Data Analysis of Investment

A personality-based stock recommendation engine that bridges psychometrics with financial data. This application analyzes user personality traits via a condensed Likert survey to recommend stocks, providing financial data, recent news, and an interactive AI agent for a more personalized experience.

Overview

This project explores the intersection of behavioral psychology and investing. It replaces traditional "risk tolerance" questionnaires with a psychometric approach, mapping fundamental personality traits to market sectors and specific tickers.

How it works:

1. Survey: User takes a quick 9-item personality assessment.
2. Recommendation: Stocks are suggested based on the user's psychographic profile.
3. Stock View: Users view 5-year price history, P/E ratios, dividend yields, and business summaries.
4. Agent Interaction: Users can chat with an AI agent (powered by LLMs + Tavily Search) to ask follow-up questions about the recommendations.

Methodology

While the interface is simple, the backend uses a rigorous data reduction pipeline derived from the "Young People Survey" dataset:

1. Factor Analysis: We started with a raw dataset of 135 survey questions. Using statistical Factor Analysis, we identified the core latent variables, distilling the survey down to 27 key questions.
2. LLM Semantic Mapping: We utilized a Large Language Model to further condense these 27 questions into just 9 high-impact items, ensuring the survey remains engaging without sacrificing predictive power.
3. Real-Time Data: The app integrates pulls data from Yahoo Finance for stock data, Finnhub for weekly news sentiment, and Tavily for live web-search capabilities during the chat and recommendation.

Features

• Psychometric Profiling: A streamlined 9-question Likert scale interface.
• Financial Dashboard: Displays current price, 5-year history charts, P/E ratio, Dividend Yield, and 24hr Volume.
• News Integration: Fetches relevant news articles from the past 7 days via Finnhub.
• RAG-Powered Chat: Discuss portfolio choices with an AI agent that has access to live web search.

Prerequisites

• Python 3.12+
• uv (Recommended) or anything which can set up a python environment from a requirements.txt file.
• API Keys:
  • GROQ API Key (or a local LM Studio server running)
  • Finnhub API Key (for stock data & news)
  • Tavily API Key (for agent web search)

Installation & Setup

1. Clone the repository:

   git clone https://github.com/leonardo-schneider/Factor-9.git
   cd Factor-9

2. Install dependencies:

   uv sync

3. Configure Environment:

   Create a .env file in the root directory and add your keys:

   GROQ_API_KEY="your_groq_key_here"
   FINNHUB_API_KEY="your_finnhub_key_here"
   TAVILY_API_KEY="your_tavily_key_here"

Usage

To launch the Streamlit application:

uv run streamlit run main.py

Note: If you are not using uv, ensure your virtual environment is activated and run streamlit run main.py.

Project Structure

.
├── main.py                         # Application entry point (Streamlit)
├── pyproject.toml                  # Dependency definitions
├── README.md                       # This file
├── requirements.txt                # Dependencies for non-uv managers
├── resources                       # Static assets
│   ├── personality_factors.json
│   ├── prompt_tenets.txt
│   └── survey_traits.json
└── uv.lock                         # Exact dependency versions

---

This project was created for Programming for Data Science at New College of Florida as an exploration of agentic AI and data reduction techniques.