docs: add v1.13.0 portfolio management implementation plan

2026-09-12 14:34:18 +08:00 · 2026-06-12 20:17:43 +08:00
parent 303964ea67
commit 6b6b7c66a0
1 changed files with 956 additions and 0 deletions
@@ -0,0 +1,956 @@
+# v1.13.0 组合管理实施计划
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development or superpowers:executing-plans.
+
+**Goal:** 实现 portfolio/ 模块（权重优化器 + 风险模型 + 再平衡引擎），使 easy-tdx 具备因子选股→组合构建→绩效评估的完整闭环。
+
+**Architecture:** 分三层：optimizer.py（权重分配）→ risk.py（风险度量）→ rebalance.py（多期调仓回测）。每层可独立使用，RebalanceEngine 串联三者。依赖因子模块的 FactorEngine.compute_cross_section() 作为截面得分来源。
+
+**Tech Stack:** Python 3.10+, numpy, pandas, scipy（可选，MeanVariance 降级为等权）
+
+**Design Spec:** `docs/superpowers/specs/2026-06-12-quantitative-factor-engine-design.md` Section 4
+
+---
+
+## File Structure
+
+| Action | Path | Responsibility |
+|--------|------|----------------|
+| Create | `src/easy_tdx/portfolio/__init__.py` | 公开 API |
+| Create | `src/easy_tdx/portfolio/types.py` | PortfolioState, RebalanceResult |
+| Create | `src/easy_tdx/portfolio/optimizer.py` | WeightOptimizer ABC + 4 个优化器 + 注册表 |
+| Create | `src/easy_tdx/portfolio/risk.py` | RiskModel (协方差/组合风险) |
+| Create | `src/easy_tdx/portfolio/rebalance.py` | RebalanceEngine (多期调仓) |
+| Modify | `src/easy_tdx/cli/cmd_factor.py` | 添加 portfolio 子命令 |
+| Modify | `pyproject.toml` | bump → 1.13.0 |
+| Create | `tests/unit/test_portfolio_optimizer.py` | 优化器测试 |
+| Create | `tests/unit/test_portfolio_risk.py` | 风险模型测试 |
+| Create | `tests/unit/test_portfolio_rebalance.py` | 再平衡集成测试 |
+
+---
+
+### Task 1: portfolio/types.py + optimizer.py
+
+**Files:**
+- Create: `src/easy_tdx/portfolio/types.py`
+- Create: `src/easy_tdx/portfolio/optimizer.py`
+- Create: `src/easy_tdx/portfolio/__init__.py`（初始版本）
+- Test: `tests/unit/test_portfolio_optimizer.py`
+
+#### types.py
+
+```python
+"""组合管理数据结构。"""
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+import pandas as pd
+
+
+@dataclass
+class PortfolioState:
+    """组合状态快照。"""
+
+    date: int
+    weights: dict[str, float]
+    holdings: dict[str, float]
+    cash: float
+    total_value: float
+    positions_count: int
+
+
+@dataclass
+class RebalanceResult:
+    """再平衡结果。"""
+
+    rebalance_dates: list[int]
+    states: list[PortfolioState]
+    trades: pd.DataFrame
+    equity_curve: pd.DataFrame
+    performance: dict[str, float]
+```
+
+#### optimizer.py
+
+```python
+"""权重优化器。"""
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+
+import numpy as np
+import pandas as pd
+
+
+class WeightOptimizer(ABC):
+    """权重优化器基类。"""
+
+    @abstractmethod
+    def optimize(
+        self,
+        factor_scores: pd.DataFrame,  # columns: [code, score]
+        n_stocks: int = 50,
+        **kwargs: object,
+    ) -> dict[str, float]:
+        """返回 {code: weight}，权重和为 1.0。"""
+        ...
