aria-code 4.1.3__py3-none-any.whl

packages/quant_engine/stochastic/kelly_criterion.py

@@ -0,0 +1,181 @@
+"""
+Kelly Criterion Implementation
+包含：
+  1. 连续时间凯利公式 (Continuous Time)
+  2. 多资产矩阵形式 (Multi-Asset Matrix Form) — 含协方差收缩
+  3. 带方差不确定性的 Robust 凯利 (Fractional Kelly)
+  4. from_returns() — 直接从历史收益率矩阵构建 Kelly 权重 (LedoitWolf 收缩)
+
+修复 (v2.0):
+  - multi_asset_kelly: 默认启用 Tikhonov 正则化防止奇异协方差矩阵
+  - from_returns(): 优先使用 LedoitWolf 协方差估计，回退到样本协方差
+"""
+
+import numpy as np
+import pandas as pd
+from typing import Dict, List, Optional, Union
+from scipy.optimize import minimize
+
+# LedoitWolf 协方差收缩（可选）
+try:
+    from sklearn.covariance import LedoitWolf as _LedoitWolf
+    LEDOIT_WOLF_AVAILABLE = True
+except ImportError:
+    LEDOIT_WOLF_AVAILABLE = False
+
+
+class KellyCriterion:
+    """
+    凯利公式服务
+    目标：最大化财富对数增长率
+    """
+
+    @staticmethod
+    def continuous_time_kelly(mu: float, r: float, sigma: float) -> float:
+        """
+        标准连续时间凯利公式：
+        f* = (mu - r) / sigma^2
+        """
+        if sigma <= 0:
+            return 0.0
+        return (mu - r) / (sigma ** 2)
+
+    @staticmethod
+    def multi_asset_kelly(
+        mu_vec: np.ndarray,
+        rf: float,
+        cov_matrix: np.ndarray,
+        allow_short: bool = False,
+        max_leverage: float = 1.0,
+        shrinkage: bool = True,
+        shrinkage_alpha: Optional[float] = None,
+    ) -> np.ndarray:
+        """
+        多资产凯利公式 (矩阵形式):
+        f* = Σ^(-1) * (μ - rf)
+
+        Args:
+            mu_vec:          期望收益率向量 (n,)
+            rf:              无风险收益率
+            cov_matrix:      协方差矩阵 (n, n)
+            allow_short:     是否允许做空
+            max_leverage:    最大总杠杆
+            shrinkage:       True → Tikhonov 正则化 (cov + alpha*I)
+                             有效防止近奇异协方差矩阵带来的极端权重
+            shrinkage_alpha: 正则化强度；None → 自动选为 1% of trace/n
+        """
+        n = len(mu_vec)
+        cov = np.array(cov_matrix, dtype=float)
+
+        if shrinkage:
+            if shrinkage_alpha is None:
+                # 自动正则化强度：trace 的 1 %
+                shrinkage_alpha = max(np.trace(cov) / n * 0.01, 1e-6)
+            cov = cov + shrinkage_alpha * np.eye(n)
+
+        try:
+            excess_ret = np.array(mu_vec, dtype=float) - rf
+            cov_inv    = np.linalg.inv(cov)
+            kelly_f    = cov_inv @ excess_ret
+
+            # 约束
+            if not allow_short:
+                kelly_f = np.maximum(kelly_f, 0.0)
+
+            total_leverage = float(np.sum(kelly_f))
+            if total_leverage > max_leverage:
+                kelly_f = kelly_f * (max_leverage / total_leverage)
+
+            return kelly_f
+
+        except np.linalg.LinAlgError:
+            # 即使加了正则化还是奇异 → 退化为等权
+            return np.ones(n) / n
+
+    @classmethod
+    def from_returns(
+        cls,
+        returns_df: pd.DataFrame,
+        rf: float = 0.02,
+        allow_short: bool = False,
+        max_leverage: float = 1.0,
+        fraction: float = 1.0,
+        use_ledoit_wolf: bool = True,
+    ) -> Dict[str, float]:
+        """
+        直接从历史日收益率矩阵计算 Kelly 权重。
+
+        协方差估计优先级：
+          1. LedoitWolf 收缩估计（sklearn，若可用）
+          2. 样本协方差（兜底）
+
+        Args:
+            returns_df:      每列为一个资产的日收益率 DataFrame
+            rf:              年化无风险收益率（自动转为日化）
+            allow_short:     是否允许做空
+            max_leverage:    最大杠杆
+            fraction:        分段凯利系数（0 < f ≤ 1.0）
+            use_ledoit_wolf: 是否优先使用 LedoitWolf 收缩
+
+        Returns:
+            {symbol: weight} 字典
+        """
+        symbols = list(returns_df.columns)
+        R       = returns_df.dropna().values  # (T, n)
+        dt      = 1.0 / 252
+
+        mu_daily = R.mean(axis=0)
+        mu_ann   = mu_daily / dt          # 日化 → 年化
+        rf_daily = rf * dt
+
+        # -- 协方差矩阵 --
+        if use_ledoit_wolf and LEDOIT_WOLF_AVAILABLE:
+            lw  = _LedoitWolf().fit(R)
+            cov_daily = lw.covariance_           # (n, n)
+        else:
+            cov_daily = np.cov(R, rowvar=False)  # 样本协方差
+
+        cov_ann = cov_daily / dt  # 日化协方差 → 年化
+
+        weights = cls.multi_asset_kelly(
+            mu_vec      = mu_ann,
+            rf          = rf,
+            cov_matrix  = cov_ann,
+            allow_short = allow_short,
+            max_leverage= max_leverage,
+            shrinkage   = True,  # always apply Tikhonov on top
+        )
+
+        # 应用分段凯利
+        weights = weights * fraction
+
+        return dict(zip(symbols, weights.tolist()))
+
+    @staticmethod
+    def robust_log_kelly(
+        returns_df: pd.DataFrame,
+        rf: float = 0.02,
+        fraction: float = 0.5,
+    ) -> Dict[str, float]:
+        """
+        基于历史分布的非参数对数凯利 (含分段凯利/Fractional Kelly)
+        """
+        symbols      = returns_df.columns
+        n            = len(symbols)
+        hist_returns = returns_df.values
+        dt           = 1 / 252
+
+        def obj(w: np.ndarray) -> float:
+            port_rets = np.dot(hist_returns, w)
+            growth    = 1.0 + port_rets + (1.0 - np.sum(w)) * (rf * dt)
+            if np.any(growth <= 0):
+                return 1e10
+            return -float(np.mean(np.log(growth)))
+
+        cons   = [{"type": "ineq", "fun": lambda x: 1.0 - np.sum(x)}]
+        bounds = [(0, 1) for _ in range(n)]
+        res    = minimize(obj, np.ones(n) / n, bounds=bounds, constraints=cons)
+
+        # 应用分段凯利 (Robust 修正)
+        final_weights = res.x * fraction
+        return dict(zip(symbols, final_weights.tolist()))

