Narratives Contextualizing Numeric Disclosures: Insights From Earnings Calls

Abstract

We investigate how narrative disclosures during earnings conference calls (ECCs) provide context for quantitative numeric disclosures in quarterly earnings releases, enhancing their informativeness. Drawing on a large sample of 34,918 quarterly ECCs from 1621 US‐listed firms from 2007 to 2020, we extract interpretable textual attributes—such as tone, specificity, novelty, and thematic content—from introductions, analyst questions, and management answers. Using stochastic gradient boosting, we model the nonlinear and interactive effects of these attributes with firm fundamentals and financial metrics. Our results reveal that narrative disclosures substantially improve the ability of firm fundamentals and financial metrics to explain market responses, both immediately and over longer horizons. We complement our analysis with Shapley additive explanations (SHAP) to identify which textual features most influence investor reactions and how their effects vary across contexts. Immediate market responses are primarily attributable to direct cues, such as the tone, while longer term responses increasingly reflect more complex content, such as strategic themes and support activities. These effects are especially pronounced when earnings surprises are either very negative or close to zero, and for firms with lower institutional ownership. They have also become more salient over time, suggesting that shifts in investor attention and analytical sophistication may be reshaping how financial narratives are processed. Our results contribute to understanding the contextual role of narrative disclosures in financial communication and illustrate the value of explainable AI in accounting research.