mirror of
https://ghfast.top/https://github.com/aeroxw/easy-tdx.git
synced 2026-09-12 16:54:17 +08:00
Justin Gu
239ff46ac8
用户反馈：3 个策略各自最大回撤仅 45.53%/40.16%/16.89%，组合却显示 83.76%。 根因：_build_combined_equity 的 drawdown_pct 分母误用 initial（固定初始资金）， 净值大涨后（总收益 545%，峰值≈6.45×初始）绝对回撤额除以小初始值被等比放大。 改为 drawdown / peak（相对当时峰值，0~1），与单标的 PortfolioTracker 一致。 连带修复卡玛比率（年化/最大回撤）。其余指标经核对均正确。加回归守卫。
259 lines
11 KiB
Python
259 lines
11 KiB
Python
"""单元测试：多策略资金分仓组合回测引擎（MultiStrategyEngine）。
|
||
|
||
覆盖：
|
||
- 基本多策略回测（2~3 个策略，各跑各的 df，合并曲线）
|
||
- 资金均分（1/N）
|
||
- individual_results 的 key 格式 "{label}@{symbol}"
|
||
- 合并净值曲线列结构 + 日期并集对齐
|
||
- 空策略列表兜底
|
||
- 同标的不同策略可区分
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import numpy as np
|
||
import pandas as pd
|
||
|
||
from easy_tdx.backtest.multi_strategy_engine import (
|
||
MultiStrategyEngine,
|
||
StrategySlot,
|
||
)
|
||
from easy_tdx.backtest.strategy import Strategy
|
||
|
||
|
||
class SimpleBuyStrategy(Strategy):
|
||
"""简单策略：bar 5 买入，bar 30 卖出。"""
|
||
|
||
def init(self) -> None:
|
||
pass
|
||
|
||
def next(self) -> None:
|
||
if self._bar_index == 5 and self.position["size"] == 0:
|
||
self.buy(size=0)
|
||
elif self._bar_index == 30 and self.position["size"] > 0:
|
||
self.sell(size=0)
|
||
|
||
|
||
class HoldStrategy(Strategy):
|
||
"""从不交易的策略（净值曲线恒等于初始资金）。"""
|
||
|
||
def init(self) -> None:
|
||
pass
|
||
|
||
def next(self) -> None:
|
||
pass
|
||
|
||
|
||
def _make_df(n: int = 100, seed: int = 42, start: str = "2024-01-01") -> pd.DataFrame:
|
||
"""生成随机 OHLCV DataFrame（与 test_portfolio_engine 同构造方式）。"""
|
||
rng = np.random.default_rng(seed)
|
||
close = 100.0 + np.cumsum(rng.normal(0, 1, n))
|
||
high = close + rng.uniform(0, 1, n)
|
||
low = close - rng.uniform(0, 1, n)
|
||
open_ = low + rng.uniform(0, high - low, n)
|
||
vol = rng.integers(1_000_000, 10_000_000, n).astype(float)
|
||
return pd.DataFrame(
|
||
{
|
||
"datetime": pd.date_range(start, periods=n, freq="D"),
|
||
"open": open_,
|
||
"high": high,
|
||
"low": low,
|
||
"close": close,
|
||
"vol": vol,
|
||
"amount": vol * close,
|
||
}
|
||
)
|
||
|
||
|
||
class TestMultiStrategyEngine:
|
||
def test_basic_run_two_strategies(self) -> None:
|
||
"""两个策略各跑各的 df，应产出合并结果。"""
|
||
slots = [
|
||
StrategySlot("双均线", "SH:601088", SimpleBuyStrategy(), _make_df(100, seed=42)),
|
||
StrategySlot("RSI", "SZ:000001", SimpleBuyStrategy(), _make_df(100, seed=99)),
|
||
]
|
||
engine = MultiStrategyEngine(slots, total_cash=1_000_000)
|
||
result = engine.run()
|
||
|
||
# individual_results 的 key 形如 "{label}@{symbol}"
|
||
assert set(result.individual_results.keys()) == {
|
||
"双均线@SH:601088",
|
||
"RSI@SZ:000001",
|
||
}
|
||
# 整体绩效含基本字段
|
||
assert "total_return" in result.total_performance
|
||
assert result.total_performance["total_stocks"] == 2
|
||
assert result.total_performance["total_cash"] == 1_000_000
|
||
|
||
def test_total_performance_has_full_metrics(self) -> None:
|
||
"""组合整体绩效应含完整 19 项指标（夏普/回撤/胜率/盈亏比等），与单标的同口径。"""
|
||
slots = [
|
||
StrategySlot("双均线", "SH:601088", SimpleBuyStrategy(), _make_df(100, seed=42)),
|
||
StrategySlot("RSI", "SZ:000001", SimpleBuyStrategy(), _make_df(100, seed=99)),
|
||
]
|
||
