Skip to content

Neural Language Processing Project: Earning Calls Sentiment Analysis

Notifications You must be signed in to change notification settings

mtbrr26/Natural_Language_Processing_project

Repository files navigation

Natural Language Processing: Sentiment Analysis of Earnings Call Transcripts

How Do CEOs and CFOs Modify Their Sentiment in Times of Economic Crisis?

By Mathieu Breier, Edward Monbiot, Amber Walker
Date: March 18th, 2024

Abstract

This project explores the dynamic shifts in sentiment expressed by CEOs and CFOs during economic crises and assesses the accuracy of their communications in such times. We analyzed earnings call transcripts from four distinct companies, applying sentiment analysis to texts from both CEOs and CFOs. The sentiment scores obtained were then regressed against share price data to gauge the correlation with actual company performance.

Section 1. Web Scraping Earnings Calls Transcripts

Efficient and Scalable Data Scraping from fool.com

Section Overview

This section focuses on implementing web scraping techniques to gather large datasets efficiently. It provides a scalable framework for data collection, essential for data-driven decision-making processes.

Description
  • Objective: Focused on scraping financial data related to cruises and airlines.
  • Scraping Breakdown:
    • Cruises Scraping: Data scraping from: Royal Caribbean and Norwegian Cruise Line.
    • Airlines Scraping: Data scraping from: Delta Air Lines and Southwest Airlines.
Libraries Used:
  • NumPy
  • Pandas
  • Selenium
  • BeautifulSoup
Highlights:
  • Implementation of a flexible and scalable data collection pipeline using Selenium.
  • Scraped data stored in Pandas dataframes
  • Verification steps to ensure all quaterly earnings transcripts have been completely scraped
  • Company transcripts transformed to CSV files for Data Analysis

Section 2. Sentiment Analysis of Earnings Calls

  • Objective: Analyzing sentiment in company earnings call transcripts during the COVID-19 economic crises.
Contents:
  • Preprocessing: Prepare earning calls transcripts for analysis (lowercasing, stemming, removal of HTML tags, stopwords and extra spaces etc.)
  • Sentiment Analysis: Create pipeline to do sentiment analysis, primary tool: Loughran McDonald Sentiment Analyzer.
  • Sentiment Over Time: Analyzing CEO & CFO sentiment scores during COVID-19 crisis and plot against share price and trading volumes.
  • Regression Analysis: OLS regression to explore relationships between CEO and CFO sentiment and market data.
Libraries Used:
  • Yahoo Finance
  • Seaborn
  • BeautifulSoup
  • Selenium
  • Scikit-learn
Highlights:
  • Creation of a preprocessing pipeline for earnings call transcripts.
  • Aggregated word cloud generation using TF-IDF.
  • Detailed sentiment analysis with a focus on economic downturns.
  • Visualisation of CEO and CFO sentiment trends over time.
  • Regression analysis to quantify the relationship between sentiment and stock performance.

Project Conclusion

Our study revealed that CEOs typically showed higher sentiment scores, but this difference narrowed during economic crises. Notably, CFOs consistently provided a more accurate reflection of company performance relative to share prices.

Further Study

We recommend incorporating time lags into sentiment scores for a deeper analysis of their influence on future share prices and exploring Vector Autoregressive (VAR) models for a more comprehensive understanding of time-dependent variables interactions. Additionally, refining sentiment score computation by normalizing for speech length and style and implementing TF-IDF weighting will allow for a more nuanced and equitable sentiment analysis.

About

Neural Language Processing Project: Earning Calls Sentiment Analysis

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •