因子择时策略：动态因子暴露的量化体系

# 构建宏观经济状态指标
import pandas as pd
import numpy as np

# 1. 获取宏观经济数据
macro_data = pd.DataFrame({
    'gdp_growth': pd.read_csv('gdp_growth.csv', index_col=0, parse_dates=True),
    'inflation': pd.read_csv('inflation.csv', index_col=0, parse_dates=True),
    'interest_rate': pd.read_csv('interest_rate.csv', index_col=0, parse_dates=True),
    'credit_spread': pd.read_csv('credit_spread.csv', index_col=0, parse_dates=True)
})

# 2. 标准化并构建综合指标
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
macro_normalized = scaler.fit_transform(macro_data)

# 3. 用PCA提取主要宏观经济成分
from sklearn.decomposition import PCA
pca = PCA(n_components=2)
macro_pca = pca.fit_transform(macro_normalized)

# 4. 根据PCA得分划分经济状态
def classify_macro_state(pca_scores):
    """将PCA得分映射为经济状态"""
    state1, state2 = pca_scores[:, 0], pca_scores[:, 1]
    
    if state1 > 0 and state2 > 0:
        return "复苏期"
    elif state1 > 0 and state2 <= 0:
        return "过热期"
    elif state1 <= 0 and state2 > 0:
        return "滞胀期"
    else:
        return "衰退期"

macro_data['macro_state'] = classify_macro_state(macro_pca)

# 使用随机森林预测各因子未来收益
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import TimeSeriesSplit

# 1. 准备特征矩阵
features = pd.DataFrame({
    'macro_state': macro_data['macro_state'].map({
        '复苏期': 0, '过热期': 1, '滞胀期': 2, '衰退期': 3
    }),
    'vix': vix_data['close'],
    'yield_curve': yield_curve_data['10y-2y'],
    'credit_spread': macro_data['credit_spread'],
    'momentum_12m': calculate_momentum(returns, period=252)
})

# 2. 准备目标变量（各因子未来收益）
factor_returns = calculate_factor_returns(stock_data, factors=['value', 'momentum', 'quality'])

# 3. 时间序列交叉验证
tscv = TimeSeriesSplit(n_splits=5)

for factor in ['value', 'momentum', 'quality', 'low_vol']:
    y = factor_returns[f'{factor}_return'].shift(-1)  # 预测下一期收益
    
    model = RandomForestRegressor(n_estimators=100, random_state=42)
    
    for train_idx, test_idx in tscv.split(features):
        X_train, X_test = features.iloc[train_idx], features.iloc[test_idx]
        y_train, y_test = y.iloc[train_idx], y.iloc[test_idx]
        
        model.fit(X_train, y_train)
        predictions = model.predict(X_test)
        
    # 4. 根据预测结果调整因子权重
    factor_weight = predict_factor_weight(model, features_latest)

# 1. 设置因子权重调整约束
class FactorTimingOptimizer:
    def __init__(self, base_weights, max_turnover=0.1, min_weight=0.05):
        self.base_weights = base_weights  # 静态因子权重作为基准
        self.max_turnover = max_turnover  # 最大换手率约束
        self.min_weight = min_weight      # 最小因子暴露约束
        
    def optimize_weights(self, predicted_returns, cov_matrix):
        """在约束条件下优化因子权重"""
        n_factors = len(self.base_weights)
        
        # 使用CVXPY进行约束优化
        import cvxpy as cp
        
        weights = cp.Variable(n_factors)
        
        # 目标函数：最大化预测收益 - 风险惩罚
        objective = cp.Maximize(
            predicted_returns @ weights - 
            0.5 * cp.quad_form(weights, cov_matrix)
        )
        
        # 约束条件
        constraints = [
            cp.sum(weights) == 1,  # 全投资
            weights >= self.min_weight,  # 最小暴露
            cp.norm(weights - self.base_weights) <= self.max_turnover  # 换手率约束
        ]
        
        problem = cp.Problem(objective, constraints)
        problem.solve()
        
        return weights.value