fix(backtest): max drawdown now correctly measures peak-to-trough percentage

Previous formula was: max(absolute_drawdown) / initial_capital, which
exceeds 100% when the portfolio grows then drops (e.g. from 600k to 300k
on a 100k initial = 300% drawdown, which is nonsensical).

Fixed to use drawdown_pct (drawdown / peak) which is always in [0, 1].
This correctly measures the maximum percentage drop from the highest
equity peak, matching the standard financial definition.

Also added regression test: test_max_drawdown_never_exceeds_100_pct.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/easy_tdx/backtest/engine.py

@@ -289,7 +289,7 @@ class BacktestEngine:
             BacktestResult with empty DataFrames
         """
         perf = PerformanceAnalyzer(
-            pd.DataFrame(columns=["total", "drawdown"]),
+            pd.DataFrame(columns=["total", "drawdown", "drawdown_pct"]),
             pd.DataFrame(columns=["direction", "pnl", "rejected"]),
         ).compute()
 

src/easy_tdx/backtest/performance.py

@@ -94,9 +94,9 @@ class PerformanceAnalyzer:
         n = len(daily_ret)
         annual_return = (1 + total_return) ** (self.ANNUAL_DAYS / n) - 1
 
-        # 3. 最大回撤
+        # 3. 最大回撤（从峰值的最大跌幅百分比，0~1 之间）
-        max_drawdown_value = np.max(drawdown)
+        drawdown_pct = self._equity_curve["drawdown_pct"].to_numpy()
-        max_drawdown = max_drawdown_value / total[0] if total[0] != 0 else 0
+        max_drawdown = float(np.max(drawdown_pct))
 
         # 4. 最大回撤持续时间
         max_dd_duration = self._compute_max_dd_duration(total, drawdown)

tests/unit/test_backtest_performance.py

@@ -31,11 +31,13 @@ def _make_equity_curve(n: int = 252, total_return: float = 0.1) -> pd.DataFrame:
     # 计算回撤
     peak = np.maximum.accumulate(total)
     drawdown = peak - total
+    drawdown_pct = np.divide(drawdown, peak, out=np.zeros_like(drawdown), where=(peak != 0))
 
     return pd.DataFrame({
         "datetime": np.arange(n),
         "total": total,
         "drawdown": drawdown,
+        "drawdown_pct": drawdown_pct,
     })
 
 
@@ -65,6 +67,35 @@ def test_total_return() -> None:
     assert abs(metrics["total_return"] - 0.1) < 0.01
 
 
+def test_max_drawdown_never_exceeds_100_pct() -> None:
+    """测试最大回撤永远不超过 100%（从峰值的跌幅）。"""
+    # 模拟先涨 5 倍再腰斩的资金曲线
+    total = np.concatenate([
+        np.linspace(100000, 600000, 126),   # 涨到 60 万
+        np.linspace(600000, 300000, 126),   # 跌到 30 万
+    ])
+    peak = np.maximum.accumulate(total)
+    drawdown = peak - total
+    drawdown_pct = np.divide(drawdown, peak, out=np.zeros_like(drawdown), where=(peak != 0))
+
+    equity = pd.DataFrame({
+        "datetime": np.arange(252),
+        "total": total,
+        "drawdown": drawdown,
+        "drawdown_pct": drawdown_pct,
+    })
+    trades = _make_trades()
+
+    analyzer = PerformanceAnalyzer(equity, trades)
+    metrics = analyzer.compute()
+
+    # 最大回撤 = 从峰值跌 50%（30万 / 60万），不应超过 1.0
+    assert 0.0 <= metrics["max_drawdown"] <= 1.0, (
+        f"max_drawdown = {metrics['max_drawdown']:.2%}, should be in [0, 100%]"
+    )
+    assert abs(metrics["max_drawdown"] - 0.5) < 0.01
+
+
 def test_max_drawdown_zero_when_monotonic() -> None:
     """测试单调递增时最大回撤接近 0。"""
     equity = _make_equity_curve(n=252, total_return=0.1)
@@ -163,7 +194,7 @@ def test_all_keys_present() -> None:
 
 def test_empty_equity_curve() -> None:
     """测试空资金曲线返回全零指标。"""
-    equity = pd.DataFrame({"total": [], "drawdown": []})
+    equity = pd.DataFrame({"total": [], "drawdown": [], "drawdown_pct": []})
     trades = _make_trades()
 
     analyzer = PerformanceAnalyzer(equity, trades)
@@ -175,7 +206,7 @@ def test_empty_equity_curve() -> None:
 
 def test_single_point_equity_curve() -> None:
     """测试只有一个点的资金曲线返回全零指标。"""
-    equity = pd.DataFrame({"total": [100000], "drawdown": [0]})
+    equity = pd.DataFrame({"total": [100000], "drawdown": [0], "drawdown_pct": [0.0]})
     trades = _make_trades()
 
     analyzer = PerformanceAnalyzer(equity, trades)