Financial Services Innovation Lab, Programming Task for Summer Research

This project analyzes SEC-EDGAR 10-K filings for the five Big Tech companies (Amazon, Apple, Google, Meta, and Microsoft), focusing on extracting and presenting insights from the management's discussion and analysis of the firm's performance. The goal is to show users what each firm's management thinks of the firm's performance, helping users gauge the company's future prospects and investment potential.

Overview:

• Task 1: Download Data from the SEC-EDGAR

  • Downloading Filings

    • Utilizes Python's sec_edgar_downloader package to download 10-K filings
    • Users can select any of the five Big Tech companies to download filings.

• Task 2: Text Analysis

  • Natural Language Processing

    • Uses Python's transformers package, specifically the AutoTokenizer and AutoModelForSequenceClassification based on the ProsusAI/Finbert model, designed for financial sentiment analysis
    • Employs nltk (Natural Language Toolkit) for additional NLP tasks, such as tokenizing text into sentences
    • Extracts key information from the filings (i.e. management's discussion and analysis of financial conditions, operations, and market risks) and classifies those sentences as positive, negative, or neutral
    • Focuses on management's discussion to ensure reasonable running time while maintaining as much core information as possible

  • Visualization

    • Calculates the ratio of positive to total strongly sentimental sentences for each year and visually displays this trend across the years
    • Uses linear regression to predict trends in management's reviews for the next five years

  • Value to Users

    • Provides users with a holistic view of the firm's performance from the management's perspective
    • Helps users understand the management's outlook and make informed decisions regarding investments or purchasing products from the company

• Task 3: Construct and Deploy Simple App

  • Note: The app may take about 10-15 minutes to run due to the large volume of text and the NLP/LLM models running in the backend.

  • Local Deployment

    • Run the following command in terminal, in the Fintech-Assignment directory.

      streamlit run streamlit_app.py
      

  • Web Deployment

    • Access at https://fintech-assignment.streamlit.app/

  • Why Streamlit

    • Easy to use, lightweight, and customizable
    • Specifically designed for deploying data science and machine learning applications

• Tools, Libraries, and Frameworks Used

  • Programming Languages: Python, CSS
  • IDE: VS Code
  • Libraries: sec_edgar_downloader, transformers, nltk, torch, beautifulsoup4, lxml, matplotlib, numpy, Streamlit

• Installation (to run the app LOCALLY)

  • Clone the repository:

    git clone https://github.com/sliu750/Fintech-Assignment.git

  • Install the required packages:

    pip install -r requirements.txt

  • Now you can run the application locally.

• Video of Local Deployment: https://www.youtube.com/watch?v=q57IdgLte4w