Dissertations, Theses, and Capstone Projects

Date of Degree

6-2024

Document Type

Dissertation

Degree Name

Ph.D.

Program

Economics

Advisor

Sebastiano Manzan

Committee Members

Lilia Maliar

Christos Giannikos

Subject Categories

Econometrics | Economics | Finance

Keywords

BERT Models, Boosting Algorithms, Narrative Economics, Sentiment Analysis, Stock Returns, Twitter Influence Metric

Abstract

Chapter 1 - Narrative Economics and the U.S. Stock Market: Insights from Twitter: This chapter examines the impact of economic narratives disseminated by influential Twitter users on the returns of the S&P500 index. The users are prominent economists working as researchers or policy advisors, institutional investors, and media analysts whose tweets have the potential to become viral and provide credible interpretations of events. To quantify the narratives, I introduce a measure of content propagation that accounts for the sentiment of each tweet adjusted by the influence of the user on the Twitter platform. The measure significantly predicts S&P 500 index returns, particularly amidst the COVID-19 pandemic, even after controlling for COVID-specific news. Further analysis shows that the impact of narratives varies across different market episodes, with the highest impact observed during market downturns. The findings underscore market participants’ tendency to depend on expert opinions during uncertain times.

Chapter 2 - Narrative Themes and Stock Market Trends: An In-Depth Analysis: This chapter explores the predictive influence of individual narratives on stock market trends. By employing state-of-the-art natural language processing techniques for narrative extraction and machine learning methods for predictive analysis, this study identifies key narratives—particularly those related to stock market and COVID-19—as significant predictors of market returns. Furthermore, it employs a phased approach using the XGBoost machine learning method. This involves comparing a baseline model containing historical data against multiple models that incrementally include separate narratives. This procedure resembles a Granger causality framework within the XGBoost context.

Chapter 3 - A Causal Analysis of Twitter Narratives on Stock Market Fluctuations: This chapter diverges from the predictive approach of the preceding chapters, delving into the causal relationship between Twitter narratives and stock market returns through the novel PCMCI+ (Peter-Clark Momentary Conditional Independence +) machine learning methodology for causal inference. It reveals a significant causal link between stock market-related narratives and same-day market performance and highlighting the immediate influence of COVID-19 narratives on market openings. Post-COVID analysis shows that COVID narratives overshadow the impact of sociopolitical and economy-related narratives, while economic narratives are a significant cause for same-day returns before the COVID-19 outbreak. The study also underscores the intricate interplay of daily narratives.

Download

This work is embargoed and will be available for download on Tuesday, November 03, 2026

Share

COinS