Self-Evolving Trading Intelligence — genetic algorithms discover strategies you never would.
FinClaw doesn't need you to write strategies — its genetic algorithm discovers and evolves them autonomously across 484 factor dimensions, then validates them with walk-forward testing and Monte Carlo simulation.
This project is for educational and research purposes only. Not financial advice. Past performance does not guarantee future results. Always paper trade first.
pip install -e .
# See everything in action — zero API keys needed
finclaw demo
# Download crypto market data
finclaw download-crypto --coins BTC,ETH,SOL --days 365
# Evolve a crypto strategy with genetic algorithms
finclaw evolve --market crypto --generations 50
# Real-time quotes
finclaw quote BTC/USDT
finclaw quote AAPLThat's it. No API keys, no exchange accounts, no config files.
finclaw demo — full feature showcase
███████╗██╗███╗ ██╗ ██████╗██╗ █████╗ ██╗ ██╗
██╔════╝██║████╗ ██║██╔════╝██║ ██╔══██╗██║ ██║
█████╗ ██║██╔██╗ ██║██║ ██║ ███████║██║ █╗ ██║
██╔══╝ ██║██║╚██╗██║██║ ██║ ██╔══██║██║███╗██║
██║ ██║██║ ╚████║╚██████╗███████╗██║ ██║╚███╔███╔╝
╚═╝ ╚═╝╚═╝ ╚═══╝ ╚═════╝╚══════╝╚═╝ ╚═╝ ╚══╝╚══╝
AI-Powered Financial Intelligence Engine
🎬 FinClaw Demo — All features, zero config
━━━ 📊 Real-Time Quotes ━━━
Symbol Price Change % Trend
────────────────────────────────────────────────────────────
AAPL 189.84 +2.31 +1.23% ▇█▇▆▅▅▅▄▄▄▃▃▂▁ ▁▁▁
NVDA 875.28 +15.67 +1.82% ▄▄▆▅▅▃▂▂ ▂▃▃▁▄▅▆▅▇▆
TSLA 175.21 -3.45 -1.93% ▁▁▁▂▄▄▄▄▄▂▃▃▃▄▃▄▅▇
MSFT 415.50 +1.02 +0.25% █▇▇▆▄▅▅▅▅▄▅▄▂▂ ▁
━━━ 🚀 Backtest: Momentum Strategy on AAPL ━━━
Strategy: +75.7% (+32.5%/yr) Buy&Hold: +67.7%
Alpha: +8.0% Sharpe: 1.85
MaxDD: -8.3% Win Rate: 63%
━━━ 🤖 AI Features ━━━
MCP Server — Expose FinClaw as tools for Claude, Cursor, VS Code
Copilot — Interactive AI financial assistant
Strategy AI — Natural language → trading code
finclaw quote BTC/USDT — real-time crypto quote
BTC/USDT $68828.00 -3.53%
Bid: 68828.0 Ask: 68828.1 Vol: 455,860,493
finclaw evolve --market crypto --generations 50 — strategy evolution
🧬 Evolution Engine — Crypto Market
Population: 30 | Mutation Rate: 0.3 | Elite: 5
Gen 1 │ Best: 0.342 │ Avg: 0.118 │ Sharpe: 0.89 │ ░░░░░░░░░░
Gen 5 │ Best: 0.567 │ Avg: 0.234 │ Sharpe: 1.12 │ ██░░░░░░░░
Gen 10 │ Best: 0.723 │ Avg: 0.389 │ Sharpe: 1.45 │ ████░░░░░░
Gen 25 │ Best: 0.891 │ Avg: 0.512 │ Sharpe: 1.87 │ ██████░░░░
Gen 50 │ Best: 0.934 │ Avg: 0.601 │ Sharpe: 2.14 │ ████████░░
✅ Best DNA saved to evolution_results/best_gen50.json
Most quant tools make you write the strategy. FinClaw evolves strategies for you.
| FinClaw | Freqtrade | Jesse | FinRL / Qlib | |
|---|---|---|---|---|
| Strategy design | GA evolves 484-dim DNA | You write rules | You write rules | DRL trains agent |
| Runs 24/7 | Strategy itself evolves | Bot runs, strategy static | Bot runs, strategy static | Training is offline |
| Walk-forward validation | ✅ Built-in (70/30 + Monte Carlo) | ❌ Plugin needed | ❌ Plugin needed | |
| Anti-overfitting | Arena competition + bias detection | Basic cross-validation | Basic | Varies |
| Zero API keys to start | ✅ pip install && finclaw demo |
❌ Needs exchange keys | ❌ Needs keys | ❌ Needs data setup |
| Market coverage | Crypto + A-shares + US stocks | Crypto only | Crypto only | A-shares (Qlib) |
| MCP server (AI agents) | ✅ Claude / Cursor / VS Code | ❌ | ❌ | ❌ |
| Factor library | 484 factors, auto-weighted | ~50 manual indicators | Manual indicators | Alpha158 (Qlib) |
- Self-evolving factors — A genetic algorithm mutates, breeds, and selects strategy DNA across 484 dimensions. No human decides the signal weights — natural selection does.
