cyb50-quant/market-regime-identifier/market_regime_hmm.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
市场环境识别器 (Market Regime Identifier)
基于HMM隐马尔可夫模型的市场状态识别系统

状态定义：
- 状态0（震荡）：价格波动大但无明显方向，Hurst指数≈0.5，自相关性低
- 状态1（趋势）：价格持续单向运动，Hurst指数>0.6，高自相关
- 状态2（反转）：超买/超卖后的V型反转，RSI极端值后的快速回归

作者: OpenClaw
日期: 2026-03-06
"""

import numpy as np
import pandas as pd
from hmmlearn.hmm import GaussianHMM
from scipy import stats
import warnings
warnings.filterwarnings('ignore')

# ==================== 特征工程 ====================

def calculate_hurst(prices, max_lag=100):
    """
    计算Hurst指数
    H ≈ 0.5: 随机游走（震荡）
    H > 0.6: 趋势性
    H < 0.4: 均值回归
    """
    lags = range(2, min(max_lag, len(prices)//4))
    tau = [np.std(np.subtract(prices[lag:], prices[:-lag])) for lag in lags]
    
    if len(tau) < 2 or any(t <= 0 for t in tau):
        return 0.5
    
    reg = np.polyfit(np.log(lags), np.log(tau), 1)
    return reg[0]

def calculate_rsi(prices, period=14):
    """计算RSI指标（标准 Wilder 方法）"""
    prices = pd.Series(prices)
    delta = prices.diff()
    gain = delta.clip(lower=0)
    loss = (-delta).clip(lower=0)

    avg_gain = gain.ewm(alpha=1/period, adjust=False, min_periods=period).mean()
    avg_loss = loss.ewm(alpha=1/period, adjust=False, min_periods=period).mean()

    rs = avg_gain / (avg_loss + 1e-10)
    rsi = 100 - (100 / (1 + rs))
    return rsi.fillna(50).to_numpy()

def extract_features(df):
    """
    提取特征向量 X_t
    X_t = [收益率标准差(5日), 价格动量(10日), 波动率比率(短/长), 成交量变化率, 日内趋势强度]
    """
    features = pd.DataFrame(index=df.index)
    
    # 1. 收益率标准差(5日)
    returns = df['close'].pct_change()
    features['ret_std_5'] = returns.rolling(5).std() * np.sqrt(252)
    
    # 2. 价格动量(10日)
    features['momentum_10'] = (df['close'] / df['close'].shift(10) - 1) * 100
    
    # 3. 波动率比率(短/长)
    vol_short = returns.rolling(5).std()
    vol_long = returns.rolling(20).std()
    features['vol_ratio'] = vol_short / (vol_long + 1e-10)
    
    # 4. 成交量变化率
    features['volume_change'] = df['volume'].pct_change() * 100
    
    # 5. 日内趋势强度
    features['intraday_trend'] = ((df['close'] - df['open']) / (df['high'] - df['low'] + 1e-10)) * 100
    
    # 6. Hurst指数(额外特征)
    features['hurst'] = df['close'].rolling(100).apply(calculate_hurst, raw=True)
    
    # 7. RSI
    features['rsi'] = calculate_rsi(df['close'].values)
    
    # 8. 自相关性
    features['autocorr'] = returns.rolling(20).apply(lambda x: x.autocorr(lag=1) if len(x) > 1 else 0)
    
    # 填充缺失值
    features = features.ffill().fillna(0)
    
    return features

# ==================== HMM模型 ====================

class MarketRegimeHMM:
    """市场环境HMM模型"""
    
    # 状态名称
    STATE_NAMES = {
        0: '震荡',
        1: '趋势',
        2: '反转'
    }
    
    def __init__(self, n_components=3, n_iter=100):
        # 先验转移概率矩阵
        self.PRIOR_TRANSITION = np.array([
            [0.85, 0.10, 0.05],  # 震荡 -> 震荡/趋势/反转
            [0.15, 0.80, 0.05],  # 趋势 -> 震荡/趋势/反转
            [0.20, 0.10, 0.70]   # 反转 -> 震荡/趋势/反转
        ])
        
        self.model = GaussianHMM(
            n_components=n_components,
            covariance_type='full',
            n_iter=n_iter,
            random_state=42,
            init_params='mc'  # 只初始化均值和协方差，不初始化转移矩阵
        )
        self.is_fitted = False
        
    def fit(self, features):
        """训练HMM模型"""
        print("训练HMM模型...")
        
