第11章 回测深入¶
本章详细介绍 Lean 回测系统的配置、使用和结果分析。
11.1 回测配置¶
11.1.1 config.json 配置文件¶
注意:引擎的 config.json 中没有 start-date/end-date/initial-cash 等回测日期与资金配置键。回测的起止日期和初始资金由算法 Initialize() 中的 SetStartDate()/SetEndDate()/SetCash() 决定(见 11.1.3 节)。config.json 中与回测相关的常见键如下:
{
// 运行环境:"backtesting"(回测)、"live-paper" 等
"environment": "backtesting",
// 数据文件夹
"data-folder": "./Data",
// 算法配置
"algorithm-type-name": "MyStrategy",
"algorithm-language": "Python",
"algorithm-location": "MyStrategy.py",
// 是否实盘模式
"live-mode": false,
// 历史数据提供商(值为类名,此处为内置默认值)
"data-provider": "QuantConnect.Lean.Engine.DataFeeds.DefaultDataProvider"
}
11.1.2 使用 Lean CLI 配置回测¶
# 基本回测(项目名是位置参数,指向项目目录)
lean backtest "MyStrategy"
# 指定结果输出目录
lean backtest "MyStrategy" --output "./backtests/custom"
# 指定历史数据提供商(值为类名)
lean backtest "MyStrategy" \
--data-provider-historical "QuantConnect.Lean.Engine.DataFeeds.DefaultDataProvider"
注意:
lean backtest没有--start/--end/--initial-cash选项,回测时间与初始资金由算法Initialize()中的SetStartDate()/SetEndDate()/SetCash()决定。
11.1.3 程序化配置¶
from AlgorithmImports import *
class ConfigurableStrategy(QCAlgorithm):
"""
可配置的策略
"""
def Initialize(self):
# 回测日期与初始资金在 Initialize() 中设置(也可从参数读取后转换)
self.SetStartDate(2020, 1, 1)
self.SetEndDate(2023, 12, 31)
self.SetCash(100000)
# 从参数读取可覆盖的值(GetParameter 返回 str,需自行转换)
# initial_cash = int(self.GetParameter("initial-cash") or 100000)
# self.SetCash(initial_cash)
self.spy = self.AddEquity("SPY", Resolution.Daily).Symbol
11.2 回测执行模式¶
11.2.1 数据归一化模式¶
def Initialize(self):
spy = self.AddEquity("SPY", Resolution.Daily)
# 数据归一化模式
# Adjusted: 使用复权价格(推荐)
spy.SetDataNormalizationMode(DataNormalizationMode.Adjusted)
# Raw: 使用原始价格(包含分红拆股)
# spy.SetDataNormalizationMode(DataNormalizationMode.Raw)
# SplitAdjusted: 仅调整拆股
# spy.SetDataNormalizationMode(DataNormalizationMode.SplitAdjusted)
# TotalReturn: 包含分红的总收益
# spy.SetDataNormalizationMode(DataNormalizationMode.TotalReturn)
11.2.2 填充模式¶
def Initialize(self):
# 默认填充数据(fillForward 默认为 True)
spy = self.AddEquity("SPY", Resolution.Daily, fillForward=True)
# 不填充
spy = self.AddEquity("SPY", Resolution.Daily, fillForward=False)
11.2.3 延长时间¶
def Initialize(self):
# 包含盘前盘后数据(extendedMarketHours 默认为 False)
spy = self.AddEquity("SPY", Resolution.Minute, extendedMarketHours=True)
11.3 公司行为处理¶
11.3.1 分红处理¶
def OnDividends(self, dividends):
"""
分红事件处理
Args:
dividends: Dividends(DataDictionary[Symbol, Dividend])
"""
for dividend in dividends.Values:
self.Debug(f"Dividend: {dividend.Symbol.Value}")
self.Debug(f" Distribution: ${dividend.Distribution}")
self.Debug(f" Reference Price: ${dividend.ReferencePrice}")
# 分红会自动计入账户现金
# 如果持有股票,现金分红会自动添加
11.3.2 拆股处理¶
def OnSplits(self, splits):
"""
拆股事件处理
Args:
splits: Splits(DataDictionary[Symbol, Split])
"""
for split in splits.Values:
self.Debug(f"Split: {split.Symbol.Value}")
self.Debug(f" Type: {split.Type}")
self.Debug(f" Split Factor: {split.SplitFactor}")
self.Debug(f" Reference Price: ${split.ReferencePrice}")
