Trading System Toolkit (TS)

MyTS is a Python-based framework for designing, testing, and deploying trading strategies. It integrates Large Language Models (LLMs) for strategy generation and utilizes robust backtesting and data handling capabilities.

Project Overview

This project provides a suite of tools for quantitative trading, including:

• Strategy Design & Generation:
  • A Streamlit web application (app.py) for manually designing strategies or using LLMs to generate strategy ideas and code.
  • Agent-based system (main_autogen.py and agents/) that uses an OrchestratorAgent to manage a team of specialized agents (e.g., CodeAgent, DataProvisioningAgent, ReviewerAgent, BacktesterAgent) for an automated strategy development pipeline.
• Data Handling:
  • Fetching historical market data using yfinance (via agents/data_provisioning_agent.py and common_logic/intraday/intelligent_data_fetcher.py).
  • Support for fetching data from Interactive Brokers (via tools/GetIBKRData.py).
  • Data cleaning and validation processes.
• Backtesting:
  • A core backtesting engine built on the Backtrader library (common_logic/backtesting_services/backtrader_runner.py).
  • Calculation of various performance metrics (Sharpe Ratio, PnL, Drawdown, etc.).
• Optimization:
  • Bayesian optimization of strategy parameters (facilitated by agents/bay_agent.py).
• Configuration:
  • Centralized configuration for API keys, file paths, and LLM model mappings (config.py).

Key Components

• app.py: Streamlit web interface for interactive strategy design, data fetching, and backtesting.
• main_autogen.py: Main entry point for the automated, agent-based strategy development workflow.
• config.py: Central configuration file for the system.
• agents/: Contains the definitions for various autonomous agents:
  • OrchestratorAgent: Manages the overall strategy development workflow.
  • DataProvisioningAgent: Fetches and saves market data.
  • CodeAgent: Generates strategy code based on prompts and schemas.
  • ReviewerAgent: Reviews generated code for quality and security.
  • BacktesterAgent: Executes backtests using the Backtrader engine.
  • BayAgent: Performs Bayesian optimization of strategy parameters.
  • Other specialized agents for UI interaction, strategy design, etc.
• common_logic/: Houses shared business logic:
  • backtesting_services/: Core Backtrader integration and runner.
  • data_provisioning/: Modules related to data fetching and preparation (though some yfinance logic is also in agents or intraday).
  • intraday/: Logic specific to intraday data handling and strategy generation, used by app.py.
  • strategies/: (Typically) Stores strategy code or templates.
  • templates/: Strategy code templates.
• tools/: Utility scripts, including the Interactive Brokers data fetching tool.
• shared_work_dir/: Default directory for storing data, generated strategies, and other artifacts.
• tests/: Unit tests for critical components.

Getting Started

(TODO: Add detailed setup instructions, dependencies, and how to run the application/agent system.)

1. Prerequisites:
   • Python 3.8+
   • IBKR TWS or Gateway (if using tools/GetIBKRData.py)
   • API keys for LLM providers (e.g., Gemini) set as environment variables (see config.py).
2. Installation:

   pip install -r requirements.txt
   # (May need separate installs for specific features like ib_insync or autogen variants)

3. Running the Streamlit App:

   streamlit run app.py

4. Running the Automated Agent System:

   python main_autogen.py

Refactoring Notes

This repository has undergone significant refactoring to:

• Improve code readability and organization.
• Enhance robustness through better error handling and data validation.
• Standardize file paths and configurations.
• Remove duplicated and obsolete files.
• Add unit tests for critical components.

Further improvements can include:

• More comprehensive unit and integration testing.
• Detailed setup and usage instructions.
• Parsing strategy parameters (strategy_name, ticker, etc.) from the initial_task in OrchestratorAgent instead of hardcoding.
• Consolidating yfinance data fetching logic into a shared utility.