packages/quant_engine/stochastic/monte_carlo_advanced.py

@@ -0,0 +1,95 @@
+"""
+Advanced Monte Carlo Engine with Variance Reduction Techniques
+包含对偶变量、控制变量、拟蒙特卡罗、重要性采样
+"""
+
+import numpy as np
+import scipy.stats as stats
+from typing import Optional, Callable, Tuple, List, Dict
+from enum import Enum
+
+class VarianceReduction(Enum):
+    NONE = "none"
+    ANTITHETIC = "antithetic"
+    CONTROL_VARIATE = "control_variate"
+    QUASI_MONTE_CARLO = "quasi_mc"
+    IMPORTANCE_SAMPLING = "importance_sampling"
+
+class MonteCarloEngine:
+    """
+    高级蒙特卡罗模拟引擎
+    支持：
+      1. 路径模拟 (Path Simulation)
+      2. 衍生品定价 (Derivative Pricing)
+      3. 风险度量 (Risk Metrics: VaR, CVaR)
+    """
+    
+    @staticmethod
+    def simulate_gbm(
+        s0: float, mu: float, sigma: float, T: float, 
+        n_steps: int, n_paths: int, 
+        method: VarianceReduction = VarianceReduction.NONE
+    ) -> np.ndarray:
+        """
+        GBM 路径模拟 - 带方差缩减选项
+        """
+        dt = T / n_steps
+        
+        if method == VarianceReduction.ANTITHETIC:
+            half_paths = n_paths // 2
+            dW = np.random.normal(0, np.sqrt(dt), (half_paths, n_steps))
+            dW_full = np.concatenate([dW, -dW], axis=0)
+        elif method == VarianceReduction.QUASI_MONTE_CARLO:
+            # 使用 Sobol 序列 (拟蒙特卡罗)
+            from scipy.stats import qmc
+            sampler = qmc.Sobol(d=n_steps, scramble=True)
+            u = sampler.random(n=n_paths)
+            z = stats.norm.ppf(u)
+            dW_full = z * np.sqrt(dt)
+        else:
+            dW_full = np.random.normal(0, np.sqrt(dt), (n_paths, n_steps))
+            
+        # 路径累加
+        log_returns = (mu - 0.5 * sigma**2) * dt + sigma * dW_full
+        cumulative_log_returns = np.cumsum(log_returns, axis=1)
+        # 加入起始点
+        paths = s0 * np.exp(np.hstack([np.zeros((n_paths, 1)), cumulative_log_returns]))
+        
+        return paths
+
+    @staticmethod
+    def price_european_option(
+        s0: float, K: float, T: float, r: float, sigma: float,
+        option_type: str = "call", n_paths: int = 100000,
+        reduction: VarianceReduction = VarianceReduction.ANTITHETIC
+    ) -> Dict[str, float]:
+        """
+        蒙特卡罗定价欧式期权
+        """
+        # 使用 1 步模拟即可（对欧式期权，中间路径不影响结果）
+        if reduction == VarianceReduction.ANTITHETIC:
+            z = np.random.normal(0, 1, n_paths // 2)
+            z_full = np.concatenate([z, -z])
+        else:
+            z_full = np.random.normal(0, 1, n_paths)
+            
+        st = s0 * np.exp((r - 0.5 * sigma**2) * T + sigma * np.sqrt(T) * z_full)
+        
+        if option_type == "call":
+            payoff = np.maximum(st - K, 0)
+        else:
+            payoff = np.maximum(K - st, 0)
+            
+        price = np.exp(-r * T) * np.mean(payoff)
+        std_err = np.exp(-r * T) * np.std(payoff) / np.sqrt(n_paths)
+        
+        return {"price": float(price), "standard_error": float(std_err)}
+    
+    @staticmethod
+    def calculate_var_cvar(returns: np.ndarray, confidence: float = 0.95) -> Tuple[float, float]:
+        """
+        计算历史/模拟收益率的 VaR 和 CVaR
+        """
+        var = np.percentile(returns, (1 - confidence) * 100)
+        cvar = returns[returns <= var].mean()
+        return float(var), float(cvar)