More than optimism and pessimism: investor emotions and stock returns

Sun, Ruipei (2025) More than optimism and pessimism: investor emotions and stock returns. PhD thesis, University of Glasgow.

Abstract

This PhD thesis seeks to advance the fields of social economics and finance, asset pricing, and behavioural finance by investigating the influence of investor psychology on asset pricing through the analysis of social media data. The research spans three empirical chapters that explore how investor emotions impact the U.S. stock market, utilising both textual and visual data from StockTwits over the period 2011 to 2022, with analysis conducted at both daily and intraday frequencies. The study is distinguished by its innovative theoretical and methodological approaches. On the theoretical front, it is one of the first to incorporate multidimensional emotion models into financial research. Methodologically, it employs advanced machine learning techniques to extract reliable emotion indicators from social media platforms. The findings from all three chapters contribute significantly to the understanding of how different emotion dimensions influence stock market dynamics, providing valuable insights for investment strategists, risk managers, and policymakers concerned with the effects of investor emotion as conveyed through social media.

The first empirical chapter focuses on the impact of various dimensions of investor emotion on stock market performance, specifically analysing user-generated content on StockTwits. Through the application of machine learning algorithms, I developed a comprehensive lexicon to construct daily indices of investor emotions, guided by the Valence-Arousal-Dominance (VAD) model by Russell and Mehrabian (1977) and Plutchick’s (2001) wheel of emotions model. The analysis reveals a counter-cyclical relationship between investor pessimism and S&P 500 index returns, marked by subsequent reversals. Furthermore, there is a statistically significant positive correlation between investor arousal and dominance with stock market returns, with emotion states characterised by high arousal and dominance exerting a strong and positive influence on equity market performance in the following trading days.

In the second empirical chapter, I introduce an innovative method to forecast intraday returns of the S&P 500 index ETF using emotion analysis of StockTwits posts. This approach combines web scraping with lexicon-based emotional analysis at 30-minute intervals. The study enhances the existing emotion lexicon by integrating specific unigrams from the Loughran and McDonald (2011) and Renault (2017) word lists using the GloVeinitiated Neural Network Mapping Method. Additionally, I assign valence, arousal, and dominance scores to frequently used emojis, aligned with the VAD model, for the first time. The empirical results demonstrate that these emotion indices positively predict intraday returns in the S&P 500, though these effects often reverse within the same trading day. Notably, the emotion impact endures through at least four subsequent 30-minute trading intervals, highlighting the significant economic implications for asset management and indicating that noise traders largely drive return predictability.

The third empirical chapter extends the analysis to examine the influence of emotion dimensions on daily returns of the S&P 500 index, incorporating the examination of images alongside traditional text and emojis. Using machine learning algorithms, I construct daily indices of emotion dimensions derived from both visual and textual content on StockTwits. The findings suggest that image-based emotion indices are robust predictors of stock market returns, even when controlling for the influence of traditional texts and emojis. Moreover, images with realistic qualities evoke stronger emotion responses after excluding cartoonish visuals, leading to a more pronounced effect on market returns. This observation highlights the amplifying role of realistic images in intensifying the emotion impact of the associated textual content.

Additional tests have been conducted across the three empirical chapters, including controls for traditional news emotions, an examination of the influence of investor emotions on size factors, and a preliminary exploration of their impact on market volatility. These tests further validate the importance of analysing investor emotions through the lens of the emotion dimension model.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Due to copyright issues this thesis is not available for viewing.
• Subjects: H Social Sciences > HG Finance
• Colleges/Schools: College of Social Sciences > Adam Smith Business School > Accounting and Finance
• Supervisor's Name: Hung, Dr. Daniel and Siganos, Professor Antonios
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-84857
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 04 Feb 2025 14:33
• Last Modified: 04 Feb 2025 14:38
• Thesis DOI: 10.5525/gla.thesis.84857
• URI: https://theses.gla.ac.uk/id/eprint/84857