WIP — 项目开发中，尚未完成全部功能。
加密货币量化回测引擎。C++ 核心引擎负责高性能回测，Python 策略层（pybind11）支持快速迭代，ClickHouse 存储历史行情与回测结果。
A crypto quantitative backtesting engine. C++ core engine for high-performance backtesting, Python strategy layer (pybind11) for rapid iteration, ClickHouse for market data and backtest result storage.
┌─────────────────────────────────────────────┐
│ Python Strategy Layer │
│ (pybind11, override C++ virtual) │
├─────────────────────────────────────────────┤
│ C++ Core Engine │
│ ┌─────────┐ ┌──────────┐ ┌──────────────┐ │
│ │ Engine │ │ SimBroker│ │ Indicators │ │
│ │ Context │ │ Portfolio│ │ SMA/EMA/RSI │ │
│ │ Strategy │ │ Order │ │ MACD/Boll. │ │
│ └─────────┘ └──────────┘ └──────────────┘ │
├─────────────────────────────────────────────┤
│ Data Layer: CsvFeed / ClickHouse Pipeline │
└─────────────────────────────────────────────┘
- 事件驱动：小顶堆多币种时间归并，支持任意数量交易对同时回测
- 高性能：单核 ~770 万 bars/s（无指标），~170 万 bars/s（7 指标），详见 Benchmark
- C++/Python 混合：pybind11 导出核心类，Python 子类可直接 override C++ 虚函数
- 内置指标：SMA、EMA、RSI、MACD、Bollinger Bands，指标自动缓存
- 模拟撮合：市价单/限价单/止损单/止损限价单、maker/taker 分离费率、滑点、成交量参与率限制
- 绩效分析：Sharpe、Sortino、Calmar、profit factor、最大回撤、年化收益、胜率、净值曲线
- 数据管道：Binance REST API → ClickHouse，支持任意时间范围批量采集
- CLI 入口：
collect（数据采集）、backtest（回测）、list-data（查询数据）
- Event-driven: Min-heap time-merging across multiple symbols
- High performance: ~7.7M bars/s (no indicators), ~1.7M bars/s (7 indicators) on single core. See Benchmark
- C++/Python hybrid: pybind11 exports core classes; Python subclasses can override C++ virtual functions
- Built-in indicators: SMA, EMA, RSI, MACD, Bollinger Bands with automatic caching
- Simulated broker: Market/limit/stop-market/stop-limit orders, maker/taker fee split, slippage, volume participation limits
- Performance analytics: Sharpe, Sortino, Calmar, profit factor, max drawdown, annual return, win rate, equity curve
- Data pipeline: Binance REST API → ClickHouse, batch collection for any time range
- CLI:
collect(data collection),backtest(run backtest),list-data(query data)
# 依赖: cmake >= 3.20, g++ >= 11 (C++17), Python 3
mkdir -p build && cd build
cmake ..
make -j$(nproc)
# 运行测试
./cpp/qe_test
# 运行 benchmark
./cpp/qe_benchcd python
# 采集 Binance 1m K线数据到 ClickHouse（需要 ClickHouse 运行在 localhost:8123）
python3 cli.py collect BTCUSDT --start 2024-01-01 --end 2024-01-31
python3 cli.py collect BTCUSDT ETHUSDT --start 2024-01-01 --end 2024-12-31 --interval 1h
# 查看已采集的数据
python3 cli.py list-data
# 用双均线策略回测（默认 1 分钟 K 线）
python3 cli.py backtest --strategy dual_ma --symbols BTCUSDT \
--start 2024-01-01 --end 2024-01-31
# 指定 K 线周期：1m, 5m, 15m, 1h, 4h, 1d
python3 cli.py backtest --strategy macd_cross --symbols BTCUSDT \
--start 2024-01-01 --end 2025-01-01 --interval 1d
# 自定义参数和资金
python3 cli.py backtest --strategy dual_ma --symbols BTCUSDT \
--start 2024-01-01 --end 2025-01-01 --interval 4h \
--cash 50000 --commission 0.0004 --slippage 0.0001 \
--params fast=10 slow=30
# 多币种动量轮动
python3 cli.py backtest --strategy momentum_rotation \
--symbols BTCUSDT ETHUSDT SOLUSDT \
--start 2024-01-01 --end 2024-06-30 \
--params lookback=20
# 生成可视化图表（净值曲线 + 回撤 + 价格走势 + 绩效摘要）
python3 cli.py backtest --strategy macd_cross --symbols BTCUSDT \
--start 2024-01-01 --end 2025-01-01 --interval 1d --plot
# 指定图表输出路径
python3 cli.py backtest --strategy dual_ma --symbols BTCUSDT \
--start 2024-01-01 --end 2025-01-01 --interval 1h \
--plot -o result.png内置策略：
| 策略 | 说明 | 参数 |
|---|---|---|
dual_ma |
双均线交叉 | fast(5), slow(20) |
macd_cross |
MACD 信号交叉 | fast(12), slow(26), signal(9) |
rsi_reversal |
RSI 超买超卖反转 | period(14), oversold(30), overbought(70) |
bollinger_breakout |
布林带突破 | period(20), std_dev(2.0) |
momentum_rotation |
多币种动量轮动 | lookback(20) |
K 线周期：非 1m 周期从 klines_1m 实时聚合（argMin(open)/argMax(close) 保证 OHLC 正确）。
#include "core/engine.h"
#include "core/sim_broker.h"
#include "data/csv_feed.h"
