Stock ARA

Stock AI Research Assistant

Stock ARA(AI Research Assistant) is an innovative stock screening and portfolio optimization system that leverages a Large Language Model(LLM) and Retrieval-Augmented Generation(RAG).

This is an official implementation of the paper: A Novel Stock Screening Approach using Large Language Models and Correlation-Aware Retrieval.

Key Features

• Advanced stock screening using LLM and RAG
• User query augmentation for improved search accuracy
• Correlation-aware retrieval combining business report similarity and stock return correlation
• Portfolio optimization based on the Mean-Variance model and Black-Litterman approach
• Cost-efficient use of GPT-3.5 and GPT-4 APIs

Get Started

Technical Specifications

• Python
• OpenAI GPT API (GPT-3.5 and GPT-4o)
• PostgreSQL with pgvector and Timescale
• Redis
• Docker

Data Sources

• KRX listed stock information from KRX Market Data System
• Business reports from DART
• Stock price data from Yahoo Finance

Usage

1. Clone the repository

   git clone https://github.com/Astro36/stock-ara.git

2. Set up the required environment variables: .env

   • APP_ID: app name, used in database (ex. demo)
   • POSTGRES_PASSWORD: user password for PostgreSQL
   • OPENAI_API_KEY: see OpenAI API Key
   • OPENDART_API_KEY: see Open DART API Key (optional)
   • TELEGRAM_BOT_TOKEN: see Telegram BotFather

3. Load stock data and business reports into the database

   COPY companies FROM '/tmp/db_data/companies.csv' DELIMITER ',' CSV HEADER;
   COPY company_filings FROM '/tmp/db_data/company_filings.csv' DELIMITER ',' CSV HEADER;
   COPY assets FROM '/tmp/db_data/assets.csv' DELIMITER ',' CSV HEADER;
   COPY asset_stocks FROM '/tmp/db_data/asset_stocks.csv' DELIMITER ',' CSV HEADER;
   COPY asset_prices FROM '/tmp/db_data/db_data/asset_prices.csv' DELIMITER ',' CSV HEADER;
   
   CALL refresh_continuous_aggregate('asset_weekly_close_prices', '2020-01-01', '2024-12-31');
   REFRESH MATERIALIZED VIEW stock_market_caps;

4. Run docker compose

   docker compose up