+
+
+_OPTIMIZER_REGISTRY: dict[str, type[WeightOptimizer]] = {}
+
+
+def register_optimizer(name: str) -> type[WeightOptimizer]:
+    """注册优化器。"""
+    def wrapper(cls: type[WeightOptimizer]) -> type[WeightOptimizer]:
+        _OPTIMIZER_REGISTRY[name] = cls
+        return cls
+    return wrapper
+
+
+def get_optimizer(name: str) -> WeightOptimizer:
+    """按名称获取优化器实例。"""
+    if name not in _OPTIMIZER_REGISTRY:
+        raise ValueError(f"未知优化器: {name!r}。可用: {sorted(_OPTIMIZER_REGISTRY.keys())}")
+    return _OPTIMIZER_REGISTRY[name]()
+
+
+@register_optimizer("equal")
+class EqualWeightOptimizer(WeightOptimizer):
+    """等权 — 取 top-N 等权分配。"""
+
+    def optimize(
+        self,
+        factor_scores: pd.DataFrame,
+        n_stocks: int = 50,
+        **kwargs: object,
+    ) -> dict[str, float]:
+        top = factor_scores.nlargest(n_stocks, "score")
+        if len(top) == 0:
+            return {}
+        w = 1.0 / len(top)
+        return {row["code"]: w for _, row in top.iterrows()}
+
+
+@register_optimizer("factor_weighted")
+class FactorWeightedOptimizer(WeightOptimizer):
+    """因子加权 — 按因子得分加权。"""
+
+    def optimize(
+        self,
+        factor_scores: pd.DataFrame,
+        n_stocks: int = 50,
+        **kwargs: object,
+    ) -> dict[str, float]:
+        top = factor_scores.nlargest(n_stocks, "score")
+        if len(top) == 0:
+            return {}
+        scores = top["score"].to_numpy(dtype=np.float64)
+        # 截断极端值：top/bottom 5% 拉回到分位数
+        if len(scores) > 10:
+            q95 = np.percentile(scores, 95)
+            q05 = np.percentile(scores, 5)
+            scores = np.clip(scores, q05, q95)
+        # 确保正值
+        scores = scores - scores.min() + 1e-8
+        total = scores.sum()
+        if total == 0:
+            w = 1.0 / len(top)
+            return {row["code"]: w for _, row in top.iterrows()}
+        weights = scores / total
+        return {row["code"]: float(weights[i]) for i, (_, row) in enumerate(top.iterrows())}
+
+
+@register_optimizer("risk_parity")
+class RiskParityOptimizer(WeightOptimizer):
+    """风险平价 — 每只股票贡献相等风险。需要 volatility 列。"""
+
+    def optimize(
+        self,
+        factor_scores: pd.DataFrame,
+        n_stocks: int = 50,
+        **kwargs: object,
+    ) -> dict[str, float]:
+        top = factor_scores.nlargest(n_stocks, "score")
+        if len(top) == 0:
+            return {}
+
+        # 用 score 的倒数作为波动率代理（score 越高 = 波动率越低 = 权重越大）
+        # 或如果提供了 volatility 列，直接使用
+        if "volatility" in top.columns:
+            vol = top["volatility"].to_numpy(dtype=np.float64)
+        else:
+            # 用 score 均值/绝对score 作为稳定性代理
+            scores = top["score"].abs().to_numpy(dtype=np.float64)
+            vol = 1.0 / (scores + 1e-8)
+
+        vol = np.maximum(vol, 1e-8)
+        inv_vol = 1.0 / vol
+        total = inv_vol.sum()
+        weights = inv_vol / total
+
+        return {row["code"]: float(weights[i]) for i, (_, row) in enumerate(top.iterrows())}
+
+
+@register_optimizer("mean_variance")
+class MeanVarianceOptimizer(WeightOptimizer):
+    """均值方差优化 — Markowitz 模型（可选 scipy）。"""
+
+    def optimize(
+        self,
+        factor_scores: pd.DataFrame,
+        n_stocks: int = 50,
+        **kwargs: object,
+    ) -> dict[str, float]:
+        # 尝试使用 scipy，降级到等权
+        try:
+            from scipy.optimize import minimize
+            return self._optimize_with_scipy(factor_scores, n_stocks, minimize)
+        except ImportError:
+            # scipy 不可用，降级为等权
+            fallback = EqualWeightOptimizer()
+            return fallback.optimize(factor_scores, n_stocks)
+
+    def _optimize_with_scipy(
+        self,
+        factor_scores: pd.DataFrame,
+        n_stocks: int,
+        minimize: object,
+    ) -> dict[str, float]:
+        from scipy.optimize import minimize as _minimize  # type: ignore[import]
+
+        top = factor_scores.nlargest(n_stocks, "score")
+        if len(top) == 0:
