Quantifying the unquantifiable: bridging sentiment analysis and financial metrics for predictive insights in stock market dynamics
No Thumbnail Available
Journal Title
Journal ISSN
Volume Title
School of Business |
Master's thesis
Author
Date
2024
Department
Major/Subject
Mcode
Degree programme
Finance
Language
en
Pages
62+6
Series
Abstract
This thesis explores the effectiveness, and the predictive capabilities of using sentiment scores derived from earnings calls and compares it against traditional financial metrics in explaining stock price developments. Utilising Natural Language Processed (NLP) pre-computed sentiment scores from S&P Global’s Capital IQ Pro, the study analyses a dataset of 43,697 events from firms listed on the NYSE and NASDAQ, spanning from Q1 2019 to Q4 2023. The primary objective is to determine whether sentiment scores provide incremental predictive value over traditional financial metrics in forecasting stock price movements, particularly in the context of post-earnings announcement drift (PEAD) and potential reversals of initial stock price reactions. The findings indicate that while sentiment scores alone do not significantly predict abnormal returns on the announcement date, they exhibit a significant negative relationship with 30-day cumulative abnormal returns (CAR). This suggests that higher positivity in earnings calls might lead to subsequent underperformance, possibly due to investor scepticism towards overly optimistic presentations amidst poor financial performance. Furthermore, a pattern resembling potential market overreactions was observed, the overreactions are more pronounced when earnings call sentiment amplifies the financial results, leading to an initial spike or drop in CAR that later corrects. Conversely, when sentiment and financial performance are misaligned, the market reaction is more subdued, potentially due to the balancing effect of mixed signals. These results underscore the role of sentiment analysis in financial markets, offering insights for academia, investors, and regulators into the dynamics between qualitative assessments and quantitative performance indicators.Description
Thesis advisor
Nyberg, PeterKeywords
sentiment analysis, stock prices, earnings calls, financial reports, cumulative abnormal returns, natural language processing, machine learning