perf = MultiStrategyEngine(slots, total_cash=1_000_000).run().total_performance
|
||
# 关键指标都应在（来自 PerformanceAnalyzer）
|
||
for key in [
|
||
"total_return",
|
||
"annual_return",
|
||
"sharpe",
|
||
"sortino",
|
||
"calmar",
|
||
"max_drawdown",
|
||
"max_dd_duration",
|
||
"volatility",
|
||
"total_trades",
|
||
"win_trades",
|
||
"lose_trades",
|
||
"win_rate",
|
||
"profit_factor",
|
||
"avg_win",
|
||
"avg_loss",
|
||
"max_win",
|
||
"max_loss",
|
||
]:
|
||
assert key in perf, f"缺少指标 {key}"
|
||
# max_drawdown 用正值约定（与单标的一致），介于 0~1
|
||
assert 0 <= perf["max_drawdown"] <= 1
|
||
# 合并净值曲线的 drawdown 也应是正值
|
||
result = MultiStrategyEngine(slots, total_cash=1_000_000).run()
|
||
assert (result.combined_equity["drawdown"] >= 0).all()
|
||
|
||
def test_max_drawdown_relative_to_peak_not_initial(self) -> None:
|
||
"""最大回撤必须相对「当时峰值」而非「初始资金」。
|
||
|
||
回归 v1.17.11/v1.17.12 的 bug：drawdown_pct 分母误用 initial（固定初始值），
|
||
导致净值大涨后回撤被严重放大（如峰值 6x 初始时，真实 45% 回撤被算成 290%）。
|
||
构造一个大涨后回撤的场景：净值为 1→6→4（即从峰值回撤 33%），验证 max_drawdown
|
||
≈ 33%（旧逻辑会算成 200%，超出 1.0）。
|
||
"""
|
||
# 构造单标的净值序列：前 50 根 close 线性涨到 6 倍，后 50 根跌到 4 倍。
|
||
# 用从不交易的 HoldStrategy，使 total ≈ initial_cash（曲线不随 close 变）……
|
||
# 不行——HoldStrategy 净值恒为初始资金，无法制造涨跌。改用直接断言合并曲线
|
||
# 的 drawdown_pct 计算逻辑：构造两段净值的合成 df 喂给 _build_combined_equity。
|
||
from easy_tdx.backtest.types import BacktestResult
|
||
|
||
# 两根等长净值曲线：均从 1.0 涨到 6.0 再跌到 4.0（各 50 根，峰值在第 50 根）
|
||
dates = pd.date_range("2024-01-01", periods=100, freq="D")
|
||
up = np.linspace(1.0, 6.0, 50) # 0→50: 1→6
|
||
down = np.linspace(6.0, 4.0, 50) # 50→100: 6→4
|
||
totals = np.concatenate([up, down]) # 峰值 6.0 在第 50 根，谷底 4.0 在末尾
|
||
ec = pd.DataFrame(
|
||
{
|
||
"datetime": dates,
|
||
"total": totals * 100_000, # 缩放到资金量级
|
||
"drawdown": np.zeros(100),
|
||
"drawdown_pct": np.zeros(100),
|
||
}
|
||
)
|
||
# 造一个空 trades/positions 的 BacktestResult 占位
|
||
empty_df = pd.DataFrame()
|
||
fake = BacktestResult(
|
||
performance={"total_return": 3.0},
|
||
equity_curve=ec,
|
||
trades=empty_df,
|
||
positions=empty_df,
|
||
config={},
|
||
)
|
||
engine = MultiStrategyEngine.__new__(MultiStrategyEngine)
|
||
combined = engine._build_combined_equity( # noqa: SLF001 — 直接测内部算法
|
||
{"A@SZ:000001": fake}, {"A@SZ:000001": 100_000.0}
|
||
)
|
||
# 真实最大回撤（相对峰值）：峰值 600000，谷底 400000，回撤 = 200000/600000 ≈ 33.3%
|
||
dd_pct = combined["drawdown_pct"].to_numpy()
|
||
max_dd = float(np.max(dd_pct))
|
||
assert 0.30 <= max_dd <= 0.36, f"max_drawdown 应≈33%，实际 {max_dd:.4f}"
|
||
# 旧 bug（除以 initial=100000）会算成 200%（200000/100000），必然 >1
|
||
assert max_dd <= 1.0, "drawdown_pct 相对峰值，绝不可能超过 100%"
|
||
|
||
def test_capital_split_equal(self) -> None:
|
||
"""资金按策略数均分：每个槽位 1/N。"""
|
||
slots = [
|
||
StrategySlot("A", "SH:601088", SimpleBuyStrategy(), _make_df(50, seed=1)),
|
||
StrategySlot("B", "SZ:000001", SimpleBuyStrategy(), _make_df(50, seed=2)),
|
||
StrategySlot("C", "SZ:000002", SimpleBuyStrategy(), _make_df(50, seed=3)),
|
||
]
|
||
engine = MultiStrategyEngine(slots, total_cash=900_000)
|
||
allocs = engine._compute_allocations() # noqa: SLF001 — 测试内部均分逻辑
|
||
assert len(allocs) == 3
|
||
assert all(v == 300_000 for v in allocs.values())
|
||
# equity_allocation 是占比，各 1/3
|
