
    
    return results, metrics

def calculate_metrics(strategy_nav, index_nav):
    """计算回测指标"""
    # 收益率
    total_return = strategy_nav.iloc[-1] - 1
    days = len(strategy_nav)
    annual_return = (1 + total_return) ** (252 / days) - 1
    
    # 指数收益
    index_return = index_nav.iloc[-1] - 1
    index_annual = (1 + index_return) ** (252 / days) - 1
    
    # 最大回撤
    running_max = strategy_nav.expanding().max()
    drawdown = (strategy_nav - running_max) / running_max
    max_drawdown = drawdown.min()
    
    # 波动率
    daily_returns = strategy_nav.pct_change().dropna()
    volatility = daily_returns.std() * np.sqrt(252)
    
    # 夏普比率（假设无风险利率3%）
    excess_return = annual_return - 0.03
    sharpe = excess_return / volatility if volatility > 0 else 0
    
    # 卡玛比率
    calmar = annual_return / abs(max_drawdown) if max_drawdown != 0 else 0
    
    # 胜率
    positive_days = (daily_returns > 0).sum()
    total_days = len(daily_returns)
    win_rate = positive_days / total_days
    
    # Beta
    index_returns = index_nav.pct_change().dropna()
    covariance = daily_returns.cov(index_returns)
    variance = index_returns.var()
    beta = covariance / variance if variance > 0 else 1
    
    # 年化超额收益
    excess_annual = annual_return - index_annual
    
    return {
        'total_return': total_return,
        'annual_return': annual_return,
        'index_return': index_return,
        'index_annual': index_annual,
        'excess_annual': excess_annual,
        'max_drawdown': max_drawdown,
        'volatility': volatility,
        'sharpe': sharpe,
        'calmar': calmar,
        'win_rate': win_rate,
        'beta': beta,
        'trading_days': days
    }

# ==================== 6. 参数优化 ====================

def grid_search(data, param_grid):
    """网格搜索最优参数 - 简化版"""
    best_score = -999
    best_params = None
    best_metrics = None
    
    # 只测试1组参数（加速演示）
    test_params = [
        {'rsrs_n': 20, 'rsrs_m': 250, 'bull_buy': 0.5, 'bull_sell': -0.7, 
         'bear_buy': 1.5, 'bull_max': 1.0, 'osc_max': 0.6, 'stop_loss': 0.10, 'min_change': 0.20},
    ]
    
    for params in test_params:
        print(f"测试参数: {params}")
        strategy = CYB50Strategy(params)
        results, metrics = backtest(data, strategy, start_date='2018-02-01', end_date='2023-12-31')
        
        # 综合评分
        score = metrics['sharpe'] * 0.4 + metrics['calmar'] * 0.4 + metrics['excess_annual'] * 2
        
        print(f"  年化: {metrics['annual_return']*100:.1f}%, 回撤: {metrics['max_drawdown']*100:.1f}%, 夏普: {metrics['sharpe']:.2f}, 评分: {score:.2f}")
        
        if score > best_score and metrics['max_drawdown'] > -0.40:
            best_score = score
            best_params = params
            best_metrics = metrics
    
    return best_params, best_metrics

# ==================== 7. 可视化 ====================

def plot_results(results, title="Backtest Results"):
    """绘制回测结果"""
    fig, axes = plt.subplots(3, 1, figsize=(12, 10))
    
    # 净值曲线
    ax1 = axes[0]
    ax1.plot(results.index, results['nav'], label='Strategy', linewidth=2)
    ax1.plot(results.index, results['index_nav'], label='Index', linewidth=1, alpha=0.7)
    ax1.set_title(f'{title} - NAV')
    ax1.set_ylabel('NAV')
    ax1.legend()
    ax1.grid(True, alpha=0.3)
    
    # 仓位变化
    ax2 = axes[1]
    ax2.fill_between(results.index, 0, results['position'], alpha=0.3, label='Position')
    ax2.set_ylabel('Position')
    ax2.set_ylim(0, 1.1)
    ax2.legend()
    ax2.grid(True, alpha=0.3)
    
