Backtester AI Agent

This project demonstrates three different implementations of a stock backtesting agent that calculates historical returns using AI:

1. Prompt Chaining Implementation (backtester_agent_prompt_chained.py)

   • Uses pure prompt chaining to guide the model through the calculation steps
   • Simpler implementation but less structured
   • Good for understanding the basic flow of the calculation

2. Function Calling Implementation (backtester_agent.py)

   • Uses OpenAI's function calling feature
   • Partially agentic approach with structured tool definitions
   • Better error handling and validation
   • More robust than prompt chaining

3. LangChain Implementation (backtester_agent_langchain.py)

   • Uses the LangChain framework for full agentic behavior
   • Most structured and maintainable implementation
   • Built-in tools and agent management
   • Best for production use and extensibility

All three implementations provide the same core functionality:

• Calculate historical returns for stocks
• Use Yahoo Finance API for data
• Handle date ranges and price calculations
• Provide detailed step-by-step explanations

Features

• Calculate historical returns for any stock
• Support for custom date ranges
• Detailed step-by-step explanations
• Error handling and validation
• Multiple implementation approaches

Setup

1. Create a virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

2. Install dependencies:

pip install -r requirements.txt

3. Set up environment variables in .env:

YAHOO_FINANCE_SEARCH_API=your_search_api_url
YAHOO_FINANCE_QUERY_API=your_query_api_url

Usage

Choose the implementation that best suits your needs:

1. For simple usage with prompt chaining:

python backtester_agent_prompt_chained.py

2. For function calling implementation:

python backtester_agent.py

3. For LangChain implementation:

python backtester_agent_langchain.py

Example query:

What would have been the return of Apple from 2020-01-01 to 2020-12-31?

Implementation Details

1. Prompt Chaining (backtester_agent_prompt_chained.py)

• Uses sequential prompts to guide the model
• Each step is handled by a separate prompt
• Simpler to understand but less robust
• Good for learning and prototyping

2. Function Calling (backtester_agent.py)

• Uses OpenAI's function calling feature
• Structured tool definitions
• Better error handling
• More robust than prompt chaining
• Good for production use

3. LangChain (backtester_agent_langchain.py)

NOTE: I couldn't get this working with the DeepSeek model in my local but tested it only with OpenAI model(gpt-4o-mini)

• Uses LangChain framework
• Full agentic behavior
• Built-in tools and agent management
• Most maintainable and extensible
• Best for complex applications

Demo

here's the working demo of this agent with this user input:

What would have been the return of Apple from 2020-01-01 to 2020-12-31?

Backtester_AI_Agent_demo_.mov

Error Handling

The script includes comprehensive error handling for:

• Invalid company names
• Date format errors
• API connection issues
• Data validation
• Calculation errors

Limitations

• Requires valid API endpoints for Yahoo Finance data
• Date range must be in YYYY-MM-DD format
• Company names must be recognizable by the Yahoo Finance API
• Historical data availability depends on the stock's listing date

Contributing

Feel free to submit issues and enhancement requests!