#include "indicator/sma.h"
class MyStrategy : public qe::Strategy {
void on_init(qe::Context& ctx) override {
auto sid = ctx.symbol("BTCUSDT");
ctx.indicator<qe::SMA>(sid, 10);
ctx.indicator<qe::SMA>(sid, 30);
}
void on_bar(qe::Context& ctx, uint16_t sid, const qe::Bar& bar) override {
auto& fast = ctx.indicator<qe::SMA>(sid, 10);
auto& slow = ctx.indicator<qe::SMA>(sid, 30);
if (!fast.ready() || !slow.ready()) return;
if (fast.value() > slow.value() && ctx.position(sid).quantity < 1e-12)
ctx.buy(sid, 0.01);
else if (fast.value() < slow.value() && ctx.position(sid).quantity > 1e-12)
ctx.sell(sid, ctx.position(sid).quantity);
}
};
int main() {
qe::Engine engine;
auto btc = engine.symbols().id("BTCUSDT");
engine.add_feed(std::make_unique<qe::CsvFeed>(btc, "data/btcusdt_1m.csv"));
qe::SimBrokerConfig cfg{.cash = 10000.0, .commission_rate = 0.0004};
engine.set_broker(std::make_unique<qe::SimBroker>(cfg));
engine.add_strategy(std::make_shared<MyStrategy>());
auto result = engine.run();
printf("Sharpe: %.2f, Return: %.2f%%\n", result.sharpe, result.total_return * 100);
}import qe
class DualMA(qe.Strategy):
def on_init(self, ctx):
sid = ctx.symbol("BTCUSDT")
ctx.sma(sid, 10)
ctx.sma(sid, 30)
def on_bar(self, ctx, symbol_id, bar):
fast = ctx.sma(symbol_id, 10)
slow = ctx.sma(symbol_id, 30)
if not fast.ready() or not slow.ready():
return
pos = ctx.position(symbol_id).quantity
if fast.value() > slow.value() and pos < 1e-12:
ctx.buy(symbol_id, 0.01)
elif fast.value() < slow.value() and pos > 1e-12:
ctx.sell(symbol_id, pos)
engine = qe.Engine()
btc = engine.symbol_id("BTCUSDT")
engine.add_feed_bars(btc, bars) # bars: list[qe.Bar]
config = qe.SimBrokerConfig()
config.cash = 10000.0
config.commission_rate = 0.0004
config.slippage = 0.0001
engine.set_broker(config)
engine.add_strategy(DualMA())
result = engine.run()
print(f"Sharpe: {result.sharpe:.2f}, Return: {result.total_return:.2%}")
print(f"Sortino: {result.sortino:.2f}, Calmar: {result.calmar:.2f}")
print(f"Max Drawdown: {result.max_drawdown:.2%}, Win Rate: {result.win_rate:.2%}")测试环境：Intel i9-13900 (单核)，-O2，GCC 13.3
[1] Throughput scaling (BuyAndHold, 1 symbol)
10K bars 1.9ms 5.4M bars/s
100K bars 13.5ms 7.4M bars/s
1M bars 129ms 7.7M bars/s
[2] Multi-symbol scaling (DualSMA 10/30, 100K bars)
1 symbol 20ms 5.1M bars/s
4 symbols 87ms 4.6M bars/s
16 symbols 378ms 4.2M bars/s
64 symbols 1.88s 3.4M bars/s
[3] Indicator overhead (7 indicators, 1 symbol)
10K bars 6.1ms 1.6M bars/s
100K bars 58ms 1.7M bars/s
500K bars 289ms 1.7M bars/s
[4] Stress test (7 indicators, 100K bars, multi-symbol)
1 symbol 61ms 1.7M bars/s
4 symbols 250ms 1.6M bars/s
16 symbols 1.05s 1.5M bars/s
quant-engine/
├── cpp/
│ ├── core/ # Engine, Context, Broker, Strategy, SymbolTable, Portfolio, Order
│ ├── data/ # Bar, DataFeed, CsvFeed
│ ├── indicator/ # SMA, EMA, RSI, MACD, Bollinger
│ ├── analyzer/ # Performance metrics (Sharpe, Sortino, Calmar, etc.)
│ ├── strategies/ # C++ strategies: DualMA, MACDCross
│ ├── cli/ # C++ CLI (direct ClickHouse backtest)
│ ├── bind/ # pybind11 module
│ ├── bench/ # Benchmark
│ └── test/ # Unit tests (66 tests)
├── python/
│ ├── cli.py # CLI entry point
│ ├── data_collector.py # Binance → ClickHouse pipeline
│ ├── plot_results.py # Backtest visualization
│ ├── freqtrade_compare.py # qe vs Freqtrade comparison
│ ├── strategies/ # Python strategies (dual_ma, macd_cross, rsi_reversal, bollinger_breakout, momentum_rotation)
│ └── ft_strategies/ # Freqtrade-compatible strategy wrappers
├── data/ # CSV test data
└── CMakeLists.txt
- Phase 1: C++ 回测引擎核心（Engine/Context/Strategy/Broker/Feed/Indicators）
- Phase 2.1: Python 策略层（pybind11）+ ClickHouse 数据管道 + CLI
- Phase 2.2: 现货回测可用（止损单、maker/taker 费率、成交量限制、Sortino/Calmar 等绩效指标）
- Phase 2.3: C++ CLI、新策略（RSI/Bollinger）、Freqtrade 对比工具、架构文档、可视化
- Phase 3: 实盘交易（Binance WebSocket + LiveBroker + 风控）
MIT