Working Papers

We study non-traditional investment research as a source of information for corporate strategic decisions, such as investments into innovation. Using a comprehensive sample of social media analyst reports from Seeking Alpha and exogenous variation in social media analysts' coverage overlaps, we show that firms are more likely to invest into technologies similar to firms covered by the same analyst. The effect varies with social media analysts' characteristics and differences in their contributed content that capture their unique information set. Overall, our results are consistent with non-traditional investment research enhancing firms' information environment as an additional source of information that guides corporate strategic decisions.

Does social media amplify bank fragility absent systemic crises (e.g., the SVB crisis)? Using a sample of U.S. commercial banks from 2009 to 2022, we show that heightened Twitter attention increases the sensitivity of non-core deposits-but not core deposits-to bank performance deterioration. This effect intensifies for banks with greater liquidity mismatch and when Twitter discussions are more influential. Neither enhanced bank transparency nor negative sentiment in social media discussions explains these results. Our findings indicate that social media is not merely an information transmitter; it heightens depositors' awareness of peer attention to banks, amplifying deposit outflow sensitivity to weak fundamentals even during calm periods.

Our research augments the expanding body of literature concerning the capability of prevalent LLM tools in supporting financial professionals. We introduce a framework to leverage ChatGPT to assess market sentiment through the analysis of social media data. We use the LLM models to construct market sentiment indicators based on Twitter tweets and use those indicators to explain Bitcoin return. Our analysis uncovers that sentiment indicators crafted with ChatGPT4o/ChatGPT3.5 significantly affect Bitcoin returns, even when accounting for a broad array of control variables and other pre-established sentiment indicators. These insights imply that ChatGPT4o/ChatGPT3.5 could empower financial professionals to discover sentiment information from Twitter tweets that were overlooked by previously introduced sentiment indicators concerning Bitcoin.

This paper relies on shocks to the CAPTCHA technology to study the effects of social media bot activity. We observe that bots drive a large amount of attention to corporate accounts around earnings announcements. In line with theoretical research, bot activity is a significant predictor of investor disagreement, which is persistent long-term. Moreover, social media bots increase analyst dispersion for the following earnings announcement. The failure of the CAPTCHA implementation technology leads to negative abnormal returns, and the failure of CAPTCHA bypassing technology leads to positive abnormal returns

Do firms' tweets improve stock price discovery at quarterly earnings announcements? We address this question using a comprehensive sample of tweets released by S&P 1500 firms at quarterly earnings announcements from 2008 through 2021. Firms' tweets are associated with stronger stock price and volume reactions to earnings announcements. In addition, firms' tweets reduce investor uncertainty, increase the timeliness and efficiency with which stock prices reflect information, and reduce the post-earnings announcement drift. We document that firms' tweets improve stock price discovery by enhancing firm visibility and increasing retail investor trading, which facilitate faster incorporation of information into stock prices. Our inferences are robust to a host of control variables, including alternative measures of media coverage, and persist in both a propensity score matched sample and an entropy balanced sample, where firms that use Twitter are matched with similar firms that do not. Our findings are of interest to regulators who wish to improve the informativeness of security prices, investors who are interested in information that affects prices and volume, and managers who seek channels to communicate with investors. 

Using events of high-profile U.S. accounting frauds, we examine investor reactions of more than 18,000 unrelated firms across 40 countries.  Despite having no direct economic ties with the U.S. fraud firm whether at the firm, industry or country level, unrelated firms experience stock price decreases during the fraud events. The negative stock price reaction is most evident among firms with higher accruals, indicating investor apprehension about potential earnings overstatements. Examining how country-level characteristics shape investor reactions worldwide, we find that the negative reactions are moderated by legal institutions such as penalties on misstatements, infrastructure facilitating news flows, and perceived similarities in accounting and auditing practices. Analyses examining high-profile accounting frauds in countries outside of the U.S. confirm the global contagion of accounting concerns. Overall, our findings highlight a far-reaching contagion effect where accounting frauds, even from an unrelated source in one country, can trigger worldwide investor concerns about similar accounting failures elsewhere.

Social media-fueled retail trading poses new risk to institutional investors. I develop a model to analyze the origin and pricing of this risk. Retail investors participate in a social network with concentrated linkages, which means their idiosyncratic sentiment shocks can lead to aggregate fluctuations in retail sentiment. These fluctuations shift investor composition, which in turn determines the price of retail sentiment risk. I calibrate the model to match price, quantity, and sentiment dynamics around the January 2021 short squeeze. Using the calibrated model, I quantify the impact of evolving social network topology on asset prices.

