Published Papers

We find that mutual funds whose managers are socially connected with firm auditors hold more shares of these firms and generate superior portfolio returns. Cross-sectional results reveal that the relation between social connections and mutual fund stockholdings is more pronounced: when the social connections are stronger, when the auditor is in a better position or has stronger incentives to acquire private information, when the fund manager exercises more power, for small audit firms, for auditors in areas with poor investor protection, and for public firms with greater business opacity or private information. Other results are consistent with fund managers electing to schedule their corporate site visits to coincide with the fieldwork of their connected auditors, as would be expected if fund managers time their visits to meet with these auditors to facilitate information transfer. Additionally, we observe associations between fund trading prior to earnings surprises and audit opinions, and the presence of social connections between fund managers and firm auditors. Finally, we show that mutual funds and firms in which they invest tend to appoint connected auditors and pay them higher fees. Collectively, we document empirical patterns that would arise if socially connected auditors and mutual fund managers share information.

We identify a specific organizational resource in brokerage houses--information sharing among analyst colleagues who cover economically related industries along a supply chain. After controlling for brokerage selection effects, we show evidence consistent with the benefit of this resource to analyst research performance. Specifically, we find that analysts whose colleagues cover more economically connected industries have better research performance, especially when their colleagues produce higher-quality research. We further show that colleagues' coverage of downstream (upstream) industries is positively related to the accuracy of only analysts' revenue (expense) forecasts and that analysts and their highly connected colleagues tend to issue earnings forecast revisions contemporaneously. Last, we find that analysts with economically connected colleagues tend to have a higher level of industry specialization. Overall, our findings suggest that analysts rely on organizational resources to produce high-quality research. Hence, a portion of their performance and reputation is not transferable across employers.

Between 2010 and 2017, Chinese investors used an investor interactive platform (IIP) to ask public companies around 2.5 million questions, the vast majority of which received a reply within two weeks. We analyze these IIP dialogues using a BERT-based algorithm and provide preliminary evidence on their causes and consequences. Our analyses show most questions reflect investors' difficulties in processing information already in the public domain. Controlling for other news, higher IIP activity is associated with increases in trading volume, return volatility, market liquidity, and price informativeness as well as decreases in bid-ask spread. Financial statement-related postings increase around the adoption of new accounting standards. Collectively, our results show that investors face significant information processing costs but that IIP activities help reduce these costs, leading to improvements in stock price formation.

Investors are central to the incorporation of firm information in capital markets, yet it is challenging to observe the particular information they use and struggle with. Lee and Zhong (2022) use online investor interactions with Chinese public firms to document evidence that investors face significant processing costs. They find that when investor interactions occur, capital markets behave as if the information environment has improved, with increased trading activity, liquidity, and timely pricing of the quarter's earnings in returns. My discussion highlights the contributions of Lee and Zhong's findings to the processing cost, retail investor, and investor interactions literatures. I also describe empirical challenges faced by this and similar studies. I encourage using the details of interactions to disentangle the nature of processing costs and to increase support for causal conclusions more generally. Finally, I note several topics related to investor inter- action that would benefit from further research.

Consumers often become "stuck in a rabbit hole" when consuming media. They may watch several YouTube videos in the same category or view several thematically similar artistic images on Instagram in a row, finding it difficult to stop. What causes individuals to choose to consume additional media on a topic that is similar to (vs. different from) what they just experienced? The authors examine a novel antecedent: the consecutive consumption of multiple similar media. After viewing multiple similar media consecutively, more consumers choose to (1) view additional similar media over dissimilar media or (2) complete a dissimilar activity entirely, even when the prior consumption pattern is externally induced. The rabbit hole effect occurs because of increased accessibility of the shared category: when a category is more accessible, people feel immersed in it and anticipate that future options within that category will be more enjoyable. The authors identify three characteristics of media consumption that contribute to the rabbit hole effect by increasing category accessibility: similarity, repetition, and consecutiveness of prior media consumption. This research contributes to literature on technology, choice, and variety seeking, and it offers implications for increasing (vs. slowing) similar consumption.

Swearing can violate norms and thereby offend consumers. Yet the prevalence of swear word use suggests that an offensiveness perspective may not fully capture their impact in marketing. This article adopts a linguistic perspective to develop and test a model of how, why, and when swear word use affects consumers in online word of mouth. In two field data sets and four experiments, the authors show that relative to reviews with no swear words, or with non-swear-word synonyms (e.g., super), reviews with swear words (e.g., damn) impact review readers. First, reviews with swear words are rated as more helpful. Second, when a swear word qualifies a desirable [undesirable] product attribute, readers' attitudes toward the product increase [decrease] (e.g., "This dishwasher is damn quiet [loud]!"). Swear words impact readers because they convey meaning about (1) the reviewer and (2) the topic (product) under discussion. These two meanings function as independent, parallel mediators that drive the observed effects. Further, these effects are moderated by swear word number and style: they do not emerge when a review contains many swear words and are stronger for uncensored and euphemistic swear words (e.g., darn) than censored swear words (e.g., d*mn). Overall, swear words in reviews provide value to readers-and review platforms-because they efficiently and effectively convey two meanings.

