Thesis

Financial sentiment beyond text : a multimodal approach to understanding financial market dynamics and investor behaviours

Conteúdo disponível para baixar

Baixar PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2025
Thesis identifier
  • T17422
Person Identifier (Local)
  • 202053449
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Over the past two decades, financial literature has extensively examined textual data to identify key drivers of market activity, with sentiment analysis emerging as a pivotal analytical tool. This thesis investigates the application of a multimodal sentiment analysis model for earnings conference calls, combining textual and audio data, to enhance the accuracy of sentiment classification and allow for deeper insights into financial behaviours to be examined. The research identifies a critical gap in existing literature, in that basic sentiment analysis methods dominate despite their underperformance when compared to state-of-the-art natural language processing (NLP) techniques (El-Haj et al. 2018). Advancing sentiment analysis in the financial domain, Chapter 4 shows that multimodal sentiment analysis significantly outperforms commonly used classifiers in extant literature, in terms of classification accuracy and forecasting capabilities. Chapter 5 reveals that the multimodal model is highly adept at forecasting Cumulative Abnormal Returns (CARs), uncovering a return reversal dynamic between sentiment and CARs that lends support to behavioural theory. This chapter also identifies that framing bias—discrepancies in how information is presented by managers and analysts—intensifies market reactions, emphasizing the role of framing in financial decision-making and providing evidence towards a potential driver of the returns reversal dynamic. Chapter 6 explores the relationship between multimodal sentiment and Cumulative Abnormal Volume (CAV). The findings demonstrate a significant association between multimodal sentiment and long-term trading volume, underscoring the impact of non-verbal communication on investor behaviour. Furthermore, it establishes that sentiment divergence, indicative of disagreement, leads to heightened volume, underscoring the market's sensitivity to conflicting signals. This research demonstrates the power of multimodal approaches in capturing nuanced financial sentiment and its implications for market behaviour. The findings advance both the technical and theoretical understanding of financial sentiment analysis, particularly highlighting the importance of incorporating audio characteristics alongside textual data in the financial domain.
Advisor / supervisor
  • Bowden, James
  • Cummins, Mark, 1968-
Resource Type
DOI

Relações

Itens

Miniatura Título Data de carga Acesso Ações
file details PDF of thesis T17422 2025-07-17 Público Baixar