This paper examines the capital market impacts of CEO podcasts-an emerging but informal communication channel. We distinguish episodes where CEOs are discussed from those where CEOs appear in-person as interview guests. CEO podcasts elicit significant market responses, including increased trading activity and abnormal returns, particularly among retail investors. However, CEO-interview podcasts generate stronger effects than mentions, likely due to their perceived intentionality and potential to foster para-social relationships. We also document heightened investor attention-measured via Wikipedia searches-especially for CEOs following interview episodes. Notably, CEO-interview podcasts following negative earnings announcements are associated with significant return reversals, suggesting they may influence perceptions through a more personal and conversational tone. Consistent with this, content analysis shows these podcasts are more positive and subjective, and less complex than corresponding earnings call transcripts. Our findings suggest that podcasts, though unconventional and potentially non-compliant with disclosure regulations, serve as an impactful medium through which CEOs can shape narratives, engage investors, and potentially soften the market's response to bad news.

We examine the simultaneous peer effects of co-workers, family, and neighbors in financialbehavior using Danish registry data. We find that neighbors exert the strongest influence, followed by co-workers and family members. Peer effects are stronger for stocks than for mutual funds, and among experienced investors. While co-workers primarily influence buying decisions, neighbors affect both buying and selling, suggesting distinct channels of influence across peer groups. A multi-layer network model formalizes our empirical results, showing that an investor's trading activity depends on her centrality within and across network layers. Our findings provide new insights into the drivers and implications of peer effects in financial markets.

Audit firms and regulators have both commented extensively on the potential for new sources of data to transform the audit process. Focusing on auditors' going-concern opinions, we use deep learning to measure the "bearishness" of posts on social media and find it strongly predicts the likelihood of firm failure. This association is incremental to other market-based signals, such as a firm's default likelihood or short interest. Interestingly, this signal appears largely orthogonal to an auditor's going-concern opinion, implying that social media predicts future events that precipitate failure not fully considered by auditors. While we fail to observe a direct association between bearishness and going concern opinions, our evidence does suggest that going concern accuracy improves with bearishness. Finally, we consider potential channels for these results and find that bearishness foreshadows difficulties in raising capital, predicting the likelihood of future credit downgrades and equity issuances. Our evidence should be informative to regulators and audit firms, both of whom are currently evaluating the usefulness of "new" data to auditors.

This article examines the 2022 South Korean bond market crisis triggered by Heungkuk Life Insurance's decision not to exercise a call option on its perpetual bonds. Although the decision was legally sound and financially rational, markets interpreted it as a betrayal of an unwritten convention, triggering a sharp collapse in bond prices and forcing a rapid capitulation. We analyze this episode through the concept of narrative performativity, highlighting how market narratives do not merely interpret information but actively constitute market reality by disciplining actors and enforcing conformity. Drawing on theories of performativity from speech act philosophy, economic sociology, and anthropological studies of mythology, we show how narratives acquire material force in shaping financial outcomes. The Heungkuk case reveals that financial markets function less as transparent information-processing devices than as arenas of myth, ritual, and power, where collective narratives are continually verified and ritually reaffirmed. Understanding financial markets in this way underscores that belief, narrative authority, and symbolic efficacy are not peripheral but central forces shaping contemporary finance.

This study investigates the formation and economic consequences of "echo chambers" within sell-side analysts' information networks. Departing from the view of analysts as independent agents, we construct a novel multilayer network that integrates three distinct channels of professional interaction: team cooperation, intra-firm affiliation, and stock co-coverage. We aggregate these layers using an optimization procedure that recovers their relative importance in shaping belief similarity, and then apply community detection to identify tightly-knit analyst cliques. We document that these cliques are structurally persistent and function as potent echo chambers: analysts within the same clique exhibit significant convergence in their forecast biases and optimism, an effect distinct from strategic herding. We also find that firms covered by a greater number of distinct cliques experience a deteriorated information environment, manifested as higher forecast dispersion and increased stock price crash risk. Our findings highlight that the social structure of information intermediaries can foster insular belief systems, with significant negative externalities for market efficiency.