+            return {}
+
+        n = len(top)
+        scores = top["score"].to_numpy(dtype=np.float64)
+
+        # 用 score 构造简化协方差矩阵（对角矩阵，方差 = 1/score²）
+        variances = 1.0 / (np.abs(scores) + 1e-8) ** 2
+        cov = np.diag(variances)
+
+        # 目标：最小化 w' Σ w，约束 sum(w)=1, 0<=w<=0.1
+        def objective(w: np.ndarray) -> float:
+            return float(w @ cov @ w)
+
+        constraints = {"type": "eq", "fun": lambda w: float(np.sum(w) - 1.0)}
+        bounds = [(0.0, 0.1)] * n
+        x0 = np.ones(n) / n
+
+        result = _minimize(objective, x0, method="SLSQP", bounds=bounds, constraints=constraints)
+
+        if result.success:
+            weights = result.x
+        else:
+            weights = np.ones(n) / n
+
+        return {row["code"]: float(weights[i]) for i, (_, row) in enumerate(top.iterrows())}
+```
+
+#### 测试
+
+```python
+# tests/unit/test_portfolio_optimizer.py
+"""Test portfolio optimizers."""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+import pytest
+
+from easy_tdx.portfolio.optimizer import (
+    EqualWeightOptimizer,
+    FactorWeightedOptimizer,
+    RiskParityOptimizer,
+    get_optimizer,
+    register_optimizer,
+)
+
+
+def _make_scores(n: int = 20, seed: int = 42) -> pd.DataFrame:
+    rng = np.random.default_rng(seed)
+    return pd.DataFrame({
+        "code": [f"{i:06d}" for i in range(n)],
+        "score": rng.normal(0.02, 0.05, n),
+    })
+
+
+class TestEqualWeight:
+    def test_weights_sum_to_one(self):
+        opt = EqualWeightOptimizer()
+        w = opt.optimize(_make_scores(), n_stocks=10)
+        assert abs(sum(w.values()) - 1.0) < 1e-8
+
+    def test_n_stocks_selected(self):
+        opt = EqualWeightOptimizer()
+        w = opt.optimize(_make_scores(), n_stocks=5)
+        assert len(w) == 5
+
+    def test_all_equal(self):
+        opt = EqualWeightOptimizer()
+        w = opt.optimize(_make_scores(), n_stocks=10)
+        values = list(w.values())
+        assert all(abs(v - values[0]) < 1e-8 for v in values)
+
+    def test_empty_input(self):
+        opt = EqualWeightOptimizer()
+        w = opt.optimize(pd.DataFrame(columns=["code", "score"]), n_stocks=5)
+        assert len(w) == 0
+
+
+class TestFactorWeighted:
+    def test_weights_sum_to_one(self):
+        opt = FactorWeightedOptimizer()
+        w = opt.optimize(_make_scores(), n_stocks=10)
+        assert abs(sum(w.values()) - 1.0) < 1e-6
+
+    def test_higher_score_higher_weight(self):
+        scores = pd.DataFrame({"code": ["A", "B", "C"], "score": [3.0, 2.0, 1.0]})
+        opt = FactorWeightedOptimizer()
+        w = opt.optimize(scores, n_stocks=3)
+        assert w["A"] > w["C"]
+
+
+class TestRiskParity:
+    def test_weights_sum_to_one(self):
+        opt = RiskParityOptimizer()
+        w = opt.optimize(_make_scores(), n_stocks=10)
+        assert abs(sum(w.values()) - 1.0) < 1e-6
+
+    def test_with_volatility_column(self):
+        scores = pd.DataFrame({
+            "code": ["A", "B", "C"],
+            "score": [1.0, 1.0, 1.0],
+            "volatility": [0.1, 0.2, 0.4],
+        })
+        opt = RiskParityOptimizer()
+        w = opt.optimize(scores, n_stocks=3)
+        # 低波动率 → 更高权重
+        assert w["A"] > w["C"]
+
+
+class TestRegistry:
+    def test_get_optimizer(self):
+        opt = get_optimizer("equal")
+        assert isinstance(opt, EqualWeightOptimizer)
+
+    def test_unknown_raises(self):
+        with pytest.raises(ValueError, match="未知优化器"):
+            get_optimizer("nonexistent")
+```
+
+#### __init__.py（初始）
+
+```python
+"""组合管理模块。"""
+```
+
+提交: `feat(portfolio): add optimizer module with 4 strategies (equal/factor_weighted/risk_parity/mean_variance)`
+
+---
+
+### Task 2: portfolio/risk.py
+
+```python
+# src/easy_tdx/portfolio/risk.py
+"""简化风险模型。"""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+
+class RiskModel:
+    """简化风险模型 — A 股够用。"""
+
+    def estimate_covariance(
