Throwing curveballs: A language‐based model of curveball questions in quarterly earnings calls uncovers their consequences and antecedents

Abstract Research Summary In evaluative contexts, evaluatees typically seek to present themselves in a favorable light, while evaluators ask penetrating questions to assess these claims. Here we develop a framework to identify curveball questions : ones that are on‐topic yet perplexing (i.e., difficult to predict) relative to past discourse. We develop a language‐based measure of curveball questions and apply it to a corpus of quarterly earnings calls. After validating this question‐level measure, we next demonstrate that a call‐level curveball measure predicts absolute returns, absolute abnormal returns, and changes in a firm's average analyst rating. Finally, we identify the types of analysts who are most likely to pose curveball questions, the types of firms that are most likely to receive them, and the conditions under which they tend to arise. Managerial Summary Even a carefully crafted presentation can be derailed by a challenging question. What makes a question challenging in ways that can be disruptive and how can such a question be measured? We propose that such questions, which we label curveballs , are on‐topic, and thus difficult to dismiss or deflect, yet difficult to predict based on prior knowledge. We harness the tools of computational linguistics to develop a measure of curveball questions and apply it to the context of quarterly earnings calls. We show that this measure predicts consequential economic outcomes and highlight the conditions under which curveball questions tend to arise. Our measurement strategy can be readily extended to other evaluative contexts such as job interviews and venture capital pitches.