Working Papers

We study whether real-time, unscripted communication by fund managers  persuades public-market investors. Using live-streamed sessions by Chinese ETF managers, we apply a multimodal ML framework to quantify facial, vocal, and textual affect and summarize them in Affective Delivery Index (ADI). A one-standard-deviation increase in ADI raises next-day (next-week) fund flows by 0.17 pp (0.6 pp), with no predictability for subsequent returns, consistent with persuasion rather than information. Facial expressions are the dominant modality. Effects are stronger in down markets, for ETFs with higher retail ownership, and remain after controlling for contemporaneous/past returns and flows as well as day, ETF-index-week, fund-month, family-week, and fund-day-of-the-week fixed effects.  We identify exogenous variation in ADI using instrumental-variables approach. Overall, our results indicate that live manager delivery reallocates capital in the short run and moves trading and prices even absent new information.

Employing a novel dataset of 304,071 U.S. news articles covering 44 countries, we investigate the impact of U.S. media-driven narratives of foreign countries on the cross-border merger and acquisition (M&A) activity of U.S. firms. We find that media narratives significantly affect the M&A decisions of corporate managers and corresponding investor reactions. The effect is driven by negative narratives while positive narratives have no impact. Specifically, a more negative narrative toward a foreign country leads U.S. firms to be more likely to abandon previously initiated M&A attempts, less likely to initiate cross-border M&As within that country and elicits more negative stock market reactions to initiated M&As within that country. Distinct from the informational content of news coverage, we provide evidence that negative narratives per se influence M&A decisions, indicating that media-driven narratives of foreign countries shape both corporate and investor behavior in international markets.

We examine whether public perception of the Securities and Exchange Commission (SEC) influences engagement with U.S. financial markets by developing a novel measure of SEC perception derived from tweets mentioning the SEC. Using this measure, we first document considerable time-series variation in the public's views of the primary financial market regulator; they hold a positive, neutral, and negative perception 29%, 58%, and 13% of the time, respectively. Then we examine the influence of SEC perception on the general public's engagementoperationalized with retail investor trading activities-with the stock market. We find that retail trading activity is significantly associated with SEC perception: daily retail turnover is 3.6% higher during periods of positive perception and 3.4% lower during periods of negative perception, relative to neutral periods. This relationship is more pronounced for firms where SEC oversight is particularly important-small firms and those with low institutional ownership-and when there is greater consensus in public perception. We also find that SEC perception influences retail trading behavior around earnings announcements: their trading volume is positively associated with SEC perception, and they rely more heavily on earnings information during periods of positive perception, consistent with enhanced perceived credibility of SEC-regulated disclosures. Collectively, our results highlight that the public's perception of the SEC shapes market engagement and information usage.

We document the evolving corporate and investor responses to the politicization of social issues and its implications for firm value. Utilizing all tweets posted by S&P 500 companies between 2018 and 2023, we identify company-level characteristics associated with pro-LGBTQ+ tweets. We validate the passage of the Florida Parental Rights in Education Act (FPREA) in March 2022 as a salient political event that shifted both public and political sentiment against LGBTQ+ rights. Leveraging a Regression Discontinuity in Time (RDiT) framework, we document a decline in the volume of pro-LGBTQ+ tweets following March 2022, primarily driven by a reduction in neutral-toned tweets. These results are further corroborated using a difference-indifferences (DID) approach. In an event-study setting, we find that pro-LGBTQ+ tweets were associated with positive abnormal stock returns prior to the politicization event, while the market reaction became muted thereafter. This differential effect is most pronounced among companies headquartered in Democratic-leaning states. Collectively, our findings highlight the strategic considerations firms face when navigating socio-political discourse, and underscore the importance of aligning public communication with shifting political and investor climates.

This study looks at how Trump-related events affect financial markets by analyzing urgency scores and sentiment indicators based on topic models. We use all posts from Donald Trump's Truth Social account between 2022 and 2025 to see how his messages influence the volatility of S&P sector indices and other market indices. Three topic modeling methods are used: Latent Dirichlet Allocation (LDA), Joint Sentiment-Topic (JST), and a reversed version of JST (rJST). These models help classify topics and measure sentiment at the same time. Using the improved rJST model, we find clear shifts in sentiment across different topics, especially in 2024 and 2025. We identify high-volatility periods as key events and examine how Trump's posts during these times lead to abnormal returns. Our results show that different sectors react differently. Sectors such as Utilities, Materials, Information Technology, Industrials, and Financials are especially sensitive. Negative comments often cause stronger and longer market reactions than positive ones. This shows that analyzing sentiment by topic helps explain investor behavior. A 10-day event window gives the clearest picture of these market effects.