+        self,
+        returns: pd.DataFrame,
+        method: str = "shrinkage",
+        shrinkage_intensity: float = 0.5,
+        window: int = 60,
+    ) -> pd.DataFrame:
+        """协方差矩阵估计。
+
+        Args:
+            returns: columns=code, index=date 的收益率 DataFrame。
+            method: "shrinkage" | "sample"
+        """
+        if len(returns) < 2:
+            codes = returns.columns.tolist() if len(returns.columns) > 0 else []
+            return pd.DataFrame(np.eye(len(codes)), index=codes, columns=codes)
+
+        # 只用最近 window 行
+        if len(returns) > window:
+            returns = returns.iloc[-window:]
+
+        sample_cov = returns.cov()
+
+        if method == "shrinkage":
+            # Ledoit-Wolf 简化版：收缩到对角矩阵
+            target = pd.DataFrame(
+                np.diag(np.diag(sample_cov.to_numpy())),
+                index=sample_cov.index,
+                columns=sample_cov.columns,
+            )
+            shrunk = (1 - shrinkage_intensity) * sample_cov + shrinkage_intensity * target
+            return shrunk
+
+        return sample_cov
+
+    def portfolio_risk(
+        self,
+        weights: dict[str, float],
+        cov_matrix: pd.DataFrame,
+    ) -> dict[str, float]:
+        """组合风险指标。
+
+        Returns:
+            total_volatility, max_risk_contribution, n_positions
+        """
+        codes = [c for c in weights if c in cov_matrix.columns]
+        if not codes:
+            return {"total_volatility": 0.0, "max_risk_contribution": 0.0, "n_positions": 0}
+
+        w = np.array([weights[c] for c in codes])
+        cov_sub = cov_matrix.loc[codes, codes].to_numpy()
+
+        # 总波动率（年化）
+        var = float(w @ cov_sub @ w)
+        total_vol = np.sqrt(max(0, var)) * np.sqrt(252)
+
+        # 边际风险贡献
+        marginal = cov_sub @ w
+        risk_contrib = np.abs(w * marginal)
+        total_rc = risk_contrib.sum()
+        max_rc = float(risk_contrib.max() / total_rc) if total_rc > 0 else 0.0
+
+        return {
+            "total_volatility": total_vol,
+            "max_risk_contribution": max_rc,
+            "n_positions": len(codes),
+        }
+```
+
+测试:
+
+```python
+# tests/unit/test_portfolio_risk.py
+"""Test RiskModel."""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+from easy_tdx.portfolio.risk import RiskModel
+
+
+def _make_returns(n_dates: int = 100, n_stocks: int = 5, seed: int = 42) -> pd.DataFrame:
+    rng = np.random.default_rng(seed)
+    codes = [f"{i:06d}" for i in range(n_stocks)]
+    data = rng.normal(0.001, 0.02, (n_dates, n_stocks))
+    return pd.DataFrame(data, columns=codes)
+
+
+class TestCovarianceEstimation:
+    def test_shape(self):
+        rm = RiskModel()
+        ret = _make_returns()
+        cov = rm.estimate_covariance(ret)
+        assert cov.shape == (5, 5)
+
+    def test_symmetric(self):
+        rm = RiskModel()
+        ret = _make_returns()
+        cov = rm.estimate_covariance(ret)
+        assert np.allclose(cov.to_numpy(), cov.to_numpy().T)
+
+    def test_shrinkage_vs_sample(self):
+        rm = RiskModel()
+        ret = _make_returns()
+        shrunk = rm.estimate_covariance(ret, method="shrinkage")
+        sample = rm.estimate_covariance(ret, method="sample")
+        # 收缩矩阵的非对角元素应更小
+        off_diag_shrunk = shrunk.values[~np.eye(len(shrunk), dtype=bool)]
+        off_diag_sample = sample.values[~np.eye(len(sample), dtype=bool)]
+        assert np.abs(off_diag_shrunk).mean() <= np.abs(off_diag_sample).mean()
+
+
+class TestPortfolioRisk:
+    def test_total_volatility(self):
+        rm = RiskModel()
+        ret = _make_returns()
+        cov = rm.estimate_covariance(ret)
+        weights = {"000000": 0.5, "000001": 0.5}
+        risk = rm.portfolio_risk(weights, cov)
+        assert risk["total_volatility"] > 0
+        assert risk["n_positions"] == 2
+
+    def test_empty_weights(self):
+        rm = RiskModel()
+        cov = pd.DataFrame()