        X = features.values
        
        # 先验状态分布（均匀分布）
        self.model.startprob_ = np.array([1/3, 1/3, 1/3])
        
        # 使用先验转移概率初始化
        self.model.transmat_ = self.PRIOR_TRANSITION.copy()
        
        # 拟合模型
        self.model.fit(X)
        self.is_fitted = True
        
        print(f"模型收敛: {self.model.monitor_.converged}")
        print(f"迭代次数: {self.model.n_iter}")
        print("\n学习到的转移概率矩阵:")
        print(self.model.transmat_.round(3))
        
        return self
    
    def predict(self, features):
        """预测状态序列"""
        if not self.is_fitted:
            raise ValueError("模型尚未训练，请先调用fit()")
        
        X = features.values
        states = self.model.predict(X)
        
        # 计算状态概率
        state_probs = self.model.predict_proba(X)
        
        return states, state_probs
    
    def get_current_regime(self, features):
        """获取当前市场状态"""
        states, probs = self.predict(features)
        current_state = states[-1]
        current_prob = probs[-1]
        
        return {
            'state': current_state,
            'state_name': self.STATE_NAMES[current_state],
            'probabilities': {
                self.STATE_NAMES[i]: current_prob[i] 
                for i in range(len(self.STATE_NAMES))
            },
            'confidence': current_prob[current_state]
        }

# ==================== 策略切换逻辑 ====================

class StrategySelector:
    """基于市场状态的策略选择器"""
    
    STRATEGY_CONFIG = {
        0: {  # 震荡
            'name': '均值回归',
            'action': 'RSI超买超卖交易',
            'position_size': 0.5,  # 降低仓位
            'stop_loss': '2N',
            'description': '关闭趋势策略，使用RSI超买(>70)超卖(<30)信号'
        },
        1: {  # 趋势
            'name': '海龟趋势',
            'action': '全速运行',
            'position_size': 1.0,  # 全仓位
            'stop_loss': '2N',
            'description': '增加仓位，突破20日高低点交易'
        },
        2: {  # 反转
            'name': '反向/观望',
            'action': '反向信号或空仓',
            'position_size': 0.3,  # 最小仓位
            'stop_loss': '1N',  # 收紧止损
            'description': '反向信号或观望，收紧止损'
        }
    }
    
    @classmethod
    def get_strategy(cls, state):
        """根据状态获取策略配置"""
        return cls.STRATEGY_CONFIG.get(state, cls.STRATEGY_CONFIG[0])
    
    @classmethod
    def generate_signal(cls, state, rsi_value, price, ma20):
        """生成交易信号"""
        strategy = cls.get_strategy(state)
        
        signal = {
            'state': state,
            'strategy': strategy['name'],
            'position_size': strategy['position_size'],
            'action': 'HOLD'
        }
        
        if state == 0:  # 震荡 - RSI均值回归
            if rsi_value < 30:
                signal['action'] = 'BUY'
                signal['reason'] = 'RSI超卖'
            elif rsi_value > 70:
                signal['action'] = 'SELL'
                signal['reason'] = 'RSI超买'
                
        elif state == 1:  # 趋势 - 突破系统
            if price > ma20 * 1.02:
                signal['action'] = 'BUY'
                signal['reason'] = '突破20日均线2%'
            elif price < ma20 * 0.98:
                signal['action'] = 'SELL'
                signal['reason'] = '跌破20日均线2%'
                
        elif state == 2:  # 反转 - 反向或观望
            if rsi_value > 70:
                signal['action'] = 'SELL'
                signal['reason'] = '超买后反转'
            elif rsi_value < 30:
                signal['action'] = 'BUY'
                signal['reason'] = '超卖后反转'
            else:
                signal['action'] = 'HOLD'
                signal['reason'] = '观望'
        
        return signal

# ==================== 模型评估 ====================

def evaluate_model(hmm, features, true_states=None):
    """
    评估模型性能
    
    由于真实状态未知，使用以下指标：
    1. 对数似然值
    2. AIC/BIC
    3. 状态持续时间合理性
    4. 状态与价格行为的对应关系
    """
    X = features.values
    
    # 计算对数似然
    log_likelihood = hmm.model.score(X)
    
    # 计算AIC和BIC
    n_params = hmm.model.n_components * (hmm.model.n_features + hmm.model.n_features * (hmm.model.n_features + 1) / 2) + hmm.model.n_components * hmm.model.n_components
    n_samples = len(X)
    aic = -2 * log_likelihood + 2 * n_params
    bic = -2 * log_likelihood + n_params * np.log(n_samples)
    
    print(f"\n模型评估指标:")
    print(f"对数似然: {log_likelihood:.2f}")
    print(f"AIC: {aic:.2f}")
    print(f"BIC: {bic:.2f}")
    
    # 预测状态
    states, probs = hmm.predict(features)
    
    # 统计状态分布
    state_counts = pd.Series(states).value_counts().sort_index()
    state_pct = (state_counts / len(states) * 100).round(2)
    
    print(f"\n状态分布:")
    for state_id, state_name in hmm.STATE_NAMES.items():
        count = state_counts.get(state_id, 0)
        pct = state_pct.get(state_id, 0)
        print(f"  {state_name}: {count}天 ({pct}%)")
    
