Swing Trading Bot — Dual-Track Project

Track 1: Technical architecture for a realistic, production-grade swing trading engine
Track 2: Income modeling — what it actually takes to make $5-15K/year supplemental

Why Swing Trading (Not Day Trading)

	Day Trading	Swing Trading
Holding period	Seconds to minutes	1-10 days
Competition	HFT firms, market makers with nanosecond infrastructure	Fund managers, other systematic traders
Latency sensitivity	Extreme (<1ms edge matters)	Moderate (seconds are fine)
Transaction costs	Compound rapidly, dominate edge	Manageable with 2-5 trades/week
PDT rule	$25K minimum or blocked	Cash account works, no PDT anxiety
ML approach	Deep learning mostly fails at intraday	XGBoost/LightGBM on daily/weekly factors works
Realistic edge	Nearly impossible for retail	Small but achievable

Project Structure

swing-trading-bot/
├── README.md                    # You are here
├── docs/
│   ├── architecture.md          # Full technical architecture
│   └── income-model.md          # Realistic income math & capital planning
├── config/
│   └── config.yaml              # Bot configuration
├── src/
│   ├── main.py                  # Entry point
│   ├── data/
│   │   ├── fetcher.py           # Data acquisition (Alpaca, Polygon, Yahoo)
│   │   └── features.py          # Factor engineering pipeline
│   ├── strategies/
│   │   ├── base.py              # Strategy interface
│   │   ├── mean_reversion.py    # Pairs/mean reversion strategies
│   │   ├── momentum.py          # Trend/momentum strategies
│   │   └── ml_strategy.py       # XGBoost/LightGBM signal strategy
│   ├── backtesting/
│   │   └── engine.py            # Walk-forward backtesting engine
│   ├── execution/
│   │   └── broker.py            # Broker API (Alpaca)
│   ├── risk/
│   │   └── manager.py           # Position sizing, circuit breakers, drawdown limits
│   └── utils/
│       ├── metrics.py           # Sharpe, Sortino, max drawdown, profit factor
│       └── logging_config.py    # Structured logging
└── requirements.txt

Quick Start

# Clone
git clone https://github.com/aisavvyguy/swing-trading-bot
cd swing-trading-bot

# Install dependencies
pip install --break-system-packages -r requirements.txt

# Paper trade (free — no real money)
python src/main.py --mode paper --strategy mean_reversion

# Live trading (requires Alpaca API keys)
python src/main.py --mode live --strategy ml_signal

The Dual-Track Philosophy

Track 1 (Engineering): Build this as a legitimate software project. Clean architecture, testable components, proper data pipeline, walk-forward backtesting. Even if the strategies never make money, you'll have built production-grade ML infrastructure that demonstrates serious engineering skill — valuable for any AI/ML role.

Track 2 (Income): Read docs/income-model.md before risking a dollar. The math is brutal but honest. This project is designed to help you find a small, sustainable edge — not to promise riches. Target: $5-15K/year supplemental income from $50-100K capital at 10-15% annual returns.

Key Design Decisions

Decision	Choice	Why
ML framework	XGBoost + LightGBM	Only ML approach with documented real-world quant success
Data source	Alpaca (free tier) + Yahoo Finance (supplement)	Low cost, production-grade API
Broker	Alpaca	Best retail algo trading API, paper trading, Python SDK
Backtesting	Custom walk-forward engine	Off-the-shelf libraries (backtrader, zipline) hide slippage assumptions
Strategies	Mean reversion + momentum + ML signals	Diversified across regimes
Risk management	Half-Kelly, volatility-adjusted, circuit breakers	Surviving bad years is more important than maximizing good years
Language	Python 3.12+	Ecosystem: pandas, numpy, sklearn, xgboost, lightgbm, alpaca-py

Important Warnings

⚠️ This is educational infrastructure, not financial advice. The starter strategies are deliberately simple and untuned. Finding a real edge requires months of research, feature engineering, and out-of-sample validation.

⚠️ Never trade money you can't afford to lose. Start with paper trading. Stay there for 6+ months. Then start with $5,000. Scale only after consistent profitability over multiple market regimes.

⚠️ Past performance does not predict future results. Walk-forward backtesting helps but doesn't eliminate regime risk. Every strategy eventually encounters a market it wasn't designed for.

License

MIT — see LICENSE file.