+        risk = rm.portfolio_risk({}, cov)
+        assert risk["total_volatility"] == 0.0
+```
+
+提交: `feat(portfolio): add RiskModel with shrinkage covariance and portfolio risk decomposition`
+
+---
+
+### Task 3: portfolio/rebalance.py — 再平衡引擎
+
+```python
+# src/easy_tdx/portfolio/rebalance.py
+"""多期调仓回测引擎。"""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+from easy_tdx.factor.engine import FactorEngine
+from easy_tdx.portfolio.optimizer import WeightOptimizer
+from easy_tdx.portfolio.types import PortfolioState, RebalanceResult
+
+
+class RebalanceEngine:
+    """多期调仓回测引擎。
+
+    管道: FactorEngine → 截面得分 → Optimizer → 目标权重 → 交易成本 → 持仓跟踪
+    """
+
+    def __init__(
+        self,
+        optimizer: WeightOptimizer,
+        factor_name: str = "momentum_20d",
+        n_stocks: int = 50,
+        rebalance_freq: str = "M",
+        commission: float = 0.0003,
+        slippage: float = 0.001,
+        cash: float = 1_000_000,
+    ) -> None:
+        self._optimizer = optimizer
+        self._factor_name = factor_name
+        self._n_stocks = n_stocks
+        self._rebalance_freq = rebalance_freq
+        self._commission = commission
+        self._slippage = slippage
+        self._cash = cash
+
+    def _get_rebalance_dates(self, dates: pd.DatetimeIndex) -> list[int]:
+        """根据频率确定调仓日期。"""
+        freq_map = {"W": "W-MON", "M": "ME", "Q": "QE"}
+        freq = freq_map.get(self._rebalance_freq, "ME")
+
+        series = pd.Series(dates, index=dates)
+        grouped = series.groupby(series.dt.to_period(freq))
+        rebalance_dates = [group.iloc[-1] for _, group in grouped if len(group) > 0]
+        return [int(d.strftime("%Y%m%d")) for d in rebalance_dates]
+
+    def run(
+        self,
+        data: dict[str, pd.DataFrame],
+        start_date: int | None = None,
+        end_date: int | None = None,
+    ) -> RebalanceResult:
+        """执行多期回测。"""
+        if not data:
+            return self._empty_result()
+
+        factor_engine = FactorEngine()
+
+        # 收集所有日期
+        all_dates: list[pd.Timestamp] = []
+        for df in data.values():
+            if "datetime" in df.columns:
+                all_dates.extend(df["datetime"].tolist())
+        if not all_dates:
+            return self._empty_result()
+
+        all_dates = sorted(set(all_dates))
+        date_ints = [int(d.strftime("%Y%m%d")) for d in all_dates]
+
+        # 日期范围过滤
+        if start_date:
+            all_dates = [d for d, di in zip(all_dates, date_ints) if di >= start_date]
+        if end_date:
+            all_dates = [d for d in all_dates if int(d.strftime("%Y%m%d")) <= end_date]
+
+        if not all_dates:
+            return self._empty_result()
+
+        # 确定调仓日期
+        dt_index = pd.DatetimeIndex(all_dates)
+        rebalance_dates = self._get_rebalance_dates(dt_index)
+        rebalance_set = set(rebalance_dates)
+
+        # 模拟
+        cash = self._cash
+        holdings: dict[str, float] = {}  # code -> shares
+        states: list[PortfolioState] = []
+        trades_list: list[dict[str, object]] = []
+        equity_records: list[dict[str, object]] = []
+
+        for dt in all_dates:
+            date_int = int(dt.strftime("%Y%m%d"))
+            is_rebalance = date_int in rebalance_set
+
+            # 获取当天各股票价格
+            prices: dict[str, float] = {}
+            for code, df in data.items():
+                if "datetime" in df.columns:
+                    row = df[df["datetime"] == dt]
+                    if not row.empty:
+                        prices[code] = float(row["close"].iloc[0])
+
+            # 计算持仓市值
+            position_value = sum(holdings.get(c, 0) * prices.get(c, 0) for c in holdings)
+            total_value = cash + position_value
+
+            if is_rebalance and total_value > 0:
+                # 计算截面因子得分
+                scores_df = factor_engine.compute_cross_section(
+                    data, [self._factor_name], date=date_int
