🌐 Available Versions: 🇰🇷 한국어 (Korean) | 🇯🇵 日本語 (Japanese)

---

LLM-based ESG-Focused Portfolio Optimization Service📊🌱

🏆🥈Excellence Award(2nd) at the 8th Industry-Academia Software Project Exhibition of Kwangwoon University
🏆🥈Excellence Award(2nd) at the 2024 Graduation Exhibition of Kwangwoon University College of AI Convergence

This repository hosts the LLM-based ESG-Focused Portfolio Optimization Service project. By integrating ESG (Environmental, Social, Governance) criteria and advanced portfolio optimization techniques, this service provides users with a personalized ESG-driven investment portfolio. Using a large language model (LLM) to process and evaluate text data, the service enables investors to assess companies' ESG scores and generate optimized portfolios based on their preferences.

Project Introduce focused on UI/UX

If you want to watch UI/UX introducing Video, Click the below video Thumbnail.

(Sorry but please right-click the Thumbnail and select 'Open link in new tab', if you want keep this tab.)

Table of Contents

1. Project Overview
2. Key Features
3. Installation
4. Usage
5. Project Structure
6. Technologies Used
7. Requirements
8. Future Extensions
9. New Features
10. Version History
11. Contributing
12. Support and Feedback
13. License
14. About

Project Overview

In recent years, ESG has become a critical metric for assessing a company's long-term sustainability and stability. However, traditional ESG evaluation methods often lack transparency and accessibility for non-expert investors. This project addresses these issues by providing a user-centered, transparent ESG evaluation system using an LLM, which performs text-based assessments and applies the Black-Litterman model for portfolio optimization. Through this, the system generates customized portfolios that balance ESG factors and financial returns according to user preferences.

Core Objectives

1. Technical Implementation: Develop a pipeline using LLM for ESG assessment and produce an optimized portfolio.
2. User-Centered Service: Provide personalized portfolio construction with ESG evaluation based on user-selected criteria and preferences.

Key Features

1. Text Data Collection and Processing

   • Collect articles related to selected companies from various sources using Selenium and BeautifulSoup, storing the data in MongoDB for efficient management.
   • Preprocess the collected data, anonymizing company names and filtering irrelevant content to ensure high data quality.

2. LLM-based ESG Scoring

   • Use an LLM (via the OpenAI API) to evaluate collected articles based on ESG relevance, extracting and labeling content by ESG factors and sentiment.
   • Fine-tune a KoElectra model for classification, allowing independent ESG assessments without API dependency.

3. Comprehensive ESG Evaluation

   • Utilize user input for adjusting weights on ESG factors (environmental, social, governance) to customize scoring.
   • Adjust scores dynamically based on investment style preferences: financial-centered, ESG-centered, or balanced.
   • Apply a scoring model that integrates multiple ESG rating agencies' criteria to build a comprehensive score, enabling evaluation for companies outside major agency coverage.

4. Enhanced Portfolio Optimization via Black-Litterman Model

   • Implement year-by-year optimization approach using ESG data from multiple rating agencies (MSCI, S&P, Sustainalytics, ISS, ESG기준원).
   • Utilize the Ledoit-Wolf method for robust covariance matrix estimation, improving numerical stability.
   • Apply user trust weights to ESG scores, creating a personalized view vector for the Black-Litterman model.
   • Calculate optimized portfolio weights by maximizing the Sharpe ratio through quadratic programming with scipy.optimize.
   • Store and analyze yearly investment weights to track portfolio performance over time (2020-2025).
   • Implement robust error handling for data gaps and numerical computation issues.

5. User Interface with Streamlit

   • A user-friendly interface allows users to input ESG preferences and visualize optimized portfolios.
   • Interactive sliders for ESG weights and investment styles, providing an intuitive way to construct personalized portfolios.
   • Visualization of portfolio metrics, including expected return, volatility, and Sharpe ratio.

Installation

1. Clone the Repository:

   git clone https://github.com/fairyofdata/LLM-ESG-POS.git
   cd LLM-ESG-POS

2. Install Required Packages:

   pip install -r requirements.txt

3. Run Streamlit Application:

   streamlit run app.py

Usage

1. Data Collection: Configure company tickers and data sources for article collection.
2. Set ESG Preferences: Use the Streamlit UI to define weights for environmental, social, and governance factors.
3. Investment Style Selection: Choose from financial-centered, ESG-centered, or balanced portfolio options.
4. View Portfolio Results: See portfolio composition and performance metrics, including expected return and volatility.

Project Structure

├── data/                   # Data and ESG score tables
├── src/                    # Source code for ESG scoring and portfolio optimization
│   ├── data_processing.py   # LLM-based text processing
│   ├── esg_scoring.py       # ESG scoring functions
│   ├── portfolio_optimization.py  # Optimization with Black-Litterman
│   └── ui/                  # Streamlit UI code
├── app.py                  # Main application script
├── README.md               # Project documentation
└── requirements.txt        # List of dependencies

Technologies Used

• Data Processing: Python, Pandas, NumPy
• Web Scraping: Selenium, BeautifulSoup4
• Database: MongoDB
• Machine Learning: scikit-learn, KoElectra
• Portfolio Optimization: Black-Litterman model, SciPy, cvxopt
• Data Visualization: Plotly, Matplotlib
• UI/UX: Streamlit
• LLM Integration: OpenAI API

Requirements

• Python 3.8+
• scikit-learn 1.0.0+
• pandas 1.3.0+
• numpy 1.20.0+
• streamlit 1.10.0+
• yfinance 0.1.70+
• FinanceDataReader 0.9.0+

Future Extensions

1. Expanded ESG Evaluation: Integrate additional ESG rating sources and add real-time news updates for responsive scoring.
2. Scalability for Non-listed Companies: Develop scoring mechanisms that leverage LLMs to assess startups and non-public companies.
3. Improved Constraints in Optimization: Include sector-specific constraints for better portfolio diversification.
4. Time-Series Portfolio Analysis: Extend the year-by-year optimization approach to provide historical performance comparisons and trend analysis.

New Features (2024 Update)

• Year-by-Year Portfolio Optimization: Track investment performance across multiple years
• Multi-Agency ESG Data Integration: Comprehensive scoring from five major ESG rating agencies
• Enhanced Stability: Improved covariance estimation and numerical computation techniques
• Interactive Portfolio Visualization: Detailed insights into portfolio composition and performance

Version History

• v2.0 (March 2025): Enhanced optimization algorithm and multi-agency ESG data integration
• v1.0 (November 2024): Initial release with basic ESG scoring and portfolio optimization

Contributing

Contributions are welcome! If you'd like to improve this project, please follow these steps:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes and commit (git commit -m 'Add amazing feature')
4. Push to your branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Support and Feedback

If you encounter any issues or have suggestions for improvements, please submit an issue on our GitHub repository.

License

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

About

KWU 8th Industry-Academic Cooperation SW Project & AI Convergence Graduation Project

• Adviser: Professor Cho Min-soo (Department of Information Convergence at Kwangwoon Univ.)
• Applied company: Billions Lab (Dr. Soo-jin Pyo)
• Contributor: Team KWargs (Baek Ji-heon (PM), Kim Na-yeon, Jang Han-jae)