- Walk-forward validation — Every backtest uses a 70/30 train/test split with Monte Carlo simulation (1,000 iterations, p < 0.05). This is what institutional quants use, not simple in-sample backtesting.
- Multi-market — Crypto (via ccxt, 100+ exchanges), China A-shares (AKShare + BaoStock), US stocks (Yahoo Finance). One engine, all markets.
- AI-native — Ship with an MCP server so Claude, Cursor, and VS Code can query quotes, run backtests, and analyze portfolios natively.
┌──────────────────────────────────────────────────────┐
│ Evolution Engine (Core) │
│ Genetic Algorithm → Mutate → Backtest → Select │
│ │
│ Input: 484 factors × weights = DNA │
│ Output: Walk-forward validated strategy │
├──────────────────────────────────────────────────────┤
│ Factor Sources │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐ │
│ │Technical │ │Sentiment │ │ DRL │ │ Davis │ │
│ │ 284 │ │ News │ │Q-learning│ │Double │ │
│ │ general │ │ EN / ZH │ │ signals │ │ Play │ │
│ └────┬─────┘ └────┬─────┘ └────┬─────┘ └───┬────┘ │
│ └─────────────┴────────────┴────────────┘ │
│ + 200 crypto-specific factors │
│ All → compute() → [0, 1] │
├──────────────────────────────────────────────────────┤
│ Quality Assurance │
│ ┌────────────┐ ┌─────────────┐ ┌────────────────┐ │
│ │ Arena │ │ Bias │ │ Walk-Forward │ │
│ │Competition │ │ Detection │ │ + Monte Carlo│ │
│ └────────────┘ └─────────────┘ └────────────────┘ │
├──────────────────────────────────────────────────────┤
│ Execution Layer │
│ Paper Trading → Live Trading → 100+ Exchanges │
└──────────────────────────────────────────────────────┘
284 general factors + 200 crypto-specific factors, organized across categories:
| Category | Count | Examples |
|---|---|---|
| Crypto-Specific | 200 | Funding rate proxy, session effects, whale detection, liquidation cascade |
| Momentum | 14 | ROC, acceleration, trend strength, quality momentum |
| Volume & Flow | 13 | OBV, smart money, volume-price divergence, Wyckoff VSA |
| Volatility | 13 | ATR, Bollinger squeeze, regime detection, vol-of-vol |
| Mean Reversion | 12 | Z-score, rubber band, Keltner position |
| Trend Following | 14 | ADX, EMA golden cross, higher-highs-lows, MA fan |
| Qlib Alpha158 | 11 | KMID, KSFT, CNTD, CORD, SUMP (Microsoft Qlib compatible) |
| Quality Filter | 11 | Earnings momentum proxy, relative strength, resilience |
| Risk Warning | 11 | Consecutive losses, death cross, gap-down, limit-down |
| Top Escape | 10 | Distribution detection, climax volume, smart money exit |
| Price Structure | 10 | Candlestick patterns, support/resistance, pivot points |
| Davis Double Play | 8 | Revenue acceleration, tech moat, supply exhaustion |
| Alpha | 10 | Various alpha factor implementations |
| Gap Analysis | 8 | Gap fill, gap momentum, gap reversal |
| Market Breadth | 5 | Advance-decline, sector rotation, new highs/lows |
| News Sentiment | 2 | EN/ZH keyword sentiment score + momentum |
| DRL Signals | 2 | Q-learning buy probability + state value estimate |
| ... and more | 130 | Fundamental proxy, pullback, bottom confirmation, regime, etc. |
Design principle: Every signal — technical, sentiment, DRL, fundamental — is expressed as a factor returning
[0, 1]. The evolution engine decides the weight. No human bias in signal mixing.
The genetic algorithm continuously discovers optimal strategies:
- Seed — Initialize population with diverse factor weight configurations
- Evaluate — Backtest each DNA with walk-forward validation
- Select — Keep the top performers
- Mutate — Random weight perturbations, crossover, factor addition/removal
- Repeat — Run 24/7 on your machine
# Crypto (primary use case)
finclaw evolve --market crypto --generations 50
# A-shares
finclaw evolve --market cn --generations 50
# With custom parameters
finclaw evolve --market crypto --population 50 --mutation-rate 0.2 --elite 10| Market | Generation | Annual Return | Sharpe | Max Drawdown |
|---|---|---|---|---|
| A-Shares | Gen 89 | 2,756% | 6.56 | 26.5% |
| Crypto | Gen 19 | 16,066% | 12.19 | 7.2% |
⚠️ These are in-sample backtest results on historical data. Real-world performance will be significantly lower. Walk-Forward out-of-sample validation is now enabled by default — always check OOS metrics before trusting any evolved strategy. Usefinclaw check-backtestto verify results andfinclaw paperto paper trade before risking real capital.