+                )
+
+                if not scores_df.empty and scores_df[self._factor_name].notna().any():
+                    scores_df = scores_df.rename(columns={self._factor_name: "score"})
+                    scores_df = scores_df[["code", "score"]].dropna(subset=["score"])
+
+                    # 优化权重
+                    target_weights = self._optimizer.optimize(scores_df, n_stocks=self._n_stocks)
+                else:
+                    target_weights = {}
+
+                # 执行调仓
+                if target_weights:
+                    trades_list, cash, holdings = self._rebalance(
+                        target_weights, prices, total_value, date_int, trades_list, cash, holdings
+                    )
+
+            # 记录状态
+            position_value = sum(holdings.get(c, 0) * prices.get(c, 0) for c in holdings)
+            total_value = cash + position_value
+            weights = {}
+            if total_value > 0:
+                for c in holdings:
+                    if holdings[c] > 0 and c in prices:
+                        weights[c] = holdings[c] * prices[c] / total_value
+
+            states.append(PortfolioState(
+                date=date_int,
+                weights=weights,
+                holdings=dict(holdings),
+                cash=cash,
+                total_value=total_value,
+                positions_count=len([s for s in holdings.values() if s > 0]),
+            ))
+
+            equity_records.append({
+                "datetime": date_int,
+                "total": total_value,
+                "cash": cash,
+                "position_value": position_value,
+            })
+
+        # 构建结果
+        equity_curve = pd.DataFrame(equity_records)
+        trades_df = pd.DataFrame(trades_list) if trades_list else pd.DataFrame(
+            columns=["datetime", "direction", "code", "shares", "price", "cost"]
+        )
+
+        performance = self._compute_performance(equity_curve)
+
+        return RebalanceResult(
+            rebalance_dates=rebalance_dates,
+            states=states,
+            trades=trades_df,
+            equity_curve=equity_curve,
+            performance=performance,
+        )
+
+    def _rebalance(
+        self,
+        target_weights: dict[str, float],
+        prices: dict[str, float],
+        total_value: float,
+        date_int: int,
+        trades_list: list[dict[str, object]],
+        cash: float,
+        holdings: dict[str, float],
+    ) -> tuple[list[dict[str, object]], float, dict[str, float]]:
+        """执行一次调仓。"""
+        # 先卖出不在目标组合中的持仓
+        for code in list(holdings.keys()):
+            if code not in target_weights and holdings[code] > 0:
+                price = prices.get(code, 0)
+                if price > 0:
+                    sell_value = holdings[code] * price
+                    cost = sell_value * (self._commission + self._slippage)
+                    cash += sell_value - cost
+                    trades_list.append({
+                        "datetime": date_int,
+                        "direction": "SELL",
+                        "code": code,
+                        "shares": holdings[code],
+                        "price": price,
+                        "cost": cost,
+                    })
+                del holdings[code]
+
+        # 调整目标持仓
+        new_holdings: dict[str, float] = {}
+        total_cost = 0.0
+        for code, weight in target_weights.items():
+            price = prices.get(code, 0)
+            if price <= 0:
+                continue
+            target_value = total_value * weight
+            shares = int(target_value / price / 100) * 100  # 100 股整手
+            if shares > 0:
+                new_holdings[code] = shares
+                trade_value = shares * price
+                cost = trade_value * (self._commission + self._slippage)
+                total_cost += trade_value + cost
+                trades_list.append({
