ClawQuant Trader

功能 Features

量化研究全链路工具箱 Everything You Need for Quant Research

从数据抓取到部署执行的一站式能力，天然支持 Agent 调用与 JSON 输出。A complete toolkit from data ingestion to deployment with AI-agent-native JSON output.

📊

批量回测Batch Backtesting

支持 N 策略 × M 标的 × K 周期并行运行，基于 ProcessPoolExecutor 实现真正 CPU 并行。Run N strategies × M symbols × K periods in parallel with ProcessPoolExecutor.

🎯

稳定性评分Stability Scoring

5 维综合评分（0-100），评估质量、风险、稳健性、成本敏感度与过度交易惩罚。5-dimensional composite score (0–100) across quality, risk, robustness, cost sensitivity, and overtrade.

🔧

参数优化Parameter Optimization

支持网格搜索与随机扫描，结合 Walk-forward 滚动训练/测试提升泛化能力。Grid search or random sweep with walk-forward validation and rolling windows.

📡

机会雷达Opportunity Radar

实时扫描观察列表信号，并给出历史上下文与风险解释。Real-time signal scanning across your watchlist with context and risk explanation.

📝

专业报告Professional Reports

每次运行生成 JSON + Markdown + 图表，包含权益曲线、回撤与交易点位。Generate JSON + Markdown + charts per run with equity, drawdown, and trades.

🚀

模拟/实盘部署Paper & Live Deploy

通过安全开关和一键平仓机制，将验证后的策略用于模拟或实盘执行。Deploy validated strategies for paper or live trading with built-in safety controls.

🦞

Agent 原生AI Agent-First

所有命令支持 --json 结构化输出，并提供 6 个 YAML Skill 供 Agent 直接调用。All commands output structured JSON with --json and 6 YAML skills.

🔄

可复现性Full Reproducibility

每次运行都会记录策略版本、参数哈希、数据哈希与 Python 环境，实现结果可复现。Each run snapshots strategy version, params hash, data hash, and Python environment.

🛡

风控机制Risk Controls

包含仓位上限、回撤止损、冷却周期、下单频率限制；下一根开盘成交模型避免未来函数。Position limits, drawdown stops, cooldowns, rate limits, and next-open fill model.

内置策略 Built-in Strategies

7 个策略开箱即用 7 Strategies, Ready to Go

从被动定投到趋势突破，支持通过 BaseStrategy 快速扩展自定义策略。From passive DCA to momentum breakouts, extensible via BaseStrategy.

策略Strategy	类型Type	说明Description	关键参数Key Parameters
dca	被动Passive	定投策略：按固定间隔买入Dollar Cost Averaging — fixed buys at intervals	`buy_interval, buy_amount_usdt`
ma_crossover	趋势Trend	均线交叉策略（SMA/EMA）Moving Average Crossover (SMA/EMA)	`fast_period, slow_period, ma_type`
macd	趋势Trend	MACD 金叉/死叉信号MACD crossover signals	`fast_period, slow_period, signal_period`
breakout	趋势Trend	唐奇安通道突破Donchian Channel Breakout	`lookback`
rsi_reversal	均值回归Reversion	RSI 反转信号RSI-based reversal signals	`rsi_period, oversold, overbought`
bollinger_bands	均值回归Reversion	布林带突破/回归Bollinger Bands breakout / reversion	`bb_period, bb_std`
grid	均值回归Reversion	网格交易：低买高卖Grid Trading — buy dips, sell rallies	`grid_count, grid_spacing_pct`

快速开始 Quick Start

4 步完成首次回测 Up and Running in 4 Steps

不到一分钟，从安装到报告生成。From zero to report in under a minute.

步骤 01STEP 01

安装Install

通过 pip 从 PyPI 安装。Install from PyPI with pip.

pip install clawquant

步骤 02STEP 02

拉取数据Pull Data

抓取 OHLCV 数据并自动缓存为 Parquet。Fetch OHLCV data and cache as Parquet.

clawquant data pull BTC/USDT --days 30

步骤 03STEP 03

运行回测Run Backtest

一次批量对比多个策略。Batch test multiple strategies at once.

clawquant backtest batch dca,ma_crossover --symbols BTC/USDT

步骤 04STEP 04

生成报告Generate Report

输出指标、图表与稳定性评分。Get metrics, charts, and stability scores.

clawquant report generate <run_id>

命令参考 CLI Reference

6 大命令组，17+ 子命令 6 Command Groups, 17+ Subcommands

全部命令支持 --json 结构化输出；配合 --help 查看细节。Every command supports structured JSON output.