Traditional backtesting evaluates each strategy in isolation — overfitted strategies score well on historical data but fail live. FinClaw's Arena Mode (inspired by FinEvo) solves this:
┌──────────────────────────────────────────┐
│ Arena: Shared Market Sim │
│ │
│ DNA-1 ──┐ │
│ DNA-2 ──┤── Same OHLCV data │
│ DNA-3 ──┤── Same initial capital │
│ DNA-4 ──┤── Price impact when crowded │
│ DNA-5 ──┘── Ranked by final P&L │
│ │
│ Overfitted DNA → poor rank → penalized │
└──────────────────────────────────────────┘
- Multiple DNA strategies trade simultaneously in the same simulated market
- Crowding penalty: When >50% of DNA strategies buy the same signal, price impact kicks in
- Overfitted strategies that only work in isolation get penalized by arena ranking
Catch common backtesting pitfalls before trusting results:
python -m src.evolution.bias_cli --all| Check | What it catches |
|---|---|
| Look-ahead bias | Factors that accidentally peek at future data |
| Data snooping | DNA that performs 3x+ better on train vs test (overfit) |
| Survivorship bias | Assets that delisted during the backtest period |
FinClaw ships with 70+ subcommands. Here are the essentials:
# Quotes & Data
finclaw quote AAPL # US stock quote
finclaw quote BTC/USDT # Crypto quote (via ccxt)
finclaw history NVDA # Historical data
finclaw download-crypto # Download crypto OHLCV data
finclaw exchanges list # Show supported exchanges
# Evolution & Backtesting
finclaw evolve --market crypto # Run genetic algorithm evolution
finclaw backtest -t AAPL # Backtest a strategy on a stock
finclaw check-backtest # Verify backtest results
# Analysis & Tools
finclaw analyze TSLA # Technical analysis
finclaw screen # Stock screener
finclaw risk-report # Portfolio risk analysis
finclaw sentiment # Market sentiment
finclaw demo # Full feature demo
finclaw doctor # Check your environment
# AI Features
finclaw copilot # AI financial assistant
finclaw generate-strategy # Natural language → strategy
finclaw mcp serve # MCP server for AI agents
# Paper Trading
finclaw paper # Paper trading mode
finclaw paper-report # Paper trading resultsRun finclaw --help for the full list.
Expose FinClaw as tools for Claude, Cursor, VS Code, or any MCP-compatible client:
{
"mcpServers": {
"finclaw": {
"command": "finclaw",
"args": ["mcp", "serve"]
}
}
}10 tools available: get_quote, get_history, list_exchanges, run_backtest, analyze_portfolio, get_indicators, screen_stocks, get_sentiment, compare_strategies, get_funding_rates.
| Market | Source | API Key Needed? |
|---|---|---|
| Cryptocurrency | ccxt (100+ exchanges) | No (public data) |
| US Stocks | Yahoo Finance | No |
| China A-Shares | AKShare + BaoStock | No |
| News Sentiment | CryptoCompare + AKShare | No |
cd dashboard && npm install && npm run dev
# Open http://localhost:3000- Real-time prices (Crypto, US stocks, A-Shares)
- TradingView professional charts
- Portfolio tracker with live P&L
- Stock screener with filters + CSV export
- AI chat assistant (OpenAI, Anthropic, DeepSeek, Ollama)
- Walk-forward validation (70/30 train/test split)
- Monte Carlo simulation (1,000 iterations, p-value < 0.05)
- Bootstrap 95% confidence intervals
- Arena competition (multi-DNA market simulation)
- Bias detection (lookahead, snooping, survivorship)
- Factor IC/IR analysis with decay curves
- Factor orthogonality matrix (auto-prune redundant factors)
- Turnover penalty in fitness function
- 5,600+ automated tests
- 484-factor evolution engine
- Walk-forward validation + Monte Carlo
- Arena competition mode
- Bias detection suite
- News sentiment + DRL factors
- Davis Double Play factors
- Paper trading infrastructure
- MCP server for AI agents
- DEX execution (Uniswap V3 / Arbitrum)
- Multi-timeframe support (1h/4h/1d)
- Foundation model for price sequences
git clone https://github.com/NeuZhou/finclaw.git
cd finclaw && pip install -e ".[dev]"
pytestWe welcome contributions! See CONTRIBUTING.md for guidelines.
Ways to contribute:
- 🐛 Report bugs
- 💡 Request features
- 🔧 Submit pull requests
- 📝 Improve documentation
- ⭐ Star the repo if you find it useful
FinClaw is a research and education tool. Key limitations:
- Free data sources — subject to delays, gaps, and API rate limits
- Simplified backtesting — does not model order book depth, partial fills, or market microstructure
- In-sample bias — evolved strategies may not perform similarly out-of-sample; always check walk-forward OOS results
- Dry-run first — always validate strategies with paper trading before risking real capital
For production trading, combine with proper risk management and position sizing.