# 拆股会自动处理
# 持仓数量和价格会自动调整
11.3.3 退市处理¶
def OnDelistings(self, delistings):
"""
退市事件处理
Args:
delistings: Delistings(DataDictionary[Symbol, Delisting])
"""
for delisting in delistings.Values:
self.Debug(f"Delisting: {delisting.Symbol.Value}")
self.Debug(f" Type: {delisting.Type}")
# 平仓(Liquidate 没有 reason 关键字参数)
if self.Portfolio[delisting.Symbol].Invested:
self.Liquidate(delisting.Symbol)
11.4 回测结果分析¶
11.4.1 基本统计指标¶
def Initialize(self):
# Portfolio 没有 StartingCash 属性,初始资金在 Initialize 中保存
self.starting_cash = self.Portfolio.TotalPortfolioValue
def OnEndOfAlgorithm(self):
"""算法结束时分析结果"""
# 基本统计
total_return = (self.Portfolio.TotalPortfolioValue /
self.starting_cash - 1)
# 从 Statistics 获取更多信息
# self.Statistics 是 StatisticsResults,只有 TotalPerformance/RollingPerformances/Summary 三个成员
# 数值指标位于 TotalPerformance.PortfolioStatistics,Summary 是字符串字典
portfolio_stats = self.Statistics.TotalPerformance.PortfolioStatistics
summary = self.Statistics.Summary
self.Debug("=" * 60)
self.Debug("BACKTEST RESULTS")
self.Debug("=" * 60)
self.Debug(f"Starting Capital: ${self.starting_cash:,.2f}")
self.Debug(f"Final Portfolio Value: ${self.Portfolio.TotalPortfolioValue:,.2f}")
self.Debug(f"Total Return: {total_return:.2%}")
self.Debug(f"Annual Return: {portfolio_stats.CompoundingAnnualReturn:.2%}")
self.Debug(f"Sharpe Ratio: {portfolio_stats.SharpeRatio:.2f}")
self.Debug(f"Sortino Ratio: {portfolio_stats.SortinoRatio:.2f}")
self.Debug(f"Max Drawdown: {portfolio_stats.Drawdown:.2%}")
self.Debug(f"Win Rate: {portfolio_stats.WinRate:.2%}")
self.Debug(f"Total Orders: {summary['Total Orders']}")
self.Debug("=" * 60)
11.4.2 详细性能指标¶
| 指标 | 说明 | 计算方式 |
|---|---|---|
| Total Return | 总收益率 | (最终值 - 初始值) / 初始值 |
| Compounding Annual Return | 复合年化收益率 | (最终值/初始值)^(1/年数) - 1 |
| Sharpe Ratio | 夏普比率 | 年化收益 / 年化波动率 |
| Sortino Ratio | 索提诺比率 | 年化收益 / 下行波动率 |
| Drawdown | 回撤 | 从峰值到谷底的跌幅 |
| Win Rate | 胜率 | 盈利交易次数 / 总交易次数 |
| Profit-Loss Ratio | 盈亏比 | 平均每笔盈利金额 / 平均每笔亏损金额 |
| Annual Variance | 年化方差 | 日收益率方差 × 252 |
11.4.3 生成 HTML 报告¶
# 生成回测报告(--backtest-results 指向回测输出目录中的结果 JSON,
# --report-destination 指定报告输出路径)
lean report \
--backtest-results "MyStrategy/backtests/2023-01-01_12-00-00/MyStrategy.json" \
--report-destination "./reports/report.html"
11.4.4 自定义分析¶
import pandas as pd
def OnEndOfAlgorithm(self):
"""自定义分析
注意:QCAlgorithm 没有内置的 CalculateMonthlyReturns/CalculateHoldingPeriods
方法,需要基于自己记录的数据(或 self.History 拉取的行情)自行计算。
"""
# self.daily_values 为 OnData 中逐日记录的 {日期: 组合总值}
equity = pd.Series(self.daily_values).sort_index()
# 计算月度收益:取每月最后一个净值,再计算环比变化
monthly_last = equity.groupby(equity.index.to_period("M")).last()
monthly_returns = monthly_last.pct_change().dropna()
self.Debug("\n=== Monthly Returns ===")
for month, return_pct in monthly_returns.items():
self.Debug(f"{month}: {return_pct:.2%}")
11.5 回测调试技巧¶
11.5.1 详细日志¶
def Initialize(self):
# 设置详细日志级别
self.Log("Initializing strategy...")
