Files

Abstract

This paper investigates the use of sentiment analysis and topic modeling to uncover hidden themes and sentiments in annual reports of publicly traded companies, offering a novel way to assess market trends that may not be immediately visible through traditional analysis. Sentiment analysis using BERT tracks sentiment trends across industries and the broader market over time, focusing on reports from companies of Dow Jones Index and top firms in semiconductors, IT, and energy industries. Topic modeling using LDA identifies evolving topics in reports from the leading companies in each industry. These findings are then compared with stock price movements to assess their predictive power. The results indicate a R-squared correlation of around 0.5 between the sentiment and stock performance in the semiconductor and information technology sectors, while the energy sector shows a R-squared correlation of around 0.1. Topic modeling effectively captures industry-specific developments and legal challenges, though it overlooks significant events at times. This study demonstrates the potential of NLP to enhance the analysis of qualitative data in annual reports, offering researchers and analysts a new method to assess company performance and market trends.

Details

Actions

from
to
Export