+                    "datetime": date_int,
+                    "direction": "BUY",
+                    "code": code,
+                    "shares": shares,
+                    "price": price,
+                    "cost": cost,
+                })
+
+        cash = total_value - sum(
+            new_holdings.get(c, 0) * prices.get(c, 0) for c in new_holdings
+        )
+        holdings.clear()
+        holdings.update(new_holdings)
+
+        return trades_list, cash, holdings
+
+    def _compute_performance(self, equity_curve: pd.DataFrame) -> dict[str, float]:
+        """计算绩效指标。"""
+        if len(equity_curve) < 2:
+            return {"total_return": 0.0, "annual_return": 0.0, "max_drawdown": 0.0, "sharpe": 0.0}
+
+        total = equity_curve["total"].to_numpy()
+        total_return = (total[-1] / total[0]) - 1
+
+        # 年化收益
+        n_days = len(total)
+        annual_return = (1 + total_return) ** (252 / max(n_days, 1)) - 1
+
+        # 最大回撤
+        peak = np.maximum.accumulate(total)
+        drawdown = (total - peak) / peak
+        max_drawdown = float(np.min(drawdown))
+
+        # 夏普比率
+        daily_ret = np.diff(total) / total[:-1]
+        daily_ret = daily_ret[~np.isnan(daily_ret)]
+        if len(daily_ret) > 1 and np.std(daily_ret) > 0:
+            sharpe = float(np.mean(daily_ret) / np.std(daily_ret) * np.sqrt(252))
+        else:
+            sharpe = 0.0
+
+        return {
+            "total_return": total_return,
+            "annual_return": annual_return,
+            "max_drawdown": max_drawdown,
+            "sharpe": sharpe,
+            "total_trades": len(equity_curve),
+        }
+
+    def _empty_result(self) -> RebalanceResult:
+        return RebalanceResult(
+            rebalance_dates=[],
+            states=[],
+            trades=pd.DataFrame(columns=["datetime", "direction", "code", "shares", "price", "cost"]),
+            equity_curve=pd.DataFrame(columns=["datetime", "total", "cash", "position_value"]),
+            performance={"total_return": 0.0, "annual_return": 0.0, "max_drawdown": 0.0, "sharpe": 0.0},
+        )
+```
+
+测试:
+
+```python
+# tests/unit/test_portfolio_rebalance.py
+"""Test RebalanceEngine."""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+import pytest
+
+from easy_tdx.portfolio.optimizer import EqualWeightOptimizer, FactorWeightedOptimizer
+from easy_tdx.portfolio.rebalance import RebalanceEngine
+
+
+def _make_market(n_stocks: int = 10, n_days: int = 120, seed: int = 42) -> dict[str, pd.DataFrame]:
+    """模拟一个微型市场。"""
+    rng = np.random.default_rng(seed)
+    data = {}
+    for i in range(n_stocks):
+        close = 10.0 + np.cumsum(rng.normal(0.01, 0.5, n_days))
+        close = np.maximum(close, 1.0)
+        data[f"{i:06d}"] = pd.DataFrame({
+            "datetime": pd.date_range("2024-01-01", periods=n_days, freq="D"),
+            "open": close,
+            "high": close + 0.3,
+            "low": close - 0.3,
+            "close": close,
+            "vol": rng.integers(1e5, 1e7, n_days).astype(float),
+            "amount": close * 1e6,
+        })
+    return data
+
+
+class TestRebalanceEngine:
+    def test_basic_run(self):
+        engine = RebalanceEngine(
+            optimizer=EqualWeightOptimizer(),
+            factor_name="momentum_20d",
+            n_stocks=5,
+            rebalance_freq="M",
+            cash=1_000_000,
+        )
+        data = _make_market()
+        result = engine.run(data, start_date=20240101, end_date=20240430)
+
+        assert len(result.states) > 0
+        assert len(result.rebalance_dates) > 0
+        assert len(result.equity_curve) > 0
+        assert "total_return" in result.performance
+
+    def test_with_factor_weighted(self):
+        engine = RebalanceEngine(
+            optimizer=FactorWeightedOptimizer(),
+            factor_name="momentum_20d",
+            n_stocks=5,
+            rebalance_freq="M",
+        )
+        data = _make_market()
+        result = engine.run(data, start_date=20240101, end_date=20240430)
+        assert len(result.states) > 0
+
+    def test_empty_data(self):
+        engine = RebalanceEngine(optimizer=EqualWeightOptimizer())
+        result = engine.run({})
+        assert result.performance["total_return"] == 0.0
+
+    def test_equity_curve_monotonic_dates(self):
+        engine = RebalanceEngine(optimizer=EqualWeightOptimizer(), rebalance_freq="M")
+        data = _make_market()
+        result = engine.run(data, start_date=20240101, end_date=20240430)
+        dates = result.equity_curve["datetime"].tolist()
+        assert dates == sorted(dates)
+
+    def test_trades_recorded(self):
+        engine = RebalanceEngine(
+            optimizer=EqualWeightOptimizer(),
+            n_stocks=3,
+            rebalance_freq="M",
+        )
+        data = _make_market()
+        result = engine.run(data, start_date=20240101, end_date=20240430)
+        assert len(result.trades) > 0
+        assert "BUY" in result.trades["direction"].values
+```
+
+提交: `feat(portfolio): add RebalanceEngine with multi-period portfolio backtesting`
+
+---
+
+### Task 4: portfolio/__init__.py + CLI + 版本号
+
+#### __init__.py
+
+```python
+"""组合管理模块。"""
+from easy_tdx.portfolio.optimizer import (
+    EqualWeightOptimizer,
+    FactorWeightedOptimizer,
+    MeanVarianceOptimizer,
+    RiskParityOptimizer,
+    WeightOptimizer,
+    get_optimizer,
+)
+from easy_tdx.portfolio.rebalance import RebalanceEngine
+from easy_tdx.portfolio.risk import RiskModel
+from easy_tdx.portfolio.types import PortfolioState, RebalanceResult
+
+__all__ = [
+    "WeightOptimizer",
+    "EqualWeightOptimizer",
+    "FactorWeightedOptimizer",
+    "RiskParityOptimizer",
+    "MeanVarianceOptimizer",
+    "get_optimizer",
+    "RiskModel",
+    "RebalanceEngine",
+    "PortfolioState",
+    "RebalanceResult",
+]
+```
+
+#### CLI: 在 cli/__init__.py 注册 portfolio 命令
+
+先检查是否已有 portfolio 命令（backtest 模块有一个 portfolio 子命令）。如果有，跳过 CLI 修改，只更新 portfolio/__init__.py。
+
+如果没有冲突，创建 src/easy_tdx/cli/cmd_portfolio.py:
+
+```python
+"""组合管理 CLI 命令。"""
+from __future__ import annotations
+
+import click
+
+
+@click.group("portfolio")
+def portfolio() -> None:
+    """组合管理工具。"""
+    pass
+
+
+@portfolio.command("backtest")
+@click.argument("factor_name")
+@click.option("--n-stocks", default=50, type=int, help="持仓数量")
+@click.option("--rebalance-freq", default="M", help="调仓频率: W/M/Q")
+@click.option("--optimizer", "opt_name", default="equal", help="优化器: equal/factor_weighted/risk_parity/mean_variance")
+@click.option("--cash", default=1000000.0, type=float, help="初始资金")
+def portfolio_backtest(factor_name: str, n_stocks: int, rebalance_freq: str, opt_name: str, cash: float) -> None:
+    """运行组合回测。
+
+    示例：
+
+      easy-tdx portfolio backtest momentum_20d
+
+      easy-tdx portfolio backtest rsi_14 --n-stocks 10 --optimizer factor_weighted
+    """
+    import json
+    click.echo(json.dumps({
+        "message": "portfolio backtest 需要行情数据，请使用 Python API",
+        "example": (
+            f"from easy_tdx.portfolio import RebalanceEngine, EqualWeightOptimizer\n"
+            f"from easy_tdx.factor import FactorEngine\n"
+            f"engine = RebalanceEngine(\n"
+            f"    optimizer=EqualWeightOptimizer(),\n"
+            f"    factor_name='{factor_name}',\n"
+            f"    n_stocks={n_stocks},\n"
+            f"    rebalance_freq='{rebalance_freq}',\n"
+            f"    cash={cash},\n"
+            f")\n"
+            f"result = engine.run(data, start_date=20230101, end_date=20240101)\n"
+            f"print(f'年化收益={{result.performance[\"annual_return\"]:.2%}}')"
+        ),
+    }, ensure_ascii=False, indent=2))
+```
+
+在 cli/__init__.py 中导入并注册。
+
+#### pyproject.toml: version = "1.13.0"
+
+#### 运行全量测试 + ruff + 提交
+
+提交: `feat(portfolio): add portfolio module exports, CLI command, bump v1.13.0`