self.Debug(f"Start Date: {self.StartDate}")
self.Debug(f"End Date: {self.EndDate}")
self.Debug(f"Initial Cash: ${self.Portfolio.Cash}")
def OnData(self, data):
# 记录关键决策点
if self.ShouldBuy(data):
self.Log(f"Buy signal at {self.Time}")
self.Debug(f"Price: {self.Securities[self.spy].Price}")
self.Debug(f"RSI: {self.rsi.Current.Value}")
11.5.2 使用 Plot 绘制¶
def Initialize(self):
# 自定义图表(PlotIndicator 只用于绘制指标对象;
# 绘制数值请使用 Plot(chart, seriesName, value))
self.AddChart(Chart("Strategy"))
def OnData(self, data):
# 绘制信号数值
if self.ShouldBuy(data):
self.Plot("Strategy", "Signal", 1)
elif self.ShouldSell(data):
self.Plot("Strategy", "Signal", -1)
11.5.3 保存中间结果¶
import json
def OnData(self, data):
# 保存每日状态(状态持久化使用 ObjectStore,
# 没有 SaveState/GetState 方法)
if self.Time.day == 1:
state = {
'portfolio_value': str(self.Portfolio.TotalPortfolioValue),
'cash': str(self.Portfolio.Cash),
'invested': str(self.Portfolio.Invested)
}
self.ObjectStore.Save("mystrategy/state", json.dumps(state))
def OnEndOfAlgorithm(self):
# 读取保存的状态
if self.ObjectStore.ContainsKey("mystrategy/state"):
state = json.loads(self.ObjectStore.Read("mystrategy/state"))
self.Debug(f"Final state: {state}")
11.6 回测最佳实践¶
11.6.1 样本内外测试¶
class OutOfSampleTestStrategy(QCAlgorithm):
"""
样本内外测试策略
"""
def Initialize(self):
# 样本内训练期:2020-2021
self.SetStartDate(2020, 1, 1)
self.SetEndDate(2021, 12, 31)
self.in_sample = True
def OnData(self, data):
# 样本期内:训练和优化参数
if self.in_sample and self.Time.date() == datetime(2021, 12, 31).date():
self.TrainModel()
self.Debug("Training complete. Switching to out-of-sample testing.")
self.in_sample = False
# 样本外:使用固定参数测试
if not self.in_sample:
self.TradeWithFixedParams()
11.6.2 滚动窗口回测¶
class RollingWindowBacktest(QCAlgorithm):
"""
滚动窗口回测
每6个月重新训练模型
"""
def Initialize(self):
self.SetStartDate(2020, 1, 1)
self.SetEndDate(2023, 12, 31)
# 每6个月重新训练
self.Schedule.On(
self.DateRules.Every(DayOfWeek.Sunday),
self.TimeRules.Every(TimeSpan.FromDays(30)),
lambda: self.CheckRetrain()
)
self.last_retrain = self.Time
def CheckRetrain(self):
"""检查是否需要重新训练"""
if (self.Time - self.last_retrain).days >= 180:
self.RetrainModel()
self.last_retrain = self.Time
11.6.3 避免常见陷阱¶
class RobustStrategy(QCAlgorithm):
"""健壮的策略示例"""
def Initialize(self):
self.SetStartDate(2020, 1, 1)
self.SetEndDate(2023, 12, 31)
self.SetCash(100000)
self.spy = self.AddEquity("SPY", Resolution.Daily).Symbol
# 1. 设置预热避免初始数据不足
self.SetWarmUp(TimeSpan.FromDays(60))
# 2. 使用调整后价格避免分红影响
self.Securities[self.spy].SetDataNormalizationMode(
DataNormalizationMode.Adjusted
)
def OnData(self, data):
# 3. 检查数据可用性
if not data.ContainsKey(self.spy):
return
# 4. 确认指标准备就绪
if not self.indicator.IsReady:
return
# 5. 检查证券可交易
if not self.Securities[self.spy].IsTradable:
return
# 6. 确认账户有足够资金
if self.Portfolio.Cash < self.Securities[self.spy].Price:
return
# 执行交易逻辑
self.ExecuteStrategy()
11.7 完整示例:带详细分析的回测¶
from AlgorithmImports import *
from datetime import datetime, timedelta
class DetailedBacktestAnalysis(QCAlgorithm):
"""
带详细分析的回测策略
"""
def Initialize(self):
self.SetStartDate(2020, 1, 1)
self.SetEndDate(2023, 12, 31)
self.SetCash(100000)
# Portfolio 没有 StartingCash 属性,在此保存初始资金
self.starting_cash = self.Portfolio.TotalPortfolioValue
self.spy = self.AddEquity("SPY", Resolution.Daily).Symbol
# 设置数据归一化
self.Securities[self.spy].SetDataNormalizationMode(
DataNormalizationMode.Adjusted
)
# 指标
self.sma_fast = self.SMA(self.spy, 20)
self.sma_slow = self.SMA(self.spy, 50)
self.SetWarmUp(TimeSpan.FromDays(60))
# 记录交易
self.trades = []
self.daily_values = {}
def OnData(self, data):
# 记录每日组合价值
self.daily_values[self.Time.date()] = self.Portfolio.TotalPortfolioValue
