Trading Analysis App

A professional-grade trading analysis application built with Python, Poetry, and clean architecture. It leverages LLMs (via LiteLLM) and yfinance to provide a comprehensive analysis of any stock.

Features

1. Company Health Analysis: Detailed look at cash, debt, market valuation, and growth projections.
2. Market Value Evaluation: Assessment of whether the stock is undervalued, overvalued, or fairly valued based on its health.
3. Technical Analysis: Automatic identification of key support and resistance levels.
4. Social Sentiment: Analysis of recent news and social sentiment (Good/Bad).
5. News Summary: Bullish and bearish summary of the latest key data.
6. Multi-Judge Perspective: Evaluation from three professional viewpoints:
   • Professional Trader: Focused on price action, technicals, and regime alignment.
   • Professional Analyst: Focused on fundamentals, valuation, and peer comparison.
   • Professional Risk Manager: Focused on downside risks (VaR), drawdown, and portfolio impact.
7. Final Recommendation: A definitive "Buy", "Hold", or "Sell" response for the final user.

Support for any LLM

The app uses litellm, allowing you to use any major LLM provider:

• Cloud Providers: OpenAI (GPT-4), Anthropic (Claude), Google (Gemini), etc.
• Local Providers:
  • Ollama: Use --model ollama/llama3 (ensure Ollama is running).
  • LM Studio: Use --model lm_studio/your-model-name (ensure LM Studio's local server is running).

Setup

1. Install dependencies using Poetry:

   poetry install

2. Set up your environment variables: Copy .env.example to .env and add your API keys.

   cp .env.example .env

Usage

Run the analysis for any stock ticker:

poetry run python main.py AAPL

You can also specify a different model:

poetry run python main.py TSLA --model anthropic/claude-3-5-sonnet-20240620

Token Usage & Costs Analysis

A single analysis run involves 7 LLM calls to process different aspects of the stock (Health, Value, Sentiment, 3 Judges, and Final Resolver).

Estimated Token Consumption

For a standard stock analysis (e.g., AAPL):

• Input Tokens: ~10,000 - 15,000 tokens (includes financials, news titles, quantitative reports, and previous judge reasoning).
• Output Tokens: ~2,000 - 3,000 tokens (includes reasoning, summaries, and structured JSON decisions).

Estimated Cost per Run

Based on standard model pricing (per 1M tokens):

Model Class	Example Models	Est. Cost / Run
Flagship	GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro	$0.06 - $0.12
Efficient	GPT-4o-mini, Gemini 1.5 Flash	<$0.005
Local	Ollama (Llama 3), LM Studio	Free ($0.00)

Note: Gemini models often have a free tier via Google AI Studio for limited usage.

Cost Optimization Tips

1. Use Efficient Models: For daily tracking, gpt-4o-mini or gemini-1.5-flash provide excellent results at a fraction of the cost.
2. Local Inference: Use Ollama or LM Studio to run models locally on your hardware for zero token cost.
3. Selective Analysis: The modular architecture allows for future expansion where specific engines can be disabled to save tokens.

Architecture

The project follows Clean Architecture principles:

• domain/: Core business logic, models, and interfaces.
• infrastructure/: External implementations (Data providers, LLM services).
• application/: Use cases and orchestration.
• main.py: Entry point and CLI.

License

This project is licensed under the MIT License - see the LICENSE file for details.