💾

数据 datadata

clawquant data pull

从交易所拉取 OHLCV，自动缓存为 ParquetFetch OHLCV from exchanges, auto-cache as Parquet

clawquant data inspect

数据质量检查：缺口、重复、异常值Quality checks — gaps, duplicates, outliers

clawquant data cache-status

查看缓存文件与元数据View cached files and metadata

🧠

策略 strategystrategy

clawquant strategy list

列出全部可用策略List all available strategies

clawquant strategy validate

校验策略接口规范Verify strategy interface compliance

clawquant strategy scaffold

生成新策略模板Generate a new strategy template

⚡

回测 backtestbacktest

clawquant backtest run

单策略 × 单标的回测Single strategy × single symbol backtest

clawquant backtest batch

多策略 × 多标的批量回测Multi-strategy × multi-symbol batch runs

clawquant backtest sweep

参数网格/随机搜索优化Parameter grid/random search optimization

clawquant backtest walkforward

滚动窗口 Walk-forward 验证Walk-forward validation with rolling splits

📡

雷达 radarradar

clawquant radar scan

扫描观察列表当前信号Scan current signals across watchlist

clawquant radar explain

结合上下文与风险解释机会Explain opportunity with context & risk

📄

报告 reportreport

clawquant report generate

生成单次运行的 JSON + Markdown + 图表JSON + Markdown + charts for a run

clawquant report batch

横向对比多个运行结果Compare multiple runs side-by-side

🚀

部署 deploydeploy

clawquant deploy paper

启动模拟交易（paper）Start paper (simulated) trading

clawquant deploy live

启动实盘交易（需确认）Start live trading (requires confirmation)

clawquant deploy status / stop / flatten

管理运行中的部署任务Manage active deployments

输出示例 Output Examples

一次运行会产出什么 What You Get After a Run

每次回测都会生成结构化指标与完整报告目录。Each run generates structured metrics and report artifacts.

report.json (excerpt)
{
  "strategy": "ma_crossover",
  "symbol": "BTC/USDT",
  "total_return": 0.1847,
  "sharpe_ratio": 1.82,
  "max_drawdown": -0.0923,
  "win_rate": 0.58,
  "total_trades": 24,
  "stability_score": 72.5,
  "score_breakdown": {
    "quality": 78,
    "risk": 71,
    "robustness": 65,
    "cost_sens": 80,
    "overtrade": 90
  }
}

runs/<run_id>/
# Report directory structure
runs/
  ma_crossover_BTC_USDT_20260305_a1b2c3d4/
    run_meta.json     # Config snapshot
    result.json       # Raw trades + equity
    report.json       # Full metrics + score
    report.md         # Human-readable report
    equity.png        # Equity curve chart
    drawdown.png      # Drawdown visualization
    trades.png        # Entry/exit markers

最佳实践 Best Practices

让结果更可靠的实战建议 Tips for Reliable Results

规避常见误区，提高回测结论的可信度。Avoid pitfalls and get more reliable backtest conclusions.

💾 数据Data

回测前先运行 data inspectAlways run data inspect before backtesting
建议至少使用 30 天数据Use at least 30 days of data for meaningful results
检查缺口与异常值Check for gaps and outliers in fetched data
周期要匹配策略频率Match interval to strategy timeframe

⏱ 区间与切分Periods & Splits

参数调优建议使用 Walk-forwardUse walk-forward validation for parameter tuning
覆盖不同市场状态测试Test across multiple market regimes
避免只对单一区间过拟合Avoid overfitting to a single time period
Walk-forward 建议至少 3 个切分Use 3+ splits for walk-forward tests

🔄 工作流Workflow

先用 batch 做策略对比Start with batch to compare strategies
只对优胜策略做参数扫描Sweep parameters on top performers only
部署前进行 Walk-forward 验证Validate with walk-forward before deploying
实盘前先跑模拟盘Paper trade before going live

⚠ 常见误区Pitfalls to Avoid

不要忽略稳定性评分的分项Don't ignore the stability score breakdown
交易样本太少时不要盲目优化Don't optimize on too few trades
不要跳过成本敏感度检查Don't skip the cost sensitivity check
未模拟验证不要直接上实盘Don't deploy without paper trading first