||
result = engine.run()
|
||
assert all(abs(v - 1 / 3) < 1e-9 for v in result.equity_allocation.values())
|
||
|
||
def test_combined_equity_has_expected_columns(self) -> None:
|
||
"""合并净值曲线应有 datetime/total/drawdown/drawdown_pct 列。"""
|
||
slots = [
|
||
StrategySlot("A", "SH:601088", SimpleBuyStrategy(), _make_df(60, seed=7)),
|
||
]
|
||
engine = MultiStrategyEngine(slots, total_cash=500_000)
|
||
result = engine.run()
|
||
cols = set(result.combined_equity.columns)
|
||
assert {"datetime", "total", "drawdown", "drawdown_pct"} <= cols
|
||
assert len(result.combined_equity) > 0
|
||
|
||
def test_combined_equity_aligns_disjoint_dates(self) -> None:
|
||
"""两个策略日期范围不同时，合并曲线应按并集对齐（ffill）。"""
|
||
# 策略 A 跑 2024-01 起 60 根，策略 B 跑 2024-03 起 60 根
|
||
df_a = _make_df(60, seed=1, start="2024-01-01")
|
||
df_b = _make_df(60, seed=2, start="2024-03-01")
|
||
slots = [
|
||
StrategySlot("A", "SH:601088", SimpleBuyStrategy(), df_a),
|
||
StrategySlot("B", "SZ:000001", SimpleBuyStrategy(), df_b),
|
||
]
|
||
engine = MultiStrategyEngine(slots, total_cash=1_000_000)
|
||
result = engine.run()
|
||
# 合并曲线长度应至少覆盖两个范围的最晚结束日（并集）
|
||
assert len(result.combined_equity) >= 60
|
||
|
||
def test_empty_strategies_returns_empty_result(self) -> None:
|
||
"""空策略列表应返回空结果，不抛异常。"""
|
||
engine = MultiStrategyEngine([], total_cash=1_000_000)
|
||
result = engine.run()
|
||
assert result.individual_results == {}
|
||
assert result.total_performance["total_return"] == 0.0
|
||
# combined_equity 为带表头的空 DataFrame
|
||
assert len(result.combined_equity) == 0
|
||
assert set(result.combined_equity.columns) == {
|
||
"datetime",
|
||
"total",
|
||
"drawdown",
|
||
"drawdown_pct",
|
||
}
|
||
|
||
def test_same_symbol_different_strategies_distinguished(self) -> None:
|
||
"""同标的不同策略应能区分（key 含 label）。"""
|
||
df = _make_df(60, seed=5)
|
||
slots = [
|
||
StrategySlot("双均线", "SH:601088", SimpleBuyStrategy(), df.copy()),
|
||
StrategySlot("RSI", "SH:601088", HoldStrategy(), df.copy()),
|
||
]
|
||
engine = MultiStrategyEngine(slots, total_cash=1_000_000)
|
||
result = engine.run()
|
||
# 两个 key 不同，都带同一 symbol
|
||
assert "双均线@SH:601088" in result.individual_results
|
||
assert "RSI@SH:601088" in result.individual_results
|
||
|
||
def test_hold_strategy_keeps_initial_capital(self) -> None:
|
||
"""从不交易的策略，其净值曲线末值应等于初始分得资金。"""
|
||
slots = [
|
||
StrategySlot("Hold", "SH:601088", HoldStrategy(), _make_df(40, seed=1)),
|
||
]
|
||
engine = MultiStrategyEngine(slots, total_cash=1_000_000)
|
||
result = engine.run()
|
||
ec = result.individual_results["Hold@SH:601088"].equity_curve
|
||
# 不交易 → 末值 ≈ 初始资金 1_000_000（单策略拿全部）
|
||
assert abs(ec["total"].iloc[-1] - 1_000_000) < 1.0
|
||
|
||
def test_to_dict_serializable(self) -> None:
|
||
"""to_dict 应产出 JSON 兼容结构（含 individual_results / combined_equity）。"""
|
||
slots = [
|
||
StrategySlot("A", "SH:601088", SimpleBuyStrategy(), _make_df(50, seed=1)),
|
||
]
|
||
result = MultiStrategyEngine(slots, total_cash=500_000).run()
|
||
d = result.to_dict()
|
||
assert "total_performance" in d
|
||
assert "individual_results" in d
|
||
assert "combined_equity" in d
|
||
assert isinstance(d["individual_results"]["A@SH:601088"], dict)
|