Skip to content

Latest commit

 

History

History
187 lines (132 loc) · 4.67 KB

AUTOMATION_README.md

File metadata and controls

187 lines (132 loc) · 4.67 KB

Automated Trading System

This system automates the ChatGPT Micro-Cap trading experiment by integrating LLM APIs to generate trading decisions automatically.

Features

  • Automated Prompt Generation: Creates daily trading prompts with current portfolio data
  • LLM Integration: Supports OpenAI GPT API
  • Trade Execution: Parses LLM responses and executes recommended trades
  • Risk Management: Includes confidence thresholds and dry-run modes
  • Logging: Saves all LLM responses and trading decisions

Setup

1. Install Dependencies

pip install -r requirements.txt
pip install openai

2. Get API Key

3. Set Environment Variable

# For OpenAI
export OPENAI_API_KEY="your-openai-api-key"

Usage

The simple_automation.py script provides automated trading decisions:

# Basic usage with OpenAI
python simple_automation.py --api-key YOUR_OPENAI_KEY

# Using environment variable
export OPENAI_API_KEY="your-key"
python simple_automation.py

# Dry run (no actual trades executed)
python simple_automation.py --dry-run

# Custom model
python simple_automation.py --model gpt-3.5-turbo

How It Works

1. Portfolio Analysis

  • Loads current portfolio state from CSV files
  • Calculates cash balance and total equity
  • Formats holdings data for LLM consumption

2. Prompt Generation

  • Creates structured prompts with portfolio data
  • Includes trading rules and constraints
  • Requests JSON-formatted responses

3. LLM Processing

  • Sends prompts to chosen LLM API
  • Receives trading recommendations
  • Parses JSON responses for trade details

4. Trade Execution

  • Validates trade recommendations
  • Checks cash availability and position limits
  • Executes approved trades (or logs in dry-run mode)

Configuration

LLM Settings

  • Temperature: 0.3 (lower for more consistent decisions)
  • Max Tokens: 1500-2000 (adjust based on model)
  • Model: GPT-4 recommended for best results

Trading Rules

  • Maximum position size: 10% of portfolio
  • Minimum confidence threshold: 70%
  • Cash reserve: Minimum $500
  • Micro-cap focus: <$300M market cap

Output Files

LLM Responses

  • llm_responses.jsonl: All LLM interactions and responses
  • automated_trades.jsonl: All automated trading decisions

Portfolio Updates

  • Uses existing CSV files from the original trading script
  • Maintains compatibility with manual trading system

Safety Features

Dry Run Mode

Always test with --dry-run first to see what trades would be executed:

python simple_automation.py --dry-run

Confidence Thresholds

The system includes confidence scoring to avoid low-quality recommendations.

Error Handling

  • Graceful handling of API failures
  • JSON parsing error recovery
  • Invalid trade validation

Integration with Existing System

The automation system is designed to work alongside the existing manual trading system:

  1. Same Data Files: Uses the same CSV files as the manual system
  2. Compatible Format: Maintains the same portfolio and trade log formats
  3. Fallback Support: Can switch between automated and manual modes

Example Workflow

# 1. Test the system with dry run
python simple_automation.py --dry-run

# 2. Run automated trading
python simple_automation.py

# 3. Review results
cat "Start Your Own/llm_responses.jsonl" | tail -1

# 4. Check portfolio updates
python "Start Your Own/Trading_Script.py"

Troubleshooting

Common Issues

  1. API Key Not Found

    export OPENAI_API_KEY="your-key"

  2. JSON Parsing Errors

    • Check LLM response format
    • Verify model supports structured output
    • Try different temperature settings

  3. Trade Execution Failures

    • Check cash availability
    • Verify ticker symbols
    • Review position size limits

Debug Mode

Add verbose logging by modifying the scripts to print more details about the LLM interactions.

Customization

Custom Prompts

Modify the generate_trading_prompt() function to customize the prompts sent to the LLM.

Different Models

Experiment with different models:

  • gpt-4: Best performance, higher cost
  • gpt-3.5-turbo: Good performance, lower cost

Trading Rules

Adjust risk parameters in the configuration or modify the validation logic.

Security Notes

  • Never commit API keys to version control
  • Use environment variables for sensitive data
  • Consider rate limiting for API calls
  • Monitor API usage and costs

Support

For issues or questions:

  1. Check the troubleshooting section
  2. Review the LLM response logs
  3. Test with dry-run mode first
  4. Verify API key permissions and quotas