    # 计算平均状态持续时间
    state_durations = []
    current_state = states[0]
    duration = 1
    
    for s in states[1:]:
        if s == current_state:
            duration += 1
        else:
            state_durations.append((current_state, duration))
            current_state = s
            duration = 1
    state_durations.append((current_state, duration))
    
    print(f"\n平均状态持续时间:")
    for state_id in range(3):
        durations = [d for s, d in state_durations if s == state_id]
        if durations:
            avg_duration = np.mean(durations)
            print(f"  {hmm.STATE_NAMES[state_id]}: {avg_duration:.1f}天")
    
    return {
        'log_likelihood': log_likelihood,
        'aic': aic,
        'bic': bic,
        'state_distribution': state_counts.to_dict(),
        'states': states,
        'state_probs': probs
    }

# ==================== 主程序 ====================

def main():
    """主程序"""
    print("="*70)
    print("市场环境识别器 (Market Regime Identifier)")
    print("基于HMM隐马尔可夫模型")
    print("="*70)
    
    # 示例：使用随机数据演示
    print("\n注意：这是演示版本，请使用真实数据运行")
    print("数据格式要求：DataFrame包含 'open', 'high', 'low', 'close', 'volume' 列")
    
    # 生成示例数据
    np.random.seed(42)
    n_days = 500
    dates = pd.date_range('2023-01-01', periods=n_days, freq='B')
    
    # 模拟价格走势（包含趋势、震荡、反转三种状态）
    price = 100
    prices = []
    
    for i in range(n_days):
        # 模拟不同状态
        if i < 150:  # 趋势
            price *= (1 + np.random.normal(0.001, 0.01))
        elif i < 300:  # 震荡
            price *= (1 + np.random.normal(0, 0.015))
        else:  # 反转
            if i < 375:
                price *= (1 + np.random.normal(-0.002, 0.012))
            else:
                price *= (1 + np.random.normal(0.002, 0.012))
        prices.append(price)
    
    df = pd.DataFrame({
        'open': prices + np.random.normal(0, 0.5, n_days),
        'high': np.array(prices) + np.abs(np.random.normal(1, 0.5, n_days)),
        'low': np.array(prices) - np.abs(np.random.normal(1, 0.5, n_days)),
        'close': prices,
        'volume': np.random.randint(1000000, 5000000, n_days)
    }, index=dates)
    
    print(f"\n示例数据: {len(df)}天")
    print(f"日期范围: {df.index[0].date()} ~ {df.index[-1].date()}")
    
    # 特征提取
    print("\n提取特征...")
    features = extract_features(df)
    
    # 选择训练特征（核心5个）
    feature_cols = ['ret_std_5', 'momentum_10', 'vol_ratio', 'volume_change', 'intraday_trend']
    X_train = features[feature_cols].dropna()
    
    print(f"特征矩阵: {X_train.shape}")
    
    # 训练HMM模型
    hmm = MarketRegimeHMM(n_components=3, n_iter=100)
    hmm.fit(X_train)
    
    # 预测状态
    states, probs = hmm.predict(X_train)
    
    # 评估模型
    eval_results = evaluate_model(hmm, X_train)
    
    # 获取当前状态
    current_regime = hmm.get_current_regime(X_train)
    
    print("\n" + "="*70)
    print("当前市场状态识别")
    print("="*70)
    print(f"状态: {current_regime['state_name']} (状态{current_regime['state']})")
    print(f"置信度: {current_regime['confidence']:.2%}")
    print("\n状态概率分布:")
    for name, prob in current_regime['probabilities'].items():
        bar = '#' * int(prob * 20)
        print(f"  {name:6s}: {prob:.2%} {bar}")
    
    # 策略建议
    strategy = StrategySelector.get_strategy(current_regime['state'])
    current_rsi = features['rsi'].iloc[-1]
    current_price = df['close'].iloc[-1]
    current_ma20 = df['close'].rolling(20).mean().iloc[-1]
    
    signal = StrategySelector.generate_signal(
        current_regime['state'], 
        current_rsi, 
        current_price, 
        current_ma20
    )
    
    print("\n" + "="*70)
    print("策略建议")
    print("="*70)
    print(f"推荐策略: {strategy['name']}")
    print(f"操作策略: {strategy['action']}")
    print(f"仓位建议: {strategy['position_size']*100:.0f}%")
    print(f"止损设置: {strategy['stop_loss']}")
    print(f"描述: {strategy['description']}")
    
    print("\n交易信号:")
    print(f"  动作: {signal['action']}")
    if 'reason' in signal:
        print(f"  原因: {signal['reason']}")
    
    print("\n" + "="*70)
    print("使用说明:")
    print("="*70)
    print("1. 准备真实市场数据(2017-2025年)")
    print("2. 调用 extract_features(df) 提取特征")
    print("3. 使用 MarketRegimeHMM 训练模型")
    print("4. 根据 get_current_regime() 结果切换策略")
    print("\n验证要求: 状态识别准确率 > 72%")
    print("="*70)

if __name__ == "__main__":
    main()