Social media has amplified the reach of financial influencers known as "finfluencers, " who share stock recommendations on platforms like YouTube. Understanding their influence requires analyzing multimodal signals like tone, delivery style, and facial expressions, which extend beyond text-based financial analysis. We introduce VideoConviction, a multimodal dataset with 6,000+ expert annotations, produced through 457 hours of human effort, to benchmark multimodal large language models (MLLMs) and text-based large language models (LLMs) in financial discourse. Our results show that while multimodal inputs improve stock ticker extraction (e.g., extracting Apple's ticker AAPL), both MLLMs and LLMs struggle to distinguish investment actions and conviction-the strength of belief conveyed through confident delivery and detailed reasoning-often misclassifying general commentary as definitive recommendations. While high-conviction recommendations perform better than low-conviction ones, they still underperform the popular S&P 500 index fund. An inverse strategy-betting against finfluencer recommendations-outperforms the S&P 500 by 6.8% in annual returns but carries greater risk (Sharpe ratio of 0.41 vs. 0.65). Our benchmark enables a diverse evaluation of multimodal tasks, comparing model performance on both full video and segmented video inputs. This enables deeper advancements in multimodal financial research. Our code, dataset, and evaluation leaderboard are available under the CC BY-NC 4.0 license. 

This paper explores the relationship between investor sentiment and stock price volatility, focusing on Tesla Inc. as a case study. The study analyses how social media activity, public perception, and news coverage contribute to short-term price fluctuations. Using Tesla's stock data and public events between 2020-2023, we demonstrate how sentiment-not just fundamentals-can trigger significant price swings. The findings support behavioural finance theories and suggest further exploration using sentiment analysis tools.

We investigate whether differences in information among investors about the environmental, social, and governance quality of firms ("ESG divergence") impact the liquidity and volume of the firms' stocks. We find that, when ESG investors are more interested in a stock and overweight it in their portfolios, lower ESG divergence decreases spreads and increases volume, a finding that is in line with the implications of rational information asymmetry models. When ESG investors are less interested in a stock and underweight it, lower divergence means lower volume and larger spreads, consistent with predictions of differences of opinion models. The social dimension of ESG appears to be the main driver behind the relationships we uncover.

Using proprietary account data, we study the intermediation of blocks sold by insiders. We find that dealers acquire information during the negotiation process, which they exploit when selling the shares on the exchange. Prior to a 2017 regulatory change, dealers unwound insider blocks more quickly than non-insider blocks, particularly when future stock returns are less favorable. These shares were primarily sold to small retail investors. Evidence from message boards suggests that the sentiment of these investors could have been manipulated. The 2017 regulation reduced excess dealer profits on insider blocks and curtailed the exploitation of retail investors.

Fake social media accounts, or bots, are often used to spread or amplify content on social media platforms. I examine the effect of this artificial amplification on capital markets. I obtain a large dataset of posts from the social media platform Twitter (now X) and use several methods to identify accounts that are likely bots. The descriptive analysis suggests bot accounts produce about a third of all tweets mentioning company tickers. Bot tweets are associated with higher market activity, particularly among retail traders. Bot tweets are generally not associated with return reversals, but those with stock price-related content are. Bot tweets are generally associated with increased liquidity, but those with earnings-related content are associated with reduced liquidity, consistent with higher informed trading. Using an exogenous shock where Twitter suspended a large number of bots, I find evidence consistent with bots impacting capital markets. Overall, the results suggest that fake social media accounts may amplify both helpful and harmful signals to investors.

This study sheds light on the "social" (relational) dimension of social media. We present novel evidence that firms use positive peer tweets as a form of corporate social signaling to indicate economic ties with other firms. By analyzing firms that share positive information about their peers, we examine the determinants, information content, and capital market consequences of positive peer tweeting. We find that positive peer tweeting is associated with economic relatedness, the demand for legitimacy, and litigation risk. These tweets predict future inter-firm collaborations, conveying information not readily available through traditional disclosures channels. Furthermore, positive peer tweets generate significantly positive market reactions for both the tweeting and tweeted firms. However, this corporate social signaling is not without costs: tweeting firms experience negative market reactions when the firms they have tweeted about subsequently release negative news. These findings highlight social media's social dimension beyond its well-studied disclosure and dissemination roles.

This paper develops daily market-wide sentiment and attention indexes derived from millions of posts across major investor social media platforms. We find that sentiment extrapolates from past market-wide returns and exhibits a strong reversal. In contrast, attention predicts negative returns as a continuation of previous trends. The two indexes have distinct predictions for aggregate trading: abnormal trading rises when sentiment is low and attention is high. To identify the drivers of attention and sentiment, we use a shock to data sharing networks: We find sentiment spreads through real firm connections while attention does not. Moreover, attention rises after abnormally high trading, while sentiment rises after abnormally high returns. This extrapolative return pattern is asymmetric, primarily driven by negative market jumps. These findings provide new evidence on the daily market dynamics of sentiment and attention.