GIFfluence: A Visual Approach to Investor Sentiment and the Stock Market
Abstract
or prosaic. GIFs are also more succinct than text or still images. This brevity enhances the emotion intensity and appeal of a message, a phenomenon highlighted by Potter et al. (2014). GIFs offer a unique immediacy of experience, being instantly understandable and effective at expressing emotions and telling stories. Bakhshi et al. (2016) found that GIFs are more engaging than text or static images for these reasons. Given the overwhelming amount of information available to investors with limited attention spans, these features of GIFs make them an efficient way to communicate and capture sentiment. Finally, GIFs also have the ability to depict sequences of event, making them particularly suited to conveying understandings of past or future events. This allows GIFs to represent simple mental models or narratives about the stock market (Shiller (2017), Hirshleifer (2020), and Andre, Schirmer and Wohlfart (2024)). For example, a GIF of a rocket launching towards the moon can be used to represent an anticipated rapid rise in the price of a stock, thereby inducing or reflecting investor sentiment. Based on these considerations, we use GIFs to construct a novel proxy for investor sentiment. Specifically, we introduce a daily aggregate market-level investor sentiment index, GIFsentiment, derived from GIFs embedded in messages posted on Stocktwits.com, a leading online platform for sharing opinions about stocks and financial assets. We then examine the relation between GIFsentiment and market outcomes, comparing it to other established sentiment proxies from previous literature to assess its incremental predictive power. We first examine whether GIFsentiment is associated with sentiment proxies from the previous literature. Behavioral theories predict that an investor sentiment proxy will be positively associated with contemporaneous returns as overvaluation grows and will negatively predict returns in subsequent periods when overvaluation is corrected. To evaluate whethe