# 交易逻辑
if self.sma_fast.Current.Value > self.sma_slow.Current.Value:
if not self.Portfolio[self.spy].IsLong:
self.SetHoldings(self.spy, 1)
self.RecordTrade("BUY", self.Securities[self.spy].Price)
else:
if self.Portfolio[self.spy].IsLong:
self.Liquidate(self.spy)
self.RecordTrade("SELL", self.Securities[self.spy].Price)
def RecordTrade(self, direction, price):
"""记录交易"""
self.trades.append({
'time': self.Time,
'direction': direction,
'price': price,
'portfolio_value': self.Portfolio.TotalPortfolioValue
})
def OnEndOfAlgorithm(self):
"""输出详细分析"""
self.PrintBasicStats()
self.PrintTradeAnalysis()
self.PrintMonthlyReturns()
self.PrintDrawdownAnalysis()
def PrintBasicStats(self):
"""打印基本统计"""
total_return = (self.Portfolio.TotalPortfolioValue /
self.starting_cash - 1)
self.Debug("\n" + "="*60)
self.Debug("BASIC STATISTICS")
self.Debug("="*60)
self.Debug(f"Starting Capital: ${self.starting_cash:,.2f}")
self.Debug(f"Final Capital: ${self.Portfolio.TotalPortfolioValue:,.2f}")
self.Debug(f"Total Return: {total_return:.2%}")
self.Debug(f"Annual Return: "
f"{self.Statistics.TotalPerformance.PortfolioStatistics.CompoundingAnnualReturn:.2%}")
self.Debug(f"Sharpe Ratio: "
f"{self.Statistics.TotalPerformance.PortfolioStatistics.SharpeRatio:.2f}")
self.Debug(f"Max Drawdown: "
f"{self.Statistics.TotalPerformance.PortfolioStatistics.Drawdown:.2%}")
self.Debug(f"Total Trades: {len(self.trades)}")
def PrintTradeAnalysis(self):
"""打印交易分析"""
if not self.trades:
return
buy_trades = [t for t in self.trades if t['direction'] == 'BUY']
sell_trades = [t for t in self.trades if t['direction'] == 'SELL']
self.Debug("\n" + "="*60)
self.Debug("TRADE ANALYSIS")
self.Debug("="*60)
self.Debug(f"Buy Trades: {len(buy_trades)}")
self.Debug(f"Sell Trades: {len(sell_trades)}")
# 计算每笔交易收益
returns = []
for i in range(0, len(self.trades)-1, 2):
if i+1 < len(self.trades):
entry = self.trades[i]
exit = self.trades[i+1]
trade_return = (exit['price'] / entry['price'] - 1) * 100
returns.append(trade_return)
if returns:
self.Debug(f"Average Trade Return: {sum(returns)/len(returns):.2f}%")
self.Debug(f"Winning Trades: {sum(1 for r in returns if r > 0)}")
self.Debug(f"Losing Trades: {sum(1 for r in returns if r < 0)}")
def PrintMonthlyReturns(self):
"""打印月度收益"""
monthly = {}
for date, value in self.daily_values.items():
month_key = date.strftime("%Y-%m")
if month_key not in monthly:
monthly[month_key] = []
monthly[month_key].append(value)
self.Debug("\n" + "="*60)
self.Debug("MONTHLY RETURNS")
self.Debug("="*60)
for month in sorted(monthly.keys()):
values = monthly[month]
month_return = (values[-1] / values[0] - 1) * 100
self.Debug(f"{month}: {month_return:.2f}%")
def PrintDrawdownAnalysis(self):
"""打印回撤分析"""
self.Debug("\n" + "="*60)
self.Debug("DRAWDOWN ANALYSIS")
self.Debug("="*60)
values = list(self.daily_values.values())
peaks = []
drawdowns = []
running_max = values[0]
for value in values:
if value > running_max:
running_max = value
peaks.append(value)
drawdown = (value - running_max) / running_max * 100
drawdowns.append(drawdown)
self.Debug(f"Max Drawdown: {min(drawdowns):.2f}%")
self.Debug(f"Average Drawdown: {sum(drawdowns)/len(drawdowns):.2f}%")
# 计算回撤持续时间
in_drawdown = False
drawdown_periods = []
current_period = 0
for dd in drawdowns:
if dd < 0:
in_drawdown = True
current_period += 1
elif in_drawdown:
in_drawdown = False
drawdown_periods.append(current_period)
current_period = 0
if drawdown_periods:
self.Debug(f"Avg Drawdown Duration: {sum(drawdown_periods)/len(drawdown_periods):.0f} days")
self.Debug(f"Max Drawdown Duration: {max(drawdown_periods)} days")