The Cost of Asking Friendly Questions in Earnings Conference Calls
Solo-authored Job Market Paper
Abstract: Why do analysts ask unfriendly questions in conference calls despite the risk of losing management access? I examine this question by testing whether the friendliness of analysts influences management team’s narrative control in Q&A section. Specifically, I hypothesize that CEOs who face a friendlier analyst pool dominate answers because CEOs are more willing to engage directly when they anticipate less confrontational questioning. Empirically, I use brokerage house mergers as plausibly exogenous shocks to the analyst pool, and measure CEOs’ answer dominance as whether the CEO contributes twice more than other participating managers. I find that when an analyst leaves, the CEOs are less likely to dominate the answers when the analyst used to provide friendlier questions. I further find that, CEO answer dominance is not associated with additional information, but is associated with more obfuscation, forward-looking language, and topic shifting. Additionally, I find that the positive association between obfuscation and illiquidity is stronger when the CEO dominates Q&A answers. Collectively, these results show that when analysts avoid unfriendly questions, the CEO becomes dominant in answers, which creates greater information opacity to market participants.
Generative AI, Investor Composition, and Information Barriers
Coauthored with Encarna Guillamón and María Gutiérrez
Best Paper Award, 2025 ASEPUC Doctoral Colloquium
Abstract: Whether generative AI narrows or widens informational advantages across investor types remains unclear. We study how managers of firms with different investor bases adjust their qualitative disclosure after the introduction of generative AI. Using the release of ChatGPT 3.5 as a common shock to information-processing technology, we decompose linguistic complexity into an information component (related to business complexity) and an obfuscation component (the residual). On average, the information component decreases after ChatGPT's release. The reduction is larger for firms with greater retail exposure and more concentrated institutional ownership, and smaller when index funds account for a larger share of mutual fund holdings and when analyst following is higher. By contrast, we find no systematic change in the obfuscation component and no differential response by overall institutional ownership. Overall, post-ChatGPT disclosure changes vary systematically with investor composition, consistent with a disclosure-supply response that may preserve sophisticated investors' relative informational advantages.
Conference Call Narratives after Voting Setback
Coauthored with Encarna Guillamón
Best Paper Award and AJICEDE Special Mention, 2026 ASEPUC Doctoral Colloquium
Abstract: Why is a failed proposal in the annual shareholder’s meeting likely to be approved next year? We study managers' communications during conference calls after they are defeated by investors' votes from the previous meeting. Using the Aggregated Regression Discontinuity Design (ARDD), we find that when managers fail at least one proposal in the previous meeting, they are less likely to pick bearish analysts to ask the first question. Moreover, negative-tone questions during the Q&A session receive shorter responses from managers. Our findings imply managers' incentives through conference call conversations to win back support from voters, implying that persuasion is an important factor in conference call narrative strategy.