    # 回撤
    ax3 = axes[2]
    running_max = results['nav'].expanding().max()
    drawdown = (results['nav'] - running_max) / running_max
    ax3.fill_between(results.index, drawdown, 0, alpha=0.3, color='red')
    ax3.set_ylabel('Drawdown')
    ax3.set_xlabel('Date')
    ax3.grid(True, alpha=0.3)
    
    plt.tight_layout()
    return fig

# ==================== 8. 主程序 ====================

def main():
    print("="*60)
    print("创业板50指数量化交易策略回测")
    print("="*60)
    
    # 加载真实数据
    print("\n[1] 加载真实数据...")
    data = load_real_data()
    print(f"数据区间: {data.index[0]} ~ {data.index[-1]}, 共{len(data)}个交易日")
    
    # 划分训练集和验证集
    train_end = '2023-12-31'
    val_start = '2024-01-01'
    
    # 训练阶段（参数优化）
    print("\n[2] 训练阶段：参数优化 (2018-2023)...")
    best_params, train_metrics = grid_search(data, None)
    
    print(f"\n最优参数:")
    for k, v in best_params.items():
        print(f"  {k}: {v}")
    
    print(f"\n训练集表现 (2018-2023):")
    print(f"  策略年化收益: {train_metrics['annual_return']*100:.2f}%")
    print(f"  指数年化收益: {train_metrics['index_annual']*100:.2f}%")
    print(f"  超额收益: {train_metrics['excess_annual']*100:.2f}%")
    print(f"  最大回撤: {train_metrics['max_drawdown']*100:.2f}%")
    print(f"  夏普比率: {train_metrics['sharpe']:.2f}")
    print(f"  卡玛比率: {train_metrics['calmar']:.2f}")
    print(f"  胜率: {train_metrics['win_rate']*100:.1f}%")
    print(f"  Beta: {train_metrics['beta']:.2f}")
    
    # 使用最优参数回测训练集（获取完整结果）
    strategy = CYB50Strategy(best_params)
    train_results, _ = backtest(data, strategy, start_date='2018-02-01', end_date=train_end)
    
    # 验证阶段（样本外）
    print(f"\n[3] 验证阶段：样本外测试 (2024-2025)...")
    strategy_val = CYB50Strategy(best_params)
    val_results, val_metrics = backtest(data, strategy_val, start_date=val_start, end_date='2025-12-31')
    
    print(f"\n验证集表现 (2024-2025):")
    print(f"  策略年化收益: {val_metrics['annual_return']*100:.2f}%")
    print(f"  指数年化收益: {val_metrics['index_annual']*100:.2f}%")
    print(f"  超额收益: {val_metrics['excess_annual']*100:.2f}%")
    print(f"  最大回撤: {val_metrics['max_drawdown']*100:.2f}%")
    print(f"  夏普比率: {val_metrics['sharpe']:.2f}")
    print(f"  卡玛比率: {val_metrics['calmar']:.2f}")
    
    # 过拟合检测
    sharpe_decay = (train_metrics['sharpe'] - val_metrics['sharpe']) / train_metrics['sharpe'] if train_metrics['sharpe'] != 0 else 0
    print(f"\n[4] 过拟合检测:")
    print(f"  夏普比率衰减: {sharpe_decay*100:.1f}%")
    if sharpe_decay > 0.5:
        print("  ⚠️ 警告：可能存在严重过拟合")
    elif sharpe_decay > 0.3:
        print("  ⚠️ 注意：轻度过拟合，建议简化参数")
    else:
        print("  ✓ 无过拟合，策略稳健")
    
    # 保存结果
    print(f"\n[5] 保存结果...")
    train_results.to_csv('train_results.csv')
    val_results.to_csv('val_results.csv')
    print("  训练集结果: train_results.csv")
    print("  验证集结果: val_results.csv")
    
    # 绘图
    print(f"\n[6] 生成图表...")
    fig1 = plot_results(train_results, "Training Set (2018-2023)")
    fig1.savefig('train_backtest.png', dpi=150, bbox_inches='tight')
    print("  训练集图表: train_backtest.png")
    
    fig2 = plot_results(val_results, "Validation Set (2024-2025)")
    fig2.savefig('val_backtest.png', dpi=150, bbox_inches='tight')
    print("  验证集图表: val_backtest.png")
    
    print("\n" + "="*60)
    print("回测完成")
    print("="*60)

if __name__ == "__main__":
    main()