This research examines the interaction effect of two dimensions of preference on social contagion: preference similarity between a consumer (i.e., who seeks recommendation) and a peer (i.e., who potentially provides recommendation) and the fit of an experience good with the consumer's preference. For empirical analyses, the authors collected rich information from Last.fm, a music social networking website, including individual users' music play histories, friendship information, social tags (i.e., user-generated keywords associated with artists and songs), and new song profiles. The results show that consumers' trial of a song that fits less with their preference is influenced more by peers with similar preferences. By contrast, consumers' trial of a song that fits more with their preference is influenced more by peers with dissimilar preferences. This research enriches the understanding of the nuanced role of preference in social contagion and offers managerial implications to better leverage social dynamics.

The authors study how strangers become friends within an evolving online social network. By modeling the coevolution of individual users' friendship tie formations (when and with whom) and their concurrent online activities, the authors uncover important drivers underlying individuals' friendship decisions and, at the same time, quantify the resulting peer effects on individuals' actions. They estimate their model using a novel data set capturing the continuous development of a network and users' entire action histories within the network. The results reveal that similarity (homophily) with a potential friend, the properties of a potential friend's network, and the potential friend's domain expertise all play a role in friendship formation. Via prediction exercises, the authors find that stimulating anime watching is the most effective sitewide intervention, which leads to the highest overall site traffic and the largest number of active users, and that recommending a friend of a friend as a potential friend is the most effective strategy in stimulating friendship tie formation. In contrast to the common finding for static networks, the results indicate that seeding to users with the most friends is not the most effective strategy to increase users' activity levels in an evolving network.

Recommender systems on online platforms are often accused of polarizing user attention and consumption. The authors examine this phenomenon using a quasi-experiment conducted by Zhihu, the largest online knowledge-sharing platform (or Q&A community) in China. Zhihu originally used a content-based filtering algorithm, which recommends content to users on the basis of the topics to which each user has subscribed. After more than a year, Zhihu moved to a social filtering algorithm, which recommends content with which users' social connections are already engaged. The authors find that this algorithm change increased the creation of social ties by approximately 15% but decreased question subscriptions by 20% and answer contributions by 23%. The authors show that users' increased social interests mainly involved following popular users, leading to a greater concentration of social interests on the platform. However, users' topical interests became less concentrated, as popular topics received significantly fewer subscribers than unpopular topics. The authors explain these findings by exploring the underlying mechanism. They show that compared with content-based filtering algorithms, social filtering algorithms are more likely to expose general users to content consumed by their followees, who are more interested in niche topics than general users are.

Marketers frequently create social media content (i.e., firm-generated content; FGC) to ignite interest in new movies. Thus, there is a clear need to understand the magnitude and heterogeneity of the effect of FGC on movie demand and associated user-generated content (UGC). The authors empirically examine the complex interactions among FGC, UGC, and sales using social media (tweet) data that are normally available to firms. They investigate two potential mechanisms by which FGC may drive box office revenues: (1) a direct mechanism, such that users who see FGC directly drive revenue, and (2) an indirect "ripple effect," by which FGC increases movie-related UGC, which then drives consumption. By analyzing 145,502 firm-generated and 5.9 million user-generated Twitter posts associated with 159 movies, the authors find a positive and significant effect of FGC on movie sales, which UGC fully mediates, which supports the indirect ripple effect reasoning. Impressions of FGC by followers of firm accounts, as opposed to nonfollowers of firm accounts, mainly drive the effect of FGC on UGC. In addition, FGC by movie accounts is more effective than that by actors and studios. Firms' regular posts with a movie-specific hashtag are more effective than replies, retweets, and posts without the hashtag. The finding of the ripple effect suggests that movie executives should focus on creating FGC that sparks conversations among followers when new movies are released.

Consumers often communicate their attitudes and opinions with others, and such word of mouth has an important impact on what others think, buy, and do. But might the way consumers communicate their attitudes (i.e., through speaking or writing) shape the attitudes they express? And, as a result, the impact of what they share? While a great deal of research has begun to examine drivers of word of mouth, there has been less attention to how communication modality might shape sharing. Six studies, conducted in the laboratory and field, demonstrate that compared to speaking, writing leads consumers to express less emotional attitudes. The effect is driven by deliberation. Writing offers more time to deliberate about what to say, which reduces emotionality. The studies also demonstrate a downstream consequence of this effect: by shaping the attitudes expressed, the modality consumers communicate through can influence the impact of their communication. This work sheds light on word of mouth, effects of communication modality, and the role of language in communication.

Social media may encourage novel ways of signaling that involve different purchase types (experiential vs. material), signaling frequencies (multiple vs. single signals), and other features unique to social media (e.g., hashtags). This work examines how purchase signals are received on social media and how these signaling variations affect signal receivers' perceptions of the authenticity of social media posts as well as the overall impressions receivers form of the signal sender. Data collected across six experiments show multiple material purchase signals lead to more negative impressions compared to multiple experiential purchase signals. Signal receivers perceive multiple material purchase posts as less authentic, which dampens their impressions of the signal sender. In line with this mechanism, the impression premium of experiential purchase signals disappears when receivers use other cues (monetary mentions, other users' comments, and marketer associations via hashtags) to infer a signal's lack of authenticity. Additional data also document downstream consequences on engagement. This work contributes theoretically to research in both signaling and social media and improves the understanding of substantive situations in which consumers' objectives of curating a positive image and creating engagement with their posts, collide with marketers' objectives of encouraging user-generated content and word of mouth.